Particleswarm connection to parallel workers
1 次查看(过去 30 天)
显示 更早的评论
When 'UseParallel' is set to true for particleswarm optimization, it seems that the optimizer needs to be 'Connected to parallel workers' quickly enough by obtaining a value for the objective function for each worker or particle. Otherwise I've been getting the error:
Error Caused by:
Error using fcnvectorizer
Function to evaluate must be represented as a string scalar, character vector, or function_handle object.
Failure in user-supplied fitness function evaluation. GA cannot continue.)
Is there a way to increase that time for all parallel workers to be connected for objective functions that take a long time to evaluate?
1 个评论
Sam Marshalik
2023-9-6
@ss_19, When the optimization is run in parallel there is some data transfer that needs to occur to send data from the MATLAB client to the parallel process workers. I have not seen a scenario where that data transfer has caused failures like this.
Can you include the code you are running to generate this error?
Does this error occur when you run it in serial? If there is an issue with the transfer component, you can try using parpool("Threads", NumWorkers) to start thread based workers, which should not need to transfer any data between the MATLAB client and the workers.
回答(0 个)
另请参阅
类别
在 Help Center 和 File Exchange 中查找有关 Optimization Toolbox 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!