[WIP] Resolved #195 - removing sklearn.cross_validation DeprecationWarning issue

[dvro]

Adapted StratifiedKFold import doing the following:

import sklearn

if hasattr(sklearn, 'model_selection'):
    from sklearn.model_selection import StratifiedKFold
else:
    from sklearn.cross_validation import StratifiedKFold

now, the following command:
from imblearn.ensemble import EasyEnsemble

do not raise the following deprecation warning:

DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.
  "This module will be removed in 0.20.", DeprecationWarning)

[coveralls]

Coverage decreased (-0.03%) to 98.754% when pulling c3900a6 on dvro:deprecation_warning into 7a5afeb on scikit-learn-contrib:master.

[glemaitre]

@dvro thanks, can you check the problem with appveyor

[coveralls]

Coverage decreased (-0.03%) to 98.754% when pulling 69848ef on dvro:deprecation_warning into 7a5afeb on scikit-learn-contrib:master.

[coveralls]

Coverage decreased (-0.05%) to 98.727% when pulling 2b653d0 on dvro:deprecation_warning into 7a5afeb on scikit-learn-contrib:master.

[dvro]

@chkoar @glemaitre this is just a quickfix, not a definitive solution (IMO, we should support sklearn 0.18.X forward, only).

[glemaitre]

LGTM. @dvro can you had your contribution in the doc/whats_new.rst

@chkoar can you make the merging as MRG+1

[chkoar]

I think that importing the iterator locally would avoid the double check.

[glemaitre]

Not sure that you follow the pep8 standard doing so.

[chkoar]

In general yes. But pep8 it is not a panacea. For these purposes I think that's ok. For example check here a local import https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/base.py#L348

[glemaitre]

Ok, but we need to keep track somewhere such that we don't forget about it. I would say in the TODO or whatsnew would be the right place.

[chkoar]

Another option could be that dvro@4225dff

[coveralls]

Coverage decreased (-0.02%) to 98.787% when pulling 2136348 on dvro:deprecation_warning into 62f6d2f on scikit-learn-contrib:master.

[dvro]

@chkoar that's a cool approach, however, if one tries to the cv_split method anywhere else and the the check_X_y doesn't pass, it won't be clear to the user that this is the problem. maybe using the if else inside the cv_split method?

[chkoar]

I don't mind. Make your changes to see the diff. However, if there is a need in the future to use this approach elsewhere we'll pull out this function to a utils or something module. It is a private function.

[coveralls]

Coverage decreased (-0.05%) to 98.761% when pulling c8260ed on dvro:deprecation_warning into 62f6d2f on scikit-learn-contrib:master.

[glemaitre]

LGTM.

[glemaitre]

@dvro @chkoar If you can in the whatsnew then we can merge this PR

[chkoar]

@glemaitre he has changed the title of the PR back to [WIP] so lets wait