Skip to content

UnderSamplingClassifier

UnderSamplingClassifier

Bases: ClassifierMixin, MetaEstimatorMixin, BaseEstimator

Under Sampling Classifier

It trains an ensemble of estimators on several folds obtained by using all samples of the minority class and undersampling the rest of classes.

It implements the strategy described here.

It is compatible with having a resampler before it, as long as the resampler only performs a partial reduction of the imbalance problem.

Parameters:

Name Type Description Default
estimator estimator object

An estimator object implementing fit and predict_proba

 required
n_jobs int

The number of jobs to use for the computation

 None

Attributes:

Name Type Description
estimators_ list

list of fitted estimators used for predictions.
classes_ array

Class labels
n_classes_ int

Number of classes.
label_encoder_ LabelEncoder object

LabelEncoder object used to encode multiclass labels
n_features_in_ int

Number of features seen during fit. Only defined if the underlying estimator exposes such an attribute when fit.
feature_names_in_ ndarray

Names of features seen during fit. Only defined if the underlying estimator exposes such an attribute when fit.
Source code in nestedcvtraining/under_sampling_classifier.py 
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
class UnderSamplingClassifier(
    ClassifierMixin, MetaEstimatorMixin, BaseEstimator
):
    """ Under Sampling Classifier

    It trains an ensemble of estimators on several folds obtained
    by using all samples of the minority class and undersampling
    the rest of classes.

    It implements the strategy described [here](http://proceedings.mlr.press/v94/ksieniewicz18a/ksieniewicz18a.pdf).

    It is compatible with having a resampler before it, as long as the resampler only
    performs a partial reduction of the imbalance problem.

    Args:
        estimator (estimator object): An estimator
            object implementing `fit` and `predict_proba`

        n_jobs (int, optional): The number
            of jobs to use for the computation

    Attributes:
        estimators_ (list): list of fitted estimators
            used for predictions.
        classes_ (array): Class labels
        n_classes_ (int): Number of classes.
        label_encoder_ (LabelEncoder object): LabelEncoder object
            used to encode multiclass labels
        n_features_in_ (int): Number of features seen during `fit`. Only defined if the
            underlying estimator exposes such an attribute when fit.
        feature_names_in_ (ndarray): Names of features seen during `fit`.
            Only defined if the underlying estimator exposes such an attribute when fit.
    """

    def __init__(self, estimator, *, max_k_under_sampling=5, n_jobs=None):
        self.estimator = estimator
        self.max_k_under_sampling = max_k_under_sampling
        self.n_jobs = n_jobs

    def fit(self, X, y):
        """Fit underlying estimators by undersampling all classes but the minority class.
        Args:
            X (array-like of shape (n_samples, n_features) ): Features
            y (array-like of shape (n_samples,) ): Targets
        Returns:
            self : object
                Instance of fitted estimator.
        """
        self.label_encoder_ = LabelEncoder()
        Y = self.label_encoder_.fit_transform(y)
        self.classes_ = self.label_encoder_.classes_

        counter_classes = Counter(Y)
        minority_class_key = counter_classes.most_common()[-1][0]

        minority_class_idxs = np.where(y == minority_class_key)[0]
        rest_idxs = np.where(y != minority_class_key)[0]

        # K is the imbalanced ratio round to int (with a minimum of 2 and a max of max_k_undersamling)
        imbalance_ratio = (
                len(rest_idxs) / len(minority_class_idxs)
        )
        k_majority_class = int(np.around(imbalance_ratio))
        k_majority_class = k_majority_class if k_majority_class < self.max_k_under_sampling else self.max_k_under_sampling
        k_majority_class = k_majority_class if k_majority_class > 2 else 2

        def under_sampling_iterator():
            splitter = StratifiedKFold(n_splits=k_majority_class)
            for _, index in splitter.split(rest_idxs, y[rest_idxs]):
                idxs = np.concatenate([minority_class_idxs, rest_idxs[index]])
                yield idxs

        self.estimators_ = Parallel(n_jobs=self.n_jobs)(
            delayed(_fit_estimator)(
                self.estimator,
                X[idxs],
                y[idxs]
            )
            for idxs in under_sampling_iterator()
        )

        if hasattr(self.estimators_[0], "n_features_in_"):
            self.n_features_in_ = self.estimators_[0].n_features_in_
        if hasattr(self.estimators_[0], "feature_names_in_"):
            self.feature_names_in_ = self.estimators_[0].feature_names_in_

        return self

    def predict(self, X):
        check_is_fitted(self)
        proba = self.predict_proba(X)
        return self.classes_.take(np.argmax(proba, axis=1), axis=0)

    def predict_proba(self, X):
        check_is_fitted(self)
        mean_proba = np.zeros((X.shape[0], len(self.classes_)))
        for classifier in self.estimators_:
            proba = classifier.predict_proba(X)
            mean_proba += proba
        mean_proba /= len(self.estimators_)
        return mean_proba

    @property
    def n_classes_(self):
        """Number of classes."""
        return len(self.classes_)

fit 

fit(X, y)

Fit underlying estimators by undersampling all classes but the minority class.

Parameters:

Name Type Description Default
X array-like of shape (n_samples, n_features)

Features

 required
y array-like of shape (n_samples,)

Targets

 required

Returns:

Name Type Description
self

object Instance of fitted estimator.
Source code in nestedcvtraining/under_sampling_classifier.py 
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
def fit(self, X, y):
    """Fit underlying estimators by undersampling all classes but the minority class.
    Args:
        X (array-like of shape (n_samples, n_features) ): Features
        y (array-like of shape (n_samples,) ): Targets
    Returns:
        self : object
            Instance of fitted estimator.
    """
    self.label_encoder_ = LabelEncoder()
    Y = self.label_encoder_.fit_transform(y)
    self.classes_ = self.label_encoder_.classes_

    counter_classes = Counter(Y)
    minority_class_key = counter_classes.most_common()[-1][0]

    minority_class_idxs = np.where(y == minority_class_key)[0]
    rest_idxs = np.where(y != minority_class_key)[0]

    # K is the imbalanced ratio round to int (with a minimum of 2 and a max of max_k_undersamling)
    imbalance_ratio = (
            len(rest_idxs) / len(minority_class_idxs)
    )
    k_majority_class = int(np.around(imbalance_ratio))
    k_majority_class = k_majority_class if k_majority_class < self.max_k_under_sampling else self.max_k_under_sampling
    k_majority_class = k_majority_class if k_majority_class > 2 else 2

    def under_sampling_iterator():
        splitter = StratifiedKFold(n_splits=k_majority_class)
        for _, index in splitter.split(rest_idxs, y[rest_idxs]):
            idxs = np.concatenate([minority_class_idxs, rest_idxs[index]])
            yield idxs

    self.estimators_ = Parallel(n_jobs=self.n_jobs)(
        delayed(_fit_estimator)(
            self.estimator,
            X[idxs],
            y[idxs]
        )
        for idxs in under_sampling_iterator()
    )

    if hasattr(self.estimators_[0], "n_features_in_"):
        self.n_features_in_ = self.estimators_[0].n_features_in_
    if hasattr(self.estimators_[0], "feature_names_in_"):
        self.feature_names_in_ = self.estimators_[0].feature_names_in_

    return self

n_classes_ property 

n_classes_()

Number of classes.

Source code in nestedcvtraining/under_sampling_classifier.py 
120
121
122
123
@property
def n_classes_(self):
    """Number of classes."""
    return len(self.classes_)