Fair-Over-Sampling/common_utils.py at main · dd1github/Fair-Over-Sampling · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
# Metrics function
#Code adapted from https://github.com/Trusted-AI/AIF360


from collections import OrderedDict
from aif360.metrics import ClassificationMetric
import numpy as np

def compute_metrics(dataset_true, dataset_pred,
                    unprivileged_groups, privileged_groups,
                    disp = True):
    """ Compute the key metrics """
    classified_metric_pred = ClassificationMetric(dataset_true,
                            dataset_pred,
                            unprivileged_groups=unprivileged_groups,
                            privileged_groups=privileged_groups)
    metrics = OrderedDict()
    metrics["Balanced accuracy"] = 0.5*(classified_metric_pred.true_positive_rate()+
                            classified_metric_pred.true_negative_rate())

    metrics["Average odds difference"] = classified_metric_pred.average_odds_difference()

    metrics["Absolute average odds difference"] = classified_metric_pred.average_abs_odds_difference()
    metrics["True positive rate difference"] = classified_metric_pred.true_positive_rate_difference()
    metrics["True negative rate difference"] = classified_metric_pred.true_negative_rate_difference()

    metrics["Equal opportunity difference"] = classified_metric_pred.equal_opportunity_difference()

    metrics["Fair utility"] = metrics["Balanced accuracy"] * .5 * \
        ((1-np.abs(metrics["True positive rate difference"])) + \
         (1-np.abs(metrics["True negative rate difference"])))

    if disp:
        for k in metrics:
            print("%s = %.4f" % (k, metrics[k]))

    return metrics