/FairInfl

Primary LanguagePythonMIT LicenseMIT

Understanding Instance-Level Impact of Fairness Constraints

This code is a Jax implementation of the ICML 2022 paper: Understanding Instance-Level Impact of Fairness Constraints

The code structure is as follows:

  • train.py - the code to train a model subject to fairness constraints
  • test.py - utils to evaluate the metrics
  • models.py - specify the models
  • metrics.py - the loss function, fairness constraints, accuracy, and fairness measures
  • data.py - data loaders
  • scores.py - compute the fairness influence scores
  • gradients.py - utility function for easy gradients
  • recorder.py - recording the results
  • utils.py - other miscellaneous functions

This code is partially adapted from the Github repo Data Diet.