Fair Multiple Decision Making Through Soft Interventions

This repository is the official implementation of Fair Multiple Decision Making Through Soft Interventions.

Overview

• data contrains the data
• models contains the codes for four models.
• synthetic contains the experiments on synthetic dataset.
• adult contains the experiments on Adult dataset.
• results contains the generated result files.

Usage

• To reproduce the experiments on synthetic dataset, get into the synthetic directory and run following commands:

python synthetic_main.py --model unconstrained_model             # run Unconstrained method
python synthetic_main.py --model separate_constrained_model      # run Separate method
python synthetic_main.py --model serial_constrained_model        # run Serial method
python synthetic_main.py --model joint_constrained_model         # run Joint method

• To reproduce the experiments on Adult dataset, get into the adult directory and run following commands:

python adult_processing.py                                       # generate preprocessed data

python adult_main.py --model unconstrained_model                 # run Unconstrained method
python adult_main.py --model separate_constrained_model          # run Separate method
python adult_main.py --model serial_constrained_model            # run Serial method
python adult_main.py --model joint_constrained_model             # run Joint method

Requirements and dependencies

This project was designed with Python 3.7. We can not guarantee and have not tested compability with Python 2.7.

Required python packages:

• numpy
• scipy
• pandas
• logging
• argparse
• scikit-learn
• imblearn
• cvxpy
• pytorch

Results

Citation

@article{hu2020fair,
  title={Fair Multiple Decision Making Through Soft Interventions},
  author={Hu, Yaowei and Wu, Yongkai and Zhang, Lu and Wu, Xintao},
  journal={Advances in Neural Information Processing Systems},
  volume={33},
  year={2020}
}