Source code for simpleml.models.classifiers.sklearn.ensemble

"""
Wrapper module around `sklearn.ensemble`
"""

[docs]__author__ = "Elisha Yadgaran"


import logging

from sklearn.ensemble import (
    AdaBoostClassifier,
    BaggingClassifier,
    ExtraTreesClassifier,
    GradientBoostingClassifier,
    HistGradientBoostingClassifier,
    RandomForestClassifier,
    VotingClassifier,
)

from simpleml.models.classifiers.external_models import ClassificationExternalModelMixin

from .base_sklearn_classifier import SklearnClassifier

[docs]LOGGER = logging.getLogger(__name__)

"""
AdaBoost Classifier
"""

# TODO: experiment with try excepts on different estimator types (feature_importances_, coefficients, ...)


[docs]class WrappedSklearnAdaBoostClassifier(
    AdaBoostClassifier, ClassificationExternalModelMixin
):
    # Boosted classifier doesnt have an easy way to aggregate feature metadata
    pass


[docs]class SklearnAdaBoostClassifier(SklearnClassifier):
[docs]    def _create_external_model(self, **kwargs):
        return WrappedSklearnAdaBoostClassifier(**kwargs)


"""
Bagging Classfier
"""


[docs]class WrappedSklearnBaggingClassifier(
    BaggingClassifier, ClassificationExternalModelMixin
):
    # Bagging classifier doesnt have an easy way to aggregate feature metadata
    pass


[docs]class SklearnBaggingClassifier(SklearnClassifier):
[docs]    def _create_external_model(self, **kwargs):
        return WrappedSklearnBaggingClassifier(**kwargs)


"""
Extra Trees Classifier
"""


[docs]class WrappedSklearnExtraTreesClassifier(
    ExtraTreesClassifier, ClassificationExternalModelMixin
):
[docs]    def get_feature_metadata(self, features, **kwargs):
        feature_importances = self.feature_importances_.squeeze()
        if features is None or len(features) < len(feature_importances):
            LOGGER.warning(
                "Fewer feature names than features passed, defaulting to numbered list"
            )
            features = range(len(feature_importances))
        return dict(zip(features, feature_importances))


[docs]class SklearnExtraTreesClassifier(SklearnClassifier):
[docs]    def _create_external_model(self, **kwargs):
        return WrappedSklearnExtraTreesClassifier(**kwargs)


"""
Gradient Boosting Classifier
"""


[docs]class WrappedSklearnGradientBoostingClassifier(
    GradientBoostingClassifier, ClassificationExternalModelMixin
):
[docs]    def get_feature_metadata(self, features, **kwargs):
        feature_importances = self.feature_importances_.squeeze()
        if features is None or len(features) < len(feature_importances):
            LOGGER.warning(
                "Fewer feature names than features passed, defaulting to numbered list"
            )
            features = range(len(feature_importances))
        return dict(zip(features, feature_importances))


[docs]class SklearnGradientBoostingClassifier(SklearnClassifier):
[docs]    def _create_external_model(self, **kwargs):
        return WrappedSklearnGradientBoostingClassifier(**kwargs)


"""
Random Forest Classifier
"""


[docs]class WrappedSklearnRandomForestClassifier(
    RandomForestClassifier, ClassificationExternalModelMixin
):
[docs]    def get_feature_metadata(self, features, **kwargs):
        feature_importances = self.feature_importances_.squeeze()
        if features is None or len(features) < len(feature_importances):
            LOGGER.warning(
                "Fewer feature names than features passed, defaulting to numbered list"
            )
            features = range(len(feature_importances))
        return dict(zip(features, feature_importances))


[docs]class SklearnRandomForestClassifier(SklearnClassifier):
[docs]    def _create_external_model(self, **kwargs):
        return WrappedSklearnRandomForestClassifier(**kwargs)


"""
Voting Classifier
"""


[docs]class WrappedSklearnVotingClassifier(
    VotingClassifier, ClassificationExternalModelMixin
):
    # Voting classifier doesnt have an easy way to aggregate feature metadata
    pass


[docs]class SklearnVotingClassifier(SklearnClassifier):
[docs]    def _create_external_model(self, **kwargs):
        return WrappedSklearnVotingClassifier(**kwargs)


"""
Histogram-based Gradient Boosting Classifier
"""


[docs]class WrappedSklearnHistGradientBoostingClassifier(
    HistGradientBoostingClassifier, ClassificationExternalModelMixin
):
    # Voting classifier doesnt have an easy way to aggregate feature metadata
    pass


[docs]class SklearnHistGradientBoostingClassifier(SklearnClassifier):
[docs]    def _create_external_model(self, **kwargs):
        return WrappedSklearnHistGradientBoostingClassifier(**kwargs)