Networkx Reference

NetworkX Reference
Release 3.0rc2.dev0
Aric Hagberg, Dan Schult, Pieter Swart
Dec 11, 2022

CONTENTS
1 Introduction 1
1.1 NetworkX Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Graph Creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Graph Reporting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.6 Drawing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.7 Data Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Graph types 7
2.1 Which graph class should I use? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Basic graph types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Graph Views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
2.4 Core Views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
2.5 Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
2.6 Backends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
3 Algorithms 169
3.1 Approximations and Heuristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
3.2 Assortativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
3.3 Asteroidal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
3.4 Bipartite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
3.5 Boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
3.6 Bridges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
3.7 Centrality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
3.8 Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
3.9 Chordal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
3.10 Clique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
3.11 Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
3.12 Coloring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
3.13 Communicability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
3.14 Communities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
3.15 Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
3.16 Connectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
3.17 Cores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
3.18 Covering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
3.19 Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
3.20 Cuts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
3.21 D-Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
3.22 Directed Acyclic Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
i
3.23 Distance Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
3.24 Distance-Regular Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
3.25 Dominance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
3.26 Dominating Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
3.27 Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
3.28 Eulerian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
3.29 Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
3.30 Graph Hashing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
3.31 Graphical degree sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
3.32 Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
3.33 Hybrid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
3.34 Isolates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
3.35 Isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
3.36 Link Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
3.37 Link Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
3.38 Lowest Common Ancestor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
3.39 Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536
3.40 Minors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540
3.41 Maximal independent set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 548
3.42 non-randomness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550
3.43 Moral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551
3.44 Node Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552
3.45 Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
3.46 Planarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
3.47 Planar Drawing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 610
3.48 Graph Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 611
3.49 Reciprocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 614
3.50 Regular . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 615
3.51 Rich Club . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 617
3.52 Shortest Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 618
3.53 Similarity Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665
3.54 Simple Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 677
3.55 Small-world . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684
3.56 s metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 687
3.57 Sparsifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 688
3.58 Structural holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689
3.59 Summarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 692
3.60 Swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696
3.61 Threshold Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 699
3.62 Tournament . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 701
3.63 Traversal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705
3.64 Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723
3.65 Triads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 747
3.66 Vitality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 753
3.67 Voronoi cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754
3.68 Wiener index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 755
4 Functions 757
4.1 Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 757
4.2 Nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 766
4.3 Edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 768
4.4 Self loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 769
4.5 Attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 771
4.6 Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 777
ii
4.7 Freezing graph structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 778
5 Graph generators 781

5.1 Atlas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 781
5.2 Classic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783
5.3 Expanders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 793
5.4 Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 795
5.5 Small . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 798
5.6 Random Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 810
5.7 Duplication Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 825
5.8 Degree Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 827
5.9 Random Clustered . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 833
5.10 Directed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 835
5.11 Geometric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 839
5.12 Line Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 849
5.13 Ego Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 852
5.14 Stochastic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 853
5.15 AS graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 853
5.16 Intersection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854
5.17 Social Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 856
5.18 Community . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 857
5.19 Spectral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 868
5.20 Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 869
5.21 Non Isomorphic Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 872
5.22 Triads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873
5.23 Joint Degree Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 874
5.24 Mycielski . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 878
5.25 Harary Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 879
5.26 Cographs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 881
5.27 Interval Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 882
5.28 Sudoku . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 883
6 Linear algebra 885

6.1 Graph Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 885
6.2 Laplacian Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 887
6.3 Bethe Hessian Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 891
6.4 Algebraic Connectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 892
6.5 Attribute Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 896
6.6 Modularity Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 900
6.7 Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 902
7 Converting to and from other data formats 907

7.1 To NetworkX Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 907
7.2 Dictionaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 908
7.3 Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 910
7.4 Numpy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 912
7.5 Scipy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 917
7.6 Pandas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 920
8 Relabeling nodes 927

8.1 Relabeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 927
9 Reading and writing graphs 931

9.1 Adjacency List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 931
9.2 Multiline Adjacency List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 935
iii
9.3 Edge List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 940
9.4 GEXF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 947
9.5 GML . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 950
9.6 GraphML . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 956
9.7 JSON . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 961
9.8 LEDA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 970
9.9 SparseGraph6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 971
9.10 Pajek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 979
9.11 Matrix Market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 981
10 Drawing 985
10.1 Matplotlib . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 985
10.2 Graphviz AGraph (dot) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1001
10.3 Graphviz with pydot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1005
10.4 Graph Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1008
11 Randomness 1019
12 Exceptions 1021
13 Utilities 1023
13.1 Helper Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1023
13.2 Data Structures and Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1027
13.3 Random Sequence Generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1028
13.4 Decorators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1030
13.5 Cuthill-Mckee Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1042
13.6 Mapped Queue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1044
14 Glossary 1047
A Tutorial 1049
A.1 Creating a graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1049
A.2 Nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1049
A.3 Edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1050
A.4 Examining elements of a graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1051
A.5 Removing elements from a graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1051
A.6 Using the graph constructors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1052
A.7 What to use as nodes and edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1052
A.8 Accessing edges and neighbors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1052
A.9 Adding attributes to graphs, nodes, and edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053
A.10 Directed graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1054
A.11 Multigraphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1055
A.12 Graph generators and graph operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1055
A.13 Analyzing graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1057
A.14 Drawing graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1057
Bibliography 1061
Python Module Index 1085
Index 1089
iv
CHAPTER
ONE
INTRODUCTION
The structure of NetworkX can be seen by the organization of its source code. The package provides classes for graph ob-
jects, generators to create standard graphs, IO routines for reading in existing datasets, algorithms to analyze the resulting
networks and some basic drawing tools.
Most of the NetworkX API is provided by functions which take a graph object as an argument. Methods of the graph
object are limited to basic manipulation and reporting. This provides modularity of code and documentation. It also
makes it easier for newcomers to learn about the package in stages. The source code for each module is meant to be easy
to read and reading this Python code is actually a good way to learn more about network algorithms, but we have put a
lot of effort into making the documentation sufficient and friendly. If you have suggestions or questions please contact us
by joining the NetworkX Google group.
Classes are named using CamelCase (capital letters at the start of each word). functions, methods and variable names
are lower_case_underscore (lowercase with an underscore representing a space between words).
1.1 NetworkX Basics
After starting Python, import the networkx module with (the recommended way)
>>> import networkx as nx
To save repetition, in the documentation we assume that NetworkX has been imported this way.
If importing networkx fails, it means that Python cannot find the installed module. Check your installation and your
PYTHONPATH.
The following basic graph types are provided as Python classes:
Graph
This class implements an undirected graph. It ignores multiple edges between two nodes. It does allow self-loop
edges between a node and itself.
DiGraph
Directed graphs, that is, graphs with directed edges. Provides operations common to directed graphs, (a subclass
of Graph).
MultiGraph
A flexible graph class that allows multiple undirected edges between pairs of nodes. The additional flexibility leads
to some degradation in performance, though usually not significant.
MultiDiGraph
A directed version of a MultiGraph.
Empty graph-like objects are created with
1
NetworkX Reference, Release 3.0rc2.dev0
>>> G = nx.Graph()
>>> G = nx.DiGraph()
>>> G = nx.MultiGraph()
>>> G = nx.MultiDiGraph()
All graph classes allow any hashable object as a node. Hashable objects include strings, tuples, integers, and more.
Arbitrary edge attributes such as weights and labels can be associated with an edge.
The graph internal data structures are based on an adjacency list representation and implemented using Python dictionary
datastructures. The graph adjacency structure is implemented as a Python dictionary of dictionaries; the outer dictionary
is keyed by nodes to values that are themselves dictionaries keyed by neighboring node to the edge attributes associated
with that edge. This “dict-of-dicts” structure allows fast addition, deletion, and lookup of nodes and neighbors in large
graphs. The underlying datastructure is accessed directly by methods (the programming interface “API”) in the class
definitions. All functions, on the other hand, manipulate graph-like objects solely via those API methods and not by
acting directly on the datastructure. This design allows for possible replacement of the ‘dicts-of-dicts’-based datastructure
with an alternative datastructure that implements the same methods.
1.2 Graphs
The first choice to be made when using NetworkX is what type of graph object to use. A graph (network) is a collection
of nodes together with a collection of edges that are pairs of nodes. Attributes are often associated with nodes and/or
edges. NetworkX graph objects come in different flavors depending on two main properties of the network:
• Directed: Are the edges directed? Does the order of the edge pairs (u, v) matter? A directed graph is specified
by the “Di” prefix in the class name, e.g. DiGraph(). We make this distinction because many classical graph
properties are defined differently for directed graphs.
• Multi-edges: Are multiple edges allowed between each pair of nodes? As you might imagine, multiple edges re-
quires a different data structure, though clever users could design edge data attributes to support this functionality.
We provide a standard data structure and interface for this type of graph using the prefix “Multi”, e.g., Multi-
Graph().
The basic graph classes are named: Graph, DiGraph, MultiGraph, and MultiDiGraph
1.2.1 Nodes and Edges
The next choice you have to make when specifying a graph is what kinds of nodes and edges to use.
If the topology of the network is all you care about then using integers or strings as the nodes makes sense and you need
not worry about edge data. If you have a data structure already in place to describe nodes you can simply use that structure
as your nodes provided it is hashable. If it is not hashable you can use a unique identifier to represent the node and assign
the data as a node attribute.
Edges often have data associated with them. Arbitrary data can be associated with edges as an edge attribute. If the data
is numeric and the intent is to represent a weighted graph then use the ‘weight’ keyword for the attribute. Some of the
graph algorithms, such as Dijkstra’s shortest path algorithm, use this attribute name by default to get the weight for each
edge.
Attributes can be assigned to an edge by using keyword/value pairs when adding edges. You can use any keyword to name
your attribute and can then query the edge data using that attribute keyword.
Once you’ve decided how to encode the nodes and edges, and whether you have an undirected/directed graph with or
without multiedges you are ready to build your network.
2 Chapter 1. Introduction
1.3 Graph Creation
NetworkX graph objects can be created in one of three ways:

• Graph generators—standard algorithms to create network topologies.
• Importing data from pre-existing (usually file) sources.
• Adding edges and nodes explicitly.
Explicit addition and removal of nodes/edges is the easiest to describe. Each graph object supplies methods to manipulate
the graph. For example,

>>> G = nx.Graph()
>>> G.add_edge(1, 2) # default edge data=1
>>> G.add_edge(2, 3, weight=0.9) # specify edge data
Edge attributes can be anything:
>>> import math

>>> G.add_edge('y', 'x', function=math.cos)
>>> G.add_node(math.cos) # any hashable can be a node
You can add many edges at one time:
>>> elist = [(1, 2), (2, 3), (1, 4), (4, 2)]
>>> G.add_edges_from(elist)
>>> elist = [('a', 'b', 5.0), ('b', 'c', 3.0), ('a', 'c', 1.0), ('c', 'd', 7.3)]
>>> G.add_weighted_edges_from(elist)
See the Tutorial for more examples.

Some basic graph operations such as union and intersection are described in the operators module documentation.
Graph generators such as binomial_graph() and erdos_renyi_graph() are provided in the graph generators
subpackage.
For importing network data from formats such as GML, GraphML, edge list text files see the reading and writing graphs
subpackage.
1.4 Graph Reporting
Class views provide basic reporting of nodes, neighbors, edges and degree. These views provide iteration over the proper-
ties as well as membership queries and data attribute lookup. The views refer to the graph data structure so changes to the
graph are reflected in the views. This is analogous to dictionary views in Python 3. If you want to change the graph while
iterating you will need to use e.g. for e in list(G.edges):. The views provide set-like operations, e.g. union
and intersection, as well as dict-like lookup and iteration of the data attributes using G.edges[u, v]['color'] and
for e, datadict in G.edges.items():. Methods G.edges.items() and G.edges.values() are
familiar from python dicts. In addition G.edges.data() provides specific attribute iteration e.g. for e, e_color
in G.edges.data('color'):.
The basic graph relationship of an edge can be obtained in two ways. One can look for neighbors of a node or one can
look for edges. We jokingly refer to people who focus on nodes/neighbors as node-centric and people who focus on edges
as edge-centric. The designers of NetworkX tend to be node-centric and view edges as a relationship between nodes.
You can see this by our choice of lookup notation like G[u] providing neighbors (adjacency) while edge lookup is G.
edges[u, v]. Most data structures for sparse graphs are essentially adjacency lists and so fit this perspective. In the
1.3. Graph Creation 3

end, of course, it doesn’t really matter which way you examine the graph. G.edges removes duplicate representations
of undirected edges while neighbor reporting across all nodes will naturally report both directions.
Any properties that are more complicated than edges, neighbors and degree are provided by functions. For example nx.
triangles(G, n) gives the number of triangles which include node n as a vertex. These functions are grouped in
the code and documentation under the term algorithms.
1.5 Algorithms
A number of graph algorithms are provided with NetworkX. These include shortest path, and breadth first search (see
traversal), clustering and isomorphism algorithms and others. There are many that we have not developed yet too. If you
implement a graph algorithm that might be useful for others please let us know through the NetworkX Google group or
the Github Developer Zone.
As an example here is code to use Dijkstra’s algorithm to find the shortest weighted path:
>>> G = nx.Graph()
>>> e = [('a', 'b', 0.3), ('b', 'c', 0.9), ('a', 'c', 0.5), ('c', 'd', 1.2)]
>>> G.add_weighted_edges_from(e)
>>> print(nx.dijkstra_path(G, 'a', 'd'))
['a', 'c', 'd']
1.6 Drawing
While NetworkX is not designed as a network drawing tool, we provide a simple interface to drawing packages and some
simple layout algorithms. We interface to the excellent Graphviz layout tools like dot and neato with the (suggested)
pygraphviz package or the pydot interface. Drawing can be done using external programs or the Matplotlib Python
package. Interactive GUI interfaces are possible, though not provided. The drawing tools are provided in the module
drawing.
The basic drawing functions essentially place the nodes on a scatterplot using the positions you provide via a dictionary
or the positions are computed with a layout function. The edges are lines between those dots.
>>> import matplotlib.pyplot as plt

>>> G = nx.cubical_graph()
>>> subax1 = plt.subplot(121)
>>> nx.draw(G) # default spring_layout
>>> nx.draw(G, pos=nx.circular_layout(G), node_color='r', edge_color='b')
See the examples for more ideas.
1.7 Data Structure
NetworkX uses a “dictionary of dictionaries of dictionaries” as the basic network data structure. This allows fast lookup
with reasonable storage for large sparse networks. The keys are nodes so G[u] returns an adjacency dictionary keyed by
neighbor to the edge attribute dictionary. A view of the adjacency data structure is provided by the dict-like object G.
adj as e.g. for node, nbrsdict in G.adj.items():. The expression G[u][v] returns the edge attribute
dictionary itself. A dictionary of lists would have also been possible, but not allow fast edge detection nor convenient
storage of edge data.
Advantages of dict-of-dicts-of-dicts data structure:
• Find edges and remove edges with two dictionary look-ups.
• Prefer to “lists” because of fast lookup with sparse storage.
• Prefer to “sets” since data can be attached to edge.
• G[u][v] returns the edge attribute dictionary.
• n in G tests if node n is in graph G.
• for n in G: iterates through the graph.
• for nbr in G[n]: iterates through neighbors.
1.7. Data Structure 5

As an example, here is a representation of an undirected graph with the edges (A, B) and (B, C).
>>> G = nx.Graph()
>>> G.add_edge('A', 'B')
>>> G.add_edge('B', 'C')
>>> print(G.adj)
{'A': {'B': {}}, 'B': {'A': {}, 'C': {}}, 'C': {'B': {}}}
The data structure gets morphed slightly for each base graph class. For DiGraph two dict-of-dicts-of-dicts structures are
provided, one for successors (G.succ) and one for predecessors (G.pred). For MultiGraph/MultiDiGraph we use a
dict-of-dicts-of-dicts-of-dicts1 where the third dictionary is keyed by an edge key identifier to the fourth dictionary which
contains the edge attributes for that edge between the two nodes.
Graphs provide two interfaces to the edge data attributes: adjacency and edges. So G[u][v]['width'] is the same
as G.edges[u, v]['width'].
>>> G = nx.Graph()
>>> G.add_edge(1, 2, color='red', weight=0.84, size=300)
>>> print(G[1][2]['size'])
300
>>> print(G.edges[1, 2]['color'])
red
• ;
• ;
• .
1 “It’s dictionaries all the way down.”
CHAPTER
TWO
GRAPH TYPES
NetworkX provides data structures and methods for storing graphs.

All NetworkX graph classes allow (hashable) Python objects as nodes and any Python object can be assigned as an edge
attribute.
The choice of graph class depends on the structure of the graph you want to represent.
2.1 Which graph class should I use?
Networkx Class Type Self-loops allowed Parallel edges allowed

Graph undirected Yes No
DiGraph directed Yes No
MultiGraph undirected Yes Yes
MultiDiGraph directed Yes Yes
2.2 Basic graph types
2.2.1 Graph—Undirected graphs with self loops
Overview
class Graph(incoming_graph_data=None, **attr)

Base class for undirected graphs.
A Graph stores nodes and edges with optional data, or attributes.
Graphs hold undirected edges. Self loops are allowed but multiple (parallel) edges are not.
Nodes can be arbitrary (hashable) Python objects with optional key/value attributes, except that None is not allowed
as a node.
Edges are represented as links between nodes with optional key/value attributes.
Parameters
incoming_graph_data
[input graph (optional, default: None)] Data to initialize graph. If None (default) an empty
graph is created. The data can be any format that is supported by the to_networkx_graph()
function, currently including edge list, dict of dicts, dict of lists, NetworkX graph, 2D NumPy
array, SciPy sparse matrix, or PyGraphviz graph.
7
attr
[keyword arguments, optional (default= no attributes)] Attributes to add to graph as key=value
pairs.
See also:
DiGraph
MultiGraph
MultiDiGraph
Examples
Create an empty graph structure (a “null graph”) with no nodes and no edges.
>>> G = nx.Graph()
G can be grown in several ways.

Nodes:
Add one node at a time:
>>> G.add_node(1)
Add the nodes from any container (a list, dict, set or even the lines from a file or the nodes from another graph).
>>> G.add_nodes_from([2, 3])

>>> G.add_nodes_from(range(100, 110))
>>> H = nx.path_graph(10)
>>> G.add_nodes_from(H)
In addition to strings and integers any hashable Python object (except None) can represent a node, e.g. a customized
node object, or even another Graph.
>>> G.add_node(H)
Edges:
G can also be grown by adding edges.
Add one edge,
>>> G.add_edge(1, 2)
a list of edges,
>>> G.add_edges_from([(1, 2), (1, 3)])
or a collection of edges,
>>> G.add_edges_from(H.edges)
If some edges connect nodes not yet in the graph, the nodes are added automatically. There are no errors when
adding nodes or edges that already exist.
Attributes:
8 Chapter 2. Graph types

Each graph, node, and edge can hold key/value attribute pairs in an associated attribute dictionary (the keys must
be hashable). By default these are empty, but can be added or changed using add_edge, add_node or direct ma-
nipulation of the attribute dictionaries named graph, node and edge respectively.
>>> G = nx.Graph(day="Friday")
>>> G.graph
{'day': 'Friday'}
Add node attributes using add_node(), add_nodes_from() or G.nodes
>>> G.add_node(1, time="5pm")

>>> G.add_nodes_from([3], time="2pm")
>>> G.nodes[1]
{'time': '5pm'}
>>> G.nodes[1]["room"] = 714 # node must exist already to use G.nodes
>>> del G.nodes[1]["room"] # remove attribute
>>> list(G.nodes(data=True))
[(1, {'time': '5pm'}), (3, {'time': '2pm'})]
Add edge attributes using add_edge(), add_edges_from(), subscript notation, or G.edges.
>>> G.add_edge(1, 2, weight=4.7)

>>> G.add_edges_from([(3, 4), (4, 5)], color="red")
>>> G.add_edges_from([(1, 2, {"color": "blue"}), (2, 3, {"weight": 8})])
>>> G[1][2]["weight"] = 4.7
>>> G.edges[1, 2]["weight"] = 4
Warning: we protect the graph data structure by making G.edges a read-only dict-like structure. However, you
can assign to attributes in e.g. G.edges[1, 2]. Thus, use 2 sets of brackets to add/change data attributes:
G.edges[1, 2]['weight'] = 4 (For multigraphs: MG.edges[u, v, key][name] = value).
Shortcuts:
Many common graph features allow python syntax to speed reporting.
>>> 1 in G # check if node in graph

True
>>> [n for n in G if n < 3] # iterate through nodes
[1, 2]
>>> len(G) # number of nodes in graph
5
Often the best way to traverse all edges of a graph is via the neighbors. The neighbors are reported as an adjacency-
dict G.adj or G.adjacency()
>>> for n, nbrsdict in G.adjacency():

... for nbr, eattr in nbrsdict.items():
... if "weight" in eattr:
... # Do something useful with the edges
... pass
But the edges() method is often more convenient:
>>> for u, v, weight in G.edges.data("weight"):

... if weight is not None:
... pass
Reporting:
2.2. Basic graph types 9

Simple graph information is obtained using object-attributes and methods. Reporting typically provides views in-
stead of containers to reduce memory usage. The views update as the graph is updated similarly to dict-views. The
objects nodes, edges and adj provide access to data attributes via lookup (e.g. nodes[n], edges[u, v],
adj[u][v]) and iteration (e.g. nodes.items(), nodes.data('color'), nodes.data('color',
default='blue') and similarly for edges) Views exist for nodes, edges, neighbors()/adj and de-
gree.
For details on these and other miscellaneous methods, see below.
Subclasses (Advanced):
The Graph class uses a dict-of-dict-of-dict data structure. The outer dict (node_dict) holds adjacency information
keyed by node. The next dict (adjlist_dict) represents the adjacency information and holds edge data keyed by
neighbor. The inner dict (edge_attr_dict) represents the edge data and holds edge attribute values keyed by attribute
names.
Each of these three dicts can be replaced in a subclass by a user defined dict-like object. In general, the dict-like
features should be maintained but extra features can be added. To replace one of the dicts create a new graph class
by changing the class(!) variable holding the factory for that dict-like structure.
node_dict_factory
[function, (default: dict)] Factory function to be used to create the dict containing node attributes, keyed by
node id. It should require no arguments and return a dict-like object
node_attr_dict_factory: function, (default: dict)
Factory function to be used to create the node attribute dict which holds attribute values keyed by attribute
name. It should require no arguments and return a dict-like object
adjlist_outer_dict_factory
[function, (default: dict)] Factory function to be used to create the outer-most dict in the data structure that
holds adjacency info keyed by node. It should require no arguments and return a dict-like object.
adjlist_inner_dict_factory
[function, (default: dict)] Factory function to be used to create the adjacency list dict which holds edge data
keyed by neighbor. It should require no arguments and return a dict-like object
edge_attr_dict_factory
[function, (default: dict)] Factory function to be used to create the edge attribute dict which holds attribute
values keyed by attribute name. It should require no arguments and return a dict-like object.
graph_attr_dict_factory
[function, (default: dict)] Factory function to be used to create the graph attribute dict which holds attribute
Typically, if your extension doesn’t impact the data structure all methods will inherit without issue except:
to_directed/to_undirected. By default these methods create a DiGraph/Graph class and you proba-
bly want them to create your extension of a DiGraph/Graph. To facilitate this we define two class variables that
you can set in your subclass.
to_directed_class
[callable, (default: DiGraph or MultiDiGraph)] Class to create a new graph structure in the to_directed
method. If None, a NetworkX class (DiGraph or MultiDiGraph) is used.
to_undirected_class
[callable, (default: Graph or MultiGraph)] Class to create a new graph structure in the to_undirected
method. If None, a NetworkX class (Graph or MultiGraph) is used.
Subclassing Example
Create a low memory graph class that effectively disallows edge attributes by using a single attribute dict for all
edges. This reduces the memory used, but you lose edge attributes.

>>> class ThinGraph(nx.Graph):

... all_edge_dict = {"weight": 1}
...
... def single_edge_dict(self):
... return self.all_edge_dict
...
... edge_attr_dict_factory = single_edge_dict
>>> G = ThinGraph()
>>> G[2][1]
{'weight': 1}
>>> G[2][1] is G[2][2]
True
Methods
Adding and removing nodes and edges
Graph.__init__([incoming_graph_data]) Initialize a graph with edges, name, or graph attributes.

Graph.add_node(node_for_adding, **attr) Add a single node node_for_adding and update
node attributes.
Graph.add_nodes_from(nodes_for_adding, Add multiple nodes.
**attr)
Graph.remove_node(n) Remove node n.
Graph.remove_nodes_from(nodes) Remove multiple nodes.
Graph.add_edge(u_of_edge, v_of_edge, **attr) Add an edge between u and v.
Graph.add_edges_from(ebunch_to_add, **attr) Add all the edges in ebunch_to_add.
Graph.add_weighted_edges_from(ebunch_to_add) Add weighted edges in ebunch_to_add with specified
weight attr
Graph.remove_edge(u, v) Remove the edge between u and v.
Graph.remove_edges_from(ebunch) Remove all edges specified in ebunch.
Graph.update([edges, nodes]) Update the graph using nodes/edges/graphs as input.
Graph.clear() Remove all nodes and edges from the graph.
Graph.clear_edges() Remove all edges from the graph without altering nodes.
Graph.__init__
Graph.__init__(incoming_graph_data=None, **attr)
Initialize a graph with edges, name, or graph attributes.
Parameters
incoming_graph_data
graph is created. The data can be an edge list, or any NetworkX graph object. If the corre-
sponding optional Python packages are installed the data can also be a 2D NumPy array, a
SciPy sparse array, or a PyGraphviz graph.
attr
pairs.

See also:
convert
Examples
>>> G = nx.Graph() # or DiGraph, MultiGraph, MultiDiGraph, etc

>>> G = nx.Graph(name="my graph")
>>> e = [(1, 2), (2, 3), (3, 4)] # list of edges
>>> G = nx.Graph(e)
Arbitrary graph attribute pairs (key=value) may be assigned
>>> G = nx.Graph(e, day="Friday")

>>> G.graph
{'day': 'Friday'}
Graph.add_node
Graph.add_node(node_for_adding, **attr)
Add a single node node_for_adding and update node attributes.
Parameters
node_for_adding
[node] A node can be any hashable Python object except None.
attr
[keyword arguments, optional] Set or change node attributes using key=value.
See also:
add_nodes_from
Notes
A hashable object is one that can be used as a key in a Python dictionary. This includes strings, numbers, tuples of
strings and numbers, etc.
On many platforms hashable items also include mutables such as NetworkX Graphs, though one should be careful
that the hash doesn’t change on mutables.
Examples

>>> G.add_node(1)
>>> G.add_node("Hello")
>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
>>> G.add_node(K3)
>>> G.number_of_nodes()
3
Use keywords set/change node attributes:

>>> G.add_node(1, size=10)

>>> G.add_node(3, weight=0.4, UTM=("13S", 382871, 3972649))
Graph.add_nodes_from
Graph.add_nodes_from(nodes_for_adding, **attr)
Add multiple nodes.
Parameters
nodes_for_adding
[iterable container] A container of nodes (list, dict, set, etc.). OR A container of (node, at-
tribute dict) tuples. Node attributes are updated using the attribute dict.
attr
[keyword arguments, optional (default= no attributes)] Update attributes for all nodes in nodes.
Node attributes specified in nodes as a tuple take precedence over attributes specified via key-
word arguments.
See also:
add_node
Examples

>>> G.add_nodes_from("Hello")
>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
>>> G.add_nodes_from(K3)
>>> sorted(G.nodes(), key=str)
[0, 1, 2, 'H', 'e', 'l', 'o']
Use keywords to update specific node attributes for every node.
>>> G.add_nodes_from([1, 2], size=10)

>>> G.add_nodes_from([3, 4], weight=0.4)
Use (node, attrdict) tuples to update attributes for specific nodes.
>>> G.add_nodes_from([(1, dict(size=11)), (2, {"color": "blue"})])

>>> G.nodes[1]["size"]
11
>>> H = nx.Graph()
>>> H.add_nodes_from(G.nodes(data=True))
>>> H.nodes[1]["size"]
11

Graph.remove_node
Graph.remove_node(n)
Remove node n.
Removes the node n and all adjacent edges. Attempting to remove a non-existent node will raise an exception.
Parameters
n
[node] A node in the graph
Raises
NetworkXError
If n is not in the graph.
See also:
remove_nodes_from
Examples
>>> G = nx.path_graph(3) # or DiGraph, MultiGraph, MultiDiGraph, etc

>>> list(G.edges)
[(0, 1), (1, 2)]
>>> G.remove_node(1)
>>> list(G.edges)
[]
Graph.remove_nodes_from
Graph.remove_nodes_from(nodes)
Remove multiple nodes.
Parameters
nodes
[iterable container] A container of nodes (list, dict, set, etc.). If a node in the container is not
in the graph it is silently ignored.
See also:
remove_node
Examples

>>> e = list(G.nodes)
>>> e
[0, 1, 2]
>>> G.remove_nodes_from(e)
>>> list(G.nodes)
[]

Graph.add_edge
Graph.add_edge(u_of_edge, v_of_edge, **attr)

Add an edge between u and v.
The nodes u and v will be automatically added if they are not already in the graph.
Edge attributes can be specified with keywords or by directly accessing the edge’s attribute dictionary. See examples
below.
Parameters
u_of_edge, v_of_edge
[nodes] Nodes can be, for example, strings or numbers. Nodes must be hashable (and not
None) Python objects.
attr
[keyword arguments, optional] Edge data (or labels or objects) can be assigned using keyword
arguments.
See also:
add_edges_from
add a collection of edges
Notes
Adding an edge that already exists updates the edge data.

Many NetworkX algorithms designed for weighted graphs use an edge attribute (by default weight) to hold a
numerical value.
Examples
The following all add the edge e=(1, 2) to graph G:

>>> e = (1, 2)
>>> G.add_edge(1, 2) # explicit two-node form
>>> G.add_edge(*e) # single edge as tuple of two nodes
>>> G.add_edges_from([(1, 2)]) # add edges from iterable container
Associate data to edges using keywords:
>>> G.add_edge(1, 2, weight=3)

>>> G.add_edge(1, 3, weight=7, capacity=15, length=342.7)
For non-string attribute keys, use subscript notation.
>>> G[1][2].update({0: 5})
>>> G.edges[1, 2].update({0: 5})

Graph.add_edges_from
Graph.add_edges_from(ebunch_to_add, **attr)
Add all the edges in ebunch_to_add.
Parameters
ebunch_to_add
[container of edges] Each edge given in the container will be added to the graph. The edges
must be given as 2-tuples (u, v) or 3-tuples (u, v, d) where d is a dictionary containing edge
data.
attr
arguments.
See also:
add_edge
add a single edge
add_weighted_edges_from
convenient way to add weighted edges
Notes
Adding the same edge twice has no effect but any edge data will be updated when each duplicate edge is added.
Edge attributes specified in an ebunch take precedence over attributes specified via keyword arguments.
Examples

>>> G.add_edges_from([(0, 1), (1, 2)]) # using a list of edge tuples
>>> e = zip(range(0, 3), range(1, 4))
>>> G.add_edges_from(e) # Add the path graph 0-1-2-3
Associate data to edges
>>> G.add_edges_from([(1, 2), (2, 3)], weight=3)

>>> G.add_edges_from([(3, 4), (1, 4)], label="WN2898")
Graph.add_weighted_edges_from
Graph.add_weighted_edges_from(ebunch_to_add, weight='weight', **attr)

Add weighted edges in ebunch_to_add with specified weight attr
Parameters
ebunch_to_add
[container of edges] Each edge given in the list or container will be added to the graph. The
edges must be given as 3-tuples (u, v, w) where w is a number.
weight
[string, optional (default= ‘weight’)] The attribute name for the edge weights to be added.

attr
[keyword arguments, optional (default= no attributes)] Edge attributes to add/update for all
edges.
See also:
add_edge
add a single edge
add_edges_from
add multiple edges
Notes
Adding the same edge twice for Graph/DiGraph simply updates the edge data. For MultiGraph/MultiDiGraph,
duplicate edges are stored.
Examples

>>> G.add_weighted_edges_from([(0, 1, 3.0), (1, 2, 7.5)])
Graph.remove_edge
Graph.remove_edge(u, v)
Remove the edge between u and v.
Parameters
u, v
[nodes] Remove the edge between nodes u and v.
Raises
NetworkXError
If there is not an edge between u and v.
See also:
remove_edges_from
remove a collection of edges
Examples
>>> G = nx.path_graph(4) # or DiGraph, etc

>>> G.remove_edge(0, 1)
>>> e = (1, 2)
>>> G.remove_edge(*e) # unpacks e from an edge tuple
>>> e = (2, 3, {"weight": 7}) # an edge with attribute data
>>> G.remove_edge(*e[:2]) # select first part of edge tuple

Graph.remove_edges_from
Graph.remove_edges_from(ebunch)
Remove all edges specified in ebunch.
Parameters
ebunch: list or container of edge tuples
Each edge given in the list or container will be removed from the graph. The edges can be:
• 2-tuples (u, v) edge between u and v.
• 3-tuples (u, v, k) where k is ignored.
See also:
remove_edge
remove a single edge
Notes
Will fail silently if an edge in ebunch is not in the graph.
Examples

>>> ebunch = [(1, 2), (2, 3)]
>>> G.remove_edges_from(ebunch)
Graph.update
Graph.update(edges=None, nodes=None)
Update the graph using nodes/edges/graphs as input.
Like dict.update, this method takes a graph as input, adding the graph’s nodes and edges to this graph. It can also
take two inputs: edges and nodes. Finally it can take either edges or nodes. To specify only nodes the keyword
nodes must be used.
The collections of edges and nodes are treated similarly to the add_edges_from/add_nodes_from methods. When
iterated, they should yield 2-tuples (u, v) or 3-tuples (u, v, datadict).
Parameters
edges
[Graph object, collection of edges, or None] The first parameter can be a graph or some edges.
If it has attributes nodes and edges, then it is taken to be a Graph-like object and those
attributes are used as collections of nodes and edges to be added to the graph. If the first
parameter does not have those attributes, it is treated as a collection of edges and added to the
graph. If the first argument is None, no edges are added.
nodes
[collection of nodes, or None] The second parameter is treated as a collection of nodes to be
added to the graph unless it is None. If edges is None and nodes is None an
exception is raised. If the first parameter is a Graph, then nodes is ignored.

See also:
add_edges_from
add multiple edges to a graph
add_nodes_from
add multiple nodes to a graph
Notes
It you want to update the graph using an adjacency structure it is straightforward to obtain the edges/nodes from
adjacency. The following examples provide common cases, your adjacency may be slightly different and require
tweaks of these examples:
>>> # dict-of-set/list/tuple
>>> adj = {1: {2, 3}, 2: {1, 3}, 3: {1, 2}}
>>> e = [(u, v) for u, nbrs in adj.items() for v in nbrs]
>>> G.update(edges=e, nodes=adj)
>>> DG = nx.DiGraph()
>>> # dict-of-dict-of-attribute
>>> adj = {1: {2: 1.3, 3: 0.7}, 2: {1: 1.4}, 3: {1: 0.7}}
>>> e = [
... (u, v, {"weight": d})
... for u, nbrs in adj.items()
... for v, d in nbrs.items()
... ]
>>> DG.update(edges=e, nodes=adj)
>>> # dict-of-dict-of-dict
>>> adj = {1: {2: {"weight": 1.3}, 3: {"color": 0.7, "weight": 1.2}}}
>>> e = [
... (u, v, {"weight": d})
... ]
>>> # predecessor adjacency (dict-of-set)

>>> pred = {1: {2, 3}, 2: {3}, 3: {3}}
>>> e = [(v, u) for u, nbrs in pred.items() for v in nbrs]
>>> # MultiGraph dict-of-dict-of-dict-of-attribute

>>> MDG = nx.MultiDiGraph()
>>> adj = {
... 1: {2: {0: {"weight": 1.3}, 1: {"weight": 1.2}}},
... 3: {2: {0: {"weight": 0.7}}},
... }
>>> e = [
... (u, v, ekey, d)
... for v, keydict in nbrs.items()
... for ekey, d in keydict.items()
... ]
>>> MDG.update(edges=e)

Examples
>>> G = nx.path_graph(5)
>>> G.update(nx.complete_graph(range(4, 10)))
>>> from itertools import combinations
>>> edges = (
... (u, v, {"power": u * v})
... for u, v in combinations(range(10, 20), 2)
... if u * v < 225
... )
>>> nodes = [1000] # for singleton, use a container
>>> G.update(edges, nodes)
Graph.clear
Graph.clear()
Remove all nodes and edges from the graph.
This also removes the name, and all graph, node, and edge attributes.
Examples

>>> G.clear()
>>> list(G.nodes)
[]
>>> list(G.edges)
[]
Graph.clear_edges
Graph.clear_edges()
Remove all edges from the graph without altering nodes.
Examples

>>> G.clear_edges()
>>> list(G.nodes)
[0, 1, 2, 3]
>>> list(G.edges)
[]

Reporting nodes edges and neighbors
Graph.nodes A NodeView of the Graph as G.nodes or G.nodes().

Graph.__iter__() Iterate over the nodes.
Graph.has_node(n) Returns True if the graph contains the node n.
Graph.__contains__(n) Returns True if n is a node, False otherwise.
Graph.edges An EdgeView of the Graph as G.edges or G.edges().
Graph.has_edge(u, v) Returns True if the edge (u, v) is in the graph.
Graph.get_edge_data(u, v[, default]) Returns the attribute dictionary associated with edge (u,
v).
Graph.neighbors(n) Returns an iterator over all neighbors of node n.
Graph.adj Graph adjacency object holding the neighbors of each
node.
Graph.__getitem__(n) Returns a dict of neighbors of node n.
Graph.adjacency() Returns an iterator over (node, adjacency dict) tuples for
all nodes.
Graph.nbunch_iter([nbunch]) Returns an iterator over nodes contained in nbunch that
are also in the graph.
Graph.nodes
property Graph.nodes
A NodeView of the Graph as G.nodes or G.nodes().
Can be used as G.nodes for data lookup and for set-like operations. Can also be used as G.
nodes(data='color', default=None) to return a NodeDataView which reports specific node data
but no set operations. It presents a dict-like interface as well with G.nodes.items() iterating over (node,
nodedata) 2-tuples and G.nodes[3]['foo'] providing the value of the foo attribute for node 3. In
addition, a view G.nodes.data('foo') provides a dict-like interface to the foo attribute of each node.
G.nodes.data('foo', default=1) provides a default for nodes that do not have attribute foo.
Parameters
data
[string or bool, optional (default=False)] The node attribute returned in 2-tuple (n, ddict[data]).
If True, return entire node attribute dict as (n, ddict). If False, return just the nodes n.
default
[value, optional (default=None)] Value used for nodes that don’t have the requested attribute.
Only relevant if data is not True or False.
Returns
NodeView
Allows set-like operations over the nodes as well as node attribute dict lookup and calling to
get a NodeDataView. A NodeDataView iterates over (n, data) and has no set operations.
A NodeView iterates over n and includes set operations.
When called, if data is False, an iterator over nodes. Otherwise an iterator of 2-tuples (node,
attribute value) where the attribute is specified in data. If data is True then the attribute
becomes the entire data dictionary.

Notes
If your node data is not needed, it is simpler and equivalent to use the expression for n in G, or list(G).
Examples
There are two simple ways of getting a list of all nodes in the graph:
>>> list(G.nodes)
[0, 1, 2]
>>> list(G)
[0, 1, 2]
To get the node data along with the nodes:

>>> G.nodes[0]["foo"] = "bar"
[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
>>> list(G.nodes.data())
[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
>>> list(G.nodes(data="foo"))
[(0, 'bar'), (1, None), (2, None)]
>>> list(G.nodes.data("foo"))
[(0, 'bar'), (1, None), (2, None)]
>>> list(G.nodes(data="time"))
[(0, None), (1, '5pm'), (2, None)]
>>> list(G.nodes.data("time"))
[(0, None), (1, '5pm'), (2, None)]
>>> list(G.nodes(data="time", default="Not Available"))

[(0, 'Not Available'), (1, '5pm'), (2, 'Not Available')]
>>> list(G.nodes.data("time", default="Not Available"))
If some of your nodes have an attribute and the rest are assumed to have a default attribute value you can create a
dictionary from node/attribute pairs using the default keyword argument to guarantee the value is never None:
>>> G = nx.Graph()
>>> G.add_node(0)
>>> G.add_node(1, weight=2)
>>> dict(G.nodes(data="weight", default=1))
{0: 1, 1: 2, 2: 3}

Graph.__iter__
Graph.__iter__()
Iterate over the nodes. Use: ‘for n in G’.
Returns
niter
[iterator] An iterator over all nodes in the graph.
Examples

>>> [n for n in G]
[0, 1, 2, 3]
>>> list(G)
[0, 1, 2, 3]
Graph.has_node
Graph.has_node(n)
Returns True if the graph contains the node n.
Identical to n in G
Parameters
n
[node]
Examples

>>> G.has_node(0)
True
It is more readable and simpler to use
>>> 0 in G
True
Graph.__contains__
Graph.__contains__(n)
Returns True if n is a node, False otherwise. Use: ‘n in G’.

Examples

>>> 1 in G
True
Graph.edges
property Graph.edges
An EdgeView of the Graph as G.edges or G.edges().
edges(self, nbunch=None, data=False, default=None)
The EdgeView provides set-like operations on the edge-tuples as well as edge attribute lookup. When called, it also
provides an EdgeDataView object which allows control of access to edge attributes (but does not provide set-like
operations). Hence, G.edges[u, v]['color'] provides the value of the color attribute for edge (u, v)
while for (u, v, c) in G.edges.data('color', default='red'): iterates through all the
edges yielding the color attribute with default 'red' if no color attribute exists.
Parameters
nbunch
[single node, container, or all nodes (default= all nodes)] The view will only report edges from
these nodes.
data
[string or bool, optional (default=False)] The edge attribute returned in 3-tuple (u, v,
ddict[data]). If True, return edge attribute dict in 3-tuple (u, v, ddict). If False, return 2-tuple
(u, v).
default
[value, optional (default=None)] Value used for edges that don’t have the requested attribute.
Returns
edges
[EdgeView] A view of edge attributes, usually it iterates over (u, v) or (u, v, d) tuples of edges,
but can also be used for attribute lookup as edges[u, v]['foo'].
Notes
Nodes in nbunch that are not in the graph will be (quietly) ignored. For directed graphs this returns the out-edges.
Examples
>>> G = nx.path_graph(3) # or MultiGraph, etc

>>> [e for e in G.edges]
[(0, 1), (1, 2), (2, 3)]
>>> G.edges.data() # default data is {} (empty dict)
EdgeDataView([(0, 1, {}), (1, 2, {}), (2, 3, {'weight': 5})])
>>> G.edges.data("weight", default=1)
EdgeDataView([(0, 1, 1), (1, 2, 1), (2, 3, 5)])
(continues on next page)

(continued from previous page)

>>> G.edges([0, 3]) # only edges from these nodes
EdgeDataView([(0, 1), (3, 2)])
>>> G.edges(0) # only edges from node 0
EdgeDataView([(0, 1)])
Graph.has_edge
Graph.has_edge(u, v)
Returns True if the edge (u, v) is in the graph.
This is the same as v in G[u] without KeyError exceptions.
Parameters
u, v
Returns
edge_ind
[bool] True if edge is in the graph, False otherwise.
Examples

>>> G.has_edge(0, 1) # using two nodes
True
>>> e = (0, 1)
>>> G.has_edge(*e) # e is a 2-tuple (u, v)
True
>>> e = (0, 1, {"weight": 7})
>>> G.has_edge(*e[:2]) # e is a 3-tuple (u, v, data_dictionary)
True
The following syntax are equivalent:
>>> G.has_edge(0, 1)
True
>>> 1 in G[0] # though this gives KeyError if 0 not in G
True
Graph.get_edge_data
Graph.get_edge_data(u, v, default=None)
Returns the attribute dictionary associated with edge (u, v).
This is identical to G[u][v] except the default is returned instead of an exception if the edge doesn’t exist.
Parameters
u, v
[nodes]

default: any Python object (default=None)

Value to return if the edge (u, v) is not found.
Returns
edge_dict
[dictionary] The edge attribute dictionary.
Examples

>>> G[0][1]
{}
Warning: Assigning to G[u][v] is not permitted. But it is safe to assign attributes G[u][v]['foo']
>>> G[0][1]["weight"] = 7
>>> G[0][1]["weight"]
7
>>> G[1][0]["weight"]
7

>>> G.get_edge_data(0, 1) # default edge data is {}
{}
>>> e = (0, 1)
>>> G.get_edge_data(*e) # tuple form
{}
>>> G.get_edge_data("a", "b", default=0) # edge not in graph, return 0
0
Graph.neighbors
Graph.neighbors(n)
Returns an iterator over all neighbors of node n.
This is identical to iter(G[n])
Parameters
n
Returns
neighbors
[iterator] An iterator over all neighbors of node n
Raises
NetworkXError
If the node n is not in the graph.

Notes
Alternate ways to access the neighbors are G.adj[n] or G[n]:

>>> G.add_edge("a", "b", weight=7)
>>> G["a"]
AtlasView({'b': {'weight': 7}})
>>> [n for n in G[0]]
[1]
Examples

>>> [n for n in G.neighbors(0)]
[1]
Graph.adj
property Graph.adj
Graph adjacency object holding the neighbors of each node.
This object is a read-only dict-like structure with node keys and neighbor-dict values. The neighbor-dict is keyed
by neighbor to the edge-data-dict. So G.adj[3][2]['color'] = 'blue' sets the color of the edge (3,
2) to "blue".
Iterating over G.adj behaves like a dict. Useful idioms include for nbr, datadict in G.adj[n].
items():.
The neighbor information is also provided by subscripting the graph. So for nbr, foovalue in G[node].
data('foo', default=1): works.
For directed graphs, G.adj holds outgoing (successor) info.
Graph.__getitem__
Graph.__getitem__(n)
Returns a dict of neighbors of node n. Use: ‘G[n]’.
Parameters
n
[node] A node in the graph.
Returns
adj_dict
[dictionary] The adjacency dictionary for nodes connected to n.

Notes
G[n] is the same as G.adj[n] and similar to G.neighbors(n) (which is an iterator over G.adj[n])
Examples

>>> G[0]
AtlasView({1: {}})
Graph.adjacency
Graph.adjacency()
Returns an iterator over (node, adjacency dict) tuples for all nodes.
For directed graphs, only outgoing neighbors/adjacencies are included.
Returns
adj_iter
[iterator] An iterator over (node, adjacency dictionary) for all nodes in the graph.
Examples

>>> [(n, nbrdict) for n, nbrdict in G.adjacency()]
[(0, {1: {}}), (1, {0: {}, 2: {}}), (2, {1: {}, 3: {}}), (3, {2: {}})]
Graph.nbunch_iter
Graph.nbunch_iter(nbunch=None)
Returns an iterator over nodes contained in nbunch that are also in the graph.
The nodes in nbunch are checked for membership in the graph and if not are silently ignored.
Parameters
nbunch
[single node, container, or all nodes (default= all nodes)] The view will only report edges inci-
dent to these nodes.
Returns
niter
[iterator] An iterator over nodes in nbunch that are also in the graph. If nbunch is None, iterate
over all nodes in the graph.
Raises
NetworkXError
If nbunch is not a node or sequence of nodes. If a node in nbunch is not hashable.
See also:
Graph.__iter__

Notes
When nbunch is an iterator, the returned iterator yields values directly from nbunch, becoming exhausted when
nbunch is exhausted.
To test whether nbunch is a single node, one can use “if nbunch in self:”, even after processing with this routine.
If nbunch is not a node or a (possibly empty) sequence/iterator or None, a NetworkXError is raised. Also, if
any object in nbunch is not hashable, a NetworkXError is raised.
Counting nodes edges and neighbors
Graph.order() Returns the number of nodes in the graph.

Graph.number_of_nodes() Returns the number of nodes in the graph.
Graph.__len__() Returns the number of nodes in the graph.
Graph.degree A DegreeView for the Graph as G.degree or G.degree().
Graph.size([weight]) Returns the number of edges or total of all edge weights.
Graph.number_of_edges([u, v]) Returns the number of edges between two nodes.
Graph.order
Graph.order()
Returns the number of nodes in the graph.
Returns
nnodes
[int] The number of nodes in the graph.
See also:
number_of_nodes
identical method
__len__
identical method
Examples

>>> G.order()
3

Graph.number_of_nodes
Graph.number_of_nodes()
Returns
nnodes
See also:
order
identical method
__len__
identical method
Examples

3
Graph.__len__
Graph.__len__()
Returns the number of nodes in the graph. Use: ‘len(G)’.
Returns
nnodes
See also:
number_of_nodes
identical method
order
identical method
Examples

>>> len(G)
4

Graph.degree
property Graph.degree
A DegreeView for the Graph as G.degree or G.degree().
The node degree is the number of edges adjacent to the node. The weighted node degree is the sum of the edge
weights for edges incident to that node.
This object provides an iterator for (node, degree) as well as lookup for the degree for a single node.
Parameters
nbunch
weight
[string or None, optional (default=None)] The name of an edge attribute that holds the numer-
ical value used as a weight. If None, then each edge has weight 1. The degree is the sum of
the edge weights adjacent to the node.
Returns
DegreeView or int
If multiple nodes are requested (the default), returns a DegreeView mapping nodes to their
degree. If a single node is requested, returns the degree of the node as an integer.
Examples

>>> G.degree[0] # node 0 has degree 1
1
>>> list(G.degree([0, 1, 2]))
[(0, 1), (1, 2), (2, 2)]
Graph.size
Graph.size(weight=None)
Returns the number of edges or total of all edge weights.
Parameters
weight
[string or None, optional (default=None)] The edge attribute that holds the numerical value
used as a weight. If None, then each edge has weight 1.
Returns
size
[numeric] The number of edges or (if weight keyword is provided) the total weight sum.
If weight is None, returns an int. Otherwise a float (or more general numeric if the weights are
more general).
See also:
number_of_edges

Examples

>>> G.size()
3

>>> G.add_edge("b", "c", weight=4)
>>> G.size()
2
>>> G.size(weight="weight")
6.0
Graph.number_of_edges
Graph.number_of_edges(u=None, v=None)
Returns the number of edges between two nodes.
Parameters
u, v
[nodes, optional (default=all edges)] If u and v are specified, return the number of edges be-
tween u and v. Otherwise return the total number of all edges.
Returns
nedges
[int] The number of edges in the graph. If nodes u and v are specified return the number of
edges between those nodes. If the graph is directed, this only returns the number of edges
from u to v.
See also:
size
Examples
For undirected graphs, this method counts the total number of edges in the graph:
>>> G.number_of_edges()
3
If you specify two nodes, this counts the total number of edges joining the two nodes:
>>> G.number_of_edges(0, 1)
1
For directed graphs, this method can count the total number of directed edges from u to v:


1
Making copies and subgraphs
Graph.copy([as_view]) Returns a copy of the graph.

Graph.to_undirected([as_view]) Returns an undirected copy of the graph.
Graph.to_directed([as_view]) Returns a directed representation of the graph.
Graph.subgraph(nodes) Returns a SubGraph view of the subgraph induced on
nodes.
Graph.edge_subgraph(edges) Returns the subgraph induced by the specified edges.
Graph.copy
Graph.copy(as_view=False)
Returns a copy of the graph.
The copy method by default returns an independent shallow copy of the graph and attributes. That is, if an attribute
is a container, that container is shared by the original an the copy. Use Python’s copy.deepcopy for new
containers.
If as_view is True then a view is returned instead of a copy.
Parameters
as_view
[bool, optional (default=False)] If True, the returned graph-view provides a read-only view of
the original graph without actually copying any data.
Returns
G
[Graph] A copy of the graph.
See also:
to_directed
return a directed copy of the graph.
Notes
All copies reproduce the graph structure, but data attributes may be handled in different ways. There are four types
of copies of a graph that people might want.
Deepcopy – A “deepcopy” copies the graph structure as well as all data attributes and any objects they might
contain. The entire graph object is new so that changes in the copy do not affect the original object. (see Python’s
copy.deepcopy)
Data Reference (Shallow) – For a shallow copy the graph structure is copied but the edge, node and graph attribute
dicts are references to those in the original graph. This saves time and memory but could cause confusion if you
change an attribute in one graph and it changes the attribute in the other. NetworkX does not provide this level of
shallow copy.

Independent Shallow – This copy creates new independent attribute dicts and then does a shallow copy of the
attributes. That is, any attributes that are containers are shared between the new graph and the original. This is
exactly what dict.copy() provides. You can obtain this style copy using:
>>> H = G.copy()
>>> H = G.copy(as_view=False)
>>> H = nx.Graph(G)
>>> H = G.__class__(G)
Fresh Data – For fresh data, the graph structure is copied while new empty data attribute dicts are created. The
resulting graph is independent of the original and it has no edge, node or graph attributes. Fresh copies are not
enabled. Instead use:
>>> H = G.__class__()
>>> H.add_nodes_from(G)
>>> H.add_edges_from(G.edges)
View – Inspired by dict-views, graph-views act like read-only versions of the original graph, providing a copy of
the original structure without requiring any memory for copying the information.
See the Python copy module for more information on shallow and deep copies, https://docs.python.org/3/library/
copy.html.
Examples

>>> H = G.copy()
Graph.to_undirected
Graph.to_undirected(as_view=False)
Returns an undirected copy of the graph.
Parameters
as_view
[bool (optional, default=False)] If True return a view of the original undirected graph.
Returns
G
[Graph/MultiGraph] A deepcopy of the graph.
See also:
Graph, copy, add_edge, add_edges_from

Notes
This returns a “deepcopy” of the edge, node, and graph attributes which attempts to completely copy all of the data
and references.
This is in contrast to the similar G = nx.DiGraph(D) which returns a shallow copy of the data.
copy.html.
Warning: If you have subclassed DiGraph to use dict-like objects in the data structure, those changes do not transfer
to the Graph created by this method.
Examples

>>> H = G.to_directed()
>>> list(H.edges)
[(0, 1), (1, 0)]
>>> G2 = H.to_undirected()
>>> list(G2.edges)
[(0, 1)]
Graph.to_directed
Graph.to_directed(as_view=False)
Returns a directed representation of the graph.
Returns
G
[DiGraph] A directed graph with the same name, same nodes, and with each edge (u, v, data)
replaced by two directed edges (u, v, data) and (v, u, data).
Notes
and references.
This is in contrast to the similar D=DiGraph(G) which returns a shallow copy of the data.
copy.html.
Warning: If you have subclassed Graph to use dict-like objects in the data structure, those changes do not transfer
to the DiGraph created by this method.

Examples
>>> G = nx.Graph() # or MultiGraph, etc

>>> list(H.edges)
[(0, 1), (1, 0)]
If already directed, return a (deep) copy
>>> G = nx.DiGraph() # or MultiDiGraph, etc

>>> list(H.edges)
[(0, 1)]
Graph.subgraph
Graph.subgraph(nodes)
Returns a SubGraph view of the subgraph induced on nodes.
The induced subgraph of the graph contains the nodes in nodes and the edges between those nodes.
Parameters
nodes
[list, iterable] A container of nodes which will be iterated through once.
Returns
G
[SubGraph View] A subgraph view of the graph. The graph structure cannot be changed but
node/edge attributes can and are shared with the original graph.
Notes
The graph, edge and node attributes are shared with the original graph. Changes to the graph structure is ruled out
by the view, but changes to attributes are reflected in the original graph.
To create a subgraph with its own copy of the edge/node attributes use: G.subgraph(nodes).copy()
For an inplace reduction of a graph to a subgraph you can remove nodes: G.remove_nodes_from([n for n in G if n
not in set(nodes)])
Subgraph views are sometimes NOT what you want. In most cases where you want to do more than simply look at
the induced edges, it makes more sense to just create the subgraph as its own graph with code like:
# Create a subgraph SG based on a (possibly multigraph) G

SG = G.__class__()
SG.add_nodes_from((n, G.nodes[n]) for n in largest_wcc)
if SG.is_multigraph():
SG.add_edges_from((n, nbr, key, d)
for n, nbrs in G.adj.items() if n in largest_wcc
for nbr, keydict in nbrs.items() if nbr in largest_wcc
for key, d in keydict.items())
else:


SG.add_edges_from((n, nbr, d)
for nbr, d in nbrs.items() if nbr in largest_wcc)
SG.graph.update(G.graph)
Examples

>>> H = G.subgraph([0, 1, 2])
>>> list(H.edges)
[(0, 1), (1, 2)]
Graph.edge_subgraph
Graph.edge_subgraph(edges)
Returns the subgraph induced by the specified edges.
The induced subgraph contains each edge in edges and each node incident to any one of those edges.
Parameters
edges
[iterable] An iterable of edges in this graph.
Returns
G
[Graph] An edge-induced subgraph of this graph with the same edge attributes.
Notes
The graph, edge, and node attributes in the returned subgraph view are references to the corresponding attributes
in the original graph. The view is read-only.
To create a full graph version of the subgraph with its own copy of the edge or node attributes, use:
G.edge_subgraph(edges).copy()
Examples
>>> H = G.edge_subgraph([(0, 1), (3, 4)])
>>> list(H.nodes)
[0, 1, 3, 4]
>>> list(H.edges)
[(0, 1), (3, 4)]

2.2.2 DiGraph—Directed graphs with self loops
Overview
class DiGraph(incoming_graph_data=None, **attr)

Base class for directed graphs.
A DiGraph stores nodes and edges with optional data, or attributes.
DiGraphs hold directed edges. Self loops are allowed but multiple (parallel) edges are not.
Nodes can be arbitrary (hashable) Python objects with optional key/value attributes. By convention None is not
used as a node.
Parameters
incoming_graph_data
attr
pairs.
See also:
Graph
MultiGraph
MultiDiGraph
Examples

Nodes:
>>> G.add_node(1)


>>> G.add_node(H)
Edges:
Add one edge,
a list of edges,
>>> G.add_edges_from([(1, 2), (1, 3)])
If some edges connect nodes not yet in the graph, the nodes are added automatically. There are no errors when
adding nodes or edges that already exist.
Attributes:
>>> G = nx.DiGraph(day="Friday")
>>> G.graph
{'day': 'Friday'}

>>> G.nodes[1]
{'time': '5pm'}
>>> G.nodes[1]["room"] = 714
[(1, {'time': '5pm'}), (3, {'time': '2pm'})]
>>> G.add_edge(1, 2, weight=4.7)

>>> G.add_edges_from([(3, 4), (4, 5)], color="red")
>>> G.add_edges_from([(1, 2, {"color": "blue"}), (2, 3, {"weight": 8})])
>>> G[1][2]["weight"] = 4.7
>>> G.edges[1, 2]["weight"] = 4
Warning: we protect the graph data structure by making G.edges[1, 2] a read-only dict-like structure. How-
ever, you can assign to attributes in e.g. G.edges[1, 2]. Thus, use 2 sets of brackets to add/change data
attributes: G.edges[1, 2]['weight'] = 4 (For multigraphs: MG.edges[u, v, key][name] =
value).
Shortcuts:


True
[1, 2]
5
dict G.adj or G.adjacency()

... for nbr, eattr in nbrsdict.items():
... pass
But the edges reporting object is often more convenient:
>>> for u, v, weight in G.edges(data="weight"):

... pass
Reporting:
Simple graph information is obtained using object-attributes and methods. Reporting usually provides views in-
stead of containers to reduce memory usage. The views update as the graph is updated similarly to dict-views. The
objects nodes, edges and adj provide access to data attributes via lookup (e.g. nodes[n], edges[u, v],
gree.
The Graph class uses a dict-of-dict-of-dict data structure. The outer dict (node_dict) holds adjacency information
keyed by node. The next dict (adjlist_dict) represents the adjacency information and holds edge data keyed by
neighbor. The inner dict (edge_attr_dict) represents the edge data and holds edge attribute values keyed by attribute
names.
Each of these three dicts can be replaced in a subclass by a user defined dict-like object. In general, the
dict-like features should be maintained but extra features can be added. To replace one of the dicts create a
new graph class by changing the class(!) variable holding the factory for that dict-like structure. The vari-
able names are node_dict_factory, node_attr_dict_factory, adjlist_inner_dict_factory, adjlist_outer_dict_factory,
edge_attr_dict_factory and graph_attr_dict_factory.
node_dict_factory

[function, optional (default: dict)] Factory function to be used to create the adjacency list dict which holds
edge data keyed by neighbor. It should require no arguments and return a dict-like object
[function, optional (default: dict)] Factory function to be used to create the edge attribute dict which holds
attribute values keyed by attribute name. It should require no arguments and return a dict-like object.
Typically, if your extension doesn’t impact the data structure all methods will inherited without issue except:
to_directed_class
to_undirected_class
Subclassing Example

...
...
>>> G = ThinGraph()
>>> G[2][1]
{'weight': 1}
>>> G[2][1] is G[2][2]
True

Methods
DiGraph.__init__([incoming_graph_data]) Initialize a graph with edges, name, or graph attributes.

DiGraph.add_node(node_for_adding, **attr) Add a single node node_for_adding and update
node attributes.
DiGraph.add_nodes_from(nodes_for_adding, Add multiple nodes.
**attr)
DiGraph.remove_node(n) Remove node n.
DiGraph.remove_nodes_from(nodes) Remove multiple nodes.
DiGraph.add_edge(u_of_edge, v_of_edge, **attr) Add an edge between u and v.
DiGraph.add_edges_from(ebunch_to_add, Add all the edges in ebunch_to_add.
**attr)
DiGraph.add_weighted_edges_from(ebunch_to_add) Add weighted edges in ebunch_to_add with specified
weight attr
DiGraph.remove_edge(u, v) Remove the edge between u and v.
DiGraph.remove_edges_from(ebunch) Remove all edges specified in ebunch.
DiGraph.update([edges, nodes]) Update the graph using nodes/edges/graphs as input.
DiGraph.clear() Remove all nodes and edges from the graph.
DiGraph.clear_edges() Remove all edges from the graph without altering nodes.
DiGraph.__init__
DiGraph.__init__(incoming_graph_data=None, **attr)
Parameters
incoming_graph_data
attr
pairs.
See also:
convert

Examples

>>> e = [(1, 2), (2, 3), (3, 4)] # list of edges
>>> G = nx.Graph(e)

>>> G.graph
{'day': 'Friday'}
DiGraph.add_node
DiGraph.add_node(node_for_adding, **attr)
Parameters
node_for_adding
attr
See also:
add_nodes_from
Notes
Examples

>>> G.add_node(1)
>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
>>> G.add_node(K3)
3


DiGraph.add_nodes_from
DiGraph.add_nodes_from(nodes_for_adding, **attr)
Add multiple nodes.
Parameters
nodes_for_adding
attr
word arguments.
See also:
add_node
Examples

>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
[0, 1, 2, 'H', 'e', 'l', 'o']


11
>>> H = nx.Graph()
11
DiGraph.remove_node
DiGraph.remove_node(n)
Remove node n.
Parameters
n

Raises
NetworkXError
See also:
remove_nodes_from
Examples

>>> list(G.edges)
[(0, 1), (1, 2)]
>>> list(G.edges)
[]
DiGraph.remove_nodes_from
DiGraph.remove_nodes_from(nodes)
Parameters
nodes
See also:
remove_node
Examples

>>> e
[0, 1, 2]
>>> list(G.nodes)
[]

DiGraph.add_edge
DiGraph.add_edge(u_of_edge, v_of_edge, **attr)

below.
Parameters
attr
arguments.
See also:
add_edges_from
Notes

Many NetworkX algorithms designed for weighted graphs use an edge attribute (by default weight) to hold a
numerical value.
Examples

>>> e = (1, 2)

>>> G[1][2].update({0: 5})
>>> G.edges[1, 2].update({0: 5})

DiGraph.add_edges_from
DiGraph.add_edges_from(ebunch_to_add, **attr)
Parameters
ebunch_to_add
data.
attr
arguments.
See also:
add_edge
add a single edge
Notes
Examples

>>> e = zip(range(0, 3), range(1, 4))

DiGraph.add_weighted_edges_from
DiGraph.add_weighted_edges_from(ebunch_to_add, weight='weight', **attr)

Parameters
ebunch_to_add
weight

attr
edges.
See also:
add_edge
add a single edge
add_edges_from
add multiple edges
Notes
Examples

DiGraph.remove_edge
DiGraph.remove_edge(u, v)
Parameters
u, v
Raises
NetworkXError
See also:
remove_edges_from
Examples
>>> G = nx.Graph() # or DiGraph, etc

>>> nx.add_path(G, [0, 1, 2, 3])
>>> e = (1, 2)

DiGraph.remove_edges_from
DiGraph.remove_edges_from(ebunch)
Parameters
See also:
remove_edge
Notes
Examples

>>> ebunch = [(1, 2), (2, 3)]
DiGraph.update
DiGraph.update(edges=None, nodes=None)
nodes must be used.
Parameters
edges
nodes

See also:
add_edges_from
add_nodes_from
Notes
>>> adj = {1: {2, 3}, 2: {1, 3}, 3: {1, 2}}
>>> adj = {1: {2: 1.3, 3: 0.7}, 2: {1: 1.4}, 3: {1: 0.7}}
>>> e = [
... (u, v, {"weight": d})
... ]
>>> e = [
... (u, v, {"weight": d})
... ]

>>> pred = {1: {2, 3}, 2: {3}, 3: {3}}

>>> adj = {
... 1: {2: {0: {"weight": 1.3}, 1: {"weight": 1.2}}},
... 3: {2: {0: {"weight": 0.7}}},
... }
>>> e = [
... (u, v, ekey, d)
... ]

Examples
>>> edges = (
... (u, v, {"power": u * v})
... if u * v < 225
... )
DiGraph.clear
DiGraph.clear()
Examples

>>> G.clear()
>>> list(G.nodes)
[]
>>> list(G.edges)
[]
DiGraph.clear_edges
DiGraph.clear_edges()
Examples

>>> G.clear_edges()
>>> list(G.nodes)
[0, 1, 2, 3]
>>> list(G.edges)
[]

DiGraph.nodes A NodeView of the Graph as G.nodes or G.nodes().

DiGraph.__iter__() Iterate over the nodes.
DiGraph.has_node(n) Returns True if the graph contains the node n.
DiGraph.__contains__(n) Returns True if n is a node, False otherwise.
DiGraph.edges An OutEdgeView of the DiGraph as G.edges or
G.edges().
DiGraph.out_edges An OutEdgeView of the DiGraph as G.edges or
G.edges().
DiGraph.in_edges A view of the in edges of the graph as G.in_edges or
G.in_edges().
DiGraph.has_edge(u, v) Returns True if the edge (u, v) is in the graph.
DiGraph.get_edge_data(u, v[, default]) Returns the attribute dictionary associated with edge (u,
v).
DiGraph.neighbors(n) Returns an iterator over successor nodes of n.
DiGraph.adj Graph adjacency object holding the neighbors of each
node.
DiGraph.__getitem__(n) Returns a dict of neighbors of node n.
DiGraph.successors(n) Returns an iterator over successor nodes of n.
DiGraph.succ Graph adjacency object holding the successors of each
node.
DiGraph.predecessors(n) Returns an iterator over predecessor nodes of n.
DiGraph.pred Graph adjacency object holding the predecessors of each
node.
DiGraph.adjacency() Returns an iterator over (node, adjacency dict) tuples for
all nodes.
DiGraph.nbunch_iter([nbunch]) Returns an iterator over nodes contained in nbunch that
DiGraph.nodes
property DiGraph.nodes
Parameters
data
default
Returns

NodeView
Notes
Examples
>>> list(G.nodes)
[0, 1, 2]
>>> list(G)
[0, 1, 2]

[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
[(0, 'bar'), (1, None), (2, None)]
[(0, 'bar'), (1, None), (2, None)]
[(0, None), (1, '5pm'), (2, None)]
[(0, None), (1, '5pm'), (2, None)]

>>> G = nx.Graph()
>>> G.add_node(0)


{0: 1, 1: 2, 2: 3}
DiGraph.__iter__
DiGraph.__iter__()
Returns
niter
Examples

>>> [n for n in G]
[0, 1, 2, 3]
>>> list(G)
[0, 1, 2, 3]
DiGraph.has_node
DiGraph.has_node(n)
Identical to n in G
Parameters
n
[node]
Examples

>>> G.has_node(0)
True
>>> 0 in G
True

DiGraph.__contains__
DiGraph.__contains__(n)
Examples

>>> 1 in G
True
DiGraph.edges
property DiGraph.edges
An OutEdgeView of the DiGraph as G.edges or G.edges().
The OutEdgeView provides set-like operations on the edge-tuples as well as edge attribute lookup. When called,
it also provides an EdgeDataView object which allows control of access to edge attributes (but does not provide
set-like operations). Hence, G.edges[u, v]['color'] provides the value of the color attribute for edge
(u, v) while for (u, v, c) in G.edges.data('color', default='red'): iterates through
all the edges yielding the color attribute with default 'red' if no color attribute exists.
Parameters
nbunch
these nodes.
data
(u, v).
default
Returns
edges
[OutEdgeView] A view of edge attributes, usually it iterates over (u, v) or (u, v, d) tuples of
edges, but can also be used for attribute lookup as edges[u, v]['foo'].
See also:
in_edges, out_edges

Notes
Examples

>>> nx.add_path(G, [0, 1, 2])
[(0, 1), (1, 2), (2, 3)]
OutEdgeDataView([(0, 1, {}), (1, 2, {}), (2, 3, {'weight': 5})])
OutEdgeDataView([(0, 1, 1), (1, 2, 1), (2, 3, 5)])
>>> G.edges([0, 2]) # only edges originating from these nodes
OutEdgeDataView([(0, 1), (2, 3)])
OutEdgeDataView([(0, 1)])
DiGraph.out_edges
property DiGraph.out_edges
The OutEdgeView provides set-like operations on the edge-tuples as well as edge attribute lookup. When called,
set-like operations). Hence, G.edges[u, v]['color'] provides the value of the color attribute for edge
(u, v) while for (u, v, c) in G.edges.data('color', default='red'): iterates through
all the edges yielding the color attribute with default 'red' if no color attribute exists.
Parameters
nbunch
these nodes.
data
(u, v).
default
Returns
edges
See also:
in_edges, out_edges

Notes
Examples

>>> nx.add_path(G, [0, 1, 2])
[(0, 1), (1, 2), (2, 3)]
DiGraph.in_edges
property DiGraph.in_edges
A view of the in edges of the graph as G.in_edges or G.in_edges().
in_edges(self, nbunch=None, data=False, default=None):
Parameters
nbunch
data
(u, v).
default
Returns
in_edges
[InEdgeView or InEdgeDataView] A view of edge attributes, usually it iterates over (u, v) or (u,
v, d) tuples of edges, but can also be used for attribute lookup as edges[u, v]['foo'].
Examples
..

>>> G.add_edge(1, 2, color='blue')

..
>>> G.in_edges()
..
InEdgeView([(1, 2)])
>>> G.in_edges(nbunch=2)
..
InEdgeDataView([(1, 2)])
See also:
edges
DiGraph.has_edge
DiGraph.has_edge(u, v)
Parameters
u, v
Returns
edge_ind
Examples

True
>>> e = (0, 1)
True
>>> e = (0, 1, {"weight": 7})
True
True
True

DiGraph.get_edge_data
DiGraph.get_edge_data(u, v, default=None)
Parameters
u, v
[nodes]
Returns
edge_dict
Examples

>>> G[0][1]
{}
>>> G[0][1]["weight"] = 7
>>> G[0][1]["weight"]
7
>>> G[1][0]["weight"]
7

{}
>>> e = (0, 1)
{}
0
DiGraph.neighbors
DiGraph.neighbors(n)
Returns an iterator over successor nodes of n.
A successor of n is a node m such that there exists a directed edge from n to m.
Parameters
n
Raises

NetworkXError
See also:
predecessors
Notes
neighbors() and successors() are the same.
DiGraph.adj
property DiGraph.adj
by neighbor to the edge-data-dict. So G.adj[3][2]['color'] = 'blue' sets the color of the edge (3,
2) to "blue".
items():.
DiGraph.__getitem__
DiGraph.__getitem__(n)
Parameters
n
Returns
adj_dict
Notes

Examples

>>> G[0]
AtlasView({1: {}})
DiGraph.successors
DiGraph.successors(n)
Parameters
n
Raises
NetworkXError
See also:
predecessors
Notes
DiGraph.succ
property DiGraph.succ
Graph adjacency object holding the successors of each node.
by neighbor to the edge-data-dict. So G.succ[3][2]['color'] = 'blue' sets the color of the edge (3,
2) to "blue".
Iterating over G.succ behaves like a dict. Useful idioms include for nbr, datadict in G.succ[n].
items():. A data-view not provided by dicts also exists: for nbr, foovalue in G.succ[node].
data('foo'): and a default can be set via a default argument to the data method.
For directed graphs, G.adj is identical to G.succ.

DiGraph.predecessors
DiGraph.predecessors(n)
Returns an iterator over predecessor nodes of n.
A predecessor of n is a node m such that there exists a directed edge from m to n.
Parameters
n
Raises
NetworkXError
See also:
successors
DiGraph.pred
property DiGraph.pred
Graph adjacency object holding the predecessors of each node.
by neighbor to the edge-data-dict. So G.pred[2][3]['color'] = 'blue' sets the color of the edge (3,
2) to "blue".
Iterating over G.pred behaves like a dict. Useful idioms include for nbr, datadict in G.pred[n].
items():. A data-view not provided by dicts also exists: for nbr, foovalue in G.pred[node].
data('foo'): A default can be set via a default argument to the data method.
DiGraph.adjacency
DiGraph.adjacency()
Returns
adj_iter

Examples

[(0, {1: {}}), (1, {0: {}, 2: {}}), (2, {1: {}, 3: {}}), (3, {2: {}})]
DiGraph.nbunch_iter
DiGraph.nbunch_iter(nbunch=None)
Parameters
nbunch
Returns
niter
Raises
NetworkXError
See also:
Graph.__iter__
Notes
DiGraph.order() Returns the number of nodes in the graph.

DiGraph.number_of_nodes() Returns the number of nodes in the graph.
DiGraph.__len__() Returns the number of nodes in the graph.
DiGraph.degree A DegreeView for the Graph as G.degree or G.degree().
DiGraph.in_degree An InDegreeView for (node, in_degree) or in_degree for
single node.
DiGraph.out_degree An OutDegreeView for (node, out_degree)
DiGraph.size([weight]) Returns the number of edges or total of all edge weights.
DiGraph.number_of_edges([u, v]) Returns the number of edges between two nodes.

DiGraph.order
DiGraph.order()
Returns
nnodes
See also:
number_of_nodes
identical method
__len__
identical method
Examples

>>> G.order()
3
DiGraph.number_of_nodes
DiGraph.number_of_nodes()
Returns
nnodes
See also:
order
identical method
__len__
identical method
Examples

3

DiGraph.__len__
DiGraph.__len__()
Returns
nnodes
See also:
number_of_nodes
identical method
order
identical method
Examples

>>> len(G)
4
DiGraph.degree
property DiGraph.degree
Parameters
nbunch
weight
Returns
DiDegreeView or int
If multiple nodes are requested (the default), returns a DiDegreeView mapping nodes to
their degree. If a single node is requested, returns the degree of the node as an integer.
See also:
in_degree, out_degree

Examples
>>> G = nx.DiGraph() # or MultiDiGraph

>>> nx.add_path(G, [0, 1, 2, 3])
>>> G.degree(0) # node 0 with degree 1
1
>>> list(G.degree([0, 1, 2]))
[(0, 1), (1, 2), (2, 2)]
DiGraph.in_degree
property DiGraph.in_degree
An InDegreeView for (node, in_degree) or in_degree for single node.
The node in_degree is the number of edges pointing to the node. The weighted node degree is the sum of the edge
This object provides an iteration over (node, in_degree) as well as lookup for the degree for a single node.
Parameters
nbunch
weight
Returns
If a single node is requested
deg
[int] In-degree of the node
OR if multiple nodes are requested
nd_iter
[iterator] The iterator returns two-tuples of (node, in-degree).
See also:
degree, out_degree
Examples
>>> nx.add_path(G, [0, 1, 2, 3])
>>> G.in_degree(0) # node 0 with degree 0
0
>>> list(G.in_degree([0, 1, 2]))
[(0, 0), (1, 1), (2, 1)]

DiGraph.out_degree
property DiGraph.out_degree
An OutDegreeView for (node, out_degree)
The node out_degree is the number of edges pointing out of the node. The weighted node degree is the sum of the
edge weights for edges incident to that node.
This object provides an iterator over (node, out_degree) as well as lookup for the degree for a single node.
Parameters
nbunch
weight
Returns
deg
[int] Out-degree of the node
nd_iter
[iterator] The iterator returns two-tuples of (node, out-degree).
See also:
degree, in_degree
Examples
>>> nx.add_path(G, [0, 1, 2, 3])
>>> G.out_degree(0) # node 0 with degree 1
1
>>> list(G.out_degree([0, 1, 2]))
[(0, 1), (1, 1), (2, 1)]
DiGraph.size
DiGraph.size(weight=None)
Parameters
weight
Returns

size
more general).
See also:
number_of_edges
Examples

>>> G.size()
3

>>> G.size()
2
6.0
DiGraph.number_of_edges
DiGraph.number_of_edges(u=None, v=None)
Parameters
u, v
[nodes, optional (default=all edges)] If u and v are specified, return the number of edges be-
tween u and v. Otherwise return the total number of all edges.
Returns
nedges
from u to v.
See also:
size

Examples
3
1
1
DiGraph.copy([as_view]) Returns a copy of the graph.

DiGraph.to_undirected([reciprocal, as_view]) Returns an undirected representation of the digraph.
DiGraph.to_directed([as_view]) Returns a directed representation of the graph.
DiGraph.subgraph(nodes) Returns a SubGraph view of the subgraph induced on
nodes.
DiGraph.edge_subgraph(edges) Returns the subgraph induced by the specified edges.
DiGraph.reverse([copy]) Returns the reverse of the graph.
DiGraph.copy
DiGraph.copy(as_view=False)
containers.
Parameters
as_view
Returns
G
See also:

to_directed
Notes
copy.deepcopy)
shallow copy.
>>> H = G.copy()
>>> H = nx.Graph(G)
>>> H = G.__class__(G)
>>> H = G.__class__()
copy.html.
Examples

>>> H = G.copy()

DiGraph.to_undirected
DiGraph.to_undirected(reciprocal=False, as_view=False)
Returns an undirected representation of the digraph.
Parameters
reciprocal
[bool (optional)] If True only keep edges that appear in both directions in the original digraph.
as_view
[bool (optional, default=False)] If True return an undirected view of the original directed graph.
Returns
G
[Graph] An undirected graph with the same name and nodes and with edge (u, v, data) if either
(u, v, data) or (v, u, data) is in the digraph. If both edges exist in digraph and their edge data
is different, only one edge is created with an arbitrary choice of which edge data to use. You
must check and correct for this manually if desired.
See also:
Notes
If edges in both directions (u, v) and (v, u) exist in the graph, attributes for the new undirected edge will be a
combination of the attributes of the directed edges. The edge data is updated in the (arbitrary) order that the edges
are encountered. For more customized control of the edge attributes use add_edge().
and references.
This is in contrast to the similar G=DiGraph(D) which returns a shallow copy of the data.
copy.html.
Warning: If you have subclassed DiGraph to use dict-like objects in the data structure, those changes do not transfer
to the Graph created by this method.
Examples

>>> list(H.edges)
[(0, 1), (1, 0)]
>>> list(G2.edges)
[(0, 1)]

DiGraph.to_directed
DiGraph.to_directed(as_view=False)
Returns
G
Notes
and references.
copy.html.
Warning: If you have subclassed Graph to use dict-like objects in the data structure, those changes do not transfer
to the DiGraph created by this method.
Examples

>>> list(H.edges)
[(0, 1), (1, 0)]

>>> list(H.edges)
[(0, 1)]
DiGraph.subgraph
DiGraph.subgraph(nodes)
Parameters
nodes
Returns

G
Notes
not in set(nodes)])

SG = G.__class__()
else:
Examples

>>> H = G.subgraph([0, 1, 2])
>>> list(H.edges)
[(0, 1), (1, 2)]
DiGraph.edge_subgraph
DiGraph.edge_subgraph(edges)
Parameters
edges
Returns
G

Notes
Examples
>>> H = G.edge_subgraph([(0, 1), (3, 4)])
>>> list(H.nodes)
[0, 1, 3, 4]
>>> list(H.edges)
[(0, 1), (3, 4)]
DiGraph.reverse
DiGraph.reverse(copy=True)
Returns the reverse of the graph.
The reverse is a graph with the same nodes and edges but with the directions of the edges reversed.
Parameters
copy
[bool optional (default=True)] If True, return a new DiGraph holding the reversed edges. If
False, the reverse graph is created using a view of the original graph.
2.2.3 MultiGraph—Undirected graphs with self loops and parallel edges
Overview
class MultiGraph(incoming_graph_data=None, multigraph_input=None, **attr)

An undirected graph class that can store multiedges.
Multiedges are multiple edges between two nodes. Each edge can hold optional data or attributes.
A MultiGraph holds undirected edges. Self loops are allowed.
used as a node.
Edges are represented as links between nodes with optional key/value attributes, in a MultiGraph each edge has a
key to distinguish between multiple edges that have the same source and destination nodes.
Parameters
incoming_graph_data
array, SciPy sparse array, or PyGraphviz graph.

multigraph_input
[bool or None (default None)] Note: Only used when incoming_graph_data is a dict.
If True, incoming_graph_data is assumed to be a dict-of-dict-of-dict-of-dict structure
keyed by node to neighbor to edge keys to edge data for multi-edges. A NetworkXError is
raised if this is not the case. If False, to_networkx_graph() is used to try to determine
the dict’s graph data structure as either a dict-of-dict-of-dict keyed by node to neighbor to edge
data, or a dict-of-iterable keyed by node to neighbors. If None, the treatment for True is tried,
but if it fails, the treatment for False is tried.
attr
pairs.
See also:
Graph
DiGraph
MultiDiGraph
Examples

Nodes:
>>> G.add_node(1)

>>> G.add_node(H)
Edges:
Add one edge,
>>> key = G.add_edge(1, 2)
a list of edges,
>>> keys = G.add_edges_from([(1, 2), (1, 3)])

>>> keys = G.add_edges_from(H.edges)
If some edges connect nodes not yet in the graph, the nodes are added automatically. If an edge already exists,
an additional edge is created and stored using a key to identify the edge. By default the key is the lowest unused
integer.
>>> keys = G.add_edges_from([(4, 5, {"route": 28}), (4, 5, {"route": 37})])
>>> G[4]
AdjacencyView({3: {0: {}}, 5: {0: {}, 1: {'route': 28}, 2: {'route': 37}}})
Attributes:
>>> G = nx.MultiGraph(day="Friday")
>>> G.graph
{'day': 'Friday'}

>>> G.nodes[1]
{'time': '5pm'}
>>> G.nodes[1]["room"] = 714
[(1, {'time': '5pm'}), (3, {'time': '2pm'})]

>>> key = G.add_edge(1, 2, weight=4.7)
>>> keys = G.add_edges_from([(3, 4), (4, 5)], color="red")
>>> keys = G.add_edges_from([(1, 2, {"color": "blue"}), (2, 3, {"weight": 8})])
>>> G[1][2][0]["weight"] = 4.7
>>> G.edges[1, 2, 0]["weight"] = 4
Warning: we protect the graph data structure by making G.edges[1, 2, 0] a read-only dict-like structure.
However, you can assign to attributes in e.g. G.edges[1, 2, 0]. Thus, use 2 sets of brackets to add/change
data attributes: G.edges[1, 2, 0]['weight'] = 4.
Shortcuts:
True
[1, 2]
5
>>> G[1] # adjacency dict-like view mapping neighbor -> edge key -> edge␣
,→attributes
AdjacencyView({2: {0: {'weight': 4}, 1: {'color': 'blue'}}})
dict G.adj or G.adjacency().


... for nbr, keydict in nbrsdict.items():
... for key, eattr in keydict.items():
... pass
>>> for u, v, keys, weight in G.edges(data="weight", keys=True):

... pass
Reporting:
Simple graph information is obtained using methods and object-attributes. Reporting usually provides views instead
of containers to reduce memory usage. The views update as the graph is updated similarly to dict-views. The ob-
jects nodes, edges and adj provide access to data attributes via lookup (e.g. nodes[n], edges[u, v, k],
gree.
The MultiGraph class uses a dict-of-dict-of-dict-of-dict data structure. The outer dict (node_dict) holds adja-
cency information keyed by node. The next dict (adjlist_dict) represents the adjacency information and holds
edge_key dicts keyed by neighbor. The edge_key dict holds each edge_attr dict keyed by edge key. The inner dict
(edge_attr_dict) represents the edge data and holds edge attribute values keyed by attribute names.
Each of these four dicts in the dict-of-dict-of-dict-of-dict structure can be replaced by a user defined dict-like object.
In general, the dict-like features should be maintained but extra features can be added. To replace one of the dicts
create a new graph class by changing the class(!) variable holding the factory for that dict-like structure. The vari-
edge_key_dict_factory, edge_attr_dict_factory and graph_attr_dict_factory.
node_dict_factory
[function, (default: dict)] Factory function to be used to create the adjacency list dict which holds multiedge
key dicts keyed by neighbor. It should require no arguments and return a dict-like object.
edge_key_dict_factory
[function, (default: dict)] Factory function to be used to create the edge key dict which holds edge data keyed
by edge key. It should require no arguments and return a dict-like object.

to_directed_class
to_undirected_class
Subclassing Example

...
...
>>> G = ThinGraph()
>>> G[2][1]
{'weight': 1}
>>> G[2][1] is G[2][2]
True
Methods

MultiGraph.__init__([incoming_graph_data, Initialize a graph with edges, name, or graph attributes.

...])
MultiGraph.add_node(node_for_adding, **attr)Add a single node node_for_adding and update
node attributes.
MultiGraph.add_nodes_from(nodes_for_adding, Add multiple nodes.
...)
MultiGraph.remove_node(n) Remove node n.
MultiGraph.remove_nodes_from(nodes) Remove multiple nodes.
MultiGraph.add_edge(u_for_edge, v_for_edge) Add an edge between u and v.
MultiGraph.add_edges_from(ebunch_to_add, Add all the edges in ebunch_to_add.
**attr)
MultiGraph.add_weighted_edges_from(ebunch_to_add)
Add weighted edges in ebunch_to_add with specified
weight attr
MultiGraph.new_edge_key(u, v) Returns an unused key for edges between nodes u and v.
MultiGraph.remove_edge(u, v[, key]) Remove an edge between u and v.
MultiGraph.remove_edges_from(ebunch) Remove all edges specified in ebunch.
MultiGraph.update([edges, nodes]) Update the graph using nodes/edges/graphs as input.
MultiGraph.clear() Remove all nodes and edges from the graph.
MultiGraph.clear_edges() Remove all edges from the graph without altering nodes.
MultiGraph.__init__
MultiGraph.__init__(incoming_graph_data=None, multigraph_input=None, **attr)

Parameters
incoming_graph_data
[input graph] Data to initialize graph. If incoming_graph_data=None (default) an empty graph
is created. The data can be an edge list, or any NetworkX graph object. If the corresponding
optional Python packages are installed the data can also be a 2D NumPy array, a SciPy sparse
array, or a PyGraphviz graph.
multigraph_input
attr
pairs.
See also:
convert

Examples
>>> G = nx.MultiGraph(name="my graph")
>>> e = [(1, 2), (1, 2), (2, 3), (3, 4)] # list of edges
>>> G = nx.MultiGraph(e)
>>> G = nx.MultiGraph(e, day="Friday")

>>> G.graph
{'day': 'Friday'}
MultiGraph.add_node
MultiGraph.add_node(node_for_adding, **attr)
Parameters
node_for_adding
attr
See also:
add_nodes_from
Notes
Examples

>>> G.add_node(1)
>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
>>> G.add_node(K3)
3


MultiGraph.add_nodes_from
MultiGraph.add_nodes_from(nodes_for_adding, **attr)
Add multiple nodes.
Parameters
nodes_for_adding
attr
word arguments.
See also:
add_node
Examples

>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
[0, 1, 2, 'H', 'e', 'l', 'o']


11
>>> H = nx.Graph()
11
MultiGraph.remove_node
MultiGraph.remove_node(n)
Remove node n.
Parameters
n

Raises
NetworkXError
See also:
remove_nodes_from
Examples

>>> list(G.edges)
[(0, 1), (1, 2)]
>>> list(G.edges)
[]
MultiGraph.remove_nodes_from
MultiGraph.remove_nodes_from(nodes)
Parameters
nodes
See also:
remove_node
Examples

>>> e
[0, 1, 2]
>>> list(G.nodes)
[]

MultiGraph.add_edge
MultiGraph.add_edge(u_for_edge, v_for_edge, key=None, **attr)

below.
Parameters
u_for_edge, v_for_edge
key
[hashable identifier, optional (default=lowest unused integer)] Used to distinguish multiedges
between a pair of nodes.
attr
arguments.
Returns
The edge key assigned to the edge.
See also:
add_edges_from
Notes
To replace/update edge data, use the optional key argument to identify a unique edge. Otherwise a new edge will
be created.
NetworkX algorithms designed for weighted graphs cannot use multigraphs directly because it is not clear how to
handle multiedge weights. Convert to Graph using edge attribute ‘weight’ to enable weighted graph algorithms.
Default keys are generated using the method new_edge_key(). This method can be overridden by subclassing
the base class and providing a custom new_edge_key() method.
Examples
The following each add an additional edge e=(1, 2) to graph G:
>>> e = (1, 2)
>>> ekey = G.add_edge(1, 2) # explicit two-node form
1
[2]

>>> ekey = G.add_edge(1, 2, weight=3)

>>> ekey = G.add_edge(1, 2, key=0, weight=4) # update data for key=0
>>> ekey = G.add_edge(1, 3, weight=7, capacity=15, length=342.7)
>>> ekey = G.add_edge(1, 2)

>>> G[1][2][0].update({0: 5})
>>> G.edges[1, 2, 0].update({0: 5})
MultiGraph.add_edges_from
MultiGraph.add_edges_from(ebunch_to_add, **attr)
Parameters
ebunch_to_add
can be:
• 2-tuples (u, v) or
• 3-tuples (u, v, d) for an edge data dict d, or
• 3-tuples (u, v, k) for not iterable key k, or
• 4-tuples (u, v, k, d) for an edge with data and key k
attr
arguments.
Returns
A list of edge keys assigned to the edges in ebunch.
See also:
add_edge
add a single edge
Notes

Examples

>>> e = zip(range(0, 3), range(1, 4))

MultiGraph.add_weighted_edges_from
MultiGraph.add_weighted_edges_from(ebunch_to_add, weight='weight', **attr)

Parameters
ebunch_to_add
weight
attr
edges.
See also:
add_edge
add a single edge
add_edges_from
add multiple edges
Notes
Examples


MultiGraph.new_edge_key
MultiGraph.new_edge_key(u, v)
Returns an unused key for edges between nodes u and v.
The nodes u and v do not need to be already in the graph.
Parameters
u, v
[nodes]
Returns
key
[int]
Notes
In the standard MultiGraph class the new key is the number of existing edges between u and v (increased if nec-
essary to ensure unused). The first edge will have key 0, then 1, etc. If an edge is removed further new_edge_keys
may not be in this order.
MultiGraph.remove_edge
MultiGraph.remove_edge(u, v, key=None)
Remove an edge between u and v.
Parameters
u, v
[nodes] Remove an edge between nodes u and v.
key
[hashable identifier, optional (default=None)] Used to distinguish multiple edges between a
pair of nodes. If None, remove a single edge between u and v. If there are multiple edges,
removes the last edge added in terms of insertion order.
Raises
NetworkXError
If there is not an edge between u and v, or if there is no edge with the specified key.
See also:
remove_edges_from

Examples
>>> nx.add_path(G, [0, 1, 2, 3])
>>> e = (1, 2)
For multiple edges
>>> G = nx.MultiGraph() # or MultiDiGraph, etc

>>> G.add_edges_from([(1, 2), (1, 2), (1, 2)]) # key_list returned
[0, 1, 2]
When key=None (the default), edges are removed in the opposite order that they were added:
>>> G.edges(keys=True)
MultiEdgeView([(1, 2, 0), (1, 2, 1)])
>>> G.remove_edge(2, 1) # edges are not directed
MultiEdgeView([(1, 2, 0)])
For edges with keys
>>> G.add_edge(1, 2, key="first")
'first'
>>> G.add_edge(1, 2, key="second")
'second'
>>> G.remove_edge(1, 2, key="first")
MultiEdgeView([(1, 2, 'second')])
MultiGraph.remove_edges_from
MultiGraph.remove_edges_from(ebunch)
Parameters
• 2-tuples (u, v) A single edge between u and v is removed.
• 3-tuples (u, v, key) The edge identified by key is removed.
• 4-tuples (u, v, key, data) where data is ignored.
See also:
remove_edge

Notes
Examples

>>> ebunch = [(1, 2), (2, 3)]
Removing multiple copies of edges
>>> keys = G.add_edges_from([(1, 2), (1, 2), (1, 2)])
>>> G.remove_edges_from([(1, 2), (2, 1)]) # edges aren't directed
>>> list(G.edges())
[(1, 2)]
>>> G.remove_edges_from([(1, 2), (1, 2)]) # silently ignore extra copy
>>> list(G.edges) # now empty graph
[]
When the edge is a 2-tuple (u, v) but there are multiple edges between u and v in the graph, the most recent
edge (in terms of insertion order) is removed.
>>> for key in ("x", "y", "a"):
... k = G.add_edge(0, 1, key=key)
MultiEdgeView([(0, 1, 'x'), (0, 1, 'y'), (0, 1, 'a')])
>>> G.remove_edges_from([(0, 1)])
MultiEdgeView([(0, 1, 'x'), (0, 1, 'y')])
MultiGraph.update
MultiGraph.update(edges=None, nodes=None)
nodes must be used.
Parameters
edges

nodes
See also:
add_edges_from
add_nodes_from
Notes
>>> adj = {1: {2, 3}, 2: {1, 3}, 3: {1, 2}}
>>> adj = {1: {2: 1.3, 3: 0.7}, 2: {1: 1.4}, 3: {1: 0.7}}
>>> e = [
... (u, v, {"weight": d})
... ]
>>> e = [
... (u, v, {"weight": d})
... ]

>>> pred = {1: {2, 3}, 2: {3}, 3: {3}}

>>> adj = {
... 1: {2: {0: {"weight": 1.3}, 1: {"weight": 1.2}}},
... 3: {2: {0: {"weight": 0.7}}},
... }
>>> e = [
... (u, v, ekey, d)


... ]
Examples
>>> edges = (
... (u, v, {"power": u * v})
... if u * v < 225
... )
MultiGraph.clear
MultiGraph.clear()
Examples

>>> G.clear()
>>> list(G.nodes)
[]
>>> list(G.edges)
[]
MultiGraph.clear_edges
MultiGraph.clear_edges()

Examples

>>> G.clear_edges()
>>> list(G.nodes)
[0, 1, 2, 3]
>>> list(G.edges)
[]
MultiGraph.nodes A NodeView of the Graph as G.nodes or G.nodes().

MultiGraph.__iter__() Iterate over the nodes.
MultiGraph.has_node(n) Returns True if the graph contains the node n.
MultiGraph.__contains__(n) Returns True if n is a node, False otherwise.
MultiGraph.edges Returns an iterator over the edges.
MultiGraph.has_edge(u, v[, key]) Returns True if the graph has an edge between nodes u
and v.
MultiGraph.get_edge_data(u, v[, key, default]) Returns the attribute dictionary associated with edge (u,
v, key).
MultiGraph.neighbors(n) Returns an iterator over all neighbors of node n.
MultiGraph.adj Graph adjacency object holding the neighbors of each
node.
MultiGraph.__getitem__(n) Returns a dict of neighbors of node n.
MultiGraph.adjacency() Returns an iterator over (node, adjacency dict) tuples for
all nodes.
MultiGraph.nbunch_iter([nbunch]) Returns an iterator over nodes contained in nbunch that
MultiGraph.nodes
property MultiGraph.nodes
Parameters
data
default
Returns

NodeView
Notes
Examples
>>> list(G.nodes)
[0, 1, 2]
>>> list(G)
[0, 1, 2]

[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
[(0, 'bar'), (1, None), (2, None)]
[(0, 'bar'), (1, None), (2, None)]
[(0, None), (1, '5pm'), (2, None)]
[(0, None), (1, '5pm'), (2, None)]

>>> G = nx.Graph()
>>> G.add_node(0)


{0: 1, 1: 2, 2: 3}
MultiGraph.__iter__
MultiGraph.__iter__()
Returns
niter
Examples

>>> [n for n in G]
[0, 1, 2, 3]
>>> list(G)
[0, 1, 2, 3]
MultiGraph.has_node
MultiGraph.has_node(n)
Identical to n in G
Parameters
n
[node]
Examples

>>> G.has_node(0)
True
>>> 0 in G
True

MultiGraph.__contains__
MultiGraph.__contains__(n)
Examples

>>> 1 in G
True
MultiGraph.edges
property MultiGraph.edges
Returns an iterator over the edges.
edges(self, nbunch=None, data=False, keys=False, default=None)
The MultiEdgeView provides set-like operations on the edge-tuples as well as edge attribute lookup. When called,
set-like operations). Hence, G.edges[u, v, k]['color'] provides the value of the color attribute for the
edge from u to v with key k while for (u, v, k, c) in G.edges(data='color', keys=True,
default="red"): iterates through all the edges yielding the color attribute with default 'red' if no color
attribute exists.
Edges are returned as tuples with optional data and keys in the order (node, neighbor, key, data). If keys=True
is not provided, the tuples will just be (node, neighbor, data), but multiple tuples with the same node and neighbor
will be generated when multiple edges exist between two nodes.
Parameters
nbunch
these nodes.
data
(u, v).
keys
[bool, optional (default=False)] If True, return edge keys with each edge, creating (u, v, k)
tuples or (u, v, k, d) tuples if data is also requested.
default
Returns
edges
[MultiEdgeView] A view of edge attributes, usually it iterates over (u, v) (u, v, k) or (u, v, k, d)
tuples of edges, but can also be used for attribute lookup as edges[u, v, k]['foo'].

Notes
Examples
>>> nx.add_path(G, [0, 1, 2])
>>> key = G.add_edge(2, 3, weight=5)
>>> key2 = G.add_edge(2, 1, weight=2) # multi-edge
>>> [e for e in G.edges()]
[(0, 1), (1, 2), (1, 2), (2, 3)]
MultiEdgeDataView([(0, 1, {}), (1, 2, {}), (1, 2, {'weight': 2}), (2, 3, {'weight
,→': 5})])

MultiEdgeDataView([(0, 1, 1), (1, 2, 1), (1, 2, 2), (2, 3, 5)])
>>> G.edges(keys=True) # default keys are integers
MultiEdgeView([(0, 1, 0), (1, 2, 0), (1, 2, 1), (2, 3, 0)])
>>> G.edges.data(keys=True)
MultiEdgeDataView([(0, 1, 0, {}), (1, 2, 0, {}), (1, 2, 1, {'weight': 2}), (2, 3,␣
,→0, {'weight': 5})])
>>> G.edges.data("weight", default=1, keys=True)

MultiEdgeDataView([(0, 1, 0, 1), (1, 2, 0, 1), (1, 2, 1, 2), (2, 3, 0, 5)])
>>> G.edges([0, 3]) # Note ordering of tuples from listed sources
MultiEdgeDataView([(0, 1), (3, 2)])
>>> G.edges([0, 3, 2, 1]) # Note ordering of tuples
MultiEdgeDataView([(0, 1), (3, 2), (2, 1), (2, 1)])
>>> G.edges(0)
MultiEdgeDataView([(0, 1)])
MultiGraph.has_edge
MultiGraph.has_edge(u, v, key=None)
Returns True if the graph has an edge between nodes u and v.
This is the same as v in G[u] or key in G[u][v] without KeyError exceptions.
Parameters
u, v
[nodes] Nodes can be, for example, strings or numbers.
key
[hashable identifier, optional (default=None)] If specified return True only if the edge with key
is found.
Returns
edge_ind

Examples
Can be called either using two nodes u, v, an edge tuple (u, v), or an edge tuple (u, v, key).
>>> G = nx.MultiGraph() # or MultiDiGraph
>>> nx.add_path(G, [0, 1, 2, 3])
True
>>> e = (0, 1)
True
>>> G.add_edge(0, 1, key="a")
'a'
>>> G.has_edge(0, 1, key="a") # specify key
True
>>> G.has_edge(1, 0, key="a") # edges aren't directed
True
>>> e = (0, 1, "a")
>>> G.has_edge(*e) # e is a 3-tuple (u, v, 'a')
True

True
>>> 1 in G[0] # though this gives :exc:`KeyError` if 0 not in G
True
>>> 0 in G[1] # other order; also gives :exc:`KeyError` if 0 not in G
True
MultiGraph.get_edge_data
MultiGraph.get_edge_data(u, v, key=None, default=None)

Returns the attribute dictionary associated with edge (u, v, key).
If a key is not provided, returns a dictionary mapping edge keys to attribute dictionaries for each edge between u
and v.
This is identical to G[u][v][key] except the default is returned instead of an exception is the edge doesn’t exist.
Parameters
u, v
[nodes]
default
[any Python object (default=None)] Value to return if the specific edge (u, v, key) is not found,
OR if there are no edges between u and v and no key is specified.
key
[hashable identifier, optional (default=None)] Return data only for the edge with specified key,
as an attribute dictionary (rather than a dictionary mapping keys to attribute dictionaries).
Returns
edge_dict
[dictionary] The edge attribute dictionary, OR a dictionary mapping edge keys to attribute
dictionaries for each of those edges if no specific key is provided (even if there’s only one edge
between u and v).

Examples

>>> key = G.add_edge(0, 1, key="a", weight=7)
>>> G[0][1]["a"] # key='a'
{'weight': 7}
>>> G.edges[0, 1, "a"] # key='a'
{'weight': 7}
Warning: we protect the graph data structure by making G.edges and G[1][2] read-only dict-like structures.
However, you can assign values to attributes in e.g. G.edges[1, 2, 'a'] or G[1][2]['a'] using an
additional bracket as shown next. You need to specify all edge info to assign to the edge data associated with an
edge.
>>> G[0][1]["a"]["weight"] = 10
>>> G.edges[0, 1, "a"]["weight"] = 10
>>> G[0][1]["a"]["weight"]
10
>>> G.edges[1, 0, "a"]["weight"]
10

>>> nx.add_path(G, [0, 1, 2, 3])
>>> G.edges[0, 1, 0]["weight"] = 5
>>> G.get_edge_data(0, 1)
{0: {'weight': 5}}
>>> e = (0, 1)
{0: {'weight': 5}}
>>> G.get_edge_data(3, 0) # edge not in graph, returns None
>>> G.get_edge_data(3, 0, default=0) # edge not in graph, return default
0
>>> G.get_edge_data(1, 0, 0) # specific key gives back
{'weight': 5}
MultiGraph.neighbors
MultiGraph.neighbors(n)
Returns an iterator over all neighbors of node n.
This is identical to iter(G[n])
Parameters
n
Returns
neighbors
[iterator] An iterator over all neighbors of node n
Raises
NetworkXError
If the node n is not in the graph.

Notes
Alternate ways to access the neighbors are G.adj[n] or G[n]:

>>> G["a"]
AtlasView({'b': {'weight': 7}})
>>> [n for n in G[0]]
[1]
Examples

>>> [n for n in G.neighbors(0)]
[1]
MultiGraph.adj
property MultiGraph.adj
This object is a read-only dict-like structure with node keys and neighbor-dict values. The neighbor-dict is keyed by
neighbor to the edgekey-data-dict. So G.adj[3][2][0]['color'] = 'blue' sets the color of the edge
(3, 2, 0) to "blue".
Iterating over G.adj behaves like a dict. Useful idioms include for nbr, edgesdict in G.adj[n].
items():.
The neighbor information is also provided by subscripting the graph.
Examples
>>> e = [(1, 2), (1, 2), (1, 3), (3, 4)] # list of edges
>>> G = nx.MultiGraph(e)
>>> G.edges[1, 2, 0]["weight"] = 3
>>> result = set()
>>> for edgekey, data in G[1][2].items():
... result.add(data.get('weight', 1))
>>> result
{1, 3}

MultiGraph.__getitem__
MultiGraph.__getitem__(n)
Parameters
n
Returns
adj_dict
Notes
Examples

>>> G[0]
AtlasView({1: {}})
MultiGraph.adjacency
MultiGraph.adjacency()
Returns
adj_iter
Examples

[(0, {1: {}}), (1, {0: {}, 2: {}}), (2, {1: {}, 3: {}}), (3, {2: {}})]

MultiGraph.nbunch_iter
MultiGraph.nbunch_iter(nbunch=None)
Parameters
nbunch
Returns
niter
Raises
NetworkXError
See also:
Graph.__iter__
Notes
MultiGraph.order() Returns the number of nodes in the graph.

MultiGraph.number_of_nodes() Returns the number of nodes in the graph.
MultiGraph.__len__() Returns the number of nodes in the graph.
MultiGraph.degree A DegreeView for the Graph as G.degree or G.degree().
MultiGraph.size([weight]) Returns the number of edges or total of all edge weights.
MultiGraph.number_of_edges([u, v]) Returns the number of edges between two nodes.

MultiGraph.order
MultiGraph.order()
Returns
nnodes
See also:
number_of_nodes
identical method
__len__
identical method
Examples

>>> G.order()
3
MultiGraph.number_of_nodes
MultiGraph.number_of_nodes()
Returns
nnodes
See also:
order
identical method
__len__
identical method
Examples

3

MultiGraph.__len__
MultiGraph.__len__()
Returns
nnodes
See also:
number_of_nodes
identical method
order
identical method
Examples

>>> len(G)
4
MultiGraph.degree
property MultiGraph.degree
Parameters
nbunch
weight
Returns
MultiDegreeView or int
If multiple nodes are requested (the default), returns a MultiDegreeView mapping nodes
to their degree. If a single node is requested, returns the degree of the node as an integer.

Examples

>>> nx.add_path(G, [0, 1, 2, 3])
1
>>> list(G.degree([0, 1]))
[(0, 1), (1, 2)]
MultiGraph.size
MultiGraph.size(weight=None)
Parameters
weight
Returns
size
more general).
See also:
number_of_edges
Examples

>>> G.size()
3

>>> G.size()
2
6.0

MultiGraph.number_of_edges
MultiGraph.number_of_edges(u=None, v=None)
Parameters
u, v
[nodes, optional (Gefault=all edges)] If u and v are specified, return the number of edges
between u and v. Otherwise return the total number of all edges.
Returns
nedges
from u to v.
See also:
size
Examples
For undirected multigraphs, this method counts the total number of edges in the graph:
>>> G.add_edges_from([(0, 1), (0, 1), (1, 2)])
[0, 1, 0]
3
2
For directed multigraphs, this method can count the total number of directed edges from u to v:
>>> G.add_edges_from([(0, 1), (0, 1), (1, 0)])
[0, 1, 0]
2
1

MultiGraph.copy([as_view]) Returns a copy of the graph.

MultiGraph.to_undirected([as_view]) Returns an undirected copy of the graph.
MultiGraph.to_directed([as_view]) Returns a directed representation of the graph.
MultiGraph.subgraph(nodes) Returns a SubGraph view of the subgraph induced on
nodes.
MultiGraph.edge_subgraph(edges) Returns the subgraph induced by the specified edges.
MultiGraph.copy
MultiGraph.copy(as_view=False)
containers.
Parameters
as_view
Returns
G
See also:
to_directed
Notes
copy.deepcopy)
shallow copy.

>>> H = G.copy()
>>> H = nx.Graph(G)
>>> H = G.__class__(G)
>>> H = G.__class__()
copy.html.
Examples

>>> H = G.copy()
MultiGraph.to_undirected
MultiGraph.to_undirected(as_view=False)
Returns an undirected copy of the graph.
Returns
G
[Graph/MultiGraph] A deepcopy of the graph.
See also:
copy, add_edge, add_edges_from
Notes
and references.
This is in contrast to the similar G = nx.MultiGraph(D) which returns a shallow copy of the data.
copy.html.
Warning: If you have subclassed MultiGraph to use dict-like objects in the data structure, those changes do not
transfer to the MultiGraph created by this method.

Examples
>>> G = nx.MultiGraph([(0, 1), (0, 1), (1, 2)])

>>> list(H.edges)
[(0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1), (1, 2, 0), (2, 1, 0)]
>>> list(G2.edges)
[(0, 1, 0), (0, 1, 1), (1, 2, 0)]
MultiGraph.to_directed
MultiGraph.to_directed(as_view=False)
Returns
G
[MultiDiGraph] A directed graph with the same name, same nodes, and with each edge (u, v,
k, data) replaced by two directed edges (u, v, k, data) and (v, u, k, data).
Notes
and references.
This is in contrast to the similar D=MultiDiGraph(G) which returns a shallow copy of the data.
copy.html.
transfer to the MultiDiGraph created by this method.
Examples
0
1
>>> list(H.edges)
[(0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1)]
0
>>> list(H.edges)
[(0, 1, 0)]

MultiGraph.subgraph
MultiGraph.subgraph(nodes)
Parameters
nodes
Returns
G
Notes
not in set(nodes)])

SG = G.__class__()
else:
Examples

>>> H = G.subgraph([0, 1, 2])
>>> list(H.edges)
[(0, 1), (1, 2)]

MultiGraph.edge_subgraph
MultiGraph.edge_subgraph(edges)
Parameters
edges
Returns
G
Notes
Examples
>>> H = G.edge_subgraph([(0, 1), (3, 4)])
>>> list(H.nodes)
[0, 1, 3, 4]
>>> list(H.edges)
[(0, 1), (3, 4)]
2.2.4 MultiDiGraph—Directed graphs with self loops and parallel edges
Overview
class MultiDiGraph(incoming_graph_data=None, multigraph_input=None, **attr)

A directed graph class that can store multiedges.
Multiedges are multiple edges between two nodes. Each edge can hold optional data or attributes.
A MultiDiGraph holds directed edges. Self loops are allowed.
used as a node.
Parameters

incoming_graph_data
multigraph_input
attr
pairs.
See also:
Graph
DiGraph
MultiGraph
Examples

Nodes:
>>> G.add_node(1)

>>> G.add_node(H)
Edges:
Add one edge,

>>> key = G.add_edge(1, 2)
a list of edges,
>>> keys = G.add_edges_from([(1, 2), (1, 3)])
>>> keys = G.add_edges_from(H.edges)
If some edges connect nodes not yet in the graph, the nodes are added automatically. If an edge already exists,
an additional edge is created and stored using a key to identify the edge. By default the key is the lowest unused
integer.
>>> keys = G.add_edges_from([(4, 5, dict(route=282)), (4, 5, dict(route=37))])

>>> G[4]
AdjacencyView({5: {0: {}, 1: {'route': 282}, 2: {'route': 37}}})
Attributes:
>>> G = nx.MultiDiGraph(day="Friday")
>>> G.graph
{'day': 'Friday'}

>>> G.nodes[1]
{'time': '5pm'}
>>> G.nodes[1]["room"] = 714
[(1, {'time': '5pm'}), (3, {'time': '2pm'})]
>>> key = G.add_edge(1, 2, weight=4.7)

>>> keys = G.add_edges_from([(3, 4), (4, 5)], color="red")
>>> keys = G.add_edges_from([(1, 2, {"color": "blue"}), (2, 3, {"weight": 8})])
>>> G[1][2][0]["weight"] = 4.7
>>> G.edges[1, 2, 0]["weight"] = 4
Warning: we protect the graph data structure by making G.edges[1, 2, 0] a read-only dict-like structure.
However, you can assign to attributes in e.g. G.edges[1, 2, 0]. Thus, use 2 sets of brackets to add/change
data attributes: G.edges[1, 2, 0]['weight'] = 4 (for multigraphs the edge key is required: MG.
edges[u, v, key][name] = value).
Shortcuts:


True
[1, 2]
5
>>> G[1] # adjacency dict-like view mapping neighbor -> edge key -> edge␣
,→attributes
AdjacencyView({2: {0: {'weight': 4}, 1: {'color': 'blue'}}})
Often the best way to traverse all edges of a graph is via the neighbors. The neighbors are available as an adjacency-
view G.adj object or via the method G.adjacency().

... for nbr, keydict in nbrsdict.items():
... for key, eattr in keydict.items():
... pass
>>> for u, v, keys, weight in G.edges(data="weight", keys=True):

... pass
Reporting:
Simple graph information is obtained using methods and object-attributes. Reporting usually provides views instead
of containers to reduce memory usage. The views update as the graph is updated similarly to dict-views. The ob-
jects nodes, edges and adj provide access to data attributes via lookup (e.g. nodes[n], edges[u, v, k],
gree.
The MultiDiGraph class uses a dict-of-dict-of-dict-of-dict structure. The outer dict (node_dict) holds adjacency in-
formation keyed by node. The next dict (adjlist_dict) represents the adjacency information and holds edge_key dicts
keyed by neighbor. The edge_key dict holds each edge_attr dict keyed by edge key. The inner dict (edge_attr_dict)
represents the edge data and holds edge attribute values keyed by attribute names.
Each of these four dicts in the dict-of-dict-of-dict-of-dict structure can be replaced by a user defined dict-like object.
In general, the dict-like features should be maintained but extra features can be added. To replace one of the dicts
create a new graph class by changing the class(!) variable holding the factory for that dict-like structure. The vari-
edge_key_dict_factory, edge_attr_dict_factory and graph_attr_dict_factory.
node_dict_factory

[function, (default: dict)] Factory function to be used to create the adjacency list dict which holds multiedge
key dicts keyed by neighbor. It should require no arguments and return a dict-like object.
edge_key_dict_factory
[function, (default: dict)] Factory function to be used to create the edge key dict which holds edge data keyed
by edge key. It should require no arguments and return a dict-like object.
to_directed_class
to_undirected_class
Subclassing Example

...
...
>>> G = ThinGraph()
>>> G[2][1]
{'weight': 1}
>>> G[2][1] is G[2][2]
True

Methods
Adding and Removing Nodes and Edges
MultiDiGraph.__init__([incoming_graph_data, Initialize a graph with edges, name, or graph attributes.

...])
MultiDiGraph.add_node(node_for_adding, Add a single node node_for_adding and update
**attr) node attributes.
MultiDiGraph.add_nodes_from(...) Add multiple nodes.
MultiDiGraph.remove_node(n) Remove node n.
MultiDiGraph.remove_nodes_from(nodes) Remove multiple nodes.
MultiDiGraph.add_edge(u_for_edge, Add an edge between u and v.
v_for_edge)
MultiDiGraph.add_edges_from(ebunch_to_add, Add all the edges in ebunch_to_add.
...)
MultiDiGraph.add_weighted_edges_from(...[, Add weighted edges in ebunch_to_add with specified
...]) weight attr
MultiDiGraph.new_edge_key(u, v) Returns an unused key for edges between nodes u and v.
MultiDiGraph.remove_edge(u, v[, key]) Remove an edge between u and v.
MultiDiGraph.remove_edges_from(ebunch) Remove all edges specified in ebunch.
MultiDiGraph.update([edges, nodes]) Update the graph using nodes/edges/graphs as input.
MultiDiGraph.clear() Remove all nodes and edges from the graph.
MultiDiGraph.clear_edges() Remove all edges from the graph without altering nodes.
MultiDiGraph.__init__
MultiDiGraph.__init__(incoming_graph_data=None, multigraph_input=None, **attr)

Parameters
incoming_graph_data
[input graph] Data to initialize graph. If incoming_graph_data=None (default) an empty graph
is created. The data can be an edge list, or any NetworkX graph object. If the corresponding
optional Python packages are installed the data can also be a 2D NumPy array, a SciPy sparse
array, or a PyGraphviz graph.
multigraph_input
attr
pairs.
See also:
convert

Examples

>>> e = [(1, 2), (2, 3), (3, 4)] # list of edges
>>> G = nx.Graph(e)

>>> G.graph
{'day': 'Friday'}
MultiDiGraph.add_node
MultiDiGraph.add_node(node_for_adding, **attr)
Parameters
node_for_adding
attr
See also:
add_nodes_from
Notes
Examples

>>> G.add_node(1)
>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
>>> G.add_node(K3)
3


MultiDiGraph.add_nodes_from
MultiDiGraph.add_nodes_from(nodes_for_adding, **attr)
Add multiple nodes.
Parameters
nodes_for_adding
attr
word arguments.
See also:
add_node
Examples

>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
[0, 1, 2, 'H', 'e', 'l', 'o']


11
>>> H = nx.Graph()
11
MultiDiGraph.remove_node
MultiDiGraph.remove_node(n)
Remove node n.
Parameters
n

Raises
NetworkXError
See also:
remove_nodes_from
Examples

>>> list(G.edges)
[(0, 1), (1, 2)]
>>> list(G.edges)
[]
MultiDiGraph.remove_nodes_from
MultiDiGraph.remove_nodes_from(nodes)
Parameters
nodes
See also:
remove_node
Examples

>>> e
[0, 1, 2]
>>> list(G.nodes)
[]

MultiDiGraph.add_edge
MultiDiGraph.add_edge(u_for_edge, v_for_edge, key=None, **attr)

below.
Parameters
u_for_edge, v_for_edge
key
[hashable identifier, optional (default=lowest unused integer)] Used to distinguish multiedges
between a pair of nodes.
attr
arguments.
Returns
The edge key assigned to the edge.
See also:
add_edges_from
Notes
To replace/update edge data, use the optional key argument to identify a unique edge. Otherwise a new edge will
be created.
NetworkX algorithms designed for weighted graphs cannot use multigraphs directly because it is not clear how to
handle multiedge weights. Convert to Graph using edge attribute ‘weight’ to enable weighted graph algorithms.
Examples
>>> e = (1, 2)
>>> key = G.add_edge(1, 2) # explicit two-node form
1
[2]


>>> key = G.add_edge(1, 2, key=0, weight=4) # update data for key=0
>>> key = G.add_edge(1, 3, weight=7, capacity=15, length=342.7)

>>> G[1][2][0].update({0: 5})
>>> G.edges[1, 2, 0].update({0: 5})
MultiDiGraph.add_edges_from
MultiDiGraph.add_edges_from(ebunch_to_add, **attr)
Parameters
ebunch_to_add
can be:
• 2-tuples (u, v) or
• 3-tuples (u, v, d) for an edge data dict d, or
• 3-tuples (u, v, k) for not iterable key k, or
• 4-tuples (u, v, k, d) for an edge with data and key k
attr
arguments.
Returns
A list of edge keys assigned to the edges in ebunch.
See also:
add_edge
add a single edge
Notes

Examples

>>> e = zip(range(0, 3), range(1, 4))

MultiDiGraph.add_weighted_edges_from
MultiDiGraph.add_weighted_edges_from(ebunch_to_add, weight='weight', **attr)

Parameters
ebunch_to_add
weight
attr
edges.
See also:
add_edge
add a single edge
add_edges_from
add multiple edges
Notes
Examples


MultiDiGraph.new_edge_key
MultiDiGraph.new_edge_key(u, v)
Returns an unused key for edges between nodes u and v.
The nodes u and v do not need to be already in the graph.
Parameters
u, v
[nodes]
Returns
key
[int]
Notes
In the standard MultiGraph class the new key is the number of existing edges between u and v (increased if nec-
essary to ensure unused). The first edge will have key 0, then 1, etc. If an edge is removed further new_edge_keys
may not be in this order.
MultiDiGraph.remove_edge
MultiDiGraph.remove_edge(u, v, key=None)
Remove an edge between u and v.
Parameters
u, v
[nodes] Remove an edge between nodes u and v.
key
[hashable identifier, optional (default=None)] Used to distinguish multiple edges between a
pair of nodes. If None, remove a single edge between u and v. If there are multiple edges,
removes the last edge added in terms of insertion order.
Raises
NetworkXError
If there is not an edge between u and v, or if there is no edge with the specified key.
See also:
remove_edges_from

Examples
>>> nx.add_path(G, [0, 1, 2, 3])
>>> e = (1, 2)
For multiple edges
>>> G.add_edges_from([(1, 2), (1, 2), (1, 2)]) # key_list returned
[0, 1, 2]
When key=None (the default), edges are removed in the opposite order that they were added:
OutMultiEdgeView([(1, 2, 0), (1, 2, 1)])
For edges with keys
>>> G.add_edge(1, 2, key="first")
'first'
>>> G.add_edge(1, 2, key="second")
'second'
>>> G.remove_edge(1, 2, key="first")
OutMultiEdgeView([(1, 2, 'second')])
MultiDiGraph.remove_edges_from
MultiDiGraph.remove_edges_from(ebunch)
Parameters
• 2-tuples (u, v) A single edge between u and v is removed.
• 3-tuples (u, v, key) The edge identified by key is removed.
• 4-tuples (u, v, key, data) where data is ignored.
See also:
remove_edge

Notes
Examples

>>> ebunch = [(1, 2), (2, 3)]
Removing multiple copies of edges
>>> keys = G.add_edges_from([(1, 2), (1, 2), (1, 2)])
>>> G.remove_edges_from([(1, 2), (2, 1)]) # edges aren't directed
>>> list(G.edges())
[(1, 2)]
>>> G.remove_edges_from([(1, 2), (1, 2)]) # silently ignore extra copy
>>> list(G.edges) # now empty graph
[]
When the edge is a 2-tuple (u, v) but there are multiple edges between u and v in the graph, the most recent
edge (in terms of insertion order) is removed.
>>> for key in ("x", "y", "a"):
... k = G.add_edge(0, 1, key=key)
MultiEdgeView([(0, 1, 'x'), (0, 1, 'y'), (0, 1, 'a')])
>>> G.remove_edges_from([(0, 1)])
MultiEdgeView([(0, 1, 'x'), (0, 1, 'y')])
MultiDiGraph.update
MultiDiGraph.update(edges=None, nodes=None)
nodes must be used.
Parameters
edges

nodes
See also:
add_edges_from
add_nodes_from
Notes
>>> adj = {1: {2, 3}, 2: {1, 3}, 3: {1, 2}}
>>> adj = {1: {2: 1.3, 3: 0.7}, 2: {1: 1.4}, 3: {1: 0.7}}
>>> e = [
... (u, v, {"weight": d})
... ]
>>> e = [
... (u, v, {"weight": d})
... ]

>>> pred = {1: {2, 3}, 2: {3}, 3: {3}}

>>> adj = {
... 1: {2: {0: {"weight": 1.3}, 1: {"weight": 1.2}}},
... 3: {2: {0: {"weight": 0.7}}},
... }
>>> e = [
... (u, v, ekey, d)


... ]
Examples
>>> edges = (
... (u, v, {"power": u * v})
... if u * v < 225
... )
MultiDiGraph.clear
MultiDiGraph.clear()
Examples

>>> G.clear()
>>> list(G.nodes)
[]
>>> list(G.edges)
[]
MultiDiGraph.clear_edges
MultiDiGraph.clear_edges()

Examples

>>> G.clear_edges()
>>> list(G.nodes)
[0, 1, 2, 3]
>>> list(G.edges)
[]
MultiDiGraph.nodes A NodeView of the Graph as G.nodes or G.nodes().

MultiDiGraph.__iter__() Iterate over the nodes.
MultiDiGraph.has_node(n) Returns True if the graph contains the node n.
MultiDiGraph.__contains__(n) Returns True if n is a node, False otherwise.
MultiDiGraph.edges An OutMultiEdgeView of the Graph as G.edges or
G.edges().
MultiDiGraph.out_edges An OutMultiEdgeView of the Graph as G.edges or
G.edges().
MultiDiGraph.in_edges A view of the in edges of the graph as G.in_edges or
G.in_edges().
MultiDiGraph.has_edge(u, v[, key]) Returns True if the graph has an edge between nodes u
and v.
MultiDiGraph.get_edge_data(u, v[, key, de- Returns the attribute dictionary associated with edge (u,
fault]) v, key).
MultiDiGraph.neighbors(n) Returns an iterator over successor nodes of n.
MultiDiGraph.adj Graph adjacency object holding the neighbors of each
node.
MultiDiGraph.__getitem__(n) Returns a dict of neighbors of node n.
MultiDiGraph.successors(n) Returns an iterator over successor nodes of n.
MultiDiGraph.succ Graph adjacency object holding the successors of each
node.
MultiDiGraph.predecessors(n) Returns an iterator over predecessor nodes of n.
MultiDiGraph.succ Graph adjacency object holding the successors of each
node.
MultiDiGraph.adjacency() Returns an iterator over (node, adjacency dict) tuples for
all nodes.
MultiDiGraph.nbunch_iter([nbunch]) Returns an iterator over nodes contained in nbunch that
MultiDiGraph.nodes
property MultiDiGraph.nodes

Parameters
data
default
Returns
NodeView
Notes
Examples
>>> list(G.nodes)
[0, 1, 2]
>>> list(G)
[0, 1, 2]

[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
[(0, 'bar'), (1, None), (2, None)]
[(0, 'bar'), (1, None), (2, None)]
[(0, None), (1, '5pm'), (2, None)]
[(0, None), (1, '5pm'), (2, None)]


>>> G = nx.Graph()
>>> G.add_node(0)
{0: 1, 1: 2, 2: 3}
MultiDiGraph.__iter__
MultiDiGraph.__iter__()
Returns
niter
Examples

>>> [n for n in G]
[0, 1, 2, 3]
>>> list(G)
[0, 1, 2, 3]
MultiDiGraph.has_node
MultiDiGraph.has_node(n)
Identical to n in G
Parameters
n
[node]

Examples

>>> G.has_node(0)
True

>>> 0 in G
True
MultiDiGraph.__contains__
MultiDiGraph.__contains__(n)
Examples

>>> 1 in G
True
MultiDiGraph.edges
property MultiDiGraph.edges
An OutMultiEdgeView of the Graph as G.edges or G.edges().
The OutMultiEdgeView provides set-like operations on the edge-tuples as well as edge attribute lookup. When
called, it also provides an EdgeDataView object which allows control of access to edge attributes (but does not
provide set-like operations). Hence, G.edges[u, v, k]['color'] provides the value of the color attribute
for the edge from u to v with key k while for (u, v, k, c) in G.edges(data='color', de-
fault='red', keys=True): iterates through all the edges yielding the color attribute with default 'red'
if no color attribute exists.
will be generated when multiple edges between two nodes exist.
Parameters
nbunch
these nodes.
data
(u, v).
keys
[bool, optional (default=False)] If True, return edge keys with each edge, creating (u, v, k, d)
tuples when data is also requested (the default) and (u, v, k) tuples when data is not requested.

default
Returns
edges
[OutMultiEdgeView] A view of edge attributes, usually it iterates over (u, v) (u, v, k) or (u, v, k,
d) tuples of edges, but can also be used for attribute lookup as edges[u, v, k]['foo'].
See also:
in_edges, out_edges
Notes
Examples
>>> nx.add_path(G, [0, 1, 2])
>>> key2 = G.add_edge(1, 2) # second edge between these nodes
[(0, 1), (1, 2), (1, 2), (2, 3)]
>>> list(G.edges(data=True)) # default data is {} (empty dict)
[(0, 1, {}), (1, 2, {}), (1, 2, {}), (2, 3, {'weight': 5})]
>>> list(G.edges(data="weight", default=1))
[(0, 1, 1), (1, 2, 1), (1, 2, 1), (2, 3, 5)]
>>> list(G.edges(keys=True)) # default keys are integers
[(0, 1, 0), (1, 2, 0), (1, 2, 1), (2, 3, 0)]
>>> list(G.edges(data=True, keys=True))
[(0, 1, 0, {}), (1, 2, 0, {}), (1, 2, 1, {}), (2, 3, 0, {'weight': 5})]
>>> list(G.edges(data="weight", default=1, keys=True))
[(0, 1, 0, 1), (1, 2, 0, 1), (1, 2, 1, 1), (2, 3, 0, 5)]
>>> list(G.edges([0, 2]))
[(0, 1), (2, 3)]
>>> list(G.edges(0))
[(0, 1)]
[(1, 2), (1, 2)]
MultiDiGraph.out_edges
property MultiDiGraph.out_edges
An OutMultiEdgeView of the Graph as G.edges or G.edges().
The OutMultiEdgeView provides set-like operations on the edge-tuples as well as edge attribute lookup. When
provide set-like operations). Hence, G.edges[u, v, k]['color'] provides the value of the color attribute

for the edge from u to v with key k while for (u, v, k, c) in G.edges(data='color', de-
fault='red', keys=True): iterates through all the edges yielding the color attribute with default 'red'
if no color attribute exists.
will be generated when multiple edges between two nodes exist.
Parameters
nbunch
these nodes.
data
(u, v).
keys
[bool, optional (default=False)] If True, return edge keys with each edge, creating (u, v, k, d)
tuples when data is also requested (the default) and (u, v, k) tuples when data is not requested.
default
Returns
edges
[OutMultiEdgeView] A view of edge attributes, usually it iterates over (u, v) (u, v, k) or (u, v, k,
d) tuples of edges, but can also be used for attribute lookup as edges[u, v, k]['foo'].
See also:
in_edges, out_edges
Notes
Examples
>>> nx.add_path(G, [0, 1, 2])
>>> key2 = G.add_edge(1, 2) # second edge between these nodes
[(0, 1), (1, 2), (1, 2), (2, 3)]
>>> list(G.edges(data=True)) # default data is {} (empty dict)
[(0, 1, {}), (1, 2, {}), (1, 2, {}), (2, 3, {'weight': 5})]
>>> list(G.edges(data="weight", default=1))
[(0, 1, 1), (1, 2, 1), (1, 2, 1), (2, 3, 5)]
>>> list(G.edges(keys=True)) # default keys are integers
[(0, 1, 0), (1, 2, 0), (1, 2, 1), (2, 3, 0)]
>>> list(G.edges(data=True, keys=True))
[(0, 1, 0, {}), (1, 2, 0, {}), (1, 2, 1, {}), (2, 3, 0, {'weight': 5})]


>>> list(G.edges(data="weight", default=1, keys=True))
[(0, 1, 0, 1), (1, 2, 0, 1), (1, 2, 1, 1), (2, 3, 0, 5)]
>>> list(G.edges([0, 2]))
[(0, 1), (2, 3)]
[(0, 1)]
[(1, 2), (1, 2)]
MultiDiGraph.in_edges
property MultiDiGraph.in_edges
in_edges(self, nbunch=None, data=False, keys=False, default=None)
Parameters
nbunch
data
(u, v).
keys
[bool, optional (default=False)] If True, return edge keys with each edge, creating 3-tuples (u,
v, k) or with data, 4-tuples (u, v, k, d).
default
Returns
in_edges
[InMultiEdgeView or InMultiEdgeDataView] A view of edge attributes, usually it iterates over
(u, v) or (u, v, k) or (u, v, k, d) tuples of edges, but can also be used for attribute lookup as
edges[u, v, k]['foo'].
See also:
edges

MultiDiGraph.has_edge
MultiDiGraph.has_edge(u, v, key=None)
Returns True if the graph has an edge between nodes u and v.
This is the same as v in G[u] or key in G[u][v] without KeyError exceptions.
Parameters
u, v
[nodes] Nodes can be, for example, strings or numbers.
key
[hashable identifier, optional (default=None)] If specified return True only if the edge with key
is found.
Returns
edge_ind
Examples
Can be called either using two nodes u, v, an edge tuple (u, v), or an edge tuple (u, v, key).

>>> nx.add_path(G, [0, 1, 2, 3])
True
>>> e = (0, 1)
True
>>> G.add_edge(0, 1, key="a")
'a'
>>> G.has_edge(0, 1, key="a") # specify key
True
>>> G.has_edge(1, 0, key="a") # edges aren't directed
True
>>> e = (0, 1, "a")
>>> G.has_edge(*e) # e is a 3-tuple (u, v, 'a')
True
True
>>> 1 in G[0] # though this gives :exc:`KeyError` if 0 not in G
True
>>> 0 in G[1] # other order; also gives :exc:`KeyError` if 0 not in G
True

MultiDiGraph.get_edge_data
MultiDiGraph.get_edge_data(u, v, key=None, default=None)

Returns the attribute dictionary associated with edge (u, v, key).
If a key is not provided, returns a dictionary mapping edge keys to attribute dictionaries for each edge between u
and v.
This is identical to G[u][v][key] except the default is returned instead of an exception is the edge doesn’t exist.
Parameters
u, v
[nodes]
default
[any Python object (default=None)] Value to return if the specific edge (u, v, key) is not found,
OR if there are no edges between u and v and no key is specified.
key
[hashable identifier, optional (default=None)] Return data only for the edge with specified key,
as an attribute dictionary (rather than a dictionary mapping keys to attribute dictionaries).
Returns
edge_dict
[dictionary] The edge attribute dictionary, OR a dictionary mapping edge keys to attribute
dictionaries for each of those edges if no specific key is provided (even if there’s only one edge
between u and v).
Examples

>>> key = G.add_edge(0, 1, key="a", weight=7)
>>> G[0][1]["a"] # key='a'
{'weight': 7}
>>> G.edges[0, 1, "a"] # key='a'
{'weight': 7}
Warning: we protect the graph data structure by making G.edges and G[1][2] read-only dict-like structures.
However, you can assign values to attributes in e.g. G.edges[1, 2, 'a'] or G[1][2]['a'] using an
additional bracket as shown next. You need to specify all edge info to assign to the edge data associated with an
edge.
>>> G[0][1]["a"]["weight"] = 10
>>> G.edges[0, 1, "a"]["weight"] = 10
>>> G[0][1]["a"]["weight"]
10
>>> G.edges[1, 0, "a"]["weight"]
10

>>> nx.add_path(G, [0, 1, 2, 3])
>>> G.edges[0, 1, 0]["weight"] = 5
>>> G.get_edge_data(0, 1)
{0: {'weight': 5}}
>>> e = (0, 1)


{0: {'weight': 5}}
>>> G.get_edge_data(3, 0) # edge not in graph, returns None
>>> G.get_edge_data(3, 0, default=0) # edge not in graph, return default
0
>>> G.get_edge_data(1, 0, 0) # specific key gives back
{'weight': 5}
MultiDiGraph.neighbors
MultiDiGraph.neighbors(n)
Parameters
n
Raises
NetworkXError
See also:
predecessors
Notes
MultiDiGraph.adj
property MultiDiGraph.adj
by neighbor to the edgekey-dict. So G.adj[3][2][0]['color'] = 'blue' sets the color of the edge
(3, 2, 0) to "blue".
items():.

MultiDiGraph.__getitem__
MultiDiGraph.__getitem__(n)
Parameters
n
Returns
adj_dict
Notes
Examples

>>> G[0]
AtlasView({1: {}})
MultiDiGraph.successors
MultiDiGraph.successors(n)
Parameters
n
Raises
NetworkXError
See also:
predecessors

Notes
MultiDiGraph.succ
property MultiDiGraph.succ
by neighbor to the edgekey-dict. So G.adj[3][2][0]['color'] = 'blue' sets the color of the edge
(3, 2, 0) to "blue".
items():.
For directed graphs, G.succ is identical to G.adj.
MultiDiGraph.predecessors
MultiDiGraph.predecessors(n)
Parameters
n
Raises
NetworkXError
See also:
successors
MultiDiGraph.adjacency
MultiDiGraph.adjacency()
Returns
adj_iter

Examples

[(0, {1: {}}), (1, {0: {}, 2: {}}), (2, {1: {}, 3: {}}), (3, {2: {}})]
MultiDiGraph.nbunch_iter
MultiDiGraph.nbunch_iter(nbunch=None)
Parameters
nbunch
Returns
niter
Raises
NetworkXError
See also:
Graph.__iter__
Notes

MultiDiGraph.order() Returns the number of nodes in the graph.

MultiDiGraph.number_of_nodes() Returns the number of nodes in the graph.
MultiDiGraph.__len__() Returns the number of nodes in the graph.
MultiDiGraph.degree A DegreeView for the Graph as G.degree or G.degree().
MultiDiGraph.in_degree A DegreeView for (node, in_degree) or in_degree for sin-
gle node.
MultiDiGraph.out_degree Returns an iterator for (node, out-degree) or out-degree
for single node.
MultiDiGraph.size([weight]) Returns the number of edges or total of all edge weights.
MultiDiGraph.number_of_edges([u, v]) Returns the number of edges between two nodes.
MultiDiGraph.order
MultiDiGraph.order()
Returns
nnodes
See also:
number_of_nodes
identical method
__len__
identical method
Examples

>>> G.order()
3
MultiDiGraph.number_of_nodes
MultiDiGraph.number_of_nodes()
Returns
nnodes
See also:
order
identical method

__len__
identical method
Examples

3
MultiDiGraph.__len__
MultiDiGraph.__len__()
Returns
nnodes
See also:
number_of_nodes
identical method
order
identical method
Examples

>>> len(G)
4
MultiDiGraph.degree
property MultiDiGraph.degree
Parameters
nbunch
weight

Returns
DiMultiDegreeView or int
If multiple nodes are requested (the default), returns a DiMultiDegreeView mapping
nodes to their degree. If a single node is requested, returns the degree of the node as an
integer.
See also:
out_degree, in_degree
Examples
>>> nx.add_path(G, [0, 1, 2, 3])
1
>>> list(G.degree([0, 1, 2]))
[(0, 1), (1, 2), (2, 2)]
>>> G.add_edge(0, 1) # parallel edge
1
>>> list(G.degree([0, 1, 2])) # parallel edges are counted
[(0, 2), (1, 3), (2, 2)]
MultiDiGraph.in_degree
property MultiDiGraph.in_degree
A DegreeView for (node, in_degree) or in_degree for single node.
The node in-degree is the number of edges pointing in to the node. The weighted node degree is the sum of the
Parameters
nbunch
weight
used as a weight. If None, then each edge has weight 1. The degree is the sum of the edge
weights adjacent to the node.
Returns
deg
[int] Degree of the node
nd_iter
See also:
degree, out_degree

Examples
>>> nx.add_path(G, [0, 1, 2, 3])
0
[(0, 0), (1, 1), (2, 1)]
1
>>> list(G.in_degree([0, 1, 2])) # parallel edges counted
[(0, 0), (1, 2), (2, 1)]
MultiDiGraph.out_degree
property MultiDiGraph.out_degree
Returns an iterator for (node, out-degree) or out-degree for single node.
out_degree(self, nbunch=None, weight=None)
The node out-degree is the number of edges pointing out of the node. This function returns the out-degree for a
single node or an iterator for a bunch of nodes or if nothing is passed as argument.
Parameters
nbunch
weight
weights.
Returns
deg
[int] Degree of the node
nd_iter
See also:
degree, in_degree

Examples
>>> nx.add_path(G, [0, 1, 2, 3])
1
[(0, 1), (1, 1), (2, 1)]
1
>>> list(G.out_degree([0, 1, 2])) # counts parallel edges
[(0, 2), (1, 1), (2, 1)]
MultiDiGraph.size
MultiDiGraph.size(weight=None)
Parameters
weight
Returns
size
more general).
See also:
number_of_edges
Examples

>>> G.size()
3

>>> G.size()
2
6.0

MultiDiGraph.number_of_edges
MultiDiGraph.number_of_edges(u=None, v=None)
Parameters
u, v
[nodes, optional (Gefault=all edges)] If u and v are specified, return the number of edges
Returns
nedges
from u to v.
See also:
size
Examples
For undirected multigraphs, this method counts the total number of edges in the graph:
>>> G.add_edges_from([(0, 1), (0, 1), (1, 2)])
[0, 1, 0]
3
2
For directed multigraphs, this method can count the total number of directed edges from u to v:
>>> G.add_edges_from([(0, 1), (0, 1), (1, 0)])
[0, 1, 0]
2
1

MultiDiGraph.copy([as_view]) Returns a copy of the graph.

MultiDiGraph.to_undirected([reciprocal, Returns an undirected representation of the digraph.
as_view])
MultiDiGraph.to_directed([as_view]) Returns a directed representation of the graph.
MultiDiGraph.subgraph(nodes) Returns a SubGraph view of the subgraph induced on
nodes.
MultiDiGraph.edge_subgraph(edges) Returns the subgraph induced by the specified edges.
MultiDiGraph.reverse([copy]) Returns the reverse of the graph.
MultiDiGraph.copy
MultiDiGraph.copy(as_view=False)
containers.
Parameters
as_view
Returns
G
See also:
to_directed
Notes
copy.deepcopy)
shallow copy.

>>> H = G.copy()
>>> H = nx.Graph(G)
>>> H = G.__class__(G)
>>> H = G.__class__()
copy.html.
Examples

>>> H = G.copy()
MultiDiGraph.to_undirected
MultiDiGraph.to_undirected(reciprocal=False, as_view=False)
Parameters
reciprocal
as_view
[bool (optional, default=False)] If True return an undirected view of the original directed graph.
Returns
G
[MultiGraph] An undirected graph with the same name and nodes and with edge (u, v, data)
if either (u, v, data) or (v, u, data) is in the digraph. If both edges exist in digraph and their
edge data is different, only one edge is created with an arbitrary choice of which edge data to
use. You must check and correct for this manually if desired.
See also:
MultiGraph, copy, add_edge, add_edges_from

Notes
and references.
copy.html.
Warning: If you have subclassed MultiDiGraph to use dict-like objects in the data structure, those changes do not
transfer to the MultiGraph created by this method.
Examples

>>> list(H.edges)
[(0, 1), (1, 0)]
>>> list(G2.edges)
[(0, 1)]
MultiDiGraph.to_directed
MultiDiGraph.to_directed(as_view=False)
Returns
G
[MultiDiGraph] A directed graph with the same name, same nodes, and with each edge (u, v,
k, data) replaced by two directed edges (u, v, k, data) and (v, u, k, data).
Notes
and references.
copy.html.
transfer to the MultiDiGraph created by this method.

Examples
0
1
>>> list(H.edges)
[(0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1)]
0
>>> list(H.edges)
[(0, 1, 0)]
MultiDiGraph.subgraph
MultiDiGraph.subgraph(nodes)
Parameters
nodes
Returns
G
Notes
not in set(nodes)])

SG = G.__class__()


else:
Examples

>>> H = G.subgraph([0, 1, 2])
>>> list(H.edges)
[(0, 1), (1, 2)]
MultiDiGraph.edge_subgraph
MultiDiGraph.edge_subgraph(edges)
Parameters
edges
Returns
G
Notes
Examples
>>> H = G.edge_subgraph([(0, 1), (3, 4)])
>>> list(H.nodes)
[0, 1, 3, 4]
>>> list(H.edges)
[(0, 1), (3, 4)]

MultiDiGraph.reverse
MultiDiGraph.reverse(copy=True)
Parameters
copy
Note: NetworkX uses dicts to store the nodes and neighbors in a graph. So the reporting of nodes and edges for the
base graph classes may not necessarily be consistent across versions and platforms; however, the reporting for CPython
is consistent across platforms and versions after 3.6.
2.3 Graph Views
View of Graphs as SubGraph, Reverse, Directed, Undirected.

In some algorithms it is convenient to temporarily morph a graph to exclude some nodes or edges. It should be better to
do that via a view than to remove and then re-add. In other algorithms it is convenient to temporarily morph a graph to
reverse directed edges, or treat a directed graph as undirected, etc. This module provides those graph views.
The resulting views are essentially read-only graphs that report data from the orignal graph object. We provide an attribute
G._graph which points to the underlying graph object.
Note: Since graphviews look like graphs, one can end up with view-of-view-of-view chains. Be careful with chains
because they become very slow with about 15 nested views. For the common simple case of node induced subgraphs
created from the graph class, we short-cut the chain by returning a subgraph of the original graph directly rather than a
subgraph of a subgraph. We are careful not to disrupt any edge filter in the middle subgraph. In general, determining how
to short-cut the chain is tricky and much harder with restricted_views than with induced subgraphs. Often it is easiest to
use .copy() to avoid chains.
generic_graph_view(G[, create_using])
subgraph_view(G[, filter_node, filter_edge]) View of G applying a filter on nodes and edges.

reverse_view(G) View of G with edge directions reversed
2.3.1 generic_graph_view
generic_graph_view(G, create_using=None)

2.3.2 subgraph_view
subgraph_view(G, filter_node=<function no_filter>, filter_edge=<function no_filter>)

View of G applying a filter on nodes and edges.
subgraph_view provides a read-only view of the input graph that excludes nodes and edges based on the
outcome of two filter functions filter_node and filter_edge.
The filter_node function takes one argument — the node — and returns True if the node should be included
in the subgraph, and False if it should not be included.
The filter_edge function takes two (or three arguments if G is a multi-graph) — the nodes describing an edge,
plus the edge-key if parallel edges are possible — and returns True if the edge should be included in the subgraph,
and False if it should not be included.
Both node and edge filter functions are called on graph elements as they are queried, meaning there is no up-front
cost to creating the view.
Parameters
G
[networkx.Graph] A directed/undirected graph/multigraph
filter_node
[callable, optional] A function taking a node as input, which returns True if the node should
appear in the view.
filter_edge
[callable, optional] A function taking as input the two nodes describing an edge (plus the edge-
key if G is a multi-graph), which returns True if the edge should appear in the view.
Returns
graph
[networkx.Graph] A read-only graph view of the input graph.
Examples
Filter functions operate on the node, and return True if the node should appear in the view:
>>> def filter_node(n1):
... return n1 != 5
...
>>> view = nx.subgraph_view(G, filter_node=filter_node)
>>> view.nodes()
NodeView((0, 1, 2, 3, 4))
We can use a closure pattern to filter graph elements based on additional data — for example, filtering on edge data
attached to the graph:
>>> G[3][4]["cross_me"] = False
>>> def filter_edge(n1, n2):
... return G[n1][n2].get("cross_me", True)
...
>>> view = nx.subgraph_view(G, filter_edge=filter_edge)
>>> view.edges()
EdgeView([(0, 1), (1, 2), (2, 3), (4, 5)])
2.3. Graph Views 151

>>> view = nx.subgraph_view(G, filter_node=filter_node, filter_edge=filter_edge,)

>>> view.nodes()
NodeView((0, 1, 2, 3, 4))
>>> view.edges()
EdgeView([(0, 1), (1, 2), (2, 3)])
2.3.3 reverse_view
reverse_view(G)
View of G with edge directions reversed
reverse_view returns a read-only view of the input graph where edge directions are reversed.
Identical to digraph.reverse(copy=False)
Parameters
G
[networkx.DiGraph]
Returns
graph
[networkx.DiGraph]
Examples
>>> G.edges()
OutEdgeView([(1, 2), (2, 3)])
>>> view = nx.reverse_view(G)

>>> view.edges()
OutEdgeView([(2, 1), (3, 2)])
2.4 Core Views
Views of core data structures such as nested Mappings (e.g. dict-of-dicts). These Views often restrict element access,
with either the entire view or layers of nested mappings being read-only.

AtlasView(d) An AtlasView is a Read-only Mapping of Mappings.

AdjacencyView(d) An AdjacencyView is a Read-only Map of Maps of Maps.
MultiAdjacencyView(d) An MultiAdjacencyView is a Read-only Map of Maps of
Maps of Maps.
UnionAtlas(succ, pred) A read-only union of two atlases (dict-of-dict).
UnionAdjacency(succ, pred) A read-only union of dict Adjacencies as a Map of Maps
of Maps.
UnionMultiInner(succ, pred) A read-only union of two inner dicts of MultiAdjacencies.
UnionMultiAdjacency(succ, pred) A read-only union of two dict MultiAdjacencies.
FilterAtlas(d, NODE_OK)
FilterAdjacency(d, NODE_OK, EDGE_OK)
FilterMultiInner(d, NODE_OK, EDGE_OK)
FilterMultiAdjacency(d, NODE_OK,
EDGE_OK)
2.4.1 networkx.classes.coreviews.AtlasView
class AtlasView(d)
An AtlasView is a Read-only Mapping of Mappings.
It is a View into a dict-of-dict data structure. The inner level of dict is read-write. But the outer level is read-only.
See also:
AdjacencyView
View into dict-of-dict-of-dict
MultiAdjacencyView
View into dict-of-dict-of-dict-of-dict
__init__(d)
Methods
copy()
get(k[,d])
items()
keys()
values()
2.4. Core Views 153

AtlasView.copy
AtlasView.copy()
AtlasView.get
AtlasView.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
AtlasView.items
AtlasView.items() → a set-like object providing a view on D's items
AtlasView.keys
AtlasView.keys() → a set-like object providing a view on D's keys
AtlasView.values
AtlasView.values() → an object providing a view on D's values
2.4.2 networkx.classes.coreviews.AdjacencyView
class AdjacencyView(d)
An AdjacencyView is a Read-only Map of Maps of Maps.
It is a View into a dict-of-dict-of-dict data structure. The inner level of dict is read-write. But the outer levels are
read-only.
See also:
AtlasView
View into dict-of-dict
MultiAdjacencyView
__init__(d)

Methods
copy()
get(k[,d])
items()
keys()
values()
AdjacencyView.copy
AdjacencyView.copy()
AdjacencyView.get
AdjacencyView.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
AdjacencyView.items
AdjacencyView.items() → a set-like object providing a view on D's items
AdjacencyView.keys
AdjacencyView.keys() → a set-like object providing a view on D's keys
AdjacencyView.values
AdjacencyView.values() → an object providing a view on D's values
2.4.3 networkx.classes.coreviews.MultiAdjacencyView
class MultiAdjacencyView(d)
An MultiAdjacencyView is a Read-only Map of Maps of Maps of Maps.
It is a View into a dict-of-dict-of-dict-of-dict data structure. The inner level of dict is read-write. But the outer
levels are read-only.
See also:
AtlasView
AdjacencyView
2.4. Core Views 155

__init__(d)
Methods
copy()
get(k[,d])
items()
keys()
values()
MultiAdjacencyView.copy
MultiAdjacencyView.copy()
MultiAdjacencyView.get
MultiAdjacencyView.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
MultiAdjacencyView.items
MultiAdjacencyView.items() → a set-like object providing a view on D's items
MultiAdjacencyView.keys
MultiAdjacencyView.keys() → a set-like object providing a view on D's keys
MultiAdjacencyView.values
MultiAdjacencyView.values() → an object providing a view on D's values
2.4.4 networkx.classes.coreviews.UnionAtlas
class UnionAtlas(succ, pred)

A read-only union of two atlases (dict-of-dict).
The two dict-of-dicts represent the inner dict of an Adjacency: G.succ[node] and G.pred[node]. The
inner level of dict of both hold attribute key:value pairs and is read-write. But the outer level is read-only.
See also:

UnionAdjacency
UnionMultiAdjacency
__init__(succ, pred)
Methods
copy()
get(k[,d])
items()
keys()
values()
UnionAtlas.copy
UnionAtlas.copy()
UnionAtlas.get
UnionAtlas.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
UnionAtlas.items
UnionAtlas.items() → a set-like object providing a view on D's items
UnionAtlas.keys
UnionAtlas.keys() → a set-like object providing a view on D's keys
UnionAtlas.values
UnionAtlas.values() → an object providing a view on D's values
2.4. Core Views 157

2.4.5 networkx.classes.coreviews.UnionAdjacency
class UnionAdjacency(succ, pred)

A read-only union of dict Adjacencies as a Map of Maps of Maps.
The two input dict-of-dict-of-dicts represent the union of G.succ and G.pred. Return values are UnionAtlas
The inner level of dict is read-write. But the middle and outer levels are read-only.
succ : a dict-of-dict-of-dict {node: nbrdict} pred : a dict-of-dict-of-dict {node: nbrdict} The keys for the two dicts
should be the same
See also:
UnionAtlas
UnionMultiAdjacency
Methods
copy()
get(k[,d])
items()
keys()
values()
UnionAdjacency.copy
UnionAdjacency.copy()
UnionAdjacency.get
UnionAdjacency.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.

UnionAdjacency.items
UnionAdjacency.items() → a set-like object providing a view on D's items
UnionAdjacency.keys
UnionAdjacency.keys() → a set-like object providing a view on D's keys
UnionAdjacency.values
UnionAdjacency.values() → an object providing a view on D's values
2.4.6 networkx.classes.coreviews.UnionMultiInner
class UnionMultiInner(succ, pred)

A read-only union of two inner dicts of MultiAdjacencies.
The two input dict-of-dict-of-dicts represent the union of G.succ[node] and G.pred[node] for MultiDi-
Graphs. Return values are UnionAtlas. The inner level of dict is read-write. But the outer levels are read-only.
See also:
UnionAtlas
UnionAdjacency
UnionMultiAdjacency
Methods
copy()
get(k[,d])
items()
keys()
values()
2.4. Core Views 159

UnionMultiInner.copy
UnionMultiInner.copy()
UnionMultiInner.get
UnionMultiInner.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
UnionMultiInner.items
UnionMultiInner.items() → a set-like object providing a view on D's items
UnionMultiInner.keys
UnionMultiInner.keys() → a set-like object providing a view on D's keys
UnionMultiInner.values
UnionMultiInner.values() → an object providing a view on D's values
2.4.7 networkx.classes.coreviews.UnionMultiAdjacency
class UnionMultiAdjacency(succ, pred)

A read-only union of two dict MultiAdjacencies.
The two input dict-of-dict-of-dict-of-dicts represent the union of G.succ and G.pred for MultiDiGraphs. Re-
turn values are UnionAdjacency. The inner level of dict is read-write. But the outer levels are read-only.
See also:
UnionAtlas
UnionMultiInner

Methods
copy()
get(k[,d])
items()
keys()
values()
UnionMultiAdjacency.copy
UnionMultiAdjacency.copy()
UnionMultiAdjacency.get
UnionMultiAdjacency.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
UnionMultiAdjacency.items
UnionMultiAdjacency.items() → a set-like object providing a view on D's items
UnionMultiAdjacency.keys
UnionMultiAdjacency.keys() → a set-like object providing a view on D's keys
UnionMultiAdjacency.values
UnionMultiAdjacency.values() → an object providing a view on D's values
2.4.8 networkx.classes.coreviews.FilterAtlas
class FilterAtlas(d, NODE_OK)
__init__(d, NODE_OK)
2.4. Core Views 161

Methods
get(k[,d])
items()
keys()
values()
FilterAtlas.get
FilterAtlas.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
FilterAtlas.items
FilterAtlas.items() → a set-like object providing a view on D's items
FilterAtlas.keys
FilterAtlas.keys() → a set-like object providing a view on D's keys
FilterAtlas.values
FilterAtlas.values() → an object providing a view on D's values
2.4.9 networkx.classes.coreviews.FilterAdjacency
class FilterAdjacency(d, NODE_OK, EDGE_OK)
__init__(d, NODE_OK, EDGE_OK)
Methods
get(k[,d])
items()
keys()
values()

FilterAdjacency.get
FilterAdjacency.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
FilterAdjacency.items
FilterAdjacency.items() → a set-like object providing a view on D's items
FilterAdjacency.keys
FilterAdjacency.keys() → a set-like object providing a view on D's keys
FilterAdjacency.values
FilterAdjacency.values() → an object providing a view on D's values
2.4.10 networkx.classes.coreviews.FilterMultiInner
class FilterMultiInner(d, NODE_OK, EDGE_OK)
Methods
get(k[,d])
items()
keys()
values()
FilterMultiInner.get
FilterMultiInner.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
2.4. Core Views 163

FilterMultiInner.items
FilterMultiInner.items() → a set-like object providing a view on D's items
FilterMultiInner.keys
FilterMultiInner.keys() → a set-like object providing a view on D's keys
FilterMultiInner.values
FilterMultiInner.values() → an object providing a view on D's values
2.4.11 networkx.classes.coreviews.FilterMultiAdjacency
class FilterMultiAdjacency(d, NODE_OK, EDGE_OK)
Methods
get(k[,d])
items()
keys()
values()
FilterMultiAdjacency.get
FilterMultiAdjacency.get(k[, d ]) → D[k] if k in D, else d. d defaults to None.
FilterMultiAdjacency.items
FilterMultiAdjacency.items() → a set-like object providing a view on D's items

FilterMultiAdjacency.keys
FilterMultiAdjacency.keys() → a set-like object providing a view on D's keys
FilterMultiAdjacency.values
FilterMultiAdjacency.values() → an object providing a view on D's values
2.5 Filters
Note: Filters can be used with views to restrict the view (or expand it). They can filter nodes or filter edges. These
examples are intended to help you build new ones. They may instead contain all the filters you ever need.
Filter factories to hide or show sets of nodes and edges.

These filters return the function used when creating SubGraph.
no_filter(*items)
hide_nodes(nodes)
hide_edges(edges)
hide_diedges(edges)
hide_multidiedges(edges)
hide_multiedges(edges)
show_nodes(nodes)
show_edges(edges)
show_diedges(edges)
show_multidiedges(edges)
show_multiedges(edges)
2.5. Filters 165

2.5.1 no_filter
no_filter(*items)
2.5.2 hide_nodes
hide_nodes(nodes)
2.5.3 hide_edges
hide_edges(edges)
2.5.4 hide_diedges
hide_diedges(edges)
2.5.5 hide_multidiedges
hide_multidiedges(edges)
2.5.6 hide_multiedges
hide_multiedges(edges)
2.5.7 networkx.classes.filters.show_nodes
class show_nodes(nodes)
__init__(nodes)
Methods
2.5.8 show_edges
show_edges(edges)

2.5.9 show_diedges
show_diedges(edges)
2.5.10 show_multidiedges
show_multidiedges(edges)
2.5.11 show_multiedges
show_multiedges(edges)
2.6 Backends
Note: This is an experimental feature to dispatch your computations to an alternate backend like GraphBLAS, instead
of using pure Python dictionaries for computation. Things will change and break in the future!
Code to support various backends in a plugin dispatch architecture.
2.6.1 Create a Dispatcher
To be a valid plugin, a package must register an entry_point of networkx.plugins with a key pointing to the handler.
For example, ` entry_points={'networkx.plugins': 'sparse = networkx_plugin_sparse'}
`
The plugin must create a Graph-like object which contains an attribute __networkx_plugin__ with a value of the
entry point name.
Continuing the example above: ``` class WrappedSparse:
__networkx_plugin__ = “sparse” …
```
When a dispatchable NetworkX algorithm encounters a Graph-like object with a __networkx_plugin__ attribute,
it will look for the associated dispatch object in the entry_points, load it, and dispatch the work to it.
2.6.2 Testing
To assist in validating the backend algorithm implementations, if an environment variable NET-

WORKX_GRAPH_CONVERT is set to a registered plugin keys, the dispatch machinery will automatically convert
regular networkx Graphs and DiGraphs to the backend equivalent by calling <backend dispatcher>.
convert_from_nx(G, weight=weight, name=name).
The converted object is then passed to the backend implementation of the algorithm. The result is then passed to
<backend dispatcher>.convert_to_nx(result, name=name) to convert back to a form expected
by the NetworkX tests.
2.6. Backends 167

By defining convert_from_nx and convert_to_nx methods and setting the environment variable, NetworkX
will automatically route tests on dispatchable algorithms to the backend, allowing the full networkx test suite to be run
against the backend implementation.
Example pytest invocation: NETWORKX_GRAPH_CONVERT=sparse pytest –pyargs networkx
Dispatchable algorithms which are not implemented by the backend will cause a pytest.xfail(), giving some
indication that not all tests are working, while avoiding causing an explicit failure.
A special on_start_tests(items) function may be defined by the backend. It will be called with the list of
NetworkX tests discovered. Each item is a pytest.Node object. If the backend does not support the test, that test can be
marked as xfail.
_dispatch([func, name]) Dispatches to a backend algorithm when the first argu-

ment is a backend graph-like object.
2.6.3 _dispatch
_dispatch(func=None, *, name=None)
Dispatches to a backend algorithm when the first argument is a backend graph-like object.

CHAPTER
THREE
ALGORITHMS
3.1 Approximations and Heuristics
Approximations of graph properties and Heuristic methods for optimization.
Warning: These functions are not imported in the top-level of networkx
These functions can be accessed using networkx.approximation.function_name

They can be imported using from networkx.algorithms import approximation or from networkx.
algorithms.approximation import function_name
3.1.1 Connectivity
Fast approximation for node connectivity
all_pairs_node_connectivity(G[, nbunch, Compute node connectivity between all pairs of nodes.

cutoff])
local_node_connectivity(G, source, target[, Compute node connectivity between source and target.
...])
node_connectivity(G[, s, t]) Returns an approximation for node connectivity for a
graph or digraph G.
all_pairs_node_connectivity
all_pairs_node_connectivity(G, nbunch=None, cutoff=None)

Compute node connectivity between all pairs of nodes.
Pairwise or local node connectivity between two distinct and nonadjacent nodes is the minimum number of nodes
that must be removed (minimum separating cutset) to disconnect them. By Menger’s theorem, this is equal to the
number of node independent paths (paths that share no nodes other than source and target). Which is what we
compute in this function.
This algorithm is a fast approximation that gives an strict lower bound on the actual number of node independent
paths between two nodes [1]. It works for both directed and undirected graphs.
Parameters
G
[NetworkX graph]
169
nbunch: container
Container of nodes. If provided node connectivity will be computed only over pairs of nodes
in nbunch.
cutoff
[integer] Maximum node connectivity to consider. If None, the minimum degree of source or
target is used as a cutoff in each pair of nodes. Default value None.
Returns
K
[dictionary] Dictionary, keyed by source and target, of pairwise node connectivity
See also:
local_node_connectivity
node_connectivity
References
[1]
Examples
A 3 node cycle with one extra node attached has connectivity 2 between all nodes in the cycle and connectivity 1
between the extra node and the rest:
>>> G = nx.cycle_graph(3)
>>> import pprint # for nice dictionary formatting
>>> pprint.pprint(nx.all_pairs_node_connectivity(G))
{0: {1: 2, 2: 2, 3: 1},
1: {0: 2, 2: 2, 3: 1},
2: {0: 2, 1: 2, 3: 1},
3: {0: 1, 1: 1, 2: 1}}
local_node_connectivity(G, source, target, cutoff=None)

Compute node connectivity between source and target.
Pairwise or local node connectivity between two distinct and nonadjacent nodes is the minimum number of nodes
that must be removed (minimum separating cutset) to disconnect them. By Menger’s theorem, this is equal to the
number of node independent paths (paths that share no nodes other than source and target). Which is what we
compute in this function.
This algorithm is a fast approximation that gives an strict lower bound on the actual number of node independent
paths between two nodes [1]. It works for both directed and undirected graphs.
Parameters
G
[NetworkX graph]
source
[node] Starting node for node connectivity
170 Chapter 3. Algorithms

target
[node] Ending node for node connectivity
cutoff
[integer] Maximum node connectivity to consider. If None, the minimum degree of source or
target is used as a cutoff. Default value None.
Returns
k: integer
pairwise node connectivity
See also:
node_connectivity
Notes
This algorithm [1] finds node independents paths between two nodes by computing their shortest path using BFS,
marking the nodes of the path found as ‘used’ and then searching other shortest paths excluding the nodes marked
as used until no more paths exist. It is not exact because a shortest path could use nodes that, if the path were
longer, may belong to two different node independent paths. Thus it only guarantees an strict lower bound on node
connectivity.
Note that the authors propose a further refinement, losing accuracy and gaining speed, which is not implemented
yet.
References
[1]
Examples
>>> # Platonic octahedral graph has node connectivity 4

>>> # for each non adjacent node pair
>>> from networkx.algorithms import approximation as approx
>>> G = nx.octahedral_graph()
>>> approx.local_node_connectivity(G, 0, 5)
4
node_connectivity
node_connectivity(G, s=None, t=None)

Returns an approximation for node connectivity for a graph or digraph G.
Node connectivity is equal to the minimum number of nodes that must be removed to disconnect G or render it
trivial. By Menger’s theorem, this is equal to the number of node independent paths (paths that share no nodes
other than source and target).
If source and target nodes are provided, this function returns the local node connectivity: the minimum number of
nodes that must be removed to break all paths from source to target in G.
3.1. Approximations and Heuristics 171

This algorithm is based on a fast approximation that gives an strict lower bound on the actual number of node
independent paths between two nodes [1]. It works for both directed and undirected graphs.
Parameters
G
[NetworkX graph] Undirected graph
s
[node] Source node. Optional. Default value: None.
t
[node] Target node. Optional. Default value: None.
Returns
K
[integer] Node connectivity of G, or local node connectivity if source and target are provided.
See also:
Notes
This algorithm [1] finds node independents paths between two nodes by computing their shortest path using BFS,
marking the nodes of the path found as ‘used’ and then searching other shortest paths excluding the nodes marked
as used until no more paths exist. It is not exact because a shortest path could use nodes that, if the path were
longer, may belong to two different node independent paths. Thus it only guarantees an strict lower bound on node
connectivity.
References
[1]
Examples
>>> # Platonic octahedral graph is 4-node-connected

>>> G = nx.octahedral_graph()
>>> approx.node_connectivity(G)
4
3.1.2 K-components
Fast approximation for k-component structure
k_components(G[, min_density]) Returns the approximate k-component structure of a

graph G.

k_components
k_components(G, min_density=0.95)
Returns the approximate k-component structure of a graph G.
A k-component is a maximal subgraph of a graph G that has, at least, node connectivity k: we need to remove
at least k nodes to break it into more components. k-components have an inherent hierarchical structure because
they are nested in terms of connectivity: a connected graph can contain several 2-components, each of which can
contain one or more 3-components, and so forth.
This implementation is based on the fast heuristics to approximate the k-component structure of a graph [1].
Which, in turn, it is based on a fast approximation algorithm for finding good lower bounds of the number of node
independent paths between two nodes [2].
Parameters
G
min_density
[Float] Density relaxation threshold. Default value 0.95
Returns
k_components
[dict] Dictionary with connectivity level k as key and a list of sets of nodes that form a k-
component of level k as values.
Raises
NetworkXNotImplemented
If G is directed.
See also:
k_components
Notes
The logic of the approximation algorithm for computing the k-component structure [1] is based on repeatedly
applying simple and fast algorithms for k-cores and biconnected components in order to narrow down the number
of pairs of nodes over which we have to compute White and Newman’s approximation algorithm for finding node
independent paths [2]. More formally, this algorithm is based on Whitney’s theorem, which states an inclusion
relation among node connectivity, edge connectivity, and minimum degree for any graph G. This theorem implies
that every k-component is nested inside a k-edge-component, which in turn, is contained in a k-core. Thus, this
algorithm computes node independent paths among pairs of nodes in each biconnected part of each k-core, and
repeats this procedure for each k from 3 to the maximal core number of a node in the input graph.
Because, in practice, many nodes of the core of level k inside a bicomponent actually are part of a component of
level k, the auxiliary graph needed for the algorithm is likely to be very dense. Thus, we use a complement graph
data structure (see AntiGraph) to save memory. AntiGraph only stores information of the edges that are not
present in the actual auxiliary graph. When applying algorithms to this complement graph data structure, it behaves
as if it were the dense version.

References
[1], [2], [3]
Examples
>>> # Petersen graph has 10 nodes and it is triconnected, thus all

>>> # nodes are in a single component on all three connectivity levels
>>> from networkx.algorithms import approximation as apxa
>>> G = nx.petersen_graph()
>>> k_components = apxa.k_components(G)
3.1.3 Clique
Functions for computing large cliques and maximum independent sets.
maximum_independent_set(G) Returns an approximate maximum independent set.

max_clique(G) Find the Maximum Clique
clique_removal(G) Repeatedly remove cliques from the graph.
large_clique_size(G) Find the size of a large clique in a graph.
maximum_independent_set
maximum_independent_set(G)
Returns an approximate maximum independent set.
Independent set or stable set is a set of vertices in a graph, no two of which are adjacent. That is, it is a set I of
vertices such that for every two vertices in I, there is no edge connecting the two. Equivalently, each edge in the
graph has at most one endpoint in I. The size of an independent set is the number of vertices it contains [1].
A maximum independent set is a largest independent set for a given graph G and its size is denoted α(G). The
problem of finding such a set is called the maximum independent set problem and is an NP-hard optimization
problem. As such, it is unlikely that there exists an efficient algorithm for finding a maximum independent set of a
graph.
The Independent Set algorithm is based on [2].
Parameters
G
Returns
iset
[Set] The apx-maximum independent set
Raises
If the graph is directed or is a multigraph.

Notes
Finds the O(|V |/(log|V |)2 ) apx of independent set in the worst case.
References
[1], [2]
max_clique
max_clique(G)
Find the Maximum Clique
Finds the O(|V |/(log|V |)2 ) apx of maximum clique/independent set in the worst case.
Parameters
G
Returns
clique
[set] The apx-maximum clique of the graph
Raises
Notes
A clique in an undirected graph G = (V, E) is a subset of the vertex set C subseteq V such that for every two
vertices in C there exists an edge connecting the two. This is equivalent to saying that the subgraph induced by C
is complete (in some cases, the term clique may also refer to the subgraph).
A maximum clique is a clique of the largest possible size in a given graph. The clique number omega(G) of a
graph G is the number of vertices in a maximum clique in G. The intersection number of G is the smallest number
of cliques that together cover all edges of G.
https://en.wikipedia.org/wiki/Maximum_clique
References
[1]

clique_removal
clique_removal(G)
Repeatedly remove cliques from the graph.
Results in a O(|V |/(log |V |)2 ) approximation of maximum clique and independent set. Returns the largest inde-
pendent set found, along with found maximal cliques.
Parameters
G
Returns
max_ind_cliques
[(set, list) tuple] 2-tuple of Maximal Independent Set and list of maximal cliques (sets).
Raises
References
[1]
large_clique_size
large_clique_size(G)
Find the size of a large clique in a graph.
A clique is a subset of nodes in which each pair of nodes is adjacent. This function is a heuristic for finding the size
of a large clique in the graph.
Parameters
G
[NetworkX graph]
Returns
k: integer
The size of a large clique in the graph.
Raises
See also:
networkx.algorithms.approximation.clique.max_clique()
A function that returns an approximate maximum clique with a guarantee on the approximation ratio.
networkx.algorithms.clique
Functions for finding the exact maximum clique in a graph.

Notes
This implementation is from [1]. Its worst case time complexity is O(nd2 ), where n is the number of nodes in the
graph and d is the maximum degree.
This function is a heuristic, which means it may work well in practice, but there is no rigorous mathematical
guarantee on the ratio between the returned number and the actual largest clique size in the graph.
References
[1]
3.1.4 Clustering
average_clustering(G[, trials, seed]) Estimates the average clustering coefficient of G.
average_clustering
average_clustering(G, trials=1000, seed=None)

Estimates the average clustering coefficient of G.
The local clustering of each node in G is the fraction of triangles that actually exist over all possible triangles in its
neighborhood. The average clustering coefficient of a graph G is the mean of local clusterings.
This function finds an approximate average clustering coefficient for G by repeating n times (defined in trials)
the following experiment: choose a node at random, choose two of its neighbors at random, and check if they are
connected. The approximate coefficient is the fraction of triangles found over the number of trials [1].
Parameters
G
[NetworkX graph]
trials
[integer] Number of trials to perform (default 1000).
seed
[integer, random_state, or None (default)] Indicator of random number generation state. See
Randomness.
Returns
c
[float] Approximated average clustering coefficient.

References
[1]
Examples
>>> from networkx.algorithms import approximation

>>> G = nx.erdos_renyi_graph(10, 0.2, seed=10)
>>> approximation.average_clustering(G, trials=1000, seed=10)
0.214
3.1.5 Distance Measures
Distance measures approximated metrics.
diameter(G[, seed]) Returns a lower bound on the diameter of the graph G.
diameter
diameter(G, seed=None)
Returns a lower bound on the diameter of the graph G.
The function computes a lower bound on the diameter (i.e., the maximum eccentricity) of a directed or undirected
graph G. The procedure used varies depending on the graph being directed or not.
If G is an undirected graph, then the function uses the 2-sweep algorithm [1]. The main idea is to pick the
farthest node from a random node and return its eccentricity.
Otherwise, if G is a directed graph, the function uses the 2-dSweep algorithm [2], The procedure starts by
selecting a random source node s from which it performs a forward and a backward BFS. Let a1 and a2 be the
farthest nodes in the forward and backward cases, respectively. Then, it computes the backward eccentricity of a1
using a backward BFS and the forward eccentricity of a2 using a forward BFS. Finally, it returns the best lower
bound between the two.
In both cases, the time complexity is linear with respect to the size of G.
Parameters
G
[NetworkX graph]
seed
Randomness.
Returns
d
[integer] Lower Bound on the Diameter of G
Raises
NetworkXError
If the graph is empty or If the graph is undirected and not connected or If the graph is directed
and not strongly connected.

See also:
networkx.algorithms.distance_measures.diameter
References
[1], [2]
3.1.6 Dominating Set
Functions for finding node and edge dominating sets.

A dominating set for an undirected graph G with vertex set V and edge set E is a subset D of V such that every vertex
not in D is adjacent to at least one member of D. An edge dominating set is a subset F of E such that every edge not in F
is incident to an endpoint of at least one edge in F.
min_weighted_dominating_set(G[, weight]) Returns a dominating set that approximates the minimum

weight node dominating set.
min_edge_dominating_set(G) Returns minimum cardinality edge dominating set.
min_weighted_dominating_set
min_weighted_dominating_set(G, weight=None)
Returns a dominating set that approximates the minimum weight node dominating set.
Parameters
G
[NetworkX graph] Undirected graph.
weight
[string] The node attribute storing the weight of an node. If provided, the node attribute with
this key must be a number for each node. If not provided, each node is assumed to have weight
one.
Returns
min_weight_dominating_set
[set] A set of nodes, the sum of whose weights is no more than (log w(V)) w(V^*),
where w(V) denotes the sum of the weights of each node in the graph and w(V^*) denotes
the sum of the weights of each node in the minimum weight dominating set.
Notes
This algorithm computes an approximate minimum weighted dominating set for the graph G. The returned solution
has weight (log w(V)) w(V^*), where w(V) denotes the sum of the weights of each node in the graph and
w(V^*) denotes the sum of the weights of each node in the minimum weight dominating set for the graph.
This implementation of the algorithm runs in O(m) time, where m is the number of edges in the graph.

References
[1]
min_edge_dominating_set
min_edge_dominating_set(G)
Returns minimum cardinality edge dominating set.
Parameters
G
Returns
min_edge_dominating_set
[set] Returns a set of dominating edges whose size is no more than 2 * OPT.
Notes
The algorithm computes an approximate solution to the edge dominating set problem. The result is no more than
2 * OPT in terms of size of the set. Runtime of the algorithm is O(|E|).
3.1.7 Matching
Given a graph G = (V,E), a matching M in G is a set of pairwise non-adjacent edges; that is, no two edges share a common
vertex.
Wikipedia: Matching
min_maximal_matching(G) Returns the minimum maximal matching of G.
min_maximal_matching
min_maximal_matching(G)
Returns the minimum maximal matching of G. That is, out of all maximal matchings of the graph G, the smallest
is returned.
Parameters
G
Returns
min_maximal_matching
[set] Returns a set of edges such that no two edges share a common endpoint and every edge
not in the set shares some common endpoint in the set. Cardinality will be 2*OPT in the worst
case.

Notes
The algorithm computes an approximate solution fo the minimum maximal cardinality matching problem. The
solution is no more than 2 * OPT in size. Runtime is O(|E|).
References
[1]
3.1.8 Ramsey
Ramsey numbers.
ramsey_R2(G) Compute the largest clique and largest independent set in

G.
ramsey_R2
ramsey_R2(G)
Compute the largest clique and largest independent set in G.
This can be used to estimate bounds for the 2-color Ramsey number R(2;s,t) for G.
This is a recursive implementation which could run into trouble for large recursions. Note that self-loop edges are
ignored.
Parameters
G
Returns
max_pair
[(set, set) tuple] Maximum clique, Maximum independent set.
Raises
3.1.9 Steiner Tree
metric_closure(G[, weight]) Return the metric closure of a graph.

steiner_tree(G, terminal_nodes[, weight, method]) Return an approximation to the minimum Steiner tree of
a graph.

metric_closure
metric_closure(G, weight='weight')
Return the metric closure of a graph.
The metric closure of a graph G is the complete graph in which each edge is weighted by the shortest path distance
between the nodes in G .
Parameters
G
[NetworkX graph]
Returns
NetworkX graph
Metric closure of the graph G.
steiner_tree
steiner_tree(G, terminal_nodes, weight='weight', method=None)

Return an approximation to the minimum Steiner tree of a graph.
The minimum Steiner tree of G w.r.t a set of terminal_nodes (also S) is a tree within G that spans those nodes
and has minimum size (sum of edge weights) among all such trees.
The approximation algorithm is specified with the method keyword argument. All three available algorithms
produce a tree whose weight is within a (2 - (2 / l)) factor of the weight of the optimal Steiner tree, where l is the
minimum number of leaf nodes across all possible Steiner trees.
• kou [2] (runtime O(|S||V |2 )) computes the minimum spanning tree of
the subgraph of the metric closure of G induced by the terminal nodes, where the metric closure of G is the
complete graph in which each edge is weighted by the shortest path distance between the nodes in G. * mehlhorn
[3] (runtime O(|E| + |V | log |V |)) modifies Kou et al.’s algorithm, beginning by finding the closest terminal node
for each non-terminal. This data is used to create a complete graph containing only the terminal nodes, in which
edge is weighted with the shortest path distance between them. The algorithm then proceeds in the same way as
Kou et al..
Parameters
G
[NetworkX graph]
terminal_nodes
[list] A list of terminal nodes for which minimum steiner tree is to be found.
weight
[string (default = ‘weight’)] Use the edge attribute specified by this string as the edge weight.
Any edge attribute not present defaults to 1.
method
[string, optional (default = ‘kou’)] The algorithm to use to approximate the Steiner tree. Sup-
ported options: ‘kou’, ‘mehlhorn’. Other inputs produce a ValueError.
Returns
NetworkX graph
Approximation to the minimum steiner tree of G induced by terminal_nodes .

Notes
For multigraphs, the edge between two nodes with minimum weight is the edge put into the Steiner tree.
References
[1], [2], [3]
3.1.10 Traveling Salesman
Travelling Salesman Problem (TSP)
Implementation of approximate algorithms for solving and approximating the TSP problem.
Categories of algorithms which are implemented:
• Christofides (provides a 3/2-approximation of TSP)
• Greedy
• Simulated Annealing (SA)
• Threshold Accepting (TA)
• Asadpour Asymmetric Traveling Salesman Algorithm
The Travelling Salesman Problem tries to find, given the weight (distance) between all points where a salesman has to
visit, the route so that:
• The total distance (cost) which the salesman travels is minimized.
• The salesman returns to the starting point.
• Note that for a complete graph, the salesman visits each point once.
The function travelling_salesman_problem allows for incomplete graphs by finding all-pairs shortest paths,
effectively converting the problem to a complete graph problem. It calls one of the approximate methods on that problem
and then converts the result back to the original graph using the previously found shortest paths.
TSP is an NP-hard problem in combinatorial optimization, important in operations research and theoretical computer
science.
http://en.wikipedia.org/wiki/Travelling_salesman_problem
christofides(G[, weight, tree]) Approximate a solution of the traveling salesman problem

traveling_salesman_problem(G[, weight, ...]) Find the shortest path in G connecting specified nodes
greedy_tsp(G[, weight, source]) Return a low cost cycle starting at source and its cost.
simulated_annealing_tsp(G, init_cycle[, ...]) Returns an approximate solution to the traveling salesman
problem.
threshold_accepting_tsp(G, init_cycle[, ...]) Returns an approximate solution to the traveling salesman
problem.
asadpour_atsp(G[, weight, seed, source]) Returns an approximate solution to the traveling salesman
problem.

christofides
christofides(G, weight='weight', tree=None)

Approximate a solution of the traveling salesman problem
Compute a 3/2-approximation of the traveling salesman problem in a complete undirected graph using Christofides
[1] algorithm.
Parameters
G
[Graph] G should be a complete weighted undirected graph. The distance between all pairs of
nodes should be included.
weight
[string, optional (default=”weight”)] Edge data key corresponding to the edge weight. If any
edge does not have this attribute the weight is set to 1.
tree
[NetworkX graph or None (default: None)] A minimum spanning tree of G. Or, if None, the
minimum spanning tree is computed using networkx.minimum_spanning_tree()
Returns
list
List of nodes in G along a cycle with a 3/2-approximation of the minimal Hamiltonian cycle.
References
[1]
traveling_salesman_problem
traveling_salesman_problem(G, weight='weight', nodes=None, cycle=True, method=None)

Find the shortest path in G connecting specified nodes
This function allows approximate solution to the traveling salesman problem on networks that are not complete
graphs and/or where the salesman does not need to visit all nodes.
This function proceeds in two steps. First, it creates a complete graph using the all-pairs shortest_paths between
nodes in nodes. Edge weights in the new graph are the lengths of the paths between each pair of nodes in
the original graph. Second, an algorithm (default: christofides for undirected and asadpour_atsp for
directed) is used to approximate the minimal Hamiltonian cycle on this new graph. The available algorithms are:
• christofides
• greedy_tsp
• simulated_annealing_tsp
• threshold_accepting_tsp
• asadpour_atsp
Once the Hamiltonian Cycle is found, this function post-processes to accommodate the structure of the original
graph. If cycle is False, the biggest weight edge is removed to make a Hamiltonian path. Then each edge on
the new complete graph used for that analysis is replaced by the shortest_path between those nodes on the original
graph.
Parameters

G
[NetworkX graph] A possibly weighted graph
nodes
[collection of nodes (default=G.nodes)] collection (list, set, etc.) of nodes to visit
weight
cycle
[bool (default: True)] Indicates whether a cycle should be returned, or a path. Note: the cycle
is the approximate minimal cycle. The path simply removes the biggest edge in that cycle.
method
[function (default: None)] A function that returns a cycle on all nodes and approximates the
solution to the traveling salesman problem on a complete graph. The returned cycle is then
used to find a corresponding solution on G. method should be callable; take inputs G, and
weight; and return a list of nodes along the cycle.
Provided options include christofides(), greedy_tsp(), simu-
lated_annealing_tsp() and threshold_accepting_tsp().
If method is None: use christofides() for undirected G and thresh-
old_accepting_tsp() for directed G.
To specify parameters for these provided functions, construct lambda functions that state the
specific value. method must have 2 inputs. (See examples).
Returns
list
List of nodes in G along a path with an approximation of the minimal path through nodes.
Raises
NetworkXError
If G is a directed graph it has to be strongly connected or the complete version cannot be
generated.
Examples
>>> tsp = nx.approximation.traveling_salesman_problem

>>> G[4][5]["weight"] = 5 # all other weights are 1
>>> tsp(G, nodes=[3, 6])
[3, 2, 1, 0, 8, 7, 6, 7, 8, 0, 1, 2, 3]
>>> path = tsp(G, cycle=False)
>>> path in ([4, 3, 2, 1, 0, 8, 7, 6, 5], [5, 6, 7, 8, 0, 1, 2, 3, 4])
True
Build (curry) your own function to provide parameter values to the methods.
>>> SA_tsp = nx.approximation.simulated_annealing_tsp

>>> method = lambda G, wt: SA_tsp(G, "greedy", weight=wt, temp=500)
>>> path = tsp(G, cycle=False, method=method)
>>> path in ([4, 3, 2, 1, 0, 8, 7, 6, 5], [5, 6, 7, 8, 0, 1, 2, 3, 4])
True

greedy_tsp
greedy_tsp(G, weight='weight', source=None)

Return a low cost cycle starting at source and its cost.
This approximates a solution to the traveling salesman problem. It finds a cycle of all the nodes that a salesman can
visit in order to visit many nodes while minimizing total distance. It uses a simple greedy algorithm. In essence,
this function returns a large cycle given a source point for which the total cost of the cycle is minimized.
Parameters
G
[Graph] The Graph should be a complete weighted undirected graph. The distance between
all pairs of nodes should be included.
weight
source
[node, optional (default: first node in list(G))] Starting node. If None, defaults to
next(iter(G))
Returns
cycle
[list of nodes] Returns the cycle (list of nodes) that a salesman can follow to minimize total
weight of the trip.
Raises
NetworkXError
If G is not complete, the algorithm raises an exception.
Notes
This implementation of a greedy algorithm is based on the following:

• The algorithm adds a node to the solution at every iteration.
• The algorithm selects a node not already in the cycle whose connection to the previous node adds the least
cost to the cycle.
A greedy algorithm does not always give the best solution. However, it can construct a first feasible solution which
can be passed as a parameter to an iterative improvement algorithm such as Simulated Annealing, or Threshold
Accepting.
Time complexity: It has a running time O(|V |2 )

Examples

>>> G.add_weighted_edges_from({
... ("A", "B", 3), ("A", "C", 17), ("A", "D", 14), ("B", "A", 3),
... ("B", "C", 12), ("B", "D", 16), ("C", "A", 13),("C", "B", 12),
... ("C", "D", 4), ("D", "A", 14), ("D", "B", 15), ("D", "C", 2)
... })
>>> cycle = approx.greedy_tsp(G, source="D")
>>> cost = sum(G[n][nbr]["weight"] for n, nbr in nx.utils.pairwise(cycle))
>>> cycle
['D', 'C', 'B', 'A', 'D']
>>> cost
31
simulated_annealing_tsp
simulated_annealing_tsp(G, init_cycle, weight='weight', source=None, temp=100, move='1-1',

max_iterations=10, N_inner=100, alpha=0.01, seed=None)
Returns an approximate solution to the traveling salesman problem.
This function uses simulated annealing to approximate the minimal cost cycle through the nodes. Starting from
a suboptimal solution, simulated annealing perturbs that solution, occasionally accepting changes that make the
solution worse to escape from a locally optimal solution. The chance of accepting such changes decreases over the
iterations to encourage an optimal result. In summary, the function returns a cycle starting at source for which
the total cost is minimized. It also returns the cost.
The chance of accepting a proposed change is related to a parameter called the temperature (annealing has a physical
analogue of steel hardening as it cools). As the temperature is reduced, the chance of moves that increase cost goes
down.
Parameters
G
init_cycle
[list of all nodes or “greedy”] The initial solution (a cycle through all nodes returning
to the start). This argument has no default to make you think about it. If “greedy”,
use greedy_tsp(G, weight). Other common starting cycles are list(G) +
[next(iter(G))] or the final result of simulated_annealing_tsp when doing
threshold_accepting_tsp.
weight
source
next(iter(G))
temp
[int, optional (default=100)] The algorithm’s temperature parameter. It represents the initial
value of temperature

move
[“1-1” or “1-0” or function, optional (default=”1-1”)] Indicator of what move to use when
finding new trial solutions. Strings indicate two special built-in moves:
• “1-1”: 1-1 exchange which transposes the position of two elements of the current solution.
The function called is swap_two_nodes(). For example if we apply 1-1 exchange in
the solution A = [3, 2, 1, 4, 3] we can get the following by the transposition of 1
and 4 elements: A' = [3, 2, 4, 1, 3]
• “1-0”: 1-0 exchange which moves an node in the solution to a new position. The function
called is move_one_node(). For example if we apply 1-0 exchange in the solution A =
[3, 2, 1, 4, 3] we can transfer the fourth element to the second position: A' =
[3, 4, 2, 1, 3]
You may provide your own functions to enact a move from one solution to a neighbor solution.
The function must take the solution as input along with a seed input to control random number
generation (see the seed input here). Your function should maintain the solution as a cycle
with equal first and last node and all others appearing once. Your function should return the
new solution.
max_iterations
[int, optional (default=10)] Declared done when this number of consecutive iterations of the
outer loop occurs without any change in the best cost solution.
N_inner
[int, optional (default=100)] The number of iterations of the inner loop.
alpha
[float between (0, 1), optional (default=0.01)] Percentage of temperature decrease in each
iteration of outer loop
seed
Randomness.
Returns
cycle
weight of the trip.
Raises
NetworkXError
If G is not complete the algorithm raises an exception.
Notes
Simulated Annealing is a metaheuristic local search algorithm. The main characteristic of this algorithm is that
it accepts even solutions which lead to the increase of the cost in order to escape from low quality local optimal
solutions.
This algorithm needs an initial solution. If not provided, it is constructed by a simple greedy algorithm. At every
iteration, the algorithm selects thoughtfully a neighbor solution. Consider c(x) cost of current solution and c(x′ ) cost
of a neighbor solution. If c(x′ ) − c(x) <= 0 then the neighbor solution becomes the current solution for the next
iteration. Otherwise, the algorithm accepts the neighbor solution with probability p = exp−([c(x′ )−c(x)]/temp).
Otherwise the current solution is retained.
temp is a parameter of the algorithm and represents temperature.

Time complexity: For Ni iterations of the inner loop and No iterations of the outer loop, this algorithm has running
time O(Ni ∗ No ∗ |V |).
For more information and how the algorithm is inspired see: http://en.wikipedia.org/wiki/Simulated_annealing
Examples

... ("A", "B", 3), ("A", "C", 17), ("A", "D", 14), ("B", "A", 3),
... ("B", "C", 12), ("B", "D", 16), ("C", "A", 13),("C", "B", 12),
... ("C", "D", 4), ("D", "A", 14), ("D", "B", 15), ("D", "C", 2)
... })
>>> cycle = approx.simulated_annealing_tsp(G, "greedy", source="D")
>>> cycle
['D', 'C', 'B', 'A', 'D']
>>> cost
31
>>> incycle = ["D", "B", "A", "C", "D"]
>>> cycle = approx.simulated_annealing_tsp(G, incycle, source="D")
>>> cycle
['D', 'C', 'B', 'A', 'D']
>>> cost
31
threshold_accepting_tsp
threshold_accepting_tsp(G, init_cycle, weight='weight', source=None, threshold=1, move='1-1',

max_iterations=10, N_inner=100, alpha=0.1, seed=None)
This function uses threshold accepting methods to approximate the minimal cost cycle through the nodes. Starting
from a suboptimal solution, threshold accepting methods perturb that solution, accepting any changes that make the
solution no worse than increasing by a threshold amount. Improvements in cost are accepted, but so are changes
leading to small increases in cost. This allows the solution to leave suboptimal local minima in solution space. The
threshold is decreased slowly as iterations proceed helping to ensure an optimum. In summary, the function returns
a cycle starting at source for which the total cost is minimized.
Parameters
G
init_cycle
[list or “greedy”] The initial solution (a cycle through all nodes returning to the
start). This argument has no default to make you think about it. If “greedy”, use
greedy_tsp(G, weight). Other common starting cycles are list(G) +
[next(iter(G))] or the final result of simulated_annealing_tsp when doing
threshold_accepting_tsp.

weight
source
next(iter(G))
threshold
[int, optional (default=1)] The algorithm’s threshold parameter. It represents the initial thresh-
old’s value
move
[“1-1” or “1-0” or function, optional (default=”1-1”)] Indicator of what move to use when
finding new trial solutions. Strings indicate two special built-in moves:
• “1-1”: 1-1 exchange which transposes the position of two elements of the current solution.
The function called is swap_two_nodes(). For example if we apply 1-1 exchange in
the solution A = [3, 2, 1, 4, 3] we can get the following by the transposition of 1
and 4 elements: A' = [3, 2, 4, 1, 3]
• “1-0”: 1-0 exchange which moves an node in the solution to a new position. The function
called is move_one_node(). For example if we apply 1-0 exchange in the solution A =
[3, 2, 1, 4, 3] we can transfer the fourth element to the second position: A' =
[3, 4, 2, 1, 3]
You may provide your own functions to enact a move from one solution to a neighbor solution.
The function must take the solution as input along with a seed input to control random number
generation (see the seed input here). Your function should maintain the solution as a cycle
with equal first and last node and all others appearing once. Your function should return the
new solution.
max_iterations
[int, optional (default=10)] Declared done when this number of consecutive iterations of the
outer loop occurs without any change in the best cost solution.
N_inner
[int, optional (default=100)] The number of iterations of the inner loop.
alpha
[float between (0, 1), optional (default=0.1)] Percentage of threshold decrease when there is at
least one acceptance of a neighbor solution. If no inner loop moves are accepted the threshold
remains unchanged.
seed
Randomness.
Returns
cycle
weight of the trip.
Raises
NetworkXError
If G is not complete the algorithm raises an exception.
See also:

simulated_annealing_tsp
Notes
Threshold Accepting is a metaheuristic local search algorithm. The main characteristic of this algorithm is that
it accepts even solutions which lead to the increase of the cost in order to escape from low quality local optimal
solutions.
This algorithm needs an initial solution. This solution can be constructed by a simple greedy algorithm. At every
iteration, it selects thoughtfully a neighbor solution. Consider c(x) cost of current solution and c(x′ ) cost of neighbor
solution. If c(x′ ) − c(x) <= threshold then the neighbor solution becomes the current solution for the next
iteration, where the threshold is named threshold.
In comparison to the Simulated Annealing algorithm, the Threshold Accepting algorithm does not accept very low
quality solutions (due to the presence of the threshold value). In the case of Simulated Annealing, even a very low
quality solution can be accepted with probability p.
Time complexity: It has a running time O(m ∗ n ∗ |V |) where m and n are the number of times the outer and
inner loop run respectively.
For more information and how algorithm is inspired see: https://doi.org/10.1016/0021-9991(90)90201-B
Examples

... ("A", "B", 3), ("A", "C", 17), ("A", "D", 14), ("B", "A", 3),
... ("B", "C", 12), ("B", "D", 16), ("C", "A", 13),("C", "B", 12),
... ("C", "D", 4), ("D", "A", 14), ("D", "B", 15), ("D", "C", 2)
... })
>>> cycle = approx.threshold_accepting_tsp(G, "greedy", source="D")
>>> cycle
['D', 'C', 'B', 'A', 'D']
>>> cost
31
>>> incycle = ["D", "B", "A", "C", "D"]
>>> cycle = approx.threshold_accepting_tsp(G, incycle, source="D")
>>> cycle
['D', 'C', 'B', 'A', 'D']
>>> cost
31

asadpour_atsp
asadpour_atsp(G, weight='weight', seed=None, source=None)

This approximate solution is one of the best known approximations for the asymmetric traveling salesman problem
developed by Asadpour et al, [1]. The algorithm first solves the Held-Karp relaxation to find a lower bound for
the weight of the cycle. Next, it constructs an exponential distribution of undirected spanning trees where the
probability of an edge being in the tree corresponds to the weight of that edge using a maximum entropy rounding
scheme. Next we sample that distribution 2⌈ln n⌉ times and save the minimum sampled tree once the direction of
the arcs is added back to the edges. Finally, we augment then short circuit that graph to find the approximate tour
for the salesman.
Parameters
G
[nx.DiGraph] The graph should be a complete weighted directed graph. The distance between
all paris of nodes should be included and the triangle inequality should hold. That is, the direct
edge between any two nodes should be the path of least cost.
weight
seed
Randomness.
source
[node label (default=`None`)] If given, return the cycle starting and ending at the given node.
Returns
cycle
[list of nodes] Returns the cycle (list of nodes) that a salesman can follow to minimize the total
weight of the trip.
Raises
NetworkXError
If G is not complete or has less than two nodes, the algorithm raises an exception.
NetworkXError
If source is not None and is not a node in G, the algorithm raises an exception.
If G is an undirected graph.
References
[1]

Examples

>>> import networkx.algorithms.approximation as approx
>>> G = nx.complete_graph(3, create_using=nx.DiGraph)
>>> nx.set_edge_attributes(G, {(0, 1): 2, (1, 2): 2, (2, 0): 2, (0, 2): 1, (2,␣
,→1): 1, (1, 0): 1}, "weight")
>>> tour = approx.asadpour_atsp(G,source=0)

>>> tour
[0, 2, 1, 0]
3.1.11 Treewidth
Functions for computing treewidth decomposition.

Treewidth of an undirected graph is a number associated with the graph. It can be defined as the size of the largest vertex
set (bag) in a tree decomposition of the graph minus one.
Wikipedia: Treewidth
The notions of treewidth and tree decomposition have gained their attractiveness partly because many graph and network
problems that are intractable (e.g., NP-hard) on arbitrary graphs become efficiently solvable (e.g., with a linear time
algorithm) when the treewidth of the input graphs is bounded by a constant [1] [2].
There are two different functions for computing a tree decomposition: treewidth_min_degree() and
treewidth_min_fill_in().
treewidth_min_degree(G) Returns a treewidth decomposition using the Minimum

Degree heuristic.
treewidth_min_fill_in(G) Returns a treewidth decomposition using the Minimum
Fill-in heuristic.
treewidth_min_degree
treewidth_min_degree(G)
Returns a treewidth decomposition using the Minimum Degree heuristic.
The heuristic chooses the nodes according to their degree, i.e., first the node with the lowest degree is chosen, then
the graph is updated and the corresponding node is removed. Next, a new node with the lowest degree is chosen,
and so on.
Parameters
G
[NetworkX graph]
Returns
Treewidth decomposition
[(int, Graph) tuple] 2-tuple with treewidth and the corresponding decomposed tree.

treewidth_min_fill_in
treewidth_min_fill_in(G)
Returns a treewidth decomposition using the Minimum Fill-in heuristic.
The heuristic chooses a node from the graph, where the number of edges added turning the neighbourhood of the
chosen node into clique is as small as possible.
Parameters
G
[NetworkX graph]
Returns
Treewidth decomposition
[(int, Graph) tuple] 2-tuple with treewidth and the corresponding decomposed tree.
3.1.12 Vertex Cover
Functions for computing an approximate minimum weight vertex cover.

A vertex cover is a subset of nodes such that each edge in the graph is incident to at least one node in the subset.
min_weighted_vertex_cover(G[, weight]) Returns an approximate minimum weighted vertex cover.
min_weighted_vertex_cover
min_weighted_vertex_cover(G, weight=None)
Returns an approximate minimum weighted vertex cover.
The set of nodes returned by this function is guaranteed to be a vertex cover, and the total weight of the set is
guaranteed to be at most twice the total weight of the minimum weight vertex cover. In other words,
w(S) ≤ 2 ∗ w(S ∗ ),
where S is the vertex cover returned by this function, S ∗ is the vertex cover of minimum weight out of all vertex
covers of the graph, and w is the function that computes the sum of the weights of each node in that given set.
Parameters
G
[NetworkX graph]
weight
[string, optional (default = None)] If None, every node has weight 1. If a string, use this node
attribute as the node weight. A node without this attribute is assumed to have weight 1.
Returns
min_weighted_cover
[set] Returns a set of nodes whose weight sum is no more than twice the weight sum of the
minimum weight vertex cover.

Notes
For a directed graph, a vertex cover has the same definition: a set of nodes such that each edge in the graph is
incident to at least one node in the set. Whether the node is the head or tail of the directed edge is ignored.
This is the local-ratio algorithm for computing an approximate vertex cover. The algorithm greedily reduces the
costs over edges, iteratively building a cover. The worst-case runtime of this implementation is O(m log n), where
n is the number of nodes and m the number of edges in the graph.
References
[1]
3.1.13 Max Cut
randomized_partitioning(G[, seed, p, weight]) Compute a random partitioning of the graph nodes and its
cut value.
one_exchange(G[, initial_cut, seed, weight]) Compute a partitioning of the graphs nodes and the cor-
responding cut value.
randomized_partitioning
randomized_partitioning(G, seed=None, p=0.5, weight=None)

Compute a random partitioning of the graph nodes and its cut value.
A partitioning is calculated by observing each node and deciding to add it to the partition with probability p,
returning a random cut and its corresponding value (the sum of weights of edges connecting different partitions).
Parameters
G
[NetworkX graph]
seed
Randomness.
p
[scalar] Probability for each node to be part of the first partition. Should be in [0,1]
weight
[object] Edge attribute key to use as weight. If not specified, edges have weight one.
Returns
cut_size
[scalar] Value of the minimum cut.
partition
[pair of node sets] A partitioning of the nodes that defines a minimum cut.

one_exchange
one_exchange(G, initial_cut=None, seed=None, weight=None)

Compute a partitioning of the graphs nodes and the corresponding cut value.
Use a greedy one exchange strategy to find a locally maximal cut and its value, it works by finding the best node (one
that gives the highest gain to the cut value) to add to the current cut and repeats this process until no improvement
can be made.
Parameters
G
[networkx Graph] Graph to find a maximum cut for.
initial_cut
[set] Cut to use as a starting point. If not supplied the algorithm starts with an empty cut.
seed
Randomness.
weight
Returns
cut_value
[scalar] Value of the maximum cut.
partition
[pair of node sets] A partitioning of the nodes that defines a maximum cut.
3.2 Assortativity
3.2.1 Assortativity
degree_assortativity_coefficient(G[, x, Compute degree assortativity of graph.

y, ...])
attribute_assortativity_coefficient(G, Compute assortativity for node attributes.
attribute)
numeric_assortativity_coefficient(G, Compute assortativity for numerical node attributes.
attribute)
degree_pearson_correlation_coefficient(G[,
Compute degree assortativity of graph.
...])

degree_assortativity_coefficient
degree_assortativity_coefficient(G, x='out', y='in', weight=None, nodes=None)

Assortativity measures the similarity of connections in the graph with respect to the node degree.
Parameters
G
[NetworkX graph]
x: string (‘in’,’out’)
The degree type for source node (directed graphs only).
y: string (‘in’,’out’)
The degree type for target node (directed graphs only).
weight: string or None, optional (default=None)
The edge attribute that holds the numerical value used as a weight. If None, then each edge
has weight 1. The degree is the sum of the edge weights adjacent to the node.
nodes: list or iterable (optional)
Compute degree assortativity only for nodes in container. The default is all nodes.
Returns
r
[float] Assortativity of graph by degree.
See also:
attribute_assortativity_coefficient
numeric_assortativity_coefficient
degree_mixing_dict
degree_mixing_matrix
Notes
This computes Eq. (21) in Ref. [1] , where e is the joint probability distribution (mixing matrix) of the degrees.
If G is directed than the matrix e is the joint probability of the user-specified degree type for the source and target.
References
[1], [2]
Examples
>>> r = nx.degree_assortativity_coefficient(G)
>>> print(f"{r:3.1f}")
-0.5
3.2. Assortativity 197

attribute_assortativity_coefficient
attribute_assortativity_coefficient(G, attribute, nodes=None)

Compute assortativity for node attributes.
Assortativity measures the similarity of connections in the graph with respect to the given attribute.
Parameters
G
[NetworkX graph]
attribute
[string] Node attribute key
Compute attribute assortativity for nodes in container. The default is all nodes.
Returns
r: float
Assortativity of graph for given attribute
Notes
This computes Eq. (2) in Ref. [1] , (trace(M)-sum(M^2))/(1-sum(M^2)), where M is the joint probability distri-
bution (mixing matrix) of the specified attribute.
References
[1]
Examples
>>> G = nx.Graph()
>>> G.add_nodes_from([0, 1], color="red")
>>> G.add_nodes_from([2, 3], color="blue")
>>> G.add_edges_from([(0, 1), (2, 3)])
>>> print(nx.attribute_assortativity_coefficient(G, "color"))
1.0
numeric_assortativity_coefficient
numeric_assortativity_coefficient(G, attribute, nodes=None)

Compute assortativity for numerical node attributes.
Assortativity measures the similarity of connections in the graph with respect to the given numeric attribute.
Parameters
G
[NetworkX graph]
attribute
[string] Node attribute key.


Compute numeric assortativity only for attributes of nodes in container. The default is all
nodes.
Returns
r: float
Assortativity of graph for given attribute
Notes
This computes Eq. (21) in Ref. [1] , which is the Pearson correlation coefficient of the specified (scalar valued)
attribute across edges.
References
[1]
Examples
>>> G = nx.Graph()
>>> G.add_edges_from([(0, 1), (2, 3)])
>>> print(nx.numeric_assortativity_coefficient(G, "size"))
1.0
degree_pearson_correlation_coefficient
degree_pearson_correlation_coefficient(G, x='out', y='in', weight=None, nodes=None)

Assortativity measures the similarity of connections in the graph with respect to the node degree.
This is the same as degree_assortativity_coefficient but uses the potentially faster scipy.stats.pearsonr function.
Parameters
G
[NetworkX graph]
Compute pearson correlation of degrees only for specified nodes. The default is all nodes.
Returns

r
[float] Assortativity of graph by degree.
Notes
This calls scipy.stats.pearsonr.
References
[1], [2]
Examples
>>> r = nx.degree_pearson_correlation_coefficient(G)
>>> print(f"{r:3.1f}")
-0.5
3.2.2 Average neighbor degree
average_neighbor_degree(G[, source, target, Returns the average degree of the neighborhood of each
...]) node.
average_neighbor_degree
average_neighbor_degree(G, source='out', target='out', nodes=None, weight=None)

Returns the average degree of the neighborhood of each node.
In an undirected graph, the neighborhood N(i) of node i contains the nodes that are connected to i by an edge.
For directed graphs, N(i) is defined according to the parameter source:
• if source is ‘in’, then N(i) consists of predecessors of node i.
• if source is ‘out’, then N(i) consists of successors of node i.
• if source is ‘in+out’, then N(i) is both predecessors and successors.
The average neighborhood degree of a node i is
1 ∑
knn,i = kj
|N (i)|
j∈N (i)
where N(i) are the neighbors of node i and k_j is the degree of node j which belongs to N(i). For weighted
graphs, an analogous measure can be defined [1],
w 1 ∑
knn,i = wij kj
si
j∈N (i)
where s_i is the weighted degree of node i, w_{ij} is the weight of the edge that links i and j and N(i) are
the neighbors of node i.

Parameters
G
[NetworkX graph]
source
[string (“in”|”out”|”in+out”), optional (default=”out”)] Directed graphs only. Use “in”- or
“out”-neighbors of source node.
target
[string (“in”|”out”|”in+out”), optional (default=”out”)] Directed graphs only. Use “in”- or
“out”-degree for target node.
nodes
[list or iterable, optional (default=G.nodes)] Compute neighbor degree only for specified nodes.
weight
Returns
d: dict
A dictionary keyed by node to the average degree of its neighbors.
Raises
NetworkXError
If either source or target are not one of ‘in’, ‘out’, or ‘in+out’. If either source or
target is passed for an undirected graph.
See also:
average_degree_connectivity
References
[1]
Examples
>>> G.edges[0, 1]["weight"] = 5
>>> G.edges[2, 3]["weight"] = 3
>>> nx.average_neighbor_degree(G)
{0: 2.0, 1: 1.5, 2: 1.5, 3: 2.0}
>>> nx.average_neighbor_degree(G, weight="weight")
{0: 2.0, 1: 1.1666666666666667, 2: 1.25, 3: 2.0}
>>> nx.add_path(G, [0, 1, 2, 3])
>>> nx.average_neighbor_degree(G, source="in", target="in")
{0: 0.0, 1: 0.0, 2: 1.0, 3: 1.0}
>>> nx.average_neighbor_degree(G, source="out", target="out")

{0: 1.0, 1: 1.0, 2: 0.0, 3: 0.0}

3.2.3 Average degree connectivity
average_degree_connectivity(G[, source, Compute the average degree connectivity of graph.

...])
average_degree_connectivity
average_degree_connectivity(G, source='in+out', target='in+out', nodes=None, weight=None)

Compute the average degree connectivity of graph.
The average degree connectivity is the average nearest neighbor degree of nodes with degree k. For weighted
graphs, an analogous measure can be computed using the weighted average neighbors degree defined in [1], for a
node i, as
w 1 ∑
knn,i = wij kj
si
j∈N (i)
where s_i is the weighted degree of node i, w_{ij} is the weight of the edge that links i and j, and N(i) are
the neighbors of node i.
Parameters
G
[NetworkX graph]
source
[“in”|”out”|”in+out” (default:”in+out”)] Directed graphs only. Use “in”- or “out”-degree for
source node.
target
[“in”|”out”|”in+out” (default:”in+out”] Directed graphs only. Use “in”- or “out”-degree for
target node.
nodes
[list or iterable (optional)] Compute neighbor connectivity for these nodes. The default is all
nodes.
weight
Returns
d
[dict] A dictionary keyed by degree k with the value of average connectivity.
Raises
NetworkXError
If either source or target are not one of ‘in’, ‘out’, or ‘in+out’. If either source or
target is passed for an undirected graph.
See also:
average_neighbor_degree

References
[1]
Examples
>>> G.edges[1, 2]["weight"] = 3
>>> nx.average_degree_connectivity(G)
{1: 2.0, 2: 1.5}
>>> nx.average_degree_connectivity(G, weight="weight")
{1: 2.0, 2: 1.75}
3.2.4 Mixing
attribute_mixing_matrix(G, attribute[, ...]) Returns mixing matrix for attribute.

degree_mixing_matrix(G[, x, y, weight, ...]) Returns mixing matrix for attribute.
attribute_mixing_dict(G, attribute[, nodes, ...]) Returns dictionary representation of mixing matrix for at-
tribute.
degree_mixing_dict(G[, x, y, weight, nodes, ...]) Returns dictionary representation of mixing matrix for
degree.
mixing_dict(xy[, normalized]) Returns a dictionary representation of mixing matrix.
attribute_mixing_matrix
attribute_mixing_matrix(G, attribute, nodes=None, mapping=None, normalized=True)

Returns mixing matrix for attribute.
Parameters
G
[graph] NetworkX graph object.
attribute
Use only nodes in container to build the matrix. The default is all nodes.
mapping
[dictionary, optional] Mapping from node attribute to integer index in matrix. If not specified,
an arbitrary ordering will be used.
normalized
[bool (default=True)] Return counts if False or probabilities if True.
Returns
m: numpy array
Counts or joint probability of occurrence of attribute pairs.

Notes
If each node has a unique attribute value, the unnormalized mixing matrix will be equal to the adjacency matrix. To
get a denser mixing matrix, the rounding can be performed to form groups of nodes with equal values. For example,
the exact height of persons in cm (180.79155222, 163.9080892, 163.30095355, 167.99016217, 168.21590163,
…) can be rounded to (180, 163, 163, 168, 168, …).
Definitions of attribute mixing matrix vary on whether the matrix should include rows for attribute values that don’t
arise. Here we do not include such empty-rows. But you can force them to appear by inputting a mapping that
includes those values.
Examples
>>> gender = {0: 'male', 1: 'female', 2: 'female'}
>>> nx.set_node_attributes(G, gender, 'gender')
>>> mapping = {'male': 0, 'female': 1}
>>> mix_mat = nx.attribute_mixing_matrix(G, 'gender', mapping=mapping)
>>> # mixing from male nodes to female nodes
>>> mix_mat[mapping['male'], mapping['female']]
0.25
degree_mixing_matrix
degree_mixing_matrix(G, x='out', y='in', weight=None, nodes=None, normalized=True, mapping=None)

Returns mixing matrix for attribute.
Parameters
G
Build the matrix using only nodes in container. The default is all nodes.
normalized
[bool (default=True)] Return counts if False or probabilities if True.
mapping
[dictionary, optional] Mapping from node degree to integer index in matrix. If not specified,
an arbitrary ordering will be used.
Returns
m: numpy array
Counts, or joint probability, of occurrence of node degree.

Notes
Definitions of degree mixing matrix vary on whether the matrix should include rows for degree values that don’t
arise. Here we do not include such empty-rows. But you can force them to appear by inputting a mapping that
includes those values. See examples.
Examples
>>> G = nx.star_graph(3)
>>> mix_mat = nx.degree_mixing_matrix(G)
>>> mix_mat[0, 1] # mixing from node degree 1 to node degree 3
0.5
If you want every possible degree to appear as a row, even if no nodes have that degree, use mapping as follows,
>>> max_degree = max(deg for n, deg in G.degree)

>>> mapping = {x: x for x in range(max_degree + 1)} # identity mapping
>>> mix_mat = nx.degree_mixing_matrix(G, mapping=mapping)
>>> mix_mat[3, 1] # mixing from node degree 3 to node degree 1
0.5
attribute_mixing_dict
attribute_mixing_dict(G, attribute, nodes=None, normalized=False)

Returns dictionary representation of mixing matrix for attribute.
Parameters
G
attribute
Unse nodes in container to build the dict. The default is all nodes.
normalized
[bool (default=False)] Return counts if False or probabilities if True.
Returns
d
[dictionary] Counts or joint probability of occurrence of attribute pairs.
Examples
>>> G = nx.Graph()
>>> G.add_nodes_from([0, 1], color="red")
>>> G.add_nodes_from([2, 3], color="blue")
>>> d = nx.attribute_mixing_dict(G, "color")
>>> print(d["red"]["blue"])
1


>>> print(d["blue"]["red"]) # d symmetric for undirected graphs
1
degree_mixing_dict
degree_mixing_dict(G, x='out', y='in', weight=None, nodes=None, normalized=False)

Returns dictionary representation of mixing matrix for degree.
Parameters
G
normalized
Returns
d: dictionary
Counts or joint probability of occurrence of degree pairs.
mixing_dict
mixing_dict(xy, normalized=False)
Returns a dictionary representation of mixing matrix.
Parameters
xy
[list or container of two-tuples] Pairs of (x,y) items.
attribute
[string] Node attribute key
normalized
Returns
d: dictionary
Counts or Joint probability of occurrence of values in xy.

3.2.5 Pairs
node_attribute_xy(G, attribute[, nodes]) Returns iterator of node-attribute pairs for all edges in G.
node_degree_xy(G[, x, y, weight, nodes]) Generate node degree-degree pairs for edges in G.
node_attribute_xy
node_attribute_xy(G, attribute, nodes=None)

Returns iterator of node-attribute pairs for all edges in G.
Parameters
G: NetworkX graph
attribute: key
The node attribute key.
Use only edges that are incident to specified nodes. The default is all nodes.
Returns
(x, y): 2-tuple
Generates 2-tuple of (attribute, attribute) values.
Notes
For undirected graphs each edge is produced twice, once for each edge representation (u, v) and (v, u), with the
exception of self-loop edges which only appear once.
Examples
>>> G.add_node(1, color="red")
>>> G.add_node(2, color="blue")
>>> list(nx.node_attribute_xy(G, "color"))
[('red', 'blue')]
node_degree_xy
node_degree_xy(G, x='out', y='in', weight=None, nodes=None)

Generate node degree-degree pairs for edges in G.
Parameters
G: NetworkX graph


Use only edges that are adjacency to specified nodes. The default is all nodes.
Returns
(x, y): 2-tuple
Generates 2-tuple of (degree, degree) values.
Notes
For undirected graphs each edge is produced twice, once for each edge representation (u, v) and (v, u), with the
exception of self-loop edges which only appear once.
Examples
>>> list(nx.node_degree_xy(G, x="out", y="in"))
[(1, 1)]
>>> list(nx.node_degree_xy(G, x="in", y="out"))
[(0, 0)]
3.3 Asteroidal
Algorithms for asteroidal triples and asteroidal numbers in graphs.

An asteroidal triple in a graph G is a set of three non-adjacent vertices u, v and w such that there exist a path between
any two of them that avoids closed neighborhood of the third. More formally, v_j, v_k belongs to the same connected
component of G - N[v_i], where N[v_i] denotes the closed neighborhood of v_i. A graph which does not contain any
asteroidal triples is called an AT-free graph. The class of AT-free graphs is a graph class for which many NP-complete
problems are solvable in polynomial time. Amongst them, independent set and coloring.
is_at_free(G) Check if a graph is AT-free.

find_asteroidal_triple(G) Find an asteroidal triple in the given graph.
3.3.1 is_at_free
is_at_free(G)
Check if a graph is AT-free.
The method uses the find_asteroidal_triple method to recognize an AT-free graph. If no asteroidal
triple is found the graph is AT-free and True is returned. If at least one asteroidal triple is found the graph is not
AT-free and False is returned.
Parameters
G
[NetworkX Graph] The graph to check whether is AT-free or not.

Returns
bool
True if G is AT-free and False otherwise.
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (1, 2), (1, 3), (1, 4), (4, 5)])
>>> nx.is_at_free(G)
True
>>> nx.is_at_free(G)
False
3.3.2 find_asteroidal_triple
find_asteroidal_triple(G)
Find an asteroidal triple in the given graph.
An asteroidal triple is a triple of non-adjacent vertices such that there exists a path between any two of them which
avoids the closed neighborhood of the third. It checks all independent triples of vertices and whether they are an
asteroidal triple or not. This is done with the help of a data structure called a component structure. A component
structure encodes information about which vertices belongs to the same connected component when the closed
neighborhood of a given vertex is removed from the graph. The algorithm used to check is the trivial one, outlined
in [1], which has a runtime of O(|V ||E+|V ||E|), where the second term is the creation of the component structure.
Parameters
G
[NetworkX Graph] The graph to check whether is AT-free or not
Returns
list or None
An asteroidal triple is returned as a list of nodes. If no asteroidal triple exists, i.e. the graph is
AT-free, then None is returned. The returned value depends on the certificate parameter. The
default option is a bool which is True if the graph is AT-free, i.e. the given graph contains no
asteroidal triples, and False otherwise, i.e. if the graph contains at least one asteroidal triple.
Notes
The component structure and the algorithm is described in [1]. The current implementation implements the trivial
algorithm for simple graphs.
3.3. Asteroidal 209

References
[1]
3.4 Bipartite
This module provides functions and operations for bipartite graphs. Bipartite graphs B = (U, V, E) have two node
sets U,V and edges in E that only connect nodes from opposite sets. It is common in the literature to use an spatial analogy
referring to the two node sets as top and bottom nodes.
The bipartite algorithms are not imported into the networkx namespace at the top level so the easiest way to use them is
with:
>>> from networkx.algorithms import bipartite
NetworkX does not have a custom bipartite graph class but the Graph() or DiGraph() classes can be used to represent
bipartite graphs. However, you have to keep track of which set each node belongs to, and make sure that there is no edge
between nodes of the same set. The convention used in NetworkX is to use a node attribute named bipartite with
values 0 or 1 to identify the sets each node belongs to. This convention is not enforced in the source code of bipartite
functions, it’s only a recommendation.
For example:
>>> B = nx.Graph()
>>> # Add nodes with the node attribute "bipartite"
>>> B.add_nodes_from([1, 2, 3, 4], bipartite=0)
>>> B.add_nodes_from(["a", "b", "c"], bipartite=1)
>>> # Add edges only between nodes of opposite node sets
>>> B.add_edges_from([(1, "a"), (1, "b"), (2, "b"), (2, "c"), (3, "c"), (4, "a")])
Many algorithms of the bipartite module of NetworkX require, as an argument, a container with all the nodes that belong
to one set, in addition to the bipartite graph B. The functions in the bipartite package do not check that the node set is
actually correct nor that the input graph is actually bipartite. If B is connected, you can find the two node sets using a
two-coloring algorithm:
>>> nx.is_connected(B)
True
>>> bottom_nodes, top_nodes = bipartite.sets(B)
However, if the input graph is not connected, there are more than one possible colorations. This is the reason why we
require the user to pass a container with all nodes of one bipartite node set as an argument to most bipartite functions.
In the face of ambiguity, we refuse the temptation to guess and raise an AmbiguousSolution Exception if the input
graph for bipartite.sets is disconnected.
Using the bipartite node attribute, you can easily get the two node sets:
>>> top_nodes = {n for n, d in B.nodes(data=True) if d["bipartite"] == 0}

>>> bottom_nodes = set(B) - top_nodes
So you can easily use the bipartite algorithms that require, as an argument, a container with all nodes that belong to one
node set:
>>> print(round(bipartite.density(B, bottom_nodes), 2))

0.5
>>> G = bipartite.projected_graph(B, top_nodes)

All bipartite graph generators in NetworkX build bipartite graphs with the bipartite node attribute. Thus, you can
use the same approach:
>>> RB = bipartite.random_graph(5, 7, 0.2)

>>> RB_top = {n for n, d in RB.nodes(data=True) if d["bipartite"] == 0}
>>> RB_bottom = set(RB) - RB_top
>>> list(RB_top)
[0, 1, 2, 3, 4]
>>> list(RB_bottom)
[5, 6, 7, 8, 9, 10, 11]
For other bipartite graph generators see Generators.
3.4.1 Basic functions
is_bipartite(G) Returns True if graph G is bipartite, False if not.

is_bipartite_node_set(G, nodes) Returns True if nodes and G/nodes are a bipartition of G.
sets(G[, top_nodes]) Returns bipartite node sets of graph G.
color(G) Returns a two-coloring of the graph.
density(B, nodes) Returns density of bipartite graph B.
degrees(B, nodes[, weight]) Returns the degrees of the two node sets in the bipartite
graph B.
is_bipartite
is_bipartite(G)
Returns True if graph G is bipartite, False if not.
Parameters
G
[NetworkX graph]
See also:
color, is_bipartite_node_set
Examples

>>> print(bipartite.is_bipartite(G))
True
3.4. Bipartite 211

is_bipartite_node_set
is_bipartite_node_set(G, nodes)
Returns True if nodes and G/nodes are a bipartition of G.
Parameters
G
[NetworkX graph]
nodes: list or container
Check if nodes are a one of a bipartite set.
Notes
An exception is raised if the input nodes are not distinct, because in this case some bipartite algorithms will yield
incorrect results. For connected graphs the bipartite sets are unique. This function handles disconnected graphs.
Examples

>>> X = set([1, 3])
>>> bipartite.is_bipartite_node_set(G, X)
True
sets
sets(G, top_nodes=None)
Returns bipartite node sets of graph G.
Raises an exception if the graph is not bipartite or if the input graph is disconnected and thus more than one valid
solution exists. See bipartite documentation for further details on how bipartite graphs are handled in
NetworkX.
Parameters
G
[NetworkX graph]
top_nodes
[container, optional] Container with all nodes in one bipartite node set. If not supplied it will
be computed. But if more than one solution exists an exception will be raised.
Returns
X
[set] Nodes from one side of the bipartite graph.
Y
[set] Nodes from the other side.
Raises

AmbiguousSolution
Raised if the input bipartite graph is disconnected and no container with all nodes in one
bipartite set is provided. When determining the nodes in each bipartite set more than one valid
solution is possible if the input graph is disconnected.
NetworkXError
Raised if the input graph is not bipartite.
See also:
color
Examples

>>> X, Y = bipartite.sets(G)
>>> list(X)
[0, 2]
>>> list(Y)
[1, 3]
color
color(G)
Returns a two-coloring of the graph.
Raises an exception if the graph is not bipartite.
Parameters
G
[NetworkX graph]
Returns
color
[dictionary] A dictionary keyed by node with a 1 or 0 as data for each node color.
Raises
NetworkXError
If the graph is not two-colorable.
Examples

>>> c = bipartite.color(G)
>>> print(c)
{0: 1, 1: 0, 2: 1, 3: 0}
You can use this to set a node attribute indicating the biparite set:
3.4. Bipartite 213

>>> nx.set_node_attributes(G, c, "bipartite")

>>> print(G.nodes[0]["bipartite"])
1
>>> print(G.nodes[1]["bipartite"])
0
density
density(B, nodes)
Returns density of bipartite graph B.
Parameters
B
[NetworkX graph]
Nodes in one node set of the bipartite graph.
Returns
d
[float] The bipartite density
See also:
color
Notes
The container of nodes passed as argument must contain all nodes in one of the two bipartite node sets to avoid
ambiguity in the case of disconnected graphs. See bipartite documentation for further details on how
bipartite graphs are handled in NetworkX.
Examples

>>> G = nx.complete_bipartite_graph(3, 2)
>>> X = set([0, 1, 2])
>>> bipartite.density(G, X)
1.0
>>> Y = set([3, 4])
>>> bipartite.density(G, Y)
1.0

degrees
degrees(B, nodes, weight=None)

Returns the degrees of the two node sets in the bipartite graph B.
Parameters
B
[NetworkX graph]
Nodes in one node set of the bipartite graph.
weight
weights adjacent to the node.
Returns
(degX,degY)
[tuple of dictionaries] The degrees of the two bipartite sets as dictionaries keyed by node.
See also:
color, density
Notes
The container of nodes passed as argument must contain all nodes in one of the two bipartite node sets to avoid
ambiguity in the case of disconnected graphs. See bipartite documentation for further details on how
bipartite graphs are handled in NetworkX.
Examples

>>> Y = set([3, 4])
>>> degX, degY = bipartite.degrees(G, Y)
>>> dict(degX)
{0: 2, 1: 2, 2: 2}
3.4.2 Edgelist
Read and write NetworkX graphs as bipartite edge lists.
3.4. Bipartite 215

Format
You can read or write three formats of edge lists with these functions.
Node pairs with no data:
1 2
Python dictionary as data:
1 2 {'weight':7, 'color':'green'}
Arbitrary data:
1 2 7 green
For each edge (u, v) the node u is assigned to part 0 and the node v to part 1.
generate_edgelist(G[, delimiter, data]) Generate a single line of the bipartite graph G in edge list
format.
write_edgelist(G, path[, comments, ...]) Write a bipartite graph as a list of edges.
parse_edgelist(lines[, comments, delimiter, ...]) Parse lines of an edge list representation of a bipartite
graph.
read_edgelist(path[, comments, delimiter, ...]) Read a bipartite graph from a list of edges.
generate_edgelist
generate_edgelist(G, delimiter=' ', data=True)

Generate a single line of the bipartite graph G in edge list format.
Parameters
G
[NetworkX graph] The graph is assumed to have node attribute part set to 0,1 representing
the two graph parts
delimiter
[string, optional] Separator for node labels
data
[bool or list of keys] If False generate no edge data. If True use a dictionary representation of
edge data. If a list of keys use a list of data values corresponding to the keys.
Returns
lines
[string] Lines of data in adjlist format.

Examples

>>> G.add_nodes_from([0, 2], bipartite=0)
>>> G[1][2]["weight"] = 3
>>> G[2][3]["capacity"] = 12
>>> for line in bipartite.generate_edgelist(G, data=False):
... print(line)
0 1
2 1
2 3
>>> for line in bipartite.generate_edgelist(G):

... print(line)
0 1 {}
2 1 {'weight': 3}
2 3 {'capacity': 12}
>>> for line in bipartite.generate_edgelist(G, data=["weight"]):

... print(line)
0 1
2 1 3
2 3
write_edgelist
write_edgelist(G, path, comments='#', delimiter=' ', data=True, encoding='utf-8')

Write a bipartite graph as a list of edges.
Parameters
G
[Graph] A NetworkX bipartite graph
path
[file or string] File or filename to write. If a file is provided, it must be opened in ‘wb’ mode.
Filenames ending in .gz or .bz2 will be compressed.
comments
[string, optional] The character used to indicate the start of a comment
delimiter
[string, optional] The string used to separate values. The default is whitespace.
data
[bool or list, optional] If False write no edge data. If True write a string representation of the
edge data dictionary.. If a list (or other iterable) is provided, write the keys specified in the list.
encoding: string, optional
Specify which encoding to use when writing file.
See also:
write_edgelist
generate_edgelist
3.4. Bipartite 217

Examples
>>> nx.write_edgelist(G, "test.edgelist")
>>> fh = open("test.edgelist", "wb")
>>> nx.write_edgelist(G, fh)
>>> nx.write_edgelist(G, "test.edgelist.gz")
>>> nx.write_edgelist(G, "test.edgelist.gz", data=False)
>>> G = nx.Graph()
>>> G.add_edge(1, 2, weight=7, color="red")
>>> nx.write_edgelist(G, "test.edgelist", data=False)
>>> nx.write_edgelist(G, "test.edgelist", data=["color"])
>>> nx.write_edgelist(G, "test.edgelist", data=["color", "weight"])
parse_edgelist
parse_edgelist(lines, comments='#', delimiter=None, create_using=None, nodetype=None, data=True)

Parse lines of an edge list representation of a bipartite graph.
Parameters
lines
[list or iterator of strings] Input data in edgelist format
comments
[string, optional] Marker for comment lines
delimiter
create_using: NetworkX graph container, optional
Use given NetworkX graph for holding nodes or edges.
nodetype
[Python type, optional] Convert nodes to this type.
data
[bool or list of (label,type) tuples] If False generate no edge data or if True use a dictionary
representation of edge data or a list tuples specifying dictionary key names and types for edge
data.
Returns
G: NetworkX Graph
The bipartite graph corresponding to lines

Examples
Edgelist with no data:

>>> lines = ["1 2", "2 3", "3 4"]
>>> G = bipartite.parse_edgelist(lines, nodetype=int)
>>> sorted(G.nodes())
[1, 2, 3, 4]
>>> sorted(G.nodes(data=True))
[(1, {'bipartite': 0}), (2, {'bipartite': 0}), (3, {'bipartite': 0}), (4, {
,→'bipartite': 1})]
>>> sorted(G.edges())
[(1, 2), (2, 3), (3, 4)]
Edgelist with data in Python dictionary representation:
>>> lines = ["1 2 {'weight':3}", "2 3 {'weight':27}", "3 4 {'weight':3.0}"]

>>> G = bipartite.parse_edgelist(lines, nodetype=int)
[1, 2, 3, 4]
>>> sorted(G.edges(data=True))
[(1, 2, {'weight': 3}), (2, 3, {'weight': 27}), (3, 4, {'weight': 3.0})]
Edgelist with data in a list:
>>> lines = ["1 2 3", "2 3 27", "3 4 3.0"]

>>> G = bipartite.parse_edgelist(lines, nodetype=int, data=(("weight", float),))
[1, 2, 3, 4]
>>> sorted(G.edges(data=True))
[(1, 2, {'weight': 3.0}), (2, 3, {'weight': 27.0}), (3, 4, {'weight': 3.0})]
read_edgelist
read_edgelist(path, comments='#', delimiter=None, create_using=None, nodetype=None, data=True,

edgetype=None, encoding='utf-8')
Read a bipartite graph from a list of edges.
Parameters
path
[file or string] File or filename to read. If a file is provided, it must be opened in ‘rb’ mode.
Filenames ending in .gz or .bz2 will be uncompressed.
comments
[string, optional] The character used to indicate the start of a comment.
delimiter
create_using
[Graph container, optional,] Use specified container to build graph. The default is net-
workx.Graph, an undirected graph.
nodetype
[int, float, str, Python type, optional] Convert node data from strings to specified type
3.4. Bipartite 219

data
[bool or list of (label,type) tuples] Tuples specifying dictionary key names and types for edge
data
edgetype
[int, float, str, Python type, optional OBSOLETE] Convert edge data from strings to specified
type and use as ‘weight’
Specify which encoding to use when reading file.
Returns
G
[graph] A networkx Graph or other type specified with create_using
See also:
parse_edgelist
Notes
Since nodes must be hashable, the function nodetype must return hashable types (e.g. int, float, str, frozenset - or
tuples of those, etc.)
Examples

>>> bipartite.write_edgelist(G, "test.edgelist")
>>> G = bipartite.read_edgelist("test.edgelist")
>>> fh = open("test.edgelist", "rb")

>>> G = bipartite.read_edgelist(fh)
>>> fh.close()
>>> G = bipartite.read_edgelist("test.edgelist", nodetype=int)
>>> textline = "1 2 3"

>>> fh = open("test.edgelist", "w")
>>> d = fh.write(textline)
>>> fh.close()
>>> G = bipartite.read_edgelist(
... "test.edgelist", nodetype=int, data=(("weight", float),)
... )
>>> list(G)
[1, 2]
>>> list(G.edges(data=True))
[(1, 2, {'weight': 3.0})]
See parse_edgelist() for more examples of formatting.

3.4.3 Matching
Provides functions for computing maximum cardinality matchings and minimum weight full matchings in a bipartite
graph.
If you don’t care about the particular implementation of the maximum matching algorithm, simply use the maxi-
mum_matching(). If you do care, you can import one of the named maximum matching algorithms directly.
For example, to find a maximum matching in the complete bipartite graph with two vertices on the left and three vertices
on the right:
>>> left, right = nx.bipartite.sets(G)
>>> list(left)
[0, 1]
>>> list(right)
[2, 3, 4]
>>> nx.bipartite.maximum_matching(G)
{0: 2, 1: 3, 2: 0, 3: 1}
The dictionary returned by maximum_matching() includes a mapping for vertices in both the left and right vertex
sets.
Similarly, minimum_weight_full_matching() produces, for a complete weighted bipartite graph, a matching
whose cardinality is the cardinality of the smaller of the two partitions, and for which the sum of the weights of the edges
included in the matching is minimal.
eppstein_matching(G[, top_nodes]) Returns the maximum cardinality matching of the bipar-

tite graph G.
hopcroft_karp_matching(G[, top_nodes]) Returns the maximum cardinality matching of the bipar-
tite graph G.
to_vertex_cover(G, matching[, top_nodes]) Returns the minimum vertex cover corresponding to the
given maximum matching of the bipartite graph G.
maximum_matching(G[, top_nodes]) Returns the maximum cardinality matching in the given
bipartite graph.
minimum_weight_full_matching(G[, Returns a minimum weight full matching of the bipartite
top_nodes, ...]) graph G.
eppstein_matching
eppstein_matching(G, top_nodes=None)
Returns the maximum cardinality matching of the bipartite graph G.
Parameters
G
[NetworkX graph] Undirected bipartite graph
top_nodes
[container] Container with all nodes in one bipartite node set. If not supplied it will be com-
puted. But if more than one solution exists an exception will be raised.
Returns
matches
[dictionary] The matching is returned as a dictionary, matching, such that matching[v]
== w if node v is matched to node w. Unmatched nodes do not occur as a key in matching.
3.4. Bipartite 221

Raises
AmbiguousSolution
See also:
hopcroft_karp_matching
Notes
This function is implemented with David Eppstein’s version of the algorithm Hopcroft–Karp algorithm (see
hopcroft_karp_matching()), which originally appeared in the Python Algorithms and Data Structures
library (PADS).
See bipartite documentation for further details on how bipartite graphs are handled in NetworkX.
hopcroft_karp_matching(G, top_nodes=None)
Returns the maximum cardinality matching of the bipartite graph G.
A matching is a set of edges that do not share any nodes. A maximum cardinality matching is a matching with the
most edges possible. It is not always unique. Finding a matching in a bipartite graph can be treated as a networkx
flow problem.
The functions hopcroft_karp_matching and maximum_matching are aliases of the same function.
Parameters
G
top_nodes
[container of nodes] Container with all nodes in one bipartite node set. If not supplied it will
be computed. But if more than one solution exists an exception will be raised.
Returns
matches
[dictionary] The matching is returned as a dictionary, matches, such that matches[v]
== w if node v is matched to node w. Unmatched nodes do not occur as a key in matches.
Raises
AmbiguousSolution
See also:
maximum_matching
eppstein_matching

Notes
This function is implemented with the Hopcroft–Karp matching algorithm for bipartite graphs.
References
[1]
to_vertex_cover
to_vertex_cover(G, matching, top_nodes=None)

Returns the minimum vertex cover corresponding to the given maximum matching of the bipartite graph G.
Parameters
G
matching
[dictionary] A dictionary whose keys are vertices in G and whose values are the distinct
neighbors comprising the maximum matching for G, as returned by, for example, maxi-
mum_matching(). The dictionary must represent the maximum matching.
top_nodes
puted. But if more than one solution exists an exception will be raised.
Returns
vertex_cover
[set] The minimum vertex cover in G.
Raises
AmbiguousSolution
Notes
This function is implemented using the procedure guaranteed by Konig’s theorem, which proves an equivalence
between a maximum matching and a minimum vertex cover in bipartite graphs.
Since a minimum vertex cover is the complement of a maximum independent set for any graph, one can compute
the maximum independent set of a bipartite graph this way:
>>> matching = nx.bipartite.maximum_matching(G)
>>> vertex_cover = nx.bipartite.to_vertex_cover(G, matching)
>>> independent_set = set(G) - vertex_cover
>>> print(list(independent_set))
[2, 3, 4]
3.4. Bipartite 223

maximum_matching
maximum_matching(G, top_nodes=None)
Returns the maximum cardinality matching in the given bipartite graph.
This function is simply an alias for hopcroft_karp_matching().
minimum_weight_full_matching
minimum_weight_full_matching(G, top_nodes=None, weight='weight')

Returns a minimum weight full matching of the bipartite graph G.
Let G = ((U, V ), E) be a weighted bipartite graph with real weights w : E → R. This function then produces a
matching M ⊆ E with cardinality
|M | = min(|U |, |V |),
∑
which minimizes the sum of the weights of the edges included in the matching, e∈M w(e), or raises an error if
no such matching exists.
When |U | = |V |, this is commonly referred to as a perfect matching; here, since we allow |U | and |V | to differ,
we follow Karp [1] and refer to the matching as full.
Parameters
G
top_nodes
puted.
weight
[string, optional (default=’weight’)] The edge data key used to provide each value in the matrix.
Returns
matches
[dictionary] The matching is returned as a dictionary, matches, such that matches[v]
== w if node v is matched to node w. Unmatched nodes do not occur as a key in matches.
Raises
ValueError
Raised if no full matching exists.
ImportError
Raised if SciPy is not available.
Notes
The problem of determining a minimum weight full matching is also known as the rectangular linear assignment
problem. This implementation defers the calculation of the assignment to SciPy.

References
[1]
3.4.4 Matrix
biadjacency_matrix(G, row_order[, ...]) Returns the biadjacency matrix of the bipartite graph G.
from_biadjacency_matrix(A[, create_using, Creates a new bipartite graph from a biadjacency matrix
...]) given as a SciPy sparse array.
biadjacency_matrix
biadjacency_matrix(G, row_order, column_order=None, dtype=None, weight='weight', format='csr')

Returns the biadjacency matrix of the bipartite graph G.
Let G = (U, V, E) be a bipartite graph with node sets U = u_{1},...,u_{r} and V = v_{1},..
.,v_{s}. The biadjacency matrix [1] is the r x s matrix B in which b_{i,j} = 1 if, and only if, (u_i,
v_j) in E. If the parameter weight is not None and matches the name of an edge attribute, its value is used
instead of 1.
Parameters
G
[graph] A NetworkX graph
row_order
[list of nodes] The rows of the matrix are ordered according to the list of nodes.
column_order
[list, optional] The columns of the matrix are ordered according to the list of nodes. If col-
umn_order is None, then the ordering of columns is arbitrary.
dtype
[NumPy data-type, optional] A valid NumPy dtype used to initialize the array. If None, then
the NumPy default is used.
weight
[string or None, optional (default=’weight’)] The edge data key used to provide each value in
the matrix. If None, then each edge has weight 1.
format
[str in {‘bsr’, ‘csr’, ‘csc’, ‘coo’, ‘lil’, ‘dia’, ‘dok’}] The type of the matrix to be returned (default
‘csr’). For some algorithms different implementations of sparse matrices can perform better.
See [2] for details.
Returns
M
[SciPy sparse array] Biadjacency matrix representation of the bipartite graph G.
See also:
adjacency_matrix
from_biadjacency_matrix
3.4. Bipartite 225

Notes
No attempt is made to check that the input graph is bipartite.

For directed bipartite graphs only successors are considered as neighbors. To obtain an adjacency matrix with ones
(or weight values) for both predecessors and successors you have to generate two biadjacency matrices where the
rows of one of them are the columns of the other, and then add one to the transpose of the other.
References
[1], [2]
from_biadjacency_matrix
from_biadjacency_matrix(A, create_using=None, edge_attribute='weight')

Creates a new bipartite graph from a biadjacency matrix given as a SciPy sparse array.
Parameters
A: scipy sparse array
A biadjacency matrix representation of a graph
create_using: NetworkX graph
Use specified graph for result. The default is Graph()
edge_attribute: string
Name of edge attribute to store matrix numeric value. The data will have the same type as the
matrix entry (int, float, (real,imag)).
See also:
biadjacency_matrix
from_numpy_array
Notes
The nodes are labeled with the attribute bipartite set to an integer 0 or 1 representing membership in part 0
or part 1 of the bipartite graph.
If create_using is an instance of networkx.MultiGraph or networkx.MultiDiGraph and the
entries of A are of type int, then this function returns a multigraph (of the same type as create_using) with
parallel edges. In this case, edge_attribute will be ignored.
References
[1] https://en.wikipedia.org/wiki/Adjacency_matrix#Adjacency_matrix_of_a_bipartite_graph

3.4.5 Projections
One-mode (unipartite) projections of bipartite graphs.
projected_graph(B, nodes[, multigraph]) Returns the projection of B onto one of its node sets.
weighted_projected_graph(B, nodes[, ratio])
Returns a weighted projection of B onto one of its node
sets.
collaboration_weighted_projected_graph(B,
Newman's weighted projection of B onto one of its node
nodes) sets.
overlap_weighted_projected_graph(B, Overlap weighted projection of B onto one of its node sets.
nodes[, ...])
generic_weighted_projected_graph(B, Weighted projection of B with a user-specified weight
nodes[, ...]) function.
projected_graph
projected_graph(B, nodes, multigraph=False)

Returns the projection of B onto one of its node sets.
Returns the graph G that is the projection of the bipartite graph B onto the specified nodes. They retain their
attributes and are connected in G if they have a common neighbor in B.
Parameters
B
[NetworkX graph] The input graph should be bipartite.
nodes
[list or iterable] Nodes to project onto (the “bottom” nodes).
multigraph: bool (default=False)
If True return a multigraph where the multiple edges represent multiple shared neighbors. They
edge key in the multigraph is assigned to the label of the neighbor.
Returns
Graph
[NetworkX graph or multigraph] A graph that is the projection onto the given nodes.
See also:
is_bipartite
sets
weighted_projected_graph
collaboration_weighted_projected_graph
overlap_weighted_projected_graph
generic_weighted_projected_graph
3.4. Bipartite 227

Notes
No attempt is made to verify that the input graph B is bipartite. Returns a simple graph that is the projection of the
bipartite graph B onto the set of nodes given in list nodes. If multigraph=True then a multigraph is returned with
an edge for every shared neighbor.
Directed graphs are allowed as input. The output will also then be a directed graph with edges if there is a directed
path between the nodes.
The graph and node properties are (shallow) copied to the projected graph.
Examples

>>> B = nx.path_graph(4)
>>> G = bipartite.projected_graph(B, [1, 3])
>>> list(G)
[1, 3]
>>> list(G.edges())
[(1, 3)]
If nodes a, and b are connected through both nodes 1 and 2 then building a multigraph results in two edges in the
projection onto [a, b]:
>>> B = nx.Graph()
>>> B.add_edges_from([("a", 1), ("b", 1), ("a", 2), ("b", 2)])
>>> G = bipartite.projected_graph(B, ["a", "b"], multigraph=True)
>>> print([sorted((u, v)) for u, v in G.edges()])
[['a', 'b'], ['a', 'b']]
weighted_projected_graph(B, nodes, ratio=False)

Returns a weighted projection of B onto one of its node sets.
The weighted projected graph is the projection of the bipartite network B onto the specified nodes with weights
representing the number of shared neighbors or the ratio between actual shared neighbors and possible shared
neighbors if ratio is True [1]. The nodes retain their attributes and are connected in the resulting graph if
they have an edge to a common node in the original graph.
Parameters
B
nodes
[list or iterable] Distinct nodes to project onto (the “bottom” nodes).
ratio: Bool (default=False)
If True, edge weight is the ratio between actual shared neighbors and maximum possible shared
neighbors (i.e., the size of the other node set). If False, edges weight is the number of shared
neighbors.
Returns

Graph
[NetworkX graph] A graph that is the projection onto the given nodes.
See also:
is_bipartite
sets
projected_graph
Notes
No attempt is made to verify that the input graph B is bipartite, or that the input nodes are distinct. However, if the
length of the input nodes is greater than or equal to the nodes in the graph B, an exception is raised. If the nodes
are not distinct but don’t raise this error, the output weights will be incorrect. The graph and node properties are
(shallow) copied to the projected graph.
References
[1]
Examples

>>> G = bipartite.weighted_projected_graph(B, [1, 3])
>>> list(G)
[1, 3]
[(1, 3, {'weight': 1})]
>>> G = bipartite.weighted_projected_graph(B, [1, 3], ratio=True)
[(1, 3, {'weight': 0.5})]
collaboration_weighted_projected_graph(B, nodes)
Newman’s weighted projection of B onto one of its node sets.
The collaboration weighted projection is the projection of the bipartite network B onto the specified nodes with
weights assigned using Newman’s collaboration model [1]:
∑ δk δk
u v
wu,v =
dk − 1
k
where u and v are nodes from the bottom bipartite node set, and k is a node of the top node set. The value d_k
is the degree of node k in the bipartite network and delta_{u}^{k} is 1 if node u is linked to node k in the
original bipartite graph or 0 otherwise.
3.4. Bipartite 229

The nodes retain their attributes and are connected in the resulting graph if have an edge to a common node in the
original bipartite graph.
Parameters
B
nodes
Returns
Graph
See also:
is_bipartite
sets
projected_graph
Notes
No attempt is made to verify that the input graph B is bipartite. The graph and node properties are (shallow) copied
to the projected graph.
References
[1]
Examples

>>> B.add_edge(1, 5)
>>> G = bipartite.collaboration_weighted_projected_graph(B, [0, 2, 4, 5])
>>> list(G)
[0, 2, 4, 5]
>>> for edge in sorted(G.edges(data=True)):
... print(edge)
...
(0, 2, {'weight': 0.5})
(0, 5, {'weight': 0.5})
(2, 4, {'weight': 1.0})
(2, 5, {'weight': 0.5})

overlap_weighted_projected_graph(B, nodes, jaccard=True)

Overlap weighted projection of B onto one of its node sets.
The overlap weighted projection is the projection of the bipartite network B onto the specified nodes with weights
representing the Jaccard index between the neighborhoods of the two nodes in the original bipartite network [1]:
|N (u) ∩ N (v)|
wv,u =
|N (u) ∪ N (v)|
or if the parameter ‘jaccard’ is False, the fraction of common neighbors by minimum of both nodes degree in the
original bipartite graph [1]:
|N (u) ∩ N (v)|
wv,u =
min(|N (u)|, |N (v)|)
The nodes retain their attributes and are connected in the resulting graph if have an edge to a common node in the
original bipartite graph.
Parameters
B
nodes
jaccard: Bool (default=True)
Returns
Graph
See also:
is_bipartite
sets
projected_graph
Notes
3.4. Bipartite 231

References
[1]
Examples

>>> nodes = [0, 2, 4]
>>> G = bipartite.overlap_weighted_projected_graph(B, nodes)
>>> list(G)
[0, 2, 4]
[(0, 2, {'weight': 0.5}), (2, 4, {'weight': 0.5})]
>>> G = bipartite.overlap_weighted_projected_graph(B, nodes, jaccard=False)
[(0, 2, {'weight': 1.0}), (2, 4, {'weight': 1.0})]
generic_weighted_projected_graph(B, nodes, weight_function=None)

Weighted projection of B with a user-specified weight function.
The bipartite network B is projected on to the specified nodes with weights computed by a user-specified function.
This function must accept as a parameter the neighborhood sets of two nodes and return an integer or a float.
The nodes retain their attributes and are connected in the resulting graph if they have an edge to a common node
in the original graph.
Parameters
B
nodes
weight_function
[function] This function must accept as parameters the same input graph that this function,
and two nodes; and return an integer or a float. The default function computes the number of
shared neighbors.
Returns
Graph
See also:
is_bipartite
sets
projected_graph

Notes
Examples

>>> # Define some custom weight functions
>>> def jaccard(G, u, v):
... unbrs = set(G[u])
... vnbrs = set(G[v])
... return float(len(unbrs & vnbrs)) / len(unbrs | vnbrs)
...
>>> def my_weight(G, u, v, weight="weight"):
... w = 0
... for nbr in set(G[u]) & set(G[v]):
... w += G[u][nbr].get(weight, 1) + G[v][nbr].get(weight, 1)
... return w
...
>>> # A complete bipartite graph with 4 nodes and 4 edges
>>> B = nx.complete_bipartite_graph(2, 2)
>>> # Add some arbitrary weight to the edges
>>> for i, (u, v) in enumerate(B.edges()):
... B.edges[u, v]["weight"] = i + 1
...
>>> for edge in B.edges(data=True):
... print(edge)
...
(0, 2, {'weight': 1})
(0, 3, {'weight': 2})
(1, 2, {'weight': 3})
(1, 3, {'weight': 4})
>>> # By default, the weight is the number of shared neighbors
>>> G = bipartite.generic_weighted_projected_graph(B, [0, 1])
>>> print(list(G.edges(data=True)))
[(0, 1, {'weight': 2})]
>>> # To specify a custom weight function use the weight_function parameter
>>> G = bipartite.generic_weighted_projected_graph(
... B, [0, 1], weight_function=jaccard
... )
[(0, 1, {'weight': 1.0})]
>>> G = bipartite.generic_weighted_projected_graph(
... B, [0, 1], weight_function=my_weight
... )
[(0, 1, {'weight': 10})]
3.4. Bipartite 233

3.4.6 Spectral
Spectral bipartivity measure.
spectral_bipartivity(G[, nodes, weight]) Returns the spectral bipartivity.
spectral_bipartivity
spectral_bipartivity(G, nodes=None, weight='weight')

Returns the spectral bipartivity.
Parameters
G
[NetworkX graph]
nodes
[list or container optional(default is all nodes)] Nodes to return value of spectral bipartivity
contribution.
weight
[string or None optional (default = ‘weight’)] Edge data key to use for edge weights. If None,
weights set to 1.
Returns
sb
[float or dict] A single number if the keyword nodes is not specified, or a dictionary keyed by
node with the spectral bipartivity contribution of that node as the value.
See also:
color
Notes
This implementation uses Numpy (dense) matrices which are not efficient for storing large sparse graphs.
References
[1]
Examples

>>> bipartite.spectral_bipartivity(G)
1.0

3.4.7 Clustering
Functions for computing clustering of pairs
clustering(G[, nodes, mode]) Compute a bipartite clustering coefficient for nodes.

average_clustering(G[, nodes, mode]) Compute the average bipartite clustering coefficient.
latapy_clustering(G[, nodes, mode]) Compute a bipartite clustering coefficient for nodes.
robins_alexander_clustering(G) Compute the bipartite clustering of G.
clustering
clustering(G, nodes=None, mode='dot')

Compute a bipartite clustering coefficient for nodes.
The bipartie clustering coefficient is a measure of local density of connections defined as [1]:
∑
v∈N (N (u)) cuv
cu =
|N (N (u))|
where N(N(u)) are the second order neighbors of u in G excluding u, and c_{uv} is the pairwise clustering
coefficient between nodes u and v.
The mode selects the function for c_{uv} which can be:
dot:
|N (u) ∩ N (v)|
cuv =
|N (u) ∪ N (v)|
min:
|N (u) ∩ N (v)|
cuv =
min(|N (u)|, |N (v)|)
max:
|N (u) ∩ N (v)|
cuv =
max(|N (u)|, |N (v)|)
Parameters
G
[graph] A bipartite graph
nodes
[list or iterable (optional)] Compute bipartite clustering for these nodes. The default is all
nodes in G.
mode
[string] The pariwise bipartite clustering method to be used in the computation. It must be
“dot”, “max”, or “min”.
Returns
clustering
[dictionary] A dictionary keyed by node with the clustering coefficient value.
See also:
robins_alexander_clustering
average_clustering
networkx.algorithms.cluster.square_clustering
3.4. Bipartite 235

References
[1]
Examples

>>> G = nx.path_graph(4) # path graphs are bipartite
>>> c = bipartite.clustering(G)
>>> c[0]
0.5
>>> c = bipartite.clustering(G, mode="min")
>>> c[0]
1.0
average_clustering
average_clustering(G, nodes=None, mode='dot')

Compute the average bipartite clustering coefficient.
A clustering coefficient for the whole graph is the average,
1 ∑
C= cv ,
n
v∈G
where n is the number of nodes in G.

Similar measures for the two bipartite sets can be defined [1]
1 ∑
CX = cv ,
|X|
v∈X
where X is a bipartite set of G.

Parameters
G
[graph] a bipartite graph
nodes
[list or iterable, optional] A container of nodes to use in computing the average. The nodes
should be either the entire graph (the default) or one of the bipartite sets.
mode
[string] The pariwise bipartite clustering method. It must be “dot”, “max”, or “min”
Returns
clustering
[float] The average bipartite clustering for the given set of nodes or the entire graph if no nodes
are specified.
See also:
clustering

Notes
The container of nodes passed to this function must contain all of the nodes in one of the bipartite sets (“top” or
“bottom”) in order to compute the correct average bipartite clustering coefficients. See bipartite documen-
tation for further details on how bipartite graphs are handled in NetworkX.
References
[1]
Examples

>>> G = nx.star_graph(3) # star graphs are bipartite
>>> bipartite.average_clustering(G)
0.75
>>> X, Y = bipartite.sets(G)
>>> bipartite.average_clustering(G, X)
0.0
>>> bipartite.average_clustering(G, Y)
1.0
latapy_clustering
latapy_clustering(G, nodes=None, mode='dot')

Compute a bipartite clustering coefficient for nodes.
The bipartie clustering coefficient is a measure of local density of connections defined as [1]:
∑
v∈N (N (u)) cuv
cu =
|N (N (u))|
where N(N(u)) are the second order neighbors of u in G excluding u, and c_{uv} is the pairwise clustering
coefficient between nodes u and v.
The mode selects the function for c_{uv} which can be:
dot:
|N (u) ∩ N (v)|
cuv =
|N (u) ∪ N (v)|
min:
|N (u) ∩ N (v)|
cuv =
min(|N (u)|, |N (v)|)
max:
|N (u) ∩ N (v)|
cuv =
max(|N (u)|, |N (v)|)
Parameters
G
3.4. Bipartite 237

nodes
[list or iterable (optional)] Compute bipartite clustering for these nodes. The default is all
nodes in G.
mode
[string] The pariwise bipartite clustering method to be used in the computation. It must be
“dot”, “max”, or “min”.
Returns
clustering
[dictionary] A dictionary keyed by node with the clustering coefficient value.
See also:
average_clustering
References
[1]
Examples

>>> G = nx.path_graph(4) # path graphs are bipartite
>>> c = bipartite.clustering(G)
>>> c[0]
0.5
>>> c = bipartite.clustering(G, mode="min")
>>> c[0]
1.0
robins_alexander_clustering(G)
Compute the bipartite clustering of G.
Robins and Alexander [1] defined bipartite clustering coefficient as four times the number of four cycles C_4
divided by the number of three paths L_3 in a bipartite graph:
4 ∗ C4
CC4 =
L3
Parameters
G
[graph] a bipartite graph
Returns
clustering
[float] The Robins and Alexander bipartite clustering for the input graph.
See also:

latapy_clustering
References
[1]
Examples

>>> G = nx.davis_southern_women_graph()
>>> print(round(bipartite.robins_alexander_clustering(G), 3))
0.468
3.4.8 Redundancy
Node redundancy for bipartite graphs.
node_redundancy(G[, nodes]) Computes the node redundancy coefficients for the nodes
in the bipartite graph G.
node_redundancy
node_redundancy(G, nodes=None)
Computes the node redundancy coefficients for the nodes in the bipartite graph G.
The redundancy coefficient of a node v is the fraction of pairs of neighbors of v that are both linked to other nodes.
In a one-mode projection these nodes would be linked together even if v were not there.
More formally, for any vertex v, the redundancy coefficient of `v` is defined by
|{{u, w} ⊆ N (v), ∃v ′ ̸= v, (v ′ , u) ∈ E and (v ′ , w) ∈ E}|
rc(v) = |N (v)|(|N (v)|−1)
,
2
where N(v) is the set of neighbors of v in G.

Parameters
G
nodes
[list or iterable (optional)] Compute redundancy for these nodes. The default is all nodes in G.
Returns
redundancy
[dictionary] A dictionary keyed by node with the node redundancy value.
Raises
NetworkXError
If any of the nodes in the graph (or in nodes, if specified) has (out-)degree less than two
(which would result in division by zero, according to the definition of the redundancy coeffi-
cient).
3.4. Bipartite 239

References
[1]
Examples
Compute the redundancy coefficient of each node in a graph:

>>> rc = bipartite.node_redundancy(G)
>>> rc[0]
1.0
Compute the average redundancy for the graph:

>>> sum(rc.values()) / len(G)
1.0
Compute the average redundancy for a set of nodes:

>>> nodes = [0, 2]
>>> sum(rc[n] for n in nodes) / len(nodes)
1.0
3.4.9 Centrality
closeness_centrality(G, nodes[, normalized]) Compute the closeness centrality for nodes in a bipartite
network.
degree_centrality(G, nodes) Compute the degree centrality for nodes in a bipartite net-
work.
betweenness_centrality(G, nodes) Compute betweenness centrality for nodes in a bipartite
network.
closeness_centrality
closeness_centrality(G, nodes, normalized=True)

Compute the closeness centrality for nodes in a bipartite network.
The closeness of a node is the distance to all other nodes in the graph or in the case that the graph is not connected
to all other nodes in the connected component containing that node.
Parameters
G
[graph] A bipartite network

nodes
[list or container] Container with all nodes in one bipartite node set.
normalized
[bool, optional] If True (default) normalize by connected component size.
Returns
closeness
[dictionary] Dictionary keyed by node with bipartite closeness centrality as the value.
See also:
betweenness_centrality
degree_centrality
sets()
is_bipartite()
Notes
The nodes input parameter must contain all nodes in one bipartite node set, but the dictionary returned contains all
nodes from both node sets. See bipartite documentation for further details on how bipartite graphs are
handled in NetworkX.
Closeness centrality is normalized by the minimum distance possible. In the bipartite case the minimum distance
for a node in one bipartite node set is 1 from all nodes in the other node set and 2 from all other nodes in its own
set [1]. Thus the closeness centrality for node v in the two bipartite sets U with n nodes and V with m nodes is
m + 2(n − 1)
cv = , forv ∈ U,
d
n + 2(m − 1)
cv = , forv ∈ V,
d
where d is the sum of the distances from v to all other nodes.
Higher values of closeness indicate higher centrality.
As in the unipartite case, setting normalized=True causes the values to normalized further to n-1 / size(G)-1 where n
is the number of nodes in the connected part of graph containing the node. If the graph is not completely connected,
this algorithm computes the closeness centrality for each connected part separately.
References
[1]
degree_centrality
degree_centrality(G, nodes)
Compute the degree centrality for nodes in a bipartite network.
The degree centrality for a node v is the fraction of nodes connected to it.
Parameters
G
[graph] A bipartite network
3.4. Bipartite 241

nodes
Returns
centrality
[dictionary] Dictionary keyed by node with bipartite degree centrality as the value.
See also:
sets()
is_bipartite()
Notes
nodes from both bipartite node sets. See bipartite documentation for further details on how bipartite
graphs are handled in NetworkX.
For unipartite networks, the degree centrality values are normalized by dividing by the maximum possible degree
(which is n-1 where n is the number of nodes in G).
In the bipartite case, the maximum possible degree of a node in a bipartite node set is the number of nodes in the
opposite node set [1]. The degree centrality for a node v in the bipartite sets U with n nodes and V with m nodes is
deg(v)
dv = , forv ∈ U,
m
deg(v)
dv = , forv ∈ V,
n
where deg(v) is the degree of node v.
References
[1]
betweenness_centrality(G, nodes)
Compute betweenness centrality for nodes in a bipartite network.
Betweenness centrality of a node v is the sum of the fraction of all-pairs shortest paths that pass through v.
Values of betweenness are normalized by the maximum possible value which for bipartite graphs is limited by the
relative size of the two node sets [1].
Let n be the number of nodes in the node set U and m be the number of nodes in the node set V, then nodes in U
are normalized by dividing by
1 2
[m (s + 1)2 + m(s + 1)(2t − s − 1) − t(2s − t + 3)],
2
where
s = (n − 1) ÷ m, t = (n − 1) mod m,

and nodes in V are normalized by dividing by

1 2
[n (p + 1)2 + n(p + 1)(2r − p − 1) − r(2p − r + 3)],
2
where,
p = (m − 1) ÷ n, r = (m − 1) mod n.
Parameters
G
nodes
Returns
betweenness
[dictionary] Dictionary keyed by node with bipartite betweenness centrality as the value.
See also:
degree_centrality
sets()
is_bipartite()
Notes
nodes from both node sets. See bipartite documentation for further details on how bipartite graphs are
handled in NetworkX.
References
[1]
3.4.10 Generators
Generators and functions for bipartite graphs.
3.4. Bipartite 243

complete_bipartite_graph(n1, n2[, cre- Returns the complete bipartite graph K_{n_1,n_2}.

ate_using])
configuration_model(aseq, bseq[, ...]) Returns a random bipartite graph from two given degree
sequences.
havel_hakimi_graph(aseq, bseq[, create_using]) Returns a bipartite graph from two given degree se-
quences using a Havel-Hakimi style construction.
reverse_havel_hakimi_graph(aseq, bseq[, ...]) Returns a bipartite graph from two given degree se-
quences using a Havel-Hakimi style construction.
alternating_havel_hakimi_graph(aseq, Returns a bipartite graph from two given degree se-
bseq[, ...]) quences using an alternating Havel-Hakimi style con-
struction.
preferential_attachment_graph(aseq, p[, Create a bipartite graph with a preferential attachment
...]) model from a given single degree sequence.
random_graph(n, m, p[, seed, directed]) Returns a bipartite random graph.
gnmk_random_graph(n, m, k[, seed, directed]) Returns a random bipartite graph G_{n,m,k}.
complete_bipartite_graph
complete_bipartite_graph(n1, n2, create_using=None)

Returns the complete bipartite graph K_{n_1,n_2}.
The graph is composed of two partitions with nodes 0 to (n1 - 1) in the first and nodes n1 to (n1 + n2 - 1) in the
second. Each node in the first is connected to each node in the second.
Parameters
n1, n2
[integer or iterable container of nodes] If integers, nodes are from range(n1) and
range(n1, n1 + n2). If a container, the elements are the nodes.
create_using
[NetworkX graph instance, (default: nx.Graph)] Return graph of this type.
Notes
Nodes are the integers 0 to n1 + n2 - 1 unless either n1 or n2 are containers of nodes. If only one of n1 or n2
are integers, that integer is replaced by range of that integer.
The nodes are assigned the attribute ‘bipartite’ with the value 0 or 1 to indicate which bipartite set the node belongs
to.
This function is not imported in the main namespace. To use it use nx.bipartite.complete_bipartite_graph
configuration_model
configuration_model(aseq, bseq, create_using=None, seed=None)

Returns a random bipartite graph from two given degree sequences.
Parameters
aseq
[list] Degree sequence for node set A.

bseq
[list] Degree sequence for node set B.
create_using
[NetworkX graph instance, optional] Return graph of this type.
seed
Randomness.
The graph is composed of two partitions. Set A has nodes 0 to
(len(aseq) - 1) and set B has nodes len(aseq) to (len(bseq) - 1).
Nodes from set A are connected to nodes in set B by choosing
randomly from the possible free stubs, one in A and one in B.
Notes
The sum of the two sequences must be equal: sum(aseq)=sum(bseq) If no graph type is specified use MultiGraph
with parallel edges. If you want a graph with no parallel edges use create_using=Graph() but then the resulting
degree sequences might not be exact.
to.
This function is not imported in the main namespace. To use it use nx.bipartite.configuration_model
havel_hakimi_graph
havel_hakimi_graph(aseq, bseq, create_using=None)

Returns a bipartite graph from two given degree sequences using a Havel-Hakimi style construction.
The graph is composed of two partitions. Set A has nodes 0 to (len(aseq) - 1) and set B has nodes len(aseq) to
(len(bseq) - 1). Nodes from the set A are connected to nodes in the set B by connecting the highest degree nodes
in set A to the highest degree nodes in set B until all stubs are connected.
Parameters
aseq
bseq
create_using
Notes
to.
This function is not imported in the main namespace. To use it use nx.bipartite.havel_hakimi_graph
3.4. Bipartite 245

reverse_havel_hakimi_graph
reverse_havel_hakimi_graph(aseq, bseq, create_using=None)

Returns a bipartite graph from two given degree sequences using a Havel-Hakimi style construction.
(len(bseq) - 1). Nodes from set A are connected to nodes in the set B by connecting the highest degree nodes in
set A to the lowest degree nodes in set B until all stubs are connected.
Parameters
aseq
bseq
create_using
Notes
to.
This function is not imported in the main namespace. To use it use nx.bipartite.reverse_havel_hakimi_graph
alternating_havel_hakimi_graph
alternating_havel_hakimi_graph(aseq, bseq, create_using=None)

Returns a bipartite graph from two given degree sequences using an alternating Havel-Hakimi style construction.
(len(bseq) - 1). Nodes from the set A are connected to nodes in the set B by connecting the highest degree nodes
in set A to alternatively the highest and the lowest degree nodes in set B until all stubs are connected.
Parameters
aseq
bseq
create_using

Notes
to.
This function is not imported in the main namespace. To use it use nx.bipartite.alternating_havel_hakimi_graph
preferential_attachment_graph
preferential_attachment_graph(aseq, p, create_using=None, seed=None)

Create a bipartite graph with a preferential attachment model from a given single degree sequence.
The graph is composed of two partitions. Set A has nodes 0 to (len(aseq) - 1) and set B has nodes starting with
node len(aseq). The number of nodes in set B is random.
Parameters
aseq
p
[float] Probability that a new bottom node is added.
create_using
seed
Randomness.
Notes
to.
This function is not imported in the main namespace. To use it use nx.bipartite.preferential_attachment_graph
References
[1], [2]
random_graph
random_graph(n, m, p, seed=None, directed=False)

Returns a bipartite random graph.
This is a bipartite version of the binomial (Erdős-Rényi) graph. The graph is composed of two partitions. Set A
has nodes 0 to (n - 1) and set B has nodes n to (n + m - 1).
Parameters
3.4. Bipartite 247

n
[int] The number of nodes in the first bipartite set.
m
[int] The number of nodes in the second bipartite set.
p
[float] Probability for edge creation.
seed
Randomness.
directed
[bool, optional (default=False)] If True return a directed graph
See also:
gnp_random_graph, configuration_model
Notes
The bipartite random graph algorithm chooses each of the n*m (undirected) or 2*nm (directed) possible edges
with probability p.
This algorithm is O(n + m) where m is the expected number of edges.
to.
This function is not imported in the main namespace. To use it use nx.bipartite.random_graph
References
[1]
gnmk_random_graph
gnmk_random_graph(n, m, k, seed=None, directed=False)

Returns a random bipartite graph G_{n,m,k}.
Produces a bipartite graph chosen randomly out of the set of all graphs with n top nodes, m bottom nodes, and k
edges. The graph is composed of two sets of nodes. Set A has nodes 0 to (n - 1) and set B has nodes n to (n + m -
1).
Parameters
n
[int] The number of nodes in the first bipartite set.
m
[int] The number of nodes in the second bipartite set.
k
[int] The number of edges

seed
Randomness.
directed
See also:
gnm_random_graph
Notes
If k > m * n then a complete bipartite graph is returned.

This graph is a bipartite version of the G_{nm} random graph model.
to.
This function is not imported in the main namespace. To use it use nx.bipartite.gnmk_random_graph
Examples
from nx.algorithms import bipartite G = bipartite.gnmk_random_graph(10,20,50)
3.4.11 Covering
Functions related to graph covers.
min_edge_cover(G[, matching_algorithm]) Returns a set of edges which constitutes the minimum

edge cover of the graph.
min_edge_cover
min_edge_cover(G, matching_algorithm=None)
Returns a set of edges which constitutes the minimum edge cover of the graph.
The smallest edge cover can be found in polynomial time by finding a maximum matching and extending it greedily
so that all nodes are covered.
Parameters
G
[NetworkX graph] An undirected bipartite graph.
matching_algorithm
[function] A function that returns a maximum cardinality matching in a given bipartite graph.
The function must take one input, the graph G, and return a dictionary mapping each node to its
mate. If not specified, hopcroft_karp_matching() will be used. Other possibilities
include eppstein_matching(),
Returns
3.4. Bipartite 249

set
A set of the edges in a minimum edge cover of the graph, given as pairs of nodes. It contains
both the edges (u, v) and (v, u) for given nodes u and v among the edges of minimum
edge cover.
Notes
An edge cover of a graph is a set of edges such that every node of the graph is incident to at least one edge of the
set. A minimum edge cover is an edge covering of smallest cardinality.
Due to its implementation, the worst-case running time of this algorithm is bounded by the worst-case running time
of the function matching_algorithm.
3.5 Boundary
Routines to find the boundary of a set of nodes.

An edge boundary is a set of edges, each of which has exactly one endpoint in a given set of nodes (or, in the case of
directed graphs, the set of edges whose source node is in the set).
A node boundary of a set S of nodes is the set of (out-)neighbors of nodes in S that are outside S.
edge_boundary(G, nbunch1[, nbunch2, data, ...]) Returns the edge boundary of nbunch1.
node_boundary(G, nbunch1[, nbunch2]) Returns the node boundary of nbunch1.
3.5.1 edge_boundary
edge_boundary(G, nbunch1, nbunch2=None, data=False, keys=False, default=None)

Returns the edge boundary of nbunch1.
The edge boundary of a set S with respect to a set T is the set of edges (u, v) such that u is in S and v is in T. If T
is not specified, it is assumed to be the set of all nodes not in S.
Parameters
G
[NetworkX graph]
nbunch1
[iterable] Iterable of nodes in the graph representing the set of nodes whose edge boundary
will be returned. (This is the set S from the definition above.)
nbunch2
[iterable] Iterable of nodes representing the target (or “exterior”) set of nodes. (This is the set
T from the definition above.) If not specified, this is assumed to be the set of all nodes in G
not in nbunch1.
keys
[bool] This parameter has the same meaning as in MultiGraph.edges().
data
[bool or object] This parameter has the same meaning as in MultiGraph.edges().
default
[object] This parameter has the same meaning as in MultiGraph.edges().

Returns
iterator
An iterator over the edges in the boundary of nbunch1 with respect to nbunch2. If keys,
data, or default are specified and G is a multigraph, then edges are returned with keys
and/or data, as in MultiGraph.edges().
Notes
Any element of nbunch that is not in the graph G will be ignored.

nbunch1 and nbunch2 are usually meant to be disjoint, but in the interest of speed and generality, that is not
required here.
3.5.2 node_boundary
node_boundary(G, nbunch1, nbunch2=None)

Returns the node boundary of nbunch1.
The node boundary of a set S with respect to a set T is the set of nodes v in T such that for some u in S, there is an
edge joining u to v. If T is not specified, it is assumed to be the set of all nodes not in S.
Parameters
G
[NetworkX graph]
nbunch1
[iterable] Iterable of nodes in the graph representing the set of nodes whose node boundary
will be returned. (This is the set S from the definition above.)
nbunch2
[iterable] Iterable of nodes representing the target (or “exterior”) set of nodes. (This is the set
T from the definition above.) If not specified, this is assumed to be the set of all nodes in G
not in nbunch1.
Returns
set
The node boundary of nbunch1 with respect to nbunch2.
Notes
Any element of nbunch that is not in the graph G will be ignored.

nbunch1 and nbunch2 are usually meant to be disjoint, but in the interest of speed and generality, that is not
required here.
3.5. Boundary 251

3.6 Bridges
Bridge-finding algorithms.
bridges(G[, root]) Generate all bridges in a graph.

has_bridges(G[, root]) Decide whether a graph has any bridges.
local_bridges(G[, with_span, weight]) Iterate over local bridges of G optionally computing the
span
3.6.1 bridges
bridges(G, root=None)
Generate all bridges in a graph.
A bridge in a graph is an edge whose removal causes the number of connected components of the graph to increase.
Equivalently, a bridge is an edge that does not belong to any cycle. Bridges are also known as cut-edges, isthmuses,
or cut arcs.
Parameters
G
[undirected graph]
root
[node (optional)] A node in the graph G. If specified, only the bridges in the connected com-
ponent containing this node will be returned.
Yields
e
[edge] An edge in the graph whose removal disconnects the graph (or causes the number of
connected components to increase).
Raises
NodeNotFound
If root is not in the graph G.
If G is a directed graph.
Notes
This is an implementation of the algorithm described in [1]. An edge is a bridge if and only if it is not contained
in any chain. Chains are found using the networkx.chain_decomposition() function.
The algorithm described in [1] requires a simple graph. If the provided graph is a multigraph, we convert it to a
simple graph and verify that any bridges discovered by the chain decomposition algorithm are not multi-edges.
Ignoring polylogarithmic factors, the worst-case time complexity is the same as the networkx.
chain_decomposition() function, O(m + n), where n is the number of nodes in the graph and m is
the number of edges.

References
[1]
Examples
The barbell graph with parameter zero has a single bridge:
>>> G = nx.barbell_graph(10, 0)
>>> list(nx.bridges(G))
[(9, 10)]
3.6.2 has_bridges
has_bridges(G, root=None)
Decide whether a graph has any bridges.
A bridge in a graph is an edge whose removal causes the number of connected components of the graph to increase.
Parameters
G
[undirected graph]
root
[node (optional)] A node in the graph G. If specified, only the bridges in the connected com-
ponent containing this node will be considered.
Returns
bool
Whether the graph (or the connected component containing root) has any bridges.
Raises
NodeNotFound
If G is a directed graph.
Notes
This implementation uses the networkx.bridges() function, so it shares its worst-case time complexity,
O(m + n), ignoring polylogarithmic factors, where n is the number of nodes in the graph and m is the number of
edges.
3.6. Bridges 253

Examples
The barbell graph with parameter zero has a single bridge:
>>> nx.has_bridges(G)
True
On the other hand, the cycle graph has no bridges:
>>> nx.has_bridges(G)
False
3.6.3 local_bridges
local_bridges(G, with_span=True, weight=None)

Iterate over local bridges of G optionally computing the span
A local bridge is an edge whose endpoints have no common neighbors. That is, the edge is not part of a triangle in
the graph.
The span of a local bridge is the shortest path length between the endpoints if the local bridge is removed.
Parameters
G
[undirected graph]
with_span
[bool] If True, yield a 3-tuple (u, v, span)
weight
[function, string or None (default: None)] If function, used to compute edge weights for the
span. If string, the edge data attribute used in calculating span. If None, all edges have weight
1.
Yields
e
[edge] The local bridges as an edge 2-tuple of nodes (u, v) or as a 3-tuple (u, v, span)
when with_span is True.
Raises
If G is a directed graph or multigraph.

Examples
A cycle graph has every edge a local bridge with span N-1.
>>> (0, 8, 8) in set(nx.local_bridges(G))
True
3.7 Centrality
3.7.1 Degree
degree_centrality(G) Compute the degree centrality for nodes.

in_degree_centrality(G) Compute the in-degree centrality for nodes.
out_degree_centrality(G) Compute the out-degree centrality for nodes.
degree_centrality
degree_centrality(G)
Compute the degree centrality for nodes.
The degree centrality for a node v is the fraction of nodes it is connected to.
Parameters
G
[graph] A networkx graph
Returns
nodes
[dictionary] Dictionary of nodes with degree centrality as the value.
See also:
betweenness_centrality, load_centrality, eigenvector_centrality
Notes
The degree centrality values are normalized by dividing by the maximum possible degree in a simple graph n-1
For multigraphs or graphs with self loops the maximum degree might be higher than n-1 and values of degree
centrality greater than 1 are possible.
3.7. Centrality 255

Examples
>>> G = nx.Graph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3)])
>>> nx.degree_centrality(G)
{0: 1.0, 1: 1.0, 2: 0.6666666666666666, 3: 0.6666666666666666}
in_degree_centrality
in_degree_centrality(G)
Compute the in-degree centrality for nodes.
The in-degree centrality for a node v is the fraction of nodes its incoming edges are connected to.
Parameters
G
Returns
nodes
[dictionary] Dictionary of nodes with in-degree centrality as values.
Raises
If G is undirected.
See also:
degree_centrality, out_degree_centrality
Notes
Examples
>>> G = nx.DiGraph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3)])
>>> nx.in_degree_centrality(G)
{0: 0.0, 1: 0.3333333333333333, 2: 0.6666666666666666, 3: 0.6666666666666666}

out_degree_centrality
out_degree_centrality(G)
Compute the out-degree centrality for nodes.
The out-degree centrality for a node v is the fraction of nodes its outgoing edges are connected to.
Parameters
G
Returns
nodes
[dictionary] Dictionary of nodes with out-degree centrality as values.
Raises
If G is undirected.
See also:
degree_centrality, in_degree_centrality
Notes
Examples
>>> G = nx.DiGraph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3)])
>>> nx.out_degree_centrality(G)
{0: 1.0, 1: 0.6666666666666666, 2: 0.0, 3: 0.0}
3.7.2 Eigenvector
eigenvector_centrality(G[, max_iter, tol, ...]) Compute the eigenvector centrality for the graph G.
eigenvector_centrality_numpy(G[, weight, Compute the eigenvector centrality for the graph G.
...])
katz_centrality(G[, alpha, beta, max_iter, ...]) Compute the Katz centrality for the nodes of the graph G.
katz_centrality_numpy(G[, alpha, beta, ...]) Compute the Katz centrality for the graph G.
3.7. Centrality 257

eigenvector_centrality
eigenvector_centrality(G, max_iter=100, tol=1e-06, nstart=None, weight=None)

Compute the eigenvector centrality for the graph G.
Eigenvector centrality computes the centrality for a node based on the centrality of its neighbors. The eigenvector
centrality for node i is the i-th element of the vector x defined by the equation
Ax = λx
where A is the adjacency matrix of the graph G with eigenvalue λ. By virtue of the Perron–Frobenius theorem,
there is a unique solution x, all of whose entries are positive, if λ is the largest eigenvalue of the adjacency matrix
A ([2]).
Parameters
G
max_iter
[integer, optional (default=100)] Maximum number of iterations in power method.
tol
[float, optional (default=1.0e-6)] Error tolerance used to check convergence in power method
iteration.
nstart
[dictionary, optional (default=None)] Starting value of eigenvector iteration for each node.
weight
[None or string, optional (default=None)] If None, all edge weights are considered equal. Oth-
erwise holds the name of the edge attribute used as weight. In this measure the weight is
interpreted as the connection strength.
Returns
nodes
[dictionary] Dictionary of nodes with eigenvector centrality as the value.
Raises
NetworkXPointlessConcept
If the graph G is the null graph.
NetworkXError
If each value in nstart is zero.
PowerIterationFailedConvergence
If the algorithm fails to converge to the specified tolerance within the specified number of
iterations of the power iteration method.
See also:
eigenvector_centrality_numpy
pagerank
hits

Notes
The measure was introduced by [1] and is discussed in [2].

The power iteration method is used to compute the eigenvector and convergence is not guaranteed. Our method
stops after max_iter iterations or when the change in the computed vector between two iterations is smaller than
an error tolerance of G.number_of_nodes() * tol. This implementation uses (A + I) rather than the
adjacency matrix A because it shifts the spectrum to enable discerning the correct eigenvector even for networks
with multiple dominant eigenvalues.
For directed graphs this is “left” eigenvector centrality which corresponds to the in-edges in the graph. For out-edges
eigenvector centrality first reverse the graph with G.reverse().
References
[1], [2]
Examples
>>> centrality = nx.eigenvector_centrality(G)
>>> sorted((v, f"{c:0.2f}") for v, c in centrality.items())
[(0, '0.37'), (1, '0.60'), (2, '0.60'), (3, '0.37')]
eigenvector_centrality_numpy(G, weight=None, max_iter=50, tol=0)

Compute the eigenvector centrality for the graph G.
Eigenvector centrality computes the centrality for a node based on the centrality of its neighbors. The eigenvector
centrality for node i is
Ax = λx
where A is the adjacency matrix of the graph G with eigenvalue λ. By virtue of the Perron–Frobenius theorem,
there is a unique and positive solution if λ is the largest eigenvalue associated with the eigenvector of the adjacency
matrix A ([2]).
Parameters
G
weight
[None or string, optional (default=None)] The name of the edge attribute used as weight. If
None, all edge weights are considered equal. In this measure the weight is interpreted as the
connection strength.
max_iter
tol
[float, optional (default=1.0e-6)] Relative accuracy for eigenvalues (stopping criterion). The
default value of 0 implies machine precision.
Returns
3.7. Centrality 259

nodes
[dictionary] Dictionary of nodes with eigenvector centrality as the value.
Raises
If the graph G is the null graph.
See also:
pagerank
hits
Notes
The measure was introduced by [1].

This algorithm uses the SciPy sparse eigenvalue solver (ARPACK) to find the largest eigenvalue/eigenvector pair.
For directed graphs this is “left” eigenvector centrality which corresponds to the in-edges in the graph. For out-edges
eigenvector centrality first reverse the graph with G.reverse().
References
[1], [2]
Examples
>>> centrality = nx.eigenvector_centrality_numpy(G)
>>> print([f"{node} {centrality[node]:0.2f}" for node in centrality])
['0 0.37', '1 0.60', '2 0.60', '3 0.37']
katz_centrality
katz_centrality(G, alpha=0.1, beta=1.0, max_iter=1000, tol=1e-06, nstart=None, normalized=True,

weight=None)
Compute the Katz centrality for the nodes of the graph G.
Katz centrality computes the centrality for a node based on the centrality of its neighbors. It is a generalization of
the eigenvector centrality. The Katz centrality for node i is
∑
xi = α Aij xj + β,
j
where A is the adjacency matrix of graph G with eigenvalues λ.

The parameter β controls the initial centrality and
1
α< .
λmax

Katz centrality computes the relative influence of a node within a network by measuring the number of the im-
mediate neighbors (first degree nodes) and also all other nodes in the network that connect to the node under
consideration through these immediate neighbors.
Extra weight can be provided to immediate neighbors through the parameter β. Connections made with distant
neighbors are, however, penalized by an attenuation factor α which should be strictly less than the inverse largest
eigenvalue of the adjacency matrix in order for the Katz centrality to be computed correctly. More information is
provided in [1].
Parameters
G
[graph] A NetworkX graph.
alpha
[float] Attenuation factor
beta
[scalar or dictionary, optional (default=1.0)] Weight attributed to the immediate neighborhood.
If not a scalar, the dictionary must have an value for every node.
max_iter
tol
[float, optional (default=1.0e-6)] Error tolerance used to check convergence in power method
iteration.
nstart
[dictionary, optional] Starting value of Katz iteration for each node.
normalized
[bool, optional (default=True)] If True normalize the resulting values.
weight
erwise holds the name of the edge attribute used as weight. In this measure the weight is
interpreted as the connection strength.
Returns
nodes
[dictionary] Dictionary of nodes with Katz centrality as the value.
Raises
NetworkXError
If the parameter beta is not a scalar but lacks a value for at least one node
See also:
katz_centrality_numpy
pagerank
hits
3.7. Centrality 261

Notes
Katz centrality was introduced by [2].

This algorithm it uses the power method to find the eigenvector corresponding to the largest eigenvalue of the
adjacency matrix of G. The parameter alpha should be strictly less than the inverse of largest eigenvalue of the
adjacency matrix for the algorithm to converge. You can use max(nx.adjacency_spectrum(G)) to get
λmax the largest eigenvalue of the adjacency matrix. The iteration will stop after max_iter iterations or an error
tolerance of number_of_nodes(G) * tol has been reached.
When α = 1/λmax and β = 0, Katz centrality is the same as eigenvector centrality.
For directed graphs this finds “left” eigenvectors which corresponds to the in-edges in the graph. For out-edges
Katz centrality first reverse the graph with G.reverse().
References
[1], [2]
Examples
>>> import math

>>> phi = (1 + math.sqrt(5)) / 2.0 # largest eigenvalue of adj matrix
>>> centrality = nx.katz_centrality(G, 1 / phi - 0.01)
>>> for n, c in sorted(centrality.items()):
... print(f"{n} {c:.2f}")
0 0.37
1 0.60
2 0.60
3 0.37
katz_centrality_numpy
katz_centrality_numpy(G, alpha=0.1, beta=1.0, normalized=True, weight=None)

Compute the Katz centrality for the graph G.
Katz centrality computes the centrality for a node based on the centrality of its neighbors. It is a generalization of
the eigenvector centrality. The Katz centrality for node i is
∑
xi = α Aij xj + β,
j
where A is the adjacency matrix of graph G with eigenvalues λ.

The parameter β controls the initial centrality and
1
α< .
λmax
Katz centrality computes the relative influence of a node within a network by measuring the number of the im-
mediate neighbors (first degree nodes) and also all other nodes in the network that connect to the node under
consideration through these immediate neighbors.
Extra weight can be provided to immediate neighbors through the parameter β. Connections made with distant
neighbors are, however, penalized by an attenuation factor α which should be strictly less than the inverse largest

eigenvalue of the adjacency matrix in order for the Katz centrality to be computed correctly. More information is
provided in [1].
Parameters
G
alpha
[float] Attenuation factor
beta
[scalar or dictionary, optional (default=1.0)] Weight attributed to the immediate neighborhood.
If not a scalar the dictionary must have an value for every node.
normalized
[bool] If True normalize the resulting values.
weight
[None or string, optional] If None, all edge weights are considered equal. Otherwise holds the
name of the edge attribute used as weight. In this measure the weight is interpreted as the
connection strength.
Returns
nodes
[dictionary] Dictionary of nodes with Katz centrality as the value.
Raises
NetworkXError
If the parameter beta is not a scalar but lacks a value for at least one node
See also:
katz_centrality
pagerank
hits
Notes
Katz centrality was introduced by [2].

This algorithm uses a direct linear solver to solve the above equation. The parameter alpha should be strictly less
than the inverse of largest eigenvalue of the adjacency matrix for there to be a solution. You can use max(nx.
adjacency_spectrum(G)) to get λmax the largest eigenvalue of the adjacency matrix.
When α = 1/λmax and β = 0, Katz centrality is the same as eigenvector centrality.
For directed graphs this finds “left” eigenvectors which corresponds to the in-edges in the graph. For out-edges
Katz centrality first reverse the graph with G.reverse().
3.7. Centrality 263

References
[1], [2]
Examples
>>> import math

>>> phi = (1 + math.sqrt(5)) / 2.0 # largest eigenvalue of adj matrix
>>> centrality = nx.katz_centrality_numpy(G, 1 / phi)
>>> for n, c in sorted(centrality.items()):
... print(f"{n} {c:.2f}")
0 0.37
1 0.60
2 0.60
3 0.37
3.7.3 Closeness
closeness_centrality(G[, u, distance, ...]) Compute closeness centrality for nodes.

incremental_closeness_centrality(G, Incremental closeness centrality for nodes.
edge[, ...])
closeness_centrality(G, u=None, distance=None, wf_improved=True)

Compute closeness centrality for nodes.
Closeness centrality [1] of a node u is the reciprocal of the average shortest path distance to u over all n-1 reachable
nodes.
n−1
C(u) = ∑n−1 ,
v=1 d(v, u)
where d(v, u) is the shortest-path distance between v and u, and n-1 is the number of nodes reachable from
u. Notice that the closeness distance function computes the incoming distance to u for directed graphs. To use
outward distance, act on G.reverse().
Notice that higher values of closeness indicate higher centrality.
Wasserman and Faust propose an improved formula for graphs with more than one connected component. The
result is “a ratio of the fraction of actors in the group who are reachable, to the average distance” from the reachable
actors [2]. You might think this scale factor is inverted but it is not. As is, nodes from small components receive a
smaller closeness value. Letting N denote the number of nodes in the graph,
n−1 n−1
CW F (u) = ∑ ,
N − 1 n−1
v=1 d(v, u)
Parameters
G

u
[node, optional] Return only the value for node u
distance
[edge attribute key, optional (default=None)] Use the specified edge attribute as the edge dis-
tance in shortest path calculations. If None (the default) all edges have a distance of 1. Absent
edge attributes are assigned a distance of 1. Note that no check is performed to ensure that
edges have the provided attribute.
wf_improved
[bool, optional (default=True)] If True, scale by the fraction of nodes reachable. This gives
the Wasserman and Faust improved formula. For single component graphs it is the same as
the original formula.
Returns
nodes
[dictionary] Dictionary of nodes with closeness centrality as the value.
See also:

degree_centrality, incremental_closeness_centrality
Notes
The closeness centrality is normalized to (n-1)/(|G|-1) where n is the number of nodes in the connected
part of graph containing the node. If the graph is not completely connected, this algorithm computes the closeness
centrality for each connected part separately scaled by that parts size.
If the ‘distance’ keyword is set to an edge attribute key then the shortest-path length will be computed using Dijkstra’s
algorithm with that edge attribute as the edge weight.
The closeness centrality uses inward distance to a node, not outward. If you want to use outword distances apply
the function to G.reverse()
In NetworkX 2.2 and earlier a bug caused Dijkstra’s algorithm to use the outward distance rather than the inward
distance. If you use a ‘distance’ keyword and a DiGraph, your results will change between v2.2 and v2.3.
References
[1], [2]
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3)])
>>> nx.closeness_centrality(G)
{0: 1.0, 1: 1.0, 2: 0.75, 3: 0.75}
3.7. Centrality 265

incremental_closeness_centrality
incremental_closeness_centrality(G, edge, prev_cc=None, insertion=True, wf_improved=True)

Incremental closeness centrality for nodes.
Compute closeness centrality for nodes using level-based work filtering as described in Incremental Algorithms for
Closeness Centrality by Sariyuce et al.
Level-based work filtering detects unnecessary updates to the closeness centrality and filters them out.
— From “Incremental Algorithms for Closeness Centrality”:
Theorem 1: Let G = (V, E) be a graph and u and v be two vertices in V such that there is no edge (u, v) in E. Let
G′ = (V, E ∪ uv) Then cc[s] = cc′ [s] if and only if |dG(s, u) − dG(s, v)| ≤ 1.
Where dG(u, v) denotes the length of the shortest path between two vertices u, v in a graph G, cc[s] is the closeness
centrality for a vertex s in V, and cc’[s] is the closeness centrality for a vertex s in V, with the (u, v) edge added. —
We use Theorem 1 to filter out updates when adding or removing an edge. When adding an edge (u, v), we compute
the shortest path lengths from all other nodes to u and to v before the node is added. When removing an edge, we
compute the shortest path lengths after the edge is removed. Then we apply Theorem 1 to use previously computed
closeness centrality for nodes where |dG(s, u) − dG(s, v)| ≤ 1. This works only for undirected, unweighted
graphs; the distance argument is not supported.
Closeness centrality [1] of a node u is the reciprocal of the sum of the shortest path distances from u to all n-1
other nodes. Since the sum of distances depends on the number of nodes in the graph, closeness is normalized by
the sum of minimum possible distances n-1.
n−1
C(u) = ∑n−1 ,
v=1 d(v, u)
where d(v, u) is the shortest-path distance between v and u, and n is the number of nodes in the graph.
Notice that higher values of closeness indicate higher centrality.
Parameters
G
edge
[tuple] The modified edge (u, v) in the graph.
prev_cc
[dictionary] The previous closeness centrality for all nodes in the graph.
insertion
[bool, optional] If True (default) the edge was inserted, otherwise it was deleted from the graph.
wf_improved
[bool, optional (default=True)] If True, scale by the fraction of nodes reachable. This gives
the Wasserman and Faust improved formula. For single component graphs it is the same as
the original formula.
Returns
nodes
[dictionary] Dictionary of nodes with closeness centrality as the value.
See also:

degree_centrality, closeness_centrality

Notes
The closeness centrality is normalized to (n-1)/(|G|-1) where n is the number of nodes in the connected
part of graph containing the node. If the graph is not completely connected, this algorithm computes the closeness
centrality for each connected part separately.
References
[1], [2]
3.7.4 Current Flow Closeness
current_flow_closeness_centrality(G[, Compute current-flow closeness centrality for nodes.

...])
information_centrality(G[, weight, dtype, ...]) Compute current-flow closeness centrality for nodes.
current_flow_closeness_centrality
current_flow_closeness_centrality(G, weight=None, dtype=<class 'float'>, solver='lu')

Compute current-flow closeness centrality for nodes.
Current-flow closeness centrality is variant of closeness centrality based on effective resistance between nodes in a
network. This metric is also known as information centrality.
Parameters
G
weight
erwise holds the name of the edge attribute used as weight. The weight reflects the capacity or
the strength of the edge.
dtype: data type (default=float)
Default data type for internal matrices. Set to np.float32 for lower memory consumption.
solver: string (default=’lu’)
Type of linear solver to use for computing the flow matrix. Options are “full” (uses most
memory), “lu” (recommended), and “cg” (uses least memory).
Returns
nodes
[dictionary] Dictionary of nodes with current flow closeness centrality as the value.
See also:
3.7. Centrality 267

Notes
The algorithm is from Brandes [1].

See also [2] for the original definition of information centrality.
References
[1], [2]
information_centrality
information_centrality(G, weight=None, dtype=<class 'float'>, solver='lu')

Compute current-flow closeness centrality for nodes.
Current-flow closeness centrality is variant of closeness centrality based on effective resistance between nodes in a
network. This metric is also known as information centrality.
Parameters
G
weight
erwise holds the name of the edge attribute used as weight. The weight reflects the capacity or
the strength of the edge.
dtype: data type (default=float)
Returns
nodes
[dictionary] Dictionary of nodes with current flow closeness centrality as the value.
See also:
Notes
The algorithm is from Brandes [1].

See also [2] for the original definition of information centrality.

References
[1], [2]
3.7.5 (Shortest Path) Betweenness
betweenness_centrality(G[, k, normalized, ...]) Compute the shortest-path betweenness centrality for

nodes.
betweenness_centrality_subset(G, sources, Compute betweenness centrality for a subset of nodes.
...)
edge_betweenness_centrality(G[, k, ...]) Compute betweenness centrality for edges.
edge_betweenness_centrality_subset(G, Compute betweenness centrality for edges for a subset of
...[, ...]) nodes.
betweenness_centrality(G, k=None, normalized=True, weight=None, endpoints=False, seed=None)

Compute the shortest-path betweenness centrality for nodes.
Betweenness centrality of a node v is the sum of the fraction of all-pairs shortest paths that pass through v
∑ σ(s, t|v)
cB (v) =
σ(s, t)
s,t∈V
where V is the set of nodes, σ(s, t) is the number of shortest (s, t)-paths, and σ(s, t|v) is the number of those
paths passing through some node v other than s, t. If s = t, σ(s, t) = 1, and if v ∈ s, t, σ(s, t|v) = 0 [2].
Parameters
G
k
[int, optional (default=None)] If k is not None use k node samples to estimate betweenness.
The value of k <= n where n is the number of nodes in the graph. Higher values give better
approximation.
normalized
[bool, optional] If True the betweenness values are normalized by 2/((n-1)(n-2)) for
graphs, and 1/((n-1)(n-2)) for directed graphs where n is the number of nodes in G.
weight
erwise holds the name of the edge attribute used as weight. Weights are used to calculate
weighted shortest paths, so they are interpreted as distances.
endpoints
[bool, optional] If True include the endpoints in the shortest path counts.
seed
Randomness. Note that this is only used if k is not None.
Returns
3.7. Centrality 269

nodes
[dictionary] Dictionary of nodes with betweenness centrality as the value.
See also:
edge_betweenness_centrality
load_centrality
Notes
The algorithm is from Ulrik Brandes [1]. See [4] for the original first published version and [2] for details on
algorithms for variations and related metrics.
For approximate betweenness calculations set k=#samples to use k nodes (“pivots”) to estimate the betweenness
values. For an estimate of the number of pivots needed see [3].
For weighted graphs the edge weights must be greater than zero. Zero edge weights can produce an infinite number
of equal length paths between pairs of nodes.
The total number of paths between source and target is counted differently for directed and undirected graphs.
Directed paths are easy to count. Undirected paths are tricky: should a path from “u” to “v” count as 1 undirected
path or as 2 directed paths?
For betweenness_centrality we report the number of undirected paths when G is undirected.
For betweenness_centrality_subset the reporting is different. If the source and target subsets are the same, then
we want to count undirected paths. But if the source and target subsets differ – for example, if sources is {0} and
targets is {1}, then we are only counting the paths in one direction. They are undirected paths but we are counting
them in a directed way. To count them as undirected paths, each should count as half a path.
References
[1], [2], [3], [4]
betweenness_centrality_subset
betweenness_centrality_subset(G, sources, targets, normalized=False, weight=None)

Compute betweenness centrality for a subset of nodes.
∑ σ(s, t|v)
cB (v) =
σ(s, t)
s∈S,t∈T
where S is the set of sources, T is the set of targets, σ(s, t) is the number of shortest (s, t)-paths, and σ(s, t|v)
is the number of those paths passing through some node v other than s, t. If s = t, σ(s, t) = 1, and if v ∈ s, t,
σ(s, t|v) = 0 [2].
Parameters
G
sources: list of nodes
Nodes to use as sources for shortest paths in betweenness
targets: list of nodes
Nodes to use as targets for shortest paths in betweenness

normalized
[bool, optional] If True the betweenness values are normalized by 2/((n − 1)(n − 2)) for
graphs, and 1/((n − 1)(n − 2)) for directed graphs where n is the number of nodes in G.
weight
Returns
nodes
See also:
load_centrality
Notes
The basic algorithm is from [1].

The normalization might seem a little strange but it is designed to make betweenness_centrality(G) be the same as
betweenness_centrality_subset(G,sources=G.nodes(),targets=G.nodes()).
Directed paths are easy to count. Undirected paths are tricky: should a path from “u” to “v” count as 1 undirected
path or as 2 directed paths?
For betweenness_centrality we report the number of undirected paths when G is undirected.
For betweenness_centrality_subset the reporting is different. If the source and target subsets are the same, then
we want to count undirected paths. But if the source and target subsets differ – for example, if sources is {0} and
targets is {1}, then we are only counting the paths in one direction. They are undirected paths but we are counting
them in a directed way. To count them as undirected paths, each should count as half a path.
References
[1], [2]
edge_betweenness_centrality(G, k=None, normalized=True, weight=None, seed=None)

Compute betweenness centrality for edges.
Betweenness centrality of an edge e is the sum of the fraction of all-pairs shortest paths that pass through e
∑ σ(s, t|e)
cB (e) =
σ(s, t)
s,t∈V
where V is the set of nodes, σ(s, t) is the number of shortest (s, t)-paths, and σ(s, t|e) is the number of those
paths passing through edge e [2].
3.7. Centrality 271

Parameters
G
k
[int, optional (default=None)] If k is not None use k node samples to estimate betweenness.
The value of k <= n where n is the number of nodes in the graph. Higher values give better
approximation.
normalized
[bool, optional] If True the betweenness values are normalized by 2/(n(n − 1)) for graphs,
and 1/(n(n − 1)) for directed graphs where n is the number of nodes in G.
weight
seed
Randomness. Note that this is only used if k is not None.
Returns
edges
[dictionary] Dictionary of edges with betweenness centrality as the value.
See also:
edge_load
Notes
The algorithm is from Ulrik Brandes [1].

References
[1], [2]
edge_betweenness_centrality_subset
edge_betweenness_centrality_subset(G, sources, targets, normalized=False, weight=None)

Compute betweenness centrality for edges for a subset of nodes.
∑ σ(s, t|e)
cB (v) =
σ(s, t)
s∈S,t∈T
where S is the set of sources, T is the set of targets, σ(s, t) is the number of shortest (s, t)-paths, and σ(s, t|e) is
the number of those paths passing through edge e [2].
Parameters

G
[graph] A networkx graph.
Nodes to use as sources for shortest paths in betweenness
Nodes to use as targets for shortest paths in betweenness
normalized
[bool, optional] If True the betweenness values are normalized by 2/(n(n-1)) for graphs,
and 1/(n(n-1)) for directed graphs where n is the number of nodes in G.
weight
Returns
edges
[dictionary] Dictionary of edges with Betweenness centrality as the value.
See also:
edge_load
Notes
The basic algorithm is from [1].

The normalization might seem a little strange but it is the same as in
edge_betweenness_centrality() and is designed to make edge_betweenness_centrality(G) be the same as
edge_betweenness_centrality_subset(G,sources=G.nodes(),targets=G.nodes()).
References
[1], [2]
3.7.6 Current Flow Betweenness
current_flow_betweenness_centrality(G[, Compute current-flow betweenness centrality for nodes.

...])
edge_current_flow_betweenness_centrality(G)
Compute current-flow betweenness centrality for edges.
approximate_current_flow_betweenness_centrality(G)
Compute the approximate current-flow betweenness cen-
trality for nodes.
current_flow_betweenness_centrality_subset(G,
Compute current-flow betweenness centrality for subsets
...) of nodes.
edge_current_flow_betweenness_centrality_subset(G,
Compute current-flow betweenness centrality for edges
...) using subsets of nodes.
3.7. Centrality 273

current_flow_betweenness_centrality
current_flow_betweenness_centrality(G, normalized=True, weight=None, dtype=<class 'float'>,

solver='full')
Compute current-flow betweenness centrality for nodes.
Current-flow betweenness centrality uses an electrical current model for information spreading in contrast to be-
tweenness centrality which uses shortest paths.
Current-flow betweenness centrality is also known as random-walk betweenness centrality [2].
Parameters
G
normalized
[bool, optional (default=True)] If True the betweenness values are normalized by 2/[(n-1)(n-
2)] where n is the number of nodes in G.
weight
[string or None, optional (default=None)] Key for edge data used as the edge weight. If None,
then use 1 as each edge weight. The weight reflects the capacity or the strength of the edge.
dtype
[data type (float)] Default data type for internal matrices. Set to np.float32 for lower memory
consumption.
solver
[string (default=’full’)] Type of linear solver to use for computing the flow matrix. Options are
“full” (uses most memory), “lu” (recommended), and “cg” (uses least memory).
Returns
nodes
See also:
approximate_current_flow_betweenness_centrality
edge_current_flow_betweenness_centrality
Notes
Current-flow betweenness can be computed in O(I(n − 1) + mn log n) time [1], where I(n − 1) is the time
needed√to compute the inverse Laplacian. For a full matrix this is O(n3 ) but using sparse methods you can achieve
O(nm k) where k is the Laplacian matrix condition number.
The space required is O(nw) where w is the width of the sparse Laplacian matrix. Worse case is w = n for O(n2 ).
If the edges have a ‘weight’ attribute they will be used as weights in this algorithm. Unspecified weights are set to
1.

References
[1], [2]
edge_current_flow_betweenness_centrality(G, normalized=True, weight=None, dtype=<class

'float'>, solver='full')
Compute current-flow betweenness centrality for edges.
Parameters
G
normalized
weight
dtype
[data type (default=float)] Default data type for internal matrices. Set to np.float32 for lower
memory consumption.
solver
Returns
nodes
[dictionary] Dictionary of edge tuples with betweenness centrality as the value.
Raises
NetworkXError
The algorithm does not support DiGraphs. If the input graph is an instance of DiGraph class,
NetworkXError is raised.
See also:
3.7. Centrality 275

Notes
1.
References
[1], [2]
approximate_current_flow_betweenness_centrality(G, normalized=True, weight=None,

dtype=<class 'float'>, solver='full',
epsilon=0.5, kmax=10000, seed=None)
Compute the approximate current-flow betweenness centrality for nodes.
Approximates the current-flow betweenness centrality within absolute error of epsilon with high probability [1].
Parameters
G
normalized
weight
dtype
[data type (float)] Default data type for internal matrices. Set to np.float32 for lower memory
consumption.
solver
epsilon: float
Absolute error tolerance.
kmax: int
Maximum number of sample node pairs to use for approximation.
seed
Randomness.
Returns
nodes

See also:
Notes
√
The running time is O((1/ϵ2 )m k log n) and the space required is O(m) for n nodes and m edges.
1.
References
[1]
current_flow_betweenness_centrality_subset
current_flow_betweenness_centrality_subset(G, sources, targets, normalized=True, weight=None,

dtype=<class 'float'>, solver='lu')
Compute current-flow betweenness centrality for subsets of nodes.
Parameters
G
Nodes to use as sources for current
Nodes to use as sinks for current
normalized
[bool, optional (default=True)] If True the betweenness values are normalized by b=b/(n-1)(n-
2) where n is the number of nodes in G.
weight
dtype: data type (float)
Returns
nodes
See also:
3.7. Centrality 277

Notes
1.
References
[1], [2]
edge_current_flow_betweenness_centrality_subset
edge_current_flow_betweenness_centrality_subset(G, sources, targets, normalized=True,

weight=None, dtype=<class 'float'>,
solver='lu')
Compute current-flow betweenness centrality for edges using subsets of nodes.
Parameters
G
Nodes to use as sources for current
Nodes to use as sinks for current
normalized
weight
dtype: data type (float)

Returns
nodes
[dict] Dictionary of edge tuples with betweenness centrality as the value.
See also:
Notes
1.
References
[1], [2]
3.7.7 Communicability Betweenness
communicability_betweenness_centrality(G)
Returns subgraph communicability for all pairs of nodes
in G.
communicability_betweenness_centrality
communicability_betweenness_centrality(G)
Returns subgraph communicability for all pairs of nodes in G.
Communicability betweenness measure makes use of the number of walks connecting every pair of nodes as the
basis of a betweenness centrality measure.
Parameters
G: graph
Returns
nodes
[dictionary] Dictionary of nodes with communicability betweenness as the value.
Raises
NetworkXError
If the graph is not undirected and simple.
3.7. Centrality 279

Notes
Let G=(V,E) be a simple undirected graph with n nodes and m edges, and A denote the adjacency matrix of G.
Let G(r)=(V,E(r)) be the graph resulting from removing all edges connected to node r but not the node itself.
The adjacency matrix for G(r) is A+E(r), where E(r) has nonzeros only in row and column r.
The subraph betweenness of a node r is [1]
1 ∑ ∑ Gprq
ωr = ̸ q, q ̸= r,
,p =
C p q Gpq
where G_{prq}=(e^{A}_{pq} - (e^{A+E(r)})_{pq} is the number of walks involving node r,

G_{pq}=(e^{A})_{pq} is the number of closed walks starting at node p and ending at node q, and
C=(n-1)^{2}-(n-1) is a normalization factor equal to the number of terms in the sum.
The resulting omega_{r} takes values between zero and one. The lower bound cannot be attained for a connected
graph, and the upper bound is attained in the star graph.
References
[1]
Examples
>>> G = nx.Graph([(0, 1), (1, 2), (1, 5), (5, 4), (2, 4), (2, 3), (4, 3), (3, 6)])
>>> cbc = nx.communicability_betweenness_centrality(G)
>>> print([f"{node} {cbc[node]:0.2f}" for node in sorted(cbc)])
['0 0.03', '1 0.45', '2 0.51', '3 0.45', '4 0.40', '5 0.19', '6 0.03']
3.7.8 Group Centrality
group_betweenness_centrality(G, C[, ...]) Compute the group betweenness centrality for a group of
nodes.
group_closeness_centrality(G, S[, weight]) Compute the group closeness centrality for a group of
nodes.
group_degree_centrality(G, S) Compute the group degree centrality for a group of nodes.
group_in_degree_centrality(G, S) Compute the group in-degree centrality for a group of
nodes.
group_out_degree_centrality(G, S) Compute the group out-degree centrality for a group of
nodes.
prominent_group(G, k[, weight, C, ...]) Find the prominent group of size k in graph G.

group_betweenness_centrality
group_betweenness_centrality(G, C, normalized=True, weight=None, endpoints=False)

Compute the group betweenness centrality for a group of nodes.
Group betweenness centrality of a group of nodes C is the sum of the fraction of all-pairs shortest paths that pass
through any vertex in C
∑ σ(s, t|v)
cB (v) =
σ(s, t)
s,t∈V
where V is the set of nodes, σ(s, t) is the number of shortest (s, t)-paths, and σ(s, t|C) is the number of those
paths passing through some node in group C. Note that (s, t) are not members of the group (V − C is the set of
nodes in V that are not in C).
Parameters
G
C
[list or set or list of lists or list of sets] A group or a list of groups containing nodes which
belong to G, for which group betweenness centrality is to be calculated.
normalized
[bool, optional (default=True)] If True, group betweenness is normalized by 1/
((|V|-|C|)(|V|-|C|-1)) where |V| is the number of nodes in G and |C| is the
number of nodes in C.
weight
erwise holds the name of the edge attribute used as weight. The weight of an edge is treated
as the length or distance between the two sides.
endpoints
[bool, optional (default=False)] If True include the endpoints in the shortest path counts.
Returns
betweenness
[list of floats or float] If C is a single group then return a float. If C is a list with several groups
then return a list of group betweenness centralities.
Raises
NodeNotFound
If node(s) in C are not present in G.
See also:
3.7. Centrality 281

Notes
Group betweenness centrality is described in [1] and its importance discussed in [3]. The initial implementation
of the algorithm is mentioned in [2]. This function uses an improved algorithm presented in [4].
The number of nodes in the group must be a maximum of n - 2 where n is the total number of nodes in the graph.
Directed paths between “u” and “v” are counted as two possible paths (one each direction) while undirected paths
between “u” and “v” are counted as one path. Said another way, the sum in the expression above is over all s !=
t for directed graphs and for s < t for undirected graphs.
References
[1], [2], [3], [4]
group_closeness_centrality
group_closeness_centrality(G, S, weight=None)
Compute the group closeness centrality for a group of nodes.
Group closeness centrality of a group of nodes S is a measure of how close the group is to the other nodes in the
graph.
|V − S|
cclose (S) = ∑
v∈V −S dS,v
dS,v = minu∈S (du,v )
where V is the set of nodes, dS,v is the distance of the group S from v defined as above. (V − S is the set of nodes
in V that are not in S).
Parameters
G
S
[list or set] S is a group of nodes which belong to G, for which group closeness centrality is to
be calculated.
weight
Returns
closeness
[float] Group closeness centrality of the group S.
Raises
NodeNotFound
If node(s) in S are not present in G.
See also:

Notes
The measure was introduced in [1]. The formula implemented here is described in [2].
Higher values of closeness indicate greater centrality.
It is assumed that 1 / 0 is 0 (required in the case of directed graphs, or when a shortest path length is 0).
For directed graphs, the incoming distance is utilized here. To use the outward distance, act on G.reverse().
References
[1], [2]
group_degree_centrality
group_degree_centrality(G, S)
Compute the group degree centrality for a group of nodes.
Group degree centrality of a group of nodes S is the fraction of non-group members connected to group members.
Parameters
G
S
[list or set] S is a group of nodes which belong to G, for which group degree centrality is to be
calculated.
Returns
centrality
[float] Group degree centrality of the group S.
Raises
NetworkXError
If node(s) in S are not in G.
See also:
degree_centrality
group_in_degree_centrality
group_out_degree_centrality
3.7. Centrality 283

Notes
The measure was introduced in [1].

References
[1]
group_in_degree_centrality(G, S)
Compute the group in-degree centrality for a group of nodes.
Group in-degree centrality of a group of nodes S is the fraction of non-group members connected to group members
by incoming edges.
Parameters
G
S
[list or set] S is a group of nodes which belong to G, for which group in-degree centrality is to
be calculated.
Returns
centrality
[float] Group in-degree centrality of the group S.
Raises
If G is undirected.
NodeNotFound
See also:
degree_centrality
Notes
G.neighbors(i) gives nodes with an outward edge from i, in a DiGraph, so for group in-degree centrality, the
reverse graph is used.

group_out_degree_centrality(G, S)
Compute the group out-degree centrality for a group of nodes.
Group out-degree centrality of a group of nodes S is the fraction of non-group members connected to group
members by outgoing edges.
Parameters
G
S
[list or set] S is a group of nodes which belong to G, for which group in-degree centrality is to
be calculated.
Returns
centrality
[float] Group out-degree centrality of the group S.
Raises
If G is undirected.
NodeNotFound
See also:
degree_centrality
Notes
G.neighbors(i) gives nodes with an outward edge from i, in a DiGraph, so for group out-degree centrality,
the graph itself is used.
prominent_group
prominent_group(G, k, weight=None, C=None, endpoints=False, normalized=True, greedy=False)

Find the prominent group of size k in graph G. The prominence of the group is evaluated by the group betweenness
centrality.
Group betweenness centrality of a group of nodes C is the sum of the fraction of all-pairs shortest paths that pass
through any vertex in C
∑ σ(s, t|v)
cB (v) =
σ(s, t)
s,t∈V
where V is the set of nodes, σ(s, t) is the number of shortest (s, t)-paths, and σ(s, t|C) is the number of those
paths passing through some node in group C. Note that (s, t) are not members of the group (V − C is the set of
nodes in V that are not in C).
3.7. Centrality 285

Parameters
G
k
[int] The number of nodes in the group.
normalized
[bool, optional (default=True)] If True, group betweenness is normalized by 1/
((|V|-|C|)(|V|-|C|-1)) where |V| is the number of nodes in G and |C| is the
number of nodes in C.
weight
endpoints
[bool, optional (default=False)] If True include the endpoints in the shortest path counts.
C
[list or set, optional (default=None)] list of nodes which won’t be candidates of the prominent
group.
greedy
[bool, optional (default=False)] Using a naive greedy algorithm in order to find non-optimal
prominent group. For scale free networks the results are negligibly below the optimal results.
Returns
max_GBC
[float] The group betweenness centrality of the prominent group.
max_group
[list] The list of nodes in the prominent group.
Raises
NodeNotFound
If node(s) in C are not present in G.
See also:
betweenness_centrality, group_betweenness_centrality
Notes
Group betweenness centrality is described in [1] and its importance discussed in [3]. The algorithm is described
in [2] and is based on techniques mentioned in [4].
The number of nodes in the group must be a maximum of n - 2 where n is the total number of nodes in the
graph.
Directed paths between “u” and “v” are counted as two possible paths (one each direction) while undirected paths
between “u” and “v” are counted as one path. Said another way, the sum in the expression above is over all s !=
t for directed graphs and for s < t for undirected graphs.

References
[1], [2], [3], [4]
3.7.9 Load
load_centrality(G[, v, cutoff, normalized, ...]) Compute load centrality for nodes.

edge_load_centrality(G[, cutoff]) Compute edge load.
load_centrality
load_centrality(G, v=None, cutoff=None, normalized=True, weight=None)

Compute load centrality for nodes.
The load centrality of a node is the fraction of all shortest paths that pass through that node.
Parameters
G
[graph] A networkx graph.
normalized
weight
[None or string, optional (default=None)] If None, edge weights are ignored. Otherwise holds
the name of the edge attribute used as weight. The weight of an edge is treated as the length
or distance between the two sides.
cutoff
[bool, optional (default=None)] If specified, only consider paths of length <= cutoff.
Returns
nodes
[dictionary] Dictionary of nodes with centrality as the value.
See also:
Notes
Load centrality is slightly different than betweenness. It was originally introduced by [2]. For this load algorithm
see [1].
3.7. Centrality 287

References
[1], [2]
edge_load_centrality
edge_load_centrality(G, cutoff=False)
Compute edge load.
WARNING: This concept of edge load has not been analysed or discussed outside of NetworkX that we know of.
It is based loosely on load_centrality in the sense that it counts the number of shortest paths which cross each edge.
This function is for demonstration and testing purposes.
Parameters
G
cutoff
[bool, optional (default=False)] If specified, only consider paths of length <= cutoff.
Returns
A dict keyed by edge 2-tuple to the number of shortest paths
which use that edge. Where more than one path is shortest
the count is divided equally among paths.
3.7.10 Subgraph
subgraph_centrality(G) Returns subgraph centrality for each node in G.

subgraph_centrality_exp(G) Returns the subgraph centrality for each node of G.
estrada_index(G) Returns the Estrada index of a the graph G.
subgraph_centrality
subgraph_centrality(G)
Returns subgraph centrality for each node in G.
Subgraph centrality of a node n is the sum of weighted closed walks of all lengths starting and ending at node n.
The weights decrease with path length. Each closed walk is associated with a connected subgraph ([1]).
Parameters
G: graph
Returns
nodes
[dictionary] Dictionary of nodes with subgraph centrality as the value.
Raises
NetworkXError
See also:

subgraph_centrality_exp
Alternative algorithm of the subgraph centrality for each node of G.
Notes
This version of the algorithm computes eigenvalues and eigenvectors of the adjacency matrix.
Subgraph centrality of a node u in G can be found using a spectral decomposition of the adjacency matrix [1],
∑
N
SC(u) = (vju )2 eλj ,
j=1
where v_j is an eigenvector of the adjacency matrix A of G corresponding to the eigenvalue lambda_j.
References
[1]
Examples
(Example from [1]) >>> G = nx.Graph( … [ … (1, 2), … (1, 5), … (1, 8), … (2, 3), … (2, 8), … (3, 4), … (3, 6),
… (4, 5), … (4, 7), … (5, 6), … (6, 7), … (7, 8), … ] … ) >>> sc = nx.subgraph_centrality(G) >>> print([f”{node}
{sc[node]:0.2f}” for node in sorted(sc)]) [‘1 3.90’, ‘2 3.90’, ‘3 3.64’, ‘4 3.71’, ‘5 3.64’, ‘6 3.71’, ‘7 3.64’, ‘8 3.90’]
subgraph_centrality_exp
subgraph_centrality_exp(G)
Returns the subgraph centrality for each node of G.
Subgraph centrality of a node n is the sum of weighted closed walks of all lengths starting and ending at node n.
The weights decrease with path length. Each closed walk is associated with a connected subgraph ([1]).
Parameters
G: graph
Returns
nodes:dictionary
Dictionary of nodes with subgraph centrality as the value.
Raises
NetworkXError
See also:
subgraph_centrality
Alternative algorithm of the subgraph centrality for each node of G.
3.7. Centrality 289

Notes
This version of the algorithm exponentiates the adjacency matrix.

The subgraph centrality of a node u in G can be found using the matrix exponential of the adjacency matrix of G
[1],
SC(u) = (eA )uu .
References
[1]
Examples
(Example from [1]) >>> G = nx.Graph( … [ … (1, 2), … (1, 5), … (1, 8), … (2, 3), … (2, 8), … (3, 4), …
(3, 6), … (4, 5), … (4, 7), … (5, 6), … (6, 7), … (7, 8), … ] … ) >>> sc = nx.subgraph_centrality_exp(G) >>>
print([f”{node} {sc[node]:0.2f}” for node in sorted(sc)]) [‘1 3.90’, ‘2 3.90’, ‘3 3.64’, ‘4 3.71’, ‘5 3.64’, ‘6 3.71’, ‘7
3.64’, ‘8 3.90’]
estrada_index
estrada_index(G)
Returns the Estrada index of a the graph G.
The Estrada Index is a topological index of folding or 3D “compactness” ([1]).
Parameters
G: graph
Returns
estrada index: float
Raises
NetworkXError
Notes
Let G=(V,E) be a simple undirected graph with n nodes and let

lambda_{1}leqlambda_{2}leqcdotslambda_{n} be a non-increasing ordering of the eigenval-
ues of its adjacency matrix A. The Estrada index is ([1], [2])
∑
n
EE(G) = e λj .
j=1

References
[1], [2]
Examples
>>> G = nx.Graph([(0, 1), (1, 2), (1, 5), (5, 4), (2, 4), (2, 3), (4, 3), (3, 6)])
>>> ei = nx.estrada_index(G)
>>> print(f"{ei:0.5}")
20.55
3.7.11 Harmonic Centrality
harmonic_centrality(G[, nbunch, distance, ...]) Compute harmonic centrality for nodes.
harmonic_centrality
harmonic_centrality(G, nbunch=None, distance=None, sources=None)

Compute harmonic centrality for nodes.
Harmonic centrality [1] of a node u is the sum of the reciprocal of the shortest path distances from all other nodes
to u
∑ 1
C(u) =
d(v, u)
v̸=u
where d(v, u) is the shortest-path distance between v and u.

If sources is given as an argument, the returned harmonic centrality values are calculated as the sum of the
reciprocals of the shortest path distances from the nodes specified in sources to u instead of from all nodes to
u.
Notice that higher values indicate higher centrality.
Parameters
G
nbunch
[container (default: all nodes in G)] Container of nodes for which harmonic centrality values
are calculated.
sources
[container (default: all nodes in G)] Container of nodes v over which reciprocal distances are
computed. Nodes not in G are silently ignored.
distance
[edge attribute key, optional (default=None)] Use the specified edge attribute as the edge dis-
tance in shortest path calculations. If None, then each edge will have distance equal to 1.
Returns
nodes
[dictionary] Dictionary of nodes with harmonic centrality as the value.
3.7. Centrality 291

See also:

degree_centrality, closeness_centrality
Notes
If the ‘distance’ keyword is set to an edge attribute key then the shortest-path length will be computed using Dijkstra’s
algorithm with that edge attribute as the edge weight.
References
[1]
3.7.12 Dispersion
dispersion(G[, u, v, normalized, alpha, b, c]) Calculate dispersion between u and v in G.
dispersion
dispersion(G, u=None, v=None, normalized=True, alpha=1.0, b=0.0, c=0.0)

Calculate dispersion between u and v in G.
A link between two actors (u and v) has a high dispersion when their mutual ties (s and t) are not well connected
with each other.
Parameters
G
u
[node, optional] The source for the dispersion score (e.g. ego node of the network).
v
[node, optional] The target of the dispersion score if specified.
normalized
[bool] If True (default) normalize by the embededness of the nodes (u and v).
alpha, b, c
[float] Parameters for the normalization procedure. When normalized is True, the disper-
sion value is normalized by:
result = ((dispersion + b) ** alpha) / (embeddedness + c)
as long as the denominator is nonzero.

Returns
nodes
[dictionary] If u (v) is specified, returns a dictionary of nodes with dispersion score for all
“target” (“source”) nodes. If neither u nor v is specified, returns a dictionary of dictionaries for
all nodes ‘u’ in the graph with a dispersion score for each node ‘v’.

Notes
This implementation follows Lars Backstrom and Jon Kleinberg [1]. Typical usage would be to run dispersion on
the ego network Gu if u were specified. Running dispersion() with neither u nor v specified can take some
time to complete.
References
[1]
3.7.13 Reaching
local_reaching_centrality(G, v[, paths, ...]) Returns the local reaching centrality of a node in a di-
rected graph.
global_reaching_centrality(G[, weight, ...]) Returns the global reaching centrality of a directed graph.
local_reaching_centrality
local_reaching_centrality(G, v, paths=None, weight=None, normalized=True)

Returns the local reaching centrality of a node in a directed graph.
The local reaching centrality of a node in a directed graph is the proportion of other nodes reachable from that node
[1].
Parameters
G
[DiGraph] A NetworkX DiGraph.
v
[node] A node in the directed graph G.
paths
[dictionary (default=None)] If this is not None it must be a dictionary representation of single-
source shortest paths, as computed by, for example, networkx.shortest_path() with
source node v. Use this keyword argument if you intend to invoke this function many times
but don’t want the paths to be recomputed each time.
weight
[None or string, optional (default=None)] Attribute to use for edge weights. If None, each
edge weight is assumed to be one. A higher weight implies a stronger connection between
nodes and a shorter path length.
normalized
[bool, optional (default=True)] Whether to normalize the edge weights by the total sum of edge
weights.
Returns
h
[float] The local reaching centrality of the node v in the graph G.
See also:
global_reaching_centrality
3.7. Centrality 293

References
[1]
Examples
>>> G.add_edges_from([(1, 2), (1, 3)])
>>> nx.local_reaching_centrality(G, 3)
0.0
>>> nx.local_reaching_centrality(G, 3)
0.5
global_reaching_centrality
global_reaching_centrality(G, weight=None, normalized=True)

Returns the global reaching centrality of a directed graph.
The global reaching centrality of a weighted directed graph is the average over all nodes of the difference between
the local reaching centrality of the node and the greatest local reaching centrality of any node in the graph [1]. For
more information on the local reaching centrality, see local_reaching_centrality(). Informally, the
local reaching centrality is the proportion of the graph that is reachable from the neighbors of the node.
Parameters
G
[DiGraph] A networkx DiGraph.
weight
[None or string, optional (default=None)] Attribute to use for edge weights. If None, each
edge weight is assumed to be one. A higher weight implies a stronger connection between
nodes and a shorter path length.
normalized
[bool, optional (default=True)] Whether to normalize the edge weights by the total sum of edge
weights.
Returns
h
[float] The global reaching centrality of the graph.
See also:
local_reaching_centrality

References
[1]
Examples
>>> nx.global_reaching_centrality(G)
1.0
>>> nx.global_reaching_centrality(G)
0.75
3.7.14 Percolation
percolation_centrality(G[, attribute, ...]) Compute the percolation centrality for nodes.
percolation_centrality
percolation_centrality(G, attribute='percolation', states=None, weight=None)

Compute the percolation centrality for nodes.
Percolation centrality of a node v, at a given time, is defined as the proportion of ‘percolated paths’ that go through
that node.
This measure quantifies relative impact of nodes based on their topological connectivity, as well as their percolation
states.
Percolation states of nodes are used to depict network percolation scenarios (such as during infection transmission
in a social network of individuals, spreading of computer viruses on computer networks, or transmission of disease
over a network of towns) over time. In this measure usually the percolation state is expressed as a decimal between
0.0 and 1.0.
When all nodes are in the same percolated state this measure is equivalent to betweenness centrality.
Parameters
G
attribute
[None or string, optional (default=’percolation’)] Name of the node attribute to use for perco-
lation state, used if states is None.
states
[None or dict, optional (default=None)] Specify percolation states for the nodes, nodes as keys
states as values.
weight
3.7. Centrality 295

Returns
nodes
[dictionary] Dictionary of nodes with percolation centrality as the value.
See also:
Notes
The algorithm is from Mahendra Piraveenan, Mikhail Prokopenko, and Liaquat Hossain [1] Pair dependecies are
calculated and accumulated using [2]
References
[1], [2]
3.7.15 Second Order Centrality
second_order_centrality(G) Compute the second order centrality for nodes of G.
second_order_centrality
second_order_centrality(G)
Compute the second order centrality for nodes of G.
The second order centrality of a given node is the standard deviation of the return times to that node of a perpetual
random walk on G:
Parameters
G
[graph] A NetworkX connected and undirected graph.
Returns
nodes
[dictionary] Dictionary keyed by node with second order centrality as the value.
Raises
NetworkXException
If the graph G is empty, non connected or has negative weights.
See also:

Notes
Lower values of second order centrality indicate higher centrality.

The algorithm is from Kermarrec, Le Merrer, Sericola and Trédan [1].
This code implements the analytical version of the algorithm, i.e., there is no simulation of a random walk process
involved. The random walk is here unbiased (corresponding to eq 6 of the paper [1]), thus the centrality values are
the standard deviations for random walk return times on the transformed input graph G (equal in-degree at each
nodes by adding self-loops).
Complexity of this implementation, made to run locally on a single machine, is O(n^3), with n the size of G, which
makes it viable only for small graphs.
References
[1]
Examples
>>> soc = nx.second_order_centrality(G)
>>> print(sorted(soc.items(), key=lambda x: x[1])[0][0]) # pick first id
0
3.7.16 Trophic
trophic_levels(G[, weight]) Compute the trophic levels of nodes.

trophic_differences(G[, weight]) Compute the trophic differences of the edges of a directed
graph.
trophic_incoherence_parameter(G[, weight, Compute the trophic incoherence parameter of a graph.
...])
trophic_levels
trophic_levels(G, weight='weight')
Compute the trophic levels of nodes.
The trophic level of a node i is
1 ∑
si = 1 + aij sj
kiin j
where kiin is the in-degree of i

∑
kiin = aij
j
and nodes with kiin = 0 have si = 1 by convention.

These are calculated using the method outlined in Levine [1].
3.7. Centrality 297

Parameters
G
[DiGraph] A directed networkx graph
Returns
nodes
[dict] Dictionary of nodes with trophic level as the value.
References
[1]
trophic_differences
trophic_differences(G, weight='weight')
Compute the trophic differences of the edges of a directed graph.
The trophic difference xi j for each edge is defined in Johnson et al. [1] as:
xi j = s j − s i
Where si is the trophic level of node i.

Parameters
G
Returns
diffs
[dict] Dictionary of edges with trophic differences as the value.
References
[1]
trophic_incoherence_parameter
trophic_incoherence_parameter(G, weight='weight', cannibalism=False)

Compute the trophic incoherence parameter of a graph.
Trophic coherence is defined as the homogeneity of the distribution of trophic distances: the more similar, the
more coherent. This is measured by the standard deviation of the trophic differences and referred to as the trophic
incoherence parameter q by [1].
Parameters
G
cannibalism: Boolean
If set to False, self edges are not considered in the calculation
Returns

trophic_incoherence_parameter
[float] The trophic coherence of a graph
References
[1]
3.7.17 VoteRank
voterank(G[, number_of_nodes]) Select a list of influential nodes in a graph using VoteRank

algorithm
voterank
voterank(G, number_of_nodes=None)
Select a list of influential nodes in a graph using VoteRank algorithm
VoteRank [1] computes a ranking of the nodes in a graph G based on a voting scheme. With VoteRank, all nodes
vote for each of its in-neighbours and the node with the highest votes is elected iteratively. The voting ability of
out-neighbors of elected nodes is decreased in subsequent turns.
Parameters
G
number_of_nodes
[integer, optional] Number of ranked nodes to extract (default all nodes).
Returns
voterank
[list] Ordered list of computed seeds. Only nodes with positive number of votes are returned.
Notes
Each edge is treated independently in case of multigraphs.
References
[1]
3.7. Centrality 299

Examples
>>> G = nx.Graph([(0, 1), (0, 2), (0, 3), (1, 4)])

>>> nx.voterank(G)
[0, 1]
The algorithm can be used both for undirected and directed graphs. However, the directed version is different in two
ways: (i) nodes only vote for their in-neighbors and (ii) only the voting ability of elected node and its out-neighbors
are updated:
>>> G = nx.DiGraph([(0, 1), (2, 1), (2, 3), (3, 4)])

>>> nx.voterank(G)
[2, 3]
3.8 Chains
Functions for finding chains in a graph.
chain_decomposition(G[, root]) Returns the chain decomposition of a graph.
3.8.1 chain_decomposition
chain_decomposition(G, root=None)
Returns the chain decomposition of a graph.
The chain decomposition of a graph with respect a depth-first search tree is a set of cycles or paths derived from the
set of fundamental cycles of the tree in the following manner. Consider each fundamental cycle with respect to the
given tree, represented as a list of edges beginning with the nontree edge oriented away from the root of the tree.
For each fundamental cycle, if it overlaps with any previous fundamental cycle, just take the initial non-overlapping
segment, which is a path instead of a cycle. Each cycle or path is called a chain. For more information, see [1].
Parameters
G
[undirected graph]
root
[node (optional)] A node in the graph G. If specified, only the chain decomposition for the
connected component containing this node will be returned. This node indicates the root of
the depth-first search tree.
Yields
chain
[list] A list of edges representing a chain. There is no guarantee on the orientation of the edges
in each chain (for example, if a chain includes the edge joining nodes 1 and 2, the chain may
include either (1, 2) or (2, 1)).
Raises
NodeNotFound

Notes
The worst-case running time of this implementation is linear in the number of nodes and number of edges [1].
References
[1]
Examples
>>> G = nx.Graph([(0, 1), (1, 4), (3, 4), (3, 5), (4, 5)])
>>> list(nx.chain_decomposition(G))
[[(4, 5), (5, 3), (3, 4)]]
3.9 Chordal
Algorithms for chordal graphs.

A graph is chordal if every cycle of length at least 4 has a chord (an edge joining two nodes not adjacent in the cycle).
https://en.wikipedia.org/wiki/Chordal_graph
is_chordal(G) Checks whether G is a chordal graph.

chordal_graph_cliques(G) Returns all maximal cliques of a chordal graph.
chordal_graph_treewidth(G) Returns the treewidth of the chordal graph G.
complete_to_chordal_graph(G) Return a copy of G completed to a chordal graph
find_induced_nodes(G, s, t[, treewidth_bound]) Returns the set of induced nodes in the path from s to t.
3.9.1 is_chordal
is_chordal(G)
Checks whether G is a chordal graph.
A graph is chordal if every cycle of length at least 4 has a chord (an edge joining two nodes not adjacent in the
cycle).
Parameters
G
Returns
chordal
[bool] True if G is a chordal graph and False otherwise.
Raises
The algorithm does not support DiGraph, MultiGraph and MultiDiGraph.
3.9. Chordal 301

Notes
The routine tries to go through every node following maximum cardinality search. It returns False when it finds
that the separator for any node is not a clique. Based on the algorithms in [1].
References
[1]
Examples
>>> e = [
... (1, 2),
... (1, 3),
... (2, 3),
... (2, 4),
... (3, 4),
... (3, 5),
... (3, 6),
... (4, 5),
... (4, 6),
... (5, 6),
... ]
>>> G = nx.Graph(e)
>>> nx.is_chordal(G)
True
3.9.2 chordal_graph_cliques
chordal_graph_cliques(G)
Returns all maximal cliques of a chordal graph.
The algorithm breaks the graph in connected components and performs a maximum cardinality search in each
component to get the cliques.
Parameters
G
Yields
frozenset of nodes
Maximal cliques, each of which is a frozenset of nodes in G. The order of cliques is arbitrary.
Raises
NetworkXError
The algorithm does not support DiGraph, MultiGraph and MultiDiGraph. The algorithm can
only be applied to chordal graphs. If the input graph is found to be non-chordal, a Net-
workXError is raised.

Examples
>>> e = [
... (1, 2),
... (1, 3),
... (2, 3),
... (2, 4),
... (3, 4),
... (3, 5),
... (3, 6),
... (4, 5),
... (4, 6),
... (5, 6),
... (7, 8),
... ]
>>> G = nx.Graph(e)
>>> G.add_node(9)
>>> cliques = [c for c in chordal_graph_cliques(G)]
>>> cliques[0]
frozenset({1, 2, 3})
3.9.3 chordal_graph_treewidth
chordal_graph_treewidth(G)
Returns the treewidth of the chordal graph G.
Parameters
G
Returns
treewidth
[int] The size of the largest clique in the graph minus one.
Raises
NetworkXError
The algorithm does not support DiGraph, MultiGraph and MultiDiGraph. The algorithm can
only be applied to chordal graphs. If the input graph is found to be non-chordal, a Net-
References
[1]
3.9. Chordal 303

Examples
>>> e = [
... (1, 2),
... (1, 3),
... (2, 3),
... (2, 4),
... (3, 4),
... (3, 5),
... (3, 6),
... (4, 5),
... (4, 6),
... (5, 6),
... (7, 8),
... ]
>>> G = nx.Graph(e)
>>> G.add_node(9)
>>> nx.chordal_graph_treewidth(G)
3
3.9.4 complete_to_chordal_graph
complete_to_chordal_graph(G)
Return a copy of G completed to a chordal graph
Adds edges to a copy of G to create a chordal graph. A graph G=(V,E) is called chordal if for each cycle with
length bigger than 3, there exist two non-adjacent nodes connected by an edge (called a chord).
Parameters
G
Returns
H
[NetworkX graph] The chordal enhancement of G
alpha
[Dictionary] The elimination ordering of nodes of G
Notes
There are different approaches to calculate the chordal enhancement of a graph. The algorithm used here is called
MCS-M and gives at least minimal (local) triangulation of graph. Note that this triangulation is not necessarily a
global minimum.
https://en.wikipedia.org/wiki/Chordal_graph

References
[1]
Examples
>>> from networkx.algorithms.chordal import complete_to_chordal_graph

>>> G = nx.wheel_graph(10)
>>> H, alpha = complete_to_chordal_graph(G)
3.9.5 find_induced_nodes
find_induced_nodes(G, s, t, treewidth_bound=9223372036854775807)
Returns the set of induced nodes in the path from s to t.
Parameters
G
[graph] A chordal NetworkX graph
s
[node] Source node to look for induced nodes
t
[node] Destination node to look for induced nodes
treewidth_bound: float
Maximum treewidth acceptable for the graph H. The search for induced nodes will end as soon
as the treewidth_bound is exceeded.
Returns
induced_nodes
[Set of nodes] The set of induced nodes in the path from s to t in G
Raises
NetworkXError
The algorithm does not support DiGraph, MultiGraph and MultiDiGraph. If the input graph is
an instance of one of these classes, a NetworkXError is raised. The algorithm can only be
applied to chordal graphs. If the input graph is found to be non-chordal, a NetworkXError
is raised.
Notes
G must be a chordal graph and (s,t) an edge that is not in G.

If a treewidth_bound is provided, the search for induced nodes will end as soon as the treewidth_bound is exceeded.
The algorithm is inspired by Algorithm 4 in [1]. A formal definition of induced node can also be found on that
reference.
3.9. Chordal 305

References
[1]
Examples
>>> G = nx.Graph()
>>> G = nx.generators.classic.path_graph(10)
>>> induced_nodes = nx.find_induced_nodes(G, 1, 9, 2)
>>> sorted(induced_nodes)
[1, 2, 3, 4, 5, 6, 7, 8, 9]
3.10 Clique
Functions for finding and manipulating cliques.

Finding the largest clique in a graph is NP-complete problem, so most of these algorithms have an exponential running
time; for more information, see the Wikipedia article on the clique problem [1].
enumerate_all_cliques(G) Returns all cliques in an undirected graph.

find_cliques(G[, nodes]) Returns all maximal cliques in an undirected graph.
find_cliques_recursive(G[, nodes]) Returns all maximal cliques in a graph.
make_max_clique_graph(G[, create_using]) Returns the maximal clique graph of the given graph.
make_clique_bipartite(G[, fpos, ...]) Returns the bipartite clique graph corresponding to G.
graph_clique_number(G[, cliques]) Returns the clique number of the graph.
graph_number_of_cliques(G[, cliques]) Returns the number of maximal cliques in the graph.
node_clique_number(G[, nodes, cliques, ...]) Returns the size of the largest maximal clique containing
each given node.
number_of_cliques(G[, nodes, cliques]) Returns the number of maximal cliques for each node.
cliques_containing_node(G[, nodes, cliques]) Returns a list of cliques containing the given node.
max_weight_clique(G[, weight]) Find a maximum weight clique in G.
3.10.1 enumerate_all_cliques
enumerate_all_cliques(G)
Returns all cliques in an undirected graph.
This function returns an iterator over cliques, each of which is a list of nodes. The iteration is ordered by cardinality
of the cliques: first all cliques of size one, then all cliques of size two, etc.
Parameters
G
[NetworkX graph] An undirected graph.
Returns
iterator
An iterator over cliques, each of which is a list of nodes in G. The cliques are ordered according
to size.

Notes
To obtain a list of all cliques, use list(enumerate_all_cliques(G)). However, be aware that in the
worst-case, the length of this list can be exponential in the number of nodes in the graph (for example, when the
graph is the complete graph). This function avoids storing all cliques in memory by only keeping current candidate
node lists in memory during its search.
The implementation is adapted from the algorithm by Zhang, et al. (2005) [1] to output all cliques discovered.
This algorithm ignores self-loops and parallel edges, since cliques are not conventionally defined with such edges.
References
[1]
3.10.2 find_cliques
find_cliques(G, nodes=None)
Returns all maximal cliques in an undirected graph.
For each node n, a maximal clique for n is a largest complete subgraph containing n. The largest maximal clique is
sometimes called the maximum clique.
This function returns an iterator over cliques, each of which is a list of nodes. It is an iterative implementation, so
should not suffer from recursion depth issues.
This function accepts a list of nodes and only the maximal cliques containing all of these nodes are returned.
It can considerably speed up the running time if some specific cliques are desired.
Parameters
G
nodes
[list, optional (default=None)] If provided, only yield maximal cliques containing all nodes in
nodes. If nodes isn’t a clique itself, a ValueError is raised.
Returns
iterator
An iterator over maximal cliques, each of which is a list of nodes in G. If nodes is provided,
only the maximal cliques containing all the nodes in nodes are returned. The order of cliques
is arbitrary.
Raises
ValueError
If nodes is not a clique.
See also:
find_cliques_recursive
A recursive version of the same algorithm.
3.10. Clique 307

Notes
To obtain a list of all maximal cliques, use list(find_cliques(G)). However, be aware that in the worst-
case, the length of this list can be exponential in the number of nodes in the graph. This function avoids storing all
cliques in memory by only keeping current candidate node lists in memory during its search.
This implementation is based on the algorithm published by Bron and Kerbosch (1973) [1], as adapted by Tomita,
Tanaka and Takahashi (2006) [2] and discussed in Cazals and Karande (2008) [3]. It essentially unrolls the
recursion used in the references to avoid issues of recursion stack depth (for a recursive implementation, see
find_cliques_recursive()).
References
[1], [2], [3]
3.10.3 find_cliques_recursive
find_cliques_recursive(G, nodes=None)
Returns all maximal cliques in a graph.
For each node v, a maximal clique for v is a largest complete subgraph containing v. The largest maximal clique is
sometimes called the maximum clique.
This function returns an iterator over cliques, each of which is a list of nodes. It is a recursive implementation, so
may suffer from recursion depth issues, but is included for pedagogical reasons. For a non-recursive implementation,
see find_cliques().
This function accepts a list of nodes and only the maximal cliques containing all of these nodes are returned.
It can considerably speed up the running time if some specific cliques are desired.
Parameters
G
[NetworkX graph]
nodes
[list, optional (default=None)] If provided, only yield maximal cliques containing all nodes in
nodes. If nodes isn’t a clique itself, a ValueError is raised.
Returns
iterator
An iterator over maximal cliques, each of which is a list of nodes in G. If nodes is provided,
only the maximal cliques containing all the nodes in nodes are yielded. The order of cliques
is arbitrary.
Raises
ValueError
If nodes is not a clique.
See also:
find_cliques
An iterative version of the same algorithm.

Notes
To obtain a list of all maximal cliques, use list(find_cliques_recursive(G)). However, be aware

that in the worst-case, the length of this list can be exponential in the number of nodes in the graph. This function
avoids storing all cliques in memory by only keeping current candidate node lists in memory during its search.
This implementation is based on the algorithm published by Bron and Kerbosch (1973) [1], as adapted by Tomita,
Tanaka and Takahashi (2006) [2] and discussed in Cazals and Karande (2008) [3]. For a non-recursive implemen-
tation, see find_cliques().
References
[1], [2], [3]
3.10.4 make_max_clique_graph
make_max_clique_graph(G, create_using=None)
Returns the maximal clique graph of the given graph.
The nodes of the maximal clique graph of G are the cliques of G and an edge joins two cliques if the cliques are
not disjoint.
Parameters
G
[NetworkX graph]
create_using
[NetworkX graph constructor, optional (default=nx.Graph)] Graph type to create. If graph
instance, then cleared before populated.
Returns
NetworkX graph
A graph whose nodes are the cliques of G and whose edges join two cliques if they are not
disjoint.
Notes
This function behaves like the following code:
import networkx as nx
G = nx.make_clique_bipartite(G)
cliques = [v for v in G.nodes() if G.nodes[v]['bipartite'] == 0]
G = nx.bipartite.projected_graph(G, cliques)
G = nx.relabel_nodes(G, {-v: v - 1 for v in G})
It should be faster, though, since it skips all the intermediate steps.
3.10. Clique 309

3.10.5 make_clique_bipartite
make_clique_bipartite(G, fpos=None, create_using=None, name=None)

Returns the bipartite clique graph corresponding to G.
In the returned bipartite graph, the “bottom” nodes are the nodes of G and the “top” nodes represent the maximal
cliques of G. There is an edge from node v to clique C in the returned graph if and only if v is an element of C.
Parameters
G
fpos
[bool] If True or not None, the returned graph will have an additional attribute, pos, a dictio-
nary mapping node to position in the Euclidean plane.
create_using
Returns
NetworkX graph
A bipartite graph whose “bottom” set is the nodes of the graph G, whose “top” set is the cliques
of G, and whose edges join nodes of G to the cliques that contain them.
The nodes of the graph G have the node attribute ‘bipartite’ set to 1 and the nodes representing
cliques have the node attribute ‘bipartite’ set to 0, as is the convention for bipartite graphs in
NetworkX.
3.10.6 graph_clique_number
graph_clique_number(G, cliques=None)
Returns the clique number of the graph.
The clique number of a graph is the size of the largest clique in the graph.
Parameters
G
cliques
[list] A list of cliques, each of which is itself a list of nodes. If not specified, the list of all
cliques will be computed, as by find_cliques().
Returns
int
The size of the largest clique in G.

Notes
You should provide cliques if you have already computed the list of maximal cliques, in order to avoid an
exponential time search for maximal cliques.
3.10.7 graph_number_of_cliques
graph_number_of_cliques(G, cliques=None)
Returns the number of maximal cliques in the graph.
Parameters
G
cliques
[list] A list of cliques, each of which is itself a list of nodes. If not specified, the list of all
cliques will be computed, as by find_cliques().
Returns
int
The number of maximal cliques in G.
Notes
You should provide cliques if you have already computed the list of maximal cliques, in order to avoid an
exponential time search for maximal cliques.
3.10.8 node_clique_number
node_clique_number(G, nodes=None, cliques=None, separate_nodes=False)

Returns the size of the largest maximal clique containing each given node.
Returns a single or list depending on input nodes. An optional list of cliques can be input if already computed.
Parameters
G
cliques
[list, optional (default=None)] A list of cliques, each of which is itself a list of nodes. If not
specified, the list of all cliques will be computed using find_cliques().
Returns
int or dict
If nodes is a single node, returns the size of the largest maximal clique in G containing that
node. Otherwise return a dict keyed by node to the size of the largest maximal clique containing
that node.
See also:
find_cliques
find_cliques yields the maximal cliques of G. It accepts a nodes argument which restricts consideration to
maximal cliques containing all the given nodes. The search for the cliques is optimized for nodes.
3.10. Clique 311

3.10.9 number_of_cliques
number_of_cliques(G, nodes=None, cliques=None)

Returns the number of maximal cliques for each node.
Returns a single or list depending on input nodes. Optional list of cliques can be input if already computed.
3.10.10 cliques_containing_node
cliques_containing_node(G, nodes=None, cliques=None)

Returns a list of cliques containing the given node.
Returns a single list or list of lists depending on input nodes. Optional list of cliques can be input if already
computed.
3.10.11 max_weight_clique
max_weight_clique(G, weight='weight')
Find a maximum weight clique in G.
A clique in a graph is a set of nodes such that every two distinct nodes are adjacent. The weight of a clique is the
sum of the weights of its nodes. A maximum weight clique of graph G is a clique C in G such that no clique in G
has weight greater than the weight of C.
Parameters
G
weight
[string or None, optional (default=’weight’)] The node attribute that holds the integer value
used as a weight. If None, then each node has weight 1.
Returns
clique
[list] the nodes of a maximum weight clique
weight
[int] the weight of a maximum weight clique
Notes
The implementation is recursive, and therefore it may run into recursion depth issues if G contains a clique whose
number of nodes is close to the recursion depth limit.
At each search node, the algorithm greedily constructs a weighted independent set cover of part of the graph in
order to find a small set of nodes on which to branch. The algorithm is very similar to the algorithm of Tavares et
al. [1], other than the fact that the NetworkX version does not use bitsets. This style of algorithm for maximum
weight clique (and maximum weight independent set, which is the same problem but on the complement graph)
has a decades-long history. See Algorithm B of Warren and Hicks [2] and the references in that paper.

References
[1], [2]
3.11 Clustering
Algorithms to characterize the number of triangles in a graph.
triangles(G[, nodes]) Compute the number of triangles.

transitivity(G) Compute graph transitivity, the fraction of all possible tri-
angles present in G.
clustering(G[, nodes, weight]) Compute the clustering coefficient for nodes.
average_clustering(G[, nodes, weight, ...]) Compute the average clustering coefficient for the graph
G.
square_clustering(G[, nodes]) Compute the squares clustering coefficient for nodes.
generalized_degree(G[, nodes]) Compute the generalized degree for nodes.
3.11.1 triangles
triangles(G, nodes=None)
Compute the number of triangles.
Finds the number of triangles that include a node as one vertex.
Parameters
G
nodes
[container of nodes, optional (default= all nodes in G)] Compute triangles for nodes in this
container.
Returns
out
[dictionary] Number of triangles keyed by node label.
Notes
When computing triangles for the entire graph each triangle is counted three times, once at each node. Self loops
are ignored.
3.11. Clustering 313

Examples
>>> G = nx.complete_graph(5)
>>> print(nx.triangles(G, 0))
6
>>> print(nx.triangles(G))
{0: 6, 1: 6, 2: 6, 3: 6, 4: 6}
>>> print(list(nx.triangles(G, (0, 1)).values()))
[6, 6]
3.11.2 transitivity
transitivity(G)
Compute graph transitivity, the fraction of all possible triangles present in G.
Possible triangles are identified by the number of “triads” (two edges with a shared vertex).
The transitivity is
#triangles
T =3 .
#triads
Parameters
G
[graph]
Returns
out
[float] Transitivity
Examples
>>> print(nx.transitivity(G))
1.0
3.11.3 clustering
clustering(G, nodes=None, weight=None)

Compute the clustering coefficient for nodes.
For unweighted graphs, the clustering of a node u is the fraction of possible triangles through that node that exist,
2T (u)
cu = ,
deg(u)(deg(u) − 1)
where T (u) is the number of triangles through node u and deg(u) is the degree of u.
For weighted graphs, there are several ways to define clustering [1]. the one used here is defined as the geometric
average of the subgraph edge weights [2],
1 ∑
cu = (ŵuv ŵuw ŵvw )1/3 .
deg(u)(deg(u) − 1)) vw

The edge weights ŵuv are normalized by the maximum weight in the network ŵuv = wuv / max(w).
The value of cu is assigned to 0 if deg(u) < 2.
Additionally, this weighted definition has been generalized to support negative edge weights [3].
For directed graphs, the clustering is similarly defined as the fraction of all possible directed triangles or geometric
average of the subgraph edge weights for unweighted and weighted directed graph respectively [4].
2
cu = T (u),
deg tot (u)(deg tot (u) − 1) − 2deg ↔ (u)
where T (u) is the number of directed triangles through node u, deg tot (u) is the sum of in degree and out degree
of u and deg ↔ (u) is the reciprocal degree of u.
Parameters
G
[graph]
nodes
[container of nodes, optional (default=all nodes in G)] Compute clustering for nodes in this
container.
weight
Returns
out
[float, or dictionary] Clustering coefficient at specified nodes
Notes
Self loops are ignored.
References
[1], [2], [3], [4]
Examples
>>> print(nx.clustering(G, 0))
1.0
>>> print(nx.clustering(G))
{0: 1.0, 1: 1.0, 2: 1.0, 3: 1.0, 4: 1.0}

3.11.4 average_clustering
average_clustering(G, nodes=None, weight=None, count_zeros=True)

Compute the average clustering coefficient for the graph G.
The clustering coefficient for the graph is the average,
1 ∑
C= cv ,
n
v∈G

Parameters
G
[graph]
nodes
[container of nodes, optional (default=all nodes in G)] Compute average clustering for nodes
in this container.
weight
count_zeros
[bool] If False include only the nodes with nonzero clustering in the average.
Returns
avg
[float] Average clustering
Notes
This is a space saving routine; it might be faster to use the clustering function to get a list and then take the average.
Self loops are ignored.
References
[1], [2]
Examples
>>> print(nx.average_clustering(G))
1.0

3.11.5 square_clustering
square_clustering(G, nodes=None)
Compute the squares clustering coefficient for nodes.
For each node return the fraction of possible squares that exist at the node [1]
∑ kv ∑ kv
w=u+1 qv (u, w)
C4 (v) = ∑kv ∑u=1
kv
,
u=1 w=u+1 [av (u, w) + qv (u, w)]
where qv (u, w) are the number of common neighbors of u and w other than v (ie squares), and av (u, w) =
(ku − (1 + qv (u, w) + θuv )) + (kw − (1 + qv (u, w) + θuw )), where θuw = 1 if u and w are connected and 0
otherwise. [2]
Parameters
G
[graph]
nodes
[container of nodes, optional (default=all nodes in G)] Compute clustering for nodes in this
container.
Returns
c4
[dictionary] A dictionary keyed by node with the square clustering coefficient value.
Notes
While C3 (v) (triangle clustering) gives the probability that two neighbors of node v are connected with each other,
C4 (v) is the probability that two neighbors of node v share a common neighbor different from v. This algorithm
can be applied to both bipartite and unipartite networks.
References
[1], [2]
Examples
>>> print(nx.square_clustering(G, 0))
1.0
>>> print(nx.square_clustering(G))
{0: 1.0, 1: 1.0, 2: 1.0, 3: 1.0, 4: 1.0}

3.11.6 generalized_degree
generalized_degree(G, nodes=None)
Compute the generalized degree for nodes.
For each node, the generalized degree shows how many edges of given triangle multiplicity the node is connected
to. The triangle multiplicity of an edge is the number of triangles an edge participates in. The generalized degree
(0) (N −2) (j)
of node i can be written as a vector ki = (ki , . . . , ki ) where ki is the number of edges attached to node
i that participate in j triangles.
Parameters
G
[graph]
nodes
[container of nodes, optional (default=all nodes in G)] Compute the generalized degree for
nodes in this container.
Returns
out
[Counter, or dictionary of Counters] Generalized degree of specified nodes. The Counter is
keyed by edge triangle multiplicity.
Notes
In a network of N nodes, the highest triangle multiplicty an edge can have is N-2.
The return value does not include a zero entry if no edges of a particular triangle multiplicity are present.
(0) (N −2)
The number of triangles node i is attached to can be recovered from the generalized degree ki = (ki , . . . , ki )
(1) (2) (N −2)
by (ki + 2ki + . . . + (N − 2)ki )/2.
References
[1]
Examples
>>> print(nx.generalized_degree(G, 0))
Counter({3: 4})
>>> print(nx.generalized_degree(G))
{0: Counter({3: 4}), 1: Counter({3: 4}), 2: Counter({3: 4}), 3: Counter({3: 4}),␣
,→4: Counter({3: 4})}
To recover the number of triangles attached to a node:
>>> k1 = nx.generalized_degree(G, 0)
>>> sum([k * v for k, v in k1.items()]) / 2 == nx.triangles(G, 0)
True

3.12 Coloring
greedy_color(G[, strategy, interchange]) Color a graph using various strategies of greedy graph col-
oring.
equitable_color(G, num_colors) Provides equitable (r + 1)-coloring for nodes of G in O(r
* n^2) time if deg(G) <= r.
3.12.1 greedy_color
greedy_color(G, strategy='largest_first', interchange=False)

Color a graph using various strategies of greedy graph coloring.
Attempts to color a graph using as few colors as possible, where no neighbours of a node can have same color as
the node itself. The given strategy determines the order in which nodes are colored.
The strategies are described in [1], and smallest-last is based on [2].
Parameters
G
[NetworkX graph]
strategy
[string or function(G, colors)] A function (or a string representing a function) that provides the
coloring strategy, by returning nodes in the ordering they should be colored. G is the graph,
and colors is a dictionary of the currently assigned colors, keyed by nodes. The function
must return an iterable over all the nodes in G.
If the strategy function is an iterator generator (that is, a function with yield statements),
keep in mind that the colors dictionary will be updated after each yield, since this function
chooses colors greedily.
If strategy is a string, it must be one of the following, each of which represents one of the
built-in strategy functions.
• 'largest_first'
• 'random_sequential'
• 'smallest_last'
• 'independent_set'
• 'connected_sequential_bfs'
• 'connected_sequential_dfs'
• 'connected_sequential' (alias for the previous strategy)
• 'saturation_largest_first'
• 'DSATUR' (alias for the previous strategy)
interchange: bool
Will use the color interchange algorithm described by [3] if set to True.
Note that saturation_largest_first and independent_set do not work with
interchange. Furthermore, if you use interchange with your own strategy function, you cannot
rely on the values in the colors argument.
3.12. Coloring 319

Returns
A dictionary with keys representing nodes and values representing
corresponding coloring.
Raises
If strategy is saturation_largest_first or independent_set and in-
terchange is True.
References
[1], [2], [3]
Examples
>>> d = nx.coloring.greedy_color(G, strategy="largest_first")
>>> d in [{0: 0, 1: 1, 2: 0, 3: 1}, {0: 1, 1: 0, 2: 1, 3: 0}]
True
3.12.2 equitable_color
equitable_color(G, num_colors)
Provides equitable (r + 1)-coloring for nodes of G in O(r * n^2) time if deg(G) <= r. The algorithm is described
in [1].
Attempts to color a graph using r colors, where no neighbors of a node can have same color as the node itself and
the number of nodes with each color differ by at most 1.
Parameters
G
[networkX graph] The nodes of this graph will be colored.
num_colors
[number of colors to use] This number must be at least one more than the maximum degree
of nodes in the graph.
Returns
A dictionary with keys representing nodes and values representing
corresponding coloring.
Raises
NetworkXAlgorithmError
If the maximum degree of the graph G is greater than num_colors.

References
[1]
Examples
>>> d = nx.coloring.equitable_color(G, num_colors=3)
>>> nx.algorithms.coloring.equitable_coloring.is_equitable(G, d)
True
Some node ordering strategies are provided for use with greedy_color().
strategy_connected_sequential(G, colors[, Returns an iterable over nodes in G in the order given by

...]) a breadth-first or depth-first traversal.
strategy_connected_sequential_dfs(G, Returns an iterable over nodes in G in the order given by
colors) a depth-first traversal.
strategy_connected_sequential_bfs(G, Returns an iterable over nodes in G in the order given by
colors) a breadth-first traversal.
strategy_independent_set(G, colors) Uses a greedy independent set removal strategy to deter-
mine the colors.
strategy_largest_first(G, colors) Returns a list of the nodes of G in decreasing order by
degree.
strategy_random_sequential(G, colors[, Returns a random permutation of the nodes of G as a list.
seed])
strategy_saturation_largest_first(G, Iterates over all the nodes of G in "saturation order" (also
colors) known as "DSATUR").
strategy_smallest_last(G, colors) Returns a deque of the nodes of G, "smallest" last.
3.12.3 strategy_connected_sequential
strategy_connected_sequential(G, colors, traversal='bfs')

Returns an iterable over nodes in G in the order given by a breadth-first or depth-first traversal.
traversal must be one of the strings 'dfs' or 'bfs', representing depth-first traversal or breadth-first traver-
sal, respectively.
The generated sequence has the property that for each node except the first, at least one neighbor appeared earlier
in the sequence.
G is a NetworkX graph. colors is ignored.
3.12.4 strategy_connected_sequential_dfs
strategy_connected_sequential_dfs(G, colors)
Returns an iterable over nodes in G in the order given by a depth-first traversal.
in the sequence.
3.12. Coloring 321

3.12.5 strategy_connected_sequential_bfs
strategy_connected_sequential_bfs(G, colors)
Returns an iterable over nodes in G in the order given by a breadth-first traversal.
in the sequence.
3.12.6 strategy_independent_set
strategy_independent_set(G, colors)
Uses a greedy independent set removal strategy to determine the colors.
This function updates colors in-place and return None, unlike the other strategy functions in this module.
This algorithm repeatedly finds and removes a maximal independent set, assigning each node in the set an unused
color.
G is a NetworkX graph.
This strategy is related to strategy_smallest_last(): in that strategy, an independent set of size one is
chosen at each step instead of a maximal independent set.
3.12.7 strategy_largest_first
strategy_largest_first(G, colors)
Returns a list of the nodes of G in decreasing order by degree.
3.12.8 strategy_random_sequential
strategy_random_sequential(G, colors, seed=None)

Returns a random permutation of the nodes of G as a list.
seed
[integer, random_state, or None (default)] Indicator of random number generation state. See Randomness.
3.12.9 strategy_saturation_largest_first
strategy_saturation_largest_first(G, colors)
Iterates over all the nodes of G in “saturation order” (also known as “DSATUR”).
G is a NetworkX graph. colors is a dictionary mapping nodes of G to colors, for those nodes that have already
been colored.

3.12.10 strategy_smallest_last
strategy_smallest_last(G, colors)
Returns a deque of the nodes of G, “smallest” last.
Specifically, the degrees of each node are tracked in a bucket queue. From this, the node of minimum degree is
repeatedly popped from the graph, updating its neighbors’ degrees.
This implementation of the strategy runs in O(n + m) time (ignoring polylogarithmic factors), where n is the
number of nodes and m is the number of edges.
This strategy is related to strategy_independent_set(): if we interpret each node removed as an inde-
pendent set of size one, then this strategy chooses an independent set of size one instead of a maximal independent
set.
3.13 Communicability
Communicability.
communicability(G) Returns communicability between all pairs of nodes in G.

communicability_exp(G) Returns communicability between all pairs of nodes in G.
3.13.1 communicability
communicability(G)
Returns communicability between all pairs of nodes in G.
The communicability between pairs of nodes in G is the sum of walks of different lengths starting at node u and
ending at node v.
Parameters
G: graph
Returns
comm: dictionary of dictionaries
Dictionary of dictionaries keyed by nodes with communicability as the value.
Raises
NetworkXError
See also:
communicability_exp
Communicability between all pairs of nodes in G using spectral decomposition.
Communicability betweeness centrality for each node in G.
3.13. Communicability 323

Notes
This algorithm uses a spectral decomposition of the adjacency matrix. Let G=(V,E) be a simple undirected graph.
Using the connection between the powers of the adjacency matrix and the number of walks in the graph, the
communicability between nodes u and v based on the graph spectrum is [1]
∑
n
C(u, v) = ϕj (u)ϕj (v)eλj ,
j=1
where phi_{j}(u) is the urm{th} element of the jrm{th} orthonormal eigenvector of the adjacency matrix
associated with the eigenvalue lambda_{j}.
References
[1]
Examples
>>> G = nx.Graph([(0, 1), (1, 2), (1, 5), (5, 4), (2, 4), (2, 3), (4, 3), (3, 6)])
>>> c = nx.communicability(G)
3.13.2 communicability_exp
communicability_exp(G)
Returns communicability between all pairs of nodes in G.
Communicability between pair of node (u,v) of node in G is the sum of walks of different lengths starting at node
u and ending at node v.
Parameters
G: graph
Returns
comm: dictionary of dictionaries
Dictionary of dictionaries keyed by nodes with communicability as the value.
Raises
NetworkXError
See also:
communicability
Communicability between pairs of nodes in G.
Communicability betweeness centrality for each node in G.

Notes
This algorithm uses matrix exponentiation of the adjacency matrix.

Let G=(V,E) be a simple undirected graph. Using the connection between the powers of the adjacency matrix and
the number of walks in the graph, the communicability between nodes u and v is [1],
C(u, v) = (eA )uv ,
where A is the adjacency matrix of G.
References
[1]
Examples
>>> G = nx.Graph([(0, 1), (1, 2), (1, 5), (5, 4), (2, 4), (2, 3), (4, 3), (3, 6)])
>>> c = nx.communicability_exp(G)
3.14 Communities
Functions for computing and measuring community structure.

The functions in this class are not imported into the top-level networkx namespace. You can access these functions by
importing the networkx.algorithms.community module, then accessing the functions as attributes of com-
munity. For example:
>>> from networkx.algorithms import community

>>> communities_generator = community.girvan_newman(G)
>>> top_level_communities = next(communities_generator)
>>> next_level_communities = next(communities_generator)
>>> sorted(map(sorted, next_level_communities))
[[0, 1, 2, 3, 4], [5], [6, 7, 8, 9, 10]]
3.14.1 Bipartitions
Functions for computing the Kernighan–Lin bipartition algorithm.
kernighan_lin_bisection(G[, partition, ...]) Partition a graph into two blocks using the Kernighan–Lin
algorithm.
3.14. Communities 325

kernighan_lin_bisection
kernighan_lin_bisection(G, partition=None, max_iter=10, weight='weight', seed=None)

Partition a graph into two blocks using the Kernighan–Lin algorithm.
This algorithm partitions a network into two sets by iteratively swapping pairs of nodes to reduce the edge cut
between the two sets. The pairs are chosen according to a modified form of Kernighan-Lin, which moves node
individually, alternating between sides to keep the bisection balanced.
Parameters
G
[graph]
partition
[tuple] Pair of iterables containing an initial partition. If not specified, a random balanced
partition is used.
max_iter
[int] Maximum number of times to attempt swaps to find an improvemement before giving up.
weight
[key] Edge data key to use as weight. If None, the weights are all set to one.
seed
Randomness. Only used if partition is None
Returns
partition
[tuple] A pair of sets of nodes representing the bipartition.
Raises
NetworkXError
If partition is not a valid partition of the nodes of the graph.
References
[1]
3.14.2 K-Clique
k_clique_communities(G, k[, cliques]) Find k-clique communities in graph using the percolation
method.

k_clique_communities
k_clique_communities(G, k, cliques=None)
Find k-clique communities in graph using the percolation method.
A k-clique community is the union of all cliques of size k that can be reached through adjacent (sharing k-1 nodes)
k-cliques.
Parameters
G
[NetworkX graph]
k
[int] Size of smallest clique
cliques: list or generator
Precomputed cliques (use networkx.find_cliques(G))
Returns
Yields sets of nodes, one for each k-clique community.
References
[1]
Examples
>>> from networkx.algorithms.community import k_clique_communities

>>> K5 = nx.convert_node_labels_to_integers(G, first_label=2)
>>> G.add_edges_from(K5.edges())
>>> c = list(k_clique_communities(G, 4))
>>> sorted(list(c[0]))
[0, 1, 2, 3, 4, 5, 6]
>>> list(k_clique_communities(G, 6))
[]
3.14.3 Modularity-based communities
Functions for detecting communities based on modularity.
greedy_modularity_communities(G[, weight, Find communities in G using greedy modularity maxi-

...]) mization.
naive_greedy_modularity_communities(G[, Find communities in G using greedy modularity maxi-
...]) mization.

greedy_modularity_communities
greedy_modularity_communities(G, weight=None, resolution=1, cutoff=1, best_n=None)

Find communities in G using greedy modularity maximization.
This function uses Clauset-Newman-Moore greedy modularity maximization [2] to find the community partition
with the largest modularity.
Greedy modularity maximization begins with each node in its own community and repeatedly joins the pair of
communities that lead to the largest modularity until no futher increase in modularity is possible (a maximum).
Two keyword arguments adjust the stopping condition. cutoff is a lower limit on the number of communities so
you can stop the process before reaching a maximum (used to save computation time). best_n is an upper limit
on the number of communities so you can make the process continue until at most n communities remain even if
the maximum modularity occurs for more. To obtain exactly n communities, set both cutoff and best_n to n.
This function maximizes the generalized modularity, where resolution is the resolution parameter, often ex-
pressed as γ. See modularity().
Parameters
G
[NetworkX graph]
weight
resolution
[float, optional (default=1)] If resolution is less than 1, modularity favors larger communities.
Greater than 1 favors smaller communities.
cutoff
[int, optional (default=1)] A minimum number of communities below which the merging pro-
cess stops. The process stops at this number of communities even if modularity is not maxi-
mized. The goal is to let the user stop the process early. The process stops before the cutoff if
it finds a maximum of modularity.
best_n
[int or None, optional (default=None)] A maximum number of communities above which the
merging process will not stop. This forces community merging to continue after modularity
starts to decrease until best_n communities remain. If None, don’t force it to continue
beyond a maximum.
Returns
communities: list
A list of frozensets of nodes, one for each community. Sorted by length with largest commu-
nities first.
Raises
ValueError
[If the cutoff or best_n value is not in the range] [1, G.number_of_nodes()],
or if best_n < cutoff.
See also:
modularity

References
[1], [2], [3], [4]
Examples
>>> from networkx.algorithms.community import greedy_modularity_communities

>>> G = nx.karate_club_graph()
>>> c = greedy_modularity_communities(G)
>>> sorted(c[0])
[8, 14, 15, 18, 20, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33]
naive_greedy_modularity_communities
naive_greedy_modularity_communities(G, resolution=1, weight=None)

Find communities in G using greedy modularity maximization.
This implementation is O(n^4), much slower than alternatives, but it is provided as an easy-to-understand reference
implementation.
Greedy modularity maximization begins with each node in its own community and joins the pair of communities
that most increases modularity until no such pair exists.
This function maximizes the generalized modularity, where resolution is the resolution parameter, often ex-
pressed as γ. See modularity().
Parameters
G
[NetworkX graph]
resolution
[float (default=1)] If resolution is less than 1, modularity favors larger communities. Greater
than 1 favors smaller communities.
weight
Returns
list
A list of sets of nodes, one for each community. Sorted by length with largest communities
first.
See also:
greedy_modularity_communities
modularity

Examples
>>> from networkx.algorithms.community import \

... naive_greedy_modularity_communities
>>> c = naive_greedy_modularity_communities(G)
>>> sorted(c[0])
[8, 14, 15, 18, 20, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33]
3.14.4 Tree partitioning
Lukes Algorithm for exact optimal weighted tree partitioning.
lukes_partitioning(G, max_size[, ...]) Optimal partitioning of a weighted tree using the Lukes
algorithm.
lukes_partitioning
lukes_partitioning(G, max_size, node_weight=None, edge_weight=None)

Optimal partitioning of a weighted tree using the Lukes algorithm.
This algorithm partitions a connected, acyclic graph featuring integer node weights and float edge weights. The
resulting clusters are such that the total weight of the nodes in each cluster does not exceed max_size and that the
weight of the edges that are cut by the partition is minimum. The algorithm is based on LUKES[1].
Parameters
G
[graph]
max_size
[int] Maximum weight a partition can have in terms of sum of node_weight for all nodes in the
partition
edge_weight
[key] Edge data key to use as weight. If None, the weights are all set to one.
node_weight
[key] Node data key to use as weight. If None, the weights are all set to one. The data must
be int.
Returns
partition
[list] A list of sets of nodes representing the clusters of the partition.
Raises
NotATree
If G is not a tree.
TypeError
If any of the values of node_weight is not int.

References
3.14.5 Label propagation
Label propagation community detection algorithms.
asyn_lpa_communities(G[, weight, seed]) Returns communities in G as detected by asynchronous

label propagation.
label_propagation_communities(G) Generates community sets determined by label propaga-
tion
asyn_lpa_communities
asyn_lpa_communities(G, weight=None, seed=None)

Returns communities in G as detected by asynchronous label propagation.
The asynchronous label propagation algorithm is described in [1]. The algorithm is probabilistic and the found
communities may vary on different executions.
The algorithm proceeds as follows. After initializing each node with a unique label, the algorithm repeatedly sets
the label of a node to be the label that appears most frequently among that nodes neighbors. The algorithm halts
when each node has the label that appears most frequently among its neighbors. The algorithm is asynchronous
because each node is updated without waiting for updates on the remaining nodes.
This generalized version of the algorithm in [1] accepts edge weights.
Parameters
G
[Graph]
weight
[string] The edge attribute representing the weight of an edge. If None, each edge is assumed to
have weight one. In this algorithm, the weight of an edge is used in determining the frequency
with which a label appears among the neighbors of a node: a higher weight means the label
appears more often.
seed
Randomness.
Returns
communities
[iterable] Iterable of communities given as sets of nodes.

Notes
Edge weight attributes must be numerical.
References
[1]
label_propagation_communities
label_propagation_communities(G)
Generates community sets determined by label propagation
Finds communities in G using a semi-synchronous label propagation method [1]. This method combines the ad-
vantages of both the synchronous and asynchronous models. Not implemented for directed graphs.
Parameters
G
[graph] An undirected NetworkX graph.
Returns
communities
[iterable] A dict_values object that contains a set of nodes for each community.
Raises
If the graph is directed
References
[1]
3.14.6 Louvain Community Detection
Function for detecting communities based on Louvain Community Detection Algorithm
louvain_communities(G[, weight, resolution, ...]) Find the best partition of a graph using the Louvain Com-
munity Detection Algorithm.
louvain_partitions(G[, weight, resolution, ...]) Yields partitions for each level of the Louvain Community
Detection Algorithm

louvain_communities
louvain_communities(G, weight='weight', resolution=1, threshold=1e-07, seed=None)

Find the best partition of a graph using the Louvain Community Detection Algorithm.
Louvain Community Detection Algorithm is a simple method to extract the community structure of a network.
This is a heuristic method based on modularity optimization. [1]
The algorithm works in 2 steps. On the first step it assigns every node to be in its own community and then for each
node it tries to find the maximum positive modularity gain by moving each node to all of its neighbor communities.
If no positive gain is achieved the node remains in its original community.
The modularity gain obtained by moving an isolated node i into a community C can easily be calculated by the
following formula (combining [1] [2] and some algebra):
ki,in Σtot · ki
∆Q = −γ
2m 2m2
where m is the size of the graph, ki,in is the sum of the weights of the links from i to nodes in C, ki is the sum of
the weights of the links incident to node i, Σtot is the sum of the weights of the links incident to nodes in C and γ
is the resolution parameter.
For the directed case the modularity gain can be computed using this formula according to [3]
ki,in tot + ki · Σtot

k out · Σin in out
∆Q = −γ i
m m2
where kiout , kiin are the outer and inner weighted degrees of node i and Σin out
tot , Σtot are the sum of in-going and
out-going links incident to nodes in C.
The first phase continues until no individual move can improve the modularity.
The second phase consists in building a new network whose nodes are now the communities found in the first phase.
To do so, the weights of the links between the new nodes are given by the sum of the weight of the links between
nodes in the corresponding two communities. Once this phase is complete it is possible to reapply the first phase
creating bigger communities with increased modularity.
The above two phases are executed until no modularity gain is achieved (or is less than the threshold).
Parameters
G
[NetworkX graph]
weight
[string or None, optional (default=”weight”)] The name of an edge attribute that holds the
numerical value used as a weight. If None then each edge has weight 1.
resolution
[float, optional (default=1)] If resolution is less than 1, the algorithm favors larger communities.
Greater than 1 favors smaller communities
threshold
[float, optional (default=0.0000001)] Modularity gain threshold for each level. If the gain of
modularity between 2 levels of the algorithm is less than the given threshold then the algorithm
stops and returns the resulting communities.
seed
Randomness.
Returns

list
A list of sets (partition of G). Each set represents one community and contains all the nodes
that constitute it.
See also:
louvain_partitions
Notes
The order in which the nodes are considered can affect the final output. In the algorithm the ordering happens using
a random shuffle.
References
[1], [2], [3]
Examples

>>> import networkx.algorithms.community as nx_comm
>>> nx_comm.louvain_communities(G, seed=123)
[{0, 4, 5, 7, 9}, {1, 2, 3, 6, 8}]
louvain_partitions
louvain_partitions(G, weight='weight', resolution=1, threshold=1e-07, seed=None)

Yields partitions for each level of the Louvain Community Detection Algorithm
Louvain Community Detection Algorithm is a simple method to extract the community structure of a network.
This is a heuristic method based on modularity optimization. [1]
The partitions at each level (step of the algorithm) form a dendogram of communities. A dendrogram is a diagram
representing a tree and each level represents a partition of the G graph. The top level contains the smallest commu-
nities and as you traverse to the bottom of the tree the communities get bigger and the overal modularity increases
making the partition better.
Each level is generated by executing the two phases of the Louvain Community Detection Algorithm.
Parameters
G
[NetworkX graph]
weight
[string or None, optional (default=”weight”)] The name of an edge attribute that holds the
numerical value used as a weight. If None then each edge has weight 1.
resolution
[float, optional (default=1)] If resolution is less than 1, the algorithm favors larger communities.
Greater than 1 favors smaller communities

threshold
[float, optional (default=0.0000001)] Modularity gain threshold for each level. If the gain of
modularity between 2 levels of the algorithm is less than the given threshold then the algorithm
stops and returns the resulting communities.
seed
Randomness.
Yields
list
A list of sets (partition of G). Each set represents one community and contains all the nodes
that constitute it.
See also:
louvain_communities
References
[1]
3.14.7 Fluid Communities
Asynchronous Fluid Communities algorithm for community detection.
asyn_fluidc(G, k[, max_iter, seed]) Returns communities in G as detected by Fluid Commu-

nities algorithm.
asyn_fluidc
asyn_fluidc(G, k, max_iter=100, seed=None)

Returns communities in G as detected by Fluid Communities algorithm.
The asynchronous fluid communities algorithm is described in [1]. The algorithm is based on the simple idea
of fluids interacting in an environment, expanding and pushing each other. Its initialization is random, so found
communities may vary on different executions.
The algorithm proceeds as follows. First each of the initial k communities is initialized in a random vertex in the
graph. Then the algorithm iterates over all vertices in a random order, updating the community of each vertex
based on its own community and the communities of its neighbours. This process is performed several times until
convergence. At all times, each community has a total density of 1, which is equally distributed among the vertices
it contains. If a vertex changes of community, vertex densities of affected communities are adjusted immediately.
When a complete iteration over all vertices is done, such that no vertex changes the community it belongs to, the
algorithm has converged and returns.
This is the original version of the algorithm described in [1]. Unfortunately, it does not support weighted graphs
yet.
Parameters
G
[Graph]

k
[integer] The number of communities to be found.
max_iter
[integer] The number of maximum iterations allowed. By default 100.
seed
Randomness.
Returns
communities
[iterable] Iterable of communities given as sets of nodes.
Notes
k variable is not an optional argument.
References
[1]
3.14.8 Measuring partitions
Functions for measuring the quality of a partition (into communities).
modularity(G, communities[, weight, resolution]) Returns the modularity of the given partition of the graph.
partition_quality(G, partition) Returns the coverage and performance of a partition of
G.
modularity
modularity(G, communities, weight='weight', resolution=1)

Returns the modularity of the given partition of the graph.
Modularity is defined in [1] as
( )
1 ∑ ki kj
Q= Aij − γ δ(ci , cj )
2m ij 2m
where m is the number of edges, A is the adjacency matrix of G, ki is the degree of i, γ is the resolution parameter,
and δ(ci , cj ) is 1 if i and j are in the same community else 0.
According to [2] (and verified by some algebra) this can be reduced to
[ ( )2 ]
∑n
Lc kc
Q= −γ
c=1
m 2m
where the sum iterates over all communities c, m is the number of edges, Lc is the number of intra-community
links for community c, kc is the sum of degrees of the nodes in community c, and γ is the resolution parameter.
The resolution parameter sets an arbitrary tradeoff between intra-group edges and inter-group edges. More com-
plex grouping patterns can be discovered by analyzing the same network with multiple values of gamma and then

combining the results [3]. That said, it is very common to simply use gamma=1. More on the choice of gamma is
in [4].
The second formula is the one actually used in calculation of the modularity. For directed graphs the second formula
replaces kc with kcin kcout .
Parameters
G
[NetworkX Graph]
communities
[list or iterable of set of nodes] These node sets must represent a partition of G’s nodes.
weight
[string or None, optional (default=”weight”)] The edge attribute that holds the numerical value
used as a weight. If None or an edge does not have that attribute, then that edge has weight 1.
resolution
[float (default=1)] If resolution is less than 1, modularity favors larger communities. Greater
than 1 favors smaller communities.
Returns
Q
[float] The modularity of the paritition.
Raises
NotAPartition
If communities is not a partition of the nodes of G.
References
[1], [2], [3], [4]
Examples
>>> import networkx.algorithms.community as nx_comm

>>> nx_comm.modularity(G, [{0, 1, 2}, {3, 4, 5}])
0.35714285714285715
>>> nx_comm.modularity(G, nx_comm.label_propagation_communities(G))
0.35714285714285715
partition_quality
partition_quality(G, partition)
Returns the coverage and performance of a partition of G.
The coverage of a partition is the ratio of the number of intra-community edges to the total number of edges in the
graph.
The performance of a partition is the number of intra-community edges plus inter-community non-edges divided
by the total number of potential edges.
This algorithm has complexity O(C 2 + L) where C is the number of communities and L is the number of links.

Parameters
G
[NetworkX graph]
partition
[sequence] Partition of the nodes of G, represented as a sequence of sets of nodes (blocks).
Each block of the partition represents a community.
Returns
(float, float)
The (coverage, performance) tuple of the partition, as defined above.
Raises
NetworkXError
If partition is not a valid partition of the nodes of G.
Notes
If G is a multigraph;
• for coverage, the multiplicity of edges is counted
• for performance, the result is -1 (total number of possible edges is not defined)
References
[1]
3.14.9 Partitions via centrality measures
Functions for computing communities based on centrality notions.
girvan_newman(G[, most_valuable_edge]) Finds communities in a graph using the Girvan–Newman

method.
girvan_newman
girvan_newman(G, most_valuable_edge=None)
Finds communities in a graph using the Girvan–Newman method.
Parameters
G
[NetworkX graph]
most_valuable_edge
[function] Function that takes a graph as input and outputs an edge. The edge returned by this
function will be recomputed and removed at each iteration of the algorithm.
If not specified, the edge with the highest networkx.
edge_betweenness_centrality() will be used.
Returns

iterator
Iterator over tuples of sets of nodes in G. Each set of node is a community, each tuple is a
sequence of communities at a particular level of the algorithm.
Notes
The Girvan–Newman algorithm detects communities by progressively removing edges from the original graph. The
algorithm removes the “most valuable” edge, traditionally the edge with the highest betweenness centrality, at each
step. As the graph breaks down into pieces, the tightly knit community structure is exposed and the result can be
depicted as a dendrogram.
Examples
To get the first pair of communities:

>>> comp = girvan_newman(G)
>>> tuple(sorted(c) for c in next(comp))
([0, 1, 2, 3, 4], [5, 6, 7, 8, 9])
To get only the first k tuples of communities, use itertools.islice():

>>> import itertools
>>> k = 2
>>> for communities in itertools.islice(comp, k):
... print(tuple(sorted(c) for c in communities))
...
([0, 1, 2, 3], [4, 5, 6, 7])
([0, 1], [2, 3], [4, 5, 6, 7])
To stop getting tuples of communities once the number of communities is greater than k, use itertools.
takewhile():
>>> k = 4
>>> limited = itertools.takewhile(lambda c: len(c) <= k, comp)
>>> for communities in limited:
... print(tuple(sorted(c) for c in communities))
...
([0, 1, 2, 3], [4, 5, 6, 7])
([0, 1], [2, 3], [4, 5, 6, 7])
([0, 1], [2, 3], [4, 5], [6, 7])
To just choose an edge to remove based on the weight:

>>> from operator import itemgetter
>>> edges = G.edges()
>>> nx.set_edge_attributes(G, {(u, v): v for u, v in edges}, "weight")
>>> def heaviest(G):
... u, v, w = max(G.edges(data="weight"), key=itemgetter(2))


... return (u, v)
...
>>> comp = girvan_newman(G, most_valuable_edge=heaviest)
([0, 1, 2, 3, 4, 5, 6, 7, 8], [9])
To utilize edge weights when choosing an edge with, for example, the highest betweenness centrality:
>>> from networkx import edge_betweenness_centrality as betweenness

>>> def most_central_edge(G):
... centrality = betweenness(G, weight="weight")
... return max(centrality, key=centrality.get)
...
>>> comp = girvan_newman(G, most_valuable_edge=most_central_edge)
([0, 1, 2, 3, 4], [5, 6, 7, 8, 9])
To specify a different ranking algorithm for edges, use the most_valuable_edge keyword argument:
>>> from networkx import edge_betweenness_centrality

>>> from random import random
>>> def most_central_edge(G):
... centrality = edge_betweenness_centrality(G)
... max_cent = max(centrality.values())
... # Scale the centrality values so they are between 0 and 1,
... # and add some random noise.
... centrality = {e: c / max_cent for e, c in centrality.items()}
... # Add some random noise.
... centrality = {e: c + random() for e, c in centrality.items()}
... return max(centrality, key=centrality.get)
...
>>> comp = girvan_newman(G, most_valuable_edge=most_central_edge)
3.14.10 Validating partitions
Helper functions for community-finding algorithms.
is_partition(G, communities) Returns True if communities is a partition of the

nodes of G.
is_partition
is_partition(G, communities)
Returns True if communities is a partition of the nodes of G.
A partition of a universe set is a family of pairwise disjoint sets whose union is the entire universe set.
Parameters
G
[NetworkX graph.]

communities
[list or iterable of sets of nodes] If not a list, the iterable is converted internally to a list. If it
is an iterator it is exhausted.
3.15 Components
3.15.1 Connectivity
is_connected(G) Returns True if the graph is connected, False otherwise.

number_connected_components(G) Returns the number of connected components.
connected_components(G) Generate connected components.
node_connected_component(G, n) Returns the set of nodes in the component of graph con-
taining node n.
is_connected
is_connected(G)
Returns True if the graph is connected, False otherwise.
Parameters
G
[NetworkX Graph] An undirected graph.
Returns
connected
[bool] True if the graph is connected, false otherwise.
Raises
If G is directed.
See also:
is_strongly_connected
is_weakly_connected
is_semiconnected
is_biconnected
connected_components
Notes
For undirected graphs only.
3.15. Components 341

Examples
>>> print(nx.is_connected(G))
True
number_connected_components
number_connected_components(G)
Returns the number of connected components.
Parameters
G
Returns
n
[integer] Number of connected components
See also:
number_weakly_connected_components
number_strongly_connected_components
Notes
Examples
>>> G = nx.Graph([(0, 1), (1, 2), (5, 6), (3, 4)])

>>> nx.number_connected_components(G)
3
connected_components(G)
Generate connected components.
Parameters
G
[NetworkX graph] An undirected graph
Returns
comp
[generator of sets] A generator of sets of nodes, one for each component of G.
Raises

If G is directed.
See also:
strongly_connected_components
weakly_connected_components
Notes
Examples
Generate a sorted list of connected components, largest first.
>>> nx.add_path(G, [10, 11, 12])
>>> [len(c) for c in sorted(nx.connected_components(G), key=len, reverse=True)]
[4, 3]
If you only want the largest connected component, it’s more efficient to use max instead of sort.
>>> largest_cc = max(nx.connected_components(G), key=len)
To create the induced subgraph of each component use:
>>> S = [G.subgraph(c).copy() for c in nx.connected_components(G)]
node_connected_component
node_connected_component(G, n)
Returns the set of nodes in the component of graph containing node n.
Parameters
G
n
[node label] A node in G
Returns
comp
[set] A set of nodes in the component of G containing node n.
Raises
If G is directed.
See also:

Notes
Examples
>>> G = nx.Graph([(0, 1), (1, 2), (5, 6), (3, 4)])

>>> nx.node_connected_component(G, 0) # nodes of component that contains node 0
{0, 1, 2}
3.15.2 Strong connectivity
is_strongly_connected(G) Test directed graph for strong connectivity.

number_strongly_connected_components(G) Returns number of strongly connected components in
graph.
strongly_connected_components(G) Generate nodes in strongly connected components of
graph.
strongly_connected_components_recursive(G)
Generate nodes in strongly connected components of
graph.
kosaraju_strongly_connected_components(G[,
Generate nodes in strongly connected components of
...]) graph.
condensation(G[, scc]) Returns the condensation of G.
is_strongly_connected(G)
Test directed graph for strong connectivity.
A directed graph is strongly connected if and only if every vertex in the graph is reachable from every other vertex.
Parameters
G
[NetworkX Graph] A directed graph.
Returns
connected
[bool] True if the graph is strongly connected, False otherwise.
Raises
If G is undirected.
See also:
is_weakly_connected
is_semiconnected
is_connected
is_biconnected

Notes
For directed graphs only.
Examples
>>> G = nx.DiGraph([(0, 1), (1, 2), (2, 3), (3, 0), (2, 4), (4, 2)])
>>> nx.is_strongly_connected(G)
True
>>> nx.is_strongly_connected(G)
False
number_strongly_connected_components(G)
Returns number of strongly connected components in graph.
Parameters
G
[NetworkX graph] A directed graph.
Returns
n
[integer] Number of strongly connected components
Raises
If G is undirected.
See also:
Notes
Examples
>>> G = nx.DiGraph([(0, 1), (1, 2), (2, 0), (2, 3), (4, 5), (3, 4), (5, 6), (6,␣
,→3), (6, 7)])
>>> nx.number_strongly_connected_components(G)
3

strongly_connected_components(G)
Generate nodes in strongly connected components of graph.
Parameters
G
Returns
comp
[generator of sets] A generator of sets of nodes, one for each strongly connected component
of G.
Raises
If G is undirected.
See also:
kosaraju_strongly_connected_components
Notes
Uses Tarjan’s algorithm[R827335e01166-1]_ with Nuutila’s modifications[R827335e01166-2]_. Nonrecursive

version of algorithm.
References
[1], [2]
Examples
Generate a sorted list of strongly connected components, largest first.
>>> G = nx.cycle_graph(4, create_using=nx.DiGraph())

>>> nx.add_cycle(G, [10, 11, 12])
>>> [
... len(c)
... for c in sorted(nx.strongly_connected_components(G), key=len,␣
,→reverse=True)
... ]
[4, 3]
If you only want the largest component, it’s more efficient to use max instead of sort.
>>> largest = max(nx.strongly_connected_components(G), key=len)

strongly_connected_components_recursive
strongly_connected_components_recursive(G)
Recursive version of algorithm.
Parameters
G
Returns
comp
of G.
Raises
If G is undirected.
See also:
Notes
Uses Tarjan’s algorithm[Re7cb971df765-1]_ with Nuutila’s modifications[Re7cb971df765-2]_.
References
[1], [2]
Examples

>>> nx.add_cycle(G, [10, 11, 12])
>>> [
... len(c)
... for c in sorted(
... nx.strongly_connected_components_recursive(G), key=len, reverse=True
... )
... ]
[4, 3]
>>> largest = max(nx.strongly_connected_components_recursive(G), key=len)
To create the induced subgraph of the components use: >>> S = [G.subgraph(c).copy() for c in
nx.weakly_connected_components(G)]

kosaraju_strongly_connected_components
kosaraju_strongly_connected_components(G, source=None)
Parameters
G
Returns
comp
of G.
Raises
If G is undirected.
See also:
Notes
Uses Kosaraju’s algorithm.
Examples

>>> nx.add_cycle(G, [10, 11, 12])
>>> [
... len(c)
... for c in sorted(
... nx.kosaraju_strongly_connected_components(G), key=len, reverse=True
... )
... ]
[4, 3]
>>> largest = max(nx.kosaraju_strongly_connected_components(G), key=len)

condensation
condensation(G, scc=None)
Returns the condensation of G.
The condensation of G is the graph with each of the strongly connected components contracted into a single node.
Parameters
G
[NetworkX DiGraph] A directed graph.
scc: list or generator (optional, default=None)
Strongly connected components. If provided, the elements in scc must partition the nodes in
G. If not provided, it will be calculated as scc=nx.strongly_connected_components(G).
Returns
C
[NetworkX DiGraph] The condensation graph C of G. The node labels are integers corre-
sponding to the index of the component in the list of strongly connected components of G.
C has a graph attribute named ‘mapping’ with a dictionary mapping the original nodes to the
nodes in C to which they belong. Each node in C also has a node attribute ‘members’ with the
set of original nodes in G that form the SCC that the node in C represents.
Raises
If G is undirected.
Notes
After contracting all strongly connected components to a single node, the resulting graph is a directed acyclic graph.
Examples
Contracting two sets of strongly connected nodes into two distinct SCC using the barbell graph.
>>> G = nx.DiGraph(G)
>>> H = nx.condensation(G)
>>> H.nodes.data()
NodeDataView({0: {'members': {0, 1, 2, 3}}, 1: {'members': {4, 5, 6, 7}}})
>>> H.graph['mapping']
{0: 0, 1: 0, 2: 0, 3: 0, 4: 1, 5: 1, 6: 1, 7: 1}
Contracting a complete graph into one single SCC.
>>> G = nx.complete_graph(7, create_using=nx.DiGraph)

>>> H = nx.condensation(G)
>>> H.nodes
NodeView((0,))
>>> H.nodes.data()
NodeDataView({0: {'members': {0, 1, 2, 3, 4, 5, 6}}})

3.15.3 Weak connectivity
is_weakly_connected(G) Test directed graph for weak connectivity.

number_weakly_connected_components(G) Returns the number of weakly connected components in
G.
weakly_connected_components(G) Generate weakly connected components of G.
is_weakly_connected
is_weakly_connected(G)
Test directed graph for weak connectivity.
A directed graph is weakly connected if and only if the graph is connected when the direction of the edge between
nodes is ignored.
Note that if a graph is strongly connected (i.e. the graph is connected even when we account for directionality), it
is by definition weakly connected as well.
Parameters
G
Returns
connected
[bool] True if the graph is weakly connected, False otherwise.
Raises
If G is undirected.
See also:
is_semiconnected
is_connected
is_biconnected
Notes

Examples
>>> G = nx.DiGraph([(0, 1), (2, 1)])

>>> G.add_node(3)
>>> nx.is_weakly_connected(G) # node 3 is not connected to the graph
False
>>> nx.is_weakly_connected(G)
True
number_weakly_connected_components(G)
Returns the number of weakly connected components in G.
Parameters
G
Returns
n
[integer] Number of weakly connected components
Raises
If G is undirected.
See also:
Notes
Examples
>>> G = nx.DiGraph([(0, 1), (2, 1), (3, 4)])

>>> nx.number_weakly_connected_components(G)
2

weakly_connected_components(G)
Generate weakly connected components of G.
Parameters
G
[NetworkX graph] A directed graph
Returns
comp
[generator of sets] A generator of sets of nodes, one for each weakly connected component of
G.
Raises
If G is undirected.
See also:
Notes
Examples
Generate a sorted list of weakly connected components, largest first.
>>> G = nx.path_graph(4, create_using=nx.DiGraph())

>>> nx.add_path(G, [10, 11, 12])
>>> [
... len(c)
... for c in sorted(nx.weakly_connected_components(G), key=len, reverse=True)
... ]
[4, 3]
If you only want the largest component, it’s more efficient to use max instead of sort:
>>> largest_cc = max(nx.weakly_connected_components(G), key=len)

3.15.4 Attracting components
is_attracting_component(G) Returns True if G consists of a single attracting compo-

nent.
number_attracting_components(G) Returns the number of attracting components in G.
attracting_components(G) Generates the attracting components in G.
is_attracting_component
is_attracting_component(G)
Returns True if G consists of a single attracting component.
Parameters
G
[DiGraph, MultiDiGraph] The graph to be analyzed.
Returns
attracting
[bool] True if G has a single attracting component. Otherwise, False.
Raises
If the input graph is undirected.
See also:
attracting_components
number_attracting_components
number_attracting_components(G)
Returns the number of attracting components in G.
Parameters
G
Returns
n
[int] The number of attracting components in G.
Raises
See also:

attracting_components(G)
Generates the attracting components in G.
An attracting component in a directed graph G is a strongly connected component with the property that a random
walker on the graph will never leave the component, once it enters the component.
The nodes in attracting components can also be thought of as recurrent nodes. If a random walker enters the
attractor containing the node, then the node will be visited infinitely often.
To obtain induced subgraphs on each component use: (G.subgraph(c).copy() for c in attract-
ing_components(G))
Parameters
G
Returns
attractors
[generator of sets] A generator of sets of nodes, one for each attracting component of G.
Raises
See also:
3.15.5 Biconnected components
is_biconnected(G) Returns True if the graph is biconnected, False otherwise.

biconnected_components(G) Returns a generator of sets of nodes, one set for each bi-
connected component of the graph
biconnected_component_edges(G) Returns a generator of lists of edges, one list for each bi-
connected component of the input graph.
articulation_points(G) Yield the articulation points, or cut vertices, of a graph.
is_biconnected
is_biconnected(G)
Returns True if the graph is biconnected, False otherwise.
A graph is biconnected if, and only if, it cannot be disconnected by removing only one node (and all edges incident
on that node). If removing a node increases the number of disconnected components in the graph, that node is
called an articulation point, or cut vertex. A biconnected graph has no articulation points.
Parameters
G
Returns

biconnected
[bool] True if the graph is biconnected, False otherwise.
Raises
If the input graph is not undirected.
See also:
biconnected_components
articulation_points
biconnected_component_edges
is_weakly_connected
is_connected
is_semiconnected
Notes
The algorithm to find articulation points and biconnected components is implemented using a non-recursive depth-
first-search (DFS) that keeps track of the highest level that back edges reach in the DFS tree. A node n is an
articulation point if, and only if, there exists a subtree rooted at n such that there is no back edge from any successor
of n that links to a predecessor of n in the DFS tree. By keeping track of all the edges traversed by the DFS we can
obtain the biconnected components because all edges of a bicomponent will be traversed consecutively between
articulation points.
References
[1]
Examples
>>> print(nx.is_biconnected(G))
False
True
biconnected_components(G)
Returns a generator of sets of nodes, one set for each biconnected component of the graph
Biconnected components are maximal subgraphs such that the removal of a node (and all edges incident on that
node) will not disconnect the subgraph. Note that nodes may be part of more than one biconnected component.
Those nodes are articulation points, or cut vertices. The removal of articulation points will increase the number of
connected components of the graph.
Notice that by convention a dyad is considered a biconnected component.
Parameters

G
Returns
nodes
[generator] Generator of sets of nodes, one set for each biconnected component.
Raises
See also:
is_biconnected
articulation_points
k_components
this function is a special case where k=2
bridge_components
similar to this function, but is defined using 2-edge-connectivity instead of 2-node-connectivity.
Notes
References
[1]
Examples
>>> G = nx.lollipop_graph(5, 1)
False
>>> bicomponents = list(nx.biconnected_components(G))
>>> len(bicomponents)
2
True
>>> bicomponents = list(nx.biconnected_components(G))
>>> len(bicomponents)
1
You can generate a sorted list of biconnected components, largest first, using sort.

>>> [len(c) for c in sorted(nx.biconnected_components(G), key=len, reverse=True)]
[5, 2]
If you only want the largest connected component, it’s more efficient to use max instead of sort.
>>> Gc = max(nx.biconnected_components(G), key=len)
To create the components as subgraphs use: (G.subgraph(c).copy() for c in bicon-

nected_components(G))
biconnected_component_edges(G)
Returns a generator of lists of edges, one list for each biconnected component of the input graph.
Biconnected components are maximal subgraphs such that the removal of a node (and all edges incident on that
node) will not disconnect the subgraph. Note that nodes may be part of more than one biconnected component.
Those nodes are articulation points, or cut vertices. However, each edge belongs to one, and only one, biconnected
component.
Parameters
G
Returns
edges
[generator of lists] Generator of lists of edges, one list for each bicomponent.
Raises
See also:
is_biconnected
articulation_points
Notes

References
[1]
Examples
False
>>> bicomponents_edges = list(nx.biconnected_component_edges(G))
>>> len(bicomponents_edges)
5
True
>>> bicomponents_edges = list(nx.biconnected_component_edges(G))
>>> len(bicomponents_edges)
1
articulation_points
articulation_points(G)
Yield the articulation points, or cut vertices, of a graph.
An articulation point or cut vertex is any node whose removal (along with all its incident edges) increases the number
of connected components of a graph. An undirected connected graph without articulation points is biconnected.
Articulation points belong to more than one biconnected component of a graph.
Parameters
G
Yields
node
An articulation point in the graph.
Raises
See also:
is_biconnected

Notes
References
[1]
Examples
False
>>> len(list(nx.articulation_points(G)))
4
True
>>> len(list(nx.articulation_points(G)))
0
3.15.6 Semiconnectedness
is_semiconnected(G[, topo_order]) Returns True if the graph is semiconnected, False other-

wise.
is_semiconnected
is_semiconnected(G, topo_order=None)
Returns True if the graph is semiconnected, False otherwise.
A graph is semiconnected if, and only if, for any pair of nodes, either one is reachable from the other, or they are
mutually reachable.
Parameters
G
topo_order: list or tuple, optional
A topological order for G (if None, the function will compute one)
Returns
semiconnected
[bool] True if the graph is semiconnected, False otherwise.
Raises

If the graph is empty.
See also:
is_weakly_connected
is_connected
is_biconnected
Examples

>>> print(nx.is_semiconnected(G))
True
>>> G = nx.DiGraph([(1, 2), (3, 2)])
>>> print(nx.is_semiconnected(G))
False
3.16 Connectivity
Connectivity and cut algorithms
3.16.1 Edge-augmentation
Algorithms for finding k-edge-augmentations

A k-edge-augmentation is a set of edges, that once added to a graph, ensures that the graph is k-edge-connected; i.e. the
graph cannot be disconnected unless k or more edges are removed. Typically, the goal is to find the augmentation with
minimum weight. In general, it is not guaranteed that a k-edge-augmentation exists.
See Also
edge_kcomponents : algorithms for finding k-edge-connected components connectivity : algorithms for de-
termening edge connectivity.
k_edge_augmentation(G, k[, avail, weight, ...]) Finds set of edges to k-edge-connect G.

is_k_edge_connected(G, k) Tests to see if a graph is k-edge-connected.
is_locally_k_edge_connected(G, s, t, k) Tests to see if an edge in a graph is locally k-edge-
connected.

k_edge_augmentation
k_edge_augmentation(G, k, avail=None, weight=None, partial=False)

Finds set of edges to k-edge-connect G.
Adding edges from the augmentation to G make it impossible to disconnect G unless k or more edges are removed.
This function uses the most efficient function available (depending on the value of k and if the problem is weighted
or unweighted) to search for a minimum weight subset of available edges that k-edge-connects G. In general,
finding a k-edge-augmentation is NP-hard, so solutions are not guaranteed to be minimal. Furthermore, a k-edge-
augmentation may not exist.
Parameters
G
k
[integer] Desired edge connectivity
avail
[dict or a set of 2 or 3 tuples] The available edges that can be used in the augmentation.
If unspecified, then all edges in the complement of G are available. Otherwise, each item is an
available edge (with an optional weight).
In the unweighted case, each item is an edge (u, v).
In the weighted case, each item is a 3-tuple (u, v, d) or a dict with items (u, v):
d. The third item, d, can be a dictionary or a real number. If d is a dictionary d[weight]
correspondings to the weight.
weight
[string] key to use to find weights if avail is a set of 3-tuples where the third item in each
tuple is a dictionary.
partial
[boolean] If partial is True and no feasible k-edge-augmentation exists, then all a partial k-
edge-augmentation is generated. Adding the edges in a partial augmentation to G, minimizes
the number of k-edge-connected components and maximizes the edge connectivity between
those components. For details, see partial_k_edge_augmentation().
Yields
edge
[tuple] Edges that, once added to G, would cause G to become k-edge-connected. If partial is
False, an error is raised if this is not possible. Otherwise, generated edges form a partial aug-
mentation, which k-edge-connects any part of G where it is possible, and maximally connects
the remaining parts.
Raises
NetworkXUnfeasible
If partial is False and no k-edge-augmentation exists.
If the input graph is directed or a multigraph.
ValueError:
If k is less than 1
3.16. Connectivity 361

Notes
When k=1 this returns an optimal solution.

When k=2 and avail is None, this returns an optimal solution. Otherwise when k=2, this returns a 2-
approximation of the optimal solution.
For k>3, this problem is NP-hard and this uses a randomized algorithm that
produces a feasible solution, but provides no guarantees on the solution weight.
Examples
>>> # Unweighted cases

>>> G = nx.path_graph((1, 2, 3, 4))
>>> G.add_node(5)
>>> sorted(nx.k_edge_augmentation(G, k=1))
[(1, 5)]
[(1, 5), (5, 4)]
[(1, 4), (1, 5), (2, 5), (3, 5), (4, 5)]
>>> complement = list(nx.k_edge_augmentation(G, k=5, partial=True))
>>> G.add_edges_from(complement)
>>> nx.edge_connectivity(G)
4
>>> # Weighted cases

>>> G = nx.path_graph((1, 2, 3, 4))
>>> G.add_node(5)
>>> # avail can be a tuple with a dict
>>> avail = [(1, 5, {"weight": 11}), (2, 5, {"weight": 10})]
>>> sorted(nx.k_edge_augmentation(G, k=1, avail=avail, weight="weight"))
[(2, 5)]
>>> # or avail can be a 3-tuple with a real number
>>> avail = [(1, 5, 11), (2, 5, 10), (4, 3, 1), (4, 5, 51)]
>>> sorted(nx.k_edge_augmentation(G, k=2, avail=avail))
[(1, 5), (2, 5), (4, 5)]
>>> # or avail can be a dict
>>> avail = {(1, 5): 11, (2, 5): 10, (4, 3): 1, (4, 5): 51}
>>> sorted(nx.k_edge_augmentation(G, k=2, avail=avail))
[(1, 5), (2, 5), (4, 5)]
>>> # If augmentation is infeasible, then a partial solution can be found
>>> avail = {(1, 5): 11}
>>> sorted(nx.k_edge_augmentation(G, k=2, avail=avail, partial=True))
[(1, 5)]

is_k_edge_connected
is_k_edge_connected(G, k)
Tests to see if a graph is k-edge-connected.
Is it impossible to disconnect the graph by removing fewer than k edges? If so, then G is k-edge-connected.
Parameters
G
k
[integer] edge connectivity to test for
Returns
boolean
True if G is k-edge-connected.
See also:
is_locally_k_edge_connected()
Examples
>>> nx.is_k_edge_connected(G, k=1)
True
>>> nx.is_k_edge_connected(G, k=2)
False
is_locally_k_edge_connected
is_locally_k_edge_connected(G, s, t, k)
Tests to see if an edge in a graph is locally k-edge-connected.
Is it impossible to disconnect s and t by removing fewer than k edges? If so, then s and t are locally k-edge-connected
in G.
Parameters
G
s
[node] Source node
t
[node] Target node
k
[integer] local edge connectivity for nodes s and t
Returns
boolean
True if s and t are locally k-edge-connected in G.

See also:
is_k_edge_connected()
Examples
>>> from networkx.algorithms.connectivity import is_locally_k_edge_connected

>>> is_locally_k_edge_connected(G, 5, 15, k=1)
True
False
True
3.16.2 K-edge-components
Algorithms for finding k-edge-connected components and subgraphs.

A k-edge-connected component (k-edge-cc) is a maximal set of nodes in G, such that all pairs of node have an edge-
connectivity of at least k.
A k-edge-connected subgraph (k-edge-subgraph) is a maximal set of nodes in G, such that the subgraph of G defined by
the nodes has an edge-connectivity at least k.
k_edge_components(G, k) Generates nodes in each maximal k-edge-connected com-

ponent in G.
k_edge_subgraphs(G, k) Generates nodes in each maximal k-edge-connected sub-
graph in G.
bridge_components(G) Finds all bridge-connected components G.
EdgeComponentAuxGraph() A simple algorithm to find all k-edge-connected compo-
nents in a graph.
k_edge_components
k_edge_components(G, k)
Generates nodes in each maximal k-edge-connected component in G.
Parameters
G
[NetworkX graph]
k
[Integer] Desired edge connectivity
Returns
k_edge_components
[a generator of k-edge-ccs. Each set of returned nodes] will have k-edge-connectivity in the
graph G.
Raises

If the input graph is a multigraph.
ValueError:
If k is less than 1
See also:
local_edge_connectivity()
k_edge_subgraphs()
similar to this function, but the subgraph defined by the nodes must also have k-edge-connectivity.
k_components()
similar to this function, but uses node-connectivity instead of edge-connectivity
Notes
Attempts to use the most efficient implementation available based on k. If k=1, this is simply connected components
for directed graphs and connected components for undirected graphs. If k=2 on an efficient bridge connected
component algorithm from _[1] is run based on the chain decomposition. Otherwise, the algorithm from _[2] is
used.
References
[1], [2]
Examples
>>> import itertools as it

>>> from networkx.utils import pairwise
>>> paths = [
... (1, 2, 4, 3, 1, 4),
... (5, 6, 7, 8, 5, 7, 8, 6),
... ]
>>> G = nx.Graph()
>>> G.add_nodes_from(it.chain(*paths))
>>> G.add_edges_from(it.chain(*[pairwise(path) for path in paths]))
>>> # note this returns {1, 4} unlike k_edge_subgraphs
>>> sorted(map(sorted, nx.k_edge_components(G, k=3)))
[[1, 4], [2], [3], [5, 6, 7, 8]]
k_edge_subgraphs
k_edge_subgraphs(G, k)
Generates nodes in each maximal k-edge-connected subgraph in G.
Parameters
G
[NetworkX graph]
k

Returns
k_edge_subgraphs
[a generator of k-edge-subgraphs] Each k-edge-subgraph is a maximal set of nodes that defines
a subgraph of G that is k-edge-connected.
Raises
If the input graph is a multigraph.
ValueError:
If k is less than 1
See also:
edge_connectivity()
k_edge_components()
similar to this function, but nodes only need to have k-edge-connctivity within the graph G and the subgraphs
might not be k-edge-connected.
Notes
Attempts to use the most efficient implementation available based on k. If k=1, or k=2 and the graph is undirected,
then this simply calls k_edge_components. Otherwise the algorithm from _[1] is used.
References
[1]
Examples

>>> paths = [
... (1, 2, 4, 3, 1, 4),
... (5, 6, 7, 8, 5, 7, 8, 6),
... ]
>>> G = nx.Graph()
>>> # note this does not return {1, 4} unlike k_edge_components
>>> sorted(map(sorted, nx.k_edge_subgraphs(G, k=3)))
[[1], [2], [3], [4], [5, 6, 7, 8]]

bridge_components
bridge_components(G)
Finds all bridge-connected components G.
Parameters
G
[NetworkX undirected graph]
Returns
bridge_components
[a generator of 2-edge-connected components]
Raises
If the input graph is directed or a multigraph.
See also:
k_edge_subgraphs()
this function is a special case for an undirected graph where k=2.
biconnected_components()
similar to this function, but is defined using 2-node-connectivity instead of 2-edge-connectivity.
Notes
Bridge-connected components are also known as 2-edge-connected components.
Examples
>>> # The barbell graph with parameter zero has a single bridge
>>> from networkx.algorithms.connectivity.edge_kcomponents import bridge_
,→components
>>> sorted(map(sorted, bridge_components(G)))

[[0, 1, 2, 3, 4], [5, 6, 7, 8, 9]]
networkx.algorithms.connectivity.edge_kcomponents.EdgeComponentAuxGraph
class EdgeComponentAuxGraph
A simple algorithm to find all k-edge-connected components in a graph.
Constructing the AuxillaryGraph (which may take some time) allows for the k-edge-ccs to be found in linear time
for arbitrary k.

Notes
This implementation is based on [1]. The idea is to construct an auxiliary graph from which the k-edge-ccs can be
extracted in linear time. The auxiliary graph is constructed in O(|V | · F ) operations, where F is the complexity of
max flow. Querying the components takes an additional O(|V |) operations. This algorithm can be slow for large
graphs, but it handles an arbitrary k and works for both directed and undirected inputs.
The undirected case for k=1 is exactly connected components. The undirected case for k=2 is exactly bridge
connected components. The directed case for k=1 is exactly strongly connected components.
References
[1]
Examples

>>> from networkx.algorithms.connectivity import EdgeComponentAuxGraph
>>> # Build an interesting graph with multiple levels of k-edge-ccs
>>> paths = [
... (1, 2, 3, 4, 1, 3, 4, 2), # a 3-edge-cc (a 4 clique)
... (5, 6, 7, 5), # a 2-edge-cc (a 3 clique)
... (1, 5), # combine first two ccs into a 1-edge-cc
... (0,), # add an additional disconnected 1-edge-cc
... ]
>>> G = nx.Graph()
>>> # Constructing the AuxGraph takes about O(n ** 4)
>>> aux_graph = EdgeComponentAuxGraph.construct(G)
>>> # Once constructed, querying takes O(n)
>>> sorted(map(sorted, aux_graph.k_edge_components(k=1)))
[[0], [1, 2, 3, 4, 5, 6, 7]]
[[0], [1, 2, 3, 4], [5, 6, 7]]
[[0], [1, 2, 3, 4], [5], [6], [7]]
[[0], [1], [2], [3], [4], [5], [6], [7]]
The auxiliary graph is primarilly used for k-edge-ccs but it can also speed up the queries of k-edge-subgraphs by
refining the search space.
>>> from networkx.algorithms.connectivity import EdgeComponentAuxGraph
>>> paths = [
... (1, 2, 4, 3, 1, 4),
... ]
>>> G = nx.Graph()
>>> aux_graph = EdgeComponentAuxGraph.construct(G)
>>> sorted(map(sorted, aux_graph.k_edge_subgraphs(k=3)))


[[1], [2], [3], [4]]
[[1, 4], [2], [3]]
__init__(*args, **kwargs)
Methods
construct(G) Builds an auxiliary graph encoding edge-connectivity

between nodes.
k_edge_components(k) Queries the auxiliary graph for k-edge-connected
components.
k_edge_subgraphs(k) Queries the auxiliary graph for k-edge-connected sub-
graphs.
EdgeComponentAuxGraph.construct
classmethod EdgeComponentAuxGraph.construct(G)
Builds an auxiliary graph encoding edge-connectivity between nodes.
Parameters
G
[NetworkX graph]
Notes
Given G=(V, E), initialize an empty auxiliary graph A. Choose an arbitrary source node s. Initialize a set N
of available nodes (that can be used as the sink). The algorithm picks an arbitrary node t from N - {s}, and
then computes the minimum st-cut (S, T) with value w. If G is directed the minimum of the st-cut or the
ts-cut is used instead. Then, the edge (s, t) is added to the auxiliary graph with weight w. The algorithm is
called recursively first using S as the available nodes and s as the source, and then using T and t. Recursion
stops when the source is the only available node.
EdgeComponentAuxGraph.k_edge_components
EdgeComponentAuxGraph.k_edge_components(k)
Queries the auxiliary graph for k-edge-connected components.
Parameters
k
Returns
k_edge_components
[a generator of k-edge-ccs]

Notes
Given the auxiliary graph, the k-edge-connected components can be determined in linear time by removing
all edges with weights less than k from the auxiliary graph. The resulting connected components are the
k-edge-ccs in the original graph.
EdgeComponentAuxGraph.k_edge_subgraphs
EdgeComponentAuxGraph.k_edge_subgraphs(k)
Queries the auxiliary graph for k-edge-connected subgraphs.
Parameters
k
Returns
k_edge_subgraphs
[a generator of k-edge-subgraphs]
Notes
Refines the k-edge-ccs into k-edge-subgraphs. The running time is more than O(|V |).
For single values of k it is faster to use nx.k_edge_subgraphs. But for multiple values of k, it can be
faster to build AuxGraph and then use this method.
3.16.3 K-node-components
Moody and White algorithm for k-components
k_components(G[, flow_func]) Returns the k-component structure of a graph G.
k_components
k_components(G, flow_func=None)
Returns the k-component structure of a graph G.
A k-component is a maximal subgraph of a graph G that has, at least, node connectivity k: we need to remove
at least k nodes to break it into more components. k-components have an inherent hierarchical structure because
they are nested in terms of connectivity: a connected graph can contain several 2-components, each of which can
contain one or more 3-components, and so forth.
Parameters
G
[NetworkX graph]
flow_func
[function] Function to perform the underlying flow computations. Default value ed-
monds_karp(). This function performs better in sparse graphs with right tailed degree
distributions. shortest_augmenting_path() will perform better in denser graphs.

Returns
k_components
[dict] Dictionary with all connectivity levels k in the input Graph as keys and a list of sets of
nodes that form a k-component of level k as values.
Raises
If the input graph is directed.
See also:
node_connectivity
all_node_cuts
special case of this function when k=2
k_edge_components
similar to this function, but uses edge-connectivity instead of node-connectivity
Notes
Moody and White [1] (appendix A) provide an algorithm for identifying k-components in a graph, which is
based on Kanevsky’s algorithm [2] for finding all minimum-size node cut-sets of a graph (implemented in
all_node_cuts() function):
1. Compute node connectivity, k, of the input graph G.
2. Identify all k-cutsets at the current level of connectivity using Kanevsky’s algorithm.
3. Generate new graph components based on the removal of these cutsets. Nodes in a cutset belong to both sides
of the induced cut.
4. If the graph is neither complete nor trivial, return to 1; else end.
This implementation also uses some heuristics (see [3] for details) to speed up the computation.
References
[1], [2], [3]
Examples
>>> # Petersen graph has 10 nodes and it is triconnected, thus all

>>> # nodes are in a single component on all three connectivity levels
>>> k_components = nx.k_components(G)

3.16.4 K-node-cutsets
Kanevsky all minimum node k cutsets algorithm.
all_node_cuts(G[, k, flow_func]) Returns all minimum k cutsets of an undirected graph G.
all_node_cuts
all_node_cuts(G, k=None, flow_func=None)

Returns all minimum k cutsets of an undirected graph G.
This implementation is based on Kanevsky’s algorithm [1] for finding all minimum-size node cut-sets of an undi-
rected graph G; ie the set (or sets) of nodes of cardinality equal to the node connectivity of G. Thus if removed,
would break G into two or more connected components.
Parameters
G
k
[Integer] Node connectivity of the input graph. If k is None, then it is computed. Default
value: None.
flow_func
[function] Function to perform the underlying flow computations. Default value is ed-
Returns
cuts
[a generator of node cutsets] Each node cutset has cardinality equal to the node connectivity
of the input graph.
See also:
node_connectivity
edmonds_karp
shortest_augmenting_path
Notes
This implementation is based on the sequential algorithm for finding all minimum-size separating vertex sets in a
graph [1]. The main idea is to compute minimum cuts using local maximum flow computations among a set of
nodes of highest degree and all other non-adjacent nodes in the Graph. Once we find a minimum cut, we add an
edge between the high degree node and the target node of the local maximum flow computation to make sure that
we will not find that minimum cut again.

References
[1]
Examples
>>> # A two-dimensional grid graph has 4 cutsets of cardinality 2

>>> G = nx.grid_2d_graph(5, 5)
>>> cutsets = list(nx.all_node_cuts(G))
>>> len(cutsets)
4
>>> all(2 == len(cutset) for cutset in cutsets)
True
>>> nx.node_connectivity(G)
2
3.16.5 Flow-based disjoint paths
Flow based node and edge disjoint paths.
edge_disjoint_paths(G, s, t[, flow_func, ...]) Returns the edges disjoint paths between source and tar-
get.
node_disjoint_paths(G, s, t[, flow_func, ...]) Computes node disjoint paths between source and target.
edge_disjoint_paths
edge_disjoint_paths(G, s, t, flow_func=None, cutoff=None, auxiliary=None, residual=None)

Returns the edges disjoint paths between source and target.
Edge disjoint paths are paths that do not share any edge. The number of edge disjoint paths between source and
target is equal to their edge connectivity.
Parameters
G
[NetworkX graph]
s
[node] Source node for the flow.
t
[node] Sink node for the flow.
flow_func
[function] A function for computing the maximum flow among a pair of nodes. The function
has to accept at least three parameters: a Digraph, a source node, and a target node. And return
a residual network that follows NetworkX conventions (see maximum_flow() for details).
If flow_func is None, the default maximum flow function (edmonds_karp()) is used. The
choice of the default function may change from version to version and should not be relied on.
Default value: None.

cutoff
[integer or None (default: None)] Maximum number of paths to yield. If specified, the max-
imum flow algorithm will terminate when the flow value reaches or exceeds the cutoff. This
only works for flows that support the cutoff parameter (most do) and is ignored otherwise.
auxiliary
[NetworkX DiGraph] Auxiliary digraph to compute flow based edge connectivity. It has to
have a graph attribute called mapping with a dictionary mapping node names in G and in the
auxiliary digraph. If provided it will be reused instead of recreated. Default value: None.
residual
[NetworkX DiGraph] Residual network to compute maximum flow. If provided it will be
reused instead of recreated. Default value: None.
Returns
paths
[generator] A generator of edge independent paths.
Raises
NetworkXNoPath
If there is no path between source and target.
NetworkXError
If source or target are not in the graph G.
See also:
node_disjoint_paths()
edge_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()
shortest_augmenting_path()
Notes
This is a flow based implementation of edge disjoint paths. We compute the maximum flow between source and
target on an auxiliary directed network. The saturated edges in the residual network after running the maximum
flow algorithm correspond to edge disjoint paths between source and target in the original network. This function
handles both directed and undirected graphs, and can use all flow algorithms from NetworkX flow package.
Examples
We use in this example the platonic icosahedral graph, which has node edge connectivity 5, thus there are 5 edge
disjoint paths between any pair of nodes.
>>> G = nx.icosahedral_graph()
>>> len(list(nx.edge_disjoint_paths(G, 0, 6)))
5
If you need to compute edge disjoint paths on several pairs of nodes in the same graph, it is recommended that you
reuse the data structures that NetworkX uses in the computation: the auxiliary digraph for edge connectivity, and
the residual network for the underlying maximum flow computation.

Example of how to compute edge disjoint paths among all pairs of nodes of the platonic icosahedral graph reusing
the data structures.

>>> # You also have to explicitly import the function for
>>> # building the auxiliary digraph from the connectivity package
>>> from networkx.algorithms.connectivity import build_auxiliary_edge_connectivity
>>> H = build_auxiliary_edge_connectivity(G)
>>> # And the function for building the residual network from the
>>> # flow package
>>> from networkx.algorithms.flow import build_residual_network
>>> # Note that the auxiliary digraph has an edge attribute named capacity
>>> R = build_residual_network(H, "capacity")
>>> result = {n: {} for n in G}
>>> # Reuse the auxiliary digraph and the residual network by passing them
>>> # as arguments
>>> for u, v in itertools.combinations(G, 2):
... k = len(list(nx.edge_disjoint_paths(G, u, v, auxiliary=H, residual=R)))
... result[u][v] = k
>>> all(result[u][v] == 5 for u, v in itertools.combinations(G, 2))
True
You can also use alternative flow algorithms for computing edge disjoint paths. For instance, in dense net-
works the algorithm shortest_augmenting_path() will usually perform better than the default ed-
monds_karp() which is faster for sparse networks with highly skewed degree distributions. Alternative flow
functions have to be explicitly imported from the flow package.
>>> from networkx.algorithms.flow import shortest_augmenting_path

>>> len(list(nx.edge_disjoint_paths(G, 0, 6, flow_func=shortest_augmenting_path)))
5
node_disjoint_paths
node_disjoint_paths(G, s, t, flow_func=None, cutoff=None, auxiliary=None, residual=None)

Computes node disjoint paths between source and target.
Node disjoint paths are paths that only share their first and last nodes. The number of node independent paths
between two nodes is equal to their local node connectivity.
Parameters
G
[NetworkX graph]
s
[node] Source node.
t
[node] Target node.
flow_func
If flow_func is None, the default maximum flow function (edmonds_karp()) is used. See
below for details. The choice of the default function may change from version to version and
should not be relied on. Default value: None.

cutoff
[integer or None (default: None)] Maximum number of paths to yield. If specified, the max-
imum flow algorithm will terminate when the flow value reaches or exceeds the cutoff. This
only works for flows that support the cutoff parameter (most do) and is ignored otherwise.
auxiliary
[NetworkX DiGraph] Auxiliary digraph to compute flow based node connectivity. It has to
residual
Returns
paths
[generator] Generator of node disjoint paths.
Raises
NetworkXNoPath
If there is no path between source and target.
NetworkXError
If source or target are not in the graph G.
See also:
edge_disjoint_paths()
node_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()
Notes
This is a flow based implementation of node disjoint paths. We compute the maximum flow between source and
target on an auxiliary directed network. The saturated edges in the residual network after running the maximum
flow algorithm correspond to node disjoint paths between source and target in the original network. This function
handles both directed and undirected graphs, and can use all flow algorithms from NetworkX flow package.
Examples
We use in this example the platonic icosahedral graph, which has node connectivity 5, thus there are 5 node disjoint
paths between any pair of non neighbor nodes.
>>> len(list(nx.node_disjoint_paths(G, 0, 6)))
5
If you need to compute node disjoint paths between several pairs of nodes in the same graph, it is recommended that
you reuse the data structures that NetworkX uses in the computation: the auxiliary digraph for node connectivity
and node cuts, and the residual network for the underlying maximum flow computation.
Example of how to compute node disjoint paths reusing the data structures:


>>> from networkx.algorithms.connectivity import build_auxiliary_node_connectivity
>>> H = build_auxiliary_node_connectivity(G)
>>> # flow package
>>> # as arguments
>>> len(list(nx.node_disjoint_paths(G, 0, 6, auxiliary=H, residual=R)))
5
You can also use alternative flow algorithms for computing node disjoint paths. For instance, in dense net-
>>> len(list(nx.node_disjoint_paths(G, 0, 6, flow_func=shortest_augmenting_path)))
5
3.16.6 Flow-based Connectivity
Flow based connectivity algorithms
average_node_connectivity(G[, flow_func]) Returns the average connectivity of a graph G.

all_pairs_node_connectivity(G[, nbunch, Compute node connectivity between all pairs of nodes of
...]) G.
edge_connectivity(G[, s, t, flow_func, cutoff]) Returns the edge connectivity of the graph or digraph G.
local_edge_connectivity(G, s, t[, ...]) Returns local edge connectivity for nodes s and t in G.
local_node_connectivity(G, s, t[, ...]) Computes local node connectivity for nodes s and t.
node_connectivity(G[, s, t, flow_func]) Returns node connectivity for a graph or digraph G.
average_node_connectivity
average_node_connectivity(G, flow_func=None)
Returns the average connectivity of a graph G.
The average connectivity bar{kappa} of a graph G is the average of local node connectivity over all pairs of
nodes of G [1] .
∑
u,v κG (u, v)
κ̄(G) = ( n)
2
Parameters
G
flow_func
has to accept at least three parameters: a Digraph, a source node, and a target node. And

return a residual network that follows NetworkX conventions (see maximum_flow() for
details). If flow_func is None, the default maximum flow function (edmonds_karp()) is
used. See local_node_connectivity() for details. The choice of the default function
may change from version to version and should not be relied on. Default value: None.
Returns
K
[float] Average node connectivity
See also:
local_node_connectivity()
node_connectivity()
edge_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()
References
[1]
all_pairs_node_connectivity(G, nbunch=None, flow_func=None)

Compute node connectivity between all pairs of nodes of G.
Parameters
G
nbunch: container
Container of nodes. If provided node connectivity will be computed only over pairs of nodes
in nbunch.
flow_func
Returns
all_pairs
[dict] A dictionary with node connectivity between all pairs of nodes in G, or in nbunch if
provided.
See also:

edge_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()
edge_connectivity
edge_connectivity(G, s=None, t=None, flow_func=None, cutoff=None)

Returns the edge connectivity of the graph or digraph G.
The edge connectivity is equal to the minimum number of edges that must be removed to disconnect G or render
it trivial. If source and target nodes are provided, this function returns the local edge connectivity: the minimum
number of edges that must be removed to break all paths from source to target in G.
Parameters
G
[NetworkX graph] Undirected or directed graph
s
t
flow_func
cutoff
[integer, float, or None (default: None)] If specified, the maximum flow algorithm will termi-
nate when the flow value reaches or exceeds the cutoff. This only works for flows that support
the cutoff parameter (most do) and is ignored otherwise.
Returns
K
[integer] Edge connectivity for G, or local edge connectivity if source and target were provided
See also:
node_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()
k_edge_components()
k_edge_subgraphs()

Notes
This is a flow based implementation of global edge connectivity. For undirected graphs the algorithm works by
finding a ‘small’ dominating set of nodes of G (see algorithm 7 in [1] ) and computing local maximum flow (see
local_edge_connectivity()) between an arbitrary node in the dominating set and the rest of nodes in it.
This is an implementation of algorithm 6 in [1] . For directed graphs, the algorithm does n calls to the maximum
flow function. This is an implementation of algorithm 8 in [1] .
References
[1]
Examples
>>> # Platonic icosahedral graph is 5-edge-connected

>>> nx.edge_connectivity(G)
5
You can use alternative flow algorithms for the underlying maximum flow computation. In dense networks the al-
gorithm shortest_augmenting_path() will usually perform better than the default edmonds_karp(),
which is faster for sparse networks with highly skewed degree distributions. Alternative flow functions have to be
explicitly imported from the flow package.

>>> nx.edge_connectivity(G, flow_func=shortest_augmenting_path)
5
If you specify a pair of nodes (source and target) as parameters, this function returns the value of local edge
connectivity.
>>> nx.edge_connectivity(G, 3, 7)
5
If you need to perform several local computations among different pairs of nodes on the same graph, it
is recommended that you reuse the data structures used in the maximum flow computations. See lo-
cal_edge_connectivity() for details.
local_edge_connectivity
local_edge_connectivity(G, s, t, flow_func=None, auxiliary=None, residual=None, cutoff=None)

Returns local edge connectivity for nodes s and t in G.
Local edge connectivity for two nodes s and t is the minimum number of edges that must be removed to disconnect
them.
This is a flow based implementation of edge connectivity. We compute the maximum flow on an auxiliary digraph
build from the original network (see below for details). This is equal to the local edge connectivity because the
value of a maximum s-t-flow is equal to the capacity of a minimum s-t-cut (Ford and Fulkerson theorem) [1] .
Parameters
G
[NetworkX graph] Undirected or directed graph

s
[node] Source node
t
[node] Target node
flow_func
auxiliary
[NetworkX DiGraph] Auxiliary digraph for computing flow based edge connectivity. If pro-
vided it will be reused instead of recreated. Default value: None.
residual
cutoff
Returns
K
[integer] local edge connectivity for nodes s and t.
See also:
edge_connectivity()
node_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()
Notes
This is a flow based implementation of edge connectivity. We compute the maximum flow using, by default, the
edmonds_karp() algorithm on an auxiliary digraph build from the original input graph:
If the input graph is undirected, we replace each edge (u,`v`) with two reciprocal arcs (u, v) and (v, u) and then
we set the attribute ‘capacity’ for each arc to 1. If the input graph is directed we simply add the ‘capacity’ attribute.
This is an implementation of algorithm 1 in [1].
The maximum flow in the auxiliary network is equal to the local edge connectivity because the value of a maximum
s-t-flow is equal to the capacity of a minimum s-t-cut (Ford and Fulkerson theorem).

References
[1]
Examples
This function is not imported in the base NetworkX namespace, so you have to explicitly import it from the con-
nectivity package:
>>> from networkx.algorithms.connectivity import local_edge_connectivity
We use in this example the platonic icosahedral graph, which has edge connectivity 5.
>>> local_edge_connectivity(G, 0, 6)
5
If you need to compute local connectivity on several pairs of nodes in the same graph, it is recommended that you
Example of how to compute local edge connectivity among all pairs of nodes of the platonic icosahedral graph
reusing the data structures.

>>> # flow package
>>> result = dict.fromkeys(G, dict())
>>> # as parameters
... k = local_edge_connectivity(G, u, v, auxiliary=H, residual=R)
True
You can also use alternative flow algorithms for computing edge connectivity. For instance, in dense networks the al-
gorithm shortest_augmenting_path() will usually perform better than the default edmonds_karp()

>>> local_edge_connectivity(G, 0, 6, flow_func=shortest_augmenting_path)
5

local_node_connectivity(G, s, t, flow_func=None, auxiliary=None, residual=None, cutoff=None)

Computes local node connectivity for nodes s and t.
Local node connectivity for two non adjacent nodes s and t is the minimum number of nodes that must be removed
(along with their incident edges) to disconnect them.
This is a flow based implementation of node connectivity. We compute the maximum flow on an auxiliary digraph
build from the original input graph (see below for details).
Parameters
G
s
[node] Source node
t
[node] Target node
flow_func
auxiliary
residual
cutoff
Returns
K
[integer] local node connectivity for nodes s and t
See also:
node_connectivity()
minimum_node_cut()
maximum_flow()
edmonds_karp()
preflow_push()

Notes
This is a flow based implementation of node connectivity. We compute the maximum flow using, by default, the
edmonds_karp() algorithm (see: maximum_flow()) on an auxiliary digraph build from the original input
graph:
For an undirected graph G having n nodes and m edges we derive a directed graph H with 2n nodes and 2m+n arcs
by replacing each original node v with two nodes v_A, v_B linked by an (internal) arc in H. Then for each edge
(u, v) in G we add two arcs (u_B, v_A) and (v_B, u_A) in H. Finally we set the attribute capacity = 1 for each
arc in H [1] .
For a directed graph G having n nodes and m arcs we derive a directed graph H with 2n nodes and m+n arcs by
replacing each original node v with two nodes v_A, v_B linked by an (internal) arc (v_A, v_B) in H. Then for
each arc (u, v) in G we add one arc (u_B, v_A) in H. Finally we set the attribute capacity = 1 for each arc in H.
This is equal to the local node connectivity because the value of a maximum s-t-flow is equal to the capacity of a
minimum s-t-cut.
References
[1]
Examples
nectivity package:
>>> from networkx.algorithms.connectivity import local_node_connectivity
We use in this example the platonic icosahedral graph, which has node connectivity 5.
>>> local_node_connectivity(G, 0, 6)
5
If you need to compute local connectivity on several pairs of nodes in the same graph, it is recommended that you
reuse the data structures that NetworkX uses in the computation: the auxiliary digraph for node connectivity, and
Example of how to compute local node connectivity among all pairs of nodes of the platonic icosahedral graph
reusing the data structures.
...
>>> # flow package
>>> # as parameters


... k = local_node_connectivity(G, u, v, auxiliary=H, residual=R)
...
True
You can also use alternative flow algorithms for computing node connectivity. For instance, in dense net-

>>> local_node_connectivity(G, 0, 6, flow_func=shortest_augmenting_path)
5
node_connectivity
node_connectivity(G, s=None, t=None, flow_func=None)

Returns node connectivity for a graph or digraph G.
Node connectivity is equal to the minimum number of nodes that must be removed to disconnect G or render it
trivial. If source and target nodes are provided, this function returns the local node connectivity: the minimum
number of nodes that must be removed to break all paths from source to target in G.
Parameters
G
s
t
flow_func
Returns
K
[integer] Node connectivity of G, or local node connectivity if source and target are provided.
See also:
edge_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()

Notes
This is a flow based implementation of node connectivity. The algorithm works by solving O((n − δ − 1 + δ(δ −
1)/2)) maximum flow problems on an auxiliary digraph. Where δ is the minimum degree of G. For details about
the auxiliary digraph and the computation of local node connectivity see local_node_connectivity().
This implementation is based on algorithm 11 in [1].
References
[1]
Examples
>>> # Platonic icosahedral graph is 5-node-connected

>>> nx.node_connectivity(G)
5

>>> nx.node_connectivity(G, flow_func=shortest_augmenting_path)
5
If you specify a pair of nodes (source and target) as parameters, this function returns the value of local node
connectivity.
>>> nx.node_connectivity(G, 3, 7)
5
cal_node_connectivity() for details.
3.16.7 Flow-based Minimum Cuts
Flow based cut algorithms
minimum_edge_cut(G[, s, t, flow_func]) Returns a set of edges of minimum cardinality that dis-

connects G.
minimum_node_cut(G[, s, t, flow_func]) Returns a set of nodes of minimum cardinality that dis-
connects G.
minimum_st_edge_cut(G, s, t[, flow_func, ...]) Returns the edges of the cut-set of a minimum (s, t)-cut.
minimum_st_node_cut(G, s, t[, flow_func, ...]) Returns a set of nodes of minimum cardinality that dis-
connect source from target in G.

minimum_edge_cut
minimum_edge_cut(G, s=None, t=None, flow_func=None)

Returns a set of edges of minimum cardinality that disconnects G.
If source and target nodes are provided, this function returns the set of edges of minimum cardinality that, if
removed, would break all paths among source and target in G. If not, it returns a set of edges of minimum cardinality
that disconnects G.
Parameters
G
[NetworkX graph]
s
t
flow_func
Returns
cutset
[set] Set of edges that, if removed, would disconnect G. If source and target nodes are provided,
the set contains the edges that if removed, would destroy all paths between source and target.
See also:
minimum_st_edge_cut()
minimum_node_cut()
stoer_wagner()
node_connectivity()
edge_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()
Notes
This is a flow based implementation of minimum edge cut. For undirected graphs the algorithm works by finding a
‘small’ dominating set of nodes of G (see algorithm 7 in [1]) and computing the maximum flow between an arbitrary
node in the dominating set and the rest of nodes in it. This is an implementation of algorithm 6 in [1]. For directed
graphs, the algorithm does n calls to the max flow function. The function raises an error if the directed graph is not
weakly connected and returns an empty set if it is weakly connected. It is an implementation of algorithm 8 in [1].

References
[1]
Examples
>>> # Platonic icosahedral graph has edge connectivity 5

>>> len(nx.minimum_edge_cut(G))
5

>>> len(nx.minimum_edge_cut(G, flow_func=shortest_augmenting_path))
5
If you specify a pair of nodes (source and target) as parameters, this function returns the value of local edge
connectivity.
>>> nx.edge_connectivity(G, 3, 7)
5
cal_edge_connectivity() for details.
minimum_node_cut
minimum_node_cut(G, s=None, t=None, flow_func=None)

Returns a set of nodes of minimum cardinality that disconnects G.
If source and target nodes are provided, this function returns the set of nodes of minimum cardinality that, if
removed, would destroy all paths among source and target in G. If not, it returns a set of nodes of minimum
cardinality that disconnects G.
Parameters
G
[NetworkX graph]
s
t
flow_func

Returns
cutset
[set] Set of nodes that, if removed, would disconnect G. If source and target nodes are provided,
the set contains the nodes that if removed, would destroy all paths between source and target.
See also:
minimum_st_node_cut()
minimum_cut()
minimum_edge_cut()
stoer_wagner()
node_connectivity()
edge_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()
Notes
This is a flow based implementation of minimum node cut. The algorithm is based in solving a number of maximum
flow computations to determine the capacity of the minimum cut on an auxiliary directed network that corresponds
to the minimum node cut of G. It handles both directed and undirected graphs. This implementation is based on
algorithm 11 in [1].
References
[1]
Examples
>>> # Platonic icosahedral graph has node connectivity 5

>>> node_cut = nx.minimum_node_cut(G)
>>> len(node_cut)
5
>>> node_cut == nx.minimum_node_cut(G, flow_func=shortest_augmenting_path)
True
If you specify a pair of nodes (source and target) as parameters, this function returns a local st node cut.
>>> len(nx.minimum_node_cut(G, 3, 7))
5

If you need to perform several local st cuts among different pairs of nodes on the same graph, it is recommended
that you reuse the data structures used in the maximum flow computations. See minimum_st_node_cut()
for details.
minimum_st_edge_cut
minimum_st_edge_cut(G, s, t, flow_func=None, auxiliary=None, residual=None)

Returns the edges of the cut-set of a minimum (s, t)-cut.
This function returns the set of edges of minimum cardinality that, if removed, would destroy all paths among
source and target in G. Edge weights are not considered. See minimum_cut() for computing minimum cuts
considering edge weights.
Parameters
G
[NetworkX graph]
s
t
auxiliary
flow_func
node_connectivity() for details. The choice of the default function may change from
version to version and should not be relied on. Default value: None.
residual
Returns
cutset
[set] Set of edges that, if removed from the graph, will disconnect it.
See also:
minimum_cut()
minimum_node_cut()
minimum_edge_cut()
stoer_wagner()
node_connectivity()
edge_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()

Examples
nectivity package:
>>> from networkx.algorithms.connectivity import minimum_st_edge_cut
We use in this example the platonic icosahedral graph, which has edge connectivity 5.
>>> len(minimum_st_edge_cut(G, 0, 6))
5
If you need to compute local edge cuts on several pairs of nodes in the same graph, it is recommended that you
Example of how to compute local edge cuts among all pairs of nodes of the platonic icosahedral graph reusing the
data structures.

>>> # flow package
>>> # as parameters
... k = len(minimum_st_edge_cut(G, u, v, auxiliary=H, residual=R))
True
You can also use alternative flow algorithms for computing edge cuts. For instance, in dense networks the algorithm
shortest_augmenting_path() will usually perform better than the default edmonds_karp() which is
faster for sparse networks with highly skewed degree distributions. Alternative flow functions have to be explicitly
imported from the flow package.

>>> len(minimum_st_edge_cut(G, 0, 6, flow_func=shortest_augmenting_path))
5

minimum_st_node_cut
minimum_st_node_cut(G, s, t, flow_func=None, auxiliary=None, residual=None)

Returns a set of nodes of minimum cardinality that disconnect source from target in G.
This function returns the set of nodes of minimum cardinality that, if removed, would destroy all paths among
source and target in G.
Parameters
G
[NetworkX graph]
s
[node] Source node.
t
[node] Target node.
flow_func
auxiliary
residual
Returns
cutset
[set] Set of nodes that, if removed, would destroy all paths between source and target in G.
See also:
minimum_node_cut()
minimum_edge_cut()
stoer_wagner()
node_connectivity()
edge_connectivity()
maximum_flow()
edmonds_karp()
preflow_push()

Notes
This is a flow based implementation of minimum node cut. The algorithm is based in solving a number of maximum
flow computations to determine the capacity of the minimum cut on an auxiliary directed network that corresponds
to the minimum node cut of G. It handles both directed and undirected graphs. This implementation is based on
algorithm 11 in [1].
References
[1]
Examples
nectivity package:
>>> from networkx.algorithms.connectivity import minimum_st_node_cut
We use in this example the platonic icosahedral graph, which has node connectivity 5.
>>> len(minimum_st_node_cut(G, 0, 6))
5
If you need to compute local st cuts between several pairs of nodes in the same graph, it is recommended that you
reuse the data structures that NetworkX uses in the computation: the auxiliary digraph for node connectivity and
node cuts, and the residual network for the underlying maximum flow computation.
Example of how to compute local st node cuts reusing the data structures:

>>> # flow package
>>> # as parameters
>>> len(minimum_st_node_cut(G, 0, 6, auxiliary=H, residual=R))
5
You can also use alternative flow algorithms for computing minimum st node cuts. For instance, in dense net-

>>> len(minimum_st_node_cut(G, 0, 6, flow_func=shortest_augmenting_path))
5

3.16.8 Stoer-Wagner minimum cut
Stoer-Wagner minimum cut algorithm.
stoer_wagner(G[, weight, heap]) Returns the weighted minimum edge cut using the Stoer-
Wagner algorithm.
stoer_wagner
stoer_wagner(G, weight='weight', heap=<class 'networkx.utils.heaps.BinaryHeap'>)

Returns the weighted minimum edge cut using the Stoer-Wagner algorithm.
Determine the minimum edge cut of a connected graph using the Stoer-Wagner algorithm. In weighted cases, all
weights must be nonnegative.
The running time of the algorithm depends on the type of heaps used:
Type of heap Running time

Binary heap O(n(m + n) log n)
2
Fibonacci heap O(nm√+ n log n)
log log n
Pairing heap O(22 nm + n2 log n)
Parameters
G
[NetworkX graph] Edges of the graph are expected to have an attribute named by the weight
parameter below. If this attribute is not present, the edge is considered to have unit weight.
weight
[string] Name of the weight attribute of the edges. If the attribute is not present, unit weight
is assumed. Default value: ‘weight’.
heap
[class] Type of heap to be used in the algorithm. It should be a subclass of MinHeap or
implement a compatible interface.
If a stock heap implementation is to be used, BinaryHeap is recommended over Pair-
ingHeap for Python implementations without optimized attribute accesses (e.g., CPython)
despite a slower asymptotic running time. For Python implementations with optimized at-
tribute accesses (e.g., PyPy), PairingHeap provides better performance. Default value:
BinaryHeap.
Returns
cut_value
[integer or float] The sum of weights of edges in a minimum cut.
partition
[pair of node lists] A partitioning of the nodes that defines a minimum cut.
Raises
If the graph is directed or a multigraph.
NetworkXError
If the graph has less than two nodes, is not connected or has a negative-weighted edge.

Examples
>>> G = nx.Graph()
>>> G.add_edge("x", "a", weight=3)
>>> G.add_edge("x", "b", weight=1)
>>> G.add_edge("a", "c", weight=3)
>>> G.add_edge("b", "d", weight=4)
>>> G.add_edge("d", "e", weight=2)
>>> G.add_edge("c", "y", weight=2)
>>> G.add_edge("e", "y", weight=3)
>>> cut_value, partition = nx.stoer_wagner(G)
>>> cut_value
4
3.16.9 Utils for flow-based connectivity
Utilities for connectivity package
build_auxiliary_edge_connectivity(G) Auxiliary digraph for computing flow based edge connec-

tivity
build_auxiliary_node_connectivity(G) Creates a directed graph D from an undirected graph G
to compute flow based node connectivity.
build_auxiliary_edge_connectivity
build_auxiliary_edge_connectivity(G)
Auxiliary digraph for computing flow based edge connectivity
If the input graph is undirected, we replace each edge (u,`v`) with two reciprocal arcs (u, v) and (v, u) and then
we set the attribute ‘capacity’ for each arc to 1. If the input graph is directed we simply add the ‘capacity’ attribute.
Part of algorithm 1 in [1] .
References
[1]
build_auxiliary_node_connectivity
build_auxiliary_node_connectivity(G)
Creates a directed graph D from an undirected graph G to compute flow based node connectivity.
For an undirected graph G having n nodes and m edges we derive a directed graph D with 2n nodes and 2m+n arcs
by replacing each original node v with two nodes vA, vB linked by an (internal) arc in D. Then for each edge (u,
v) in G we add two arcs (uB, vA) and (vB, uA) in D. Finally we set the attribute capacity = 1 for each arc in D [1].
For a directed graph having n nodes and m arcs we derive a directed graph D with 2n nodes and m+n arcs by
replacing each original node v with two nodes vA, vB linked by an (internal) arc (vA, vB) in D. Then for each arc
(u, v) in G we add one arc (uB, vA) in D. Finally we set the attribute capacity = 1 for each arc in D.
A dictionary with a mapping between nodes in the original graph and the auxiliary digraph is stored as a graph
attribute: H.graph[‘mapping’].

References
[1]
3.17 Cores
Find the k-cores of a graph.

The k-core is found by recursively pruning nodes with degrees less than k.
See the following references for details:
An O(m) Algorithm for Cores Decomposition of Networks Vladimir Batagelj and Matjaz Zaversnik, 2003. https://arxiv.
org/abs/cs.DS/0310049
Generalized Cores Vladimir Batagelj and Matjaz Zaversnik, 2002. https://arxiv.org/pdf/cs/0202039
For directed graphs a more general notion is that of D-cores which looks at (k, l) restrictions on (in, out) degree. The (k,
k) D-core is the k-core.
D-cores: Measuring Collaboration of Directed Graphs Based on Degeneracy Christos Giatsidis, Dimitrios M. Thilikos,
Michalis Vazirgiannis, ICDM 2011. http://www.graphdegeneracy.org/dcores_ICDM_2011.pdf
Multi-scale structure and topological anomaly detection via a new network statistic: The onion decomposition L. Hébert-
Dufresne, J. A. Grochow, and A. Allard Scientific Reports 6, 31708 (2016) http://doi.org/10.1038/srep31708
core_number(G) Returns the core number for each vertex.

k_core(G[, k, core_number]) Returns the k-core of G.
k_shell(G[, k, core_number]) Returns the k-shell of G.
k_crust(G[, k, core_number]) Returns the k-crust of G.
k_corona(G, k[, core_number]) Returns the k-corona of G.
k_truss(G, k) Returns the k-truss of G.
onion_layers(G) Returns the layer of each vertex in an onion decomposi-
tion of the graph.
3.17.1 core_number
core_number(G)
Returns the core number for each vertex.
A k-core is a maximal subgraph that contains nodes of degree k or more.
The core number of a node is the largest value k of a k-core containing that node.
Parameters
G
[NetworkX graph] A graph or directed graph
Returns
core_number
[dictionary] A dictionary keyed by node to the core number.
Raises
NetworkXError
The k-core is not implemented for graphs with self loops or parallel edges.

Notes
Not implemented for graphs with parallel edges or self loops.

For directed graphs the node degree is defined to be the in-degree + out-degree.
References
[1]
3.17.2 k_core
k_core(G, k=None, core_number=None)

Returns the k-core of G.
A k-core is a maximal subgraph that contains nodes of degree k or more.
Parameters
G
k
[int, optional] The order of the core. If not specified return the main core.
core_number
[dictionary, optional] Precomputed core numbers for the graph G.
Returns
G
[NetworkX graph] The k-core subgraph
Raises
NetworkXError
The k-core is not defined for graphs with self loops or parallel edges.
See also:
core_number
Notes
The main core is the core with the largest degree.

Graph, node, and edge attributes are copied to the subgraph.
3.17. Cores 397

References
[1]
3.17.3 k_shell
k_shell(G, k=None, core_number=None)

Returns the k-shell of G.
The k-shell is the subgraph induced by nodes with core number k. That is, nodes in the k-core that are not in the
(k+1)-core.
Parameters
G
[NetworkX graph] A graph or directed graph.
k
[int, optional] The order of the shell. If not specified return the outer shell.
core_number
Returns
G
[NetworkX graph] The k-shell subgraph
Raises
NetworkXError
The k-shell is not implemented for graphs with self loops or parallel edges.
See also:
core_number
k_corona
Notes
This is similar to k_corona but in that case only neighbors in the k-core are considered.

References
[1]
3.17.4 k_crust
k_crust(G, k=None, core_number=None)

Returns the k-crust of G.
The k-crust is the graph G with the edges of the k-core removed and isolated nodes found after the removal of
edges are also removed.
Parameters
G
[NetworkX graph] A graph or directed graph.
k
[int, optional] The order of the shell. If not specified return the main crust.
core_number
Returns
G
[NetworkX graph] The k-crust subgraph
Raises
NetworkXError
The k-crust is not implemented for graphs with self loops or parallel edges.
See also:
core_number
Notes
This definition of k-crust is different than the definition in [1]. The k-crust in [1] is equivalent to the k+1 crust of
this algorithm.
3.17. Cores 399

References
[1]
3.17.5 k_corona
k_corona(G, k, core_number=None)
Returns the k-corona of G.
The k-corona is the subgraph of nodes in the k-core which have exactly k neighbours in the k-core.
Parameters
G
k
[int] The order of the corona.
core_number
Returns
G
[NetworkX graph] The k-corona subgraph
Raises
NetworkXError
The k-cornoa is not defined for graphs with self loops or parallel edges.
See also:
core_number
Notes

References
[1]

3.17.6 k_truss
k_truss(G, k)
Returns the k-truss of G.
The k-truss is the maximal induced subgraph of G which contains at least three vertices where every edge is incident
to at least k-2 triangles.
Parameters
G
k
[int] The order of the truss
Returns
H
[NetworkX graph] The k-truss subgraph
Raises
NetworkXError
The k-truss is not defined for graphs with self loops or parallel edges or directed graphs.
Notes
A k-clique is a (k-2)-truss and a k-truss is a (k+1)-core.

Not implemented for digraphs or graphs with parallel edges or self loops.
K-trusses were originally defined in [2] which states that the k-truss is the maximal induced subgraph where each
edge belongs to at least k-2 triangles. A more recent paper, [1], uses a slightly different definition requiring that
each edge belong to at least k triangles. This implementation uses the original definition of k-2 triangles.
References
[1], [2]
3.17.7 onion_layers
onion_layers(G)
Returns the layer of each vertex in an onion decomposition of the graph.
The onion decomposition refines the k-core decomposition by providing information on the internal organization
of each k-shell. It is usually used alongside the core numbers.
Parameters
G
[NetworkX graph] A simple graph without self loops or parallel edges
Returns
3.17. Cores 401

od_layers
[dictionary] A dictionary keyed by vertex to the onion layer. The layers are contiguous integers
starting at 1.
Raises
NetworkXError
The onion decomposition is not implemented for graphs with self loops or parallel edges or for
directed graphs.
See also:
core_number
Notes

Not implemented for directed graphs.
References
[1], [2]
3.18 Covering
Functions related to graph covers.
min_edge_cover(G[, matching_algorithm]) Returns the min cardinality edge cover of the graph as a
set of edges.
is_edge_cover(G, cover) Decides whether a set of edges is a valid edge cover of the
graph.
3.18.1 min_edge_cover
min_edge_cover(G, matching_algorithm=None)
Returns the min cardinality edge cover of the graph as a set of edges.
A smallest edge cover can be found in polynomial time by finding a maximum matching and extending it greedily
so that all nodes are covered. This function follows that process. A maximum matching algorithm can be specified
for the first step of the algorithm. The resulting set may return a set with one 2-tuple for each edge, (the usual case)
or with both 2-tuples (u, v) and (v, u) for each edge. The latter is only done when a bipartite matching
algorithm is specified as matching_algorithm.
Parameters
G
matching_algorithm
[function] A function that returns a maximum cardinality matching for G. The function

must take one input, the graph G, and return either a set of edges (with only one direc-
tion for the pair of nodes) or a dictionary mapping each node to its mate. If not speci-
fied, max_weight_matching() is used. Common bipartite matching functions include
hopcroft_karp_matching() or eppstein_matching().
Returns
min_cover
[set] A set of the edges in a minimum edge cover in the form of tuples. It contains only one of
the equivalent 2-tuples (u, v) and (v, u) for each edge. If a bipartite method is used to
compute the matching, the returned set contains both the 2-tuples (u, v) and (v, u) for
each edge of a minimum edge cover.
Notes
set. The minimum edge cover is an edge covering of smallest cardinality.
Due to its implementation, the worst-case running time of this algorithm is bounded by the worst-case running time
of the function matching_algorithm.
Minimum edge cover for G can also be found using the min_edge_covering function in networkx.
algorithms.bipartite.covering which is simply this function with a default matching algorithm of
hopcraft_karp_matching()
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3)])
>>> sorted(nx.min_edge_cover(G))
[(2, 1), (3, 0)]
3.18.2 is_edge_cover
is_edge_cover(G, cover)
Decides whether a set of edges is a valid edge cover of the graph.
Given a set of edges, whether it is an edge covering can be decided if we just check whether all nodes of the graph
has an edge from the set, incident on it.
Parameters
G
[NetworkX graph] An undirected bipartite graph.
cover
[set] Set of edges to be checked.
Returns
bool
Whether the set of edges is a valid edge cover of the graph.
3.18. Covering 403

Notes
set.
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3)])
>>> cover = {(2, 1), (3, 0)}
>>> nx.is_edge_cover(G, cover)
True
3.19 Cycles
cycle_basis(G[, root]) Returns a list of cycles which form a basis for cycles of G.
simple_cycles(G) Find simple cycles (elementary circuits) of a directed
graph.
recursive_simple_cycles(G) Find simple cycles (elementary circuits) of a directed
graph.
find_cycle(G[, source, orientation]) Returns a cycle found via depth-first traversal.
minimum_cycle_basis(G[, weight]) Returns a minimum weight cycle basis for G
3.19.1 cycle_basis
cycle_basis(G, root=None)
Returns a list of cycles which form a basis for cycles of G.
A basis for cycles of a network is a minimal collection of cycles such that any cycle in the network can be written
as a sum of cycles in the basis. Here summation of cycles is defined as “exclusive or” of the edges. Cycle bases are
useful, e.g. when deriving equations for electric circuits using Kirchhoff’s Laws.
Parameters
G
[NetworkX Graph]
root
[node, optional] Specify starting node for basis.
Returns
A list of cycle lists. Each cycle list is a list of nodes
which forms a cycle (loop) in G.
See also:
simple_cycles

Notes
This is adapted from algorithm CACM 491 [1].
References
[1]
Examples
>>> G = nx.Graph()
>>> nx.add_cycle(G, [0, 1, 2, 3])
>>> nx.add_cycle(G, [0, 3, 4, 5])
>>> print(nx.cycle_basis(G, 0))
[[3, 4, 5, 0], [1, 2, 3, 0]]
3.19.2 simple_cycles
simple_cycles(G)
Find simple cycles (elementary circuits) of a directed graph.
A simple cycle, or elementary circuit, is a closed path where no node appears twice. Two elementary
circuits are distinct if they are not cyclic permutations of each other.
This is a nonrecursive, iterator/generator version of Johnson’s algorithm [1]. There may be better algorithms for
some cases [2] [3].
Parameters
G
[NetworkX DiGraph] A directed graph
Yields
list of nodes
Each cycle is represented by a list of nodes along the cycle.
See also:
cycle_basis
Notes
The implementation follows pp. 79-80 in [1].

The time complexity is O((n + e)(c + 1)) for n nodes, e edges and c elementary circuits.
3.19. Cycles 405

References
[1], [2], [3]
Examples
>>> edges = [(0, 0), (0, 1), (0, 2), (1, 2), (2, 0), (2, 1), (2, 2)]
>>> G = nx.DiGraph(edges)
>>> sorted(nx.simple_cycles(G))
[[0], [0, 1, 2], [0, 2], [1, 2], [2]]
To filter the cycles so that they don’t include certain nodes or edges, copy your graph and eliminate those nodes or
edges before calling. For example, to exclude self-loops from the above example:
>>> H = G.copy()
>>> H.remove_edges_from(nx.selfloop_edges(G))
>>> sorted(nx.simple_cycles(H))
[[0, 1, 2], [0, 2], [1, 2]]
3.19.3 recursive_simple_cycles
recursive_simple_cycles(G)
Find simple cycles (elementary circuits) of a directed graph.
A simple cycle, or elementary circuit, is a closed path where no node appears twice. Two elementary
circuits are distinct if they are not cyclic permutations of each other.
This version uses a recursive algorithm to build a list of cycles. You should probably use the iterator version called
simple_cycles(). Warning: This recursive version uses lots of RAM! It appears in NetworkX for pedagogical value.
Parameters
G
Returns
A list of cycles, where each cycle is represented by a list of nodes
along the cycle.
Example:
>>> edges = [(0, 0), (0, 1), (0, 2), (1, 2), (2, 0), (2, 1), (2, 2)]
..
>>> G = nx.DiGraph(edges)
..
>>> nx.recursive_simple_cycles(G)
..
[[0], [2], [0, 1, 2], [0, 2], [1, 2]]
See also:
simple_cycles, cycle_basis

Notes
The implementation follows pp. 79-80 in [1].

The time complexity is O((n + e)(c + 1)) for n nodes, e edges and c elementary circuits.
References
[1]
3.19.4 find_cycle
find_cycle(G, source=None, orientation=None)

Returns a cycle found via depth-first traversal.
The cycle is a list of edges indicating the cyclic path. Orientation of directed edges is controlled by orientation.
Parameters
G
[graph] A directed/undirected graph/multigraph.
source
[node, list of nodes] The node from which the traversal begins. If None, then a source is chosen
arbitrarily and repeatedly until all edges from each node in the graph are searched.
orientation
[None | ‘original’ | ‘reverse’ | ‘ignore’ (default: None)] For directed graphs and directed multi-
graphs, edge traversals need not respect the original orientation of the edges. When set to
‘reverse’ every edge is traversed in the reverse direction. When set to ‘ignore’, every edge is
treated as undirected. When set to ‘original’, every edge is treated as directed. In all three
cases, the yielded edge tuples add a last entry to indicate the direction in which that edge was
traversed. If orientation is None, the yielded edge has no direction indicated. The direction is
respected, but not reported.
Returns
edges
[directed edges] A list of directed edges indicating the path taken for the loop. If no cycle is
found, then an exception is raised. For graphs, an edge is of the form (u, v) where u and v
are the tail and head of the edge as determined by the traversal. For multigraphs, an edge is of
the form (u, v, key), where key is the key of the edge. When the graph is directed, then
u and v are always in the order of the actual directed edge. If orientation is not None then the
edge tuple is extended to include the direction of traversal (‘forward’ or ‘reverse’) on that edge.
Raises
NetworkXNoCycle
If no cycle was found.
See also:
simple_cycles
3.19. Cycles 407

Examples
In this example, we construct a DAG and find, in the first call, that there are no directed cycles, and so an exception
is raised. In the second call, we ignore edge orientations and find that there is an undirected cycle. Note that the
second call finds a directed cycle while effectively traversing an undirected graph, and so, we found an “undirected
cycle”. This means that this DAG structure does not form a directed tree (which is also known as a polytree).
>>> G = nx.DiGraph([(0, 1), (0, 2), (1, 2)])

>>> nx.find_cycle(G, orientation="original")
Traceback (most recent call last):
...
networkx.exception.NetworkXNoCycle: No cycle found.
>>> list(nx.find_cycle(G, orientation="ignore"))
[(0, 1, 'forward'), (1, 2, 'forward'), (0, 2, 'reverse')]
3.19.5 minimum_cycle_basis
minimum_cycle_basis(G, weight=None)
Returns a minimum weight cycle basis for G
Minimum weight means a cycle basis for which the total weight (length for unweighted graphs) of all the cycles is
minimum.
Parameters
G
[NetworkX Graph]
weight: string
name of the edge attribute to use for edge weights
Returns
A list of cycle lists. Each cycle list is a list of nodes
which forms a cycle (loop) in G. Note that the nodes are not
necessarily returned in a order by which they appear in the cycle
See also:
simple_cycles, cycle_basis
Examples
>>> G = nx.Graph()
>>> nx.add_cycle(G, [0, 1, 2, 3])
>>> nx.add_cycle(G, [0, 3, 4, 5])
>>> print([sorted(c) for c in nx.minimum_cycle_basis(G)])
[[0, 1, 2, 3], [0, 3, 4, 5]]
References:
[1] Kavitha, Telikepalli, et al. “An O(m^2n) Algorithm for Minimum Cycle Basis of Graphs.” http:
//link.springer.com/article/10.1007/s00453-007-9064-z [2] de Pina, J. 1995. Applications of shortest path
methods. Ph.D. thesis, University of Amsterdam, Netherlands

3.20 Cuts
Functions for finding and evaluating cuts in a graph.
boundary_expansion(G, S) Returns the boundary expansion of the set S.

conductance(G, S[, T, weight]) Returns the conductance of two sets of nodes.
cut_size(G, S[, T, weight]) Returns the size of the cut between two sets of nodes.
edge_expansion(G, S[, T, weight]) Returns the edge expansion between two node sets.
mixing_expansion(G, S[, T, weight]) Returns the mixing expansion between two node sets.
node_expansion(G, S) Returns the node expansion of the set S.
normalized_cut_size(G, S[, T, weight]) Returns the normalized size of the cut between two sets
of nodes.
volume(G, S[, weight]) Returns the volume of a set of nodes.
3.20.1 boundary_expansion
boundary_expansion(G, S)
Returns the boundary expansion of the set S.
The boundary expansion is the quotient of the size of the node boundary and the cardinality of S. [1]
Parameters
G
[NetworkX graph]
S
[collection] A collection of nodes in G.
Returns
number
The boundary expansion of the set S.
See also:
edge_expansion
mixing_expansion
node_expansion
References
[1]
3.20. Cuts 409

3.20.2 conductance
conductance(G, S, T=None, weight=None)

Returns the conductance of two sets of nodes.
The conductance is the quotient of the cut size and the smaller of the volumes of the two sets. [1]
Parameters
G
[NetworkX graph]
S
T
weight
Returns
number
The conductance between the two sets S and T.
See also:
cut_size
edge_expansion
normalized_cut_size
volume
References
[1]
3.20.3 cut_size
cut_size(G, S, T=None, weight=None)

Returns the size of the cut between two sets of nodes.
A cut is a partition of the nodes of a graph into two sets. The cut size is the sum of the weights of the edges
“between” the two sets of nodes.
Parameters
G
[NetworkX graph]
S
T
[collection] A collection of nodes in G. If not specified, this is taken to be the set complement
of S.
weight

Returns
number
Total weight of all edges from nodes in set S to nodes in set T (and, in the case of directed
graphs, all edges from nodes in T to nodes in S).
Notes
In a multigraph, the cut size is the total weight of edges including multiplicity.
Examples
In the graph with two cliques joined by a single edges, the natural bipartition of the graph into two blocks, one for
each clique, yields a cut of weight one:
>>> S = {0, 1, 2}
>>> T = {3, 4, 5}
>>> nx.cut_size(G, S, T)
1
Each parallel edge in a multigraph is counted when determining the cut size:
>>> G = nx.MultiGraph(["ab", "ab"])

>>> S = {"a"}
>>> T = {"b"}
>>> nx.cut_size(G, S, T)
2
3.20.4 edge_expansion
edge_expansion(G, S, T=None, weight=None)

Returns the edge expansion between two node sets.
The edge expansion is the quotient of the cut size and the smaller of the cardinalities of the two sets. [1]
Parameters
G
[NetworkX graph]
S
T
weight
Returns
number
The edge expansion between the two sets S and T.
See also:
3.20. Cuts 411

boundary_expansion
mixing_expansion
node_expansion
References
[1]
3.20.5 mixing_expansion
mixing_expansion(G, S, T=None, weight=None)

Returns the mixing expansion between two node sets.
The mixing expansion is the quotient of the cut size and twice the number of edges in the graph. [1]
Parameters
G
[NetworkX graph]
S
T
weight
Returns
number
The mixing expansion between the two sets S and T.
See also:
boundary_expansion
edge_expansion
node_expansion
References
[1]
3.20.6 node_expansion
node_expansion(G, S)
Returns the node expansion of the set S.
The node expansion is the quotient of the size of the node boundary of S and the cardinality of S. [1]
Parameters
G
[NetworkX graph]

S
Returns
number
The node expansion of the set S.
See also:
boundary_expansion
edge_expansion
mixing_expansion
References
[1]
3.20.7 normalized_cut_size
normalized_cut_size(G, S, T=None, weight=None)

Returns the normalized size of the cut between two sets of nodes.
The normalized cut size is the cut size times the sum of the reciprocal sizes of the volumes of the two sets. [1]
Parameters
G
[NetworkX graph]
S
T
weight
Returns
number
The normalized cut size between the two sets S and T.
See also:
conductance
cut_size
edge_expansion
volume
3.20. Cuts 413

Notes
In a multigraph, the cut size is the total weight of edges including multiplicity.
References
[1]
3.20.8 volume
volume(G, S, weight=None)
Returns the volume of a set of nodes.
The volume of a set S is the sum of the (out-)degrees of nodes in S (taking into account parallel edges in multigraphs).
[1]
Parameters
G
[NetworkX graph]
S
weight
Returns
number
The volume of the set of nodes represented by S in the graph G.
See also:
conductance
cut_size
edge_expansion
edge_boundary
normalized_cut_size
References
[1]
3.21 D-Separation
Algorithm for testing d-separation in DAGs.

d-separation is a test for conditional independence in probability distributions that can be factorized using DAGs. It is a
purely graphical test that uses the underlying graph and makes no reference to the actual distribution parameters. See [1]
for a formal definition.
The implementation is based on the conceptually simple linear time algorithm presented in [2]. Refer to [3], [4] for a
couple of alternative algorithms.

Here, we provide a brief overview of d-separation and related concepts that are relevant for understanding it:
3.21.1 Blocking paths
Before we overview, we introduce the following terminology to describe paths:

• “open” path: A path between two nodes that can be traversed
• “blocked” path: A path between two nodes that cannot be traversed
A collider is a triplet of nodes along a path that is like the following: ... u -> c <- v ...), where ‘c’ is a
common successor of u and v. A path through a collider is considered “blocked”. When a node that is a collider, or a
descendant of a collider is included in the d-separating set, then the path through that collider node is “open”. If the path
through the collider node is open, then we will call this node an open collider.
The d-separation set blocks the paths between u and v. If you include colliders, or their descendant nodes in the d-
separation set, then those colliders will open up, enabling a path to be traversed if it is not blocked some other way.
3.21.2 Illustration of D-separation with examples
For a pair of two nodes, u and v, all paths are considered open if there is a path between u and v that is not blocked.
That means, there is an open path between u and v that does not encounter a collider, or a variable in the d-separating
set.
For example, if the d-separating set is the empty set, then the following paths are unblocked between u and v:
• u <- z -> v
• u -> w -> … -> z -> v
If for example, ‘z’ is in the d-separating set, then ‘z’ blocks those paths between u and v.
Colliders block a path by default if they and their descendants are not included in the d-separating set. An example of a
path that is blocked when the d-separating set is empty is:
• u -> w -> … -> z <- v
because ‘z’ is a collider in this path and ‘z’ is not in the d-separating set. However, if ‘z’ or a descendant of ‘z’ is included
in the d-separating set, then the path through the collider at ‘z’ (… -> z <- …) is now “open”.
D-separation is concerned with blocking all paths between u and v. Therefore, a d-separating set between u and v is one
where all paths are blocked.
3.21.3 D-separation and its applications in probability
D-separation is commonly used in probabilistic graphical models. D-separation connects the idea of probabilistic “depen-
dence” with separation in a graph. If one assumes the causal Markov condition [5], then d-separation implies conditional
independence in probability distributions.
3.21. D-Separation 415

3.21.4 Examples
>>>
>>> # HMM graph with five states and observation nodes
... g = nx.DiGraph()
>>> g.add_edges_from(
... [
... ("S1", "S2"),
... ("S2", "S3"),
... ("S3", "S4"),
... ("S4", "S5"),
... ("S1", "O1"),
... ("S2", "O2"),
... ("S3", "O3"),
... ("S4", "O4"),
... ("S5", "O5"),
... ]
... )
>>>
>>> # states/obs before 'S3' are d-separated from states/obs after 'S3'
... nx.d_separated(g, {"S1", "S2", "O1", "O2"}, {"S4", "S5", "O4", "O5"}, {"S3"})
True
3.21.5 References
d_separated(G, x, y, z) Return whether node sets x and y are d-separated by z.
3.21.6 d_separated
d_separated(G, x, y, z)
Return whether node sets x and y are d-separated by z.
Parameters
G
[graph] A NetworkX DAG.
x
[set] First set of nodes in G.
y
[set] Second set of nodes in G.
z
[set] Set of conditioning nodes in G. Can be empty set.
Returns
b
[bool] A boolean that is true if x is d-separated from y given z in G.
Raises
NetworkXError
The d-separation test is commonly used with directed graphical models which are acyclic.
Accordingly, the algorithm raises a NetworkXError if the input graph is not a DAG.

NodeNotFound
If any of the input nodes are not found in the graph, a NodeNotFound exception is raised.
Notes
A d-separating set in a DAG is a set of nodes that blocks all paths between the two sets. Nodes in z block a path
if they are part of the path and are not a collider, or a descendant of a collider. A collider structure along a path is
... -> c <- ... where c is the collider node.
https://en.wikipedia.org/wiki/Bayesian_network#d-separation
3.22 Directed Acyclic Graphs
Algorithms for directed acyclic graphs (DAGs).

Note that most of these functions are only guaranteed to work for DAGs. In general, these functions do not check for
acyclic-ness, so it is up to the user to check for that.
ancestors(G, source) Returns all nodes having a path to source in G.

descendants(G, source) Returns all nodes reachable from source in G.
topological_sort(G) Returns a generator of nodes in topologically sorted order.
topological_generations(G) Stratifies a DAG into generations.
all_topological_sorts(G) Returns a generator of _all_ topological sorts of the di-
rected graph G.
lexicographical_topological_sort(G[, Generate the nodes in the unique lexicographical topolog-
key]) ical sort order.
is_directed_acyclic_graph(G) Returns True if the graph G is a directed acyclic graph
(DAG) or False if not.
is_aperiodic(G) Returns True if G is aperiodic.
transitive_closure(G[, reflexive]) Returns transitive closure of a graph
transitive_closure_dag(G[, topo_order]) Returns the transitive closure of a directed acyclic graph.
transitive_reduction(G) Returns transitive reduction of a directed graph
antichains(G[, topo_order]) Generates antichains from a directed acyclic graph
(DAG).
dag_longest_path(G[, weight, ...]) Returns the longest path in a directed acyclic graph
(DAG).
dag_longest_path_length(G[, weight, ...]) Returns the longest path length in a DAG
dag_to_branching(G) Returns a branching representing all (overlapping) paths
from root nodes to leaf nodes in the given directed acyclic
graph.
3.22. Directed Acyclic Graphs 417

3.22.1 ancestors
ancestors(G, source)
Returns all nodes having a path to source in G.
Parameters
G
[NetworkX Graph]
source
[node in G]
Returns
set()
The ancestors of source in G
Raises
NetworkXError
If node source is not in G.
See also:
descendants
Examples
>>> DG = nx.path_graph(5, create_using=nx.DiGraph)

>>> sorted(nx.ancestors(DG, 2))
[0, 1]
The source node is not an ancestor of itself, but can be included manually:
>>> sorted(nx.ancestors(DG, 2) | {2})

[0, 1, 2]
3.22.2 descendants
descendants(G, source)
Returns all nodes reachable from source in G.
Parameters
G
[NetworkX Graph]
source
[node in G]
Returns
set()
The descendants of source in G
Raises

NetworkXError
If node source is not in G.
See also:
ancestors
Examples
>>> DG = nx.path_graph(5, create_using=nx.DiGraph)

>>> sorted(nx.descendants(DG, 2))
[3, 4]
The source node is not a descendant of itself, but can be included manually:
>>> sorted(nx.descendants(DG, 2) | {2})

[2, 3, 4]
3.22.3 topological_sort
topological_sort(G)
Returns a generator of nodes in topologically sorted order.
A topological sort is a nonunique permutation of the nodes of a directed graph such that an edge from u to v implies
that u appears before v in the topological sort order. This ordering is valid only if the graph has no directed cycles.
Parameters
G
[NetworkX digraph] A directed acyclic graph (DAG)
Yields
nodes
Yields the nodes in topological sorted order.
Raises
NetworkXError
Topological sort is defined for directed graphs only. If the graph G is undirected, a Net-
NetworkXUnfeasible
If G is not a directed acyclic graph (DAG) no topological sort exists and a NetworkXUn-
feasible exception is raised. This can also be raised if G is changed while the returned
iterator is being processed
RuntimeError
If G is changed while the returned iterator is being processed.
See also:
is_directed_acyclic_graph, lexicographical_topological_sort

Notes
This algorithm is based on a description and proof in “Introduction to Algorithms: A Creative Approach” [1] .
References
[1]
Examples
To get the reverse order of the topological sort:
>>> DG = nx.DiGraph([(1, 2), (2, 3)])

>>> list(reversed(list(nx.topological_sort(DG))))
[3, 2, 1]
If your DiGraph naturally has the edges representing tasks/inputs and nodes representing people/processes that
initiate tasks, then topological_sort is not quite what you need. You will have to change the tasks to nodes with
dependence reflected by edges. The result is a kind of topological sort of the edges. This can be done with
networkx.line_graph() as follows:
>>> list(nx.topological_sort(nx.line_graph(DG)))
[(1, 2), (2, 3)]
3.22.4 topological_generations
topological_generations(G)
Stratifies a DAG into generations.
A topological generation is node collection in which ancestors of a node in each generation are guaranteed to be
in a previous generation, and any descendants of a node are guaranteed to be in a following generation. Nodes are
guaranteed to be in the earliest possible generation that they can belong to.
Parameters
G
Yields
sets of nodes
Yields sets of nodes representing each generation.
Raises
NetworkXError
Generations are defined for directed graphs only. If the graph G is undirected, a NetworkX-
Error is raised.
NetworkXUnfeasible
If G is not a directed acyclic graph (DAG) no topological generations exist and a NetworkX-
Unfeasible exception is raised. This can also be raised if G is changed while the returned
RuntimeError

See also:
topological_sort
Notes
The generation in which a node resides can also be determined by taking the max-path-distance from
the node to the farthest leaf node. That value can be obtained with this function using enumer-
ate(topological_generations(G)).
Examples
>>> DG = nx.DiGraph([(2, 1), (3, 1)])

>>> [sorted(generation) for generation in nx.topological_generations(DG)]
[[2, 3], [1]]
3.22.5 all_topological_sorts
all_topological_sorts(G)
Returns a generator of _all_ topological sorts of the directed graph G.
A topological sort is a nonunique permutation of the nodes such that an edge from u to v implies that u appears
before v in the topological sort order.
Parameters
G
Yields
topological_sort_order
[list] a list of nodes in G, representing one of the topological sort orders
Raises
If G is not directed
NetworkXUnfeasible
If G is not acyclic
Notes
Implements an iterative version of the algorithm given in [1].

References
[1]
Examples
To enumerate all topological sorts of directed graph:
>>> DG = nx.DiGraph([(1, 2), (2, 3), (2, 4)])

>>> list(nx.all_topological_sorts(DG))
[[1, 2, 4, 3], [1, 2, 3, 4]]
3.22.6 lexicographical_topological_sort
lexicographical_topological_sort(G, key=None)
Generate the nodes in the unique lexicographical topological sort order.
Generates a unique ordering of nodes by first sorting topologically (for which there are often multiple valid order-
ings) and then additionally by sorting lexicographically.
A topological sort arranges the nodes of a directed graph so that the upstream node of each directed edge precedes
the downstream node. It is always possible to find a solution for directed graphs that have no cycles. There may be
more than one valid solution.
Lexicographical sorting is just sorting alphabetically. It is used here to break ties in the topological sort and to
determine a single, unique ordering. This can be useful in comparing sort results.
The lexicographical order can be customized by providing a function to the key= parameter. The definition of the
key function is the same as used in python’s built-in sort(). The function takes a single argument and returns a
key to use for sorting purposes.
Lexicographical sorting can fail if the node names are un-sortable. See the example below. The solution is to
provide a function to the key= argument that returns sortable keys.
Parameters
G
key
[function, optional] A function of one argument that converts a node name to a comparison
key. It defines and resolves ambiguities in the sort order. Defaults to the identity function.
Yields
nodes
Yields the nodes of G in lexicographical topological sort order.
Raises
NetworkXError
Topological sort is defined for directed graphs only. If the graph G is undirected, a Net-
NetworkXUnfeasible
If G is not a directed acyclic graph (DAG) no topological sort exists and a NetworkXUn-
feasible exception is raised. This can also be raised if G is changed while the returned

RuntimeError
TypeError
Results from un-sortable node names. Consider using key= parameter to resolve ambiguities
in the sort order.
See also:
topological_sort
Notes
This algorithm is based on a description and proof in “Introduction to Algorithms: A Creative Approach” [1] .
References
[1]
Examples
>>> DG = nx.DiGraph([(2, 1), (2, 5), (1, 3), (1, 4), (5, 4)])
>>> list(nx.lexicographical_topological_sort(DG))
[2, 1, 3, 5, 4]
>>> list(nx.lexicographical_topological_sort(DG, key=lambda x: -x))
[2, 5, 1, 4, 3]
The sort will fail for any graph with integer and string nodes. Comparison of integer to strings is not defined in
python. Is 3 greater or less than ‘red’?
>>> DG = nx.DiGraph([(1, 'red'), (3, 'red'), (1, 'green'), (2, 'blue')])

>>> list(nx.lexicographical_topological_sort(DG))
...
TypeError: '<' not supported between instances of 'str' and 'int'
...
Incomparable nodes can be resolved using a key function. This example function allows comparison of integers
and strings by returning a tuple where the first element is True for str, False otherwise. The second element is
the node name. This groups the strings and integers separately so they can be compared only among themselves.
>>> key = lambda node: (isinstance(node, str), node)

>>> list(nx.lexicographical_topological_sort(DG, key=key))
[1, 2, 3, 'blue', 'green', 'red']

3.22.7 is_directed_acyclic_graph
is_directed_acyclic_graph(G)
Returns True if the graph G is a directed acyclic graph (DAG) or False if not.
Parameters
G
[NetworkX graph]
Returns
bool
True if G is a DAG, False otherwise
See also:
topological_sort
Examples
Undirected graph:
>>> G = nx.Graph([(1, 2), (2, 3)])

>>> nx.is_directed_acyclic_graph(G)
False
Directed graph with cycle:
>>> G = nx.DiGraph([(1, 2), (2, 3), (3, 1)])

False
Directed acyclic graph:
>>> G = nx.DiGraph([(1, 2), (2, 3)])

True
3.22.8 is_aperiodic
is_aperiodic(G)
Returns True if G is aperiodic.
A directed graph is aperiodic if there is no integer k > 1 that divides the length of every cycle in the graph.
Parameters
G
Returns
bool
True if the graph is aperiodic False otherwise
Raises

NetworkXError
Notes
This uses the method outlined in [1], which runs in O(m) time given m edges in G. Note that a graph is not
aperiodic if it is acyclic as every integer trivial divides length 0 cycles.
References
[1]
Examples
A graph consisting of one cycle, the length of which is 2. Therefore k = 2 divides the length of every cycle in the
graph and thus the graph is not aperiodic:
>>> DG = nx.DiGraph([(1, 2), (2, 1)])

>>> nx.is_aperiodic(DG)
False
A graph consisting of two cycles: one of length 2 and the other of length 3. The cycle lengths are coprime, so there
is no single value of k where k > 1 that divides each cycle length and therefore the graph is aperiodic:
>>> DG = nx.DiGraph([(1, 2), (2, 3), (3, 1), (1, 4), (4, 1)])
True
A graph consisting of two cycles: one of length 2 and the other of length 4. The lengths of the cycles share a
common factor k = 2, and therefore the graph is not aperiodic:
>>> DG = nx.DiGraph([(1, 2), (2, 1), (3, 4), (4, 5), (5, 6), (6, 3)])
False
An acyclic graph, therefore the graph is not aperiodic:
>>> DG = nx.DiGraph([(1, 2), (2, 3)])

False
3.22.9 transitive_closure
transitive_closure(G, reflexive=False)
Returns transitive closure of a graph
The transitive closure of G = (V,E) is a graph G+ = (V,E+) such that for all v, w in V there is an edge (v, w) in E+
if and only if there is a path from v to w in G.
Handling of paths from v to v has some flexibility within this definition. A reflexive transitive closure creates a
self-loop for the path from v to v of length 0. The usual transitive closure creates a self-loop only if a cycle exists
(a path from v to v with length > 0). We also allow an option for no self-loops.

Parameters
G
[NetworkX Graph] A directed/undirected graph/multigraph.
reflexive
[Bool or None, optional (default: False)] Determines when cycles create self-loops in the Tran-
sitive Closure. If True, trivial cycles (length 0) create self-loops. The result is a reflexive transi-
tive closure of G. If False (the default) non-trivial cycles create self-loops. If None, self-loops
are not created.
Returns
NetworkX graph
The transitive closure of G
Raises
NetworkXError
If reflexive not in {None, True, False}
References
[1]
Examples
The treatment of trivial (i.e. length 0) cycles is controlled by the reflexive parameter.
Trivial (i.e. length 0) cycles do not create self-loops when reflexive=False (the default):
>>> DG = nx.DiGraph([(1, 2), (2, 3)])

>>> TC = nx.transitive_closure(DG, reflexive=False)
>>> TC.edges()
OutEdgeView([(1, 2), (1, 3), (2, 3)])
However, nontrivial (i.e. length greater then 0) cycles create self-loops when reflexive=False (the default):
>>> DG = nx.DiGraph([(1, 2), (2, 3), (3, 1)])

>>> TC = nx.transitive_closure(DG, reflexive=False)
>>> TC.edges()
OutEdgeView([(1, 2), (1, 3), (1, 1), (2, 3), (2, 1), (2, 2), (3, 1), (3, 2), (3,␣
,→3)])
Trivial cycles (length 0) create self-loops when reflexive=True:
>>> DG = nx.DiGraph([(1, 2), (2, 3)])

>>> TC = nx.transitive_closure(DG, reflexive=True)
>>> TC.edges()
OutEdgeView([(1, 2), (1, 1), (1, 3), (2, 3), (2, 2), (3, 3)])
And the third option is not to create self-loops at all when reflexive=None:
>>> DG = nx.DiGraph([(1, 2), (2, 3), (3, 1)])

>>> TC = nx.transitive_closure(DG, reflexive=None)
>>> TC.edges()
OutEdgeView([(1, 2), (1, 3), (2, 3), (2, 1), (3, 1), (3, 2)])

3.22.10 transitive_closure_dag
transitive_closure_dag(G, topo_order=None)
Returns the transitive closure of a directed acyclic graph.
This function is faster than the function transitive_closure, but fails if the graph has a cycle.
The transitive closure of G = (V,E) is a graph G+ = (V,E+) such that for all v, w in V there is an edge (v, w) in E+
if and only if there is a non-null path from v to w in G.
Parameters
G
[NetworkX DiGraph] A directed acyclic graph (DAG)
Returns
NetworkX DiGraph
The transitive closure of G
Raises
NetworkXUnfeasible
If G has a cycle
Notes
This algorithm is probably simple enough to be well-known but I didn’t find a mention in the literature.
Examples
>>> DG = nx.DiGraph([(1, 2), (2, 3)])

>>> TC = nx.transitive_closure_dag(DG)
>>> TC.edges()
OutEdgeView([(1, 2), (1, 3), (2, 3)])
3.22.11 transitive_reduction
transitive_reduction(G)
Returns transitive reduction of a directed graph
The transitive reduction of G = (V,E) is a graph G- = (V,E-) such that for all v,w in V there is an edge (v,w) in E-
if and only if (v,w) is in E and there is no path from v to w in G with length greater than 1.
Parameters
G
Returns

NetworkX DiGraph
The transitive reduction of G
Raises
NetworkXError
If G is not a directed acyclic graph (DAG) transitive reduction is not uniquely defined and a
NetworkXError exception is raised.
References
https://en.wikipedia.org/wiki/Transitive_reduction
Examples
To perform transitive reduction on a DiGraph:
>>> DG = nx.DiGraph([(1, 2), (2, 3), (1, 3)])

>>> TR = nx.transitive_reduction(DG)
>>> list(TR.edges)
[(1, 2), (2, 3)]
To avoid unnecessary data copies, this implementation does not return a DiGraph with node/edge data. To perform
transitive reduction on a DiGraph and transfer node/edge data:
>>> DG.add_edges_from([(1, 2), (2, 3), (1, 3)], color='red')
>>> TR = nx.transitive_reduction(DG)
>>> TR.add_nodes_from(DG.nodes(data=True))
>>> TR.add_edges_from((u, v, DG.edges[u, v]) for u, v in TR.edges)
>>> list(TR.edges(data=True))
[(1, 2, {'color': 'red'}), (2, 3, {'color': 'red'})]
3.22.12 antichains
antichains(G, topo_order=None)
Generates antichains from a directed acyclic graph (DAG).
An antichain is a subset of a partially ordered set such that any two elements in the subset are incomparable.
Parameters
G
Yields
antichain
[list] a list of nodes in G representing an antichain
Raises

NetworkXUnfeasible
If G contains a cycle
Notes
This function was originally developed by Peter Jipsen and Franco Saliola for the SAGE project. It’s included in
NetworkX with permission from the authors. Original SAGE code at:
https://github.com/sagemath/sage/blob/master/src/sage/combinat/posets/hasse_diagram.py
References
[1]
Examples
>>> DG = nx.DiGraph([(1, 2), (1, 3)])

>>> list(nx.antichains(DG))
[[], [3], [2], [2, 3], [1]]
3.22.13 dag_longest_path
dag_longest_path(G, weight='weight', default_weight=1, topo_order=None)

Returns the longest path in a directed acyclic graph (DAG).
If G has edges with weight attribute the edge data are used as weight values.
Parameters
G
weight
[str, optional] Edge data key to use for weight
default_weight
[int, optional] The weight of edges that do not have a weight attribute
Returns
list
Longest path
Raises
See also:
dag_longest_path_length

Examples
>>> DG = nx.DiGraph([(0, 1, {'cost':1}), (1, 2, {'cost':1}), (0, 2, {'cost':42})])

>>> list(nx.all_simple_paths(DG, 0, 2))
[[0, 1, 2], [0, 2]]
>>> nx.dag_longest_path(DG)
[0, 1, 2]
>>> nx.dag_longest_path(DG, weight="cost")
[0, 2]
In the case where multiple valid topological orderings exist, topo_order can be used to specify a specific or-
dering:
>>> DG = nx.DiGraph([(0, 1), (0, 2)])

>>> sorted(nx.all_topological_sorts(DG)) # Valid topological orderings
[[0, 1, 2], [0, 2, 1]]
>>> nx.dag_longest_path(DG, topo_order=[0, 1, 2])
[0, 1]
>>> nx.dag_longest_path(DG, topo_order=[0, 2, 1])
[0, 2]
3.22.14 dag_longest_path_length
dag_longest_path_length(G, weight='weight', default_weight=1)

Returns the longest path length in a DAG
Parameters
G
weight
[string, optional] Edge data key to use for weight
default_weight
[int, optional] The weight of edges that do not have a weight attribute
Returns
int
Longest path length
Raises
See also:
dag_longest_path

Examples
>>> DG = nx.DiGraph([(0, 1, {'cost':1}), (1, 2, {'cost':1}), (0, 2, {'cost':42})])

>>> list(nx.all_simple_paths(DG, 0, 2))
[[0, 1, 2], [0, 2]]
>>> nx.dag_longest_path_length(DG)
2
>>> nx.dag_longest_path_length(DG, weight="cost")
42
3.22.15 dag_to_branching
dag_to_branching(G)
Returns a branching representing all (overlapping) paths from root nodes to leaf nodes in the given directed acyclic
graph.
As described in networkx.algorithms.tree.recognition, a branching is a directed forest in which
each node has at most one parent. In other words, a branching is a disjoint union of arborescences. For this function,
each node of in-degree zero in G becomes a root of one of the arborescences, and there will be one leaf node for
each distinct path from that root to a leaf node in G.
Each node v in G with k parents becomes k distinct nodes in the returned branching, one for each parent, and the
sub-DAG rooted at v is duplicated for each copy. The algorithm then recurses on the children of each copy of v.
Parameters
G
[NetworkX graph] A directed acyclic graph.
Returns
DiGraph
The branching in which there is a bijection between root-to-leaf paths in G (in which multiple
paths may share the same leaf) and root-to-leaf paths in the branching (in which there is a
unique path from a root to a leaf).
Each node has an attribute ‘source’ whose value is the original node to which this node corre-
sponds. No other graph, node, or edge attributes are copied into this new graph.
Raises
If G is not directed, or if G is a multigraph.
HasACycle
If G is not acyclic.

Notes
This function is not idempotent in the sense that the node labels in the returned branching may be uniquely generated
each time the function is invoked. In fact, the node labels may not be integers; in order to relabel the nodes to be
more readable, you can use the networkx.convert_node_labels_to_integers() function.
The current implementation of this function uses networkx.prefix_tree(), so it is subject to the limitations
of that function.
Examples
To examine which nodes in the returned branching were produced by which original node in the directed acyclic
graph, we can collect the mapping from source node to new nodes into a dictionary. For example, consider the
directed diamond graph:
>>> from collections import defaultdict

>>> from operator import itemgetter
>>>
>>> G = nx.DiGraph(nx.utils.pairwise("abd"))
>>> G.add_edges_from(nx.utils.pairwise("acd"))
>>> B = nx.dag_to_branching(G)
>>>
>>> sources = defaultdict(set)
>>> for v, source in B.nodes(data="source"):
... sources[source].add(v)
>>> len(sources["a"])
1
>>> len(sources["d"])
2
To copy node attributes from the original graph to the new graph, you can use a dictionary like the one constructed
in the above example:
>>> for source, nodes in sources.items():

... for v in nodes:
... B.nodes[v].update(G.nodes[source])
3.23 Distance Measures
Graph diameter, radius, eccentricity and other properties.
barycenter(G[, weight, attr, sp]) Calculate barycenter of a connected graph, optionally

with edge weights.
center(G[, e, usebounds, weight]) Returns the center of the graph G.
diameter(G[, e, usebounds, weight]) Returns the diameter of the graph G.
eccentricity(G[, v, sp, weight]) Returns the eccentricity of nodes in G.
periphery(G[, e, usebounds, weight]) Returns the periphery of the graph G.
radius(G[, e, usebounds, weight]) Returns the radius of the graph G.
resistance_distance(G, nodeA, nodeB[, ...]) Returns the resistance distance between node A and node
B on graph G.

3.23.1 barycenter
barycenter(G, weight=None, attr=None, sp=None)

Calculate barycenter of a connected graph, optionally with edge weights.
The barycenter a connected graph G is the subgraph induced by the set of its nodes v minimizing the objective
function
∑
dG (u, v),
u∈V (G)
where dG is the (possibly weighted) path length. The barycenter is also called the median. See [?], p. 78.
Parameters
G
[networkx.Graph] The connected graph G.
weight
[str, optional] Passed through to shortest_path_length().
attr
[str, optional] If given, write the value of the objective function to each node’s attr at-
tribute. Otherwise do not store the value.
sp
[dict of dicts, optional] All pairs shortest path lengths as a dictionary of dictionaries
Returns
list
Nodes of G that induce the barycenter of G.
Raises
NetworkXNoPath
If G is disconnected. G may appear disconnected to barycenter() if sp is given but is
missing shortest path lengths for any pairs.
ValueError
If sp and weight are both given.
See also:
center
periphery
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (1, 4), (3, 4), (3, 5), (4, 5)])
>>> nx.barycenter(G)
[1, 3, 4]
3.23. Distance Measures 433

3.23.2 center
center(G, e=None, usebounds=False, weight=None)

Returns the center of the graph G.
The center is the set of nodes with eccentricity equal to radius.
Parameters
G
[NetworkX graph] A graph
e
[eccentricity dictionary, optional] A precomputed dictionary of eccentricities.
weight
[string, function, or None] If this is a string, then edge weights will be accessed via the edge
attribute with this key (that is, the weight of the edge joining u to v will be G.edges[u,
v][weight]). If no such edge attribute exists, the weight of the edge is assumed to be one.
If this is a function, the weight of an edge is the value returned by the function. The function
must accept exactly three positional arguments: the two endpoints of an edge and the dictionary
of edge attributes for that edge. The function must return a number.
If this is None, every edge has weight/distance/cost 1.
Weights stored as floating point values can lead to small round-off errors in distances. Use
integer weights to avoid this.
Weights should be positive, since they are distances.
Returns
c
[list] List of nodes in center
See also:
barycenter
periphery
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (1, 4), (3, 4), (3, 5), (4, 5)])
>>> list(nx.center(G))
[1, 3, 4]
3.23.3 diameter
diameter(G, e=None, usebounds=False, weight=None)

Returns the diameter of the graph G.
The diameter is the maximum eccentricity.
Parameters
G

e
weight
Returns
d
[integer] Diameter of graph
See also:
eccentricity
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (1, 4), (3, 4), (3, 5), (4, 5)])
>>> nx.diameter(G)
3
3.23.4 eccentricity
eccentricity(G, v=None, sp=None, weight=None)

Returns the eccentricity of nodes in G.
The eccentricity of a node v is the maximum distance from v to all other nodes in G.
Parameters
G
v
[node, optional] Return value of specified node
sp
[dict of dicts, optional] All pairs shortest path lengths as a dictionary of dictionaries
weight
[string, function, or None (default=None)] If this is a string, then edge weights will be accessed
via the edge attribute with this key (that is, the weight of the edge joining u to v will be
G.edges[u, v][weight]). If no such edge attribute exists, the weight of the edge is
assumed to be one.

Returns
ecc
[dictionary] A dictionary of eccentricity values keyed by node.
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (1, 4), (3, 4), (3, 5), (4, 5)])
>>> dict(nx.eccentricity(G))
{1: 2, 2: 3, 3: 2, 4: 2, 5: 3}
>>> dict(nx.eccentricity(G, v=[1, 5])) # This returns the eccentrity of node 1 &␣
,→5
{1: 2, 5: 3}
3.23.5 periphery
periphery(G, e=None, usebounds=False, weight=None)

Returns the periphery of the graph G.
The periphery is the set of nodes with eccentricity equal to the diameter.
Parameters
G
e
weight
Returns

p
[list] List of nodes in periphery
See also:
barycenter
center
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (1, 4), (3, 4), (3, 5), (4, 5)])
>>> nx.periphery(G)
[2, 5]
3.23.6 radius
radius(G, e=None, usebounds=False, weight=None)

Returns the radius of the graph G.
The radius is the minimum eccentricity.
Parameters
G
e
weight
Returns
r
[integer] Radius of graph

Examples
>>> G = nx.Graph([(1, 2), (1, 3), (1, 4), (3, 4), (3, 5), (4, 5)])
>>> nx.radius(G)
2
3.23.7 resistance_distance
resistance_distance(G, nodeA, nodeB, weight=None, invert_weight=True)

Returns the resistance distance between node A and node B on graph G.
The resistance distance between two nodes of a graph is akin to treating the graph as a grid of resistorses with a
resistance equal to the provided weight.
If weight is not provided, then a weight of 1 is used for all edges.
Parameters
G
nodeA
[node] A node within graph G.
nodeB
[node] A node within graph G, exclusive of Node A.
weight
[string or None, optional (default=None)] The edge data key used to compute the resistance
distance. If None, then each edge has weight 1.
invert_weight
[boolean (default=True)] Proper calculation of resistance distance requires building the Lapla-
cian matrix with the reciprocal of the weight. Not required if the weight is already inverted.
Weight cannot be zero.
Returns
rd
[float] Value of effective resistance distance
Notes
Overviews are provided in [1] and [2]. Additional details on computational methods, proofs of properties, and
corresponding MATLAB codes are provided in [3].

References
[1], [2], [3]
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (1, 4), (3, 4), (3, 5), (4, 5)])
>>> nx.resistance_distance(G, 1, 3)
0.625
3.24 Distance-Regular Graphs
is_distance_regular(G) Returns True if the graph is distance regular, False other-

wise.
is_strongly_regular(G) Returns True if and only if the given graph is strongly
regular.
intersection_array(G) Returns the intersection array of a distance-regular graph.
global_parameters(b, c) Returns global parameters for a given intersection array.
3.24.1 is_distance_regular
is_distance_regular(G)
Returns True if the graph is distance regular, False otherwise.
A connected graph G is distance-regular if for any nodes x,y and any integers i,j=0,1,…,d (where d is the graph
diameter), the number of vertices at distance i from x and distance j from y depends only on i,j and the graph
distance between x and y, independently of the choice of x and y.
Parameters
G: Networkx graph (undirected)
Returns
bool
True if the graph is Distance Regular, False otherwise
See also:
intersection_array, global_parameters
3.24. Distance-Regular Graphs 439

Notes
For undirected and simple graphs only
References
[1], [2]
Examples
>>> G = nx.hypercube_graph(6)
>>> nx.is_distance_regular(G)
True
3.24.2 is_strongly_regular
is_strongly_regular(G)
Returns True if and only if the given graph is strongly regular.
An undirected graph is strongly regular if
• it is regular,
• each pair of adjacent vertices has the same number of neighbors in common,
• each pair of nonadjacent vertices has the same number of neighbors in common.
Each strongly regular graph is a distance-regular graph. Conversely, if a distance-regular graph has di-
ameter two, then it is a strongly regular graph. For more information on distance-regular graphs, see
is_distance_regular().
Parameters
G
Returns
bool
Whether G is strongly regular.
Examples
The cycle graph on five vertices is strongly regular. It is two-regular, each pair of adjacent vertices has no shared
neighbors, and each pair of nonadjacent vertices has one shared neighbor:
>>> nx.is_strongly_regular(G)
True

3.24.3 intersection_array
intersection_array(G)
Returns the intersection array of a distance-regular graph.
Given a distance-regular graph G with integers b_i, c_i,i = 0,….,d such that for any 2 vertices x,y in G at a distance
i=d(x,y), there are exactly c_i neighbors of y at a distance of i-1 from x and b_i neighbors of y at a distance of i+1
from x.
A distance regular graph’s intersection array is given by, [b_0,b_1,…..b_{d-1};c_1,c_2,…..c_d]
Parameters
G: Networkx graph (undirected)
Returns
b,c: tuple of lists
See also:
global_parameters
References
[1]
Examples
>>> nx.intersection_array(G)
([5, 2, 1], [1, 2, 5])
3.24.4 global_parameters
global_parameters(b, c)
Returns global parameters for a given intersection array.
Given a distance-regular graph G with integers b_i, c_i,i = 0,….,d such that for any 2 vertices x,y in G at a distance
i=d(x,y), there are exactly c_i neighbors of y at a distance of i-1 from x and b_i neighbors of y at a distance of i+1
from x.
Thus, a distance regular graph has the global parameters, [[c_0,a_0,b_0],[c_1,a_1,b_1],……,[c_d,a_d,b_d]] for
the intersection array [b_0,b_1,…..b_{d-1};c_1,c_2,…..c_d] where a_i+b_i+c_i=k , k= degree of every vertex.
Parameters
b
[list]
c
[list]
Returns
iterable
An iterable over three tuples.
3.24. Distance-Regular Graphs 441

See also:
intersection_array
References
[1]
Examples
>>> G = nx.dodecahedral_graph()
>>> b, c = nx.intersection_array(G)
>>> list(nx.global_parameters(b, c))
[(0, 0, 3), (1, 0, 2), (1, 1, 1), (1, 1, 1), (2, 0, 1), (3, 0, 0)]
3.25 Dominance
Dominance algorithms.
immediate_dominators(G, start) Returns the immediate dominators of all nodes of a di-

rected graph.
dominance_frontiers(G, start) Returns the dominance frontiers of all nodes of a directed
graph.
3.25.1 immediate_dominators
immediate_dominators(G, start)
Returns the immediate dominators of all nodes of a directed graph.
Parameters
G
[a DiGraph or MultiDiGraph] The graph where dominance is to be computed.
start
[node] The start node of dominance computation.
Returns
idom
[dict keyed by nodes] A dict containing the immediate dominators of each node reachable from
start.
Raises
If G is undirected.
NetworkXError
If start is not in G.

Notes
Except for start, the immediate dominators are the parents of their corresponding nodes in the dominator tree.
References
[1]
Examples
>>> G = nx.DiGraph([(1, 2), (1, 3), (2, 5), (3, 4), (4, 5)])
>>> sorted(nx.immediate_dominators(G, 1).items())
[(1, 1), (2, 1), (3, 1), (4, 3), (5, 1)]
3.25.2 dominance_frontiers
dominance_frontiers(G, start)
Returns the dominance frontiers of all nodes of a directed graph.
Parameters
G
[a DiGraph or MultiDiGraph] The graph where dominance is to be computed.
start
[node] The start node of dominance computation.
Returns
df
[dict keyed by nodes] A dict containing the dominance frontiers of each node reachable from
start as lists.
Raises
If G is undirected.
NetworkXError
If start is not in G.
References
[1]
3.25. Dominance 443

Examples
>>> G = nx.DiGraph([(1, 2), (1, 3), (2, 5), (3, 4), (4, 5)])
>>> sorted((u, sorted(df)) for u, df in nx.dominance_frontiers(G, 1).items())
[(1, []), (2, [5]), (3, [5]), (4, [5]), (5, [])]
3.26 Dominating Sets
Functions for computing dominating sets in a graph.
dominating_set(G[, start_with]) Finds a dominating set for the graph G.

is_dominating_set(G, nbunch) Checks if nbunch is a dominating set for G.
3.26.1 dominating_set
dominating_set(G, start_with=None)
Finds a dominating set for the graph G.
A dominating set for a graph with node set V is a subset D of V such that every node not in D is adjacent to at least
one member of D [1].
Parameters
G
[NetworkX graph]
start_with
[node (default=None)] Node to use as a starting point for the algorithm.
Returns
D
[set] A dominating set for G.
See also:
is_dominating_set
Notes
This function is an implementation of algorithm 7 in [2] which finds some dominating set, not necessarily the
smallest one.

References
[1], [2]
3.26.2 is_dominating_set
is_dominating_set(G, nbunch)
Checks if nbunch is a dominating set for G.
A dominating set for a graph with node set V is a subset D of V such that every node not in D is adjacent to at least
one member of D [1].
Parameters
G
[NetworkX graph]
nbunch
[iterable] An iterable of nodes in the graph G.
See also:
dominating_set
References
[1]
3.27 Efficiency
Provides functions for computing the efficiency of nodes and graphs.
efficiency(G, u, v) Returns the efficiency of a pair of nodes in a graph.

local_efficiency(G) Returns the average local efficiency of the graph.
global_efficiency(G) Returns the average global efficiency of the graph.
3.27.1 efficiency
efficiency(G, u, v)
Returns the efficiency of a pair of nodes in a graph.
The efficiency of a pair of nodes is the multiplicative inverse of the shortest path distance between the nodes [1].
Returns 0 if no path between nodes.
Parameters
G
[networkx.Graph] An undirected graph for which to compute the average local efficiency.
u, v
[node] Nodes in the graph G.
Returns
3.27. Efficiency 445

float
Multiplicative inverse of the shortest path distance between the nodes.
See also:
local_efficiency
global_efficiency
Notes
Edge weights are ignored when computing the shortest path distances.
References
[1]
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3)])
>>> nx.efficiency(G, 2, 3) # this gives efficiency for node 2 and 3
0.5
3.27.2 local_efficiency
local_efficiency(G)
Returns the average local efficiency of the graph.
The efficiency of a pair of nodes in a graph is the multiplicative inverse of the shortest path distance between the
nodes. The local efficiency of a node in the graph is the average global efficiency of the subgraph induced by the
neighbors of the node. The average local efficiency is the average of the local efficiencies of each node [1].
Parameters
G
[networkx.Graph] An undirected graph for which to compute the average local efficiency.
Returns
float
The average local efficiency of the graph.
See also:
global_efficiency

Notes
References
[1]
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3)])
>>> nx.local_efficiency(G)
0.9166666666666667
3.27.3 global_efficiency
global_efficiency(G)
Returns the average global efficiency of the graph.
The efficiency of a pair of nodes in a graph is the multiplicative inverse of the shortest path distance between the
nodes. The average global efficiency of a graph is the average efficiency of all pairs of nodes [1].
Parameters
G
[networkx.Graph] An undirected graph for which to compute the average global effi-
ciency.
Returns
float
The average global efficiency of the graph.
See also:
local_efficiency
Notes
References
[1]
3.27. Efficiency 447

Examples
>>> G = nx.Graph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3)])
>>> round(nx.global_efficiency(G), 12)
0.916666666667
3.28 Eulerian
Eulerian circuits and graphs.
is_eulerian(G) Returns True if and only if G is Eulerian.

eulerian_circuit(G[, source, keys]) Returns an iterator over the edges of an Eulerian circuit
in G.
eulerize(G) Transforms a graph into an Eulerian graph.
is_semieulerian(G) Return True iff G is semi-Eulerian.
has_eulerian_path(G[, source]) Return True iff G has an Eulerian path.
eulerian_path(G[, source, keys]) Return an iterator over the edges of an Eulerian path in G.
3.28.1 is_eulerian
is_eulerian(G)
Returns True if and only if G is Eulerian.
A graph is Eulerian if it has an Eulerian circuit. An Eulerian circuit is a closed walk that includes each edge of a
graph exactly once.
Graphs with isolated vertices (i.e. vertices with zero degree) are not considered to have Eulerian circuits. Therefore,
if the graph is not connected (or not strongly connected, for directed graphs), this function returns False.
Parameters
G
[NetworkX graph] A graph, either directed or undirected.
Examples
>>> nx.is_eulerian(nx.DiGraph({0: [3], 1: [2], 2: [3], 3: [0, 1]}))

True
>>> nx.is_eulerian(nx.complete_graph(5))
True
>>> nx.is_eulerian(nx.petersen_graph())
False
If you prefer to allow graphs with isolated vertices to have Eulerian circuits, you can first remove such vertices and
then call is_eulerian as below example shows.
>>> G = nx.Graph([(0, 1), (1, 2), (0, 2)])

>>> G.add_node(3)
>>> nx.is_eulerian(G)
False

>>> G.remove_nodes_from(list(nx.isolates(G)))
>>> nx.is_eulerian(G)
True
3.28.2 eulerian_circuit
eulerian_circuit(G, source=None, keys=False)

Returns an iterator over the edges of an Eulerian circuit in G.
An Eulerian circuit is a closed walk that includes each edge of a graph exactly once.
Parameters
G
[NetworkX graph] A graph, either directed or undirected.
source
[node, optional] Starting node for circuit.
keys
[bool] If False, edges generated by this function will be of the form (u, v). Otherwise,
edges will be of the form (u, v, k). This option is ignored unless G is a multigraph.
Returns
edges
[iterator] An iterator over edges in the Eulerian circuit.
Raises
NetworkXError
If the graph is not Eulerian.
See also:
is_eulerian
Notes
This is a linear time implementation of an algorithm adapted from [1].

For general information about Euler tours, see [2].
References
[1], [2]
3.28. Eulerian 449

Examples
To get an Eulerian circuit in an undirected graph:
>>> list(nx.eulerian_circuit(G))
[(0, 2), (2, 1), (1, 0)]
>>> list(nx.eulerian_circuit(G, source=1))
[(1, 2), (2, 0), (0, 1)]
To get the sequence of vertices in an Eulerian circuit:
>>> [u for u, v in nx.eulerian_circuit(G)]

[0, 2, 1]
3.28.3 eulerize
eulerize(G)
Transforms a graph into an Eulerian graph.
If G is Eulerian the result is G as a MultiGraph, otherwise the result is a smallest (in terms of the number of edges)
multigraph whose underlying simple graph is G.
Parameters
G
Returns
G
[NetworkX multigraph]
Raises
NetworkXError
If the graph is not connected.
See also:
is_eulerian
eulerian_circuit
References
[1], [2], [3]

Examples
>>> H = nx.eulerize(G)
>>> nx.is_eulerian(H)
True
3.28.4 is_semieulerian
is_semieulerian(G)
Return True iff G is semi-Eulerian.
G is semi-Eulerian if it has an Eulerian path but no Eulerian circuit.
See also:
has_eulerian_path
is_eulerian
3.28.5 has_eulerian_path
has_eulerian_path(G, source=None)
Return True iff G has an Eulerian path.
An Eulerian path is a path in a graph which uses each edge of a graph exactly once. If source is specified, then
this function checks whether an Eulerian path that starts at node source exists.
A directed graph has an Eulerian path iff:
• at most one vertex has out_degree - in_degree = 1,
• at most one vertex has in_degree - out_degree = 1,
• every other vertex has equal in_degree and out_degree,
• and all of its vertices belong to a single connected component of the underlying undirected graph.
If source is not None, an Eulerian path starting at source exists if no other node has out_degree - in_degree
= 1. This is equivalent to either there exists an Eulerian circuit or source has out_degree - in_degree = 1 and the
conditions above hold.
An undirected graph has an Eulerian path iff:
• exactly zero or two vertices have odd degree,
• and all of its vertices belong to a single connected component.
If source is not None, an Eulerian path starting at source exists if either there exists an Eulerian circuit or
source has an odd degree and the conditions above hold.
Graphs with isolated vertices (i.e. vertices with zero degree) are not considered to have an Eulerian path. Therefore,
if the graph is not connected (or not strongly connected, for directed graphs), this function returns False.
Parameters
G
[NetworkX Graph] The graph to find an euler path in.
3.28. Eulerian 451

source
[node, optional] Starting node for path.
Returns
Bool
[True if G has an Eulerian path.]
See also:
is_eulerian
eulerian_path
Examples
If you prefer to allow graphs with isolated vertices to have Eulerian path, you can first remove such vertices and
then call has_eulerian_path as below example shows.
>>> G = nx.Graph([(0, 1), (1, 2), (0, 2)])

>>> G.add_node(3)
>>> nx.has_eulerian_path(G)
False
>>> nx.has_eulerian_path(G)
True
3.28.6 eulerian_path
eulerian_path(G, source=None, keys=False)

Return an iterator over the edges of an Eulerian path in G.
Parameters
G
[NetworkX Graph] The graph in which to look for an eulerian path.
source
[node or None (default: None)] The node at which to start the search. None means search over
all starting nodes.
keys
[Bool (default: False)] Indicates whether to yield edge 3-tuples (u, v, edge_key). The default
yields edge 2-tuples
Yields
Edge tuples along the eulerian path.
Warning: If source provided is not the start node of an Euler path
will raise error even if an Euler Path exists.

3.29 Flows
3.29.1 Maximum Flow
maximum_flow(flowG, _s, _t[, capacity, ...]) Find a maximum single-commodity flow.

maximum_flow_value(flowG, _s, _t[, ...]) Find the value of maximum single-commodity flow.
minimum_cut(flowG, _s, _t[, capacity, flow_func]) Compute the value and the node partition of a minimum
(s, t)-cut.
minimum_cut_value(flowG, _s, _t[, capacity, ...]) Compute the value of a minimum (s, t)-cut.
maximum_flow
maximum_flow(flowG, _s, _t, capacity='capacity', flow_func=None, **kwargs)

Find a maximum single-commodity flow.
Parameters
flowG
[NetworkX graph] Edges of the graph are expected to have an attribute called ‘capacity’. If
this attribute is not present, the edge is considered to have infinite capacity.
_s
_t
capacity
[string] Edges of the graph G are expected to have an attribute capacity that indicates how
much flow the edge can support. If this attribute is not present, the edge is considered to have
infinite capacity. Default value: ‘capacity’.
flow_func
[function] A function for computing the maximum flow among a pair of nodes in a capacitated
graph. The function has to accept at least three parameters: a Graph or Digraph, a source
node, and a target node. And return a residual network that follows NetworkX conventions
(see Notes). If flow_func is None, the default maximum flow function (preflow_push())
is used. See below for alternative algorithms. The choice of the default function may change
from version to version and should not be relied on. Default value: None.
kwargs
[Any other keyword parameter is passed to the function that] computes the maximum flow.
Returns
flow_value
[integer, float] Value of the maximum flow, i.e., net outflow from the source.
flow_dict
[dict] A dictionary containing the value of the flow that went through each edge.
Raises
NetworkXError
The algorithm does not support MultiGraph and MultiDiGraph. If the input graph is an in-
stance of one of these two classes, a NetworkXError is raised.
3.29. Flows 453

NetworkXUnbounded
If the graph has a path of infinite capacity, the value of a feasible flow on the graph is unbounded
above and the function raises a NetworkXUnbounded.
See also:
maximum_flow_value()
minimum_cut()
minimum_cut_value()
edmonds_karp()
preflow_push()
Notes
The function used in the flow_func parameter has to return a residual network that follows NetworkX conventions:
The residual network R from an input graph G has the same nodes as G. R is a DiGraph that contains a pair of edges
(u, v) and (v, u) iff (u, v) is not a self-loop, and at least one of (u, v) and (v, u) exists in G.
For each edge (u, v) in R, R[u][v]['capacity'] is equal to the capacity of (u, v) in G if it exists in
G or zero otherwise. If the capacity is infinite, R[u][v]['capacity'] will have a high arbitrary finite value
that does not affect the solution of the problem. This value is stored in R.graph['inf']. For each edge (u,
v) in R, R[u][v]['flow'] represents the flow function of (u, v) and satisfies R[u][v]['flow'] ==
-R[v][u]['flow'].
The flow value, defined as the total flow into t, the sink, is stored in R.graph['flow_value']. Reachabil-
ity to t using only edges (u, v) such that R[u][v]['flow'] < R[u][v]['capacity'] induces a
minimum s-t cut.
Specific algorithms may store extra data in R.
The function should supports an optional boolean parameter value_only. When True, it can optionally terminate
the algorithm as soon as the maximum flow value and the minimum cut can be determined.
Examples
>>> G.add_edge("x", "a", capacity=3.0)
>>> G.add_edge("x", "b", capacity=1.0)
>>> G.add_edge("a", "c", capacity=3.0)
>>> G.add_edge("b", "c", capacity=5.0)
>>> G.add_edge("b", "d", capacity=4.0)
>>> G.add_edge("d", "e", capacity=2.0)
>>> G.add_edge("c", "y", capacity=2.0)
>>> G.add_edge("e", "y", capacity=3.0)
maximum_flow returns both the value of the maximum flow and a dictionary with all flows.
>>> flow_value, flow_dict = nx.maximum_flow(G, "x", "y")

>>> flow_value
3.0
>>> print(flow_dict["x"]["b"])
1.0
You can also use alternative algorithms for computing the maximum flow by using the flow_func parameter.


>>> flow_value == nx.maximum_flow(G, "x", "y", flow_func=shortest_augmenting_
,→path)[
... 0
... ]
True
maximum_flow_value
maximum_flow_value(flowG, _s, _t, capacity='capacity', flow_func=None, **kwargs)

Find the value of maximum single-commodity flow.
Parameters
flowG
_s
_t
capacity
flow_func
kwargs
Returns
flow_value
[integer, float] Value of the maximum flow, i.e., net outflow from the source.
Raises
NetworkXError
NetworkXUnbounded
See also:
3.29. Flows 455

maximum_flow()
minimum_cut()
minimum_cut_value()
edmonds_karp()
preflow_push()
Notes
-R[v][u]['flow'].
minimum s-t cut.
Examples
maximum_flow_value computes only the value of the maximum flow:
>>> flow_value = nx.maximum_flow_value(G, "x", "y")

>>> flow_value
3.0
You can also use alternative algorithms for computing the maximum flow by using the flow_func parameter.

>>> flow_value == nx.maximum_flow_value(
... G, "x", "y", flow_func=shortest_augmenting_path
... )
True

minimum_cut
minimum_cut(flowG, _s, _t, capacity='capacity', flow_func=None, **kwargs)

Compute the value and the node partition of a minimum (s, t)-cut.
Use the max-flow min-cut theorem, i.e., the capacity of a minimum capacity cut is equal to the flow value of a
maximum flow.
Parameters
flowG
_s
_t
capacity
flow_func
kwargs
Returns
cut_value
[integer, float] Value of the minimum cut.
partition
[pair of node sets] A partitioning of the nodes that defines a minimum cut.
Raises
NetworkXUnbounded
If the graph has a path of infinite capacity, all cuts have infinite capacity and the function raises
a NetworkXError.
See also:
maximum_flow()
minimum_cut_value()
edmonds_karp()
preflow_push()
3.29. Flows 457

Notes
-R[v][u]['flow'].
minimum s-t cut.
Examples
minimum_cut computes both the value of the minimum cut and the node partition:
>>> cut_value, partition = nx.minimum_cut(G, "x", "y")

>>> reachable, non_reachable = partition
‘partition’ here is a tuple with the two sets of nodes that define the minimum cut. You can compute the cut set of
edges that induce the minimum cut as follows:
>>> cutset = set()

>>> for u, nbrs in ((n, G[n]) for n in reachable):
... cutset.update((u, v) for v in nbrs if v in non_reachable)
>>> print(sorted(cutset))
[('c', 'y'), ('x', 'b')]
>>> cut_value == sum(G.edges[u, v]["capacity"] for (u, v) in cutset)
True
You can also use alternative algorithms for computing the minimum cut by using the flow_func parameter.

>>> cut_value == nx.minimum_cut(G, "x", "y", flow_func=shortest_augmenting_
,→path)[0]
True

minimum_cut_value
minimum_cut_value(flowG, _s, _t, capacity='capacity', flow_func=None, **kwargs)

Compute the value of a minimum (s, t)-cut.
Use the max-flow min-cut theorem, i.e., the capacity of a minimum capacity cut is equal to the flow value of a
maximum flow.
Parameters
flowG
_s
_t
capacity
flow_func
kwargs
Returns
cut_value
[integer, float] Value of the minimum cut.
Raises
NetworkXUnbounded
If the graph has a path of infinite capacity, all cuts have infinite capacity and the function raises
a NetworkXError.
See also:
maximum_flow()
minimum_cut()
edmonds_karp()
preflow_push()
3.29. Flows 459

Notes
-R[v][u]['flow'].
minimum s-t cut.
Examples
minimum_cut_value computes only the value of the minimum cut:
>>> cut_value = nx.minimum_cut_value(G, "x", "y")

>>> cut_value
3.0
You can also use alternative algorithms for computing the minimum cut by using the flow_func parameter.

>>> cut_value == nx.minimum_cut_value(
... G, "x", "y", flow_func=shortest_augmenting_path
... )
True

3.29.2 Edmonds-Karp
edmonds_karp(G, s, t[, capacity, residual, ...]) Find a maximum single-commodity flow using the
Edmonds-Karp algorithm.
edmonds_karp
edmonds_karp(G, s, t, capacity='capacity', residual=None, value_only=False, cutoff=None)

Find a maximum single-commodity flow using the Edmonds-Karp algorithm.
This function returns the residual network resulting after computing the maximum flow. See below for details about
the conventions NetworkX uses for defining residual networks.
This algorithm has a running time of O(nm2 ) for n nodes and m edges.
Parameters
G
s
t
capacity
residual
[NetworkX graph] Residual network on which the algorithm is to be executed. If None, a new
residual network is created. Default value: None.
value_only
[bool] If True compute only the value of the maximum flow. This parameter will be ignored
by this algorithm because it is not applicable.
cutoff
[integer, float] If specified, the algorithm will terminate when the flow value reaches or exceeds
the cutoff. In this case, it may be unable to immediately determine a minimum cut. Default
value: None.
Returns
R
[NetworkX DiGraph] Residual network after computing the maximum flow.
Raises
NetworkXError
NetworkXUnbounded
3.29. Flows 461

See also:
maximum_flow()
minimum_cut()
preflow_push()
Notes
-R[v][u]['flow'].
The flow value, defined as the total flow into t, the sink, is stored in R.graph['flow_value']. If
cutoff is not specified, reachability to t using only edges (u, v) such that R[u][v]['flow'] <
R[u][v]['capacity'] induces a minimum s-t cut.
Examples
>>> from networkx.algorithms.flow import edmonds_karp
The functions that implement flow algorithms and output a residual network, such as this one, are not imported to
the base NetworkX namespace, so you have to explicitly import them from the flow package.
>>> R = edmonds_karp(G, "x", "y")
>>> flow_value
3.0
>>> flow_value == R.graph["flow_value"]
True

3.29.3 Shortest Augmenting Path
shortest_augmenting_path(G, s, t[, ...]) Find a maximum single-commodity flow using the short-
est augmenting path algorithm.
shortest_augmenting_path
shortest_augmenting_path(G, s, t, capacity='capacity', residual=None, value_only=False, two_phase=False,

cutoff=None)
Find a maximum single-commodity flow using the shortest augmenting path algorithm.
This algorithm has a running time of O(n2 m) for n nodes and m edges.
Parameters
G
s
t
capacity
residual
value_only
two_phase
[bool] If True, a two-phase variant is used. The two-phase variant improves the running time
on unit-capacity networks from O(nm) to O(min(n2/3 , m1/2 )m). Default value: False.
cutoff
value: None.
Returns
R
Raises
3.29. Flows 463

NetworkXError
NetworkXUnbounded
See also:
maximum_flow()
minimum_cut()
edmonds_karp()
preflow_push()
Notes
-R[v][u]['flow'].
Examples
>>> R = shortest_augmenting_path(G, "x", "y")
>>> flow_value
3.0
True

3.29.4 Preflow-Push
preflow_push(G, s, t[, capacity, residual, ...]) Find a maximum single-commodity flow using the
highest-label preflow-push algorithm.
preflow_push
preflow_push(G, s, t, capacity='capacity', residual=None, global_relabel_freq=1, value_only=False)

Find a maximum single-commodity flow using the highest-label preflow-push algorithm.
√
This algorithm has a running time of O(n2 m) for n nodes and m edges.
Parameters
G
s
t
capacity
residual
global_relabel_freq
[integer, float] Relative frequency of applying the global relabeling heuristic to speed up the
algorithm. If it is None, the heuristic is disabled. Default value: 1.
value_only
[bool] If False, compute a maximum flow; otherwise, compute a maximum preflow which is
enough for computing the maximum flow value. Default value: False.
Returns
R
Raises
NetworkXError
NetworkXUnbounded
See also:
3.29. Flows 465

maximum_flow()
minimum_cut()
edmonds_karp()
Notes
(u, v) and (v, u) iff (u, v) is not a self-loop, and at least one of (u, v) and (v, u) exists in G. For
each node u in R, R.nodes[u]['excess'] represents the difference between flow into u and flow out of u.
-R[v][u]['flow'].
minimum s-t cut.
Examples
>>> from networkx.algorithms.flow import preflow_push
>>> R = preflow_push(G, "x", "y")
True
>>> # preflow_push also stores the maximum flow value
>>> # in the excess attribute of the sink node t
>>> flow_value == R.nodes["y"]["excess"]
True
>>> # For some problems, you might only want to compute a
>>> # maximum preflow.
>>> R = preflow_push(G, "x", "y", value_only=True)
True
>>> flow_value == R.nodes["y"]["excess"]
True

3.29.5 Dinitz
dinitz(G, s, t[, capacity, residual, ...]) Find a maximum single-commodity flow using Dinitz' al-
gorithm.
dinitz
dinitz(G, s, t, capacity='capacity', residual=None, value_only=False, cutoff=None)

Find a maximum single-commodity flow using Dinitz’ algorithm.
This algorithm has a running time of O(n2 m) for n nodes and m edges [1].
Parameters
G
s
t
capacity
residual
value_only
cutoff
value: None.
Returns
R
Raises
NetworkXError
NetworkXUnbounded
3.29. Flows 467

See also:
maximum_flow()
minimum_cut()
preflow_push()
Notes
-R[v][u]['flow'].
References
[1]
Examples
>>> from networkx.algorithms.flow import dinitz
>>> R = dinitz(G, "x", "y")
>>> flow_value
3.0
True

3.29.6 Boykov-Kolmogorov
boykov_kolmogorov(G, s, t[, capacity, ...]) Find a maximum single-commodity flow using Boykov-
Kolmogorov algorithm.
boykov_kolmogorov
boykov_kolmogorov(G, s, t, capacity='capacity', residual=None, value_only=False, cutoff=None)

Find a maximum single-commodity flow using Boykov-Kolmogorov algorithm.
This algorithm has worse case complexity O(n2 m|C|) for n nodes, m edges, and |C| the cost of the minimum
cut [1]. This implementation uses the marking heuristic defined in [2] which improves its running time in many
practical problems.
Parameters
G
s
t
capacity
residual
value_only
cutoff
value: None.
Returns
R
Raises
NetworkXError
3.29. Flows 469

NetworkXUnbounded
See also:
maximum_flow()
minimum_cut()
preflow_push()
Notes
-R[v][u]['flow'].
References
[1], [2]
Examples
>>> from networkx.algorithms.flow import boykov_kolmogorov
>>> R = boykov_kolmogorov(G, "x", "y")
>>> flow_value
3.0
True

A nice feature of the Boykov-Kolmogorov algorithm is that a partition of the nodes that defines a minimum cut
can be easily computed based on the search trees used during the algorithm. These trees are stored in the graph
attribute trees of the residual network.
>>> source_tree, target_tree = R.graph["trees"]

>>> partition = (set(source_tree), set(G) - set(source_tree))
Or equivalently:
>>> partition = (set(G) - set(target_tree), set(target_tree))
3.29.7 Gomory-Hu Tree
gomory_hu_tree(G[, capacity, flow_func]) Returns the Gomory-Hu tree of an undirected graph G.
gomory_hu_tree
gomory_hu_tree(G, capacity='capacity', flow_func=None)

Returns the Gomory-Hu tree of an undirected graph G.
A Gomory-Hu tree of an undirected graph with capacities is a weighted tree that represents the minimum s-t cuts
for all s-t pairs in the graph.
It only requires n-1 minimum cut computations instead of the obvious n(n-1)/2. The tree represents all s-t
cuts as the minimum cut value among any pair of nodes is the minimum edge weight in the shortest path between
the two nodes in the Gomory-Hu tree.
The Gomory-Hu tree also has the property that removing the edge with the minimum weight in the shortest path
between any two nodes leaves two connected components that form a partition of the nodes in G that defines the
minimum s-t cut.
See Examples section below for details.
Parameters
G
capacity
flow_func
[function] Function to perform the underlying flow computations. Default value ed-
Returns
Tree
[NetworkX graph] A NetworkX graph representing the Gomory-Hu tree of the input graph.
Raises
Raised if the input graph is directed.
3.29. Flows 471

NetworkXError
Raised if the input graph is an empty Graph.
See also:
minimum_cut()
maximum_flow()
Notes
This implementation is based on Gusfield approach [1] to compute Comory-Hu trees, which does not require node
contractions and has the same computational complexity than the original method.
References
[1]
Examples
>>> nx.set_edge_attributes(G, 1, "capacity")
>>> T = nx.gomory_hu_tree(G)
>>> # The value of the minimum cut between any pair
... # of nodes in G is the minimum edge weight in the
... # shortest path between the two nodes in the
... # Gomory-Hu tree.
... def minimum_edge_weight_in_shortest_path(T, u, v):
... path = nx.shortest_path(T, u, v, weight="weight")
... return min((T[u][v]["weight"], (u, v)) for (u, v) in zip(path, path[1:]))
>>> u, v = 0, 33
>>> cut_value, edge = minimum_edge_weight_in_shortest_path(T, u, v)
>>> cut_value
10
>>> nx.minimum_cut_value(G, u, v)
10
>>> # The Comory-Hu tree also has the property that removing the
... # edge with the minimum weight in the shortest path between
... # any two nodes leaves two connected components that form
... # a partition of the nodes in G that defines the minimum s-t
... # cut.
... cut_value, edge = minimum_edge_weight_in_shortest_path(T, u, v)
>>> T.remove_edge(*edge)
>>> U, V = list(nx.connected_components(T))
>>> # Thus U and V form a partition that defines a minimum cut
... # between u and v in G. You can compute the edge cut set,
... # that is, the set of edges that if removed from G will
... # disconnect u from v in G, with this information:
... cutset = set()
>>> for x, nbrs in ((n, G[n]) for n in U):
... cutset.update((x, y) for y in nbrs if y in V)
>>> # Because we have set the capacities of all edges to 1
... # the cutset contains ten edges
... len(cutset)
10


>>> # You can use any maximum flow algorithm for the underlying
... # flow computations using the argument flow_func
... from networkx.algorithms import flow
>>> T = nx.gomory_hu_tree(G, flow_func=flow.boykov_kolmogorov)
>>> cut_value, edge = minimum_edge_weight_in_shortest_path(T, u, v)
>>> cut_value
10
>>> nx.minimum_cut_value(G, u, v, flow_func=flow.boykov_kolmogorov)
10
3.29.8 Utils
build_residual_network(G, capacity) Build a residual network and initialize a zero flow.
build_residual_network
build_residual_network(G, capacity)
Build a residual network and initialize a zero flow.
-R[v][u]['flow'].
3.29.9 Network Simplex
network_simplex(G[, demand, capacity, weight]) Find a minimum cost flow satisfying all demands in di-
graph G.
min_cost_flow_cost(G[, demand, capacity, Find the cost of a minimum cost flow satisfying all de-
weight]) mands in digraph G.
min_cost_flow(G[, demand, capacity, weight]) Returns a minimum cost flow satisfying all demands in
digraph G.
cost_of_flow(G, flowDict[, weight]) Compute the cost of the flow given by flowDict on graph
G.
max_flow_min_cost(G, s, t[, capacity, weight]) Returns a maximum (s, t)-flow of minimum cost.
3.29. Flows 473

network_simplex
network_simplex(G, demand='demand', capacity='capacity', weight='weight')

Find a minimum cost flow satisfying all demands in digraph G.
This is a primal network simplex algorithm that uses the leaving arc rule to prevent cycling.
G is a digraph with edge costs and capacities and in which nodes have demand, i.e., they want to send or receive
some amount of flow. A negative demand means that the node wants to send flow, a positive demand means that
the node want to receive flow. A flow on the digraph G satisfies all demand if the net flow into each node is equal
to the demand of that node.
Parameters
G
[NetworkX graph] DiGraph on which a minimum cost flow satisfying all demands is to be
found.
demand
[string] Nodes of the graph G are expected to have an attribute demand that indicates how
much flow a node wants to send (negative demand) or receive (positive demand). Note that
the sum of the demands should be 0 otherwise the problem in not feasible. If this attribute is
not present, a node is considered to have 0 demand. Default value: ‘demand’.
capacity
weight
[string] Edges of the graph G are expected to have an attribute weight that indicates the cost
incurred by sending one unit of flow on that edge. If not present, the weight is considered to
be 0. Default value: ‘weight’.
Returns
flowCost
[integer, float] Cost of a minimum cost flow satisfying all demands.
flowDict
[dictionary] Dictionary of dictionaries keyed by nodes such that flowDict[u][v] is the flow edge
(u, v).
Raises
NetworkXError
This exception is raised if the input graph is not directed or not connected.
NetworkXUnfeasible
This exception is raised in the following situations:
• The sum of the demands is not zero. Then, there is no flow satisfying all demands.
• There is no flow satisfying all demand.
NetworkXUnbounded
This exception is raised if the digraph G has a cycle of negative cost and infinite capacity.
Then, the cost of a flow satisfying all demands is unbounded below.
See also:
cost_of_flow, max_flow_min_cost, min_cost_flow, min_cost_flow_cost

Notes
This algorithm is not guaranteed to work if edge weights or demands are floating point numbers (overflows and
roundoff errors can cause problems). As a workaround you can use integer numbers by multiplying the relevant
edge attributes by a convenient constant factor (eg 100).
References
[1], [2]
Examples
A simple example of a min cost flow problem.
>>> G.add_node("a", demand=-5)
>>> G.add_node("d", demand=5)
>>> G.add_edge("a", "b", weight=3, capacity=4)
>>> G.add_edge("a", "c", weight=6, capacity=10)
>>> G.add_edge("b", "d", weight=1, capacity=9)
>>> G.add_edge("c", "d", weight=2, capacity=5)
>>> flowCost, flowDict = nx.network_simplex(G)
>>> flowCost
24
>>> flowDict
{'a': {'b': 4, 'c': 1}, 'd': {}, 'b': {'d': 4}, 'c': {'d': 1}}
The mincost flow algorithm can also be used to solve shortest path problems. To find the shortest path between two
nodes u and v, give all edges an infinite capacity, give node u a demand of -1 and node v a demand a 1. Then run
the network simplex. The value of a min cost flow will be the distance between u and v and edges carrying positive
flow will indicate the path.
>>> G.add_weighted_edges_from(
... [
... ("s", "u", 10),
... ("s", "x", 5),
... ("u", "v", 1),
... ("u", "x", 2),
... ("v", "y", 1),
... ("x", "u", 3),
... ("x", "v", 5),
... ("x", "y", 2),
... ("y", "s", 7),
... ("y", "v", 6),
... ]
... )
>>> G.add_node("s", demand=-1)
>>> G.add_node("v", demand=1)
>>> flowCost, flowDict = nx.network_simplex(G)
>>> flowCost == nx.shortest_path_length(G, "s", "v", weight="weight")
True
>>> sorted([(u, v) for u in flowDict for v in flowDict[u] if flowDict[u][v] > 0])
[('s', 'x'), ('u', 'v'), ('x', 'u')]
3.29. Flows 475


>>> nx.shortest_path(G, "s", "v", weight="weight")
['s', 'x', 'u', 'v']
It is possible to change the name of the attributes used for the algorithm.
>>> G.add_node("p", spam=-4)
>>> G.add_node("q", spam=2)
>>> G.add_node("a", spam=-2)
>>> G.add_node("d", spam=-1)
>>> G.add_node("t", spam=2)
>>> G.add_node("w", spam=3)
>>> G.add_edge("p", "q", cost=7, vacancies=5)
>>> G.add_edge("p", "a", cost=1, vacancies=4)
>>> G.add_edge("q", "d", cost=2, vacancies=3)
>>> G.add_edge("t", "q", cost=1, vacancies=2)
>>> G.add_edge("a", "t", cost=2, vacancies=4)
>>> G.add_edge("d", "w", cost=3, vacancies=4)
>>> G.add_edge("t", "w", cost=4, vacancies=1)
>>> flowCost, flowDict = nx.network_simplex(
... G, demand="spam", capacity="vacancies", weight="cost"
... )
>>> flowCost
37
>>> flowDict
{'p': {'q': 2, 'a': 2}, 'q': {'d': 1}, 'a': {'t': 4}, 'd': {'w': 2}, 't': {'q': 1,
,→ 'w': 1}, 'w': {}}
min_cost_flow_cost
min_cost_flow_cost(G, demand='demand', capacity='capacity', weight='weight')

Find the cost of a minimum cost flow satisfying all demands in digraph G.
Parameters
G
found.
demand
capacity
weight

Returns
flowCost
[integer, float] Cost of a minimum cost flow satisfying all demands.
Raises
NetworkXError
NetworkXUnfeasible
NetworkXUnbounded
See also:
cost_of_flow, max_flow_min_cost, min_cost_flow, network_simplex
Notes
Examples
>>> flowCost = nx.min_cost_flow_cost(G)
>>> flowCost
24
3.29. Flows 477

min_cost_flow
min_cost_flow(G, demand='demand', capacity='capacity', weight='weight')

Returns a minimum cost flow satisfying all demands in digraph G.
Parameters
G
found.
demand
capacity
weight
Returns
flowDict
(u, v).
Raises
NetworkXError
NetworkXUnfeasible
NetworkXUnbounded
See also:
cost_of_flow, max_flow_min_cost, min_cost_flow_cost, network_simplex

Notes
Examples
>>> flowDict = nx.min_cost_flow(G)
cost_of_flow
cost_of_flow(G, flowDict, weight='weight')

Compute the cost of the flow given by flowDict on graph G.
Note that this function does not check for the validity of the flow flowDict. This function will fail if the graph G
and the flow don’t have the same edge set.
Parameters
G
found.
weight
flowDict
(u, v).
Returns
cost
[Integer, float] The total cost of the flow. This is given by the sum over all edges of the product
of the edge’s flow and the edge’s weight.
See also:
max_flow_min_cost, min_cost_flow, min_cost_flow_cost, network_simplex
3.29. Flows 479

Notes
max_flow_min_cost
max_flow_min_cost(G, s, t, capacity='capacity', weight='weight')

Returns a maximum (s, t)-flow of minimum cost.
G is a digraph with edge costs and capacities. There is a source node s and a sink node t. This function finds a
maximum flow from s to t whose total cost is minimized.
Parameters
G
found.
s: node label
Source of the flow.
t: node label
Destination of the flow.
capacity: string
Edges of the graph G are expected to have an attribute capacity that indicates how much flow
the edge can support. If this attribute is not present, the edge is considered to have infinite
capacity. Default value: ‘capacity’.
weight: string
Edges of the graph G are expected to have an attribute weight that indicates the cost incurred
by sending one unit of flow on that edge. If not present, the weight is considered to be 0.
Default value: ‘weight’.
Returns
flowDict: dictionary
Dictionary of dictionaries keyed by nodes such that flowDict[u][v] is the flow edge (u, v).
Raises
NetworkXError
NetworkXUnbounded
This exception is raised if there is an infinite capacity path from s to t in G. In this case there
is no maximum flow. This exception is also raised if the digraph G has a cycle of negative cost
and infinite capacity. Then, the cost of a flow is unbounded below.
See also:
cost_of_flow, min_cost_flow, min_cost_flow_cost, network_simplex

Notes
Examples
>>> G.add_edges_from(
... [
... (1, 2, {"capacity": 12, "weight": 4}),
... (1, 3, {"capacity": 20, "weight": 6}),
... (2, 3, {"capacity": 6, "weight": -3}),
... (2, 6, {"capacity": 14, "weight": 1}),
... (3, 4, {"weight": 9}),
... (3, 5, {"capacity": 10, "weight": 5}),
... (4, 2, {"capacity": 19, "weight": 13}),
... (4, 5, {"capacity": 4, "weight": 0}),
... (5, 7, {"capacity": 28, "weight": 2}),
... (6, 5, {"capacity": 11, "weight": 1}),
... (6, 7, {"weight": 8}),
... (7, 4, {"capacity": 6, "weight": 6}),
... ]
... )
>>> mincostFlow = nx.max_flow_min_cost(G, 1, 7)
>>> mincost = nx.cost_of_flow(G, mincostFlow)
>>> mincost
373
>>> from networkx.algorithms.flow import maximum_flow
>>> maxFlow = maximum_flow(G, 1, 7)[1]
>>> nx.cost_of_flow(G, maxFlow) >= mincost
True
>>> mincostFlowValue = sum((mincostFlow[u][7] for u in G.predecessors(7))) - sum(
... (mincostFlow[7][v] for v in G.successors(7))
... )
>>> mincostFlowValue == nx.maximum_flow_value(G, 1, 7)
True
3.29. Flows 481

3.29.10 Capacity Scaling Minimum Cost Flow
capacity_scaling(G[, demand, capacity, ...]) Find a minimum cost flow satisfying all demands in di-
graph G.
capacity_scaling
capacity_scaling(G, demand='demand', capacity='capacity', weight='weight', heap=<class

'networkx.utils.heaps.BinaryHeap'>)
Find a minimum cost flow satisfying all demands in digraph G.
This is a capacity scaling successive shortest augmenting path algorithm.
Parameters
G
[NetworkX graph] DiGraph or MultiDiGraph on which a minimum cost flow satisfying all
demands is to be found.
demand
capacity
weight
heap
[class] Type of heap to be used in the algorithm. It should be a subclass of MinHeap or
implement a compatible interface.
If a stock heap implementation is to be used, BinaryHeap is recommended over Pair-
ingHeap for Python implementations without optimized attribute accesses (e.g., CPython)
despite a slower asymptotic running time. For Python implementations with optimized at-
tribute accesses (e.g., PyPy), PairingHeap provides better performance. Default value:
BinaryHeap.
Returns
flowCost
[integer] Cost of a minimum cost flow satisfying all demands.
flowDict
[dictionary] If G is a digraph, a dict-of-dicts keyed by nodes such that flowDict[u][v] is the

flow on edge (u, v). If G is a MultiDiGraph, a dict-of-dicts-of-dicts keyed by nodes so that

flowDict[u][v][key] is the flow on edge (u, v, key).
Raises
NetworkXError
This exception is raised if the input graph is not directed, not connected.
NetworkXUnfeasible
NetworkXUnbounded
See also:
network_simplex()
Notes
This algorithm does not work if edge weights are floating-point numbers.
Examples
>>> flowCost, flowDict = nx.capacity_scaling(G)
>>> flowCost
24
>>> flowDict
{'a': {'b': 4, 'c': 1}, 'd': {}, 'b': {'d': 4}, 'c': {'d': 1}}
It is possible to change the name of the attributes used for the algorithm.
>>> G.add_node("p", spam=-4)
>>> G.add_node("q", spam=2)
>>> G.add_node("a", spam=-2)
>>> G.add_node("d", spam=-1)
>>> G.add_node("t", spam=2)
>>> G.add_node("w", spam=3)
>>> G.add_edge("p", "q", cost=7, vacancies=5)
>>> G.add_edge("p", "a", cost=1, vacancies=4)
>>> G.add_edge("q", "d", cost=2, vacancies=3)
>>> G.add_edge("t", "q", cost=1, vacancies=2)
3.29. Flows 483


>>> G.add_edge("a", "t", cost=2, vacancies=4)
>>> G.add_edge("d", "w", cost=3, vacancies=4)
>>> G.add_edge("t", "w", cost=4, vacancies=1)
>>> flowCost, flowDict = nx.capacity_scaling(
... G, demand="spam", capacity="vacancies", weight="cost"
... )
>>> flowCost
37
>>> flowDict
{'p': {'q': 2, 'a': 2}, 'q': {'d': 1}, 'a': {'t': 4}, 'd': {'w': 2}, 't': {'q': 1,
,→ 'w': 1}, 'w': {}}
3.30 Graph Hashing
Functions for hashing graphs to strings. Isomorphic graphs should be assigned identical hashes. For now, only Weisfeiler-
Lehman hashing is implemented.
weisfeiler_lehman_graph_hash(G[, Return Weisfeiler Lehman (WL) graph hash.

edge_attr, ...])
weisfeiler_lehman_subgraph_hashes(G[, Return a dictionary of subgraph hashes by node.
...])
3.30.1 weisfeiler_lehman_graph_hash
weisfeiler_lehman_graph_hash(G, edge_attr=None, node_attr=None, iterations=3, digest_size=16)

Return Weisfeiler Lehman (WL) graph hash.
The function iteratively aggregates and hashes neighbourhoods of each node. After each node’s neighbors are
hashed to obtain updated node labels, a hashed histogram of resulting labels is returned as the final hash.
Hashes are identical for isomorphic graphs and strong guarantees that non-isomorphic graphs will get different
hashes. See [1] for details.
If no node or edge attributes are provided, the degree of each node is used as its initial label. Otherwise, node
and/or edge labels are used to compute the hash.
Parameters
G: graph
The graph to be hashed. Can have node and/or edge attributes. Can also have no attributes.
edge_attr: string, default=None
The key in edge attribute dictionary to be used for hashing. If None, edge labels are ignored.
node_attr: string, default=None
The key in node attribute dictionary to be used for hashing. If None, and no edge_attr given,
use the degrees of the nodes as labels.
iterations: int, default=3
Number of neighbor aggregations to perform. Should be larger for larger graphs.
digest_size: int, default=16
Size (in bits) of blake2b hash digest to use for hashing node labels.

Returns
h
[string] Hexadecimal string corresponding to hash of the input graph.
See also:
weisfeiler_lehman_subgraph_hashes
Notes
To return the WL hashes of each subgraph of a graph, use weisfeiler_lehman_subgraph_hashes

Similarity between hashes does not imply similarity between graphs.
References
[1]
Examples
Two graphs with edge attributes that are isomorphic, except for differences in the edge labels.
>>> G1 = nx.Graph()
>>> G1.add_edges_from(
... [
... (1, 2, {"label": "A"}),
... (2, 3, {"label": "A"}),
... (3, 1, {"label": "A"}),
... (1, 4, {"label": "B"}),
... ]
... )
>>> G2 = nx.Graph()
>>> G2.add_edges_from(
... [
... (5, 6, {"label": "B"}),
... (6, 7, {"label": "A"}),
... (7, 5, {"label": "A"}),
... (7, 8, {"label": "A"}),
... ]
... )
Omitting the edge_attr option, results in identical hashes.
>>> nx.weisfeiler_lehman_graph_hash(G1)
'7bc4dde9a09d0b94c5097b219891d81a'
>>> nx.weisfeiler_lehman_graph_hash(G2)
'7bc4dde9a09d0b94c5097b219891d81a'
With edge labels, the graphs are no longer assigned the same hash digest.
>>> nx.weisfeiler_lehman_graph_hash(G1, edge_attr="label")

'c653d85538bcf041d88c011f4f905f10'
>>> nx.weisfeiler_lehman_graph_hash(G2, edge_attr="label")
'3dcd84af1ca855d0eff3c978d88e7ec7'
3.30. Graph Hashing 485

3.30.2 weisfeiler_lehman_subgraph_hashes
weisfeiler_lehman_subgraph_hashes(G, edge_attr=None, node_attr=None, iterations=3, digest_size=16)

Return a dictionary of subgraph hashes by node.
The dictionary is keyed by node to a list of hashes in increasingly sized induced subgraphs containing the nodes
within 2*k edges of the key node for increasing integer k until all nodes are included.
The function iteratively aggregates and hashes neighbourhoods of each node. This is achieved for each step by
replacing for each node its label from the previous iteration with its hashed 1-hop neighborhood aggregate. The
new node label is then appended to a list of node labels for each node.
To aggregate neighborhoods at each step for a node n, all labels of nodes adjacent to n are concatenated. If the
edge_attr parameter is set, labels for each neighboring node are prefixed with the value of this attribute along
the connecting edge from this neighbor to node n. The resulting string is then hashed to compress this information
into a fixed digest size.
Thus, at the i`thiterationnodeswithin : math : `2i distance influence any given hashed node label. We can
therefore say that at depth i for node n we have a hash for a subgraph induced by the 2i-hop neighborhood of n.
Can be used to to create general Weisfeiler-Lehman graph kernels, or generate features for graphs or nodes, for
example to generate ‘words’ in a graph as seen in the ‘graph2vec’ algorithm. See [1] & [2] respectively for details.
Hashes are identical for isomorphic subgraphs and there exist strong guarantees that non-isomorphic graphs will
get different hashes. See [1] for details.
If no node or edge attributes are provided, the degree of each node is used as its initial label. Otherwise, node
and/or edge labels are used to compute the hash.
Parameters
G: graph
The graph to be hashed. Can have node and/or edge attributes. Can also have no attributes.
edge_attr: string, default=None
The key in edge attribute dictionary to be used for hashing. If None, edge labels are ignored.
node_attr: string, default=None
The key in node attribute dictionary to be used for hashing. If None, and no edge_attr given,
use the degrees of the nodes as labels.
iterations: int, default=3
Number of neighbor aggregations to perform. Should be larger for larger graphs.
digest_size: int, default=16
Size (in bits) of blake2b hash digest to use for hashing node labels. The default size is 16 bits
Returns
node_subgraph_hashes
[dict] A dictionary with each key given by a node in G, and each value given by the subgraph
hashes in order of depth from the key node.
See also:
weisfeiler_lehman_graph_hash

Notes
To hash the full graph when subgraph hashes are not needed, use weisfeiler_lehman_graph_hash for
efficiency.
Similarity between hashes does not imply similarity between graphs.
References
[1], [2]
Examples
Finding similar nodes in different graphs:
>>> G1 = nx.Graph()
>>> G1.add_edges_from([
... (1, 2), (2, 3), (2, 4), (3, 5), (4, 6), (5, 7), (6, 7)
... ])
>>> G2 = nx.Graph()
>>> G2.add_edges_from([
... (1, 3), (2, 3), (1, 6), (1, 5), (4, 6)
... ])
>>> g1_hashes = nx.weisfeiler_lehman_subgraph_hashes(G1, iterations=3, digest_
,→size=8)
>>> g2_hashes = nx.weisfeiler_lehman_subgraph_hashes(G2, iterations=3, digest_

,→size=8)
Even though G1 and G2 are not isomorphic (they have different numbers of edges), the hash sequence of depth 3
for node 1 in G1 and node 5 in G2 are similar:
>>> g1_hashes[1]
['a93b64973cfc8897', 'db1b43ae35a1878f', '57872a7d2059c1c0']
>>> g2_hashes[5]
['a93b64973cfc8897', 'db1b43ae35a1878f', '1716d2a4012fa4bc']
The first 2 WL subgraph hashes match. From this we can conclude that it’s very likely the neighborhood of 4 hops
around these nodes are isomorphic: each iteration aggregates 1-hop neighbourhoods meaning hashes at depth n are
influenced by every node within 2n hops.
However the neighborhood of 6 hops is no longer isomorphic since their 3rd hash does not match.
These nodes may be candidates to be classified together since their local topology is similar.
3.31 Graphical degree sequence
Test sequences for graphiness.
3.31. Graphical degree sequence 487

is_graphical(sequence[, method]) Returns True if sequence is a valid degree sequence.

is_digraphical(in_sequence, out_sequence) Returns True if some directed graph can realize the in-
and out-degree sequences.
is_multigraphical(sequence) Returns True if some multigraph can realize the sequence.
is_pseudographical(sequence) Returns True if some pseudograph can realize the se-
quence.
is_valid_degree_sequence_havel_hakimi(...)Returns True if deg_sequence can be realized by a simple
graph.
is_valid_degree_sequence_erdos_gallai(...)Returns True if deg_sequence can be realized by a simple
graph.
3.31.1 is_graphical
is_graphical(sequence, method='eg')
Returns True if sequence is a valid degree sequence.
A degree sequence is valid if some graph can realize it.
Parameters
sequence
[list or iterable container] A sequence of integer node degrees
method
[“eg” | “hh” (default: ‘eg’)] The method used to validate the degree sequence. “eg” corresponds
to the Erdős-Gallai algorithm [EG1960], [choudum1986], and “hh” to the Havel-Hakimi al-
gorithm [havel1955], [hakimi1962], [CL1996].
Returns
valid
[bool] True if the sequence is a valid degree sequence and False if not.
References
[EG1960], [choudum1986], [havel1955], [hakimi1962], [CL1996]
Examples
>>> sequence = (d for n, d in G.degree())
>>> nx.is_graphical(sequence)
True

3.31.2 is_digraphical
is_digraphical(in_sequence, out_sequence)
Returns True if some directed graph can realize the in- and out-degree sequences.
Parameters
in_sequence
[list or iterable container] A sequence of integer node in-degrees
out_sequence
[list or iterable container] A sequence of integer node out-degrees
Returns
valid
[bool] True if in and out-sequences are digraphic False if not.
Notes
This algorithm is from Kleitman and Wang [1]. The worst case runtime is O(s × log n) where s and n are the sum
and length of the sequences respectively.
References
[1]
3.31.3 is_multigraphical
is_multigraphical(sequence)
Returns True if some multigraph can realize the sequence.
Parameters
sequence
[list] A list of integers
Returns
valid
[bool] True if deg_sequence is a multigraphic degree sequence and False if not.
Notes
The worst-case run time is O(n) where n is the length of the sequence.

References
[1]
3.31.4 is_pseudographical
is_pseudographical(sequence)
Returns True if some pseudograph can realize the sequence.
Every nonnegative integer sequence with an even sum is pseudographical (see [1]).
Parameters
sequence
[list or iterable container] A sequence of integer node degrees
Returns
valid
[bool] True if the sequence is a pseudographic degree sequence and False if not.
Notes
The worst-case run time is O(n) where n is the length of the sequence.
References
[1]
3.31.5 is_valid_degree_sequence_havel_hakimi
is_valid_degree_sequence_havel_hakimi(deg_sequence)
Returns True if deg_sequence can be realized by a simple graph.
The validation proceeds using the Havel-Hakimi theorem [havel1955], [hakimi1962], [CL1996]. Worst-case run
time is O(s) where s is the sum of the sequence.
Parameters
deg_sequence
[list] A list of integers where each element specifies the degree of a node in a graph.
Returns
valid
[bool] True if deg_sequence is graphical and False if not.

Notes
The ZZ condition says that for the sequence d if
(max(d) + min(d) + 1)2

|d| >=
4 ∗ min(d)
then d is graphical. This was shown in Theorem 6 in [1].
References
[1], [havel1955], [hakimi1962], [CL1996]
3.31.6 is_valid_degree_sequence_erdos_gallai
is_valid_degree_sequence_erdos_gallai(deg_sequence)
Returns True if deg_sequence can be realized by a simple graph.
The validation is done using the Erdős-Gallai theorem [EG1960].
Parameters
deg_sequence
[list] A list of integers
Returns
valid
[bool] True if deg_sequence is graphical and False if not.
Notes
This implementation uses an equivalent form of the Erdős-Gallai criterion. Worst-case run time is O(n) where n
is the length of the sequence.
Specifically, a sequence d is graphical if and only if the sum of the sequence is even and for all strong indices k in
the sequence,
∑
k ∑
n ∑
k−1 ∑
k−1
di ≤ k(k − 1) + min(di , k) = k(n − 1) − (k nj − jnj )
i=1 j=k+1 j=0 j=0
A strong index k is any index where d_k >= k and the value n_j is the number of occurrences of j in d. The maximal
strong index is called the Durfee index.
This particular rearrangement comes from the proof of Theorem 3 in [2].
The ZZ condition says that for the sequence d if
(max(d) + min(d) + 1)2

|d| >=
4 ∗ min(d)
then d is graphical. This was shown in Theorem 6 in [2].

References
[1], [2], [EG1960]
3.32 Hierarchy
Flow Hierarchy.
flow_hierarchy(G[, weight]) Returns the flow hierarchy of a directed network.
3.32.1 flow_hierarchy
flow_hierarchy(G, weight=None)
Returns the flow hierarchy of a directed network.
Flow hierarchy is defined as the fraction of edges not participating in cycles in a directed graph [1].
Parameters
G
[DiGraph or MultiDiGraph] A directed graph
weight
[key,optional (default=None)] Attribute to use for node weights. If None the weight defaults
to 1.
Returns
h
[float] Flow hierarchy value
Notes
The algorithm described in [1] computes the flow hierarchy through exponentiation of the adjacency matrix. This
function implements an alternative approach that finds strongly connected components. An edge is in a cycle if and
only if it is in a strongly connected component, which can be found in O(m) time using Tarjan’s algorithm.
References
[1]
3.33 Hybrid
Provides functions for finding and testing for locally (k, l)-connected graphs.
kl_connected_subgraph(G, k, l[, low_memory, Returns the maximum locally (k, l)-connected sub-
...]) graph of G.
is_kl_connected(G, k, l[, low_memory]) Returns True if and only if G is locally (k, l)-
connected.

3.33.1 kl_connected_subgraph
kl_connected_subgraph(G, k, l, low_memory=False, same_as_graph=False)

Returns the maximum locally (k, l)-connected subgraph of G.
A graph is locally (k, l)-connected if for each edge (u, v) in the graph there are at least l edge-disjoint paths
of length at most k joining u to v.
Parameters
G
[NetworkX graph] The graph in which to find a maximum locally (k, l)-connected sub-
graph.
k
[integer] The maximum length of paths to consider. A higher number means a looser connec-
tivity requirement.
l
[integer] The number of edge-disjoint paths. A higher number means a stricter connectivity
requirement.
low_memory
[bool] If this is True, this function uses an algorithm that uses slightly more time but less
memory.
same_as_graph
[bool] If True then return a tuple of the form (H, is_same), where H is the maximum
locally (k, l)-connected subgraph and is_same is a Boolean representing whether G is
locally (k, l)-connected (and hence, whether H is simply a copy of the input graph G).
Returns
NetworkX graph or two-tuple
If same_as_graph is True, then this function returns a two-tuple as described above. Oth-
erwise, it returns only the maximum locally (k, l)-connected subgraph.
See also:
is_kl_connected
References
[1]
3.33.2 is_kl_connected
is_kl_connected(G, k, l, low_memory=False)
Returns True if and only if G is locally (k, l)-connected.
A graph is locally (k, l)-connected if for each edge (u, v) in the graph there are at least l edge-disjoint paths
of length at most k joining u to v.
Parameters
G
[NetworkX graph] The graph to test for local (k, l)-connectedness.
3.33. Hybrid 493

k
[integer] The maximum length of paths to consider. A higher number means a looser connec-
tivity requirement.
l
[integer] The number of edge-disjoint paths. A higher number means a stricter connectivity
requirement.
low_memory
[bool] If this is True, this function uses an algorithm that uses slightly more time but less
memory.
Returns
bool
Whether the graph is locally (k, l)-connected subgraph.
See also:
kl_connected_subgraph
References
[1]
3.34 Isolates
Functions for identifying isolate (degree zero) nodes.
is_isolate(G, n) Determines whether a node is an isolate.

isolates(G) Iterator over isolates in the graph.
number_of_isolates(G) Returns the number of isolates in the graph.
3.34.1 is_isolate
is_isolate(G, n)
Determines whether a node is an isolate.
An isolate is a node with no neighbors (that is, with degree zero). For directed graphs, this means no in-neighbors
and no out-neighbors.
Parameters
G
[NetworkX graph]
n
[node] A node in G.
Returns
is_isolate
[bool] True if and only if n has no neighbors.

Examples
>>> G = nx.Graph()
>>> G.add_node(3)
>>> nx.is_isolate(G, 2)
False
>>> nx.is_isolate(G, 3)
True
3.34.2 isolates
isolates(G)
Iterator over isolates in the graph.
Parameters
G
[NetworkX graph]
Returns
iterator
An iterator over the isolates of G.
Examples
To get a list of all isolates of a graph, use the list constructor:
>>> G = nx.Graph()
>>> G.add_node(3)
>>> list(nx.isolates(G))
[3]
To remove all isolates in the graph, first create a list of the isolates, then use Graph.remove_nodes_from():
>>> list(G)
[1, 2]
For digraphs, isolates have zero in-degree and zero out_degre:
>>> G = nx.DiGraph([(0, 1), (1, 2)])

>>> G.add_node(3)
>>> list(nx.isolates(G))
[3]
3.34. Isolates 495

3.34.3 number_of_isolates
number_of_isolates(G)
Returns the number of isolates in the graph.
Parameters
G
[NetworkX graph]
Returns
int
The number of degree zero nodes in the graph G.
3.35 Isomorphism
is_isomorphic(G1, G2[, node_match, Returns True if the graphs G1 and G2 are isomorphic and
edge_match]) False otherwise.
could_be_isomorphic(G1, G2) Returns False if graphs are definitely not isomorphic.
fast_could_be_isomorphic(G1, G2) Returns False if graphs are definitely not isomorphic.
faster_could_be_isomorphic(G1, G2) Returns False if graphs are definitely not isomorphic.
3.35.1 is_isomorphic
is_isomorphic(G1, G2, node_match=None, edge_match=None)

Returns True if the graphs G1 and G2 are isomorphic and False otherwise.
Parameters
G1, G2: graphs
The two graphs G1 and G2 must be the same type.
node_match
[callable] A function that returns True if node n1 in G1 and n2 in G2 should be considered
equal during the isomorphism test. If node_match is not specified then node attributes are not
considered.
The function will be called like
node_match(G1.nodes[n1], G2.nodes[n2]).
That is, the function will receive the node attribute dictionaries for n1 and n2 as inputs.
edge_match
[callable] A function that returns True if the edge attribute dictionary for the pair of nodes
(u1, v1) in G1 and (u2, v2) in G2 should be considered equal during the isomorphism test. If
edge_match is not specified then edge attributes are not considered.
edge_match(G1[u1][v1], G2[u2][v2]).

That is, the function will receive the edge attribute dictionaries of the edges under considera-
tion.
See also:
numerical_node_match, numerical_edge_match, numerical_multiedge_match

categorical_node_match, categorical_edge_match, categorical_multiedge_match
Notes
Uses the vf2 algorithm [1].
References
[1]
Examples
>>> import networkx.algorithms.isomorphism as iso
For digraphs G1 and G2, using ‘weight’ edge attribute (default: 1)
>>> G1 = nx.DiGraph()
>>> G2 = nx.DiGraph()
>>> nx.add_path(G1, [1, 2, 3, 4], weight=1)
>>> em = iso.numerical_edge_match("weight", 1)
>>> nx.is_isomorphic(G1, G2) # no weights considered
True
>>> nx.is_isomorphic(G1, G2, edge_match=em) # match weights
False
For multidigraphs G1 and G2, using ‘fill’ node attribute (default: ‘’)
>>> G1 = nx.MultiDiGraph()
>>> G2 = nx.MultiDiGraph()
>>> G1.add_nodes_from([1, 2, 3], fill="red")
>>> G2.add_nodes_from([10, 20, 30, 40], fill="red")
>>> nx.add_path(G1, [1, 2, 3, 4], weight=3, linewidth=2.5)
>>> nm = iso.categorical_node_match("fill", "red")
>>> nx.is_isomorphic(G1, G2, node_match=nm)
True
For multidigraphs G1 and G2, using ‘weight’ edge attribute (default: 7)
>>> G1.add_edge(1, 2, weight=7)

1
>>> G2.add_edge(10, 20)
1
>>> em = iso.numerical_multiedge_match("weight", 7, rtol=1e-6)
>>> nx.is_isomorphic(G1, G2, edge_match=em)
True
3.35. Isomorphism 497

For multigraphs G1 and G2, using ‘weight’ and ‘linewidth’ edge attributes with default values 7 and 2.5. Also using
‘fill’ node attribute with default value ‘red’.
>>> em = iso.numerical_multiedge_match(["weight", "linewidth"], [7, 2.5])

>>> nm = iso.categorical_node_match("fill", "red")
>>> nx.is_isomorphic(G1, G2, edge_match=em, node_match=nm)
True
3.35.2 could_be_isomorphic
could_be_isomorphic(G1, G2)
Returns False if graphs are definitely not isomorphic. True does NOT guarantee isomorphism.
Parameters
G1, G2
[graphs] The two graphs G1 and G2 must be the same type.
Notes
Checks for matching degree, triangle, and number of cliques sequences.
3.35.3 fast_could_be_isomorphic
fast_could_be_isomorphic(G1, G2)
Returns False if graphs are definitely not isomorphic.
True does NOT guarantee isomorphism.
Parameters
G1, G2
Notes
Checks for matching degree and triangle sequences.
3.35.4 faster_could_be_isomorphic
faster_could_be_isomorphic(G1, G2)
Returns False if graphs are definitely not isomorphic.
True does NOT guarantee isomorphism.
Parameters
G1, G2

Notes
Checks for matching degree sequences.
3.35.5 VF2++
VF2++ Algorithm
An implementation of the VF2++ algorithm for Graph Isomorphism testing.

The simplest interface to use this module is to call:
vf2pp_is_isomorphic: to check whether two graphs are isomorphic. vf2pp_isomorphism: to obtain the node
mapping between two graphs, in case they are isomorphic. vf2pp_all_isomorphisms: to generate all possible
mappings between two graphs, if isomorphic.
Introduction
The VF2++ algorithm, follows a similar logic to that of VF2, while also introducing new easy-to-check cutting rules and
determining the optimal access order of nodes. It is also implemented in a non-recursive manner, which saves both time
and space, when compared to its previous counterpart.
The optimal node ordering is obtained after taking into consideration both the degree but also the label rarity of each
node. This way we place the nodes that are more likely to match, first in the order, thus examining the most promising
branches in the beginning. The rules also consider node labels, making it easier to prune unfruitful branches early in the
process.
Examples
Suppose G1 and G2 are Isomorphic Graphs. Verification is as follows:

Without node labels:

>>> G1 = nx.path_graph(4)
>>> nx.vf2pp_is_isomorphic(G1, G2, node_label=None)
True
>>> nx.vf2pp_isomorphism(G1, G2, node_label=None)
{1: 1, 2: 2, 0: 0, 3: 3}
With node labels:
>>> mapped = {1: 1, 2: 2, 3: 3, 0: 0}
>>> nx.set_node_attributes(G1, dict(zip(G1, ["blue", "red", "green", "yellow"])),
,→"label")
>>> nx.set_node_attributes(G2, dict(zip([mapped[u] for u in G1], ["blue", "red",

,→"green", "yellow"])), "label")
>>> nx.vf2pp_is_isomorphic(G1, G2, node_label="label")

True
>>> nx.vf2pp_isomorphism(G1, G2, node_label="label")
{1: 1, 2: 2, 0: 0, 3: 3}

vf2pp_is_isomorphic(G1, G2[, node_label, ...]) Examines whether G1 and G2 are isomorphic.

vf2pp_all_isomorphisms(G1, G2[, node_label, Yields all the possible mappings between G1 and G2.
...])
vf2pp_isomorphism(G1, G2[, node_label, ...]) Return an isomorphic mapping between G1 and G2 if it
exists.
vf2pp_is_isomorphic
vf2pp_is_isomorphic(G1, G2, node_label=None, default_label=None)

Examines whether G1 and G2 are isomorphic.
Parameters
G1, G2
[NetworkX Graph or MultiGraph instances.] The two graphs to check for isomorphism.
node_label
[str, optional] The name of the node attribute to be used when comparing nodes. The default
is None, meaning node attributes are not considered in the comparison. Any node that doesn’t
have the node_label attribute uses default_label instead.
default_label
[scalar] Default value to use when a node doesn’t have an attribute named node_label.
Default is None.
Returns
bool
True if the two graphs are isomorphic, False otherwise.
vf2pp_all_isomorphisms
vf2pp_all_isomorphisms(G1, G2, node_label=None, default_label=None)

Yields all the possible mappings between G1 and G2.
Parameters
G1, G2
node_label
default_label
Default is None.
Yields
dict
Isomorphic mapping between the nodes in G1 and G2.

vf2pp_isomorphism
vf2pp_isomorphism(G1, G2, node_label=None, default_label=None)

Return an isomorphic mapping between G1 and G2 if it exists.
Parameters
G1, G2
node_label
default_label
Default is None.
Returns
dict or None
Node mapping if the two graphs are isomorphic. None otherwise.
3.35.6 Tree Isomorphism
An algorithm for finding if two undirected trees are isomorphic, and if so returns an isomorphism between the two sets
of nodes.
This algorithm uses a routine to tell if two rooted trees (trees with a specified root node) are isomorphic, which may be
independently useful.
This implements an algorithm from: The Design and Analysis of Computer Algorithms by Aho, Hopcroft, and Ullman
Addison-Wesley Publishing 1974 Example 3.2 pp. 84-86.
A more understandable version of this algorithm is described in: Homework Assignment 5 McGill University SOCS
308-250B, Winter 2002 by Matthew Suderman http://crypto.cs.mcgill.ca/~crepeau/CS250/2004/HW5+.pdf
rooted_tree_isomorphism(t1, root1, t2, root2) Given two rooted trees t1 and t2, with roots root1 and
root2 respectivly this routine will determine if they are
isomorphic.
tree_isomorphism(t1, t2) Given two undirected (or free) trees t1 and t2, this rou-
tine will determine if they are isomorphic.
rooted_tree_isomorphism
rooted_tree_isomorphism(t1, root1, t2, root2)

Given two rooted trees t1 and t2, with roots root1 and root2 respectivly this routine will determine if they
are isomorphic.
These trees may be either directed or undirected, but if they are directed, all edges should flow from the root.
It returns the isomorphism, a mapping of the nodes of t1 onto the nodes of t2, such that two trees are then
identical.
Note that two trees may have more than one isomorphism, and this routine just returns one valid mapping.
Parameters

`t1`
[NetworkX graph] One of the trees being compared
`root1`
[a node of t1 which is the root of the tree]
`t2`
[undirected NetworkX graph] The other tree being compared
`root2`
[a node of t2 which is the root of the tree]
This is a subroutine used to implement `tree_isomorphism`, but will
be somewhat faster if you already have rooted trees.
Returns
isomorphism
[list] A list of pairs in which the left element is a node in t1 and the right element is a node
in t2. The pairs are in arbitrary order. If the nodes in one tree is mapped to the names in the
other, then trees will be identical. Note that an isomorphism will not necessarily be unique.
If t1 and t2 are not isomorphic, then it returns the empty list.
tree_isomorphism
tree_isomorphism(t1, t2)
Given two undirected (or free) trees t1 and t2, this routine will determine if they are isomorphic. It returns the
isomorphism, a mapping of the nodes of t1 onto the nodes of t2, such that two trees are then identical.
Note that two trees may have more than one isomorphism, and this routine just returns one valid mapping.
Parameters
t1
[undirected NetworkX graph] One of the trees being compared
t2
[undirected NetworkX graph] The other tree being compared
Returns
isomorphism
[list] A list of pairs in which the left element is a node in t1 and the right element is a node
in t2. The pairs are in arbitrary order. If the nodes in one tree is mapped to the names in the
other, then trees will be identical. Note that an isomorphism will not necessarily be unique.
If t1 and t2 are not isomorphic, then it returns the empty list.
Notes
This runs in O(n*log(n)) time for trees with n nodes.

3.35.7 Advanced Interfaces
VF2 Algorithm
An implementation of VF2 algorithm for graph isomorphism testing.

The simplest interface to use this module is to call networkx.is_isomorphic().
Introduction
The GraphMatcher and DiGraphMatcher are responsible for matching graphs or directed graphs in a predetermined
manner. This usually means a check for an isomorphism, though other checks are also possible. For example, a subgraph
of one graph can be checked for isomorphism to a second graph.
Matching is done via syntactic feasibility. It is also possible to check for semantic feasibility. Feasibility, then, is defined
as the logical AND of the two functions.
To include a semantic check, the (Di)GraphMatcher class should be subclassed, and the semantic_feasibility() function
should be redefined. By default, the semantic feasibility function always returns True. The effect of this is that semantics
are not considered in the matching of G1 and G2.
Examples
Suppose G1 and G2 are isomorphic graphs. Verification is as follows:
>>> from networkx.algorithms import isomorphism

>>> GM = isomorphism.GraphMatcher(G1, G2)
>>> GM.is_isomorphic()
True
GM.mapping stores the isomorphism mapping from G1 to G2.
>>> GM.mapping
{0: 0, 1: 1, 2: 2, 3: 3}
Suppose G1 and G2 are isomorphic directed graphs. Verification is as follows:
>>> G1 = nx.path_graph(4, create_using=nx.DiGraph())

>>> G2 = nx.path_graph(4, create_using=nx.DiGraph())
>>> DiGM = isomorphism.DiGraphMatcher(G1, G2)
>>> DiGM.is_isomorphic()
True
DiGM.mapping stores the isomorphism mapping from G1 to G2.
>>> DiGM.mapping
{0: 0, 1: 1, 2: 2, 3: 3}

Subgraph Isomorphism
Graph theory literature can be ambiguous about the meaning of the above statement, and we seek to clarify it now.
In the VF2 literature, a mapping M is said to be a graph-subgraph isomorphism iff M is an isomorphism between G2
and a subgraph of G1. Thus, to say that G1 and G2 are graph-subgraph isomorphic is to say that a subgraph of G1 is
isomorphic to G2.
Other literature uses the phrase ‘subgraph isomorphic’ as in ‘G1 does not have a subgraph isomorphic to G2’. Another
use is as an in adverb for isomorphic. Thus, to say that G1 and G2 are subgraph isomorphic is to say that a subgraph of
G1 is isomorphic to G2.
Finally, the term ‘subgraph’ can have multiple meanings. In this context, ‘subgraph’ always means a ‘node-induced sub-
graph’. Edge-induced subgraph isomorphisms are not directly supported, but one should be able to perform the check
by making use of nx.line_graph(). For subgraphs which are not induced, the term ‘monomorphism’ is preferred over
‘isomorphism’.
Let G=(N,E) be a graph with a set of nodes N and set of edges E.
If G’=(N’,E’) is a subgraph, then:
N’ is a subset of N E’ is a subset of E
If G’=(N’,E’) is a node-induced subgraph, then:
N’ is a subset of N E’ is the subset of edges in E relating nodes in N’
If G’=(N’,E’) is an edge-induced subgraph, then:
N’ is the subset of nodes in N related by edges in E’ E’ is a subset of E
If G’=(N’,E’) is a monomorphism, then:
N’ is a subset of N E’ is a subset of the set of edges in E relating nodes in N’
Note that if G’ is a node-induced subgraph of G, then it is always a subgraph monomorphism of G, but the opposite is not
always true, as a monomorphism can have fewer edges.
References
[1] Luigi P. Cordella, Pasquale Foggia, Carlo Sansone, Mario Vento,

“A (Sub)Graph Isomorphism Algorithm for Matching Large Graphs”, IEEE Transactions on Pattern Analysis and
Machine Intelligence, vol. 26, no. 10, pp. 1367-1372, Oct., 2004. http://ieeexplore.ieee.org/iel5/34/29305/
01323804.pdf
[2] L. P. Cordella, P. Foggia, C. Sansone, M. Vento, “An Improved
Algorithm for Matching Large Graphs”, 3rd IAPR-TC15 Workshop on Graph-based Representations in Pattern
Recognition, Cuen, pp. 149-159, 2001. https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.101.5342
See Also
syntactic_feasibility(), semantic_feasibility()

Notes
The implementation handles both directed and undirected graphs as well as multigraphs.
In general, the subgraph isomorphism problem is NP-complete whereas the graph isomorphism problem is most likely
not NP-complete (although no polynomial-time algorithm is known to exist).
Graph Matcher
GraphMatcher.__init__(G1, G2[, node_match, Initialize graph matcher.

...])
GraphMatcher.initialize() Reinitializes the state of the algorithm.
GraphMatcher.is_isomorphic() Returns True if G1 and G2 are isomorphic graphs.
GraphMatcher.subgraph_is_isomorphic() Returns True if a subgraph of G1 is isomorphic to G2.
GraphMatcher.isomorphisms_iter() Generator over isomorphisms between G1 and G2.
GraphMatcher.subgraph_isomorphisms_iter() Generator over isomorphisms between a subgraph of G1
and G2.
GraphMatcher.candidate_pairs_iter() Iterator over candidate pairs of nodes in G1 and G2.
GraphMatcher.match() Extends the isomorphism mapping.
GraphMatcher.semantic_feasibility(G1_node, Returns True if mapping G1_node to G2_node is seman-
...) tically feasible.
GraphMatcher.syntactic_feasibility(G1_node,Returns True if adding (G1_node, G2_node) is syntacti-
...) cally feasible.
GraphMatcher.__init__
GraphMatcher.__init__(G1, G2, node_match=None, edge_match=None)

Initialize graph matcher.
Parameters
G1, G2: graph
The graphs to be tested.
node_match: callable
A function that returns True iff node n1 in G1 and n2 in G2 should be considered equal during
the isomorphism test. The function will be called like:
node_match(G1.nodes[n1], G2.nodes[n2])
That is, the function will receive the node attribute dictionaries of the nodes under considera-
tion. If None, then no attributes are considered when testing for an isomorphism.
edge_match: callable
A function that returns True iff the edge attribute dictionary for the pair of nodes (u1, v1) in
G1 and (u2, v2) in G2 should be considered equal during the isomorphism test. The function
will be called like:
edge_match(G1[u1][v1], G2[u2][v2])

GraphMatcher.initialize
GraphMatcher.initialize()
Reinitializes the state of the algorithm.
This method should be redefined if using something other than GMState. If only subclassing GraphMatcher, a
redefinition is not necessary.
GraphMatcher.is_isomorphic
GraphMatcher.is_isomorphic()
Returns True if G1 and G2 are isomorphic graphs.
GraphMatcher.subgraph_is_isomorphic
GraphMatcher.subgraph_is_isomorphic()
Returns True if a subgraph of G1 is isomorphic to G2.
GraphMatcher.isomorphisms_iter
GraphMatcher.isomorphisms_iter()
Generator over isomorphisms between G1 and G2.
GraphMatcher.subgraph_isomorphisms_iter
GraphMatcher.subgraph_isomorphisms_iter()
Generator over isomorphisms between a subgraph of G1 and G2.
GraphMatcher.candidate_pairs_iter
GraphMatcher.candidate_pairs_iter()
Iterator over candidate pairs of nodes in G1 and G2.
GraphMatcher.match
GraphMatcher.match()
Extends the isomorphism mapping.
This function is called recursively to determine if a complete isomorphism can be found between G1 and G2. It
cleans up the class variables after each recursive call. If an isomorphism is found, we yield the mapping.

GraphMatcher.semantic_feasibility
GraphMatcher.semantic_feasibility(G1_node, G2_node)
Returns True if mapping G1_node to G2_node is semantically feasible.
GraphMatcher.syntactic_feasibility
GraphMatcher.syntactic_feasibility(G1_node, G2_node)
Returns True if adding (G1_node, G2_node) is syntactically feasible.
This function returns True if it is adding the candidate pair to the current partial isomorphism/monomorphism
mapping is allowable. The addition is allowable if the inclusion of the candidate pair does not make it impossible
for an isomorphism/monomorphism to be found.
DiGraph Matcher
DiGraphMatcher.__init__(G1, G2[, ...]) Initialize graph matcher.

DiGraphMatcher.initialize() Reinitializes the state of the algorithm.
DiGraphMatcher.is_isomorphic() Returns True if G1 and G2 are isomorphic graphs.
DiGraphMatcher.subgraph_is_isomorphic() Returns True if a subgraph of G1 is isomorphic to G2.
DiGraphMatcher.isomorphisms_iter() Generator over isomorphisms between G1 and G2.
DiGraphMatcher.subgraph_isomorphisms_iter()
Generator over isomorphisms between a subgraph of G1
and G2.
DiGraphMatcher.candidate_pairs_iter() Iterator over candidate pairs of nodes in G1 and G2.
DiGraphMatcher.match() Extends the isomorphism mapping.
DiGraphMatcher.semantic_feasibility(G1_node,
Returns True if mapping G1_node to G2_node is seman-
...) tically feasible.
DiGraphMatcher.syntactic_feasibility(...) Returns True if adding (G1_node, G2_node) is syntacti-
cally feasible.
DiGraphMatcher.__init__
DiGraphMatcher.__init__(G1, G2, node_match=None, edge_match=None)

Initialize graph matcher.
Parameters
G1, G2
[graph] The graphs to be tested.
node_match
[callable] A function that returns True iff node n1 in G1 and n2 in G2 should be considered
equal during the isomorphism test. The function will be called like:
node_match(G1.nodes[n1], G2.nodes[n2])
That is, the function will receive the node attribute dictionaries of the nodes under considera-
edge_match
[callable] A function that returns True iff the edge attribute dictionary for the pair of nodes

(u1, v1) in G1 and (u2, v2) in G2 should be considered equal during the isomorphism test.
The function will be called like:
edge_match(G1[u1][v1], G2[u2][v2])
DiGraphMatcher.initialize
DiGraphMatcher.initialize()
Reinitializes the state of the algorithm.
This method should be redefined if using something other than DiGMState. If only subclassing GraphMatcher, a
redefinition is not necessary.
DiGraphMatcher.is_isomorphic
DiGraphMatcher.is_isomorphic()
Returns True if G1 and G2 are isomorphic graphs.
DiGraphMatcher.subgraph_is_isomorphic
DiGraphMatcher.subgraph_is_isomorphic()
Returns True if a subgraph of G1 is isomorphic to G2.
DiGraphMatcher.isomorphisms_iter
DiGraphMatcher.isomorphisms_iter()
Generator over isomorphisms between G1 and G2.
DiGraphMatcher.subgraph_isomorphisms_iter
DiGraphMatcher.subgraph_isomorphisms_iter()
Generator over isomorphisms between a subgraph of G1 and G2.
DiGraphMatcher.candidate_pairs_iter
DiGraphMatcher.candidate_pairs_iter()
Iterator over candidate pairs of nodes in G1 and G2.

DiGraphMatcher.match
DiGraphMatcher.match()
Extends the isomorphism mapping.
This function is called recursively to determine if a complete isomorphism can be found between G1 and G2. It
cleans up the class variables after each recursive call. If an isomorphism is found, we yield the mapping.
DiGraphMatcher.semantic_feasibility
DiGraphMatcher.semantic_feasibility(G1_node, G2_node)
Returns True if mapping G1_node to G2_node is semantically feasible.
DiGraphMatcher.syntactic_feasibility
DiGraphMatcher.syntactic_feasibility(G1_node, G2_node)
Returns True if adding (G1_node, G2_node) is syntactically feasible.
This function returns True if it is adding the candidate pair to the current partial isomorphism/monomorphism
mapping is allowable. The addition is allowable if the inclusion of the candidate pair does not make it impossible
for an isomorphism/monomorphism to be found.
Match helpers
categorical_node_match(attr, default) Returns a comparison function for a categorical node at-

tribute.
categorical_edge_match(attr, default) Returns a comparison function for a categorical edge at-
tribute.
categorical_multiedge_match(attr, default) Returns a comparison function for a categorical edge at-
tribute.
numerical_node_match(attr, default[, rtol, atol]) Returns a comparison function for a numerical node at-
tribute.
numerical_edge_match(attr, default[, rtol, atol]) Returns a comparison function for a numerical edge at-
tribute.
numerical_multiedge_match(attr, default[, ...]) Returns a comparison function for a numerical edge at-
tribute.
generic_node_match(attr, default, op) Returns a comparison function for a generic attribute.
generic_edge_match(attr, default, op) Returns a comparison function for a generic attribute.
generic_multiedge_match(attr, default, op) Returns a comparison function for a generic attribute.

categorical_node_match
categorical_node_match(attr, default)
Returns a comparison function for a categorical node attribute.
The value(s) of the attr(s) must be hashable and comparable via the == operator since they are placed into a set([])
object. If the sets from G1 and G2 are the same, then the constructed function returns True.
Parameters
attr
[string | list] The categorical node attribute to compare, or a list of categorical node attributes
to compare.
default
[value | list] The default value for the categorical node attribute, or a list of default values for
the categorical node attributes.
Returns
match
[function] The customized, categorical node_match function.
Examples

>>> nm = iso.categorical_node_match("size", 1)
>>> nm = iso.categorical_node_match(["color", "size"], ["red", 2])
categorical_edge_match
categorical_edge_match(attr, default)
Returns a comparison function for a categorical edge attribute.
Parameters
attr
[string | list] The categorical edge attribute to compare, or a list of categorical edge attributes
to compare.
default
[value | list] The default value for the categorical edge attribute, or a list of default values for
the categorical edge attributes.
Returns
match
[function] The customized, categorical edge_match function.

Examples

>>> nm = iso.categorical_edge_match("size", 1)
>>> nm = iso.categorical_edge_match(["color", "size"], ["red", 2])
categorical_multiedge_match
categorical_multiedge_match(attr, default)
Returns a comparison function for a categorical edge attribute.
Parameters
attr
[string | list] The categorical edge attribute to compare, or a list of categorical edge attributes
to compare.
default
[value | list] The default value for the categorical edge attribute, or a list of default values for
the categorical edge attributes.
Returns
match
[function] The customized, categorical edge_match function.
Examples

>>> nm = iso.categorical_multiedge_match("size", 1)
>>> nm = iso.categorical_multiedge_match(["color", "size"], ["red", 2])
numerical_node_match
numerical_node_match(attr, default, rtol=1e-05, atol=1e-08)

Returns a comparison function for a numerical node attribute.
The value(s) of the attr(s) must be numerical and sortable. If the sorted list of values from G1 and G2 are the same
within some tolerance, then the constructed function returns True.
Parameters
attr
[string | list] The numerical node attribute to compare, or a list of numerical node attributes to
compare.
default
[value | list] The default value for the numerical node attribute, or a list of default values for
the numerical node attributes.
rtol
[float] The relative error tolerance.

atol
[float] The absolute error tolerance.
Returns
match
[function] The customized, numerical node_match function.
Examples

>>> nm = iso.numerical_node_match("weight", 1.0)
>>> nm = iso.numerical_node_match(["weight", "linewidth"], [0.25, 0.5])
numerical_edge_match
numerical_edge_match(attr, default, rtol=1e-05, atol=1e-08)

Returns a comparison function for a numerical edge attribute.
Parameters
attr
[string | list] The numerical edge attribute to compare, or a list of numerical edge attributes to
compare.
default
[value | list] The default value for the numerical edge attribute, or a list of default values for
the numerical edge attributes.
rtol
atol
Returns
match
[function] The customized, numerical edge_match function.
Examples

>>> nm = iso.numerical_edge_match("weight", 1.0)
>>> nm = iso.numerical_edge_match(["weight", "linewidth"], [0.25, 0.5])

numerical_multiedge_match
numerical_multiedge_match(attr, default, rtol=1e-05, atol=1e-08)

Returns a comparison function for a numerical edge attribute.
Parameters
attr
[string | list] The numerical edge attribute to compare, or a list of numerical edge attributes to
compare.
default
[value | list] The default value for the numerical edge attribute, or a list of default values for
the numerical edge attributes.
rtol
atol
Returns
match
[function] The customized, numerical edge_match function.
Examples

>>> nm = iso.numerical_multiedge_match("weight", 1.0)
>>> nm = iso.numerical_multiedge_match(["weight", "linewidth"], [0.25, 0.5])
generic_node_match
generic_node_match(attr, default, op)

Returns a comparison function for a generic attribute.
The value(s) of the attr(s) are compared using the specified operators. If all the attributes are equal, then the
constructed function returns True.
Parameters
attr
[string | list] The node attribute to compare, or a list of node attributes to compare.
default
[value | list] The default value for the node attribute, or a list of default values for the node
attributes.
op
[callable | list] The operator to use when comparing attribute values, or a list of operators to
use when comparing values for each attribute.
Returns

match
[function] The customized, generic node_match function.
Examples
>>> from operator import eq

>>> from math import isclose
>>> from networkx.algorithms.isomorphism import generic_node_match
>>> nm = generic_node_match("weight", 1.0, isclose)
>>> nm = generic_node_match("color", "red", eq)
>>> nm = generic_node_match(["weight", "color"], [1.0, "red"], [isclose, eq])
generic_edge_match
generic_edge_match(attr, default, op)

constructed function returns True.
Parameters
attr
[string | list] The edge attribute to compare, or a list of edge attributes to compare.
default
[value | list] The default value for the edge attribute, or a list of default values for the edge
attributes.
op
Returns
match
[function] The customized, generic edge_match function.
Examples

>>> from networkx.algorithms.isomorphism import generic_edge_match
>>> nm = generic_edge_match("weight", 1.0, isclose)
>>> nm = generic_edge_match("color", "red", eq)
>>> nm = generic_edge_match(["weight", "color"], [1.0, "red"], [isclose, eq])

generic_multiedge_match
generic_multiedge_match(attr, default, op)

constructed function returns True. Potentially, the constructed edge_match function can be slow since it must
verify that no isomorphism exists between the multiedges before it returns False.
Parameters
attr
[string | list] The edge attribute to compare, or a list of node attributes to compare.
default
[value | list] The default value for the edge attribute, or a list of default values for the dgeat-
tributes.
op
Returns
match
[function] The customized, generic edge_match function.
Examples

>>> from networkx.algorithms.isomorphism import generic_node_match
>>> nm = generic_node_match("weight", 1.0, isclose)
>>> nm = generic_node_match("color", "red", eq)
>>> nm = generic_node_match(["weight", "color"], [1.0, "red"], [isclose, eq])
...
ISMAGS Algorithm
Provides a Python implementation of the ISMAGS algorithm. [1]

It is capable of finding (subgraph) isomorphisms between two graphs, taking the symmetry of the subgraph into account.
In most cases the VF2 algorithm is faster (at least on small graphs) than this implementation, but in some cases there is an
exponential number of isomorphisms that are symmetrically equivalent. In that case, the ISMAGS algorithm will provide
only one solution per symmetry group.
>>> petersen = nx.petersen_graph()

>>> ismags = nx.isomorphism.ISMAGS(petersen, petersen)
>>> isomorphisms = list(ismags.isomorphisms_iter(symmetry=False))
>>> len(isomorphisms)
120
>>> isomorphisms = list(ismags.isomorphisms_iter(symmetry=True))
>>> answer = [{0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9}]
>>> answer == isomorphisms
True

In addition, this implementation also provides an interface to find the largest common induced subgraph [2] between any
two graphs, again taking symmetry into account. Given graph and subgraph the algorithm will remove nodes from
the subgraph until subgraph is isomorphic to a subgraph of graph. Since only the symmetry of subgraph is
taken into account it is worth thinking about how you provide your graphs:
>>> graph1 = nx.path_graph(4)
>>> graph2 = nx.star_graph(3)
>>> ismags = nx.isomorphism.ISMAGS(graph1, graph2)
>>> ismags.is_isomorphic()
False
>>> largest_common_subgraph = list(ismags.largest_common_subgraph())
>>> answer = [{1: 0, 0: 1, 2: 2}, {2: 0, 1: 1, 3: 2}]
>>> answer == largest_common_subgraph
True
>>> ismags2 = nx.isomorphism.ISMAGS(graph2, graph1)
>>> largest_common_subgraph = list(ismags2.largest_common_subgraph())
>>> answer = [
... {1: 0, 0: 1, 2: 2},
... {1: 0, 0: 1, 3: 2},
... {2: 0, 0: 1, 1: 2},
... {2: 0, 0: 1, 3: 2},
... {3: 0, 0: 1, 1: 2},
... {3: 0, 0: 1, 2: 2},
... ]
True
However, when not taking symmetry into account, it doesn’t matter:

>>> largest_common_subgraph = list(ismags.largest_common_subgraph(symmetry=False))
>>> answer = [
... {1: 0, 0: 1, 2: 2},
... {1: 0, 2: 1, 0: 2},
... {2: 0, 1: 1, 3: 2},
... {2: 0, 3: 1, 1: 2},
... {1: 0, 0: 1, 2: 3},
... {1: 0, 2: 1, 0: 3},
... {2: 0, 1: 1, 3: 3},
... {2: 0, 3: 1, 1: 3},
... {1: 0, 0: 2, 2: 3},
... {1: 0, 2: 2, 0: 3},
... {2: 0, 1: 2, 3: 3},
... {2: 0, 3: 2, 1: 3},
... ]
True
>>> largest_common_subgraph = list(ismags2.largest_common_subgraph(symmetry=False))
>>> answer = [
... {1: 0, 0: 1, 2: 2},
... {1: 0, 0: 1, 3: 2},
... {2: 0, 0: 1, 1: 2},
... {2: 0, 0: 1, 3: 2},
... {3: 0, 0: 1, 1: 2},
... {3: 0, 0: 1, 2: 2},
... {1: 1, 0: 2, 2: 3},
... {1: 1, 0: 2, 3: 3},
... {2: 1, 0: 2, 1: 3},
... {2: 1, 0: 2, 3: 3},


... {3: 1, 0: 2, 1: 3},
... {3: 1, 0: 2, 2: 3},
... ]
True
Notes
• The current implementation works for undirected graphs only. The algorithm in general should work for directed
graphs as well though.
• Node keys for both provided graphs need to be fully orderable as well as hashable.
• Node and edge equality is assumed to be transitive: if A is equal to B, and B is equal to C, then A is equal to C.
References
ISMAGS object
ISMAGS(graph, subgraph[, node_match, ...]) Implements the ISMAGS subgraph matching algorith.
networkx.algorithms.isomorphism.ISMAGS
class ISMAGS(graph, subgraph, node_match=None, edge_match=None, cache=None)

Implements the ISMAGS subgraph matching algorith. [1] ISMAGS stands for “Index-based Subgraph Matching
Algorithm with General Symmetries”. As the name implies, it is symmetry aware and will only generate non-
symmetric isomorphisms.
Notes
The implementation imposes additional conditions compared to the VF2 algorithm on the graphs provided and the
comparison functions (node_equality and edge_equality):
• Node keys in both graphs must be orderable as well as hashable.
• Equality must be transitive: if A is equal to B, and B is equal to C, then A must be equal to C.

References
[1]
Attributes
graph: networkx.Graph
subgraph: networkx.Graph
node_equality: collections.abc.Callable
The function called to see if two nodes should be considered equal. It’s signature looks like
this: f(graph1: networkx.Graph, node1, graph2: networkx.Graph,
node2) -> bool. node1 is a node in graph1, and node2 a node in graph2. Con-
structed from the argument node_match.
edge_equality: collections.abc.Callable
The function called to see if two edges should be considered equal. It’s signature looks like
this: f(graph1: networkx.Graph, edge1, graph2: networkx.Graph,
edge2) -> bool. edge1 is an edge in graph1, and edge2 an edge in graph2.
Constructed from the argument edge_match.
__init__(graph, subgraph, node_match=None, edge_match=None, cache=None)
Parameters
graph: networkx.Graph
subgraph: networkx.Graph
node_match: collections.abc.Callable or None
Function used to determine whether two nodes are equivalent. Its signature should look like
f(n1: dict, n2: dict) -> bool, with n1 and n2 node property dicts. See also
categorical_node_match() and friends. If None, all nodes are considered equal.
edge_match: collections.abc.Callable or None
Function used to determine whether two edges are equivalent. Its signature should look like
f(e1: dict, e2: dict) -> bool, with e1 and e2 edge property dicts. See also
categorical_edge_match() and friends. If None, all edges are considered equal.
cache: collections.abc.Mapping
A cache used for caching graph symmetries.

Methods
analyze_symmetry(graph, node_partitions, ...) Find a minimal set of permutations and corre-

sponding co-sets that describe the symmetry of
graph, given the node and edge equalities given by
node_partitions and edge_colors, respec-
tively.
find_isomorphisms([symmetry]) Find all subgraph isomorphisms between subgraph and
graph
is_isomorphic([symmetry]) Returns True if graph is isomorphic to subgraph
and False otherwise.
isomorphisms_iter([symmetry]) Does the same as find_isomorphisms() if
graph and subgraph have the same number of
nodes.
largest_common_subgraph([symmetry]) Find the largest common induced subgraphs between
subgraph and graph.
subgraph_is_isomorphic([symmetry]) Returns True if a subgraph of graph is isomorphic to
subgraph and False otherwise.
subgraph_isomorphisms_iter([symmetry]) Alternative name for find_isomorphisms().
ISMAGS.analyze_symmetry
ISMAGS.analyze_symmetry(graph, node_partitions, edge_colors)

Find a minimal set of permutations and corresponding co-sets that describe the symmetry of graph, given
the node and edge equalities given by node_partitions and edge_colors, respectively.
Parameters
graph
[networkx.Graph] The graph whose symmetry should be analyzed.
node_partitions
[list of sets] A list of sets containining node keys. Node keys in the same set are considered
equivalent. Every node key in graph should be in exactly one of the sets. If all nodes are
equivalent, this should be [set(graph.nodes)].
edge_colors
[dict mapping edges to their colors] A dict mapping every edge in graph to its corresponding
color. Edges with the same color are considered equivalent. If all edges are equivalent, this
should be {e: 0 for e in graph.edges}.
Returns
set[frozenset]
The found permutations. This is a set of frozensets of pairs of node keys which can be
exchanged without changing subgraph.
dict[collections.abc.Hashable, set[collections.abc.Hashable]]
The found co-sets. The co-sets is a dictionary of {node key: set of node keys}.
Every key-value pair describes which values can be interchanged without changing nodes
less than key.

ISMAGS.find_isomorphisms
ISMAGS.find_isomorphisms(symmetry=True)
Find all subgraph isomorphisms between subgraph and graph
Finds isomorphisms where subgraph <= graph.
Parameters
symmetry: bool
Whether symmetry should be taken into account. If False, found isomorphisms may be
symmetrically equivalent.
Yields
dict
The found isomorphism mappings of {graph_node: subgraph_node}.
ISMAGS.is_isomorphic
ISMAGS.is_isomorphic(symmetry=False)
Returns True if graph is isomorphic to subgraph and False otherwise.
Returns
bool
ISMAGS.isomorphisms_iter
ISMAGS.isomorphisms_iter(symmetry=True)
Does the same as find_isomorphisms() if graph and subgraph have the same number of nodes.
ISMAGS.largest_common_subgraph
ISMAGS.largest_common_subgraph(symmetry=True)
Find the largest common induced subgraphs between subgraph and graph.
Parameters
symmetry: bool
Whether symmetry should be taken into account. If False, found largest common subgraphs
may be symmetrically equivalent.
Yields
dict
The found isomorphism mappings of {graph_node: subgraph_node}.

ISMAGS.subgraph_is_isomorphic
ISMAGS.subgraph_is_isomorphic(symmetry=False)
Returns True if a subgraph of graph is isomorphic to subgraph and False otherwise.
Returns
bool
ISMAGS.subgraph_isomorphisms_iter
ISMAGS.subgraph_isomorphisms_iter(symmetry=True)
Alternative name for find_isomorphisms().
3.36 Link Analysis
3.36.1 PageRank
PageRank analysis of graph structure.
pagerank(G[, alpha, personalization, ...]) Returns the PageRank of the nodes in the graph.
google_matrix(G[, alpha, personalization, ...]) Returns the Google matrix of the graph.
pagerank
pagerank(G, alpha=0.85, personalization=None, max_iter=100, tol=1e-06, nstart=None, weight='weight',

dangling=None)
Returns the PageRank of the nodes in the graph.
PageRank computes a ranking of the nodes in the graph G based on the structure of the incoming links. It was
originally designed as an algorithm to rank web pages.
Parameters
G
[graph] A NetworkX graph. Undirected graphs will be converted to a directed graph with two
directed edges for each undirected edge.
alpha
[float, optional] Damping parameter for PageRank, default=0.85.
personalization: dict, optional
The “personalization vector” consisting of a dictionary with a key some subset of graph nodes
and personalization value each of those. At least one personalization value must be non-zero.
If not specfiied, a nodes personalization value will be zero. By default, a uniform distribution
is used.
max_iter
[integer, optional] Maximum number of iterations in power method eigenvalue solver.
tol
[float, optional] Error tolerance used to check convergence in power method solver.
3.36. Link Analysis 521

nstart
[dictionary, optional] Starting value of PageRank iteration for each node.
weight
[key, optional] Edge data key to use as weight. If None weights are set to 1.
dangling: dict, optional
The outedges to be assigned to any “dangling” nodes, i.e., nodes without any outedges. The
dict key is the node the outedge points to and the dict value is the weight of that outedge. By
default, dangling nodes are given outedges according to the personalization vector (uniform
if not specified). This must be selected to result in an irreducible transition matrix (see notes
under google_matrix). It may be common to have the dangling dict to be the same as the
personalization dict.
Returns
pagerank
[dictionary] Dictionary of nodes with PageRank as value
Raises
See also:
google_matrix
Notes
The eigenvector calculation is done by the power iteration method and has no guarantee of convergence. The
iteration will stop after an error tolerance of len(G) * tol has been reached. If the number of iterations exceed
max_iter, a networkx.exception.PowerIterationFailedConvergence exception is raised.
The PageRank algorithm was designed for directed graphs but this algorithm does not check if the input graph is
directed and will execute on undirected graphs by converting each edge in the directed graph to two edges.
References
[1], [2]
Examples
>>> G = nx.DiGraph(nx.path_graph(4))
>>> pr = nx.pagerank(G, alpha=0.9)

google_matrix
google_matrix(G, alpha=0.85, personalization=None, nodelist=None, weight='weight', dangling=None)

Returns the Google matrix of the graph.
Parameters
G
[graph] A NetworkX graph. Undirected graphs will be converted to a directed graph with two
directed edges for each undirected edge.
alpha
[float] The damping factor.
personalization: dict, optional
The “personalization vector” consisting of a dictionary with a key some subset of graph nodes
and personalization value each of those. At least one personalization value must be non-zero.
If not specfiied, a nodes personalization value will be zero. By default, a uniform distribution
is used.
nodelist
[list, optional] The rows and columns are ordered according to the nodes in nodelist. If nodelist
is None, then the ordering is produced by G.nodes().
weight
[key, optional] Edge data key to use as weight. If None weights are set to 1.
dangling: dict, optional
The outedges to be assigned to any “dangling” nodes, i.e., nodes without any outedges. The
dict key is the node the outedge points to and the dict value is the weight of that outedge. By
default, dangling nodes are given outedges according to the personalization vector (uniform
if not specified) This must be selected to result in an irreducible transition matrix (see notes
below). It may be common to have the dangling dict to be the same as the personalization dict.
Returns
A
[2D NumPy ndarray] Google matrix of the graph
See also:
pagerank
Notes
The array returned represents the transition matrix that describes the Markov chain used in PageRank. For PageR-
ank to converge to a unique solution (i.e., a unique stationary distribution in a Markov chain), the transition matrix
must be irreducible. In other words, it must be that there exists a path between every pair of nodes in the graph, or
else there is the potential of “rank sinks.”
This implementation works with Multi(Di)Graphs. For multigraphs the weight between two nodes is set to be the
sum of all edge weights between those nodes.
3.36. Link Analysis 523

3.36.2 Hits
Hubs and authorities analysis of graph structure.
hits(G[, max_iter, tol, nstart, normalized]) Returns HITS hubs and authorities values for nodes.
hits
hits(G, max_iter=100, tol=1e-08, nstart=None, normalized=True)

Returns HITS hubs and authorities values for nodes.
The HITS algorithm computes two numbers for a node. Authorities estimates the node value based on the incoming
links. Hubs estimates the node value based on outgoing links.
Parameters
G
max_iter
[integer, optional] Maximum number of iterations in power method.
tol
[float, optional] Error tolerance used to check convergence in power method iteration.
nstart
[dictionary, optional] Starting value of each node for power method iteration.
normalized
[bool (default=True)] Normalize results by the sum of all of the values.
Returns
(hubs,authorities)
[two-tuple of dictionaries] Two dictionaries keyed by node containing the hub and authority
values.
Raises
Notes
The eigenvector calculation is done by the power iteration method and has no guarantee of convergence. The
iteration will stop after max_iter iterations or an error tolerance of number_of_nodes(G)*tol has been reached.
The HITS algorithm was designed for directed graphs but this algorithm does not check if the input graph is directed
and will execute on undirected graphs.

References
[1], [2]
Examples
>>> h, a = nx.hits(G)
3.37 Link Prediction
Link prediction algorithms.
resource_allocation_index(G[, ebunch]) Compute the resource allocation index of all node pairs
in ebunch.
jaccard_coefficient(G[, ebunch]) Compute the Jaccard coefficient of all node pairs in
ebunch.
adamic_adar_index(G[, ebunch]) Compute the Adamic-Adar index of all node pairs in
ebunch.
preferential_attachment(G[, ebunch]) Compute the preferential attachment score of all node
pairs in ebunch.
cn_soundarajan_hopcroft(G[, ebunch, commu- Count the number of common neighbors of all node pairs
nity]) in ebunch
ra_index_soundarajan_hopcroft(G[, ebunch, Compute the resource allocation index of all node pairs
...]) in ebunch using community information.
within_inter_cluster(G[, ebunch, delta, ...]) Compute the ratio of within- and inter-cluster common
neighbors of all node pairs in ebunch.
common_neighbor_centrality(G[, ebunch, al- Return the CCPA score for each pair of nodes.
pha])
3.37.1 resource_allocation_index
resource_allocation_index(G, ebunch=None)
Compute the resource allocation index of all node pairs in ebunch.
Resource allocation index of u and v is defined as
∑ 1
|Γ(w)|
w∈Γ(u)∩Γ(v)
where Γ(u) denotes the set of neighbors of u.

Parameters
G
[graph] A NetworkX undirected graph.
ebunch
[iterable of node pairs, optional (default = None)] Resource allocation index will be computed
for each pair of nodes given in the iterable. The pairs must be given as 2-tuples (u, v) where u
and v are nodes in the graph. If ebunch is None then all non-existent edges in the graph will
be used. Default value: None.
3.37. Link Prediction 525

Returns
piter
[iterator] An iterator of 3-tuples in the form (u, v, p) where (u, v) is a pair of nodes and p is
their resource allocation index.
References
[1]
Examples
>>> preds = nx.resource_allocation_index(G, [(0, 1), (2, 3)])
>>> for u, v, p in preds:
... print(f"({u}, {v}) -> {p:.8f}")
(0, 1) -> 0.75000000
(2, 3) -> 0.75000000
3.37.2 jaccard_coefficient
jaccard_coefficient(G, ebunch=None)
Compute the Jaccard coefficient of all node pairs in ebunch.
Jaccard coefficient of nodes u and v is defined as
|Γ(u) ∩ Γ(v)|
|Γ(u) ∪ Γ(v)|

Parameters
G
ebunch
[iterable of node pairs, optional (default = None)] Jaccard coefficient will be computed for
each pair of nodes given in the iterable. The pairs must be given as 2-tuples (u, v) where u and
v are nodes in the graph. If ebunch is None then all non-existent edges in the graph will be
used. Default value: None.
Returns
piter
their Jaccard coefficient.

References
[1]
Examples
>>> preds = nx.jaccard_coefficient(G, [(0, 1), (2, 3)])
... print(f"({u}, {v}) -> {p:.8f}")
(0, 1) -> 0.60000000
(2, 3) -> 0.60000000
3.37.3 adamic_adar_index
adamic_adar_index(G, ebunch=None)
Compute the Adamic-Adar index of all node pairs in ebunch.
Adamic-Adar index of u and v is defined as
∑ 1
log |Γ(w)|
w∈Γ(u)∩Γ(v)
where Γ(u) denotes the set of neighbors of u. This index leads to zero-division for nodes only connected via
self-loops. It is intended to be used when no self-loops are present.
Parameters
G
[graph] NetworkX undirected graph.
ebunch
[iterable of node pairs, optional (default = None)] Adamic-Adar index will be computed for
Returns
piter
their Adamic-Adar index.
References
[1]

Examples
>>> preds = nx.adamic_adar_index(G, [(0, 1), (2, 3)])
... print(f"({u}, {v}) -> {p:.8f}")
(0, 1) -> 2.16404256
(2, 3) -> 2.16404256
3.37.4 preferential_attachment
preferential_attachment(G, ebunch=None)
Compute the preferential attachment score of all node pairs in ebunch.
Preferential attachment score of u and v is defined as
|Γ(u)||Γ(v)|

Parameters
G
ebunch
[iterable of node pairs, optional (default = None)] Preferential attachment score will be com-
puted for each pair of nodes given in the iterable. The pairs must be given as 2-tuples (u, v)
where u and v are nodes in the graph. If ebunch is None then all non-existent edges in the
graph will be used. Default value: None.
Returns
piter
their preferential attachment score.
References
[1]
Examples
>>> preds = nx.preferential_attachment(G, [(0, 1), (2, 3)])
... print(f"({u}, {v}) -> {p}")
(0, 1) -> 16
(2, 3) -> 16

3.37.5 cn_soundarajan_hopcroft
cn_soundarajan_hopcroft(G, ebunch=None, community='community')
Count the number of common neighbors of all node pairs in ebunch

using community information.
For two nodes u and v, this function computes the number of common neighbors and bonus one for each common
neighbor belonging to the same community as u and v. Mathematically,
∑
|Γ(u) ∩ Γ(v)| + f (w)
w∈Γ(u)∩Γ(v)
where f (w) equals 1 if w belongs to the same community as u and v or 0 otherwise and Γ(u) denotes the set of
neighbors of u.
Parameters
G
ebunch
[iterable of node pairs, optional (default = None)] The score will be computed for each pair of
nodes given in the iterable. The pairs must be given as 2-tuples (u, v) where u and v are nodes
in the graph. If ebunch is None then all non-existent edges in the graph will be used. Default
value: None.
community
[string, optional (default = ‘community’)] Nodes attribute name containing the community
information. G[u][community] identifies which community u belongs to. Each node belongs
to at most one community. Default value: ‘community’.
Returns
piter
their score.
References
[1]
Examples
>>> G.nodes[0]["community"] = 0
>>> preds = nx.cn_soundarajan_hopcroft(G, [(0, 2)])
... print(f"({u}, {v}) -> {p}")
(0, 2) -> 2

3.37.6 ra_index_soundarajan_hopcroft
ra_index_soundarajan_hopcroft(G, ebunch=None, community='community')

Compute the resource allocation index of all node pairs in ebunch using community information.
For two nodes u and v, this function computes the resource allocation index considering only common neighbors
belonging to the same community as u and v. Mathematically,
∑ f (w)
|Γ(w)|
w∈Γ(u)∩Γ(v)
where f (w) equals 1 if w belongs to the same community as u and v or 0 otherwise and Γ(u) denotes the set of
neighbors of u.
Parameters
G
ebunch
[iterable of node pairs, optional (default = None)] The score will be computed for each pair of
nodes given in the iterable. The pairs must be given as 2-tuples (u, v) where u and v are nodes
in the graph. If ebunch is None then all non-existent edges in the graph will be used. Default
value: None.
community
Returns
piter
their score.
References
[1]
Examples
>>> G = nx.Graph()
>>> G.add_edges_from([(0, 1), (0, 2), (1, 3), (2, 3)])
>>> preds = nx.ra_index_soundarajan_hopcroft(G, [(0, 3)])
... print(f"({u}, {v}) -> {p:.8f}")
(0, 3) -> 0.50000000

3.37.7 within_inter_cluster
within_inter_cluster(G, ebunch=None, delta=0.001, community='community')

Compute the ratio of within- and inter-cluster common neighbors of all node pairs in ebunch.
For two nodes u and v, if a common neighbor w belongs to the same community as them, w is considered as
within-cluster common neighbor of u and v. Otherwise, it is considered as inter-cluster common neighbor of u
and v. The ratio between the size of the set of within- and inter-cluster common neighbors is defined as the WIC
measure. [1]
Parameters
G
ebunch
[iterable of node pairs, optional (default = None)] The WIC measure will be computed for
delta
[float, optional (default = 0.001)] Value to prevent division by zero in case there is no inter-
cluster common neighbor between two nodes. See [1] for details. Default value: 0.001.
community
Returns
piter
their WIC measure.
References
[1]
Examples
>>> G = nx.Graph()
>>> G.add_edges_from([(0, 1), (0, 2), (0, 3), (1, 4), (2, 4), (3, 4)])
>>> preds = nx.within_inter_cluster(G, [(0, 4)])
... print(f"({u}, {v}) -> {p:.8f}")
(0, 4) -> 1.99800200
>>> preds = nx.within_inter_cluster(G, [(0, 4)], delta=0.5)
... print(f"({u}, {v}) -> {p:.8f}")
(0, 4) -> 1.33333333

3.37.8 common_neighbor_centrality
common_neighbor_centrality(G, ebunch=None, alpha=0.8)

Return the CCPA score for each pair of nodes.
Compute the Common Neighbor and Centrality based Parameterized Algorithm(CCPA) score of all node pairs in
ebunch.
CCPA score of u and v is defined as
N
α · (|Γ(u)∩Γ(v)|) + (1 − α) ·
duv
where Γ(u) denotes the set of neighbors of u, Γ(v) denotes the set of neighbors of v, α is parameter varies between
[0,1], N denotes total number of nodes in the Graph and duv denotes shortest distance between u and v.
This algorithm is based on two vital properties of nodes, namely the number of common neighbors and their
centrality. Common neighbor refers to the common nodes between two nodes. Centrality refers to the prestige that
a node enjoys in a network.
See also:
common_neighbors()
Parameters
G
ebunch
[iterable of node pairs, optional (default = None)] Preferential attachment score will be com-
puted for each pair of nodes given in the iterable. The pairs must be given as 2-tuples (u, v)
where u and v are nodes in the graph. If ebunch is None then all non-existent edges in the
graph will be used. Default value: None.
alpha
[Parameter defined for participation of Common Neighbor] and Centrality Algorithm share.
Values for alpha should normally be between 0 and 1. Default value set to 0.8 because author
found better performance at 0.8 for all the dataset. Default value: 0.8
Returns
piter
their Common Neighbor and Centrality based Parameterized Algorithm(CCPA) score.
References
[1]

Examples
>>> preds = nx.common_neighbor_centrality(G, [(0, 1), (2, 3)])
... print(f"({u}, {v}) -> {p}")
(0, 1) -> 3.4000000000000004
(2, 3) -> 3.4000000000000004
3.38 Lowest Common Ancestor
Algorithms for finding the lowest common ancestor of trees and DAGs.
all_pairs_lowest_common_ancestor(G[, Return the lowest common ancestor of all pairs or the pro-
pairs]) vided pairs
tree_all_pairs_lowest_common_ancestor(G[,Yield the lowest common ancestor for sets of pairs in a
...]) tree.
lowest_common_ancestor(G, node1, node2[, ...]) Compute the lowest common ancestor of the given pair
of nodes.
3.38.1 all_pairs_lowest_common_ancestor
all_pairs_lowest_common_ancestor(G, pairs=None)
Return the lowest common ancestor of all pairs or the provided pairs
Parameters
G
[NetworkX directed graph]
pairs
[iterable of pairs of nodes, optional (default: all pairs)] The pairs of nodes of interest. If None,
will find the LCA of all pairs of nodes.
Yields
((node1, node2), lca)
[2-tuple] Where lca is least common ancestor of node1 and node2. Note that for the default
case, the order of the node pair is not considered, e.g. you will not get both (a, b) and (b,
a)
Raises
If G is null.
NetworkXError
If G is not a DAG.
See also:
lowest_common_ancestor
3.38. Lowest Common Ancestor 533

Notes
Only defined on non-null directed acyclic graphs.
Examples
The default behavior is to yield the lowest common ancestor for all possible combinations of nodes in G, including
self-pairings:
>>> G = nx.DiGraph([(0, 1), (0, 3), (1, 2)])

>>> dict(nx.all_pairs_lowest_common_ancestor(G))
{(0, 0): 0, (0, 1): 0, (0, 3): 0, (0, 2): 0, (1, 1): 1, (1, 3): 0, (1, 2): 1, (3,␣
,→3): 3, (3, 2): 0, (2, 2): 2}
The pairs argument can be used to limit the output to only the specified node pairings:
>>> dict(nx.all_pairs_lowest_common_ancestor(G, pairs=[(1, 2), (2, 3)]))

{(1, 2): 1, (2, 3): 0}
3.38.2 tree_all_pairs_lowest_common_ancestor
tree_all_pairs_lowest_common_ancestor(G, root=None, pairs=None)

Yield the lowest common ancestor for sets of pairs in a tree.
Parameters
G
[NetworkX directed graph (must be a tree)]
root
[node, optional (default: None)] The root of the subtree to operate on. If None, assume the
entire graph has exactly one source and use that.
pairs
[iterable or iterator of pairs of nodes, optional (default: None)] The pairs of interest. If None,
Defaults to all pairs of nodes under root that have a lowest common ancestor.
Returns
lcas
[generator of tuples ((u, v), lca) where u and v are nodes] in pairs and lca is their
lowest common ancestor.
See also:
all_pairs_lowest_common_ancestor
similar routine for general DAGs
lowest_common_ancestor
just a single pair for general DAGs

Notes
Only defined on non-null trees represented with directed edges from parents to children. Uses Tarjan’s off-line
lowest-common-ancestors algorithm. Runs in time O(4 × (V + E + P )) time, where 4 is the largest value of the
inverse Ackermann function likely to ever come up in actual use, and P is the number of pairs requested (or V 2 if
all are needed).
Tarjan, R. E. (1979), “Applications of path compression on balanced trees”, Journal of the ACM 26 (4): 690-715,
doi:10.1145/322154.322161.
Examples
>>> import pprint

>>> G = nx.DiGraph([(1, 3), (2, 4), (1, 2)])
>>> pprint.pprint(dict(nx.tree_all_pairs_lowest_common_ancestor(G)))
{(1, 1): 1,
(2, 1): 1,
(2, 2): 2,
(3, 1): 1,
(3, 2): 1,
(3, 3): 3,
(3, 4): 1,
(4, 1): 1,
(4, 2): 2,
(4, 4): 4}
We can also use pairs argument to specify the pairs of nodes for which we want to compute lowest common
ancestors. Here is an example:
>>> dict(nx.tree_all_pairs_lowest_common_ancestor(G, pairs=[(1, 4), (2, 3)]))

{(2, 3): 1, (1, 4): 1}
3.38.3 lowest_common_ancestor
lowest_common_ancestor(G, node1, node2, default=None)

Compute the lowest common ancestor of the given pair of nodes.
Parameters
G
[NetworkX directed graph]
node1, node2
[nodes in the graph.]
default
[object] Returned if no common ancestor between node1 and node2
Returns
The lowest common ancestor of node1 and node2,
or default if they have no common ancestors.
See also:
all_pairs_lowest_common_ancestor
3.38. Lowest Common Ancestor 535

Examples
>>> nx.add_path(G, (0, 1, 2, 3))
>>> nx.add_path(G, (0, 4, 3))
>>> nx.lowest_common_ancestor(G, 2, 4)
0
3.39 Matching
Functions for computing and verifying matchings in a graph.
is_matching(G, matching) Return True if matching is a valid matching of G

is_maximal_matching(G, matching) Return True if matching is a maximal matching of G
is_perfect_matching(G, matching) Return True if matching is a perfect matching for G
maximal_matching(G) Find a maximal matching in the graph.
max_weight_matching(G[, maxcardinality, Compute a maximum-weighted matching of G.
weight])
min_weight_matching(G[, weight]) Computing a minimum-weight maximal matching of G.
3.39.1 is_matching
is_matching(G, matching)
Return True if matching is a valid matching of G
A matching in a graph is a set of edges in which no two distinct edges share a common endpoint. Each node is
incident to at most one edge in the matching. The edges are said to be independent.
Parameters
G
[NetworkX graph]
matching
[dict or set] A dictionary or set representing a matching. If a dictionary, it must have match-
ing[u] == v and matching[v] == u for each edge (u, v) in the matching. If a
set, it must have elements of the form (u, v), where (u, v) is an edge in the matching.
Returns
bool
Whether the given set or dictionary represents a valid matching in the graph.
Raises
NetworkXError
If the proposed matching has an edge to a node not in G. Or if the matching is not a collection
of 2-tuple edges.

Examples
>>> G = nx.Graph([(1, 2), (1, 3), (2, 3), (2, 4), (3, 5), (4, 5)])
>>> nx.is_maximal_matching(G, {1: 3, 2: 4}) # using dict to represent matching
True
>>> nx.is_matching(G, {(1, 3), (2, 4)}) # using set to represent matching
True
3.39.2 is_maximal_matching
is_maximal_matching(G, matching)
Return True if matching is a maximal matching of G
A maximal matching in a graph is a matching in which adding any edge would cause the set to no longer be a valid
matching.
Parameters
G
[NetworkX graph]
matching
Returns
bool
Whether the given set or dictionary represents a valid maximal matching in the graph.
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (2, 3), (3, 4), (3, 5)])
>>> nx.is_maximal_matching(G, {(1, 2), (3, 4)})
True
3.39.3 is_perfect_matching
is_perfect_matching(G, matching)
Return True if matching is a perfect matching for G
A perfect matching in a graph is a matching in which exactly one edge is incident upon each vertex.
Parameters
G
[NetworkX graph]
matching
3.39. Matching 537

Returns
bool
Whether the given set or dictionary represents a valid perfect matching in the graph.
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (2, 3), (2, 4), (3, 5), (4, 5), (4, 6)])
>>> my_match = {1: 2, 3: 5, 4: 6}
>>> nx.is_perfect_matching(G, my_match)
True
3.39.4 maximal_matching
maximal_matching(G)
Find a maximal matching in the graph.
A matching is a subset of edges in which no node occurs more than once. A maximal matching cannot add more
edges and still be a matching.
Parameters
G
Returns
matching
[set] A maximal matching of the graph.
Notes
The algorithm greedily selects a maximal matching M of the graph G (i.e. no superset of M exists). It runs in
O(|E|) time.
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (2, 3), (2, 4), (3, 5), (4, 5)])
>>> sorted(nx.maximal_matching(G))
[(1, 2), (3, 5)]
3.39.5 max_weight_matching
max_weight_matching(G, maxcardinality=False, weight='weight')

Compute a maximum-weighted matching of G.
A matching is a subset of edges in which no node occurs more than once. The weight of a matching is the sum of
the weights of its edges. A maximal matching cannot add more edges and still be a matching. The cardinality of a
matching is the number of matched edges.
Parameters

G
maxcardinality: bool, optional (default=False)
If maxcardinality is True, compute the maximum-cardinality matching with maximum weight
among all maximum-cardinality matchings.
weight: string, optional (default=’weight’)
Edge data key corresponding to the edge weight. If key not found, uses 1 as weight.
Returns
matching
[set] A maximal matching of the graph.
Notes
If G has edges with weight attributes the edge data are used as weight values else the weights are assumed to be 1.
This function takes time O(number_of_nodes ** 3).
If all edge weights are integers, the algorithm uses only integer computations. If floating point weights are used,
the algorithm could return a slightly suboptimal matching due to numeric precision errors.
This method is based on the “blossom” method for finding augmenting paths and the “primal-dual” method for
finding a matching of maximum weight, both methods invented by Jack Edmonds [1].
Bipartite graphs can also be matched using the functions present in networkx.algorithms.bipartite.
matching.
References
[1]
Examples
>>> G = nx.Graph()
>>> edges = [(1, 2, 6), (1, 3, 2), (2, 3, 1), (2, 4, 7), (3, 5, 9), (4, 5, 3)]
>>> G.add_weighted_edges_from(edges)
>>> sorted(nx.max_weight_matching(G))
[(2, 4), (5, 3)]
3.39.6 min_weight_matching
min_weight_matching(G, weight='weight')
Computing a minimum-weight maximal matching of G.
Use the maximum-weight algorithm with edge weights subtracted from the maximum weight of all edges.
A matching is a subset of edges in which no node occurs more than once. The weight of a matching is the sum of
the weights of its edges. A maximal matching cannot add more edges and still be a matching. The cardinality of a
matching is the number of matched edges.
This method replaces the edge weights with 1 plus the maximum edge weight minus the original edge weight.
new_weight = (max_weight + 1) - edge_weight
3.39. Matching 539

then runs max_weight_matching() with the new weights. The max weight matching with these new weights
corresponds to the min weight matching using the original weights. Adding 1 to the max edge weight keeps all edge
weights positive and as integers if they started as integers.
You might worry that adding 1 to each weight would make the algorithm favor matchings with more edges. But we
use the parameter maxcardinality=True in max_weight_matching to ensure that the number of edges
in the competing matchings are the same and thus the optimum does not change due to changes in the number of
edges.
Read the documentation of max_weight_matching for more information.
Parameters
G
Edge data key corresponding to the edge weight. If key not found, uses 1 as weight.
Returns
matching
[set] A minimal weight matching of the graph.
See also:
max_weight_matching
3.40 Minors
Subpackages related to graph-minor problems.

In graph theory, an undirected graph H is called a minor of the graph G if H can be formed from G by deleting edges and
vertices and by contracting edges [1].
3.40.1 References
contracted_edge(G, edge[, self_loops, copy]) Returns the graph that results from contracting the speci-
fied edge.
contracted_nodes(G, u, v[, self_loops, copy]) Returns the graph that results from contracting u and v.
identified_nodes(G, u, v[, self_loops, copy]) Returns the graph that results from contracting u and v.
equivalence_classes(iterable, relation) Returns equivalence classes of relation when applied
to iterable.
quotient_graph(G, partition[, ...]) Returns the quotient graph of G under the specified equiv-
alence relation on nodes.

3.40.2 contracted_edge
contracted_edge(G, edge, self_loops=True, copy=True)

Returns the graph that results from contracting the specified edge.
Edge contraction identifies the two endpoints of the edge as a single node incident to any edge that was incident to
the original two nodes. A graph that results from edge contraction is called a minor of the original graph.
Parameters
G
[NetworkX graph] The graph whose edge will be contracted.
edge
[tuple] Must be a pair of nodes in G.
self_loops
[Boolean] If this is True, any edges (including edge) joining the endpoints of edge in G
become self-loops on the new node in the returned graph.
copy
[Boolean (default True)] If this is True, a the contraction will be performed on a copy of G,
otherwise the contraction will happen in place.
Returns
Networkx graph
A new graph object of the same type as G (leaving G unmodified) with endpoints of edge
identified in a single node. The right node of edge will be merged into the left one, so only
the left one will appear in the returned graph.
Raises
ValueError
If edge is not an edge in G.
See also:
contracted_nodes
quotient_graph
Examples
Attempting to contract two nonadjacent nodes yields an error:
>>> nx.contracted_edge(G, (1, 3))
...
ValueError: Edge (1, 3) does not exist in graph G; cannot contract it
Contracting two adjacent nodes in the cycle graph on n nodes yields the cycle graph on n - 1 nodes:
>>> C5 = nx.cycle_graph(5)
>>> C4 = nx.cycle_graph(4)
>>> M = nx.contracted_edge(C5, (0, 1), self_loops=False)
>>> nx.is_isomorphic(M, C4)
True
3.40. Minors 541

3.40.3 contracted_nodes
contracted_nodes(G, u, v, self_loops=True, copy=True)

Returns the graph that results from contracting u and v.
Node contraction identifies the two nodes as a single node incident to any edge that was incident to the original two
nodes.
Parameters
G
[NetworkX graph] The graph whose nodes will be contracted.
u, v
[nodes] Must be nodes in G.
self_loops
[Boolean] If this is True, any edges joining u and v in G become self-loops on the new node
in the returned graph.
copy
[Boolean] If this is True (default True), make a copy of G and return that instead of directly
changing G.
Returns
Networkx graph
If Copy is True, A new graph object of the same type as G (leaving G unmodified) with u and
v identified in a single node. The right node v will be merged into the node u, so only u will
appear in the returned graph. If copy is False, Modifies G with u and v identified in a single
node. The right node v will be merged into the node u, so only u will appear in the returned
graph.
See also:
contracted_edge
quotient_graph
Notes
For multigraphs, the edge keys for the realigned edges may not be the same as the edge keys for the old edges. This
is natural because edge keys are unique only within each pair of nodes.
For non-multigraphs where u and v are adjacent to a third node w, the edge (v, w) will be contracted into the edge
(u, w) with its attributes stored into a “contraction” attribute.
This function is also available as identified_nodes.

Examples
Contracting two nonadjacent nodes of the cycle graph on four nodes C_4 yields the path graph (ignoring parallel
edges):
>>> M = nx.contracted_nodes(G, 1, 3)
>>> P3 = nx.path_graph(3)
>>> nx.is_isomorphic(M, P3)
True
>>> G = nx.MultiGraph(P3)
>>> M.edges
MultiEdgeView([(0, 1, 0), (0, 1, 1)])
>>> G = nx.Graph([(1, 2), (2, 2)])

>>> H = nx.contracted_nodes(G, 1, 2, self_loops=False)
>>> list(H.nodes())
[1]
>>> list(H.edges())
[(1, 1)]
3.40.4 identified_nodes
identified_nodes(G, u, v, self_loops=True, copy=True)

Returns the graph that results from contracting u and v.
Node contraction identifies the two nodes as a single node incident to any edge that was incident to the original two
nodes.
Parameters
G
[NetworkX graph] The graph whose nodes will be contracted.
u, v
[nodes] Must be nodes in G.
self_loops
[Boolean] If this is True, any edges joining u and v in G become self-loops on the new node
copy
[Boolean] If this is True (default True), make a copy of G and return that instead of directly
changing G.
Returns
Networkx graph
If Copy is True, A new graph object of the same type as G (leaving G unmodified) with u and
v identified in a single node. The right node v will be merged into the node u, so only u will
appear in the returned graph. If copy is False, Modifies G with u and v identified in a single
node. The right node v will be merged into the node u, so only u will appear in the returned
graph.
See also:
3.40. Minors 543

contracted_edge
quotient_graph
Notes
For multigraphs, the edge keys for the realigned edges may not be the same as the edge keys for the old edges. This
is natural because edge keys are unique only within each pair of nodes.
For non-multigraphs where u and v are adjacent to a third node w, the edge (v, w) will be contracted into the edge
(u, w) with its attributes stored into a “contraction” attribute.
This function is also available as identified_nodes.
Examples
Contracting two nonadjacent nodes of the cycle graph on four nodes C_4 yields the path graph (ignoring parallel
edges):
>>> P3 = nx.path_graph(3)
>>> nx.is_isomorphic(M, P3)
True
>>> G = nx.MultiGraph(P3)
>>> M.edges
MultiEdgeView([(0, 1, 0), (0, 1, 1)])
>>> G = nx.Graph([(1, 2), (2, 2)])

>>> H = nx.contracted_nodes(G, 1, 2, self_loops=False)
>>> list(H.nodes())
[1]
>>> list(H.edges())
[(1, 1)]
3.40.5 equivalence_classes
equivalence_classes(iterable, relation)
Returns equivalence classes of relation when applied to iterable.
The equivalence classes, or blocks, consist of objects from iterable which are all equivalent. They are defined
to be equivalent if the relation function returns True when passed any two objects from that class, and False
otherwise. To define an equivalence relation the function must be reflexive, symmetric and transitive.
Parameters
iterable
[list, tuple, or set] An iterable of elements/nodes.
relation
[function] A Boolean-valued function that implements an equivalence relation (reflexive, sym-
metric, transitive binary relation) on the elements of iterable - it must take two elements
and return True if they are related, or False if not.

Returns
set of frozensets
A set of frozensets representing the partition induced by the equivalence relation function
relation on the elements of iterable. Each member set in the return set represents an
equivalence class, or block, of the partition.
Duplicate elements will be ignored so it makes the most sense for iterable to be a set.
Notes
This function does not check that relation represents an equivalence relation. You can check that your equiv-
alence classes provide a partition using is_partition.
Examples
Let X be the set of integers from 0 to 9, and consider an equivalence relation R on X of congruence modulo 3: this
means that two integers x and y in X are equivalent under R if they leave the same remainder when divided by 3,
i.e. (x - y) mod 3 = 0.
The equivalence classes of this relation are {0, 3, 6, 9}, {1, 4, 7}, {2, 5, 8}: 0, 3, 6, 9 are all
divisible by 3 and leave zero remainder; 1, 4, 7 leave remainder 1; while 2, 5 and 8 leave remainder 2. We can
see this by calling equivalence_classes with X and a function implementation of R.
>>> X = set(range(10))
>>> def mod3(x, y): return (x - y) % 3 == 0
>>> equivalence_classes(X, mod3)
{frozenset({1, 4, 7}), frozenset({8, 2, 5}), frozenset({0, 9, 3, 6})}
3.40.6 quotient_graph
quotient_graph(G, partition, edge_relation=None, node_data=None, edge_data=None, relabel=False,

create_using=None)
Returns the quotient graph of G under the specified equivalence relation on nodes.
Parameters
G
[NetworkX graph] The graph for which to return the quotient graph with the specified node
relation.
partition
[function, or dict or list of lists, tuples or sets] If a function, this function must represent an
equivalence relation on the nodes of G. It must take two arguments u and v and return True
exactly when u and v are in the same equivalence class. The equivalence classes form the nodes
If a dict of lists/tuples/sets, the keys can be any meaningful block labels, but the values must
be the block lists/tuples/sets (one list/tuple/set per block), and the blocks must form a valid
partition of the nodes of the graph. That is, each node must be in exactly one block of the
partition.
If a list of sets, the list must form a valid partition of the nodes of the graph. That is, each node
must be in exactly one block of the partition.
3.40. Minors 545

edge_relation
[Boolean function with two arguments] This function must represent an edge relation on the
blocks of the partition of G. It must take two arguments, B and C, each one a set of nodes,
and return True exactly when there should be an edge joining block B to block C in the returned
graph.
If edge_relation is not specified, it is assumed to be the following relation. Block B is
related to block C if and only if some node in B is adjacent to some node in C, according to
the edge set of G.
edge_data
[function] This function takes two arguments, B and C, each one a set of nodes, and must return
a dictionary representing the edge data attributes to set on the edge joining B and C, should
there be an edge joining B and C in the quotient graph (if no such edge occurs in the quotient
graph as determined by edge_relation, then the output of this function is ignored).
If the quotient graph would be a multigraph, this function is not applied, since the edge data
from each edge in the graph G appears in the edges of the quotient graph.
node_data
[function] This function takes one argument, B, a set of nodes in G, and must return a dictionary
representing the node data attributes to set on the node representing B in the quotient graph.
If None, the following node attributes will be set:
• ‘graph’, the subgraph of the graph G that this block represents,
• ‘nnodes’, the number of nodes in this block,
• ‘nedges’, the number of edges within this block,
• ‘density’, the density of the subgraph of G that this block represents.
relabel
[bool] If True, relabel the nodes of the quotient graph to be nonnegative integers. Otherwise,
the nodes are identified with frozenset instances representing the blocks given in par-
tition.
create_using
Returns
NetworkX graph
The quotient graph of G under the equivalence relation specified by partition. If the
partition were given as a list of set instances and relabel is False, each node will be a
frozenset corresponding to the same set.
Raises
NetworkXException
If the given partition is not a valid partition of the nodes of G.

References
[1]
Examples
The quotient graph of the complete bipartite graph under the “same neighbors” equivalence relation is K_2. Under
this relation, two nodes are equivalent if they are not adjacent but have the same neighbor set.
>>> same_neighbors = lambda u, v: (
... u not in G[v] and v not in G[u] and G[u] == G[v]
... )
>>> Q = nx.quotient_graph(G, same_neighbors)
>>> K2 = nx.complete_graph(2)
>>> nx.is_isomorphic(Q, K2)
True
The quotient graph of a directed graph under the “same strongly connected component” equivalence relation is the
condensation of the graph (see condensation()). This example comes from the Wikipedia article `Strongly
connected component`_.
>>> edges = [
... "ab",
... "be",
... "bf",
... "bc",
... "cg",
... "cd",
... "dc",
... "dh",
... "ea",
... "ef",
... "fg",
... "gf",
... "hd",
... "hf",
... ]
>>> G.add_edges_from(tuple(x) for x in edges)
>>> components = list(nx.strongly_connected_components(G))
>>> sorted(sorted(component) for component in components)
[['a', 'b', 'e'], ['c', 'd', 'h'], ['f', 'g']]
>>>
>>> C = nx.condensation(G, components)
>>> component_of = C.graph["mapping"]
>>> same_component = lambda u, v: component_of[u] == component_of[v]
>>> Q = nx.quotient_graph(G, same_component)
>>> nx.is_isomorphic(C, Q)
True
Node identification can be represented as the quotient of a graph under the equivalence relation that places the two
nodes in one block and each other node in its own singleton block.
>>> K24 = nx.complete_bipartite_graph(2, 4)

>>> K34 = nx.complete_bipartite_graph(3, 4)
3.40. Minors 547


>>> C = nx.contracted_nodes(K34, 1, 2)
>>> nodes = {1, 2}
>>> is_contracted = lambda u, v: u in nodes and v in nodes
>>> Q = nx.quotient_graph(K34, is_contracted)
>>> nx.is_isomorphic(Q, C)
True
>>> nx.is_isomorphic(Q, K24)
True
The blockmodeling technique described in [1] can be implemented as a quotient graph.
>>> partition = [{0, 1}, {2, 3}, {4, 5}]
>>> M = nx.quotient_graph(G, partition, relabel=True)
>>> list(M.edges())
[(0, 1), (1, 2)]
Here is the sample example but using partition as a dict of block sets.
>>> partition = {0: {0, 1}, 2: {2, 3}, 4: {4, 5}}
>>> M = nx.quotient_graph(G, partition, relabel=True)
>>> list(M.edges())
[(0, 1), (1, 2)]
Partitions can be represented in various ways:

0. a list/tuple/set of block lists/tuples/sets
1. a dict with block labels as keys and blocks lists/tuples/sets as values
2. a dict with block lists/tuples/sets as keys and block labels as values
3. a function from nodes in the original iterable to block labels
4. an equivalence relation function on the target iterable
As quotient_graph is designed to accept partitions represented as (0), (1) or (4) only, the equiv-
alence_classes function can be used to get the partitions in the right form, in order to call quo-
tient_graph.
3.41 Maximal independent set
Algorithm to find a maximal (not maximum) independent set.
maximal_independent_set(G[, nodes, seed]) Returns a random maximal independent set guaranteed to

contain a given set of nodes.

3.41.1 maximal_independent_set
maximal_independent_set(G, nodes=None, seed=None)

Returns a random maximal independent set guaranteed to contain a given set of nodes.
An independent set is a set of nodes such that the subgraph of G induced by these nodes contains no edges. A max-
imal independent set is an independent set such that it is not possible to add a new node and still get an independent
set.
Parameters
G
[NetworkX graph]
nodes
[list or iterable] Nodes that must be part of the independent set. This set of nodes must be
independent.
seed
Randomness.
Returns
indep_nodes
[list] List of nodes that are part of a maximal independent set.
Raises
NetworkXUnfeasible
If the nodes in the provided list are not part of the graph or do not form an independent set,
an exception is raised.
If G is directed.
Notes
This algorithm does not solve the maximum independent set problem.
Examples
>>> nx.maximal_independent_set(G)
[4, 0, 2]
>>> nx.maximal_independent_set(G, [1])
[1, 3]
3.41. Maximal independent set 549

3.42 non-randomness
Computation of graph non-randomness
non_randomness(G[, k, weight]) Compute the non-randomness of graph G.
3.42.1 non_randomness
non_randomness(G, k=None, weight='weight')

Compute the non-randomness of graph G.
The first returned value nr is the sum of non-randomness values of all edges within the graph (where the non-
randomness of an edge tends to be small when the two nodes linked by that edge are from two different commu-
nities).
The second computed value nr_rd is a relative measure that indicates to what extent graph G is different from
random graphs in terms of probability. When it is close to 0, the graph tends to be more likely generated by an
Erdos Renyi model.
Parameters
G
[NetworkX graph] Graph must be symmetric, connected, and without self-loops.
k
[int] The number of communities in G. If k is not set, the function will use a default community
detection algorithm to set it.
weight
ical value used as a weight. If None, then each edge has weight 1, i.e., the graph is binary.
Returns
non-randomness
[(float, float) tuple] Non-randomness, Relative non-randomness w.r.t. Erdos Renyi random
graphs.
Raises
NetworkXException
if the input graph is not connected.
NetworkXError
if the input graph contains self-loops.
Notes
This computes Eq. (4.4) and (4.5) in Ref. [1].

If a weight field is passed, this algorithm will use the eigenvalues of the weighted adjacency matrix to compute Eq.
(4.4) and (4.5).

References
[1]
Examples
>>> nr, nr_rd = nx.non_randomness(G, 2)
>>> nr, nr_rd = nx.non_randomness(G, 2, 'weight')
3.43 Moral
Function for computing the moral graph of a directed graph.
moral_graph(G) Return the Moral Graph
3.43.1 moral_graph
moral_graph(G)
Return the Moral Graph
Returns the moralized graph of a given directed graph.
Parameters
G
[NetworkX graph] Directed graph
Returns
H
[NetworkX graph] The undirected moralized graph of G
Raises
If G is undirected.
Notes
A moral graph is an undirected graph H = (V, E) generated from a directed Graph, where if a node has more than
one parent node, edges between these parent nodes are inserted and all directed edges become undirected.
https://en.wikipedia.org/wiki/Moral_graph
3.43. Moral 551

References
[1]
Examples
>>> G = nx.DiGraph([(1, 2), (2, 3), (2, 5), (3, 4), (4, 3)])
>>> G_moral = nx.moral_graph(G)
>>> G_moral.edges()
EdgeView([(1, 2), (2, 3), (2, 5), (2, 4), (3, 4)])
3.44 Node Classification
This module provides the functions for node classification problem.

The functions in this module are not imported into the top level networkx namespace. You can access these functions
by importing the networkx.algorithms.node_classification modules, then accessing the functions as
attributes of node_classification. For example:
>>> from networkx.algorithms import node_classification

>>> G.edges()
EdgeView([(0, 1), (1, 2), (2, 3)])
>>> G.nodes[0]["label"] = "A"
>>> G.nodes[3]["label"] = "B"
>>> node_classification.harmonic_function(G)
['A', 'A', 'B', 'B']
3.44.1 References
Zhu, X., Ghahramani, Z., & Lafferty, J. (2003, August). Semi-supervised learning using gaussian fields and harmonic
functions. In ICML (Vol. 3, pp. 912-919).
harmonic_function(G[, max_iter, label_name]) Node classification by Harmonic function

local_and_global_consistency(G[, alpha, Node classification by Local and Global Consistency
...])
3.44.2 harmonic_function
harmonic_function(G, max_iter=30, label_name='label')

Node classification by Harmonic function
Function for computing Harmonic function algorithm by Zhu et al.
Parameters
G
[NetworkX Graph]
max_iter
[int] maximum number of iterations allowed

label_name
[string] name of target labels to predict
Returns
predicted
[list] List of length len(G) with the predicted labels for each node.
Raises
NetworkXError
If no nodes in G have attribute label_name.
References
Zhu, X., Ghahramani, Z., & Lafferty, J. (2003, August). Semi-supervised learning using gaussian fields and har-
monic functions. In ICML (Vol. 3, pp. 912-919).
Examples

>>> G.nodes(data=True)
NodeDataView({0: {'label': 'A'}, 1: {}, 2: {}, 3: {'label': 'B'}})
>>> G.edges()
EdgeView([(0, 1), (1, 2), (2, 3)])
>>> predicted = node_classification.harmonic_function(G)
>>> predicted
['A', 'A', 'B', 'B']
3.44.3 local_and_global_consistency
local_and_global_consistency(G, alpha=0.99, max_iter=30, label_name='label')

Node classification by Local and Global Consistency
Function for computing Local and global consistency algorithm by Zhou et al.
Parameters
G
[NetworkX Graph]
alpha
[float] Clamping factor
max_iter
[int] Maximum number of iterations allowed
label_name
[string] Name of target labels to predict
Returns
predicted
[list] List of length len(G) with the predicted labels for each node.
3.44. Node Classification 553

Raises
NetworkXError
If no nodes in G have attribute label_name.
References
Zhou, D., Bousquet, O., Lal, T. N., Weston, J., & Schölkopf, B. (2004). Learning with local and global consistency.
Advances in neural information processing systems, 16(16), 321-328.
Examples

>>> G.nodes(data=True)
NodeDataView({0: {'label': 'A'}, 1: {}, 2: {}, 3: {'label': 'B'}})
>>> G.edges()
EdgeView([(0, 1), (1, 2), (2, 3)])
>>> predicted = node_classification.local_and_global_consistency(G)
>>> predicted
['A', 'A', 'B', 'B']
3.45 Operators
Unary operations on graphs
complement(G) Returns the graph complement of G.

reverse(G[, copy]) Returns the reverse directed graph of G.
3.45.1 complement
complement(G)
Returns the graph complement of G.
Parameters
G
Returns
GC
[A new graph.]

Notes
Note that complement does not create self-loops and also does not produce parallel edges for MultiGraphs.
Graph, node, and edge data are not propagated to the new graph.
Examples
>>> G = nx.Graph([(1, 2), (1, 3), (2, 3), (3, 4), (3, 5)])
>>> G_complement = nx.complement(G)
>>> G_complement.edges() # This shows the edges of the complemented graph
EdgeView([(1, 4), (1, 5), (2, 4), (2, 5), (4, 5)])
3.45.2 reverse
reverse(G, copy=True)
Returns the reverse directed graph of G.
Parameters
G
[directed graph] A NetworkX directed graph
copy
[bool] If True, then a new graph is returned. If False, then the graph is reversed in place.
Returns
H
[directed graph] The reversed G.
Raises
NetworkXError
If graph is undirected.
Examples
>>> G = nx.DiGraph([(1, 2), (1, 3), (2, 3), (3, 4), (3, 5)])
>>> G_reversed = nx.reverse(G)
>>> G_reversed.edges()
OutEdgeView([(2, 1), (3, 1), (3, 2), (4, 3), (5, 3)])
Operations on graphs including union, intersection, difference.
3.45. Operators 555

compose(G, H) Compose graph G with H by combining nodes and edges

into a single graph.
union(G, H[, rename]) Combine graphs G and H.
disjoint_union(G, H) Combine graphs G and H.
intersection(G, H) Returns a new graph that contains only the nodes and the
edges that exist in both G and H.
difference(G, H) Returns a new graph that contains the edges that exist in
G but not in H.
symmetric_difference(G, H) Returns new graph with edges that exist in either G or H
but not both.
full_join(G, H[, rename]) Returns the full join of graphs G and H.
3.45.3 compose
compose(G, H)
Compose graph G with H by combining nodes and edges into a single graph.
The node sets and edges sets do not need to be disjoint.
Composing preserves the attributes of nodes and edges. Attribute values from H take precedent over attribute
values from G.
Parameters
G, H
Returns
C: A new graph with the same type as G
See also:
update()
union
disjoint_union
Notes
It is recommended that G and H be either both directed or both undirected.

For MultiGraphs, the edges are identified by incident nodes AND edge-key. This can cause surprises (i.e., edge
(1, 2) may or may not be the same in two graphs) if you use MultiGraph without keeping track of edge keys.
If combining the attributes of common nodes is not desired, consider union(), which raises an exception for name
collisions.

Examples
>>> G = nx.Graph([(0, 1), (0, 2)])

>>> H = nx.Graph([(0, 1), (1, 2)])
>>> R = nx.compose(G, H)
>>> R.nodes
NodeView((0, 1, 2))
>>> R.edges
EdgeView([(0, 1), (0, 2), (1, 2)])
By default, the attributes from H take precedent over attributes from G. If you prefer another way of combining
attributes, you can update them after the compose operation:
>>> G = nx.Graph([(0, 1, {'weight': 2.0}), (3, 0, {'weight': 100.0})])

>>> H = nx.Graph([(0, 1, {'weight': 10.0}), (1, 2, {'weight': -1.0})])
>>> nx.set_node_attributes(G, {0: 'dark', 1: 'light', 3: 'black'}, name='color')
>>> nx.set_node_attributes(H, {0: 'green', 1: 'orange', 2: 'yellow'}, name='color
,→')
>>> GcomposeH = nx.compose(G, H)
Normally, color attribute values of nodes of GcomposeH come from H. We can workaround this as follows:
>>> node_data = {n: G.nodes[n]['color'] + " " + H.nodes[n]['color'] for n in G.

,→nodes & H.nodes}
>>> nx.set_node_attributes(GcomposeH, node_data, 'color')

>>> print(GcomposeH.nodes[0]['color'])
dark green
>>> print(GcomposeH.nodes[3]['color'])
black
Similarly, we can update edge attributes after the compose operation in a way we prefer:
>>> edge_data = {e: G.edges[e]['weight'] * H.edges[e]['weight'] for e in G.edges &

,→ H.edges}
>>> nx.set_edge_attributes(GcomposeH, edge_data, 'weight')

>>> print(GcomposeH.edges[(0, 1)]['weight'])
20.0
>>> print(GcomposeH.edges[(3, 0)]['weight'])

100.0
3.45.4 union
union(G, H, rename=())
Combine graphs G and H. The names of nodes must be unique.
A name collision between the graphs will raise an exception.
A renaming facility is provided to avoid name collisions.
Parameters
G, H
3.45. Operators 557

rename
[iterable , optional] Node names of G and H can be changed by specifying the tuple
rename=(‘G-‘,’H-’) (for example). Node “u” in G is then renamed “G-u” and “v” in H is re-
named “H-v”.
Returns
U
[A union graph with the same type as G.]
See also:
compose
update()
disjoint_union
Notes
To combine graphs that have common nodes, consider compose(G, H) or the method, Graph.update().
disjoint_union() is similar to union() except that it avoids name clashes by relabeling the nodes with sequential
integers.
Edge and node attributes are propagated from G and H to the union graph. Graph attributes are also propagated,
but if they are present in both G and H, then the value from H is used.
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (1, 2)])

>>> H = nx.Graph([(0, 1), (0, 3), (1, 3), (1, 2)])
>>> U = nx.union(G, H, rename=("G", "H"))
>>> U.nodes
NodeView(('G0', 'G1', 'G2', 'H0', 'H1', 'H3', 'H2'))
>>> U.edges
EdgeView([('G0', 'G1'), ('G0', 'G2'), ('G1', 'G2'), ('H0', 'H1'), ('H0', 'H3'), (
,→'H1', 'H3'), ('H1', 'H2')])
3.45.5 disjoint_union
disjoint_union(G, H)
Combine graphs G and H. The nodes are assumed to be unique (disjoint).
This algorithm automatically relabels nodes to avoid name collisions.
Parameters
G,H
Returns
U
[A union graph with the same type as G.]
See also:

union
compose
update()
Notes
A new graph is created, of the same class as G. It is recommended that G and H be either both directed or both
undirected.
The nodes of G are relabeled 0 to len(G)-1, and the nodes of H are relabeled len(G) to len(G)+len(H)-1.
Renumbering forces G and H to be disjoint, so no exception is ever raised for a name collision. To preserve the
check for common nodes, use union().
Edge and node attributes are propagated from G and H to the union graph. Graph attributes are also propagated,
but if they are present in both G and H, then the value from H is used.
To combine graphs that have common nodes, consider compose(G, H) or the method, Graph.update().
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (1, 2)])

>>> H = nx.Graph([(0, 3), (1, 2), (2, 3)])
>>> G.nodes[0]["key1"] = 5
>>> H.nodes[0]["key2"] = 10
>>> U = nx.disjoint_union(G, H)
>>> U.nodes(data=True)
NodeDataView({0: {'key1': 5}, 1: {}, 2: {}, 3: {'key2': 10}, 4: {}, 5: {}, 6: {}})
>>> U.edges
EdgeView([(0, 1), (0, 2), (1, 2), (3, 4), (4, 6), (5, 6)])
3.45.6 intersection
intersection(G, H)
Returns a new graph that contains only the nodes and the edges that exist in both G and H.
Parameters
G,H
[graph] A NetworkX graph. G and H can have different node sets but must be both graphs or
both multigraphs.
Returns
GH
[A new graph with the same type as G.]
Raises
NetworkXError
If one is a MultiGraph and the other one is a graph.
3.45. Operators 559

Notes
Attributes from the graph, nodes, and edges are not copied to the new graph. If you want a new graph of the
intersection of G and H with the attributes (including edge data) from G use remove_nodes_from() as follows
>>> R = G.copy()
>>> R.remove_nodes_from(n for n in G if n not in H)
>>> R.remove_edges_from(e for e in G.edges if e not in H.edges)
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (1, 2)])

>>> H = nx.Graph([(0, 3), (1, 2), (2, 3)])
>>> R = nx.intersection(G, H)
>>> R.nodes
NodeView((0, 1, 2))
>>> R.edges
EdgeView([(1, 2)])
3.45.7 difference
difference(G, H)
Returns a new graph that contains the edges that exist in G but not in H.
The node sets of H and G must be the same.
Parameters
G,H
[graph] A NetworkX graph. G and H must have the same node sets.
Returns
D
Notes
Attributes from the graph, nodes, and edges are not copied to the new graph. If you want a new graph of the
difference of G and H with the attributes (including edge data) from G use remove_nodes_from() as follows:
>>> R = G.copy()
>>> R.remove_nodes_from(n for n in G if n in H)

Examples
>>> G = nx.Graph([(0, 1), (0, 2), (1, 2), (1, 3)])

>>> H = nx.Graph([(0, 1), (1, 2), (0, 3)])
>>> R = nx.difference(G, H)
>>> R.nodes
NodeView((0, 1, 2, 3))
>>> R.edges
EdgeView([(0, 2), (1, 3)])
3.45.8 symmetric_difference
symmetric_difference(G, H)
Returns new graph with edges that exist in either G or H but not both.
The node sets of H and G must be the same.
Parameters
G,H
[graph] A NetworkX graph. G and H must have the same node sets.
Returns
D
Notes
Attributes from the graph, nodes, and edges are not copied to the new graph.
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (1, 2), (1, 3)])

>>> H = nx.Graph([(0, 1), (1, 2), (0, 3)])
>>> R = nx.symmetric_difference(G, H)
>>> R.nodes
NodeView((0, 1, 2, 3))
>>> R.edges
EdgeView([(0, 2), (0, 3), (1, 3)])
3.45.9 full_join
full_join(G, H, rename=(None, None))

Returns the full join of graphs G and H.
Full join is the union of G and H in which all edges between G and H are added. The node sets of G and H must
be disjoint, otherwise an exception is raised.
Parameters
G, H
3.45. Operators 561

rename
[tuple , default=(None, None)] Node names of G and H can be changed by specifying the
tuple rename=(‘G-‘,’H-’) (for example). Node “u” in G is then renamed “G-u” and “v” in H is
renamed “H-v”.
Returns
U
[The full join graph with the same type as G.]
See also:
union
disjoint_union
Notes
It is recommended that G and H be either both directed or both undirected.

If G is directed, then edges from G to H are added as well as from H to G.
Note that full_join() does not produce parallel edges for MultiGraphs.
The full join operation of graphs G and H is the same as getting their complement, performing a disjoint union,
and finally getting the complement of the resulting graph.
Graph, edge, and node attributes are propagated from G and H to the union graph. If a graph attribute is present
in both G and H the value from H is used.
Examples
>>> G = nx.Graph([(0, 1), (0, 2)])

>>> H = nx.Graph([(3, 4)])
>>> R = nx.full_join(G, H, rename=("G", "H"))
>>> R.nodes
NodeView(('G0', 'G1', 'G2', 'H3', 'H4'))
>>> R.edges
EdgeView([('G0', 'G1'), ('G0', 'G2'), ('G0', 'H3'), ('G0', 'H4'), ('G1', 'H3'), (
,→'G1', 'H4'), ('G2', 'H3'), ('G2', 'H4'), ('H3', 'H4')])
Operations on many graphs.
compose_all(graphs) Returns the composition of all graphs.

union_all(graphs[, rename]) Returns the union of all graphs.
disjoint_union_all(graphs) Returns the disjoint union of all graphs.
intersection_all(graphs) Returns a new graph that contains only the nodes and the
edges that exist in all graphs.

3.45.10 compose_all
compose_all(graphs)
Returns the composition of all graphs.
Composition is the simple union of the node sets and edge sets. The node sets of the supplied graphs need not be
disjoint.
Parameters
graphs
[iterable] Iterable of NetworkX graphs
Returns
C
[A graph with the same type as the first graph in list]
Raises
ValueError
If graphs is an empty list.
Notes
It is recommended that the supplied graphs be either all directed or all undirected.
Graph, edge, and node attributes are propagated to the union graph. If a graph attribute is present in multiple
graphs, then the value from the last graph in the list with that attribute is used.
3.45.11 union_all
union_all(graphs, rename=())
Returns the union of all graphs.
The graphs must be disjoint, otherwise an exception is raised.
Parameters
graphs
rename
[iterable , optional] Node names of graphs can be changed by specifying the tuple rename=(‘G-
‘,’H-’) (for example). Node “u” in G is then renamed “G-u” and “v” in H is renamed “H-v”.
Infinite generators (like itertools.count) are also supported.
Returns
U
[a graph with the same type as the first graph in list]
Raises
ValueError
See also:
3.45. Operators 563

union
disjoint_union_all
Notes
To force a disjoint union with node relabeling, use disjoint_union_all(G,H) or convert_node_labels_to integers().
3.45.12 disjoint_union_all
disjoint_union_all(graphs)
Returns the disjoint union of all graphs.
This operation forces distinct integer node labels starting with 0 for the first graph in the list and numbering con-
secutively.
Parameters
graphs
Returns
U
[A graph with the same type as the first graph in list]
Raises
ValueError
Notes
It is recommended that the graphs be either all directed or all undirected.

3.45.13 intersection_all
intersection_all(graphs)
Returns a new graph that contains only the nodes and the edges that exist in all graphs.
Parameters
graphs
Returns
R
[A new graph with the same type as the first graph in list]
Raises

ValueError
Notes
Attributes from the graph, nodes, and edges are not copied to the new graph.
Graph products.
cartesian_product(G, H) Returns the Cartesian product of G and H.

lexicographic_product(G, H) Returns the lexicographic product of G and H.
rooted_product(G, H, root) Return the rooted product of graphs G and H rooted at
root in H.
strong_product(G, H) Returns the strong product of G and H.
tensor_product(G, H) Returns the tensor product of G and H.
power(G, k) Returns the specified power of a graph.
corona_product(G, H) Returns the Corona product of G and H.
3.45.14 cartesian_product
cartesian_product(G, H)
Returns the Cartesian product of G and H.
The Cartesian product P of the graphs G and H has a node set that is the Cartesian product of the node sets,
V (P ) = V (G) × V (H). P has an edge ((u, v), (x, y)) if and only if either u is equal to x and both v and y are
adjacent in H or if v is equal to y and both u and x are adjacent in G.
Parameters
G, H: graphs
Networkx graphs.
Returns
P: NetworkX graph
The Cartesian product of G and H. P will be a multi-graph if either G or H is a multi-graph.
Will be a directed if G and H are directed, and undirected if G and H are undirected.
Raises
NetworkXError
If G and H are not both directed or both undirected.
Notes
Node attributes in P are two-tuple of the G and H node attributes. Missing attributes are assigned None.
3.45. Operators 565

Examples
>>> G = nx.Graph()
>>> H = nx.Graph()
>>> G.add_node(0, a1=True)
>>> H.add_node("a", a2="Spam")
>>> P = nx.cartesian_product(G, H)
>>> list(P)
[(0, 'a')]
Edge attributes and edge keys (for multigraphs) are also copied to the new product graph
3.45.15 lexicographic_product
lexicographic_product(G, H)
Returns the lexicographic product of G and H.
The lexicographical product P of the graphs G and H has a node set that is the Cartesian product of the node sets,
V (P ) = V (G) × V (H). P has an edge ((u, v), (x, y)) if and only if (u, v) is an edge in G or u == v and (x, y)
is an edge in H.
Parameters
G, H: graphs
Networkx graphs.
Returns
P: NetworkX graph
Raises
NetworkXError
Notes
Examples
>>> G = nx.Graph()
>>> H = nx.Graph()
>>> P = nx.lexicographic_product(G, H)
>>> list(P)
[(0, 'a')]

3.45.16 rooted_product
rooted_product(G, H, root)
Return the rooted product of graphs G and H rooted at root in H.
A new graph is constructed representing the rooted product of the inputted graphs, G and H, with a root in H. A
rooted product duplicates H for each nodes in G with the root of H corresponding to the node in G. Nodes are
renamed as the direct product of G and H. The result is a subgraph of the cartesian product.
Parameters
G,H
root
[node] A node in H
Returns
R
[The rooted product of G and H with a specified root in H]
Notes
The nodes of R are the Cartesian Product of the nodes of G and H. The nodes of G and H are not relabeled.
3.45.17 strong_product
strong_product(G, H)
Returns the strong product of G and H.
The strong product P of the graphs G and H has a node set that is the Cartesian product of the node sets, V (P ) =
V (G) × V (H). P has an edge ((u, v), (x, y)) if and only if u == v and (x, y) is an edge in H, or x == y and
(u, v) is an edge in G, or (u, v) is an edge in G and (x, y) is an edge in H.
Parameters
G, H: graphs
Networkx graphs.
Returns
P: NetworkX graph
Raises
NetworkXError
3.45. Operators 567

Notes
Examples
>>> G = nx.Graph()
>>> H = nx.Graph()
>>> P = nx.strong_product(G, H)
>>> list(P)
[(0, 'a')]
3.45.18 tensor_product
tensor_product(G, H)
Returns the tensor product of G and H.
The tensor product P of the graphs G and H has a node set that is the tensor product of the node sets, V (P ) =
V (G) × V (H). P has an edge ((u, v), (x, y)) if and only if (u, x) is an edge in G and (v, y) is an edge in H.
Tensor product is sometimes also referred to as the categorical product, direct product, cardinal product or con-
junction.
Parameters
G, H: graphs
Networkx graphs.
Returns
P: NetworkX graph
The tensor product of G and H. P will be a multi-graph if either G or H is a multi-graph, will
be a directed if G and H are directed, and undirected if G and H are undirected.
Raises
NetworkXError
Notes

Examples
>>> G = nx.Graph()
>>> H = nx.Graph()
>>> P = nx.tensor_product(G, H)
>>> list(P)
[(0, 'a')]
3.45.19 power
power(G, k)
Returns the specified power of a graph.
The k`thpowerof asimplegraph : math : `G, denoted Gk , is a graph on the same set of nodes in which two
distinct nodes u and v are adjacent in Gk if and only if the shortest path distance between u and v in G is at most
k.
Parameters
G
[graph] A NetworkX simple graph object.
k
[positive integer] The power to which to raise the graph G.
Returns
NetworkX simple graph
G to the power k.
Raises
ValueError
If the exponent k is not positive.
If G is not a simple graph.
Notes
This definition of “power graph” comes from Exercise 3.1.6 of Graph Theory by Bondy and Murty [1].
References
[1]
3.45. Operators 569

Examples
The number of edges will never decrease when taking successive powers:
>>> list(nx.power(G, 2).edges)
[(0, 1), (0, 2), (1, 2), (1, 3), (2, 3)]
>>> list(nx.power(G, 3).edges)
[(0, 1), (0, 2), (0, 3), (1, 2), (1, 3), (2, 3)]
The k`th power of a cycle graph on *n* nodes is the complete graph on *n*
nodes, if `k is at least n // 2:
>>> H = nx.complete_graph(5)
>>> nx.is_isomorphic(nx.power(G, 2), H)
True
>>> H = nx.complete_graph(8)
>>> nx.is_isomorphic(nx.power(G, 4), H)
True
3.45.20 corona_product
corona_product(G, H)
Returns the Corona product of G and H.
The corona product of G and H is the graph C = G ◦ H obtained by taking one copy of G, called the center
graph, |V (G)| copies of H, called the outer graph, and making the i-th vertex of G adjacent to every vertex of the
i-th copy of H, where 1≤i≤|V (G)|.
Parameters
G, H: NetworkX graphs
The graphs to take the carona product of. G is the center graph and H is the outer graph
Returns
C: NetworkX graph
The Corona product of G and H.
Raises
NetworkXError
References
[1] M. Tavakoli, F. Rahbarnia, and A. R. Ashrafi,

“Studying the corona product of graphs under some graph invariants,” Transactions on Combinatorics, vol.
3, no. 3, pp. 43–49, Sep. 2014, doi: 10.22108/toc.2014.5542.
[2] A. Faraji, “Corona Product in Graph Theory,” Ali Faraji, May 11, 2021.
https://blog.alifaraji.ir/math/graph-theory/corona-product.html (accessed Dec. 07, 2021).

Examples
>>> C = nx.corona_product(G, H)
>>> list(C)
[0, 1, 2, 3, (0, 0), (0, 1), (1, 0), (1, 1), (2, 0), (2, 1), (3, 0), (3, 1)]
>>> print(C)
Graph with 12 nodes and 16 edges
3.46 Planarity
check_planarity(G[, counterexample]) Check if a graph is planar and return a counterexample or

an embedding.
is_planar(G) Returns True if and only if G is planar.
PlanarEmbedding([incoming_graph_data]) Represents a planar graph with its planar embedding.
3.46.1 check_planarity
check_planarity(G, counterexample=False)
Check if a graph is planar and return a counterexample or an embedding.
A graph is planar iff it can be drawn in a plane without any edge intersections.
Parameters
G
[NetworkX graph]
counterexample
[bool] A Kuratowski subgraph (to proof non planarity) is only returned if set to true.
Returns
(is_planar, certificate)
[(bool, NetworkX graph) tuple] is_planar is true if the graph is planar. If the graph is planar
certificate is a PlanarEmbedding otherwise it is a Kuratowski subgraph.
See also:
is_planar
Check for planarity without creating a PlanarEmbedding or counterexample.
3.46. Planarity 571

Notes
A (combinatorial) embedding consists of cyclic orderings of the incident edges at each vertex. Given such an
embedding there are multiple approaches discussed in literature to drawing the graph (subject to various constraints,
e.g. integer coordinates), see e.g. [2].
The planarity check algorithm and extraction of the combinatorial embedding is based on the Left-Right Planarity
Test [1].
A counterexample is only generated if the corresponding parameter is set, because the complexity of the coun-
terexample generation is higher.
References
[1], [2]
Examples
>>> G = nx.Graph([(0, 1), (0, 2)])

>>> is_planar, P = nx.check_planarity(G)
>>> print(is_planar)
True
When G is planar, a PlanarEmbedding instance is returned:
>>> P.get_data()
{0: [1, 2], 1: [0], 2: [0]}
3.46.2 is_planar
is_planar(G)
Returns True if and only if G is planar.
A graph is planar iff it can be drawn in a plane without any edge intersections.
Parameters
G
[NetworkX graph]
Returns
bool
Whether the graph is planar.
See also:
check_planarity
Check if graph is planar and return a PlanarEmbedding instance if True.

Examples
>>> G = nx.Graph([(0, 1), (0, 2)])

>>> nx.is_planar(G)
True
>>> nx.is_planar(nx.complete_graph(5))
False
3.46.3 networkx.algorithms.planarity.PlanarEmbedding
class PlanarEmbedding(incoming_graph_data=None, **attr)

Represents a planar graph with its planar embedding.
The planar embedding is given by a combinatorial embedding.
Note: check_planarity is the preferred way to check if a graph is planar.
Neighbor ordering:
In comparison to a usual graph structure, the embedding also stores the order of all neighbors for every vertex. The
order of the neighbors can be given in clockwise (cw) direction or counterclockwise (ccw) direction. This order is
stored as edge attributes in the underlying directed graph. For the edge (u, v) the edge attribute ‘cw’ is set to the
neighbor of u that follows immediately after v in clockwise direction.
In order for a PlanarEmbedding to be valid it must fulfill multiple conditions. It is possible to check if these
conditions are fulfilled with the method check_structure(). The conditions are:
• Edges must go in both directions (because the edge attributes differ)
• Every edge must have a ‘cw’ and ‘ccw’ attribute which corresponds to a correct planar embedding.
• A node with non zero degree must have a node attribute ‘first_nbr’.
As long as a PlanarEmbedding is invalid only the following methods should be called:
• add_half_edge_ccw()
• add_half_edge_cw()
• connect_components()
• add_half_edge_first()
Even though the graph is a subclass of nx.DiGraph, it can still be used for algorithms that require undirected graphs,
because the method is_directed() is overridden. This is possible, because a valid PlanarGraph must have
edges in both directions.
Half edges:
In methods like add_half_edge_ccw the term “half-edge” is used, which is a term that is used in doubly
connected edge lists. It is used to emphasize that the edge is only in one direction and there exists another half-
edge in the opposite direction. While conventional edges always have two faces (including outer face) next to them,
it is possible to assign each half-edge exactly one face. For a half-edge (u, v) that is orientated such that u is below
v then the face that belongs to (u, v) is to the right of this half-edge.
See also:
is_planar
Preferred way to check if an existing graph is planar.
3.46. Planarity 573

check_planarity
A convenient way to create a PlanarEmbedding. If not planar, it returns a subgraph that shows this.
Examples
Create an embedding of a star graph (compare nx.star_graph(3)):
>>> G = nx.PlanarEmbedding()
>>> G.add_half_edge_cw(0, 1, None)
>>> G.add_half_edge_cw(0, 2, 1)
>>> G.add_half_edge_cw(0, 3, 2)
Alternatively the same embedding can also be defined in counterclockwise orientation. The following results in
exactly the same PlanarEmbedding:
>>> G = nx.PlanarEmbedding()
>>> G.add_half_edge_ccw(0, 1, None)
>>> G.add_half_edge_ccw(0, 3, 1)
>>> G.add_half_edge_ccw(0, 2, 3)
After creating a graph, it is possible to validate that the PlanarEmbedding object is correct:
>>> G.check_structure()
__init__(incoming_graph_data=None, **attr)
Parameters
incoming_graph_data
attr
[keyword arguments, optional (default= no attributes)] Attributes to add to graph as
key=value pairs.
See also:
convert

Examples

>>> e = [(1, 2), (2, 3), (3, 4)] # list of edges
>>> G = nx.Graph(e)

>>> G.graph
{'day': 'Friday'}
Methods
add_edge(u_of_edge, v_of_edge, **attr) Add an edge between u and v.

add_edges_from(ebunch_to_add, **attr) Add all the edges in ebunch_to_add.
add_half_edge_ccw(start_node, end_node, ...) Adds a half-edge from start_node to end_node.
add_half_edge_cw(start_node, end_node, ...) Adds a half-edge from start_node to end_node.
add_half_edge_first(start_node, end_node) The added half-edge is inserted at the first position in
the order.
add_node(node_for_adding, **attr) Add a single node node_for_adding and update
node attributes.
add_nodes_from(nodes_for_adding, **attr) Add multiple nodes.
add_weighted_edges_from(ebunch_to_add[, Add weighted edges in ebunch_to_add with spec-
weight]) ified weight attr
adjacency() Returns an iterator over (node, adjacency dict) tuples
for all nodes.
check_structure() Runs without exceptions if this object is valid.
clear() Remove all nodes and edges from the graph.
clear_edges() Remove all edges from the graph without altering
nodes.
connect_components(v, w) Adds half-edges for (v, w) and (w, v) at some position.
copy([as_view]) Returns a copy of the graph.
edge_subgraph(edges) Returns the subgraph induced by the specified edges.
get_data() Converts the adjacency structure into a better readable
structure.
get_edge_data(u, v[, default]) Returns the attribute dictionary associated with edge
(u, v).
has_edge(u, v) Returns True if the edge (u, v) is in the graph.
has_node(n) Returns True if the graph contains the node n.
has_predecessor(u, v) Returns True if node u has predecessor v.
has_successor(u, v) Returns True if node u has successor v.
is_directed() A valid PlanarEmbedding is undirected.
is_multigraph() Returns True if graph is a multigraph, False otherwise.
nbunch_iter([nbunch]) Returns an iterator over nodes contained in nbunch that
neighbors(n) Returns an iterator over successor nodes of n.
neighbors_cw_order(v) Generator for the neighbors of v in clockwise order.
next_face_half_edge(v, w) Returns the following half-edge left of a face.
continues on next page
3.46. Planarity 575

Table 1 – continued from previous page

number_of_edges([u, v]) Returns the number of edges between two nodes.
number_of_nodes() Returns the number of nodes in the graph.
order() Returns the number of nodes in the graph.
predecessors(n) Returns an iterator over predecessor nodes of n.
remove_edge(u, v) Remove the edge between u and v.
remove_edges_from(ebunch) Remove all edges specified in ebunch.
remove_node(n) Remove node n.
remove_nodes_from(nodes) Remove multiple nodes.
reverse([copy]) Returns the reverse of the graph.
set_data(data) Inserts edges according to given sorted neighbor list.
size([weight]) Returns the number of edges or total of all edge
weights.
subgraph(nodes) Returns a SubGraph view of the subgraph induced on
nodes.
successors(n) Returns an iterator over successor nodes of n.
to_directed([as_view]) Returns a directed representation of the graph.
to_directed_class() Returns the class to use for empty directed copies.
to_undirected([reciprocal, as_view]) Returns an undirected representation of the digraph.
to_undirected_class() Returns the class to use for empty undirected copies.
traverse_face(v, w[, mark_half_edges]) Returns nodes on the face that belong to the half-edge
(v, w).
update([edges, nodes]) Update the graph using nodes/edges/graphs as input.
PlanarEmbedding.add_edge
PlanarEmbedding.add_edge(u_of_edge, v_of_edge, **attr)

Edge attributes can be specified with keywords or by directly accessing the edge’s attribute dictionary. See
examples below.
Parameters
attr
arguments.
See also:
add_edges_from

Notes

Many NetworkX algorithms designed for weighted graphs use an edge attribute (by default weight) to hold
a numerical value.
Examples

>>> e = (1, 2)

>>> G[1][2].update({0: 5})
>>> G.edges[1, 2].update({0: 5})
PlanarEmbedding.add_edges_from
PlanarEmbedding.add_edges_from(ebunch_to_add, **attr)
Parameters
ebunch_to_add
data.
attr
arguments.
See also:
add_edge
add a single edge
3.46. Planarity 577

Notes
Adding the same edge twice has no effect but any edge data will be updated when each duplicate edge is
added.
Examples

>>> e = zip(range(0, 3), range(1, 4))

PlanarEmbedding.add_half_edge_ccw
PlanarEmbedding.add_half_edge_ccw(start_node, end_node, reference_neighbor)

Adds a half-edge from start_node to end_node.
The half-edge is added counter clockwise next to the existing half-edge (start_node, reference_neighbor).
Parameters
start_node
[node] Start node of inserted edge.
end_node
[node] End node of inserted edge.
reference_neighbor: node
End node of reference edge.
Raises
NetworkXException
If the reference_neighbor does not exist.
See also:
add_half_edge_cw
connect_components
add_half_edge_first

PlanarEmbedding.add_half_edge_cw
PlanarEmbedding.add_half_edge_cw(start_node, end_node, reference_neighbor)

Adds a half-edge from start_node to end_node.
The half-edge is added clockwise next to the existing half-edge (start_node, reference_neighbor).
Parameters
start_node
[node] Start node of inserted edge.
end_node
[node] End node of inserted edge.
reference_neighbor: node
End node of reference edge.
Raises
NetworkXException
If the reference_neighbor does not exist.
See also:
add_half_edge_ccw
connect_components
add_half_edge_first
PlanarEmbedding.add_half_edge_first
PlanarEmbedding.add_half_edge_first(start_node, end_node)
The added half-edge is inserted at the first position in the order.
Parameters
start_node
[node]
end_node
[node]
See also:
add_half_edge_ccw
add_half_edge_cw
connect_components
3.46. Planarity 579

PlanarEmbedding.add_node
PlanarEmbedding.add_node(node_for_adding, **attr)
Parameters
node_for_adding
attr
See also:
add_nodes_from
Notes
A hashable object is one that can be used as a key in a Python dictionary. This includes strings, numbers,
tuples of strings and numbers, etc.
On many platforms hashable items also include mutables such as NetworkX Graphs, though one should be
careful that the hash doesn’t change on mutables.
Examples

>>> G.add_node(1)
>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
>>> G.add_node(K3)
3

PlanarEmbedding.add_nodes_from
PlanarEmbedding.add_nodes_from(nodes_for_adding, **attr)
Add multiple nodes.
Parameters
nodes_for_adding
[iterable container] A container of nodes (list, dict, set, etc.). OR A container of (node,
attribute dict) tuples. Node attributes are updated using the attribute dict.
attr
[keyword arguments, optional (default= no attributes)] Update attributes for all nodes in
nodes. Node attributes specified in nodes as a tuple take precedence over attributes specified
via keyword arguments.

See also:
add_node
Examples

>>> K3 = nx.Graph([(0, 1), (1, 2), (2, 0)])
[0, 1, 2, 'H', 'e', 'l', 'o']


11
>>> H = nx.Graph()
11
PlanarEmbedding.add_weighted_edges_from
PlanarEmbedding.add_weighted_edges_from(ebunch_to_add, weight='weight', **attr)

Parameters
ebunch_to_add
weight
attr
edges.
See also:
add_edge
add a single edge
add_edges_from
add multiple edges
3.46. Planarity 581

Notes
Adding the same edge twice for Graph/DiGraph simply updates the edge data. For Multi-
Graph/MultiDiGraph, duplicate edges are stored.
Examples

PlanarEmbedding.adjacency
PlanarEmbedding.adjacency()
Returns
adj_iter
Examples

[(0, {1: {}}), (1, {0: {}, 2: {}}), (2, {1: {}, 3: {}}), (3, {2: {}})]
PlanarEmbedding.check_structure
PlanarEmbedding.check_structure()
Runs without exceptions if this object is valid.
Checks that the following properties are fulfilled:
• Edges go in both directions (because the edge attributes differ).
• Every edge has a ‘cw’ and ‘ccw’ attribute which corresponds to a correct planar embedding.
• A node with a degree larger than 0 has a node attribute ‘first_nbr’.
Running this method verifies that the underlying Graph must be planar.
Raises
NetworkXException
This exception is raised with a short explanation if the PlanarEmbedding is invalid.

PlanarEmbedding.clear
PlanarEmbedding.clear()
Examples

>>> G.clear()
>>> list(G.nodes)
[]
>>> list(G.edges)
[]
PlanarEmbedding.clear_edges
PlanarEmbedding.clear_edges()
Examples

>>> G.clear_edges()
>>> list(G.nodes)
[0, 1, 2, 3]
>>> list(G.edges)
[]
PlanarEmbedding.connect_components
PlanarEmbedding.connect_components(v, w)
Adds half-edges for (v, w) and (w, v) at some position.
This method should only be called if v and w are in different components, or it might break the embed-
ding. This especially means that if connect_components(v, w) is called it is not allowed to call
connect_components(w, v) afterwards. The neighbor orientations in both directions are all set cor-
rectly after the first call.
Parameters
v
[node]
w
[node]
See also:
add_half_edge_ccw
add_half_edge_cw
add_half_edge_first
3.46. Planarity 583

PlanarEmbedding.copy
PlanarEmbedding.copy(as_view=False)
The copy method by default returns an independent shallow copy of the graph and attributes. That is, if an
attribute is a container, that container is shared by the original an the copy. Use Python’s copy.deepcopy
for new containers.
Parameters
as_view
[bool, optional (default=False)] If True, the returned graph-view provides a read-only view
of the original graph without actually copying any data.
Returns
G
See also:
to_directed
Notes
All copies reproduce the graph structure, but data attributes may be handled in different ways. There are four
types of copies of a graph that people might want.
contain. The entire graph object is new so that changes in the copy do not affect the original object. (see
Python’s copy.deepcopy)
Data Reference (Shallow) – For a shallow copy the graph structure is copied but the edge, node and graph
attribute dicts are references to those in the original graph. This saves time and memory but could cause
confusion if you change an attribute in one graph and it changes the attribute in the other. NetworkX does
not provide this level of shallow copy.
attributes. That is, any attributes that are containers are shared between the new graph and the original. This
is exactly what dict.copy() provides. You can obtain this style copy using:
>>> H = G.copy()
>>> H = nx.Graph(G)
>>> H = G.__class__(G)
Fresh Data – For fresh data, the graph structure is copied while new empty data attribute dicts are created.
The resulting graph is independent of the original and it has no edge, node or graph attributes. Fresh copies
are not enabled. Instead use:
>>> H = G.__class__()

View – Inspired by dict-views, graph-views act like read-only versions of the original graph, providing a copy
of the original structure without requiring any memory for copying the information.
See the Python copy module for more information on shallow and deep copies, https://docs.python.org/3/
library/copy.html.
Examples

>>> H = G.copy()
PlanarEmbedding.edge_subgraph
PlanarEmbedding.edge_subgraph(edges)
Parameters
edges
Returns
G
Notes
The graph, edge, and node attributes in the returned subgraph view are references to the corresponding at-
tributes in the original graph. The view is read-only.
Examples
>>> H = G.edge_subgraph([(0, 1), (3, 4)])
>>> list(H.nodes)
[0, 1, 3, 4]
>>> list(H.edges)
[(0, 1), (3, 4)]
3.46. Planarity 585

PlanarEmbedding.get_data
PlanarEmbedding.get_data()
Converts the adjacency structure into a better readable structure.
Returns
embedding
[dict] A dict mapping all nodes to a list of neighbors sorted in clockwise order.
See also:
set_data
PlanarEmbedding.get_edge_data
PlanarEmbedding.get_edge_data(u, v, default=None)
Parameters
u, v
[nodes]
Returns
edge_dict
Examples

>>> G[0][1]
{}
>>> G[0][1]["weight"] = 7
>>> G[0][1]["weight"]
7
>>> G[1][0]["weight"]
7

{}
>>> e = (0, 1)
{}
0

PlanarEmbedding.has_edge
PlanarEmbedding.has_edge(u, v)
Parameters
u, v
Returns
edge_ind
Examples

True
>>> e = (0, 1)
True
>>> e = (0, 1, {"weight": 7})
True
True
True
PlanarEmbedding.has_node
PlanarEmbedding.has_node(n)
Identical to n in G
Parameters
n
[node]
3.46. Planarity 587

Examples

>>> G.has_node(0)
True
>>> 0 in G
True
PlanarEmbedding.has_predecessor
PlanarEmbedding.has_predecessor(u, v)
Returns True if node u has predecessor v.
This is true if graph has the edge u<-v.
PlanarEmbedding.has_successor
PlanarEmbedding.has_successor(u, v)
Returns True if node u has successor v.
This is true if graph has the edge u->v.
PlanarEmbedding.is_directed
PlanarEmbedding.is_directed()
A valid PlanarEmbedding is undirected.
All reverse edges are contained, i.e. for every existing half-edge (v, w) the half-edge in the opposite direction
(w, v) is also contained.
PlanarEmbedding.is_multigraph
PlanarEmbedding.is_multigraph()
Returns True if graph is a multigraph, False otherwise.
PlanarEmbedding.nbunch_iter
PlanarEmbedding.nbunch_iter(nbunch=None)
Parameters
nbunch
[single node, container, or all nodes (default= all nodes)] The view will only report edges
incident to these nodes.
Returns

niter
[iterator] An iterator over nodes in nbunch that are also in the graph. If nbunch is None,
iterate over all nodes in the graph.
Raises
NetworkXError
See also:
Graph.__iter__
Notes
When nbunch is an iterator, the returned iterator yields values directly from nbunch, becoming exhausted
when nbunch is exhausted.
To test whether nbunch is a single node, one can use “if nbunch in self:”, even after processing with this
routine.
If nbunch is not a node or a (possibly empty) sequence/iterator or None, a NetworkXError is raised. Also,
if any object in nbunch is not hashable, a NetworkXError is raised.
PlanarEmbedding.neighbors
PlanarEmbedding.neighbors(n)
Parameters
n
Raises
NetworkXError
See also:
predecessors
Notes
3.46. Planarity 589

PlanarEmbedding.neighbors_cw_order
PlanarEmbedding.neighbors_cw_order(v)
Generator for the neighbors of v in clockwise order.
Parameters
v
[node]
Yields
node
PlanarEmbedding.next_face_half_edge
PlanarEmbedding.next_face_half_edge(v, w)
Returns the following half-edge left of a face.
Parameters
v
[node]
w
[node]
Returns
half-edge
[tuple]
PlanarEmbedding.number_of_edges
PlanarEmbedding.number_of_edges(u=None, v=None)
Parameters
u, v
[nodes, optional (default=all edges)] If u and v are specified, return the number of edges
Returns
nedges
from u to v.
See also:
size

Examples
3
1
1
PlanarEmbedding.number_of_nodes
PlanarEmbedding.number_of_nodes()
Returns
nnodes
See also:
order
identical method
__len__
identical method
Examples

3
3.46. Planarity 591

PlanarEmbedding.order
PlanarEmbedding.order()
Returns
nnodes
See also:
number_of_nodes
identical method
__len__
identical method
Examples

>>> G.order()
3
PlanarEmbedding.predecessors
PlanarEmbedding.predecessors(n)
Parameters
n
Raises
NetworkXError
See also:
successors
PlanarEmbedding.remove_edge
PlanarEmbedding.remove_edge(u, v)
Parameters
u, v
Raises

NetworkXError
See also:
remove_edges_from
Examples
>>> G = nx.Graph() # or DiGraph, etc

>>> nx.add_path(G, [0, 1, 2, 3])
>>> e = (1, 2)
PlanarEmbedding.remove_edges_from
PlanarEmbedding.remove_edges_from(ebunch)
Parameters
See also:
remove_edge
Notes
Examples

>>> ebunch = [(1, 2), (2, 3)]
3.46. Planarity 593

PlanarEmbedding.remove_node
PlanarEmbedding.remove_node(n)
Remove node n.
Parameters
n
Raises
NetworkXError
See also:
remove_nodes_from
Examples

>>> list(G.edges)
[(0, 1), (1, 2)]
>>> list(G.edges)
[]
PlanarEmbedding.remove_nodes_from
PlanarEmbedding.remove_nodes_from(nodes)
Parameters
nodes
See also:
remove_node
Examples

>>> e
[0, 1, 2]
>>> list(G.nodes)
[]

PlanarEmbedding.reverse
PlanarEmbedding.reverse(copy=True)
Parameters
copy
PlanarEmbedding.set_data
PlanarEmbedding.set_data(data)
Inserts edges according to given sorted neighbor list.
The input format is the same as the output format of get_data().
Parameters
data
[dict] A dict mapping all nodes to a list of neighbors sorted in clockwise order.
See also:
get_data
PlanarEmbedding.size
PlanarEmbedding.size(weight=None)
Parameters
weight
Returns
size
If weight is None, returns an int. Otherwise a float (or more general numeric if the weights
are more general).
See also:
number_of_edges
3.46. Planarity 595

Examples

>>> G.size()
3

>>> G.size()
2
6.0
PlanarEmbedding.subgraph
PlanarEmbedding.subgraph(nodes)
Parameters
nodes
Returns
G
Notes
The graph, edge and node attributes are shared with the original graph. Changes to the graph structure is
ruled out by the view, but changes to attributes are reflected in the original graph.
For an inplace reduction of a graph to a subgraph you can remove nodes: G.remove_nodes_from([n for n in
G if n not in set(nodes)])
Subgraph views are sometimes NOT what you want. In most cases where you want to do more than simply
look at the induced edges, it makes more sense to just create the subgraph as its own graph with code like:

SG = G.__class__()
else:


Examples

>>> H = G.subgraph([0, 1, 2])
>>> list(H.edges)
[(0, 1), (1, 2)]
PlanarEmbedding.successors
PlanarEmbedding.successors(n)
Parameters
n
Raises
NetworkXError
See also:
predecessors
Notes
PlanarEmbedding.to_directed
PlanarEmbedding.to_directed(as_view=False)
Returns
G
3.46. Planarity 597

Notes
This returns a “deepcopy” of the edge, node, and graph attributes which attempts to completely copy all of
the data and references.
library/copy.html.
Warning: If you have subclassed Graph to use dict-like objects in the data structure, those changes do not
transfer to the DiGraph created by this method.
Examples

>>> list(H.edges)
[(0, 1), (1, 0)]

>>> list(H.edges)
[(0, 1)]
PlanarEmbedding.to_directed_class
PlanarEmbedding.to_directed_class()
Returns the class to use for empty directed copies.
If you subclass the base classes, use this to designate what directed class to use for to_directed() copies.
PlanarEmbedding.to_undirected
PlanarEmbedding.to_undirected(reciprocal=False, as_view=False)
Parameters
reciprocal
as_view
[bool (optional, default=False)] If True return an undirected view of the original directed
graph.
Returns
G
[Graph] An undirected graph with the same name and nodes and with edge (u, v, data) if
either (u, v, data) or (v, u, data) is in the digraph. If both edges exist in digraph and their

edge data is different, only one edge is created with an arbitrary choice of which edge data
to use. You must check and correct for this manually if desired.
See also:
Notes
If edges in both directions (u, v) and (v, u) exist in the graph, attributes for the new undirected edge will be a
combination of the attributes of the directed edges. The edge data is updated in the (arbitrary) order that the
edges are encountered. For more customized control of the edge attributes use add_edge().
This returns a “deepcopy” of the edge, node, and graph attributes which attempts to completely copy all of
the data and references.
This is in contrast to the similar G=DiGraph(D) which returns a shallow copy of the data.
library/copy.html.
Warning: If you have subclassed DiGraph to use dict-like objects in the data structure, those changes do not
transfer to the Graph created by this method.
Examples

>>> list(H.edges)
[(0, 1), (1, 0)]
>>> list(G2.edges)
[(0, 1)]
PlanarEmbedding.to_undirected_class
PlanarEmbedding.to_undirected_class()
Returns the class to use for empty undirected copies.
If you subclass the base classes, use this to designate what directed class to use for to_directed() copies.
PlanarEmbedding.traverse_face
PlanarEmbedding.traverse_face(v, w, mark_half_edges=None)
Returns nodes on the face that belong to the half-edge (v, w).
The face that is traversed lies to the right of the half-edge (in an orientation where v is below w).
Optionally it is possible to pass a set to which all encountered half edges are added. Before calling this method,
this set must not include any half-edges that belong to the face.
Parameters
v
[node] Start node of half-edge.
3.46. Planarity 599

w
[node] End node of half-edge.
mark_half_edges: set, optional
Set to which all encountered half-edges are added.
Returns
face
[list] A list of nodes that lie on this face.
PlanarEmbedding.update
PlanarEmbedding.update(edges=None, nodes=None)
Like dict.update, this method takes a graph as input, adding the graph’s nodes and edges to this graph. It can
also take two inputs: edges and nodes. Finally it can take either edges or nodes. To specify only nodes the
keyword nodes must be used.
The collections of edges and nodes are treated similarly to the add_edges_from/add_nodes_from methods.
When iterated, they should yield 2-tuples (u, v) or 3-tuples (u, v, datadict).
Parameters
edges
[Graph object, collection of edges, or None] The first parameter can be a graph or some
edges. If it has attributes nodes and edges, then it is taken to be a Graph-like object and
those attributes are used as collections of nodes and edges to be added to the graph. If the
first parameter does not have those attributes, it is treated as a collection of edges and added
to the graph. If the first argument is None, no edges are added.
nodes
[collection of nodes, or None] The second parameter is treated as a collection of nodes to
be added to the graph unless it is None. If edges is None and nodes is None an
See also:
add_edges_from
add_nodes_from
Notes
It you want to update the graph using an adjacency structure it is straightforward to obtain the edges/nodes
from adjacency. The following examples provide common cases, your adjacency may be slightly different and
require tweaks of these examples:
>>> adj = {1: {2, 3}, 2: {1, 3}, 3: {1, 2}}

>>> adj = {1: {2: 1.3, 3: 0.7}, 2: {1: 1.4}, 3: {1: 0.7}}
>>> e = [
... (u, v, {"weight": d})
... ]
>>> e = [
... (u, v, {"weight": d})
... ]

>>> pred = {1: {2, 3}, 2: {3}, 3: {3}}

>>> adj = {
... 1: {2: {0: {"weight": 1.3}, 1: {"weight": 1.2}}},
... 3: {2: {0: {"weight": 0.7}}},
... }
>>> e = [
... (u, v, ekey, d)
... ]
Examples
>>> edges = (
... (u, v, {"power": u * v})
... if u * v < 225
... )
3.46. Planarity 601

Attributes
adj Graph adjacency object holding the neighbors of each

node.
degree A DegreeView for the Graph as G.degree or
G.degree().
edges An OutEdgeView of the DiGraph as G.edges or
G.edges().
in_degree An InDegreeView for (node, in_degree) or in_degree
for single node.
in_edges A view of the in edges of the graph as G.in_edges or
G.in_edges().
name String identifier of the graph.
nodes A NodeView of the Graph as G.nodes or G.nodes().
out_degree An OutDegreeView for (node, out_degree)
out_edges An OutEdgeView of the DiGraph as G.edges or
G.edges().
pred Graph adjacency object holding the predecessors of
each node.
succ Graph adjacency object holding the successors of each
node.
PlanarEmbedding.adj
property PlanarEmbedding.adj
This object is a read-only dict-like structure with node keys and neighbor-dict values. The neighbor-dict is
keyed by neighbor to the edge-data-dict. So G.adj[3][2]['color'] = 'blue' sets the color of the
edge (3, 2) to "blue".
items():.
The neighbor information is also provided by subscripting the graph. So for nbr, foovalue in
G[node].data('foo', default=1): works.
PlanarEmbedding.degree
property PlanarEmbedding.degree
The node degree is the number of edges adjacent to the node. The weighted node degree is the sum of the
Parameters
nbunch

weight
[string or None, optional (default=None)] The name of an edge attribute that holds the nu-
merical value used as a weight. If None, then each edge has weight 1. The degree is the sum
of the edge weights adjacent to the node.
Returns
DiDegreeView or int
If multiple nodes are requested (the default), returns a DiDegreeView mapping nodes to
their degree. If a single node is requested, returns the degree of the node as an integer.
See also:
in_degree, out_degree
Examples
>>> G = nx.DiGraph() # or MultiDiGraph

>>> nx.add_path(G, [0, 1, 2, 3])
1
>>> list(G.degree([0, 1, 2]))
[(0, 1), (1, 2), (2, 2)]
PlanarEmbedding.edges
property PlanarEmbedding.edges
The OutEdgeView provides set-like operations on the edge-tuples as well as edge attribute lookup. When
provide set-like operations). Hence, G.edges[u, v]['color'] provides the value of the color attribute
for edge (u, v) while for (u, v, c) in G.edges.data('color', default='red'):
iterates through all the edges yielding the color attribute with default 'red' if no color attribute exists.
Parameters
nbunch
from these nodes.
data
ddict[data]). If True, return edge attribute dict in 3-tuple (u, v, ddict). If False, return
2-tuple (u, v).
default
Returns
edges
3.46. Planarity 603

See also:
in_edges, out_edges
Notes
Nodes in nbunch that are not in the graph will be (quietly) ignored. For directed graphs this returns the
out-edges.
Examples

>>> nx.add_path(G, [0, 1, 2])
[(0, 1), (1, 2), (2, 3)]
PlanarEmbedding.in_degree
property PlanarEmbedding.in_degree
An InDegreeView for (node, in_degree) or in_degree for single node.
The node in_degree is the number of edges pointing to the node. The weighted node degree is the sum of the
This object provides an iteration over (node, in_degree) as well as lookup for the degree for a single node.
Parameters
nbunch
weight
Returns
deg
[int] In-degree of the node
nd_iter

See also:
degree, out_degree
Examples
>>> nx.add_path(G, [0, 1, 2, 3])
0
[(0, 0), (1, 1), (2, 1)]
PlanarEmbedding.in_edges
property PlanarEmbedding.in_edges
in_edges(self, nbunch=None, data=False, default=None):
Parameters
nbunch
data
2-tuple (u, v).
default
Returns
in_edges
[InEdgeView or InEdgeDataView] A view of edge attributes, usually it iterates over (u,
v) or (u, v, d) tuples of edges, but can also be used for attribute lookup as edges[u,
v]['foo'].
Examples
..
>>> G.add_edge(1, 2, color='blue')

..
>>> G.in_edges()
..
3.46. Planarity 605

InEdgeView([(1, 2)])
>>> G.in_edges(nbunch=2)
..
InEdgeDataView([(1, 2)])
See also:
edges
PlanarEmbedding.name
property PlanarEmbedding.name
String identifier of the graph.
This graph attribute appears in the attribute dict G.graph keyed by the string "name". as well as an attribute
(technically a property) G.name. This is entirely user controlled.
PlanarEmbedding.nodes
property PlanarEmbedding.nodes
nodes(data='color', default=None) to return a NodeDataView which reports specific node
data but no set operations. It presents a dict-like interface as well with G.nodes.items() iterating over
(node, nodedata) 2-tuples and G.nodes[3]['foo'] providing the value of the foo attribute for
node 3. In addition, a view G.nodes.data('foo') provides a dict-like interface to the foo attribute
of each node. G.nodes.data('foo', default=1) provides a default for nodes that do not have
attribute foo.
Parameters
data
[string or bool, optional (default=False)] The node attribute returned in 2-tuple (n,
ddict[data]). If True, return entire node attribute dict as (n, ddict). If False, return just
the nodes n.
default
Returns
NodeView

Notes
Examples
>>> list(G.nodes)
[0, 1, 2]
>>> list(G)
[0, 1, 2]

[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
[(0, {'foo': 'bar'}), (1, {'time': '5pm'}), (2, {})]
[(0, 'bar'), (1, None), (2, None)]
[(0, 'bar'), (1, None), (2, None)]
[(0, None), (1, '5pm'), (2, None)]
[(0, None), (1, '5pm'), (2, None)]

If some of your nodes have an attribute and the rest are assumed to have a default attribute value you can
create a dictionary from node/attribute pairs using the default keyword argument to guarantee the value
is never None:
>>> G = nx.Graph()
>>> G.add_node(0)
{0: 1, 1: 2, 2: 3}
3.46. Planarity 607

PlanarEmbedding.out_degree
property PlanarEmbedding.out_degree
An OutDegreeView for (node, out_degree)
The node out_degree is the number of edges pointing out of the node. The weighted node degree is the sum
of the edge weights for edges incident to that node.
This object provides an iterator over (node, out_degree) as well as lookup for the degree for a single node.
Parameters
nbunch
weight
Returns
deg
[int] Out-degree of the node
nd_iter
See also:
degree, in_degree
Examples
>>> nx.add_path(G, [0, 1, 2, 3])
1
[(0, 1), (1, 1), (2, 1)]
PlanarEmbedding.out_edges
property PlanarEmbedding.out_edges
The OutEdgeView provides set-like operations on the edge-tuples as well as edge attribute lookup. When
provide set-like operations). Hence, G.edges[u, v]['color'] provides the value of the color attribute
for edge (u, v) while for (u, v, c) in G.edges.data('color', default='red'):
iterates through all the edges yielding the color attribute with default 'red' if no color attribute exists.
Parameters

nbunch
from these nodes.
data
2-tuple (u, v).
default
Returns
edges
See also:
in_edges, out_edges
Notes
Nodes in nbunch that are not in the graph will be (quietly) ignored. For directed graphs this returns the
out-edges.
Examples

>>> nx.add_path(G, [0, 1, 2])
[(0, 1), (1, 2), (2, 3)]
PlanarEmbedding.pred
property PlanarEmbedding.pred
Graph adjacency object holding the predecessors of each node.
keyed by neighbor to the edge-data-dict. So G.pred[2][3]['color'] = 'blue' sets the color of
the edge (3, 2) to "blue".
3.46. Planarity 609

Iterating over G.pred behaves like a dict. Useful idioms include for nbr, datadict in G.
pred[n].items():. A data-view not provided by dicts also exists: for nbr, foovalue in
G.pred[node].data('foo'): A default can be set via a default argument to the data method.
PlanarEmbedding.succ
property PlanarEmbedding.succ
keyed by neighbor to the edge-data-dict. So G.succ[3][2]['color'] = 'blue' sets the color of
the edge (3, 2) to "blue".
Iterating over G.succ behaves like a dict. Useful idioms include for nbr, datadict in G.
succ[n].items():. A data-view not provided by dicts also exists: for nbr, foovalue in G.
succ[node].data('foo'): and a default can be set via a default argument to the data method.
The neighbor information is also provided by subscripting the graph. So for nbr, foovalue in
G[node].data('foo', default=1): works.
For directed graphs, G.adj is identical to G.succ.
3.47 Planar Drawing
combinatorial_embedding_to_pos(embedding[, Assigns every node a (x, y) position based on the given

...]) embedding
3.47.1 combinatorial_embedding_to_pos
combinatorial_embedding_to_pos(embedding, fully_triangulate=False)
Assigns every node a (x, y) position based on the given embedding
The algorithm iteratively inserts nodes of the input graph in a certain order and rearranges previously inserted nodes
so that the planar drawing stays valid. This is done efficiently by only maintaining relative positions during the node
placements and calculating the absolute positions at the end. For more information see [1].
Parameters
embedding
[nx.PlanarEmbedding] This defines the order of the edges
fully_triangulate
[bool] If set to True the algorithm adds edges to a copy of the input embedding and makes it
chordal.
Returns
pos
[dict] Maps each node to a tuple that defines the (x, y) position

References
[1]
3.48 Graph Polynomials
Provides algorithms supporting the computation of graph polynomials.

Graph polynomials are polynomial-valued graph invariants that encode a wide variety of structural information. Examples
include the Tutte polynomial, chromatic polynomial, characteristic polynomial, and matching polynomial. An extensive
treatment is provided in [1].
For a simple example, the charpoly method can be used to compute the characteristic polynomial from the adjacency
matrix of a graph. Consider the complete graph K_4:
>>> import sympy

>>> x = sympy.Symbol("x")
>>> A = nx.adjacency_matrix(G)
>>> M = sympy.SparseMatrix(A.todense())
>>> M.charpoly(x).as_expr()
x**4 - 6*x**2 - 8*x - 3
tutte_polynomial(G) Returns the Tutte polynomial of G

chromatic_polynomial(G) Returns the chromatic polynomial of G
3.48.1 tutte_polynomial
tutte_polynomial(G)
Returns the Tutte polynomial of G
This function computes the Tutte polynomial via an iterative version of the deletion-contraction algorithm.
The Tutte polynomial T_G(x, y) is a fundamental graph polynomial invariant in two variables. It encodes a wide
array of information related to the edge-connectivity of a graph; “Many problems about graphs can be reduced to
problems of finding and evaluating the Tutte polynomial at certain values” [1]. In fact, every deletion-contraction-
expressible feature of a graph is a specialization of the Tutte polynomial [2] (see Notes for examples).
There are several equivalent definitions; here are three:
Def 1 (rank-nullity expansion): For G an undirected graph, n(G) the number of vertices of G, E the edge set of G,
V the vertex set of G, and c(A) the number of connected components of the graph with vertex set V and edge set
A [3]:
∑
TG (x, y) = (x − 1)c(A)−c(E) (y − 1)c(A)+|A|−n(G)
A∈E
Def 2 (spanning tree expansion): Let G be an undirected graph, T a spanning tree of G, and E the edge set of G.
Let E have an arbitrary strict linear order L. Let B_e be the unique minimal nonempty edge cut of E \ T ∪ e. An
edge e is internally active with respect to T and L if e is the least edge in B_e according to the linear order L. The
internal activity of T (denoted i(T)) is the number of edges in E \ T that are internally active with respect to T
and L. Let P_e be the unique path in T ∪ e whose source and target vertex are the same. An edge e is externally
3.48. Graph Polynomials 611

active with respect to T and L if e is the least edge in P_e according to the linear order L. The external activity of
T (denoted e(T)) is the number of edges in E \ T that are externally active with respect to T and L. Then [4] [5]:
∑
TG (x, y) = xi(T ) y e(T )
T a spanning tree of G
Def 3 (deletion-contraction recurrence): For G an undirected graph, G-e the graph obtained from G by deleting
edge e, G/e the graph obtained from G by contracting edge e, k(G) the number of cut-edges of G, and l(G) the
number of self-loops of G:
{
xk(G) y l(G) , if all edges are cut-edges or self-loops
TG (x, y) =
TG−e (x, y) + TG/e (x, y), otherwise, for an arbitrary edge e not a cut-edge or loop
Parameters
G
[NetworkX graph]
Returns
instance of sympy.core.add.Add
A Sympy expression representing the Tutte polynomial for G.
Notes
Some specializations of the Tutte polynomial:

• T_G(1, 1) counts the number of spanning trees of G
• T_G(1, 2) counts the number of connected spanning subgraphs of G
• T_G(2, 1) counts the number of spanning forests in G
• T_G(0, 2) counts the number of strong orientations of G
• T_G(2, 0) counts the number of acyclic orientations of G
Edge contraction is defined and deletion-contraction is introduced in [6]. Combinatorial meaning of the coefficients
is introduced in [7]. Universality, properties, and applications are discussed in [8].
Practically, up-front computation of the Tutte polynomial may be useful when users wish to repeatedly calculate
edge-connectivity-related information about one or more graphs.
References
[1], [2], [3], [4], [5], [6], [7], [8]
Examples
>>> C = nx.cycle_graph(5)
>>> nx.tutte_polynomial(C)
x**4 + x**3 + x**2 + x + y
>>> D = nx.diamond_graph()
>>> nx.tutte_polynomial(D)
x**3 + 2*x**2 + 2*x*y + x + y**2 + y

3.48.2 chromatic_polynomial
chromatic_polynomial(G)
Returns the chromatic polynomial of G
This function computes the chromatic polynomial via an iterative version of the deletion-contraction algorithm.
The chromatic polynomial X_G(x) is a fundamental graph polynomial invariant in one variable. Evaluating
X_G(k) for an natural number k enumerates the proper k-colorings of G.
There are several equivalent definitions; here are three:
Def 1 (explicit formula): For G an undirected graph, c(G) the number of connected components of G, E the edge
set of G, and G(S) the spanning subgraph of G with edge set S [1]:
∑
XG (x) = (−1)|S| xc(G(S))
S⊆E
Def 2 (interpolating polynomial): For G an undirected graph, n(G) the number of vertices of G, k_0 = 0,
and k_i the number of distinct ways to color the vertices of G with i unique colors (for i a natural number at
most n(G)), X_G(x) is the unique Lagrange interpolating polynomial of degree n(G) through the points (0,
k_0), (1, k_1), dots, (n(G), k_{n(G)}) [2].
Def 3 (chromatic recurrence): For G an undirected graph, G-e the graph obtained from G by deleting edge e, G/e
the graph obtained from G by contracting edge e, n(G) the number of vertices of G, and e(G) the number of
edges of G [3]:
{
xn(G) , if e(G) = 0
XG (x) =
XG−e (x) − XG/e (x), otherwise, for an arbitrary edge e
This formulation is also known as the Fundamental Reduction Theorem [4].

Parameters
G
[NetworkX graph]
Returns
instance of sympy.core.add.Add
A Sympy expression representing the chromatic polynomial for G.
Notes
Interpretation of the coefficients is discussed in [5]. Several special cases are listed in [2].
The chromatic polynomial is a specialization of the Tutte polynomial; in particular, X_G(x) = `T_G(x, 0)
[6].
The chromatic polynomial may take negative arguments, though evaluations may not have chromatic interpretations.
For instance, X_G(-1) enumerates the acyclic orientations of G [7].
3.48. Graph Polynomials 613

References
[1], [2], [3], [4], [5], [6], [7]
Examples
>>> C = nx.cycle_graph(5)
>>> nx.chromatic_polynomial(C)
x**5 - 5*x**4 + 10*x**3 - 10*x**2 + 4*x
>>> nx.chromatic_polynomial(G)
x**4 - 6*x**3 + 11*x**2 - 6*x
3.49 Reciprocity
Algorithms to calculate reciprocity in a directed graph.
reciprocity(G[, nodes]) Compute the reciprocity in a directed graph.

overall_reciprocity(G) Compute the reciprocity for the whole graph.
3.49.1 reciprocity
reciprocity(G, nodes=None)
Compute the reciprocity in a directed graph.
The reciprocity of a directed graph is defined as the ratio of the number of edges pointing in both directions to the
total number of edges in the graph. Formally, r = |(u, v) ∈ G|(v, u) ∈ G|/|(u, v) ∈ G|.
The reciprocity of a single node u is defined similarly, it is the ratio of the number of edges in both directions to
the total number of edges attached to node u.
Parameters
G
[graph] A networkx directed graph
nodes
[container of nodes, optional (default=whole graph)] Compute reciprocity for nodes in this
container.
Returns
out
[dictionary] Reciprocity keyed by node label.

Notes
The reciprocity is not defined for isolated nodes. In such cases this function will return None.
3.49.2 overall_reciprocity
overall_reciprocity(G)
Compute the reciprocity for the whole graph.
See the doc of reciprocity for the definition.
Parameters
G
3.50 Regular
Functions for computing and verifying regular graphs.
is_regular(G) Determines whether the graph G is a regular graph.

is_k_regular(G, k) Determines whether the graph G is a k-regular graph.
k_factor(G, k[, matching_weight]) Compute a k-factor of G
3.50.1 is_regular
is_regular(G)
Determines whether the graph G is a regular graph.
A regular graph is a graph where each vertex has the same degree. A regular digraph is a graph where the indegree
and outdegree of each vertex are equal.
Parameters
G
[NetworkX graph]
Returns
bool
Whether the given graph or digraph is regular.
Examples
>>> G = nx.DiGraph([(1, 2), (2, 3), (3, 4), (4, 1)])

>>> nx.is_regular(G)
True
3.50. Regular 615

3.50.2 is_k_regular
is_k_regular(G, k)
Determines whether the graph G is a k-regular graph.
A k-regular graph is a graph where each vertex has degree k.
Parameters
G
[NetworkX graph]
Returns
bool
Whether the given graph is k-regular.
Examples
>>> G = nx.Graph([(1, 2), (2, 3), (3, 4), (4, 1)])

>>> nx.is_k_regular(G, k=3)
False
3.50.3 k_factor
k_factor(G, k, matching_weight='weight')
Compute a k-factor of G
A k-factor of a graph is a spanning k-regular subgraph. A spanning k-regular subgraph of G is a subgraph that
contains each vertex of G and a subset of the edges of G such that each vertex has degree k.
Parameters
G
matching_weight: string, optional (default=’weight’)
Edge data key corresponding to the edge weight. Used for finding the max-weighted perfect
matching. If key not found, uses 1 as weight.
Returns
G2
[NetworkX graph] A k-factor of G
References
[1]

Examples
>>> G = nx.Graph([(1, 2), (2, 3), (3, 4), (4, 1)])

>>> G2 = nx.k_factor(G, k=1)
>>> G2.edges()
EdgeView([(1, 2), (3, 4)])
3.51 Rich Club
Functions for computing rich-club coefficients.
rich_club_coefficient(G[, normalized, Q, Returns the rich-club coefficient of the graph G.

seed])
3.51.1 rich_club_coefficient
rich_club_coefficient(G, normalized=True, Q=100, seed=None)

Returns the rich-club coefficient of the graph G.
For each degree k, the rich-club coefficient is the ratio of the number of actual to the number of potential edges for
nodes with degree greater than k:
2Ek
ϕ(k) =
Nk (Nk − 1)
where N_k is the number of nodes with degree larger than k, and E_k is the number of edges among those nodes.
Parameters
G
[NetworkX graph] Undirected graph with neither parallel edges nor self-loops.
normalized
[bool (optional)] Normalize using randomized network as in [1]
Q
[float (optional, default=100)] If normalized is True, perform Q * m double-edge swaps,
where m is the number of edges in G, to use as a null-model for normalization.
seed
Randomness.
Returns
rc
[dictionary] A dictionary, keyed by degree, with rich-club coefficient values.
3.51. Rich Club 617

Notes
The rich club definition and algorithm are found in [1]. This algorithm ignores any edge weights and is not defined
for directed graphs or graphs with parallel edges or self loops.
Estimates for appropriate values of Q are found in [2].
References
[1], [2]
Examples
>>> G = nx.Graph([(0, 1), (0, 2), (1, 2), (1, 3), (1, 4), (4, 5)])
>>> rc = nx.rich_club_coefficient(G, normalized=False, seed=42)
>>> rc[0]
0.4
3.52 Shortest Paths
Compute the shortest paths and path lengths between nodes in the graph.
These algorithms work with undirected and directed graphs.
shortest_path(G[, source, target, weight, ...]) Compute shortest paths in the graph.
all_shortest_paths(G, source, target[, ...]) Compute all shortest simple paths in the graph.
shortest_path_length(G[, source, target, ...]) Compute shortest path lengths in the graph.
average_shortest_path_length(G[, weight, Returns the average shortest path length.
method])
has_path(G, source, target) Returns True if G has a path from source to target.
3.52.1 shortest_path
shortest_path(G, source=None, target=None, weight=None, method='dijkstra')

Compute shortest paths in the graph.
Parameters
G
[NetworkX graph]
source
[node, optional] Starting node for path. If not specified, compute shortest paths for each pos-
sible starting node.
target
[node, optional] Ending node for path. If not specified, compute shortest paths to all possible
nodes.

weight
[None, string or function, optional (default = None)] If None, every edge has
weight/distance/cost 1. If a string, use this edge attribute as the edge weight. Any edge at-
tribute not present defaults to 1. If this is a function, the weight of an edge is the value re-
turned by the function. The function must accept exactly three positional arguments: the two
endpoints of an edge and the dictionary of edge attributes for that edge. The function must
return a number.
method
[string, optional (default = ‘dijkstra’)] The algorithm to use to compute the path. Supported
options: ‘dijkstra’, ‘bellman-ford’. Other inputs produce a ValueError. If weight is None,
unweighted graph methods are used, and this suggestion is ignored.
Returns
path: list or dictionary
All returned paths include both the source and target in the path.
If the source and target are both specified, return a single list of nodes in a shortest path from
the source to the target.
If only the source is specified, return a dictionary keyed by targets with a list of nodes in a
shortest path from the source to one of the targets.
If only the target is specified, return a dictionary keyed by sources with a list of nodes in a
shortest path from one of the sources to the target.
If neither the source nor target are specified return a dictionary of dictionaries with
path[source][target]=[list of nodes in path].
Raises
NodeNotFound
If source is not in G.
ValueError
If method is not among the supported options.
See also:
all_pairs_shortest_path
all_pairs_dijkstra_path
all_pairs_bellman_ford_path
single_source_shortest_path
single_source_dijkstra_path
single_source_bellman_ford_path
Notes
There may be more than one shortest path between a source and target. This returns only one of them.
3.52. Shortest Paths 619

Examples
>>> print(nx.shortest_path(G, source=0, target=4))
[0, 1, 2, 3, 4]
>>> p = nx.shortest_path(G, source=0) # target not specified
>>> p[3] # shortest path from source=0 to target=3
[0, 1, 2, 3]
>>> p = nx.shortest_path(G, target=4) # source not specified
>>> p[1] # shortest path from source=1 to target=4
[1, 2, 3, 4]
>>> p = nx.shortest_path(G) # source, target not specified
>>> p[2][4] # shortest path from source=2 to target=4
[2, 3, 4]
3.52.2 all_shortest_paths
all_shortest_paths(G, source, target, weight=None, method='dijkstra')

Compute all shortest simple paths in the graph.
Parameters
G
[NetworkX graph]
source
[node] Starting node for path.
target
[node] Ending node for path.
weight
return a number.
method
[string, optional (default = ‘dijkstra’)] The algorithm to use to compute the path lengths. Sup-
ported options: ‘dijkstra’, ‘bellman-ford’. Other inputs produce a ValueError. If weight is
None, unweighted graph methods are used, and this suggestion is ignored.
Returns
paths
[generator of lists] A generator of all paths between source and target.
Raises
ValueError
NetworkXNoPath
If target cannot be reached from source.
See also:

shortest_path
Notes
There may be many shortest paths between the source and target. If G contains zero-weight cycles, this function
will not produce all shortest paths because doing so would produce infinitely many paths of unbounded length –
instead, we only produce the shortest simple paths.
Examples
>>> G = nx.Graph()
>>> nx.add_path(G, [0, 1, 2])
>>> nx.add_path(G, [0, 10, 2])
>>> print([p for p in nx.all_shortest_paths(G, source=0, target=2)])
[[0, 1, 2], [0, 10, 2]]
3.52.3 shortest_path_length
shortest_path_length(G, source=None, target=None, weight=None, method='dijkstra')

Compute shortest path lengths in the graph.
Parameters
G
[NetworkX graph]
source
[node, optional] Starting node for path. If not specified, compute shortest path lengths using
all nodes as source nodes.
target
[node, optional] Ending node for path. If not specified, compute shortest path lengths using
all nodes as target nodes.
weight
return a number.
method
[string, optional (default = ‘dijkstra’)] The algorithm to use to compute the path length. Sup-
ported options: ‘dijkstra’, ‘bellman-ford’. Other inputs produce a ValueError. If weight is
None, unweighted graph methods are used, and this suggestion is ignored.
Returns
length: int or iterator
If the source and target are both specified, return the length of the shortest path from the source
to the target.

If only the source is specified, return a dict keyed by target to the shortest path length from the
source to that target.
If only the target is specified, return a dict keyed by source to the shortest path length from that
source to the target.
If neither the source nor target are specified, return an iterator over (source, dictionary) where
dictionary is keyed by target to shortest path length from source to that target.
Raises
NodeNotFound
NetworkXNoPath
If no path exists between source and target.
ValueError
See also:
all_pairs_shortest_path_length
all_pairs_dijkstra_path_length
all_pairs_bellman_ford_path_length
single_source_shortest_path_length
single_source_dijkstra_path_length
single_source_bellman_ford_path_length
Notes
The length of the path is always 1 less than the number of nodes involved in the path since the length measures the
number of edges followed.
For digraphs this returns the shortest directed path length. To find path lengths in the reverse direction use
G.reverse(copy=False) first to flip the edge orientation.
Examples
>>> nx.shortest_path_length(G, source=0, target=4)
4
>>> p = nx.shortest_path_length(G, source=0) # target not specified
>>> p[4]
4
>>> p = nx.shortest_path_length(G, target=4) # source not specified
>>> p[0]
4
>>> p = dict(nx.shortest_path_length(G)) # source,target not specified
>>> p[0][4]
4

3.52.4 average_shortest_path_length
average_shortest_path_length(G, weight=None, method=None)

Returns the average shortest path length.
The average shortest path length is
∑ d(s, t)
a=
n(n − 1)
s,t∈V
s̸=t
where V is the set of nodes in G, d(s, t) is the shortest path from s to t, and n is the number of nodes in G.
Changed in version 3.0: An exception is raised for directed graphs that are not strongly connected.
Parameters
G
[NetworkX graph]
weight
return a number.
method
[string, optional (default = ‘unweighted’ or ‘djikstra’)] The algorithm to use to compute the
path lengths. Supported options are ‘unweighted’, ‘dijkstra’, ‘bellman-ford’, ‘floyd-warshall’
and ‘floyd-warshall-numpy’. Other method values produce a ValueError. The default method
is ‘unweighted’ if weight is None, otherwise the default method is ‘dijkstra’.
Raises
If G is the null graph (that is, the graph on zero nodes).
NetworkXError
If G is not connected (or not strongly connected, in the case of a directed graph).
ValueError
Examples
>>> nx.average_shortest_path_length(G)
2.0
For disconnected graphs, you can compute the average shortest path length for each component
>>> G = nx.Graph([(1, 2), (3, 4)])

>>> for C in (G.subgraph(c).copy() for c in nx.connected_components(G)):
... print(nx.average_shortest_path_length(C))
1.0
1.0

3.52.5 has_path
has_path(G, source, target)

Returns True if G has a path from source to target.
Parameters
G
[NetworkX graph]
source
[node] Starting node for path
target
[node] Ending node for path
3.52.6 Advanced Interface
Shortest path algorithms for unweighted graphs.
single_source_shortest_path(G, source[, Compute shortest path between source and all other nodes
cutoff]) reachable from source.
single_source_shortest_path_length(G, Compute the shortest path lengths from source to all
source) reachable nodes.
single_target_shortest_path(G, target[, cut- Compute shortest path to target from all nodes that reach
off]) target.
single_target_shortest_path_length(G, Compute the shortest path lengths to target from all reach-
target) able nodes.
bidirectional_shortest_path(G, source, tar- Returns a list of nodes in a shortest path between source
get) and target.
all_pairs_shortest_path(G[, cutoff]) Compute shortest paths between all nodes.
all_pairs_shortest_path_length(G[, cut- Computes the shortest path lengths between all nodes in
off]) G.
predecessor(G, source[, target, cutoff, ...]) Returns dict of predecessors for the path from source to
all nodes in G.
single_source_shortest_path(G, source, cutoff=None)

Compute shortest path between source and all other nodes reachable from source.
Parameters
G
[NetworkX graph]
source
[node label] Starting node for path
cutoff
[integer, optional] Depth to stop the search. Only paths of length <= cutoff are returned.
Returns
lengths
[dictionary] Dictionary, keyed by target, of shortest paths.

See also:
shortest_path
Notes
The shortest path is not necessarily unique. So there can be multiple paths between the source and each target node,
all of which have the same ‘shortest’ length. For each target node, this function returns only one of those paths.
Examples
>>> path = nx.single_source_shortest_path(G, 0)
>>> path[4]
[0, 1, 2, 3, 4]
single_source_shortest_path_length
single_source_shortest_path_length(G, source, cutoff=None)

Compute the shortest path lengths from source to all reachable nodes.
Parameters
G
[NetworkX graph]
source
cutoff
Returns
lengths
[dict] Dict keyed by node to shortest path length to source.
See also:
shortest_path_length
Examples
>>> length = nx.single_source_shortest_path_length(G, 0)
>>> length[4]
4
>>> for node in length:
... print(f"{node}: {length[node]}")
0: 0
1: 1
2: 2
3: 3
4: 4

single_target_shortest_path
single_target_shortest_path(G, target, cutoff=None)

Compute shortest path to target from all nodes that reach target.
Parameters
G
[NetworkX graph]
target
[node label] Target node for path
cutoff
Returns
lengths
[dictionary] Dictionary, keyed by target, of shortest paths.
See also:
shortest_path, single_source_shortest_path
Notes
The shortest path is not necessarily unique. So there can be multiple paths between the source and each target node,
all of which have the same ‘shortest’ length. For each target node, this function returns only one of those paths.
Examples

>>> path = nx.single_target_shortest_path(G, 4)
>>> path[0]
[0, 1, 2, 3, 4]
single_target_shortest_path_length
single_target_shortest_path_length(G, target, cutoff=None)

Compute the shortest path lengths to target from all reachable nodes.
Parameters
G
[NetworkX graph]
target
[node] Target node for path
cutoff
Returns
lengths
[iterator] (source, shortest path length) iterator

See also:
single_source_shortest_path_length, shortest_path_length
Examples

>>> length = dict(nx.single_target_shortest_path_length(G, 4))
>>> length[0]
4
>>> for node in range(5):
0: 4
1: 3
2: 2
3: 1
4: 0
bidirectional_shortest_path
bidirectional_shortest_path(G, source, target)

Returns a list of nodes in a shortest path between source and target.
Parameters
G
[NetworkX graph]
source
[node label] starting node for path
target
[node label] ending node for path
Returns
path: list
List of nodes in a path from source to target.
Raises
NetworkXNoPath
See also:
shortest_path

Notes
This algorithm is used by shortest_path(G, source, target).
all_pairs_shortest_path(G, cutoff=None)
Compute shortest paths between all nodes.
Parameters
G
[NetworkX graph]
cutoff
[integer, optional] Depth at which to stop the search. Only paths of length at most cutoff
are returned.
Returns
lengths
[dictionary] Dictionary, keyed by source and target, of shortest paths.
See also:
floyd_warshall
Examples
>>> path = dict(nx.all_pairs_shortest_path(G))
>>> print(path[0][4])
[0, 1, 2, 3, 4]
all_pairs_shortest_path_length(G, cutoff=None)
Computes the shortest path lengths between all nodes in G.
Parameters
G
[NetworkX graph]
cutoff
[integer, optional] Depth at which to stop the search. Only paths of length at most cutoff
are returned.
Returns
lengths
[iterator] (source, dictionary) iterator with dictionary keyed by target and shortest path length
as the key value.

Notes
The iterator returned only has reachable node pairs.
Examples
>>> length = dict(nx.all_pairs_shortest_path_length(G))
>>> for node in [0, 1, 2, 3, 4]:
... print(f"1 - {node}: {length[1][node]}")
1 - 0: 1
1 - 1: 0
1 - 2: 1
1 - 3: 2
1 - 4: 3
>>> length[3][2]
1
>>> length[2][2]
0
predecessor
predecessor(G, source, target=None, cutoff=None, return_seen=None)

Returns dict of predecessors for the path from source to all nodes in G.
Parameters
G
[NetworkX graph]
source
target
[node label, optional] Ending node for path. If provided only predecessors between source and
target are returned
cutoff
return_seen
[bool, optional (default=None)] Whether to return a dictionary, keyed by node, of the level
(number of hops) to reach the node (as seen during breadth-first-search).
Returns
pred
[dictionary] Dictionary, keyed by node, of predecessors in the shortest path.
(pred, seen): tuple of dictionaries
If return_seen argument is set to True, then a tuple of dictionaries is returned. The first
element is the dictionary, keyed by node, of predecessors in the shortest path. The second
element is the dictionary, keyed by node, of the level (number of hops) to reach the node (as
seen during breadth-first-search).

Examples
>>> list(G)
[0, 1, 2, 3]
>>> nx.predecessor(G, 0)
{0: [], 1: [0], 2: [1], 3: [2]}
>>> nx.predecessor(G, 0, return_seen=True)
({0: [], 1: [0], 2: [1], 3: [2]}, {0: 0, 1: 1, 2: 2, 3: 3})
Shortest path algorithms for weighted graphs.

dijkstra_predecessor_and_distance(G, Compute weighted shortest path length and predecessors.

source)
dijkstra_path(G, source, target[, weight]) Returns the shortest weighted path from source to target
in G.
dijkstra_path_length(G, source, target[, Returns the shortest weighted path length in G from
weight]) source to target.
single_source_dijkstra(G, source[, target, ...]) Find shortest weighted paths and lengths from a source
node.
single_source_dijkstra_path(G, source[, Find shortest weighted paths in G from a source node.
...])
single_source_dijkstra_path_length(G, Find shortest weighted path lengths in G from a source
source) node.
multi_source_dijkstra(G, sources[, target, ...]) Find shortest weighted paths and lengths from a given set
of source nodes.
multi_source_dijkstra_path(G, sources[, ...]) Find shortest weighted paths in G from a given set of
source nodes.
multi_source_dijkstra_path_length(G, Find shortest weighted path lengths in G from a given set
sources) of source nodes.
all_pairs_dijkstra(G[, cutoff, weight]) Find shortest weighted paths and lengths between all
nodes.
all_pairs_dijkstra_path(G[, cutoff, weight]) Compute shortest paths between all nodes in a weighted
graph.
all_pairs_dijkstra_path_length(G[, cut- Compute shortest path lengths between all nodes in a
off, ...]) weighted graph.
bidirectional_dijkstra(G, source, target[, ...]) Dijkstra's algorithm for shortest paths using bidirectional
search.
bellman_ford_path(G, source, target[, weight]) Returns the shortest path from source to target in a
weighted graph G.
bellman_ford_path_length(G, source, target) Returns the shortest path length from source to target in a
weighted graph.
single_source_bellman_ford(G, source[, ...]) Compute shortest paths and lengths in a weighted graph
G.
single_source_bellman_ford_path(G, Compute shortest path between source and all other
source[, ...]) reachable nodes for a weighted graph.
single_source_bellman_ford_path_length(G, Compute the shortest path length between source and all
source) other reachable nodes for a weighted graph.
all_pairs_bellman_ford_path(G[, weight]) Compute shortest paths between all nodes in a weighted
graph.
all_pairs_bellman_ford_path_length(G[, Compute shortest path lengths between all nodes in a
weight]) weighted graph.
bellman_ford_predecessor_and_distance(G,Compute shortest path lengths and predecessors on short-
source) est paths in weighted graphs.
negative_edge_cycle(G[, weight, heuristic]) Returns True if there exists a negative edge cycle any-
where in G.
find_negative_cycle(G, source[, weight]) Returns a cycle with negative total weight if it exists.
goldberg_radzik(G, source[, weight]) Compute shortest path lengths and predecessors on short-
est paths in weighted graphs.
johnson(G[, weight]) Uses Johnson's Algorithm to compute shortest paths.

dijkstra_predecessor_and_distance
dijkstra_predecessor_and_distance(G, source, cutoff=None, weight='weight')

Compute weighted shortest path length and predecessors.
Uses Dijkstra’s Method to obtain the shortest weighted paths and return dictionaries of predecessors for each node
and distance for each node from the source.
Parameters
G
[NetworkX graph]
source
cutoff
[integer or float, optional] Length (sum of edge weights) at which the search is stopped. If
cutoff is provided, only return paths with summed weight <= cutoff.
weight
[string or function] If this is a string, then edge weights will be accessed via the edge at-
tribute with this key (that is, the weight of the edge joining u to v will be G.edges[u,
of edge attributes for that edge. The function must return a number or None to indicate a hidden
edge.
Returns
pred, distance
[dictionaries] Returns two dictionaries representing a list of predecessors of a node and the
distance to each node.
Raises
NodeNotFound
Notes
Edge weight attributes must be numerical. Distances are calculated as sums of weighted edges traversed.
The list of predecessors contains more than one element only when there are more than one shortest paths to the
key node.
Examples

>>> pred, dist = nx.dijkstra_predecessor_and_distance(G, 0)
>>> sorted(pred.items())
[(0, []), (1, [0]), (2, [1]), (3, [2]), (4, [3])]
>>> sorted(dist.items())
[(0, 0), (1, 1), (2, 2), (3, 3), (4, 4)]

>>> pred, dist = nx.dijkstra_predecessor_and_distance(G, 0, 1)

[(0, []), (1, [0])]
[(0, 0), (1, 1)]
dijkstra_path
dijkstra_path(G, source, target, weight='weight')

Returns the shortest weighted path from source to target in G.
Uses Dijkstra’s Method to compute the shortest weighted path between two nodes in a graph.
Parameters
G
[NetworkX graph]
source
[node] Starting node
target
[node] Ending node
weight
edge.
Returns
path
[list] List of nodes in a shortest path.
Raises
NodeNotFound
NetworkXNoPath
See also:
bidirectional_dijkstra
bellman_ford_path
single_source_dijkstra

Notes
The weight function can be used to hide edges by returning None. So weight = lambda u, v, d: 1 if
d['color']=="red" else None will find the shortest red path.
The weight function can be used to include node weights.
>>> def func(u, v, d):

... node_u_wt = G.nodes[u].get("node_weight", 1)
... node_v_wt = G.nodes[v].get("node_weight", 1)
... edge_wt = d.get("weight", 1)
... return node_u_wt / 2 + node_v_wt / 2 + edge_wt
In this example we take the average of start and end node weights of an edge and add it to the weight of the edge.
The function single_source_dijkstra() computes both path and length-of-path if you need both, use
that.
Examples
>>> print(nx.dijkstra_path(G, 0, 4))
[0, 1, 2, 3, 4]
dijkstra_path_length
dijkstra_path_length(G, source, target, weight='weight')

Returns the shortest weighted path length in G from source to target.
Uses Dijkstra’s Method to compute the shortest weighted path length between two nodes in a graph.
Parameters
G
[NetworkX graph]
source
target
weight
edge.
Returns
length
[number] Shortest path length.

Raises
NodeNotFound
NetworkXNoPath
See also:
bellman_ford_path_length
Notes
The function single_source_dijkstra() computes both path and length-of-path if you need both, use
that.
Examples
>>> nx.dijkstra_path_length(G, 0, 4)
4
single_source_dijkstra(G, source, target=None, cutoff=None, weight='weight')

Find shortest weighted paths and lengths from a source node.
Compute the shortest path length between source and all other reachable nodes for a weighted graph.
Uses Dijkstra’s algorithm to compute shortest paths and lengths between a source and all other reachable nodes in
a weighted graph.
Parameters
G
[NetworkX graph]
source
target
[node label, optional] Ending node for path
cutoff

weight
edge.
Returns
distance, path
[pair of dictionaries, or numeric and list.] If target is None, paths and lengths to all nodes
are computed. The return value is a tuple of two dictionaries keyed by target nodes. The first
dictionary stores distance to each target node. The second stores the path to each target node.
If target is not None, returns a tuple (distance, path), where distance is the distance from source
to target and path is a list representing the path from source to target.
Raises
NodeNotFound
See also:
single_source_bellman_ford
Notes
Based on the Python cookbook recipe (119466) at https://code.activestate.com/recipes/119466/
This algorithm is not guaranteed to work if edge weights are negative or are floating point numbers (overflows and
roundoff errors can cause problems).
Examples
>>> length, path = nx.single_source_dijkstra(G, 0)
>>> length[4]
4
>>> for node in [0, 1, 2, 3, 4]:
0: 0
1: 1
2: 2
3: 3
4: 4
>>> path[4]
[0, 1, 2, 3, 4]


>>> length, path = nx.single_source_dijkstra(G, 0, 1)
>>> length
1
>>> path
[0, 1]
single_source_dijkstra_path(G, source, cutoff=None, weight='weight')

Find shortest weighted paths in G from a source node.
Compute shortest path between source and all other reachable nodes for a weighted graph.
Parameters
G
[NetworkX graph]
source
cutoff
weight
edge.
Returns
paths
[dictionary] Dictionary of shortest path lengths keyed by target.
Raises
NodeNotFound
See also:
single_source_dijkstra, single_source_bellman_ford

Notes
Examples
>>> path = nx.single_source_dijkstra_path(G, 0)
>>> path[4]
[0, 1, 2, 3, 4]
single_source_dijkstra_path_length(G, source, cutoff=None, weight='weight')

Find shortest weighted path lengths in G from a source node.
Parameters
G
[NetworkX graph]
source
cutoff
weight
edge.
Returns
length
[dict] Dict keyed by node to shortest path length from source.
Raises
NodeNotFound
See also:
single_source_dijkstra, single_source_bellman_ford_path_length

Notes
Examples
>>> length = nx.single_source_dijkstra_path_length(G, 0)
>>> length[4]
4
>>> for node in [0, 1, 2, 3, 4]:
0: 0
1: 1
2: 2
3: 3
4: 4
multi_source_dijkstra
multi_source_dijkstra(G, sources, target=None, cutoff=None, weight='weight')

Find shortest weighted paths and lengths from a given set of source nodes.
Uses Dijkstra’s algorithm to compute the shortest paths and lengths between one of the source nodes and the given
target, or all other reachable nodes if not specified, for a weighted graph.
Parameters
G
[NetworkX graph]
sources
[non-empty set of nodes] Starting nodes for paths. If this is just a set containing a single node,
then all paths computed by this function will start from that node. If there are two or more
nodes in the set, the computed paths may begin from any one of the start nodes.
target
cutoff
weight
edge.
Returns

distance, path
[pair of dictionaries, or numeric and list] If target is None, returns a tuple of two dictionar-
ies keyed by node. The first dictionary stores distance from one of the source nodes. The
second stores the path from one of the sources to that node. If target is not None, returns a
tuple of (distance, path) where distance is the distance from source to target and path is a list
representing the path from source to target.
Raises
ValueError
If sources is empty.
NodeNotFound
If any of sources is not in G.
See also:
multi_source_dijkstra_path
multi_source_dijkstra_path_length
Notes
Based on the Python cookbook recipe (119466) at https://code.activestate.com/recipes/119466/
Examples
>>> length, path = nx.multi_source_dijkstra(G, {0, 4})
>>> for node in [0, 1, 2, 3, 4]:
0: 0
1: 1
2: 2
3: 1
4: 0
>>> path[1]
[0, 1]
>>> path[3]
[4, 3]
>>> length, path = nx.multi_source_dijkstra(G, {0, 4}, 1)

>>> length
1
>>> path
[0, 1]

multi_source_dijkstra_path
multi_source_dijkstra_path(G, sources, cutoff=None, weight='weight')

Find shortest weighted paths in G from a given set of source nodes.
Compute shortest path between any of the source nodes and all other reachable nodes for a weighted graph.
Parameters
G
[NetworkX graph]
sources
cutoff
weight
edge.
Returns
paths
[dictionary] Dictionary of shortest paths keyed by target.
Raises
ValueError
NodeNotFound
See also:
multi_source_dijkstra, multi_source_bellman_ford
Notes

Examples
>>> path = nx.multi_source_dijkstra_path(G, {0, 4})
>>> path[1]
[0, 1]
>>> path[3]
[4, 3]
multi_source_dijkstra_path_length
multi_source_dijkstra_path_length(G, sources, cutoff=None, weight='weight')

Find shortest weighted path lengths in G from a given set of source nodes.
Compute the shortest path length between any of the source nodes and all other reachable nodes for a weighted
graph.
Parameters
G
[NetworkX graph]
sources
cutoff
weight
edge.
Returns
length
[dict] Dict keyed by node to shortest path length to nearest source.
Raises
ValueError
NodeNotFound
See also:
multi_source_dijkstra

Notes
Examples
>>> length = nx.multi_source_dijkstra_path_length(G, {0, 4})
>>> for node in [0, 1, 2, 3, 4]:
0: 0
1: 1
2: 2
3: 1
4: 0
all_pairs_dijkstra
all_pairs_dijkstra(G, cutoff=None, weight='weight')

Find shortest weighted paths and lengths between all nodes.
Parameters
G
[NetworkX graph]
cutoff
weight
[string or function] If this is a string, then edge weights will be accessed via the edge
attribute with this key (that is, the weight of the edge joining u to v will be G.
edge[u][v][weight]). If no such edge attribute exists, the weight of the edge is assumed
to be one.
edge.
Yields
(node, (distance, path))
[(node obj, (dict, dict))] Each source node has two associated dicts. The first holds distance
keyed by target and the second holds paths keyed by target. (See single_source_dijkstra for the
source/target node terminology.) If desired you can apply dict() to this function to create
a dict keyed by source node to the two dicts.

Notes
The yielded dicts only have keys for reachable nodes.
Examples
>>> len_path = dict(nx.all_pairs_dijkstra(G))
>>> len_path[3][0][1]
2
>>> for node in [0, 1, 2, 3, 4]:
... print(f"3 - {node}: {len_path[3][0][node]}")
3 - 0: 3
3 - 1: 2
3 - 2: 1
3 - 3: 0
3 - 4: 1
>>> len_path[3][1][1]
[3, 2, 1]
>>> for n, (dist, path) in nx.all_pairs_dijkstra(G):
... print(path[1])
[0, 1]
[1]
[2, 1]
[3, 2, 1]
[4, 3, 2, 1]
all_pairs_dijkstra_path(G, cutoff=None, weight='weight')

Compute shortest paths between all nodes in a weighted graph.
Parameters
G
[NetworkX graph]
cutoff
weight
edge.
Returns
distance

See also:
floyd_warshall, all_pairs_bellman_ford_path
Notes
Examples
>>> path = dict(nx.all_pairs_dijkstra_path(G))
>>> path[0][4]
[0, 1, 2, 3, 4]
all_pairs_dijkstra_path_length
all_pairs_dijkstra_path_length(G, cutoff=None, weight='weight')

Compute shortest path lengths between all nodes in a weighted graph.
Parameters
G
[NetworkX graph]
cutoff
weight
edge.
Returns
distance
as the key value.

Notes
The dictionary returned only has keys for reachable node pairs.
Examples
>>> length = dict(nx.all_pairs_dijkstra_path_length(G))
>>> for node in [0, 1, 2, 3, 4]:
1 - 0: 1
1 - 1: 0
1 - 2: 1
1 - 3: 2
1 - 4: 3
>>> length[3][2]
1
>>> length[2][2]
0
bidirectional_dijkstra(G, source, target, weight='weight')

Dijkstra’s algorithm for shortest paths using bidirectional search.
Parameters
G
[NetworkX graph]
source
[node] Starting node.
target
[node] Ending node.
weight
edge.
Returns
length, path
[number and list] length is the distance from source to target. path is a list of nodes on a path
from source to target.
Raises
NodeNotFound
If either source or target is not in G.

NetworkXNoPath
See also:
shortest_path
shortest_path_length
Notes
In practice bidirectional Dijkstra is much more than twice as fast as ordinary Dijkstra.
Ordinary Dijkstra expands nodes in a sphere-like manner from the source. The radius of this sphere will eventually
be the length of the shortest path. Bidirectional Dijkstra will expand nodes from both the source and the target,
making two spheres of half this radius. Volume of the first sphere is pi*r*r while the others are 2*pi*r/2*r/
2, making up half the volume.
Examples
>>> length, path = nx.bidirectional_dijkstra(G, 0, 4)
>>> print(length)
4
>>> print(path)
[0, 1, 2, 3, 4]
bellman_ford_path
bellman_ford_path(G, source, target, weight='weight')

Returns the shortest path from source to target in a weighted graph G.
Parameters
G
[NetworkX graph]
source
[node] Starting node
target
[node] Ending node
weight
[string or function (default=”weight”)] If this is a string, then edge weights will be accessed
assumed to be one.

Returns
path
[list] List of nodes in a shortest path.
Raises
NodeNotFound
NetworkXNoPath
See also:
dijkstra_path, bellman_ford_path_length
Notes
Examples
>>> nx.bellman_ford_path(G, 0, 4)
[0, 1, 2, 3, 4]
bellman_ford_path_length
bellman_ford_path_length(G, source, target, weight='weight')

Returns the shortest path length from source to target in a weighted graph.
Parameters
G
[NetworkX graph]
source
target
weight
assumed to be one.

Returns
length
[number] Shortest path length.
Raises
NodeNotFound
NetworkXNoPath
See also:
dijkstra_path_length, bellman_ford_path
Notes
Examples
>>> nx.bellman_ford_path_length(G, 0, 4)
4
single_source_bellman_ford
single_source_bellman_ford(G, source, target=None, weight='weight')

Compute shortest paths and lengths in a weighted graph G.
Uses Bellman-Ford algorithm for shortest paths.
Parameters
G
[NetworkX graph]
source
target
weight
Returns

distance, path
[pair of dictionaries, or numeric and list] If target is None, returns a tuple of two dictionar-
ies keyed by node. The first dictionary stores distance from one of the source nodes. The
second stores the path from one of the sources to that node. If target is not None, returns a
tuple of (distance, path) where distance is the distance from source to target and path is a list
representing the path from source to target.
Raises
NodeNotFound
See also:
Notes
Examples
>>> length, path = nx.single_source_bellman_ford(G, 0)
>>> length[4]
4
>>> for node in [0, 1, 2, 3, 4]:
0: 0
1: 1
2: 2
3: 3
4: 4
>>> path[4]
[0, 1, 2, 3, 4]
>>> length, path = nx.single_source_bellman_ford(G, 0, 1)
>>> length
1
>>> path
[0, 1]
single_source_bellman_ford_path(G, source, weight='weight')

Compute shortest path between source and all other reachable nodes for a weighted graph.
Parameters
G
[NetworkX graph]
source

weight
assumed to be one.
Returns
paths
[dictionary] Dictionary of shortest path lengths keyed by target.
Raises
NodeNotFound
See also:
Notes
Examples
>>> path = nx.single_source_bellman_ford_path(G, 0)
>>> path[4]
[0, 1, 2, 3, 4]
single_source_bellman_ford_path_length(G, source, weight='weight')

Parameters
G
[NetworkX graph]
source
weight
assumed to be one.

Returns
length
[iterator] (target, shortest path length) iterator
Raises
NodeNotFound
See also:
Notes
Examples
>>> length = dict(nx.single_source_bellman_ford_path_length(G, 0))
>>> length[4]
4
>>> for node in [0, 1, 2, 3, 4]:
0: 0
1: 1
2: 2
3: 3
4: 4
all_pairs_bellman_ford_path(G, weight='weight')
Compute shortest paths between all nodes in a weighted graph.
Parameters
G
[NetworkX graph]
weight
assumed to be one.
Returns
distance

See also:
floyd_warshall, all_pairs_dijkstra_path
Notes
Examples
>>> path = dict(nx.all_pairs_bellman_ford_path(G))
>>> path[0][4]
[0, 1, 2, 3, 4]
all_pairs_bellman_ford_path_length(G, weight='weight')
Compute shortest path lengths between all nodes in a weighted graph.
Parameters
G
[NetworkX graph]
weight
assumed to be one.
Returns
distance
as the key value.
Notes
The dictionary returned only has keys for reachable node pairs.

Examples
>>> length = dict(nx.all_pairs_bellman_ford_path_length(G))
>>> for node in [0, 1, 2, 3, 4]:
1 - 0: 1
1 - 1: 0
1 - 2: 1
1 - 3: 2
1 - 4: 3
>>> length[3][2]
1
>>> length[2][2]
0
bellman_ford_predecessor_and_distance
bellman_ford_predecessor_and_distance(G, source, target=None, weight='weight', heuristic=False)

Compute shortest path lengths and predecessors on shortest paths in weighted graphs.
The algorithm has a running time of O(mn) where n is the number of nodes and m is the number of edges. It is
slower than Dijkstra but can handle negative edge weights.
If a negative cycle is detected, you can use find_negative_cycle() to return the cycle and examine it.
Shortest paths are not defined when a negative cycle exists because once reached, the path can cycle forever to build
up arbitrarily low weights.
Parameters
G
[NetworkX graph] The algorithm works for all types of graphs, including directed graphs and
multigraphs.
source: node label
Starting node for path
target
weight
heuristic
[bool] Determines whether to use a heuristic to early detect negative cycles at a hopefully
negligible cost.
Returns
pred, dist
[dictionaries] Returns two dictionaries keyed by node to predecessor in the path and to the
distance from the source respectively.

Raises
NodeNotFound
NetworkXUnbounded
If the (di)graph contains a negative (di)cycle, the algorithm raises an exception to indicate the
presence of the negative (di)cycle. Note: any negative weight edge in an undirected graph is a
negative cycle.
See also:
find_negative_cycle
Notes
The dictionaries returned only have keys for nodes reachable from the source.
In the case where the (di)graph is not connected, if a component not containing the source contains a negative
(di)cycle, it will not be detected.
In NetworkX v2.1 and prior, the source node had predecessor [None]. In NetworkX v2.2 this changed to the
source node having predecessor []
Examples

>>> pred, dist = nx.bellman_ford_predecessor_and_distance(G, 0)
[(0, []), (1, [0]), (2, [1]), (3, [2]), (4, [3])]
[(0, 0), (1, 1), (2, 2), (3, 3), (4, 4)]
>>> pred, dist = nx.bellman_ford_predecessor_and_distance(G, 0, 1)

[(0, []), (1, [0]), (2, [1]), (3, [2]), (4, [3])]
[(0, 0), (1, 1), (2, 2), (3, 3), (4, 4)]

>>> G[1][2]["weight"] = -7
>>> nx.bellman_ford_predecessor_and_distance(G, 0)
...
networkx.exception.NetworkXUnbounded: Negative cycle detected.

negative_edge_cycle
negative_edge_cycle(G, weight='weight', heuristic=True)

Returns True if there exists a negative edge cycle anywhere in G.
Parameters
G
[NetworkX graph]
weight
heuristic
[bool] Determines whether to use a heuristic to early detect negative cycles at a negligible cost.
In case of graphs with a negative cycle, the performance of detection increases by at least an
order of magnitude.
Returns
negative_cycle
[bool] True if a negative edge cycle exists, otherwise False.
Notes
This algorithm uses bellman_ford_predecessor_and_distance() but finds negative cycles on any component by first
adding a new node connected to every node, and starting bellman_ford_predecessor_and_distance on that node. It
then removes that extra node.
Examples

>>> print(nx.negative_edge_cycle(G))
False
>>> G[1][2]["weight"] = -7
>>> print(nx.negative_edge_cycle(G))
True

find_negative_cycle
find_negative_cycle(G, source, weight='weight')

Returns a cycle with negative total weight if it exists.
Bellman-Ford is used to find shortest_paths. That algorithm stops if there exists a negative cycle. This algorithm
picks up from there and returns the found negative cycle.
The cycle consists of a list of nodes in the cycle order. The last node equals the first to make it a cycle. You can
look up the edge weights in the original graph. In the case of multigraphs the relevant edge is the minimal weight
edge between the nodes in the 2-tuple.
If the graph has no negative cycle, a NetworkXError is raised.
Parameters
G
[NetworkX graph]
source: node label
The search for the negative cycle will start from this node.
weight
Returns
cycle
[list] A list of nodes in the order of the cycle found. The last node equals the first to indicate a
cycle.
Raises
NetworkXError
If no negative cycle is found.
Examples
>>> G.add_weighted_edges_from([(0, 1, 2), (1, 2, 2), (2, 0, 1), (1, 4, 2), (4, 0,␣
,→-5)])
>>> nx.find_negative_cycle(G, 0)
[4, 0, 1, 4]

goldberg_radzik
goldberg_radzik(G, source, weight='weight')

Compute shortest path lengths and predecessors on shortest paths in weighted graphs.
The algorithm has a running time of O(mn) where n is the number of nodes and m is the number of edges. It is
slower than Dijkstra but can handle negative edge weights.
Parameters
G
[NetworkX graph] The algorithm works for all types of graphs, including directed graphs and
multigraphs.
source: node label
Starting node for path
weight
Returns
pred, dist
[dictionaries] Returns two dictionaries keyed by node to predecessor in the path and to the
distance from the source respectively.
Raises
NodeNotFound
NetworkXUnbounded
If the (di)graph contains a negative (di)cycle, the algorithm raises an exception to indicate the
presence of the negative (di)cycle. Note: any negative weight edge in an undirected graph is a
negative cycle.
Notes
The dictionaries returned only have keys for nodes reachable from the source.
In the case where the (di)graph is not connected, if a component not containing the source contains a negative
(di)cycle, it will not be detected.

Examples

>>> pred, dist = nx.goldberg_radzik(G, 0)
[(0, None), (1, 0), (2, 1), (3, 2), (4, 3)]
[(0, 0), (1, 1), (2, 2), (3, 3), (4, 4)]

>>> G[1][2]["weight"] = -7
>>> nx.goldberg_radzik(G, 0)
...
networkx.exception.NetworkXUnbounded: Negative cycle detected.
johnson
johnson(G, weight='weight')
Uses Johnson’s Algorithm to compute shortest paths.
Johnson’s Algorithm finds a shortest path between each pair of nodes in a weighted graph even if negative weights
are present.
Parameters
G
[NetworkX graph]
weight
Returns
distance
Raises
NetworkXError
If given graph is not weighted.
See also:
floyd_warshall_predecessor_and_distance
floyd_warshall_numpy
bellman_ford_predecessor_and_distance

Notes
Johnson’s algorithm is suitable even for graphs with negative weights. It works by using the Bellman–Ford algorithm
to compute a transformation of the input graph that removes all negative weights, allowing Dijkstra’s algorithm to
be used on the transformed graph.
The time complexity of this algorithm is O(n2 log n + nm), where n is the number of nodes and m the number
of edges in the graph. For dense graphs, this may be faster than the Floyd–Warshall algorithm.
Examples
>>> graph = nx.DiGraph()

>>> graph.add_weighted_edges_from(
... [("0", "3", 3), ("0", "1", -5), ("0", "2", 2), ("1", "2", 4), ("2", "3",␣
,→1)]
... )
>>> paths = nx.johnson(graph, weight="weight")
>>> paths["0"]["2"]
['0', '1', '2']
3.52.7 Dense Graphs
Floyd-Warshall algorithm for shortest paths.
floyd_warshall(G[, weight]) Find all-pairs shortest path lengths using Floyd's algo-
rithm.
floyd_warshall_predecessor_and_distance(G[, Find all-pairs shortest path lengths using Floyd's algo-
...]) rithm.
floyd_warshall_numpy(G[, nodelist, weight]) Find all-pairs shortest path lengths using Floyd's algo-
rithm.
reconstruct_path(source, target, predecessors) Reconstruct a path from source to target us-
ing the predecessors dict as returned by
floyd_warshall
floyd_warshall(G, weight='weight')
Find all-pairs shortest path lengths using Floyd’s algorithm.
Parameters
G
[NetworkX graph]
weight: string, optional (default= ‘weight’)
Edge data key corresponding to the edge weight.
Returns
distance
[dict] A dictionary, keyed by source and target, of shortest paths distances between nodes.
See also:

Notes
Floyd’s algorithm is appropriate for finding shortest paths in dense graphs or graphs with negative weights when
Dijkstra’s algorithm fails. This algorithm can still fail if there are negative cycles. It has running time O(n3 ) with
running space of O(n2 ).
floyd_warshall_predecessor_and_distance(G, weight='weight')
Parameters
G
[NetworkX graph]
weight: string, optional (default= ‘weight’)
Returns
predecessor,distance
[dictionaries] Dictionaries, keyed by source and target, of predecessors and distances in the
shortest path.
See also:
floyd_warshall
Notes
Examples
>>> G.add_weighted_edges_from(
... [
... ("s", "u", 10),
... ("s", "x", 5),
... ("u", "v", 1),
... ("u", "x", 2),
... ("v", "y", 1),


... ("x", "u", 3),
... ("x", "v", 5),
... ("x", "y", 2),
... ("y", "s", 7),
... ("y", "v", 6),
... ]
... )
>>> predecessors, _ = nx.floyd_warshall_predecessor_and_distance(G)
>>> print(nx.reconstruct_path("s", "v", predecessors))
['s', 'x', 'u', 'v']
floyd_warshall_numpy(G, nodelist=None, weight='weight')

This algorithm for finding shortest paths takes advantage of matrix representations of a graph and works well for
dense graphs where all-pairs shortest path lengths are desired. The results are returned as a NumPy array, distance[i,
j], where i and j are the indexes of two nodes in nodelist. The entry distance[i, j] is the distance along a shortest
path from i to j. If no path exists the distance is Inf.
Parameters
G
[NetworkX graph]
nodelist
[list, optional (default=G.nodes)] The rows and columns are ordered by the nodes in nodelist.
If nodelist is None then the ordering is produced by G.nodes. Nodelist should include all nodes
in G.
Returns
distance
[2D numpy.ndarray] A numpy array of shortest path distances between nodes. If there is no
path between two nodes the value is Inf.
Raises
NetworkXError
If nodelist is not a list of the nodes in G.
Notes

reconstruct_path
reconstruct_path(source, target, predecessors)

Reconstruct a path from source to target using the predecessors dict as returned by
Parameters
source
target
predecessors: dictionary
Dictionary, keyed by source and target, of predecessors in the shortest path, as returned by
Returns
path
[list] A list of nodes containing the shortest path from source to target
If source and target are the same, an empty list is returned
See also:
Notes
This function is meant to give more applicability to the floyd_warshall_predecessor_and_distance function
3.52.8 A* Algorithm
Shortest paths and path lengths using the A* (“A star”) algorithm.
astar_path(G, source, target[, heuristic, ...]) Returns a list of nodes in a shortest path between source
and target using the A* ("A-star") algorithm.
astar_path_length(G, source, target[, ...]) Returns the length of the shortest path between source and
target using the A* ("A-star") algorithm.
astar_path
astar_path(G, source, target, heuristic=None, weight='weight')

Returns a list of nodes in a shortest path between source and target using the A* (“A-star”) algorithm.
There may be more than one shortest path. This returns only one.
Parameters
G
[NetworkX graph]
source

target
heuristic
[function] A function to evaluate the estimate of the distance from the a node to the target. The
function takes two nodes arguments and must return a number. If the heuristic is inadmissible
(if it might overestimate the cost of reaching the goal from a node), the result may not be a
shortest path. The algorithm does not support updating heuristic values for the same node due
to caching the first heuristic calculation per node.
weight
must accept exactly three positional arguments: the two endpoints of an edge and the dictio-
nary of edge attributes for that edge. The function must return a number or None to indicate
a hidden edge.
Raises
NetworkXNoPath
See also:
shortest_path, dijkstra_path
Notes
Examples
>>> print(nx.astar_path(G, 0, 4))
[0, 1, 2, 3, 4]
>>> G = nx.grid_graph(dim=[3, 3]) # nodes are two-tuples (x,y)
>>> nx.set_edge_attributes(G, {e: e[1][0] * 2 for e in G.edges()}, "cost")
>>> def dist(a, b):
... (x1, y1) = a
... (x2, y2) = b
... return ((x1 - x2) ** 2 + (y1 - y2) ** 2) ** 0.5
>>> print(nx.astar_path(G, (0, 0), (2, 2), heuristic=dist, weight="cost"))
[(0, 0), (0, 1), (0, 2), (1, 2), (2, 2)]

astar_path_length
astar_path_length(G, source, target, heuristic=None, weight='weight')

Returns the length of the shortest path between source and target using the A* (“A-star”) algorithm.
Parameters
G
[NetworkX graph]
source
target
heuristic
[function] A function to evaluate the estimate of the distance from the a node to the target. The
function takes two nodes arguments and must return a number. If the heuristic is inadmissible
(if it might overestimate the cost of reaching the goal from a node), the result may not be a
shortest path. The algorithm does not support updating heuristic values for the same node due
to caching the first heuristic calculation per node.
weight
must accept exactly three positional arguments: the two endpoints of an edge and the dictio-
nary of edge attributes for that edge. The function must return a number or None to indicate
a hidden edge.
Raises
——
NetworkXNoPath
See also:
astar_path
3.53 Similarity Measures
Functions measuring similarity using graph edit distance.

The graph edit distance is the number of edge/node changes needed to make two graphs isomorphic.
The default algorithm/implementation is sub-optimal for some graphs. The problem of finding the exact Graph Edit
Distance (GED) is NP-hard so it is often slow. If the simple interface graph_edit_distance takes too long for
your graph, try optimize_graph_edit_distance and/or optimize_edit_paths.
At the same time, I encourage capable people to investigate alternative GED algorithms, in order to improve the choices
available.
3.53. Similarity Measures 665

graph_edit_distance(G1, G2[, node_match, ...]) Returns GED (graph edit distance) between graphs G1
and G2.
optimal_edit_paths(G1, G2[, node_match, ...]) Returns all minimum-cost edit paths transforming G1 to
G2.
optimize_graph_edit_distance(G1, G2[, ...]) Returns consecutive approximations of GED (graph edit
distance) between graphs G1 and G2.
optimize_edit_paths(G1, G2[, node_match, ...]) GED (graph edit distance) calculation: advanced inter-
face.
simrank_similarity(G[, source, target, ...]) Returns the SimRank similarity of nodes in the graph G.
panther_similarity(G, source[, k, ...]) Returns the Panther similarity of nodes in the graph G to
node v.
generate_random_paths(G, sample_size[, ...]) Randomly generate sample_size paths of length
path_length.
3.53.1 graph_edit_distance
graph_edit_distance(G1, G2, node_match=None, edge_match=None, node_subst_cost=None,

node_del_cost=None, node_ins_cost=None, edge_subst_cost=None, edge_del_cost=None,
edge_ins_cost=None, roots=None, upper_bound=None, timeout=None)
Returns GED (graph edit distance) between graphs G1 and G2.
Graph edit distance is a graph similarity measure analogous to Levenshtein distance for strings. It is defined as
minimum cost of edit path (sequence of node and edge edit operations) transforming graph G1 to graph isomorphic
to G2.
Parameters
G1, G2: graphs
The two graphs G1 and G2 must be of the same type.
node_match
equal during matching.
Ignored if node_subst_cost is specified. If neither node_match nor node_subst_cost are spec-
ified then node attributes are not considered.
edge_match
[callable] A function that returns True if the edge attribute dictionaries for the pair of nodes
(u1, v1) in G1 and (u2, v2) in G2 should be considered equal during matching.
tion.
Ignored if edge_subst_cost is specified. If neither edge_match nor edge_subst_cost are speci-
fied then edge attributes are not considered.

node_subst_cost, node_del_cost, node_ins_cost

[callable] Functions that return the costs of node substitution, node deletion, and node insertion,
respectively.
The functions will be called like
node_subst_cost(G1.nodes[n1], G2.nodes[n2]), node_del_cost(G1.nodes[n1]),
node_ins_cost(G2.nodes[n2]).
That is, the functions will receive the node attribute dictionaries as inputs. The functions are
expected to return positive numeric values.
Function node_subst_cost overrides node_match if specified. If neither node_match nor
node_subst_cost are specified then default node substitution cost of 0 is used (node attributes
are not considered during matching).
If node_del_cost is not specified then default node deletion cost of 1 is used. If node_ins_cost
is not specified then default node insertion cost of 1 is used.
edge_subst_cost, edge_del_cost, edge_ins_cost
[callable] Functions that return the costs of edge substitution, edge deletion, and edge insertion,
respectively.
edge_subst_cost(G1[u1][v1], G2[u2][v2]), edge_del_cost(G1[u1][v1]),
edge_ins_cost(G2[u2][v2]).
That is, the functions will receive the edge attribute dictionaries as inputs. The functions are
Function edge_subst_cost overrides edge_match if specified. If neither edge_match nor
edge_subst_cost are specified then default edge substitution cost of 0 is used (edge attributes
If edge_del_cost is not specified then default edge deletion cost of 1 is used. If edge_ins_cost
is not specified then default edge insertion cost of 1 is used.
roots
[2-tuple] Tuple where first element is a node in G1 and the second is a node in G2. These nodes
are forced to be matched in the comparison to allow comparison between rooted graphs.
upper_bound
[numeric] Maximum edit distance to consider. Return None if no edit distance under or equal
to upper_bound exists.
timeout
[numeric] Maximum number of seconds to execute. After timeout is met, the current best
GED is returned.
See also:
optimal_edit_paths, optimize_graph_edit_distance
is_isomorphic
test for graph edit distance of 0

References
[1]
Examples
>>> G1 = nx.cycle_graph(6)
>>> G2 = nx.wheel_graph(7)
>>> nx.graph_edit_distance(G1, G2)
7.0
>>> G1 = nx.star_graph(5)
>>> G2 = nx.star_graph(5)
>>> nx.graph_edit_distance(G1, G2, roots=(0, 0))
0.0
>>> nx.graph_edit_distance(G1, G2, roots=(1, 0))
8.0
3.53.2 optimal_edit_paths
optimal_edit_paths(G1, G2, node_match=None, edge_match=None, node_subst_cost=None,

edge_ins_cost=None, upper_bound=None)
Returns all minimum-cost edit paths transforming G1 to G2.
Graph edit path is a sequence of node and edge edit operations transforming graph G1 to graph isomorphic to G2.
Edit operations include substitutions, deletions, and insertions.
Parameters
G1, G2: graphs
node_match
edge_match
tion.


respectively.
respectively.
upper_bound
[numeric] Maximum edit distance to consider.
Returns
edit_paths
[list of tuples (node_edit_path, edge_edit_path)] node_edit_path : list of tuples (u, v)
edge_edit_path : list of tuples ((u1, v1), (u2, v2))
cost
[numeric] Optimal edit path cost (graph edit distance).
See also:
graph_edit_distance, optimize_edit_paths

References
[1]
Examples
>>> paths, cost = nx.optimal_edit_paths(G1, G2)
>>> len(paths)
40
>>> cost
5.0
3.53.3 optimize_graph_edit_distance
optimize_graph_edit_distance(G1, G2, node_match=None, edge_match=None, node_subst_cost=None,

node_del_cost=None, node_ins_cost=None, edge_subst_cost=None,
edge_del_cost=None, edge_ins_cost=None, upper_bound=None)
Returns consecutive approximations of GED (graph edit distance) between graphs G1 and G2.
Graph edit distance is a graph similarity measure analogous to Levenshtein distance for strings. It is defined as
minimum cost of edit path (sequence of node and edge edit operations) transforming graph G1 to graph isomorphic
to G2.
Parameters
G1, G2: graphs
node_match
edge_match
tion.


respectively.
respectively.
upper_bound
Returns
Generator of consecutive approximations of graph edit distance.
See also:
graph_edit_distance, optimize_edit_paths
References
[1]

Examples
>>> for v in nx.optimize_graph_edit_distance(G1, G2):
... minv = v
>>> minv
7.0
3.53.4 optimize_edit_paths
optimize_edit_paths(G1, G2, node_match=None, edge_match=None, node_subst_cost=None,

edge_ins_cost=None, upper_bound=None, strictly_decreasing=True, roots=None,
timeout=None)
GED (graph edit distance) calculation: advanced interface.
Graph edit path is a sequence of node and edge edit operations transforming graph G1 to graph isomorphic to G2.
Edit operations include substitutions, deletions, and insertions.
Graph edit distance is defined as minimum cost of edit path.
Parameters
G1, G2: graphs
node_match
edge_match
tion.
respectively.


respectively.
upper_bound
strictly_decreasing
[bool] If True, return consecutive approximations of strictly decreasing cost. Otherwise, return
all edit paths of cost less than or equal to the previous minimum cost.
roots
[2-tuple] Tuple where first element is a node in G1 and the second is a node in G2. These nodes
are forced to be matched in the comparison to allow comparison between rooted graphs.
timeout
[numeric] Maximum number of seconds to execute. After timeout is met, the current best
GED is returned.
Returns
Generator of tuples (node_edit_path, edge_edit_path, cost)
node_edit_path : list of tuples (u, v) edge_edit_path : list of tuples ((u1, v1), (u2, v2)) cost :
numeric
See also:
graph_edit_distance, optimize_graph_edit_distance, optimal_edit_paths

References
[1]
3.53.5 simrank_similarity
simrank_similarity(G, source=None, target=None, importance_factor=0.9, max_iterations=1000,

tolerance=0.0001)
Returns the SimRank similarity of nodes in the graph G.
SimRank is a similarity metric that says “two objects are considered to be similar if they are referenced by similar
objects.” [1].
The pseudo-code definition from the paper is:
def simrank(G, u, v):

in_neighbors_u = G.predecessors(u)
in_neighbors_v = G.predecessors(v)
scale = C / (len(in_neighbors_u) * len(in_neighbors_v))
return scale * sum(simrank(G, w, x)
for w, x in product(in_neighbors_u,
in_neighbors_v))
where G is the graph, u is the source, v is the target, and C is a float decay or importance factor between 0 and 1.
The SimRank algorithm for determining node similarity is defined in [2].
Parameters
G
[NetworkX graph] A NetworkX graph
source
[node] If this is specified, the returned dictionary maps each node v in the graph to the simi-
larity between source and v.
target
[node] If both source and target are specified, the similarity value between source and
target is returned. If target is specified but source is not, this argument is ignored.
importance_factor
[float] The relative importance of indirect neighbors with respect to direct neighbors.
max_iterations
[integer] Maximum number of iterations.
tolerance
[float] Error tolerance used to check convergence. When an iteration of the algorithm finds
that no similarity value changes more than this amount, the algorithm halts.
Returns
similarity
[dictionary or float] If source and target are both None, this returns a dictionary of
dictionaries, where keys are node pairs and value are similarity of the pair of nodes.
If source is not None but target is, this returns a dictionary mapping node to the simi-
larity of source and that node.

If neither source nor target is None, this returns the similarity value for the given pair
of nodes.
References
[1], [2]
Examples
>>> nx.simrank_similarity(G)
{0: {0: 1.0, 1: 0.0}, 1: {0: 0.0, 1: 1.0}}
>>> nx.simrank_similarity(G, source=0)
{0: 1.0, 1: 0.0}
>>> nx.simrank_similarity(G, source=0, target=0)
1.0
The result of this function can be converted to a numpy array representing the SimRank matrix by using the node
order of the graph to determine which row and column represent each node. Other ordering of nodes is also
possible.
>>> import numpy as np

>>> sim = nx.simrank_similarity(G)
>>> np.array([[sim[u][v] for v in G] for u in G])
array([[1., 0.],
[0., 1.]])
>>> sim_1d = nx.simrank_similarity(G, source=0)
>>> np.array([sim[0][v] for v in G])
array([1., 0.])
3.53.6 panther_similarity
panther_similarity(G, source, k=5, path_length=5, c=0.5, delta=0.1, eps=None)

Returns the Panther similarity of nodes in the graph G to node v.
Panther is a similarity metric that says “two objects are considered to be similar if they frequently appear on the
same paths.” [1].
Parameters
G
source
[node] Source node for which to find the top k similar other nodes
k
[int (default = 5)] The number of most similar nodes to return
path_length
[int (default = 5)] How long the randomly generated paths should be (T in [1])
c
[float (default = 0.5)] A universal positive constant used to scale the number of sample random
paths to generate.

delta
[float (default = 0.1)] The probability that the similarity S is not an epsilon-approximation to
(R, phi), where R is the number of random paths and ϕ is the probability that an element
sampled from a set A ⊆ D, where D is the domain.
eps
[float or None (default = None)] The error bound. Per [1], a good value is sqrt(1/|E|).
Therefore, if no value is provided, the recommended computed value will be used.
Returns
similarity
[dictionary] Dictionary of nodes to similarity scores (as floats). Note: the self-similarity (i.e.,
v) will not be included in the returned dictionary.
References
[1]
Examples
>>> sim = nx.panther_similarity(G, 0)
3.53.7 generate_random_paths
generate_random_paths(G, sample_size, path_length=5, index_map=None)

Randomly generate sample_size paths of length path_length.
Parameters
G
sample_size
[integer] The number of paths to generate. This is R in [1].
path_length
[integer (default = 5)] The maximum size of the path to randomly generate. This is T in [1].
According to the paper, T >= 5 is recommended.
index_map
[dictionary, optional] If provided, this will be populated with the inverted index of nodes
mapped to the set of generated random path indices within paths.
Returns
paths
[generator of lists] Generator of sample_size paths each with length path_length.

References
[1]
Examples
Note that the return value is the list of paths:
>>> random_path = nx.generate_random_paths(G, 2)
By passing a dictionary into index_map, it will build an inverted index mapping of nodes to the paths in which
that node is present:
>>> index_map = {}
>>> random_path = nx.generate_random_paths(G, 3, index_map=index_map)
>>> paths_containing_node_0 = [random_path[path_idx] for path_idx in index_map.
,→get(0, [])]
3.54 Simple Paths
all_simple_paths(G, source, target[, cutoff]) Generate all simple paths in the graph G from source to
target.
all_simple_edge_paths(G, source, target[, ...]) Generate lists of edges for all simple paths in G from
source to target.
is_simple_path(G, nodes) Returns True if and only if nodes form a simple path in
G.
shortest_simple_paths(G, source, target[, ...]) Generate all simple paths in the graph G from source to
target,
3.54.1 all_simple_paths
all_simple_paths(G, source, target, cutoff=None)

Generate all simple paths in the graph G from source to target.
A simple path is a path with no repeated nodes.
Parameters
G
[NetworkX graph]
source
target
[nodes] Single node or iterable of nodes at which to end path
cutoff
Returns
3.54. Simple Paths 677

path_generator: generator
A generator that produces lists of simple paths. If there are no paths between the source and
target within the given cutoff the generator produces no output. If it is possible to traverse
the same sequence of nodes in multiple ways, namely through parallel edges, then it will be
returned multiple times (once for each viable edge combination).
See also:
all_shortest_paths, shortest_path, has_path
Notes
This algorithm uses a modified depth-first search to generate the paths [1]. A single path can be found in O(V +E)
time but the number of simple paths in a graph can be very large, e.g. O(n!) in the complete graph of order n.
This function does not check that a path exists between source and target. For large graphs, this may result in
very long runtimes. Consider using has_path to check that a path exists between source and target before
calling this function on large graphs.
References
[1]
Examples
This iterator generates lists of nodes:
>>> for path in nx.all_simple_paths(G, source=0, target=3):
... print(path)
...
[0, 1, 2, 3]
[0, 1, 3]
[0, 2, 1, 3]
[0, 2, 3]
[0, 3]
You can generate only those paths that are shorter than a certain length by using the cutoff keyword argument:
>>> paths = nx.all_simple_paths(G, source=0, target=3, cutoff=2)

>>> print(list(paths))
[[0, 1, 3], [0, 2, 3], [0, 3]]
To get each path as the corresponding list of edges, you can use the networkx.utils.pairwise() helper
function:
>>> paths = nx.all_simple_paths(G, source=0, target=3)

>>> for path in map(nx.utils.pairwise, paths):
... print(list(path))
[(0, 1), (1, 2), (2, 3)]
[(0, 1), (1, 3)]
[(0, 2), (2, 1), (1, 3)]
[(0, 2), (2, 3)]
[(0, 3)]

Pass an iterable of nodes as target to generate all paths ending in any of several nodes:
>>> for path in nx.all_simple_paths(G, source=0, target=[3, 2]):
... print(path)
...
[0, 1, 2]
[0, 1, 2, 3]
[0, 1, 3]
[0, 1, 3, 2]
[0, 2]
[0, 2, 1, 3]
[0, 2, 3]
[0, 3]
[0, 3, 1, 2]
[0, 3, 2]
Iterate over each path from the root nodes to the leaf nodes in a directed acyclic graph using a functional program-
ming approach:
>>> from itertools import chain

>>> from itertools import product
>>> from itertools import starmap
>>> from functools import partial
>>>
>>> chaini = chain.from_iterable
>>>
>>> G = nx.DiGraph([(0, 1), (1, 2), (0, 3), (3, 2)])
>>> roots = (v for v, d in G.in_degree() if d == 0)
>>> leaves = (v for v, d in G.out_degree() if d == 0)
>>> all_paths = partial(nx.all_simple_paths, G)
>>> list(chaini(starmap(all_paths, product(roots, leaves))))
[[0, 1, 2], [0, 3, 2]]
The same list computed using an iterative approach:
>>> G = nx.DiGraph([(0, 1), (1, 2), (0, 3), (3, 2)])

>>> leaves = (v for v, d in G.out_degree() if d == 0)
>>> all_paths = []
>>> for root in roots:
... for leaf in leaves:
... paths = nx.all_simple_paths(G, root, leaf)
... all_paths.extend(paths)
>>> all_paths
[[0, 1, 2], [0, 3, 2]]
Iterate over each path from the root nodes to the leaf nodes in a directed acyclic graph passing all leaves together
to avoid unnecessary compute:
>>> G = nx.DiGraph([(0, 1), (2, 1), (1, 3), (1, 4)])

>>> leaves = [v for v, d in G.out_degree() if d == 0]
>>> all_paths = []
>>> for root in roots:
... paths = nx.all_simple_paths(G, root, leaves)
... all_paths.extend(paths)


>>> all_paths
[[0, 1, 3], [0, 1, 4], [2, 1, 3], [2, 1, 4]]
If parallel edges offer multiple ways to traverse a given sequence of nodes, this sequence of nodes will be returned
multiple times:
>>> G = nx.MultiDiGraph([(0, 1), (0, 1), (1, 2)])

>>> list(nx.all_simple_paths(G, 0, 2))
[[0, 1, 2], [0, 1, 2]]
3.54.2 all_simple_edge_paths
all_simple_edge_paths(G, source, target, cutoff=None)

Generate lists of edges for all simple paths in G from source to target.
Parameters
G
[NetworkX graph]
source
target
[nodes] Single node or iterable of nodes at which to end path
cutoff
Returns
A generator that produces lists of simple paths. If there are no paths between the source and
target within the given cutoff the generator produces no output. For multigraphs, the list of
edges have elements of the form (u,v,k). Where k corresponds to the edge key.
See also:
all_shortest_paths, shortest_path, all_simple_paths
Notes
This algorithm uses a modified depth-first search to generate the paths [1]. A single path can be found in O(V +E)
time but the number of simple paths in a graph can be very large, e.g. O(n!) in the complete graph of order n.

References
[1]
Examples
Print the simple path edges of a Graph:
>>> g = nx.Graph([(1, 2), (2, 4), (1, 3), (3, 4)])

>>> for path in sorted(nx.all_simple_edge_paths(g, 1, 4)):
... print(path)
[(1, 2), (2, 4)]
[(1, 3), (3, 4)]
Print the simple path edges of a MultiGraph. Returned edges come with their associated keys:
>>> mg = nx.MultiGraph()
>>> mg.add_edge(1, 2, key="k0")
'k0'
'k1'
'k0'
>>> for path in sorted(nx.all_simple_edge_paths(mg, 1, 3)):
... print(path)
[(1, 2, 'k0'), (2, 3, 'k0')]
[(1, 2, 'k1'), (2, 3, 'k0')]
3.54.3 is_simple_path
is_simple_path(G, nodes)
Returns True if and only if nodes form a simple path in G.
A simple path in a graph is a nonempty sequence of nodes in which no node appears more than once in the sequence,
and each adjacent pair of nodes in the sequence is adjacent in the graph.
Parameters
G
nodes
[list] A list of one or more nodes in the graph G.
Returns
bool
Whether the given list of nodes represents a simple path in G.

Notes
An empty list of nodes is not a path but a list of one node is a path. Here’s an explanation why.
This function operates on node paths. One could also consider edge paths. There is a bijection between node paths
and edge paths.
The length of a path is the number of edges in the path, so a list of nodes of length n corresponds to a path of length
n - 1. Thus the smallest edge path would be a list of zero edges, the empty path. This corresponds to a list of one
node.
To convert between a node path and an edge path, you can use code like the following:

>>> nodes = [0, 1, 2, 3]
>>> edges = list(pairwise(nodes))
>>> edges
[(0, 1), (1, 2), (2, 3)]
>>> nodes = [edges[0][0]] + [v for u, v in edges]
>>> nodes
[0, 1, 2, 3]
Examples
>>> nx.is_simple_path(G, [2, 3, 0])
True
>>> nx.is_simple_path(G, [0, 2])
False
3.54.4 shortest_simple_paths
shortest_simple_paths(G, source, target, weight=None)
Generate all simple paths in the graph G from source to target,

starting from shortest ones.
If a weighted shortest path search is to be used, no negative weights are allowed.
Parameters
G
[NetworkX graph]
source
target
weight
[string or function] If it is a string, it is the name of the edge attribute to be used as a weight.
If it is a function, the weight of an edge is the value returned by the function. The function must
accept exactly three positional arguments: the two endpoints of an edge and the dictionary of
edge attributes for that edge. The function must return a number.

If None all edges are considered to have unit weight. Default value None.
Returns
A generator that produces lists of simple paths, in order from shortest to longest.
Raises
NetworkXNoPath
NetworkXError
If source or target nodes are not in the input graph.
If the input graph is a Multi[Di]Graph.
See also:
all_shortest_paths
shortest_path
all_simple_paths
Notes
This procedure is based on algorithm by Jin Y. Yen [1]. Finding the first K paths requires O(KN 3 ) operations.
References
[1]
Examples
>>> paths = list(nx.shortest_simple_paths(G, 0, 3))
>>> print(paths)
[[0, 1, 2, 3], [0, 6, 5, 4, 3]]
You can use this function to efficiently compute the k shortest/best paths between two nodes.
>>> from itertools import islice

>>> def k_shortest_paths(G, source, target, k, weight=None):
... return list(
... islice(nx.shortest_simple_paths(G, source, target, weight=weight), k)
... )
>>> for path in k_shortest_paths(G, 0, 3, 2):
... print(path)
[0, 1, 2, 3]
[0, 6, 5, 4, 3]

3.55 Small-world
Functions for estimating the small-world-ness of graphs.

A small world network is characterized by a small average shortest path length, and a large clustering coefficient.
Small-worldness is commonly measured with the coefficient sigma or omega.
Both coefficients compare the average clustering coefficient and shortest path length of a given graph against the same
quantities for an equivalent random or lattice graph.
For more information, see the Wikipedia article on small-world network [1].
random_reference(G[, niter, connectivity, seed]) Compute a random graph by swapping edges of a given
graph.
lattice_reference(G[, niter, D, ...]) Latticize the given graph by swapping edges.
sigma(G[, niter, nrand, seed]) Returns the small-world coefficient (sigma) of the given
graph.
omega(G[, niter, nrand, seed]) Returns the small-world coefficient (omega) of a graph
3.55.1 random_reference
random_reference(G, niter=1, connectivity=True, seed=None)

Compute a random graph by swapping edges of a given graph.
Parameters
G
[graph] An undirected graph with 4 or more nodes.
niter
[integer (optional, default=1)] An edge is rewired approximately niter times.
connectivity
[boolean (optional, default=True)] When True, ensure connectivity for the randomized graph.
seed
Randomness.
Returns
G
[graph] The randomized graph.
Notes
The implementation is adapted from the algorithm by Maslov and Sneppen (2002) [1].

References
[1]
3.55.2 lattice_reference
lattice_reference(G, niter=5, D=None, connectivity=True, seed=None)

Latticize the given graph by swapping edges.
Parameters
G
[graph] An undirected graph with 4 or more nodes.
niter
[integer (optional, default=1)] An edge is rewired approximatively niter times.
D
[numpy.array (optional, default=None)] Distance to the diagonal matrix.
connectivity
[boolean (optional, default=True)] Ensure connectivity for the latticized graph when set to
True.
seed
Randomness.
Returns
G
[graph] The latticized graph.
Notes
The implementation is adapted from the algorithm by Sporns et al. [1]. which is inspired from the original work
by Maslov and Sneppen(2002) [2].
References
[1], [2]
3.55.3 sigma
sigma(G, niter=100, nrand=10, seed=None)

Returns the small-world coefficient (sigma) of the given graph.
The small-world coefficient is defined as: sigma = C/Cr / L/Lr where C and L are respectively the average clustering
coefficient and average shortest path length of G. Cr and Lr are respectively the average clustering coefficient and
average shortest path length of an equivalent random graph.
A graph is commonly classified as small-world if sigma>1.
Parameters
3.55. Small-world 685

G
niter
[integer (optional, default=100)] Approximate number of rewiring per edge to compute the
equivalent random graph.
nrand
[integer (optional, default=10)] Number of random graphs generated to compute the average
clustering coefficient (Cr) and average shortest path length (Lr).
seed
Randomness.
Returns
sigma
[float] The small-world coefficient of G.
Notes
The implementation is adapted from Humphries et al. [1] [2].
References
[1], [2]
3.55.4 omega
omega(G, niter=5, nrand=10, seed=None)

Returns the small-world coefficient (omega) of a graph
The small-world coefficient of a graph G is:
omega = Lr/L - C/Cl
where C and L are respectively the average clustering coefficient and average shortest path length of G. Lr is
the average shortest path length of an equivalent random graph and Cl is the average clustering coefficient of an
equivalent lattice graph.
The small-world coefficient (omega) measures how much G is like a lattice or a random graph. Negative values
mean G is similar to a lattice whereas positive values mean G is a random graph. Values close to 0 mean that G
has small-world characteristics.
Parameters
G
niter: integer (optional, default=5)
Approximate number of rewiring per edge to compute the equivalent random graph.
nrand: integer (optional, default=10)
Number of random graphs generated to compute the maximal clustering coefficient (Cr) and
average shortest path length (Lr).

seed
Randomness.
Returns
omega
[float] The small-world coefficient (omega)
Notes
The implementation is adapted from the algorithm by Telesford et al. [1].
References
[1]
3.56 s metric
s_metric(G[, normalized]) Returns the s-metric of graph.
3.56.1 s_metric
s_metric(G, normalized=True)
Returns the s-metric of graph.
The s-metric is defined as the sum of the products deg(u)*deg(v) for every edge (u,v) in G. If norm is provided
construct the s-max graph and compute it’s s_metric, and return the normalized s value
Parameters
G
[graph] The graph used to compute the s-metric.
normalized
[bool (optional)] Normalize the value.
Returns
s
[float] The s-metric of the graph.
3.56. s metric 687

References
[1]
3.57 Sparsifiers
Functions for computing sparsifiers of graphs.
spanner(G, stretch[, weight, seed]) Returns a spanner of the given graph with the given
stretch.
3.57.1 spanner
spanner(G, stretch, weight=None, seed=None)

Returns a spanner of the given graph with the given stretch.
A spanner of a graph G = (V, E) with stretch t is a subgraph H = (V, E_S) such that E_S is a subset of E and the
distance between any pair of nodes in H is at most t times the distance between the nodes in G.
Parameters
G
[NetworkX graph] An undirected simple graph.
stretch
[float] The stretch of the spanner.
weight
[object] The edge attribute to use as distance.
seed
Randomness.
Returns
NetworkX graph
A spanner of the given graph with the given stretch.
Raises
ValueError
If a stretch less than 1 is given.
Notes
This function implements the spanner algorithm by Baswana and Sen, see [1].
This algorithm is a randomized las vegas algorithm: The expected running time is O(km) where k = (stretch + 1) //
2 and m is the number of edges in G. The returned graph is always a spanner of the given graph with the specified
stretch. For weighted graphs the number of edges in the spanner is O(k * n^(1 + 1 / k)) where k is defined as above
and n is the number of nodes in G. For unweighted graphs the number of edges is O(n^(1 + 1 / k) + kn).

References
[1] S. Baswana, S. Sen. A Simple and Linear Time Randomized Algorithm for Computing Sparse Spanners in
Weighted Graphs. Random Struct. Algorithms 30(4): 532-563 (2007).
3.58 Structural holes
Functions for computing measures of structural holes.
constraint(G[, nodes, weight]) Returns the constraint on all nodes in the graph G.
effective_size(G[, nodes, weight]) Returns the effective size of all nodes in the graph G.
local_constraint(G, u, v[, weight]) Returns the local constraint on the node u with respect to
the node v in the graph G.
3.58.1 constraint
constraint(G, nodes=None, weight=None)

Returns the constraint on all nodes in the graph G.
The constraint is a measure of the extent to which a node v is invested in those nodes that are themselves invested
in the neighbors of v. Formally, the constraint on v, denoted c(v), is defined by
∑
c(v) = ℓ(v, w)
w∈N (v)\{v}
where N (v) is the subset of the neighbors of v that are either predecessors or successors of v and ℓ(v, w) is the
local constraint on v with respect to w [1]. For the definition of local constraint, see local_constraint().
Parameters
G
[NetworkX graph] The graph containing v. This can be either directed or undirected.
nodes
[container, optional] Container of nodes in the graph G to compute the constraint. If None,
the constraint of every node is computed.
weight
name of the edge attribute used as weight.
Returns
dict
Dictionary with nodes as keys and the constraint on the node as values.
See also:
local_constraint
3.58. Structural holes 689

References
[1]
3.58.2 effective_size
effective_size(G, nodes=None, weight=None)

Returns the effective size of all nodes in the graph G.
The effective size of a node’s ego network is based on the concept of redundancy. A person’s ego network has
redundancy to the extent that her contacts are connected to each other as well. The nonredundant part of a person’s
relationships it’s the effective size of her ego network [1]. Formally, the effective size of a node u, denoted e(u), is
defined by
 
∑ ∑
e(u) = 1 − puw mvw 
v∈N (u)\{u} w∈N (v)
where N (u) is the set of neighbors of u and puw is the normalized mutual weight of the (directed or undirected)
edges joining u and v, for each vertex u and v [1]. And mvw is the mutual weight of v and w divided by v highest
mutual weight with any of its neighbors. The mutual weight of u and v is the sum of the weights of edges joining
them (edge weights are assumed to be one if the graph is unweighted).
For the case of unweighted and undirected graphs, Borgatti proposed a simplified formula to compute effective size
[2]
2t
e(u) = n −
n
where t is the number of ties in the ego network (not including ties to ego) and n is the number of nodes (excluding
ego).
Parameters
G
[NetworkX graph] The graph containing v. Directed graphs are treated like undirected graphs
when computing neighbors of v.
nodes
[container, optional] Container of nodes in the graph G to compute the effective size. If None,
the effective size of every node is computed.
weight
Returns
dict
Dictionary with nodes as keys and the effective size of the node as values.
See also:
constraint

Notes
Burt also defined the related concept of efficiency of a node’s ego network, which is its effective size divided by the
degree of that node [1]. So you can easily compute efficiency:
>>> G.add_edges_from([(0, 1), (0, 2), (1, 0), (2, 1)])
>>> esize = nx.effective_size(G)
>>> efficiency = {n: v / G.degree(n) for n, v in esize.items()}
References
[1], [2]
3.58.3 local_constraint
local_constraint(G, u, v, weight=None)
Returns the local constraint on the node u with respect to the node v in the graph G.
Formally, the local constraint on u with respect to v, denoted ℓ(v), is defined by
 2
∑
ℓ(u, v) = puv + puw pwv  ,
w∈N (v)
where N (v) is the set of neighbors of v and puv is the normalized mutual weight of the (directed or undirected)
edges joining u and v, for each vertex u and v [1]. The mutual weight of u and v is the sum of the weights of edges
joining them (edge weights are assumed to be one if the graph is unweighted).
Parameters
G
[NetworkX graph] The graph containing u and v. This can be either directed or undirected.
u
[node] A node in the graph G.
v
[node] A node in the graph G.
weight
Returns
float
The constraint of the node v in the graph G.
See also:
constraint
3.58. Structural holes 691

References
[1]
3.59 Summarization
Graph summarization finds smaller representations of graphs resulting in faster runtime of algorithms, reduced storage
needs, and noise reduction. Summarization has applications in areas such as visualization, pattern mining, clustering
and community detection, and more. Core graph summarization techniques are grouping/aggregation, bit-compression,
simplification/sparsification, and influence based. Graph summarization algorithms often produce either summary graphs
in the form of supergraphs or sparsified graphs, or a list of independent structures. Supergraphs are the most common
product, which consist of supernodes and original nodes and are connected by edges and superedges, which represent
aggregate edges between nodes and supernodes.
Grouping/aggregation based techniques compress graphs by representing close/connected nodes and edges in a graph by
a single node/edge in a supergraph. Nodes can be grouped together into supernodes based on their structural similarities
or proximity within a graph to reduce the total number of nodes in a graph. Edge-grouping techniques group edges into
lossy/lossless nodes called compressor or virtual nodes to reduce the total number of edges in a graph. Edge-grouping
techniques can be lossless, meaning that they can be used to re-create the original graph, or techniques can be lossy,
requiring less space to store the summary graph, but at the expense of lower recontruction accuracy of the original graph.
Bit-compression techniques minimize the amount of information needed to describe the original graph, while revealing
structural patterns in the original graph. The two-part minimum description length (MDL) is often used to represent the
model and the original graph in terms of the model. A key difference between graph compression and graph summarization
is that graph summarization focuses on finding structural patterns within the original graph, whereas graph compression
focuses on compressions the original graph to be as small as possible. NOTE: Some bit-compression methods exist solely
to compress a graph without creating a summary graph or finding comprehensible structural patterns.
Simplification/Sparsification techniques attempt to create a sparse representation of a graph by removing unimportant
nodes and edges from the graph. Sparsified graphs differ from supergraphs created by grouping/aggregation by only
containing a subset of the original nodes and edges of the original graph.
Influence based techniques aim to find a high-level description of influence propagation in a large graph. These methods
are scarce and have been mostly applied to social graphs.
dedensification is a grouping/aggregation based technique to compress the neighborhoods around high-degree nodes in
unweighted graphs by adding compressor nodes that summarize multiple edges of the same type to high-degree nodes
(nodes with a degree greater than a given threshold). Dedensification was developed for the purpose of increasing perfor-
mance of query processing around high-degree nodes in graph databases and enables direct operations on the compressed
graph. The structural patterns surrounding high-degree nodes in the original is preserved while using fewer edges and
adding a small number of compressor nodes. The degree of nodes present in the original graph is also preserved. The
current implementation of dedensification supports graphs with one edge type.
For more information on graph summarization, see Graph Summarization Methods and Applications: A Survey
dedensify(G, threshold[, prefix, copy]) Compresses neighborhoods around high-degree nodes

snap_aggregation(G, node_attributes[, ...]) Creates a summary graph based on attributes and connec-
tivity.

3.59.1 dedensify
dedensify(G, threshold, prefix=None, copy=True)

Compresses neighborhoods around high-degree nodes
Reduces the number of edges to high-degree nodes by adding compressor nodes that summarize multiple edges
of the same type to high-degree nodes (nodes with a degree greater than a given threshold). Dedensification also
has the added benefit of reducing the number of edges around high-degree nodes. The implementation currently
supports graphs with a single edge type.
Parameters
G: graph
A networkx graph
threshold: int
Minimum degree threshold of a node to be considered a high degree node. The threshold must
be greater than or equal to 2.
prefix: str or None, optional (default: None)
An optional prefix for denoting compressor nodes
copy: bool, optional (default: True)
Indicates if dedensification should be done inplace
Returns
dedensified networkx graph
[(graph, set)] 2-tuple of the dedensified graph and set of compressor nodes
Notes
According to the algorithm in [1], removes edges in a graph by compressing/decompressing the neighborhoods
around high degree nodes by adding compressor nodes that summarize multiple edges of the same type to high-
degree nodes. Dedensification will only add a compressor node when doing so will reduce the total number of
edges in the given graph. This implementation currently supports graphs with a single edge type.
References
[1]
Examples
Dedensification will only add compressor nodes when doing so would result in fewer edges:
>>> original_graph = nx.DiGraph()

>>> original_graph.add_nodes_from(
... ["1", "2", "3", "4", "5", "6", "A", "B", "C"]
... )
>>> original_graph.add_edges_from(
... [
... ("1", "C"), ("1", "B"),
... ("2", "C"), ("2", "B"), ("2", "A"),
... ("3", "B"), ("3", "A"), ("3", "6"),
... ("4", "C"), ("4", "B"), ("4", "A"),
3.59. Summarization 693


... ("5", "B"), ("5", "A"),
... ("6", "5"),
... ("A", "6")
... ]
... )
>>> c_graph, c_nodes = nx.dedensify(original_graph, threshold=2)
>>> original_graph.number_of_edges()
15
>>> c_graph.number_of_edges()
14
A dedensified, directed graph can be “densified” to reconstruct the original graph:

>>> original_graph = nx.DiGraph()
>>> original_graph.add_nodes_from(
... ["1", "2", "3", "4", "5", "6", "A", "B", "C"]
... )
>>> original_graph.add_edges_from(
... [
... ("1", "C"), ("1", "B"),
... ("2", "C"), ("2", "B"), ("2", "A"),
... ("3", "B"), ("3", "A"), ("3", "6"),
... ("4", "C"), ("4", "B"), ("4", "A"),
... ("5", "B"), ("5", "A"),
... ("6", "5"),
... ("A", "6")
... ]
... )
>>> c_graph, c_nodes = nx.dedensify(original_graph, threshold=2)
>>> # re-densifies the compressed graph into the original graph
>>> for c_node in c_nodes:
... all_neighbors = set(nx.all_neighbors(c_graph, c_node))
... out_neighbors = set(c_graph.neighbors(c_node))
... for out_neighbor in out_neighbors:
... c_graph.remove_edge(c_node, out_neighbor)
... in_neighbors = all_neighbors - out_neighbors
... for in_neighbor in in_neighbors:
... c_graph.remove_edge(in_neighbor, c_node)
... for out_neighbor in out_neighbors:
... c_graph.add_edge(in_neighbor, out_neighbor)
... c_graph.remove_node(c_node)
...
>>> nx.is_isomorphic(original_graph, c_graph)
True
3.59.2 snap_aggregation
snap_aggregation(G, node_attributes, edge_attributes=(), prefix='Supernode-', supernode_attribute='group',

superedge_attribute='types')
Creates a summary graph based on attributes and connectivity.
This function uses the Summarization by Grouping Nodes on Attributes and Pairwise edges (SNAP) algorithm
for summarizing a given graph by grouping nodes by node attributes and their edge attributes into supernodes in a
summary graph. This name SNAP should not be confused with the Stanford Network Analysis Project (SNAP).
Here is a high-level view of how this algorithm works:

1) Group nodes by node attribute values.

2) Iteratively split groups until all nodes in each group have edges to nodes in the same groups. That is, until all the
groups are homogeneous in their member nodes’ edges to other groups. For example, if all the nodes in group A
only have edge to nodes in group B, then the group is homogeneous and does not need to be split. If all nodes in
group B have edges with nodes in groups {A, C}, but some also have edges with other nodes in B, then group B is
not homogeneous and needs to be split into groups have edges with {A, C} and a group of nodes having edges with
{A, B, C}. This way, viewers of the summary graph can assume that all nodes in the group have the exact same
node attributes and the exact same edges.
3) Build the output summary graph, where the groups are represented by super-nodes. Edges represent the edges
shared between all the nodes in each respective groups.
A SNAP summary graph can be used to visualize graphs that are too large to display or visually analyze, or to
efficiently identify sets of similar nodes with similar connectivity patterns to other sets of similar nodes based on
specified node and/or edge attributes in a graph.
Parameters
G: graph
Networkx Graph to be summarized
edge_attributes: iterable, optional
An iterable of the edge attributes considered in the summarization process. If provided, unique
combinations of the attribute values found in the graph are used to determine the edge types
in the graph. If not provided, all edges are considered to be of the same type.
prefix: str
The prefix used to denote supernodes in the summary graph. Defaults to ‘Supernode-‘.
supernode_attribute: str
The node attribute for recording the supernode groupings of nodes. Defaults to ‘group’.
superedge_attribute: str
The edge attribute for recording the edge types of multiple edges. Defaults to ‘types’.
Returns
networkx.Graph: summary graph
Notes
The summary graph produced is called a maximum Attribute-edge compatible (AR-compatible) grouping. Accord-
ing to [1], an AR-compatible grouping means that all nodes in each group have the same exact node attribute values
and the same exact edges and edge types to one or more nodes in the same groups. The maximal AR-compatible
grouping is the grouping with the minimal cardinality.
The AR-compatible grouping is the most detailed grouping provided by any of the SNAP algorithms.
3.59. Summarization 695

References
[1]
Examples
SNAP aggregation takes a graph and summarizes it in the context of user-provided node and edge attributes such
that a viewer can more easily extract and analyze the information represented by the graph
>>> nodes = {
... "A": dict(color="Red"),
... "B": dict(color="Red"),
... "C": dict(color="Red"),
... "D": dict(color="Red"),
... "E": dict(color="Blue"),
... "F": dict(color="Blue"),
... }
>>> edges = [
... ("A", "E", "Strong"),
... ("B", "F", "Strong"),
... ("C", "E", "Weak"),
... ("D", "F", "Weak"),
... ]
>>> G = nx.Graph()
>>> for node in nodes:
... attributes = nodes[node]
... G.add_node(node, **attributes)
...
>>> for source, target, type in edges:
... G.add_edge(source, target, type=type)
...
>>> node_attributes = ('color', )
>>> edge_attributes = ('type', )
>>> summary_graph = nx.snap_aggregation(G, node_attributes=node_attributes, edge_
,→attributes=edge_attributes)
3.60 Swap
Swap edges in a graph.
double_edge_swap(G[, nswap, max_tries, seed]) Swap two edges in the graph while keeping the node de-
grees fixed.
directed_edge_swap(G, *[, nswap, max_tries, ...]) Swap three edges in a directed graph while keeping the
node degrees fixed.
connected_double_edge_swap(G[, nswap, ...]) Attempts the specified number of double-edge swaps in
the graph G.

3.60.1 double_edge_swap
double_edge_swap(G, nswap=1, max_tries=100, seed=None)

Swap two edges in the graph while keeping the node degrees fixed.
A double-edge swap removes two randomly chosen edges u-v and x-y and creates the new edges u-x and v-y:
u--v u v
becomes | |
x--y x y
If either the edge u-x or v-y already exist no swap is performed and another attempt is made to find a suitable edge
pair.
Parameters
G
[graph] An undirected graph
nswap
[integer (optional, default=1)] Number of double-edge swaps to perform
max_tries
[integer (optional)] Maximum number of attempts to swap edges
seed
Randomness.
Returns
G
[graph] The graph after double edge swaps.
Notes
Does not enforce any connectivity constraints.

The graph G is modified in place.
3.60.2 directed_edge_swap
directed_edge_swap(G, *, nswap=1, max_tries=100, seed=None)

Swap three edges in a directed graph while keeping the node degrees fixed.
A directed edge swap swaps three edges such that a -> b -> c -> d becomes a -> c -> b -> d. This pattern of
swapping allows all possible states with the same in- and out-degree distribution in a directed graph to be reached.
If the swap would create parallel edges (e.g. if a -> c already existed in the previous example), another attempt is
made to find a suitable trio of edges.
Parameters
G
[DiGraph] A directed graph
nswap
[integer (optional, default=1)] Number of three-edge (directed) swaps to perform
3.60. Swap 697

max_tries
[integer (optional, default=100)] Maximum number of attempts to swap edges
seed
Randomness.
Returns
G
[DiGraph] The graph after the edges are swapped.
Raises
NetworkXError
If G is not directed, or If nswap > max_tries, or If there are fewer than 4 nodes in G
NetworkXAlgorithmError
If the number of swap attempts exceeds max_tries before nswap swaps are made
Notes
Does not enforce any connectivity constraints.

The graph G is modified in place.
References
[1], [2]
3.60.3 connected_double_edge_swap
connected_double_edge_swap(G, nswap=1, _window_threshold=3, seed=None)

Attempts the specified number of double-edge swaps in the graph G.
A double-edge swap removes two randomly chosen edges (u, v) and (x, y) and creates the new edges (u,
x) and (v, y):
u--v u v
becomes | |
x--y x y
If either (u, x) or (v, y) already exist, then no swap is performed so the actual number of swapped edges is
always at most nswap.
Parameters
G
[graph] An undirected graph
nswap
[integer (optional, default=1)] Number of double-edge swaps to perform
_window_threshold
[integer] The window size below which connectedness of the graph will be checked after each
swap.

The “window” in this function is a dynamically updated integer that represents the number of
swap attempts to make before checking if the graph remains connected. It is an optimization
used to decrease the running time of the algorithm in exchange for increased complexity of
implementation.
If the window size is below this threshold, then the algorithm checks after each swap if the
graph remains connected by checking if there is a path joining the two nodes whose edge was
just removed. If the window size is above this threshold, then the algorithm performs do all
the swaps in the window and only then check if the graph is still connected.
seed
Randomness.
Returns
int
The number of successful swaps
Raises
NetworkXError
If the input graph is not connected, or if the graph has fewer than four nodes.
Notes
The initial graph G must be connected, and the resulting graph is connected. The graph G is modified in place.
References
[1]
3.61 Threshold Graphs
Threshold Graphs - Creation, manipulation and identification.
find_threshold_graph(G[, create_using]) Returns a threshold subgraph that is close to largest in G.

is_threshold_graph(G) Returns True if G is a threshold graph.
3.61.1 find_threshold_graph
find_threshold_graph(G, create_using=None)
Returns a threshold subgraph that is close to largest in G.
The threshold graph will contain the largest degree node in G.
Parameters
G
[NetworkX graph instance] An instance of Graph, or MultiDiGraph
create_using
[NetworkX graph class or None (default), optional] Type of graph to use when constructing
the threshold graph. If None, infer the appropriate graph type from the input.
3.61. Threshold Graphs 699

Returns
graph
A graph instance representing the threshold graph
References
[1]
Examples
>>> from networkx.algorithms.threshold import find_threshold_graph

>>> T = find_threshold_graph(G)
>>> T.nodes # may vary
NodeView((7, 8, 5, 6))
3.61.2 is_threshold_graph
is_threshold_graph(G)
Returns True if G is a threshold graph.
Parameters
G
[NetworkX graph instance] An instance of Graph, DiGraph, MultiGraph or Multi-
DiGraph
Returns
bool
True if G is a threshold graph, False otherwise.
References
[1]
Examples
>>> from networkx.algorithms.threshold import is_threshold_graph

>>> is_threshold_graph(G)
True
>>> is_threshold_graph(G)
False

3.62 Tournament
Functions concerning tournament graphs.

A tournament graph is a complete oriented graph. In other words, it is a directed graph in which there is exactly one
directed edge joining each pair of distinct nodes. For each function in this module that accepts a graph as input, you must
provide a tournament graph. The responsibility is on the caller to ensure that the graph is a tournament graph.
To access the functions in this module, you must access them through the networkx.algorithms.tournament
module:
>>> from networkx.algorithms import tournament

>>> G = nx.DiGraph([(0, 1), (1, 2), (2, 0)])
>>> tournament.is_tournament(G)
True
hamiltonian_path(G) Returns a Hamiltonian path in the given tournament

graph.
is_reachable(G, s, t) Decides whether there is a path from s to t in the tour-
nament.
is_strongly_connected(G) Decides whether the given tournament is strongly con-
nected.
is_tournament(G) Returns True if and only if G is a tournament.
random_tournament(n[, seed]) Returns a random tournament graph on n nodes.
score_sequence(G) Returns the score sequence for the given tournament
graph.
3.62.1 hamiltonian_path
hamiltonian_path(G)
Returns a Hamiltonian path in the given tournament graph.
Each tournament has a Hamiltonian path. If furthermore, the tournament is strongly connected, then the returned
Hamiltonian path is a Hamiltonian cycle (by joining the endpoints of the path).
Parameters
G
[NetworkX graph] A directed graph representing a tournament.
Returns
path
[list] A list of nodes which form a Hamiltonian path in G.
3.62. Tournament 701

Notes
This is a recursive implementation with an asymptotic running time of O(n2 ), ignoring multiplicative polyloga-
rithmic factors, where n is the number of nodes in the graph.
Examples

>>> G = nx.DiGraph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3), (2, 3)])
>>> tournament.hamiltonian_path(G)
[0, 1, 2, 3]
3.62.2 is_reachable
is_reachable(G, s, t)
Decides whether there is a path from s to t in the tournament.
This function is more theoretically efficient than the reachability checks than the shortest path algorithms in
networkx.algorithms.shortest_paths.
The given graph must be a tournament, otherwise this function’s behavior is undefined.
Parameters
G
s
t
Returns
bool
Whether there is a path from s to t in G.
Notes
Although this function is more theoretically efficient than the generic shortest path functions, a speedup requires
the use of parallelism. Though it may in the future, the current implementation does not use parallelism, thus you
may not see much of a speedup.
This algorithm comes from [1].

References
[1]
Examples

>>> G = nx.DiGraph([(1, 0), (1, 3), (1, 2), (2, 3), (2, 0), (3, 0)])
>>> tournament.is_reachable(G, 1, 3)
True
>>> tournament.is_reachable(G, 3, 2)
False
3.62.3 is_strongly_connected
is_strongly_connected(G)
Decides whether the given tournament is strongly connected.
This function is more theoretically efficient than the is_strongly_connected() function.
The given graph must be a tournament, otherwise this function’s behavior is undefined.
Parameters
G
Returns
bool
Whether the tournament is strongly connected.
Notes
Although this function is more theoretically efficient than the generic strong connectivity function, a speedup re-
quires the use of parallelism. Though it may in the future, the current implementation does not use parallelism,
thus you may not see much of a speedup.
This algorithm comes from [1].
References
[1]
3.62. Tournament 703

Examples

>>> G = nx.DiGraph([(0, 1), (0, 2), (0, 3), (1, 2), (1, 3), (2, 3), (3, 0)])
>>> tournament.is_strongly_connected(G)
True
>>> tournament.is_strongly_connected(G)
False
3.62.4 is_tournament
is_tournament(G)
Returns True if and only if G is a tournament.
A tournament is a directed graph, with neither self-loops nor multi-edges, in which there is exactly one directed
edge joining each pair of distinct nodes.
Parameters
G
Returns
bool
Whether the given graph is a tournament graph.
Notes
Some definitions require a self-loop on each node, but that is not the convention used here.
Examples

>>> G = nx.DiGraph([(0, 1), (1, 2), (2, 0)])
>>> tournament.is_tournament(G)
True
3.62.5 random_tournament
random_tournament(n, seed=None)
Returns a random tournament graph on n nodes.
Parameters
n
[int] The number of nodes in the returned graph.
seed
Randomness.
Returns

G
[DiGraph] A tournament on n nodes, with exactly one directed edge joining each pair of
distinct nodes.
Notes
This algorithm adds, for each pair of distinct nodes, an edge with uniformly random orientation. In other words,
binom{n}{2} flips of an unbiased coin decide the orientations of the edges in the graph.
3.62.6 score_sequence
score_sequence(G)
Returns the score sequence for the given tournament graph.
The score sequence is the sorted list of the out-degrees of the nodes of the graph.
Parameters
G
Returns
list
A sorted list of the out-degrees of the nodes of G.
Examples

>>> G = nx.DiGraph([(1, 0), (1, 3), (0, 2), (0, 3), (2, 1), (3, 2)])
>>> tournament.score_sequence(G)
[1, 1, 2, 2]
3.63 Traversal
3.63.1 Depth First Search
Basic algorithms for depth-first searching the nodes of a graph.
3.63. Traversal 705

dfs_edges(G[, source, depth_limit]) Iterate over edges in a depth-first-search (DFS).

dfs_tree(G[, source, depth_limit]) Returns oriented tree constructed from a depth-first-
search from source.
dfs_predecessors(G[, source, depth_limit]) Returns dictionary of predecessors in depth-first-search
from source.
dfs_successors(G[, source, depth_limit]) Returns dictionary of successors in depth-first-search
from source.
dfs_preorder_nodes(G[, source, depth_limit]) Generate nodes in a depth-first-search pre-ordering start-
ing at source.
dfs_postorder_nodes(G[, source, depth_limit]) Generate nodes in a depth-first-search post-ordering start-
ing at source.
dfs_labeled_edges(G[, source, depth_limit]) Iterate over edges in a depth-first-search (DFS) labeled by
type.
dfs_edges
dfs_edges(G, source=None, depth_limit=None)

Iterate over edges in a depth-first-search (DFS).
Perform a depth-first-search over the nodes of G and yield the edges in order. This may not generate all edges in G
(see edge_dfs).
Parameters
G
[NetworkX graph]
source
[node, optional] Specify starting node for depth-first search and yield edges in the component
reachable from source.
depth_limit
[int, optional (default=len(G))] Specify the maximum search depth.
Yields
edge: 2-tuple of nodes
Yields edges resulting from the depth-first-search.
See also:
dfs_preorder_nodes
dfs_postorder_nodes
dfs_labeled_edges
edge_dfs()
bfs_edges()

Notes
If a source is not specified then a source is chosen arbitrarily and repeatedly until all components in the graph are
searched.
The implementation of this function is adapted from David Eppstein’s depth-first search function in PADS [1], with
modifications to allow depth limits based on the Wikipedia article “Depth-limited search” [2].
References
[1], [2]
Examples
>>> list(nx.dfs_edges(G, source=0))
[(0, 1), (1, 2), (2, 3), (3, 4)]
>>> list(nx.dfs_edges(G, source=0, depth_limit=2))
[(0, 1), (1, 2)]
dfs_tree
dfs_tree(G, source=None, depth_limit=None)

Returns oriented tree constructed from a depth-first-search from source.
Parameters
G
[NetworkX graph]
source
[node, optional] Specify starting node for depth-first search.
depth_limit
Returns
T
[NetworkX DiGraph] An oriented tree
See also:
dfs_preorder_nodes
dfs_postorder_nodes
dfs_labeled_edges
edge_dfs
bfs_tree
3.63. Traversal 707

Examples
>>> T = nx.dfs_tree(G, source=0, depth_limit=2)
>>> list(T.edges())
[(0, 1), (1, 2)]
>>> T = nx.dfs_tree(G, source=0)
>>> list(T.edges())
[(0, 1), (1, 2), (2, 3), (3, 4)]
dfs_predecessors
dfs_predecessors(G, source=None, depth_limit=None)

Returns dictionary of predecessors in depth-first-search from source.
Parameters
G
[NetworkX graph]
source
depth_limit
Returns
pred: dict
A dictionary with nodes as keys and predecessor nodes as values.
See also:
dfs_preorder_nodes
dfs_postorder_nodes
dfs_labeled_edges
edge_dfs
bfs_tree
Notes
searched.
The implementation of this function is adapted from David Eppstein’s depth-first search function in PADS, with
modifications to allow depth limits based on the Wikipedia article “Depth-limited search”.

Examples
>>> nx.dfs_predecessors(G, source=0)
{1: 0, 2: 1, 3: 2}
>>> nx.dfs_predecessors(G, source=0, depth_limit=2)
{1: 0, 2: 1}
dfs_successors
dfs_successors(G, source=None, depth_limit=None)

Returns dictionary of successors in depth-first-search from source.
Parameters
G
[NetworkX graph]
source
depth_limit
Returns
succ: dict
A dictionary with nodes as keys and list of successor nodes as values.
See also:
dfs_preorder_nodes
dfs_postorder_nodes
dfs_labeled_edges
edge_dfs
bfs_tree
Notes
searched.
Examples
>>> nx.dfs_successors(G, source=0)
{0: [1], 1: [2], 2: [3], 3: [4]}
>>> nx.dfs_successors(G, source=0, depth_limit=2)
{0: [1], 1: [2]}
3.63. Traversal 709

dfs_preorder_nodes
dfs_preorder_nodes(G, source=None, depth_limit=None)

Generate nodes in a depth-first-search pre-ordering starting at source.
Parameters
G
[NetworkX graph]
source
[node, optional] Specify starting node for depth-first search and return nodes in the component
depth_limit
Returns
nodes: generator
A generator of nodes in a depth-first-search pre-ordering.
See also:
dfs_edges
dfs_postorder_nodes
dfs_labeled_edges
bfs_edges
Notes
searched.
Examples
>>> list(nx.dfs_preorder_nodes(G, source=0))
[0, 1, 2, 3, 4]
>>> list(nx.dfs_preorder_nodes(G, source=0, depth_limit=2))
[0, 1, 2]
dfs_postorder_nodes
dfs_postorder_nodes(G, source=None, depth_limit=None)

Generate nodes in a depth-first-search post-ordering starting at source.
Parameters
G
[NetworkX graph]

source
depth_limit
Returns
nodes: generator
A generator of nodes in a depth-first-search post-ordering.
See also:
dfs_edges
dfs_preorder_nodes
dfs_labeled_edges
edge_dfs
bfs_tree
Notes
searched.
Examples
>>> list(nx.dfs_postorder_nodes(G, source=0))
[4, 3, 2, 1, 0]
>>> list(nx.dfs_postorder_nodes(G, source=0, depth_limit=2))
[1, 0]
dfs_labeled_edges
dfs_labeled_edges(G, source=None, depth_limit=None)

Iterate over edges in a depth-first-search (DFS) labeled by type.
Parameters
G
[NetworkX graph]
source
[node, optional] Specify starting node for depth-first search and return edges in the component
depth_limit
Returns
3.63. Traversal 711

edges: generator
A generator of triples of the form (u, v, d), where (u, v) is the edge being explored in the depth-
first search and d is one of the strings ‘forward’, ‘nontree’, ‘reverse’, or ‘reverse-depth_limit’.
A ‘forward’ edge is one in which u has been visited but v has not. A ‘nontree’ edge is one in
which both u and v have been visited but the edge is not in the DFS tree. A ‘reverse’ edge
is one in which both u and v have been visited and the edge is in the DFS tree. When the
depth_limit is reached via a ‘forward’ edge, a ‘reverse’ edge is immediately generated
rather than the subtree being explored. To indicate this flavor of ‘reverse’ edge, the string
yielded is ‘reverse-depth_limit’.
See also:
dfs_edges
dfs_preorder_nodes
dfs_postorder_nodes
Notes
searched.
Examples
The labels reveal the complete transcript of the depth-first search algorithm in more detail than, for example,
dfs_edges():
>>> from pprint import pprint

>>>
>>> G = nx.DiGraph([(0, 1), (1, 2), (2, 1)])
>>> pprint(list(nx.dfs_labeled_edges(G, source=0)))
[(0, 0, 'forward'),
(0, 1, 'forward'),
(1, 2, 'forward'),
(2, 1, 'nontree'),
(1, 2, 'reverse'),
(0, 1, 'reverse'),
(0, 0, 'reverse')]
3.63.2 Breadth First Search
Basic algorithms for breadth-first searching the nodes of a graph.

bfs_edges(G, source[, reverse, depth_limit, ...]) Iterate over edges in a breadth-first-search starting at
source.
bfs_layers(G, sources) Returns an iterator of all the layers in breadth-first search
traversal.
bfs_tree(G, source[, reverse, depth_limit, ...]) Returns an oriented tree constructed from of a breadth-
first-search starting at source.
bfs_predecessors(G, source[, depth_limit, ...]) Returns an iterator of predecessors in breadth-first-search
from source.
bfs_successors(G, source[, depth_limit, ...]) Returns an iterator of successors in breadth-first-search
from source.
descendants_at_distance(G, source, distance) Returns all nodes at a fixed distance from source in
G.
bfs_edges
bfs_edges(G, source, reverse=False, depth_limit=None, sort_neighbors=None)

Iterate over edges in a breadth-first-search starting at source.
Parameters
G
[NetworkX graph]
source
[node] Specify starting node for breadth-first search; this function iterates over only those edges
in the component reachable from this node.
reverse
[bool, optional] If True traverse a directed graph in the reverse direction
depth_limit
[int, optional(default=len(G))] Specify the maximum search depth
sort_neighbors
[function] A function that takes the list of neighbors of given node as input, and returns an
iterator over these neighbors but with custom ordering.
Yields
edge: 2-tuple of nodes
Yields edges resulting from the breadth-first search.
See also:
bfs_tree
dfs_edges()
edge_bfs()
3.63. Traversal 713

Notes
The naming of this function is very similar to edge_bfs(). The difference is that edge_bfs yields edges
even if they extend back to an already explored node while this generator yields the edges of the tree that results
from a breadth-first-search (BFS) so no edges are reported if they extend to already explored nodes. That means
edge_bfs reports all edges while bfs_edges only reports those traversed by a node-based BFS. Yet another
description is that bfs_edges reports the edges traversed during BFS while edge_bfs reports all edges in the
order they are explored.
Based on the breadth-first search implementation in PADS [1] by D. Eppstein, July 2004; with modifications to
allow depth limits as described in [2].
References
[1], [2]
Examples
To get the edges in a breadth-first search:
>>> list(nx.bfs_edges(G, 0))
[(0, 1), (1, 2)]
>>> list(nx.bfs_edges(G, source=0, depth_limit=1))
[(0, 1)]
To get the nodes in a breadth-first search order:
>>> root = 2
>>> edges = nx.bfs_edges(G, root)
>>> nodes = [root] + [v for u, v in edges]
>>> nodes
[2, 1, 0]
bfs_layers
bfs_layers(G, sources)
Returns an iterator of all the layers in breadth-first search traversal.
Parameters
G
[NetworkX graph] A graph over which to find the layers using breadth-first search.
sources
[node in G or list of nodes in G] Specify starting nodes for single source or multiple sources
breadth-first search
Yields
layer: list of nodes
Yields list of nodes at the same distance from sources

Examples
>>> dict(enumerate(nx.bfs_layers(G, [0, 4])))
{0: [0, 4], 1: [1, 3], 2: [2]}
>>> H = nx.Graph()
>>> H.add_edges_from([(0, 1), (0, 2), (1, 3), (1, 4), (2, 5), (2, 6)])
>>> dict(enumerate(nx.bfs_layers(H, [1])))
{0: [1], 1: [0, 3, 4], 2: [2], 3: [5, 6]}
>>> dict(enumerate(nx.bfs_layers(H, [1, 6])))
{0: [1, 6], 1: [0, 3, 4, 2], 2: [5]}
bfs_tree
bfs_tree(G, source, reverse=False, depth_limit=None, sort_neighbors=None)

Returns an oriented tree constructed from of a breadth-first-search starting at source.
Parameters
G
[NetworkX graph]
source
[node] Specify starting node for breadth-first search
reverse
[bool, optional] If True traverse a directed graph in the reverse direction
depth_limit
sort_neighbors
Returns
T: NetworkX DiGraph
An oriented tree
See also:
dfs_tree
bfs_edges
edge_bfs
Notes
Based on http://www.ics.uci.edu/~eppstein/PADS/BFS.py by D. Eppstein, July 2004. The modifications to allow

depth limits based on the Wikipedia article “Depth-limited-search”.
3.63. Traversal 715

Examples
>>> print(list(nx.bfs_tree(G, 1).edges()))
[(1, 0), (1, 2)]
>>> H = nx.Graph()
>>> nx.add_path(H, [0, 1, 2, 3, 4, 5, 6])
>>> nx.add_path(H, [2, 7, 8, 9, 10])
>>> print(sorted(list(nx.bfs_tree(H, source=3, depth_limit=3).edges())))
[(1, 0), (2, 1), (2, 7), (3, 2), (3, 4), (4, 5), (5, 6), (7, 8)]
bfs_predecessors
bfs_predecessors(G, source, depth_limit=None, sort_neighbors=None)

Returns an iterator of predecessors in breadth-first-search from source.
Parameters
G
[NetworkX graph]
source
depth_limit
sort_neighbors
Returns
pred: iterator
(node, predecessor) iterator where predecessor is the predecessor of node in a breadth
first search starting from source.
See also:
bfs_tree
bfs_edges
edge_bfs
Notes
Based on http://www.ics.uci.edu/~eppstein/PADS/BFS.py by D. Eppstein, July 2004. The modifications to allow


Examples
>>> print(dict(nx.bfs_predecessors(G, 0)))
{1: 0, 2: 1}
>>> H = nx.Graph()
>>> H.add_edges_from([(0, 1), (0, 2), (1, 3), (1, 4), (2, 5), (2, 6)])
>>> print(dict(nx.bfs_predecessors(H, 0)))
{1: 0, 2: 0, 3: 1, 4: 1, 5: 2, 6: 2}
>>> M = nx.Graph()
>>> nx.add_path(M, [0, 1, 2, 3, 4, 5, 6])
>>> nx.add_path(M, [2, 7, 8, 9, 10])
>>> print(sorted(nx.bfs_predecessors(M, source=1, depth_limit=3)))
[(0, 1), (2, 1), (3, 2), (4, 3), (7, 2), (8, 7)]
>>> N = nx.DiGraph()
>>> nx.add_path(N, [0, 1, 2, 3, 4, 7])
>>> nx.add_path(N, [3, 5, 6, 7])
>>> print(sorted(nx.bfs_predecessors(N, source=2)))
[(3, 2), (4, 3), (5, 3), (6, 5), (7, 4)]
bfs_successors
bfs_successors(G, source, depth_limit=None, sort_neighbors=None)

Returns an iterator of successors in breadth-first-search from source.
Parameters
G
[NetworkX graph]
source
depth_limit
sort_neighbors
Returns
succ: iterator
(node, successors) iterator where successors is the non-empty list of successors of node
in a breadth first search from source. To appear in the iterator, node must have successors.
See also:
bfs_tree
bfs_edges
edge_bfs
3.63. Traversal 717

Notes
Based on http://www.ics.uci.edu/~eppstein/PADS/BFS.py by D. Eppstein, July 2004.The modifications to allow

Examples
>>> print(dict(nx.bfs_successors(G, 0)))
{0: [1], 1: [2]}
>>> H = nx.Graph()
>>> H.add_edges_from([(0, 1), (0, 2), (1, 3), (1, 4), (2, 5), (2, 6)])
>>> print(dict(nx.bfs_successors(H, 0)))
{0: [1, 2], 1: [3, 4], 2: [5, 6]}
>>> G = nx.Graph()
>>> nx.add_path(G, [0, 1, 2, 3, 4, 5, 6])
>>> nx.add_path(G, [2, 7, 8, 9, 10])
>>> print(dict(nx.bfs_successors(G, source=1, depth_limit=3)))
{1: [0, 2], 2: [3, 7], 3: [4], 7: [8]}
>>> nx.add_path(G, [0, 1, 2, 3, 4, 5])
>>> print(dict(nx.bfs_successors(G, source=3)))
{3: [4], 4: [5]}
descendants_at_distance
descendants_at_distance(G, source, distance)

Returns all nodes at a fixed distance from source in G.
Parameters
G
source
[node in G]
distance
[the distance of the wanted nodes from source]
Returns
set()
The descendants of source in G at the given distance from source
Examples
>>> nx.descendants_at_distance(G, 2, 2)
{0, 4}
>>> H = nx.DiGraph()
>>> H.add_edges_from([(0, 1), (0, 2), (1, 3), (1, 4), (2, 5), (2, 6)])
>>> nx.descendants_at_distance(H, 0, 2)
{3, 4, 5, 6}


{5}
set()
3.63.3 Beam search
Basic algorithms for breadth-first searching the nodes of a graph.
bfs_beam_edges(G, source, value[, width]) Iterates over edges in a beam search.
bfs_beam_edges
bfs_beam_edges(G, source, value, width=None)

Iterates over edges in a beam search.
The beam search is a generalized breadth-first search in which only the “best” w neighbors of the current node are
enqueued, where w is the beam width and “best” is an application-specific heuristic. In general, a beam search with
a small beam width might not visit each node in the graph.
Parameters
G
[NetworkX graph]
source
[node] Starting node for the breadth-first search; this function iterates over only those edges in
the component reachable from this node.
value
[function] A function that takes a node of the graph as input and returns a real number indi-
cating how “good” it is. A higher value means it is more likely to be visited sooner during the
search. When visiting a new node, only the width neighbors with the highest value are
enqueued (in decreasing order of value).
width
[int (default = None)] The beam width for the search. This is the number of neighbors (ordered
by value) to enqueue when visiting each new node.
Yields
edge
Edges in the beam search starting from source, given as a pair of nodes.
3.63. Traversal 719

Examples
To give nodes with, for example, a higher centrality precedence during the search, set the value function to return
the centrality value of the node:
>>> centrality = nx.eigenvector_centrality(G)
>>> source = 0
>>> width = 5
>>> for u, v in nx.bfs_beam_edges(G, source, centrality.get, width):
... print((u, v))
...
(0, 2)
(0, 1)
(0, 8)
(0, 13)
(0, 3)
(2, 32)
(1, 30)
(8, 33)
(3, 7)
(32, 31)
(31, 28)
(31, 25)
(25, 23)
(25, 24)
(23, 29)
(23, 27)
(29, 26)
3.63.4 Depth First Search on Edges
Algorithms for a depth-first traversal of edges in a graph.
edge_dfs(G[, source, orientation]) A directed, depth-first-search of edges in G, beginning at

source.
edge_dfs
edge_dfs(G, source=None, orientation=None)

A directed, depth-first-search of edges in G, beginning at source.
Yield the edges of G in a depth-first-search order continuing until all edges are generated.
Parameters
G
source
orientation

Yields
edge
[directed edge] A directed edge indicating the path taken by the depth-first traversal. For
graphs, edge is of the form (u, v) where u and v are the tail and head of the edge as
determined by the traversal. For multigraphs, edge is of the form (u, v, key), where
key is the key of the edge. When the graph is directed, then u and v are always in the order of
the actual directed edge. If orientation is not None then the edge tuple is extended to include
the direction of traversal (‘forward’ or ‘reverse’) on that edge.
See also:
dfs_edges()
Notes
The goal of this function is to visit edges. It differs from the more familiar depth-first traversal of nodes, as provided
by dfs_edges(), in that it does not stop once every node has been visited. In a directed graph with edges [(0,
1), (1, 2), (2, 1)], the edge (2, 1) would not be visited if not for the functionality provided by this function.
Examples
>>> nodes = [0, 1, 2, 3]

>>> edges = [(0, 1), (1, 0), (1, 0), (2, 1), (3, 1)]
>>> list(nx.edge_dfs(nx.Graph(edges), nodes))

[(0, 1), (1, 2), (1, 3)]
>>> list(nx.edge_dfs(nx.DiGraph(edges), nodes))

[(0, 1), (1, 0), (2, 1), (3, 1)]
>>> list(nx.edge_dfs(nx.MultiGraph(edges), nodes))

[(0, 1, 0), (1, 0, 1), (0, 1, 2), (1, 2, 0), (1, 3, 0)]
>>> list(nx.edge_dfs(nx.MultiDiGraph(edges), nodes))

[(0, 1, 0), (1, 0, 0), (1, 0, 1), (2, 1, 0), (3, 1, 0)]
>>> list(nx.edge_dfs(nx.DiGraph(edges), nodes, orientation="ignore"))

[(0, 1, 'forward'), (1, 0, 'forward'), (2, 1, 'reverse'), (3, 1, 'reverse')]
>>> list(nx.edge_dfs(nx.MultiDiGraph(edges), nodes, orientation="ignore"))

[(0, 1, 0, 'forward'), (1, 0, 0, 'forward'), (1, 0, 1, 'reverse'), (2, 1, 0,
,→'reverse'), (3, 1, 0, 'reverse')]
3.63. Traversal 721

3.63.5 Breadth First Search on Edges
Algorithms for a breadth-first traversal of edges in a graph.
edge_bfs(G[, source, orientation]) A directed, breadth-first-search of edges in G, beginning

at source.
edge_bfs
edge_bfs(G, source=None, orientation=None)

A directed, breadth-first-search of edges in G, beginning at source.
Yield the edges of G in a breadth-first-search order continuing until all edges are generated.
Parameters
G
source
orientation
Yields
edge
[directed edge] A directed edge indicating the path taken by the breadth-first-search. For
graphs, edge is of the form (u, v) where u and v are the tail and head of the edge as
determined by the traversal. For multigraphs, edge is of the form (u, v, key), where
key is the key of the edge. When the graph is directed, then u and v are always in the order of
the actual directed edge. If orientation is not None then the edge tuple is extended to include
the direction of traversal (‘forward’ or ‘reverse’) on that edge.
See also:
bfs_edges
bfs_tree
edge_dfs

Notes
The goal of this function is to visit edges. It differs from the more familiar breadth-first-search of nodes, as provided
by networkx.algorithms.traversal.breadth_first_search.bfs_edges(), in that it does
not stop once every node has been visited. In a directed graph with edges [(0, 1), (1, 2), (2, 1)], the edge (2, 1)
would not be visited if not for the functionality provided by this function.
The naming of this function is very similar to bfs_edges. The difference is that ‘edge_bfs’ yields edges even if
they extend back to an already explored node while ‘bfs_edges’ yields the edges of the tree that results from a
breadth-first-search (BFS) so no edges are reported if they extend to already explored nodes. That means ‘edge_bfs’
reports all edges while ‘bfs_edges’ only report those traversed by a node-based BFS. Yet another description is that
‘bfs_edges’ reports the edges traversed during BFS while ‘edge_bfs’ reports all edges in the order they are explored.
Examples
>>> nodes = [0, 1, 2, 3]

>>> edges = [(0, 1), (1, 0), (1, 0), (2, 0), (2, 1), (3, 1)]
>>> list(nx.edge_bfs(nx.Graph(edges), nodes))

[(0, 1), (0, 2), (1, 2), (1, 3)]
>>> list(nx.edge_bfs(nx.DiGraph(edges), nodes))

[(0, 1), (1, 0), (2, 0), (2, 1), (3, 1)]
>>> list(nx.edge_bfs(nx.MultiGraph(edges), nodes))

[(0, 1, 0), (0, 1, 1), (0, 1, 2), (0, 2, 0), (1, 2, 0), (1, 3, 0)]
>>> list(nx.edge_bfs(nx.MultiDiGraph(edges), nodes))

[(0, 1, 0), (1, 0, 0), (1, 0, 1), (2, 0, 0), (2, 1, 0), (3, 1, 0)]
>>> list(nx.edge_bfs(nx.DiGraph(edges), nodes, orientation="ignore"))

[(0, 1, 'forward'), (1, 0, 'reverse'), (2, 0, 'reverse'), (2, 1, 'reverse'), (3,␣
,→1, 'reverse')]
>>> list(nx.edge_bfs(nx.MultiDiGraph(edges), nodes, orientation="ignore"))

[(0, 1, 0, 'forward'), (1, 0, 0, 'reverse'), (1, 0, 1, 'reverse'), (2, 0, 0,
,→'reverse'), (2, 1, 0, 'reverse'), (3, 1, 0, 'reverse')]
3.64 Tree
3.64.1 Recognition
Recognition Tests
A forest is an acyclic, undirected graph, and a tree is a connected forest. Depending on the subfield, there are various
conventions for generalizing these definitions to directed graphs.
In one convention, directed variants of forest and tree are defined in an identical manner, except that the direction of the
edges is ignored. In effect, each directed edge is treated as a single undirected edge. Then, additional restrictions are
imposed to define branchings and arborescences.
3.64. Tree 723

In another convention, directed variants of forest and tree correspond to the previous convention’s branchings and arbores-
cences, respectively. Then two new terms, polyforest and polytree, are defined to correspond to the other convention’s
forest and tree.
Summarizing:
+-----------------------------+
| Convention A | Convention B |
+=============================+
| forest | polyforest |
| tree | polytree |
| branching | forest |
| arborescence | tree |
+-----------------------------+
Each convention has its reasons. The first convention emphasizes definitional similarity in that directed forests and trees
are only concerned with acyclicity and do not have an in-degree constraint, just as their undirected counterparts do not.
The second convention emphasizes functional similarity in the sense that the directed analog of a spanning tree is a
spanning arborescence. That is, take any spanning tree and choose one node as the root. Then every edge is assigned a
direction such there is a directed path from the root to every other node. The result is a spanning arborescence.
NetworkX follows convention “A”. Explicitly, these are:
undirected forest
An undirected graph with no undirected cycles.
undirected tree
A connected, undirected forest.
directed forest
A directed graph with no undirected cycles. Equivalently, the underlying graph structure (which ignores edge
orientations) is an undirected forest. In convention B, this is known as a polyforest.
directed tree
A weakly connected, directed forest. Equivalently, the underlying graph structure (which ignores edge orientations)
is an undirected tree. In convention B, this is known as a polytree.
branching
A directed forest with each node having, at most, one parent. So the maximum in-degree is equal to 1. In convention
B, this is known as a forest.
arborescence
A directed tree with each node having, at most, one parent. So the maximum in-degree is equal to 1. In convention
B, this is known as a tree.
For trees and arborescences, the adjective “spanning” may be added to designate that the graph, when considered as a
forest/branching, consists of a single tree/arborescence that includes all nodes in the graph. It is true, by definition, that
every tree/arborescence is spanning with respect to the nodes that define the tree/arborescence and so, it might seem
redundant to introduce the notion of “spanning”. However, the nodes may represent a subset of nodes from a larger
graph, and it is in this context that the term “spanning” becomes a useful notion.
is_tree(G) Returns True if G is a tree.

is_forest(G) Returns True if G is a forest.
is_arborescence(G) Returns True if G is an arborescence.
is_branching(G) Returns True if G is a branching.

is_tree
is_tree(G)
Returns True if G is a tree.
A tree is a connected graph with no undirected cycles.
For directed graphs, G is a tree if the underlying graph is a tree. The underlying graph is obtained by treating each
directed edge as a single undirected edge in a multigraph.
Parameters
G
[graph] The graph to test.
Returns
b
[bool] A boolean that is True if G is a tree.
Raises
If G is empty.
See also:
is_arborescence
Notes
In another convention, a directed tree is known as a polytree and then tree corresponds to an arborescence.
Examples
>>> G = nx.Graph()
>>> G.add_edges_from([(1, 2), (1, 3), (2, 4), (2, 5)])
>>> nx.is_tree(G) # n-1 edges
True
>>> nx.is_tree(G) # n edges
False
is_forest
is_forest(G)
Returns True if G is a forest.
A forest is a graph with no undirected cycles.
For directed graphs, G is a forest if the underlying graph is a forest. The underlying graph is obtained by treating
each directed edge as a single undirected edge in a multigraph.
Parameters
G
3.64. Tree 725

Returns
b
[bool] A boolean that is True if G is a forest.
Raises
If G is empty.
See also:
is_branching
Notes
In another convention, a directed forest is known as a polyforest and then forest corresponds to a branching.
Examples
>>> G = nx.Graph()
>>> G.add_edges_from([(1, 2), (1, 3), (2, 4), (2, 5)])
>>> nx.is_forest(G)
True
>>> nx.is_forest(G)
False
is_arborescence
is_arborescence(G)
Returns True if G is an arborescence.
An arborescence is a directed tree with maximum in-degree equal to 1.
Parameters
G
Returns
b
[bool] A boolean that is True if G is an arborescence.
See also:
is_tree

Notes
In another convention, an arborescence is known as a tree.
Examples
>>> G = nx.DiGraph([(0, 1), (0, 2), (2, 3), (3, 4)])

>>> nx.is_arborescence(G)
True
>>> G.add_edge(1, 2) # maximum in-degree is 2
>>> nx.is_arborescence(G)
False
is_branching
is_branching(G)
Returns True if G is a branching.
A branching is a directed forest with maximum in-degree equal to 1.
Parameters
G
[directed graph] The directed graph to test.
Returns
b
[bool] A boolean that is True if G is a branching.
See also:
is_forest
Notes
In another convention, a branching is also known as a forest.
Examples
>>> G = nx.DiGraph([(0, 1), (1, 2), (2, 3), (3, 4)])

>>> nx.is_branching(G)
True
>>> G.add_edge(3, 1) # maximum in-degree is 2
>>> nx.is_branching(G)
False
3.64. Tree 727

3.64.2 Branchings and Spanning Arborescences
Algorithms for finding optimum branchings and spanning arborescences.

This implementation is based on:
J. Edmonds, Optimum branchings, J. Res. Natl. Bur. Standards 71B (1967), 233–240. URL: http://archive.
org/details/jresv71Bn4p233
branching_weight(G[, attr, default]) Returns the total weight of a branching.

greedy_branching(G[, attr, default, kind, seed]) Returns a branching obtained through a greedy algorithm.
maximum_branching(G[, attr, default, ...]) Returns a maximum branching from G.
minimum_branching(G[, attr, default, ...]) Returns a minimum branching from G.
maximum_spanning_arborescence(G[, attr, Returns a maximum spanning arborescence from G.
...])
minimum_spanning_arborescence(G[, attr, Returns a minimum spanning arborescence from G.
...])
ArborescenceIterator(G[, weight, minimum, Iterate over all spanning arborescences of a graph in either
...]) increasing or decreasing cost.
Edmonds(G[, seed]) Edmonds algorithm [1] for finding optimal branchings
and spanning arborescences.
branching_weight
branching_weight(G, attr='weight', default=1)

Returns the total weight of a branching.
You must access this function through the networkx.algorithms.tree module.
Parameters
G
[DiGraph] The directed graph.
attr
[str] The attribute to use as weights. If None, then each edge will be treated equally with a
weight of 1.
default
[float] When attr is not None, then if an edge does not have that attribute, default spec-
ifies what value it should take.
Returns
weight: int or float
The total weight of the branching.

Examples
>>> G.add_weighted_edges_from([(0, 1, 2), (1, 2, 4), (2, 3, 3), (3, 4, 2)])
>>> nx.tree.branching_weight(G)
11
greedy_branching
greedy_branching(G, attr='weight', default=1, kind='max', seed=None)

Returns a branching obtained through a greedy algorithm.
This algorithm is wrong, and cannot give a proper optimal branching. However, we include it for pedagogical
reasons, as it can be helpful to see what its outputs are.
The output is a branching, and possibly, a spanning arborescence. However, it is not guaranteed to be optimal in
either case.
Parameters
G
[DiGraph] The directed graph to scan.
attr
[str] The attribute to use as weights. If None, then each edge will be treated equally with a
weight of 1.
default
[float] When attr is not None, then if an edge does not have that attribute, default spec-
ifies what value it should take.
kind
[str] The type of optimum to search for: ‘min’ or ‘max’ greedy branching.
seed
Randomness.
Returns
B
[directed graph] The greedily obtained branching.
maximum_branching
maximum_branching(G, attr='weight', default=1, preserve_attrs=False, partition=None)

Returns a maximum branching from G.
Parameters
G
[(multi)digraph-like] The graph to be searched.
attr
[str] The edge attribute used to in determining optimality.
default
[float] The value of the edge attribute used if an edge does not have the attribute attr.
3.64. Tree 729

preserve_attrs
[bool] If True, preserve the other attributes of the original graph (that are not passed to attr)
partition
[str] The key for the edge attribute containing the partition data on the graph. Edges can be
included, excluded or open using the EdgePartition enum.
Returns
B
[(multi)digraph-like] A maximum branching.
minimum_branching
minimum_branching(G, attr='weight', default=1, preserve_attrs=False, partition=None)

Returns a minimum branching from G.
Parameters
G
attr
default
preserve_attrs
partition
Returns
B
[(multi)digraph-like] A minimum branching.
maximum_spanning_arborescence
maximum_spanning_arborescence(G, attr='weight', default=1, preserve_attrs=False, partition=None)

Returns a maximum spanning arborescence from G.
Parameters
G
attr
default
preserve_attrs

partition
Returns
B
[(multi)digraph-like] A maximum spanning arborescence.
Raises
NetworkXException
If the graph does not contain a maximum spanning arborescence.
minimum_spanning_arborescence
minimum_spanning_arborescence(G, attr='weight', default=1, preserve_attrs=False, partition=None)

Returns a minimum spanning arborescence from G.
Parameters
G
attr
default
preserve_attrs
partition
Returns
B
[(multi)digraph-like] A minimum spanning arborescence.
Raises
NetworkXException
If the graph does not contain a minimum spanning arborescence.
networkx.algorithms.tree.branchings.ArborescenceIterator
class ArborescenceIterator(G, weight='weight', minimum=True, init_partition=None)

Iterate over all spanning arborescences of a graph in either increasing or decreasing cost.
3.64. Tree 731

Notes
This iterator uses the partition scheme from [1] (included edges, excluded edges and open edges). It generates
minimum spanning arborescences using a modified Edmonds’ Algorithm which respects the partition of edges. For
arborescences with the same weight, ties are broken arbitrarily.
References
[1]
__init__(G, weight='weight', minimum=True, init_partition=None)
Initialize the iterator
Parameters
G
[nx.DiGraph] The directed graph which we need to iterate trees over
weight
[String, default = “weight”] The edge attribute used to store the weight of the edge
minimum
[bool, default = True] Return the trees in increasing order while true and decreasing order
while false.
init_partition
[tuple, default = None] In the case that certain edges have to be included or excluded from the
arborescences, init_partition should be in the form (included_edges, ex-
cluded_edges) where each edges is a (u, v)-tuple inside an iterable such as a list or
set.
Methods
networkx.algorithms.tree.branchings.Edmonds
class Edmonds(G, seed=None)

Edmonds algorithm [1] for finding optimal branchings and spanning arborescences.
This algorithm can find both minimum and maximum spanning arborescences and branchings.
Notes
While this algorithm can find a minimum branching, since it isn’t required to be spanning, the minimum branching
is always from the set of negative weight edges which is most likely the empty set for most graphs.

References
[1]
__init__(G, seed=None)
Methods
find_optimum([attr, default, kind, style, ...]) Returns a branching from G.
Edmonds.find_optimum
Edmonds.find_optimum(attr='weight', default=1, kind='max', style='branching', preserve_attrs=False,

partition=None, seed=None)
Returns a branching from G.
Parameters
attr
default
kind
[{‘min’, ‘max’}] The type of optimum to search for, either ‘min’ or ‘max’.
style
[{‘branching’, ‘arborescence’}] If ‘branching’, then an optimal branching is found. If style
is ‘arborescence’, then a branching is found, such that if the branching is also an arborescence,
then the branching is an optimal spanning arborescences. A given graph G need not have an
optimal spanning arborescence.
preserve_attrs
[bool] If True, preserve the other edge attributes of the original graph (that are not the one
passed to attr)
partition
[str] The edge attribute holding edge partition data. Used in the spanning arborescence iter-
ator.
seed
Randomness.
Returns
H
[(multi)digraph] The branching.
3.64. Tree 733

3.64.3 Encoding and decoding
Functions for encoding and decoding trees.

Since a tree is a highly restricted form of graph, it can be represented concisely in several ways. This module includes
functions for encoding and decoding trees in the form of nested tuples and Prüfer sequences. The former requires a
rooted tree, whereas the latter can be applied to unrooted trees. Furthermore, there is a bijection from Prüfer sequences
to labeled trees.
from_nested_tuple(sequence[, ...]) Returns the rooted tree corresponding to the given nested
tuple.
to_nested_tuple(T, root[, canonical_form]) Returns a nested tuple representation of the given tree.
from_prufer_sequence(sequence) Returns the tree corresponding to the given Prüfer se-
quence.
to_prufer_sequence(T) Returns the Prüfer sequence of the given tree.
from_nested_tuple
from_nested_tuple(sequence, sensible_relabeling=False)
Returns the rooted tree corresponding to the given nested tuple.
The nested tuple representation of a tree is defined recursively. The tree with one node and no edges is represented
by the empty tuple, (). A tree with k subtrees is represented by a tuple of length k in which each element is the
nested tuple representation of a subtree.
Parameters
sequence
[tuple] A nested tuple representing a rooted tree.
sensible_relabeling
[bool] Whether to relabel the nodes of the tree so that nodes are labeled in increasing order
according to their breadth-first search order from the root node.
Returns
NetworkX graph
The tree corresponding to the given nested tuple, whose root node is node 0. If sensi-
ble_labeling is True, nodes will be labeled in breadth-first search order starting from
the root node.
See also:
to_nested_tuple
from_prufer_sequence

Notes
This function is not the inverse of to_nested_tuple(); the only guarantee is that the rooted trees are iso-
morphic.
Examples
Sensible relabeling ensures that the nodes are labeled from the root starting at 0:
>>> balanced = (((), ()), ((), ()))

>>> T = nx.from_nested_tuple(balanced, sensible_relabeling=True)
>>> edges = [(0, 1), (0, 2), (1, 3), (1, 4), (2, 5), (2, 6)]
>>> all((u, v) in T.edges() or (v, u) in T.edges() for (u, v) in edges)
True
to_nested_tuple
to_nested_tuple(T, root, canonical_form=False)

Returns a nested tuple representation of the given tree.
The nested tuple representation of a tree is defined recursively. The tree with one node and no edges is represented
by the empty tuple, (). A tree with k subtrees is represented by a tuple of length k in which each element is the
nested tuple representation of a subtree.
Parameters
T
[NetworkX graph] An undirected graph object representing a tree.
root
[node] The node in T to interpret as the root of the tree.
canonical_form
[bool] If True, each tuple is sorted so that the function returns a canonical form for rooted
trees. This means “lighter” subtrees will appear as nested tuples before “heavier” subtrees. In
this way, each isomorphic rooted tree has the same nested tuple representation.
Returns
tuple
A nested tuple representation of the tree.
See also:
from_nested_tuple
to_prufer_sequence
3.64. Tree 735

Notes
This function is not the inverse of from_nested_tuple(); the only guarantee is that the rooted trees are
isomorphic.
Examples
The tree need not be a balanced binary tree:
>>> T = nx.Graph()
>>> T.add_edges_from([(0, 1), (0, 2), (0, 3)])
>>> T.add_edges_from([(1, 4), (1, 5)])
>>> T.add_edges_from([(3, 6), (3, 7)])
>>> root = 0
>>> nx.to_nested_tuple(T, root)
(((), ()), (), ((), ()))
Continuing the above example, if canonical_form is True, the nested tuples will be sorted:
>>> nx.to_nested_tuple(T, root, canonical_form=True)

((), ((), ()), ((), ()))
Even the path graph can be interpreted as a tree:
>>> T = nx.path_graph(4)
>>> root = 0
>>> nx.to_nested_tuple(T, root)
((((),),),)
from_prufer_sequence(sequence)
Returns the tree corresponding to the given Prüfer sequence.
A Prüfer sequence is a list of n - 2 numbers between 0 and n - 1, inclusive. The tree corresponding to a given Prüfer
sequence can be recovered by repeatedly joining a node in the sequence with a node with the smallest potential
degree according to the sequence.
Parameters
sequence
[list] A Prüfer sequence, which is a list of n - 2 integers between zero and n - 1, inclusive.
Returns
NetworkX graph
The tree corresponding to the given Prüfer sequence.
See also:
from_nested_tuple
to_prufer_sequence

Notes
There is a bijection from labeled trees to Prüfer sequences. This function is the inverse of the
from_prufer_sequence() function.
Sometimes Prüfer sequences use nodes labeled from 1 to n instead of from 0 to n - 1. This function requires nodes
to be labeled in the latter form. You can use networkx.relabel_nodes() to relabel the nodes of your tree
to the appropriate format.
This implementation is from [1] and has a running time of O(n).
References
[1]
Examples
There is a bijection between Prüfer sequences and labeled trees, so this function is the inverse of the
to_prufer_sequence() function:
>>> edges = [(0, 3), (1, 3), (2, 3), (3, 4), (4, 5)]
>>> tree = nx.Graph(edges)
>>> sequence = nx.to_prufer_sequence(tree)
>>> sequence
[3, 3, 3, 4]
>>> tree2 = nx.from_prufer_sequence(sequence)
>>> list(tree2.edges()) == edges
True
to_prufer_sequence
to_prufer_sequence(T)
Returns the Prüfer sequence of the given tree.
A Prüfer sequence is a list of n - 2 numbers between 0 and n - 1, inclusive. The tree corresponding to a given Prüfer
sequence can be recovered by repeatedly joining a node in the sequence with a node with the smallest potential
degree according to the sequence.
Parameters
T
[NetworkX graph] An undirected graph object representing a tree.
Returns
list
The Prüfer sequence of the given tree.
Raises
If the number of nodes in T is less than two.
NotATree
If T is not a tree.
3.64. Tree 737

KeyError
If the set of nodes in T is not {0, …, n - 1}.
See also:
to_nested_tuple
Notes
There is a bijection from labeled trees to Prüfer sequences. This function is the inverse of the
from_prufer_sequence() function.
Sometimes Prüfer sequences use nodes labeled from 1 to n instead of from 0 to n - 1. This function requires
nodes to be labeled in the latter form. You can use relabel_nodes() to relabel the nodes of your tree to the
appropriate format.
This implementation is from [1] and has a running time of O(n).
References
[1]
Examples
There is a bijection between Prüfer sequences and labeled trees, so this function is the inverse of the
from_prufer_sequence() function:
>>> edges = [(0, 3), (1, 3), (2, 3), (3, 4), (4, 5)]
>>> tree = nx.Graph(edges)
>>> sequence = nx.to_prufer_sequence(tree)
>>> sequence
[3, 3, 3, 4]
>>> tree2 = nx.from_prufer_sequence(sequence)
>>> list(tree2.edges()) == edges
True
3.64.4 Operations
Operations on trees.
join(rooted_trees[, label_attribute]) Returns a new rooted tree with a root node joined with
the roots of each of the given rooted trees.

join
join(rooted_trees, label_attribute=None)
Returns a new rooted tree with a root node joined with the roots of each of the given rooted trees.
Parameters
rooted_trees
[list] A list of pairs in which each left element is a NetworkX graph object representing a tree
and each right element is the root node of that tree. The nodes of these trees will be relabeled
to integers.
label_attribute
[str] If provided, the old node labels will be stored in the new tree under this node attribute. If
not provided, the node attribute '_old' will store the original label of the node in the rooted
trees given in the input.
Returns
NetworkX graph
The rooted tree whose subtrees are the given rooted trees. The new root node is labeled
0. Each non-root node has an attribute, as described under the keyword argument la-
bel_attribute, that indicates the label of the original node in the input tree.
Notes
Graph, edge, and node attributes are propagated from the given rooted trees to the created tree. If there are any
overlapping graph attributes, those from later trees will overwrite those from earlier trees in the tuple of positional
arguments.
Examples
Join two full balanced binary trees of height h to get a full balanced binary tree of depth h + 1:
>>> h = 4
>>> left = nx.balanced_tree(2, h)
>>> right = nx.balanced_tree(2, h)
>>> joined_tree = nx.join([(left, 0), (right, 0)])
>>> nx.is_isomorphic(joined_tree, nx.balanced_tree(2, h + 1))
True
3.64.5 Spanning Trees
Algorithms for calculating min/max spanning trees/forests.
3.64. Tree 739

minimum_spanning_tree(G[, weight, ...]) Returns a minimum spanning tree or forest on an undi-

rected graph G.
maximum_spanning_tree(G[, weight, ...]) Returns a maximum spanning tree or forest on an undi-
rected graph G.
random_spanning_tree(G[, weight, ...]) Sample a random spanning tree using the edges weights
of G.
minimum_spanning_edges(G[, algorithm, ...]) Generate edges in a minimum spanning forest of an undi-
rected weighted graph.
maximum_spanning_edges(G[, algorithm, ...]) Generate edges in a maximum spanning forest of an undi-
rected weighted graph.
SpanningTreeIterator(G[, weight, minimum, Iterate over all spanning trees of a graph in either increas-
...]) ing or decreasing cost.
minimum_spanning_tree
minimum_spanning_tree(G, weight='weight', algorithm='kruskal', ignore_nan=False)

Returns a minimum spanning tree or forest on an undirected graph G.
Parameters
G
[undirected graph] An undirected graph. If G is connected, then the algorithm finds a spanning
tree. Otherwise, a spanning forest is found.
weight
[str] Data key to use for edge weights.
algorithm
[string] The algorithm to use when finding a minimum spanning tree. Valid choices are
‘kruskal’, ‘prim’, or ‘boruvka’. The default is ‘kruskal’.
ignore_nan
[bool (default: False)] If a NaN is found as an edge weight normally an exception is raised. If
ignore_nan is True then that edge is ignored instead.
Returns
G
[NetworkX Graph] A minimum spanning tree or forest.
Notes
For Borůvka’s algorithm, each edge must have a weight attribute, and each edge weight must be distinct.
For the other algorithms, if the graph edges do not have a weight attribute a default weight of 1 will be used.
There may be more than one tree with the same minimum or maximum weight. See networkx.tree.
recognition for more detailed definitions.
Isolated nodes with self-loops are in the tree as edgeless isolated nodes.

Examples
>>> T = nx.minimum_spanning_tree(G)
>>> sorted(T.edges(data=True))
[(0, 1, {}), (1, 2, {}), (2, 3, {})]
maximum_spanning_tree
maximum_spanning_tree(G, weight='weight', algorithm='kruskal', ignore_nan=False)

Returns a maximum spanning tree or forest on an undirected graph G.
Parameters
G
[undirected graph] An undirected graph. If G is connected, then the algorithm finds a spanning
weight
[str] Data key to use for edge weights.
algorithm
[string] The algorithm to use when finding a maximum spanning tree. Valid choices are
ignore_nan
Returns
G
[NetworkX Graph] A maximum spanning tree or forest.
Notes
There may be more than one tree with the same minimum or maximum weight. See networkx.tree.
recognition for more detailed definitions.
Isolated nodes with self-loops are in the tree as edgeless isolated nodes.
Examples
>>> T = nx.maximum_spanning_tree(G)
>>> sorted(T.edges(data=True))
[(0, 1, {}), (0, 3, {'weight': 2}), (1, 2, {})]
3.64. Tree 741

random_spanning_tree
random_spanning_tree(G, weight=None, *, multiplicative=True, seed=None)

Sample a random spanning tree using the edges weights of G.
This function supports two different methods for determining the probability of the graph. If multiplica-
tive=True, the probability is based on the product of edge weights, and if multiplicative=False it is
based on the sum of the edge weight. However, since it is easier to determine the total weight of all spanning
trees for the multiplicative verison, that is significantly faster and should be used if possible. Additionally, setting
weight to None will cause a spanning tree to be selected with uniform probability.
The function uses algorithm A8 in [1] .
Parameters
G
[nx.Graph] An undirected version of the original graph.
weight
[string] The edge key for the edge attribute holding edge weight.
multiplicative
[bool, default=True] If True, the probability of each tree is the product of its edge weight
over the sum of the product of all the spanning trees in the graph. If False, the probability
is the sum of its edge weight over the sum of the sum of weights for all spanning trees in the
graph.
seed
Randomness.
Returns
nx.Graph
A spanning tree using the distribution defined by the weight of the tree.
References
[1]
minimum_spanning_edges
minimum_spanning_edges(G, algorithm='kruskal', weight='weight', keys=True, data=True, ignore_nan=False)

Generate edges in a minimum spanning forest of an undirected weighted graph.
A minimum spanning tree is a subgraph of the graph (a tree) with the minimum sum of edge weights. A spanning
forest is a union of the spanning trees for each connected component of the graph.
Parameters
G
[undirected Graph] An undirected graph. If G is connected, then the algorithm finds a spanning
algorithm
[string] The algorithm to use when finding a minimum spanning tree. Valid choices are

weight
[string] Edge data key to use for weight (default ‘weight’).
keys
[bool] Whether to yield edge key in multigraphs in addition to the edge. If G is not a multigraph,
this is ignored.
data
[bool, optional] If True yield the edge data along with the edge.
ignore_nan
Returns
edges
[iterator] An iterator over edges in a maximum spanning tree of G. Edges connecting nodes u
and v are represented as tuples: (u, v, k, d) or (u, v, k) or (u, v, d) or (u,
v)
If G is a multigraph, keys indicates whether the edge key k will be reported in the third
position in the edge tuple. data indicates whether the edge datadict d will appear at the end
of the edge tuple.
If G is not a multigraph, the tuples are (u, v, d) if data is True or (u, v) if data is
False.
Notes
Modified code from David Eppstein, April 2006 http://www.ics.uci.edu/~eppstein/PADS/
Examples
>>> from networkx.algorithms import tree
Find minimum spanning edges by Kruskal’s algorithm
>>> mst = tree.minimum_spanning_edges(G, algorithm="kruskal", data=False)
>>> edgelist = list(mst)
>>> sorted(sorted(e) for e in edgelist)
[[0, 1], [1, 2], [2, 3]]
Find minimum spanning edges by Prim’s algorithm
>>> mst = tree.minimum_spanning_edges(G, algorithm="prim", data=False)
[[0, 1], [1, 2], [2, 3]]
3.64. Tree 743

maximum_spanning_edges
maximum_spanning_edges(G, algorithm='kruskal', weight='weight', keys=True, data=True, ignore_nan=False)

Generate edges in a maximum spanning forest of an undirected weighted graph.
A maximum spanning tree is a subgraph of the graph (a tree) with the maximum possible sum of edge weights. A
spanning forest is a union of the spanning trees for each connected component of the graph.
Parameters
G
[undirected Graph] An undirected graph. If G is connected, then the algorithm finds a spanning
algorithm
[string] The algorithm to use when finding a maximum spanning tree. Valid choices are
weight
[string] Edge data key to use for weight (default ‘weight’).
keys
[bool] Whether to yield edge key in multigraphs in addition to the edge. If G is not a multigraph,
this is ignored.
data
[bool, optional] If True yield the edge data along with the edge.
ignore_nan
Returns
edges
[iterator] An iterator over edges in a maximum spanning tree of G. Edges connecting nodes u
and v are represented as tuples: (u, v, k, d) or (u, v, k) or (u, v, d) or (u,
v)
If G is a multigraph, keys indicates whether the edge key k will be reported in the third
position in the edge tuple. data indicates whether the edge datadict d will appear at the end
of the edge tuple.
If G is not a multigraph, the tuples are (u, v, d) if data is True or (u, v) if data is
False.
Notes
Modified code from David Eppstein, April 2006 http://www.ics.uci.edu/~eppstein/PADS/

Examples
>>> from networkx.algorithms import tree
Find maximum spanning edges by Kruskal’s algorithm

>>> mst = tree.maximum_spanning_edges(G, algorithm="kruskal", data=False)
[[0, 1], [0, 3], [1, 2]]
Find maximum spanning edges by Prim’s algorithm

>>> G.add_edge(0, 3, weight=2) # assign weight 2 to edge 0-3
>>> mst = tree.maximum_spanning_edges(G, algorithm="prim", data=False)
[[0, 1], [0, 3], [2, 3]]
networkx.algorithms.tree.mst.SpanningTreeIterator
class SpanningTreeIterator(G, weight='weight', minimum=True, ignore_nan=False)

Iterate over all spanning trees of a graph in either increasing or decreasing cost.
Notes
This iterator uses the partition scheme from [1] (included edges, excluded edges and open edges) as well as a mod-
ified Kruskal’s Algorithm to generate minimum spanning trees which respect the partition of edges. For spanning
trees with the same weight, ties are broken arbitrarily.
References
[1]
__init__(G, weight='weight', minimum=True, ignore_nan=False)
Initialize the iterator
Parameters
G
[nx.Graph] The directed graph which we need to iterate trees over
weight
[String, default = “weight”] The edge attribute used to store the weight of the edge
minimum
[bool, default = True] Return the trees in increasing order while true and decreasing order
while false.
ignore_nan
[bool, default = False] If a NaN is found as an edge weight normally an exception is raised.
If ignore_nan is True then that edge is ignored instead.
3.64. Tree 745

Methods
3.64.6 Decomposition
Function for computing a junction tree of a graph.
junction_tree(G) Returns a junction tree of a given graph.
junction_tree
junction_tree(G)
Returns a junction tree of a given graph.
A junction tree (or clique tree) is constructed from a (un)directed graph G. The tree is constructed based on a
moralized and triangulated version of G. The tree’s nodes consist of maximal cliques and sepsets of the revised
graph. The sepset of two cliques is the intersection of the nodes of these cliques, e.g. the sepset of (A,B,C) and
(A,C,E,F) is (A,C). These nodes are often called “variables” in this literature. The tree is bipartitie with each sepset
connected to its two cliques.
Junction Trees are not unique as the order of clique consideration determines which sepsets are included.
The junction tree algorithm consists of five steps [1]:
1. Moralize the graph
2. Triangulate the graph
3. Find maximal cliques
4. Build the tree from cliques, connecting cliques with shared nodes, set edge-weight to number of shared vari-
ables
5. Find maximum spanning tree
Parameters
G
[networkx.Graph] Directed or undirected graph.
Returns
junction_tree
[networkx.Graph] The corresponding junction tree of G.
Raises
Raised if G is an instance of MultiGraph or MultiDiGraph.

References
[1], [2]
3.64.7 Exceptions
Functions for encoding and decoding trees.

Since a tree is a highly restricted form of graph, it can be represented concisely in several ways. This module includes
functions for encoding and decoding trees in the form of nested tuples and Prüfer sequences. The former requires a
rooted tree, whereas the latter can be applied to unrooted trees. Furthermore, there is a bijection from Prüfer sequences
to labeled trees.
NotATree Raised when a function expects a tree (that is, a connected

undirected graph with no cycles) but gets a non-tree graph
as input instead.
NotATree
exception NotATree
Raised when a function expects a tree (that is, a connected undirected graph with no cycles) but gets a non-tree
graph as input instead.
3.65 Triads
Functions for analyzing triads of a graph.
triadic_census(G[, nodelist]) Determines the triadic census of a directed graph.

random_triad(G[, seed]) Returns a random triad from a directed graph.
triads_by_type(G) Returns a list of all triads for each triad type in a directed
graph.
triad_type(G) Returns the sociological triad type for a triad.
is_triad(G) Returns True if the graph G is a triad, else False.
all_triads(G) A generator of all possible triads in G.
all_triplets(G) Returns a generator of all possible sets of 3 nodes in a
DiGraph.
3.65.1 triadic_census
triadic_census(G, nodelist=None)
Determines the triadic census of a directed graph.
The triadic census is a count of how many of the 16 possible types of triads are present in a directed graph. If a
list of nodes is passed, then only those triads are taken into account which have elements of nodelist in them.
Parameters
G
[digraph] A NetworkX DiGraph
3.65. Triads 747

nodelist
[list] List of nodes for which you want to calculate triadic census
Returns
census
[dict] Dictionary with triad type as keys and number of occurrences as values.
Raises
ValueError
If nodelist contains duplicate nodes or nodes not in G. If you want to ignore this you can
preprocess with set(nodelist) & G.nodes
See also:
triad_graph
Notes
This algorithm has complexity O(m) where m is the number of edges in the graph.
References
[1]
Examples
>>> G = nx.DiGraph([(1, 2), (2, 3), (3, 1), (3, 4), (4, 1), (4, 2)])
>>> triadic_census = nx.triadic_census(G)
>>> for key, value in triadic_census.items():
... print(f"{key}: {value}")
...
003: 0
012: 0
102: 0
021D: 0
021U: 0
021C: 0
111D: 0
111U: 0
030T: 2
030C: 2
201: 0
120D: 0
120U: 0
120C: 0
210: 0
300: 0

3.65.2 random_triad
random_triad(G, seed=None)
Returns a random triad from a directed graph.
Parameters
G
seed
Randomness.
Returns
G2
[subgraph] A randomly selected triad (order-3 NetworkX DiGraph)
Examples
>>> G = nx.DiGraph([(1, 2), (1, 3), (2, 3), (3, 1), (5, 6), (5, 4), (6, 7)])
>>> triad = nx.random_triad(G, seed=1)
>>> triad.edges
OutEdgeView([(1, 2)])
3.65.3 triads_by_type
triads_by_type(G)
Returns a list of all triads for each triad type in a directed graph. There are exactly 16 different types of triads
possible. Suppose 1, 2, 3 are three nodes, they will be classified as a particular triad type if their connections are
as follows:
• 003: 1, 2, 3
• 012: 1 -> 2, 3
• 102: 1 <-> 2, 3
• 021D: 1 <- 2 -> 3
• 021U: 1 -> 2 <- 3
• 021C: 1 -> 2 -> 3
• 111D: 1 <-> 2 <- 3
• 111U: 1 <-> 2 -> 3
• 030T: 1 -> 2 -> 3, 1 -> 3
• 030C: 1 <- 2 <- 3, 1 -> 3
• 201: 1 <-> 2 <-> 3
• 120D: 1 <- 2 -> 3, 1 <-> 3
• 120U: 1 -> 2 <- 3, 1 <-> 3
• 120C: 1 -> 2 -> 3, 1 <-> 3
3.65. Triads 749

• 210: 1 -> 2 <-> 3, 1 <-> 3

• 300: 1 <-> 2 <-> 3, 1 <-> 3
Refer to the example gallery for visual examples of the triad types.
Parameters
G
Returns
tri_by_type
[dict] Dictionary with triad types as keys and lists of triads as values.
References
[1]
Examples
>>> G = nx.DiGraph([(1, 2), (1, 3), (2, 3), (3, 1), (5, 6), (5, 4), (6, 7)])
>>> dict = nx.triads_by_type(G)
>>> dict['120C'][0].edges()
OutEdgeView([(1, 2), (1, 3), (2, 3), (3, 1)])
>>> dict['012'][0].edges()
OutEdgeView([(1, 2)])
3.65.4 triad_type
triad_type(G)
Returns the sociological triad type for a triad.
Parameters
G
[digraph] A NetworkX DiGraph with 3 nodes
Returns
triad_type
[str] A string identifying the triad type
Notes
There can be 6 unique edges in a triad (order-3 DiGraph) (so 2^^6=64 unique triads given 3 nodes). These 64 triads
each display exactly 1 of 16 topologies of triads (topologies can be permuted). These topologies are identified by
the following notation:
{m}{a}{n}{type} (for example: 111D, 210, 102)
Here:
{m} = number of mutual ties (takes 0, 1, 2, 3); a mutual tie is (0,1)
AND (1,0)

{a} = number of assymmetric ties (takes 0, 1, 2, 3); an assymmetric tie

is (0,1) BUT NOT (1,0) or vice versa
{n} = number of null ties (takes 0, 1, 2, 3); a null tie is NEITHER
(0,1) NOR (1,0)
{type} = a letter (takes U, D, C, T) corresponding to up, down, cyclical
and transitive. This is only used for topologies that can have more than one form (eg: 021D and 021U).
References
[1]
Examples
>>> G = nx.DiGraph([(1, 2), (2, 3), (3, 1)])

>>> nx.triad_type(G)
'030C'
>>> nx.triad_type(G)
'120C'
3.65.5 is_triad
is_triad(G)
Returns True if the graph G is a triad, else False.
Parameters
G
[graph] A NetworkX Graph
Returns
istriad
[boolean] Whether G is a valid triad
Examples
>>> G = nx.DiGraph([(1, 2), (2, 3), (3, 1)])

>>> nx.is_triad(G)
True
>>> nx.is_triad(G)
False
3.65. Triads 751

3.65.6 all_triads
all_triads(G)
A generator of all possible triads in G.
Parameters
G
Returns
all_triads
[generator of DiGraphs] Generator of triads (order-3 DiGraphs)
Examples
>>> G = nx.DiGraph([(1, 2), (2, 3), (3, 1), (3, 4), (4, 1), (4, 2)])
>>> for triad in nx.all_triads(G):
... print(triad.edges)
[(1, 2), (2, 3), (3, 1)]
[(1, 2), (4, 1), (4, 2)]
[(3, 1), (3, 4), (4, 1)]
[(2, 3), (3, 4), (4, 2)]
3.65.7 all_triplets
all_triplets(G)
Returns a generator of all possible sets of 3 nodes in a DiGraph.
Parameters
G
Returns
triplets
[generator of 3-tuples] Generator of tuples of 3 nodes
Examples
>>> G = nx.DiGraph([(1, 2), (2, 3), (3, 4)])

>>> list(nx.all_triplets(G))
[(1, 2, 3), (1, 2, 4), (1, 3, 4), (2, 3, 4)]

3.66 Vitality
Vitality measures.
closeness_vitality(G[, node, weight, ...]) Returns the closeness vitality for nodes in the graph.
3.66.1 closeness_vitality
closeness_vitality(G, node=None, weight=None, wiener_index=None)

Returns the closeness vitality for nodes in the graph.
The closeness vitality of a node, defined in Section 3.6.2 of [1], is the change in the sum of distances between all
node pairs when excluding that node.
Parameters
G
[NetworkX graph] A strongly-connected graph.
weight
[string] The name of the edge attribute used as weight. This is passed directly to the
wiener_index() function.
node
[object] If specified, only the closeness vitality for this node will be returned. Otherwise, a
dictionary mapping each node to its closeness vitality will be returned.
Returns
dictionary or float
If node is None, this function returns a dictionary with nodes as keys and closeness vitality as
the value. Otherwise, it returns only the closeness vitality for the specified node.
The closeness vitality of a node may be negative infinity if removing that node would disconnect
the graph.
Other Parameters
wiener_index
[number] If you have already computed the Wiener index of the graph G, you can provide that
value here. Otherwise, it will be computed for you.
See also:
References
[1]
3.66. Vitality 753

Examples
>>> nx.closeness_vitality(G)
{0: 2.0, 1: 2.0, 2: 2.0}
3.67 Voronoi cells
Functions for computing the Voronoi cells of a graph.
voronoi_cells(G, center_nodes[, weight]) Returns the Voronoi cells centered at center_nodes

with respect to the shortest-path distance metric.
3.67.1 voronoi_cells
voronoi_cells(G, center_nodes, weight='weight')

Returns the Voronoi cells centered at center_nodes with respect to the shortest-path distance metric.
If C is a set of nodes in the graph and c is an element of C, the Voronoi cell centered at a node c is the set of all
nodes v that are closer to c than to any other center node in C with respect to the shortest-path distance metric. [1]
For directed graphs, this will compute the “outward” Voronoi cells, as defined in [1], in which distance is measured
from the center nodes to the target node. For the “inward” Voronoi cells, use the DiGraph.reverse() method
to reverse the orientation of the edges before invoking this function on the directed graph.
Parameters
G
[NetworkX graph]
center_nodes
[set] A nonempty set of nodes in the graph G that represent the center of the Voronoi cells.
weight
[string or function] The edge attribute (or an arbitrary function) representing the
weight of an edge. This keyword argument is as described in the documentation for
multi_source_dijkstra_path(), for example.
Returns
dictionary
A mapping from center node to set of all nodes in the graph closer to that center node than to
any other center node. The keys of the dictionary are the element of center_nodes, and
the values of the dictionary form a partition of the nodes of G.
Raises
ValueError
If center_nodes is empty.

References
[1]
Examples
To get only the partition of the graph induced by the Voronoi cells, take the collection of all values in the returned
dictionary:
>>> center_nodes = {0, 3}
>>> cells = nx.voronoi_cells(G, center_nodes)
>>> partition = set(map(frozenset, cells.values()))
>>> sorted(map(sorted, partition))
[[0, 1], [2, 3, 4, 5]]
3.68 Wiener index
Functions related to the Wiener index of a graph.
wiener_index(G[, weight]) Returns the Wiener index of the given graph.
3.68.1 wiener_index
wiener_index(G, weight=None)
Returns the Wiener index of the given graph.
The Wiener index of a graph is the sum of the shortest-path distances between each pair of reachable nodes. For
pairs of nodes in undirected graphs, only one orientation of the pair is counted.
Parameters
G
[NetworkX graph]
weight
[object] The edge attribute to use as distance when computing shortest-path distances. This is
passed directly to the networkx.shortest_path_length() function.
Returns
float
The Wiener index of the graph G.
Raises
NetworkXError
If the graph G is not connected.
3.68. Wiener index 755

Notes
If a pair of nodes is not reachable, the distance is assumed to be infinity. This means that for graphs that are not
strongly-connected, this function returns inf.
The Wiener index is not usually defined for directed graphs, however this function uses the natural generalization
of the Wiener index to directed graphs.
Examples
The Wiener index of the (unweighted) complete graph on n nodes equals the number of pairs of the n nodes, since
each pair of nodes is at distance one:
>>> n = 10
>>> G = nx.complete_graph(n)
>>> nx.wiener_index(G) == n * (n - 1) / 2
True
Graphs that are not strongly-connected have infinite Wiener index:
>>> G = nx.empty_graph(2)
>>> nx.wiener_index(G)
inf

CHAPTER
FOUR
FUNCTIONS
Functional interface to graph methods and assorted utilities.
4.1 Graph
degree(G[, nbunch, weight]) Returns a degree view of single node or of nbunch of

nodes.
degree_histogram(G) Returns a list of the frequency of each degree value.
density(G) Returns the density of a graph.
create_empty_copy(G[, with_data]) Returns a copy of the graph G with all of the edges re-
moved.
is_directed(G) Return True if graph is directed.
to_directed(graph) Returns a directed view of the graph graph.
to_undirected(graph) Returns an undirected view of the graph graph.
is_empty(G) Returns True if G has no edges.
add_star(G_to_add_to, nodes_for_star, **attr) Add a star to Graph G_to_add_to.
add_path(G_to_add_to, nodes_for_path, **attr) Add a path to the Graph G_to_add_to.
add_cycle(G_to_add_to, nodes_for_cycle, **attr) Add a cycle to the Graph G_to_add_to.
subgraph(G, nbunch) Returns the subgraph induced on nodes in nbunch.
subgraph_view(G[, filter_node, filter_edge]) View of G applying a filter on nodes and edges.
induced_subgraph(G, nbunch) Returns a SubGraph view of G showing only nodes in
nbunch.
restricted_view(G, nodes, edges) Returns a view of G with hidden nodes and edges.
reverse_view(G) View of G with edge directions reversed
edge_subgraph(G, edges) Returns a view of the subgraph induced by the specified
edges.
4.1.1 degree
degree(G, nbunch=None, weight=None)

Returns a degree view of single node or of nbunch of nodes. If nbunch is omitted, then return degrees of all nodes.
757
4.1.2 degree_histogram
degree_histogram(G)
Returns a list of the frequency of each degree value.
Parameters
G
[Networkx graph] A graph
Returns
hist
[list] A list of frequencies of degrees. The degree values are the index in the list.
Notes
Note: the bins are width one, hence len(list) can be large (Order(number_of_edges))
4.1.3 density
density(G)
Returns the density of a graph.
The density for undirected graphs is
2m
d= ,
n(n − 1)
and for directed graphs is

m
d= ,
n(n − 1)
where n is the number of nodes and m is the number of edges in G.
Notes
The density is 0 for a graph without edges and 1 for a complete graph. The density of multigraphs can be higher
than 1.
Self loops are counted in the total number of edges so graphs with self loops can have density higher than 1.
4.1.4 create_empty_copy
create_empty_copy(G, with_data=True)
Returns a copy of the graph G with all of the edges removed.
Parameters
G
with_data
[bool (default=True)] Propagate Graph and Nodes data to the new graph.
758 Chapter 4. Functions

See also:
empty_graph
4.1.5 is_directed
is_directed(G)
Return True if graph is directed.
4.1.6 to_directed
to_directed(graph)
Returns a directed view of the graph graph.
Identical to graph.to_directed(as_view=True) Note that graph.to_directed defaults to as_view=False while
this function always provides a view.
4.1.7 to_undirected
to_undirected(graph)
Returns an undirected view of the graph graph.
Identical to graph.to_undirected(as_view=True) Note that graph.to_undirected defaults to as_view=False
while this function always provides a view.
4.1.8 is_empty
is_empty(G)
Returns True if G has no edges.
Parameters
G
Returns
bool
True if G has no edges, and False otherwise.
Notes
An empty graph can have nodes but not edges. The empty graph with zero nodes is known as the null graph. This
is an O(n) operation where n is the number of nodes in the graph.
4.1. Graph 759

4.1.9 add_star
add_star(G_to_add_to, nodes_for_star, **attr)

Add a star to Graph G_to_add_to.
The first node in nodes_for_star is the middle of the star. It is connected to all other nodes.
Parameters
G_to_add_to
nodes_for_star
[iterable container] A container of nodes.
attr
[keyword arguments, optional (default= no attributes)] Attributes to add to every edge in star.
See also:
add_path, add_cycle
Examples
>>> G = nx.Graph()
>>> nx.add_star(G, [0, 1, 2, 3])
>>> nx.add_star(G, [10, 11, 12], weight=2)
4.1.10 add_path
add_path(G_to_add_to, nodes_for_path, **attr)

Add a path to the Graph G_to_add_to.
Parameters
G_to_add_to
nodes_for_path
[iterable container] A container of nodes. A path will be constructed from the nodes (in order)
and added to the graph.
attr
[keyword arguments, optional (default= no attributes)] Attributes to add to every edge in path.
See also:
add_star, add_cycle

Examples
>>> G = nx.Graph()
>>> nx.add_path(G, [0, 1, 2, 3])
>>> nx.add_path(G, [10, 11, 12], weight=7)
4.1.11 add_cycle
add_cycle(G_to_add_to, nodes_for_cycle, **attr)

Add a cycle to the Graph G_to_add_to.
Parameters
G_to_add_to
nodes_for_cycle: iterable container
A container of nodes. A cycle will be constructed from the nodes (in order) and added to the
graph.
attr
[keyword arguments, optional (default= no attributes)] Attributes to add to every edge in cycle.
See also:
add_path, add_star
Examples

>>> nx.add_cycle(G, [0, 1, 2, 3])
>>> nx.add_cycle(G, [10, 11, 12], weight=7)
4.1.12 subgraph
subgraph(G, nbunch)
Returns the subgraph induced on nodes in nbunch.
Parameters
G
nbunch
[list, iterable] A container of nodes that will be iterated through once (thus it should be an
iterator or be iterable). Each element of the container should be a valid node type: any hashable
type except None. If nbunch is None, return all edges data in the graph. Nodes in nbunch that
are not in the graph will be (quietly) ignored.
4.1. Graph 761

Notes
subgraph(G) calls G.subgraph()
4.1.13 subgraph_view
subgraph_view(G, filter_node=<function no_filter>, filter_edge=<function no_filter>)

View of G applying a filter on nodes and edges.
subgraph_view provides a read-only view of the input graph that excludes nodes and edges based on the
outcome of two filter functions filter_node and filter_edge.
The filter_node function takes one argument — the node — and returns True if the node should be included
in the subgraph, and False if it should not be included.
The filter_edge function takes two (or three arguments if G is a multi-graph) — the nodes describing an edge,
plus the edge-key if parallel edges are possible — and returns True if the edge should be included in the subgraph,
and False if it should not be included.
Both node and edge filter functions are called on graph elements as they are queried, meaning there is no up-front
cost to creating the view.
Parameters
G
[networkx.Graph] A directed/undirected graph/multigraph
filter_node
[callable, optional] A function taking a node as input, which returns True if the node should
appear in the view.
filter_edge
[callable, optional] A function taking as input the two nodes describing an edge (plus the edge-
key if G is a multi-graph), which returns True if the edge should appear in the view.
Returns
graph
[networkx.Graph] A read-only graph view of the input graph.
Examples
Filter functions operate on the node, and return True if the node should appear in the view:
>>> def filter_node(n1):

... return n1 != 5
...
>>> view = nx.subgraph_view(G, filter_node=filter_node)
>>> view.nodes()
NodeView((0, 1, 2, 3, 4))
We can use a closure pattern to filter graph elements based on additional data — for example, filtering on edge data
attached to the graph:

>>> G[3][4]["cross_me"] = False

>>> def filter_edge(n1, n2):
... return G[n1][n2].get("cross_me", True)
...
>>> view = nx.subgraph_view(G, filter_edge=filter_edge)
>>> view.edges()
EdgeView([(0, 1), (1, 2), (2, 3), (4, 5)])
>>> view = nx.subgraph_view(G, filter_node=filter_node, filter_edge=filter_edge,)

>>> view.nodes()
NodeView((0, 1, 2, 3, 4))
>>> view.edges()
EdgeView([(0, 1), (1, 2), (2, 3)])
4.1.14 induced_subgraph
induced_subgraph(G, nbunch)
Returns a SubGraph view of G showing only nodes in nbunch.
The induced subgraph of a graph on a set of nodes N is the graph with nodes N and edges from G which have both
ends in N.
Parameters
G
[NetworkX Graph]
nbunch
[node, container of nodes or None (for all nodes)]
Returns
subgraph
[SubGraph View] A read-only view of the subgraph in G induced by the nodes. Changes to
the graph G will be reflected in the view.
Notes
To create a mutable subgraph with its own copies of nodes edges and attributes use subgraph.copy() or
Graph(subgraph)
For an inplace reduction of a graph to a subgraph you can remove nodes: G.remove_nodes_from(n in G
if n not in set(nbunch))
If you are going to compute subgraphs of your subgraphs you could end up with a chain of views that can be very
slow once the chain has about 15 views in it. If they are all induced subgraphs, you can short-cut the chain by
making them all subgraphs of the original graph. The graph class method G.subgraph does this when G is a
subgraph. In contrast, this function allows you to choose to build chains or not, as you wish. The returned subgraph
is a view on G.
4.1. Graph 763

Examples

>>> H = nx.induced_subgraph(G, [0, 1, 3])
>>> list(H.edges)
[(0, 1)]
>>> list(H.nodes)
[0, 1, 3]
4.1.15 restricted_view
restricted_view(G, nodes, edges)

Returns a view of G with hidden nodes and edges.
The resulting subgraph filters out node nodes and edges edges. Filtered out nodes also filter out any of their
edges.
Parameters
G
[NetworkX Graph]
nodes
[iterable] An iterable of nodes. Nodes not present in G are ignored.
edges
[iterable] An iterable of edges. Edges not present in G are ignored.
Returns
subgraph
[SubGraph View] A read-only restricted view of G filtering out nodes and edges. Changes to
G are reflected in the view.
Notes
Graph(subgraph)
If you create a subgraph of a subgraph recursively you may end up with a chain of subgraph views. Such chains can
get quite slow for lengths near 15. To avoid long chains, try to make your subgraph based on the original graph. We
do not rule out chains programmatically so that odd cases like an edge_subgraph of a restricted_view
can be created.
Examples
>>> H = nx.restricted_view(G, [0], [(1, 2), (3, 4)])
>>> list(H.nodes)
[1, 2, 3, 4]
>>> list(H.edges)
[(2, 3)]

4.1.16 reverse_view
reverse_view(G)
View of G with edge directions reversed
reverse_view returns a read-only view of the input graph where edge directions are reversed.
Identical to digraph.reverse(copy=False)
Parameters
G
[networkx.DiGraph]
Returns
graph
[networkx.DiGraph]
Examples
>>> G.edges()
OutEdgeView([(1, 2), (2, 3)])
>>> view = nx.reverse_view(G)

>>> view.edges()
OutEdgeView([(2, 1), (3, 2)])
4.1.17 edge_subgraph
edge_subgraph(G, edges)
Returns a view of the subgraph induced by the specified edges.
The induced subgraph contains each edge in edges and each node incident to any of those edges.
Parameters
G
[NetworkX Graph]
edges
[iterable] An iterable of edges. Edges not present in G are ignored.
Returns
subgraph
[SubGraph View] A read-only edge-induced subgraph of G. Changes to G are reflected in the
view.
4.1. Graph 765

Notes
Graph(subgraph)
If you create a subgraph of a subgraph recursively you can end up with a chain of subgraphs that becomes very
slow with about 15 nested subgraph views. Luckily the edge_subgraph filter nests nicely so you can use the original
graph as G in this function to avoid chains. We do not rule out chains programmatically so that odd cases like an
edge_subgraph of a restricted_view can be created.
Examples
>>> H = G.edge_subgraph([(0, 1), (3, 4)])
>>> list(H.nodes)
[0, 1, 3, 4]
>>> list(H.edges)
[(0, 1), (3, 4)]
4.2 Nodes
nodes(G) Returns an iterator over the graph nodes.

number_of_nodes(G) Returns the number of nodes in the graph.
neighbors(G, n) Returns a list of nodes connected to node n.
all_neighbors(graph, node) Returns all of the neighbors of a node in the graph.
non_neighbors(graph, node) Returns the non-neighbors of the node in the graph.
common_neighbors(G, u, v) Returns the common neighbors of two nodes in a graph.
4.2.1 nodes
nodes(G)
Returns an iterator over the graph nodes.
4.2.2 number_of_nodes
number_of_nodes(G)

4.2.3 neighbors
neighbors(G, n)
Returns a list of nodes connected to node n.
4.2.4 all_neighbors
all_neighbors(graph, node)
Returns all of the neighbors of a node in the graph.
If the graph is directed returns predecessors as well as successors.
Parameters
graph
[NetworkX graph] Graph to find neighbors.
node
[node] The node whose neighbors will be returned.
Returns
neighbors
[iterator] Iterator of neighbors
4.2.5 non_neighbors
non_neighbors(graph, node)
Returns the non-neighbors of the node in the graph.
Parameters
graph
[NetworkX graph] Graph to find neighbors.
node
[node] The node whose neighbors will be returned.
Returns
non_neighbors
[iterator] Iterator of nodes in the graph that are not neighbors of the node.
4.2.6 common_neighbors
common_neighbors(G, u, v)
Returns the common neighbors of two nodes in a graph.
Parameters
G
u, v
[nodes] Nodes in the graph.
Returns
4.2. Nodes 767

cnbors
[iterator] Iterator of common neighbors of u and v in the graph.
Raises
NetworkXError
If u or v is not a node in the graph.
Examples
>>> sorted(nx.common_neighbors(G, 0, 1))
[2, 3, 4]
4.3 Edges
edges(G[, nbunch]) Returns an edge view of edges incident to nodes in

nbunch.
number_of_edges(G) Returns the number of edges in the graph.
density(G) Returns the density of a graph.
non_edges(graph) Returns the non-existent edges in the graph.
4.3.1 edges
edges(G, nbunch=None)
Returns an edge view of edges incident to nodes in nbunch.
Return all edges if nbunch is unspecified or nbunch=None.
For digraphs, edges=out_edges
4.3.2 number_of_edges
number_of_edges(G)
Returns the number of edges in the graph.
4.3.3 non_edges
non_edges(graph)
Returns the non-existent edges in the graph.
Parameters
graph
[NetworkX graph.] Graph to find non-existent edges.
Returns
non_edges
[iterator] Iterator of edges that are not in the graph.

4.4 Self loops
selfloop_edges(G[, data, keys, default]) Returns an iterator over selfloop edges.

number_of_selfloops(G) Returns the number of selfloop edges.
nodes_with_selfloops(G) Returns an iterator over nodes with self loops.
4.4.1 selfloop_edges
selfloop_edges(G, data=False, keys=False, default=None)

Returns an iterator over selfloop edges.
A selfloop edge has the same node at both ends.
Parameters
G
data
[string or bool, optional (default=False)] Return selfloop edges as two tuples (u, v) (data=False)
or three-tuples (u, v, datadict) (data=True) or three-tuples (u, v, datavalue) (data=’attrname’)
keys
[bool, optional (default=False)] If True, return edge keys with each edge.
default
Returns
edgeiter
[iterator over edge tuples] An iterator over all selfloop edges.
See also:
nodes_with_selfloops, number_of_selfloops
Examples
>>> G = nx.MultiGraph() # or Graph, DiGraph, MultiDiGraph, etc

>>> list(nx.selfloop_edges(G))
[(1, 1)]
>>> list(nx.selfloop_edges(G, data=True))
[(1, 1, {})]
>>> list(nx.selfloop_edges(G, keys=True))
[(1, 1, 0)]
>>> list(nx.selfloop_edges(G, keys=True, data=True))
[(1, 1, 0, {})]
4.4. Self loops 769

4.4.2 number_of_selfloops
number_of_selfloops(G)
Returns the number of selfloop edges.
A selfloop edge has the same node at both ends.
Returns
nloops
[int] The number of selfloops.
See also:
nodes_with_selfloops, selfloop_edges
Examples

>>> nx.number_of_selfloops(G)
1
4.4.3 nodes_with_selfloops
nodes_with_selfloops(G)
Returns an iterator over nodes with self loops.
A node with a self loop has an edge with both ends adjacent to that node.
Returns
nodelist
[iterator] A iterator over nodes with self loops.
See also:
selfloop_edges, number_of_selfloops
Examples

>>> list(nx.nodes_with_selfloops(G))
[1]

4.5 Attributes
is_weighted(G[, edge, weight]) Returns True if G has weighted edges.

is_negatively_weighted(G[, edge, weight]) Returns True if G has negatively weighted edges.
set_node_attributes(G, values[, name]) Sets node attributes from a given value or dictionary of
values.
get_node_attributes(G, name) Get node attributes from graph
set_edge_attributes(G, values[, name]) Sets edge attributes from a given value or dictionary of
values.
get_edge_attributes(G, name) Get edge attributes from graph
4.5.1 is_weighted
is_weighted(G, edge=None, weight='weight')

Returns True if G has weighted edges.
Parameters
G
edge
[tuple, optional] A 2-tuple specifying the only edge in G that will be tested. If None, then every
edge in G is tested.
weight: string, optional
The attribute name used to query for edge weights.
Returns
bool
A boolean signifying if G, or the specified edge, is weighted.
Raises
NetworkXError
If the specified edge does not exist.
Examples
>>> nx.is_weighted(G)
False
>>> nx.is_weighted(G, (2, 3))
False
>>> nx.is_weighted(G)
True
4.5. Attributes 771

4.5.2 is_negatively_weighted
is_negatively_weighted(G, edge=None, weight='weight')

Returns True if G has negatively weighted edges.
Parameters
G
edge
[tuple, optional] A 2-tuple specifying the only edge in G that will be tested. If None, then every
edge in G is tested.
weight: string, optional
The attribute name used to query for edge weights.
Returns
bool
A boolean signifying if G, or the specified edge, is negatively weighted.
Raises
NetworkXError
Examples
>>> G = nx.Graph()
>>> G.add_edges_from([(1, 3), (2, 4), (2, 6)])
>>> nx.is_negatively_weighted(G, (1, 2))
False
>>> G[2][4]["weight"] = -2
>>> nx.is_negatively_weighted(G)
True
>>> edges = [("0", "3", 3), ("0", "1", -5), ("1", "0", -2)]
>>> G.add_weighted_edges_from(edges)
>>> nx.is_negatively_weighted(G)
True
4.5.3 set_node_attributes
set_node_attributes(G, values, name=None)

Sets node attributes from a given value or dictionary of values.
Warning: The call order of arguments values and name switched between v1.x & v2.x.
Parameters
G
[NetworkX Graph]

values
[scalar value, dict-like] What the node attribute should be set to. If values is not a dictionary,
then it is treated as a single attribute value that is then applied to every node in G. This means
that if you provide a mutable object, like a list, updates to that object will be reflected in the
node attribute for every node. The attribute name will be name.
If values is a dict or a dict of dict, it should be keyed by node to either an attribute value or
a dict of attribute key/value pairs used to update the node’s attributes.
name
[string (optional, default=None)] Name of the node attribute to set if values is a scalar.
Examples
After computing some property of the nodes of a graph, you may want to assign a node attribute to store the value
of that property for each node:
>>> bb = nx.betweenness_centrality(G)
>>> isinstance(bb, dict)
True
>>> nx.set_node_attributes(G, bb, "betweenness")
>>> G.nodes[1]["betweenness"]
1.0
If you provide a list as the second argument, updates to the list will be reflected in the node attribute for each node:
>>> labels = []
>>> nx.set_node_attributes(G, labels, "labels")
>>> labels.append("foo")
>>> G.nodes[0]["labels"]
['foo']
['foo']
['foo']
If you provide a dictionary of dictionaries as the second argument, the outer dictionary is assumed to be keyed by
node to an inner dictionary of node attributes for that node:
>>> attrs = {0: {"attr1": 20, "attr2": "nothing"}, 1: {"attr2": 3}}
>>> nx.set_node_attributes(G, attrs)
>>> G.nodes[0]["attr1"]
20
'nothing'
3
>>> G.nodes[2]
{}
Note that if the dictionary contains nodes that are not in G, the values are silently ignored:
>>> G = nx.Graph()
>>> G.add_node(0)
4.5. Attributes 773


>>> nx.set_node_attributes(G, {0: "red", 1: "blue"}, name="color")
>>> G.nodes[0]["color"]
'red'
>>> 1 in G.nodes
False
4.5.4 get_node_attributes
get_node_attributes(G, name)
Get node attributes from graph
Parameters
G
[NetworkX Graph]
name
[string] Attribute name
Returns
Dictionary of attributes keyed by node.
Examples
>>> G = nx.Graph()
>>> G.add_nodes_from([1, 2, 3], color="red")
>>> color = nx.get_node_attributes(G, "color")
>>> color[1]
'red'
4.5.5 set_edge_attributes
set_edge_attributes(G, values, name=None)

Sets edge attributes from a given value or dictionary of values.
Warning: The call order of arguments values and name switched between v1.x & v2.x.
Parameters
G
[NetworkX Graph]
values
[scalar value, dict-like] What the edge attribute should be set to. If values is not a dictionary,
then it is treated as a single attribute value that is then applied to every edge in G. This means
that if you provide a mutable object, like a list, updates to that object will be reflected in the
edge attribute for each edge. The attribute name will be name.
If values is a dict or a dict of dict, it should be keyed by edge tuple to either an attribute value
or a dict of attribute key/value pairs used to update the edge’s attributes. For multigraphs, the

edge tuples must be of the form (u, v, key), where u and v are nodes and key is the
edge key. For non-multigraphs, the keys must be tuples of the form (u, v).
name
[string (optional, default=None)] Name of the edge attribute to set if values is a scalar.
Examples
After computing some property of the edges of a graph, you may want to assign a edge attribute to store the value
of that property for each edge:
>>> bb = nx.edge_betweenness_centrality(G, normalized=False)
>>> nx.set_edge_attributes(G, bb, "betweenness")
>>> G.edges[1, 2]["betweenness"]
2.0
If you provide a list as the second argument, updates to the list will be reflected in the edge attribute for each edge:
>>> labels = []
>>> nx.set_edge_attributes(G, labels, "labels")
>>> labels.append("foo")
>>> G.edges[0, 1]["labels"]
['foo']
>>> G.edges[1, 2]["labels"]
['foo']
If you provide a dictionary of dictionaries as the second argument, the entire dictionary will be used to update edge
attributes:
>>> attrs = {(0, 1): {"attr1": 20, "attr2": "nothing"}, (1, 2): {"attr2": 3}}
>>> nx.set_edge_attributes(G, attrs)
>>> G[0][1]["attr1"]
20
>>> G[0][1]["attr2"]
'nothing'
>>> G[1][2]["attr2"]
3
The attributes of one Graph can be used to set those of another.

>>> nx.set_edge_attributes(H, G.edges)
Note that if the dict contains edges that are not in G, they are silently ignored:
>>> G = nx.Graph([(0, 1)])
>>> nx.set_edge_attributes(G, {(1, 2): {"weight": 2.0}})
>>> (1, 2) in G.edges()
False
For multigraphs, the values dict is expected to be keyed by 3-tuples including the edge key:
>>> MG = nx.MultiGraph()
>>> edges = [(0, 1), (0, 1)]
>>> MG.add_edges_from(edges) # Returns list of edge keys
4.5. Attributes 775


[0, 1]
>>> attributes = {(0, 1, 0): {"cost": 21}, (0, 1, 1): {"cost": 7}}
>>> nx.set_edge_attributes(MG, attributes)
>>> MG[0][1][0]["cost"]
21
>>> MG[0][1][1]["cost"]
7
If MultiGraph attributes are desired for a Graph, you must convert the 3-tuple multiedge to a 2-tuple edge and the
last multiedge’s attribute value will overwrite the previous values. Continuing from the previous case we get:
>>> H = nx.path_graph([0, 1, 2])

>>> nx.set_edge_attributes(H, {(u, v): ed for u, v, ed in MG.edges.data()})
>>> nx.get_edge_attributes(H, "cost")
{(0, 1): 7}
4.5.6 get_edge_attributes
get_edge_attributes(G, name)
Get edge attributes from graph
Parameters
G
[NetworkX Graph]
name
[string] Attribute name
Returns
Dictionary of attributes keyed by edge. For (di)graphs, the keys are
2-tuples of the form: (u, v). For multi(di)graphs, the keys are 3-tuples of
the form: (u, v, key).

Examples
>>> G = nx.Graph()
>>> nx.add_path(G, [1, 2, 3], color="red")
>>> color = nx.get_edge_attributes(G, "color")
>>> color[(1, 2)]
'red'
4.6 Paths
is_path(G, path) Returns whether or not the specified path exists.

path_weight(G, path, weight) Returns total cost associated with specified path and
weight
4.6.1 is_path
is_path(G, path)
Returns whether or not the specified path exists.
For it to return True, every node on the path must exist and each consecutive pair must be connected via one or
more edges.
Parameters
G
path
[list] A list of nodes which defines the path to traverse
Returns
bool
True if path is a valid path in G
4.6.2 path_weight
path_weight(G, path, weight)

Returns total cost associated with specified path and weight
Parameters
G
path: list
A list of node labels which defines the path to traverse
weight: string
A string indicating which edge attribute to use for path cost
Returns
4.6. Paths 777

cost: int or float

An integer or a float representing the total cost with respect to the specified weight of the
specified path
Raises
NetworkXNoPath
4.7 Freezing graph structure
freeze(G) Modify graph to prevent further change by adding or re-

moving nodes or edges.
is_frozen(G) Returns True if graph is frozen.
4.7.1 freeze
freeze(G)
Modify graph to prevent further change by adding or removing nodes or edges.
Node and edge data can still be modified.
Parameters
G
See also:
is_frozen
Notes
To “unfreeze” a graph you must make a copy by creating a new graph object:
>>> graph = nx.path_graph(4)

>>> frozen_graph = nx.freeze(graph)
>>> unfrozen_graph = nx.Graph(frozen_graph)
>>> nx.is_frozen(unfrozen_graph)
False
Examples
>>> G = nx.freeze(G)
>>> try:
... G.add_edge(4, 5)
... except nx.NetworkXError as err:
... print(str(err))
Frozen graph can't be modified

4.7.2 is_frozen
is_frozen(G)
Returns True if graph is frozen.
Parameters
G
See also:
freeze
4.7. Freezing graph structure 779


CHAPTER
FIVE
GRAPH GENERATORS
5.1 Atlas
Generators for the small graph atlas.
graph_atlas(i) Returns graph number i from the Graph Atlas.

graph_atlas_g() Returns the list of all graphs with up to seven nodes named
in the Graph Atlas.
5.1.1 graph_atlas
graph_atlas(i)
Returns graph number i from the Graph Atlas.
For more information, see graph_atlas_g().
Parameters
i
[int] The index of the graph from the atlas to get. The graph at index 0 is assumed to be the
null graph.
Returns
list
A list of Graph objects, the one at index i corresponding to the graph i in the Graph Atlas.
See also:
graph_atlas_g
Notes
The time required by this function increases linearly with the argument i, since it reads a large file sequentially in
order to generate the graph [1].
781
References
[1]
5.1.2 graph_atlas_g
graph_atlas_g()
Returns the list of all graphs with up to seven nodes named in the Graph Atlas.
The graphs are listed in increasing order by
1. number of nodes,
2. number of edges,
3. degree sequence (for example 111223 < 112222),
4. number of automorphisms,
in that order, with three exceptions as described in the Notes section below. This causes the list to correspond with
the index of the graphs in the Graph Atlas [atlas], with the first graph, G[0], being the null graph.
Returns
list
A list of Graph objects, the one at index i corresponding to the graph i in the Graph Atlas.
See also:
graph_atlas
Notes
This function may be expensive in both time and space, since it reads a large file sequentially in order to populate
the list.
Although the NetworkX atlas functions match the order of graphs given in the “Atlas of Graphs” book, there
are (at least) three errors in the ordering described in the book. The following three pairs of nodes violate the
lexicographically nondecreasing sorted degree sequence rule:
• graphs 55 and 56 with degree sequences 001111 and 000112,
• graphs 1007 and 1008 with degree sequences 3333444 and 3333336,
• graphs 1012 and 1213 with degree sequences 1244555 and 1244456.
References
[atlas]
782 Chapter 5. Graph generators

5.2 Classic
Generators for some classic graphs.

The typical graph builder function is called as follows:
returning the complete graph on n nodes labeled 0, .., 99 as a simple graph. Except for empty_graph, all the functions
in this module return a Graph class (i.e. a simple, undirected graph).
balanced_tree(r, h[, create_using]) Returns the perfectly balanced r-ary tree of height h.
barbell_graph(m1, m2[, create_using]) Returns the Barbell Graph: two complete graphs con-
nected by a path.
binomial_tree(n[, create_using]) Returns the Binomial Tree of order n.
complete_graph(n[, create_using]) Return the complete graph K_n with n nodes.
complete_multipartite_graph(*subset_sizes) Returns the complete multipartite graph with the specified
subset sizes.
circular_ladder_graph(n[, create_using]) Returns the circular ladder graph CLn of length n.
circulant_graph(n, offsets[, create_using]) Returns the circulant graph Cin (x1 , x2 , ..., xm ) with n
nodes.
cycle_graph(n[, create_using]) Returns the cycle graph Cn of cyclically connected nodes.
dorogovtsev_goltsev_mendes_graph(n[, ...]) Returns the hierarchically constructed Dorogovtsev-
Goltsev-Mendes graph.
empty_graph([n, create_using, default]) Returns the empty graph with n nodes and zero edges.
full_rary_tree(r, n[, create_using]) Creates a full r-ary tree of n nodes.
ladder_graph(n[, create_using]) Returns the Ladder graph of length n.
lollipop_graph(m, n[, create_using]) Returns the Lollipop Graph; K_m connected to P_n.
null_graph([create_using]) Returns the Null graph with no nodes or edges.
path_graph(n[, create_using]) Returns the Path graph P_n of linearly connected nodes.
star_graph(n[, create_using]) Return the star graph
trivial_graph([create_using]) Return the Trivial graph with one node (with label 0) and
no edges.
turan_graph(n, r) Return the Turan Graph
wheel_graph(n[, create_using]) Return the wheel graph
5.2.1 balanced_tree
balanced_tree(r, h, create_using=None)
Returns the perfectly balanced r-ary tree of height h.
Parameters
r
[int] Branching factor of the tree; each node will have r children.
h
[int] Height of the tree.
create_using
Returns
5.2. Classic 783

G
[NetworkX graph] A balanced r-ary tree of height h.
Notes
This is the rooted tree where all leaves are at distance h from the root. The root has degree r and all other internal
nodes have degree r + 1.
Node labels are integers, starting from zero.
A balanced tree is also known as a complete r-ary tree.
5.2.2 barbell_graph
barbell_graph(m1, m2, create_using=None)

Returns the Barbell Graph: two complete graphs connected by a path.
For m1 > 1 and m2 >= 0.
Two identical complete graphs Km1 form the left and right bells, and are connected by a path Pm2 .
The 2*m1+m2 nodes are numbered
0, ..., m1-1 for the left barbell, m1, ..., m1+m2-1 for the path, and m1+m2, ..., 2*m1+m2-1
for the right barbell.
The 3 subgraphs are joined via the edges (m1-1, m1) and (m1+m2-1, m1+m2). If m2=0, this is merely two
complete graphs joined together.
This graph is an extremal example in David Aldous and Jim Fill’s e-text on Random Walks on Graphs.
5.2.3 binomial_tree
binomial_tree(n, create_using=None)
Returns the Binomial Tree of order n.
The binomial tree of order 0 consists of a single node. A binomial tree of order k is defined recursively by linking
two binomial trees of order k-1: the root of one is the leftmost child of the root of the other.
Parameters
n
[int] Order of the binomial tree.
create_using
Returns
G
[NetworkX graph] A binomial tree of 2n nodes and 2n − 1 edges.

5.2.4 complete_graph
complete_graph(n, create_using=None)
Return the complete graph K_n with n nodes.
A complete graph on n nodes means that all pairs of distinct nodes have an edge connecting them.
Parameters
n
[int or iterable container of nodes] If n is an integer, nodes are from range(n). If n is a container
of nodes, those nodes appear in the graph. Warning: n is not checked for duplicates and if
present the resulting graph may not be as desired. Make sure you have no duplicates.
create_using
Examples
>>> len(G)
9
>>> G.size()
36
>>> G = nx.complete_graph(range(11, 14))
>>> list(G.nodes())
[11, 12, 13]
>>> G = nx.complete_graph(4, nx.DiGraph())
>>> G.is_directed()
True
5.2.5 complete_multipartite_graph
complete_multipartite_graph(*subset_sizes)
Returns the complete multipartite graph with the specified subset sizes.
Parameters
subset_sizes
[tuple of integers or tuple of node iterables] The arguments can either all be integer number
of nodes or they can all be iterables of nodes. If integers, they represent the number of nodes
in each subset of the multipartite graph. If iterables, each is used to create the nodes for that
subset. The length of subset_sizes is the number of subsets.
Returns
G
[NetworkX Graph] Returns the complete multipartite graph with the specified subsets.
For each node, the node attribute ‘subset’ is an integer indicating which subset contains the
node.
See also:
complete_bipartite_graph
5.2. Classic 785

Notes
This function generalizes several other graph builder functions.

• If no subset sizes are given, this returns the null graph.
• If a single subset size n is given, this returns the empty graph on n nodes.
• If two subset sizes m and n are given, this returns the complete bipartite graph on m + n nodes.
• If subset sizes 1 and n are given, this returns the star graph on n + 1 nodes.
Examples
Creating a complete tripartite graph, with subsets of one, two, and three nodes, respectively.
>>> G = nx.complete_multipartite_graph(1, 2, 3)
>>> [G.nodes[u]["subset"] for u in G]
[0, 1, 1, 2, 2, 2]
[(0, 1), (0, 2), (0, 3), (0, 4), (0, 5)]
[(2, 0), (2, 3), (2, 4), (2, 5)]
[(4, 0), (4, 1), (4, 2)]
>>> G = nx.complete_multipartite_graph("a", "bc", "def")

>>> [G.nodes[u]["subset"] for u in sorted(G)]
[0, 1, 1, 2, 2, 2]
5.2.6 circular_ladder_graph
circular_ladder_graph(n, create_using=None)
Returns the circular ladder graph CLn of length n.
CLn consists of two concentric n-cycles in which each of the n pairs of concentric nodes are joined by an edge.
Node labels are the integers 0 to n-1
5.2.7 circulant_graph
circulant_graph(n, offsets, create_using=None)

Returns the circulant graph Cin (x1 , x2 , ..., xm ) with n nodes.
The circulant graph Cin (x1 , ..., xm ) consists of n nodes 0, ..., n − 1 such that node i is connected to nodes (i + x)
mod n and (i − x) mod n for all x in x1 , ..., xm . Thus Cin (1) is a cycle graph.
Parameters
n
[integer] The number of nodes in the graph.
offsets
[list of integers] A list of node offsets, x1 up to xm , as described above.

create_using
Returns
NetworkX Graph of type create_using
Examples
Many well-known graph families are subfamilies of the circulant graphs; for example, to create the cycle graph on
n points, we connect every node to nodes on either side (with offset plus or minus one). For n = 10,
>>> G = nx.circulant_graph(10, [1])

>>> edges = [
... (0, 9),
... (0, 1),
... (1, 2),
... (2, 3),
... (3, 4),
... (4, 5),
... (5, 6),
... (6, 7),
... (7, 8),
... (8, 9),
... ]
...
>>> sorted(edges) == sorted(G.edges())
True
Similarly, we can create the complete graph on 5 points with the set of offsets [1, 2]:
>>> G = nx.circulant_graph(5, [1, 2])

>>> edges = [
... (0, 1),
... (0, 2),
... (0, 3),
... (0, 4),
... (1, 2),
... (1, 3),
... (1, 4),
... (2, 3),
... (2, 4),
... (3, 4),
... ]
...
>>> sorted(edges) == sorted(G.edges())
True
5.2. Classic 787

5.2.8 cycle_graph
cycle_graph(n, create_using=None)
Returns the cycle graph Cn of cyclically connected nodes.
Cn is a path with its two end-nodes connected.
Parameters
n
[int or iterable container of nodes] If n is an integer, nodes are from range(n). If n is a
container of nodes, those nodes appear in the graph. Warning: n is not checked for duplicates
and if present the resulting graph may not be as desired. Make sure you have no duplicates.
create_using
Notes
If create_using is directed, the direction is in increasing order.
5.2.9 dorogovtsev_goltsev_mendes_graph
dorogovtsev_goltsev_mendes_graph(n, create_using=None)
Returns the hierarchically constructed Dorogovtsev-Goltsev-Mendes graph.
n is the generation. See: arXiv:/cond-mat/0112143 by Dorogovtsev, Goltsev and Mendes.
5.2.10 empty_graph
empty_graph(n=0, create_using=None, default=<class 'networkx.classes.graph.Graph'>)

Returns the empty graph with n nodes and zero edges.
Parameters
n
[int or iterable container of nodes (default = 0)] If n is an integer, nodes are from range(n).
If n is a container of nodes, those nodes appear in the graph.
create_using
[Graph Instance, Constructor or None] Indicator of type of graph to return. If a Graph-type
instance, then clear and use it. If None, use the default constructor. If a constructor, call
it to create an empty graph.
default
[Graph constructor (optional, default = nx.Graph)] The constructor to use if create_using
is None. If None, then nx.Graph is used. This is used when passing an unknown cre-
ate_using value through your home-grown function to empty_graph and you want a
default constructor other than nx.Graph.

Notes
The variable create_using should be a Graph Constructor or a “graph”-like object. Constructors, e.g. nx.Graph
or nx.MultiGraph will be used to create the returned graph. “graph”-like objects will be cleared (nodes and
edges will be removed) and refitted as an empty “graph” with nodes specified in n. This capability is useful for
specifying the class-nature of the resulting empty “graph” (i.e. Graph, DiGraph, MyWeirdGraphClass, etc.).
The variable create_using has three main uses: Firstly, the variable create_using can be used to create an empty
digraph, multigraph, etc. For example,
>>> n = 10
>>> G = nx.empty_graph(n, create_using=nx.DiGraph)
will create an empty digraph on n nodes.

Secondly, one can pass an existing graph (digraph, multigraph, etc.) via create_using. For example, if G is an
existing graph (resp. digraph, multigraph, etc.), then empty_graph(n, create_using=G) will empty G (i.e. delete all
nodes and edges using G.clear()) and then add n nodes and zero edges, and return the modified graph.
Thirdly, when constructing your home-grown graph creation function you can use empty_graph to construct the
graph by passing a user defined create_using to empty_graph. In this case, if you want the default constructor to
be other than nx.Graph, specify default.
>>> def mygraph(n, create_using=None):

... G = nx.empty_graph(n, create_using, nx.MultiGraph)
... G.add_edges_from([(0, 1), (0, 1)])
... return G
>>> G = mygraph(3)
>>> G.is_multigraph()
True
>>> G = mygraph(3, nx.Graph)
>>> G.is_multigraph()
False
See also create_empty_copy(G).
Examples
>>> G = nx.empty_graph(10)
10
0
>>> G = nx.empty_graph("ABC")
3
>>> sorted(G)
['A', 'B', 'C']
5.2. Classic 789

5.2.11 full_rary_tree
full_rary_tree(r, n, create_using=None)
Creates a full r-ary tree of n nodes.
Sometimes called a k-ary, n-ary, or m-ary tree. “… all non-leaf nodes have exactly r children and all levels are full
except for some rightmost position of the bottom level (if a leaf at the bottom level is missing, then so are all of the
leaves to its right.” [1]
Parameters
r
[int] branching factor of the tree
n
[int] Number of nodes in the tree
create_using
Returns
G
[networkx Graph] An r-ary tree with n nodes
References
[1]
5.2.12 ladder_graph
ladder_graph(n, create_using=None)
Returns the Ladder graph of length n.
This is two paths of n nodes, with each pair connected by a single edge.
Node labels are the integers 0 to 2*n - 1.
5.2.13 lollipop_graph
lollipop_graph(m, n, create_using=None)
Returns the Lollipop Graph; K_m connected to P_n.
This is the Barbell Graph without the right barbell.
Parameters
m, n
[int or iterable container of nodes (default = 0)] If an integer, nodes are from range(m) and
range(m,m+n). If a container of nodes, those nodes appear in the graph. Warning: m and
n are not checked for duplicates and if present the resulting graph may not be as desired. Make
sure you have no duplicates.
The nodes for m appear in the complete graph Km and the nodes for n appear in the path Pn

create_using
Notes
The 2 subgraphs are joined via an edge (m-1, m). If n=0, this is merely a complete graph.
(This graph is an extremal example in David Aldous and Jim Fill’s etext on Random Walks on Graphs.)
5.2.14 null_graph
null_graph(create_using=None)
Returns the Null graph with no nodes or edges.
See empty_graph for the use of create_using.
5.2.15 path_graph
path_graph(n, create_using=None)
Returns the Path graph P_n of linearly connected nodes.
Parameters
n
[int or iterable] If an integer, nodes are 0 to n - 1. If an iterable of nodes, in the order they
appear in the path. Warning: n is not checked for duplicates and if present the resulting graph
may not be as desired. Make sure you have no duplicates.
create_using
5.2.16 star_graph
star_graph(n, create_using=None)
Return the star graph
The star graph consists of one center node connected to n outer nodes.
Parameters
n
[int or iterable] If an integer, node labels are 0 to n with center 0. If an iterable of nodes, the
center is the first. Warning: n is not checked for duplicates and if present the resulting graph
create_using
5.2. Classic 791

Notes
The graph has n+1 nodes for integer n. So star_graph(3) is the same as star_graph(range(4)).
5.2.17 trivial_graph
trivial_graph(create_using=None)
Return the Trivial graph with one node (with label 0) and no edges.
5.2.18 turan_graph
turan_graph(n, r)
Return the Turan Graph
The Turan Graph is a complete multipartite graph on n nodes with r disjoint subsets. That is, edges connect each
node to every node not in its subset.
Given n and r, we create a complete multipartite graph with r − (n mod r) partitions of size n/r, rounded down,
and n mod r partitions of size n/r + 1, rounded down.
Parameters
n
[int] The number of nodes.
r
[int] The number of partitions. Must be less than or equal to n.
Notes
Must satisfy 1 <= r <= n. The graph has (r − 1)(n2 )/(2r) edges, rounded down.
5.2.19 wheel_graph
wheel_graph(n, create_using=None)
Return the wheel graph
The wheel graph consists of a hub node connected to a cycle of (n-1) nodes.
Parameters
n
[int or iterable] If an integer, node labels are 0 to n with center 0. If an iterable of nodes, the
center is the first. Warning: n is not checked for duplicates and if present the resulting graph
create_using
Node labels are the integers 0 to n - 1.

5.3 Expanders
Provides explicit constructions of expander graphs.
margulis_gabber_galil_graph(n[, cre- Returns the Margulis-Gabber-Galil undirected Multi-

ate_using]) Graph on n^2 nodes.
chordal_cycle_graph(p[, create_using]) Returns the chordal cycle graph on p nodes.
paley_graph(p[, create_using]) Returns the Paley (p-1)/2-regular graph on p nodes.
5.3.1 margulis_gabber_galil_graph
margulis_gabber_galil_graph(n, create_using=None)
Returns the Margulis-Gabber-Galil undirected MultiGraph on n^2 nodes.
The undirected MultiGraph is regular with degree 8. Nodes are integer pairs. The second-largest eigenvalue of the
adjacency matrix of the graph is at most 5 sqrt{2}, regardless of n.
Parameters
n
[int] Determines the number of nodes in the graph: n^2.
create_using
[NetworkX graph constructor, optional (default MultiGraph)] Graph type to create. If graph
Returns
G
[graph] The constructed undirected multigraph.
Raises
NetworkXError
If the graph is directed or not a multigraph.
5.3.2 chordal_cycle_graph
chordal_cycle_graph(p, create_using=None)
Returns the chordal cycle graph on p nodes.
The returned graph is a cycle graph on p nodes with chords joining each vertex x to its inverse modulo p. This
graph is a (mildly explicit) 3-regular expander [1].
p must be a prime number.
Parameters
p
[a prime number] The number of vertices in the graph. This also indicates where the chordal
edges in the cycle will be created.
create_using
Returns
5.3. Expanders 793

G
[graph] The constructed undirected multigraph.
Raises
NetworkXError
If create_using indicates directed or not a multigraph.
References
[1]
5.3.3 paley_graph
paley_graph(p, create_using=None)
Returns the Paley (p-1)/2-regular graph on p nodes.
The returned graph is a graph on Z/pZ with edges between x and y if and only if x-y is a nonzero square in Z/pZ.
If p = 1 mod 4, -1 is a square in Z/pZ and therefore x-y is a square if and only if y-x is also a square, i.e the edges
in the Paley graph are symmetric.
If p = 3 mod 4, -1 is not a square in Z/pZ and therefore either x-y or y-x is a square in Z/pZ but not both.
Note that a more general definition of Paley graphs extends this construction to graphs over q=p^n vertices, by
using the finite field F_q instead of Z/pZ. This construction requires to compute squares in general finite fields and
is not what is implemented here (i.e paley_graph(25) does not return the true Paley graph associated with 5^2).
Parameters
p
[int, an odd prime number.]
create_using
Returns
G
[graph] The constructed directed graph.
Raises
NetworkXError
If the graph is a multigraph.
References
Chapter 13 in B. Bollobas, Random Graphs. Second edition. Cambridge Studies in Advanced Mathematics, 73.
Cambridge University Press, Cambridge (2001).

5.4 Lattice
Functions for generating grid graphs and lattices

The grid_2d_graph(), triangular_lattice_graph(), and hexagonal_lattice_graph() func-
tions correspond to the three regular tilings of the plane, the square, triangular, and hexagonal tilings, respectively.
grid_graph() and hypercube_graph() are similar for arbitrary dimensions. Useful relevant discussion can
be found about Triangular Tiling, and Square, Hex and Triangle Grids
grid_2d_graph(m, n[, periodic, create_using]) Returns the two-dimensional grid graph.

grid_graph(dim[, periodic]) Returns the n-dimensional grid graph.
hexagonal_lattice_graph(m, n[, periodic, ...]) Returns an m by n hexagonal lattice graph.
hypercube_graph(n) Returns the n-dimensional hypercube graph.
triangular_lattice_graph(m, n[, periodic, ...]) Returns the m by n triangular lattice graph.
5.4.1 grid_2d_graph
grid_2d_graph(m, n, periodic=False, create_using=None)

Returns the two-dimensional grid graph.
The grid graph has each node connected to its four nearest neighbors.
Parameters
m, n
[int or iterable container of nodes] If an integer, nodes are from range(n). If a container,
elements become the coordinate of the nodes.
periodic
[bool or iterable] If periodic is True, both dimensions are periodic. If False, none are
periodic. If periodic is iterable, it should yield 2 bool values indicating whether the 1st and
2nd axes, respectively, are periodic.
create_using
Returns
NetworkX graph
The (possibly periodic) grid graph of the specified dimensions.
5.4.2 grid_graph
grid_graph(dim, periodic=False)
Returns the n-dimensional grid graph.
The dimension n is the length of the list dim and the size in each dimension is the value of the corresponding list
element.
Parameters
dim
[list or tuple of numbers or iterables of nodes] ‘dim’ is a tuple or list with, for each dimension,
either a number that is the size of that dimension or an iterable of nodes for that dimension.
The dimension of the grid_graph is the length of dim.
5.4. Lattice 795

periodic
[bool or iterable] If periodic is True, all dimensions are periodic. If False all dimensions
are not periodic. If periodic is iterable, it should yield dim bool values each of which
indicates whether the corresponding axis is periodic.
Returns
NetworkX graph
The (possibly periodic) grid graph of the specified dimensions.
Examples
To produce a 2 by 3 by 4 grid graph, a graph on 24 nodes:
>>> from networkx import grid_graph

>>> G = grid_graph(dim=(2, 3, 4))
>>> len(G)
24
>>> G = grid_graph(dim=(range(7, 9), range(3, 6)))
>>> len(G)
6
5.4.3 hexagonal_lattice_graph
hexagonal_lattice_graph(m, n, periodic=False, with_positions=True, create_using=None)

Returns an m by n hexagonal lattice graph.
The hexagonal lattice graph is a graph whose nodes and edges are the hexagonal tiling of the plane.
The returned graph will have m rows and n columns of hexagons. Odd numbered columns are shifted up relative
to even numbered columns.
Positions of nodes are computed by default or with_positions is True. Node positions creating the
standard embedding in the plane with sidelength 1 and are stored in the node attribute ‘pos’. pos = nx.
get_node_attributes(G, 'pos') creates a dict ready for drawing.
Parameters
m
[int] The number of rows of hexagons in the lattice.
n
[int] The number of columns of hexagons in the lattice.
periodic
[bool] Whether to make a periodic grid by joining the boundary vertices. For this to work n
must be even and both n > 1 and m > 1. The periodic connections create another row and
column of hexagons so these graphs have fewer nodes as boundary nodes are identified.
with_positions
[bool (default: True)] Store the coordinates of each node in the graph node attribute ‘pos’. The
coordinates provide a lattice with vertical columns of hexagons offset to interleave and cover
the plane. Periodic positions shift the nodes vertically in a nonlinear way so the edges don’t
overlap so much.
create_using
instance, then cleared before populated. If graph is directed, edges will point up or right.

Returns
NetworkX graph
The m by n hexagonal lattice graph.
5.4.4 hypercube_graph
hypercube_graph(n)
Returns the n-dimensional hypercube graph.
The nodes are the integers between 0 and 2 ** n - 1, inclusive.
For more information on the hypercube graph, see the Wikipedia article Hypercube graph.
Parameters
n
[int] The dimension of the hypercube. The number of nodes in the graph will be 2 ** n.
Returns
NetworkX graph
The hypercube graph of dimension n.
5.4.5 triangular_lattice_graph
triangular_lattice_graph(m, n, periodic=False, with_positions=True, create_using=None)

Returns the m by n triangular lattice graph.
The triangular lattice graph is a two-dimensional grid graph in which each square unit has a diagonal edge (each
grid unit has a chord).
The returned graph has m rows and n columns of triangles. Rows and columns include both triangles pointing up
and down. Rows form a strip of constant height. Columns form a series of diamond shapes, staggered with the
columns on either side. Another way to state the size is that the nodes form a grid of m+1 rows and (n + 1) //
2 columns. The odd row nodes are shifted horizontally relative to the even rows.
Directed graph types have edges pointed up or right.
Positions of nodes are computed by default or with_positions is True. The position of each node
(embedded in a euclidean plane)
√ is stored in the graph using equilateral triangles with sidelength 1. The height
between rows of nodes is thus (3)/2. Nodes lie in the first quadrant with the node (0, 0) at the origin.
Parameters
m
[int] The number of rows in the lattice.
n
[int] The number of columns in the lattice.
periodic
[bool (default: False)] If True, join the boundary vertices of the grid using periodic boundary
conditions. The join between boundaries is the final row and column of triangles. This means
there is one row and one column fewer nodes for the periodic lattice. Periodic lattices require
m >= 3, n >= 5 and are allowed but misaligned if m or n are odd
5.4. Lattice 797

with_positions
[bool (default: True)] Store the coordinates of each node in the graph node attribute ‘pos’.
The coordinates provide a lattice with equilateral triangles. Periodic positions shift the nodes
vertically in a nonlinear way so the edges don’t overlap so much.
create_using
Returns
NetworkX graph
The m by n triangular lattice graph.
5.5 Small
Various small and named graphs, together with some compact generators.
LCF_graph(n, shift_list, repeats[, create_using]) Return the cubic graph specified in LCF notation.
bull_graph([create_using]) Returns the Bull Graph
chvatal_graph([create_using]) Returns the Chvátal Graph
cubical_graph([create_using]) Returns the 3-regular Platonic Cubical Graph
desargues_graph([create_using]) Returns the Desargues Graph
diamond_graph([create_using]) Returns the Diamond graph
dodecahedral_graph([create_using]) Returns the Platonic Dodecahedral graph.
frucht_graph([create_using]) Returns the Frucht Graph.
heawood_graph([create_using]) Returns the Heawood Graph, a (3,6) cage.
hoffman_singleton_graph() Returns the Hoffman-Singleton Graph.
house_graph([create_using]) Returns the House graph (square with triangle on top)
house_x_graph([create_using]) Returns the House graph with a cross inside the house
square.
icosahedral_graph([create_using]) Returns the Platonic Icosahedral graph.
krackhardt_kite_graph([create_using]) Returns the Krackhardt Kite Social Network.
moebius_kantor_graph([create_using]) Returns the Moebius-Kantor graph.
octahedral_graph([create_using]) Returns the Platonic Octahedral graph.
pappus_graph() Returns the Pappus graph.
petersen_graph([create_using]) Returns the Petersen graph.
sedgewick_maze_graph([create_using]) Return a small maze with a cycle.
tetrahedral_graph([create_using]) Returns the 3-regular Platonic Tetrahedral graph.
truncated_cube_graph([create_using]) Returns the skeleton of the truncated cube.
truncated_tetrahedron_graph([create_using]) Returns the skeleton of the truncated Platonic tetrahe-
dron.
tutte_graph([create_using]) Returns the Tutte graph.

5.5.1 LCF_graph
LCF_graph(n, shift_list, repeats, create_using=None)

Return the cubic graph specified in LCF notation.
LCF notation (LCF=Lederberg-Coxeter-Fruchte) is a compressed notation used in the generation of various cubic
Hamiltonian graphs of high symmetry. See, for example, dodecahedral_graph, desargues_graph, heawood_graph
and pappus_graph below.
n (number of nodes)
The starting graph is the n-cycle with nodes 0,…,n-1. (The null graph is returned if n < 0.)
shift_list = [s1,s2,..,sk], a list of integer shifts mod n,
repeats
integer specifying the number of times that shifts in shift_list are successively applied to each v_current in
the n-cycle to generate an edge between v_current and v_current+shift mod n.
For v1 cycling through the n-cycle a total of k*repeats with shift cycling through shiftlist repeats times connect v1
with v1+shift mod n
The utility graph K3,3
>>> G = nx.LCF_graph(6, [3, -3], 3)
The Heawood graph
>>> G = nx.LCF_graph(14, [5, -5], 7)
See http://mathworld.wolfram.com/LCFNotation.html for a description and references.
5.5.2 bull_graph
bull_graph(create_using=None)
Returns the Bull Graph
The Bull Graph has 5 nodes and 5 edges. It is a planar undirected graph in the form of a triangle with two disjoint
pendant edges [1] The name comes from the triangle and pendant edges representing respectively the body and legs
of a bull.
Parameters
create_using
Returns
G
[networkx Graph] A bull graph with 5 nodes
5.5. Small 799

References
[1]
5.5.3 chvatal_graph
chvatal_graph(create_using=None)
Returns the Chvátal Graph
The Chvátal Graph is an undirected graph with 12 nodes and 24 edges [1]. It has 370 distinct (directed) Hamiltonian
cycles, giving a unique generalized LCF notation of order 4, two of order 6 , and 43 of order 1 [2].
Parameters
create_using
Returns
G
[networkx Graph] The Chvátal graph with 12 nodes and 24 edges
References
[1], [2]
5.5.4 cubical_graph
cubical_graph(create_using=None)
Returns the 3-regular Platonic Cubical Graph
The skeleton of the cube (the nodes and edges) form a graph, with 8 nodes, and 12 edges. It is a special case of
the hypercube graph. It is one of 5 Platonic graphs, each a skeleton of its Platonic solid [1]. Such graphs arise in
parallel processing in computers.
Parameters
create_using
Returns
G
[networkx Graph] A cubical graph with 8 nodes and 12 edges

References
[1]
5.5.5 desargues_graph
desargues_graph(create_using=None)
Returns the Desargues Graph
The Desargues Graph is a non-planar, distance-transitive cubic graph with 20 nodes and 30 edges [1]. It is a
symmetric graph. It can be represented in LCF notation as [5,-5,9,-9]^5 [2].
Parameters
create_using
Returns
G
[networkx Graph] Desargues Graph with 20 nodes and 30 edges
References
[1], [2]
5.5.6 diamond_graph
diamond_graph(create_using=None)
Returns the Diamond graph
The Diamond Graph is planar undirected graph with 4 nodes and 5 edges. It is also sometimes known as the double
triangle graph or kite graph [1].
Parameters
create_using
Returns
G
[networkx Graph] Diamond Graph with 4 nodes and 5 edges
5.5. Small 801

References
[1]
5.5.7 dodecahedral_graph
dodecahedral_graph(create_using=None)
Returns the Platonic Dodecahedral graph.
The dodecahedral graph has 20 nodes and 30 edges. The skeleton of the dodecahedron forms a graph. It is one of
5 Platonic graphs [1]. It can be described in LCF notation as: [10, 7, 4, -4, -7, 10, -4, 7, -7,
4]^2 [2].
Parameters
create_using
Returns
G
[networkx Graph] Dodecahedral Graph with 20 nodes and 30 edges
References
[1], [2]
5.5.8 frucht_graph
frucht_graph(create_using=None)
Returns the Frucht Graph.
The Frucht Graph is the smallest cubical graph whose automorphism group consists only of the identity element
[1]. It has 12 nodes and 18 edges and no nontrivial symmetries. It is planar and Hamiltonian [2].
Parameters
create_using
Returns
G
[networkx Graph] Frucht Graph with 12 nodes and 18 edges

References
[1], [2]
5.5.9 heawood_graph
heawood_graph(create_using=None)
Returns the Heawood Graph, a (3,6) cage.
The Heawood Graph is an undirected graph with 14 nodes and 21 edges, named after Percy John Heawood [1]. It
is cubic symmetric, nonplanar, Hamiltonian, and can be represented in LCF notation as [5,-5]^7 [2]. It is the
unique (3,6)-cage: the regular cubic graph of girth 6 with minimal number of vertices [3].
Parameters
create_using
Returns
G
[networkx Graph] Heawood Graph with 14 nodes and 21 edges
References
[1], [2], [3]
5.5.10 hoffman_singleton_graph
hoffman_singleton_graph()
Returns the Hoffman-Singleton Graph.
The Hoffman–Singleton graph is a symmetrical undirected graph with 50 nodes and 175 edges. All indices lie in
Z % 5: that is, the integers mod 5 [1]. It is the only regular graph of vertex degree 7, diameter 2, and girth 5. It
is the unique (7,5)-cage graph and Moore graph, and contains many copies of the Petersen graph [2].
Returns
G
[networkx Graph] Hoffman–Singleton Graph with 50 nodes and 175 edges
Notes
Constructed from pentagon and pentagram as follows: Take five pentagons Ph and five pentagrams Qi . Join vertex
j of Ph to vertex h·i + j of Qi [3].
5.5. Small 803

References
[1], [2], [3]
5.5.11 house_graph
house_graph(create_using=None)
Returns the House graph (square with triangle on top)
The house graph is a simple undirected graph with 5 nodes and 6 edges [1].
Parameters
create_using
Returns
G
[networkx Graph] House graph in the form of a square with a triangle on top
References
[1]
5.5.12 house_x_graph
house_x_graph(create_using=None)
Returns the House graph with a cross inside the house square.
The House X-graph is the House graph plus the two edges connecting diagonally opposite vertices of the square
base. It is also one of the two graphs obtained by removing two edges from the pentatope graph [1].
Parameters
create_using
Returns
G
[networkx Graph] House graph with diagonal vertices connected
References
[1]

5.5.13 icosahedral_graph
icosahedral_graph(create_using=None)
Returns the Platonic Icosahedral graph.
The icosahedral graph has 12 nodes and 30 edges. It is a Platonic graph whose nodes have the connectivity of the
icosahedron. It is undirected, regular and Hamiltonian [1].
Parameters
create_using
Returns
G
[networkx Graph] Icosahedral graph with 12 nodes and 30 edges.
References
[1]
5.5.14 krackhardt_kite_graph
krackhardt_kite_graph(create_using=None)
Returns the Krackhardt Kite Social Network.
A 10 actor social network introduced by David Krackhardt to illustrate different centrality measures [1].
Parameters
create_using
Returns
G
[networkx Graph] Krackhardt Kite graph with 10 nodes and 18 edges
Notes
The traditional labeling is: Andre=1, Beverley=2, Carol=3, Diane=4, Ed=5, Fernando=6, Garth=7, Heather=8,
Ike=9, Jane=10.
5.5. Small 805

References
[1]
5.5.15 moebius_kantor_graph
moebius_kantor_graph(create_using=None)
Returns the Moebius-Kantor graph.
The Möbius-Kantor graph is the cubic symmetric graph on 16 nodes. Its LCF notation is [5,-5]^8, and it is iso-
morphic to the generalized Petersen graph [1].
Parameters
create_using
Returns
G
[networkx Graph] Moebius-Kantor graph
References
[1]
5.5.16 octahedral_graph
octahedral_graph(create_using=None)
Returns the Platonic Octahedral graph.
The octahedral graph is the 6-node 12-edge Platonic graph having the connectivity of the octahedron [1]. If 6
couples go to a party, and each person shakes hands with every person except his or her partner, then this graph
describes the set of handshakes that take place; for this reason it is also called the cocktail party graph [2].
Parameters
create_using
Returns
G
[networkx Graph] Octahedral graph

References
[1], [2]
5.5.17 pappus_graph
pappus_graph()
Returns the Pappus graph.
The Pappus graph is a cubic symmetric distance-regular graph with 18 nodes and 27 edges. It is Hamiltonian and
can be represented in LCF notation as [5,7,-7,7,-7,-5]^3 [1].
Returns
G
[networkx Graph] Pappus graph
References
[1]
5.5.18 petersen_graph
petersen_graph(create_using=None)
Returns the Petersen graph.
The Peterson graph is a cubic, undirected graph with 10 nodes and 15 edges [1]. Julius Petersen constructed
the graph as the smallest counterexample against the claim that a connected bridgeless cubic graph has an edge
colouring with three colours [2].
Parameters
create_using
Returns
G
[networkx Graph] Petersen graph
References
[1], [2]
5.5. Small 807

5.5.19 sedgewick_maze_graph
sedgewick_maze_graph(create_using=None)
Return a small maze with a cycle.
This is the maze used in Sedgewick, 3rd Edition, Part 5, Graph Algorithms, Chapter 18, e.g. Figure 18.2 and
following [1]. Nodes are numbered 0,..,7
Parameters
create_using
Returns
G
[networkx Graph] Small maze with a cycle
References
[1]
5.5.20 tetrahedral_graph
tetrahedral_graph(create_using=None)
Returns the 3-regular Platonic Tetrahedral graph.
Tetrahedral graph has 4 nodes and 6 edges. It is a special case of the complete graph, K4, and wheel graph, W4.
It is one of the 5 platonic graphs [1].
Parameters
create_using
Returns
G
[networkx Graph] Tetrahedral Grpah
References
[1]

5.5.21 truncated_cube_graph
truncated_cube_graph(create_using=None)
Returns the skeleton of the truncated cube.
The truncated cube is an Archimedean solid with 14 regular faces (6 octagonal and 8 triangular), 36 edges and 24
nodes [1]. The truncated cube is created by truncating (cutting off) the tips of the cube one third of the way into
each edge [2].
Parameters
create_using
Returns
G
[networkx Graph] Skeleton of the truncated cube
References
[1], [2]
5.5.22 truncated_tetrahedron_graph
truncated_tetrahedron_graph(create_using=None)
Returns the skeleton of the truncated Platonic tetrahedron.
The truncated tetrahedron is an Archimedean solid with 4 regular hexagonal faces, 4 equilateral triangle faces, 12
nodes and 18 edges. It can be constructed by truncating all 4 vertices of a regular tetrahedron at one third of the
original edge length [1].
Parameters
create_using
Returns
G
[networkx Graph] Skeleton of the truncated tetrahedron
References
[1]
5.5. Small 809

5.5.23 tutte_graph
tutte_graph(create_using=None)
Returns the Tutte graph.
The Tutte graph is a cubic polyhedral, non-Hamiltonian graph. It has 46 nodes and 69 edges. It is a counterexample
to Tait’s conjecture that every 3-regular polyhedron has a Hamiltonian cycle. It can be realized geometrically from
a tetrahedron by multiply truncating three of its vertices [1].
Parameters
create_using
Returns
G
[networkx Graph] Tutte graph
References
[1]
5.6 Random Graphs
Generators for random graphs.

fast_gnp_random_graph(n, p[, seed, directed]) Returns a Gn,p random graph, also known as an Erdős-
Rényi graph or a binomial graph.
gnp_random_graph(n, p[, seed, directed]) Returns a Gn,p random graph, also known as an Erdős-
dense_gnm_random_graph(n, m[, seed]) Returns a Gn,m random graph.
gnm_random_graph(n, m[, seed, directed]) Returns a Gn,m random graph.
erdos_renyi_graph(n, p[, seed, directed]) Returns a Gn,p random graph, also known as an Erdős-
binomial_graph(n, p[, seed, directed]) Returns a Gn,p random graph, also known as an Erdős-
newman_watts_strogatz_graph(n, k, p[, seed]) Returns a Newman–Watts–Strogatz small-world graph.
watts_strogatz_graph(n, k, p[, seed]) Returns a Watts–Strogatz small-world graph.
connected_watts_strogatz_graph(n, k, p[, Returns a connected Watts–Strogatz small-world graph.
...])
random_regular_graph(d, n[, seed]) Returns a random d-regular graph on n nodes.
barabasi_albert_graph(n, m[, seed, ...]) Returns a random graph using Barabási–Albert preferen-
tial attachment
dual_barabasi_albert_graph(n, m1, m2, p[, Returns a random graph using dual Barabási–Albert pref-
...]) erential attachment
extended_barabasi_albert_graph(n, m, p, Returns an extended Barabási–Albert model graph.
q[, ...])
powerlaw_cluster_graph(n, m, p[, seed]) Holme and Kim algorithm for growing graphs with pow-
erlaw degree distribution and approximate average clus-
tering.
random_kernel_graph(n, kernel_integral[, ...]) Returns an random graph based on the specified kernel.
random_lobster(n, p1, p2[, seed]) Returns a random lobster graph.
random_shell_graph(constructor[, seed]) Returns a random shell graph for the constructor given.
random_powerlaw_tree(n[, gamma, seed, tries]) Returns a tree with a power law degree distribution.
random_powerlaw_tree_sequence(n[, gamma, Returns a degree sequence for a tree with a power law
...]) distribution.
random_kernel_graph(n, kernel_integral[, ...]) Returns an random graph based on the specified kernel.
5.6.1 fast_gnp_random_graph
fast_gnp_random_graph(n, p, seed=None, directed=False)

Returns a Gn,p random graph, also known as an Erdős-Rényi graph or a binomial graph.
Parameters
n
p
seed
Randomness.
directed
[bool, optional (default=False)] If True, this function returns a directed graph.
See also:
5.6. Random Graphs 811

gnp_random_graph
Notes
The Gn,p graph algorithm chooses each of the [n(n − 1)]/2 (undirected) or n(n − 1) (directed) possible edges
with probability p.
This algorithm [1] runs in O(n + m) time, where m is the expected number of edges, which equals pn(n − 1)/2.
This should be faster than gnp_random_graph() when p is small and the expected number of edges is small
(that is, the graph is sparse).
References
[1]
5.6.2 gnp_random_graph
gnp_random_graph(n, p, seed=None, directed=False)

The Gn,p model chooses each of the possible edges with probability p.
Parameters
n
p
seed
Randomness.
directed
See also:
fast_gnp_random_graph
Notes
This algorithm [2] runs in O(n2 ) time. For sparse graphs (that is, for small values of p),
fast_gnp_random_graph() is a faster algorithm.
binomial_graph() and erdos_renyi_graph() are aliases for gnp_random_graph().
>>> nx.binomial_graph is nx.gnp_random_graph

True
>>> nx.erdos_renyi_graph is nx.gnp_random_graph
True

References
[1], [2]
5.6.3 dense_gnm_random_graph
dense_gnm_random_graph(n, m, seed=None)
Returns a Gn,m random graph.
In the Gn,m model, a graph is chosen uniformly at random from the set of all graphs with n nodes and m edges.
This algorithm should be faster than gnm_random_graph() for dense graphs.
Parameters
n
m
[int] The number of edges.
seed
Randomness.
See also:
gnm_random_graph
Notes
Algorithm by Keith M. Briggs Mar 31, 2006. Inspired by Knuth’s Algorithm S (Selection sampling technique), in
section 3.4.2 of [1].
References
[1]
5.6.4 gnm_random_graph
gnm_random_graph(n, m, seed=None, directed=False)

Returns a Gn,m random graph.
In the Gn,m model, a graph is chosen uniformly at random from the set of all graphs with n nodes and m edges.
This algorithm should be faster than dense_gnm_random_graph() for sparse graphs.
Parameters
n
m
[int] The number of edges.

seed
Randomness.
directed
See also:
dense_gnm_random_graph
5.6.5 erdos_renyi_graph
erdos_renyi_graph(n, p, seed=None, directed=False)

Parameters
n
p
seed
Randomness.
directed
See also:
Notes

True
True

References
[1], [2]
5.6.6 binomial_graph
binomial_graph(n, p, seed=None, directed=False)

Parameters
n
p
seed
Randomness.
directed
See also:
Notes

True
True
References
[1], [2]

5.6.7 newman_watts_strogatz_graph
newman_watts_strogatz_graph(n, k, p, seed=None)
Returns a Newman–Watts–Strogatz small-world graph.
Parameters
n
k
[int] Each node is joined with its k nearest neighbors in a ring topology.
p
[float] The probability of adding a new edge for each edge.
seed
Randomness.
See also:
watts_strogatz_graph
Notes
First create a ring over n nodes [1]. Then each node in the ring is connected with its k nearest neighbors (or
k − 1 neighbors if k is odd). Then shortcuts are created by adding new edges as follows: for each edge (u, v) in
the underlying “n-ring with k nearest neighbors” with probability p add a new edge (u, w) with randomly-chosen
existing node w. In contrast with watts_strogatz_graph(), no edges are removed.
References
[1]
5.6.8 watts_strogatz_graph
watts_strogatz_graph(n, k, p, seed=None)
Returns a Watts–Strogatz small-world graph.
Parameters
n
[int] The number of nodes
k
p
[float] The probability of rewiring each edge
seed
Randomness.
See also:

newman_watts_strogatz_graph
connected_watts_strogatz_graph
Notes
First create a ring over n nodes [1]. Then each node in the ring is joined to its k nearest neighbors (or k − 1
neighbors if k is odd). Then shortcuts are created by replacing some edges as follows: for each edge (u, v) in the
underlying “n-ring with k nearest neighbors” with probability p replace it with a new edge (u, w) with uniformly
random choice of existing node w.
In contrast with newman_watts_strogatz_graph(), the random rewiring does not increase the number of
edges. The rewired graph is not guaranteed to be connected as in connected_watts_strogatz_graph().
References
[1]
5.6.9 connected_watts_strogatz_graph
connected_watts_strogatz_graph(n, k, p, tries=100, seed=None)

Returns a connected Watts–Strogatz small-world graph.
Attempts to generate a connected graph by repeated generation of Watts–Strogatz small-world graphs. An exception
is raised if the maximum number of tries is exceeded.
Parameters
n
k
p
[float] The probability of rewiring each edge
tries
[int] Number of attempts to generate a connected graph.
seed
Randomness.
See also:
newman_watts_strogatz_graph
watts_strogatz_graph

Notes
First create a ring over n nodes [1]. Then each node in the ring is joined to its k nearest neighbors (or k − 1
neighbors if k is odd). Then shortcuts are created by replacing some edges as follows: for each edge (u, v) in the
underlying “n-ring with k nearest neighbors” with probability p replace it with a new edge (u, w) with uniformly
random choice of existing node w. The entire process is repeated until a connected graph results.
References
[1]
5.6.10 random_regular_graph
random_regular_graph(d, n, seed=None)
Returns a random d-regular graph on n nodes.
The resulting graph has no self-loops or parallel edges.
Parameters
d
[int] The degree of each node.
n
[integer] The number of nodes. The value of n × d must be even.
seed
Randomness.
Raises
NetworkXError
If n × d is odd or d is greater than or equal to n.
Notes
The nodes are numbered from 0 to n − 1.

Kim and Vu’s paper [2] shows that this algorithm samples in an asymptotically uniform way from the space of
random graphs when d = O(n1/3−ϵ ).
References
[1], [2]

5.6.11 barabasi_albert_graph
barabasi_albert_graph(n, m, seed=None, initial_graph=None)

Returns a random graph using Barabási–Albert preferential attachment
A graph of n nodes is grown by attaching new nodes each with m edges that are preferentially attached to existing
nodes with high degree.
Parameters
n
[int] Number of nodes
m
[int] Number of edges to attach from a new node to existing nodes
seed
Randomness.
initial_graph
[Graph or None (default)] Initial network for Barabási–Albert algorithm. It should be a con-
nected graph for most use cases. A copy of initial_graph is used. If None, starts from
a star graph on (m+1) nodes.
Returns
G
[Graph]
Raises
NetworkXError
If m does not satisfy 1 <= m < n, or the initial graph number of nodes m0 does not satisfy
m <= m0 <= n.
References
[1]
5.6.12 dual_barabasi_albert_graph
dual_barabasi_albert_graph(n, m1, m2, p, seed=None, initial_graph=None)

Returns a random graph using dual Barabási–Albert preferential attachment
A graph of n nodes is grown by attaching new nodes each with either m1 edges (with probability p) or m2 edges
(with probability 1 − p) that are preferentially attached to existing nodes with high degree.
Parameters
n
m1
[int] Number of edges to link each new node to existing nodes with probability p
m2
[int] Number of edges to link each new node to existing nodes with probability 1 − p

p
[float] The probability of attaching m1 edges (as opposed to m2 edges)
seed
Randomness.
initial_graph
[Graph or None (default)] Initial network for Barabási–Albert algorithm. A copy of ini-
tial_graph is used. It should be connected for most use cases. If None, starts from an star
graph on max(m1, m2) + 1 nodes.
Returns
G
[Graph]
Raises
NetworkXError
If m1 and m2 do not satisfy 1 <= m1,m2 < n, or p does not satisfy 0 <= p <= 1, or
the initial graph number of nodes m0 does not satisfy m1, m2 <= m0 <= n.
References
[1]
5.6.13 extended_barabasi_albert_graph
extended_barabasi_albert_graph(n, m, p, q, seed=None)
Returns an extended Barabási–Albert model graph.
An extended Barabási–Albert model graph is a random graph constructed using preferential attachment. The
extended model allows new edges, rewired edges or new nodes. Based on the probabilities p and q with p + q < 1,
the growing behavior of the graph is determined as:
1) With p probability, m new edges are added to the graph, starting from randomly chosen existing nodes and
attached preferentially at the other end.
2) With q probability, m existing edges are rewired by randomly choosing an edge and rewiring one end to a
preferentially chosen node.
3) With (1 − p − q) probability, m new nodes are added to the graph with edges attached preferentially.
When p = q = 0, the model behaves just like the Barabási–Alber model.
Parameters
n
m
[int] Number of edges with which a new node attaches to existing nodes
p
[float] Probability value for adding an edge between existing nodes. p + q < 1
q
[float] Probability value of rewiring of existing edges. p + q < 1

seed
Randomness.
Returns
G
[Graph]
Raises
NetworkXError
If m does not satisfy 1 <= m < n or 1 >= p + q
References
[1]
5.6.14 powerlaw_cluster_graph
powerlaw_cluster_graph(n, m, p, seed=None)
Holme and Kim algorithm for growing graphs with powerlaw degree distribution and approximate average clus-
tering.
Parameters
n
[int] the number of nodes
m
[int] the number of random edges to add for each new node
p
[float,] Probability of adding a triangle after adding a random edge
seed
Randomness.
Raises
NetworkXError
If m does not satisfy 1 <= m <= n or p does not satisfy 0 <= p <= 1.
Notes
The average clustering has a hard time getting above a certain cutoff that depends on m. This cutoff is often quite
low. The transitivity (fraction of triangles to possible triangles) seems to decrease with network size.
It is essentially the Barabási–Albert (BA) growth model with an extra step that each random edge is followed by a
chance of making an edge to one of its neighbors too (and thus a triangle).
This algorithm improves on BA in the sense that it enables a higher average clustering to be attained if desired.
It seems possible to have a disconnected graph with this algorithm since the initial m nodes may not be all linked
to a new node on the first iteration like the BA model.

References
[1]
5.6.15 random_kernel_graph
random_kernel_graph(n, kernel_integral, kernel_root=None, seed=None)

Returns an random graph based on the specified kernel.
The algorithm chooses each of the [n(n − 1)]/2 possible edges with probability specified by a kernel κ(x, y) [1].
The kernel κ(x, y) must be a symmetric (in x, y), non-negative, bounded function.
Parameters
n
kernel_integral
[function] Function that returns the definite integral of the kernel κ(x, y), F (y, a, b) :=
∫b
a
κ(x, y)dx
kernel_root: function (optional)
Function that returns the root b of the equation F (y, a, b) = r. If None, the root is found
using scipy.optimize.brentq() (this requires SciPy).
seed
Randomness.
See also:
gnp_random_graph
expected_degree_graph
Notes
The kernel is specified through its definite integral which must be provided as one of the arguments. If the integral
and root of the kernel integral can be found in O(1) time then this algorithm runs in time O(n + m) where m is
the expected number of edges [2].
The nodes are set to integers from 0 to n − 1.
References
[1], [2]

Examples
Generate an Erdős–Rényi random graph G(n, c/n), with kernel κ(x, y) = c where c is the mean expected degree.
>>> def integral(u, w, z):

... return c * (z - w)
>>> def root(u, w, r):
... return r / c + w
>>> c = 1
>>> graph = nx.random_kernel_graph(1000, integral, root)
5.6.16 random_lobster
random_lobster(n, p1, p2, seed=None)

Returns a random lobster graph.
A lobster is a tree that reduces to a caterpillar when pruning all leaf nodes. A caterpillar is a tree that reduces to a
path graph when pruning all leaf nodes; setting p2 to zero produces a caterpillar.
This implementation iterates on the probabilities p1 and p2 to add edges at levels 1 and 2, respectively. Graphs
are therefore constructed iteratively with uniform randomness at each level rather than being selected uniformly at
random from the set of all possible lobsters.
Parameters
n
[int] The expected number of nodes in the backbone
p1
[float] Probability of adding an edge to the backbone
p2
[float] Probability of adding an edge one level beyond backbone
seed
Randomness.
Raises
NetworkXError
If p1 or p2 parameters are >= 1 because the while loops would never finish.
5.6.17 random_shell_graph
random_shell_graph(constructor, seed=None)
Returns a random shell graph for the constructor given.
Parameters
constructor
[list of three-tuples] Represents the parameters for a shell, starting at the center shell. Each
element of the list must be of the form (n, m, d), where n is the number of nodes in the
shell, m is the number of edges in the shell, and d is the ratio of inter-shell (next) edges to
intra-shell edges. If d is zero, there will be no intra-shell edges, and if d is one there will be
all possible intra-shell edges.

seed
Randomness.
Examples
>>> constructor = [(10, 20, 0.8), (20, 40, 0.8)]

>>> G = nx.random_shell_graph(constructor)
5.6.18 random_powerlaw_tree
random_powerlaw_tree(n, gamma=3, seed=None, tries=100)

Returns a tree with a power law degree distribution.
Parameters
n
gamma
[float] Exponent of the power law.
seed
Randomness.
tries
[int] Number of attempts to adjust the sequence to make it a tree.
Raises
NetworkXError
If no valid sequence is found within the maximum number of attempts.
Notes
A trial power law degree sequence is chosen and then elements are swapped with new elements from a powerlaw
distribution until the sequence makes a tree (by checking, for example, that the number of edges is one smaller
than the number of nodes).
5.6.19 random_powerlaw_tree_sequence
random_powerlaw_tree_sequence(n, gamma=3, seed=None, tries=100)

Returns a degree sequence for a tree with a power law distribution.
Parameters
n
[int,] The number of nodes.
gamma
[float] Exponent of the power law.

seed
Randomness.
tries
[int] Number of attempts to adjust the sequence to make it a tree.
Raises
NetworkXError
If no valid sequence is found within the maximum number of attempts.
Notes
A trial power law degree sequence is chosen and then elements are swapped with new elements from a power law
distribution until the sequence makes a tree (by checking, for example, that the number of edges is one smaller
than the number of nodes).
5.7 Duplication Divergence
Functions for generating graphs based on the “duplication” method.

These graph generators start with a small initial graph then duplicate nodes and (partially) duplicate their edges. These
functions are generally inspired by biological networks.
duplication_divergence_graph(n, p[, seed]) Returns an undirected graph using the duplication-

divergence model.
partial_duplication_graph(N, n, p, q[, seed]) Returns a random graph using the partial duplication
model.
5.7.1 duplication_divergence_graph
duplication_divergence_graph(n, p, seed=None)
Returns an undirected graph using the duplication-divergence model.
A graph of n nodes is created by duplicating the initial nodes and retaining edges incident to the original nodes
with a retention probability p.
Parameters
n
[int] The desired number of nodes in the graph.
p
[float] The probability for retaining the edge of the replicated node.
seed
Randomness.
Returns
G
[Graph]
Raises
5.7. Duplication Divergence 825

NetworkXError
If p is not a valid probability. If n is less than 2.
Notes
This algorithm appears in [1].

This implementation disallows the possibility of generating disconnected graphs.
References
[1]
5.7.2 partial_duplication_graph
partial_duplication_graph(N, n, p, q, seed=None)
Returns a random graph using the partial duplication model.
Parameters
N
[int] The total number of nodes in the final graph.
n
[int] The number of nodes in the initial clique.
p
[float] The probability of joining each neighbor of a node to the duplicate node. Must be a
number in the between zero and one, inclusive.
q
[float] The probability of joining the source node to the duplicate node. Must be a number in
the between zero and one, inclusive.
seed
Randomness.
Notes
A graph of nodes is grown by creating a fully connected graph of size n. The following procedure is then repeated
until a total of N nodes have been reached.
1. A random node, u, is picked and a new node, v, is created.
2. For each neighbor of u an edge from the neighbor to v is created with probability p.
3. An edge from u to v is created with probability q.
This algorithm appears in [1].
This implementation allows the possibility of generating disconnected graphs.

References
[1]
5.8 Degree Sequence
Generate graphs with a given degree sequence or expected degree sequence.
configuration_model(deg_sequence[, ...]) Returns a random graph with the given degree sequence.
directed_configuration_model(...[, ...]) Returns a directed_random graph with the given degree
sequences.
expected_degree_graph(w[, seed, selfloops]) Returns a random graph with given expected degrees.
havel_hakimi_graph(deg_sequence[, cre- Returns a simple graph with given degree sequence con-
ate_using]) structed using the Havel-Hakimi algorithm.
directed_havel_hakimi_graph(in_deg_sequence,Returns a directed graph with the given degree sequences.
...)
degree_sequence_tree(deg_sequence[, ...]) Make a tree for the given degree sequence.
random_degree_sequence_graph(sequence[, Returns a simple random graph with the given degree se-
...]) quence.
5.8.1 configuration_model
configuration_model(deg_sequence, create_using=None, seed=None)

Returns a random graph with the given degree sequence.
The configuration model generates a random pseudograph (graph with parallel edges and self loops) by randomly
assigning edges to match the given degree sequence.
Parameters
deg_sequence
[list of nonnegative integers] Each list entry corresponds to the degree of a node.
create_using
seed
Randomness.
Returns
G
[MultiGraph] A graph with the specified degree sequence. Nodes are labeled starting at 0 with
an index corresponding to the position in deg_sequence.
Raises
NetworkXError
If the degree sequence does not have an even sum.
See also:
is_graphical
5.8. Degree Sequence 827

Notes
As described by Newman [1].

A non-graphical degree sequence (not realizable by some simple graph) is allowed since this function returns graphs
with self loops and parallel edges. An exception is raised if the degree sequence does not have an even sum.
This configuration model construction process can lead to duplicate edges and loops. You can remove the self-
loops and parallel edges (see below) which will likely result in a graph that doesn’t have the exact degree sequence
specified.
The density of self-loops and parallel edges tends to decrease as the number of nodes increases. However, typically
the number of self-loops will approach a Poisson distribution with a nonzero mean, and similarly for the number
of parallel edges. Consider a node with k stubs. The probability of being joined to another stub of the same node
is basically (k - 1) / N, where k is the degree and N is the number of nodes. So the probability of a self-loop scales
like c / N for some constant c. As N grows, this means we expect c self-loops. Similarly for parallel edges.
References
[1]
Examples
You can create a degree sequence following a particular distribution by using the one of the distribution functions
in random_sequence (or one of your own). For example, to create an undirected multigraph on one hundred
nodes with degree sequence chosen from the power law distribution:
>>> sequence = nx.random_powerlaw_tree_sequence(100, tries=5000)

>>> G = nx.configuration_model(sequence)
>>> len(G)
100
>>> actual_degrees = [d for v, d in G.degree()]
>>> actual_degrees == sequence
True
The returned graph is a multigraph, which may have parallel edges. To remove any parallel edges from the returned
graph:
>>> G = nx.Graph(G)
Similarly, to remove self-loops:
>>> G.remove_edges_from(nx.selfloop_edges(G))
5.8.2 directed_configuration_model
directed_configuration_model(in_degree_sequence, out_degree_sequence, create_using=None,

seed=None)
Returns a directed_random graph with the given degree sequences.
The configuration model generates a random directed pseudograph (graph with parallel edges and self loops) by
randomly assigning edges to match the given degree sequences.
Parameters

in_degree_sequence
[list of nonnegative integers] Each list entry corresponds to the in-degree of a node.
out_degree_sequence
[list of nonnegative integers] Each list entry corresponds to the out-degree of a node.
create_using
[NetworkX graph constructor, optional (default MultiDiGraph)] Graph type to create. If graph
seed
Randomness.
Returns
G
[MultiDiGraph] A graph with the specified degree sequences. Nodes are labeled starting at 0
with an index corresponding to the position in deg_sequence.
Raises
NetworkXError
If the degree sequences do not have the same sum.
See also:
configuration_model
Notes
Algorithm as described by Newman [1].

with self loops and parallel edges. An exception is raised if the degree sequences does not have the same sum.
This configuration model construction process can lead to duplicate edges and loops. You can remove the self-
specified. This “finite-size effect” decreases as the size of the graph increases.
References
[1]
Examples
One can modify the in- and out-degree sequences from an existing directed graph in order to create a new directed
graph. For example, here we modify the directed path graph:
>>> D = nx.DiGraph([(0, 1), (1, 2), (2, 3)])

>>> din = list(d for n, d in D.in_degree())
>>> dout = list(d for n, d in D.out_degree())
>>> din.append(1)
>>> dout[0] = 2
>>> # We now expect an edge from node 0 to a new node, node 3.
... D = nx.directed_configuration_model(din, dout)

The returned graph is a directed multigraph, which may have parallel edges. To remove any parallel edges from
the returned graph:
>>> D = nx.DiGraph(D)
Similarly, to remove self-loops:

>>> D.remove_edges_from(nx.selfloop_edges(D))
5.8.3 expected_degree_graph
expected_degree_graph(w, seed=None, selfloops=True)

Returns a random graph with given expected degrees.
Given a sequence of expected degrees W = (w0 , w1 , . . . , wn−1 ) of length n this algorithm assigns an edge between
node u and node v with probability
wu wv
puv = ∑ .
k wk
Parameters
w
[list] The list of expected degrees.
selfloops: bool (default=True)
Set to False to remove the possibility of self-loop edges.
seed
Randomness.
Returns
Graph
Notes
The nodes have integer labels corresponding to index of expected degrees input sequence.
The complexity of this algorithm is O(n + m) where n is the number of nodes and m is the expected number of
edges.
The model in [1] includes the possibility of self-loop edges. Set selfloops=False to produce a graph without self
loops.
For finite graphs this model doesn’t produce exactly the given expected degree sequence. Instead the expected
degrees are as follows.
For the case without self loops (selfloops=False),
∑ ( )
wu
E[deg(u)] = puv = wu 1− ∑ .
v̸=u k wk
NetworkX uses the standard convention that a self-loop edge counts 2 in the degree of a node, so with self loops
(selfloops=True),
∑ ( )
wu
E[deg(u)] = puv + 2puu = wu 1 + ∑ .
v̸=u k wk

References
[1], [2]
Examples
>>> z = [10 for i in range(100)]

>>> G = nx.expected_degree_graph(z)
5.8.4 havel_hakimi_graph
havel_hakimi_graph(deg_sequence, create_using=None)
Returns a simple graph with given degree sequence constructed using the Havel-Hakimi algorithm.
Parameters
deg_sequence: list of integers
Each integer corresponds to the degree of a node (need not be sorted).
create_using
instance, then cleared before populated. Directed graphs are not allowed.
Raises
NetworkXException
For a non-graphical degree sequence (i.e. one not realizable by some simple graph).
Notes
The Havel-Hakimi algorithm constructs a simple graph by successively connecting the node of highest degree to
other nodes of highest degree, resorting remaining nodes by degree, and repeating the process. The resulting
graph has a high degree-associativity. Nodes are labeled 1,.., len(deg_sequence), corresponding to their position in
deg_sequence.
The basic algorithm is from Hakimi [1] and was generalized by Kleitman and Wang [2].
References
[1], [2]
5.8.5 directed_havel_hakimi_graph
directed_havel_hakimi_graph(in_deg_sequence, out_deg_sequence, create_using=None)

Returns a directed graph with the given degree sequences.
Parameters
in_deg_sequence
[list of integers] Each list entry corresponds to the in-degree of a node.
out_deg_sequence
[list of integers] Each list entry corresponds to the out-degree of a node.

create_using
[NetworkX graph constructor, optional (default DiGraph)] Graph type to create. If graph
Returns
G
[DiGraph] A graph with the specified degree sequences. Nodes are labeled starting at 0 with
an index corresponding to the position in deg_sequence
Raises
NetworkXError
If the degree sequences are not digraphical.
See also:
configuration_model
Notes
Algorithm as described by Kleitman and Wang [1].
References
[1]
5.8.6 degree_sequence_tree
degree_sequence_tree(deg_sequence, create_using=None)
Make a tree for the given degree sequence.
A tree has #nodes-#edges=1 so the degree sequence must have len(deg_sequence)-sum(deg_sequence)/2=1
5.8.7 random_degree_sequence_graph
random_degree_sequence_graph(sequence, seed=None, tries=10)

Returns a simple random graph with the given degree sequence.
If the maximum degree dm in the sequence is O(m1/4 ) then the algorithm produces almost uniform random graphs
in O(mdm ) time where m is the number of edges.
Parameters
sequence
[list of integers] Sequence of degrees
seed
Randomness.
tries
[int, optional] Maximum number of tries to create a graph
Returns

G
[Graph] A graph with the specified degree sequence. Nodes are labeled starting at 0 with an
index corresponding to the position in the sequence.
Raises
NetworkXUnfeasible
If the degree sequence is not graphical.
NetworkXError
If a graph is not produced in specified number of tries
See also:
is_graphical, configuration_model
Notes
The generator algorithm [1] is not guaranteed to produce a graph.
References
[1]
Examples
>>> sequence = [1, 2, 2, 3]

>>> G = nx.random_degree_sequence_graph(sequence, seed=42)
>>> sorted(d for n, d in G.degree())
[1, 2, 2, 3]
5.9 Random Clustered
Generate graphs with given degree and triangle sequence.
random_clustered_graph(joint_degree_sequence) Generate a random graph with the given joint independent

edge degree and triangle degree sequence.
5.9.1 random_clustered_graph
random_clustered_graph(joint_degree_sequence, create_using=None, seed=None)

Generate a random graph with the given joint independent edge degree and triangle degree sequence.
This uses a configuration model-like approach to generate a random graph (with parallel edges and self-loops) by
randomly assigning edges to match the given joint degree sequence.
The joint degree sequence is a list of pairs of integers of the form [(d1,i , d1,t ), . . . , (dn,i , dn,t )]. According to this
list, vertex u is a member of du,t triangles and has du,i other edges. The number du,t is the triangle degree of u
and the number du,i is the independent edge degree.
Parameters
5.9. Random Clustered 833

joint_degree_sequence
[list of integer pairs] Each list entry corresponds to the independent edge degree and triangle
degree of a node.
create_using
seed
Randomness.
Returns
G
[MultiGraph] A graph with the specified degree sequence. Nodes are labeled starting at 0 with
an index corresponding to the position in deg_sequence.
Raises
NetworkXError
If the independent edge degree sequence sum is not even or the triangle degree sequence sum
is not divisible by 3.
Notes
As described by Miller [1] (see also Newman [2] for an equivalent description).
with self loops and parallel edges. An exception is raised if the independent degree sequence does not have an even
sum or the triangle degree sequence sum is not divisible by 3.
This configuration model-like construction process can lead to duplicate edges and loops. You can remove the self-
specified. This “finite-size effect” decreases as the size of the graph increases.
References
[1], [2]
Examples
>>> deg = [(1, 0), (1, 0), (1, 0), (2, 0), (1, 0), (2, 1), (0, 1), (0, 1)]
>>> G = nx.random_clustered_graph(deg)
To remove parallel edges:
>>> G = nx.Graph(G)
To remove self loops:
>>> G.remove_edges_from(nx.selfloop_edges(G))

5.10 Directed
Generators for some directed graphs, including growing network (GN) graphs and scale-free graphs.
gn_graph(n[, kernel, create_using, seed]) Returns the growing network (GN) digraph with n nodes.
gnr_graph(n, p[, create_using, seed]) Returns the growing network with redirection (GNR) di-
graph with n nodes and redirection probability p.
gnc_graph(n[, create_using, seed]) Returns the growing network with copying (GNC) di-
graph with n nodes.
random_k_out_graph(n, k, alpha[, ...]) Returns a random k-out graph with preferential attach-
ment.
scale_free_graph(n[, alpha, beta, gamma, ...]) Returns a scale-free directed graph.
5.10.1 gn_graph
gn_graph(n, kernel=None, create_using=None, seed=None)

Returns the growing network (GN) digraph with n nodes.
The GN graph is built by adding nodes one at a time with a link to one previously added node. The target node for
the link is chosen with probability based on degree. The default attachment kernel is a linear function of the degree
of a node.
The graph is always a (directed) tree.
Parameters
n
[int] The number of nodes for the generated graph.
kernel
[function] The attachment kernel.
create_using
seed
Randomness.
References
[1]
5.10. Directed 835

Examples
To create the undirected GN graph, use the to_directed() method:
>>> D = nx.gn_graph(10) # the GN graph

>>> G = D.to_undirected() # the undirected version
To specify an attachment kernel, use the kernel keyword argument:
>>> D = nx.gn_graph(10, kernel=lambda x: x ** 1.5) # A_k = k^1.5
5.10.2 gnr_graph
gnr_graph(n, p, create_using=None, seed=None)

Returns the growing network with redirection (GNR) digraph with n nodes and redirection probability p.
The GNR graph is built by adding nodes one at a time with a link to one previously added node. The previous
target node is chosen uniformly at random. With probabiliy p the link is instead “redirected” to the successor node
of the target.
The graph is always a (directed) tree.
Parameters
n
p
[float] The redirection probability.
create_using
seed
Randomness.
References
[1]
Examples
To create the undirected GNR graph, use the to_directed() method:
>>> D = nx.gnr_graph(10, 0.5) # the GNR graph

>>> G = D.to_undirected() # the undirected version

5.10.3 gnc_graph
gnc_graph(n, create_using=None, seed=None)

Returns the growing network with copying (GNC) digraph with n nodes.
The GNC graph is built by adding nodes one at a time with a link to one previously added node (chosen uniformly
at random) and to all of that node’s successors.
Parameters
n
create_using
seed
Randomness.
References
[1]
5.10.4 random_k_out_graph
random_k_out_graph(n, k, alpha, self_loops=True, seed=None)

Returns a random k-out graph with preferential attachment.
A random k-out graph with preferential attachment is a multidigraph generated by the following algorithm.
1. Begin with an empty digraph, and initially set each node to have weight alpha.
2. Choose a node u with out-degree less than k uniformly at random.
3. Choose a node v from with probability proportional to its weight.
4. Add a directed edge from u to v, and increase the weight of v by one.
5. If each node has out-degree k, halt, otherwise repeat from step 2.
For more information on this model of random graph, see [1].
Parameters
n
[int] The number of nodes in the returned graph.
k
[int] The out-degree of each node in the returned graph.
alpha
[float] A positive float representing the initial weight of each vertex. A higher number means
that in step 3 above, nodes will be chosen more like a true uniformly random sample, and a
lower number means that nodes are more likely to be chosen as their in-degree increases. If
this parameter is not positive, a ValueError is raised.
self_loops
[bool] If True, self-loops are allowed when generating the graph.
5.10. Directed 837

seed
Randomness.
Returns
MultiDiGraph
A k-out-regular multidigraph generated according to the above algorithm.
Raises
ValueError
If alpha is not positive.
Notes
The returned multidigraph may not be strongly connected, or even weakly connected.
References
[1]: Peterson, Nicholas R., and Boris Pittel.

“Distance between two random k-out digraphs, with and without preferential attachment.” arXiv preprint
arXiv:1311.5961 (2013). <https://arxiv.org/abs/1311.5961>
5.10.5 scale_free_graph
scale_free_graph(n, alpha=0.41, beta=0.54, gamma=0.05, delta_in=0.2, delta_out=0, create_using=None,

seed=None, initial_graph=None)
Returns a scale-free directed graph.
Parameters
n
[integer] Number of nodes in graph
alpha
[float] Probability for adding a new node connected to an existing node chosen randomly ac-
cording to the in-degree distribution.
beta
[float] Probability for adding an edge between two existing nodes. One existing node is chosen
randomly according the in-degree distribution and the other chosen randomly according to the
out-degree distribution.
gamma
[float] Probability for adding a new node connected to an existing node chosen randomly ac-
cording to the out-degree distribution.
delta_in
[float] Bias for choosing nodes from in-degree distribution.
delta_out
[float] Bias for choosing nodes from out-degree distribution.
create_using
[NetworkX graph constructor, optional] The default is a MultiDiGraph 3-cycle. If a graph

instance, use it without clearing first. If a graph constructor, call it to construct an empty
graph.
Deprecated since version 3.0: create_using is deprecated, use initial_graph instead.
seed
Randomness.
initial_graph
[MultiDiGraph instance, optional] Build the scale-free graph starting from this initial Multi-
DiGraph, if provided.
Returns
MultiDiGraph
Notes
The sum of alpha, beta, and gamma must be 1.
References
[1]
Examples
Create a scale-free graph on one hundred nodes:
>>> G = nx.scale_free_graph(100)
5.11 Geometric
Generators for geometric graphs.
geometric_edges(G, radius[, p]) Returns edge list of node pairs within radius of each
other.
geographical_threshold_graph(n, theta[, ...]) Returns a geographical threshold graph.
navigable_small_world_graph(n[, p, q, r, ...]) Returns a navigable small-world graph.
random_geometric_graph(n, radius[, dim, ...]) Returns a random geometric graph in the unit cube of di-
mensions dim.
soft_random_geometric_graph(n, radius[, ...]) Returns a soft random geometric graph in the unit cube.
thresholded_random_geometric_graph(n, Returns a thresholded random geometric graph in the unit
...[, ...]) cube.
waxman_graph(n[, beta, alpha, L, domain, ...]) Returns a Waxman random graph.
5.11. Geometric 839

5.11.1 geometric_edges
geometric_edges(G, radius, p=2)

Returns edge list of node pairs within radius of each other.
Parameters
G
[networkx graph] The graph from which to generate the edge list. The nodes in G should have
an attribute pos corresponding to the node position, which is used to compute the distance to
other nodes.
radius
[scalar] The distance threshold. Edges are included in the edge list if the distance between the
two nodes is less than radius.
p
[scalar, default=2] The Minkowski distance metric used to compute distances. The default
value is 2, i.e. Euclidean distance.
Returns
edges
[list] List of edges whose distances are less than radius
Notes
Radius uses Minkowski distance metric p. If scipy is available, scipy.spatial.cKDTree is used to speed
computation.
Examples
Create a graph with nodes that have a “pos” attribute representing 2D coordinates.
>>> G = nx.Graph()
>>> G.add_nodes_from([
... (0, {"pos": (0, 0)}),
... (1, {"pos": (3, 0)}),
... (2, {"pos": (8, 0)}),
... ])
>>> nx.geometric_edges(G, radius=1)
[]
[(0, 1)]
[(0, 1), (1, 2)]
[(0, 1), (0, 2), (1, 2)]

5.11.2 geographical_threshold_graph
geographical_threshold_graph(n, theta, dim=2, pos=None, weight=None, metric=None, p_dist=None,

seed=None)
Returns a geographical threshold graph.
The geographical threshold graph model places n nodes uniformly at random in a rectangular domain. Each node
u is assigned a weight wu . Two nodes u and v are joined by an edge if
(wu + wv )pdist (r) ≥ θ
where r is the distance between u and v, p_dist is any function of r, and θ as the threshold parameter. p_dist
is used to give weight to the distance between nodes when deciding whether or not they should be connected. The
larger p_dist is, the more prone nodes separated by r are to be connected, and vice versa.
Parameters
n
[int or iterable] Number of nodes or iterable of nodes
theta: float
Threshold value
dim
[int, optional] Dimension of graph
pos
[dict] Node positions as a dictionary of tuples keyed by node.
weight
[dict] Node weights as a dictionary of numbers keyed by node.
metric
[function] A metric on vectors of numbers (represented as lists or tuples). This must be a
function that accepts two lists (or tuples) as input and yields a number as output. The function
must also satisfy the four requirements of a metric. Specifically, if d is the function and x, y,
and z are vectors in the graph, then d must satisfy
1. d(x, y) ≥ 0,
2. d(x, y) = 0 if and only if x = y,
3. d(x, y) = d(y, x),
4. d(x, z) ≤ d(x, y) + d(y, z).
If this argument is not specified, the Euclidean distance metric is used.
p_dist
[function, optional] Any function used to give weight to the distance between nodes when
deciding whether or not they should be connected. p_dist was originally conceived as a
probability density function giving the probability of connecting two nodes that are of metric
distance r apart. The implementation here allows for more arbitrary definitions of p_dist
that do not need to correspond to valid probability density functions. The scipy.stats
package has many probability density functions implemented and tools for custom probability
density definitions, and passing the .pdf method of scipy.stats distributions can be used here.
If p_dist=None (the default), the exponential function r−2 is used.
seed
Randomness.
5.11. Geometric 841

Returns
Graph
A random geographic threshold graph, undirected and without self-loops.
Each node has a node attribute pos that stores the position of that node in Euclidean space
as provided by the pos keyword argument or, if pos was not provided, as generated by this
function. Similarly, each node has a node attribute weight that stores the weight of that node
as provided or as generated.
Notes
If weights are not specified they are assigned to nodes by drawing randomly from the exponential distribution with
rate parameter λ = 1. To specify weights from a different distribution, use the weight keyword argument:
>>> import random

>>> n = 20
>>> w = {i: random.expovariate(5.0) for i in range(n)}
>>> G = nx.geographical_threshold_graph(20, 50, weight=w)
If node positions are not specified they are randomly assigned from the uniform distribution.
References
[1], [2]
Examples
Specify an alternate distance metric using the metric keyword argument. For example, to use the taxicab metric
instead of the default Euclidean metric:
>>> dist = lambda x, y: sum(abs(a - b) for a, b in zip(x, y))

>>> G = nx.geographical_threshold_graph(10, 0.1, metric=dist)
5.11.3 navigable_small_world_graph
navigable_small_world_graph(n, p=1, q=1, r=2, dim=2, seed=None)

Returns a navigable small-world graph.
A navigable small-world graph is a directed grid with additional long-range connections that are chosen randomly.
[…] we begin with a set of nodes […] that are identified with the set of lattice points in an n × n square,
{(i, j) : i ∈ {1, 2, . . . , n}, j ∈ {1, 2, . . . , n}}, and we define the lattice distance between two nodes
(i, j) and (k, l) to be the number of “lattice steps” separating them: d((i, j), (k, l)) = |k − i| + |l − j|.
For a universal constant p >= 1, the node u has a directed edge to every other node within lattice
distance p—these are its local contacts. For universal constants q >= 0 and r >= 0 we also construct
directed edges from u to q other nodes (the long-range contacts) using independent random trials; the
i`thdirectededgef rom : math : ù has endpoint v with probability proportional to [d(u, v)]−r .
---[1]
Parameters

n
[int] The length of one side of the lattice; the number of nodes in the graph is therefore n2 .
p
[int] The diameter of short range connections. Each node is joined with every other node
within this lattice distance.
q
[int] The number of long-range connections for each node.
r
[float] Exponent for decaying probability of connections. The probability of connecting to a
node at lattice distance d is 1/dr .
dim
[int] Dimension of grid
seed
Randomness.
References
[1]
5.11.4 random_geometric_graph
random_geometric_graph(n, radius, dim=2, pos=None, p=2, seed=None)

Returns a random geometric graph in the unit cube of dimensions dim.
The random geometric graph model places n nodes uniformly at random in the unit cube. Two nodes are joined
by an edge if the distance between the nodes is at most radius.
Edges are determined using a KDTree when SciPy is available. This reduces the time complexity from O(n2 ) to
O(n).
Parameters
n
radius: float
Distance threshold value
dim
pos
[dict, optional] A dictionary keyed by node with node positions as values.
p
[float, optional] Which Minkowski distance metric to use. p has to meet the condition 1 <=
p <= infinity.
If this argument is not specified, the L2 metric (the Euclidean distance metric), p = 2 is used.
This should not be confused with the p of an Erdős-Rényi random graph, which represents
probability.
5.11. Geometric 843

seed
Randomness.
Returns
Graph
A random geometric graph, undirected and without self-loops. Each node has a node attribute
'pos' that stores the position of that node in Euclidean space as provided by the pos keyword
argument or, if pos was not provided, as generated by this function.
Notes
This uses a k-d tree to build the graph.

The pos keyword argument can be used to specify node positions so you can create an arbitrary distribution and
domain for positions.
For example, to use a 2D Gaussian distribution of node positions with mean (0, 0) and standard deviation 2:
>>> import random

>>> n = 20
>>> pos = {i: (random.gauss(0, 2), random.gauss(0, 2)) for i in range(n)}
>>> G = nx.random_geometric_graph(n, 0.2, pos=pos)
References
[1]
Examples
Create a random geometric graph on twenty nodes where nodes are joined by an edge if their distance is at most
0.1:
>>> G = nx.random_geometric_graph(20, 0.1)
5.11.5 soft_random_geometric_graph
soft_random_geometric_graph(n, radius, dim=2, pos=None, p=2, p_dist=None, seed=None)

Returns a soft random geometric graph in the unit cube.
The soft random geometric graph [1] model places n nodes uniformly at random in the unit cube in dimension
dim. Two nodes of distance, dist, computed by the p-Minkowski distance metric are joined by an edge with
probability p_dist if the computed distance metric value of the nodes is at most radius, otherwise they are
not joined.
Edges within radius of each other are determined using a KDTree when SciPy is available. This reduces the
time complexity from O(n2 ) to O(n).
Parameters
n

radius: float
dim
pos
p
[float, optional] Which Minkowski distance metric to use. p has to meet the condition 1 <=
p <= infinity.
probability.
p_dist
[function, optional] A probability density function computing the probability of connecting
two nodes that are of distance, dist, computed by the Minkowski distance metric. The proba-
bility density function, p_dist, must be any function that takes the metric value as input and
outputs a single probability value between 0-1. The scipy.stats package has many probability
distribution functions implemented and tools for custom probability distribution definitions
[2], and passing the .pdf method of scipy.stats distributions can be used here. If the probabil-
ity function, p_dist, is not supplied, the default function is an exponential distribution with
rate parameter λ = 1.
seed
Randomness.
Returns
Graph
A soft random geometric graph, undirected and without self-loops. Each node has a node
attribute 'pos' that stores the position of that node in Euclidean space as provided by the
pos keyword argument or, if pos was not provided, as generated by this function.
Notes

For example, to use a 2D Gaussian distribution of node positions with mean (0, 0) and standard deviation 2
The scipy.stats package can be used to define the probability distribution with the .pdf method used as p_dist.
>>> import random

>>> import math
>>> n = 100
>>> p_dist = lambda dist: math.exp(-dist)
>>> G = nx.soft_random_geometric_graph(n, 0.2, pos=pos, p_dist=p_dist)
5.11. Geometric 845

References
[1], [2]
Examples
Default Graph:
G = nx.soft_random_geometric_graph(50, 0.2)
Custom Graph:
Create a soft random geometric graph on 100 uniformly distributed nodes where nodes are joined by an edge with
probability computed from an exponential distribution with rate parameter λ = 1 if their Euclidean distance is at
most 0.2.
5.11.6 thresholded_random_geometric_graph
thresholded_random_geometric_graph(n, radius, theta, dim=2, pos=None, weight=None, p=2,

seed=None)
Returns a thresholded random geometric graph in the unit cube.
The thresholded random geometric graph [1] model places n nodes uniformly at random in the unit cube of dimen-
sions dim. Each node u is assigned a weight wu . Two nodes u and v are joined by an edge if they are within the
maximum connection distance, radius computed by the p-Minkowski distance and the summation of weights
wu + wv is greater than or equal to the threshold parameter theta.
Edges within radius of each other are determined using a KDTree when SciPy is available. This reduces the
time complexity from O(n2 ) to O(n).
Parameters
n
radius: float
theta: float
Threshold value
dim
pos
weight
[dict, optional] Node weights as a dictionary of numbers keyed by node.
p
[float, optional (default 2)] Which Minkowski distance metric to use. p has to meet the con-
dition 1 <= p <= infinity.
probability.

seed
Randomness.
Returns
Graph
A thresholded random geographic graph, undirected and without self-loops.
Each node has a node attribute 'pos' that stores the position of that node in Euclidean space
as provided by the pos keyword argument or, if pos was not provided, as generated by this
function. Similarly, each node has a nodethre attribute 'weight' that stores the weight of
that node as provided or as generated.
Notes

For example, to use a 2D Gaussian distribution of node positions with mean (0, 0) and standard deviation 2
If weights are not specified they are assigned to nodes by drawing randomly from the exponential distribution with
rate parameter λ = 1. To specify weights from a different distribution, use the weight keyword argument:
::
>>> import random

>>> import math
>>> n = 50
>>> w = {i: random.expovariate(5.0) for i in range(n)}
>>> G = nx.thresholded_random_geometric_graph(n, 0.2, 0.1, 2, pos, w)
References
[1]
Examples
Default Graph:
G = nx.thresholded_random_geometric_graph(50, 0.2, 0.1)
Custom Graph:
Create a thresholded random geometric graph on 50 uniformly distributed nodes where nodes are joined by an
edge if their sum weights drawn from a exponential distribution with rate = 5 are >= theta = 0.1 and their Euclidean
distance is at most 0.2.
5.11. Geometric 847

5.11.7 waxman_graph
waxman_graph(n, beta=0.4, alpha=0.1, L=None, domain=(0, 0, 1, 1), metric=None, seed=None)

Returns a Waxman random graph.
The Waxman random graph model places n nodes uniformly at random in a rectangular domain. Each pair of
nodes at distance d is joined by an edge with probability
p = β exp(−d/αL).
This function implements both Waxman models, using the L keyword argument.
• Waxman-1: if L is not specified, it is set to be the maximum distance between any pair of nodes.
• Waxman-2: if L is specified, the distance between a pair of nodes is chosen uniformly at random from the
interval [0, L].
Parameters
n
beta: float
Model parameter
alpha: float
Model parameter
L
[float, optional] Maximum distance between nodes. If not specified, the actual distance is
calculated.
domain
[four-tuple of numbers, optional] Domain size, given as a tuple of the form (x_min,
y_min, x_max, y_max).
metric
[function] A metric on vectors of numbers (represented as lists or tuples). This must be a
function that accepts two lists (or tuples) as input and yields a number as output. The function
must also satisfy the four requirements of a metric. Specifically, if d is the function and x, y,
and z are vectors in the graph, then d must satisfy
1. d(x, y) ≥ 0,
2. d(x, y) = 0 if and only if x = y,
3. d(x, y) = d(y, x),
4. d(x, z) ≤ d(x, y) + d(y, z).
If this argument is not specified, the Euclidean distance metric is used.
seed
Randomness.
Returns
Graph
A random Waxman graph, undirected and without self-loops. Each node has a node attribute
'pos' that stores the position of that node in Euclidean space as generated by this function.

Notes
Starting in NetworkX 2.0 the parameters alpha and beta align with their usual roles in the probability distribution.
In earlier versions their positions in the expression were reversed. Their position in the calling sequence reversed
as well to minimize backward incompatibility.
References
[1]
Examples
Specify an alternate distance metric using the metric keyword argument. For example, to use the “taxicab metric”
instead of the default Euclidean metric:
>>> dist = lambda x, y: sum(abs(a - b) for a, b in zip(x, y))

>>> G = nx.waxman_graph(10, 0.5, 0.1, metric=dist)
5.12 Line Graph
Functions for generating line graphs.
line_graph(G[, create_using]) Returns the line graph of the graph or digraph G.

inverse_line_graph(G) Returns the inverse line graph of graph G.
5.12.1 line_graph
line_graph(G, create_using=None)
Returns the line graph of the graph or digraph G.
The line graph of a graph G has a node for each edge in G and an edge joining those nodes if the two edges in
G share a common node. For directed graphs, nodes are adjacent exactly when the edges they represent form a
directed path of length two.
The nodes of the line graph are 2-tuples of nodes in the original graph (or 3-tuples for multigraphs, with the key of
the edge as the third element).
For information about self-loops and more discussion, see the Notes section below.
Parameters
G
[graph] A NetworkX Graph, DiGraph, MultiGraph, or MultiDigraph.
create_using
Returns
L
[graph] The line graph of G.
5.12. Line Graph 849

Notes
Graph, node, and edge data are not propagated to the new graph. For undirected graphs, the nodes in G must be
sortable, otherwise the constructed line graph may not be correct.
Self-loops in undirected graphs
For an undirected graph G without multiple edges, each edge can be written as a set {u, v}. Its line graph L
has the edges of G as its nodes. If x and y are two nodes in L, then {x, y} is an edge in L if and only if the
intersection of x and y is nonempty. Thus, the set of all edges is determined by the set of all pairwise intersections
of edges in G.
Trivially, every edge in G would have a nonzero intersection with itself, and so every node in L should have a
self-loop. This is not so interesting, and the original context of line graphs was with simple graphs, which had no
self-loops or multiple edges. The line graph was also meant to be a simple graph and thus, self-loops in L are not
part of the standard definition of a line graph. In a pairwise intersection matrix, this is analogous to excluding the
diagonal entries from the line graph definition.
Self-loops and multiple edges in G add nodes to L in a natural way, and do not require any fundamental changes
to the definition. It might be argued that the self-loops we excluded before should now be included. However, the
self-loops are still “trivial” in some sense and thus, are usually excluded.
Self-loops in directed graphs
For a directed graph G without multiple edges, each edge can be written as a tuple (u, v). Its line graph L has
the edges of G as its nodes. If x and y are two nodes in L, then (x, y) is an edge in L if and only if the tail of x
matches the head of y, for example, if x = (a, b) and y = (b, c) for some vertices a, b, and c in G.
Due to the directed nature of the edges, it is no longer the case that every edge in G should have a self-loop in L.
Now, the only time self-loops arise is if a node in G itself has a self-loop. So such self-loops are no longer “trivial”
but instead, represent essential features of the topology of G. For this reason, the historical development of line
digraphs is such that self-loops are included. When the graph G has multiple edges, once again only superficial
changes are required to the definition.
References
• Harary, Frank, and Norman, Robert Z., “Some properties of line digraphs”, Rend. Circ. Mat. Palermo, II.
Ser. 9 (1960), 161–168.
• Hemminger, R. L.; Beineke, L. W. (1978), “Line graphs and line digraphs”, in Beineke, L. W.; Wilson, R.
J., Selected Topics in Graph Theory, Academic Press Inc., pp. 271–305.
Examples
>>> L = nx.line_graph(G)
>>> print(sorted(map(sorted, L.edges()))) # makes a 3-clique, K3
[[(0, 1), (0, 2)], [(0, 1), (0, 3)], [(0, 2), (0, 3)]]
Edge attributes from G are not copied over as node attributes in L, but attributes can be copied manually:
>>> G.add_edges_from((u, v, {"tot": u+v}) for u, v in G.edges)
>>> G.edges(data=True)
EdgeDataView([(0, 1, {'tot': 1}), (1, 2, {'tot': 3}), (2, 3, {'tot': 5})])
>>> H = nx.line_graph(G)


>>> H.add_nodes_from((node, G.edges[node]) for node in H)
>>> H.nodes(data=True)
NodeDataView({(0, 1): {'tot': 1}, (2, 3): {'tot': 5}, (1, 2): {'tot': 3}})
5.12.2 inverse_line_graph
inverse_line_graph(G)
Returns the inverse line graph of graph G.
If H is a graph, and G is the line graph of H, such that G = L(H). Then H is the inverse line graph of G.
Not all graphs are line graphs and these do not have an inverse line graph. In these cases this function raises a
NetworkXError.
Parameters
G
[graph] A NetworkX Graph
Returns
H
[graph] The inverse line graph of G.
Raises
If G is directed or a multigraph
NetworkXError
If G is not a line graph
Notes
This is an implementation of the Roussopoulos algorithm.

If G consists of multiple components, then the algorithm doesn’t work. You should invert every component seper-
ately:
>>> P4 = nx.Graph([("a", "b"), ("b", "c"), ("c", "d")])
>>> G = nx.union(K5, P4)
>>> root_graphs = []
>>> for comp in nx.connected_components(G):
... root_graphs.append(nx.inverse_line_graph(G.subgraph(comp)))
>>> len(root_graphs)
2
5.12. Line Graph 851

References
• Roussopolous, N, “A max {m, n} algorithm for determining the graph H from its line graph G”, Information
Processing Letters 2, (1973), 108–112.
5.13 Ego Graph
Ego graph.
ego_graph(G, n[, radius, center, ...]) Returns induced subgraph of neighbors centered at node
n within a given radius.
5.13.1 ego_graph
ego_graph(G, n, radius=1, center=True, undirected=False, distance=None)

Returns induced subgraph of neighbors centered at node n within a given radius.
Parameters
G
[graph] A NetworkX Graph or DiGraph
n
[node] A single node
radius
[number, optional] Include all neighbors of distance<=radius from n.
center
[bool, optional] If False, do not include center node in graph
undirected
[bool, optional] If True use both in- and out-neighbors of directed graphs.
distance
[key, optional] Use specified edge data key as distance. For example, setting distance=’weight’
will use the edge weight to measure the distance from the node n.
Notes
For directed graphs D this produces the “out” neighborhood or successors. If you want the neighborhood of pre-
decessors first reverse the graph with D.reverse(). If you want both directions use the keyword argument undi-
rected=True.
Node, edge, and graph attributes are copied to the returned subgraph.

5.14 Stochastic
Functions for generating stochastic graphs from a given weighted directed graph.
stochastic_graph(G[, copy, weight]) Returns a right-stochastic representation of directed

graph G.
5.14.1 stochastic_graph
stochastic_graph(G, copy=True, weight='weight')

Returns a right-stochastic representation of directed graph G.
A right-stochastic graph is a weighted digraph in which for each node, the sum of the weights of all the out-edges
of that node is 1. If the graph is already weighted (for example, via a ‘weight’ edge attribute), the reweighting takes
that into account.
Parameters
G
[directed graph] A DiGraph or MultiDiGraph.
copy
[boolean, optional] If this is True, then this function returns a new graph with the stochastic
reweighting. Otherwise, the original graph is modified in-place (and also returned, for conve-
nience).
weight
[edge attribute key (optional, default=’weight’)] Edge attribute key used for reading the existing
weight and setting the new weight. If no attribute with this key is found for an edge, then the
edge weight is assumed to be 1. If an edge has a weight, it must be a positive number.
5.15 AS graph
Generates graphs resembling the Internet Autonomous System network
random_internet_as_graph(n[, seed]) Generates a random undirected graph resembling the In-

ternet AS network
5.15.1 random_internet_as_graph
random_internet_as_graph(n, seed=None)
Generates a random undirected graph resembling the Internet AS network
Parameters
n: integer in [1000, 10000]
Number of graph nodes
seed
Randomness.
5.14. Stochastic 853

Returns
G: Networkx Graph object
A randomly generated undirected graph
Notes
This algorithm returns an undirected graph resembling the Internet Autonomous System (AS) network, it uses the
approach by Elmokashfi et al. [1] and it grants the properties described in the related paper [1].
Each node models an autonomous system, with an attribute ‘type’ specifying its kind; tier-1 (T), mid-level (M),
customer (C) or content-provider (CP). Each edge models an ADV communication link (hence, bidirectional) with
attributes:
• type: transit|peer, the kind of commercial agreement between nodes;
• customer: <node id>, the identifier of the node acting as customer (‘none’ if type is peer).
References
[1]
5.16 Intersection
Generators for random intersection graphs.
uniform_random_intersection_graph(n, m, Returns a uniform random intersection graph.

p[, ...])
k_random_intersection_graph(n, m, k[, Returns a intersection graph with randomly chosen at-
seed]) tribute sets for each node that are of equal size (k).
general_random_intersection_graph(n, m, Returns a random intersection graph with independent
p[, ...]) probabilities for connections between node and attribute
sets.
5.16.1 uniform_random_intersection_graph
uniform_random_intersection_graph(n, m, p, seed=None)
Returns a uniform random intersection graph.
Parameters
n
[int] The number of nodes in the first bipartite set (nodes)
m
[int] The number of nodes in the second bipartite set (attributes)
p
[float] Probability of connecting nodes between bipartite sets
seed
Randomness.

See also:
gnp_random_graph
References
[1], [2]
5.16.2 k_random_intersection_graph
k_random_intersection_graph(n, m, k, seed=None)
Returns a intersection graph with randomly chosen attribute sets for each node that are of equal size (k).
Parameters
n
m
k
[float] Size of attribute set to assign to each node.
seed
Randomness.
See also:
gnp_random_graph, uniform_random_intersection_graph
References
[1]
5.16.3 general_random_intersection_graph
general_random_intersection_graph(n, m, p, seed=None)
Returns a random intersection graph with independent probabilities for connections between node and attribute
sets.
Parameters
n
m
p
[list of floats of length m] Probabilities for connecting nodes to each attribute
5.16. Intersection 855

seed
Randomness.
See also:
gnp_random_graph, uniform_random_intersection_graph
References
[1]
5.17 Social Networks
Famous social networks.
karate_club_graph() Returns Zachary's Karate Club graph.

davis_southern_women_graph() Returns Davis Southern women social network.
florentine_families_graph() Returns Florentine families graph.
les_miserables_graph() Returns coappearance network of characters in the novel
Les Miserables.
5.17.1 karate_club_graph
karate_club_graph()
Returns Zachary’s Karate Club graph.
Each node in the returned graph has a node attribute ‘club’ that indicates the name of the club to which the member
represented by that node belongs, either ‘Mr. Hi’ or ‘Officer’. Each edge has a weight based on the number of
contexts in which that edge’s incident node members interacted.
References
[1]
Examples
To get the name of the club to which a node belongs:
>>> G.nodes[5]["club"]
'Mr. Hi'
>>> G.nodes[9]["club"]
'Officer'

5.17.2 davis_southern_women_graph
davis_southern_women_graph()
Returns Davis Southern women social network.
This is a bipartite graph.
References
[1]
5.17.3 florentine_families_graph
florentine_families_graph()
Returns Florentine families graph.
References
[1]
5.17.4 les_miserables_graph
les_miserables_graph()
Returns coappearance network of characters in the novel Les Miserables.
References
[1]
5.18 Community
Generators for classes of graphs used in studying social networks.
caveman_graph(l, k) Returns a caveman graph of l cliques of size k.

connected_caveman_graph(l, k) Returns a connected caveman graph of l cliques of size
k.
gaussian_random_partition_graph(n, s, v, Generate a Gaussian random partition graph.
...)
LFR_benchmark_graph(n, tau1, tau2, mu[, ...]) Returns the LFR benchmark graph.
planted_partition_graph(l, k, p_in, p_out[, Returns the planted l-partition graph.
...])
random_partition_graph(sizes, p_in, p_out[, Returns the random partition graph with a partition of
...]) sizes.
relaxed_caveman_graph(l, k, p[, seed]) Returns a relaxed caveman graph.
ring_of_cliques(num_cliques, clique_size) Defines a "ring of cliques" graph.
stochastic_block_model(sizes, p[, nodelist, ...]) Returns a stochastic block model graph.
windmill_graph(n, k) Generate a windmill graph.
5.18. Community 857

5.18.1 caveman_graph
caveman_graph(l, k)
Returns a caveman graph of l cliques of size k.
Parameters
l
[int] Number of cliques
k
[int] Size of cliques
Returns
G
[NetworkX Graph] caveman graph
See also:
connected_caveman_graph
Notes
This returns an undirected graph, it can be converted to a directed graph using nx.to_directed(), or a multi-
graph using nx.MultiGraph(nx.caveman_graph(l, k)). Only the undirected version is described in
[1] and it is unclear which of the directed generalizations is most useful.
References
[1]
Examples
>>> G = nx.caveman_graph(3, 3)
5.18.2 connected_caveman_graph
connected_caveman_graph(l, k)
Returns a connected caveman graph of l cliques of size k.
The connected caveman graph is formed by creating n cliques of size k, then a single edge in each clique is rewired
to a node in an adjacent clique.
Parameters
l
[int] number of cliques
k
[int] size of cliques (k at least 2 or NetworkXError is raised)
Returns

G
[NetworkX Graph] connected caveman graph
Raises
NetworkXError
If the size of cliques k is smaller than 2.
Notes
This returns an undirected graph, it can be converted to a directed graph using nx.to_directed(), or a multi-
graph using nx.MultiGraph(nx.caveman_graph(l, k)). Only the undirected version is described in
[1] and it is unclear which of the directed generalizations is most useful.
References
[1]
Examples
>>> G = nx.connected_caveman_graph(3, 3)
5.18.3 gaussian_random_partition_graph
gaussian_random_partition_graph(n, s, v, p_in, p_out, directed=False, seed=None)

Generate a Gaussian random partition graph.
A Gaussian random partition graph is created by creating k partitions each with a size drawn from a normal dis-
tribution with mean s and variance s/v. Nodes are connected within clusters with probability p_in and between
clusters with probability p_out[1]
Parameters
n
[int] Number of nodes in the graph
s
[float] Mean cluster size
v
[float] Shape parameter. The variance of cluster size distribution is s/v.
p_in
[float] Probabilty of intra cluster connection.
p_out
[float] Probability of inter cluster connection.
directed
[boolean, optional default=False] Whether to create a directed graph or not
seed
Randomness.
Returns
5.18. Community 859

G
[NetworkX Graph or DiGraph] gaussian random partition graph
Raises
NetworkXError
If s is > n If p_in or p_out is not in [0,1]
See also:
random_partition_graph
Notes
Note the number of partitions is dependent on s,v and n, and that the last partition may be considerably smaller, as
it is sized to simply fill out the nodes [1]
References
[1]
Examples
>>> G = nx.gaussian_random_partition_graph(100, 10, 10, 0.25, 0.1)

>>> len(G)
100
5.18.4 LFR_benchmark_graph
LFR_benchmark_graph(n, tau1, tau2, mu, average_degree=None, min_degree=None, max_degree=None,

min_community=None, max_community=None, tol=1e-07, max_iters=500, seed=None)
Returns the LFR benchmark graph.
This algorithm proceeds as follows:
1) Find a degree sequence with a power law distribution, and minimum value min_degree, which has ap-
proximate average degree average_degree. This is accomplished by either
a) specifying min_degree and not average_degree,
b) specifying average_degree and not min_degree, in which case a suitable minimum degree will
be found.
max_degree can also be specified, otherwise it will be set to n. Each node u will have µdeg(u) edges
joining it to nodes in communities other than its own and (1 − µ)deg(u) edges joining it to nodes in its own
community.
2) Generate community sizes according to a power law distribution with exponent tau2. If min_community
and max_community are not specified they will be selected to be min_degree and max_degree,
respectively. Community sizes are generated until the sum of their sizes equals n.
3) Each node will be randomly assigned a community with the condition that the community is large enough for
the node’s intra-community degree, (1 − µ)deg(u) as described in step 2. If a community grows too large,
a random node will be selected for reassignment to a new community, until all nodes have been assigned a
community.

4) Each node u then adds (1 − µ)deg(u) intra-community edges and µdeg(u) inter-community edges.
Parameters
n
[int] Number of nodes in the created graph.
tau1
[float] Power law exponent for the degree distribution of the created graph. This value must
be strictly greater than one.
tau2
[float] Power law exponent for the community size distribution in the created graph. This value
must be strictly greater than one.
mu
[float] Fraction of inter-community edges incident to each node. This value must be in the
interval [0, 1].
average_degree
[float] Desired average degree of nodes in the created graph. This value must be in the interval
[0, n]. Exactly one of this and min_degree must be specified, otherwise a NetworkXEr-
ror is raised.
min_degree
[int] Minimum degree of nodes in the created graph. This value must be in the interval [0, n].
Exactly one of this and average_degree must be specified, otherwise a NetworkXEr-
ror is raised.
max_degree
[int] Maximum degree of nodes in the created graph. If not specified, this is set to n, the total
number of nodes in the graph.
min_community
[int] Minimum size of communities in the graph. If not specified, this is set to min_degree.
max_community
[int] Maximum size of communities in the graph. If not specified, this is set to n, the total
number of nodes in the graph.
tol
[float] Tolerance when comparing floats, specifically when comparing average degree values.
max_iters
[int] Maximum number of iterations to try to create the community sizes, degree distribution,
and community affiliations.
seed
Randomness.
Returns
G
[NetworkX graph] The LFR benchmark graph generated according to the specified parame-
ters.
Each node in the graph has a node attribute 'community' that stores the community (that
is, the set of nodes) that includes it.
Raises
5.18. Community 861

NetworkXError
If any of the parameters do not meet their upper and lower bounds:
• tau1 and tau2 must be strictly greater than 1.
• mu must be in [0, 1].
• max_degree must be in {1, …, n}.
• min_community and max_community must be in {0, …, n}.
If not exactly one of average_degree and min_degree is specified.
If min_degree is not specified and a suitable min_degree cannot be found.
ExceededMaxIterations
If a valid degree sequence cannot be created within max_iters number of iterations.
If a valid set of community sizes cannot be created within max_iters number of iterations.
If a valid community assignment cannot be created within 10 * n * max_iters number
of iterations.
Notes
This algorithm differs slightly from the original way it was presented in [1].
1) Rather than connecting the graph via a configuration model then rewiring to match the intra-community and
inter-community degrees, we do this wiring explicitly at the end, which should be equivalent.
2) The code posted on the author’s website [2] calculates the random power law distributed variables and their
average using continuous approximations, whereas we use the discrete distributions here as both degree and
community size are discrete.
Though the authors describe the algorithm as quite robust, testing during development indicates that a somewhat
narrower parameter set is likely to successfully produce a graph. Some suggestions have been provided in the event
of exceptions.
References
[1], [2]
Examples
Basic usage:
>>> from networkx.generators.community import LFR_benchmark_graph

>>> n = 250
>>> tau1 = 3
>>> tau2 = 1.5
>>> mu = 0.1
>>> G = LFR_benchmark_graph(
... n, tau1, tau2, mu, average_degree=5, min_community=20, seed=10
... )
Continuing the example above, you can get the communities from the node attributes of the graph:
>>> communities = {frozenset(G.nodes[v]["community"]) for v in G}

5.18.5 planted_partition_graph
planted_partition_graph(l, k, p_in, p_out, seed=None, directed=False)

Returns the planted l-partition graph.
This model partitions a graph with n=l*k vertices in l groups with k vertices each. Vertices of the same group are
linked with a probability p_in, and vertices of different groups are linked with probability p_out.
Parameters
l
[int] Number of groups
k
[int] Number of vertices in each group
p_in
[float] probability of connecting vertices within a group
p_out
[float] probability of connected vertices between groups
seed
Randomness.
directed
[bool,optional (default=False)] If True return a directed graph
Returns
G
[NetworkX Graph or DiGraph] planted l-partition graph
Raises
NetworkXError
If p_in,p_out are not in [0,1] or
See also:
random_partition_model
References
[1], [2]
Examples
>>> G = nx.planted_partition_graph(4, 3, 0.5, 0.1, seed=42)
5.18. Community 863

5.18.6 random_partition_graph
random_partition_graph(sizes, p_in, p_out, seed=None, directed=False)

Returns the random partition graph with a partition of sizes.
A partition graph is a graph of communities with sizes defined by s in sizes. Nodes in the same group are connected
with probability p_in and nodes of different groups are connected with probability p_out.
Parameters
sizes
[list of ints] Sizes of groups
p_in
[float] probability of edges with in groups
p_out
[float] probability of edges between groups
directed
[boolean optional, default=False] Whether to create a directed graph
seed
Randomness.
Returns
G
[NetworkX Graph or DiGraph] random partition graph of size sum(gs)
Raises
NetworkXError
If p_in or p_out is not in [0,1]
Notes
This is a generalization of the planted-l-partition described in [1]. It allows for the creation of groups of any size.
The partition is store as a graph attribute ‘partition’.
References
[1]
Examples
>>> G = nx.random_partition_graph([10, 10, 10], 0.25, 0.01)

>>> len(G)
30
>>> partition = G.graph["partition"]
>>> len(partition)
3

5.18.7 relaxed_caveman_graph
relaxed_caveman_graph(l, k, p, seed=None)
Returns a relaxed caveman graph.
A relaxed caveman graph starts with l cliques of size k. Edges are then randomly rewired with probability p to
link different cliques.
Parameters
l
[int] Number of groups
k
p
[float] Probabilty of rewiring each edge.
seed
Randomness.
Returns
G
[NetworkX Graph] Relaxed Caveman Graph
Raises
NetworkXError
If p is not in [0,1]
References
[1]
Examples
>>> G = nx.relaxed_caveman_graph(2, 3, 0.1, seed=42)
5.18.8 ring_of_cliques
ring_of_cliques(num_cliques, clique_size)
Defines a “ring of cliques” graph.
A ring of cliques graph is consisting of cliques, connected through single links. Each clique is a complete graph.
Parameters
num_cliques
clique_size
Returns
5.18. Community 865

G
[NetworkX Graph] ring of cliques graph
Raises
NetworkXError
If the number of cliques is lower than 2 or if the size of cliques is smaller than 2.
See also:
connected_caveman_graph
Notes
The connected_caveman_graph graph removes a link from each clique to connect it with the next clique.
Instead, the ring_of_cliques graph simply adds the link without removing any link from the cliques.
Examples
>>> G = nx.ring_of_cliques(8, 4)
5.18.9 stochastic_block_model
stochastic_block_model(sizes, p, nodelist=None, seed=None, directed=False, selfloops=False, sparse=True)

Returns a stochastic block model graph.
This model partitions the nodes in blocks of arbitrary sizes, and places edges between pairs of nodes independently,
with a probability that depends on the blocks.
Parameters
sizes
[list of ints] Sizes of blocks
p
[list of list of floats] Element (r,s) gives the density of edges going from the nodes of group r
to nodes of group s. p must match the number of groups (len(sizes) == len(p)), and it must be
symmetric if the graph is undirected.
nodelist
[list, optional] The block tags are assigned according to the node identifiers in nodelist. If
nodelist is None, then the ordering is the range [0,sum(sizes)-1].
seed
Randomness.
directed
[boolean optional, default=False] Whether to create a directed graph or not.
selfloops
[boolean optional, default=False] Whether to include self-loops or not.
sparse: boolean optional, default=True
Use the sparse heuristic to speed up the generator.
Returns

g
[NetworkX Graph or DiGraph] Stochastic block model graph of size sum(sizes)
Raises
NetworkXError
If probabilities are not in [0,1]. If the probability matrix is not square (directed case). If the
probability matrix is not symmetric (undirected case). If the sizes list does not match nodelist
or the probability matrix. If nodelist contains duplicate.
See also:
random_partition_graph
planted_partition_graph
gaussian_random_partition_graph
gnp_random_graph
References
[1]
Examples
>>> sizes = [75, 75, 300]

>>> probs = [[0.25, 0.05, 0.02], [0.05, 0.35, 0.07], [0.02, 0.07, 0.40]]
>>> g = nx.stochastic_block_model(sizes, probs, seed=0)
>>> len(g)
450
>>> H = nx.quotient_graph(g, g.graph["partition"], relabel=True)
>>> for v in H.nodes(data=True):
... print(round(v[1]["density"], 3))
...
0.245
0.348
0.405
>>> for v in H.edges(data=True):
... print(round(1.0 * v[2]["weight"] / (sizes[v[0]] * sizes[v[1]]), 3))
...
0.051
0.022
0.07
5.18.10 windmill_graph
windmill_graph(n, k)
Generate a windmill graph. A windmill graph is a graph of n cliques each of size k that are all joined at one node.
It can be thought of as taking a disjoint union of n cliques of size k, selecting one point from each, and contracting
all of the selected points. Alternatively, one could generate n cliques of size k-1 and one node that is connected
to all other nodes in the graph.
Parameters
n
5.18. Community 867

k
Returns
G
[NetworkX Graph] windmill graph with n cliques of size k
Raises
NetworkXError
If the number of cliques is less than two If the size of the cliques are less than two
Notes
The node labeled 0 will be the node connected to all other nodes. Note that windmill graphs are usually denoted
Wd(k,n), so the parameters are in the opposite order as the parameters of this method.
Examples
>>> G = nx.windmill_graph(4, 5)
5.19 Spectral
Generates graphs with a given eigenvector structure
spectral_graph_forge(G, alpha[, ...]) Returns a random simple graph with spectrum resembling
that of G
5.19.1 spectral_graph_forge
spectral_graph_forge(G, alpha, transformation='identity', seed=None)

Returns a random simple graph with spectrum resembling that of G
This algorithm, called Spectral Graph Forge (SGF), computes the eigenvectors of a given graph adjacency matrix,
filters them and builds a random graph with a similar eigenstructure. SGF has been proved to be particularly useful
for synthesizing realistic social networks and it can also be used to anonymize graph sensitive data.
Parameters
G
[Graph]
alpha
[float] Ratio representing the percentage of eigenvectors of G to consider, values in [0,1].
transformation
[string, optional] Represents the intended matrix linear transformation, possible values are
‘identity’ and ‘modularity’
seed
[integer, random_state, or None (default)] Indicator of numpy random number generation
state. See Randomness.

Returns
H
[Graph] A graph with a similar eigenvector structure of the input one.
Raises
NetworkXError
If transformation has a value different from ‘identity’ or ‘modularity’
Notes
Spectral Graph Forge (SGF) generates a random simple graph resembling the global properties of the given one. It
leverages the low-rank approximation of the associated adjacency matrix driven by the alpha precision parameter.
SGF preserves the number of nodes of the input graph and their ordering. This way, nodes of output graphs
resemble the properties of the input one and attributes can be directly mapped.
It considers the graph adjacency matrices which can optionally be transformed to other symmetric real matrices
(currently transformation options include identity and modularity). The modularity transformation, in the sense of
Newman’s modularity matrix allows the focusing on community structure related properties of the graph.
SGF applies a low-rank approximation whose fixed rank is computed from the ratio alpha of the input graph
adjacency matrix dimension. This step performs a filtering on the input eigenvectors similar to the low pass filtering
common in telecommunications.
The filtered values (after truncation) are used as input to a Bernoulli sampling for constructing a random adjacency
matrix.
References
Examples
>>> H = nx.spectral_graph_forge(G, 0.3)
>>>
5.20 Trees
Functions for generating trees.
random_tree(n[, seed, create_using]) Returns a uniformly random tree on n nodes.

prefix_tree(paths) Creates a directed prefix tree from a list of paths.
5.20. Trees 869

5.20.1 random_tree
random_tree(n, seed=None, create_using=None)

Returns a uniformly random tree on n nodes.
Parameters
n
[int] A positive integer representing the number of nodes in the tree.
seed
Randomness.
create_using
Returns
NetworkX graph
A tree, given as an undirected graph, whose nodes are numbers in the set {0, …, n - 1}.
Raises
If n is zero (because the null graph is not a tree).
Notes
The current implementation of this function generates a uniformly random Prüfer sequence then converts that to
a tree via the from_prufer_sequence() function. Since there is a bijection between Prüfer sequences of
length n - 2 and trees on n nodes, the tree is chosen uniformly at random from the set of all trees on n nodes.
Examples
>>> tree = nx.random_tree(n=10, seed=0)

>>> print(nx.forest_str(tree, sources=[0]))
╙── 0
├── 3
└── 4
├── 6
│ ├── 1
│ ├── 2
│ └── 7
│ └── 8
│ └── 5
└── 9
>>> tree = nx.random_tree(n=10, seed=0, create_using=nx.DiGraph)

>>> print(nx.forest_str(tree))
╙── 0
├─╼ 3
└─╼ 4
├─╼ 6
│ ├─╼ 1


│ ├─╼ 2
│ └─╼ 7
│ └─╼ 8
│ └─╼ 5
└─╼ 9
5.20.2 prefix_tree
prefix_tree(paths)
Creates a directed prefix tree from a list of paths.
Usually the paths are described as strings or lists of integers.
A “prefix tree” represents the prefix structure of the strings. Each node represents a prefix of some string. The root
represents the empty prefix with children for the single letter prefixes which in turn have children for each double
letter prefix starting with the single letter corresponding to the parent node, and so on.
More generally the prefixes do not need to be strings. A prefix refers to the start of a sequence. The root has
children for each one element prefix and they have children for each two element prefix that starts with the one
element sequence of the parent, and so on.
Note that this implementation uses integer nodes with an attribute. Each node has an attribute “source” whose
value is the original element of the path to which this node corresponds. For example, suppose paths consists of
one path: “can”. Then the nodes [1, 2, 3] which represent this path have “source” values “c”, “a” and “n”.
All the descendants of a node have a common prefix in the sequence/path associated with that node. From the
returned tree, the prefix for each node can be constructed by traversing the tree up to the root and accumulating
the “source” values along the way.
The root node is always 0 and has “source” attribute None. The root is the only node with in-degree zero. The nil
node is always -1 and has “source” attribute "NIL". The nil node is the only node with out-degree zero.
Parameters
paths: iterable of paths
An iterable of paths which are themselves sequences. Matching prefixes among these se-
quences are identified with nodes of the prefix tree. One leaf of the tree is associated with
each path. (Identical paths are associated with the same leaf of the tree.)
Returns
tree: DiGraph
A directed graph representing an arborescence consisting of the prefix tree generated by
paths. Nodes are directed “downward”, from parent to child. A special “synthetic” root
node is added to be the parent of the first node in each path. A special “synthetic” leaf node,
the “nil” node -1, is added to be the child of all nodes representing the last element in a path.
(The addition of this nil node technically makes this not an arborescence but a directed acyclic
graph; removing the nil node makes it an arborescence.)
5.20. Trees 871

Notes
The prefix tree is also known as a trie.
Examples
Create a prefix tree from a list of strings with common prefixes:

>>> paths = ["ab", "abs", "ad"]
>>> T = nx.prefix_tree(paths)
>>> list(T.edges)
[(0, 1), (1, 2), (1, 4), (2, -1), (2, 3), (3, -1), (4, -1)]
The leaf nodes can be obtained as predecessors of the nil node:

>>> root, NIL = 0, -1
>>> list(T.predecessors(NIL))
[2, 3, 4]
To recover the original paths that generated the prefix tree, traverse up the tree from the node -1 to the node 0:
>>> recovered = []
>>> for v in T.predecessors(NIL):
... prefix = ""
... while v != root:
... prefix = str(T.nodes[v]["source"]) + prefix
... v = next(T.predecessors(v)) # only one predecessor
... recovered.append(prefix)
>>> sorted(recovered)
['ab', 'abs', 'ad']
5.21 Non Isomorphic Trees
Implementation of the Wright, Richmond, Odlyzko and McKay (WROM) algorithm for the enumeration of all non-
isomorphic free trees of a given order. Rooted trees are represented by level sequences, i.e., lists in which the i-th
element specifies the distance of vertex i to the root.
nonisomorphic_trees(order[, create]) Returns a list of nonisomporphic trees

number_of_nonisomorphic_trees(order) Returns the number of nonisomorphic trees
5.21.1 nonisomorphic_trees
nonisomorphic_trees(order, create='graph')
Returns a list of nonisomporphic trees
Parameters
order
[int] order of the desired tree(s)
create
[graph or matrix (default=”Graph)] If graph is selected a list of trees will be returned, if matrix
is selected a list of adjancency matrix will be returned

Returns
G
[List of NetworkX Graphs]
M
[List of Adjacency matrices]
5.21.2 number_of_nonisomorphic_trees
number_of_nonisomorphic_trees(order)
Returns the number of nonisomorphic trees
Parameters
order
[int] order of the desired tree(s)
Returns
length
[Number of nonisomorphic graphs for the given order]
5.22 Triads
Functions that generate the triad graphs, that is, the possible digraphs on three nodes.
triad_graph(triad_name) Returns the triad graph with the given name.
5.22.1 triad_graph
triad_graph(triad_name)
Returns the triad graph with the given name.
Each string in the following tuple is a valid triad name:
('003', '012', '102', '021D', '021U', '021C', '111D', '111U',

'030T', '030C', '201', '120D', '120U', '120C', '210', '300')
Each triad name corresponds to one of the possible valid digraph on three nodes.
Parameters
triad_name
[string] The name of a triad, as described above.
Returns
DiGraph
The digraph on three nodes with the given name. The nodes of the graph are the single-
character strings ‘a’, ‘b’, and ‘c’.
Raises
ValueError
If triad_name is not the name of a triad.
5.22. Triads 873

See also:
triadic_census
5.23 Joint Degree Sequence
Generate graphs with a given joint degree and directed joint degree
is_valid_joint_degree(joint_degrees) Checks whether the given joint degree dictionary is real-

izable.
joint_degree_graph(joint_degrees[, seed]) Generates a random simple graph with the given joint de-
gree dictionary.
is_valid_directed_joint_degree(in_degrees, Checks whether the given directed joint degree input is
...) realizable
directed_joint_degree_graph(in_degrees, ...) Generates a random simple directed graph with the joint
degree.
5.23.1 is_valid_joint_degree
is_valid_joint_degree(joint_degrees)
Checks whether the given joint degree dictionary is realizable.
A joint degree dictionary is a dictionary of dictionaries, in which entry joint_degrees[k][l] is an integer
representing the number of edges joining nodes of degree k with nodes of degree l. Such a dictionary is realizable
as a simple graph if and only if the following conditions are satisfied.
• each entry must be an integer,
• the total number of nodes of degree k, computed by sum(joint_degrees[k].values()) / k,
must be an integer,
• the total number of edges joining nodes of degree k with nodes of degree l cannot exceed the total number
of possible edges,
• each diagonal entry joint_degrees[k][k] must be even (this is a convention assumed by the
joint_degree_graph() function).
Parameters
joint_degrees
[dictionary of dictionary of integers] A joint degree dictionary in which entry
joint_degrees[k][l] is the number of edges joining nodes of degree k with
nodes of degree l.
Returns
bool
Whether the given joint degree dictionary is realizable as a simple graph.

References
[1], [2]
5.23.2 joint_degree_graph
joint_degree_graph(joint_degrees, seed=None)
Generates a random simple graph with the given joint degree dictionary.
Parameters
joint_degrees
[dictionary of dictionary of integers] A joint degree dictionary in which entry
joint_degrees[k][l] is the number of edges joining nodes of degree k with
nodes of degree l.
seed
Randomness.
Returns
G
[Graph] A graph with the specified joint degree dictionary.
Raises
NetworkXError
If joint_degrees dictionary is not realizable.
Notes
In each iteration of the “while loop” the algorithm picks two disconnected nodes v and w, of degree k and l cor-
respondingly, for which joint_degrees[k][l] has not reached its target yet. It then adds edge (v, w) and
increases the number of edges in graph G by one.
The intelligence of the algorithm lies in the fact that it is always possible to add an edge between such disconnected
nodes v and w, even if one or both nodes do not have free stubs. That is made possible by executing a “neighbor
switch”, an edge rewiring move that releases a free stub while keeping the joint degree of G the same.
The algorithm continues for E (number of edges) iterations of the “while loop”, at the which point all entries of the
given joint_degrees[k][l] have reached their target values and the construction is complete.
References
Examples
>>> joint_degrees = {
... 1: {4: 1},
... 2: {2: 2, 3: 2, 4: 2},
... 3: {2: 2, 4: 1},
... 4: {1: 1, 2: 2, 3: 1},
... }
>>> G = nx.joint_degree_graph(joint_degrees)
>>>
5.23. Joint Degree Sequence 875

5.23.3 is_valid_directed_joint_degree
is_valid_directed_joint_degree(in_degrees, out_degrees, nkk)

Checks whether the given directed joint degree input is realizable
Parameters
in_degrees
[list of integers] in degree sequence contains the in degrees of nodes.
out_degrees
[list of integers] out degree sequence contains the out degrees of nodes.
nkk
[dictionary of dictionary of integers] directed joint degree dictionary. for nodes of out degree
k (first level of dict) and nodes of in degree l (seconnd level of dict) describes the number of
edges.
Returns
boolean
returns true if given input is realizable, else returns false.
Notes
Here is the list of conditions that the inputs (in/out degree sequences, nkk) need to satisfy for simple directed graph
realizability:
• Condition 0: in_degrees and out_degrees have the same length
• Condition 1: nkk[k][l] is integer for all k,l
• Condition 2: sum(nkk[k])/k = number of nodes with partition id k, is an
integer and matching degree sequence
• Condition 3: number of edges and non-chords between k and l cannot exceed
maximum possible number of edges
References
[1] B. Tillman, A. Markopoulou, C. T. Butts & M. Gjoka,

“Construction of Directed 2K Graphs”. In Proc. of KDD 2017.
5.23.4 directed_joint_degree_graph
directed_joint_degree_graph(in_degrees, out_degrees, nkk, seed=None)

Generates a random simple directed graph with the joint degree.
Parameters
degree_seq
[list of tuples (of size 3)] degree sequence contains tuples of nodes with node id, in degree and
out degree.
nkk
[dictionary of dictionary of integers] directed joint degree dictionary, for nodes of out degree

k (first level of dict) and nodes of in degree l (second level of dict) describes the number of
edges.
seed
[hashable object, optional] Seed for random number generator.
Returns
G
[Graph] A directed graph with the specified inputs.
Raises
NetworkXError
If degree_seq and nkk are not realizable as a simple directed graph.
Notes
Similarly to the undirected version: In each iteration of the “while loop” the algorithm picks two disconnected
nodes v and w, of degree k and l correspondingly, for which nkk[k][l] has not reached its target yet i.e. (for given
k,l): n_edges_add < nkk[k][l]. It then adds edge (v,w) and always increases the number of edges in graph G by
one.
The intelligence of the algorithm lies in the fact that it is always possible to add an edge between disconnected
nodes v and w, for which nkk[degree(v)][degree(w)] has not reached its target, even if one or both nodes do not
have free stubs. If either node v or w does not have a free stub, we perform a “neighbor switch”, an edge rewiring
move that releases a free stub while keeping nkk the same.
The difference for the directed version lies in the fact that neighbor switches might not be able to rewire, but in
these cases unsaturated nodes can be reassigned to use instead, see [1] for detailed description and proofs.
The algorithm continues for E (number of edges in the graph) iterations of the “while loop”, at which point all
entries of the given nkk[k][l] have reached their target values and the construction is complete.
References
[1] B. Tillman, A. Markopoulou, C. T. Butts & M. Gjoka,

“Construction of Directed 2K Graphs”. In Proc. of KDD 2017.
Examples
>>> in_degrees = [0, 1, 1, 2]

>>> out_degrees = [1, 1, 1, 1]
>>> nkk = {1: {1: 2, 2: 2}}
>>> G = nx.directed_joint_degree_graph(in_degrees, out_degrees, nkk)
>>>
5.23. Joint Degree Sequence 877

5.24 Mycielski
Functions related to the Mycielski Operation and the Mycielskian family of graphs.
mycielskian(G[, iterations]) Returns the Mycielskian of a simple, undirected graph G

mycielski_graph(n) Generator for the n_th Mycielski Graph.
5.24.1 mycielskian
mycielskian(G, iterations=1)
Returns the Mycielskian of a simple, undirected graph G
The Mycielskian of graph preserves a graph’s triangle free property while increasing the chromatic number by 1.
The Mycielski Operation on a graph, G = (V, E), constructs a new graph with 2|V | + 1 nodes and 3|E| + |V |
edges.
The construction is as follows:
Let V = 0,∪..., n∪− 1. Construct another vertex set U = n, ..., 2n and a vertex, w. Construct a new graph, M, with
vertices U V w. For edges, (u, v) ∈ E add edges (u, v), (u, v + n), and (u + n, v) to M. Finally, for all
vertices u ∈ U , add edge (u, w) to M.
The Mycielski Operation can be done multiple times by repeating the above process iteratively.
More information can be found at https://en.wikipedia.org/wiki/Mycielskian
Parameters
G
[graph] A simple, undirected NetworkX graph
iterations
[int] The number of iterations of the Mycielski operation to perform on G. Defaults to 1. Must
be a non-negative integer.
Returns
M
[graph] The Mycielskian of G after the specified number of iterations.
Notes
Graph, node, and edge data are not necessarily propagated to the new graph.
5.24.2 mycielski_graph
mycielski_graph(n)
Generator for the n_th Mycielski Graph.
The Mycielski family of graphs is an infinite set of graphs. M1 is the singleton graph, M2 is two vertices with an
edge, and, for i > 2, Mi is the Mycielskian of Mi−1 .
More information can be found at http://mathworld.wolfram.com/MycielskiGraph.html
Parameters

n
[int] The desired Mycielski Graph.
Returns
M
[graph] The n_th Mycielski Graph
Notes
The first graph in the Mycielski sequence is the singleton graph. The Mycielskian of this graph is not the P2 graph,
but rather the P2 graph with an extra, isolated vertex. The second Mycielski graph is the P2 graph, so the first two
are hard coded. The remaining graphs are generated using the Mycielski operation.
5.25 Harary Graph
Generators for Harary graphs

This module gives two generators for the Harary graph, which was introduced by the famous mathematician Frank Harary
in his 1962 work [H]. The first generator gives the Harary graph that maximizes the node connectivity with given number
of nodes and given number of edges. The second generator gives the Harary graph that minimizes the number of edges
in the graph with given node connectivity and number of nodes.
5.25.1 References
hnm_harary_graph(n, m[, create_using]) Returns the Harary graph with given numbers of nodes
and edges.
hkn_harary_graph(k, n[, create_using]) Returns the Harary graph with given node connectivity
and node number.
5.25.2 hnm_harary_graph
hnm_harary_graph(n, m, create_using=None)
Returns the Harary graph with given numbers of nodes and edges.
The Harary graph Hn,m is the graph that maximizes node connectivity with n nodes and m edges.
This maximum node connectivity is known to be floor(2m/n). [1]
Parameters
n: integer
The number of nodes the generated graph is to contain
m: integer
The number of edges the generated graph is to contain
create_using
[NetworkX graph constructor, optional Graph type] to create (default=nx.Graph). If graph
Returns
5.25. Harary Graph 879

NetworkX graph
The Harary graph Hn,m .
See also:
hkn_harary_graph
Notes
This algorithm runs in O(m) time. It is implemented by following the Reference [2].
References
[1], [2]
5.25.3 hkn_harary_graph
hkn_harary_graph(k, n, create_using=None)
Returns the Harary graph with given node connectivity and node number.
The Harary graph Hk,n is the graph that minimizes the number of edges needed with given node connectivity k
and node number n.
This smallest number of edges is known to be ceil(kn/2) [1].
Parameters
k: integer
The node connectivity of the generated graph
n: integer
The number of nodes the generated graph is to contain
create_using
[NetworkX graph constructor, optional Graph type] to create (default=nx.Graph). If graph
Returns
NetworkX graph
The Harary graph Hk,n .
See also:
hnm_harary_graph

Notes
This algorithm runs in O(kn) time. It is implemented by following the Reference [2].
References
[1], [2]
5.26 Cographs
Generators for cographs

A cograph is a graph containing no path on four vertices. Cographs or P4 -free graphs can be obtained from a single vertex
by disjoint union and complementation operations.
5.26.1 References
random_cograph(n[, seed]) Returns a random cograph with 2n nodes.
5.26.2 random_cograph
random_cograph(n, seed=None)
Returns a random cograph with 2n nodes.
A cograph is a graph containing no path on four vertices. Cographs or P4 -free graphs can be obtained from a single
vertex by disjoint union and complementation operations.
This generator starts off from a single vertex and performes disjoint union and full join operations on itself. The
decision on which operation will take place is random.
Parameters
n
[int] The order of the cograph.
seed
Randomness.
Returns
G
[A random graph containing no path on four vertices.]
See also:
full_join
union
5.26. Cographs 881

References
[1]
5.27 Interval Graph
Generators for interval graph.
interval_graph(intervals) Generates an interval graph for a list of intervals given.
5.27.1 interval_graph
interval_graph(intervals)
Generates an interval graph for a list of intervals given.
In graph theory, an interval graph is an undirected graph formed from a set of closed intervals on the real line, with
a vertex for each interval and an edge between vertices whose intervals intersect. It is the intersection graph of the
intervals.
More information can be found at: https://en.wikipedia.org/wiki/Interval_graph
Parameters
intervals
[a sequence of intervals, say (l, r) where l is the left end,]
and r is the right end of the closed interval.
Returns
G
[networkx graph]
Raises
TypeError
if intervals contains None or an element which is not collections.abc.Sequence or not a
length of 2.
ValueError
if intervals contains an interval such that min1 > max1 where min1,max1 = interval
Examples
>>> intervals = [(-2, 3), [1, 4], (2, 3), (4, 6)]
>>> G = nx.interval_graph(intervals)
>>> sorted(G.edges)
[((-2, 3), (1, 4)), ((-2, 3), (2, 3)), ((1, 4), (2, 3)), ((1, 4), (4, 6))]

5.28 Sudoku
Generator for Sudoku graphs

This module gives a generator for n-Sudoku graphs. It can be used to develop algorithms for solving or generating Sudoku
puzzles.
A completed Sudoku grid is a 9x9 array of integers between 1 and 9, with no number appearing twice in the same row,
column, or 3x3 box.
864 371 259

325 849 761
971 265 843
436 192 587

198 657 432
257 483 916
689 734 125

713 528 694
542 916 378
The Sudoku graph is an undirected graph with 81 vertices, corresponding to the cells of a Sudoku grid. It is a regular
graph of degree 20. Two distinct vertices are adjacent if and only if the corresponding cells belong to the same row,
column, or box. A completed Sudoku grid corresponds to a vertex coloring of the Sudoku graph with nine colors.
More generally, the n-Sudoku graph is a graph with n^4 vertices, corresponding to the cells of an n^2 by n^2 grid. Two
distinct vertices are adjacent if and only if they belong to the same row, column, or n by n box.
5.28.1 References
sudoku_graph([n]) Returns the n-Sudoku graph.
5.28.2 sudoku_graph
sudoku_graph(n=3)
Returns the n-Sudoku graph. The default value of n is 3.
The n-Sudoku graph is a graph with n^4 vertices, corresponding to the cells of an n^2 by n^2 grid. Two distinct
vertices are adjacent if and only if they belong to the same row, column, or n-by-n box.
Parameters
n: integer
The order of the Sudoku graph, equal to the square root of the number of rows. The default is
3.
5.28. Sudoku 883

Returns
NetworkX graph
The n-Sudoku graph Sud(n).
References
[1], [2], [3]
Examples
>>> G = nx.sudoku_graph()
81
810
>>> sorted(G.neighbors(42))
[6, 15, 24, 33, 34, 35, 36, 37, 38, 39, 40, 41, 43, 44, 51, 52, 53, 60, 69, 78]
>>> G = nx.sudoku_graph(2)
16
56

CHAPTER
SIX
LINEAR ALGEBRA
6.1 Graph Matrix
Adjacency matrix and incidence matrix of graphs.
adjacency_matrix(G[, nodelist, dtype, weight]) Returns adjacency matrix of G.

incidence_matrix(G[, nodelist, edgelist, ...]) Returns incidence matrix of G.
6.1.1 adjacency_matrix
adjacency_matrix(G, nodelist=None, dtype=None, weight='weight')

Returns adjacency matrix of G.
Parameters
G
nodelist
dtype
[NumPy data-type, optional] The desired data-type for the array. If None, then the NumPy
default is used.
weight
[string or None, optional (default=’weight’)] The edge data key used to provide each value in
Returns
A
[SciPy sparse array] Adjacency matrix representation of G.
See also:
to_numpy_array
to_scipy_sparse_array
to_dict_of_dicts
adjacency_spectrum
885
Notes
For directed graphs, entry i,j corresponds to an edge from i to j.

If you want a pure Python adjacency matrix representation try networkx.convert.to_dict_of_dicts which will return
a dictionary-of-dictionaries format that can be addressed as a sparse matrix.
For MultiGraph/MultiDiGraph with parallel edges the weights are summed. See to_numpy_array for other
options.
The convention used for self-loop edges in graphs is to assign the diagonal matrix entry value to the edge weight
attribute (or the number 1 if the edge has no weight attribute). If the alternate convention of doubling the edge
weight is desired the resulting SciPy sparse array can be modified as follows:
>>> G = nx.Graph([(1, 1)])

>>> A = nx.adjacency_matrix(G)
>>> print(A.todense())
[[1]]
>>> A.setdiag(A.diagonal() * 2)
[[2]]
6.1.2 incidence_matrix
incidence_matrix(G, nodelist=None, edgelist=None, oriented=False, weight=None)

Returns incidence matrix of G.
The incidence matrix assigns each row to a node and each column to an edge. For a standard incidence matrix
a 1 appears wherever a row’s node is incident on the column’s edge. For an oriented incidence matrix each edge
is assigned an orientation (arbitrarily for undirected and aligning to direction for directed). A -1 appears for the
source (tail) of an edge and 1 for the destination (head) of the edge. The elements are zero otherwise.
Parameters
G
nodelist
[list, optional (default= all nodes in G)] The rows are ordered according to the nodes in nodelist.
If nodelist is None, then the ordering is produced by G.nodes().
edgelist
[list, optional (default= all edges in G)] The columns are ordered according to the edges in
edgelist. If edgelist is None, then the ordering is produced by G.edges().
oriented: bool, optional (default=False)
If True, matrix elements are +1 or -1 for the head or tail node respectively of each edge. If
False, +1 occurs at both nodes.
weight
[string or None, optional (default=None)] The edge data key used to provide each value in the
matrix. If None, then each edge has weight 1. Edge weights, if used, should be positive so that
the orientation can provide the sign.
Returns
A
[SciPy sparse array] The incidence matrix of G.
886 Chapter 6. Linear algebra

Notes
For MultiGraph/MultiDiGraph, the edges in edgelist should be (u,v,key) 3-tuples.

“Networks are the best discrete model for so many problems in applied mathematics” [1].
References
[1]
6.2 Laplacian Matrix
Laplacian matrix of graphs.
laplacian_matrix(G[, nodelist, weight]) Returns the Laplacian matrix of G.

normalized_laplacian_matrix(G[, nodelist, Returns the normalized Laplacian matrix of G.
...])
directed_laplacian_matrix(G[, nodelist, ...]) Returns the directed Laplacian matrix of G.
directed_combinatorial_laplacian_matrix(G[, Return the directed combinatorial Laplacian matrix of G.
...])
6.2.1 laplacian_matrix
laplacian_matrix(G, nodelist=None, weight='weight')

Returns the Laplacian matrix of G.
The graph Laplacian is the matrix L = D - A, where A is the adjacency matrix and D is the diagonal matrix of node
degrees.
Parameters
G
nodelist
weight
[string or None, optional (default=’weight’)] The edge data key used to compute each value in
Returns
L
[SciPy sparse array] The Laplacian matrix of G.
See also:
to_numpy_array
normalized_laplacian_matrix
laplacian_spectrum
6.2. Laplacian Matrix 887

Notes
For MultiGraph, the edges weights are summed.
Examples
For graphs with multiple connected components, L is permutation-similar to a block diagonal matrix where each
block is the respective Laplacian matrix for each component.
>>> G = nx.Graph([(1, 2), (2, 3), (4, 5)])

>>> print(nx.laplacian_matrix(G).toarray())
[[ 1 -1 0 0 0]
[-1 2 -1 0 0]
[ 0 -1 1 0 0]
[ 0 0 0 1 -1]
[ 0 0 0 -1 1]]
6.2.2 normalized_laplacian_matrix
normalized_laplacian_matrix(G, nodelist=None, weight='weight')

Returns the normalized Laplacian matrix of G.
The normalized graph Laplacian is the matrix
N = D−1/2 LD−1/2
where L is the graph Laplacian and D is the diagonal matrix of node degrees [1].
Parameters
G
nodelist
weight
Returns
N
[SciPy sparse array] The normalized Laplacian matrix of G.
See also:
laplacian_matrix
normalized_laplacian_spectrum

Notes
For MultiGraph, the edges weights are summed. See to_numpy_array() for other options.
If the Graph contains selfloops, D is defined as diag(sum(A, 1)), where A is the adjacency matrix [2].
References
[1], [2]
6.2.3 directed_laplacian_matrix
directed_laplacian_matrix(G, nodelist=None, weight='weight', walk_type=None, alpha=0.95)

Returns the directed Laplacian matrix of G.
The graph directed Laplacian is the matrix
L = I − (Φ1/2 P Φ−1/2 + Φ−1/2 P T Φ1/2 )/2
where I is the identity matrix, P is the transition matrix of the graph, and Phi a matrix with the Perron vector of
P in the diagonal and zeros elsewhere [1].
Depending on the value of walk_type, P can be the transition matrix induced by a random walk, a lazy random
walk, or a random walk with teleportation (PageRank).
Parameters
G
[DiGraph] A NetworkX graph
nodelist
weight
walk_type
[string or None, optional (default=None)] If None, P is selected depending on the properties
of the graph. Otherwise is one of ‘random’, ‘lazy’, or ‘pagerank’
alpha
[real] (1 - alpha) is the teleportation probability used with pagerank
Returns
L
[NumPy matrix] Normalized Laplacian of G.
See also:
laplacian_matrix
6.2. Laplacian Matrix 889

Notes
Only implemented for DiGraphs
References
[1]
6.2.4 directed_combinatorial_laplacian_matrix
directed_combinatorial_laplacian_matrix(G, nodelist=None, weight='weight', walk_type=None,

alpha=0.95)
Return the directed combinatorial Laplacian matrix of G.
The graph directed combinatorial Laplacian is the matrix
L = Φ − (ΦP + P T Φ)/2
where P is the transition matrix of the graph and Phi a matrix with the Perron vector of P in the diagonal and
zeros elsewhere [1].
Depending on the value of walk_type, P can be the transition matrix induced by a random walk, a lazy random
walk, or a random walk with teleportation (PageRank).
Parameters
G
[DiGraph] A NetworkX graph
nodelist
weight
walk_type
[string or None, optional (default=None)] If None, P is selected depending on the properties
of the graph. Otherwise is one of ‘random’, ‘lazy’, or ‘pagerank’
alpha
[real] (1 - alpha) is the teleportation probability used with pagerank
Returns
L
[NumPy matrix] Combinatorial Laplacian of G.
See also:
laplacian_matrix

Notes
Only implemented for DiGraphs
References
[1]
6.3 Bethe Hessian Matrix
Bethe Hessian or deformed Laplacian matrix of graphs.
bethe_hessian_matrix(G[, r, nodelist]) Returns the Bethe Hessian matrix of G.
6.3.1 bethe_hessian_matrix
bethe_hessian_matrix(G, r=None, nodelist=None)

Returns the Bethe Hessian matrix of G.
The Bethe Hessian is a family of matrices parametrized by r, defined as H(r) = (r^2 - 1) I - r A + D where A is the
adjacency matrix, D is the diagonal matrix of node degrees, and I is the identify matrix. It is equal to the graph
laplacian when the regularizer r = 1.
The default choice of regularizer should be the ratio [2]
(∑ )−1 (∑ )
rm = ki ki2 − 1
Parameters
G
[Graph] A NetworkX graph
r
[float] Regularizer parameter
nodelist
Returns
H
[scipy.sparse.csr_array] The Bethe Hessian matrix of G, with parameter r.
See also:
bethe_hessian_spectrum
adjacency_matrix
laplacian_matrix
6.3. Bethe Hessian Matrix 891

References
[1], [2]
Examples
>>> k = [3, 2, 2, 1, 0]
>>> G = nx.havel_hakimi_graph(k)
>>> H = nx.bethe_hessian_matrix(G)
>>> H.toarray()
array([[ 3.5625, -1.25 , -1.25 , -1.25 , 0. ],
[-1.25 , 2.5625, -1.25 , 0. , 0. ],
[-1.25 , -1.25 , 2.5625, 0. , 0. ],
[-1.25 , 0. , 0. , 1.5625, 0. ],
[ 0. , 0. , 0. , 0. , 0.5625]])
6.4 Algebraic Connectivity
Algebraic connectivity and Fiedler vectors of undirected graphs.
algebraic_connectivity(G[, weight, ...]) Returns the algebraic connectivity of an undirected graph.

fiedler_vector(G[, weight, normalized, tol, ...]) Returns the Fiedler vector of a connected undirected
graph.
spectral_ordering(G[, weight, normalized, ...]) Compute the spectral_ordering of a graph.
6.4.1 algebraic_connectivity
algebraic_connectivity(G, weight='weight', normalized=False, tol=1e-08, method='tracemin_pcg',

seed=None)
Returns the algebraic connectivity of an undirected graph.
The algebraic connectivity of a connected undirected graph is the second smallest eigenvalue of its Laplacian matrix.
Parameters
G
weight
[object, optional (default: None)] The data key used to determine the weight of each edge. If
None, then each edge has unit weight.
normalized
[bool, optional (default: False)] Whether the normalized Laplacian matrix is used.
tol
[float, optional (default: 1e-8)] Tolerance of relative residual in eigenvalue computation.
method
[string, optional (default: ‘tracemin_pcg’)] Method of eigenvalue computation. It must be one
of the tracemin options shown below (TraceMIN), ‘lanczos’ (Lanczos iteration) or ‘lobpcg’
(LOBPCG).

The TraceMIN algorithm uses a linear system solver. The following values allow specifying
the solver to be used.
Value Solver
‘tracemin_pcg’ Preconditioned conjugate gradient method
‘tracemin_lu’ LU factorization
seed
Randomness.
Returns
algebraic_connectivity
[float] Algebraic connectivity.
Raises
If G is directed.
NetworkXError
If G has less than two nodes.
See also:
laplacian_matrix
Notes
Edge weights are interpreted by their absolute values. For MultiGraph’s, weights of parallel edges are summed.
Zero-weighted edges are ignored.
Examples
For undirected graphs algebraic connectivity can tell us if a graph is connected or not G is connected iff alge-
braic_connectivity(G) > 0:
>>> nx.algebraic_connectivity(G) > 0
True
>>> G.add_node(10) # G is no longer connected
>>> nx.algebraic_connectivity(G) > 0
False
6.4. Algebraic Connectivity 893

6.4.2 fiedler_vector
fiedler_vector(G, weight='weight', normalized=False, tol=1e-08, method='tracemin_pcg', seed=None)

Returns the Fiedler vector of a connected undirected graph.
The Fiedler vector of a connected undirected graph is the eigenvector corresponding to the second smallest eigen-
value of the Laplacian matrix of the graph.
Parameters
G
weight
normalized
tol
method
(LOBPCG).
Value Solver
seed
Randomness.
Returns
fiedler_vector
[NumPy array of floats.] Fiedler vector.
Raises
If G is directed.
NetworkXError
If G has less than two nodes or is not connected.
See also:
laplacian_matrix

Notes
Examples
Given a connected graph the signs of the values in the Fiedler vector can be used to partition the graph into two
components.
>>> nx.fiedler_vector(G, normalized=True, seed=1)
array([-0.32864129, -0.32864129, -0.32864129, -0.32864129, -0.26072899,
0.26072899, 0.32864129, 0.32864129, 0.32864129, 0.32864129])
The connected components are the two 5-node cliques of the barbell graph.
6.4.3 spectral_ordering
spectral_ordering(G, weight='weight', normalized=False, tol=1e-08, method='tracemin_pcg', seed=None)

Compute the spectral_ordering of a graph.
The spectral ordering of a graph is an ordering of its nodes where nodes in the same weakly connected components
appear contiguous and ordered by their corresponding elements in the Fiedler vector of the component.
Parameters
G
[NetworkX graph] A graph.
weight
normalized
tol
method
(LOBPCG).
Value Solver
seed
Randomness.
6.4. Algebraic Connectivity 895

Returns
spectral_ordering
[NumPy array of floats.] Spectral ordering of nodes.
Raises
NetworkXError
If G is empty.
See also:
laplacian_matrix
Notes
6.5 Attribute Matrices
Functions for constructing matrix-like objects from graph attributes.
attr_matrix(G[, edge_attr, node_attr, ...]) Returns the attribute matrix using attributes from G as a
numpy array.
attr_sparse_matrix(G[, edge_attr, ...]) Returns a SciPy sparse array using attributes from G.
6.5.1 attr_matrix
attr_matrix(G, edge_attr=None, node_attr=None, normalized=False, rc_order=None, dtype=None, order=None)

Returns the attribute matrix using attributes from G as a numpy array.
If only G is passed in, then the adjacency matrix is constructed.
Let A be a discrete set of values for the node attribute node_attr. Then the elements of A represent the rows and
columns of the constructed matrix. Now, iterate through every edge e=(u,v) in G and consider the value of the edge
attribute edge_attr. If ua and va are the values of the node attribute node_attr for u and v, respectively,
then the value of the edge attribute is added to the matrix element at (ua, va).
Parameters
G
[graph] The NetworkX graph used to construct the attribute matrix.
edge_attr
[str, optional] Each element of the matrix represents a running total of the specified edge at-
tribute for edges whose node attributes correspond to the rows/cols of the matrix. The attribute
must be present for all edges in the graph. If no attribute is specified, then we just count the
number of edges whose node attributes correspond to the matrix element.
node_attr
[str, optional] Each row and column in the matrix represents a particular value of the node
attribute. The attribute must be present for all nodes in the graph. Note, the values of this
attribute should be reliably hashable. So, float values are not recommended. If no attribute is
specified, then the rows and columns will be the nodes of the graph.

normalized
[bool, optional] If True, then each row is normalized by the summation of its values.
rc_order
[list, optional] A list of the node attribute values. This list specifies the ordering of rows and
columns of the array. If no ordering is provided, then the ordering will be random (and also,
a return value).
Returns
M
[2D NumPy ndarray] The attribute matrix.
ordering
[list] If rc_order was specified, then only the attribute matrix is returned. However, if
rc_order was None, then the ordering used to construct the matrix is returned as well.
Other Parameters
dtype
[NumPy data-type, optional] A valid NumPy dtype used to initialize the array. Keep in mind
certain dtypes can yield unexpected results if the array is to be normalized. The parameter is
passed to numpy.zeros(). If unspecified, the NumPy default is used.
order
[{‘C’, ‘F’}, optional] Whether to store multidimensional data in C- or Fortran-contiguous (row-
or column-wise) order in memory. This parameter is passed to numpy.zeros(). If unspecified,
Examples
Construct an adjacency matrix:
>>> G = nx.Graph()
>>> G.add_edge(0, 1, thickness=1, weight=3)
>>> G.add_edge(0, 2, thickness=2)
>>> nx.attr_matrix(G, rc_order=[0, 1, 2])
array([[0., 1., 1.],
[1., 0., 1.],
[1., 1., 0.]])
Alternatively, we can obtain the matrix describing edge thickness.
>>> nx.attr_matrix(G, edge_attr="thickness", rc_order=[0, 1, 2])

array([[0., 1., 2.],
[1., 0., 3.],
[2., 3., 0.]])
We can also color the nodes and ask for the probability distribution over all edges (u,v) describing:
Pr(v has color Y | u has color X)
>>> G.nodes[0]["color"] = "red"

>>> G.nodes[2]["color"] = "blue"
>>> rc = ["red", "blue"]
>>> nx.attr_matrix(G, node_attr="color", normalized=True, rc_order=rc)
6.5. Attribute Matrices 897


array([[0.33333333, 0.66666667],
[1. , 0. ]])
For example, the above tells us that for all edges (u,v):
Pr( v is red | u is red) = 1/3 Pr( v is blue | u is red) = 2/3
Pr( v is red | u is blue) = 1 Pr( v is blue | u is blue) = 0
Finally, we can obtain the total weights listed by the node colors.
>>> nx.attr_matrix(G, edge_attr="weight", node_attr="color", rc_order=rc)

array([[3., 2.],
[2., 0.]])
Thus, the total weight over all edges (u,v) with u and v having colors:
(red, red) is 3 # the sole contribution is from edge (0,1) (red, blue) is 2 # contributions from edges (0,2)
and (1,2) (blue, red) is 2 # same as (red, blue) since graph is undirected (blue, blue) is 0 # there are no
edges with blue endpoints
6.5.2 attr_sparse_matrix
attr_sparse_matrix(G, edge_attr=None, node_attr=None, normalized=False, rc_order=None, dtype=None)

Returns a SciPy sparse array using attributes from G.
If only G is passed in, then the adjacency matrix is constructed.
Let A be a discrete set of values for the node attribute node_attr. Then the elements of A represent the rows and
columns of the constructed matrix. Now, iterate through every edge e=(u,v) in G and consider the value of the edge
attribute edge_attr. If ua and va are the values of the node attribute node_attr for u and v, respectively,
then the value of the edge attribute is added to the matrix element at (ua, va).
Parameters
G
[graph] The NetworkX graph used to construct the NumPy matrix.
edge_attr
[str, optional] Each element of the matrix represents a running total of the specified edge at-
tribute for edges whose node attributes correspond to the rows/cols of the matirx. The attribute
must be present for all edges in the graph. If no attribute is specified, then we just count the
number of edges whose node attributes correspond to the matrix element.
node_attr
[str, optional] Each row and column in the matrix represents a particular value of the node
attribute. The attribute must be present for all nodes in the graph. Note, the values of this
attribute should be reliably hashable. So, float values are not recommended. If no attribute is
specified, then the rows and columns will be the nodes of the graph.
normalized
[bool, optional] If True, then each row is normalized by the summation of its values.
rc_order
[list, optional] A list of the node attribute values. This list specifies the ordering of rows and
columns of the array. If no ordering is provided, then the ordering will be random (and also,
a return value).

Returns
M
[SciPy sparse array] The attribute matrix.
ordering
[list] If rc_order was specified, then only the matrix is returned. However, if rc_order
was None, then the ordering used to construct the matrix is returned as well.
Other Parameters
dtype
[NumPy data-type, optional] A valid NumPy dtype used to initialize the array. Keep in mind
certain dtypes can yield unexpected results if the array is to be normalized. The parameter is
passed to numpy.zeros(). If unspecified, the NumPy default is used.
Examples
Construct an adjacency matrix:
>>> G = nx.Graph()
>>> G.add_edge(0, 1, thickness=1, weight=3)
>>> M = nx.attr_sparse_matrix(G, rc_order=[0, 1, 2])
>>> M.toarray()
array([[0., 1., 1.],
[1., 0., 1.],
[1., 1., 0.]])
Alternatively, we can obtain the matrix describing edge thickness.
>>> M = nx.attr_sparse_matrix(G, edge_attr="thickness", rc_order=[0, 1, 2])

>>> M.toarray()
array([[0., 1., 2.],
[1., 0., 3.],
[2., 3., 0.]])
We can also color the nodes and ask for the probability distribution over all edges (u,v) describing:
Pr(v has color Y | u has color X)

>>> G.nodes[2]["color"] = "blue"
>>> rc = ["red", "blue"]
>>> M = nx.attr_sparse_matrix(G, node_attr="color", normalized=True, rc_order=rc)
>>> M.toarray()
array([[0.33333333, 0.66666667],
[1. , 0. ]])
For example, the above tells us that for all edges (u,v):
Pr( v is red | u is red) = 1/3 Pr( v is blue | u is red) = 2/3
Pr( v is red | u is blue) = 1 Pr( v is blue | u is blue) = 0
Finally, we can obtain the total weights listed by the node colors.
6.5. Attribute Matrices 899

>>> M = nx.attr_sparse_matrix(G, edge_attr="weight", node_attr="color", rc_

,→order=rc)
>>> M.toarray()
array([[3., 2.],
[2., 0.]])
Thus, the total weight over all edges (u,v) with u and v having colors:
(red, red) is 3 # the sole contribution is from edge (0,1) (red, blue) is 2 # contributions from edges (0,2)
and (1,2) (blue, red) is 2 # same as (red, blue) since graph is undirected (blue, blue) is 0 # there are no
edges with blue endpoints
6.6 Modularity Matrices
Modularity matrix of graphs.
modularity_matrix(G[, nodelist, weight]) Returns the modularity matrix of G.

directed_modularity_matrix(G[, nodelist, Returns the directed modularity matrix of G.
weight])
6.6.1 modularity_matrix
modularity_matrix(G, nodelist=None, weight=None)

Returns the modularity matrix of G.
The modularity matrix is the matrix B = A - <A>, where A is the adjacency matrix and <A> is the average adjacency
matrix, assuming that the graph is described by the configuration model.
More specifically, the element B_ij of B is defined as
ki kj
Aij −
2m
where k_i is the degree of node i, and where m is the number of edges in the graph. When weight is set to a name
of an attribute edge, Aij, k_i, k_j and m are computed using its value.
Parameters
G
[Graph] A NetworkX graph
nodelist
weight
used for the edge weight. If None then all edge weights are 1.
Returns
B
[Numpy array] The modularity matrix of G.
See also:

to_numpy_array
modularity_spectrum
adjacency_matrix
directed_modularity_matrix
References
[1]
Examples
>>> k = [3, 2, 2, 1, 0]
>>> G = nx.havel_hakimi_graph(k)
>>> B = nx.modularity_matrix(G)
6.6.2 directed_modularity_matrix
directed_modularity_matrix(G, nodelist=None, weight=None)

Returns the directed modularity matrix of G.
The modularity matrix is the matrix B = A - <A>, where A is the adjacency matrix and <A> is the expected
adjacency matrix, assuming that the graph is described by the configuration model.
More specifically, the element B_ij of B is defined as
Bij = Aij − kiout kjin /m
where kiin is the in degree of node i, and kjout is the out degree of node j, with m the number of edges in the graph.
When weight is set to a name of an attribute edge, Aij, k_i, k_j and m are computed using its value.
Parameters
G
[DiGraph] A NetworkX DiGraph
nodelist
weight
Returns
B
[Numpy array] The modularity matrix of G.
See also:
to_numpy_array
modularity_spectrum
adjacency_matrix
modularity_matrix
6.6. Modularity Matrices 901

Notes
NetworkX defines the element A_ij of the adjacency matrix as 1 if there is a link going from node i to node j.
Leicht and Newman use the opposite definition. This explains the different expression for B_ij.
References
[1]
Examples
... (
... (1, 2),
... (1, 3),
... (3, 1),
... (3, 2),
... (3, 5),
... (4, 5),
... (4, 6),
... (5, 4),
... (5, 6),
... (6, 4),
... )
... )
>>> B = nx.directed_modularity_matrix(G)
6.7 Spectrum
Eigenvalue spectrum of graphs.
adjacency_spectrum(G[, weight]) Returns eigenvalues of the adjacency matrix of G.

laplacian_spectrum(G[, weight]) Returns eigenvalues of the Laplacian of G
bethe_hessian_spectrum(G[, r]) Returns eigenvalues of the Bethe Hessian matrix of G.
normalized_laplacian_spectrum(G[, Return eigenvalues of the normalized Laplacian of G
weight])
modularity_spectrum(G) Returns eigenvalues of the modularity matrix of G.
6.7.1 adjacency_spectrum
adjacency_spectrum(G, weight='weight')
Returns eigenvalues of the adjacency matrix of G.
Parameters
G

weight
Returns
evals
[NumPy array] Eigenvalues
See also:
adjacency_matrix
Notes
For MultiGraph/MultiDiGraph, the edges weights are summed. See to_numpy_array for other options.
6.7.2 laplacian_spectrum
laplacian_spectrum(G, weight='weight')
Returns eigenvalues of the Laplacian of G
Parameters
G
weight
Returns
evals
See also:
laplacian_matrix
Notes
For MultiGraph/MultiDiGraph, the edges weights are summed. See to_numpy_array() for other options.
Examples
The multiplicity of 0 as an eigenvalue of the laplacian matrix is equal to the number of connected components of
G.
>>> G = nx.Graph() # Create a graph with 5 nodes and 3 connected components

>>> G.add_nodes_from(range(5))
>>> G.add_edges_from([(0, 2), (3, 4)])
>>> nx.laplacian_spectrum(G)
array([0., 0., 0., 2., 2.])
6.7. Spectrum 903

6.7.3 bethe_hessian_spectrum
bethe_hessian_spectrum(G, r=None)
Returns eigenvalues of the Bethe Hessian matrix of G.
Parameters
G
[Graph] A NetworkX Graph or DiGraph
r
[float] Regularizer parameter
Returns
evals
See also:
bethe_hessian_matrix
References
[1]
6.7.4 normalized_laplacian_spectrum
normalized_laplacian_spectrum(G, weight='weight')
Return eigenvalues of the normalized Laplacian of G
Parameters
G
weight
Returns
evals
See also:
normalized_laplacian_matrix

Notes
For MultiGraph/MultiDiGraph, the edges weights are summed. See to_numpy_array for other options.
6.7.5 modularity_spectrum
modularity_spectrum(G)
Returns eigenvalues of the modularity matrix of G.
Parameters
G
[Graph] A NetworkX Graph or DiGraph
Returns
evals
See also:
modularity_matrix
References
[1]
6.7. Spectrum 905


CHAPTER
SEVEN
CONVERTING TO AND FROM OTHER DATA FORMATS
7.1 To NetworkX Graph
Functions to convert NetworkX graphs to and from other formats.

The preferred way of converting data to a NetworkX graph is through the graph constructor. The constructor calls the
to_networkx_graph() function which attempts to guess the input type and convert it automatically.
7.1.1 Examples
Create a graph with a single edge from a dictionary of dictionaries
>>> d = {0: {1: 1}} # dict-of-dicts single edge (0,1)

>>> G = nx.Graph(d)
7.1.2 See Also
nx_agraph, nx_pydot
to_networkx_graph(data[, create_using, ...]) Make a NetworkX graph from a known data structure.
7.1.3 to_networkx_graph
to_networkx_graph(data, create_using=None, multigraph_input=False)

Make a NetworkX graph from a known data structure.
The preferred way to call this is automatically from the class constructor
>>> d = {0: {1: {"weight": 1}}} # dict-of-dicts single edge (0,1)

>>> G = nx.Graph(d)
instead of the equivalent
>>> G = nx.from_dict_of_dicts(d)
Parameters
data
[object to be converted]
907
Current known types are:

any NetworkX graph dict-of-dicts dict-of-lists container (e.g. set, list, tuple) of edges iter-
ator (e.g. itertools.chain) that produces edges generator of edges Pandas DataFrame (row
per edge) 2D numpy array scipy sparse array pygraphviz agraph
create_using
multigraph_input
[bool (default False)] If True and data is a dict_of_dicts, try to create a multigraph assuming
dict_of_dict_of_lists. If data and create_using are both multigraphs then create a multigraph
from a multigraph.
7.2 Dictionaries
to_dict_of_dicts(G[, nodelist, edge_data]) Returns adjacency representation of graph as a dictionary

of dictionaries.
from_dict_of_dicts(d[, create_using, ...]) Returns a graph from a dictionary of dictionaries.
7.2.1 to_dict_of_dicts
to_dict_of_dicts(G, nodelist=None, edge_data=None)

Returns adjacency representation of graph as a dictionary of dictionaries.
Parameters
G
nodelist
[list] Use only nodes specified in nodelist
edge_data
[scalar, optional] If provided, the value of the dictionary will be set to edge_data for all
edges. Usual values could be 1 or True. If edge_data is None (the default), the edge-
data in G is used, resulting in a dict-of-dict-of-dicts. If G is a MultiGraph, the result will be
a dict-of-dict-of-dict-of-dicts. See Notes for an approach to customize handling edge data.
edge_data should not be a container.
Returns
dod
[dict] A nested dictionary representation of G. Note that the level of nesting depends on the
type of G and the value of edge_data (see Examples).
See also:
from_dict_of_dicts, to_dict_of_lists
908 Chapter 7. Converting to and from other data formats

Notes
For a more custom approach to handling edge data, try:

dod = {
n: {
nbr: custom(n, nbr, dd) for nbr, dd in nbrdict.items()
}
for n, nbrdict in G.adj.items()
}
where custom returns the desired edge data for each edge between n and nbr, given existing edge data dd.
Examples
>>> nx.to_dict_of_dicts(G)
{0: {1: {}}, 1: {0: {}, 2: {}}, 2: {1: {}}}
Edge data is preserved by default (edge_data=None), resulting in dict-of-dict-of-dicts where the innermost
dictionary contains the edge data:
>>> G = nx.Graph()
... [
... (0, 1, {'weight': 1.0}),
... (1, 2, {'weight': 2.0}),
... (2, 0, {'weight': 1.0}),
... ]
... )
>>> d = nx.to_dict_of_dicts(G)
>>> d
{0: {1: {'weight': 1.0}, 2: {'weight': 1.0}},
1: {0: {'weight': 1.0}, 2: {'weight': 2.0}},
2: {1: {'weight': 2.0}, 0: {'weight': 1.0}}}
>>> d[1][2]['weight']
2.0
If edge_data is not None, edge data in the original graph (if any) is replaced:
>>> d = nx.to_dict_of_dicts(G, edge_data=1)
>>> d
{0: {1: 1, 2: 1}, 1: {0: 1, 2: 1}, 2: {1: 1, 0: 1}}
>>> d[1][2]
1
This also applies to MultiGraphs: edge data is preserved by default:

>>> G.add_edge(0, 1, key='a', weight=1.0)
'a'
>>> G.add_edge(0, 1, key='b', weight=5.0)
'b'
>>> d = nx.to_dict_of_dicts(G)
>>> d
{0: {1: {'a': {'weight': 1.0}, 'b': {'weight': 5.0}}},
7.2. Dictionaries 909


1: {0: {'a': {'weight': 1.0}, 'b': {'weight': 5.0}}}}
>>> d[0][1]['b']['weight']
5.0
But multi edge data is lost if edge_data is not None:
>>> d = nx.to_dict_of_dicts(G, edge_data=10)

>>> d
{0: {1: 10}, 1: {0: 10}}
7.2.2 from_dict_of_dicts
from_dict_of_dicts(d, create_using=None, multigraph_input=False)

Returns a graph from a dictionary of dictionaries.
Parameters
d
[dictionary of dictionaries] A dictionary of dictionaries adjacency representation.
create_using
multigraph_input
[bool (default False)] When True, the dict d is assumed to be a dict-of-dict-of-dict-of-dict
structure keyed by node to neighbor to edge keys to edge data for multi-edges. Otherwise this
routine assumes dict-of-dict-of-dict keyed by node to neighbor to edge data.
Examples
>>> dod = {0: {1: {"weight": 1}}} # single edge (0,1)

>>> G = nx.from_dict_of_dicts(dod)
or
>>> G = nx.Graph(dod) # use Graph constructor
7.3 Lists
to_dict_of_lists(G[, nodelist]) Returns adjacency representation of graph as a dictionary

of lists.
from_dict_of_lists(d[, create_using]) Returns a graph from a dictionary of lists.
to_edgelist(G[, nodelist]) Returns a list of edges in the graph.
from_edgelist(edgelist[, create_using]) Returns a graph from a list of edges.

7.3.1 to_dict_of_lists
to_dict_of_lists(G, nodelist=None)
Returns adjacency representation of graph as a dictionary of lists.
Parameters
G
nodelist
Notes
Completely ignores edge data for MultiGraph and MultiDiGraph.
7.3.2 from_dict_of_lists
from_dict_of_lists(d, create_using=None)
Returns a graph from a dictionary of lists.
Parameters
d
[dictionary of lists] A dictionary of lists adjacency representation.
create_using
Examples
>>> dol = {0: [1]} # single edge (0,1)

>>> G = nx.from_dict_of_lists(dol)
or
>>> G = nx.Graph(dol) # use Graph constructor
7.3.3 to_edgelist
to_edgelist(G, nodelist=None)
Returns a list of edges in the graph.
Parameters
G
nodelist
7.3. Lists 911

7.3.4 from_edgelist
from_edgelist(edgelist, create_using=None)
Returns a graph from a list of edges.
Parameters
edgelist
[list or iterator] Edge tuples
create_using
Examples
>>> edgelist = [(0, 1)] # single edge (0,1)

>>> G = nx.from_edgelist(edgelist)
or
>>> G = nx.Graph(edgelist) # use Graph constructor
7.4 Numpy
Functions to convert NetworkX graphs to and from common data containers like numpy arrays, scipy sparse arrays, and
pandas DataFrames.
The preferred way of converting data to a NetworkX graph is through the graph constructor. The constructor calls the
to_networkx_graph function which attempts to guess the input type and convert it automatically.
7.4.1 Examples
Create a 10 node random graph from a numpy array

>>> rng = np.random.default_rng()
>>> a = rng.integers(low=0, high=2, size=(10, 10))
>>> DG = nx.from_numpy_array(a, create_using=nx.DiGraph)
or equivalently:
>>> DG = nx.DiGraph(a)
which calls from_numpy_array internally based on the type of a.

7.4.2 See Also
nx_agraph, nx_pydot
to_numpy_array(G[, nodelist, dtype, order, ...]) Returns the graph adjacency matrix as a NumPy array.
from_numpy_array(A[, parallel_edges, ...]) Returns a graph from a 2D NumPy array.
7.4.3 to_numpy_array
to_numpy_array(G, nodelist=None, dtype=None, order=None, multigraph_weight=<built-in function sum>,

weight='weight', nonedge=0.0)
Returns the graph adjacency matrix as a NumPy array.
Parameters
G
[graph] The NetworkX graph used to construct the NumPy array.
nodelist
[list, optional] The rows and columns are ordered according to the nodes in nodelist. If
nodelist is None, then the ordering is produced by G.nodes().
dtype
[NumPy data type, optional] A NumPy data type used to initialize the array. If None, then
the NumPy default is used. The dtype can be structured if weight=None, in which case the
dtype field names are used to look up edge attributes. The result is a structured array where
each named field in the dtype corresponds to the adjaceny for that edge attribute. See examples
for details.
order
[{‘C’, ‘F’}, optional] Whether to store multidimensional data in C- or Fortran-contiguous (row-
or column-wise) order in memory. If None, then the NumPy default is used.
multigraph_weight
[callable, optional] An function that determines how weights in multigraphs are handled. The
function should accept a sequence of weights and return a single value. The default is to sum
the weights of the multiple edges.
weight
[string or None optional (default = ‘weight’)] The edge attribute that holds the numerical value
used for the edge weight. If an edge does not have that attribute, then the value 1 is used
instead. weight must be None if a structured dtype is used.
nonedge
[array_like (default = 0.0)] The value used to represent non-edges in the adjaceny matrix.
The array values corresponding to nonedges are typically set to zero. However, this could be
undesirable if there are array values corresponding to actual edges that also have the value zero.
If so, one might prefer nonedges to have some other value, such as nan.
Returns
A
[NumPy ndarray] Graph adjacency matrix
Raises
NetworkXError
If dtype is a structured dtype and G is a multigraph
7.4. Numpy 913

ValueError
If dtype is a structured dtype and weight is not None
See also:
from_numpy_array
Notes
For directed graphs, entry i, j corresponds to an edge from i to j.

Entries in the adjacency matrix are given by the weight edge attribute. When an edge does not have a weight
attribute, the value of the entry is set to the number 1. For multiple (parallel) edges, the values of the entries are
determined by the multigraph_weight parameter. The default is to sum the weight attributes for each of the
parallel edges.
When nodelist does not contain every node in G, the adjacency matrix is built from the subgraph of G that is
induced by the nodes in nodelist.
The convention used for self-loop edges in graphs is to assign the diagonal array entry value to the weight attribute
of the edge (or the number 1 if the edge has no weight attribute). If the alternate convention of doubling the edge
weight is desired the resulting NumPy array can be modified as follows:

>>> G = nx.Graph([(1, 1)])
>>> A = nx.to_numpy_array(G)
>>> A
array([[1.]])
>>> A[np.diag_indices_from(A)] *= 2
>>> A
array([[2.]])
Examples
0
0
0
1
>>> nx.to_numpy_array(G, nodelist=[0, 1, 2])
array([[0., 2., 0.],
[1., 0., 0.],
[0., 0., 4.]])
When nodelist argument is used, nodes of G which do not appear in the nodelist and their edges are not
included in the adjacency matrix. Here is an example:
>>> G = nx.Graph()


>>> nx.to_numpy_array(G, nodelist=[1, 2, 3])
array([[0., 1., 1.],
[1., 0., 0.],
[1., 0., 0.]])
This function can also be used to create adjacency matrices for multiple edge attributes with structured dtypes:
>>> G = nx.Graph()
>>> G.add_edge(1, 2, cost=5)
>>> G.add_edge(2, 3, weight=3, cost=-4.0)
>>> dtype = np.dtype([("weight", int), ("cost", float)])
>>> A = nx.to_numpy_array(G, dtype=dtype, weight=None)
>>> A["weight"]
array([[ 0, 10, 0, 0],
[10, 0, 1, 0],
[ 0, 1, 0, 3],
[ 0, 0, 3, 0]])
>>> A["cost"]
array([[ 0., 1., 0., 0.],
[ 1., 0., 5., 0.],
[ 0., 5., 0., -4.],
[ 0., 0., -4., 0.]])
As stated above, the argument “nonedge” is useful especially when there are actually edges with weight 0 in the
graph. Setting a nonedge value different than 0, makes it much clearer to differentiate such 0-weighted edges and
actual nonedge values.
>>> G = nx.Graph()
>>> nx.to_numpy_array(G, nonedge=-1.)
array([[-1., 2., -1., 1.],
[ 2., -1., 0., -1.],
[-1., 0., -1., 0.],
[ 1., -1., 0., -1.]])
7.4.4 from_numpy_array
from_numpy_array(A, parallel_edges=False, create_using=None)

Returns a graph from a 2D NumPy array.
The 2D NumPy array is interpreted as an adjacency matrix for the graph.
Parameters
A
[a 2D numpy.ndarray] An adjacency matrix representation of a graph
parallel_edges
[Boolean] If this is True, create_using is a multigraph, and A is an integer array, then
entry (i, j) in the array is interpreted as the number of parallel edges joining vertices i and j in
7.4. Numpy 915

the graph. If it is False, then the entries in the array are interpreted as the weight of a single
edge joining the vertices.
create_using
See also:
to_numpy_array
Notes
For directed graphs, explicitly mention create_using=nx.DiGraph, and entry i,j of A corresponds to an edge from
i to j.
If create_using is networkx.MultiGraph or networkx.MultiDiGraph, parallel_edges is
True, and the entries of A are of type int, then this function returns a multigraph (of the same type as cre-
ate_using) with parallel edges.
If create_using indicates an undirected multigraph, then only the edges indicated by the upper triangle of the
array A will be added to the graph.
If the NumPy array has a single data type for each array entry it will be converted to an appropriate Python data
type.
If the NumPy array has a user-specified compound data type the names of the data fields will be used as attribute
keys in the resulting NetworkX graph.
Examples
Simple integer weights on edges:

>>> A = np.array([[1, 1], [2, 1]])
>>> G = nx.from_numpy_array(A)
EdgeDataView([(0, 0, {'weight': 1}), (0, 1, {'weight': 2}), (1, 1, {'weight': 1}
,→)])
If create_using indicates a multigraph and the array has only integer entries and parallel_edges is False,
then the entries will be treated as weights for edges joining the nodes (without creating parallel edges):
>>> A = np.array([[1, 1], [1, 2]])

>>> G = nx.from_numpy_array(A, create_using=nx.MultiGraph)
>>> G[1][1]
AtlasView({0: {'weight': 2}})
If create_using indicates a multigraph and the array has only integer entries and parallel_edges is True,
then the entries will be treated as the number of parallel edges joining those two vertices:
>>> A = np.array([[1, 1], [1, 2]])

>>> temp = nx.MultiGraph()
>>> G = nx.from_numpy_array(A, parallel_edges=True, create_using=temp)
>>> G[1][1]
AtlasView({0: {'weight': 1}, 1: {'weight': 1}})

User defined compound data type on edges:
>>> dt = [("weight", float), ("cost", int)]

>>> A = np.array([[(1.0, 2)]], dtype=dt)
>>> G = nx.from_numpy_array(A)
>>> G.edges()
EdgeView([(0, 0)])
>>> G[0][0]["cost"]
2
>>> G[0][0]["weight"]
1.0
7.5 Scipy
to_scipy_sparse_array(G[, nodelist, dtype, ...]) Returns the graph adjacency matrix as a SciPy sparse ar-
ray.
from_scipy_sparse_array(A[, parallel_edges, Creates a new graph from an adjacency matrix given as a
...]) SciPy sparse array.
7.5.1 to_scipy_sparse_array
to_scipy_sparse_array(G, nodelist=None, dtype=None, weight='weight', format='csr')

Returns the graph adjacency matrix as a SciPy sparse array.
Parameters
G
[graph] The NetworkX graph used to construct the sparse matrix.
nodelist
dtype
[NumPy data-type, optional] A valid NumPy dtype used to initialize the array. If None, then
weight
[string or None optional (default=’weight’)] The edge attribute that holds the numerical value
format
[str in {‘bsr’, ‘csr’, ‘csc’, ‘coo’, ‘lil’, ‘dia’, ‘dok’}] The type of the matrix to be returned (default
‘csr’). For some algorithms different implementations of sparse matrices can perform better.
See [1] for details.
Returns
A
[SciPy sparse array] Graph adjacency matrix.
7.5. Scipy 917

Notes
For directed graphs, matrix entry i,j corresponds to an edge from i to j.

The matrix entries are populated using the edge attribute held in parameter weight. When an edge does not have
that attribute, the value of the entry is 1.
For multiple edges the matrix values are the sums of the edge weights.
When nodelist does not contain every node in G, the adjacency matrix is built from the subgraph of G that is
induced by the nodes in nodelist.
The convention used for self-loop edges in graphs is to assign the diagonal matrix entry value to the weight attribute
weight is desired the resulting SciPy sparse array can be modified as follows:
>>> G = nx.Graph([(1, 1)])

>>> A = nx.to_scipy_sparse_array(G)
[[1]]
>>> A.setdiag(A.diagonal() * 2)
>>> print(A.toarray())
[[2]]
References
[1]
Examples
0
0
0
1
>>> S = nx.to_scipy_sparse_array(G, nodelist=[0, 1, 2])
>>> print(S.toarray())
[[0 2 0]
[1 0 0]
[0 0 4]]

7.5.2 from_scipy_sparse_array
from_scipy_sparse_array(A, parallel_edges=False, create_using=None, edge_attribute='weight')

Creates a new graph from an adjacency matrix given as a SciPy sparse array.
Parameters
A: scipy.sparse array
An adjacency matrix representation of a graph
parallel_edges
[Boolean] If this is True, create_using is a multigraph, and A is an integer matrix, then
entry (i, j) in the matrix is interpreted as the number of parallel edges joining vertices i and j in
the graph. If it is False, then the entries in the matrix are interpreted as the weight of a single
edge joining the vertices.
create_using
edge_attribute: string
Name of edge attribute to store matrix numeric value. The data will have the same type as the
matrix entry (int, float, (real,imag)).
Notes
For directed graphs, explicitly mention create_using=nx.DiGraph, and entry i,j of A corresponds to an edge from
i to j.
If create_using is networkx.MultiGraph or networkx.MultiDiGraph, parallel_edges is
True, and the entries of A are of type int, then this function returns a multigraph (constructed from cre-
ate_using) with parallel edges. In this case, edge_attribute will be ignored.
If create_using indicates an undirected multigraph, then only the edges indicated by the upper triangle of the
matrix A will be added to the graph.
Examples
>>> import scipy as sp

>>> import scipy.sparse # call as sp.sparse
>>> A = sp.sparse.eye(2, 2, 1)
>>> G = nx.from_scipy_sparse_array(A)
If create_using indicates a multigraph and the matrix has only integer entries and parallel_edges is
False, then the entries will be treated as weights for edges joining the nodes (without creating parallel edges):
>>> A = sp.sparse.csr_array([[1, 1], [1, 2]])

>>> G = nx.from_scipy_sparse_array(A, create_using=nx.MultiGraph)
>>> G[1][1]
AtlasView({0: {'weight': 2}})
If create_using indicates a multigraph and the matrix has only integer entries and parallel_edges is
True, then the entries will be treated as the number of parallel edges joining those two vertices:
7.5. Scipy 919

>>> A = sp.sparse.csr_array([[1, 1], [1, 2]])

>>> G = nx.from_scipy_sparse_array(
... A, parallel_edges=True, create_using=nx.MultiGraph
... )
>>> G[1][1]
AtlasView({0: {'weight': 1}, 1: {'weight': 1}})
7.6 Pandas
to_pandas_adjacency(G[, nodelist, dtype, ...]) Returns the graph adjacency matrix as a Pandas
DataFrame.
from_pandas_adjacency(df[, create_using]) Returns a graph from Pandas DataFrame.
to_pandas_edgelist(G[, source, target, ...]) Returns the graph edge list as a Pandas DataFrame.
from_pandas_edgelist(df[, source, target, ...]) Returns a graph from Pandas DataFrame containing an
edge list.
7.6.1 to_pandas_adjacency
to_pandas_adjacency(G, nodelist=None, dtype=None, order=None, multigraph_weight=<built-in function

sum>, weight='weight', nonedge=0.0)
Returns the graph adjacency matrix as a Pandas DataFrame.
Parameters
G
[graph] The NetworkX graph used to construct the Pandas DataFrame.
nodelist
multigraph_weight
[{sum, min, max}, optional] An operator that determines how weights in multigraphs are han-
dled. The default is to sum the weights of the multiple edges.
weight
[string or None, optional] The edge attribute that holds the numerical value used for the edge
weight. If an edge does not have that attribute, then the value 1 is used instead.
nonedge
[float, optional] The matrix values corresponding to nonedges are typically set to zero. How-
ever, this could be undesirable if there are matrix values corresponding to actual edges that
also have the value zero. If so, one might prefer nonedges to have some other value, such as
nan.
Returns
df
[Pandas DataFrame] Graph adjacency matrix

Notes
For directed graphs, entry i,j corresponds to an edge from i to j.

The DataFrame entries are assigned to the weight edge attribute. When an edge does not have a weight attribute,
the value of the entry is set to the number 1. For multiple (parallel) edges, the values of the entries are determined
by the ‘multigraph_weight’ parameter. The default is to sum the weight attributes for each of the parallel edges.
When nodelist does not contain every node in G, the matrix is built from the subgraph of G that is induced by
the nodes in nodelist.
The convention used for self-loop edges in graphs is to assign the diagonal matrix entry value to the weight attribute
weight is desired the resulting Pandas DataFrame can be modified as follows:
>>> import pandas as pd

>>> pd.options.display.max_columns = 20
>>> G = nx.Graph([(1, 1)])
>>> df = nx.to_pandas_adjacency(G, dtype=int)
>>> df
1
1 1
>>> df.values[np.diag_indices_from(df)] *= 2
>>> df
1
1 2
Examples
0
0
0
1
>>> nx.to_pandas_adjacency(G, nodelist=[0, 1, 2], dtype=int)
0 1 2
0 0 2 0
1 1 0 0
2 0 0 4
7.6.2 from_pandas_adjacency
from_pandas_adjacency(df, create_using=None)
Returns a graph from Pandas DataFrame.
The Pandas DataFrame is interpreted as an adjacency matrix for the graph.
Parameters
df
[Pandas DataFrame] An adjacency matrix representation of a graph
7.6. Pandas 921

create_using
See also:
to_pandas_adjacency
Notes
For directed graphs, explicitly mention create_using=nx.DiGraph, and entry i,j of df corresponds to an edge from
i to j.
If df has a single data type for each entry it will be converted to an appropriate Python data type.
If df has a user-specified compound data type the names of the data fields will be used as attribute keys in the
resulting NetworkX graph.
Examples

>>> df = pd.DataFrame([[1, 1], [2, 1]])
>>> df
0 1
0 1 1
1 2 1
>>> G = nx.from_pandas_adjacency(df)
>>> G.name = "Graph from pandas adjacency matrix"
>>> print(G)
Graph named 'Graph from pandas adjacency matrix' with 2 nodes and 3 edges
7.6.3 to_pandas_edgelist
to_pandas_edgelist(G, source='source', target='target', nodelist=None, dtype=None, edge_key=None)

Returns the graph edge list as a Pandas DataFrame.
Parameters
G
[graph] The NetworkX graph used to construct the Pandas DataFrame.
source
[str or int, optional] A valid column name (string or integer) for the source nodes (for the
directed case).
target
[str or int, optional] A valid column name (string or integer) for the target nodes (for the
directed case).
nodelist
[list, optional] Use only nodes specified in nodelist

dtype
[dtype, default None] Use to create the DataFrame. Data type to force. Only a single dtype is
allowed. If None, infer.
edge_key
[str or int or None, optional (default=None)] A valid column name (string or integer) for the
edge keys (for the multigraph case). If None, edge keys are not stored in the DataFrame.
Returns
df
[Pandas DataFrame] Graph edge list
Examples
>>> G = nx.Graph(
... [
... ("A", "B", {"cost": 1, "weight": 7}),
... ("C", "E", {"cost": 9, "weight": 10}),
... ]
... )
>>> df = nx.to_pandas_edgelist(G, nodelist=["A", "C"])
>>> df[["source", "target", "cost", "weight"]]
source target cost weight
0 A B 1 7
1 C E 9 10
>>> G = nx.MultiGraph([('A', 'B', {'cost': 1}), ('A', 'B', {'cost': 9})])

>>> df = nx.to_pandas_edgelist(G, nodelist=['A', 'C'], edge_key='ekey')
>>> df[['source', 'target', 'cost', 'ekey']]
source target cost ekey
0 A B 1 0
1 A B 9 1
7.6.4 from_pandas_edgelist
from_pandas_edgelist(df, source='source', target='target', edge_attr=None, create_using=None,

edge_key=None)
Returns a graph from Pandas DataFrame containing an edge list.
The Pandas DataFrame should contain at least two columns of node names and zero or more columns of edge
attributes. Each row will be processed as one edge instance.
Note: This function iterates over DataFrame.values, which is not guaranteed to retain the data type across columns
in the row. This is only a problem if your row is entirely numeric and a mix of ints and floats. In that case, all
values will be returned as floats. See the DataFrame.iterrows documentation for an example.
Parameters
df
[Pandas DataFrame] An edge list representation of a graph
source
[str or int] A valid column name (string or integer) for the source nodes (for the directed case).
target
[str or int] A valid column name (string or integer) for the target nodes (for the directed case).
7.6. Pandas 923

edge_attr
[str or int, iterable, True, or None] A valid column name (str or int) or iterable of column
names that are used to retrieve items and add them to the graph as edge attributes. If True,
all of the remaining columns will be added. If None, no edge attributes are added to the graph.
create_using
edge_key
[str or None, optional (default=None)] A valid column name for the edge keys (for a Multi-
Graph). The values in this column are used for the edge keys when adding edges if create_using
is a multigraph.
See also:
to_pandas_edgelist
Examples

>>> rng = np.random.RandomState(seed=5)
>>> ints = rng.randint(1, 11, size=(3, 2))
>>> a = ["A", "B", "C"]
>>> b = ["D", "A", "E"]
>>> df = pd.DataFrame(ints, columns=["weight", "cost"])
>>> df[0] = a
>>> df["b"] = b
>>> df[["weight", "cost", 0, "b"]]
weight cost 0 b
0 4 7 A D
1 7 1 B A
2 10 9 C E
>>> G = nx.from_pandas_edgelist(df, 0, "b", ["weight", "cost"])
>>> G["E"]["C"]["weight"]
10
>>> G["E"]["C"]["cost"]
9
>>> edges = pd.DataFrame(
... {
... "source": [0, 1, 2],
... "target": [2, 2, 3],
... "weight": [3, 4, 5],
... "color": ["red", "blue", "blue"],
... }
... )
>>> G = nx.from_pandas_edgelist(edges, edge_attr=True)
>>> G[0][2]["color"]
'red'
Build multigraph with custom keys:

>>> edges = pd.DataFrame(

... {
... "source": [0, 1, 2, 0],
... "target": [2, 2, 3, 2],
... "my_edge_key": ["A", "B", "C", "D"],
... "weight": [3, 4, 5, 6],
... "color": ["red", "blue", "blue", "blue"],
... }
... )
>>> G = nx.from_pandas_edgelist(
... edges,
... edge_key="my_edge_key",
... edge_attr=["weight", "color"],
... create_using=nx.MultiGraph(),
... )
>>> G[0][2]
AtlasView({'A': {'weight': 3, 'color': 'red'}, 'D': {'weight': 6, 'color': 'blue'}
,→})
7.6. Pandas 925


CHAPTER
EIGHT
RELABELING NODES
8.1 Relabeling
convert_node_labels_to_integers(G[, ...]) Returns a copy of the graph G with the nodes relabeled
using consecutive integers.
relabel_nodes(G, mapping[, copy]) Relabel the nodes of the graph G according to a given
mapping.
8.1.1 convert_node_labels_to_integers
convert_node_labels_to_integers(G, first_label=0, ordering='default', label_attribute=None)

Returns a copy of the graph G with the nodes relabeled using consecutive integers.
Parameters
G
first_label
[int, optional (default=0)] An integer specifying the starting offset in numbering nodes. The
new integer labels are numbered first_label, …, n-1+first_label.
ordering
[string] “default” : inherit node ordering from G.nodes() “sorted” : inherit node ordering from
sorted(G.nodes()) “increasing degree” : nodes are sorted by increasing degree “decreasing
degree” : nodes are sorted by decreasing degree
label_attribute
[string, optional (default=None)] Name of node attribute to store old label. If None no attribute
is created.
See also:
relabel_nodes
927
Notes
Node and edge attribute data are copied to the new (relabeled) graph.
There is no guarantee that the relabeling of nodes to integers will give the same two integers for two (even identical
graphs). Use the ordering argument to try to preserve the order.
8.1.2 relabel_nodes
relabel_nodes(G, mapping, copy=True)

Relabel the nodes of the graph G according to a given mapping.
The original node ordering may not be preserved if copy is False and the mapping includes overlap between
old and new labels.
Parameters
G
mapping
[dictionary] A dictionary with the old labels as keys and new labels as values. A partial mapping
is allowed. Mapping 2 nodes to a single node is allowed. Any non-node keys in the mapping
are ignored.
copy
[bool (optional, default=True)] If True return a copy, or if False relabel the nodes in place.
See also:
convert_node_labels_to_integers
Notes
Only the nodes specified in the mapping will be relabeled. Any non-node keys in the mapping are ignored.
The keyword setting copy=False modifies the graph in place. Relabel_nodes avoids naming collisions by building a
directed graph from mapping which specifies the order of relabelings. Naming collisions, such as a->b, b->c, are
ordered such that “b” gets renamed to “c” before “a” gets renamed “b”. In cases of circular mappings (e.g. a->b,
b->a), modifying the graph is not possible in-place and an exception is raised. In that case, use copy=True.
If a relabel operation on a multigraph would cause two or more edges to have the same source, target and key, the
second edge must be assigned a new key to retain all edges. The new key is set to the lowest non-negative integer not
already used as a key for edges between these two nodes. Note that this means non-numeric keys may be replaced
by numeric keys.
928 Chapter 8. Relabeling nodes

Examples
To create a new graph with nodes relabeled according to a given dictionary:
>>> sorted(G)
[0, 1, 2]
>>> mapping = {0: "a", 1: "b", 2: "c"}
>>> H = nx.relabel_nodes(G, mapping)
>>> sorted(H)
['a', 'b', 'c']
Nodes can be relabeled with any hashable object, including numbers and strings:
>>> import string

>>> G = nx.path_graph(26) # nodes are integers 0 through 25
>>> sorted(G)[:3]
[0, 1, 2]
>>> mapping = dict(zip(G, string.ascii_lowercase))
>>> G = nx.relabel_nodes(G, mapping) # nodes are characters a through z
>>> sorted(G)[:3]
['a', 'b', 'c']
>>> mapping = dict(zip(G, range(1, 27)))
>>> G = nx.relabel_nodes(G, mapping) # nodes are integers 1 through 26
>>> sorted(G)[:3]
[1, 2, 3]
To perform a partial in-place relabeling, provide a dictionary mapping only a subset of the nodes, and set the copy
keyword argument to False:
>>> G = nx.path_graph(3) # nodes 0-1-2

>>> mapping = {0: "a", 1: "b"} # 0->'a' and 1->'b'
>>> G = nx.relabel_nodes(G, mapping, copy=False)
>>> sorted(G, key=str)
[2, 'a', 'b']
A mapping can also be given as a function:
>>> H = nx.relabel_nodes(G, lambda x: x ** 2)
>>> list(H)
[0, 1, 4]
In a multigraph, relabeling two or more nodes to the same new node will retain all edges, but may change the edge
keys in the process:
>>> G.add_edge(0, 1, value="a") # returns the key for this edge
0
>>> G.add_edge(0, 2, value="b")
0
>>> G.add_edge(0, 3, value="c")
0
>>> mapping = {1: 4, 2: 4, 3: 4}
>>> H = nx.relabel_nodes(G, mapping, copy=True)
>>> print(H[0])
{4: {0: {'value': 'a'}, 1: {'value': 'b'}, 2: {'value': 'c'}}}
8.1. Relabeling 929

This works for in-place relabeling too:
>>> G = nx.relabel_nodes(G, mapping, copy=False)

>>> print(G[0])
{4: {0: {'value': 'a'}, 1: {'value': 'b'}, 2: {'value': 'c'}}}
930 Chapter 8. Relabeling nodes

CHAPTER
NINE
READING AND WRITING GRAPHS
9.1 Adjacency List
Read and write NetworkX graphs as adjacency lists.

Adjacency list format is useful for graphs without data associated with nodes or edges and for nodes that can be mean-
ingfully represented as strings.
9.1.1 Format
The adjacency list format consists of lines with node labels. The first label in a line is the source node. Further labels in
the line are considered target nodes and are added to the graph along with an edge between the source node and target
node.
The graph with edges a-b, a-c, d-e can be represented as the following adjacency list (anything following the # in a line is
a comment):
a b c # source target target

d e
read_adjlist(path[, comments, delimiter, ...]) Read graph in adjacency list format from path.
write_adjlist(G, path[, comments, ...]) Write graph G in single-line adjacency-list format to path.
parse_adjlist(lines[, comments, delimiter, ...]) Parse lines of a graph adjacency list representation.
generate_adjlist(G[, delimiter]) Generate a single line of the graph G in adjacency list for-
mat.
9.1.2 read_adjlist
read_adjlist(path, comments='#', delimiter=None, create_using=None, nodetype=None, encoding='utf-8')

Read graph in adjacency list format from path.
Parameters
path
[string or file] Filename or file handle to read. Filenames ending in .gz or .bz2 will be uncom-
pressed.
create_using
931
nodetype
comments
delimiter
[string, optional] Separator for node labels. The default is whitespace.
Returns
G: NetworkX graph
The graph corresponding to the lines in adjacency list format.
See also:
write_adjlist
Notes
This format does not store graph or node data.
Examples
>>> nx.write_adjlist(G, "test.adjlist")
>>> G = nx.read_adjlist("test.adjlist")
The path can be a filehandle or a string with the name of the file. If a filehandle is provided, it has to be opened in
‘rb’ mode.
>>> fh = open("test.adjlist", "rb")

>>> G = nx.read_adjlist(fh)
>>> nx.write_adjlist(G, "test.adjlist.gz")

>>> G = nx.read_adjlist("test.adjlist.gz")
The optional nodetype is a function to convert node strings to nodetype.

For example
>>> G = nx.read_adjlist("test.adjlist", nodetype=int)
will attempt to convert all nodes to integer type.

The optional create_using parameter indicates the type of NetworkX graph created. The default is nx.Graph,
an undirected graph. To read the data as a directed graph use
>>> G = nx.read_adjlist("test.adjlist", create_using=nx.DiGraph)
932 Chapter 9. Reading and writing graphs

9.1.3 write_adjlist
write_adjlist(G, path, comments='#', delimiter=' ', encoding='utf-8')

Write graph G in single-line adjacency-list format to path.
Parameters
G
[NetworkX graph]
path
[string or file] Filename or file handle for data output. Filenames ending in .gz or .bz2 will be
compressed.
comments
delimiter
encoding
[string, optional] Text encoding.
See also:
read_adjlist, generate_adjlist
Notes
The default delimiter=" " will result in unexpected results if node names contain whitespace characters. To
avoid this problem, specify an alternate delimiter when spaces are valid in node names. NB: This option is not
available for data that isn’t user-generated.
This format does not store graph, node, or edge data.
Examples
>>> nx.write_adjlist(G, "test.adjlist")
The path can be a filehandle or a string with the name of the file. If a filehandle is provided, it has to be opened in
‘wb’ mode.
>>> fh = open("test.adjlist", "wb")

>>> nx.write_adjlist(G, fh)
9.1. Adjacency List 933

9.1.4 parse_adjlist
parse_adjlist(lines, comments='#', delimiter=None, create_using=None, nodetype=None)

Parse lines of a graph adjacency list representation.
Parameters
lines
[list or iterator of strings] Input data in adjlist format
create_using
nodetype
comments
delimiter
Returns
G: NetworkX graph
The graph corresponding to the lines in adjacency list format.
See also:
read_adjlist
Examples
>>> lines = ["1 2 5", "2 3 4", "3 5", "4", "5"]
>>> G = nx.parse_adjlist(lines, nodetype=int)
>>> nodes = [1, 2, 3, 4, 5]
>>> all(node in G for node in nodes)
True
>>> edges = [(1, 2), (1, 5), (2, 3), (2, 4), (3, 5)]
>>> all((u, v) in G.edges() or (v, u) in G.edges() for (u, v) in edges)
True
9.1.5 generate_adjlist
generate_adjlist(G, delimiter=' ')

Generate a single line of the graph G in adjacency list format.
Parameters
G
[NetworkX graph]
delimiter
Returns

lines
See also:
write_adjlist, read_adjlist
Notes
The default delimiter=" " will result in unexpected results if node names contain whitespace characters. To
avoid this problem, specify an alternate delimiter when spaces are valid in node names.
NB: This option is not available for data that isn’t user-generated.
Examples
>>> for line in nx.generate_adjlist(G):
... print(line)
0 1 2 3
1 2 3
2 3
3 4
4 5
5 6
6
9.2 Multiline Adjacency List
Read and write NetworkX graphs as multi-line adjacency lists.

The multi-line adjacency list format is useful for graphs with nodes that can be meaningfully represented as strings. With
this format simple edge data can be stored but node or graph data is not.
9.2.1 Format
The first label in a line is the source node label followed by the node degree d. The next d lines are target node labels and
optional edge data. That pattern repeats for all nodes in the graph.
The graph with edges a-b, a-c, d-e can be represented as the following adjacency list (anything following the # in a line is
a comment):
# example.multiline-adjlist
a 2
b
c
d 1
e
9.2. Multiline Adjacency List 935

read_multiline_adjlist(path[, comments, ...]) Read graph in multi-line adjacency list format from path.
write_multiline_adjlist(G, path[, ...]) Write the graph G in multiline adjacency list format to
path
parse_multiline_adjlist(lines[, comments, Parse lines of a multiline adjacency list representation of
...]) a graph.
generate_multiline_adjlist(G[, delimiter]) Generate a single line of the graph G in multiline adja-
cency list format.
9.2.2 read_multiline_adjlist
read_multiline_adjlist(path, comments='#', delimiter=None, create_using=None, nodetype=None,

Read graph in multi-line adjacency list format from path.
Parameters
path
[string or file] Filename or file handle to read. Filenames ending in .gz or .bz2 will be uncom-
pressed.
create_using
nodetype
edgetype
[Python type, optional] Convert edge data to this type.
comments
delimiter
Returns
G: NetworkX graph
See also:
write_multiline_adjlist
Notes
This format does not store graph, node, or edge data.

Examples
>>> nx.write_multiline_adjlist(G, "test.adjlist")
>>> G = nx.read_multiline_adjlist("test.adjlist")
The path can be a file or a string with the name of the file. If a file s provided, it has to be opened in ‘rb’ mode.
>>> fh = open("test.adjlist", "rb")
>>> G = nx.read_multiline_adjlist(fh)

>>> nx.write_multiline_adjlist(G, "test.adjlist.gz")
>>> G = nx.read_multiline_adjlist("test.adjlist.gz")
The optional nodetype is a function to convert node strings to nodetype.

For example
>>> G = nx.read_multiline_adjlist("test.adjlist", nodetype=int)
will attempt to convert all nodes to integer type.

The optional edgetype is a function to convert edge data strings to edgetype.
>>> G = nx.read_multiline_adjlist("test.adjlist")
The optional create_using parameter is a NetworkX graph container. The default is Graph(), an undirected graph.
To read the data as a directed graph use
>>> G = nx.read_multiline_adjlist("test.adjlist", create_using=nx.DiGraph)
9.2.3 write_multiline_adjlist
write_multiline_adjlist(G, path, delimiter=' ', comments='#', encoding='utf-8')

Write the graph G in multiline adjacency list format to path
Parameters
G
[NetworkX graph]
path
[string or file] Filename or file handle to write to. Filenames ending in .gz or .bz2 will be
compressed.
comments
delimiter
encoding
[string, optional] Text encoding.
See also:
read_multiline_adjlist

Examples
>>> nx.write_multiline_adjlist(G, "test.adjlist")
The path can be a file handle or a string with the name of the file. If a file handle is provided, it has to be opened
in ‘wb’ mode.
>>> fh = open("test.adjlist", "wb")

>>> nx.write_multiline_adjlist(G, fh)
>>> nx.write_multiline_adjlist(G, "test.adjlist.gz")
9.2.4 parse_multiline_adjlist
parse_multiline_adjlist(lines, comments='#', delimiter=None, create_using=None, nodetype=None,

edgetype=None)
Parse lines of a multiline adjacency list representation of a graph.
Parameters
lines
[list or iterator of strings] Input data in multiline adjlist format
create_using
nodetype
edgetype
[Python type, optional] Convert edges to this type.
comments
delimiter
Returns
G: NetworkX graph
The graph corresponding to the lines in multiline adjacency list format.

Examples
>>> lines = [
... "1 2",
... "2 {'weight':3, 'name': 'Frodo'}",
... "3 {}",
... "2 1",
... "5 {'weight':6, 'name': 'Saruman'}",
... ]
>>> G = nx.parse_multiline_adjlist(iter(lines), nodetype=int)
>>> list(G)
[1, 2, 3, 5]
9.2.5 generate_multiline_adjlist
generate_multiline_adjlist(G, delimiter=' ')

Generate a single line of the graph G in multiline adjacency list format.
Parameters
G
[NetworkX graph]
delimiter
Returns
lines
[string] Lines of data in multiline adjlist format.
See also:
write_multiline_adjlist, read_multiline_adjlist
Examples
>>> for line in nx.generate_multiline_adjlist(G):
... print(line)
0 3
1 {}
2 {}
3 {}
1 2
2 {}
3 {}
2 1
3 {}
3 1
4 {}
4 1
5 {}
5 1
6 {}
6 0

9.3 Edge List
Read and write NetworkX graphs as edge lists.

The multi-line adjacency list format is useful for graphs with nodes that can be meaningfully represented as strings. With
the edgelist format simple edge data can be stored but node or graph data is not. There is no way of representing isolated
nodes unless the node has a self-loop edge.
9.3.1 Format
You can read or write three formats of edge lists with these functions.
Node pairs with no data:
1 2
Python dictionary as data:
1 2 {'weight':7, 'color':'green'}
Arbitrary data:
1 2 7 green
read_edgelist(path[, comments, delimiter, ...]) Read a graph from a list of edges.

write_edgelist(G, path[, comments, ...]) Write graph as a list of edges.
read_weighted_edgelist(path[, comments, ...]) Read a graph as list of edges with numeric weights.
write_weighted_edgelist(G, path[, comments, Write graph G as a list of edges with numeric weights.
...])
generate_edgelist(G[, delimiter, data]) Generate a single line of the graph G in edge list format.
parse_edgelist(lines[, comments, delimiter, ...]) Parse lines of an edge list representation of a graph.
9.3.2 read_edgelist
read_edgelist(path, comments='#', delimiter=None, create_using=None, nodetype=None, data=True,

Read a graph from a list of edges.
Parameters
path
comments
[string, optional] The character used to indicate the start of a comment. To specify that no
character should be treated as a comment, use comments=None.
delimiter
create_using

nodetype
data
[bool or list of (label,type) tuples] Tuples specifying dictionary key names and types for edge
data
edgetype
[int, float, str, Python type, optional OBSOLETE] Convert edge data from strings to specified
type and use as ‘weight’
Returns
G
See also:
parse_edgelist
write_edgelist
Notes
Examples
>>> nx.write_edgelist(nx.path_graph(4), "test.edgelist")

>>> G = nx.read_edgelist("test.edgelist")
>>> fh = open("test.edgelist", "rb")

>>> G = nx.read_edgelist(fh)
>>> fh.close()
>>> G = nx.read_edgelist("test.edgelist", nodetype=int)

>>> G = nx.read_edgelist("test.edgelist", create_using=nx.DiGraph)
>>> textline = "1 2 3"

>>> fh = open("test.edgelist", "w")
>>> d = fh.write(textline)
>>> fh.close()
>>> G = nx.read_edgelist("test.edgelist", nodetype=int, data=(("weight", float),))
>>> list(G)
[1, 2]
[(1, 2, {'weight': 3.0})]
See parse_edgelist() for more examples of formatting.
9.3. Edge List 941

9.3.3 write_edgelist
write_edgelist(G, path, comments='#', delimiter=' ', data=True, encoding='utf-8')

Write graph as a list of edges.
Parameters
G
path
comments
delimiter
data
[bool or list, optional] If False write no edge data. If True write a string representation of the
edge data dictionary.. If a list (or other iterable) is provided, write the keys specified in the list.
See also:
read_edgelist
write_weighted_edgelist
Examples
>>> nx.write_edgelist(G, "test.edgelist")
>>> fh = open("test.edgelist", "wb")
>>> nx.write_edgelist(G, fh)
>>> nx.write_edgelist(G, "test.edgelist.gz")
>>> nx.write_edgelist(G, "test.edgelist.gz", data=False)
>>> G = nx.Graph()
>>> G.add_edge(1, 2, weight=7, color="red")
>>> nx.write_edgelist(G, "test.edgelist", data=False)
>>> nx.write_edgelist(G, "test.edgelist", data=["color"])
>>> nx.write_edgelist(G, "test.edgelist", data=["color", "weight"])

9.3.4 read_weighted_edgelist
read_weighted_edgelist(path, comments='#', delimiter=None, create_using=None, nodetype=None,

encoding='utf-8')
Read a graph as list of edges with numeric weights.
Parameters
path
comments
[string, optional] The character used to indicate the start of a comment.
delimiter
create_using
nodetype
Returns
G
See also:
write_weighted_edgelist
Notes
Example edgelist file format.
With numeric edge data:
# read with
# >>> G=nx.read_weighted_edgelist(fh)
# source target data
a b 1
a c 3.14159
d e 42
9.3. Edge List 943

9.3.5 write_weighted_edgelist
write_weighted_edgelist(G, path, comments='#', delimiter=' ', encoding='utf-8')

Write graph G as a list of edges with numeric weights.
Parameters
G
path
comments
delimiter
See also:
read_edgelist
write_edgelist
read_weighted_edgelist
Examples
>>> G = nx.Graph()
>>> nx.write_weighted_edgelist(G, "test.weighted.edgelist")
9.3.6 generate_edgelist
generate_edgelist(G, delimiter=' ', data=True)

Generate a single line of the graph G in edge list format.
Parameters
G
[NetworkX graph]
delimiter
data
[bool or list of keys] If False generate no edge data. If True use a dictionary representation of
edge data. If a list of keys use a list of data values corresponding to the keys.
Returns
lines
See also:

write_adjlist, read_adjlist
Examples
>>> G[1][2]["weight"] = 3
>>> G[3][4]["capacity"] = 12
>>> for line in nx.generate_edgelist(G, data=False):
... print(line)
0 1
0 2
0 3
1 2
1 3
2 3
3 4
4 5
5 6
>>> for line in nx.generate_edgelist(G):

... print(line)
0 1 {}
0 2 {}
0 3 {}
1 2 {'weight': 3}
1 3 {}
2 3 {}
3 4 {'capacity': 12}
4 5 {}
5 6 {}
>>> for line in nx.generate_edgelist(G, data=["weight"]):

... print(line)
0 1
0 2
0 3
1 2 3
1 3
2 3
3 4
4 5
5 6
9.3.7 parse_edgelist
parse_edgelist(lines, comments='#', delimiter=None, create_using=None, nodetype=None, data=True)

Parse lines of an edge list representation of a graph.
Parameters
lines
[list or iterator of strings] Input data in edgelist format
comments
[string, optional] Marker for comment lines. Default is '#'. To specify that no character
9.3. Edge List 945

should be treated as a comment, use comments=None.

delimiter
[string, optional] Separator for node labels. Default is None, meaning any whitespace.
create_using
nodetype
[Python type, optional] Convert nodes to this type. Default is None, meaning no conversion
is performed.
data
[bool or list of (label,type) tuples] If False generate no edge data or if True use a dictionary
representation of edge data or a list tuples specifying dictionary key names and types for edge
data.
Returns
G: NetworkX Graph
The graph corresponding to lines
See also:
read_weighted_edgelist
Examples
Edgelist with no data:
>>> lines = ["1 2", "2 3", "3 4"]

>>> G = nx.parse_edgelist(lines, nodetype=int)
>>> list(G)
[1, 2, 3, 4]
>>> list(G.edges())
[(1, 2), (2, 3), (3, 4)]
Edgelist with data in Python dictionary representation:
>>> lines = ["1 2 {'weight': 3}", "2 3 {'weight': 27}", "3 4 {'weight': 3.0}"]
>>> G = nx.parse_edgelist(lines, nodetype=int)
>>> list(G)
[1, 2, 3, 4]
[(1, 2, {'weight': 3}), (2, 3, {'weight': 27}), (3, 4, {'weight': 3.0})]
>>> lines = ["1 2 3", "2 3 27", "3 4 3.0"]

>>> G = nx.parse_edgelist(lines, nodetype=int, data=(("weight", float),))
>>> list(G)
[1, 2, 3, 4]
[(1, 2, {'weight': 3.0}), (2, 3, {'weight': 27.0}), (3, 4, {'weight': 3.0})]

9.4 GEXF
Read and write graphs in GEXF format.
Warning: This parser uses the standard xml library present in Python, which is insecure - see library/xml for
additional information. Only parse GEFX files you trust.
GEXF (Graph Exchange XML Format) is a language for describing complex network structures, their associated data
and dynamics.
This implementation does not support mixed graphs (directed and undirected edges together).
9.4.1 Format
GEXF is an XML format. See http://gexf.net/schema.html for the specification and http://gexf.net/basic.html for exam-
ples.
read_gexf(path[, node_type, relabel, version]) Read graph in GEXF format from path.
write_gexf(G, path[, encoding, prettyprint, ...]) Write G in GEXF format to path.
generate_gexf(G[, encoding, prettyprint, ...]) Generate lines of GEXF format representation of G.
relabel_gexf_graph(G) Relabel graph using "label" node keyword for node label.
9.4.2 read_gexf
read_gexf(path, node_type=None, relabel=False, version='1.2draft')

Read graph in GEXF format from path.
“GEXF (Graph Exchange XML Format) is a language for describing complex networks structures, their associated
data and dynamics” [1].
Parameters
path
[file or string] File or file name to read. File names ending in .gz or .bz2 will be decompressed.
node_type: Python type (default: None)
Convert node ids to this type if not None.
relabel
[bool (default: False)] If True relabel the nodes to use the GEXF node “label” attribute instead
of the node “id” attribute as the NetworkX node label.
version
[string (default: 1.2draft)]
Version of GEFX File Format (see http://gexf.net/schema.html)
Supported values: “1.1draft”, “1.2draft”
Returns
graph: NetworkX graph
If no parallel edges are found a Graph or DiGraph is returned. Otherwise a MultiGraph or
MultiDiGraph is returned.
9.4. GEXF 947

Notes
References
[1]
9.4.3 write_gexf
write_gexf(G, path, encoding='utf-8', prettyprint=True, version='1.2draft')

Write G in GEXF format to path.
Node attributes are checked according to the version of the GEXF schemas used for parameters which are not user
defined, e.g. visualization ‘viz’ [2]. See example for usage.
Parameters
G
path
[file or string] File or file name to write. File names ending in .gz or .bz2 will be compressed.
encoding
[string (optional, default: ‘utf-8’)] Encoding for text data.
prettyprint
[bool (optional, default: True)] If True use line breaks and indenting in output XML.
version: string (optional, default: ‘1.2draft’)
The version of GEXF to be used for nodes attributes checking
Notes
The node id attribute is set to be the string of the node label. If you want to specify an id use set it as node data,
e.g. node[‘a’][‘id’]=1 to set the id of node ‘a’ to 1.
References
[1], [2]

Examples
>>> nx.write_gexf(G, "test.gexf")
# visualization data >>> G.nodes[0][“viz”] = {“size”: 54} >>> G.nodes[0][“viz”][“position”] = {“x”: 0, “y”: 1}
>>> G.nodes[0][“viz”][“color”] = {“r”: 0, “g”: 0, “b”: 256}
9.4.4 generate_gexf
generate_gexf(G, encoding='utf-8', prettyprint=True, version='1.2draft')

Generate lines of GEXF format representation of G.
Parameters
G
[graph]
A NetworkX graph
encoding
[string (optional, default: ‘utf-8’)]
Encoding for text data.
prettyprint
[bool (optional, default: True)]
If True use line breaks and indenting in output XML.
version
[string (default: 1.2draft)]
Version of GEFX File Format (see http://gexf.net/schema.html)
Supported values: “1.1draft”, “1.2draft”
Notes
The node id attribute is set to be the string of the node label. If you want to specify an id use set it as node data,
e.g. node[‘a’][‘id’]=1 to set the id of node ‘a’ to 1.
References
[1]
9.4. GEXF 949

Examples
>>> linefeed = chr(10) # linefeed=
>>> s = linefeed.join(nx.generate_gexf(G))
>>> for line in nx.generate_gexf(G):
... print(line)
9.4.5 relabel_gexf_graph
relabel_gexf_graph(G)
Relabel graph using “label” node keyword for node label.
Parameters
G
[graph] A NetworkX graph read from GEXF data
Returns
H
[graph] A NetworkX graph with relabeled nodes
Raises
NetworkXError
If node labels are missing or not unique while relabel=True.
Notes
This function relabels the nodes in a NetworkX graph with the “label” attribute. It also handles relabeling the
specific GEXF node attributes “parents”, and “pid”.
9.5 GML
Read graphs in GML format.

“GML, the Graph Modelling Language, is our proposal for a portable file format for graphs. GML’s key features are
portability, simple syntax, extensibility and flexibility. A GML file consists of a hierarchical key-value lists. Graphs can
be annotated with arbitrary data structures. The idea for a common file format was born at the GD’95; this proposal is the
outcome of many discussions. GML is the standard file format in the Graphlet graph editor system. It has been overtaken
and adapted by several other systems for drawing graphs.”
GML files are stored using a 7-bit ASCII encoding with any extended ASCII characters (iso8859-1) appearing as HTML
character entities. You will need to give some thought into how the exported data should interact with different languages
and even different Python versions. Re-importing from gml is also a concern.
Without specifying a stringizer/destringizer, the code is capable of writing int/float/str/dict/list
data as required by the GML specification. For writing other data types, and for reading data other than str you need
to explicitly supply a stringizer/destringizer.
For additional documentation on the GML file format, please see the GML website.
Several example graphs in GML format may be found on Mark Newman’s Network data page.

read_gml(path[, label, destringizer]) Read graph in GML format from path.

write_gml(G, path[, stringizer]) Write a graph G in GML format to the file or file handle
path.
parse_gml(lines[, label, destringizer]) Parse GML graph from a string or iterable.
generate_gml(G[, stringizer]) Generate a single entry of the graph G in GML format.
literal_destringizer(rep) Convert a Python literal to the value it represents.
literal_stringizer(value) Convert a value to a Python literal in GML representa-
tion.
9.5.1 read_gml
read_gml(path, label='label', destringizer=None)

Read graph in GML format from path.
Parameters
path
[filename or filehandle] The filename or filehandle to read from.
label
[string, optional] If not None, the parsed nodes will be renamed according to node attributes
indicated by label. Default value: ‘label’.
destringizer
[callable, optional] A destringizer that recovers values stored as strings in GML. If it
cannot convert a string to a value, a ValueError is raised. Default value : None.
Returns
G
[NetworkX graph] The parsed graph.
Raises
NetworkXError
If the input cannot be parsed.
See also:
write_gml, parse_gml
literal_destringizer
Notes
GML files are stored using a 7-bit ASCII encoding with any extended ASCII characters (iso8859-1) appearing as
HTML character entities. Without specifying a stringizer/destringizer, the code is capable of writing
int/float/str/dict/list data as required by the GML specification. For writing other data types, and for
reading data other than str you need to explicitly supply a stringizer/destringizer.
For additional documentation on the GML file format, please see the GML url.
See the module docstring networkx.readwrite.gml for more details.
9.5. GML 951

Examples
>>> nx.write_gml(G, "test.gml")
GML values are interpreted as strings by default:
>>> H = nx.read_gml("test.gml")
>>> H.nodes
NodeView(('0', '1', '2', '3'))
When a destringizer is provided, GML values are converted to the provided type. For example, integer nodes
can be recovered as shown below:
>>> J = nx.read_gml("test.gml", destringizer=int)

>>> J.nodes
NodeView((0, 1, 2, 3))
9.5.2 write_gml
write_gml(G, path, stringizer=None)

Write a graph G in GML format to the file or file handle path.
Parameters
G
[NetworkX graph] The graph to be converted to GML.
path
[filename or filehandle] The filename or filehandle to write. Files whose names end with .gz or
.bz2 will be compressed.
stringizer
[callable, optional] A stringizer which converts non-int/non-float/non-dict values into
strings. If it cannot convert a value into a string, it should raise a ValueError to indicate
that. Default value: None.
Raises
NetworkXError
If stringizer cannot convert a value into a string, or the value to convert is not a string
while stringizer is None.
See also:
read_gml, generate_gml
literal_stringizer

Notes
Graph attributes named ‘directed’, ‘multigraph’, ‘node’ or ‘edge’, node attributes named ‘id’ or ‘label’, edge attributes
named ‘source’ or ‘target’ (or ‘key’ if G is a multigraph) are ignored because these attribute names are used to encode
the graph structure.
Note that while we allow non-standard GML to be read from a file, we make sure to write GML format. In
particular, underscores are not allowed in attribute names. For additional documentation on the GML file format,
please see the GML url.
Examples
>>> nx.write_gml(G, "test.gml")
>>> nx.write_gml(G, "test.gml.gz")
9.5.3 parse_gml
parse_gml(lines, label='label', destringizer=None)

Parse GML graph from a string or iterable.
Parameters
lines
[string or iterable of strings] Data in GML format.
label
[string, optional] If not None, the parsed nodes will be renamed according to node attributes
indicated by label. Default value: ‘label’.
destringizer
[callable, optional] A destringizer that recovers values stored as strings in GML. If it
cannot convert a string to a value, a ValueError is raised. Default value : None.
Returns
G
[NetworkX graph] The parsed graph.
Raises
NetworkXError
If the input cannot be parsed.
See also:
write_gml, read_gml
9.5. GML 953

Notes
This stores nested GML attributes as dictionaries in the NetworkX graph, node, and edge attribute structures.
9.5.4 generate_gml
generate_gml(G, stringizer=None)
Generate a single entry of the graph G in GML format.
Parameters
G
[NetworkX graph] The graph to be converted to GML.
stringizer
[callable, optional] A stringizer which converts non-int/non-float/non-dict values into
strings. If it cannot convert a value into a string, it should raise a ValueError to indicate
that. Default value: None.
Returns
lines: generator of strings
Lines of GML data. Newlines are not appended.
Raises
NetworkXError
If stringizer cannot convert a value into a string, or the value to convert is not a string
while stringizer is None.
See also:
literal_stringizer
Notes
Graph attributes named ‘directed’, ‘multigraph’, ‘node’ or ‘edge’, node attributes named ‘id’ or ‘label’, edge attributes
named ‘source’ or ‘target’ (or ‘key’ if G is a multigraph) are ignored because these attribute names are used to encode
the graph structure.

Examples
>>> G = nx.Graph()
>>> G.add_node("1")
>>> print("\n".join(nx.generate_gml(G)))
graph [
node [
id 0
label "1"
]
]
>>> G = nx.MultiGraph([("a", "b"), ("a", "b")])
>>> print("\n".join(nx.generate_gml(G)))
graph [
multigraph 1
node [
id 0
label "a"
]
node [
id 1
label "b"
]
edge [
source 0
target 1
key 0
]
edge [
source 0
target 1
key 1
]
]
9.5.5 literal_destringizer
literal_destringizer(rep)
Convert a Python literal to the value it represents.
Parameters
rep
[string] A Python literal.
Returns
value
[object] The value of the Python literal.
Raises
ValueError
If rep is not a Python literal.
9.5. GML 955

9.5.6 literal_stringizer
literal_stringizer(value)
Convert a value to a Python literal in GML representation.
Parameters
value
[object] The value to be converted to GML representation.
Returns
rep
[string] A double-quoted Python literal representing value. Unprintable characters are replaced
by XML character references.
Raises
ValueError
If value cannot be converted to GML.
Notes
literal_stringizer is largely the same as repr in terms of functionality but attempts prefix unicode
and bytes literals with u and b to provide better interoperability of data generated by Python 2 and Python 3.
The original value can be recovered using the networkx.readwrite.gml.literal_destringizer()
function.
9.6 GraphML
Read and write graphs in GraphML format.
Warning: This parser uses the standard xml library present in Python, which is insecure - see library/xml for
additional information. Only parse GraphML files you trust.
This implementation does not support mixed graphs (directed and unidirected edges together), hyperedges, nested graphs,
or ports.
“GraphML is a comprehensive and easy-to-use file format for graphs. It consists of a language core to describe the
structural properties of a graph and a flexible extension mechanism to add application-specific data. Its main features
include support of
• directed, undirected, and mixed graphs,
• hypergraphs,
• hierarchical graphs,
• graphical representations,
• references to external data,
• application-specific attribute data, and
• light-weight parsers.

Unlike many other file formats for graphs, GraphML does not use a custom syntax. Instead, it is based on XML and
hence ideally suited as a common denominator for all kinds of services generating, archiving, or processing graphs.”
http://graphml.graphdrawing.org/
9.6.1 Format
GraphML is an XML format. See http://graphml.graphdrawing.org/specification.html for the specification and http:
//graphml.graphdrawing.org/primer/graphml-primer.html for examples.
read_graphml(path[, node_type, ...]) Read graph in GraphML format from path.

write_graphml(G, path[, encoding, ...]) Write G in GraphML XML format to path
generate_graphml(G[, encoding, prettyprint, ...]) Generate GraphML lines for G
parse_graphml(graphml_string[, node_type, ...]) Read graph in GraphML format from string.
9.6.2 read_graphml
read_graphml(path, node_type=<class 'str'>, edge_key_type=<class 'int'>, force_multigraph=False)

Read graph in GraphML format from path.
Parameters
path
[file or string] File or filename to write. Filenames ending in .gz or .bz2 will be compressed.
node_type: Python type (default: str)
Convert node ids to this type
edge_key_type: Python type (default: int)
Convert graphml edge ids to this type. Multigraphs use id as edge key. Non-multigraphs add
to edge attribute dict with name “id”.
force_multigraph
[bool (default: False)] If True, return a multigraph with edge keys. If False (the default) return
a multigraph when multiedges are in the graph.
Returns
If parallel edges are present or force_multigraph=True then a MultiGraph or Multi-
DiGraph is returned. Otherwise a Graph/DiGraph. The returned graph is directed if the file
indicates it should be.
Notes
Default node and edge attributes are not propagated to each node and edge. They can be obtained from G.graph
and applied to node and edge attributes if desired using something like this:
>>> default_color = G.graph["node_default"]["color"]

>>> for node, data in G.nodes(data=True):
... if "color" not in data:
... data["color"] = default_color
>>> default_color = G.graph["edge_default"]["color"]
>>> for u, v, data in G.edges(data=True):
9.6. GraphML 957


This implementation does not support mixed graphs (directed and unidirected edges together), hypergraphs, nested
graphs, or ports.
For multigraphs the GraphML edge “id” will be used as the edge key. If not specified then they “key” attribute will
be used. If there is no “key” attribute a default NetworkX multigraph edge key will be provided.
Files with the yEd “yfiles” extension can be read. The type of the node’s shape is preserved in the shape_type
node attribute.
yEd compressed files (“file.graphmlz” extension) can be read by renaming the file to “file.graphml.gz”.
9.6.3 write_graphml
write_graphml(G, path, encoding='utf-8', prettyprint=True, infer_numeric_types=False, named_key_ids=False,

edge_id_from_attribute=None)
Write G in GraphML XML format to path
This function uses the LXML framework and should be faster than the version using the xml library.
Parameters
G
path
encoding
[string (optional)] Encoding for text data.
prettyprint
[bool (optional)] If True use line breaks and indenting in output XML.
infer_numeric_types
[boolean] Determine if numeric types should be generalized. For example, if edges have both
int and float ‘weight’ attributes, we infer in GraphML that both are floats.
named_key_ids
[bool (optional)] If True use attr.name as value for key elements’ id attribute.
edge_id_from_attribute
[dict key (optional)] If provided, the graphml edge id is set by looking up the corresponding
edge data attribute keyed by this parameter. If None or the key does not exist in edge data,
the edge id is set by the edge key if G is a MultiGraph, else the edge id is left unset.

Notes
This implementation does not support mixed graphs (directed and unidirected edges together) hyperedges, nested
graphs, or ports.
Examples
>>> nx.write_graphml_lxml(G, "fourpath.graphml")
9.6.4 generate_graphml
generate_graphml(G, encoding='utf-8', prettyprint=True, named_key_ids=False, edge_id_from_attribute=None)

Generate GraphML lines for G
Parameters
G
encoding
[string (optional)] Encoding for text data.
prettyprint
[bool (optional)] If True use line breaks and indenting in output XML.
named_key_ids
[bool (optional)] If True use attr.name as value for key elements’ id attribute.
edge_id_from_attribute
[dict key (optional)] If provided, the graphml edge id is set by looking up the corresponding
edge data attribute keyed by this parameter. If None or the key does not exist in edge data,
the edge id is set by the edge key if G is a MultiGraph, else the edge id is left unset.
Notes
This implementation does not support mixed graphs (directed and unidirected edges together) hyperedges, nested
graphs, or ports.
Examples
>>> linefeed = chr(10) # linefeed =
>>> s = linefeed.join(nx.generate_graphml(G))
>>> for line in nx.generate_graphml(G):
... print(line)
9.6. GraphML 959

9.6.5 parse_graphml
parse_graphml(graphml_string, node_type=<class 'str'>, edge_key_type=<class 'int'>, force_multigraph=False)

Read graph in GraphML format from string.
Parameters
graphml_string
[string] String containing graphml information (e.g., contents of a graphml file).
node_type: Python type (default: str)
Convert node ids to this type
edge_key_type: Python type (default: int)
Convert graphml edge ids to this type. Multigraphs use id as edge key. Non-multigraphs add
to edge attribute dict with name “id”.
force_multigraph
[bool (default: False)] If True, return a multigraph with edge keys. If False (the default) return
a multigraph when multiedges are in the graph.
Returns
If no parallel edges are found a Graph or DiGraph is returned. Otherwise a MultiGraph or
MultiDiGraph is returned.
Notes
Default node and edge attributes are not propagated to each node and edge. They can be obtained from G.graph
and applied to node and edge attributes if desired using something like this:
>>> default_color = G.graph["node_default"]["color"]

>>> for node, data in G.nodes(data=True):
>>> default_color = G.graph["edge_default"]["color"]
>>> for u, v, data in G.edges(data=True):
This implementation does not support mixed graphs (directed and unidirected edges together), hypergraphs, nested
graphs, or ports.
For multigraphs the GraphML edge “id” will be used as the edge key. If not specified then they “key” attribute will
be used. If there is no “key” attribute a default NetworkX multigraph edge key will be provided.
Examples
>>> linefeed = chr(10) # linefeed =
>>> s = linefeed.join(nx.generate_graphml(G))
>>> H = nx.parse_graphml(s)

9.7 JSON
Generate and parse JSON serializable data for NetworkX graphs.

These formats are suitable for use with the d3.js examples https://d3js.org/
The three formats that you can generate with NetworkX are:
• node-link like in the d3.js example https://bl.ocks.org/mbostock/4062045
• tree like in the d3.js example https://bl.ocks.org/mbostock/4063550
• adjacency like in the d3.js example https://bost.ocks.org/mike/miserables/
node_link_data(G[, attrs, source, target, ...]) Returns data in node-link format that is suitable for JSON
serialization and use in Javascript documents.
node_link_graph(data[, directed, ...]) Returns graph from node-link data format.
adjacency_data(G[, attrs]) Returns data in adjacency format that is suitable for JSON
serialization and use in Javascript documents.
adjacency_graph(data[, directed, ...]) Returns graph from adjacency data format.
cytoscape_data(G[, name, ident]) Returns data in Cytoscape JSON format (cyjs).
cytoscape_graph(data[, name, ident]) Create a NetworkX graph from a dictionary in cytoscape
JSON format.
tree_data(G, root[, ident, children]) Returns data in tree format that is suitable for JSON seri-
alization and use in Javascript documents.
tree_graph(data[, ident, children]) Returns graph from tree data format.
9.7.1 node_link_data
node_link_data(G, attrs=None, *, source='source', target='target', name='id', key='key', link='links')

Returns data in node-link format that is suitable for JSON serialization and use in Javascript documents.
Parameters
G
[NetworkX graph]
attrs
[dict] A dictionary that contains five keys ‘source’, ‘target’, ‘name’, ‘key’ and ‘link’. The cor-
responding values provide the attribute names for storing NetworkX-internal graph data. The
values should be unique. Default value:
dict(source='source', target='target', name='id',

key='key', link='links')
If some user-defined graph data use these attribute names as data keys, they may be silently
dropped.
Deprecated since version 2.8.6: The attrs keyword argument will be replaced with
source, target, name, key and link. in networkx 3.2
If the attrs keyword and the new keywords are both used in a single function call (not
recommended) the attrs keyword argument will take precedence.
The values of the keywords must be unique.
9.7. JSON 961

source
[string] A string that provides the ‘source’ attribute name for storing NetworkX-internal graph
data.
target
[string] A string that provides the ‘target’ attribute name for storing NetworkX-internal graph
data.
name
[string] A string that provides the ‘name’ attribute name for storing NetworkX-internal graph
data.
key
[string] A string that provides the ‘key’ attribute name for storing NetworkX-internal graph
data.
link
[string] A string that provides the ‘link’ attribute name for storing NetworkX-internal graph
data.
Returns
data
[dict] A dictionary with node-link formatted data.
Raises
NetworkXError
If the values of ‘source’, ‘target’ and ‘key’ are not unique.
See also:
node_link_graph, adjacency_data, tree_data
Notes
Graph, node, and link attributes are stored in this format. Note that attribute keys will be converted to strings in
order to comply with JSON.
Attribute ‘key’ is only used for multigraphs.
To use node_link_data in conjunction with node_link_graph, the keyword names for the attributes must
match.
Examples
>>> G = nx.Graph([("A", "B")])

>>> data1 = nx.node_link_data(G)
>>> data1
{'directed': False, 'multigraph': False, 'graph': {}, 'nodes': [{'id': 'A'}, {'id
,→': 'B'}], 'links': [{'source': 'A', 'target': 'B'}]}
To serialize with JSON
>>> import json

>>> s1 = json.dumps(data1)
>>> s1


'{"directed": false, "multigraph": false, "graph": {}, "nodes": [{"id": "A"}, {"id
,→": "B"}], "links": [{"source": "A", "target": "B"}]}'
A graph can also be serialized by passing node_link_data as an encoder function. The two methods are
equivalent.
>>> s1 = json.dumps(G, default=nx.node_link_data)

>>> s1
'{"directed": false, "multigraph": false, "graph": {}, "nodes": [{"id": "A"}, {"id
,→": "B"}], "links": [{"source": "A", "target": "B"}]}'
The attribute names for storing NetworkX-internal graph data can be specified as keyword options.
>>> H = nx.gn_graph(2)
>>> data2 = nx.node_link_data(H, link="edges", source="from", target="to")
>>> data2
{'directed': True, 'multigraph': False, 'graph': {}, 'nodes': [{'id': 0}, {'id':␣
,→1}], 'edges': [{'from': 1, 'to': 0}]}
9.7.2 node_link_graph
node_link_graph(data, directed=False, multigraph=True, attrs=None, *, source='source', target='target',

name='id', key='key', link='links')
Returns graph from node-link data format. Useful for de-serialization from JSON.
Parameters
data
[dict] node-link formatted graph data
directed
[bool] If True, and direction not specified in data, return a directed graph.
multigraph
[bool] If True, and multigraph not specified in data, return a multigraph.
attrs
[dict] A dictionary that contains five keys ‘source’, ‘target’, ‘name’, ‘key’ and ‘link’. The corre-
sponding values provide the attribute names for storing NetworkX-internal graph data. Default
value:
dict(source=’source’, target=’target’, name=’id’,
key=’key’, link=’links’)
Deprecated since version 2.8.6: The attrs keyword argument will be replaced with the in-
dividual keywords: source, target, name, key and link. in networkx 3.2.
If the attrs keyword and the new keywords are both used in a single function call (not
recommended) the attrs keyword argument will take precedence.
The values of the keywords must be unique.
source
[string] A string that provides the ‘source’ attribute name for storing NetworkX-internal graph
data.
9.7. JSON 963

target
[string] A string that provides the ‘target’ attribute name for storing NetworkX-internal graph
data.
name
[string] A string that provides the ‘name’ attribute name for storing NetworkX-internal graph
data.
key
[string] A string that provides the ‘key’ attribute name for storing NetworkX-internal graph
data.
link
[string] A string that provides the ‘link’ attribute name for storing NetworkX-internal graph
data.
Returns
G
[NetworkX graph] A NetworkX graph object
See also:
node_link_data, adjacency_data, tree_data
Notes
Attribute ‘key’ is only used for multigraphs.

To use node_link_data in conjunction with node_link_graph, the keyword names for the attributes must
match.
Examples
Create data in node-link format by converting a graph.
>>> G = nx.Graph([('A', 'B')])

>>> data = nx.node_link_data(G)
>>> data
{'directed': False, 'multigraph': False, 'graph': {}, 'nodes': [{'id': 'A'}, {'id
,→': 'B'}], 'links': [{'source': 'A', 'target': 'B'}]}
Revert data in node-link format to a graph.
>>> H = nx.node_link_graph(data)
>>> print(H.edges)
[('A', 'B')]
To serialize and deserialize a graph with JSON,
>>> import json

>>> d = json.dumps(node_link_data(G))
>>> H = node_link_graph(json.loads(d))
>>> print(G.edges, H.edges)
[('A', 'B')] [('A', 'B')]

9.7.3 adjacency_data
adjacency_data(G, attrs={'id': 'id', 'key': 'key'})

Returns data in adjacency format that is suitable for JSON serialization and use in Javascript documents.
Parameters
G
[NetworkX graph]
attrs
[dict] A dictionary that contains two keys ‘id’ and ‘key’. The corresponding values provide
the attribute names for storing NetworkX-internal graph data. The values should be unique.
Default value: dict(id='id', key='key').
If some user-defined graph data use these attribute names as data keys, they may be silently
dropped.
Returns
data
[dict] A dictionary with adjacency formatted data.
Raises
NetworkXError
If values in attrs are not unique.
See also:
adjacency_graph, node_link_data, tree_data
Notes
Graph, node, and link attributes will be written when using this format but attribute keys must be strings if you
want to serialize the resulting data with JSON.
The default value of attrs will be changed in a future release of NetworkX.
Examples
>>> from networkx.readwrite import json_graph

>>> G = nx.Graph([(1, 2)])
>>> data = json_graph.adjacency_data(G)
To serialize with json
>>> import json

>>> s = json.dumps(data)
9.7. JSON 965

9.7.4 adjacency_graph
adjacency_graph(data, directed=False, multigraph=True, attrs={'id': 'id', 'key': 'key'})

Returns graph from adjacency data format.
Parameters
data
[dict] Adjacency list formatted graph data
directed
[bool] If True, and direction not specified in data, return a directed graph.
multigraph
[bool] If True, and multigraph not specified in data, return a multigraph.
attrs
[dict] A dictionary that contains two keys ‘id’ and ‘key’. The corresponding values provide
the attribute names for storing NetworkX-internal graph data. The values should be unique.
Default value: dict(id='id', key='key').
Returns
G
[NetworkX graph] A NetworkX graph object
See also:
adjacency_graph, node_link_data, tree_data
Notes
The default value of attrs will be changed in a future release of NetworkX.
Examples

>>> G = nx.Graph([(1, 2)])
>>> data = json_graph.adjacency_data(G)
>>> H = json_graph.adjacency_graph(data)
9.7.5 cytoscape_data
cytoscape_data(G, name='name', ident='id')

Returns data in Cytoscape JSON format (cyjs).
Parameters
G
[NetworkX Graph] The graph to convert to cytoscape format
name
[string] A string which is mapped to the ‘name’ node element in cyjs format. Must not have
the same value as ident.

ident
[string] A string which is mapped to the ‘id’ node element in cyjs format. Must not have the
same value as name.
Returns
data: dict
A dictionary with cyjs formatted data.
Raises
NetworkXError
If the values for name and ident are identical.
See also:
cytoscape_graph
convert a dictionary in cyjs format to a graph
References
[1]
Examples
>>> nx.cytoscape_data(G)
{'data': [],
'directed': False,
'multigraph': False,
'elements': {'nodes': [{'data': {'id': '0', 'value': 0, 'name': '0'}},
{'data': {'id': '1', 'value': 1, 'name': '1'}}],
'edges': [{'data': {'source': 0, 'target': 1}}]}}
9.7.6 cytoscape_graph
cytoscape_graph(data, name='name', ident='id')

Create a NetworkX graph from a dictionary in cytoscape JSON format.
Parameters
data
[dict] A dictionary of data conforming to cytoscape JSON format.
name
[string] A string which is mapped to the ‘name’ node element in cyjs format. Must not have
the same value as ident.
ident
[string] A string which is mapped to the ‘id’ node element in cyjs format. Must not have the
same value as name.
Returns
graph
[a NetworkX graph instance] The graph can be an instance of Graph, DiGraph, Multi-
Graph, or MultiDiGraph depending on the input data.
9.7. JSON 967

Raises
NetworkXError
If the name and ident attributes are identical.
See also:
cytoscape_data
convert a NetworkX graph to a dict in cyjs format
References
[1]
Examples
>>> data_dict = {
... 'data': [],
... 'directed': False,
... 'multigraph': False,
... 'elements': {'nodes': [{'data': {'id': '0', 'value': 0, 'name': '0'}},
... {'data': {'id': '1', 'value': 1, 'name': '1'}}],
... 'edges': [{'data': {'source': 0, 'target': 1}}]}
... }
>>> G = nx.cytoscape_graph(data_dict)
>>> G.name
''
>>> G.nodes()
NodeView((0, 1))
>>> G.nodes(data=True)[0]
{'id': '0', 'value': 0, 'name': '0'}
EdgeDataView([(0, 1, {'source': 0, 'target': 1})])
9.7.7 tree_data
tree_data(G, root, ident='id', children='children')

Returns data in tree format that is suitable for JSON serialization and use in Javascript documents.
Parameters
G
[NetworkX graph] G must be an oriented tree
root
[node] The root of the tree
ident
[string] Attribute name for storing NetworkX-internal graph data. ident must have a differ-
ent value than children. The default is ‘id’.
children
[string] Attribute name for storing NetworkX-internal graph data. children must have a
different value than ident. The default is ‘children’.
Returns

data
[dict] A dictionary with node-link formatted data.
Raises
NetworkXError
If children and ident attributes are identical.
See also:
tree_graph, node_link_data, adjacency_data
Notes
Node attributes are stored in this format but keys for attributes must be strings if you want to serialize with JSON.
Graph and edge attributes are not stored.
Examples

>>> G = nx.DiGraph([(1, 2)])
>>> data = json_graph.tree_data(G, root=1)
To serialize with json
>>> import json

>>> s = json.dumps(data)
9.7.8 tree_graph
tree_graph(data, ident='id', children='children')

Returns graph from tree data format.
Parameters
data
[dict] Tree formatted graph data
ident
[string] Attribute name for storing NetworkX-internal graph data. ident must have a differ-
ent value than children. The default is ‘id’.
children
[string] Attribute name for storing NetworkX-internal graph data. children must have a
different value than ident. The default is ‘children’.
Returns
G
[NetworkX DiGraph]
See also:
tree_data, node_link_data, adjacency_data
9.7. JSON 969

Examples

>>> G = nx.DiGraph([(1, 2)])
>>> data = json_graph.tree_data(G, root=1)
>>> H = json_graph.tree_graph(data)
9.8 LEDA
Read graphs in LEDA format.

LEDA is a C++ class library for efficient data types and algorithms.
9.8.1 Format
See http://www.algorithmic-solutions.info/leda_guide/graphs/leda_native_graph_fileformat.html
read_leda(path[, encoding]) Read graph in LEDA format from path.

parse_leda(lines) Read graph in LEDA format from string or iterable.
9.8.2 read_leda
read_leda(path, encoding='UTF-8')
Read graph in LEDA format from path.
Parameters
path
[file or string] File or filename to read. Filenames ending in .gz or .bz2 will be uncompressed.
Returns
G
[NetworkX graph]
References
[1]
Examples
G=nx.read_leda(‘file.leda’)

9.8.3 parse_leda
parse_leda(lines)
Read graph in LEDA format from string or iterable.
Parameters
lines
[string or iterable] Data in LEDA format.
Returns
G
[NetworkX graph]
References
[1]
Examples
G=nx.parse_leda(string)
9.9 SparseGraph6
Functions for reading and writing graphs in the graph6 or sparse6 file formats.
According to the author of these formats,
graph6 and sparse6 are formats for storing undirected graphs in a compact manner, using only printable
ASCII characters. Files in these formats have text type and contain one line per graph.
graph6 is suitable for small graphs, or large dense graphs. sparse6 is more space-efficient for large sparse
graphs.
---graph6 and sparse6 homepage
9.9.1 Graph6
Functions for reading and writing graphs in the graph6 format.

The graph6 file format is suitable for small graphs or large dense graphs. For large sparse graphs, use the sparse6 format.
For more information, see the graph6 homepage.
from_graph6_bytes(bytes_in) Read a simple undirected graph in graph6 format from

bytes.
read_graph6(path) Read simple undirected graphs in graph6 format from
path.
to_graph6_bytes(G[, nodes, header]) Convert a simple undirected graph to bytes in graph6 for-
mat.
write_graph6(G, path[, nodes, header]) Write a simple undirected graph to a path in graph6 for-
mat.
9.9. SparseGraph6 971

from_graph6_bytes
from_graph6_bytes(bytes_in)
Read a simple undirected graph in graph6 format from bytes.
Parameters
bytes_in
[bytes] Data in graph6 format, without a trailing newline.
Returns
G
[Graph]
Raises
NetworkXError
If bytes_in is unable to be parsed in graph6 format
ValueError
If any character c in bytes_in does not satisfy 63 <= ord(c) < 127.
See also:
read_graph6, write_graph6
References
[1]
Examples
>>> G = nx.from_graph6_bytes(b"A_")
[(0, 1)]
read_graph6
read_graph6(path)
Read simple undirected graphs in graph6 format from path.
Parameters
path
[file or string] File or filename to write.
Returns
G
[Graph or list of Graphs] If the file contains multiple lines then a list of graphs is returned
Raises
NetworkXError
If the string is unable to be parsed in graph6 format
See also:

from_graph6_bytes, write_graph6
References
[1]
Examples
You can read a graph6 file by giving the path to the file:
>>> import tempfile
>>> with tempfile.NamedTemporaryFile(delete=False) as f:
... _ = f.write(b">>graph6<<A_\n")
... _ = f.seek(0)
... G = nx.read_graph6(f.name)
>>> list(G.edges())
[(0, 1)]
You can also read a graph6 file by giving an open file-like object:
>>> import tempfile
>>> with tempfile.NamedTemporaryFile() as f:
... _ = f.write(b">>graph6<<A_\n")
... _ = f.seek(0)
... G = nx.read_graph6(f)
>>> list(G.edges())
[(0, 1)]
to_graph6_bytes
to_graph6_bytes(G, nodes=None, header=True)

Convert a simple undirected graph to bytes in graph6 format.
Parameters
G
[Graph (undirected)]
nodes: list or iterable
Nodes are labeled 0…n-1 in the order provided. If None the ordering given by G.nodes()
is used.
header: bool
If True add ‘>>graph6<<’ bytes to head of data.
Raises
ValueError
If the graph has at least 2 ** 36 nodes; the graph6 format is only defined for graphs of order
less than 2 ** 36.
See also:
from_graph6_bytes, read_graph6, write_graph6_bytes

Notes
The returned bytes end with a newline character.

The format does not support edge or node labels, parallel edges or self loops. If self loops are present they are
silently ignored.
References
[1]
Examples
>>> nx.to_graph6_bytes(nx.path_graph(2))
b'>>graph6<<A_\n'
write_graph6
write_graph6(G, path, nodes=None, header=True)

Write a simple undirected graph to a path in graph6 format.
Parameters
G
path
[str] The path naming the file to which to write the graph.
is used.
header: bool
If True add ‘>>graph6<<’ string to head of data
Raises
ValueError
If the graph has at least 2 ** 36 nodes; the graph6 format is only defined for graphs of order
less than 2 ** 36.
See also:
from_graph6_bytes, read_graph6

Notes
The function writes a newline character after writing the encoding of the graph.
The format does not support edge or node labels, parallel edges or self loops. If self loops are present they are
silently ignored.
References
[1]
Examples
You can write a graph6 file by giving the path to a file:

>>> import tempfile
... nx.write_graph6(nx.path_graph(2), f.name)
... _ = f.seek(0)
... print(f.read())
b'>>graph6<<A_\n'
9.9.2 Sparse6
Functions for reading and writing graphs in the sparse6 format.

The sparse6 file format is a space-efficient format for large sparse graphs. For small graphs or large dense graphs, use the
graph6 file format.
For more information, see the sparse6 homepage.
from_sparse6_bytes(string) Read an undirected graph in sparse6 format from string.

read_sparse6(path) Read an undirected graph in sparse6 format from path.
to_sparse6_bytes(G[, nodes, header]) Convert an undirected graph to bytes in sparse6 format.
write_sparse6(G, path[, nodes, header]) Write graph G to given path in sparse6 format.
from_sparse6_bytes
from_sparse6_bytes(string)
Read an undirected graph in sparse6 format from string.
Parameters
string
[string] Data in sparse6 format
Returns
G
[Graph]
Raises
NetworkXError
If the string is unable to be parsed in sparse6 format

See also:
read_sparse6, write_sparse6
References
[1]
Examples
>>> G = nx.from_sparse6_bytes(b":A_")
[(0, 1), (0, 1), (0, 1)]
read_sparse6
read_sparse6(path)
Read an undirected graph in sparse6 format from path.
Parameters
path
[file or string] File or filename to write.
Returns
G
[Graph/Multigraph or list of Graphs/MultiGraphs] If the file contains multiple lines then a list
of graphs is returned
Raises
NetworkXError
If the string is unable to be parsed in sparse6 format
See also:
read_sparse6, from_sparse6_bytes
References
[1]
Examples
You can read a sparse6 file by giving the path to the file:
>>> import tempfile

... _ = f.write(b">>sparse6<<:An\n")
... _ = f.seek(0)
... G = nx.read_sparse6(f.name)


>>> list(G.edges())
[(0, 1)]
You can also read a sparse6 file by giving an open file-like object:
>>> import tempfile

... _ = f.write(b">>sparse6<<:An\n")
... _ = f.seek(0)
... G = nx.read_sparse6(f)
>>> list(G.edges())
[(0, 1)]
to_sparse6_bytes
to_sparse6_bytes(G, nodes=None, header=True)

Convert an undirected graph to bytes in sparse6 format.
Parameters
G
is used.
header: bool
If True add ‘>>sparse6<<’ bytes to head of data.
Raises
If the graph is directed.
ValueError
If the graph has at least 2 ** 36 nodes; the sparse6 format is only defined for graphs of order
less than 2 ** 36.
See also:
to_sparse6_bytes, read_sparse6, write_sparse6_bytes
Notes
The returned bytes end with a newline character.

The format does not support edge or node labels.

References
[1]
Examples
>>> nx.to_sparse6_bytes(nx.path_graph(2))
b'>>sparse6<<:An\n'
write_sparse6
write_sparse6(G, path, nodes=None, header=True)

Write graph G to given path in sparse6 format.
Parameters
G
path
[file or string] File or filename to write
Nodes are labeled 0…n-1 in the order provided. If None the ordering given by G.nodes() is
used.
header: bool
If True add ‘>>sparse6<<’ string to head of data
Raises
NetworkXError
If the graph is directed
See also:
read_sparse6, from_sparse6_bytes
Notes
The format does not support edge or node labels.
References
[1]

Examples
You can write a sparse6 file by giving the path to the file:
>>> import tempfile

... nx.write_sparse6(nx.path_graph(2), f.name)
... print(f.read())
b'>>sparse6<<:An\n'
You can also write a sparse6 file by giving an open file-like object:

... nx.write_sparse6(nx.path_graph(2), f)
... _ = f.seek(0)
... print(f.read())
b'>>sparse6<<:An\n'
9.10 Pajek
Read graphs in Pajek format.

This implementation handles directed and undirected graphs including those with self loops and parallel edges.
9.10.1 Format
See http://vlado.fmf.uni-lj.si/pub/networks/pajek/doc/draweps.htm for format information.
read_pajek(path[, encoding]) Read graph in Pajek format from path.

write_pajek(G, path[, encoding]) Write graph in Pajek format to path.
parse_pajek(lines) Parse Pajek format graph from string or iterable.
generate_pajek(G) Generate lines in Pajek graph format.
9.10.2 read_pajek
read_pajek(path, encoding='UTF-8')
Read graph in Pajek format from path.
Parameters
path
[file or string] File or filename to write. Filenames ending in .gz or .bz2 will be uncompressed.
Returns
G
[NetworkX MultiGraph or MultiDiGraph.]
9.10. Pajek 979

References
Examples
>>> nx.write_pajek(G, "test.net")
>>> G = nx.read_pajek("test.net")
To create a Graph instead of a MultiGraph use

>>> G1 = nx.Graph(G)
9.10.3 write_pajek
write_pajek(G, path, encoding='UTF-8')

Write graph in Pajek format to path.
Parameters
G
[graph] A Networkx graph
path
Warning: Optional node attributes and edge attributes must be non-empty strings. Otherwise it will not be
written into the file. You will need to convert those attributes to strings if you want to keep them.
References
Examples
>>> nx.write_pajek(G, "test.net")
9.10.4 parse_pajek
parse_pajek(lines)
Parse Pajek format graph from string or iterable.
Parameters
lines
[string or iterable] Data in Pajek format.
Returns

G
[NetworkX graph]
See also:
read_pajek
9.10.5 generate_pajek
generate_pajek(G)
Generate lines in Pajek graph format.
Parameters
G
[graph] A Networkx graph
References
9.11 Matrix Market
The Matrix Market exchange format is a text-based file format described by NIST. Matrix Market supports both a co-
ordinate format for sparse matrices and an array format for dense matrices. The scipy.io module provides the
scipy.io.mmread and scipy.io.mmwrite functions to read and write data in Matrix Market format, respec-
tively. These functions work with either numpy.ndarray or scipy.sparse.coo_matrix objects depending
on whether the data is in array or coordinate format. These functions can be combined with those of NetworkX’s
convert_matrix module to read and write Graphs in Matrix Market format.
9.11.1 Examples
Reading and writing graphs using Matrix Market’s array format for dense matrices:

>>> import scipy.io # for mmread() and mmwrite()
>>> import io # Use BytesIO as a stand-in for a Python file object
>>> fh = io.BytesIO()
>>> a = nx.to_numpy_array(G)
>>> print(a)
[[0. 1. 1. 1. 1.]
[1. 0. 1. 1. 1.]
[1. 1. 0. 1. 1.]
[1. 1. 1. 0. 1.]
[1. 1. 1. 1. 0.]]
>>> # Write to file in Matrix Market array format

>>> sp.io.mmwrite(fh, a)
>>> print(fh.getvalue().decode('utf-8')) # file contents
9.11. Matrix Market 981


%%MatrixMarket matrix array real symmetric
%
5 5
0.0000000000000000e+00
1.0000000000000000e+00
1.0000000000000000e+00
1.0000000000000000e+00
1.0000000000000000e+00
0.0000000000000000e+00
1.0000000000000000e+00
1.0000000000000000e+00
1.0000000000000000e+00
0.0000000000000000e+00
1.0000000000000000e+00
1.0000000000000000e+00
0.0000000000000000e+00
1.0000000000000000e+00
0.0000000000000000e+00
>>> # Read from file

>>> fh.seek(0)
>>> H = nx.from_numpy_array(sp.io.mmread(fh))
>>> H.edges() == G.edges()
True
Reading and writing graphs using Matrix Market’s coordinate format for sparse matrices:
>>> import scipy.io # for mmread() and mmwrite()
>>> import io # Use BytesIO as a stand-in for a Python file object
>>> fh = io.BytesIO()
>>> m = nx.to_scipy_sparse_array(G)
>>> print(m)
(0, 1) 1
(1, 0) 1
(1, 2) 1
(2, 1) 1
(2, 3) 1
(3, 2) 1
(3, 4) 1
(4, 3) 1
>>> sp.io.mmwrite(fh, m)
>>> print(fh.getvalue().decode('utf-8')) # file contents
%%MatrixMarket matrix coordinate integer symmetric
%
5 5 4
2 1 1
3 2 1
4 3 1
5 4 1
>>> # Read from file

>>> fh.seek(0)
>>> H = nx.from_scipy_sparse_array(sp.io.mmread(fh))


>>> H.edges() == G.edges()
True
9.11. Matrix Market 983


CHAPTER
TEN
DRAWING
NetworkX provides basic functionality for visualizing graphs, but its main goal is to enable graph analysis rather than
perform graph visualization. In the future, graph visualization functionality may be removed from NetworkX or only
available as an add-on package.
Proper graph visualization is hard, and we highly recommend that people visualize their graphs with tools dedicated to
that task. Notable examples of dedicated and fully-featured graph visualization tools are Cytoscape, Gephi, Graphviz
and, for LaTeX typesetting, PGF/TikZ. To use these and other such tools, you should export your NetworkX graph into
a format that can be read by those tools. For example, Cytoscape can read the GraphML format, and so, networkx.
write_graphml(G, path) might be an appropriate choice.
More information on the features provided here are available at
• matplotlib: http://matplotlib.org/
• pygraphviz: http://pygraphviz.github.io/
10.1 Matplotlib
Draw networks with matplotlib.
10.1.1 Examples
>>> nx.draw(G)
10.1.2 See Also
• matplotlib
• matplotlib.pyplot.scatter()
• matplotlib.patches.FancyArrowPatch
985
draw(G[, pos, ax]) Draw the graph G with Matplotlib.

draw_networkx(G[, pos, arrows, with_labels]) Draw the graph G using Matplotlib.
draw_networkx_nodes(G, pos[, nodelist, ...]) Draw the nodes of the graph G.
draw_networkx_edges(G, pos[, edgelist, ...]) Draw the edges of the graph G.
draw_networkx_labels(G, pos[, labels, ...]) Draw node labels on the graph G.
draw_networkx_edge_labels(G, pos[, ...]) Draw edge labels.
draw_circular(G, **kwargs) Draw the graph G with a circular layout.
draw_kamada_kawai(G, **kwargs) Draw the graph G with a Kamada-Kawai force-directed
layout.
draw_planar(G, **kwargs) Draw a planar networkx graph G with planar layout.
draw_random(G, **kwargs) Draw the graph G with a random layout.
draw_spectral(G, **kwargs) Draw the graph G with a spectral 2D layout.
draw_spring(G, **kwargs) Draw the graph G with a spring layout.
draw_shell(G[, nlist]) Draw networkx graph G with shell layout.
10.1.3 draw
draw(G, pos=None, ax=None, **kwds)

Draw the graph G with Matplotlib.
Draw the graph as a simple representation with no node labels or edge labels and using the full Matplotlib figure
area and no axis labels by default. See draw_networkx() for more full-featured drawing that allows title, axis labels
etc.
Parameters
G
pos
[dictionary, optional] A dictionary with nodes as keys and positions as values. If not speci-
fied a spring layout positioning will be computed. See networkx.drawing.layout for
functions that compute node positions.
ax
[Matplotlib Axes object, optional] Draw the graph in specified Matplotlib axes.
kwds
[optional keywords] See networkx.draw_networkx() for a description of optional keywords.
See also:
draw_networkx
draw_networkx_nodes
draw_networkx_edges
draw_networkx_labels
draw_networkx_edge_labels
986 Chapter 10. Drawing

Notes
This function has the same name as pylab.draw and pyplot.draw so beware when using from networkx im-
port *
since you might overwrite the pylab.draw function.
With pyplot use

>>> nx.draw(G) # networkx draw()
>>> plt.draw() # pyplot draw()
Also see the NetworkX drawing examples at https://networkx.org/documentation/latest/auto_examples/index.

html
Examples
>>> nx.draw(G)
>>> nx.draw(G, pos=nx.spring_layout(G)) # use spring layout
10.1.4 draw_networkx
draw_networkx(G, pos=None, arrows=None, with_labels=True, **kwds)

Draw the graph G using Matplotlib.
Draw the graph with Matplotlib with options for node positions, labeling, titles, and many other drawing features.
See draw() for simple drawing without labels or axes.
Parameters
G
pos
[dictionary, optional] A dictionary with nodes as keys and positions as values. If not speci-
fied a spring layout positioning will be computed. See networkx.drawing.layout for
functions that compute node positions.
arrows
[bool or None, optional (default=None)] If None, directed graphs draw arrowheads with Fan-
cyArrowPatch, while undirected graphs draw edges via LineCollection for speed.
If True, draw arrowheads with FancyArrowPatches (bendable and stylish). If False, draw
edges using LineCollection (linear and fast). For directed graphs, if True draw arrowheads.
Note: Arrows will be the same color as edges.
arrowstyle
[str (default=’-|>’ for directed graphs)] For directed graphs, choose the style of the arrowsheads.
For undirected graphs default to ‘-’
See matplotlib.patches.ArrowStyle for more options.
arrowsize
[int or list (default=10)] For directed graphs, choose the size of the arrow head’s length
and width. A list of values can be passed in to assign a different size for arrow head’s
10.1. Matplotlib 987

length and width. See matplotlib.patches.FancyArrowPatch for attribute mu-

tation_scale for more info.
with_labels
[bool (default=True)] Set to True to draw labels on the nodes.
ax
[Matplotlib Axes object, optional] Draw the graph in the specified Matplotlib axes.
nodelist
[list (default=list(G))] Draw only specified nodes
edgelist
[list (default=list(G.edges()))] Draw only specified edges
node_size
[scalar or array (default=300)] Size of nodes. If an array is specified it must be the same length
as nodelist.
node_color
[color or array of colors (default=’#1f78b4’)] Node color. Can be a single color or a sequence
of colors with the same length as nodelist. Color can be string or rgb (or rgba) tuple of floats
from 0-1. If numeric values are specified they will be mapped to colors using the cmap and
vmin,vmax parameters. See matplotlib.scatter for more details.
node_shape
[string (default=’o’)] The shape of the node. Specification is as matplotlib.scatter marker, one
of ‘so^>v<dph8’.
alpha
[float or None (default=None)] The node and edge transparency
cmap
[Matplotlib colormap, optional] Colormap for mapping intensities of nodes
vmin,vmax
[float, optional] Minimum and maximum for node colormap scaling
linewidths
[scalar or sequence (default=1.0)] Line width of symbol border
width
[float or array of floats (default=1.0)] Line width of edges
edge_color
[color or array of colors (default=’k’)] Edge color. Can be a single color or a sequence of
colors with the same length as edgelist. Color can be string or rgb (or rgba) tuple of floats from
0-1. If numeric values are specified they will be mapped to colors using the edge_cmap and
edge_vmin,edge_vmax parameters.
edge_cmap
[Matplotlib colormap, optional] Colormap for mapping intensities of edges
edge_vmin,edge_vmax
[floats, optional] Minimum and maximum for edge colormap scaling
style
[string (default=solid line)] Edge line style e.g.: ‘-’, ‘–’, ‘-.’, ‘:’ or words like ‘solid’ or ‘dashed’.
(See matplotlib.patches.FancyArrowPatch: linestyle)
labels
[dictionary (default=None)] Node labels in a dictionary of text labels keyed by node

font_size
[int (default=12 for nodes, 10 for edges)] Font size for text labels
font_color
[string (default=’k’ black)] Font color string
font_weight
[string (default=’normal’)] Font weight
font_family
[string (default=’sans-serif’)] Font family
label
[string, optional] Label for graph legend
kwds
[optional keywords] See networkx.draw_networkx_nodes(), net-
workx.draw_networkx_edges(), and networkx.draw_networkx_labels() for a description
of optional keywords.
See also:
draw
draw_networkx_nodes
draw_networkx_edges
Notes
For directed graphs, arrows are drawn at the head end. Arrows can be turned off with keyword arrows=False.
Examples
>>> nx.draw(G)
>>> nx.draw(G, pos=nx.spring_layout(G)) # use spring layout

>>> limits = plt.axis("off") # turn off axis

html
10.1.5 draw_networkx_nodes
draw_networkx_nodes(G, pos, nodelist=None, node_size=300, node_color='#1f78b4', node_shape='o',

alpha=None, cmap=None, vmin=None, vmax=None, ax=None, linewidths=None,
edgecolors=None, label=None, margins=None)
Draw the nodes of the graph G.
This draws only the nodes of the graph G.
Parameters

G
pos
[dictionary] A dictionary with nodes as keys and positions as values. Positions should be se-
quences of length 2.
ax
nodelist
[list (default list(G))] Draw only specified nodes
node_size
[scalar or array (default=300)] Size of nodes. If an array it must be the same length as nodelist.
node_color
[color or array of colors (default=’#1f78b4’)] Node color. Can be a single color or a sequence
of colors with the same length as nodelist. Color can be string or rgb (or rgba) tuple of floats
from 0-1. If numeric values are specified they will be mapped to colors using the cmap and
vmin,vmax parameters. See matplotlib.scatter for more details.
node_shape
[string (default=’o’)] The shape of the node. Specification is as matplotlib.scatter marker, one
of ‘so^>v<dph8’.
alpha
[float or array of floats (default=None)] The node transparency. This can be a single alpha
value, in which case it will be applied to all the nodes of color. Otherwise, if it is an array, the
elements of alpha will be applied to the colors in order (cycling through alpha multiple times
if necessary).
cmap
[Matplotlib colormap (default=None)] Colormap for mapping intensities of nodes
vmin,vmax
[floats or None (default=None)] Minimum and maximum for node colormap scaling
linewidths
[[None | scalar | sequence] (default=1.0)] Line width of symbol border
edgecolors
[[None | scalar | sequence] (default = node_color)] Colors of node borders
label
[[None | string]] Label for legend
margins
[float or 2-tuple, optional] Sets the padding for axis autoscaling. Increase margin to prevent
clipping for nodes that are near the edges of an image. Values should be in the range [0, 1].
See matplotlib.axes.Axes.margins() for details. The default is None, which uses
the Matplotlib default.
Returns
matplotlib.collections.PathCollection
PathCollection of the nodes.
See also:

draw
draw_networkx
draw_networkx_edges
Examples
>>> nodes = nx.draw_networkx_nodes(G, pos=nx.spring_layout(G))

html
10.1.6 draw_networkx_edges
draw_networkx_edges(G, pos, edgelist=None, width=1.0, edge_color='k', style='solid', alpha=None,

arrowstyle=None, arrowsize=10, edge_cmap=None, edge_vmin=None,
edge_vmax=None, ax=None, arrows=None, label=None, node_size=300, nodelist=None,
node_shape='o', connectionstyle='arc3', min_source_margin=0, min_target_margin=0)
Draw the edges of the graph G.
This draws only the edges of the graph G.
Parameters
G
pos
edgelist
[collection of edge tuples (default=G.edges())] Draw only specified edges
width
[float or array of floats (default=1.0)] Line width of edges
edge_color
[color or array of colors (default=’k’)] Edge color. Can be a single color or a sequence of
colors with the same length as edgelist. Color can be string or rgb (or rgba) tuple of floats from
0-1. If numeric values are specified they will be mapped to colors using the edge_cmap and
edge_vmin,edge_vmax parameters.
style
[string or array of strings (default=’solid’)] Edge line style e.g.: ‘-’, ‘–’, ‘-.’, ‘:’ or words like ‘solid’
or ‘dashed’. Can be a single style or a sequence of styles with the same length as the edge list. If
less styles than edges are given the styles will cycle. If more styles than edges are given the styles
will be used sequentially and not be exhausted. Also, (offset, onoffseq) tuples can
be used as style instead of a strings. (See matplotlib.patches.FancyArrowPatch:
linestyle)
alpha
[float or array of floats (default=None)] The edge transparency. This can be a single alpha
value, in which case it will be applied to all specified edges. Otherwise, if it is an array, the

elements of alpha will be applied to the colors in order (cycling through alpha multiple times
if necessary).
edge_cmap
[Matplotlib colormap, optional] Colormap for mapping intensities of edges
edge_vmin,edge_vmax
[floats, optional] Minimum and maximum for edge colormap scaling
ax
arrows
[bool or None, optional (default=None)] If None, directed graphs draw arrowheads with Fan-
cyArrowPatch, while undirected graphs draw edges via LineCollection for speed.
If True, draw arrowheads with FancyArrowPatches (bendable and stylish). If False, draw
edges using LineCollection (linear and fast).
Note: Arrowheads will be the same color as edges.
arrowstyle
[str (default=’-|>’ for directed graphs)] For directed graphs and arrows==True defaults to
‘-|>’, For undirected graphs default to ‘-‘.
See matplotlib.patches.ArrowStyle for more options.
arrowsize
[int (default=10)] For directed graphs, choose the size of the arrow head’s length and width.
See matplotlib.patches.FancyArrowPatch for attribute mutation_scale
for more info.
connectionstyle
[string (default=”arc3”)] Pass the connectionstyle parameter to create curved arc of round-
ing radius rad. For example, connectionstyle=’arc3,rad=0.2’. See matplotlib.
patches.ConnectionStyle and matplotlib.patches.FancyArrowPatch
for more info.
node_size
[scalar or array (default=300)] Size of nodes. Though the nodes are not drawn with this func-
tion, the node size is used in determining edge positioning.
nodelist
[list, optional (default=G.nodes())] This provides the node order for the node_size array (if
it is an array).
node_shape
[string (default=’o’)] The marker used for nodes, used in determining edge positioning. Spec-
ification is as a matplotlib.markers marker, e.g. one of ‘so^>v<dph8’.
label
[None or string] Label for legend
min_source_margin
[int (default=0)] The minimum margin (gap) at the begining of the edge at the source.
min_target_margin
[int (default=0)] The minimum margin (gap) at the end of the edge at the target.
Returns
matplotlib.colections.LineCollection or a list of matplotlib.patches.FancyArrowPatch
If arrows=True, a list of FancyArrowPatches is returned. If arrows=False, a LineCol-

lection is returned. If arrows=None (the default), then a LineCollection is returned if G is

undirected, otherwise returns a list of FancyArrowPatches.
See also:
draw
draw_networkx
draw_networkx_nodes
Notes
For directed graphs, arrows are drawn at the head end. Arrows can be turned off with keyword arrows=False or by
passing an arrowstyle without an arrow on the end.
Be sure to include node_size as a keyword argument; arrows are drawn considering the size of nodes.
Self-loops are always drawn with FancyArrowPatch regardless of the value of arrows or whether G is di-
rected. When arrows=False or arrows=None and G is undirected, the FancyArrowPatches corresponding
to the self-loops are not explicitly returned. They should instead be accessed via the Axes.patches attribute
(see examples).
Examples
>>> edges = nx.draw_networkx_edges(G, pos=nx.spring_layout(G))
>>> G.add_edges_from([(1, 2), (1, 3), (2, 3)])
>>> arcs = nx.draw_networkx_edges(G, pos=nx.spring_layout(G))
>>> alphas = [0.3, 0.4, 0.5]
>>> for i, arc in enumerate(arcs): # change alpha values of arcs
... arc.set_alpha(alphas[i])
The FancyArrowPatches corresponding to self-loops are not always returned, but can always be accessed via the
patches attribute of the matplotlib.Axes object.

>>> fig, ax = plt.subplots()
>>> G = nx.Graph([(0, 1), (0, 0)]) # Self-loop at node 0
>>> edge_collection = nx.draw_networkx_edges(G, pos=nx.circular_layout(G), ax=ax)
>>> self_loop_fap = ax.patches[0]

html

10.1.7 draw_networkx_labels
draw_networkx_labels(G, pos, labels=None, font_size=12, font_color='k', font_family='sans-serif',

font_weight='normal', alpha=None, bbox=None, horizontalalignment='center',
verticalalignment='center', ax=None, clip_on=True)
Draw node labels on the graph G.
Parameters
G
pos
labels
[dictionary (default={n: n for n in G})] Node labels in a dictionary of text labels keyed by node.
Node-keys in labels should appear as keys in pos. If needed use: {n:lab for n,lab
in labels.items() if n in pos}
font_size
[int (default=12)] Font size for text labels
font_color
font_weight
font_family
alpha
[float or None (default=None)] The text transparency
bbox
[Matplotlib bbox, (default is Matplotlib’s ax.text default)] Specify text box properties (e.g.
shape, color etc.) for node labels.
horizontalalignment
[string (default=’center’)] Horizontal alignment {‘center’, ‘right’, ‘left’}
verticalalignment
[string (default=’center’)] Vertical alignment {‘center’, ‘top’, ‘bottom’, ‘baseline’, ‘cen-
ter_baseline’}
ax
clip_on
[bool (default=True)] Turn on clipping of node labels at axis boundaries
Returns
dict
dict of labels keyed on the nodes
See also:

draw
draw_networkx
draw_networkx_nodes
draw_networkx_edges
Examples
>>> labels = nx.draw_networkx_labels(G, pos=nx.spring_layout(G))

html
10.1.8 draw_networkx_edge_labels
draw_networkx_edge_labels(G, pos, edge_labels=None, label_pos=0.5, font_size=10, font_color='k',

font_family='sans-serif', font_weight='normal', alpha=None, bbox=None,
horizontalalignment='center', verticalalignment='center', ax=None, rotate=True,
clip_on=True)
Draw edge labels.
Parameters
G
pos
edge_labels
[dictionary (default=None)] Edge labels in a dictionary of labels keyed by edge two-tuple. Only
labels for the keys in the dictionary are drawn.
label_pos
[float (default=0.5)] Position of edge label along edge (0=head, 0.5=center, 1=tail)
font_size
[int (default=10)] Font size for text labels
font_color
font_weight
font_family
alpha
[float or None (default=None)] The text transparency
bbox
[Matplotlib bbox, optional] Specify text box properties (e.g. shape, color etc.) for edge labels.
Default is {boxstyle=’round’, ec=(1.0, 1.0, 1.0), fc=(1.0, 1.0, 1.0)}.

horizontalalignment
[string (default=’center’)] Horizontal alignment {‘center’, ‘right’, ‘left’}
verticalalignment
[string (default=’center’)] Vertical alignment {‘center’, ‘top’, ‘bottom’, ‘baseline’, ‘cen-
ter_baseline’}
ax
rotate
[bool (deafult=True)] Rotate edge labels to lie parallel to edges
clip_on
[bool (default=True)] Turn on clipping of edge labels at axis boundaries
Returns
dict
dict of labels keyed by edge
See also:
draw
draw_networkx
draw_networkx_nodes
draw_networkx_edges
Examples
>>> edge_labels = nx.draw_networkx_edge_labels(G, pos=nx.spring_layout(G))

html
10.1.9 draw_circular
draw_circular(G, **kwargs)
Draw the graph G with a circular layout.
This is a convenience function equivalent to:
nx.draw(G, pos=nx.circular_layout(G), **kwargs)
Parameters
G
kwargs
[optional keywords] See draw_networkx for a description of optional keywords.
See also:
circular_layout()

Notes
The layout is computed each time this function is called. For repeated drawing it is much more efficient to call
circular_layout directly and reuse the result:
>>> pos = nx.circular_layout(G)
>>> nx.draw(G, pos=pos) # Draw the original graph
>>> # Draw a subgraph, reusing the same node positions
>>> nx.draw(G.subgraph([0, 1, 2]), pos=pos, node_color="red")
10.1.10 draw_kamada_kawai
draw_kamada_kawai(G, **kwargs)
Draw the graph G with a Kamada-Kawai force-directed layout.
nx.draw(G, pos=nx.kamada_kawai_layout(G), **kwargs)
Parameters
G
kwargs
See also:
kamada_kawai_layout()
Notes
kamada_kawai_layout directly and reuse the result:
>>> pos = nx.kamada_kawai_layout(G)
10.1.11 draw_planar
draw_planar(G, **kwargs)
Draw a planar networkx graph G with planar layout.
nx.draw(G, pos=nx.planar_layout(G), **kwargs)

Parameters
G
[graph] A planar networkx graph
kwargs
Raises
NetworkXException
When G is not planar
See also:
planar_layout()
Notes
planar_layout directly and reuse the result:
>>> pos = nx.planar_layout(G)
10.1.12 draw_random
draw_random(G, **kwargs)
Draw the graph G with a random layout.
nx.draw(G, pos=nx.random_layout(G), **kwargs)
Parameters
G
kwargs
See also:
random_layout()

Notes
random_layout directly and reuse the result:
>>> pos = nx.random_layout(G)
10.1.13 draw_spectral
draw_spectral(G, **kwargs)
Draw the graph G with a spectral 2D layout.
nx.draw(G, pos=nx.spectral_layout(G), **kwargs)
For more information about how node positions are determined, see spectral_layout.
Parameters
G
kwargs
See also:
spectral_layout()
Notes
spectral_layout directly and reuse the result:
>>> pos = nx.spectral_layout(G)

10.1.14 draw_spring
draw_spring(G, **kwargs)
Draw the graph G with a spring layout.
nx.draw(G, pos=nx.spring_layout(G), **kwargs)
Parameters
G
kwargs
See also:
draw
spring_layout()
Notes
spring_layout is also the default layout for draw, so this function is equivalent to draw.
spring_layout directly and reuse the result:
>>> pos = nx.spring_layout(G)
10.1.15 draw_shell
draw_shell(G, nlist=None, **kwargs)

Draw networkx graph G with shell layout.
nx.draw(G, pos=nx.shell_layout(G, nlist=nlist), **kwargs)
Parameters
G
nlist
[list of list of nodes, optional] A list containing lists of nodes representing the shells. Default
is None, meaning all nodes are in a single shell. See shell_layout for details.
kwargs

See also:
shell_layout()
Notes
shell_layout directly and reuse the result:
>>> pos = nx.shell_layout(G)
10.2 Graphviz AGraph (dot)
Interface to pygraphviz AGraph class.
10.2.1 Examples
>>> A = nx.nx_agraph.to_agraph(G)
>>> H = nx.nx_agraph.from_agraph(A)
10.2.2 See Also
• Pygraphviz: http://pygraphviz.github.io/
• Graphviz: https://www.graphviz.org
• DOT Language: http://www.graphviz.org/doc/info/lang.html
from_agraph(A[, create_using]) Returns a NetworkX Graph or DiGraph from a Py-

Graphviz graph.
to_agraph(N) Returns a pygraphviz graph from a NetworkX graph N.
write_dot(G, path) Write NetworkX graph G to Graphviz dot format on path.
read_dot(path) Returns a NetworkX graph from a dot file on path.
graphviz_layout(G[, prog, root, args]) Create node positions for G using Graphviz.
pygraphviz_layout(G[, prog, root, args]) Create node positions for G using Graphviz.
10.2. Graphviz AGraph (dot) 1001

10.2.3 from_agraph
from_agraph(A, create_using=None)
Returns a NetworkX Graph or DiGraph from a PyGraphviz graph.
Parameters
A
[PyGraphviz AGraph] A graph created with PyGraphviz
create_using
[NetworkX graph constructor, optional (default=None)] Graph type to create. If graph in-
stance, then cleared before populated. If None, then the appropriate Graph type is inferred
from A.
Notes
The Graph G will have a dictionary G.graph_attr containing the default graphviz attributes for graphs, nodes and
edges.
Default node attributes will be in the dictionary G.node_attr which is keyed by node.
Edge attributes will be returned as edge data in G. With edge_attr=False the edge data will be the Graphviz edge
weight attribute or the value 1 if no edge weight attribute is found.
Examples
>>> A = nx.nx_agraph.to_agraph(K5)
>>> G = nx.nx_agraph.from_agraph(A)
10.2.4 to_agraph
to_agraph(N)
Returns a pygraphviz graph from a NetworkX graph N.
Parameters
N
[NetworkX graph] A graph created with NetworkX
Notes
If N has an dict N.graph_attr an attempt will be made first to copy properties attached to the graph (see from_agraph)
and then updated with the calling arguments if any.

Examples
>>> A = nx.nx_agraph.to_agraph(K5)
10.2.5 write_dot
write_dot(G, path)
Write NetworkX graph G to Graphviz dot format on path.
Parameters
G
path
[filename] Filename or file handle to write
Notes
To use a specific graph layout, call A.layout prior to write_dot. Note that some graphviz layouts are not
guaranteed to be deterministic, see https://gitlab.com/graphviz/graphviz/-/issues/1767 for more info.
10.2.6 read_dot
read_dot(path)
Returns a NetworkX graph from a dot file on path.
Parameters
path
[file or string] File name or file handle to read.
10.2.7 graphviz_layout
graphviz_layout(G, prog='neato', root=None, args='')

Create node positions for G using Graphviz.
Parameters
G
prog
[string] Name of Graphviz layout program
root
[string, optional] Root node for twopi layout
args
[string, optional] Extra arguments to Graphviz layout program
Returns
Dictionary of x, y, positions keyed by node.
10.2. Graphviz AGraph (dot) 1003

Notes
This is a wrapper for pygraphviz_layout.

Note that some graphviz layouts are not guaranteed to be deterministic, see https://gitlab.com/graphviz/graphviz/
-/issues/1767 for more info.
Examples
>>> pos = nx.nx_agraph.graphviz_layout(G)
>>> pos = nx.nx_agraph.graphviz_layout(G, prog="dot")
10.2.8 pygraphviz_layout
pygraphviz_layout(G, prog='neato', root=None, args='')

Create node positions for G using Graphviz.
Parameters
G
prog
[string] Name of Graphviz layout program
root
[string, optional] Root node for twopi layout
args
[string, optional] Extra arguments to Graphviz layout program
Returns
node_pos
[dict] Dictionary of x, y, positions keyed by node.
Notes
If you use complex node objects, they may have the same string representation and GraphViz could treat them as
the same node. The layout may assign both nodes a single location. See Issue #1568 If this occurs in your case,
consider relabeling the nodes just for the layout computation using something similar to:
>>> H = nx.convert_node_labels_to_integers(G, label_attribute="node_label")

>>> H_layout = nx.nx_agraph.pygraphviz_layout(G, prog="dot")
>>> G_layout = {H.nodes[n]["node_label"]: p for n, p in H_layout.items()}
Note that some graphviz layouts are not guaranteed to be deterministic, see https://gitlab.com/graphviz/graphviz/
-/issues/1767 for more info.

Examples
>>> pos = nx.nx_agraph.graphviz_layout(G, prog="dot")
10.3 Graphviz with pydot
Import and export NetworkX graphs in Graphviz dot format using pydot.
Either this module or nx_agraph can be used to interface with graphviz.
10.3.1 Examples
>>> PG = nx.nx_pydot.to_pydot(G)
>>> H = nx.nx_pydot.from_pydot(PG)
10.3.2 See Also
• pydot: https://github.com/erocarrera/pydot
• Graphviz: https://www.graphviz.org
• DOT Language: http://www.graphviz.org/doc/info/lang.html
from_pydot(P) Returns a NetworkX graph from a Pydot graph.

to_pydot(N) Returns a pydot graph from a NetworkX graph N.
write_dot(G, path) Write NetworkX graph G to Graphviz dot format on path.
read_dot(path) Returns a NetworkX MultiGraph or MultiDi-
Graph from the dot file with the passed path.
graphviz_layout(G[, prog, root]) Create node positions using Pydot and Graphviz.
pydot_layout(G[, prog, root]) Create node positions using pydot and Graphviz.
10.3.3 from_pydot
from_pydot(P)
Returns a NetworkX graph from a Pydot graph.
Parameters
P
[Pydot graph] A graph created with Pydot
Returns
G
[NetworkX multigraph] A MultiGraph or MultiDiGraph.
10.3. Graphviz with pydot 1005

Examples
>>> A = nx.nx_pydot.to_pydot(K5)
>>> G = nx.nx_pydot.from_pydot(A) # return MultiGraph
# make a Graph instead of MultiGraph >>> G = nx.Graph(nx.nx_pydot.from_pydot(A))
10.3.4 to_pydot
to_pydot(N)
Returns a pydot graph from a NetworkX graph N.
Parameters
N
Examples
>>> P = nx.nx_pydot.to_pydot(K5)
10.3.5 write_dot
write_dot(G, path)
Write NetworkX graph G to Graphviz dot format on path.
Path can be a string or a file handle.
10.3.6 read_dot
read_dot(path)
Returns a NetworkX MultiGraph or MultiDiGraph from the dot file with the passed path.
If this file contains multiple graphs, only the first such graph is returned. All graphs _except_ the first are silently
ignored.
Parameters
path
[str or file] Filename or file handle.
Returns
G
[MultiGraph or MultiDiGraph] A MultiGraph or MultiDiGraph.

Notes
Use G = nx.Graph(nx.nx_pydot.read_dot(path)) to return a Graph instead of a MultiGraph.
10.3.7 graphviz_layout
graphviz_layout(G, prog='neato', root=None)

Create node positions using Pydot and Graphviz.
Returns a dictionary of positions keyed by node.
Parameters
G
[NetworkX Graph] The graph for which the layout is computed.
prog
[string (default: ‘neato’)] The name of the GraphViz program to use for layout. Options depend
on GraphViz version but may include: ‘dot’, ‘twopi’, ‘fdp’, ‘sfdp’, ‘circo’
root
[Node from G or None (default: None)] The node of G from which to start some layout algo-
rithms.
Returns
Dictionary of (x, y) positions keyed by node.
Notes
This is a wrapper for pydot_layout.
Examples
>>> pos = nx.nx_pydot.graphviz_layout(G)
>>> pos = nx.nx_pydot.graphviz_layout(G, prog="dot")
10.3.8 pydot_layout
pydot_layout(G, prog='neato', root=None)

Create node positions using pydot and Graphviz.
Parameters
G
[Graph] NetworkX graph to be laid out.
prog
[string (default: ‘neato’)] Name of the GraphViz command to use for layout. Options depend
on GraphViz version but may include: ‘dot’, ‘twopi’, ‘fdp’, ‘sfdp’, ‘circo’
root
[Node from G or None (default: None)] The node of G from which to start some layout algo-
rithms.
10.3. Graphviz with pydot 1007

Returns
dict
Dictionary of positions keyed by node.
Notes
If you use complex node objects, they may have the same string representation and GraphViz could treat them as
the same node. The layout may assign both nodes a single location. See Issue #1568 If this occurs in your case,
consider relabeling the nodes just for the layout computation using something similar to:
H = nx.convert_node_labels_to_integers(G, label_attribute='node_label')
H_layout = nx.nx_pydot.pydot_layout(G, prog='dot')
G_layout = {H.nodes[n]['node_label']: p for n, p in H_layout.items()}
Examples
>>> pos = nx.nx_pydot.pydot_layout(G)
>>> pos = nx.nx_pydot.pydot_layout(G, prog="dot")
10.4 Graph Layout
Node positioning algorithms for graph drawing.

For random_layout() the possible resulting shape is a square of side [0, scale] (default: [0, 1]) Changing center
shifts the layout by that amount.
For the other layout routines, the extent is [center - scale, center + scale] (default: [-1, 1]).
Warning: Most layout routines have only been tested in 2-dimensions.
bipartite_layout(G, nodes[, align, scale, ...]) Position nodes in two straight lines.
circular_layout(G[, scale, center, dim]) Position nodes on a circle.
kamada_kawai_layout(G[, dist, pos, weight, ...]) Position nodes using Kamada-Kawai path-length cost-
function.
planar_layout(G[, scale, center, dim]) Position nodes without edge intersections.
random_layout(G[, center, dim, seed]) Position nodes uniformly at random in the unit square.
rescale_layout(pos[, scale]) Returns scaled position array to (-scale, scale) in all axes.
rescale_layout_dict(pos[, scale]) Return a dictionary of scaled positions keyed by node
shell_layout(G[, nlist, rotate, scale, ...]) Position nodes in concentric circles.
spring_layout(G[, k, pos, fixed, ...]) Position nodes using Fruchterman-Reingold force-
directed algorithm.
spectral_layout(G[, weight, scale, center, dim]) Position nodes using the eigenvectors of the graph Lapla-
cian.
spiral_layout(G[, scale, center, dim, ...]) Position nodes in a spiral layout.
multipartite_layout(G[, subset_key, align, ...]) Position nodes in layers of straight lines.

10.4.1 bipartite_layout
bipartite_layout(G, nodes, align='vertical', scale=1, center=None, aspect_ratio=1.3333333333333333)

Position nodes in two straight lines.
Parameters
G
[NetworkX graph or list of nodes] A position will be assigned to every node in G.
nodes
[list or container] Nodes in one node set of the bipartite graph. This set will be placed on left
or top.
align
[string (default=’vertical’)] The alignment of nodes. Vertical or horizontal.
scale
[number (default: 1)] Scale factor for positions.
center
[array-like or None] Coordinate pair around which to center the layout.
aspect_ratio
[number (default=4/3):] The ratio of the width to the height of the layout.
Returns
pos
[dict] A dictionary of positions keyed by node.
Notes
This algorithm currently only works in two dimensions and does not try to minimize edge crossings.
Examples
>>> G = nx.bipartite.gnmk_random_graph(3, 5, 10, seed=123)

>>> top = nx.bipartite.sets(G)[0]
>>> pos = nx.bipartite_layout(G, top)
10.4.2 circular_layout
circular_layout(G, scale=1, center=None, dim=2)

Position nodes on a circle.
Parameters
G
scale
center
10.4. Graph Layout 1009

dim
[int] Dimension of layout. If dim>2, the remaining dimensions are set to zero in the returned
positions. If dim<2, a ValueError is raised.
Returns
pos
[dict] A dictionary of positions keyed by node
Raises
ValueError
If dim < 2
Notes
Examples
>>> pos = nx.circular_layout(G)
10.4.3 kamada_kawai_layout
kamada_kawai_layout(G, dist=None, pos=None, weight='weight', scale=1, center=None, dim=2)

Position nodes using Kamada-Kawai path-length cost-function.
Parameters
G
dist
[dict (default=None)] A two-level dictionary of optimal distances between nodes, indexed by
source and destination node. If None, the distance is computed using shortest_path_length().
pos
[dict or None optional (default=None)] Initial positions for nodes as a dictionary with node as
keys and values as a coordinate list or tuple. If None, then use circular_layout() for dim >= 2
and a linear layout for dim == 1.
weight
used for the edge weight. If None, then all edge weights are 1.
scale
center
dim
[int] Dimension of layout.
Returns

pos
Examples
>>> pos = nx.kamada_kawai_layout(G)
10.4.4 planar_layout
planar_layout(G, scale=1, center=None, dim=2)

Position nodes without edge intersections.
Parameters
G
[NetworkX graph or list of nodes] A position will be assigned to every node in G. If G is of
type nx.PlanarEmbedding, the positions are selected accordingly.
scale
center
dim
Returns
pos
Raises
NetworkXException
If G is not planar
Examples
>>> pos = nx.planar_layout(G)
10.4.5 random_layout
random_layout(G, center=None, dim=2, seed=None)

Position nodes uniformly at random in the unit square.
For every node, a position is generated by choosing each of dim coordinates uniformly at random on the interval
[0.0, 1.0).
NumPy (http://scipy.org) is required for this function.
Parameters

G
center
dim
seed
[int, RandomState instance or None optional (default=None)] Set the random state for de-
terministic node layouts. If int, seed is the seed used by the random number generator, if
numpy.random.RandomState instance, seed is the random number generator, if None, the
random number generator is the RandomState instance used by numpy.random.
Returns
pos
Examples
>>> pos = nx.random_layout(G)
10.4.6 rescale_layout
rescale_layout(pos, scale=1)
Returns scaled position array to (-scale, scale) in all axes.
The function acts on NumPy arrays which hold position information. Each position is one row of the array. The
dimension of the space equals the number of columns. Each coordinate in one column.
To rescale, the mean (center) is subtracted from each axis separately. Then all values are scaled so that the largest
magnitude value from all axes equals scale (thus, the aspect ratio is preserved). The resulting NumPy Array is
returned (order of rows unchanged).
Parameters
pos
[numpy array] positions to be scaled. Each row is a position.
scale
[number (default: 1)] The size of the resulting extent in all directions.
Returns
pos
[numpy array] scaled positions. Each row is a position.
See also:
rescale_layout_dict

10.4.7 rescale_layout_dict
rescale_layout_dict(pos, scale=1)
Return a dictionary of scaled positions keyed by node
Parameters
pos
[A dictionary of positions keyed by node]
scale
[number (default: 1)] The size of the resulting extent in all directions.
Returns
pos
[A dictionary of positions keyed by node]
See also:
rescale_layout
Examples

>>> pos = {0: np.array((0, 0)), 1: np.array((1, 1)), 2: np.array((0.5, 0.5))}
>>> nx.rescale_layout_dict(pos)
{0: array([-1., -1.]), 1: array([1., 1.]), 2: array([0., 0.])}
>>> pos = {0: np.array((0, 0)), 1: np.array((-1, 1)), 2: np.array((-0.5, 0.5))}

>>> nx.rescale_layout_dict(pos, scale=2)
{0: array([ 2., -2.]), 1: array([-2., 2.]), 2: array([0., 0.])}
10.4.8 shell_layout
shell_layout(G, nlist=None, rotate=None, scale=1, center=None, dim=2)

Position nodes in concentric circles.
Parameters
G
nlist
[list of lists] List of node lists for each shell.
rotate
[angle in radians (default=pi/len(nlist))] Angle by which to rotate the starting position of each
shell relative to the starting position of the previous shell. To recreate behavior before v2.5 use
rotate=0.
scale
center

dim
[int] Dimension of layout, currently only dim=2 is supported. Other dimension values result
in a ValueError.
Returns
pos
Raises
ValueError
If dim != 2
Notes
Examples
>>> shells = [[0], [1, 2, 3]]
>>> pos = nx.shell_layout(G, shells)
10.4.9 spring_layout
spring_layout(G, k=None, pos=None, fixed=None, iterations=50, threshold=0.0001, weight='weight', scale=1,

center=None, dim=2, seed=None)
Position nodes using Fruchterman-Reingold force-directed algorithm.
The algorithm simulates a force-directed representation of the network treating edges as springs holding nodes
close, while treating nodes as repelling objects, sometimes called an anti-gravity force. Simulation continues until
the positions are close to an equilibrium.
There are some hard-coded values: minimal distance between nodes (0.01) and “temperature” of 0.1 to ensure
nodes don’t fly away. During the simulation, k helps determine the distance between nodes, though scale and
center determine the size and place after rescaling occurs at the end of the simulation.
Fixing some nodes doesn’t allow them to move in the simulation. It also turns off the rescaling feature at the
simulation’s end. In addition, setting scale to None turns off rescaling.
Parameters
G
k
[float (default=None)] Optimal distance between nodes. If None the distance is set to 1/sqrt(n)
where n is the number of nodes. Increase this value to move nodes farther apart.
pos
[dict or None optional (default=None)] Initial positions for nodes as a dictionary with node as
keys and values as a coordinate list or tuple. If None, then use random initial positions.
fixed
[list or None optional (default=None)] Nodes to keep fixed at initial position. Nodes not in
G.nodes are ignored. ValueError raised if fixed specified and pos not.

iterations
[int optional (default=50)] Maximum number of iterations taken
threshold: float optional (default = 1e-4)
Threshold for relative error in node position changes. The iteration stops if the error is below
this threshold.
weight
used for the edge weight. Larger means a stronger attractive force. If None, then all edge
weights are 1.
scale
[number or None (default: 1)] Scale factor for positions. Not used unless fixed is None.
If scale is None, no rescaling is performed.
center
[array-like or None] Coordinate pair around which to center the layout. Not used unless
fixed is None.
dim
seed
[int, RandomState instance or None optional (default=None)] Set the random state for de-
terministic node layouts. If int, seed is the seed used by the random number generator, if
numpy.random.RandomState instance, seed is the random number generator, if None, the
random number generator is the RandomState instance used by numpy.random.
Returns
pos
Examples
>>> pos = nx.spring_layout(G)
# The same using longer but equivalent function name >>> pos = nx.fruchterman_reingold_layout(G)
10.4.10 spectral_layout
spectral_layout(G, weight='weight', scale=1, center=None, dim=2)

Position nodes using the eigenvectors of the graph Laplacian.
Using the unnormalized Laplacian, the layout shows possible clusters of nodes which are an approximation of the
ratio cut. If dim is the number of dimensions then the positions are the entries of the dim eigenvectors corresponding
to the ascending eigenvalues starting from the second one.
Parameters
G
weight
used for the edge weight. If None, then all edge weights are 1.

scale
center
dim
Returns
pos
Notes
Directed graphs will be considered as undirected graphs when positioning the nodes.
For larger graphs (>500 nodes) this will use the SciPy sparse eigenvalue solver (ARPACK).
Examples
>>> pos = nx.spectral_layout(G)
10.4.11 spiral_layout
spiral_layout(G, scale=1, center=None, dim=2, resolution=0.35, equidistant=False)

Position nodes in a spiral layout.
Parameters
G
scale
center
dim
[int, default=2] Dimension of layout, currently only dim=2 is supported. Other dimension
values result in a ValueError.
resolution
[float, default=0.35] The compactness of the spiral layout returned. Lower values result in
more compressed spiral layouts.
equidistant
[bool, default=False] If True, nodes will be positioned equidistant from each other by decreas-
ing angle further from center. If False, nodes will be positioned at equal angles from each
other by increasing separation further from center.
Returns
pos

Raises
ValueError
If dim != 2
Notes
This algorithm currently only works in two dimensions.
Examples
>>> pos = nx.spiral_layout(G)
>>> nx.draw(G, pos=pos)
10.4.12 multipartite_layout
multipartite_layout(G, subset_key='subset', align='vertical', scale=1, center=None)

Position nodes in layers of straight lines.
Parameters
G
subset_key
[string (default=’subset’)] Key of node data to be used as layer subset.
align
[string (default=’vertical’)] The alignment of nodes. Vertical or horizontal.
scale
center
Returns
pos
[dict] A dictionary of positions keyed by node.
Notes
Network does not need to be a complete multipartite graph. As long as nodes have subset_key data, they will be
placed in the corresponding layers.

Examples
>>> G = nx.complete_multipartite_graph(28, 16, 10)

>>> pos = nx.multipartite_layout(G)

CHAPTER
ELEVEN
RANDOMNESS
Random Number Generators (RNGs) are often used when generating, drawing and computing properties or manipulating
networks. NetworkX provides functions which use one of two standard RNGs: NumPy’s package numpy.random or
Python’s built-in package random. They each provide the same algorithm for generating numbers (Mersenne Twister).
Their interfaces are similar (dangerously similar) and yet distinct. They each provide a global default instance of their
generator that is shared by all programs in a single session. For the most part you can use the RNGs as NetworkX has them
set up and you’ll get reasonable pseudorandom results (results that are statistically random, but created in a deterministic
manner).
Sometimes you want more control over how the numbers are generated. In particular, you need to set the seed of the
generator to make your results reproducible – either for scientific publication or for debugging. Both RNG packages have
easy functions to set the seed to any integer, thus determining the subsequent generated values. Since this package (and
many others) use both RNGs you may need to set the seed of both RNGs. Even if we strictly only used one of the
RNGs, you may find yourself using another package that uses the other. Setting the state of the two global RNGs is as
simple setting the seed of each RNG to an arbitrary integer:
>>> import random

>>> random.seed(246) # or any integer
>>> import numpy
>>> numpy.random.seed(4812)
Many users will be satisfied with this level of control.

For people who want even more control, we include an optional argument to functions that use an RNG. This argument
is called seed, but determines more than the seed of the RNG. It tells the function which RNG package to use, and
whether to use a global or local RNG.
>>> from networkx import path_graph, random_layout

>>> G = path_graph(9)
>>> pos = random_layout(G, seed=None) # use (either) global default RNG
>>> pos = random_layout(G, seed=42) # local RNG just for this call
>>> pos = random_layout(G, seed=numpy.random) # use numpy global RNG
>>> random_state = numpy.random.RandomState(42)
>>> pos = random_layout(G, seed=random_state) # use/reuse your own RNG
Each NetworkX function that uses an RNG was written with one RNG package in mind. It either uses random or
numpy.random by default. But some users want to only use a single RNG for all their code. This seed argument
provides a mechanism so that any function can use a numpy.random RNG even if the function is written for random.
It works as follows.
The default behavior (when seed=None) is to use the global RNG for the function’s preferred package. If seed is set
to an integer value, a local RNG is created with the indicated seed value and is used for the duration of that function
(including any calls to other functions) and then discarded. Alternatively, you can specify seed=numpy.random to
ensure that the global numpy RNG is used whether the function expects it or not. Finally, you can provide a numpy RNG
1019
to be used by the function. The RNG is then available to use in other functions or even other package like sklearn. In this
way you can use a single RNG for all random numbers in your project.
While it is possible to assign seed a random-style RNG for NetworkX functions written for the random package API,
the numpy RNG interface has too many nice features for us to ensure a random-style RNG will work in all functions.
In practice, you can do most things using only random RNGs (useful if numpy is not available). But your experience
will be richer if numpy is available.
To summarize, you can easily ignore the seed argument and use the global RNGs. You can specify to use only the
numpy global RNG with seed=numpy.random. You can use a local RNG by providing an integer seed value. And
you can provide your own numpy RNG, reusing it for all functions. It is easier to use numpy RNGs if you want a single
RNG for your computations.
1020 Chapter 11. Randomness

CHAPTER
TWELVE
EXCEPTIONS
Base exceptions and errors for NetworkX.

class NetworkXException
Base class for exceptions in NetworkX.
class NetworkXError
Exception for a serious error in NetworkX
class NetworkXPointlessConcept
Raised when a null graph is provided as input to an algorithm that cannot use it.
The null graph is sometimes considered a pointless concept [1], thus the name of the exception.
References
[1]
class NetworkXAlgorithmError
Exception for unexpected termination of algorithms.
class NetworkXUnfeasible
Exception raised by algorithms trying to solve a problem instance that has no feasible solution.
class NetworkXNoPath
Exception for algorithms that should return a path when running on graphs where such a path does not exist.
class NetworkXNoCycle
Exception for algorithms that should return a cycle when running on graphs where such a cycle does not exist.
class NodeNotFound
Exception raised if requested node is not present in the graph
class HasACycle
Raised if a graph has a cycle when an algorithm expects that it will have no cycles.
class NetworkXUnbounded
Exception raised by algorithms trying to solve a maximization or a minimization problem instance that is un-
bounded.
class NetworkXNotImplemented
Exception raised by algorithms not implemented for a type of graph.
1021
class AmbiguousSolution
Raised if more than one valid solution exists for an intermediary step of an algorithm.
In the face of ambiguity, refuse the temptation to guess. This may occur, for example, when trying to determine
the bipartite node sets in a disconnected bipartite graph when computing bipartite matchings.
class ExceededMaxIterations
Raised if a loop iterates too many times without breaking.
This may occur, for example, in an algorithm that computes progressively better approximations to a value but
exceeds an iteration bound specified by the user.
class PowerIterationFailedConvergence(num_iterations, *args, **kw)
Raised when the power iteration method fails to converge within a specified iteration limit.
num_iterations is the number of iterations that have been completed when this exception was raised.
1022 Chapter 12. Exceptions

CHAPTER
THIRTEEN
UTILITIES
13.1 Helper Functions
Miscellaneous Helpers for NetworkX.

These are not imported into the base networkx namespace but can be accessed, for example, as
>>> import networkx

>>> networkx.utils.make_list_of_ints({1, 2, 3})
[1, 2, 3]
>>> networkx.utils.arbitrary_element({5, 1, 7})
1
arbitrary_element(iterable) Returns an arbitrary element of iterable without re-

moving it.
flatten(obj[, result]) Return flattened version of (possibly nested) iterable ob-
ject.
make_list_of_ints(sequence) Return list of ints from sequence of integral numbers.
dict_to_numpy_array(d[, mapping]) Convert a dictionary of dictionaries to a numpy array with
optional mapping.
pairwise(iterable[, cyclic]) s -> (s0, s1), (s1, s2), (s2, s3), ...
groups(many_to_one) Converts a many-to-one mapping into a one-to-many
mapping.
create_random_state([random_state]) Returns a numpy.random.RandomState or
numpy.random.Generator instance depending on in-
put.
create_py_random_state([random_state]) Returns a random.Random instance depending on input.
nodes_equal(nodes1, nodes2) Check if nodes are equal.
edges_equal(edges1, edges2) Check if edges are equal.
graphs_equal(graph1, graph2) Check if graphs are equal.
1023
13.1.1 arbitrary_element
arbitrary_element(iterable)
Returns an arbitrary element of iterable without removing it.
This is most useful for “peeking” at an arbitrary element of a set, but can be used for any list, dictionary, etc., as
well.
Parameters
iterable
[abc.collections.Iterable instance] Any object that implements __iter__, e.g.
set, dict, list, tuple, etc.
Returns
The object that results from next(iter(iterable))
Raises
ValueError
If iterable is an iterator (because the current implementation of this function would con-
sume an element from the iterator).
Notes
This function does not return a random element. If iterable is ordered, sequential calls will return the same
value:
>>> l = [1, 2, 3]
>>> nx.utils.arbitrary_element(l)
1
>>> nx.utils.arbitrary_element(l)
1
Examples
Arbitrary elements from common Iterable objects:
>>> nx.utils.arbitrary_element([1, 2, 3]) # list

1
>>> nx.utils.arbitrary_element((1, 2, 3)) # tuple
1
>>> nx.utils.arbitrary_element({1, 2, 3}) # set
1
>>> d = {k: v for k, v in zip([1, 2, 3], [3, 2, 1])}
>>> nx.utils.arbitrary_element(d) # dict_keys
1
>>> nx.utils.arbitrary_element(d.values()) # dict values
3
str is also an Iterable:
>>> nx.utils.arbitrary_element("hello")
'h'
ValueError is raised if iterable is an iterator:
1024 Chapter 13. Utilities

>>> iterator = iter([1, 2, 3]) # Iterator, *not* Iterable

>>> nx.utils.arbitrary_element(iterator)
...
ValueError: cannot return an arbitrary item from an iterator
13.1.2 flatten
flatten(obj, result=None)
Return flattened version of (possibly nested) iterable object.
13.1.3 make_list_of_ints
make_list_of_ints(sequence)
Return list of ints from sequence of integral numbers.
All elements of the sequence must satisfy int(element) == element or a ValueError is raised. Sequence is iterated
through once.
If sequence is a list, the non-int values are replaced with ints. So, no new list is created
13.1.4 dict_to_numpy_array
dict_to_numpy_array(d, mapping=None)
Convert a dictionary of dictionaries to a numpy array with optional mapping.
13.1.5 pairwise
pairwise(iterable, cyclic=False)
s -> (s0, s1), (s1, s2), (s2, s3), …
13.1.6 groups
groups(many_to_one)
Converts a many-to-one mapping into a one-to-many mapping.
many_to_one must be a dictionary whose keys and values are all hashable.
The return value is a dictionary mapping values from many_to_one to sets of keys from many_to_one that
have that value.
13.1. Helper Functions 1025

Examples
>>> from networkx.utils import groups

>>> many_to_one = {"a": 1, "b": 1, "c": 2, "d": 3, "e": 3}
>>> groups(many_to_one)
{1: {'a', 'b'}, 2: {'c'}, 3: {'e', 'd'}}
13.1.7 create_random_state
create_random_state(random_state=None)
Returns a numpy.random.RandomState or numpy.random.Generator instance depending on input.
Parameters
random_state
[int or NumPy RandomState or Generator instance, optional (default=None)] If int, re-
turn a numpy.random.RandomState instance set with seed=int. if numpy.random.
RandomState instance, return it. if numpy.random.Generator instance, return it. if
None or numpy.random, return the global random number generator used by numpy.random.
13.1.8 create_py_random_state
create_py_random_state(random_state=None)
Returns a random.Random instance depending on input.
Parameters
random_state
[int or random number generator or None (default=None)] If int, return a random.Random
instance set with seed=int. if random.Random instance, return it. if None or the random
package, return the global random number generator used by random. if np.random pack-
age, return the global numpy random number generator wrapped in a PythonRandomInter-
face class. if np.random.RandomState or np.random.Generator instance, return it wrapped in
PythonRandomInterface if a PythonRandomInterface instance, return it
13.1.9 nodes_equal
nodes_equal(nodes1, nodes2)
Check if nodes are equal.
Equality here means equal as Python objects. Node data must match if included. The order of nodes is not relevant.
Parameters
nodes1, nodes2
[iterables of nodes, or (node, datadict) tuples]
Returns
bool
True if nodes are equal, False otherwise.

13.1.10 edges_equal
edges_equal(edges1, edges2)
Check if edges are equal.
Equality here means equal as Python objects. Edge data must match if included. The order of the edges is not
relevant.
Parameters
edges1, edges2
[iterables of with u, v nodes as] edge tuples (u, v), or edge tuples with data dicts (u, v, d), or
edge tuples with keys and data dicts (u, v, k, d)
Returns
bool
True if edges are equal, False otherwise.
13.1.11 graphs_equal
graphs_equal(graph1, graph2)
Check if graphs are equal.
Equality here means equal as Python objects (not isomorphism). Node, edge and graph data must match.
Parameters
graph1, graph2
[graph]
Returns
bool
True if graphs are equal, False otherwise.
13.2 Data Structures and Algorithms
Union-find data structure.
UnionFind.union(*objects) Find the sets containing the objects and merge them all.
13.2.1 UnionFind.union
UnionFind.union(*objects)
Find the sets containing the objects and merge them all.
13.2. Data Structures and Algorithms 1027

13.3 Random Sequence Generators
Utilities for generating random numbers, random sequences, and random selections.
powerlaw_sequence(n[, exponent, seed]) Return sample sequence of length n from a power law dis-
tribution.
cumulative_distribution(distribution) Returns normalized cumulative distribution from discrete
distribution.
discrete_sequence(n[, distribution, ...]) Return sample sequence of length n from a given discrete
distribution or discrete cumulative distribution.
zipf_rv(alpha[, xmin, seed]) Returns a random value chosen from the Zipf distribution.
random_weighted_sample(mapping, k[, seed]) Returns k items without replacement from a weighted
sample.
weighted_choice(mapping[, seed]) Returns a single element from a weighted sample.
13.3.1 powerlaw_sequence
powerlaw_sequence(n, exponent=2.0, seed=None)

Return sample sequence of length n from a power law distribution.
13.3.2 cumulative_distribution
cumulative_distribution(distribution)
Returns normalized cumulative distribution from discrete distribution.
13.3.3 discrete_sequence
discrete_sequence(n, distribution=None, cdistribution=None, seed=None)

Return sample sequence of length n from a given discrete distribution or discrete cumulative distribution.
One of the following must be specified.
distribution = histogram of values, will be normalized
cdistribution = normalized discrete cumulative distribution
13.3.4 zipf_rv
zipf_rv(alpha, xmin=1, seed=None)

Returns a random value chosen from the Zipf distribution.
The return value is an integer drawn from the probability distribution
x−α
p(x) = ,
ζ(α, xmin )
where ζ(α, xmin ) is the Hurwitz zeta function.

Parameters

alpha
[float] Exponent value of the distribution
xmin
[int] Minimum value
seed
Randomness.
Returns
x
[int] Random value from Zipf distribution
Raises
ValueError:
If xmin < 1 or If alpha <= 1
Notes
The rejection algorithm generates random values for a the power-law distribution in uniformly bounded expected
time dependent on parameters. See [1] for details on its operation.
References
[1]
Examples
>>> nx.utils.zipf_rv(alpha=2, xmin=3, seed=42)

8
13.3.5 random_weighted_sample
random_weighted_sample(mapping, k, seed=None)
Returns k items without replacement from a weighted sample.
The input is a dictionary of items with weights as values.
13.3.6 weighted_choice
weighted_choice(mapping, seed=None)
Returns a single element from a weighted sample.
The input is a dictionary of items with weights as values.
13.3. Random Sequence Generators 1029

13.4 Decorators
open_file(path_arg[, mode]) Decorator to ensure clean opening and closing of files.

not_implemented_for(*graph_types) Decorator to mark algorithms as not implemented
nodes_or_number(which_args) Decorator to allow number of nodes or container of nodes.
np_random_state(random_state_argument) Decorator to generate a numpy.random.
RandomState instance.
py_random_state(random_state_argument) Decorator to generate a random.Random instance (or
equiv).
argmap(func, *args[, try_finally]) A decorator to apply a map to arguments before calling
the function
13.4.1 open_file
open_file(path_arg, mode='r')
Decorator to ensure clean opening and closing of files.
Parameters
path_arg
[string or int] Name or index of the argument that is a path.
mode
[str] String for opening mode.
Returns
_open_file
[function] Function which cleanly executes the io.
Notes
Note that this decorator solves the problem when a path argument is specified as a string, but it does not handle the
situation when the function wants to accept a default of None (and then handle it).
Here is an example of how to handle this case:
@open_file("path")
def some_function(arg1, arg2, path=None):
if path is None:
fobj = tempfile.NamedTemporaryFile(delete=False)
else:
# `path` could have been a string or file object or something
# similar. In any event, the decorator has given us a file object
# and it will close it for us, if it should.
fobj = path
try:
fobj.write("blah")
finally:
if path is None:
fobj.close()

Normally, we’d want to use “with” to ensure that fobj gets closed. However, the decorator will make path a file
object for us, and using “with” would undesirably close that file object. Instead, we use a try block, as shown above.
When we exit the function, fobj will be closed, if it should be, by the decorator.
Examples
Decorate functions like this:
@open_file(0,"r")
def read_function(pathname):
pass
@open_file(1,"w")
def write_function(G, pathname):
pass
@open_file(1,"w")
def write_function(G, pathname="graph.dot"):
pass
@open_file("pathname","w")
def write_function(G, pathname="graph.dot"):
pass
@open_file("path", "w+")
def another_function(arg, **kwargs):
path = kwargs["path"]
pass
13.4.2 not_implemented_for
not_implemented_for(*graph_types)
Decorator to mark algorithms as not implemented
Parameters
graph_types
[container of strings] Entries must be one of “directed”, “undirected”, “multigraph”, or
“graph”.
Returns
_require
[function] The decorated function.
Raises
If any of the packages cannot be imported
13.4. Decorators 1031

Notes
Multiple types are joined logically with “and”. For “or” use multiple @not_implemented_for() lines.
Examples
@not_implemented_for("directed")
def sp_function(G):
pass
# rule out MultiDiGraph

@not_implemented_for("directed","multigraph")
def sp_np_function(G):
pass
# rule out all except DiGraph

@not_implemented_for("undirected")
@not_implemented_for("multigraph")
def sp_np_function(G):
pass
13.4.3 nodes_or_number
nodes_or_number(which_args)
Decorator to allow number of nodes or container of nodes.
With this decorator, the specified argument can be either a number or a container of nodes. If
it is a number, the nodes used are range(n). This allows nx.complete_graph(50)
in place of nx.complete_graph(list(range(50))). And it also allows nx.
complete_graph(any_list_of_nodes).
Parameters
which_args
[string or int or sequence of strings or ints] If string, the name of the argument to be treated.
If int, the index of the argument to be treated. If more than one node argument is allowed, can
be a list of locations.
Returns
_nodes_or_numbers
[function] Function which replaces int args with ranges.

Examples
@nodes_or_number("nodes")
def empty_graph(nodes):
# nodes is converted to a list of nodes
@nodes_or_number(0)
def empty_graph(nodes):
# nodes is converted to a list of nodes
@nodes_or_number(["m1", "m2"])
def grid_2d_graph(m1, m2, periodic=False):
# m1 and m2 are each converted to a list of nodes
@nodes_or_number([0, 1])
def grid_2d_graph(m1, m2, periodic=False):
# m1 and m2 are each converted to a list of nodes
@nodes_or_number(1)
def full_rary_tree(r, n)
# presumably r is a number. It is not handled by this decorator.
# n is converted to a list of nodes
13.4.4 np_random_state
np_random_state(random_state_argument)
Decorator to generate a numpy.random.RandomState instance.
The decorator processes the argument indicated by random_state_argument using nx.utils.
create_random_state(). The argument value can be a seed (integer), or a numpy.random.
RandomState instance or (None or numpy.random). The latter options use the glocal random number
generator used by numpy.random. The result is a numpy.random.RandomState instance.
Parameters
random_state_argument
[string or int] The name or index of the argument to be converted to a numpy.random.
RandomState instance.
Returns
_random_state
[function] Function whose random_state keyword argument is a RandomState instance.
See also:
py_random_state

Examples
@np_random_state("seed")
def random_float(seed=None):
return seed.rand()
@np_random_state(0)
def random_float(rng=None):
return rng.rand()
@np_random_state(1)
def random_array(dims, random_state=1):
return random_state.rand(*dims)
13.4.5 py_random_state
py_random_state(random_state_argument)
Decorator to generate a random.Random instance (or equiv).
The decorator processes the argument indicated by random_state_argument using nx.utils.
create_py_random_state(). The argument value can be a seed (integer), or a random number generator:
If int, return a random.Random instance set with seed=int.

If random.Random instance, return it.
If None or the `random` package, return the global random number
generator used by `random`.
If np.random package, return the global numpy random number
generator wrapped in a PythonRandomInterface class.
If np.random.RandomState instance, return it wrapped in
PythonRandomInterface
If a PythonRandomInterface instance, return it
Parameters
random_state_argument
[string or int] The name of the argument or the index of the argument in args that is to be con-
verted to the random.Random instance or numpy.random.RandomState instance that mimics
basic methods of random.Random.
Returns
_random_state
[function] Function whose random_state_argument is converted to a Random instance.
See also:
np_random_state

Examples
@py_random_state("random_state")
def random_float(random_state=None):
return random_state.rand()
@py_random_state(0)
def random_float(rng=None):
return rng.rand()
@py_random_state(1)
def random_array(dims, seed=12345):
return seed.rand(*dims)
13.4.6 networkx.utils.decorators.argmap
class argmap(func, *args, try_finally=False)

A decorator to apply a map to arguments before calling the function
This class provides a decorator that maps (transforms) arguments of the function before the function is called. Thus
for example, we have similar code in many functions to determine whether an argument is the number of nodes to
be created, or a list of nodes to be handled. The decorator provides the code to accept either – transforming the
indicated argument into a list of nodes before the actual function is called.
This decorator class allows us to process single or multiple arguments. The arguments to be processed can be
specified by string, naming the argument, or by index, specifying the item in the args list.
Parameters
func
[callable] The function to apply to arguments
*args
[iterable of (int, str or tuple)] A list of parameters, specified either as strings (their names), ints
(numerical indices) or tuples, which may contain ints, strings, and (recursively) tuples. Each
indicates which parameters the decorator should map. Tuples indicate that the map function
takes (and returns) multiple parameters in the same order and nested structure as indicated
here.
try_finally
[bool (default: False)] When True, wrap the function call in a try-finally block with code for
the finally block created by func. This is used when the map function constructs an object
(like a file handle) that requires post-processing (like closing).
Note: try_finally decorators cannot be used to decorate generator functions.
See also:
not_implemented_for
open_file
nodes_or_number
random_state
py_random_state
networkx.community.quality.require_partition
require_partition

Notes
An object of this class is callable and intended to be used when defining a decorator. Generally, a decorator takes
a function as input and constructs a function as output. Specifically, an argmap object returns the input func-
tion decorated/wrapped so that specified arguments are mapped (transformed) to new values before the decorated
function is called.
As an overview, the argmap object returns a new function with all the dunder values of the original function (like
__doc__, __name__, etc). Code for this decorated function is built based on the original function’s signature.
It starts by mapping the input arguments to potentially new values. Then it calls the decorated function with these
new values in place of the indicated arguments that have been mapped. The return value of the original function is
then returned. This new function is the function that is actually called by the user.
Three additional features are provided.
1) The code is lazily compiled. That is, the new function is returned as an object without the code compiled,
but with all information needed so it can be compiled upon it’s first invocation. This saves time on import at
the cost of additional time on the first call of the function. Subsequent calls are then just as fast as normal.
2) If the “try_finally” keyword-only argument is True, a try block follows each mapped argument, matched
on the other side of the wrapped call, by a finally block closing that mapping. We expect func to return a
2-tuple: the mapped value and a function to be called in the finally clause. This feature was included so the
open_file decorator could provide a file handle to the decorated function and close the file handle after
the function call. It even keeps track of whether to close the file handle or not based on whether it had to
open the file or the input was already open. So, the decorated function does not need to include any code to
open or close files.
3) The maps applied can process multiple arguments. For example, you could swap two arguments using a
mapping, or transform them to their sum and their difference. This was included to allow a decorator in the
quality.py module that checks that an input partition is a valid partition of the nodes of the input
graph G. In this example, the map has inputs (G, partition). After checking for a valid partition, the
map either raises an exception or leaves the inputs unchanged. Thus many functions that make this check can
use the decorator rather than copy the checking code into each function. More complicated nested argument
structures are described below.
The remaining notes describe the code structure and methods for this class in broad terms to aid in understanding
how to use it.
Instantiating an argmap object simply stores the mapping function and the input identifiers of which arguments
to map. The resulting decorator is ready to use this map to decorate any function. Calling that object (argmap.
__call__, but usually done via @my_decorator) a lazily compiled thin wrapper of the decorated function is
constructed, wrapped with the necessary function dunder attributes like __doc__ and __name__. That thinly
wrapped function is returned as the decorated function. When that decorated function is called, the thin wrapper
of code calls argmap._lazy_compile which compiles the decorated function (using argmap.compile)
and replaces the code of the thin wrapper with the newly compiled code. This saves the compilation step every
import of networkx, at the cost of compiling upon the first call to the decorated function.
When the decorated function is compiled, the code is recursively assembled using the argmap.assemble
method. The recursive nature is needed in case of nested decorators. The result of the assembly is a number
of useful objects.
sig
[the function signature of the original decorated function as] constructed by argmap.
signature(). This is constructed using inspect.signature but enhanced with attribute
strings sig_def and sig_call, and other information specific to mapping arguments of this
function. This information is used to construct a string of code defining the new decorated function.
wrapped_name
[a unique internally used name constructed by argmap] for the decorated function.

functions
[a dict of the functions used inside the code of this] decorated function, to be used as globals
in exec. This dict is recursively updated to allow for nested decorating.
mapblock
[code (as a list of strings) to map the incoming argument] values to their mapped values.
finallys
[code (as a list of strings) to provide the possibly nested] set of finally clauses if needed.
mutable_args
[a bool indicating whether the sig.args tuple should be] converted to a list so mutation can
occur.
After this recursive assembly process, the argmap.compile method constructs code (as strings) to convert the
tuple sig.args to a list if needed. It joins the defining code with appropriate indents and compiles the result.
Finally, this code is evaluated and the original wrapper’s implementation is replaced with the compiled version (see
argmap._lazy_compile for more details).
Other argmap methods include _name and _count which allow internally generated names to be unique within
a python session. The methods _flatten and _indent process the nested lists of strings into properly indented
python code ready to be compiled.
More complicated nested tuples of arguments also allowed though usually not used. For the simple 2 argument case,
the argmap input (“a”, “b”) implies the mapping function will take 2 arguments and return a 2-tuple of mapped
values. A more complicated example with argmap input ("a", ("b", "c")) requires the mapping function
take 2 inputs, with the second being a 2-tuple. It then must output the 3 mapped values in the same nested structure
(newa, (newb, newc)). This level of generality is not often needed, but was convenient to implement when
handling the multiple arguments.
Examples
Most of these examples use @argmap(...) to apply the decorator to the function defined on the next line. In the
NetworkX codebase however, argmap is used within a function to construct a decorator. That is, the decorator
defines a mapping function and then uses argmap to build and return a decorated function. A simple example is
a decorator that specifies which currency to report money. The decorator (named convert_to) would be used
like:
@convert_to("US_Dollars", "income")
def show_me_the_money(name, income):
print(f"{name} : {income}")
And the code to create the decorator might be:
def convert_to(currency, which_arg):

def _convert(amount):
if amount.currency != currency:
amount = amount.to_currency(currency)
return amount
return argmap(_convert, which_arg)
Despite this common idiom for argmap, most of the following examples use the @argmap(...) idiom to save
space.
Here’s an example use of argmap to sum the elements of two of the functions arguments. The decorated function:

@argmap(sum, "xlist", "zlist")

def foo(xlist, y, zlist):
return xlist - y + zlist
is syntactic sugar for:

x = sum(xlist)
z = sum(zlist)
return x - y + z
and is equivalent to (using argument indexes):
@argmap(sum, "xlist", 2)
or:
@argmap(sum, "zlist", 0)
Transforming functions can be applied to multiple arguments, such as:
def swap(x, y):

return y, x
# the 2-tuple tells argmap that the map `swap` has 2 inputs/outputs.
@argmap(swap, ("a", "b")):
def foo(a, b, c):
return a / b * c
is equivalent to:
def foo(a, b, c):

a, b = swap(a, b)
return a / b * c
More generally, the applied arguments can be nested tuples of strings or ints. The syntax
@argmap(some_func, ("a", ("b", "c"))) would expect some_func to accept 2 inputs
with the second expected to be a 2-tuple. It should then return 2 outputs with the second a 2-tuple. The returns
values would replace input “a” “b” and “c” respectively. Similarly for @argmap(some_func, (0, ("b",
2))).
Also, note that an index larger than the number of named parameters is allowed for variadic functions. For example:
def double(a):
return 2 * a
@argmap(double, 3)
def overflow(a, *args):
return a, args
print(overflow(1, 2, 3, 4, 5, 6)) # output is 1, (2, 3, 8, 5, 6)
Try Finally

Additionally, this argmap class can be used to create a decorator that initiates a try…finally block. The decorator
must be written to return both the transformed argument and a closing function. This feature was included to enable
the open_file decorator which might need to close the file or not depending on whether it had to open that file.
This feature uses the keyword-only try_finally argument to @argmap.
For example this map opens a file and then makes sure it is closed:
def open_file(fn):
f = open(fn)
return f, lambda: f.close()
The decorator applies that to the function foo:
@argmap(open_file, "file", try_finally=True)

def foo(file):
print(file.read())
is syntactic sugar for:
def foo(file):
file, close_file = open_file(file)
try:
print(file.read())
finally:
close_file()
and is equivalent to (using indexes):
@argmap(open_file, 0, try_finally=True)
def foo(file):
print(file.read())
Here’s an example of the try_finally feature used to create a decorator:
def my_closing_decorator(which_arg):
def _opener(path):
if path is None:
path = open(path)
fclose = path.close
else:
# assume `path` handles the closing
fclose = lambda: None
return path, fclose
return argmap(_opener, which_arg, try_finally=True)
which can then be used as:
@my_closing_decorator("file")
def fancy_reader(file=None):
# this code doesn't need to worry about closing the file
print(file.read())
Decorators with try_finally = True cannot be used with generator functions, because the finally block is eval-
uated before the generator is exhausted:
@argmap(open_file, "file", try_finally=True)

def file_to_lines(file):


for line in file.readlines():
yield line
is equivalent to:
def file_to_lines_wrapped(file):
yield line
def file_to_lines_wrapper(file):
try:
file = open_file(file)
return file_to_lines_wrapped(file)
finally:
file.close()
which behaves similarly to:
def file_to_lines_whoops(file):
file = open_file(file)
file.close()
yield line
because the finally block of file_to_lines_wrapper is executed before the caller has a chance to
exhaust the iterator.
__init__(func, *args, try_finally=False)
Methods
assemble(f) Collects components of the source for the decorated

function wrapping f.
compile(f) Compile the decorated function.
signature(f) Construct a Signature object describing f
argmap.assemble
argmap.assemble(f)
Collects components of the source for the decorated function wrapping f.
If f has multiple argmap decorators, we recursively assemble the stack of decorators into a single flattened
function.
This method is part of the compile method’s process yet separated from that method to allow recursive
processing. The outputs are strings, dictionaries and lists that collect needed info to flatten any nested argmap-
decoration.
Parameters
f
[callable] The function to be decorated. If f is argmapped, we assemble it.
Returns

sig
[argmap.Signature] The function signature as an argmap.Signature object.
wrapped_name
[str] The mangled name used to represent the wrapped function in the code being assembled.
functions
[dict] A dictionary mapping id(g) -> (mangled_name(g), g) for functions g referred to in the
code being assembled. These need to be present in the globals scope of exec when
defining the decorated function.
mapblock
[list of lists and/or strings] Code that implements mapping of parameters including any try
blocks if needed. This code will precede the decorated function call.
finallys
[list of lists and/or strings] Code that implements the finally blocks to post-process the argu-
ments (usually close any files if needed) after the decorated function is called.
mutable_args
[bool] True if the decorator needs to modify positional arguments via their indices. The
compile method then turns the argument tuple into a list so that the arguments can be mod-
ified.
argmap.compile
argmap.compile(f)
Compile the decorated function.
Called once for a given decorated function – collects the code from all argmap decorators in the stack, and
compiles the decorated function.
Much of the work done here uses the assemble method to allow recursive treatment of multiple argmap
decorators on a single decorated function. That flattens the argmap decorators, collects the source code to
construct a single decorated function, then compiles/executes/returns that function.
The source code for the decorated function is stored as an attribute _code on the function object itself.
Note that Python’s compile function requires a filename, but this code is constructed without a file, so a
fictitious filename is used to describe where the function comes from. The name is something like: “argmap
compilation 4”.
Parameters
f
[callable] The function to be decorated
Returns
func
[callable] The decorated file

argmap.signature
classmethod argmap.signature(f)
Construct a Signature object describing f
Compute a Signature so that we can write a function wrapping f with the same signature and call-type.
Parameters
f
[callable] A function to be decorated
Returns
sig
[argmap.Signature] The Signature of f
Notes
The Signature is a namedtuple with names:

name : a unique version of the name of the decorated function signature : the inspect.signature
of the decorated function def_sig : a string used as code to define the new function call_sig : a
string used as code to call the decorated function names : a dict keyed by argument name and index
to the argument’s name n_positional : the number of positional arguments in the signature args :
the name of the VAR_POSITIONAL argument if any, i.e. *theseargs kwargs : the name of the
VAR_KEYWORDS argument if any, i.e. **kwargs
These named attributes of the signature are used in assemble and compile to construct a string of source
code for the decorated function.
13.5 Cuthill-Mckee Ordering
Cuthill-McKee ordering of graph nodes to produce sparse matrices
cuthill_mckee_ordering(G[, heuristic]) Generate an ordering (permutation) of the graph nodes to

make a sparse matrix.
reverse_cuthill_mckee_ordering(G[, Generate an ordering (permutation) of the graph nodes to
heuristic]) make a sparse matrix.
13.5.1 cuthill_mckee_ordering
cuthill_mckee_ordering(G, heuristic=None)
Generate an ordering (permutation) of the graph nodes to make a sparse matrix.
Uses the Cuthill-McKee heuristic (based on breadth-first search) [1].
Parameters
G

heuristic
[function, optional] Function to choose starting node for RCM algorithm. If None a node from
a pseudo-peripheral pair is used. A user-defined function can be supplied that takes a graph
object and returns a single node.
Returns
nodes
[generator] Generator of nodes in Cuthill-McKee ordering.
See also:
reverse_cuthill_mckee_ordering
Notes
The optimal solution the bandwidth reduction is NP-complete [2].
References
[1], [2]
Examples
>>> from networkx.utils import cuthill_mckee_ordering

>>> rcm = list(cuthill_mckee_ordering(G))
>>> A = nx.adjacency_matrix(G, nodelist=rcm)
Smallest degree node as heuristic function:
>>> def smallest_degree(G):

... return min(G, key=G.degree)
>>> rcm = list(cuthill_mckee_ordering(G, heuristic=smallest_degree))
13.5.2 reverse_cuthill_mckee_ordering
reverse_cuthill_mckee_ordering(G, heuristic=None)
Generate an ordering (permutation) of the graph nodes to make a sparse matrix.
Uses the reverse Cuthill-McKee heuristic (based on breadth-first search) [1].
Parameters
G
heuristic
[function, optional] Function to choose starting node for RCM algorithm. If None a node from
a pseudo-peripheral pair is used. A user-defined function can be supplied that takes a graph
object and returns a single node.
Returns
13.5. Cuthill-Mckee Ordering 1043

nodes
[generator] Generator of nodes in reverse Cuthill-McKee ordering.
See also:
cuthill_mckee_ordering
Notes
The optimal solution the bandwidth reduction is NP-complete [2].
References
[1], [2]
Examples
>>> from networkx.utils import reverse_cuthill_mckee_ordering

>>> rcm = list(reverse_cuthill_mckee_ordering(G))
>>> A = nx.adjacency_matrix(G, nodelist=rcm)
Smallest degree node as heuristic function:
>>> def smallest_degree(G):

... return min(G, key=G.degree)
>>> rcm = list(reverse_cuthill_mckee_ordering(G, heuristic=smallest_degree))
13.6 Mapped Queue
Priority queue class with updatable priorities.
MappedQueue([data]) The MappedQueue class implements a min-heap with re-

moval and update-priority.
13.6.1 networkx.utils.mapped_queue.MappedQueue
class MappedQueue(data=None)
The MappedQueue class implements a min-heap with removal and update-priority.
The min heap uses heapq as well as custom written _siftup and _siftdown methods to allow the heap positions to be
tracked by an additional dict keyed by element to position. The smallest element can be popped in O(1) time, new
elements can be pushed in O(log n) time, and any element can be removed or updated in O(log n) time. The queue
cannot contain duplicate elements and an attempt to push an element already in the queue will have no effect.
MappedQueue complements the heapq package from the python standard library. While MappedQueue is designed
for maximum compatibility with heapq, it adds element removal, lookup, and priority update.
Parameters

data
[dict or iterable]
References
[1], [2]
Examples
A MappedQueue can be created empty, or optionally, given a dictionary of initial elements and priorities. The
methods push, pop, remove, and update operate on the queue.
>>> colors_nm = {'red':665, 'blue': 470, 'green': 550}

>>> q = MappedQueue(colors_nm)
>>> q.remove('red')
>>> q.update('green', 'violet', 400)
>>> q.push('indigo', 425)
True
>>> [q.pop().element for i in range(len(q.heap))]
['violet', 'indigo', 'blue']
A MappedQueue can also be initialized with a list or other iterable. The priority is assumed to be the sort order
of the items in the list.
>>> q = MappedQueue([916, 50, 4609, 493, 237])

>>> q.remove(493)
>>> q.update(237, 1117)
>>> [q.pop() for i in range(len(q.heap))]
[50, 916, 1117, 4609]
An exception is raised if the elements are not comparable.
>>> q = MappedQueue([100, 'a'])

...
TypeError: '<' not supported between instances of 'int' and 'str'
To avoid the exception, use a dictionary to assign priorities to the elements.
>>> q = MappedQueue({100: 0, 'a': 1 })
__init__(data=None)
Priority queue class with updatable priorities.
Methods
pop() Remove and return the smallest element in the queue.

push(elt[, priority]) Add an element to the queue.
remove(elt) Remove an element from the queue.
update(elt, new[, priority]) Replace an element in the queue with a new one.
13.6. Mapped Queue 1045

MappedQueue.pop
MappedQueue.pop()
Remove and return the smallest element in the queue.
MappedQueue.push
MappedQueue.push(elt, priority=None)
Add an element to the queue.
MappedQueue.remove
MappedQueue.remove(elt)
Remove an element from the queue.
MappedQueue.update
MappedQueue.update(elt, new, priority=None)

Replace an element in the queue with a new one.

CHAPTER
FOURTEEN
GLOSSARY
dictionary
A Python dictionary maps keys to values. Also known as “hashes”, or “associative arrays” in other programming
languages. See the Python tutorial on dictionaries.
edge
Edges are either two-tuples of nodes (u, v) or three tuples of nodes with an edge attribute dictionary (u, v,
dict).
ebunch
An iteratable container of edge tuples like a list, iterator, or file.
edge attribute
Edges can have arbitrary Python objects assigned as attributes by using keyword/value pairs when adding an edge
assigning to the G.edges[u][v] attribute dictionary for the specified edge u-v.
nbunch
An nbunch is a single node, container of nodes or None (representing all nodes). It can be a list, set, graph, etc..
To filter an nbunch so that only nodes actually in G appear, use G.nbunch_iter(nbunch).
node
A node can be any hashable Python object except None.
node attribute
Nodes can have arbitrary Python objects assigned as attributes by using keyword/value pairs when adding a node
or assigning to the G.nodes[n] attribute dictionary for the specified node n.
1047
1048 Chapter 14. Glossary

APPENDIX
TUTORIAL
This guide can help you start working with NetworkX.
A.1 Creating a graph
Create an empty graph with no nodes and no edges.

>>> G = nx.Graph()
By definition, a Graph is a collection of nodes (vertices) along with identified pairs of nodes (called edges, links, etc). In
NetworkX, nodes can be any hashable object e.g., a text string, an image, an XML object, another Graph, a customized
node object, etc.
Note: Python’s None object is not allowed to be used as a node. It determines whether optional function arguments
have been assigned in many functions.
A.2 Nodes
The graph G can be grown in several ways. NetworkX includes many graph generator functions and facilities to read and
write graphs in many formats. To get started though we’ll look at simple manipulations. You can add one node at a time,
>>> G.add_node(1)
or add nodes from any iterable container, such as a list
You can also add nodes along with node attributes if your container yields 2-tuples of the form (node,
node_attribute_dict):
>>> G.add_nodes_from([
... (4, {"color": "red"}),
... (5, {"color": "green"}),
... ])
Node attributes are discussed further below.

Nodes from one graph can be incorporated into another:
1049
G now contains the nodes of H as nodes of G. In contrast, you could use the graph H as a node in G.
>>> G.add_node(H)
The graph G now contains H as a node. This flexibility is very powerful as it allows graphs of graphs, graphs of files,
graphs of functions and much more. It is worth thinking about how to structure your application so that the nodes are
useful entities. Of course you can always use a unique identifier in G and have a separate dictionary keyed by identifier to
the node information if you prefer.
Note: You should not change the node object if the hash depends on its contents.
A.3 Edges
G can also be grown by adding one edge at a time,
>>> e = (2, 3)
>>> G.add_edge(*e) # unpack edge tuple*
by adding a list of edges,
>>> G.add_edges_from([(1, 2), (1, 3)])
or by adding any ebunch of edges. An ebunch is any iterable container of edge-tuples. An edge-tuple can be a 2-tuple of
nodes or a 3-tuple with 2 nodes followed by an edge attribute dictionary, e.g., (2, 3, {'weight': 3.1415}).
Edge attributes are discussed further below.
There are no complaints when adding existing nodes or edges. For example, after removing all nodes and edges,
>>> G.clear()
we add new nodes/edges and NetworkX quietly ignores any that are already present.
>>> G.add_edges_from([(1, 2), (1, 3)])

>>> G.add_node(1)
>>> G.add_node("spam") # adds node "spam"
>>> G.add_nodes_from("spam") # adds 4 nodes: 's', 'p', 'a', 'm'
>>> G.add_edge(3, 'm')
At this stage the graph G consists of 8 nodes and 3 edges, as can be seen by:
8
3
1050 Appendix A. Tutorial

Note: The order of adjacency reporting (e.g., G.adj, G.successors, G.predecessors) is the order of edge
addition. However, the order of G.edges is the order of the adjacencies which includes both the order of the nodes and
each node’s adjacencies. See example below:
>>> DG.add_edge(2, 1) # adds the nodes in order 2, 1
>>> DG.add_edge(1, 3)
>>> assert list(DG.successors(2)) == [1, 4]
>>> assert list(DG.edges) == [(2, 1), (2, 4), (1, 3), (1, 2)]
A.4 Examining elements of a graph
We can examine the nodes and edges. Four basic graph properties facilitate reporting: G.nodes, G.edges, G.adj
and G.degree. These are set-like views of the nodes, edges, neighbors (adjacencies), and degrees of nodes in a graph.
They offer a continually updated read-only view into the graph structure. They are also dict-like in that you can look up
node and edge data attributes via the views and iterate with data attributes using methods .items(), .data(). If
you want a specific container type instead of a view, you can specify one. Here we use lists, though sets, dicts, tuples and
other containers may be better in other contexts.
>>> list(G.nodes)
[1, 2, 3, 'spam', 's', 'p', 'a', 'm']
>>> list(G.edges)
[(1, 2), (1, 3), (3, 'm')]
>>> list(G.adj[1]) # or list(G.neighbors(1))
[2, 3]
>>> G.degree[1] # the number of edges incident to 1
2
One can specify to report the edges and degree from a subset of all nodes using an nbunch. An nbunch is any of: None
(meaning all nodes), a node, or an iterable container of nodes that is not itself a node in the graph.
>>> G.edges([2, 'm'])

EdgeDataView([(2, 1), ('m', 3)])
>>> G.degree([2, 3])
DegreeView({2: 1, 3: 2})
A.5 Removing elements from a graph
One can remove nodes and edges from the graph in a similar fashion to adding. Use methods Graph.remove_node(),
Graph.remove_nodes_from(), Graph.remove_edge() and Graph.remove_edges_from(), e.g.
>>> G.remove_nodes_from("spam")
>>> list(G.nodes)
[1, 3, 'spam']
A.4. Examining elements of a graph 1051

A.6 Using the graph constructors
Graph objects do not have to be built up incrementally - data specifying graph structure can be passed directly to the
constructors of the various graph classes. When creating a graph structure by instantiating one of the graph classes you
can specify data in several formats.
>>> H = nx.DiGraph(G) # create a DiGraph using the connections from G
>>> list(H.edges())
[(1, 2), (2, 1)]
>>> edgelist = [(0, 1), (1, 2), (2, 3)]
>>> H = nx.Graph(edgelist) # create a graph from an edge list
>>> list(H.edges())
[(0, 1), (1, 2), (2, 3)]
>>> adjacency_dict = {0: (1, 2), 1: (0, 2), 2: (0, 1)}
>>> H = nx.Graph(adjacency_dict) # create a Graph dict mapping nodes to nbrs
>>> list(H.edges())
[(0, 1), (0, 2), (1, 2)]
A.7 What to use as nodes and edges
You might notice that nodes and edges are not specified as NetworkX objects. This leaves you free to use meaningful
items as nodes and edges. The most common choices are numbers or strings, but a node can be any hashable object
(except None), and an edge can be associated with any object x using G.add_edge(n1, n2, object=x).
As an example, n1 and n2 could be protein objects from the RCSB Protein Data Bank, and x could refer to an XML
record of publications detailing experimental observations of their interaction.
We have found this power quite useful, but its abuse can lead to surprising behavior unless one is familiar with Python. If
in doubt, consider using convert_node_labels_to_integers() to obtain a more traditional graph with integer
labels.
A.8 Accessing edges and neighbors
In addition to the views Graph.edges, and Graph.adj, access to edges and neighbors is possible using subscript
notation.
>>> G = nx.Graph([(1, 2, {"color": "yellow"})])

>>> G[1] # same as G.adj[1]
AtlasView({2: {'color': 'yellow'}})
>>> G[1][2]
{'color': 'yellow'}
>>> G.edges[1, 2]
{'color': 'yellow'}
You can get/set the attributes of an edge using subscript notation if the edge already exists.
>>> G[1][3]['color'] = "blue"
>>> G.edges[1, 2]['color'] = "red"
>>> G.edges[1, 2]
{'color': 'red'}

Fast examination of all (node, adjacency) pairs is achieved using G.adjacency(), or G.adj.items(). Note that
for undirected graphs, adjacency iteration sees each edge twice.
>>> FG = nx.Graph()
>>> FG.add_weighted_edges_from([(1, 2, 0.125), (1, 3, 0.75), (2, 4, 1.2), (3, 4, 0.
,→375)])
>>> for n, nbrs in FG.adj.items():

... for nbr, eattr in nbrs.items():
... wt = eattr['weight']
... if wt < 0.5: print(f"({n}, {nbr}, {wt:.3})")
(1, 2, 0.125)
(2, 1, 0.125)
(3, 4, 0.375)
(4, 3, 0.375)
Convenient access to all edges is achieved with the edges property.
>>> for (u, v, wt) in FG.edges.data('weight'):

... if wt < 0.5:
... print(f"({u}, {v}, {wt:.3})")
(1, 2, 0.125)
(3, 4, 0.375)
A.9 Adding attributes to graphs, nodes, and edges
Attributes such as weights, labels, colors, or whatever Python object you like, can be attached to graphs, nodes, or edges.
Each graph, node, and edge can hold key/value attribute pairs in an associated attribute dictionary (the keys must be
hashable). By default these are empty, but attributes can be added or changed using add_edge, add_node or direct
manipulation of the attribute dictionaries named G.graph, G.nodes, and G.edges for a graph G.
A.9.1 Graph attributes
Assign graph attributes when creating a new graph
>>> G = nx.Graph(day="Friday")
>>> G.graph
{'day': 'Friday'}
Or you can modify attributes later
>>> G.graph['day'] = "Monday"

>>> G.graph
{'day': 'Monday'}
A.9. Adding attributes to graphs, nodes, and edges 1053

A.9.2 Node attributes
Add node attributes using add_node(), add_nodes_from(), or G.nodes
>>> G.add_node(1, time='5pm')

>>> G.add_nodes_from([3], time='2pm')
>>> G.nodes[1]
{'time': '5pm'}
>>> G.nodes[1]['room'] = 714
>>> G.nodes.data()
NodeDataView({1: {'time': '5pm', 'room': 714}, 3: {'time': '2pm'}})
Note that adding a node to G.nodes does not add it to the graph, use G.add_node() to add new nodes. Similarly
for edges.
A.9.3 Edge Attributes
Add/change edge attributes using add_edge(), add_edges_from(), or subscript notation.
>>> G.add_edge(1, 2, weight=4.7 )

>>> G.add_edges_from([(3, 4), (4, 5)], color='red')
>>> G.add_edges_from([(1, 2, {'color': 'blue'}), (2, 3, {'weight': 8})])
>>> G[1][2]['weight'] = 4.7
>>> G.edges[3, 4]['weight'] = 4.2
The special attribute weight should be numeric as it is used by algorithms requiring weighted edges.
A.10 Directed graphs
The DiGraph class provides additional methods and properties specific to directed edges, e.g., DiGraph.
out_edges, DiGraph.in_degree, DiGraph.predecessors, DiGraph.successors etc. To allow al-
gorithms to work with both classes easily, the directed versions of neighbors is equivalent to successors while
degree reports the sum of in_degree and out_degree even though that may feel inconsistent at times.
>>> DG.add_weighted_edges_from([(1, 2, 0.5), (3, 1, 0.75)])
>>> DG.out_degree(1, weight='weight')
0.5
>>> DG.degree(1, weight='weight')
1.25
>>> list(DG.successors(1))
[2]
>>> list(DG.neighbors(1))
[2]
Some algorithms work only for directed graphs and others are not well defined for directed graphs. Indeed the tendency
to lump directed and undirected graphs together is dangerous. If you want to treat a directed graph as undirected for some
measurement you should probably convert it using Graph.to_undirected() or with
>>> H = nx.Graph(G) # create an undirected graph H from a directed graph G

A.11 Multigraphs
NetworkX provides classes for graphs which allow multiple edges between any pair of nodes. The MultiGraph and
MultiDiGraph classes allow you to add the same edge twice, possibly with different edge data. This can be powerful
for some applications, but many algorithms are not well defined on such graphs. Where results are well defined, e.g.,
MultiGraph.degree() we provide the function. Otherwise you should convert to a standard graph in a way that
makes the measurement well defined.
>>> MG = nx.MultiGraph()
>>> MG.add_weighted_edges_from([(1, 2, 0.5), (1, 2, 0.75), (2, 3, 0.5)])
>>> dict(MG.degree(weight='weight'))
{1: 1.25, 2: 1.75, 3: 0.5}
>>> GG = nx.Graph()
>>> for n, nbrs in MG.adjacency():
... for nbr, edict in nbrs.items():
... minvalue = min([d['weight'] for d in edict.values()])
... GG.add_edge(n, nbr, weight = minvalue)
...
>>> nx.shortest_path(GG, 1, 3)
[1, 2, 3]
A.12 Graph generators and graph operations
In addition to constructing graphs node-by-node or edge-by-edge, they can also be generated by
A.12.1 1. Applying classic graph operations, such as:
subgraph(G, nbunch) Returns the subgraph induced on nodes in nbunch.

union(G, H[, rename]) Combine graphs G and H.
disjoint_union(G, H) Combine graphs G and H.
cartesian_product(G, H) Returns the Cartesian product of G and H.
compose(G, H) Compose graph G with H by combining nodes and edges
into a single graph.
complement(G) Returns the graph complement of G.
create_empty_copy(G[, with_data]) Returns a copy of the graph G with all of the edges re-
moved.
to_undirected(graph) Returns an undirected view of the graph graph.
to_directed(graph) Returns a directed view of the graph graph.
A.12.2 2. Using a call to one of the classic small graphs, e.g.,
petersen_graph([create_using]) Returns the Petersen graph.

tutte_graph([create_using]) Returns the Tutte graph.
sedgewick_maze_graph([create_using]) Return a small maze with a cycle.
tetrahedral_graph([create_using]) Returns the 3-regular Platonic Tetrahedral graph.
A.11. Multigraphs 1055

A.12.3 3. Using a (constructive) generator for a classic graph, e.g.,
complete_graph(n[, create_using]) Return the complete graph K_n with n nodes.

complete_bipartite_graph(n1, n2[, cre- Returns the complete bipartite graph K_{n_1,n_2}.
ate_using])
barbell_graph(m1, m2[, create_using]) Returns the Barbell Graph: two complete graphs con-
nected by a path.
lollipop_graph(m, n[, create_using]) Returns the Lollipop Graph; K_m connected to P_n.
like so:
>>> K_5 = nx.complete_graph(5)

>>> K_3_5 = nx.complete_bipartite_graph(3, 5)
>>> barbell = nx.barbell_graph(10, 10)
>>> lollipop = nx.lollipop_graph(10, 20)
A.12.4 4. Using a stochastic graph generator, e.g,
erdos_renyi_graph(n, p[, seed, directed]) Returns a Gn,p random graph, also known as an Erdős-
watts_strogatz_graph(n, k, p[, seed]) Returns a Watts–Strogatz small-world graph.
barabasi_albert_graph(n, m[, seed, ...]) Returns a random graph using Barabási–Albert preferen-
tial attachment
random_lobster(n, p1, p2[, seed]) Returns a random lobster graph.
like so:
>>> er = nx.erdos_renyi_graph(100, 0.15)

>>> ws = nx.watts_strogatz_graph(30, 3, 0.1)
>>> ba = nx.barabasi_albert_graph(100, 5)
>>> red = nx.random_lobster(100, 0.9, 0.9)
A.12.5 5. Reading a graph stored in a file using common graph formats
NetworkX supports many popular formats, such as edge lists, adjacency lists, GML, GraphML, LEDA and others.
>>> nx.write_gml(red, "path.to.file")

>>> mygraph = nx.read_gml("path.to.file")
For details on graph formats see Reading and writing graphs and for graph generator functions see Graph generators

A.13 Analyzing graphs
The structure of G can be analyzed using various graph-theoretic functions such as:
>>> G = nx.Graph()
>>> G.add_edges_from([(1, 2), (1, 3)])
>>> G.add_node("spam") # adds node "spam"
>>> list(nx.connected_components(G))
[{1, 2, 3}, {'spam'}]
>>> sorted(d for n, d in G.degree())
[0, 1, 1, 2]
>>> nx.clustering(G)
{1: 0, 2: 0, 3: 0, 'spam': 0}
Some functions with large output iterate over (node, value) 2-tuples. These are easily stored in a dict structure if you
desire.
>>> sp = dict(nx.all_pairs_shortest_path(G))
>>> sp[3]
{3: [3], 1: [3, 1], 2: [3, 1, 2]}
See Algorithms for details on graph algorithms supported.
A.14 Drawing graphs
NetworkX is not primarily a graph drawing package but basic drawing with Matplotlib as well as an interface to use
the open source Graphviz software package are included. These are part of the networkx.drawing module and will be
imported if possible.
First import Matplotlib’s plot interface (pylab works too)
To test if the import of nx_pylab was successful draw G using one of
>>> nx.draw(G, with_labels=True, font_weight='bold')
>>> nx.draw_shell(G, nlist=[range(5, 10), range(5)], with_labels=True, font_weight=
,→'bold')
A.13. Analyzing graphs 1057

0 4
8
4 3
5
5 6
1 6 9
0
3
9 7 8
2
7
2 1
when drawing to an interactive display. Note that you may need to issue a Matplotlib
>>> plt.show()
command if you are not using matplotlib in interactive mode.
>>> options = {
... 'node_color': 'black',
... 'node_size': 100,
... 'width': 3,
... }
>>> nx.draw_random(G, **options)
>>> nx.draw_circular(G, **options)
>>> nx.draw_spectral(G, **options)
>>> nx.draw_shell(G, nlist=[range(5,10), range(5)], **options)

You can find additional options via draw_networkx() and layouts via the layout module. You can use multiple
shells with draw_shell().
>>> shells = [[2, 3, 4, 5, 6], [8, 1, 0, 19, 18, 17, 16, 15, 14, 7], [9, 10, 11, 12,␣
,→13]]
>>> nx.draw_shell(G, nlist=shells, **options)
A.14. Drawing graphs 1059

To save drawings to a file, use, for example
>>> nx.draw(G)
>>> plt.savefig("path.png")
This function writes to the file path.png in the local directory. If Graphviz and PyGraphviz or pydot, are avail-
able on your system, you can also use networkx.drawing.nx_agraph.graphviz_layout or networkx.
drawing.nx_pydot.graphviz_layout to get the node positions, or write the graph in dot format for further
processing.
>>> from networkx.drawing.nx_pydot import write_dot

>>> nx.draw(G, pos=pos)
>>> write_dot(G, 'file.dot')
See Drawing for additional details.

• ;
• ;
• .

BIBLIOGRAPHY
[1] White, Douglas R., and Mark Newman. 2001 A Fast Algorithm for Node-Independent Paths. Santa Fe
Institute Working Paper #01-07-035 http://eclectic.ss.uci.edu/~drwhite/working.pdf
[1] Torrents, J. and F. Ferraro (2015) Structural Cohesion: Visualization and Heuristics for Fast Computation.
https://arxiv.org/pdf/1503.04476v1
[2] White, Douglas R., and Mark Newman (2001) A Fast Algorithm for Node-Independent Paths.
Santa Fe Institute Working Paper #01-07-035 https://www.santafe.edu/research/results/working-papers/
fast-approximation-algorithms-for-finding-node-ind
[3] Moody, J. and D. White (2003). Social cohesion and embeddedness: A hierarchical conception of social
groups. American Sociological Review 68(1), 103–28. https://doi.org/10.2307/3088904
[1] Wikipedia: Independent set
[2] Boppana, R., & Halldórsson, M. M. (1992). Approximating maximum independent sets by excluding sub-
graphs. BIT Numerical Mathematics, 32(2), 180–196. Springer.
graphs. BIT Numerical Mathematics, 32(2), 180–196. Springer. doi:10.1007/BF01994876
graphs. BIT Numerical Mathematics, 32(2), 180–196. Springer.
[1] Pattabiraman, Bharath, et al. “Fast Algorithms for the Maximum Clique Problem on Massive Graphs with
Applications to Overlapping Community Detection.” Internet Mathematics 11.4-5 (2015): 421–448. <https:
//doi.org/10.1080/15427951.2014.986778>
[1] Schank, Thomas, and Dorothea Wagner. Approximating clustering coefficient and transitivity. Universität
Karlsruhe, Fakultät für Informatik, 2004. https://doi.org/10.5445/IR/1000001239
[1] Magnien, Clémence, Matthieu Latapy, and Michel Habib. Fast computation of empirically tight bounds for
the diameter of massive graphs. Journal of Experimental Algorithmics (JEA), 2009. https://arxiv.org/pdf/
0904.2728.pdf
[2] Crescenzi, Pierluigi, Roberto Grossi, Leonardo Lanzi, and Andrea Marino. On computing the diameter
of real-world directed (weighted) graphs. International Symposium on Experimental Algorithms. Springer,
Berlin, Heidelberg, 2012. https://courses.cs.ut.ee/MTAT.03.238/2014_fall/uploads/Main/diameter.pdf
[1] Vazirani, Vijay V. Approximation Algorithms. Springer Science & Business Media, 2001.
1061
[1] Vazirani, Vijay Approximation Algorithms (2001)

[1] Steiner_tree_problem on Wikipedia. https://en.wikipedia.org/wiki/Steiner_tree_problem
[2] Kou, L., G. Markowsky, and L. Berman. 1981. ‘A Fast Algorithm for Steiner Trees’. Acta Informatica 15
(2): 141–45. https://doi.org/10.1007/BF00288961.
[3] Mehlhorn, Kurt. 1988. ‘A Faster Approximation Algorithm for the Steiner Problem in Graphs’. Information
Processing Letters 27 (3): 125–28. https://doi.org/10.1016/0020-0190(88)90066-X.
[1] Christofides, Nicos. “Worst-case analysis of a new heuristic for the travelling salesman problem.” No. RR-
388. Carnegie-Mellon Univ Pittsburgh Pa Management Sciences Research Group, 1976.
[1] A. Asadpour, M. X. Goemans, A. Madry, S. O. Gharan, and A. Saberi, An o(log n/log log n)-approximation
algorithm for the asymmetric traveling salesman problem, Operations research, 65 (2017), pp. 1043–1061
[1] Hans L. Bodlaender and Arie M. C. A. Koster. 2010. “Treewidth computations I.Upper bounds”. Inf. Com-
put. 208, 3 (March 2010),259-275. http://dx.doi.org/10.1016/j.ic.2009.03.008
[2] Hans L. Bodlaender. “Discovering Treewidth”. Institute of Information and Computing Sciences, Utrecht
University. Technical Report UU-CS-2005-018. http://www.cs.uu.nl
[3] K. Wang, Z. Lu, and J. Hicks Treewidth. https://web.archive.org/web/20210507025929/http://web.eecs.
utk.edu/~cphill25/cs594_spring2015_projects/treewidth.pdf
[1] Bar-Yehuda, R., and Even, S. (1985). “A local-ratio theorem for approximating the weighted vertex cover
problem.” Annals of Discrete Mathematics, 25, 27–46 <http://www.cs.technion.ac.il/~reuven/PDF/vc_lr.
pdf>
[1] M. E. J. Newman, Mixing patterns in networks, Physical Review E, 67 026126, 2003
[2] Foster, J.G., Foster, D.V., Grassberger, P. & Paczuski, M. Edge direction and the structure of networks,
PNAS 107, 10815-20 (2010).
[1] M. E. J. Newman, Mixing patterns in networks, Physical Review E, 67 026126, 2003
[1] M. E. J. Newman, Mixing patterns in networks Physical Review E, 67 026126, 2003
[1] M. E. J. Newman, Mixing patterns in networks Physical Review E, 67 026126, 2003
[2] Foster, J.G., Foster, D.V., Grassberger, P. & Paczuski, M. Edge direction and the structure of networks,
PNAS 107, 10815-20 (2010).
[1] A. Barrat, M. Barthélemy, R. Pastor-Satorras, and A. Vespignani, “The architecture of complex weighted
networks”. PNAS 101 (11): 3747–3752 (2004).
[1] A. Barrat, M. Barthélemy, R. Pastor-Satorras, and A. Vespignani, “The architecture of complex weighted
networks”. PNAS 101 (11): 3747–3752 (2004).
[1] Ekkehard Köhler, “Recognizing Graphs without asteroidal triples”, Journal of Discrete Algorithms 2, pages
439-452, 2004. https://www.sciencedirect.com/science/article/pii/S157086670400019X
[1] John E. Hopcroft and Richard M. Karp. “An n^{5 / 2} Algorithm for Maximum Matchings in Bipartite
Graphs” In: SIAM Journal of Computing 2.4 (1973), pp. 225–231. <https://doi.org/10.1137/0202019>.
[1] Richard Manning Karp: An algorithm to Solve the m x n Assignment Problem in Expected Time O(mn log
n). Networks, 10(2):143–152, 1980.
[1] https://en.wikipedia.org/wiki/Adjacency_matrix#Adjacency_matrix_of_a_bipartite_graph
[2] Scipy Dev. References, “Sparse Matrices”, https://docs.scipy.org/doc/scipy/reference/sparse.html
[1] Borgatti, S.P. and Halgin, D. In press. “Analyzing Affiliation Networks”. In Carrington, P. and Scott, J.
(eds) The Sage Handbook of Social Network Analysis. Sage Publications.
1062 Bibliography
[1] Scientific collaboration networks: II. Shortest paths, weighted networks, and centrality, M. E. J. Newman,
Phys. Rev. E 64, 016132 (2001).
[1] Borgatti, S.P. and Halgin, D. In press. Analyzing Affiliation Networks. In Carrington, P. and Scott, J. (eds)
The Sage Handbook of Social Network Analysis. Sage Publications.
[1] E. Estrada and J. A. Rodríguez-Velázquez, “Spectral measures of bipartivity in complex networks”, PhysRev
E 72, 046105 (2005)
[1] Latapy, Matthieu, Clémence Magnien, and Nathalie Del Vecchio (2008). Basic notions for the analysis of
large two-mode networks. Social Networks 30(1), 31–48.
[1] Robins, G. and M. Alexander (2004). Small worlds among interlocking directors: Network structure and
distance in bipartite graphs. Computational & Mathematical Organization Theory 10(1), 69–94.
[1] Borgatti, S.P. and Halgin, D. In press. “Analyzing Affiliation Networks”. In Carrington, P. and Scott,
J. (eds) The Sage Handbook of Social Network Analysis. Sage Publications. https://dx.doi.org/10.4135/
9781446294413.n28
9781446294413.n28
9781446294413.n28
[1] Guillaume, J.L. and Latapy, M., Bipartite graphs as models of complex networks. Physica A: Statistical
Mechanics and its Applications, 2006, 371(2), pp.795-813.
[2] Jean-Loup Guillaume and Matthieu Latapy, Bipartite structure of all complex networks, Inf. Process. Lett.
90, 2004, pg. 215-221 https://doi.org/10.1016/j.ipl.2004.03.007
[1] Vladimir Batagelj and Ulrik Brandes, “Efficient generation of large random networks”, Phys. Rev. E, 71,
036113, 2005.
[1] https://en.wikipedia.org/wiki/Bridge_%28graph_theory%29#Bridge-Finding_with_Chain_
Decompositions
[1] Phillip Bonacich. “Power and Centrality: A Family of Measures.” American Journal of So-
ciology 92(5):1170–1182, 1986 <http://www.leonidzhukov.net/hse/2014/socialnetworks/papers/
Bonacich-Centrality.pdf>
[2] Mark E. J. Newman. Networks: An Introduction. Oxford University Press, USA, 2010, pp. 169.
[1] Phillip Bonacich: Power and Centrality: A Family of Measures. American Journal of Sociology 92(5):1170–
1182, 1986 http://www.leonidzhukov.net/hse/2014/socialnetworks/papers/Bonacich-Centrality.pdf
[2] Mark E. J. Newman: Networks: An Introduction. Oxford University Press, USA, 2010, pp. 169.
[1] Mark E. J. Newman: Networks: An Introduction. Oxford University Press, USA, 2010, p. 720.
[2] Leo Katz: A New Status Index Derived from Sociometric Index. Psychometrika 18(1):39–43, 1953 https:
//link.springer.com/content/pdf/10.1007/BF02289026.pdf
Bibliography 1063
[1] Mark E. J. Newman: Networks: An Introduction. Oxford University Press, USA, 2010, p. 173.
[2] Leo Katz: A New Status Index Derived from Sociometric Index. Psychometrika 18(1):39–43, 1953 https:
//link.springer.com/content/pdf/10.1007/BF02289026.pdf
[1] Linton C. Freeman: Centrality in networks: I. Conceptual clarification. Social Networks 1:215-239, 1979.
https://doi.org/10.1016/0378-8733(78)90021-7
[2] pg. 201 of Wasserman, S. and Faust, K., Social Network Analysis: Methods and Applications, 1994, Cam-
bridge University Press.
[1] Freeman, L.C., 1979. Centrality in networks: I. Conceptual clarification. Social Networks 1, 215–239.
https://doi.org/10.1016/0378-8733(78)90021-7
[2] Sariyuce, A.E. ; Kaya, K. ; Saule, E. ; Catalyiirek, U.V. Incremental Algorithms for Closeness Centrality.
2013 IEEE International Conference on Big Data http://sariyuce.com/papers/bigdata13.pdf
[1] Ulrik Brandes and Daniel Fleischer, Centrality Measures Based on Current Flow. Proc. 22nd Symp. The-
oretical Aspects of Computer Science (STACS ‘05). LNCS 3404, pp. 533-544. Springer-Verlag, 2005.
https://doi.org/10.1007/978-3-540-31856-9_44
[2] Karen Stephenson and Marvin Zelen: Rethinking centrality: Methods and examples. Social Networks
11(1):1-37, 1989. https://doi.org/10.1016/0378-8733(89)90016-6
[1] Ulrik Brandes and Daniel Fleischer, Centrality Measures Based on Current Flow. Proc. 22nd Symp. The-
https://doi.org/10.1007/978-3-540-31856-9_44
[2] Karen Stephenson and Marvin Zelen: Rethinking centrality: Methods and examples. Social Networks
11(1):1-37, 1989. https://doi.org/10.1016/0378-8733(89)90016-6
[1] Ulrik Brandes: A Faster Algorithm for Betweenness Centrality. Journal of Mathematical Sociology
25(2):163-177, 2001. https://doi.org/10.1080/0022250X.2001.9990249
[2] Ulrik Brandes: On Variants of Shortest-Path Betweenness Centrality and their Generic Computation. Social
Networks 30(2):136-145, 2008. https://doi.org/10.1016/j.socnet.2007.11.001
[3] Ulrik Brandes and Christian Pich: Centrality Estimation in Large Networks. International Journal of Bifur-
cation and Chaos 17(7):2303-2318, 2007. https://dx.doi.org/10.1142/S0218127407018403
[4] Linton C. Freeman: A set of measures of centrality based on betweenness. Sociometry 40: 35–41, 1977
https://doi.org/10.2307/3033543
[1] Ulrik Brandes, A Faster Algorithm for Betweenness Centrality. Journal of Mathematical Sociology
25(2):163-177, 2001. https://doi.org/10.1080/0022250X.2001.9990249
[1] A Faster Algorithm for Betweenness Centrality. Ulrik Brandes, Journal of Mathematical Sociology
25(2):163-177, 2001. https://doi.org/10.1080/0022250X.2001.9990249
[1] Ulrik Brandes, A Faster Algorithm for Betweenness Centrality. Journal of Mathematical Sociology
25(2):163-177, 2001. https://doi.org/10.1080/0022250X.2001.9990249
[1] Centrality Measures Based on Current Flow. Ulrik Brandes and Daniel Fleischer, Proc. 22nd Symp. The-
https://doi.org/10.1007/978-3-540-31856-9_44
1064 Bibliography
[2] A measure of betweenness centrality based on random walks, M. E. J. Newman, Social Networks 27, 39-54
(2005).
https://doi.org/10.1007/978-3-540-31856-9_44
(2005).
[1] Ulrik Brandes and Daniel Fleischer: Centrality Measures Based on Current Flow. Proc. 22nd Symp. The-
https://doi.org/10.1007/978-3-540-31856-9_44
https://doi.org/10.1007/978-3-540-31856-9_44
(2005).
https://doi.org/10.1007/978-3-540-31856-9_44
(2005).
[1] Ernesto Estrada, Desmond J. Higham, Naomichi Hatano, “Communicability Betweenness in Complex Net-
works” Physica A 388 (2009) 764-774. https://arxiv.org/abs/0905.4102
[1] M G Everett and S P Borgatti: The Centrality of Groups and Classes. Journal of Mathematical Sociology.
23(3): 181-201. 1999. http://www.analytictech.com/borgatti/group_centrality.htm
Networks 30(2):136-145, 2008. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.72.9610&rep=
rep1&type=pdf
[3] Sourav Medya et. al.: Group Centrality Maximization via Network Design. SIAM International Conference
on Data Mining, SDM 2018, 126–134. https://sites.cs.ucsb.edu/~arlei/pubs/sdm18.pdf
[4] Rami Puzis, Yuval Elovici, and Shlomi Dolev. “Fast algorithm for successive computation of group be-
tweenness centrality.” https://journals.aps.org/pre/pdf/10.1103/PhysRevE.76.056709
[2] J. Zhao et. al.: Measuring and Maximizing Group Closeness Centrality over Disk Resident Graphs. WWW-
Conference Proceedings, 2014. 689-694. https://doi.org/10.1145/2567948.2579356
[2] Rami Puzis, Yuval Elovici, and Shlomi Dolev: “Finding the Most Prominent Group in Complex Net-
works” AI communications 20(4): 287-296, 2007. https://www.researchgate.net/profile/Rami_Puzis2/
publication/220308855
[3] Sourav Medya et. al.: Group Centrality Maximization via Network Design. SIAM International Conference
on Data Mining, SDM 2018, 126–134. https://sites.cs.ucsb.edu/~arlei/pubs/sdm18.pdf
Bibliography 1065
[4] Rami Puzis, Yuval Elovici, and Shlomi Dolev. “Fast algorithm for successive computation of group be-
tweenness centrality.” https://journals.aps.org/pre/pdf/10.1103/PhysRevE.76.056709
[1] Mark E. J. Newman: Scientific collaboration networks. II. Shortest paths, weighted networks, and centrality.
Physical Review E 64, 016132, 2001. http://journals.aps.org/pre/abstract/10.1103/PhysRevE.64.016132
[2] Kwang-Il Goh, Byungnam Kahng and Doochul Kim Universal behavior of Load Distribution in Scale-Free
Networks. Physical Review Letters 87(27):1–4, 2001. https://doi.org/10.1103/PhysRevLett.87.278701
[1] Ernesto Estrada, Juan A. Rodriguez-Velazquez, “Subgraph centrality in complex networks”, Physical Re-
view E 71, 056103 (2005). https://arxiv.org/abs/cond-mat/0504730
[1] Ernesto Estrada, Juan A. Rodriguez-Velazquez, “Subgraph centrality in complex networks”, Physical Re-
view E 71, 056103 (2005). https://arxiv.org/abs/cond-mat/0504730
[1] E. Estrada, “Characterization of 3D molecular structure”, Chem. Phys. Lett. 319, 713 (2000). https://doi.
org/10.1016/S0009-2614(00)00158-5
[2] José Antonio de la Peñaa, Ivan Gutman, Juan Rada, “Estimating the Estrada index”, Linear Algebra and its
Applications. 427, 1 (2007). https://doi.org/10.1016/j.laa.2007.06.020
[1] Boldi, Paolo, and Sebastiano Vigna. “Axioms for centrality.” Internet Mathematics 10.3-4 (2014): 222-262.
[1] Romantic Partnerships and the Dispersion of Social Ties: A Network Analysis of Relationship Status on
Facebook. Lars Backstrom, Jon Kleinberg. https://arxiv.org/pdf/1310.6753v1.pdf
[1] Mones, Enys, Lilla Vicsek, and Tamás Vicsek. “Hierarchy Measure for Complex Networks.” PLoS ONE 7.3
(2012): e33799. https://doi.org/10.1371/journal.pone.0033799
[1] Mones, Enys, Lilla Vicsek, and Tamás Vicsek. “Hierarchy Measure for Complex Networks.” PLoS ONE 7.3
(2012): e33799. https://doi.org/10.1371/journal.pone.0033799
[1] Mahendra Piraveenan, Mikhail Prokopenko, Liaquat Hossain Percolation Centrality: Quantifying Graph-
Theoretic Impact of Nodes during Percolation in Networks http://journals.plos.org/plosone/article?id=10.
1371/journal.pone.0053095
[2] Ulrik Brandes: A Faster Algorithm for Betweenness Centrality. Journal of Mathematical Sociology
25(2):163-177, 2001. https://doi.org/10.1080/0022250X.2001.9990249
[1] Anne-Marie Kermarrec, Erwan Le Merrer, Bruno Sericola, Gilles Trédan “Second order centrality:
Distributed assessment of nodes criticity in complex networks”, Elsevier Computer Communications
34(5):619-628, 2011.
[1] Stephen Levine (1980) J. theor. Biol. 83, 195-207
[1] Samuel Johnson, Virginia Dominguez-Garcia, Luca Donetti, Miguel A. Munoz (2014) PNAS “Trophic
coherence determines food-web stability”
[1] Samuel Johnson, Virginia Dominguez-Garcia, Luca Donetti, Miguel A. Munoz (2014) PNAS “Trophic
coherence determines food-web stability”
[1] Zhang, J.-X. et al. (2016). Identifying a set of influential spreaders in complex networks. Sci. Rep. 6, 27823;
doi: 10.1038/srep27823.
[1] Jens M. Schmidt (2013). “A simple test on 2-vertex- and 2-edge-connectivity.” Information Processing Let-
ters, 113, 241–244. Elsevier. <https://doi.org/10.1016/j.ipl.2013.01.016>
[1] R. E. Tarjan and M. Yannakakis, Simple linear-time algorithms to test chordality of graphs, test acyclicity
of hypergraphs, and selectively reduce acyclic hypergraphs, SIAM J. Comput., 13 (1984), pp. 566–579.
[1] https://en.wikipedia.org/wiki/Tree_decomposition#Treewidth
[1] Berry, Anne & Blair, Jean & Heggernes, Pinar & Peyton, Barry. (2004) Maximum Cardinality Search for
Computing Minimal Triangulations of Graphs. Algorithmica. 39. 287-298. 10.1007/s00453-004-1084-3.
1066 Bibliography
[1] Learning Bounded Treewidth Bayesian Networks. Gal Elidan, Stephen Gould; JMLR, 9(Dec):2699–2731,
2008. http://jmlr.csail.mit.edu/papers/volume9/elidan08a/elidan08a.pdf
[1] clique problem:: https://en.wikipedia.org/wiki/Clique_problem
[1] Yun Zhang, Abu-Khzam, F.N., Baldwin, N.E., Chesler, E.J., Langston, M.A., Samatova, N.F., “Genome-
Scale Computational Approaches to Memory-Intensive Applications in Systems Biology”. Supercomputing,
2005. Proceedings of the ACM/IEEE SC 2005 Conference, pp. 12, 12–18 Nov. 2005. <https://doi.org/10.
1109/SC.2005.29>.
[1] Bron, C. and Kerbosch, J. “Algorithm 457: finding all cliques of an undirected graph”. Communications of
the ACM 16, 9 (Sep. 1973), 575–577. <http://portal.acm.org/citation.cfm?doid=362342.362367>
[2] Etsuji Tomita, Akira Tanaka, Haruhisa Takahashi, “The worst-case time complexity for generating all max-
imal cliques and computational experiments”, Theoretical Computer Science, Volume 363, Issue 1, Com-
puting and Combinatorics, 10th Annual International Conference on Computing and Combinatorics (CO-
COON 2004), 25 October 2006, Pages 28–42 <https://doi.org/10.1016/j.tcs.2006.06.015>
[3] F. Cazals, C. Karande, “A note on the problem of reporting maximal cliques”, Theoretical Computer Science,
Volume 407, Issues 1–3, 6 November 2008, Pages 564–568, <https://doi.org/10.1016/j.tcs.2008.05.010>
[1] Bron, C. and Kerbosch, J. “Algorithm 457: finding all cliques of an undirected graph”. Communications of
the ACM 16, 9 (Sep. 1973), 575–577. <http://portal.acm.org/citation.cfm?doid=362342.362367>
[2] Etsuji Tomita, Akira Tanaka, Haruhisa Takahashi, “The worst-case time complexity for generating all max-
imal cliques and computational experiments”, Theoretical Computer Science, Volume 363, Issue 1, Com-
puting and Combinatorics, 10th Annual International Conference on Computing and Combinatorics (CO-
COON 2004), 25 October 2006, Pages 28–42 <https://doi.org/10.1016/j.tcs.2006.06.015>
[3] F. Cazals, C. Karande, “A note on the problem of reporting maximal cliques”, Theoretical Computer Science,
Volume 407, Issues 1–3, 6 November 2008, Pages 564–568, <https://doi.org/10.1016/j.tcs.2008.05.010>
[1] Tavares, W.A., Neto, M.B.C., Rodrigues, C.D., Michelon, P.: Um algoritmo de branch and bound para o
problema da clique máxima ponderada. Proceedings of XLVII SBPO 1 (2015).
[2] Warrent, Jeffrey S, Hicks, Illya V.: Combinatorial Branch-and-Bound for the Maximum Weight Independent
Set Problem. Technical Report, Texas A&M University (2016).
[1] Generalizations of the clustering coefficient to weighted complex networks by J. Saramäki, M. Kivelä, J.-
P. Onnela, K. Kaski, and J. Kertész, Physical Review E, 75 027105 (2007). http://jponnela.com/web_
documents/a9.pdf
[2] Intensity and coherence of motifs in weighted complex networks by J. P. Onnela, J. Saramäki, J. Kertész,
and K. Kaski, Physical Review E, 71(6), 065103 (2005).
[3] Generalization of Clustering Coefficients to Signed Correlation Networks by G. Costantini and M. Perugini,
PloS one, 9(2), e88669 (2014).
[4] Clustering in complex directed networks by G. Fagiolo, Physical Review E, 76(2), 026107 (2007).
[1] Generalizations of the clustering coefficient to weighted complex networks by J. Saramäki, M. Kivelä, J.-
P. Onnela, K. Kaski, and J. Kertész, Physical Review E, 75 027105 (2007). http://jponnela.com/web_
documents/a9.pdf
[2] Marcus Kaiser, Mean clustering coefficients: the role of isolated nodes and leafs on clustering measures for
small-world networks. https://arxiv.org/abs/0802.2512
[1] Pedro G. Lind, Marta C. González, and Hans J. Herrmann. 2005 Cycles and clustering in bipartite networks.
Physical Review E (72) 056127.
[2] Zhang, Peng et al. Clustering Coefficient and Community Structure of Bipartite Networks. Physica A: Sta-
tistical Mechanics and its Applications 387.27 (2008): 6869–6875. https://arxiv.org/abs/0710.0117v1
Bibliography 1067
[1] Networks with arbitrary edge multiplicities by V. Zlatić, D. Garlaschelli and G. Caldarelli, EPL (Europhysics
Letters), Volume 97, Number 2 (2012). https://iopscience.iop.org/article/10.1209/0295-5075/97/28005
[1] Adrian Kosowski, and Krzysztof Manuszewski, Classical Coloring of Graphs, Graph Colorings, 2-19, 2004.
ISBN 0-8218-3458-4.
[2] David W. Matula, and Leland L. Beck, “Smallest-last ordering and clustering and graph coloring algo-
rithms.” J. ACM 30, 3 (July 1983), 417–427. <https://doi.org/10.1145/2402.322385>
[3] Maciej M. Sysło, Narsingh Deo, Janusz S. Kowalik, Discrete Optimization Algorithms with Pascal Pro-
grams, 415-424, 1983. ISBN 0-486-45353-7.
[1] Kierstead, H. A., Kostochka, A. V., Mydlarz, M., & Szemerédi, E. (2010). A fast algorithm for equitable
coloring. Combinatorica, 30(2), 217-224.
[1] Ernesto Estrada, Naomichi Hatano, “Communicability in complex networks”, Phys. Rev. E 77, 036111
(2008). https://arxiv.org/abs/0707.0756
[1] Ernesto Estrada, Naomichi Hatano, “Communicability in complex networks”, Phys. Rev. E 77, 036111
(2008). https://arxiv.org/abs/0707.0756
[1] Kernighan, B. W.; Lin, Shen (1970). “An efficient heuristic procedure for partitioning graphs.” Bell Systems
Technical Journal 49: 291–307. Oxford University Press 2011.
[1] Gergely Palla, Imre Derényi, Illés Farkas1, and Tamás Vicsek, Uncovering the overlapping community
structure of complex networks in nature and society Nature 435, 814-818, 2005, doi:10.1038/nature03607
[1] Newman, M. E. J. “Networks: An Introduction”, page 224 Oxford University Press 2011.
[2] Clauset, A., Newman, M. E., & Moore, C. “Finding community structure in very large networks.” Physical
Review E 70(6), 2004.
[3] Reichardt and Bornholdt “Statistical Mechanics of Community Detection” Phys. Rev. E74, 2006.
[4] Newman, M. E. J.”Analysis of weighted networks” Physical Review E 70(5 Pt 2):056131, 2004.
[1] Raghavan, Usha Nandini, Réka Albert, and Soundar Kumara. “Near linear time algorithm to detect com-
munity structures in large-scale networks.” Physical Review E 76.3 (2007): 036106.
[1] Cordasco, G., & Gargano, L. (2010, December). Community detection via semi-synchronous label propa-
gation algorithms. In Business Applications of Social Network Analysis (BASNA), 2010 IEEE International
Workshop on (pp. 1-8). IEEE.
[1] Blondel, V.D. et al. Fast unfolding of communities in large networks. J. Stat. Mech 10008, 1-12(2008).
https://doi.org/10.1088/1742-5468/2008/10/P10008
[2] Traag, V.A., Waltman, L. & van Eck, N.J. From Louvain to Leiden: guaranteeing well-connected commu-
nities. Sci Rep 9, 5233 (2019). https://doi.org/10.1038/s41598-019-41695-z
[3] Nicolas Dugué, Anthony Perez. Directed Louvain : maximizing modularity in directed networks. [Research
Report] Université d’Orléans. 2015. hal-01231784. https://hal.archives-ouvertes.fr/hal-01231784
[1] Blondel, V.D. et al. Fast unfolding of communities in large networks. J. Stat. Mech 10008, 1-12(2008)
[1] Parés F., Garcia-Gasulla D. et al. “Fluid Communities: A Competitive and Highly Scalable Community
Detection Algorithm”. [https://arxiv.org/pdf/1703.09307.pdf].
[1] M. E. J. Newman “Networks: An Introduction”, page 224. Oxford University Press, 2011.
[2] Clauset, Aaron, Mark EJ Newman, and Cristopher Moore. “Finding community structure in very large
networks.” Phys. Rev. E 70.6 (2004). <https://arxiv.org/abs/cond-mat/0408187>
[3] Reichardt and Bornholdt “Statistical Mechanics of Community Detection” Phys. Rev. E 74, 016110, 2006.
https://doi.org/10.1103/PhysRevE.74.016110
1068 Bibliography
[4] M. E. J. Newman, “Equivalence between modularity optimization and maximum likelihood methods for
community detection” Phys. Rev. E 94, 052315, 2016. https://doi.org/10.1103/PhysRevE.94.052315
[1] Santo Fortunato. “Community Detection in Graphs”. Physical Reports, Volume 486, Issue 3–5 pp. 75–174
<https://arxiv.org/abs/0906.0612>
[1] Depth-first search and linear graph algorithms, R. Tarjan SIAM Journal of Computing 1(2):146-160,
(1972).
[2] On finding the strongly connected components in a directed graph. E. Nuutila and E. Soisalon-Soinen In-
formation Processing Letters 49(1): 9-14, (1994)..
[1] Depth-first search and linear graph algorithms, R. Tarjan SIAM Journal of Computing 1(2):146-160,
(1972).
[2] On finding the strongly connected components in a directed graph. E. Nuutila and E. Soisalon-Soinen In-
formation Processing Letters 49(1): 9-14, (1994)..
[1] Hopcroft, J.; Tarjan, R. (1973). “Efficient algorithms for graph manipulation”. Communications of the ACM
16: 372–378. doi:10.1145/362248.362272
16: 372–378. doi:10.1145/362248.362272
16: 372–378. doi:10.1145/362248.362272
16: 372–378. doi:10.1145/362248.362272
[1] https://en.wikipedia.org/wiki/Bridge_%28graph_theory%29
[2] Wang, Tianhao, et al. (2015) A simple algorithm for finding all k-edge-connected components. http:
//journals.plos.org/plosone/article?id=10.1371/journal.pone.0136264
[1] Zhou, Liu, et al. (2012) Finding maximal k-edge-connected subgraphs from a large graph. ACM Interna-
tional Conference on Extending Database Technology 2012 480-–491. https://openproceedings.org/2012/
conf/edbt/ZhouLYLCL12.pdf
[1] Wang, Tianhao, et al. (2015) A simple algorithm for finding all k-edge-connected components. http:
//journals.plos.org/plosone/article?id=10.1371/journal.pone.0136264
[1] Moody, J. and D. White (2003). Social cohesion and embeddedness: A hierarchical concep-
tion of social groups. American Sociological Review 68(1), 103–28. http://www2.asanet.org/journals/
ASRFeb03MoodyWhite.pdf
[2] Kanevsky, A. (1993). Finding all minimum-size separating vertex sets in a graph. Networks 23(6), 533–541.
http://onlinelibrary.wiley.com/doi/10.1002/net.3230230604/abstract
[3] Torrents, J. and F. Ferraro (2015). Structural Cohesion: Visualization and Heuristics for Fast Computation.
https://arxiv.org/pdf/1503.04476v1
[1] Kanevsky, A. (1993). Finding all minimum-size separating vertex sets in a graph. Networks 23(6), 533–541.
http://onlinelibrary.wiley.com/doi/10.1002/net.3230230604/abstract
[1] Beineke, L., O. Oellermann, and R. Pippert (2002). The average connectivity of a graph. Discrete mathe-
matics 252(1-3), 31-45. http://www.sciencedirect.com/science/article/pii/S0012365X01001807
[1] Abdol-Hossein Esfahanian. Connectivity Algorithms. http://www.cse.msu.edu/~cse835/Papers/Graph_
connectivity_revised.pdf
Bibliography 1069
[1] Kammer, Frank and Hanjo Taubig. Graph Connectivity. in Brandes and Erlebach, ‘Network Analysis:
Methodological Foundations’, Lecture Notes in Computer Science, Volume 3418, Springer-Verlag, 2005.
http://www.informatik.uni-augsburg.de/thi/personen/kammer/Graph_Connectivity.pdf
[1] Abdol-Hossein Esfahanian. Connectivity Algorithms. (this is a chapter, look for the reference of the book).
http://www.cse.msu.edu/~cse835/Papers/Graph_connectivity_revised.pdf
[1] Kammer, Frank and Hanjo Taubig. Graph Connectivity. in Brandes and Erlebach, ‘Network Analysis:
Methodological Foundations’, Lecture Notes in Computer Science, Volume 3418, Springer-Verlag, 2005.
https://doi.org/10.1007/978-3-540-31955-9_7
[1] An O(m) Algorithm for Cores Decomposition of Networks Vladimir Batagelj and Matjaz Zaversnik, 2003.
https://arxiv.org/abs/cs.DS/0310049
[1] An O(m) Algorithm for Cores Decomposition of Networks Vladimir Batagelj and Matjaz Zaversnik, 2003.
https://arxiv.org/abs/cs.DS/0310049
[1] A model of Internet topology using k-shell decomposition Shai Carmi, Shlomo Havlin, Scott Kirkpatrick,
Yuval Shavitt, and Eran Shir, PNAS July 3, 2007 vol. 104 no. 27 11150-11154 http://www.pnas.org/
content/104/27/11150.full
[1] A model of Internet topology using k-shell decomposition Shai Carmi, Shlomo Havlin, Scott Kirkpatrick,
Yuval Shavitt, and Eran Shir, PNAS July 3, 2007 vol. 104 no. 27 11150-11154 http://www.pnas.org/
content/104/27/11150.full
[1] k -core (bootstrap) percolation on complex networks: Critical phenomena and nonlocal effects, A. V. Golt-
sev, S. N. Dorogovtsev, and J. F. F. Mendes, Phys. Rev. E 73, 056101 (2006) http://link.aps.org/doi/10.
1103/PhysRevE.73.056101
[1] Bounds and Algorithms for k-truss. Paul Burkhardt, Vance Faber, David G. Harris, 2018. https://arxiv.org/
abs/1806.05523v2
[2] Trusses: Cohesive Subgraphs for Social Network Analysis. Jonathan Cohen, 2005.
[1] Multi-scale structure and topological anomaly detection via a new network statistic: The onion decom-
position L. Hébert-Dufresne, J. A. Grochow, and A. Allard Scientific Reports 6, 31708 (2016) http:
//doi.org/10.1038/srep31708
[2] Percolation and the effective structure of complex networks A. Allard and L. Hébert-Dufresne Physical
Review X 9, 011023 (2019) http://doi.org/10.1103/PhysRevX.9.011023
[1] Paton, K. An algorithm for finding a fundamental set of cycles of a graph. Comm. ACM 12, 9 (Sept 1969),
514-518.
[1] Finding all the elementary circuits of a directed graph. D. B. Johnson, SIAM Journal on Computing 4, no.
1, 77-84, 1975. https://doi.org/10.1137/0204007
[2] Enumerating the cycles of a digraph: a new preprocessing strategy. G. Loizou and P. Thanish, Information
Sciences, v. 27, 163-182, 1982.
1070 Bibliography
[3] A search strategy for the elementary cycles of a directed graph. J.L. Szwarcfiter and P.E. Lauer, BIT NU-
MERICAL MATHEMATICS, v. 16, no. 2, 192-204, 1976.
[1] Finding all the elementary circuits of a directed graph. D. B. Johnson, SIAM Journal on Computing 4, no.
1, 77-84, 1975. https://doi.org/10.1137/0204007
[1] Vadhan, Salil P. “Pseudorandomness.” Foundations and Trends in Theoretical Computer Science 7.1–3
(2011): 1–336. <https://doi.org/10.1561/0400000010>
[1] David Gleich. Hierarchical Directed Spectral Graph Partitioning. <https://www.cs.purdue.edu/homes/
dgleich/publications/Gleich%202005%20-%20hierarchical%20directed%20spectral.pdf>
[1] Fan Chung. Spectral Graph Theory. (CBMS Regional Conference Series in Mathematics, No. 92), Ameri-
can Mathematical Society, 1997, ISBN 0-8218-0315-8 <http://www.math.ucsd.edu/~fan/research/revised.
html>
(2011): 1–336. <https://doi.org/10.1561/0400000010>
(2011): 1–336. <https://doi.org/10.1561/0400000010>
[1] Pearl, J. (2009). Causality. Cambridge: Cambridge University Press.
[2] Darwiche, A. (2009). Modeling and reasoning with Bayesian networks. Cambridge: Cambridge University
Press.
[3] Shachter, R. D. (1998). Bayes-ball: rational pastime (for determining irrelevance and requisite information
in belief networks and influence diagrams). In , Proceedings of the Fourteenth Conference on Uncertainty
in Artificial Intelligence (pp. 480–487). San Francisco, CA, USA: Morgan Kaufmann Publishers Inc.
[4] Koller, D., & Friedman, N. (2009). Probabilistic graphical models: principles and techniques. The MIT
Press.
[5] https://en.wikipedia.org/wiki/Causal_Markov_condition
[1] Manber, U. (1989). Introduction to Algorithms - A Creative Approach. Addison-Wesley.
[1] Knuth, Donald E., Szwarcfiter, Jayme L. (1974). “A Structured Program to Generate All Topological Sorting
Arrangements” Information Processing Letters, Volume 2, Issue 6, 1974, Pages 153-157, ISSN 0020-0190,
https://doi.org/10.1016/0020-0190(74)90001-5. Elsevier (North-Holland), Amsterdam
[1] Manber, U. (1989). Introduction to Algorithms - A Creative Approach. Addison-Wesley.
[1] Jarvis, J. P.; Shier, D. R. (1996), “Graph-theoretic analysis of finite Markov chains,” in Shier, D. R.; Wal-
lenius, K. T., Applied Mathematical Modeling: A Multidisciplinary Approach, CRC Press.
[1] https://www.ics.uci.edu/~eppstein/PADS/PartialOrder.py
[1] Free Lattices, by R. Freese, J. Jezek and J. B. Nation, AMS, Vol 42, 1995, p. 226.
[1] Wikipedia “Resistance distance.” https://en.wikipedia.org/wiki/Resistance_distance
[2] E. W. Weisstein “Resistance Distance.” MathWorld–A Wolfram Web Resource https://mathworld.wolfram.
com/ResistanceDistance.html
[3] V. S. S. Vos, “Methods for determining the effective resistance.” Mestrado, Mathematisch Instituut Univer-
siteit Leiden, 2016 https://www.universiteitleiden.nl/binaries/content/assets/science/mi/scripties/master/
vos_vaya_master.pdf
Bibliography 1071
[1] Brouwer, A. E.; Cohen, A. M.; and Neumaier, A. Distance-Regular Graphs. New York: Springer-Verlag,
1989.
[2] Weisstein, Eric W. “Distance-Regular Graph.” http://mathworld.wolfram.com/Distance-RegularGraph.
html
[1] Weisstein, Eric W. “Intersection Array.” From MathWorld–A Wolfram Web Resource. http://mathworld.
wolfram.com/IntersectionArray.html
[1] Weisstein, Eric W. “Global Parameters.” From MathWorld–A Wolfram Web Resource. http://mathworld.
wolfram.com/GlobalParameters.html
[1] K. D. Cooper, T. J. Harvey, and K. Kennedy. A simple, fast dominance algorithm. Software Practice &
Experience, 4:110, 2001.
[1] K. D. Cooper, T. J. Harvey, and K. Kennedy. A simple, fast dominance algorithm. Software Practice &
Experience, 4:110, 2001.
[1] https://en.wikipedia.org/wiki/Dominating_set
[1] https://en.wikipedia.org/wiki/Dominating_set
[1] Latora, Vito, and Massimo Marchiori. “Efficient behavior of small-world networks.” Physical Review Letters
87.19 (2001): 198701. <https://doi.org/10.1103/PhysRevLett.87.198701>
[1] J. Edmonds, E. L. Johnson. Matching, Euler tours and the Chinese postman. Mathematical programming,
Volume 5, Issue 1 (1973), 111-114.
[2] https://en.wikipedia.org/wiki/Eulerian_path
[1] J. Edmonds, E. L. Johnson. Matching, Euler tours and the Chinese postman. Mathematical programming,
Volume 5, Issue 1 (1973), 111-114.
[2] https://en.wikipedia.org/wiki/Eulerian_path
[3] http://web.math.princeton.edu/math_alive/5/Notes1.pdf
[1] Dinitz’ Algorithm: The Original Version and Even’s Version. 2006. Yefim Dinitz. In Theoretical Com-
puter Science. Lecture Notes in Computer Science. Volume 3895. pp 218-240. https://doi.org/10.1007/
11685654_10
[1] Boykov, Y., & Kolmogorov, V. (2004). An experimental comparison of min-cut/max-flow algorithms for
energy minimization in vision. Pattern Analysis and Machine Intelligence, IEEE Transactions on, 26(9),
1124-1137. https://doi.org/10.1109/TPAMI.2004.60
[2] Vladimir Kolmogorov. Graph-based Algorithms for Multi-camera Reconstruction Problem. PhD thesis,
Cornell University, CS Department, 2003. pp. 109-114. https://web.archive.org/web/20170809091249/
https://pub.ist.ac.at/~vnk/papers/thesis.pdf
[1] Gusfield D: Very simple methods for all pairs network flow analysis. SIAM J Comput 19(1):143-155, 1990.
[1] Z. Kiraly, P. Kovacs. Efficient implementation of minimum-cost flow algorithms. Acta Universitatis Sapi-
entiae, Informatica 4(1):67–118. 2012.
[2] R. Barr, F. Glover, D. Klingman. Enhancement of spanning tree labeling procedures for network optimiza-
tion. INFOR 17(1):16–34. 1979.
1072 Bibliography
[1] Shervashidze, Nino, Pascal Schweitzer, Erik Jan Van Leeuwen, Kurt Mehlhorn, and Karsten M. Borgwardt.
Weisfeiler Lehman Graph Kernels. Journal of Machine Learning Research. 2011. http://www.jmlr.org/
papers/volume12/shervashidze11a/shervashidze11a.pdf
[1] Shervashidze, Nino, Pascal Schweitzer, Erik Jan Van Leeuwen, Kurt Mehlhorn, and Karsten M. Borgwardt.
Weisfeiler Lehman Graph Kernels. Journal of Machine Learning Research. 2011. http://www.jmlr.org/
papers/volume12/shervashidze11a/shervashidze11a.pdf
[2] Annamalai Narayanan, Mahinthan Chandramohan, Rajasekar Venkatesan, Lihui Chen, Yang Liu and Shan-
tanu Jaiswa. graph2vec: Learning Distributed Representations of Graphs. arXiv. 2017 https://arxiv.org/pdf/
1707.05005.pdf
[EG1960] Erdős and Gallai, Mat. Lapok 11 264, 1960.
[choudum1986] S.A. Choudum. “A simple proof of the Erdős-Gallai theorem on graph sequences.” Bulletin of the
Australian Mathematical Society, 33, pp 67-70, 1986. https://doi.org/10.1017/S0004972700002872
[havel1955] Havel, V. “A Remark on the Existence of Finite Graphs” Casopis Pest. Mat. 80, 477-480, 1955.
[hakimi1962] Hakimi, S. “On the Realizability of a Set of Integers as Degrees of the Vertices of a Graph.” SIAM J.
Appl. Math. 10, 496-506, 1962.
[CL1996] G. Chartrand and L. Lesniak, “Graphs and Digraphs”, Chapman and Hall/CRC, 1996.
[1] D.J. Kleitman and D.L. Wang Algorithms for Constructing Graphs and Digraphs with Given Valences and
Factors, Discrete Mathematics, 6(1), pp. 79-88 (1973)
[1] S. L. Hakimi. “On the realizability of a set of integers as degrees of the vertices of a linear graph”, J. SIAM,
10, pp. 496-506 (1962).
[1] F. Boesch and F. Harary. “Line removal algorithms for graphs and their degree lists”, IEEE Trans. Circuits
and Systems, CAS-23(12), pp. 778-782 (1976).
[1] I.E. Zverovich and V.E. Zverovich. “Contributions to the theory of graphic sequences”, Discrete Mathe-
matics, 105, pp. 292-303 (1992).
[havel1955] Havel, V. “A Remark on the Existence of Finite Graphs” Casopis Pest. Mat. 80, 477-480, 1955.
[hakimi1962] Hakimi, S. “On the Realizability of a Set of Integers as Degrees of the Vertices of a Graph.” SIAM J.
Appl. Math. 10, 496-506, 1962.
[CL1996] G. Chartrand and L. Lesniak, “Graphs and Digraphs”, Chapman and Hall/CRC, 1996.
[1] A. Tripathi and S. Vijay. “A note on a theorem of Erdős & Gallai”, Discrete Mathematics, 265, pp. 417-420
(2003).
[2] I.E. Zverovich and V.E. Zverovich. “Contributions to the theory of graphic sequences”, Discrete Mathe-
matics, 105, pp. 292-303 (1992).
[EG1960] Erdős and Gallai, Mat. Lapok 11 264, 1960.
[1] Luo, J.; Magee, C.L. (2011), Detecting evolving patterns of self-organizing networks by flow hierarchy mea-
surement, Complexity, Volume 16 Issue 6 53-61. DOI: 10.1002/cplx.20368 http://web.mit.edu/~cmagee/
www/documents/28-DetectingEvolvingPatterns_FlowHierarchy.pdf
[1] Chung, Fan and Linyuan Lu. “The Small World Phenomenon in Hybrid Power Law Graphs.” Complex
Networks. Springer Berlin Heidelberg, 2004. 89–104.
[1] Chung, Fan and Linyuan Lu. “The Small World Phenomenon in Hybrid Power Law Graphs.” Complex
Networks. Springer Berlin Heidelberg, 2004. 89–104.
[1] L. P. Cordella, P. Foggia, C. Sansone, M. Vento, “An Improved Algorithm for Matching Large Graphs”,
3rd IAPR-TC15 Workshop on Graph-based Representations in Pattern Recognition, Cuen, pp. 149-159,
2001. https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.101.5342
Bibliography 1073
[1] M. Houbraken, S. Demeyer, T. Michoel, P. Audenaert, D. Colle, M. Pickavet, “The Index-Based Subgraph
Matching Algorithm with General Symmetries (ISMAGS): Exploiting Symmetry for Faster Subgraph Enu-
meration”, PLoS One 9(5): e97896, 2014. https://doi.org/10.1371/journal.pone.0097896
[2] https://en.wikipedia.org/wiki/Maximum_common_induced_subgraph
[1] M. Houbraken, S. Demeyer, T. Michoel, P. Audenaert, D. Colle, M. Pickavet, “The Index-Based Subgraph
Matching Algorithm with General Symmetries (ISMAGS): Exploiting Symmetry for Faster Subgraph Enu-
meration”, PLoS One 9(5): e97896, 2014. https://doi.org/10.1371/journal.pone.0097896
[1] A. Langville and C. Meyer, “A survey of eigenvector methods of web information retrieval.” http://citeseer.
ist.psu.edu/713792.html
[2] Page, Lawrence; Brin, Sergey; Motwani, Rajeev and Winograd, Terry, The PageRank citation ranking:
Bringing order to the Web. 1999 http://dbpubs.stanford.edu:8090/pub/showDoc.Fulltext?lang=en&doc=
1999-66&format=pdf
[1] A. Langville and C. Meyer, “A survey of eigenvector methods of web information retrieval.” http://citeseer.
ist.psu.edu/713792.html
[2] Jon Kleinberg, Authoritative sources in a hyperlinked environment Journal of the ACM 46 (5): 604-32,
1999. doi:10.1145/324133.324140. http://www.cs.cornell.edu/home/kleinber/auth.pdf.
[1] T. Zhou, L. Lu, Y.-C. Zhang. Predicting missing links via local information. Eur. Phys. J. B 71 (2009) 623.
https://arxiv.org/pdf/0901.0553.pdf
[1] D. Liben-Nowell, J. Kleinberg. The Link Prediction Problem for Social Networks (2004). http://www.cs.
cornell.edu/home/kleinber/link-pred.pdf
[1] Sucheta Soundarajan and John Hopcroft. Using community information to improve the precision of link
prediction methods. In Proceedings of the 21st international conference companion on World Wide Web
(WWW ‘12 Companion). ACM, New York, NY, USA, 607-608. http://doi.acm.org/10.1145/2187980.
2188150
[1] Sucheta Soundarajan and John Hopcroft. Using community information to improve the precision of link
prediction methods. In Proceedings of the 21st international conference companion on World Wide Web
(WWW ‘12 Companion). ACM, New York, NY, USA, 607-608. http://doi.acm.org/10.1145/2187980.
2188150
[1] Jorge Carlos Valverde-Rebaza and Alneu de Andrade Lopes. Link prediction in complex networks based on
cluster information. In Proceedings of the 21st Brazilian conference on Advances in Artificial Intelligence
(SBIA’12) https://doi.org/10.1007/978-3-642-34459-6_10
[1] Ahmad, I., Akhtar, M.U., Noor, S. et al. Missing Link Prediction using Common Neighbor and Centrality
based Parameterized Algorithm. Sci Rep 10, 364 (2020). https://doi.org/10.1038/s41598-019-57304-y
[1] “Efficient Algorithms for Finding Maximum Matching in Graphs”, Zvi Galil, ACM Computing Surveys,
1986.
[1] https://en.wikipedia.org/wiki/Graph_minor
[1] Patrick Doreian, Vladimir Batagelj, and Anuska Ferligoj. Generalized Blockmodeling. Cambridge University
Press, 2004.
[1] Xiaowei Ying and Xintao Wu, On Randomness Measures for Social Networks, SIAM International Con-
ference on Data Mining. 2009
1074 Bibliography
[1] Wray L. Buntine. 1995. Chain graphs for learning. In Proceedings of the Eleventh conference on Uncertainty
in artificial intelligence (UAI’95)
[1] J. A. Bondy, U. S. R. Murty, Graph Theory. Springer, 2008.
[1] Ulrik Brandes: The Left-Right Planarity Test 2009 http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.
1.1.217.9208
[2] Takao Nishizeki, Md Saidur Rahman: Planar graph drawing Lecture Notes Series on Computing: Volume
12 2004
[1] M. Chrobak and T.H. Payne: A Linear-time Algorithm for Drawing a Planar Graph on a Grid 1989 http:
//citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.51.6677
[1] Y. Shi, M. Dehmer, X. Li, I. Gutman, “Graph Polynomials”
[1] M. Brandt, “The Tutte Polynomial.” Talking About Combinatorial Objects Seminar, 2015 https://math.
berkeley.edu/~brandtm/talks/tutte.pdf
[2] A. Björklund, T. Husfeldt, P. Kaski, M. Koivisto, “Computing the Tutte polynomial in vertex-exponential
time” 49th Annual IEEE Symposium on Foundations of Computer Science, 2008 https://ieeexplore.ieee.
org/abstract/document/4691000
[3] Y. Shi, M. Dehmer, X. Li, I. Gutman, “Graph Polynomials,” p. 14
[4] Y. Shi, M. Dehmer, X. Li, I. Gutman, “Graph Polynomials,” p. 46
[5] A. Nešetril, J. Goodall, “Graph invariants, homomorphisms, and the Tutte polynomial” https://iuuk.mff.
cuni.cz/~andrew/Tutte.pdf
[6] D. B. West, “Introduction to Graph Theory,” p. 84
[7] G. Coutinho, “A brief introduction to the Tutte polynomial” Structural Analysis of Complex Networks,
2011 https://homepages.dcc.ufmg.br/~gabriel/seminars/coutinho_tuttepolynomial_seminar.pdf
[8] J. A. Ellis-Monaghan, C. Merino, “Graph polynomials and their applications I: The Tutte polynomial” Struc-
tural Analysis of Complex Networks, 2011 https://arxiv.org/pdf/0803.3079.pdf
[2] E. W. Weisstein “Chromatic Polynomial” MathWorld–A Wolfram Web Resource https://mathworld.
wolfram.com/ChromaticPolynomial.html
[4] J. Zhang, J. Goodall, “An Introduction to Chromatic Polynomials” https://math.mit.edu/~apost/courses/18.
204_2018/Julie_Zhang_paper.pdf
[5] R. C. Read, “An Introduction to Chromatic Polynomials” Journal of Combinatorial Theory, 1968 https:
//math.berkeley.edu/~mrklug/ReadChromatic.pdf
[6] W. T. Tutte, “Graph-polynomials” Advances in Applied Mathematics, 2004 https://www.sciencedirect.
com/science/article/pii/S0196885803000411
[7] R. P. Stanley, “Acyclic orientations of graphs” Discrete Mathematics, 2006 https://math.mit.edu/~rstan/
pubs/pubfiles/18.pdf
[1] “An algorithm for computing simple k-factors.”, Meijer, Henk, Yurai Núñez-Rodríguez, and David Rap-
paport, Information processing letters, 2009.
[1] Julian J. McAuley, Luciano da Fontoura Costa, and Tibério S. Caetano, “The rich-club phenomenon across
complex network hierarchies”, Applied Physics Letters Vol 91 Issue 8, August 2007. https://arxiv.org/abs/
physics/0701290
Bibliography 1075
[2] R. Milo, N. Kashtan, S. Itzkovitz, M. E. J. Newman, U. Alon, “Uniform generation of random graphs with
arbitrary degree sequences”, 2006. https://arxiv.org/abs/cond-mat/0312028
[1] Zeina Abu-Aisheh, Romain Raveaux, Jean-Yves Ramel, Patrick Martineau. An Exact Graph Edit Distance
Algorithm for Solving Pattern Recognition Problems. 4th International Conference on Pattern Recognition
Applications and Methods 2015, Jan 2015, Lisbon, Portugal. 2015, <10.5220/0005209202710278>. <hal-
01168816> https://hal.archives-ouvertes.fr/hal-01168816
[1] https://en.wikipedia.org/wiki/SimRank
[2] G. Jeh and J. Widom. “SimRank: a measure of structural-context similarity”, In KDD’02: Proceedings
of the Eighth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp.
538–543. ACM Press, 2002.
[1] Zhang, J., Tang, J., Ma, C., Tong, H., Jing, Y., & Li, J. Panther: Fast top-k similarity search on large
networks. In Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and
Data Mining (Vol. 2015-August, pp. 1445–1454). Association for Computing Machinery. https://doi.org/
10.1145/2783258.2783267.
[1] Zhang, J., Tang, J., Ma, C., Tong, H., Jing, Y., & Li, J. Panther: Fast top-k similarity search on large
networks. In Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and
Data Mining (Vol. 2015-August, pp. 1445–1454). Association for Computing Machinery. https://doi.org/
10.1145/2783258.2783267.
[1] R. Sedgewick, “Algorithms in C, Part 5: Graph Algorithms”, Addison Wesley Professional, 3rd ed., 2001.
[1] R. Sedgewick, “Algorithms in C, Part 5: Graph Algorithms”, Addison Wesley Professional, 3rd ed., 2001.
[1] Jin Y. Yen, “Finding the K Shortest Loopless Paths in a Network”, Management Science, Vol. 17, No. 11,
Theory Series (Jul., 1971), pp. 712-716.
[1] Small-world network:: https://en.wikipedia.org/wiki/Small-world_network
[1] Maslov, Sergei, and Kim Sneppen. “Specificity and stability in topology of protein networks.” Science
296.5569 (2002): 910-913.
[1] Sporns, Olaf, and Jonathan D. Zwi. “The small world of the cerebral cortex.” Neuroinformatics 2.2 (2004):
145-162.
[2] Maslov, Sergei, and Kim Sneppen. “Specificity and stability in topology of protein networks.” Science
296.5569 (2002): 910-913.
[1] The brainstem reticular formation is a small-world, not scale-free, network M. D. Humphries, K. Gurney
and T. J. Prescott, Proc. Roy. Soc. B 2006 273, 503-511, doi:10.1098/rspb.2005.3354.
[2] Humphries and Gurney (2008). “Network ‘Small-World-Ness’: A Quantitative Method for Determining
Canonical Network Equivalence”. PLoS One. 3 (4). PMID 18446219. doi:10.1371/journal.pone.0002051.
1076 Bibliography
[1] Telesford, Joyce, Hayasaka, Burdette, and Laurienti (2011). “The Ubiquity of Small-World Networks”.
Brain Connectivity. 1 (0038): 367-75. PMC 3604768. PMID 22432451. doi:10.1089/brain.2011.0038.
[1] Lun Li, David Alderson, John C. Doyle, and Walter Willinger, Towards a Theory of Scale-Free Graphs:
Definition, Properties, and Implications (Extended Version), 2005. https://arxiv.org/abs/cond-mat/0501169
[1] Burt, Ronald S. “Structural holes and good ideas”. American Journal of Sociology (110): 349–399.
[1] Burt, Ronald S. Structural Holes: The Social Structure of Competition. Cambridge: Harvard University Press,
1995.
[2] Borgatti, S. “Structural Holes: Unpacking Burt’s Redundancy Measures” CONNECTIONS 20(1):35-38.
http://www.analytictech.com/connections/v20(1)/holes.htm
[1] Burt, Ronald S. “Structural holes and good ideas”. American Journal of Sociology (110): 349–399.
[1] Maccioni, A., & Abadi, D. J. (2016, August). Scalable pattern matching over compressed graphs via deden-
sification. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery
and Data Mining (pp. 1755-1764). http://www.cs.umd.edu/~abadi/papers/graph-dedense.pdf
[1] Y. Tian, R. A. Hankins, and J. M. Patel. Efficient aggregation for graph summarization. In Proc. 2008
ACM-SIGMOD Int. Conf. Management of Data (SIGMOD’08), pages 567–580, Vancouver, Canada, June
2008.
[1] Erdős, Péter L., et al. “A Simple Havel-Hakimi Type Algorithm to Realize Graph-
ical Degree Sequences of Directed Graphs.” ArXiv:0905.4913 [Math], Jan. 2010.
https://doi.org/10.48550/arXiv.0905.4913. Published 2010 in Elec. J. Combinatorics (17(1)). R66.
http://www.combinatorics.org/Volume_17/PDF/v17i1r66.pdf
[2] “Combinatorics - Reaching All Possible Simple Directed Graphs with a given Degree Sequence with 2-
Edge Swaps.” Mathematics Stack Exchange, https://math.stackexchange.com/questions/22272/. Accessed
30 May 2022.
[1] C. Gkantsidis and M. Mihail and E. Zegura, The Markov chain simulation method for generating connected
power law random graphs, 2003. http://citeseer.ist.psu.edu/gkantsidis03markov.html
[1] Threshold graphs: https://en.wikipedia.org/wiki/Threshold_graph
[1] Threshold graphs: https://en.wikipedia.org/wiki/Threshold_graph
[1] Tantau, Till. “A note on the complexity of the reachability problem for tournaments.” Electronic Colloquium
on Computational Complexity. 2001. <http://eccc.hpi-web.de/report/2001/092/>
[1] Tantau, Till. “A note on the complexity of the reachability problem for tournaments.” Electronic Colloquium
on Computational Complexity. 2001. <http://eccc.hpi-web.de/report/2001/092/>
[1] http://www.ics.uci.edu/~eppstein/PADS
[2] https://en.wikipedia.org/wiki/Depth-limited_search
[1] http://www.ics.uci.edu/~eppstein/PADS/BFS.py.
[2] https://en.wikipedia.org/wiki/Depth-limited_search
[1] G.K. Janssens, K. Sörensen, An algorithm to generate all spanning trees in order of increas-
ing cost, Pesquisa Operacional, 2005-08, Vol. 25 (2), p. 219-229, https://www.scielo.br/j/pope/a/
XHswBwRwJyrfL88dmMwYNWp/?lang=en
[1] J. Edmonds, Optimum Branchings, Journal of Research of the National Bureau of Standards, 1967, Vol.
71B, p.233-240, https://archive.org/details/jresv71Bn4p233
[1] Wang, Xiaodong, Lei Wang, and Yingjie Wu. “An optimal algorithm for Prufer codes.” Journal of Software
Engineering and Applications 2.02 (2009): 111. <https://doi.org/10.4236/jsea.2009.22016>
Bibliography 1077
[1] Wang, Xiaodong, Lei Wang, and Yingjie Wu. “An optimal algorithm for Prufer codes.” Journal of Software
Engineering and Applications 2.02 (2009): 111. <https://doi.org/10.4236/jsea.2009.22016>
[1] V. Kulkarni, Generating random combinatorial objects, Journal of Algorithms, 11 (1990), pp. 185–207
[1] G.K. Janssens, K. Sörensen, An algorithm to generate all spanning trees in order of increas-
ing cost, Pesquisa Operacional, 2005-08, Vol. 25 (2), p. 219-229, https://www.scielo.br/j/pope/a/
XHswBwRwJyrfL88dmMwYNWp/?lang=en
[1] Junction tree algorithm: https://en.wikipedia.org/wiki/Junction_tree_algorithm
[2] Finn V. Jensen and Frank Jensen. 1994. Optimal junction trees. In Proceedings of the Tenth international
conference on Uncertainty in artificial intelligence (UAI’94). Morgan Kaufmann Publishers Inc., San Fran-
cisco, CA, USA, 360–366.
[1] Vladimir Batagelj and Andrej Mrvar, A subquadratic triad census algorithm for large sparse networks with
small maximum degree, University of Ljubljana, http://vlado.fmf.uni-lj.si/pub/networks/doc/triads/triads.
pdf
[1] Snijders, T. (2012). “Transitivity and triads.” University of Oxford. https://web.archive.org/web/
20170830032057/http://www.stats.ox.ac.uk/~snijders/Trans_Triads_ha.pdf
[1] Snijders, T. (2012). “Transitivity and triads.” University of Oxford. https://web.archive.org/web/
20170830032057/http://www.stats.ox.ac.uk/~snijders/Trans_Triads_ha.pdf
[1] Ulrik Brandes, Thomas Erlebach (eds.). Network Analysis: Methodological Foundations. Springer, 2005.
<http://books.google.com/books?id=TTNhSm7HYrIC>
[1] Erwig, Martin. (2000), “The graph Voronoi diagram with applications.” Networks, 36: 156–163.
<dx.doi.org/10.1002/1097-0037(200010)36:3<156::AID-NET2>3.0.CO;2-L>
[1] Ronald C. Read and Robin J. Wilson, An Atlas of Graphs. Oxford University Press, 1998.
[atlas] Ronald C. Read and Robin J. Wilson, An Atlas of Graphs. Oxford University Press, 1998.
[1] An introduction to data structures and algorithms, James Andrew Storer, Birkhauser Boston 2001, (page
225).
[1] Theorem 4.4.2 in A. Lubotzky. “Discrete groups, expanding graphs and invariant measures”, volume 125
of Progress in Mathematics. Birkhäuser Verlag, Basel, 1994.
[1] https://en.wikipedia.org/wiki/Bull_graph.
[1] https://en.wikipedia.org/wiki/Chv%C3%A1tal_graph
[2] https://mathworld.wolfram.com/ChvatalGraph.html
[1] https://en.wikipedia.org/wiki/Cube#Cubical_graph
[1] https://en.wikipedia.org/wiki/Desargues_graph
[2] https://mathworld.wolfram.com/DesarguesGraph.html
[1] https://mathworld.wolfram.com/DiamondGraph.html
[1] https://en.wikipedia.org/wiki/Regular_dodecahedron#Dodecahedral_graph
[2] https://mathworld.wolfram.com/DodecahedralGraph.html
[1] https://en.wikipedia.org/wiki/Frucht_graph
[2] https://mathworld.wolfram.com/FruchtGraph.html
[1] https://en.wikipedia.org/wiki/Heawood_graph
[2] https://mathworld.wolfram.com/HeawoodGraph.html
1078 Bibliography
[3] https://www.win.tue.nl/~aeb/graphs/Heawood.html
[1] https://blogs.ams.org/visualinsight/2016/02/01/hoffman-singleton-graph/
[2] https://mathworld.wolfram.com/Hoffman-SingletonGraph.html
[3] https://en.wikipedia.org/wiki/Hoffman%E2%80%93Singleton_graph
[1] https://mathworld.wolfram.com/HouseGraph.html
[1] https://mathworld.wolfram.com/HouseGraph.html
[1] https://mathworld.wolfram.com/IcosahedralGraph.html
[1] Krackhardt, David. “Assessing the Political Landscape: Structure, Cognition, and Power in Organizations”.
Administrative Science Quarterly. 35 (2): 342–369. doi:10.2307/2393394. JSTOR 2393394. June 1990.
[1] https://en.wikipedia.org/wiki/M%C3%B6bius%E2%80%93Kantor_graph
[1] https://mathworld.wolfram.com/OctahedralGraph.html
[2] https://en.wikipedia.org/wiki/Tur%C3%A1n_graph#Special_cases
[1] https://en.wikipedia.org/wiki/Pappus_graph
[1] https://en.wikipedia.org/wiki/Petersen_graph
[2] https://www.win.tue.nl/~aeb/drg/graphs/Petersen.html
[1] Figure 18.2, Chapter 18, Graph Algorithms (3rd Ed), Sedgewick
[1] https://en.wikipedia.org/wiki/Tetrahedron#Tetrahedral_graph
[1] https://en.wikipedia.org/wiki/Truncated_cube
[2] https://www.coolmath.com/reference/polyhedra-truncated-cube
[1] https://en.wikipedia.org/wiki/Truncated_tetrahedron
[1] https://en.wikipedia.org/wiki/Tutte_graph
[1] Vladimir Batagelj and Ulrik Brandes, “Efficient generation of large random networks”, Phys. Rev. E, 71,
036113, 2005.
[1] P. Erdős and A. Rényi, On Random Graphs, Publ. Math. 6, 290 (1959).
[2] E. N. Gilbert, Random Graphs, Ann. Math. Stat., 30, 1141 (1959).
[1] Donald E. Knuth, The Art of Computer Programming, Volume 2/Seminumerical algorithms, Third Edition,
Addison-Wesley, 1997.
[1] M. E. J. Newman and D. J. Watts, Renormalization group analysis of the small-world network model, Physics
Letters A, 263, 341, 1999. https://doi.org/10.1016/S0375-9601(99)00757-4
[1] Duncan J. Watts and Steven H. Strogatz, Collective dynamics of small-world networks, Nature, 393, pp.
440–442, 1998.
[1] Duncan J. Watts and Steven H. Strogatz, Collective dynamics of small-world networks, Nature, 393, pp.
440–442, 1998.
Bibliography 1079
[1] A. Steger and N. Wormald, Generating random regular graphs quickly, Probability and Computing 8 (1999),
377-396, 1999. http://citeseer.ist.psu.edu/steger99generating.html
[2] Jeong Han Kim and Van H. Vu, Generating random regular graphs, Proceedings of the thirty-fifth ACM
symposium on Theory of computing, San Diego, CA, USA, pp 213–222, 2003. http://portal.acm.org/
citation.cfm?id=780542.780576
[1] A. L. Barabási and R. Albert “Emergence of scaling in random networks”, Science 286, pp 509-512, 1999.
[1] N. Moshiri “The dual-Barabasi-Albert model”, arXiv:1810.10538.
[1] Albert, R., & Barabási, A. L. (2000) Topology of evolving networks: local events and universality Physical
review letters, 85(24), 5234.
[1] P. Holme and B. J. Kim, “Growing scale-free networks with tunable clustering”, Phys. Rev. E, 65, 026107,
2002.
[1] Bollobás, Béla, Janson, S. and Riordan, O. “The phase transition in inhomogeneous random graphs”, Ran-
dom Structures Algorithms, 31, 3–122, 2007.
[2] Hagberg A, Lemons N (2015), “Fast Generation of Sparse Random Kernel Graphs”. PLoS ONE 10(9):
e0135177, 2015. doi:10.1371/journal.pone.0135177
[1] I. Ispolatov, P. L. Krapivsky, A. Yuryev, “Duplication-divergence model of protein interaction network”,
Phys. Rev. E, 71, 061911, 2005.
[1] Knudsen Michael, and Carsten Wiuf. “A Markov chain approach to randomly grown graphs.” Journal of
Applied Mathematics 2008. <https://doi.org/10.1155/2008/190836>
[1] M.E.J. Newman, “The structure and function of complex networks”, SIAM REVIEW 45-2, pp 167-256,
2003.
[1] Newman, M. E. J. and Strogatz, S. H. and Watts, D. J. Random graphs with arbitrary degree distributions
and their applications Phys. Rev. E, 64, 026118 (2001)
[1] Fan Chung and L. Lu, Connected components in random graphs with given expected degree sequences,
Ann. Combinatorics, 6, pp. 125-145, 2002.
[2] Joel Miller and Aric Hagberg, Efficient generation of networks with given expected degrees, in Algorithms
and Models for the Web-Graph (WAW 2011), Alan Frieze, Paul Horn, and Paweł Prałat (Eds), LNCS 6732,
pp. 115-126, 2011.
[1] Hakimi S., On Realizability of a Set of Integers as Degrees of the Vertices of a Linear Graph. I, Journal of
SIAM, 10(3), pp. 496-506 (1962)
[2] Kleitman D.J. and Wang D.L. Algorithms for Constructing Graphs and Digraphs with Given Valences and
Factors Discrete Mathematics, 6(1), pp. 79-88 (1973)
[1] D.J. Kleitman and D.L. Wang Algorithms for Constructing Graphs and Digraphs with Given Valences and
Factors Discrete Mathematics, 6(1), pp. 79-88 (1973)
[1] Moshen Bayati, Jeong Han Kim, and Amin Saberi, A sequential algorithm for generating random graphs.
Algorithmica, Volume 58, Number 4, 860-910, DOI: 10.1007/s00453-009-9340-1
[1] Joel C. Miller. “Percolation and epidemics in random clustered networks”. In: Physical review. E, Statistical,
nonlinear, and soft matter physics 80 (2 Part 1 August 2009).
[2] M. E. J. Newman. “Random Graphs with Clustering”. In: Physical Review Letters 103 (5 July 2009)
[1] P. L. Krapivsky and S. Redner, Organization of Growing Random Networks, Phys. Rev. E, 63, 066123,
2001.
[1] P. L. Krapivsky and S. Redner, Organization of Growing Random Networks, Phys. Rev. E, 63, 066123,
2001.
1080 Bibliography
[1] P. L. Krapivsky and S. Redner, Network Growth by Copying, Phys. Rev. E, 71, 036118, 2005k.},
[1] B. Bollobás, C. Borgs, J. Chayes, and O. Riordan, Directed scale-free graphs, Proceedings of the fourteenth
annual ACM-SIAM Symposium on Discrete Algorithms, 132–139, 2003.
[1] Masuda, N., Miwa, H., Konno, N.: Geographical threshold graphs with small-world and scale-free proper-
ties. Physical Review E 71, 036108 (2005)
[2] Milan Bradonjić, Aric Hagberg and Allon G. Percus, Giant component and connectivity in geographical
threshold graphs, in Algorithms and Models for the Web-Graph (WAW 2007), Antony Bonato and Fan
Chung (Eds), pp. 209–216, 2007
[1] J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on
Theory of Computing, 2000.
[1] Penrose, Mathew, Random Geometric Graphs, Oxford Studies in Probability, 5, 2003.
[1] Penrose, Mathew D. “Connectivity of soft random geometric graphs.” The Annals of Applied Probability
26.2 (2016): 986-1028.
[2] scipy.stats - https://docs.scipy.org/doc/scipy/reference/tutorial/stats.html
[1] http://cole-maclean.github.io/blog/files/thesis.pdf
[1] B. M. Waxman, Routing of multipoint connections. IEEE J. Select. Areas Commun. 6(9),(1988) 1617–1622.
[1] A. Elmokashfi, A. Kvalbein and C. Dovrolis, “On the Scalability of BGP: The Role of Topology Growth,”
in IEEE Journal on Selected Areas in Communications, vol. 28, no. 8, pp. 1250-1261, October 2010.
[1] K.B. Singer-Cohen, Random Intersection Graphs, 1995, PhD thesis, Johns Hopkins University
[2] Fill, J. A., Scheinerman, E. R., and Singer-Cohen, K. B., Random intersection graphs when m = !(n): An
equivalence theorem relating the evolution of the g(n, m, p) and g(n, p) models. Random Struct. Algorithms
16, 2 (2000), 156–176.
[1] Godehardt, E., and Jaworski, J. Two models of random intersection graphs and their applications. Electronic
Notes in Discrete Mathematics 10 (2001), 129–132.
[1] Nikoletseas, S. E., Raptopoulos, C., and Spirakis, P. G. The existence and efficient construction of large
independent sets in general random intersection graphs. In ICALP (2004), J. D´ıaz, J. Karhumäki, A.
Lepistö, and D. Sannella, Eds., vol. 3142 of Lecture Notes in Computer Science, Springer, pp. 1029–
1040.
[1] Zachary, Wayne W. “An Information Flow Model for Conflict and Fission in Small Groups.” Journal of
Anthropological Research, 33, 452–473, (1977).
[1] A. Davis, Gardner, B. B., Gardner, M. R., 1941. Deep South. University of Chicago Press, Chicago, IL.
[1] Ronald L. Breiger and Philippa E. Pattison Cumulated social roles: The duality of persons and their alge-
bras,1 Social Networks, Volume 8, Issue 3, September 1986, Pages 215-256
[1] D. E. Knuth, 1993. The Stanford GraphBase: a platform for combinatorial computing, pp. 74-87. New
York: AcM Press.
[1] Watts, D. J. ‘Networks, Dynamics, and the Small-World Phenomenon.’ Amer. J. Soc. 105, 493-527, 1999.
[1] Watts, D. J. ‘Networks, Dynamics, and the Small-World Phenomenon.’ Amer. J. Soc. 105, 493-527, 1999.
[1] Ulrik Brandes, Marco Gaertler, Dorothea Wagner, Experiments on Graph Clustering Algorithms, In the
proceedings of the 11th Europ. Symp. Algorithms, 2003.
[1] “Benchmark graphs for testing community detection algorithms”, Andrea Lancichinetti, Santo Fortunato,
and Filippo Radicchi, Phys. Rev. E 78, 046110 2008
[2] https://www.santofortunato.net/resources
Bibliography 1081
[1] A. Condon, R.M. Karp, Algorithms for graph partitioning on the planted partition model, Random Struct.
Algor. 18 (2001) 116-140.
[2] Santo Fortunato ‘Community Detection in Graphs’ Physical Reports Volume 486, Issue 3-5 p. 75-174.
https://arxiv.org/abs/0906.0612
[1] Santo Fortunato ‘Community Detection in Graphs’ Physical Reports Volume 486, Issue 3-5 p. 75-174.
https://arxiv.org/abs/0906.0612
[1] Santo Fortunato, Community Detection in Graphs, Physics Reports Volume 486, Issues 3-5, February 2010,
Pages 75-174. https://arxiv.org/abs/0906.0612
[1] Holland, P. W., Laskey, K. B., & Leinhardt, S., “Stochastic blockmodels: First steps”, Social networks,
5(2), 109-137, 1983.
[1] M. Gjoka, M. Kurant, A. Markopoulou, “2.5K Graphs: from Sampling to Generation”, IEEE Infocom,
2013.
[2] I. Stanton, A. Pinar, “Constructing and sampling graphs with a prescribed joint degree distribution”, Journal
of Experimental Algorithmics, 2012.
[H] Harary, F. “The Maximum Connectivity of a Graph.” Proc. Nat. Acad. Sci. USA 48, 1142-1146, 1962.
[1] F. T. Boesch, A. Satyanarayana, and C. L. Suffel, “A Survey of Some Network Reliability Analysis and
Synthesis Results,” Networks, pp. 99-107, 2009.
[2] Harary, F. “The Maximum Connectivity of a Graph.” Proc. Nat. Acad. Sci. USA 48, 1142-1146, 1962.
[1] Weisstein, Eric W. “Harary Graph.” From MathWorld–A Wolfram Web Resource. http://mathworld.
wolfram.com/HararyGraph.html.
[2] Harary, F. “The Maximum Connectivity of a Graph.” Proc. Nat. Acad. Sci. USA 48, 1142-1146, 1962.
[0] D.G. Corneil, H. Lerchs, L.Stewart Burlingham, “Complement reducible graphs”, Discrete Applied Math-
ematics, Volume 3, Issue 3, 1981, Pages 163-174, ISSN 0166-218X.
[1] D.G. Corneil, H. Lerchs, L.Stewart Burlingham, “Complement reducible graphs”, Discrete Applied Math-
ematics, Volume 3, Issue 3, 1981, Pages 163-174, ISSN 0166-218X.
[1] Herzberg, A. M., & Murty, M. R. (2007). Sudoku squares and chromatic polynomials. Notices of the AMS,
54(6), 708-717.
[2] Sander, Torsten (2009), “Sudoku graphs are integral”, Electronic Journal of Combinatorics, 16 (1): Note
25, 7pp, MR 2529816
[3] Wikipedia contributors. “Glossary of Sudoku.” Wikipedia, The Free Encyclopedia, 3 Dec. 2019. Web. 22
Dec. 2019.
[1] Herzberg, A. M., & Murty, M. R. (2007). Sudoku squares and chromatic polynomials. Notices of the AMS,
54(6), 708-717.
[2] Sander, Torsten (2009), “Sudoku graphs are integral”, Electronic Journal of Combinatorics, 16 (1): Note
25, 7pp, MR 2529816
[3] Wikipedia contributors. “Glossary of Sudoku.” Wikipedia, The Free Encyclopedia, 3 Dec. 2019. Web. 22
Dec. 2019.
[1] Gil Strang, Network applications: A = incidence matrix, http://videolectures.net/mit18085f07_strang_
lec03/
[1] Fan Chung-Graham, Spectral Graph Theory, CBMS Regional Conference Series in Mathematics, Number
92, 1997.
[2] Steve Butler, Interlacing For Weighted Graphs Using The Normalized Laplacian, Electronic Journal of
Linear Algebra, Volume 16, pp. 90-98, March 2007.
1082 Bibliography
[1] Fan Chung (2005). Laplacians and the Cheeger inequality for directed graphs. Annals of Combinatorics,
9(1), 2005
[1] Fan Chung (2005). Laplacians and the Cheeger inequality for directed graphs. Annals of Combinatorics,
9(1), 2005
[1] A. Saade, F. Krzakala and L. Zdeborová “Spectral Clustering of Graphs with the Bethe Hessian”, Advances
in Neural Information Processing Systems, 2014.
[2] C. M. Le, E. Levina “Estimating the number of communities in networks by spectral methods”
arXiv:1507.00827, 2015.
[1] M. E. J. Newman, “Modularity and community structure in networks”, Proc. Natl. Acad. Sci. USA, vol.
103, pp. 8577-8582, 2006.
[1] E. A. Leicht, M. E. J. Newman, “Community structure in directed networks”, Phys. Rev Lett., vol. 100, no.
11, p. 118703, 2008.
[1] A. Saade, F. Krzakala and L. Zdeborová “Spectral clustering of graphs with the bethe hessian”, Advances
in Neural Information Processing Systems. 2014.
[1] M. E. J. Newman, “Modularity and community structure in networks”, Proc. Natl. Acad. Sci. USA, vol.
103, pp. 8577-8582, 2006.
[1] Scipy Dev. References, “Sparse Matrices”, https://docs.scipy.org/doc/scipy/reference/sparse.html
[1] GEXF File Format, http://gexf.net/
[1] GEXF File Format, http://gexf.net/
[2] GEXF schema, http://gexf.net/schema.html
[1] GEXF File Format, https://gephi.org/gexf/format/
[1] Cytoscape user’s manual: http://manual.cytoscape.org/en/stable/index.html
[1] Cytoscape user’s manual: http://manual.cytoscape.org/en/stable/index.html
[1] http://www.algorithmic-solutions.info/leda_guide/graphs/leda_native_graph_fileformat.html
[1] http://www.algorithmic-solutions.info/leda_guide/graphs/leda_native_graph_fileformat.html
[1] Graph6 specification <http://users.cecs.anu.edu.au/~bdm/data/formats.html>
[1] Sparse6 specification <https://users.cecs.anu.edu.au/~bdm/data/formats.html>
[1] Graph6 specification <https://users.cecs.anu.edu.au/~bdm/data/formats.html>
[1] Harary, F. and Read, R. “Is the Null Graph a Pointless Concept?” In Graphs and Combinatorics Conference,
George Washington University. New York: Springer-Verlag, 1973.
[1] Luc Devroye, Non-Uniform Random Variate Generation, Springer-Verlag, New York, 1986.
[1] E. Cuthill and J. McKee. Reducing the bandwidth of sparse symmetric matrices, In Proc. 24th Nat. Conf.
ACM, pages 157-172, 1969. http://doi.acm.org/10.1145/800195.805928
[2] Steven S. Skiena. 1997. The Algorithm Design Manual. Springer-Verlag New York, Inc., New York, NY,
USA.
Bibliography 1083
[1] E. Cuthill and J. McKee. Reducing the bandwidth of sparse symmetric matrices, In Proc. 24th Nat. Conf.
ACM, pages 157-72, 1969. http://doi.acm.org/10.1145/800195.805928
[2] Steven S. Skiena. 1997. The Algorithm Design Manual. Springer-Verlag New York, Inc., New York, NY,
USA.
[1] Cormen, T. H., Leiserson, C. E., Rivest, R. L., & Stein, C. (2001). Introduction to algorithms second edition.
[2] Knuth, D. E. (1997). The art of computer programming (Vol. 3). Pearson Education.
1084 Bibliography
PYTHON MODULE INDEX
a networkx.algorithms.bipartite.matching,
networkx.algorithms.approximation, 169 221
networkx.algorithms.approximation.clique,networkx.algorithms.bipartite.matrix,
174 225
networkx.algorithms.bipartite.projection,
networkx.algorithms.approximation.clustering_coefficient,
177 227
networkx.algorithms.bipartite.redundancy,
networkx.algorithms.approximation.connectivity,
169 239
networkx.algorithms.bipartite.spectral,
networkx.algorithms.approximation.distance_measures,
178 234
networkx.algorithms.boundary,
networkx.algorithms.approximation.dominating_set, 250
179 networkx.algorithms.bridges, 252
networkx.algorithms.centrality, 255
networkx.algorithms.approximation.kcomponents,
172 networkx.algorithms.chains, 300
networkx.algorithms.chordal, 301
networkx.algorithms.approximation.matching,
180 networkx.algorithms.clique, 306
networkx.algorithms.approximation.maxcut,networkx.algorithms.cluster, 313
195 networkx.algorithms.coloring, 319
networkx.algorithms.approximation.ramsey,networkx.algorithms.communicability_alg,
181 323
networkx.algorithms.community,
networkx.algorithms.approximation.steinertree, 325
181 networkx.algorithms.community.asyn_fluid,
335
networkx.algorithms.approximation.traveling_salesman,
183 networkx.algorithms.community.centrality,
networkx.algorithms.approximation.treewidth, 338
193 networkx.algorithms.community.community_utils,
networkx.algorithms.approximation.vertex_cover, 340
194 networkx.algorithms.community.kclique,
networkx.algorithms.assortativity, 196 326
networkx.algorithms.asteroidal, 208 networkx.algorithms.community.kernighan_lin,
networkx.algorithms.bipartite, 210 325
networkx.algorithms.bipartite.basic, 211 networkx.algorithms.community.label_propagation,
networkx.algorithms.bipartite.centrality, 331
240 networkx.algorithms.community.louvain,
networkx.algorithms.bipartite.cluster, 332
235 networkx.algorithms.community.lukes, 330
networkx.algorithms.bipartite.covering, networkx.algorithms.community.modularity_max,
249 327
networkx.algorithms.bipartite.edgelist, networkx.algorithms.community.quality,
215 336
networkx.algorithms.bipartite.generators,networkx.algorithms.components, 341
243 networkx.algorithms.connectivity, 360
1085
networkx.algorithms.connectivity.connectivity, 533
377 networkx.algorithms.matching, 536
networkx.algorithms.connectivity.cuts, networkx.algorithms.minors, 540
386 networkx.algorithms.mis, 548
networkx.algorithms.connectivity.disjoint_paths,
networkx.algorithms.moral, 551
373 networkx.algorithms.node_classification,
networkx.algorithms.connectivity.edge_augmentation,
552
360 networkx.algorithms.non_randomness, 550
networkx.algorithms.connectivity.edge_kcomponents,
networkx.algorithms.operators.all, 562
364 networkx.algorithms.operators.binary,
networkx.algorithms.connectivity.kcomponents, 555
370 networkx.algorithms.operators.product,
networkx.algorithms.connectivity.kcutsets, 565
372 networkx.algorithms.operators.unary, 554
networkx.algorithms.connectivity.stoerwagner,
networkx.algorithms.planar_drawing, 610
394 networkx.algorithms.planarity, 571
networkx.algorithms.connectivity.utils, networkx.algorithms.polynomials, 611
395 networkx.algorithms.reciprocity, 614
networkx.algorithms.core, 396 networkx.algorithms.regular, 615
networkx.algorithms.covering, 402 networkx.algorithms.richclub, 617
networkx.algorithms.cuts, 409 networkx.algorithms.shortest_paths.astar,
networkx.algorithms.cycles, 404 663
networkx.algorithms.d_separation, 414 networkx.algorithms.shortest_paths.dense,
networkx.algorithms.dag, 417 660
networkx.algorithms.distance_measures, networkx.algorithms.shortest_paths.generic,
432 618
networkx.algorithms.distance_regular, networkx.algorithms.shortest_paths.unweighted,
439 624
networkx.algorithms.dominance, 442 networkx.algorithms.shortest_paths.weighted,
networkx.algorithms.dominating, 444 630
networkx.algorithms.efficiency_measures, networkx.algorithms.similarity, 665
445 networkx.algorithms.simple_paths, 677
networkx.algorithms.euler, 448 networkx.algorithms.smallworld, 684
networkx.algorithms.flow, 453 networkx.algorithms.smetric, 687
networkx.algorithms.graph_hashing, 484 networkx.algorithms.sparsifiers, 688
networkx.algorithms.graphical, 487 networkx.algorithms.structuralholes, 689
networkx.algorithms.hierarchy, 492 networkx.algorithms.summarization, 692
networkx.algorithms.hybrid, 492 networkx.algorithms.swap, 696
networkx.algorithms.isolate, 494 networkx.algorithms.threshold, 699
networkx.algorithms.isomorphism, 496 networkx.algorithms.tournament, 701
networkx.algorithms.isomorphism.ismags, networkx.algorithms.traversal.beamsearch,
515 719
networkx.algorithms.isomorphism.isomorphvf2,
networkx.algorithms.traversal.breadth_first_search,
503 712
networkx.algorithms.isomorphism.tree_isomorphism,
networkx.algorithms.traversal.depth_first_search,
501 705
networkx.algorithms.isomorphism.vf2pp, networkx.algorithms.traversal.edgebfs,
499 722
networkx.algorithms.link_analysis.hits_alg,
networkx.algorithms.traversal.edgedfs,
524 720
networkx.algorithms.link_analysis.pagerank_alg,
networkx.algorithms.tree.branchings, 728
521 networkx.algorithms.tree.coding, 734
networkx.algorithms.link_prediction, 525 networkx.algorithms.tree.decomposition,
networkx.algorithms.lowest_common_ancestors, 746
1086 Python Module Index

networkx.algorithms.tree.mst, 739 networkx.generators.social, 856

networkx.algorithms.tree.operations, 738 networkx.generators.spectral_graph_forge,
networkx.algorithms.tree.recognition, 868
723 networkx.generators.stochastic, 853
networkx.algorithms.triads, 747 networkx.generators.sudoku, 883
networkx.algorithms.vitality, 753 networkx.generators.trees, 869
networkx.algorithms.voronoi, 754 networkx.generators.triads, 873
networkx.algorithms.wiener, 755
l
c networkx.linalg.algebraicconnectivity,
networkx.classes.backends, 167 892
networkx.classes.coreviews, 152 networkx.linalg.attrmatrix, 896
networkx.classes.filters, 165 networkx.linalg.bethehessianmatrix, 891
networkx.classes.function, 757 networkx.linalg.graphmatrix, 885
networkx.classes.graphviews, 150 networkx.linalg.laplacianmatrix, 887
networkx.convert, 907 networkx.linalg.modularitymatrix, 900
networkx.convert_matrix, 912 networkx.linalg.spectrum, 902
d r
networkx.drawing.layout, 1008 networkx.readwrite.adjlist, 931
networkx.drawing.nx_agraph, 1001 networkx.readwrite.edgelist, 940
networkx.drawing.nx_pydot, 1005 networkx.readwrite.gexf, 947
networkx.drawing.nx_pylab, 985 networkx.readwrite.gml, 950
networkx.readwrite.graph6, 971
e networkx.readwrite.graphml, 956
networkx.exception, 1021 networkx.readwrite.json_graph, 961
networkx.readwrite.leda, 970
g networkx.readwrite.multiline_adjlist,
networkx.generators.atlas, 781 935
networkx.generators.classic, 783 networkx.readwrite.pajek, 979
networkx.generators.cographs, 881 networkx.readwrite.sparse6, 975
networkx.generators.community, 857 networkx.relabel, 927
networkx.generators.degree_seq, 827
networkx.generators.directed, 835 u
networkx.generators.duplication, 825 networkx.utils, 1023
networkx.generators.ego, 852 networkx.utils.decorators, 1030
networkx.generators.expanders, 793 networkx.utils.mapped_queue, 1044
networkx.generators.geometric, 839 networkx.utils.misc, 1023
networkx.generators.harary_graph, 879 networkx.utils.random_sequence, 1028
networkx.generators.internet_as_graphs, networkx.utils.rcm, 1042
853 networkx.utils.union_find, 1027
networkx.generators.intersection, 854
networkx.generators.interval_graph, 882
networkx.generators.joint_degree_seq,
874
networkx.generators.lattice, 795
networkx.generators.line, 849
networkx.generators.mycielski, 878
networkx.generators.nonisomorphic_trees,
872
networkx.generators.random_clustered,
833
networkx.generators.random_graphs, 810
networkx.generators.small, 798
Python Module Index 1087

1088 Python Module Index

INDEX
Non-alphabetical __len__() (MultiGraph method), 102

__contains__() (DiGraph method), 55 _dispatch() (in module networkx.classes.backends),
__contains__() (Graph method), 23 168
__contains__() (MultiDiGraph method), 129
__contains__() (MultiGraph method), 94
A
__getitem__() (DiGraph method), 60 adamic_adar_index() (in module net-
__getitem__() (Graph method), 27 workx.algorithms.link_prediction), 527
__getitem__() (MultiDiGraph method), 136 add_cycle() (in module networkx.classes.function),
__getitem__() (MultiGraph method), 99 761
__init__() (AdjacencyView method), 154 add_edge() (DiGraph method), 46
__init__() (ArborescenceIterator method), 732 add_edge() (Graph method), 15
__init__() (argmap method), 1040 add_edge() (MultiDiGraph method), 118
__init__() (AtlasView method), 153 add_edge() (MultiGraph method), 83
__init__() (DiGraph method), 42 add_edge() (PlanarEmbedding method), 576
__init__() (DiGraphMatcher method), 507 add_edges_from() (DiGraph method), 47
__init__() (EdgeComponentAuxGraph method), 369 add_edges_from() (Graph method), 16
__init__() (Edmonds method), 733 add_edges_from() (MultiDiGraph method), 119
__init__() (FilterAdjacency method), 162 add_edges_from() (MultiGraph method), 84
__init__() (FilterAtlas method), 161 add_edges_from() (PlanarEmbedding method), 577
__init__() (FilterMultiAdjacency method), 164 add_half_edge_ccw() (PlanarEmbedding method),
__init__() (FilterMultiInner method), 163 578
__init__() (Graph method), 11 add_half_edge_cw() (PlanarEmbedding method),
__init__() (GraphMatcher method), 505 579
__init__() (ISMAGS method), 518 add_half_edge_first() (PlanarEmbedding
__init__() (MappedQueue method), 1045 method), 579
__init__() (MultiAdjacencyView method), 156 add_node() (DiGraph method), 43
__init__() (MultiDiGraph method), 114 add_node() (Graph method), 12
__init__() (MultiGraph method), 79 add_node() (MultiDiGraph method), 115
__init__() (PlanarEmbedding method), 574 add_node() (MultiGraph method), 80
__init__() (show_nodes method), 166 add_node() (PlanarEmbedding method), 580
__init__() (SpanningTreeIterator method), 745 add_nodes_from() (DiGraph method), 44
__init__() (UnionAdjacency method), 158 add_nodes_from() (Graph method), 13
__init__() (UnionAtlas method), 157 add_nodes_from() (MultiDiGraph method), 116
__init__() (UnionMultiAdjacency method), 160 add_nodes_from() (MultiGraph method), 81
__init__() (UnionMultiInner method), 159 add_nodes_from() (PlanarEmbedding method), 580
__iter__() (DiGraph method), 54 add_path() (in module networkx.classes.function), 760
__iter__() (Graph method), 23 add_star() (in module networkx.classes.function), 760
__iter__() (MultiDiGraph method), 128 add_weighted_edges_from() (DiGraph method),
__iter__() (MultiGraph method), 93 47
__len__() (DiGraph method), 65 add_weighted_edges_from() (Graph method), 16
__len__() (Graph method), 30 add_weighted_edges_from() (MultiDiGraph
__len__() (MultiDiGraph method), 140 method), 120
1089
add_weighted_edges_from() (MultiGraph 169

method), 85 all_pairs_node_connectivity() (in module
add_weighted_edges_from() (PlanarEmbedding networkx.algorithms.connectivity.connectivity),
method), 581 378
adj (DiGraph property), 60 all_pairs_shortest_path() (in module net-
adj (Graph property), 27 workx.algorithms.shortest_paths.unweighted),
adj (MultiDiGraph property), 135 628
adj (MultiGraph property), 98 all_pairs_shortest_path_length()
adj (PlanarEmbedding property), 602 (in module net-
adjacency() (DiGraph method), 62 workx.algorithms.shortest_paths.unweighted),
adjacency() (Graph method), 28 628
adjacency() (MultiDiGraph method), 137 all_shortest_paths() (in module net-
adjacency() (MultiGraph method), 99 workx.algorithms.shortest_paths.generic),
adjacency() (PlanarEmbedding method), 582 620
adjacency_data() (in module net- all_simple_edge_paths() (in module net-
workx.readwrite.json_graph), 965 workx.algorithms.simple_paths), 680
adjacency_graph() (in module net- all_simple_paths() (in module net-
workx.readwrite.json_graph), 966 workx.algorithms.simple_paths), 677
adjacency_matrix() (in module net- all_topological_sorts() (in module net-
workx.linalg.graphmatrix), 885 workx.algorithms.dag), 421
adjacency_spectrum() (in module net- all_triads() (in module networkx.algorithms.triads),
workx.linalg.spectrum), 902 752
AdjacencyView (class in networkx.classes.coreviews), all_triplets() (in module net-
154 workx.algorithms.triads), 752
algebraic_connectivity() (in module net- alternating_havel_hakimi_graph() (in mod-
workx.linalg.algebraicconnectivity), 892 ule networkx.algorithms.bipartite.generators),
all_neighbors() (in module net- 246
workx.classes.function), 767 AmbiguousSolution (class in networkx), 1021
all_node_cuts() (in module net- analyze_symmetry() (ISMAGS method), 519
workx.algorithms.connectivity.kcutsets), 372 ancestors() (in module networkx.algorithms.dag), 418
all_pairs_bellman_ford_path() (in module antichains() (in module networkx.algorithms.dag),
networkx.algorithms.shortest_paths.weighted), 428
652 approximate_current_flow_betweenness_centrality()
all_pairs_bellman_ford_path_length() (in module networkx.algorithms.centrality), 276
(in module net- arbitrary_element() (in module net-
workx.algorithms.shortest_paths.weighted), workx.utils.misc), 1024
653 ArborescenceIterator (class in net-
all_pairs_dijkstra() (in module networkx.algorithms.tree.branchings), 731
workx.algorithms.shortest_paths.weighted), argmap (class in networkx.utils.decorators), 1035
643 articulation_points() (in module net-
all_pairs_dijkstra_path() (in module networkx.algorithms.components), 358
workx.algorithms.shortest_paths.weighted), asadpour_atsp() (in module net-
644 workx.algorithms.approximation.traveling_salesman),
all_pairs_dijkstra_path_length() 192
(in module net- assemble() (argmap method), 1040
workx.algorithms.shortest_paths.weighted), astar_path() (in module net-
645 workx.algorithms.shortest_paths.astar), 663
all_pairs_lowest_common_ancestor() astar_path_length() (in module net-
(in module networkx.algorithms.shortest_paths.astar), 665
workx.algorithms.lowest_common_ancestors), asyn_fluidc() (in module net-
533 workx.algorithms.community.asyn_fluid),
all_pairs_node_connectivity() 335
(in module net- asyn_lpa_communities() (in module net-
workx.algorithms.approximation.connectivity), workx.algorithms.community.label_propagation),
1090 Index
331 betweenness_centrality() (in module net-

AtlasView (class in networkx.classes.coreviews), 153 workx.algorithms.bipartite.centrality), 242
attr_matrix() (in module networkx.linalg.attrmatrix), betweenness_centrality() (in module net-
896 workx.algorithms.centrality), 269
attr_sparse_matrix() (in module net- betweenness_centrality_subset() (in module
workx.linalg.attrmatrix), 898 networkx.algorithms.centrality), 270
attracting_components() (in module net- bfs_beam_edges() (in module net-
workx.algorithms.components), 354 workx.algorithms.traversal.beamsearch), 719
attribute_assortativity_coefficient() bfs_edges() (in module net-
(in module networkx.algorithms.assortativity), workx.algorithms.traversal.breadth_first_search),
198 713
attribute_mixing_dict() (in module net- bfs_layers() (in module net-
workx.algorithms.assortativity), 205 workx.algorithms.traversal.breadth_first_search),
attribute_mixing_matrix() (in module net- 714
workx.algorithms.assortativity), 203 bfs_predecessors() (in module net-
average_clustering() (in module networkx.algorithms.traversal.breadth_first_search),
workx.algorithms.approximation.clustering_coefficient), 716
177 bfs_successors() (in module net-
average_clustering() (in module networkx.algorithms.traversal.breadth_first_search),
workx.algorithms.bipartite.cluster), 236 717
average_clustering() (in module net- bfs_tree() (in module net-
workx.algorithms.cluster), 316 workx.algorithms.traversal.breadth_first_search),
average_degree_connectivity() (in module 715
networkx.algorithms.assortativity), 202 biadjacency_matrix() (in module net-
average_neighbor_degree() (in module networkx.algorithms.bipartite.matrix), 225
workx.algorithms.assortativity), 200 biconnected_component_edges() (in module
average_node_connectivity() (in module net- networkx.algorithms.components), 357
workx.algorithms.connectivity.connectivity), 377 biconnected_components() (in module net-
average_shortest_path_length() (in module workx.algorithms.components), 355
networkx.algorithms.shortest_paths.generic), 623 bidirectional_dijkstra() (in module net-
workx.algorithms.shortest_paths.weighted),
B 646
balanced_tree() (in module net- bidirectional_shortest_path() (in module
workx.generators.classic), 783 networkx.algorithms.shortest_paths.unweighted),
barabasi_albert_graph() (in module net- 627
workx.generators.random_graphs), 819 binomial_graph() (in module net-
barbell_graph() (in module networkx.generators.random_graphs), 815
workx.generators.classic), 784 binomial_tree() (in module net-
barycenter() (in module networkx.generators.classic), 784
workx.algorithms.distance_measures), 433 bipartite_layout() (in module net-
bellman_ford_path() (in module networkx.drawing.layout), 1009
workx.algorithms.shortest_paths.weighted), boundary_expansion() (in module net-
647 workx.algorithms.cuts), 409
bellman_ford_path_length() (in module net- boykov_kolmogorov() (in module net-
workx.algorithms.shortest_paths.weighted), 648 workx.algorithms.flow), 469
bellman_ford_predecessor_and_distance() branching_weight() (in module net-
(in module networkx.algorithms.tree.branchings), 728
workx.algorithms.shortest_paths.weighted), bridge_components() (in module net-
654 workx.algorithms.connectivity.edge_kcomponents),
bethe_hessian_matrix() (in module net- 367
workx.linalg.bethehessianmatrix), 891 bridges() (in module networkx.algorithms.bridges), 252
bethe_hessian_spectrum() (in module net- build_auxiliary_edge_connectivity() (in
workx.linalg.spectrum), 904 module networkx.algorithms.connectivity.utils),
395
Index 1091
build_auxiliary_node_connectivity() (in clear() (MultiDiGraph method), 125

module networkx.algorithms.connectivity.utils), clear() (MultiGraph method), 90
395 clear() (PlanarEmbedding method), 583
build_residual_network() (in module net- clear_edges() (DiGraph method), 51
workx.algorithms.flow), 473 clear_edges() (Graph method), 20
bull_graph() (in module networkx.generators.small), clear_edges() (MultiDiGraph method), 125
799 clear_edges() (MultiGraph method), 90
clear_edges() (PlanarEmbedding method), 583
C clique_removal() (in module net-
candidate_pairs_iter() (DiGraphMatcher workx.algorithms.approximation.clique), 176
method), 508 cliques_containing_node() (in module net-
candidate_pairs_iter() (GraphMatcher workx.algorithms.clique), 312
method), 506 closeness_centrality() (in module net-
capacity_scaling() (in module networkx.algorithms.bipartite.centrality), 240
workx.algorithms.flow), 482 closeness_centrality() (in module net-
cartesian_product() (in module networkx.algorithms.centrality), 264
workx.algorithms.operators.product), 565 closeness_vitality() (in module net-
categorical_edge_match() (in module networkx.algorithms.vitality), 753
workx.algorithms.isomorphism), 510 clustering() (in module net-
categorical_multiedge_match() (in module workx.algorithms.bipartite.cluster), 235
networkx.algorithms.isomorphism), 511 clustering() (in module networkx.algorithms.cluster),
categorical_node_match() (in module net- 314
workx.algorithms.isomorphism), 510 cn_soundarajan_hopcroft() (in module net-
caveman_graph() (in module networkx.algorithms.link_prediction), 529
workx.generators.community), 858 collaboration_weighted_projected_graph()
center() (in module net- (in module net-
workx.algorithms.distance_measures), 434 workx.algorithms.bipartite.projection), 229
chain_decomposition() (in module net- color() (in module networkx.algorithms.bipartite.basic),
workx.algorithms.chains), 300 213
check_planarity() (in module net- combinatorial_embedding_to_pos() (in mod-
workx.algorithms.planarity), 571 ule networkx.algorithms.planar_drawing), 610
check_structure() (PlanarEmbedding method), common_neighbor_centrality() (in module net-
582 workx.algorithms.link_prediction), 532
chordal_cycle_graph() (in module net- common_neighbors() (in module net-
workx.generators.expanders), 793 workx.classes.function), 767
chordal_graph_cliques() (in module net- communicability() (in module net-
workx.algorithms.chordal), 302 workx.algorithms.communicability_alg), 323
chordal_graph_treewidth() (in module net- communicability_betweenness_centrality()
workx.algorithms.chordal), 303 (in module networkx.algorithms.centrality), 279
christofides() (in module net- communicability_exp() (in module net-
workx.algorithms.approximation.traveling_salesman), workx.algorithms.communicability_alg), 324
184 compile() (argmap method), 1041
chromatic_polynomial() (in module net- complement() (in module net-
workx.algorithms.polynomials), 613 workx.algorithms.operators.unary), 554
chvatal_graph() (in module net- complete_bipartite_graph() (in module net-
workx.generators.small), 800 workx.algorithms.bipartite.generators), 244
circulant_graph() (in module net- complete_graph() (in module net-
workx.generators.classic), 786 workx.generators.classic), 785
circular_ladder_graph() (in module net- complete_multipartite_graph() (in module
workx.generators.classic), 786 networkx.generators.classic), 785
circular_layout() (in module net- complete_to_chordal_graph() (in module net-
workx.drawing.layout), 1009 workx.algorithms.chordal), 304
clear() (DiGraph method), 51 compose() (in module net-
clear() (Graph method), 20 workx.algorithms.operators.binary), 556
1092 Index
compose_all() (in module net- create_random_state() (in module net-

workx.algorithms.operators.all), 563 workx.utils.misc), 1026
condensation() (in module net- cubical_graph() (in module net-
workx.algorithms.components), 349 workx.generators.small), 800
conductance() (in module networkx.algorithms.cuts), cumulative_distribution() (in module net-
410 workx.utils.random_sequence), 1028
configuration_model() (in module net- current_flow_betweenness_centrality()
workx.algorithms.bipartite.generators), 244 (in module networkx.algorithms.centrality), 274
configuration_model() (in module net- current_flow_betweenness_centrality_subset()
workx.generators.degree_seq), 827 (in module networkx.algorithms.centrality), 277
connect_components() (PlanarEmbedding current_flow_closeness_centrality() (in
method), 583 module networkx.algorithms.centrality), 267
connected_caveman_graph() (in module net- cut_size() (in module networkx.algorithms.cuts), 410
workx.generators.community), 858 cuthill_mckee_ordering() (in module net-
connected_components() (in module networkx.utils.rcm), 1042
workx.algorithms.components), 342 cycle_basis() (in module net-
connected_double_edge_swap() (in module networkx.algorithms.cycles), 404
workx.algorithms.swap), 698 cycle_graph() (in module net-
connected_watts_strogatz_graph() (in mod- workx.generators.classic), 788
ule networkx.generators.random_graphs), 817 cytoscape_data() (in module net-
constraint() (in module networkx.readwrite.json_graph), 966
workx.algorithms.structuralholes), 689 cytoscape_graph() (in module net-
construct() (EdgeComponentAuxGraph class workx.readwrite.json_graph), 967
method), 369
contracted_edge() (in module net- D
workx.algorithms.minors), 541 d_separated() (in module net-
contracted_nodes() (in module networkx.algorithms.d_separation), 416
workx.algorithms.minors), 542 dag_longest_path() (in module net-
convert_node_labels_to_integers() (in workx.algorithms.dag), 429
module networkx.relabel), 927 dag_longest_path_length() (in module net-
copy() (AdjacencyView method), 155 workx.algorithms.dag), 430
copy() (AtlasView method), 154 dag_to_branching() (in module net-
copy() (DiGraph method), 69 workx.algorithms.dag), 431
copy() (Graph method), 33 davis_southern_women_graph() (in module net-
copy() (MultiAdjacencyView method), 156 workx.generators.social), 857
copy() (MultiDiGraph method), 145 dedensify() (in module net-
copy() (MultiGraph method), 105 workx.algorithms.summarization), 693
copy() (PlanarEmbedding method), 584 degree (DiGraph property), 65
copy() (UnionAdjacency method), 158 degree (Graph property), 31
copy() (UnionAtlas method), 157 degree (MultiDiGraph property), 140
copy() (UnionMultiAdjacency method), 161 degree (MultiGraph property), 102
copy() (UnionMultiInner method), 160 degree (PlanarEmbedding property), 602
core_number() (in module networkx.algorithms.core), degree() (in module networkx.classes.function), 757
396 degree_assortativity_coefficient() (in
corona_product() (in module net- module networkx.algorithms.assortativity), 197
workx.algorithms.operators.product), 570 degree_centrality() (in module net-
cost_of_flow() (in module networkx.algorithms.bipartite.centrality), 241
workx.algorithms.flow), 479 degree_centrality() (in module net-
could_be_isomorphic() (in module networkx.algorithms.centrality), 255
workx.algorithms.isomorphism), 498 degree_histogram() (in module net-
create_empty_copy() (in module networkx.classes.function), 758
workx.classes.function), 758 degree_mixing_dict() (in module net-
create_py_random_state() (in module networkx.algorithms.assortativity), 206
workx.utils.misc), 1026
Index 1093
degree_mixing_matrix() (in module net- DiGraph (class in networkx), 38

workx.algorithms.assortativity), 204 dijkstra_path() (in module net-
degree_pearson_correlation_coefficient() workx.algorithms.shortest_paths.weighted),
(in module networkx.algorithms.assortativity), 633
199 dijkstra_path_length() (in module net-
degree_sequence_tree() (in module networkx.algorithms.shortest_paths.weighted),
workx.generators.degree_seq), 832 634
degrees() (in module net- dijkstra_predecessor_and_distance()
workx.algorithms.bipartite.basic), 215 (in module net-
dense_gnm_random_graph() (in module networkx.algorithms.shortest_paths.weighted),
workx.generators.random_graphs), 813 632
density() (in module net- dinitz() (in module networkx.algorithms.flow), 467
workx.algorithms.bipartite.basic), 214 directed_combinatorial_laplacian_matrix()
density() (in module networkx.classes.function), 758 (in module networkx.linalg.laplacianmatrix), 890
desargues_graph() (in module net- directed_configuration_model() (in module
workx.generators.small), 801 networkx.generators.degree_seq), 828
descendants() (in module networkx.algorithms.dag), directed_edge_swap() (in module net-
418 workx.algorithms.swap), 697
descendants_at_distance() (in module net- directed_havel_hakimi_graph() (in module
workx.algorithms.traversal.breadth_first_search), networkx.generators.degree_seq), 831
718 directed_joint_degree_graph() (in module
dfs_edges() (in module net- networkx.generators.joint_degree_seq), 876
workx.algorithms.traversal.depth_first_search), directed_laplacian_matrix() (in module net-
706 workx.linalg.laplacianmatrix), 889
dfs_labeled_edges() (in module net- directed_modularity_matrix() (in module net-
workx.algorithms.traversal.depth_first_search), workx.linalg.modularitymatrix), 901
711 discrete_sequence() (in module net-
dfs_postorder_nodes() (in module networkx.utils.random_sequence), 1028
workx.algorithms.traversal.depth_first_search), disjoint_union() (in module net-
710 workx.algorithms.operators.binary), 558
dfs_predecessors() (in module net- disjoint_union_all() (in module net-
workx.algorithms.traversal.depth_first_search), workx.algorithms.operators.all), 564
708 dispersion() (in module net-
dfs_preorder_nodes() (in module networkx.algorithms.centrality), 292
workx.algorithms.traversal.depth_first_search), dodecahedral_graph() (in module net-
710 workx.generators.small), 802
dfs_successors() (in module net- dominance_frontiers() (in module net-
workx.algorithms.traversal.depth_first_search), workx.algorithms.dominance), 443
709 dominating_set() (in module net-
dfs_tree() (in module networkx.algorithms.dominating), 444
workx.algorithms.traversal.depth_first_search), dorogovtsev_goltsev_mendes_graph() (in
707 module networkx.generators.classic), 788
diameter() (in module net- double_edge_swap() (in module net-
workx.algorithms.approximation.distance_measures), workx.algorithms.swap), 697
178 draw() (in module networkx.drawing.nx_pylab), 986
diameter() (in module net- draw_circular() (in module net-
workx.algorithms.distance_measures), 434 workx.drawing.nx_pylab), 996
diamond_graph() (in module net- draw_kamada_kawai() (in module net-
workx.generators.small), 801 workx.drawing.nx_pylab), 997
dict_to_numpy_array() (in module net- draw_networkx() (in module net-
workx.utils.misc), 1025 workx.drawing.nx_pylab), 987
dictionary, 1047 draw_networkx_edge_labels() (in module net-
difference() (in module networkx.drawing.nx_pylab), 995
workx.algorithms.operators.binary), 560 draw_networkx_edges() (in module net-
1094 Index
workx.drawing.nx_pylab), 991 edge_subgraph() (MultiDiGraph method), 149

draw_networkx_labels() (in module net- edge_subgraph() (MultiGraph method), 109
workx.drawing.nx_pylab), 994 edge_subgraph() (PlanarEmbedding method), 585
draw_networkx_nodes() (in module net- EdgeComponentAuxGraph (class in net-
workx.drawing.nx_pylab), 989 workx.algorithms.connectivity.edge_kcomponents),
draw_planar() (in module net- 367
workx.drawing.nx_pylab), 997 edges (DiGraph property), 55
draw_random() (in module net- edges (Graph property), 24
workx.drawing.nx_pylab), 998 edges (MultiDiGraph property), 129
draw_shell() (in module networkx.drawing.nx_pylab), edges (MultiGraph property), 94
1000 edges (PlanarEmbedding property), 603
draw_spectral() (in module net- edges() (in module networkx.classes.function), 768
workx.drawing.nx_pylab), 999 edges_equal() (in module networkx.utils.misc), 1027
draw_spring() (in module net- Edmonds (class in networkx.algorithms.tree.branchings),
workx.drawing.nx_pylab), 1000 732
dual_barabasi_albert_graph() (in module net- edmonds_karp() (in module net-
workx.generators.random_graphs), 819 workx.algorithms.flow), 461
duplication_divergence_graph() (in module effective_size() (in module net-
networkx.generators.duplication), 825 workx.algorithms.structuralholes), 690
efficiency() (in module net-
E workx.algorithms.efficiency_measures), 445
ebunch, 1047 ego_graph() (in module networkx.generators.ego), 852
eccentricity() (in module net- eigenvector_centrality() (in module net-
workx.algorithms.distance_measures), 435 workx.algorithms.centrality), 258
edge, 1047 eigenvector_centrality_numpy() (in module
edge attribute, 1047 networkx.algorithms.centrality), 259
edge_betweenness_centrality() (in module empty_graph() (in module net-
networkx.algorithms.centrality), 271 workx.generators.classic), 788
edge_betweenness_centrality_subset() (in enumerate_all_cliques() (in module net-
module networkx.algorithms.centrality), 272 workx.algorithms.clique), 306
edge_bfs() (in module net- eppstein_matching() (in module net-
workx.algorithms.traversal.edgebfs), 722 workx.algorithms.bipartite.matching), 221
edge_boundary() (in module net- equitable_color() (in module net-
workx.algorithms.boundary), 250 workx.algorithms.coloring), 320
edge_connectivity() (in module net- equivalence_classes() (in module net-
workx.algorithms.connectivity.connectivity), workx.algorithms.minors), 544
379 erdos_renyi_graph() (in module net-
edge_current_flow_betweenness_centrality() workx.generators.random_graphs), 814
(in module networkx.algorithms.centrality), 275 estrada_index() (in module net-
edge_current_flow_betweenness_centrality_subset() workx.algorithms.centrality), 290
(in module networkx.algorithms.centrality), 278 eulerian_circuit() (in module net-
edge_dfs() (in module networkx.algorithms.euler), 449
workx.algorithms.traversal.edgedfs), 720 eulerian_path() (in module net-
edge_disjoint_paths() (in module networkx.algorithms.euler), 452
workx.algorithms.connectivity.disjoint_paths), eulerize() (in module networkx.algorithms.euler), 450
373 ExceededMaxIterations (class in networkx), 1022
edge_expansion() (in module net- expected_degree_graph() (in module net-
workx.algorithms.cuts), 411 workx.generators.degree_seq), 830
edge_load_centrality() (in module net- extended_barabasi_albert_graph() (in mod-
workx.algorithms.centrality), 288 ule networkx.generators.random_graphs), 820
edge_subgraph() (DiGraph method), 73
edge_subgraph() (Graph method), 37 F
edge_subgraph() (in module net- fast_could_be_isomorphic() (in module net-
workx.classes.function), 765 workx.algorithms.isomorphism), 498
Index 1095
fast_gnp_random_graph() (in module net- from_nested_tuple() (in module net-

workx.generators.random_graphs), 811 workx.algorithms.tree.coding), 734
faster_could_be_isomorphic() (in module net- from_numpy_array() (in module net-
workx.algorithms.isomorphism), 498 workx.convert_matrix), 915
fiedler_vector() (in module net- from_pandas_adjacency() (in module net-
workx.linalg.algebraicconnectivity), 894 workx.convert_matrix), 921
FilterAdjacency (class in net- from_pandas_edgelist() (in module net-
workx.classes.coreviews), 162 workx.convert_matrix), 923
FilterAtlas (class in networkx.classes.coreviews), 161 from_prufer_sequence() (in module net-
FilterMultiAdjacency (class in networkx.algorithms.tree.coding), 736
workx.classes.coreviews), 164 from_pydot() (in module net-
FilterMultiInner (class in networkx.drawing.nx_pydot), 1005
workx.classes.coreviews), 163 from_scipy_sparse_array() (in module net-
find_asteroidal_triple() (in module networkx.convert_matrix), 919
workx.algorithms.asteroidal), 209 from_sparse6_bytes() (in module net-
find_cliques() (in module networkx.readwrite.sparse6), 975
workx.algorithms.clique), 307 frucht_graph() (in module net-
find_cliques_recursive() (in module networkx.generators.small), 802
workx.algorithms.clique), 308 full_join() (in module net-
find_cycle() (in module networkx.algorithms.cycles), workx.algorithms.operators.binary), 561
407 full_rary_tree() (in module net-
find_induced_nodes() (in module networkx.generators.classic), 790
workx.algorithms.chordal), 305
find_isomorphisms() (ISMAGS method), 520 G
find_negative_cycle() (in module net- gaussian_random_partition_graph() (in
workx.algorithms.shortest_paths.weighted), module networkx.generators.community), 859
657 general_random_intersection_graph() (in
find_optimum() (Edmonds method), 733 module networkx.generators.intersection), 855
find_threshold_graph() (in module net- generalized_degree() (in module net-
workx.algorithms.threshold), 699 workx.algorithms.cluster), 318
flatten() (in module networkx.utils.misc), 1025 generate_adjlist() (in module net-
florentine_families_graph() (in module networkx.readwrite.adjlist), 934
workx.generators.social), 857 generate_edgelist() (in module net-
flow_hierarchy() (in module networkx.algorithms.bipartite.edgelist), 216
workx.algorithms.hierarchy), 492 generate_edgelist() (in module net-
floyd_warshall() (in module networkx.readwrite.edgelist), 944
workx.algorithms.shortest_paths.dense), 660 generate_gexf() (in module net-
floyd_warshall_numpy() (in module networkx.readwrite.gexf), 949
workx.algorithms.shortest_paths.dense), 662 generate_gml() (in module networkx.readwrite.gml),
floyd_warshall_predecessor_and_distance() 954
(in module net- generate_graphml() (in module net-
workx.algorithms.shortest_paths.dense), 661 workx.readwrite.graphml), 959
freeze() (in module networkx.classes.function), 778 generate_multiline_adjlist() (in module net-
from_agraph() (in module networkx.readwrite.multiline_adjlist), 939
workx.drawing.nx_agraph), 1002 generate_pajek() (in module net-
from_biadjacency_matrix() (in module networkx.readwrite.pajek), 981
workx.algorithms.bipartite.matrix), 226 generate_random_paths() (in module net-
from_dict_of_dicts() (in module networkx.algorithms.similarity), 676
workx.convert), 910 generic_edge_match() (in module net-
from_dict_of_lists() (in module networkx.algorithms.isomorphism), 514
workx.convert), 911 generic_graph_view() (in module net-
from_edgelist() (in module networkx.convert), 912 workx.classes.graphviews), 150
from_graph6_bytes() (in module net- generic_multiedge_match() (in module net-
workx.readwrite.graph6), 972 workx.algorithms.isomorphism), 515
1096 Index
generic_node_match() (in module networkx.algorithms.flow), 471

workx.algorithms.isomorphism), 513 google_matrix() (in module net-
generic_weighted_projected_graph() (in workx.algorithms.link_analysis.pagerank_alg),
module networkx.algorithms.bipartite.projection), 523
232 Graph (class in networkx), 7
geographical_threshold_graph() (in module graph_atlas() (in module networkx.generators.atlas),
networkx.generators.geometric), 841 781
geometric_edges() (in module net- graph_atlas_g() (in module net-
workx.generators.geometric), 840 workx.generators.atlas), 782
get() (AdjacencyView method), 155 graph_clique_number() (in module net-
get() (AtlasView method), 154 workx.algorithms.clique), 310
get() (FilterAdjacency method), 163 graph_edit_distance() (in module net-
get() (FilterAtlas method), 162 workx.algorithms.similarity), 666
get() (FilterMultiAdjacency method), 164 graph_number_of_cliques() (in module net-
get() (FilterMultiInner method), 163 workx.algorithms.clique), 311
get() (MultiAdjacencyView method), 156 graphs_equal() (in module networkx.utils.misc), 1027
get() (UnionAdjacency method), 158 graphviz_layout() (in module net-
get() (UnionAtlas method), 157 workx.drawing.nx_agraph), 1003
get() (UnionMultiAdjacency method), 161 graphviz_layout() (in module net-
get() (UnionMultiInner method), 160 workx.drawing.nx_pydot), 1007
get_data() (PlanarEmbedding method), 586 greedy_branching() (in module net-
get_edge_attributes() (in module networkx.algorithms.tree.branchings), 729
workx.classes.function), 776 greedy_color() (in module net-
get_edge_data() (DiGraph method), 59 workx.algorithms.coloring), 319
get_edge_data() (Graph method), 25 greedy_modularity_communities()
get_edge_data() (MultiDiGraph method), 134 (in module net-
get_edge_data() (MultiGraph method), 96 workx.algorithms.community.modularity_max),
get_edge_data() (PlanarEmbedding method), 586 328
get_node_attributes() (in module net- greedy_tsp() (in module net-
workx.classes.function), 774 workx.algorithms.approximation.traveling_salesman),
girvan_newman() (in module net- 186
workx.algorithms.community.centrality), 338 grid_2d_graph() (in module net-
global_efficiency() (in module networkx.generators.lattice), 795
workx.algorithms.efficiency_measures), 447 grid_graph() (in module networkx.generators.lattice),
global_parameters() (in module net- 795
workx.algorithms.distance_regular), 441 group_betweenness_centrality() (in module
global_reaching_centrality() (in module net- networkx.algorithms.centrality), 281
workx.algorithms.centrality), 294 group_closeness_centrality() (in module net-
gn_graph() (in module networkx.generators.directed), workx.algorithms.centrality), 282
835 group_degree_centrality() (in module net-
gnc_graph() (in module networkx.generators.directed), workx.algorithms.centrality), 283
837 group_in_degree_centrality() (in module net-
gnm_random_graph() (in module networkx.algorithms.centrality), 284
workx.generators.random_graphs), 813 group_out_degree_centrality() (in module
gnmk_random_graph() (in module net- networkx.algorithms.centrality), 285
workx.algorithms.bipartite.generators), 248 groups() (in module networkx.utils.misc), 1025
gnp_random_graph() (in module net-
workx.generators.random_graphs), 812 H
gnr_graph() (in module networkx.generators.directed), hamiltonian_path() (in module net-
836 workx.algorithms.tournament), 701
goldberg_radzik() (in module net- harmonic_centrality() (in module net-
workx.algorithms.shortest_paths.weighted), workx.algorithms.centrality), 291
658 harmonic_function() (in module net-
gomory_hu_tree() (in module networkx.algorithms.node_classification), 552
Index 1097
has_bridges() (in module net- I

workx.algorithms.bridges), 253 icosahedral_graph() (in module net-
has_edge() (DiGraph method), 58 workx.generators.small), 805
has_edge() (Graph method), 25 identified_nodes() (in module net-
has_edge() (MultiDiGraph method), 133 workx.algorithms.minors), 543
has_edge() (MultiGraph method), 95 immediate_dominators() (in module net-
has_edge() (PlanarEmbedding method), 587 workx.algorithms.dominance), 442
has_eulerian_path() (in module net- in_degree (DiGraph property), 66
workx.algorithms.euler), 451 in_degree (MultiDiGraph property), 141
has_node() (DiGraph method), 54 in_degree (PlanarEmbedding property), 604
has_node() (Graph method), 23 in_degree_centrality() (in module net-
has_node() (MultiDiGraph method), 128 workx.algorithms.centrality), 256
has_node() (MultiGraph method), 93 in_edges (DiGraph property), 57
has_node() (PlanarEmbedding method), 587 in_edges (MultiDiGraph property), 132
has_path() (in module net- in_edges (PlanarEmbedding property), 605
workx.algorithms.shortest_paths.generic), incidence_matrix() (in module net-
624 workx.linalg.graphmatrix), 886
has_predecessor() (PlanarEmbedding method), incremental_closeness_centrality() (in
588 module networkx.algorithms.centrality), 266
has_successor() (PlanarEmbedding method), 588 induced_subgraph() (in module net-
HasACycle (class in networkx), 1021 workx.classes.function), 763
havel_hakimi_graph() (in module net- information_centrality() (in module net-
workx.algorithms.bipartite.generators), 245 workx.algorithms.centrality), 268
havel_hakimi_graph() (in module net- initialize() (DiGraphMatcher method), 508
workx.generators.degree_seq), 831 initialize() (GraphMatcher method), 506
heawood_graph() (in module net- intersection() (in module net-
workx.generators.small), 803 workx.algorithms.operators.binary), 559
hexagonal_lattice_graph() (in module net- intersection_all() (in module net-
workx.generators.lattice), 796 workx.algorithms.operators.all), 564
hide_diedges() (in module networkx.classes.filters), intersection_array() (in module net-
166 workx.algorithms.distance_regular), 441
hide_edges() (in module networkx.classes.filters), 166 interval_graph() (in module net-
hide_multidiedges() (in module networkx.generators.interval_graph), 882
workx.classes.filters), 166 inverse_line_graph() (in module net-
hide_multiedges() (in module networkx.generators.line), 851
workx.classes.filters), 166 is_aperiodic() (in module networkx.algorithms.dag),
hide_nodes() (in module networkx.classes.filters), 166 424
hits() (in module net- is_arborescence() (in module net-
workx.algorithms.link_analysis.hits_alg), 524 workx.algorithms.tree.recognition), 726
hkn_harary_graph() (in module net- is_at_free() (in module net-
workx.generators.harary_graph), 880 workx.algorithms.asteroidal), 208
hnm_harary_graph() (in module net- is_attracting_component() (in module net-
workx.generators.harary_graph), 879 workx.algorithms.components), 353
hoffman_singleton_graph() (in module net- is_biconnected() (in module net-
workx.generators.small), 803 workx.algorithms.components), 354
hopcroft_karp_matching() (in module net- is_bipartite() (in module net-
workx.algorithms.bipartite.matching), 222 workx.algorithms.bipartite.basic), 211
house_graph() (in module networkx.generators.small), is_bipartite_node_set() (in module net-
804 workx.algorithms.bipartite.basic), 212
house_x_graph() (in module net- is_branching() (in module net-
workx.generators.small), 804 workx.algorithms.tree.recognition), 727
hypercube_graph() (in module net- is_chordal() (in module net-
workx.generators.lattice), 797 workx.algorithms.chordal), 301
1098 Index
is_connected() (in module net- is_path() (in module networkx.classes.function), 777

workx.algorithms.components), 341 is_perfect_matching() (in module net-
is_digraphical() (in module networkx.algorithms.matching), 537
workx.algorithms.graphical), 489 is_planar() (in module net-
is_directed() (in module networkx.classes.function), workx.algorithms.planarity), 572
759 is_pseudographical() (in module net-
is_directed() (PlanarEmbedding method), 588 workx.algorithms.graphical), 490
is_directed_acyclic_graph() (in module net- is_reachable() (in module net-
workx.algorithms.dag), 424 workx.algorithms.tournament), 702
is_distance_regular() (in module net- is_regular() (in module net-
workx.algorithms.distance_regular), 439 workx.algorithms.regular), 615
is_dominating_set() (in module net- is_semiconnected() (in module net-
workx.algorithms.dominating), 445 workx.algorithms.components), 359
is_edge_cover() (in module net- is_semieulerian() (in module net-
workx.algorithms.covering), 403 workx.algorithms.euler), 451
is_empty() (in module networkx.classes.function), 759 is_simple_path() (in module net-
is_eulerian() (in module networkx.algorithms.euler), workx.algorithms.simple_paths), 681
448 is_strongly_connected() (in module net-
is_forest() (in module networkx.algorithms.components), 344
workx.algorithms.tree.recognition), 725 is_strongly_connected() (in module net-
is_frozen() (in module networkx.classes.function), workx.algorithms.tournament), 703
779 is_strongly_regular() (in module net-
is_graphical() (in module networkx.algorithms.distance_regular), 440
workx.algorithms.graphical), 488 is_threshold_graph() (in module net-
is_isolate() (in module networkx.algorithms.isolate), workx.algorithms.threshold), 700
494 is_tournament() (in module net-
is_isomorphic() (DiGraphMatcher method), 508 workx.algorithms.tournament), 704
is_isomorphic() (GraphMatcher method), 506 is_tree() (in module net-
is_isomorphic() (in module networkx.algorithms.tree.recognition), 725
workx.algorithms.isomorphism), 496 is_triad() (in module networkx.algorithms.triads), 751
is_isomorphic() (ISMAGS method), 520 is_valid_degree_sequence_erdos_gallai()
is_k_edge_connected() (in module net- (in module networkx.algorithms.graphical), 491
workx.algorithms.connectivity.edge_augmentation), is_valid_degree_sequence_havel_hakimi()
363 (in module networkx.algorithms.graphical), 490
is_k_regular() (in module net- is_valid_directed_joint_degree() (in mod-
workx.algorithms.regular), 616 ule networkx.generators.joint_degree_seq), 876
is_kl_connected() (in module net- is_valid_joint_degree() (in module net-
workx.algorithms.hybrid), 493 workx.generators.joint_degree_seq), 874
is_locally_k_edge_connected() is_weakly_connected() (in module net-
(in module networkx.algorithms.components), 350
workx.algorithms.connectivity.edge_augmentation), is_weighted() (in module networkx.classes.function),
363 771
is_matching() (in module net- ISMAGS (class in networkx.algorithms.isomorphism), 517
workx.algorithms.matching), 536 isolates() (in module networkx.algorithms.isolate),
is_maximal_matching() (in module net- 495
workx.algorithms.matching), 537 isomorphisms_iter() (DiGraphMatcher method),
is_multigraph() (PlanarEmbedding method), 588 508
is_multigraphical() (in module net- isomorphisms_iter() (GraphMatcher method), 506
workx.algorithms.graphical), 489 isomorphisms_iter() (ISMAGS method), 520
is_negatively_weighted() (in module net- items() (AdjacencyView method), 155
workx.classes.function), 772 items() (AtlasView method), 154
is_partition() (in module net- items() (FilterAdjacency method), 163
workx.algorithms.community.community_utils), items() (FilterAtlas method), 162
340 items() (FilterMultiAdjacency method), 164
Index 1099
items() (FilterMultiInner method), 164 karate_club_graph() (in module net-

items() (MultiAdjacencyView method), 156 workx.generators.social), 856
items() (UnionAdjacency method), 159 katz_centrality() (in module net-
items() (UnionAtlas method), 157 workx.algorithms.centrality), 260
items() (UnionMultiAdjacency method), 161 katz_centrality_numpy() (in module net-
items() (UnionMultiInner method), 160 workx.algorithms.centrality), 262
kernighan_lin_bisection() (in module net-
J workx.algorithms.community.kernighan_lin),
jaccard_coefficient() (in module net- 326
workx.algorithms.link_prediction), 526 keys() (AdjacencyView method), 155
johnson() (in module net- keys() (AtlasView method), 154
workx.algorithms.shortest_paths.weighted), keys() (FilterAdjacency method), 163
659 keys() (FilterAtlas method), 162
join() (in module networkx.algorithms.tree.operations), keys() (FilterMultiAdjacency method), 165
739 keys() (FilterMultiInner method), 164
joint_degree_graph() (in module net- keys() (MultiAdjacencyView method), 156
workx.generators.joint_degree_seq), 875 keys() (UnionAdjacency method), 159
junction_tree() (in module net- keys() (UnionAtlas method), 157
workx.algorithms.tree.decomposition), 746 keys() (UnionMultiAdjacency method), 161
keys() (UnionMultiInner method), 160
K kl_connected_subgraph() (in module net-
k_clique_communities() (in module networkx.algorithms.hybrid), 493
workx.algorithms.community.kclique), 327 kosaraju_strongly_connected_components()
k_components() (in module net- (in module networkx.algorithms.components),
workx.algorithms.approximation.kcomponents), 348
173 krackhardt_kite_graph() (in module net-
k_components() (in module networkx.generators.small), 805
workx.algorithms.connectivity.kcomponents),
370 L
k_core() (in module networkx.algorithms.core), 397 label_propagation_communities()
k_corona() (in module networkx.algorithms.core), 400 (in module net-
k_crust() (in module networkx.algorithms.core), 399 workx.algorithms.community.label_propagation),
k_edge_augmentation() (in module net- 332
workx.algorithms.connectivity.edge_augmentation), ladder_graph() (in module net-
361 workx.generators.classic), 790
k_edge_components() (EdgeComponentAuxGraph laplacian_matrix() (in module net-
method), 369 workx.linalg.laplacianmatrix), 887
k_edge_components() (in module net- laplacian_spectrum() (in module net-
workx.algorithms.connectivity.edge_kcomponents), workx.linalg.spectrum), 903
364 large_clique_size() (in module net-
k_edge_subgraphs() (EdgeComponentAuxGraph workx.algorithms.approximation.clique), 176
method), 370 largest_common_subgraph() (ISMAGS method),
k_edge_subgraphs() (in module net- 520
workx.algorithms.connectivity.edge_kcomponents), latapy_clustering() (in module net-
365 workx.algorithms.bipartite.cluster), 237
k_factor() (in module networkx.algorithms.regular), lattice_reference() (in module net-
616 workx.algorithms.smallworld), 685
k_random_intersection_graph() (in module LCF_graph() (in module networkx.generators.small),
networkx.generators.intersection), 855 799
k_shell() (in module networkx.algorithms.core), 398 les_miserables_graph() (in module net-
k_truss() (in module networkx.algorithms.core), 401 workx.generators.social), 857
kamada_kawai_layout() (in module net- lexicographic_product() (in module net-
workx.drawing.layout), 1010 workx.algorithms.operators.product), 566
1100 Index
lexicographical_topological_sort() (in match() (DiGraphMatcher method), 509

module networkx.algorithms.dag), 422 match() (GraphMatcher method), 506
LFR_benchmark_graph() (in module net- max_clique() (in module net-
workx.generators.community), 860 workx.algorithms.approximation.clique), 175
line_graph() (in module networkx.generators.line), max_flow_min_cost() (in module net-
849 workx.algorithms.flow), 480
literal_destringizer() (in module net- max_weight_clique() (in module net-
workx.readwrite.gml), 955 workx.algorithms.clique), 312
literal_stringizer() (in module net- max_weight_matching() (in module net-
workx.readwrite.gml), 956 workx.algorithms.matching), 538
load_centrality() (in module net- maximal_independent_set() (in module net-
workx.algorithms.centrality), 287 workx.algorithms.mis), 549
local_and_global_consistency() (in module maximal_matching() (in module net-
networkx.algorithms.node_classification), 553 workx.algorithms.matching), 538
local_bridges() (in module net- maximum_branching() (in module net-
workx.algorithms.bridges), 254 workx.algorithms.tree.branchings), 729
local_constraint() (in module net- maximum_flow() (in module net-
workx.algorithms.structuralholes), 691 workx.algorithms.flow), 453
local_edge_connectivity() (in module net- maximum_flow_value() (in module net-
workx.algorithms.connectivity.connectivity), workx.algorithms.flow), 455
380 maximum_independent_set() (in module net-
local_efficiency() (in module networkx.algorithms.approximation.clique), 174
workx.algorithms.efficiency_measures), 446 maximum_matching() (in module net-
local_node_connectivity() (in module networkx.algorithms.bipartite.matching), 224
workx.algorithms.approximation.connectivity), maximum_spanning_arborescence() (in module
170 networkx.algorithms.tree.branchings), 730
local_node_connectivity() (in module net- maximum_spanning_edges() (in module net-
workx.algorithms.connectivity.connectivity), workx.algorithms.tree.mst), 744
383 maximum_spanning_tree() (in module net-
local_reaching_centrality() (in module networkx.algorithms.tree.mst), 741
workx.algorithms.centrality), 293 metric_closure() (in module net-
lollipop_graph() (in module networkx.algorithms.approximation.steinertree),
workx.generators.classic), 790 182
louvain_communities() (in module net- min_cost_flow() (in module net-
workx.algorithms.community.louvain), 333 workx.algorithms.flow), 478
louvain_partitions() (in module net- min_cost_flow_cost() (in module net-
workx.algorithms.community.louvain), 334 workx.algorithms.flow), 476
lowest_common_ancestor() (in module net- min_edge_cover() (in module net-
workx.algorithms.lowest_common_ancestors), workx.algorithms.bipartite.covering), 249
535 min_edge_cover() (in module net-
lukes_partitioning() (in module networkx.algorithms.covering), 402
workx.algorithms.community.lukes), 330 min_edge_dominating_set() (in module net-
workx.algorithms.approximation.dominating_set),
M 180
make_clique_bipartite() (in module net- min_maximal_matching() (in module net-
workx.algorithms.clique), 310 workx.algorithms.approximation.matching),
make_list_of_ints() (in module net- 180
workx.utils.misc), 1025 min_weight_matching() (in module net-
make_max_clique_graph() (in module networkx.algorithms.matching), 539
workx.algorithms.clique), 309 min_weighted_dominating_set()
MappedQueue (class in networkx.utils.mapped_queue), (in module net-
1044 workx.algorithms.approximation.dominating_set),
margulis_gabber_galil_graph() (in module 179
networkx.generators.expanders), 793 min_weighted_vertex_cover() (in module net-
Index 1101
workx.algorithms.approximation.vertex_cover), 195
194 networkx.algorithms.approximation.ramsey,
minimum_branching() (in module net- 181
workx.algorithms.tree.branchings), 730 networkx.algorithms.approximation.steinertree,
minimum_cut() (in module networkx.algorithms.flow), 181
457 networkx.algorithms.approximation.traveling_sale
minimum_cut_value() (in module net- 183
workx.algorithms.flow), 459 networkx.algorithms.approximation.treewidth,
minimum_cycle_basis() (in module net- 193
workx.algorithms.cycles), 408 networkx.algorithms.approximation.vertex_cover,
minimum_edge_cut() (in module net- 194
workx.algorithms.connectivity.cuts), 387 networkx.algorithms.assortativity,
minimum_node_cut() (in module net- 196
workx.algorithms.connectivity.cuts), 388 networkx.algorithms.asteroidal, 208
minimum_spanning_arborescence() (in module networkx.algorithms.bipartite, 210
networkx.algorithms.tree.branchings), 731 networkx.algorithms.bipartite.basic,
minimum_spanning_edges() (in module net- 211
workx.algorithms.tree.mst), 742 networkx.algorithms.bipartite.centrality,
minimum_spanning_tree() (in module net- 240
workx.algorithms.tree.mst), 740 networkx.algorithms.bipartite.cluster,
minimum_st_edge_cut() (in module net- 235
workx.algorithms.connectivity.cuts), 390 networkx.algorithms.bipartite.covering,
minimum_st_node_cut() (in module net- 249
workx.algorithms.connectivity.cuts), 392 networkx.algorithms.bipartite.edgelist,
minimum_weight_full_matching() (in module 215
networkx.algorithms.bipartite.matching), 224 networkx.algorithms.bipartite.generators,
mixing_dict() (in module net- 243
workx.algorithms.assortativity), 206 networkx.algorithms.bipartite.matching,
mixing_expansion() (in module net- 221
workx.algorithms.cuts), 412 networkx.algorithms.bipartite.matrix,
modularity() (in module net- 225
workx.algorithms.community.quality), 336 networkx.algorithms.bipartite.projection,
modularity_matrix() (in module net- 227
workx.linalg.modularitymatrix), 900 networkx.algorithms.bipartite.redundancy,
modularity_spectrum() (in module net- 239
workx.linalg.spectrum), 905 networkx.algorithms.bipartite.spectral,
module 234
networkx.algorithms.approximation, networkx.algorithms.boundary, 250
169 networkx.algorithms.bridges, 252
networkx.algorithms.approximation.clique,networkx.algorithms.centrality, 255
174 networkx.algorithms.chains, 300
networkx.algorithms.approximation.clustering_coefficient,
networkx.algorithms.chordal, 301
177 networkx.algorithms.clique, 306
networkx.algorithms.approximation.connectivity, networkx.algorithms.cluster, 313
169 networkx.algorithms.coloring, 319
networkx.algorithms.approximation.distance_measures, networkx.algorithms.communicability_alg,
178 323
networkx.algorithms.approximation.dominating_set, networkx.algorithms.community, 325
179 networkx.algorithms.community.asyn_fluid,
networkx.algorithms.approximation.kcomponents, 335
172 networkx.algorithms.community.centrality,
networkx.algorithms.approximation.matching, 338
180 networkx.algorithms.community.community_utils,
networkx.algorithms.approximation.maxcut, 340
1102 Index
networkx.algorithms.community.kclique, networkx.algorithms.graphical, 487

326 networkx.algorithms.hierarchy, 492
networkx.algorithms.community.kernighan_lin,
networkx.algorithms.hybrid, 492
325 networkx.algorithms.isolate, 494
networkx.algorithms.community.label_propagation,
networkx.algorithms.isomorphism, 496
331 networkx.algorithms.isomorphism.ismags,
networkx.algorithms.community.louvain, 515
332 networkx.algorithms.isomorphism.isomorphvf2,
networkx.algorithms.community.lukes, 503
330 networkx.algorithms.isomorphism.tree_isomorphism
networkx.algorithms.community.modularity_max,
501
327 networkx.algorithms.isomorphism.vf2pp,
networkx.algorithms.community.quality, 499
336 networkx.algorithms.link_analysis.hits_alg,
networkx.algorithms.components, 341 524
networkx.algorithms.connectivity, networkx.algorithms.link_analysis.pagerank_alg,
360 521
networkx.algorithms.connectivity.connectivity,
networkx.algorithms.link_prediction,
377 525
networkx.algorithms.connectivity.cuts, networkx.algorithms.lowest_common_ancestors,
386 533
networkx.algorithms.connectivity.disjoint_paths,
networkx.algorithms.matching, 536
373 networkx.algorithms.minors, 540
networkx.algorithms.connectivity.edge_augmentation,
networkx.algorithms.mis, 548
360 networkx.algorithms.moral, 551
networkx.algorithms.connectivity.edge_kcomponents,
networkx.algorithms.node_classification,
364 552
networkx.algorithms.connectivity.kcomponents,
networkx.algorithms.non_randomness,
370 550
networkx.algorithms.connectivity.kcutsets,
networkx.algorithms.operators.all,
372 562
networkx.algorithms.connectivity.stoerwagner,
networkx.algorithms.operators.binary,
394 555
networkx.algorithms.connectivity.utils, networkx.algorithms.operators.product,
395 565
networkx.algorithms.core, 396 networkx.algorithms.operators.unary,
networkx.algorithms.covering, 402 554
networkx.algorithms.cuts, 409 networkx.algorithms.planar_drawing,
networkx.algorithms.cycles, 404 610
networkx.algorithms.d_separation, networkx.algorithms.planarity, 571
414 networkx.algorithms.polynomials, 611
networkx.algorithms.dag, 417 networkx.algorithms.reciprocity, 614
networkx.algorithms.distance_measures, networkx.algorithms.regular, 615
432 networkx.algorithms.richclub, 617
networkx.algorithms.distance_regular, networkx.algorithms.shortest_paths.astar,
439 663
networkx.algorithms.dominance, 442 networkx.algorithms.shortest_paths.dense,
networkx.algorithms.dominating, 444 660
networkx.algorithms.efficiency_measures, networkx.algorithms.shortest_paths.generic,
445 618
networkx.algorithms.euler, 448 networkx.algorithms.shortest_paths.unweighted,
networkx.algorithms.flow, 453 624
networkx.algorithms.graph_hashing, networkx.algorithms.shortest_paths.weighted,
484 630
Index 1103
networkx.algorithms.similarity, 665 networkx.generators.directed, 835

networkx.algorithms.simple_paths, networkx.generators.duplication, 825
677 networkx.generators.ego, 852
networkx.algorithms.smallworld, 684 networkx.generators.expanders, 793
networkx.algorithms.smetric, 687 networkx.generators.geometric, 839
networkx.algorithms.sparsifiers, 688 networkx.generators.harary_graph,
networkx.algorithms.structuralholes, 879
689 networkx.generators.internet_as_graphs,
networkx.algorithms.summarization, 853
692 networkx.generators.intersection,
networkx.algorithms.swap, 696 854
networkx.algorithms.threshold, 699 networkx.generators.interval_graph,
networkx.algorithms.tournament, 701 882
networkx.algorithms.traversal.beamsearch,networkx.generators.joint_degree_seq,
719 874
networkx.algorithms.traversal.breadth_first_search,
networkx.generators.lattice, 795
712 networkx.generators.line, 849
networkx.algorithms.traversal.depth_first_search,
networkx.generators.mycielski, 878
705 networkx.generators.nonisomorphic_trees,
networkx.algorithms.traversal.edgebfs, 872
722 networkx.generators.random_clustered,
networkx.algorithms.traversal.edgedfs, 833
720 networkx.generators.random_graphs,
networkx.algorithms.tree.branchings, 810
728 networkx.generators.small, 798
networkx.algorithms.tree.coding, 734 networkx.generators.social, 856
networkx.algorithms.tree.decomposition, networkx.generators.spectral_graph_forge,
746 868
networkx.algorithms.tree.mst, 739 networkx.generators.stochastic, 853
networkx.algorithms.tree.operations, networkx.generators.sudoku, 883
738 networkx.generators.trees, 869
networkx.algorithms.tree.recognition, networkx.generators.triads, 873
723 networkx.linalg.algebraicconnectivity,
networkx.algorithms.triads, 747 892
networkx.algorithms.vitality, 753 networkx.linalg.attrmatrix, 896
networkx.algorithms.voronoi, 754 networkx.linalg.bethehessianmatrix,
networkx.algorithms.wiener, 755 891
networkx.classes.backends, 167 networkx.linalg.graphmatrix, 885
networkx.classes.coreviews, 152 networkx.linalg.laplacianmatrix, 887
networkx.classes.filters, 165 networkx.linalg.modularitymatrix,
networkx.classes.function, 757 900
networkx.classes.graphviews, 150 networkx.linalg.spectrum, 902
networkx.convert, 907 networkx.readwrite.adjlist, 931
networkx.convert_matrix, 912 networkx.readwrite.edgelist, 940
networkx.drawing.layout, 1008 networkx.readwrite.gexf, 947
networkx.drawing.nx_agraph, 1001 networkx.readwrite.gml, 950
networkx.drawing.nx_pydot, 1005 networkx.readwrite.graph6, 971
networkx.drawing.nx_pylab, 985 networkx.readwrite.graphml, 956
networkx.exception, 1021 networkx.readwrite.json_graph, 961
networkx.generators.atlas, 781 networkx.readwrite.leda, 970
networkx.generators.classic, 783 networkx.readwrite.multiline_adjlist,
networkx.generators.cographs, 881 935
networkx.generators.community, 857 networkx.readwrite.pajek, 979
networkx.generators.degree_seq, 827 networkx.readwrite.sparse6, 975
1104 Index
networkx.relabel, 927 neighbors() (MultiGraph method), 97

networkx.utils, 1023 neighbors() (PlanarEmbedding method), 589
networkx.utils.decorators, 1030 neighbors_cw_order() (PlanarEmbedding
networkx.utils.mapped_queue, 1044 method), 590
networkx.utils.misc, 1023 network_simplex() (in module net-
networkx.utils.random_sequence, 1028 workx.algorithms.flow), 474
networkx.utils.rcm, 1042 NetworkXAlgorithmError (class in networkx), 1021
networkx.utils.union_find, 1027 networkx.algorithms.approximation
moebius_kantor_graph() (in module net- module, 169
workx.generators.small), 806 networkx.algorithms.approximation.clique
moral_graph() (in module net- module, 174
workx.algorithms.moral), 551 networkx.algorithms.approximation.clustering_coeffi
multi_source_dijkstra() (in module net- module, 177
workx.algorithms.shortest_paths.weighted), networkx.algorithms.approximation.connectivity
639 module, 169
multi_source_dijkstra_path() (in module net- networkx.algorithms.approximation.distance_measures
workx.algorithms.shortest_paths.weighted), 641 module, 178
multi_source_dijkstra_path_length() networkx.algorithms.approximation.dominating_set
(in module net- module, 179
workx.algorithms.shortest_paths.weighted), networkx.algorithms.approximation.kcomponents
642 module, 172
MultiAdjacencyView (class in net- networkx.algorithms.approximation.matching
workx.classes.coreviews), 155 module, 180
MultiDiGraph (class in networkx), 109 networkx.algorithms.approximation.maxcut
MultiGraph (class in networkx), 74 module, 195
multipartite_layout() (in module net- networkx.algorithms.approximation.ramsey
workx.drawing.layout), 1017 module, 181
mycielski_graph() (in module net- networkx.algorithms.approximation.steinertree
workx.generators.mycielski), 878 module, 181
mycielskian() (in module net- networkx.algorithms.approximation.traveling_salesma
workx.generators.mycielski), 878 module, 183
networkx.algorithms.approximation.treewidth
N module, 193
naive_greedy_modularity_communities() networkx.algorithms.approximation.vertex_cover
(in module net- module, 194
workx.algorithms.community.modularity_max), networkx.algorithms.assortativity
329 module, 196
name (PlanarEmbedding property), 606 networkx.algorithms.asteroidal
navigable_small_world_graph() (in module module, 208
networkx.generators.geometric), 842 networkx.algorithms.bipartite
nbunch, 1047 module, 210
nbunch_iter() (DiGraph method), 63 networkx.algorithms.bipartite.basic
nbunch_iter() (Graph method), 28 module, 211
nbunch_iter() (MultiDiGraph method), 138 networkx.algorithms.bipartite.centrality
nbunch_iter() (MultiGraph method), 100 module, 240
nbunch_iter() (PlanarEmbedding method), 588 networkx.algorithms.bipartite.cluster
negative_edge_cycle() (in module net- module, 235
workx.algorithms.shortest_paths.weighted), networkx.algorithms.bipartite.covering
656 module, 249
neighbors() (DiGraph method), 59 networkx.algorithms.bipartite.edgelist
neighbors() (Graph method), 26 module, 215
neighbors() (in module networkx.classes.function), networkx.algorithms.bipartite.generators
767 module, 243
neighbors() (MultiDiGraph method), 135 networkx.algorithms.bipartite.matching
Index 1105
module, 221 module, 377

networkx.algorithms.bipartite.matrix networkx.algorithms.connectivity.cuts
networkx.algorithms.bipartite.projectionnetworkx.algorithms.connectivity.disjoint_paths
networkx.algorithms.bipartite.redundancynetworkx.algorithms.connectivity.edge_augmentation
networkx.algorithms.bipartite.spectral networkx.algorithms.connectivity.edge_kcomponents
networkx.algorithms.boundary networkx.algorithms.connectivity.kcomponents
networkx.algorithms.bridges networkx.algorithms.connectivity.kcutsets
networkx.algorithms.centrality networkx.algorithms.connectivity.stoerwagner
networkx.algorithms.chains networkx.algorithms.connectivity.utils
networkx.algorithms.chordal networkx.algorithms.core
networkx.algorithms.clique networkx.algorithms.covering
networkx.algorithms.cluster networkx.algorithms.cuts
networkx.algorithms.coloring networkx.algorithms.cycles
networkx.algorithms.communicability_alg networkx.algorithms.d_separation
networkx.algorithms.community networkx.algorithms.dag
networkx.algorithms.community.asyn_fluidnetworkx.algorithms.distance_measures
networkx.algorithms.community.centralitynetworkx.algorithms.distance_regular
networkx.algorithms.community.community_utils
networkx.algorithms.dominance
networkx.algorithms.community.kclique networkx.algorithms.dominating
networkx.algorithms.community.kernighan_lin
networkx.algorithms.efficiency_measures
networkx.algorithms.community.label_propagation
networkx.algorithms.euler
networkx.algorithms.community.louvain networkx.algorithms.flow
networkx.algorithms.community.lukes networkx.algorithms.graph_hashing
networkx.algorithms.community.modularity_max
networkx.algorithms.graphical
networkx.algorithms.community.quality networkx.algorithms.hierarchy
networkx.algorithms.components networkx.algorithms.hybrid
networkx.algorithms.connectivity networkx.algorithms.isolate
networkx.algorithms.connectivity.connectivity
networkx.algorithms.isomorphism
1106 Index

networkx.algorithms.isomorphism.ismags networkx.algorithms.shortest_paths.unweighted
networkx.algorithms.isomorphism.isomorphvf2
networkx.algorithms.shortest_paths.weighted
networkx.algorithms.isomorphism.tree_isomorphism
networkx.algorithms.similarity
networkx.algorithms.isomorphism.vf2pp networkx.algorithms.simple_paths
networkx.algorithms.link_analysis.hits_alg
networkx.algorithms.smallworld
networkx.algorithms.link_analysis.pagerank_alg
networkx.algorithms.smetric
networkx.algorithms.link_prediction networkx.algorithms.sparsifiers
networkx.algorithms.lowest_common_ancestors
networkx.algorithms.structuralholes
networkx.algorithms.matching networkx.algorithms.summarization
networkx.algorithms.minors networkx.algorithms.swap
networkx.algorithms.mis networkx.algorithms.threshold
networkx.algorithms.moral networkx.algorithms.tournament
networkx.algorithms.node_classification networkx.algorithms.traversal.beamsearch
networkx.algorithms.non_randomness networkx.algorithms.traversal.breadth_first_search
networkx.algorithms.operators.all networkx.algorithms.traversal.depth_first_search
networkx.algorithms.operators.binary networkx.algorithms.traversal.edgebfs
networkx.algorithms.operators.product networkx.algorithms.traversal.edgedfs
networkx.algorithms.operators.unary networkx.algorithms.tree.branchings
networkx.algorithms.planar_drawing networkx.algorithms.tree.coding
networkx.algorithms.planarity networkx.algorithms.tree.decomposition
networkx.algorithms.polynomials networkx.algorithms.tree.mst
networkx.algorithms.reciprocity networkx.algorithms.tree.operations
networkx.algorithms.regular networkx.algorithms.tree.recognition
networkx.algorithms.richclub networkx.algorithms.triads
networkx.algorithms.shortest_paths.astarnetworkx.algorithms.vitality
networkx.algorithms.shortest_paths.densenetworkx.algorithms.voronoi
networkx.algorithms.shortest_paths.generic
networkx.algorithms.wiener
Index 1107

networkx.classes.backends networkx.generators.joint_degree_seq
networkx.classes.coreviews networkx.generators.lattice
networkx.classes.filters networkx.generators.line
networkx.classes.function networkx.generators.mycielski
networkx.classes.graphviews networkx.generators.nonisomorphic_trees
networkx.convert networkx.generators.random_clustered
networkx.convert_matrix networkx.generators.random_graphs
networkx.drawing.layout networkx.generators.small
networkx.drawing.nx_agraph networkx.generators.social
networkx.drawing.nx_pydot networkx.generators.spectral_graph_forge
networkx.drawing.nx_pylab networkx.generators.stochastic
NetworkXError (class in networkx), 1021 networkx.generators.sudoku
networkx.exception module, 883
module, 1021 networkx.generators.trees
NetworkXException (class in networkx), 1021 module, 869
networkx.generators.atlas networkx.generators.triads
networkx.generators.classic networkx.linalg.algebraicconnectivity
networkx.generators.cographs networkx.linalg.attrmatrix
networkx.generators.community networkx.linalg.bethehessianmatrix
networkx.generators.degree_seq networkx.linalg.graphmatrix
networkx.generators.directed networkx.linalg.laplacianmatrix
networkx.generators.duplication networkx.linalg.modularitymatrix
networkx.generators.ego networkx.linalg.spectrum
networkx.generators.expanders NetworkXNoCycle (class in networkx), 1021
module, 793 NetworkXNoPath (class in networkx), 1021
networkx.generators.geometric NetworkXNotImplemented (class in networkx), 1021
module, 839 NetworkXPointlessConcept (class in networkx),
networkx.generators.harary_graph 1021
module, 879 networkx.readwrite.adjlist
networkx.generators.internet_as_graphs module, 931
module, 853 networkx.readwrite.edgelist
networkx.generators.intersection module, 940
module, 854 networkx.readwrite.gexf
networkx.generators.interval_graph module, 947
1108 Index
networkx.readwrite.gml node_connectivity() (in module net-

module, 950 workx.algorithms.connectivity.connectivity),
networkx.readwrite.graph6 385
module, 971 node_degree_xy() (in module net-
networkx.readwrite.graphml workx.algorithms.assortativity), 207
module, 956 node_disjoint_paths() (in module net-
networkx.readwrite.json_graph workx.algorithms.connectivity.disjoint_paths),
module, 961 375
networkx.readwrite.leda node_expansion() (in module net-
module, 970 workx.algorithms.cuts), 412
networkx.readwrite.multiline_adjlist node_link_data() (in module net-
module, 935 workx.readwrite.json_graph), 961
networkx.readwrite.pajek node_link_graph() (in module net-
module, 979 workx.readwrite.json_graph), 963
networkx.readwrite.sparse6 node_redundancy() (in module net-
module, 975 workx.algorithms.bipartite.redundancy), 239
networkx.relabel NodeNotFound (class in networkx), 1021
module, 927 nodes (DiGraph property), 52
NetworkXUnbounded (class in networkx), 1021 nodes (Graph property), 21
NetworkXUnfeasible (class in networkx), 1021 nodes (MultiDiGraph property), 126
networkx.utils nodes (MultiGraph property), 91
module, 1023 nodes (PlanarEmbedding property), 606
networkx.utils.decorators nodes() (in module networkx.classes.function), 766
module, 1030 nodes_equal() (in module networkx.utils.misc), 1026
networkx.utils.mapped_queue nodes_or_number() (in module net-
module, 1044 workx.utils.decorators), 1032
networkx.utils.misc nodes_with_selfloops() (in module net-
module, 1023 workx.classes.function), 770
networkx.utils.random_sequence non_edges() (in module networkx.classes.function),
module, 1028 768
networkx.utils.rcm non_neighbors() (in module net-
module, 1042 workx.classes.function), 767
networkx.utils.union_find non_randomness() (in module net-
module, 1027 workx.algorithms.non_randomness), 550
new_edge_key() (MultiDiGraph method), 121 nonisomorphic_trees() (in module net-
new_edge_key() (MultiGraph method), 86 workx.generators.nonisomorphic_trees), 872
newman_watts_strogatz_graph() (in module normalized_cut_size() (in module net-
networkx.generators.random_graphs), 816 workx.algorithms.cuts), 413
next_face_half_edge() (PlanarEmbedding normalized_laplacian_matrix() (in module
method), 590 networkx.linalg.laplacianmatrix), 888
no_filter() (in module networkx.classes.filters), 166 normalized_laplacian_spectrum() (in module
node, 1047 networkx.linalg.spectrum), 904
node attribute, 1047 not_implemented_for() (in module net-
node_attribute_xy() (in module networkx.utils.decorators), 1031
workx.algorithms.assortativity), 207 NotATree, 747
node_boundary() (in module net- np_random_state() (in module net-
workx.algorithms.boundary), 251 workx.utils.decorators), 1033
node_clique_number() (in module net- null_graph() (in module networkx.generators.classic),
workx.algorithms.clique), 311 791
node_connected_component() (in module net- number_attracting_components() (in module
workx.algorithms.components), 343 networkx.algorithms.components), 353
node_connectivity() (in module net- number_connected_components() (in module
workx.algorithms.approximation.connectivity), networkx.algorithms.components), 342
171 number_of_cliques() (in module net-
Index 1109
workx.algorithms.clique), 312 optimize_graph_edit_distance() (in module

number_of_edges() (DiGraph method), 68 networkx.algorithms.similarity), 670
number_of_edges() (Graph method), 32 order() (DiGraph method), 64
number_of_edges() (in module net- order() (Graph method), 29
workx.classes.function), 768 order() (MultiDiGraph method), 139
number_of_edges() (MultiDiGraph method), 144 order() (MultiGraph method), 101
number_of_edges() (MultiGraph method), 104 order() (PlanarEmbedding method), 592
number_of_edges() (PlanarEmbedding method), out_degree (DiGraph property), 67
590 out_degree (MultiDiGraph property), 142
number_of_isolates() (in module net- out_degree (PlanarEmbedding property), 608
workx.algorithms.isolate), 496 out_degree_centrality() (in module net-
number_of_nodes() (DiGraph method), 64 workx.algorithms.centrality), 257
number_of_nodes() (Graph method), 30 out_edges (DiGraph property), 56
number_of_nodes() (in module net- out_edges (MultiDiGraph property), 130
workx.classes.function), 766 out_edges (PlanarEmbedding property), 608
number_of_nodes() (MultiDiGraph method), 139 overall_reciprocity() (in module net-
number_of_nodes() (MultiGraph method), 101 workx.algorithms.reciprocity), 615
number_of_nodes() (PlanarEmbedding method), overlap_weighted_projected_graph() (in
591 module networkx.algorithms.bipartite.projection),
number_of_nonisomorphic_trees() (in module 231
networkx.generators.nonisomorphic_trees), 873
number_of_selfloops() (in module net- P
workx.classes.function), 770 pagerank() (in module net-
number_strongly_connected_components() workx.algorithms.link_analysis.pagerank_alg),
(in module networkx.algorithms.components), 521
345 pairwise() (in module networkx.utils.misc), 1025
number_weakly_connected_components() (in paley_graph() (in module net-
module networkx.algorithms.components), 351 workx.generators.expanders), 794
numeric_assortativity_coefficient() (in panther_similarity() (in module net-
module networkx.algorithms.assortativity), 198 workx.algorithms.similarity), 675
numerical_edge_match() (in module net- pappus_graph() (in module net-
workx.algorithms.isomorphism), 512 workx.generators.small), 807
numerical_multiedge_match() (in module net- parse_adjlist() (in module net-
workx.algorithms.isomorphism), 513 workx.readwrite.adjlist), 934
numerical_node_match() (in module net- parse_edgelist() (in module net-
workx.algorithms.isomorphism), 511 workx.algorithms.bipartite.edgelist), 218
parse_edgelist() (in module net-
O workx.readwrite.edgelist), 945
octahedral_graph() (in module net- parse_gml() (in module networkx.readwrite.gml), 953
workx.generators.small), 806 parse_graphml() (in module net-
omega() (in module networkx.algorithms.smallworld), workx.readwrite.graphml), 960
686 parse_leda() (in module networkx.readwrite.leda),
one_exchange() (in module net- 971
workx.algorithms.approximation.maxcut), parse_multiline_adjlist() (in module net-
196 workx.readwrite.multiline_adjlist), 938
onion_layers() (in module net- parse_pajek() (in module networkx.readwrite.pajek),
workx.algorithms.core), 401 980
open_file() (in module networkx.utils.decorators), partial_duplication_graph() (in module net-
1030 workx.generators.duplication), 826
optimal_edit_paths() (in module net- partition_quality() (in module net-
workx.algorithms.similarity), 668 workx.algorithms.community.quality), 337
optimize_edit_paths() (in module net- path_graph() (in module networkx.generators.classic),
workx.algorithms.similarity), 672 791
1110 Index
path_weight() (in module networkx.classes.function), R

777 ra_index_soundarajan_hopcroft() (in module
percolation_centrality() (in module net- networkx.algorithms.link_prediction), 530
workx.algorithms.centrality), 295 radius() (in module net-
periphery() (in module networkx.algorithms.distance_measures), 437
workx.algorithms.distance_measures), 436 ramsey_R2() (in module net-
petersen_graph() (in module networkx.algorithms.approximation.ramsey),
workx.generators.small), 807 181
planar_layout() (in module net- random_clustered_graph() (in module net-
workx.drawing.layout), 1011 workx.generators.random_clustered), 833
PlanarEmbedding (class in net- random_cograph() (in module net-
workx.algorithms.planarity), 573 workx.generators.cographs), 881
planted_partition_graph() (in module net- random_degree_sequence_graph() (in module
workx.generators.community), 863 networkx.generators.degree_seq), 832
pop() (MappedQueue method), 1046 random_geometric_graph() (in module net-
power() (in module networkx.generators.geometric), 843
workx.algorithms.operators.product), 569 random_graph() (in module net-
PowerIterationFailedConvergence (class in workx.algorithms.bipartite.generators), 247
networkx), 1022 random_internet_as_graph() (in module net-
powerlaw_cluster_graph() (in module networkx.generators.internet_as_graphs), 853
workx.generators.random_graphs), 821 random_k_out_graph() (in module net-
powerlaw_sequence() (in module networkx.generators.directed), 837
workx.utils.random_sequence), 1028 random_kernel_graph() (in module net-
pred (DiGraph property), 62 workx.generators.random_graphs), 822
pred (PlanarEmbedding property), 609 random_layout() (in module net-
predecessor() (in module networkx.drawing.layout), 1011
workx.algorithms.shortest_paths.unweighted), random_lobster() (in module net-
629 workx.generators.random_graphs), 823
predecessors() (DiGraph method), 62 random_partition_graph() (in module net-
predecessors() (MultiDiGraph method), 137 workx.generators.community), 864
predecessors() (PlanarEmbedding method), 592 random_powerlaw_tree() (in module net-
preferential_attachment() (in module networkx.generators.random_graphs), 824
workx.algorithms.link_prediction), 528 random_powerlaw_tree_sequence() (in module
preferential_attachment_graph() (in module networkx.generators.random_graphs), 824
networkx.algorithms.bipartite.generators), 247 random_reference() (in module net-
prefix_tree() (in module networkx.generators.trees), workx.algorithms.smallworld), 684
871 random_regular_graph() (in module net-
preflow_push() (in module networkx.generators.random_graphs), 818
workx.algorithms.flow), 465 random_shell_graph() (in module net-
projected_graph() (in module networkx.generators.random_graphs), 823
workx.algorithms.bipartite.projection), 227 random_spanning_tree() (in module net-
prominent_group() (in module networkx.algorithms.tree.mst), 742
workx.algorithms.centrality), 285 random_tournament() (in module net-
push() (MappedQueue method), 1046 workx.algorithms.tournament), 704
py_random_state() (in module net- random_tree() (in module networkx.generators.trees),
workx.utils.decorators), 1034 870
pydot_layout() (in module net- random_triad() (in module net-
workx.drawing.nx_pydot), 1007 workx.algorithms.triads), 749
pygraphviz_layout() (in module net- random_weighted_sample() (in module net-
workx.drawing.nx_agraph), 1004 workx.utils.random_sequence), 1029
randomized_partitioning() (in module net-
Q workx.algorithms.approximation.maxcut), 195
quotient_graph() (in module net- read_adjlist() (in module net-
workx.algorithms.minors), 545 workx.readwrite.adjlist), 931
Index 1111
read_dot() (in module networkx.drawing.nx_agraph), remove_nodes_from() (MultiGraph method), 82

1003 remove_nodes_from() (PlanarEmbedding method),
read_dot() (in module networkx.drawing.nx_pydot), 594
1006 rescale_layout() (in module net-
read_edgelist() (in module networkx.drawing.layout), 1012
workx.algorithms.bipartite.edgelist), 219 rescale_layout_dict() (in module net-
read_edgelist() (in module networkx.drawing.layout), 1013
workx.readwrite.edgelist), 940 resistance_distance() (in module net-
read_gexf() (in module networkx.readwrite.gexf), 947 workx.algorithms.distance_measures), 438
read_gml() (in module networkx.readwrite.gml), 951 resource_allocation_index() (in module net-
read_graph6() (in module networkx.algorithms.link_prediction), 525
workx.readwrite.graph6), 972 restricted_view() (in module net-
read_graphml() (in module networkx.classes.function), 764
workx.readwrite.graphml), 957 reverse() (DiGraph method), 74
read_leda() (in module networkx.readwrite.leda), 970 reverse() (in module net-
read_multiline_adjlist() (in module networkx.algorithms.operators.unary), 555
workx.readwrite.multiline_adjlist), 936 reverse() (MultiDiGraph method), 150
read_pajek() (in module networkx.readwrite.pajek), reverse() (PlanarEmbedding method), 595
979 reverse_cuthill_mckee_ordering() (in mod-
read_sparse6() (in module net- ule networkx.utils.rcm), 1043
workx.readwrite.sparse6), 976 reverse_havel_hakimi_graph() (in module net-
read_weighted_edgelist() (in module networkx.algorithms.bipartite.generators), 246
workx.readwrite.edgelist), 943 reverse_view() (in module net-
reciprocity() (in module networkx.classes.function), 765
workx.algorithms.reciprocity), 614 reverse_view() (in module net-
reconstruct_path() (in module networkx.classes.graphviews), 152
workx.algorithms.shortest_paths.dense), 663 rich_club_coefficient() (in module net-
recursive_simple_cycles() (in module networkx.algorithms.richclub), 617
workx.algorithms.cycles), 406 ring_of_cliques() (in module net-
relabel_gexf_graph() (in module networkx.generators.community), 865
workx.readwrite.gexf), 950 robins_alexander_clustering() (in module
relabel_nodes() (in module networkx.relabel), 928 networkx.algorithms.bipartite.cluster), 238
relaxed_caveman_graph() (in module net- rooted_product() (in module net-
workx.generators.community), 865 workx.algorithms.operators.product), 567
remove() (MappedQueue method), 1046 rooted_tree_isomorphism() (in module net-
remove_edge() (DiGraph method), 48 workx.algorithms.isomorphism.tree_isomorphism),
remove_edge() (Graph method), 17 501
remove_edge() (MultiDiGraph method), 121
remove_edge() (MultiGraph method), 86 S
remove_edge() (PlanarEmbedding method), 592 s_metric() (in module networkx.algorithms.smetric),
remove_edges_from() (DiGraph method), 49 687
remove_edges_from() (Graph method), 18 scale_free_graph() (in module net-
remove_edges_from() (MultiDiGraph method), 122 workx.generators.directed), 838
remove_edges_from() (MultiGraph method), 87 score_sequence() (in module net-
remove_edges_from() (PlanarEmbedding method), workx.algorithms.tournament), 705
593 second_order_centrality() (in module net-
remove_node() (DiGraph method), 44 workx.algorithms.centrality), 296
remove_node() (Graph method), 14 sedgewick_maze_graph() (in module net-
remove_node() (MultiDiGraph method), 116 workx.generators.small), 808
remove_node() (MultiGraph method), 81 selfloop_edges() (in module net-
remove_node() (PlanarEmbedding method), 594 workx.classes.function), 769
remove_nodes_from() (DiGraph method), 45 semantic_feasibility() (DiGraphMatcher
remove_nodes_from() (Graph method), 14 method), 509
remove_nodes_from() (MultiDiGraph method), 117
1112 Index
semantic_feasibility() (GraphMatcher 637

method), 507 single_source_dijkstra_path_length()
set_data() (PlanarEmbedding method), 595 (in module net-
set_edge_attributes() (in module networkx.algorithms.shortest_paths.weighted),
workx.classes.function), 774 638
set_node_attributes() (in module net- single_source_shortest_path() (in module
workx.classes.function), 772 networkx.algorithms.shortest_paths.unweighted),
sets() (in module networkx.algorithms.bipartite.basic), 624
212 single_source_shortest_path_length()
shell_layout() (in module net- (in module net-
workx.drawing.layout), 1013 workx.algorithms.shortest_paths.unweighted),
shortest_augmenting_path() (in module net- 625
workx.algorithms.flow), 463 single_target_shortest_path() (in module
shortest_path() (in module net- networkx.algorithms.shortest_paths.unweighted),
workx.algorithms.shortest_paths.generic), 626
618 single_target_shortest_path_length()
shortest_path_length() (in module net- (in module net-
workx.algorithms.shortest_paths.generic), workx.algorithms.shortest_paths.unweighted),
621 626
shortest_simple_paths() (in module net- size() (DiGraph method), 67
workx.algorithms.simple_paths), 682 size() (Graph method), 31
show_diedges() (in module networkx.classes.filters), size() (MultiDiGraph method), 143
167 size() (MultiGraph method), 103
show_edges() (in module networkx.classes.filters), 166 size() (PlanarEmbedding method), 595
show_multidiedges() (in module net- snap_aggregation() (in module net-
workx.classes.filters), 167 workx.algorithms.summarization), 694
show_multiedges() (in module net- soft_random_geometric_graph() (in module
workx.classes.filters), 167 networkx.generators.geometric), 844
show_nodes (class in networkx.classes.filters), 166 spanner() (in module networkx.algorithms.sparsifiers),
sigma() (in module networkx.algorithms.smallworld), 688
685 SpanningTreeIterator (class in net-
signature() (argmap class method), 1042 workx.algorithms.tree.mst), 745
simple_cycles() (in module net- spectral_bipartivity() (in module net-
workx.algorithms.cycles), 405 workx.algorithms.bipartite.spectral), 234
simrank_similarity() (in module net- spectral_graph_forge() (in module net-
workx.algorithms.similarity), 674 workx.generators.spectral_graph_forge), 868
simulated_annealing_tsp() (in module net- spectral_layout() (in module net-
workx.algorithms.approximation.traveling_salesman), workx.drawing.layout), 1015
187 spectral_ordering() (in module net-
single_source_bellman_ford() (in module networkx.linalg.algebraicconnectivity), 895
workx.algorithms.shortest_paths.weighted), 649 spiral_layout() (in module net-
single_source_bellman_ford_path() workx.drawing.layout), 1016
(in module net- spring_layout() (in module net-
workx.algorithms.shortest_paths.weighted), workx.drawing.layout), 1014
650 square_clustering() (in module net-
single_source_bellman_ford_path_length() workx.algorithms.cluster), 317
(in module net- star_graph() (in module networkx.generators.classic),
workx.algorithms.shortest_paths.weighted), 791
651 steiner_tree() (in module net-
single_source_dijkstra() (in module networkx.algorithms.approximation.steinertree),
workx.algorithms.shortest_paths.weighted), 182
635 stochastic_block_model() (in module net-
single_source_dijkstra_path() (in module workx.generators.community), 866
networkx.algorithms.shortest_paths.weighted), stochastic_graph() (in module net-
Index 1113
workx.generators.stochastic), 853 succ (MultiDiGraph property), 137

stoer_wagner() (in module net- succ (PlanarEmbedding property), 610
workx.algorithms.connectivity.stoerwagner), successors() (DiGraph method), 61
394 successors() (MultiDiGraph method), 136
strategy_connected_sequential() (in module successors() (PlanarEmbedding method), 597
networkx.algorithms.coloring), 321 sudoku_graph() (in module net-
strategy_connected_sequential_bfs() (in workx.generators.sudoku), 883
module networkx.algorithms.coloring), 322 symmetric_difference() (in module net-
strategy_connected_sequential_dfs() (in workx.algorithms.operators.binary), 561
module networkx.algorithms.coloring), 321 syntactic_feasibility() (DiGraphMatcher
strategy_independent_set() (in module net- method), 509
workx.algorithms.coloring), 322 syntactic_feasibility() (GraphMatcher
strategy_largest_first() (in module net- method), 507
workx.algorithms.coloring), 322
strategy_random_sequential() (in module net- T
workx.algorithms.coloring), 322 tensor_product() (in module net-
strategy_saturation_largest_first() (in workx.algorithms.operators.product), 568
module networkx.algorithms.coloring), 322 tetrahedral_graph() (in module net-
strategy_smallest_last() (in module networkx.generators.small), 808
workx.algorithms.coloring), 323 threshold_accepting_tsp() (in module net-
strong_product() (in module networkx.algorithms.approximation.traveling_salesman),
workx.algorithms.operators.product), 567 189
strongly_connected_components() (in module thresholded_random_geometric_graph() (in
networkx.algorithms.components), 346 module networkx.generators.geometric), 846
strongly_connected_components_recursive() to_agraph() (in module networkx.drawing.nx_agraph),
(in module networkx.algorithms.components), 1002
347 to_dict_of_dicts() (in module networkx.convert),
subgraph() (DiGraph method), 72 908
subgraph() (Graph method), 36 to_dict_of_lists() (in module networkx.convert),
subgraph() (in module networkx.classes.function), 761 911
subgraph() (MultiDiGraph method), 148 to_directed() (DiGraph method), 72
subgraph() (MultiGraph method), 108 to_directed() (Graph method), 35
subgraph() (PlanarEmbedding method), 596 to_directed() (in module networkx.classes.function),
subgraph_centrality() (in module net- 759
workx.algorithms.centrality), 288 to_directed() (MultiDiGraph method), 147
subgraph_centrality_exp() (in module net- to_directed() (MultiGraph method), 107
workx.algorithms.centrality), 289 to_directed() (PlanarEmbedding method), 597
subgraph_is_isomorphic() (DiGraphMatcher to_directed_class() (PlanarEmbedding method),
method), 508 598
subgraph_is_isomorphic() (GraphMatcher to_edgelist() (in module networkx.convert), 911
method), 506 to_graph6_bytes() (in module net-
subgraph_is_isomorphic() (ISMAGS method), workx.readwrite.graph6), 973
521 to_nested_tuple() (in module net-
subgraph_isomorphisms_iter() (DiGraph- workx.algorithms.tree.coding), 735
Matcher method), 508 to_networkx_graph() (in module net-
subgraph_isomorphisms_iter() (Graph- workx.convert), 907
Matcher method), 506 to_numpy_array() (in module net-
subgraph_isomorphisms_iter() (ISMAGS workx.convert_matrix), 913
method), 521 to_pandas_adjacency() (in module net-
subgraph_view() (in module networkx.convert_matrix), 920
workx.classes.function), 762 to_pandas_edgelist() (in module net-
subgraph_view() (in module networkx.convert_matrix), 922
workx.classes.graphviews), 151 to_prufer_sequence() (in module net-
succ (DiGraph property), 61 workx.algorithms.tree.coding), 737
1114 Index
to_pydot() (in module networkx.drawing.nx_pydot), triadic_census() (in module net-

1006 workx.algorithms.triads), 747
to_scipy_sparse_array() (in module net- triads_by_type() (in module net-
workx.convert_matrix), 917 workx.algorithms.triads), 749
to_sparse6_bytes() (in module net- triangles() (in module networkx.algorithms.cluster),
workx.readwrite.sparse6), 977 313
to_undirected() (DiGraph method), 71 triangular_lattice_graph() (in module net-
to_undirected() (Graph method), 34 workx.generators.lattice), 797
to_undirected() (in module net- trivial_graph() (in module net-
workx.classes.function), 759 workx.generators.classic), 792
to_undirected() (MultiDiGraph method), 146 trophic_differences() (in module net-
to_undirected() (MultiGraph method), 106 workx.algorithms.centrality), 298
to_undirected() (PlanarEmbedding method), 598 trophic_incoherence_parameter() (in module
to_undirected_class() (PlanarEmbedding networkx.algorithms.centrality), 298
method), 599 trophic_levels() (in module net-
to_vertex_cover() (in module networkx.algorithms.centrality), 297
workx.algorithms.bipartite.matching), 223 truncated_cube_graph() (in module net-
topological_generations() (in module networkx.generators.small), 809
workx.algorithms.dag), 420 truncated_tetrahedron_graph() (in module
topological_sort() (in module net- networkx.generators.small), 809
workx.algorithms.dag), 419 turan_graph() (in module net-
transitive_closure() (in module networkx.generators.classic), 792
workx.algorithms.dag), 425 tutte_graph() (in module networkx.generators.small),
transitive_closure_dag() (in module net- 810
workx.algorithms.dag), 427 tutte_polynomial() (in module net-
transitive_reduction() (in module networkx.algorithms.polynomials), 611
workx.algorithms.dag), 427
transitivity() (in module net- U
workx.algorithms.cluster), 314 uniform_random_intersection_graph() (in
traveling_salesman_problem() (in module net- module networkx.generators.intersection), 854
workx.algorithms.approximation.traveling_salesman),
union() (in module net-
184 workx.algorithms.operators.binary), 557
traverse_face() (PlanarEmbedding method), 599 union() (UnionFind method), 1027
tree_all_pairs_lowest_common_ancestor() union_all() (in module net-
(in module networkx.algorithms.operators.all), 563
workx.algorithms.lowest_common_ancestors), UnionAdjacency (class in networkx.classes.coreviews),
534 158
tree_data() (in module net- UnionAtlas (class in networkx.classes.coreviews), 156
workx.readwrite.json_graph), 968 UnionMultiAdjacency (class in net-
tree_graph() (in module networkx.classes.coreviews), 160
workx.readwrite.json_graph), 969 UnionMultiInner (class in net-
tree_isomorphism() (in module networkx.classes.coreviews), 159
workx.algorithms.isomorphism.tree_isomorphism), update() (DiGraph method), 49
502 update() (Graph method), 18
treewidth_min_degree() (in module net- update() (MappedQueue method), 1046
workx.algorithms.approximation.treewidth), update() (MultiDiGraph method), 123
193 update() (MultiGraph method), 88
treewidth_min_fill_in() (in module net- update() (PlanarEmbedding method), 600
workx.algorithms.approximation.treewidth),
194 V
triad_graph() (in module net- values() (AdjacencyView method), 155
workx.generators.triads), 873 values() (AtlasView method), 154
triad_type() (in module networkx.algorithms.triads), values() (FilterAdjacency method), 163
750 values() (FilterAtlas method), 162
Index 1115
values() (FilterMultiAdjacency method), 165 write_gexf() (in module networkx.readwrite.gexf),

values() (FilterMultiInner method), 164 948
values() (MultiAdjacencyView method), 156 write_gml() (in module networkx.readwrite.gml), 952
values() (UnionAdjacency method), 159 write_graph6() (in module net-
values() (UnionAtlas method), 157 workx.readwrite.graph6), 974
values() (UnionMultiAdjacency method), 161 write_graphml() (in module net-
values() (UnionMultiInner method), 160 workx.readwrite.graphml), 958
vf2pp_all_isomorphisms() (in module net- write_multiline_adjlist() (in module net-
workx.algorithms.isomorphism.vf2pp), 500 workx.readwrite.multiline_adjlist), 937
vf2pp_is_isomorphic() (in module net- write_pajek() (in module networkx.readwrite.pajek),
workx.algorithms.isomorphism.vf2pp), 500 980
vf2pp_isomorphism() (in module net- write_sparse6() (in module net-
workx.algorithms.isomorphism.vf2pp), 501 workx.readwrite.sparse6), 978
volume() (in module networkx.algorithms.cuts), 414 write_weighted_edgelist() (in module net-
voronoi_cells() (in module networkx.readwrite.edgelist), 944
workx.algorithms.voronoi), 754
voterank() (in module networkx.algorithms.centrality), Z
299 zipf_rv() (in module networkx.utils.random_sequence),
1028
W
watts_strogatz_graph() (in module net-
workx.generators.random_graphs), 816
waxman_graph() (in module net-
workx.generators.geometric), 848
weakly_connected_components() (in module
networkx.algorithms.components), 352
weighted_choice() (in module net-
workx.utils.random_sequence), 1029
weighted_projected_graph() (in module net-
workx.algorithms.bipartite.projection), 228
weisfeiler_lehman_graph_hash() (in module
networkx.algorithms.graph_hashing), 484
weisfeiler_lehman_subgraph_hashes() (in
module networkx.algorithms.graph_hashing),
486
wheel_graph() (in module net-
workx.generators.classic), 792
wiener_index() (in module net-
workx.algorithms.wiener), 755
windmill_graph() (in module net-
workx.generators.community), 867
within_inter_cluster() (in module net-
workx.algorithms.link_prediction), 531
write_adjlist() (in module net-
workx.readwrite.adjlist), 933
write_dot() (in module networkx.drawing.nx_agraph),
1003
write_dot() (in module networkx.drawing.nx_pydot),
1006
write_edgelist() (in module net-
workx.algorithms.bipartite.edgelist), 217
write_edgelist() (in module net-
workx.readwrite.edgelist), 942
1116 Index

Networkx Reference

Uploaded by

Copyright:

Available Formats

Networkx Reference

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Networkx Reference

Uploaded by

Copyright:

Available Formats

NetworkX Reference

Aric Hagberg, Dan Schult, Pieter Swart

Dec 11, 2022

5 Graph generators 781

6 Linear algebra 885

7 Converting to and from other data formats 907

8 Relabeling nodes 927

9 Reading and writing graphs 931

Python Module Index 1085

1.1 NetworkX Basics

>>> import networkx as nx

1.2.1 Nodes and Edges

1.3 Graph Creation

NetworkX graph objects can be created in one of three ways:

>>> import networkx as nx

Edge attributes can be anything:

>>> import math

You can add many edges at one time:

See the Tutorial for more examples.

1.4 Graph Reporting

1.3. Graph Creation 3

>>> import matplotlib.pyplot as plt

See the examples for more ideas.

1.7 Data Structure

1.7. Data Structure 5

1 “It’s dictionaries all the way down.”

NetworkX provides data structures and methods for storing graphs.

2.1 Which graph class should I use?

Networkx Class Type Self-loops allowed Parallel edges allowed

2.2 Basic graph types

2.2.1 Graph—Undirected graphs with self loops

class Graph(incoming_graph_data=None, **attr)

G can be grown in several ways.

>>> G.add_nodes_from([2, 3])

>>> G.add_edges_from([(1, 2), (1, 3)])

8 Chapter 2. Graph types

Add node attributes using add_node(), add_nodes_from() or G.nodes

>>> G.add_node(1, time="5pm")

Add edge attributes using add_edge(), add_edges_from(), subscript notation, or G.edges.

>>> G.add_edge(1, 2, weight=4.7)

>>> 1 in G # check if node in graph

>>> for n, nbrsdict in G.adjacency():

But the edges() method is often more convenient:

>>> for u, v, weight in G.edges.data("weight"):

2.2. Basic graph types 9

10 Chapter 2. Graph types

>>> class ThinGraph(nx.Graph):

Adding and removing nodes and edges

Graph.__init__([incoming_graph_data]) Initialize a graph with edges, name, or graph attributes.

2.2. Basic graph types 11

>>> G = nx.Graph() # or DiGraph, MultiGraph, MultiDiGraph, etc

Arbitrary graph attribute pairs (key=value) may be assigned

>>> G = nx.Graph(e, day="Friday")

>>> G = nx.Graph() # or DiGraph, MultiGraph, MultiDiGraph, etc

Use keywords set/change node attributes:

12 Chapter 2. Graph types

>>> G.add_node(1, size=10)

>>> G = nx.Graph() # or DiGraph, MultiGraph, MultiDiGraph, etc

Use keywords to update specific node attributes for every node.

>>> G.add_nodes_from([1, 2], size=10)

Graph.init([incoming_graph_data]) Initialize a graph with edges, name, or graph attributes.