NELS empirical data analysis fairness.
Run through vscode or jupyter to walk through some examples of Stability measures on the original dataset using
Below are instructions to replicate the plots we generated. In order to generate our plots, we leveraged the infrastructure built by Friedler et. al. (https://github.com/algofairness/fairness-comparison) for their paper "A comparative study of fairness-enhancing interventions in machine learning." We added our stability metric to their existing metrics, and modified their framework to run with the NELS:88 education dataset. We are thankful that they shared their code, which allowed us to perform many metric comparisons with relatively low effort.
From the top directory, run python fairness/runner.py. Results from the run will appear in: /fairness/data/results/results/
To recreate correlation, cd into analysis/correlation-vis. Make sure the proper data files are included in the directory. Then, run: python combine_results_files.py education_Race_original.csv education_Race-Sex_original.csv education_Sex_original.csv education_cor.csv
Move the the data files from /fairness/data/results/results/ to /fairness/ Then run: python fairness/runner_analysis.py