imbalanced-calibrate - Analytical Calibration for Imbalanced Learning

This package provides sklearn-compatible calibration methods which correct model output for the bias induced by certain imbalanced learning techniques. It supports class weighting in sklearn binary classifiers (including XGBoost and LightGBM), and resampling methods from the imbalanced-learn package.

Documentation

Installation instructions, usage examples and mathematical justification can be found in the documentation.

Prerequisites

imbalanced-calibrate requires the following dependencies:

• Python (>=3.10)
• NumPy (>=2.0.2)
• Scikit-learn (>=1.6.0)

Additionally, imbalanced-calibrate requires the following optional dependencies:

• Imbalanced-learn (>=0.14.2), for calibration after using imbalanced-learn resampling methods.

Install

imbalanced-calibrate is currently available on the PyPI repository and you can install it via pip:

pip install imbalanced-calibrate

Or with uv:

uv add imbalanced-calibrate

The optional dependencies for resampling methods can be installed using pip install imbalanced-calibrate[resampling] or uv add "imbalanced-calibrate[resampling]".

Contribution

You can contribute to this package through a Pull Request, subject to appropriate unit testing and review. If you have any feature requests or suggestions, or encounter any errors or unexpected behaviour, please raise them either in issues or discussions. If you use this package in your work, please do let me know!