Hello,
As I understand it, TPOT uses DEAP library in the backend. This library lists two options of parallelizing its toolbox computations, either using SCOOP or multiprocessing libraries.
https://github.com/DEAP/deap/blob/232ed17142da1c8bb6a39179018f8278b122aada/doc/tutorials/basic/part4.rst
I am wondering whether this has been considered as an avenue for parallelising TPOT, and if yes, if it's been successful?
Thanks!
I tried to use both SCOOP and multiprocessing to parallelize TPOT but both did not work due to some functions (like lambda func) are not pickleable. So I used pathos to parallelize TPOT and it use dill instead of pickle for function serialization (check this for more details) . Check this PR #338 for a demo for parallelizing TPOT.
Hi,
Thanks! Looks good. The Travis build fails though?
Best,
Francesco
I think it is because the Travis build does not have the pathos module.
On Jan 11, 2017, at 5:54 AM, fferroni notifications@github.com wrote:
Hi,
Thanks! Looks good. The Travis build fails though?
Best,
Francesco—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub, or mute the thread.
We're closing all questions that haven't been updated in a while. Please feel free to re-open this issue if your issue persists.
Most helpful comment
I tried to use both SCOOP and multiprocessing to parallelize TPOT but both did not work due to some functions (like lambda func) are not pickleable. So I used
pathosto parallelize TPOT and it usedillinstead ofpicklefor function serialization (check this for more details) . Check this PR #338 for a demo for parallelizing TPOT.