fair_logit_estimator.py is an implementation of the fair classifier described
in Learning Fair Classifiers (Zafar et al., 2016).
To use the classifier, make sure you have numpy, scipy, and scikit-learn installed. Then simply import fair_logit_estimator.py into your project.
Only binary dependent variables in [-1, 1] format are supported. Specify some number of sensitive attributes in your training data by passing a list of column indices. See code comments for more detailed documentation.
The fair-classification folder contains a lightly modified version of mbilalzafar/fair-classification. Their testing code has been rewritten to benchmark the two implementations, proving that they produce nearly-identical output for the sample data.
To run the benchmarks:
cd fair-classification/synthetic_data_demo
python decision_boundary_demo.py