Yellowbrick is completely broken with sklearn 1.8

[edschofield]

Description

Yellowbrick has hand-rolled functions is_classifier(), is_regressor(), and is_clusterer(), defined in yellowbrick/utils/types.py, do not work with sklearn 1.8.

Scikit-Learn 1.8 removed the attribute _estimator_type on all mixin classes: ClassifierMixin, RegressorMixin, ClusterMixin, DensityMixin, and OutlierMixin. This breaks the introspection method used by Yellowbrick's functions is_classifier() and friends.

To reproduce

uv pip install scikit-learn>=1.8.0 yellowbrick

from sklearn.svm import SVC   # or any other classifier
from yellowbrick.classifier import classification_report
from yellowbrick.datasets import load_bikeshare

X, y = load_bikeshare()
vis = classification_report(SVC(), X, y)

Traceback

File /opt/conda/lib/python3.12/site-packages/yellowbrick/classifier/base.py:122, in ClassificationScoreVisualizer.__init__(self, estimator, ax, fig, classes, encoder, is_fitted, force_model, **kwargs)
    109 def __init__(
    110     self,
    111     estimator,
   (...)    119 ):
    120     # A bit of type checking
    121     if not force_model and not isclassifier(estimator):
--> 122         raise YellowbrickTypeError(
    123             "This estimator is not a classifier; "
    124             "try a regression or clustering score visualizer instead!"
    125         )
    127     # Initialize the super method.
    128     super(ClassificationScoreVisualizer, self).__init__(
    129         estimator, ax=ax, fig=fig, is_fitted=is_fitted, **kwargs
    130     )

YellowbrickTypeError: This estimator is not a classifier; try a regression or clustering score visualizer instead!

Fix

The fix is trivial: simply remove Yellowbrick's implementations if is_classifer() and friends and instead import these functions from sklearn.base.

Affected package versions

• Scikit-Learn 1.8.0
• Yellowbrick 1.5

Implication

The Yellowbrick project no longer works at all with the latest version of its primary dependency. Unless a new Yellowbrick version is released that fixes this, the project is effectively dead. This is a pity, because Yellowbrick is a nicely designed little library and could be more relevant today than ever.