Parpool in Pattern search with temp directories

3 次查看（过去 30 天）
Pantelis
Pantelis 2025-3-15
评论： Pantelis 2025-4-21
I am currently trying to parpoll in pattern search. The problem is that my objective function is editing a txt file and runs a simulation. I have tried to make a separate directory for each worker but it doesnt seem to work as expected. this is my code right now.
delete(gcp('nocreate')); % Clean up any existing pool
pool = parpool; % Start parallel pool
numWorkers = pool.NumWorkers; % Get the number of parallel workers
% Define bounds
lb = [0; 0; 0.5; 2/3; 0.8; 0.75];
ub = [1.3; 4; 2; 4/3; 1.1; 1.1];
% Define initial guess for each worker
x0_all = repmat([1 1 1 1 1 1], numWorkers, 1); % Same x0 for all workers initially
% Perturb each worker's x0 slightly to avoid conflict
rng('shuffle'); % Set random seed based on the current time
for i = 1:numWorkers
x0_all(i,:) = x0_all(i,:) + 0.1 * randn(1, 6); % Perturb by small random values
end
% Path to bin folder (adjust this path if necessary)
bin_folder = fullfile(pwd, 'bin');
% Parallel execution
parfor i = 1:numWorkers % Iterate over workers
workerID = i;
% Create unique temporary directory for each worker
tmpdir = fullfile(pwd, sprintf('tmpdir%d', workerID)); % Unique temp folder
mkdir(tmpdir); % Create directory for worker
% Copy the 'bin' folder into the worker's temporary directory
copyfile('bin', tmpdir);
% Change the current directory to the worker's temporary directory
cd(tmpdir);
% Get the initial guess for the worker
x0 = x0_all(workerID, :); % Get the initial guess for this worker
try
% Run pattern search optimization with the unique initial guess
options = optimoptions('patternsearch', ...
'UseParallel', true, ... % Enable parallel execution
'PlotFcn', @psplotbestf); % Plot best function value
[x_opt, fval] = patternsearch(@Motionview, x0, [], [], [], [], lb, ub, [], options);
% Display results
fprintf('Worker %d - Optimized variables: %s\n', workerID, mat2str(x_opt));
fprintf('Worker %d - Optimal function value: %.6f\n', workerID, fval);
catch ME
fprintf('Worker %d encountered an error: %s\n', workerID, ME.message);
end
% Clean up: Remove the worker directory after the computation
rmdir(workDir, 's'); % Delete worker's folder after work is done
end
The process continous but i get these warnings:Warning: Unable to create a parallel pool. Continuing with evaluations in serial. To avoid this warning, set the 'UseParallel' option to false.
> In setOptimFcnHandleOnWorkers (line 98)
In patternsearch (line 278)
Note that the Motionview objective function, need the necessary file which are located in bin folder, thats why its getting copied.

请先登录，再回答此问题。

回答（1 个）

Matt J
Matt J 2025-3-15
编辑：Matt J 2025-3-15
The process continous but i get these warnings:Warning: Unable to create a parallel pool. Continuing with evaluations in serial. To avoid this warning, set the 'UseParallel' option to false.
You should heed the warning and set UseParallel to false.
I have tried to make a separate directory for each worker but it doesnt seem to work as expected.
We don't know what "as expected" means here. Are the directories not getting created? Are the workers overwriting each other? If the processing is serial as the warning claims, that might explain it.
  3 个评论
显示 1更早的评论
Matt J
Matt J 2025-3-15
That is interesting. Was it fixed by setting UseParallel=true?
Pantelis
Pantelis 2025-4-21
No, I approached it differently. I implemented a piece of code within my objective function that detected which worker was involved in the process and automatically created a directory named after that worker. Inside each directory, all the necessary files were stored. This way, each worker had their own dedicated folder, avoiding any confusion and ensuring proper file organization.

请先登录，再进行评论。

类别

在 Help Center 和 File Exchange 中查找有关 Startup and Shutdown 的更多信息

标签

产品


版本

R2024b

Community Treasure Hunt

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

Start Hunting!

Translated by