parallel execution for permutation_importance

[luoq]

No description provided.

[codecov-io]

Codecov Report

Merging #244 into master will decrease coverage by 0.38%.
The diff coverage is 83.33%.

@@            Coverage Diff             @@
##           master     #244      +/-   ##
==========================================
- Coverage   97.26%   96.88%   -0.39%     
==========================================
  Files          44       44              
  Lines        2851     2855       +4     
  Branches      541      542       +1     
==========================================
- Hits         2773     2766       -7     
- Misses         41       48       +7     
- Partials       37       41       +4

Impacted Files	Coverage Δ
eli5/sklearn/permutation_importance.py	100% <100%> (ø)	⬆️
eli5/permutation_importance.py	90.62% <80%> (-9.38%)	⬇️
eli5/sklearn/utils.py	87.32% <0%> (-2.82%)	⬇️
eli5/lime/utils.py	85.13% <0%> (-1.36%)	⬇️
eli5/xgboost.py	98.17% <0%> (-1.22%)	⬇️
eli5/lime/lime.py	96% <0%> (-1%)	⬇️

[kmike]

I think you can add # type: ignore to this line to make type checks pass

[kmike]

Thanks, looks like a great feature!

Tests are failing in Python 2.7; I'm not sure what it takes to make it work. Does it at least work with n_jobs=1?

[luoq]

It seems in py2 instance method cannot be pickled in default. https://gist.github.com/luoq/703ec045827caf5a504ff97376db9c3e
https://stackoverflow.com/questions/1816958/cant-pickle-type-instancemethod-when-using-multiprocessing-pool-map?rq=1

[luoq]

It's fixed after python3.5:
https://stackoverflow.com/questions/45901505/python3-6-can-pickle-instance-method-staticmethod-classmethod-but-document-say

[thomasfrederikhoeck]

Are you still working on this? Would be very useful.

[eyadsibai]

will you drop python2 soon, so this pr gets merged?

[eyadsibai]

shall we switch this to use https://joblib.readthedocs.io/en/latest/parallel.html instead?

[aridf]

Can I ask what happened to this? Is there now a way to use eli5.sklearn.PermutationImportance() in parallel?