fairness-by-location

====================================================

Overview

Code for paper: Fairness by "Where": A Statistically-Robust and Model-Agnostic Bi-Level Learning Framework. AAAI 2022.

====================================================

Explanation of files

(Link for data: https://pitt-my.sharepoint.com/:f:/g/personal/erh108_pitt_edu/EmqOrtnsaCVFnD_PA1cFjt8BLX1zg6Ws0smAF0hr90JKjw?e=NCHtar)

• X_train.npy: all training samples extracted from the satellite-based crop monitoring dataset.
• y_train.npy: the corresponding labels for training samples.
• train_id.pickle: training samples' indices for all partitions within each candidate partitioning.
• X_test.npy: all testing samples (not overlapped with training samples).
• y_test.npy: the corresponding labels for testing samples.
• test_id.pickle: training samples' indices for all partitions within each candidate partitioning.
• results: an example model.

Explanation of the code:

Procedures

model_train.py:

1. Training a base model with training data with 300 epochs.
2. Applying stochastic and bi-level training strategies to the base model with 50 epochs.

evaluation.py:

3. Comparing the overall performance and fairness between the base model and the final model.