How to test the fairness of ML models? The 80% rule to measure disparate impact | Hacker News Reader