Visualization of Fairness Limitations in Automated Decision Systems

Project submission for COS 597Ed By Irene Fan, Thomas Schaffner, and Gautam Sharma

Dependencies

Python 3.6.3 or later

Python packages

• Django 2.0 or later
• sklearn 0.19.1 or later
• json 2.0.9 or later
• numpy 1.13.3 or later

Running locally

Launch the server with

python3 manage.py runserver

Then use any browser to navigate to

localhost:8000

Sample Data Files

Two sets of sample files are found in the testfiles directory. Both sets were generated using generate_calibrated_data.py. partially_calibrated_* files contain one protected field with three values. Scores generated for two of the values are calibrated, while scores for the final (larger) group were generated uniformly at random.

well_calibrated_* files also contain a single protected field with three values, but scores for all values of the protected attribute are calibrated.