MATLAB Answers

# multiple random samples of size k

41 views (last 30 days)
cmo on 25 Jun 2015
Commented: Star Strider on 25 Jun 2015
I would like to generate M random samples, each of size k , from the integers 1:n without replacement.
(i.e. generate an M-by-k matrix of random integers, where each row contains integers in [1,n], and within a row, no integer is repeated).
I will use this matrix to perform Monte Carlo hypothesis testing.
I can easily write this in Matlab myself., however...
I would like to use an in-built Matlab function, but I am unable to find one. Does such a function exist in Matlab?
Note: Matlab's bootstrap do not fulfill the criteria I listed above.

#### 0 Comments

Sign in to comment.

### Answers (3)

Walter Roberson on 25 Jun 2015
You could use randperm(n,k) in a loop. But randperm is not designed to generate an array. You can look into the randperm implementation and generalize it easily:
T = rand(M, n);
[~, order] = sort(T, 2);
randsamples = order(:,1:k);

#### 0 Comments

Sign in to comment.

Azzi Abdelmalek on 25 Jun 2015
Edited: Azzi Abdelmalek on 25 Jun 2015
n=6
k=3
row1=randperm(n,k)
For a matrix with M rows
n=6
k=3
M=5
A=cell2mat(arrayfun(@(x) randperm(n,k),(1:M)','un',0))

#### 0 Comments

Sign in to comment.

Star Strider on 25 Jun 2015
Use the nchoosek function. It will do what you want. The randperm function is another option that may be more appropriate to your application.

#### 2 Comments

cmo on 25 Jun 2015
Thank you, but nchoosek will not work when n is large and k is non-trivial (the space of permutations is too big)
Star Strider on 25 Jun 2015
My pleasure. The perms function has the same limitations, but it and nchoosek are the only ones I know of that produce matrices. Otherwise, you’re reduced to using a loop with randperm to create your matrix.

Sign in to comment.

Sign in to answer this question.