scikit-feature

An open-source feature selection repository in Python, originally developed by the Data Mining and Machine Learning (DMML) Lab at Arizona State University. This fork keeps the library compatible with modern versions of scikit-learn.

scikit-feature serves as a platform for feature selection application, research, and comparative study. It shares widely used feature selection algorithms and offers researchers and practitioners a convenient way to perform empirical evaluation when developing new feature selection methods.

Why scikit-feature?

• Comprehensive

  Over 35 feature selection algorithms across information-theoretic, similarity-based, sparse-learning, statistical, structure, wrapper, and streaming categories.

• Easy to use

  Clean, scikit-learn compatible API. Use any scoring function directly with SelectKBest, or drop streamwise/wrapper methods into your existing pipelines.

• Well documented

  Every algorithm has its own reference page with usage examples, parameters, and references to the original papers.

• Actively maintained

  This fork keeps the original algorithms working against current versions of scikit-learn, numpy, and pandas (Python 3.10+).

Getting started

pip install skfeature-chappers

from sklearn.datasets import load_iris
from sklearn.feature_selection import SelectKBest

from skfeature.function.statistical_based import t_score

X, y = load_iris(return_X_y=True)
selector = SelectKBest(score_func=t_score.t_score, k=2)
X_selected = selector.fit_transform(X, y)

Head over to the Getting Started guide for installation instructions and a Quick Start tutorial, or browse the algorithm reference for the full list of available methods.

Project information

• Forked project site — https://github.com/NoRaincheck/scikit-feature
• Original project site — https://github.com/jundongl/scikit-feature
• Original documentation — http://featureselection.asu.edu/

License

This project is licensed under the GNU General Public License v2.0.