/FairRanking

Primary LanguagePythonMIT LicenseMIT

FairRanking

This repository provides the code for reproducing the results obtained in the paper Long-Term Fairness Strategies in Ranking with Continuous Sensitive Attributes (currently under revision at AEQUITAS@ECAI24)

Prerequisites

  • Virtual environment with Python >=3.7
  • Packages: 
    pip install -r requirements.txt

How to run

The script run.py replicates the results presented in the paper. For Discrete Actions experiments run:

python run.py --actions discrete

For Continuous Actions experiments run:

python run.py --actions continuous

Results are stored in results_group_weights folder for Discrete Actions experiment and results_polynomial_fn for Continuous Actions experiment.
Each folder contains a records with:

  • the configuration file with all the run information config.json
  • the historical records of actions, metrics of interest, optimization and ranking (.pkl files)
  • the mean and standard deviation of the metrics (statistics.txt and statistics.json)

and a images folder contains images both in png and eps format.