FairBayes-DDP+

Codebase and Experiments for "Minimax Optimal Fair Classification with Bounded Demographic Disparity"

Codebase Overview

This repository provide simulation codes of our paper "Minimax Optimal Fair Classification with Bounded Demographic Disparity". Simulation codes for synthetic data is contained in document Synthetic_data'' and Simulation codes for real data is contained in document Real_data''.

Algorithms Considered in Read data analysis

This repository provide python realization of 5 benchmark methods of fair classification.

--FairBayes-DDP+: X. Zeng, G. Cheng, and E. Dobriban. Minimax Optimal Fair Classification with Bounded Demographic Disparity.

--KDE based constrained optimization (KDE)： J. Cho, G. Hwang, and C. Suh. A fair classifier using kernel density estimation.

--Adversarial Debiasing (ADV): B. H. Zhang, B. Lemoine, and M. Mitchell. Mitigating unwanted biases with adversarial learning.

--Post-processing through Flipping (FFP) W. Chen, Y. Klochkov, and Y. Liu. Post-hoc bias scoring is optimal for fair classification.

--Post-processing through Optimal Transport (PPOT) R. Xian, L. Yin, and H. Zhao. Fair and optimal classification via post-processing.

Related Works and Repositories

This repository has draw lessons from other open resourses.

--Codes for ADV take inspiration from the AI Fairness 360 platform: https://github.com/Trusted-AI/AIF360;

--Codes for KDE follows the original code provided by: J. Cho, G. Hwang, and C. Suh. A fair classifier using kernel density estimation;

--Codes for PPOT take inspiration from: https://github.com/rxian/fair-classification;

--Codes for PPF take inspiration from the paper: W. Chen, Y. Klochkov, and Y. Liu. Post-hoc bias scoring is optimal for fair classification.

Data

This repository uses the AdultCensus. It can be found in the Datasets folder and are loaded using dataloader.py.