ksfeatureselector 0.3.0

Scikit-learn compatible feature selection for binary and multi-class classification using the Kolmogorov-Smirnov (K-S) test, with pairwise / one-vs-rest comparison and Fisher / min / max p-value aggregation.

KSFeatureSelector is a scikit-learn compatible feature selector that ranks features by how well they separate the classes of a binary or multi-class target, using the two-sample Kolmogorov-Smirnov (K-S) test. It subclasses scikit-learn’s SelectorMixin, passes check_estimator, and plugs directly into Pipeline and GridSearchCV.

Features

• Ranks features by their K-S test p-value (lower p-value is more discriminative).

• Handles binary and multi-class targets (2 to 10 classes).

• Two class-comparison strategies for multi-class targets:

  • pairwise: K-S test between every pair of classes.

  • one-vs-rest: each class against the rest.

• Three p-value aggregation methods: fisher (default), min, max.

• Select features by a count (top_n) or a p-value threshold (top_p).

• Full scikit-learn API: fit, transform, get_support, get_feature_names_out, inverse_transform.

• A select_ks_features convenience function for quick one-off selection.

Installation

pip install ksfeatureselector

Usage

import numpy as np
from ksfeatureselector import KSFeatureSelector

rng = np.random.RandomState(0)
X = rng.normal(size=(200, 5))
y = (X[:, 0] + X[:, 1] > 0).astype(int)

selector = KSFeatureSelector(top_n=2).fit(X, y)
X_reduced = selector.transform(X)
print(selector.get_support())
print(selector.get_feature_p_values())

In a scikit-learn pipeline:

from sklearn.pipeline import Pipeline
from sklearn.linear_model import LogisticRegression

pipe = Pipeline([
    ("ks", KSFeatureSelector(top_p=0.05)),
    ("clf", LogisticRegression()),
])
pipe.fit(X, y)

Convenience function for DataFrames:

from ksfeatureselector import select_ks_features

selected = select_ks_features(
    df, x_cols=["f1", "f2", "f3"], y_var="target",
    top_p=0.01,
    aggregation_method="one-vs-rest",
    p_value_aggregation_method="min",
)

Parameters

• top_n (int, optional): keep this many top-ranked features.

• top_p (float in [0, 1], optional): keep features whose aggregated p-value is at most this value.

• aggregation_method ({"pairwise", "one-vs-rest"}): class comparison strategy for multi-class targets.

• p_value_aggregation_method ({"fisher", "min", "max"}): per-feature p-value aggregation method.

top_n and top_p are mutually exclusive. If neither is set, all features are kept (ranked by p-value).

License

BSD 3-Clause License.

Author

V Subrahmanya Raghu Ram Kishore Parupudi Email: [e-mail]