Pairwise Fairness for Ordinal Regression

This repository contains code for our AISTATS 2022 paper Pairwise Fairness for Ordinal Regression.

In order to reproduce the experiments of Section 5.1 of the paper, do the following:

1. install all required packages as specified in requirements.txt

2. run prepare_drug_consumption.py and prepare_communities_and_crime.py in order to download and prepare the data sets

3. set the parameters at the top of run_experiment.py and run the script in order to produce plots like in Fig. 1 of our paper

Remarks

1. The predictions produced by the POM model trained in Matlab are provided in Results/Predictions_MATLAB. If you have Matlab on your system, you can generate these on your own by running run_POM_model_MATLAB.m.

2. In our paper, we present results averaged over a large number of splits and for a large number of different parameters --- creating these results is quite time consuming and we did so by running our code on numerous machines in parallel. The code provided here is intended to be run on a single machine, and we recommend to run it with parameters different from the ones that we used (see the top of run_experiment.py). Note that this will result in slightly worse results than those reported in our paper.

Citation

If you publish material that uses this code, please cite our paper:

@inproceedings{kleindessner2022fairordreg,
title={Pairwise Fairness for Ordinal Regression},
author={Kleindessner, Matthäus and Samadi, Samira and Zafar, Muhammad Bilal and Kenthapadi, Krishnaram and Russell, Chris},
year={2022},
booktitle={International Conference on Artificial Intelligence and Statistics (AISTATS)}
}

Security

See CONTRIBUTING for more information.

License

This project is licensed under the Apache-2.0 License.