Judiciously Reducing Subgroup Comparisons for Learning Intersectional Fair Representations in Ranking
Set up the conda env with python 3.8. Install requirements:
conda create -n intersectional_fairness python=3.8
conda activate intersectional_fairness
pip install -r requirements.txt
For pre-processing method CIFRank install R.
Configurations are defined under: ./configs. Some configurations are dynamically updated when running the experiments. For this check the ./run_experiments/configs_<dataset>.py file.
Processed dataset files are under: DATA.
Code used to pre-process the datasets, create the train/test split is under: ./src/datasets. Each class contains dataset specific pre-processing. The syntehtic_dataset.py contains the code to generate the simulated set-up.
To run the experiments run the following command:
python ./run_experiments/configs_<dataset_name>.py --model <model_name> --optimisation_type <optimization>
To run the evaluation:
python ./evaluation/compute_extra_metrics.py
To save results of all experiments as .csv:
python ./evaluation/experiment_results.py
Complete results on the simulated set-up.
