在Matlab中随机选择大量样本的子集

Question

I would like to create a subset of n samples from a large number of N samples, n << N. I usually use randperm function in Matlab and take the first n indices . However, since the data can be very large, randperm give me error message of not enough memory.

I would like to have suggestions, how can I select a small subset out of a large number of data set without using randperm function in Matlab.

Thank you.

Answer 1

You could try to use single() to halve the datasize:

http://www.mathworks.de/help/matlab/ref/single.html

Answer 2

randi gives uniformly distributed numbers,

Ind = randi(N,[n 1]);
Observation = data(Ind); 

You could also use datasample ,

Observation = datasample(data,n,'Replace',false));

As @Shai mentioned, another option would be randsample ,

Observation = data(randsample(N,n));

Answer 3

If n is much smaller than N , a rejection method is efficient: generate a sample with possible repetitions using randi , check if there are repetitions (which is not very likely), and if so repeat:

N = 10000;
n = 100;
repeat = true;
while repeat
    sample = randi(N,1,n);
    repeat = any(sum(bsxfun(@eq, sample, sample.'))>1);
end