ml_fairness_check1.py

import numpy as np

def statistical_parity(B, m, fairness_threshold=.01):

    if len(B) != len(m):
        raise ValueError('Input arrays do not have same number of entries')

    indices_pos_class, = np.where(B == 1)
    indices_neg_class, = np.where(B == 0)

    outcomes_pos = m[indices_pos_class]
    outcomes_neg = m[indices_neg_class]

    if len(outcomes_pos) == 0:
        return None
    if len(outcomes_neg) == 0:
        return None
    stats_parity = np.abs(len(np.where(outcomes_pos == 1)) /
                           len(outcomes_pos) -
                           len(np.where(outcomes_neg == 1)) /
                           len(outcomes_neg))
    if abs(stats_parity) < fairness_threshold:
        print("The model is FAIR on Black with statistical parity {}".format(round(stats_parity,5)))
    else:
        print("The model is NOT FAIR on Black with statistical parity {}".format(round(stats_parity,5)))