gradientinstitute / fastfair

Code for "Fast Fair Regression via Efficient Approximations of Mutual Information"

Fast Fair Regression

This package provides code for the regression models and measures found in the papers,

Steinberg, D., Reid, A., O'Callaghan, S., Lattimore, F., McCalman, L., & Caetano, T. (2020). Fast fair regression via efficient approximations of mutual information. arXiv preprint arXiv:2002.06200.

and

Steinberg, D., Reid, A., & O'Callaghan, S. (2020). Fairness measures for regression via probabilistic classification. arXiv preprint arXiv:2001.06089.

We only provide the minimal code required to implement these models and measures. We have chosen not to provide our complete experimental pipeline as it depends on code we do not own, and do not have permission to redistribute.

Installation

We use python poetry for convenience. We recommend using it as you will more likely be able to get this code to work. However, we also provide a requirements.txt file if you would like to use another package manager or environment. To install this project using poetry, first install it for your system, then issue the following commands,

git clone [e-mail]:gradientinstitute/fastfair.git
poetry install
mkdir data
curl https://archive.ics.uci.edu/ml/machine-learning-databases/communities/communities.data --output data/communities.data
curl https://archive.ics.uci.edu/ml/machine-learning-databases/communities/communities.names --output data/communities.names
poetry shell

You should now be ready to use the models and measures. If you use the requirements.txt file, replace the poetry install and poetry shell commands with those appropriate for your choice of tools.

Running the code

We demonstrate the models and measures in two Jupyter notebooks.

• notebooks/measures.ipynb demonstrates the use of probabilistic classification for measuring fairness in a regression setting. It also reproduces some plots from the "Fairness measures for regression via probabilistic classification" paper.
• notebooks/regression.ipynb demonstrates the use of the fast fair regression models on the "Communities and crime" dataset (which is downloaded in the installation instructions). Each of the fairness models are run with different fairness regularisation strengths to explore how this effects predictive and fairness performance.

To run these notebooks simply type the commands,

cd notebooks
jupyter notebook regression.ipynb

Or substitute regression.ipynb with measures.ipynb on the last line.

License

We provide this code under the Apache 2.0 license. See the LICENSE file for more information.

This code is prototype/research quality only, and has not been developed to the point where it can be used reliably in production.

Copyright 2023 Gradient Institute