This is the repository for DifFair and ConFair.
conda create -n 'venv' python=3.7.0
conda activate venv
pip install -r requirements.txtDownload the folder "DataInsights" from https://github.com/microsoft/prose/tree/main/misc/CCSynth/CC and copy this folder inside your local directory of this repository.
pip install -e DataInsightsThen download the code repository and cd to your downloaded local directory "ConFair". Run one single execution of the experiments using the below command.
Execute below script to run the experiments for ConFair and DifFair in their performance over real-world data
./ exec_confair.zshExecute below script to run the experiments for ConFair and DifFair in their performance over synthetic data
./ exec_diffair.zshExecute below script to run the experiments in comparing ConFair to OMN in their performance under model-aware weights
./ exec_aware.zshExecute below script to run the experiments in comparing ConFair and DifFair in their performance without the optimization of CCs over real data
./ exec_opt_cc.zshExecute below script to run the experiments in comparing ConFair to OMN in their relationship between inpute degree and fairness improvement
./ exec_degree.zshNote that for MEPS16 dataset, you need to extract the raw data using the R scrip. See details at https://github.com/Trusted-AI/AIF360/blob/master/aif360/data/raw/meps/README.md.
Note for running CAPUCHIN, you need to install the CAPUCHIN package.
Some visualization in the paper can be found at this folder notebooks.
