Optimizing generating random number in Matlab

Question

I have a two 1 by 10^3 matrix which its elements change in each step of the loop of simulation. To find a new element in each for loop step, I need 2*1000 Gaussian random numbers. the number of loop steps is equal to 10^5. Which method is less time-consuming in this case?

creating two 10^5 by 1000 random matrix before for loop and call them inside the loop.
creating two 1 by 10^3 random matrix in each step of the loop?

Because there are lots of stuff in the for loop and if I ran the code, it takes hours to give me the output@EBH — Oliver Range, Jun 04 '17 at 11:30

score 0 · Accepted Answer · answered Jun 04 '17 at 12:22

Here is a little code to answer your question:

function rnd_time_test
rnd_test = @(n) randn(2,n,1000);
N = 1e5;
one_time = timeit(@() rnd_test(N))
loop_time = timeit(@() rnd_test_loop(N))

    function R = rnd_test_loop(n)
        for k = 1:n
            R = randn(2,1000);
        end
    end
end

and the result:

one_time =
      2.245
loop_time =
      1.6311

So, on my machine, using a loop is faster.

Optimizing generating random number in Matlab

1 Answers1