Randomly Sample entries of a matrix and return the (Row, Column) Indexes in R?

Question

I have a matrix with M rows and N columns. I need to randomly sample different locations in these matrix and return the row indexes and col indexes.

My approach: Say, I want to sample 30 percentage of entries in the matrix. Then, I iterate through the whole matrix, at each point, I toss a biased coin with heads of 30 percent probability and select the location if heads comes. Since, my data is large, this approximately selects 30% of the entries. However, I observe that this is really slow. Is there a way to speed this up? Or a better way to do it?

You need to row/col indexes or the values themselves from the matrix? — David Arenburg, May 14 '15 at 08:36
Why not the actual values? What are you going to do with the indexes? — David Arenburg, May 14 '15 at 08:38
Well, In my problem, I need to keep track of the locations of the matrix I sampled from. — dineshdileep, May 14 '15 at 08:39
@Roland So sample 30% of the rows, and sample 30% of the columns, is that what you mean? Then, I will have to generate whole combinations also, isn't it? — dineshdileep, May 14 '15 at 08:42
possible duplicate of [random sampling - matrix](http://stackoverflow.com/questions/9109778/random-sampling-matrix) — zx8754, May 14 '15 at 08:50

nicola · Accepted Answer · 2018-01-22T21:49:40.777

4

If m is your matrix, just try:

arrayInd(sample(length(m),0.3*length(m)),dim(m))

An example:

set.seed(1)
m<-matrix(ncol=6,nrow=6)
arrayInd(sample(length(m),0.3*length(m)),dim(m))      
#      [,1] [,2]
# [1,]    4    2
# [2,]    2    3
# [3,]    2    4
# [4,]    6    5
# [5,]    1    2
# [6,]    4    5
# [7,]    5    5
# [8,]    4    6
# [9,]    6    3
#[10,]    2    1

edited Jan 22 '18 at 21:49

answered May 14 '15 at 09:13

nicola

24,005
3
35
56

score 1 · Answer 2 · answered May 14 '15 at 08:58

See this example:

m=2
n=5
SampleSize=0.3

#dummy data
x <- matrix(runif(m*n),nrow=n)

#sample
set.seed(123)
temp <- x
temp[ sample(1:length(temp),round(length(temp)*SampleSize))] <- -9

#index
ix <- temp==-9

ix
#        [,1]  [,2]
# [1,] FALSE FALSE
# [2,] FALSE FALSE
# [3,]  TRUE  TRUE
# [4,]  TRUE FALSE
# [5,] FALSE FALSE

score 1 · Answer 3 · answered May 14 '15 at 09:13

1

My new favorite option:

indexSampler <- function(m, p) {
    matrix(sample(c(TRUE,FALSE), length(m), p = c(p, 1 - p), replace=TRUE), ncol(m))
}

You won't get indices, but you'll get a matrix full of TRUE/FALSE that can be used to index.

It is ridiculously fast (a factor of 1000 for a matrix of 200x200, and also significantly faster for small matrices).

answered May 14 '15 at 09:13

PascalVKooten

20,643
17
103
160

if I try to get back the indices, won't it be further slower!! – dineshdileep May 14 '15 at 10:22

Randomly Sample entries of a matrix and return the (Row, Column) Indexes in R?

3 Answers3