Parfor loop and variable length output

4 visualizzazioni (ultimi 30 giorni)
Hello everyone,
I would like to speed-up a code using a parfor loop but my function is not straightforwardly suitable for it. In particular, the problem is the for loop contains a function whose output has a variable length and I need to store all the variable outputs in a final vector.
I can predict the maximum length of the final result, but most likely I will only use one tenth of this maximum.
result = zeros(1,10000);
N_result = 0;
for i = 1:N
result_i = generate_result(i); % This has a variable number of elements
result(N_result+1:numel(result_i) = result_i;
N_result = N_result+numel(result_i);
end
I would like to use the parfor instead of a for, but I don't know how merge the variable result_i in a single output result. It would be perfect if I could create a variable length vector for each worker, and then merge those vectors.
How can I do this?
Alessandro.

Risposta accettata

Edric Ellis
Edric Ellis il 26 Feb 2015
You have several options here. Probably the simplest is to return a cell array and then concatenate the contents. It might not be terribly efficient though if N is large compared to the number of outputs you're returning (this is because cell arrays where each cell contains only a small number of elements have memory and performance overhead compared to a numeric array).
Here's an example. (Note the use of feval to work around a PARFOR restriction on using function handles)
% Here's a simple function that generates a random array of a random size:
generate_result = @(x) x * rand(1, randi([0,10]))
parfor idx = 1:7
output{idx} = feval(generate_result, idx);
end
% Now, concatenate the output elements into a single array
output = [output{:}];
  2 Commenti
Alessandro Masullo
Alessandro Masullo il 27 Feb 2015
Thank you for the answer. Just out of curiosity, is this the proper way of doing what I want to do? Does the feval affect the speed?
Thank you.
Edric Ellis
Edric Ellis il 2 Mar 2015
Modificato: Edric Ellis il 2 Mar 2015
You don't normally need to use feval - I needed it only because generate_result is a function handle, and there's a limitation with PARFOR which means you must invoke those using feval. In your case, if generate_result is an ordinary function, you shouldn't use feval.

Accedi per commentare.

Più risposte (0)

Categorie

Scopri di più su Parallel for-Loops (parfor) in Help Center e File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by