The requirements.txt file list all Python libraries used on this project, which can be installed using:
pip install -r requirements.txt
The optimization algorithm used in our approach is MONISE - Many Objective Non-Inferior Set Estimation. To use it, clone the repository https://github.com/marcosmrai/moopt and install it:
python setup.py install
If you have any problems with PuLP, install version 2.1:
pip install -Iv PuLP==2.1
In the experiments, we made comparisons with several approaches. Please clone the following repositories in the directory of this project:
- Minimax Pareto Fairness: https://github.com/natalialmg/MMPF
- AdaFair: https://github.com/iosifidisvasileios/AdaFair
- MAMO Fair: https://github.com/kirtanp/MAMO-fair
- Preferential Fairness: https://github.com/mbilalzafar/fair-classification
To use Minimax Pareto Fairness change line 9 of the file MMPF/dataset_loaders.py to: from MMPF.MinimaxParetoFair import *.
AdaFair uses a different version of sklearn. To solve compatiblity issues, replace line 40 of the AdaFair.py file with:
from sklearn.tree import BaseDecisionTree, DecisionTreeClassifier
from numpy import float32 as DTYPE
After cloning the MAMO-fair repository, rename it to MAMOfair so that its functions can be imported. Replace the file MAMOfair/metric/metrics.py with the file with the same name which is in this repository. Lastly, if there is any error while importing from MAMOfair, you may resolve it by modifying their imports, such as changing from loss.loss_class import Loss to from .loss.loss_class import Loss.
For Preferential Fairness, we modified their code to be able to run it with Python 3.8 (the original was Python 2.7). All of the required code for our experiments are in the folder fair_classification_modified/