CMSC 671

Fall 2010
Chapter 6: Constraints

Prof. Marie desJardins

Class #6 – Wednesday, September 22

1
 On Lisp
• “Tell your students they'll learn to love it, I did.”
– Mike Pickens ’03 (CMSC 471 student, Fall 2002)

2
 Today’s class
• Constraint Processing / Constraint Satisfaction Problem
(CSP) paradigm
• Algorithms for CSPs
– Backtracking (systematic search)
– Constraint propagation (k-consistency)
– Variable and value ordering heuristics
– Intelligent backtracking

3
Constraint
Satisfaction
Russell & Norvig Ch. 6

4
 Overview
• Constraint satisfaction offers a powerful problem-solving
paradigm
– View a problem as a set of variables to which we have to assign
values that satisfy a number of problem-specific constraints.
– Constraint programming, constraint satisfaction problems (CSPs),
constraint logic programming…
• Algorithms for CSPs
– Backtracking (systematic search)
– Constraint propagation (k-consistency)
– Variable and value ordering heuristics
– Backjumping and dependency-directed backtracking

5
 Informal definition of CSP
• CSP = Constraint Satisfaction Problem
• Given
(1) a finite set of variables
(2) each with a domain of possible values (often finite)
(3) a set of constraints that limit the values the variables
can take on
• A solution is an assignment of a value to each variable such
that the constraints are all satisfied.
• Tasks might be to decide if a solution exists, to find a
solution, to find all solutions, or to find the “best solution”
according to some metric (objective function).
6
 Informal example: Map coloring
• Color the following map using three colors
(red, green, blue) such that no two adjacent
regions have the same color.

D A
B
C
7
 Map coloring II
• Variables: A, B, C, D, E all of domain RGB
• Domains: RGB = {red, green, blue}
• Constraints: AB, AC,A  E, A  D, B  C, C  D, D  E
• One solution: A=red, B=green, C=blue, D=green, E=blue

E E
D A => D A
B B
C C

8
 Example: SATisfiability
• Given a set of propositions containing variables, find an
assignment of the variables to {false,true} that satisfies them.
• For example, the clauses:
– (A  B  C)  ( A  D)
– (equivalent to (C  A)  (B  D  A)
are satisfied by
A = false
B = true
C = false
D = false

9
 Real-world problems
• Scheduling • Graph layout
• Temporal reasoning • Network management
• Building design • Natural language
• Planning processing
• Optimization/satisfaction • Molecular biology /
• genomics
Vision
• VLSI design

10
 Formal definition of a constraint
network (CN)
A constraint network (CN) consists of
• a set of variables X = {x1, x2, … xn}
– each with an associated domain of values {d1, d2, … dn}.
– the domains are typically finite
• a set of constraints {c1, c2 … cm} where
– each constraint defines a predicate which is a relation
over a particular subset of X.
– e.g., Ci involves variables {Xi1, Xi2, … Xik} and defines
the relation Ri  Di1 x Di2 x … Dik
• Unary constraint: only involves one variable
• Binary constraint: only involves two variables

11
 Formal definition of a CN (cont.)

• Instantiations
– An instantiation of a subset of variables S is an
assignment of a value in its domain to each
variable in S
– An instantiation is legal iff it does not violate any
constraints.
• A solution is an instantiation of all of the
variables in the network.

12
 Typical tasks for CSP

• Solutions:
– Does a solution exist?
– Find one solution
– Find all solutions
– Given a partial instantiation, do any of the above
• Transform the CN into an equivalent CN
that is easier to solve.

13
 Binary CSP
• A binary CSP is a CSP in which all of the
constraints are binary or unary.
• Any non-binary CSP can be converted into a
binary CSP by introducing additional variables.
• A binary CSP can be represented as a constraint
graph, which has a node for each variable and an
arc between two nodes if and only there is a
constraint involving the two variables.
– Unary constraint appears as a self-referential arc

14
Example: Sudoku (of course!)
3 1

1 4

3 4 1 2

15
 Running example: Sudoku
v11 3 v13 1
• Variables and their domains
– vij is the value in the jth cell of the ith row
– Dij = D = {1, 2, 3, 4} v21 1 v23 4
• Blocks:
– B1 = {11, 12, 21, 22} 3 4 1 2
– ...
– B4 = {33, 34, 43, 44} v41 v42 4 v44
• Constraints (implicit/intensional)
– CR : i, j vij = D (every value appears in every row)
– CC : j, j vij = D (every value appears in every column)
– CB : k,  (vij | ij Bk) = D (every value appears in every block)
– Alternative representation: pairwise inequality constraints:
• IR : i, j≠j’ : vij ≠ vij’ (no value appears twice in any row)
• IC : j, i≠i’ : vij ≠ vi’j (no value appears twice in any column)
• IB : k, ij  Bk, i’j’  Bk, ij ≠ i’j’ :vij ≠ vi’j’ (no value appears twice in any block)
– Advantage of the second representation: all binary constraints!
16
 Sudoku constraint network
v11 3 v13 1
v11 v13
v21 1 v23 4

3 4 1 2 v21 v23

v41 v42 4 v44

v41 v42 v44

17
 Solving constraint problems
• Systematic search
– Generate and test
– Backtracking
• Constraint propagation (consistency)
• Variable ordering heuristics
• Value ordering heuristics
• Backjumping and dependency-directed
backtracking
21
 Systematic search: Backtracking
(a.k.a. depth-first search!)

• Consider the variables in some order

• Pick an unassigned variable and give it a provisional
value such that it is consistent with all of the constraints
• If no such assignment can be made, we’ve reached a
dead end and need to backtrack to the previous variable
• Continue this process until a solution is found or we
backtrack to the initial variable and have exhausted all
possible values

23
 Backtracking: Sudoku
Let’s try it…

v11 3 v13 1 
v21 1 v23 4
3 4 1 2
v41 v42 4 v44

24
 Problems with backtracking

• Thrashing: keep repeating the same failed

variable assignments
– Consistency checking can help
– Intelligent backtracking schemes can also help
• Inefficiency: can explore areas of the search
space that aren’t likely to succeed
– Variable ordering can help

26
 Consistency
• Node consistency
– A node X is node-consistent if every value in the domain of X is
consistent with X’s unary constraints
– A graph is node-consistent if all nodes are node-consistent
• Arc consistency
– An arc (X, Y) is arc-consistent if, for every value x of X, there is a
value y for Y that satisfies the constraint represented by the arc.
– A graph is arc-consistent if all arcs are arc-consistent.
• To create arc consistency, we perform constraint
propagation: that is, we repeatedly reduce the domain of
each variable to be consistent with its arcs

27
Constraint propagation: Sudoku
v11 v13
v11 3 v13 1
v21 1 v23 4
v21 v23
3 4 1 2
v41 v42 4 v44
y v41 v42 v44
s i ste n c
co n
e
Nod
2,4 2
4 2

2 2,3
Arc consistency 2 3
…and we didn’t even need to search!

1,2 2 3
1 2 3
28
 A famous example:
Labelling line drawings
• Waltz labelling algorithm – one of the earliest CSP applications
– Convex interior lines are labelled as +
– Concave interior lines are labeled as –
– Boundary lines are labeled as
• There are 208 labellings (most of which are impossible)
• Here are the 18 legal labellings:

30
 Labelling line drawings II
• Here are some illegal labelings:

- -
+ +
-

31
 Labelling line drawings (cont.)
• Waltz labelling algorithm: Propagate constraints repeatedly
until a solution is found

A solution for one A labelling problem

labelling problem with no solution
32
 K-consistency
• K- consistency generalizes the notion of arc
consistency to sets of more than two variables.
– A graph is K-consistent if, for legal values of any K-1
variables in the graph, and for any Kth variable Vk, there
is a legal value for Vk
• Strong K-consistency = J-consistency for all J<=K
• Node consistency = strong 1-consistency
• Arc consistency = strong 2-consistency
• Path consistency = strong 3-consistency

33
 Why do we care?

1. If we have a CSP with N variables that

is known to be strongly N-consistent,
we can solve it without backtracking
2. For any CSP that is strongly K-
consistent, if we find an appropriate
variable ordering (one with “small
enough” branching factor), we can
solve the CSP without backtracking

34
 Ordered constraint graphs
• Select a variable ordering, V1, …, Vn
• Width of a node in this OCG is the number of arcs leading
to earlier variables:
– w(Vi) = Count ( (Vi, Vk) | k < i)
• Width of the OCG is the maximum width of any node:
– w(G) = Max (w (Vi)), 1 <= i <= N
• Width of an unordered CG is the minimum width of all
orderings of that graph (“best you can do”)

35
 Tree-structured constraint graph
• A constraint tree rooted at V1 satisfies the following property:
– There exists an ordering V1, …, Vn such that every node has zero or one
parents (i.e., each node only has constraints with at most one “earlier” node
in the ordering)
V2 V3 V5
V1 V9 V10
V6
V8 V4 V7
– Also known as an ordered constraint graph with width 1
• If this constraint tree is also node- and arc-consistent (a.k.a. strongly 2-
consistent), then it can be solved without backtracking
• (More generally, if the ordered graph is strongly k-consistent, and has
width w < k, then it can be solved without backtracking.)

36
 Proof sketch for constraint trees
• Perform backtracking search in the order that satisfies the
constraint tree condition
• Every node, when instantiated, is constrained only by at
most one previous node
• Arc consistency tells us that there must be at least one legal
instantiation in this case
– (If there are no legal solutions, the arc consistency procedure will
collapse the graph – some node will have no legal instantiations)
• Keep doing this for all n nodes, and you have a legal
solution – without backtracking!

37
Backtrack-free CSPs: Proof sketch
• Given a strongly k-consistent OCG, G, with width w < k:
– Instantiate variables in order, choosing values that are consistent
with the constraints between Vi and its parents
– Each variable has at most w parents, and k-consistency tells us we
can find a legal value consistent with the values of those w parents
• Unfortunately, achieving k-consistency is hard (and can
increase the width of the graph in the process!)
• Fortunately, 2-consistency is relatively easy to achieve, so
constraint trees are easy to solve
• Unfortunately, many CGs have width greater than one (that
is, no equivalent tree), so we still need to improve search

38
 So what if we don’t have a tree?
• Answer #1: Try interleaving constraint propagation and
backtracking
• Answer #2: Try using variable-ordering heuristics to
improve search
• Answer #3: Try using value-ordering heuristics during
variable instantiation
• Answer #4: See if iterative repair works better
• Answer #5: Try using intelligent backtracking methods

39
Interleaving constraint propagation
and search
Generate and No constraint propagation: assign
Test all variable values, then test
constraints
Simple Check constraints only for variables
Backtracking “up the tree”

Forward Check constraints for immediate

Checking neighbors “down the tree”

Partial Propagate constraints forward

Lookahead “down the tree”

Full Ensure complete arc consistency

Lookahead after each instantiation (AC-3)

40
 Variable ordering
• Intuition: choose variables that are highly constrained early in
the search process; leave easy ones for later
• Minimum width ordering (MWO): identify OCG with
minimum width
• Maximum cardinality ordering: approximation of MWO
that’s cheaper to compute: order variables by decreasing
cardinality (a.k.a. degree heuristic)
• Fail first principle (FFP): choose variable with the fewest
values (a.k.a. minimum remaining values (MRV))
– Static FFP: use domain size of variables
– Dynamic FFP (search rearrangement method): At each point in the
search, select the variable with the fewest remaining values

41
 Variable ordering II
• Maximal stable set: find largest set of variables with no
constraints between them and save these for last
• Cycle-cutset tree creation: Find a set of variables that,
once instantiated, leave a tree of uninstantiated variables;
solve these, then solve the tree without backtracking
• Tree decomposition: Construct a tree-structured set of
connected subproblems

42
 Value ordering
• Intuition: Choose values that are the least constrained early
on, leaving the most legal values in later variables
• Maximal options method (a.k.a. least-constraining-value
heuristic): Choose the value that leaves the most legal
values in uninstantiated variables
• Min-conflicts: Used in iterative repair search (see below)

43
 Iterative repair
• Start with an initial complete (but invalid) assignment
• Hill climbing, simulated annealing
• Min-conflicts: Select new values that minimally conflict
with the other variables
– Use in conjunction with hill climbing or simulated annealing or…
• Local maxima strategies
– Random restart
– Random walk
– Tabu search: don’t try recently attempted values

44
 Min-conflicts heuristic
• Iterative repair method
1. Find some “reasonably good” initial solution
– E.g., in N-queens problem, use greedy search through rows, putting
each queen where it conflicts with the smallest number of previously
placed queens, breaking ties randomly
2. Find a variable in conflict (randomly)
3. Select a new value that minimizes the number of constraint
violations
– O(N) time and space
4. Repeat steps 2 and 3 until done
• Performance depends on quality and informativeness of
initial assignment; inversely related to distance to solution

45
 Intelligent backtracking
• Backjumping: if Vj fails, jump back to the variable Vi with
greatest i such that the constraint (Vi, Vj) fails (i.e., most
recently instantiated variable in conflict with Vi)
• Backchecking: keep track of incompatible value
assignments computed during backjumping
• Backmarking: keep track of which variables led to the
incompatible variable assignments for improved
backchecking

46
 Some challenges for constraint
reasoning
• What if not all constraints can be satisfied?
– Hard vs. soft constraints
– Degree of constraint satisfaction
– Cost of violating constraints
• What if constraints are of different forms?
– Symbolic constraints
– Numerical constraints [constraint solving]
– Temporal constraints
– Mixed constraints

47
 Some challenges for constraint
reasoning II
• What if constraints are represented intensionally?
– Cost of evaluating constraints (time, memory, resources)
• What if constraints, variables, and/or values change over
time?
– Dynamic constraint networks
– Temporal constraint networks
– Constraint repair
• What if you have multiple agents or systems involved in
constraint satisfaction?
– Distributed CSPs
– Localization techniques

48