Software
- FairCORELS: Open-source python module for learning fair rule lists.
- PyPI project page: here
- Source code repository: here
- Paper: original method introduced here, and presented in a demo paper at the CIKM 2021 conference
- A hands-on tutorial and examples: here
- Short description: FairCORELS is a learning algorithm building rule lists (that are inherently interpretable models) under statistical/group fairness constraints. It is an extension of the CORELS algorithm, modified to also take into account fairness considerations. More precisely, given a particular notion of fairness, FairCORELS builds a rule list minimizing CORELS' original objective function and meeting the fairness requirement.
- FairCORELSV2: New version of the FairCORELS package, specifically designed to learn optimal fair models, and embedding advanced techniques to efficiently explore the search space of fair rule lists.
- PyPI project page: here
- Source code repository: here
- Paper: here (version accepted at the CPAIOR 2022 conference)
- Short description: FairCORELSV2 embedds a novel pruning method, leveraging Mixed Integer Linear Programming to prune the search space of fair rule lists efficiently, considering jointly the objective function and the fairness constraint. It also provides a new prefix permutation map that can be used to speed up the learning process while preserving the guarantee of optimality.