Question

Currently I have a parfor loop making calculations on a struct in MATLAB. While my code is a bit long to post, it can be emulated using the following example:

a.test = [1 2 3];
result = [];

parfor i = 1:3
  c = a;
  c.test(2) = round( rand() );

  if c.test(2) == 1
     %# Store c in result
  end
end

disp(result.test) %# Should show [1 1 3]

a (and consequently c) is a very large structure, so storing every iteration is not feasible for me (due to memory constraints).

Ideally I would like to be able to store c straight to a variable that I initialise before the parfor loop. Looking at MATLAB’s Example: Using a Custom Reduction Function, I can see it is possible to store a given iteration variable (without having to store every single iteration) yet I do not fully understand their method. I’m baffled by how their function produces 2 separate output variables (seemingly) while the function only defines one output variable.

I am sure this is a very common problem but so far none of my searches have yielded any valid results.

Any help would be much appreciated.

Answer 1

The example you link to is not quite what you want. However, it is possible to store only the “good” results. Here’s a simple example:

A=zeros(0,2); %# initialize A with no rows, two columns
parfor i=1:6
   tmp = rand(1); %# create some result
   if tmp > 0.5 %# store only part of the results
      A = [A;[tmp,i]]; %# store also the iteration number, in case it matters
   end
end
>> A

A =

    0.6497    2.0000
    0.5052    4.0000
    0.9993    5.0000