11 June 2020

Here we have collected the code to accompany the NeurIPS submission "Individually Fair Gradient Boosting"

• FOR AN EXAMPLE OF BuDRO, SEE THE NOTEBOOK german/01_german_run.ipynb
• FOR A SIMPLE EXAMPLE OF BuDRO ON SYNTHETIC DATA, SEE THE SCRIPT synthetic/sim_tf_test.py - THE RESULTS CAN BE PLOTTED IN synthetic/simulated-plots.ipynb TO PRODUCE AN EQUIVALENT TO FIGURE 1 IN THE SUPPLEMENT

The packages used in the conda environment for the notebooks and scripts are in the file py37fair.yml. You also need to install the AIF360 package (from "https://github.com/IBM/AIF360") and probably the SenSR package (from "https://github.com/IBM/sensitive-subspace-robustness).

In most files, all paths have been removed and need to be replaced with your own paths.

Other directories:

german contains specific scripts and notebooks used in the experiements on the German credit data set(predict good or bad for credit).

adult contains specific scripts and Jupyter notebooks used in the experiments on the Adult data set(predict whether income exceeds $50k/yr based on Census data).

compas contains specific scripts and notebooks used in the experiments on the COMPAS data set(predicts criminal recidivism rates).

scripts contains the files needed for running BuDRO, as well as a few other scripts used in data processing.

synthetic contains the scripts needed to generate the synthetic plots from the appendix of the submission. To plot the data generated in this directory, see the notebook simulated-plots.ipynb.