def conditional_statistical_parity_aux(parameters, parameters_result):
all_satisfy = True
results = list()
for legitimate_attributes in parameters["legitimate_attributes_list"]:
satisfies, proportions = conditional_statistical_parity(
parameters["testing_set"], legitimate_attributes,
parameters["descriptions"],
parameters["maximum_acceptable_difference"],
parameters["positive_outcome"], parameters["negative_outcome"])
template = "P({}={}|L,{})".format(
parameters_result.predicted_outcome_display_name(),
parameters["positive_outcome"], "{}")
single_result = Result(satisfies)
parameters_result.add_legitimate_attributes(single_result,
legitimate_attributes,
proportions, template,
parameters["descriptions"])
results.append(single_result)
all_satisfy &= satisfies
final_result = Result(all_satisfy)
parameters_result.add_maximum_acceptable_difference(
final_result, parameters["maximum_acceptable_difference"])
for result in results:
final_result.add_element(result)
return final_result
def treatment_equality_aux(parameters, parameters_result):
satisfies, errors = treatment_equality(
parameters["metrics"], parameters["maximum_acceptable_difference"])
result = Result(satisfies)
parameters_result.add_maximum_acceptable_difference(
result, parameters["maximum_acceptable_difference"])
parameters_result.add_list(result, "FN/FP", errors, "[{}]",
parameters["descriptions"])
return result
def well_calibration_aux(parameters, parameters_result):
satisfies_list = well_calibration(parameters["probabilities_table"],
parameters["decimals"])
result = Result(True in satisfies_list)
parameters_result.add_decimals(result, parameters["decimals"])
template = "P({}={}|S=s,{})".format(parameters["outcome_name"],
parameters["positive_outcome"], "{}")
parameters_result.add_probabilities_table(
result, template, satisfies_list, parameters["descriptions"],
parameters["probabilities_table"])
return result
def overall_accuracy_equality_aux(parameters, parameters_result):
satisfies, precisions = overall_accuracy_equality(
parameters["metrics"], parameters["maximum_acceptable_difference"])
result = Result(satisfies)
parameters_result.add_maximum_acceptable_difference(
result, parameters["maximum_acceptable_difference"])
template = "P({}={}|{})".format(
parameters_result.predicted_outcome_display_name(),
parameters["outcome_name"], "{}")
parameters_result.add_list(result, "(TP+TN)/(TP+FP+TN+FN)", precisions,
template, parameters["descriptions"])
return result
def group_fairness_aux(parameters, parameters_result):
satisfies, proportions = group_fairness(
parameters["metrics"], parameters["maximum_acceptable_difference"])
result = Result(satisfies)
parameters_result.add_maximum_acceptable_difference(
result, parameters["maximum_acceptable_difference"])
template = "P({}={}|{})".format(
parameters_result.predicted_outcome_display_name(),
parameters["positive_outcome"], "{}")
parameters_result.add_list(result, "(TP+FP)/(TP+FP+TN+FN)", proportions,
template, parameters["descriptions"])
return result
def predictive_parity_aux(parameters, parameters_result):
satisfies, ppv_values = predictive_parity(
parameters["metrics"], parameters["maximum_acceptable_difference"])
result = Result(satisfies)
parameters_result.add_maximum_acceptable_difference(
result, parameters["maximum_acceptable_difference"])
template = "P({}={}|{}={},{})".format(
parameters["outcome_name"], parameters["positive_outcome"],
parameters_result.predicted_outcome_display_name(),
parameters["positive_outcome"], "{}")
parameters_result.add_list(result, "PPV", ppv_values, template,
parameters["descriptions"])
return result
def false_positive_error_rate_balance_aux(parameters, parameters_result):
satisfies, fpr_values = false_positive_error_rate_balance(
parameters["metrics"], parameters["maximum_acceptable_difference"])
result = Result(satisfies)
parameters_result.add_maximum_acceptable_difference(
result, parameters["maximum_acceptable_difference"])
template = "P({}={}|{}={},{})".format(
parameters_result.predicted_outcome_display_name(),
parameters["positive_outcome"], parameters["outcome_name"],
parameters["negative_outcome"], "{}")
parameters_result.add_list(result, "FPR", fpr_values, template,
parameters["descriptions"])
return result
def balance_for_positive_class_aux(parameters, parameters_result):
satisfies, expected_values = balance_for_positive_class(
parameters["positives_table"], parameters["metrics"],
parameters["maximum_acceptable_difference"])
result = Result(satisfies)
parameters_result.add_maximum_acceptable_difference(
result, parameters["maximum_acceptable_difference"])
template = "E({}|{}={},{})".format(
parameters_result.predicted_probability_display_name(),
parameters["outcome_name"], parameters["positive_outcome"], "{}")
parameters_result.add_list(result, "expected_values", expected_values,
template, parameters["descriptions"])
return result
def test_fairness_aux(parameters, parameters_result):
satisfies_list = test_fairness(parameters["probabilities_table"],
parameters["maximum_acceptable_difference"],
parameters["decimals"])
result = Result(True in satisfies_list)
parameters_result.add_maximum_acceptable_difference(
result, parameters["maximum_acceptable_difference"])
parameters_result.add_decimals(result, parameters["decimals"])
template = "P({}={}|S=s,{})".format(parameters["outcome_name"],
parameters["positive_outcome"], "{}")
parameters_result.add_probabilities_table(
result, template, satisfies_list, parameters["descriptions"],
parameters["probabilities_table"])
return result
def causal_discrimination_aux(parameters, parameters_result):
proportion, test_suite, confidence_reached = causal_discrimination(
parameters["testing_set"], parameters["descriptions"],
parameters["confidence"], parameters["error"] / 100,
parameters["minimum_samples_amount"], parameters["predict_outcome"])
result = Result(proportion <= MAXIMUM_FAILING_PROPORTION)
parameters_result.add_confidence(result, parameters["confidence"],
confidence_reached)
parameters_result.add_percentage(result, proportion, parameters["error"])
parameters_result.add_minimum_samples_amount(
result, parameters["minimum_samples_amount"])
parameters_result.add_test_suite(result, test_suite,
parameters["descriptions"])
return result
def equalized_odds_aux(parameters, parameters_result):
satisfies, tpr_values, fpr_values = equalized_odds(
parameters["metrics"], parameters["maximum_acceptable_difference"])
result = Result(satisfies)
parameters_result.add_maximum_acceptable_difference(
result, parameters["maximum_acceptable_difference"])
template = "P({}={}|{}={},{})".format(
parameters_result.predicted_outcome_display_name(),
parameters["positive_outcome"], parameters["outcome_name"], "{}", "{}")
tpr_template = template.format(parameters["positive_outcome"], "{}")
fpr_template = template.format(parameters["negative_outcome"], "{}")
parameters_result.add_list(result, "TPR", tpr_values, tpr_template,
parameters["descriptions"])
parameters_result.add_list(result, "FPR", fpr_values, fpr_template,
parameters["descriptions"])
return result
def fairness_through_awareness_aux(parameters, parameters_result):
proportion, failing_cases, confidence_reached = \
fairness_through_awareness(parameters["testing_set"],
parameters["calculate_individuals_distance"],
parameters["calculate_outcomes_distance"],
parameters["confidence"],
parameters["error"] / 100,
parameters["minimum_samples_amount"])
result = Result(proportion <= MAXIMUM_FAILING_PROPORTION)
parameters_result.add_confidence(result, parameters["confidence"],
confidence_reached)
parameters_result.add_percentage(result, proportion, parameters["error"])
parameters_result.add_minimum_samples_amount(
result, parameters["minimum_samples_amount"])
parameters_result.add_failing_cases(result, failing_cases)
return result
def conditional_use_accuracy_equality_aux(parameters, parameters_result):
satisfies, ppv_values, npv_values = conditional_use_accuracy_equality(
parameters["metrics"], parameters["maximum_acceptable_difference"])
result = Result(satisfies)
parameters_result.add_maximum_acceptable_difference(
result, parameters["maximum_acceptable_difference"])
template = "P({}={}|{}={},{})".format(
parameters["outcome_name"], "{}",
parameters_result.predicted_outcome_display_name(), "{}", "{}")
ppv_template = template.format(parameters["positive_outcome"],
parameters["positive_outcome"], "{}")
npv_template = template.format(parameters["negative_outcome"],
parameters["negative_outcome"], "{}")
parameters_result.add_list(result, "PPV", ppv_values, ppv_template,
parameters["descriptions"])
parameters_result.add_list(result, "NPV", npv_values, npv_template,
parameters["descriptions"])
return result