RoCourseNet: Distributionally Robust Training of a Prediction Aware Recourse Model

This repository implements paper "RoCourseNet: Distributionally Robust Training of a Prediction Aware Recourse Model" (submitted to Neurips 2022).

Requirements

1. install pytorch
2. install all dependencies

pip install -e

Training

# train RoCourseNet
python -m scripts.train_cfnet --n_steps 13 
# train CounterNet
python -m scripts.train_cfnet --model_name CounterNet
# train RoCourseNet on the German Credit dataset
python -m scripts.train_cfnet --data_name german_credit

Evaluation

Evaluate CF Baselines

python -m scripts.eval_cf_baselines

Evaluate VDS Attacker

python -m scripts.eval_attacker --data_name "loan" --model_name AdvCounterNet

Pretrained Models

See assets/weights

Useful nbdev commands

Build nbs to module

nbdev_build_lib

Update nbs from module

nbdev_update_lib

clean notebooks

nbdev_clean_nbs