ResponsiblyAI / responsibly

    return indices

def find_independence_thresholds(roc_curves, base_rates, proportions,

                                 cost_matrix):

    """Compute thresholds that achieve independence and minimize cost.

    :param roc_curves: Receiver operating characteristic (ROC)

                       by attribute.

    :type roc_curves: dict

    :param base_rates: Base rate by attribute.

    :type base_rates: dict

    :param proportions: Proportion of each attribute value.

    :type proportions: dict

    :param cost_matrix: Cost matrix by [[tn, fp], [fn, tp]].

    :type cost_matrix: sequence

    :return: Thresholds, FPR and TPR by attribute and cost value.

    :rtype: tuple

    """

    cutoffs = {}

    def total_cost_function(acceptance_rate_value):

        # todo: move demo here + multiple cost

        #       + refactor - use threshold to calculate

        #         acceptance_rate_value

        indices = get_acceptance_rate_indices(roc_curves, base_rates,

                                              acceptance_rate_value)

        total_cost = 0

        for group, roc in roc_curves.items():

            index = indices[group]

            fpr = roc[0][index]

            tpr = roc[1][index]

            group_cost = _cost_function(fpr, tpr,

                                        base_rates[group],

                                        cost_matrix)

            group_cost *= proportions[group]

            total_cost += group_cost

        return -total_cost

    acceptance_rate_min_cost = _ternary_search_float(total_cost_function,

                                                     0, 1, TRINARY_SEARCH_TOL)

    cost = total_cost_function(acceptance_rate_min_cost)

    threshold_indices = get_acceptance_rate_indices(roc_curves, base_rates,

                                                    acceptance_rate_min_cost)

    thresholds = _extract_threshold(roc_curves)

    cutoffs = {group: thresholds[threshold_index]

               for group, threshold_index

               in threshold_indices.items()}

    fpr_tpr = {group: (roc[0][threshold_indices[group]],

                       roc[1][threshold_indices[group]])

               for group, roc in roc_curves.items()}

    return cutoffs, fpr_tpr, cost, acceptance_rate_min_cost

def get_fnr_indices(roc_curves, fnr_value):

    return indices

def find_fnr_thresholds(roc_curves, base_rates, proportions,

                        cost_matrix):

    """Compute thresholds that achieve equal FNRs and minimize cost.

    Also known as **equal opportunity**.

    :param roc_curves: Receiver operating characteristic (ROC)

                       by attribute.

    :type roc_curves: dict

    :param base_rates: Base rate by attribute.

    :type base_rates: dict

    :param proportions: Proportion of each attribute value.

    :type proportions: dict

    :param cost_matrix: Cost matrix by [[tn, fp], [fn, tp]].

    :type cost_matrix: sequence

    :return: Thresholds, FPR and TPR by attribute and cost value.

    :rtype: tuple

    """

    cutoffs = {}

    def total_cost_function(fnr_value):

        # todo: move demo here + multiple cost

        indices = get_fnr_indices(roc_curves, fnr_value)

        total_cost = 0

        for group, roc in roc_curves.items():

            index = indices[group]

            fpr = roc[0][index]

            tpr = roc[1][index]

            group_cost = _cost_function(fpr, tpr,

                                        base_rates[group],

                                        cost_matrix)

            group_cost *= proportions[group]

            total_cost += group_cost

        return -total_cost

    fnr_value_min_cost = _ternary_search_float(total_cost_function,

                                               0, 1,

                                               TRINARY_SEARCH_TOL)

    threshold_indices = get_fnr_indices(roc_curves, fnr_value_min_cost)

    cost = total_cost_function(fnr_value_min_cost)

    fpr_tpr = {group: (roc[0][threshold_indices[group]],

                       roc[1][threshold_indices[group]])

               for group, roc in roc_curves.items()}

    thresholds = _extract_threshold(roc_curves)

    cutoffs = {group: thresholds[threshold_index]

               for group, threshold_index

               in threshold_indices.items()}

    return cutoffs, fpr_tpr, cost, fnr_value_min_cost

def _find_feasible_roc(roc_curves):