MATLAB: How to vectorize random permutation of data

for looppermutationstatisticsvectorization

I need to randomly permute a set of data, and I need to do it 10,000 or more times, so I need to do it efficiently. Below is an example of how I'm doing it (with randomly generated data standing in for real data). I feel like there should be a way to vectorize the permutation process instead of the for-loop I'm using, but I can't think of how to do it. I need a method that works for any number of data points–i.e., below I am permuting two data points for each hypothetical subject, but I need to generalize to three, four, etc.
data = rand(24, 2);
for j = 1:24
    perm_data(i, :) = data(i, randperm(2));
end
%Do some calculations on the permuted data here

Best Answer

You wish to randomly permute each of the rows of ‘data’. Then do this:
   (Simplified)
   [m,n] = size(data);
   [~,p] = sort(rand(m,n),2);
   perm_data = reshape(data(repmat((1-m:0).’,n,1)+p(:)*m),m,n);
Related Question