Sampling with periodic replacement

I am looking for an efficient way of doing randperm(n,k) many successive times with the same n and k. Can anyone propose something more efficient than the obvious for-loop approach below?

  M=1e5;
  n=100;
  k=10;
  
  A=nan(k,M);
  
  for i=1:M
      
     A(:,i) = randperm(n,k).'; 
      
  end

Answer 1

[Walter Roberson]

In current versions of MATLAB, randperm with small k relative to n (and perhaps other cases) uses a Fisher-Yates shuffle for efficiency.

Older versions of MATLAB use sort(rand) to extract orderings. That can be extended easily:

[~, A] = sort( rand(n, M) );

[Matt J]

Ah! Thanks. I suppose we should use mink/maxk, though, for Matlab versions that have it

[~, A] = mink( rand(n, M) , k );   ?

[Walter Roberson]

Good idea. On older systems, you would use A(1:k,:)

... Though I just did some timing tests, and using sort and indexing turns out to be measurably faster. Using sort and indexing is bout 0.125 for the parameters you indicate, vs about 0.156 for using mink.