Conversation
ryan-deak-zefr
left a comment
There was a problem hiding this comment.
LGTM. Are these changes backward compatible before sklearn 0.24?
zexuan-zhou
left a comment
There was a problem hiding this comment.
why do we use a dev version of sklearn instead of waiting for an official release?
| estimator, | ||
| estimator_params, # parameters | ||
| scoring=make_scorer(metric, greater_is_better, needs_proba), | ||
| scoring=make_scorer(score_func=metric, greater_is_better=greater_is_better, needs_proba=needs_proba), |
There was a problem hiding this comment.
Just checked. This should be backward compatible in sklearn 0.23.x
https://github.com/scikit-learn/scikit-learn/blob/0.23.1/sklearn/metrics/_scorer.py#L534
Dask dask-ml maintainers use the dev build for their CI. See below: https://github.com/dask/dask-ml/blob/7f6dcb5b1082f5ef9f53217809efac21e0232f7f/dask_ml/_compat.py#L20 Dask-ml build logs here: https://dev.azure.com/dask-dev/dask/_build/results?buildId=1514&view=results |
Yes they are I tested with 0.23.2 and everything passed except for the two marked xfail requiring the scikit-learn 0.24.dev0 nightly build. |
Got it. Thank you! |
================ 663 passed, 8 skipped, 14 xfailed, 2 xpassed, 142 warnings in 214.74s (0:03:34) =================
(dask-ml-dev-skl024)
After installing Scikit-learn nightly build into dask-ml's ci latest env via:
pip install --pre --extra-index https://pypi.anaconda.org/scipy-wheels-nightly/simple scikit-learn==0.24.dev0