0

Suppose I have the following code in MATLAB:

clc; clear;
myseed = rng(1);
c = 1; d = 2;
parfor i = 1:1000
   a = randn(200,1);
   b = randn(50,1);
   c*(sum(a)+sum(b));
end
parfor i = 1:1000
   a = randn(200,1);
   b = randn(50,1);
   d*(sum(a)+sum(b));
end

The third line in each loop above captures the processing that I apply to a and b, which is specific to the loop and which is very lengthy in my real application. The reason for repeating the randomizing of a and b is because I would like to put these 2 loops into 2 separate files that I can call from the main program. I understand there is some inefficiency here but the code would be easier to follow for me.

How do I use rng so a and b in one loop are the same as the a and b in another loop? (That is, the 1000 "random" values for a from the first loop are the same as the 1000 "random" values for a from the other loop. They don't have to be in the same order.) I tried adding rng(myseed) between the 2 loops but this didn't help.

yurnero
  • 315
  • 2
  • 9
  • @LuisMendo I confirm what you said works. I apologize for this: but I forgot that I was working with `parfor` as opposed to `for`. The solution you gave no longer works with `parfor`. Could you please help with the edited question? – yurnero May 28 '18 at 17:10

1 Answers1

1

I don't use parfor often, but the following approach (adapted from here) seems to work:

parpool('local',4); % create parallel pool of workers
spmd
    rng(123); % set the same seed in each worker
end
parfor k = 1:12
    r = rand; % generate a random number
    w = getCurrentWorker;
    fprintf('Worker %i, iteration %i, random number %f\n', w.ProcessId, k, r)
end

Note how each worker generates the same random numbers in the same order. (The iterations are randomly assigned to workers, and are run in a random order; this is by design).

Starting parallel pool (parpool) using the 'local' profile ...
connected to 4 workers.
Worker 11688, iteration 1, random number 0.275141
Worker 17188, iteration 2, random number 0.275141
Worker 18408, iteration 4, random number 0.275141
Worker 18408, iteration 3, random number 0.423046
Worker 5812, iteration 6, random number 0.275141
Worker 5812, iteration 5, random number 0.423046
Worker 11688, iteration 8, random number 0.423046
Worker 11688, iteration 7, random number 0.973406
Worker 17188, iteration 9, random number 0.423046
Worker 18408, iteration 10, random number 0.973406
Worker 5812, iteration 11, random number 0.973406
Worker 17188, iteration 12, random number 0.973406
Luis Mendo
  • 110,752
  • 13
  • 76
  • 147
  • I just tested, you need the 2 `stream` commands as in the linked Mathworks article. At least this is the case for my 2018a MATLAB. I will tick when you do. Thanks for all the help. I hope you don't mind this minor suggestion. I knowledge: you have 99 percent of the solution. :) – yurnero May 28 '18 at 17:55
  • Sorry, I can't reproduce. I just tested in R2018a and it works for me. Did you paste my code as is, on a clean Matlab session? – Luis Mendo May 29 '18 at 09:13
  • I meant for my example with 2 loops. You need the `stream` commands to ensure the 2 parloops spit out the same numbers (in possibly different ordering). Anyway, let me just tick you. – yurnero May 29 '18 at 10:56
  • I see now that I may have answered a different question from what you asked. I'm glad it helped anyway – Luis Mendo May 29 '18 at 12:04
  • You definitely did help. Thanks again! – yurnero May 29 '18 at 12:44