sklearn-deap icon indicating copy to clipboard operation
sklearn-deap copied to clipboard

Published 20 hours ago • verified publisher icon rsteca

Fixed Windows multiprocessing | Added pretty README plots | Issue with AttributeError: can't set attribute

Open flipdazed opened this issue 6 years ago • 3 comments

  • Solved problem with Windows multiprocessing
  • fixed issue with AttributeError: can't set attribute: https://github.com/rsteca/sklearn-deap/issues/44
  • added README plots

img

flipdazed avatar Jun 08 '18 15:06 flipdazed

Do you think this is ready for a merge?

rsteca avatar Jun 14 '18 21:06 rsteca

Yes I don’t plan on making more changes to do with getting it working on windows.

What I would like to do in another merge would be to change eaSimple to the CMA-ES strategy. I think it would be a lot more effective. It requires a bit of work though as the DEAP default CMA-ES strategy is only set up for continuous variables it seems.

flipdazed avatar Jun 15 '18 01:06 flipdazed

hello Alexander, thanks for fixing the code and posting examples. It does not seem parallel processing on Evolutionary Search is working properly though.

When I run sklearn out of the box GridSearchCV I see many parallel Python process (multiple entries in Task Manager) -- as many as parallel jobs I have requested (thru n_jobs) . Incidentally I can set a number greater than my cores and If I do My Win10 machine runs at 100% CPU, as is expected.

When I switch to Evolutionary Search, and I repeat the experiment all else being equal, my task manager only shows 1 instance of python, my machine runs at very low CPU and the search takes a long time to complete.

All this seems to point to some underlying problem with the way multi processing is done by Evo Search. Do you have the same. Can you get your machine running parallel processes with Evolutionary Search? Thanks

luded avatar Jun 21 '18 06:06 luded