This repository is an implementation of Fair Selective Classification via Sufficiency.

To install requirements move to the folder and run:

conda env create -f environment.yml

After a succesfull install activate the enviroment:

conda activate fair

Running the code for the first time will automatically install the datasets needed.

Please download the datasets, along with the pretrained models and place them in ./Data/ and ./Models/ respectively:

• Models trained on the datasets.

To run the model(s) evaluation on a certain dataset using default hyperparameters, run the following command:

python3 main.py --dataset_name=[adult/celeba/civil]

This loads a trained model(s), skipping the training phase.

To force training of new model(s) in the paper, run this command:

python3 main.py --dataset_name=[adult/celeba/civil] --force_train=1

For a complete list of parser arguments and hyperparameters available, see main.py.

To recreate the plots and tables, open plot.ipynb, fill in the name of the desired dataset in the designated cell and run the notebook.