Perturbation Classifiers

Perturbation Classifiers is an easy-to-use library focused on the implementation of the Perturbation-based Classifier (PerC) [1] and subconcept Perturbation-based Classifier (sPerC). The library is is based on scikit-learn, using the same method signatures: fit, predict, predict_proba and score.

Installation:

The package can be installed using the following command:

# Clone repository
git clone https://github.com/rjos/perturbation-classifiers.git
cd perturbation-classifiers/

# Installation lib
python setup.py install

Dependencies:

perturbation_classifiers is tested to work with Python 3.7. The dependencies requirements are:

• scikit-learn(>=0.24.2)
• numpy(>=1.21.2)
• scipy(>=1.7.1)
• matplotlib(>=3.4.3)
• pandas(>=1.3.2)
• gap-stat(>=2.0.1)
• gapstat-rs(>=2.0.1)

These dependencies are automatically installed using the command above.

Examples:

Here we show an example using the PerC method:

from perturbation_classifiers import PerC

# Train a PerC model
perc = PerC()
perc.fit(X_train, y_train)

# Predict new examples
perc.predict(X_test)

and here we show an example using the sPerC method:

from perturbation_classifiers.subconcept import sPerC

# Train a sPerC model
sperc = sPerC()
sperc.fit(X_train, y_train)

# Predict new examples
sperc.predict(X_test)

References: