[dask] use more specific method names on _DaskLGBMModel

[jameslamb]

The Dask sklearn estimators in lightgbm.dask inherit from both their lightgbm.sklearn equivalent and a mixin called _DaskLGBMModel.

LightGBM/python-package/lightgbm/dask.py

Line 525 in 7880b79

class DaskLGBMClassifier(LGBMClassifier, _DaskLGBMModel):

Those equivalents from lightgbm.sklearn inherit from a mix of LightGBM internal classes and some mixins brought in from sklearn.

from lightgbm.sklearn import DaskLGBMClassifier
for c in DaskLGBMClassifier.__mro__:
    print(c)

# <class 'lightgbm.dask.DaskLGBMClassifier'>
# <class 'lightgbm.sklearn.LGBMClassifier'>
# <class 'lightgbm.sklearn.LGBMModel'>
# <class 'sklearn.base.BaseEstimator'>
# <class 'sklearn.base.ClassifierMixin'>
# <class 'lightgbm.dask._DaskLGBMModel'>
# <class 'object'>

Since LGBMClassifier is the first parent class in the method resolution order ("MRO") for DaskLGBMClassifier, if it or any of the parent classes from sklearn introduce a method that conflicts with one of the method names in _DaskLGBMModel, the _DaskLGBMModel method won't be used, which could break DaskLGBMClassifier.

To protect against this happening, this PR proposes changing the method names on _DaskLGBMModel to names that are less likely to conflict with sklearn in the future. For example, it proposes changing _fit() to _lgb_dask_fit(), so that if an sklearn.base.BaseEstimator._fit() is introduced in some future version of sklearn, DaskLGBMClassifier will still use the correct method.

All of the methods changed here are internal, so this should not have any user-facing impact.

Notes for reviewers

I also considered the pattern below for absolute certainty:

class DaskLGBMClassifier(LGBMClassifier, _DaskLGBMModel)
    def fit(self, ...):
        _DaskLGBMModel._fit(self, ...)

But I don't understand the full implications of doing that, so I decided against it.

[StrikerRUS]

Makes sense! Thank you!

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.