RoCourseNet: Distributionally Robust Training of a Prediction Aware Recourse Model

This repo contains code to reproduce our paper published at CIKM 2023.

To cite this paper:

@inproceedings{guo2023rocoursenet,
author = {Guo, Hangzhi and Jia, Feiran and Chen, Jinghui and Squicciarini, Anna and Yadav, Amulya},
title = {RoCourseNet: Robust Training of a Prediction Aware Recourse Model},
year = {2023},
isbn = {9798400701245},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3583780.3615040},
doi = {10.1145/3583780.3615040},
booktitle = {Proceedings of the 32nd ACM International Conference on Information and Knowledge Management},
pages = {619–628},
numpages = {10},
keywords = {explainable artificial intelligence, adversarial machine learning, counterfactual explanation, algorithmic recourse, interpretability},
location = {Birmingham, United Kingdom},
series = {CIKM '23}
}

Install

This project uses jax-relax (a fast and scalable recourse explanation library). Ths library is highly scalable and extensible, which enables our experiments to be finished within 30 minutes. In contrast, a pytorch implementation of RoCourseNet takes around 12 hours to run.

pip install -e ".[dev]" --upgrade

Run Experiments

Running scripts.experiment.py with different arguments will reproduce results in our paper. For example,

1. Train and Evaluate RoCourseNet on Loan Application Dataset:

python -m scripts.experiment.py -d loan

2. Train and Evaluate CounterNet on Loan Application Dataset:

python -m scripts.experiment.py -m CounterNet -d loan

3. Train and Evaluate ROAR on Loan Application Dataset:

python -m scripts.experiment.py -m ROAR -d loan