Optimization in serial?
esmuigors opened this issue · 4 comments
Dear SimonBlanke,
First of all, thank You for this wonderful project!
My question is: can I somehow specify the optimizers to NOT run in parallel? My function is actually a call to external (parallelized) program, which furthermore cannot be run in multiple instances on the same machine (it is a commercial program). So I would like to prevent the optimizer from initiating next calculations while the first one is not finished.
Best regards,
Igors
Hello @esmuigors,
thank you for your kind words! :-)
I suspect this issue does not belong in this repository, because it is a question about the package itself and not about the tutorial (please correct me if I am wrong about this).
Gradient-Free-Optimizers does not have any parallel processing capabilities. Hyperactive can do parallel processing via multiprocessing, joblib and pathos.
So I will transfer this issue to Hyperactive.
About your question:
Hyperactive will automatically use parallel processing if:
- you set n_jobs > 1
- you added more than 1 search via .add_search(...)
In the first case Hyperactive adds the same search n_jobs-number of times.
In the second case Hyperactive adds different searches.
n_jobs is 1 per default. So you do not need to change this. And you should call .add_search() just once before calling .run(). To not use parallel processing you should write something like this:
# In this case Hyperactive will not use any parallel-processing package to run the optimization.
# (Even if you use population based optimization algorithms)
...
hyper = Hyperactive()
hyper.add_search(model, search_space)
hyper.run()
...Does this answer your question? Let me know if this works for your use case.
Dear @SimonBlanke
I am very sorry for hastily opening this issue (though it was the opportunity to say those kind words :) ). It turned out to be a problem completely unrelated to either gradient_free_optimizers or the Hyperactive... I didn't look at the modification times of the files, then it became obvious :'-) What a shame on me...
I hope that You answer will be useful for someone else at least :)
Best wishes,
Igors
Hello @esmuigors,
no Problem! I think the question and answer is still useful for others.
If you have more questions don't hesitate to open another issue.