MATLAB/R

Dic,onary R meetup NYC January 7, 2010

Harlan Harris harlan@harris.name @HarlanH Marck Vaisman marck@vaisman.us @wahalulu
MATLAB and the MATLAB logo are registered trademarks of The Mathworks.

About MATLAB
What is MATLAB
Commercial numerical programming language, simula,on and visualiza,on One million users (engineers, scien,sts, academics) MATrix LABoratory specializes in matrix opera,ons Mathworks - base & add-ons Open-source Octave project

MATLAB History
Developed by Cleve Moler (Math/CS Prof at UNM) in the 1970s as a higher-level numerical programming language (vs. Fortran LINPACK) Adopted by engineers for signal processing, control modeling Mul,purpose programming language

Notes
Todays focus: Compare MATLAB & R for data analysis, contrast as programming languages MATLAB is Base plus many toolboxes
Base includes: descrip,ve stats, covariance and correla,on, linear and nonlinear regression Sta,s,cs toolbox adds: dataset and category (like data.frames and factors) arrays, more visualiza,ons, distribu,ons, ANOVA, mul,variate regression, hypothesis tests

->
Interac,ve programming: Scripts and Read-Evaluate- Print Loop Similar representa,ons of data
Both use vectors/arrays as the primary data structures
Matlab is based on 2-D matricies; R is based on 1-D vectors

Both prefer vectorized func,ons to for loops Variables are declared dynamically

Can do most MATLAB func,onality in R; can do most R func,onality in MATLAB.

The basics: vectors, matrices and indexing

Task Create a row vector Create a column vector Enter a matrix A
v = [1 2 3 4] v=[1;2;3;4] or v=[1 2 3 4] v<-c(1,2,3,4) v<-c(1,2,3,4) Note: R does not distinguish between row and column vectors Enter values by row: A<-matrix(c(1,2,3,4,5,6), nrow=2, byrow=TRUE) Enter values by column: A<-matrix(c(1,4,2,5,3,6), nrow=2) v[3] or v[[3]] A[2,3] A<-cbind(a1,a2) A<-rbind(a1,a2) dim(A)<-c(m,n)

A=[1 2 3; 4 5 6]

Access third element of vector v Access element of matrix A Glue two matrices a1 and a2, same number of rows, side by side Stack two matrices a1 and a2, same number of columns Reshape* matrix A, making it an m x n matrix with elements taken columnwise from A

v(3) A(2,3) A=[a1 a2] A=[a1;a2] A=reshape(A,m,n)

Operators
Task Assignment Whole Matrix Opera,ons: Element-by-element Opera,ons: Compute A-1B Sums Logical operators (element-by- element on vectors/matrices)
= Multiplication: A*B Square the matrix: A^2 Raise to power k: A^k A.*B A./B A.^k A\B Columns of matrix: sum(A) Rows of matrix: sum(A,2) a < b, a > b, a <= b, a >= b a == b a ~= b AND: a && b OR: a || b XOR: xor(a,b) NOT: ~a <- or = A %*% B A %*% A A %*% A %*% A A*B A/B A^k A%*% solve(B) colSums(A) rowSums(A) a < b, a > b, a <= b, a >= b a == b a != b AND: a && b (short-circuit) a & b (element-wise) OR: a || b a | b XOR: xor(a,b) NOT: !a

Working with data structures

Task
Build a structure v of length n, capable of containing dierent data types in dierent elements. MATLAB: cell array R: list Create a matrix-like object with dierent named columns. MATLAB: struct array R: data.frame
v=cell(1,n) In general, cell (m,n) makes an m n cell array. Then you can do e.g.: v{1}=12 v{2}=hi there v{3}=rand(3) v<-vector(list,n) Then you can do e.g.: v[[1]]<-12 v[[2]]<-hi there v[[3]]<-matrix(runif(9),3)

avals=2*ones(1,6); v<-c(1,5,3,2,3,7) yvals=6:-1:1; v=[1 5 3 2 3 7]; d<-data.frame(cbind(a=2, d=struct(a, avals, yy=6:1), v) yy, yyvals, fac, v);

Condi,onals, control structures, loops

Task for loops over values in vector v
for i=v command1 command2 end If only one command: for (i in v) command If multiple commands: for (i in v) { command1 command2 }

If/else statement

if cond command1 command2 else command3 command4 end MATLAB also has the elseif statement.

if (cond) { command1 command2 } else { command3 command4 } R uses chained else if statements. > print(ifelse(c(T,F), 2, 3)) [1] 2 3

ifelse() func,on

Help!
Task Get help on a func,on Search the help for a word Describe a variable Show variables in environment Underlying type of variable
help fminsearch help(pmin) or ?pmin ??inverse class(a) str(a) ls() typeof(a)

lookfor inverse class(a) who whos(a)

Example: k-means clustering of Fisher Iris data

Fisher Iris Dataset sepal_length,sepal_width,petal_length,petal_width,species 5.1,3.5,1.4,0.2,setosa 4.9,3.0,1.4,0.2,setosa 4.7,3.2,1.3,0.2,setosa 4.6,3.1,1.5,0.2,setosa

Matlab and R as programming languages

Scrip,ng, real-,me analysis File-based environments Impera,ve programming style Sta,cally scoped Func,ons with mul,ple return values Evolving OOP system Can be compiled Large library of func,ons Professional developed, cost money Can embed (in) many other languages Scrip,ng, real-,me analysis Files unimportant Func,onal programming style (impure) Dynamically scoped Func,ons with named arguments, lazy evalua,on Mul,ple compe,ng OOP systems Cannot be compiled Large library of func,ons Varying quality and support Can embed (in) many other languages

Func,ons
function [a, b] = minmax(z) % one function per .m file! % assign to formal return names a = min(z) b = max(z) end minmax <- function(c, opt=12) { # functions are assigned to # variables ret <- list(min = min(z), max = max(z)) ret # last statement is # return value } # if minmax was created in current # environment x <- minmax(c(1, 30, 3)) smallest <- x$min

% if minmax.m in path [smallest, largest] = minmax([1 30 3])

Object-Oriented Programming
Formerly: objects were dened by a directory tree, with one method per le As of 2008: new classdef syntax resembles other languages S3 classes: anributes + syntax
class(object) plot.lm()

S4 classes: deni,ons + methods R.oo, proto, etc

Other notes
r.matlab package Graphics
Matlab has much bener 3-d/interac,ve graphics support R has ggplot2 and much bener sta,s,cal graphics

Addi,onal Resources
Will Dwinell, Data Mining in MATLAB Computerworld ar,cle on Cleve Moler Mathworks Matlabcentral Comparison of Data Analysis packages ( hnp://anyall.org/blog/2009/02/comparison-of-data- analysis-packages-r-matlab-scipy-excel-sas-spss- stata/) R.matlab package stackoverow

References used for this talk

David Hiebeler MATLAB/R Reference document: hnp://www.math.umaine.edu/~hiebeler/comp/ matlabR.html hnp://www.cyclismo.org/tutorial/R/index.html hnp://www.stat.berkeley.edu/~spector/R.pdf MATLAB documenta,on hnp://www.r-cookbook.com/node/23

Thank You!