Usage of parfor (parallel computing)

I have a question regarding parallel computing with MATLAB. Suppose I have the following code:

----

trials = 30;

data = zeros(trials,11,3);

for k=1:trials

      count = 1;
      for i=0:1:10
          b = simulate_network_inhibitory_testing(i/10);
          data(k,count,: ) = b;
          count = count + 1;
      end
end

save('data2301performance.mat', 'data');
---

The called function 'simulate_network_inhibitory_testing' returns a 3x1 Matrix.

I now want to parallelize this script using parfor. However, if I just substitute the first for loop by a parfor loop this does not work, because MATLAB says I am not allowed to use the data matrix in that way.

What is the problem with this and how would a work-around look like?

[Walter Roberson]

Does simulate_network_inhibitory_testing return a value involving random seeds? If not then there is no need to do repeated trials.

Remove your count loop and replace its value in the loop with i+1

data(k,i+1,:) = b;

You would have better efficiency if you moved the trial number to the last dimension and the 3 outputs of b to the first:

data(:, i+1, trial) = b;

This would allow the array to be split along complete panes.

[Stefan Depeweg]

Hello,

thanks for your quick answer!

1. Yes, each network simulation involves random seeds, the trials are neccessary.

2. Just saw that the count variable is useless, thanks :)

3. However, even if I switch the third and the last dimension, parfor still does not work. "parfor loop cannot run due to the way variable 'data' is used"

[Walter Roberson]

Use a local array to hold all the 11 x 3 results calculated in the loop, and then after the end of the inner loop, write the block in, so that the only index named in data() in the outer loop is k and the other two are : .

Answer 1

[Edric Ellis]

Walter is quite right to suggest the local array approach. Just to expand what he described:

trials = 30;
data = zeros(trials,11,3);
parfor k=1:trials
  datak = zeros(11,3);
  for i=0:1:10
      b = simulate_network_inhibitory_testing(i/10);
      datak(i+1, :) = b;
  end
  data(k,:,:) = datak;
end

[Walter Roberson]

Amazing what one can pick up by reading random cssm answers for products one doesn't even use ;-)