maddlib documentation

maddlib is a Python library for algorithmic fairness assessment and mitigation with the MADD metric. MADD, which stands for Model Absolute Density Distance, is a fairness metric based on the distance of the distributions of the probability estimates or confidence scores of machine learning classifiers or regressors (except SVM models).

Source code on GitHub and PyPI.

Can be used with popular ML libraries such as Pandas and Scikit-learn.

Note

This project is under active development.

Installation

(.venv) $ pip install maddlib

or

(.venv) $ python3 -m pip install maddlib

If you have an error, please check first that your Python version is greater than 3.10.4.

Upgrade

(.venv) $ pip install --upgrade maddlib

or

(.venv) $ python3 -m pip install --upgrade pip

Contents:

• Evaluation
  • Compute MADD
  • Display MADD results
• Optimal bandwidth
  • Generate bandwidth search space
  • Find the optimal bandwidth range
  • Display optimal bandwidth range
• Post-processing
• Released versions
  • maddlib package
  • documentation