Allows for replication of the Counterfactual Fairness paper results in Python using PyStan.
Options are:
- -do_l2: Performs the replication of the L2 (Fair K) model, which can take a while depending on computing power
- -save_l2: Saves the resultant models (or not) for the L2 (Fair K) model, which produces large-ish files (100s MBs)
Dependencies include:
- Python 3.5.5
- NumPy 1.14.3
- Pandas 0.23.0
- Scikit-learn 0.19.1
- PyStan 2.17.1.0
- StatsModels 0.9.0
To run with default settings (Perform L2 tests, but don't save the Posterior Samples), type:
python CounterFair_Emulate.py
The following results should print:
Unfair RMSE: 0.870
FTU RMSE: 0.891
Level 2 (Fair K) RMSE: 0.929
Level 3 (Fair Add) RMSE: 0.918