def get_fairlearn_data()

in src/responsibleai/rai_analyse/_score_card/_rai_insight_data.py [0:0]

• 40 lines of code
• 5 McCabe index (conditional complexity)

    def get_fairlearn_data(self):
        fairness_config = self.config["Fairness"]
        fm = {}
        dataset = self.data.get_test()

        for sensitive_feature in fairness_config["sensitive_features"]:
            fm[sensitive_feature] = {}
            fm[sensitive_feature]["metrics"] = {}
            fm[sensitive_feature]["statistics"] = {}
            topnlabels = (
                dataset[sensitive_feature].value_counts().nlargest(20).index.to_list()
            )
            for metric in fairness_config["metric"]:
                gm = self.data.get_fairlearn_grouped_metric(
                    metric, sensitive_feature, self.pos_label, self.classes
                )

                fm_lookup = {
                    "difference": gm.difference(method="between_groups"),
                    "ratio": gm.ratio(),
                }

                gmd = gm.by_group.to_dict()
                gmd = {k: gmd[k] for k in topnlabels}

                sorted_group_metric = sorted(gmd.items(), key=lambda x: x[1])

                fm[sensitive_feature]["metrics"][metric] = {
                    "kind": fairness_config["fairness_evaluation_kind"],
                    "value": fm_lookup[fairness_config["fairness_evaluation_kind"]],
                    "group_metric": OrderedDict(sorted_group_metric),
                    "group_max": next(iter(sorted_group_metric[-1:]), None),
                    "group_min": next(iter(sorted_group_metric[:1]), None),
                }

            feature_statistics = dict(
                self.data.get_feature_statistics(sensitive_feature)
            )

            fl_short_label_generator = AlphabetLabelIterator()
            for k, _ in feature_statistics.items():
                filtermap = self.data.get_test()[sensitive_feature] == k
                cohort_data = self.data.get_cohort_data(filtermap)
                cohort_data["short_label"] = next(fl_short_label_generator)
                cohort_data["pos_label"] = self.pos_label
                fm[sensitive_feature]["statistics"][k] = cohort_data

        return fm