GlobalSearch inside parfor loop

Question

Tristan Potter 2020-7-2

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/558442-globalsearch-inside-parfor-loop

回答： Sean de Wolski 2020-7-8

Hi everyone,

I would like to loop in parallel over a parameter of a numerical optimization problem that I'm solving via GlobalSearch. However, when I try to implement a simple parfor loop, I get the following error message:

"The function RUN might make an invalid workspace access inside the PARFOR loop."

Is there a simple workaround for this?

Thanks.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Sean de Wolski 2020-7-8

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/558442-globalsearch-inside-parfor-loop#answer_462848

Take everything inside the parfor loop and refactor it to a local or separate file function. Parfor loops should be as simple as possible:

parfor ii = 1:10
    r(ii) = runGS(parameter(ii))
end

And

function r = runGS(param)
    % Call global search etc.
end

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

Answer 2

Gouri Chennuru 2020-7-8

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/558442-globalsearch-inside-parfor-loop#answer_462746

Hi Tristan,

You can go through all the information provided and thus get a clear idea about the problem you are trying to solve.

I am not so sure about the model you are working on, but these might help you with the solution and eliminate the error.

If your Simulink model requires access to variables contained in a .mat file, you must load these parameters in the workspace of each worker. You must do this before the parfor-loop, and after opening parpool
if your model also requires variables defined in the body of your MATLAB script, you must use assignin or evalin to move these variables to the base workspace of each worker, in every parfor iteration, also ensure the transparency and workspace access issues in parfor loops

You can refer to the link which provides detailed information about troubleshooting variables using parfor loops and solving variable classification issues in parfor loops.

Here is the link which guides you with different parallel processing types. In case of global search it always runs the fmincon local solver, fmincon can estimate gradients by parallel finite differences.

Hope this Helps!