FaUCI: Fairness Under Constrained Injection

Description

This repo contains the experiments for the paper mentioned in the title.

Requirements

The code is written in Python 3.10.0. To install the required packages, run:

pip install -r requirements.txt

All experiments were run on a MacBook Pro Apple M1 chip with 16GB of RAM.

Organisation

The code is organised as follows:

• dataset/ contains the dataset used in the experiments along with the code for its loading and preprocessing.
• fairness/ contains the code for the fairness metrics of our method (FaUCI), cho and jiang methods.
• analysis/ contains the code for executing and gathering of the results.
• images/ contains the code to generate images (some of them are included in the paper).
• configuration.py contains the setup of the experiments.

Concerning Cho and Jiang methods, we used the code provided by the authors of the papers. The code is available at the following links:

• Cho: https://proceedings.neurips.cc/paper_files/paper/2020/file/ac3870fcad1cfc367825cda0101eee62-Supplemental.zip
• Jiang: https://github.com/zhimengj0326/GDP

Usage

To run the experiments, execute the following command:

python fairness/our/__main__.py

The results will be saved in fairness/our/log folder. Similarly, for cho and jiang methods, run:

python fairness/cho/__main__.py

python fairness/jiang/__main__.py

To analise the results for each method, run:

python analysis/our/__main__.py

python analysis/cho/__main__.py

python analysis/jiang/__main__.py

To generate comparison images, run:

python images/__main__.py