Source code for tests.metrics.privacy.test_reidentification

# Standard library
import pytest
from typing import Type, Tuple
from inspect import getfullargspec
import math

# 3rd party packages
import pandas as pd
import matplotlib.pyplot as plt

# Local packages
from clover.metrics.base import Metric
from clover.metrics.privacy import reidentification as reid

test_params = [
    {"metric_class": metric, "which_data": data}
    for metric in reid.get_metrics()
    for data in ["different_datasets", "identical_datasets"]
]
test_ids = [f"{d['metric_class'].name}-{d['which_data']}" for d in test_params]


[docs]@pytest.fixture(scope="module", params=test_params, ids=test_ids) def reidentification_metrics_results( request, df_wbcd: dict[str, pd.DataFrame], df_mock_wbcd: dict[str, pd.DataFrame], metadata_wbcd: dict, ) -> Tuple[Type[Metric], str, dict]: """ Compute the reidentification metrics in different settings. :param request: the number of continuous and categorical columns to test :param df_wbcd: the real Wisconsin Breast Cancer Dataset fixture, split into **train** and **test** sets :param df_mock_wbcd: the mock wbcd dataset fixture, contained **train** and **test** sets :param metadata_wbcd: the wbcd metadata fixture :return: a tuple containing the metric class, the dataset type and a dictionary containing the **average** scores of the metric and the **detailed** scores """ metric_class = request.param["metric_class"] which_data = request.param["which_data"] # Instance parameters d = {"random_state": 0, "sampling_frac": 1} # Select only the expected instance parameters args = getfullargspec(metric_class).args[1:] # remove self metric = metric_class(*[d[arg] for arg in args]) df_to_compare = df_mock_wbcd if which_data == "different_datasets" else df_wbcd scores = metric.compute(df_wbcd, df_to_compare, metadata_wbcd) return metric_class, which_data, scores
[docs]def test_reidentification_metrics_summary( reidentification_metrics_results: Tuple[Type[Metric], str, dict], ) -> None: """ Test the reidentification metrics average scores. :param reidentification_metrics_results: a tuple containing the metric class, the dataset type and a dictionary containing the **average** scores of the metric and the **detailed** scores :return: None """ metric, which_data, scores = reidentification_metrics_results scores = scores["average"] for submetric in metric.get_average_submetrics(): # Check the boundaries assert scores[submetric["submetric"]] >= submetric["min"] assert scores[submetric["submetric"]] <= submetric["max"] # Filter out the references if not submetric["submetric"].endswith("_train_test_ref"): # Check the target diff_to_objective = abs( scores[submetric["submetric"]] - submetric[submetric["objective"]] ) inv_obj = "min" if submetric["objective"] == "max" else "max" diff_to_inv_objective = abs( scores[submetric["submetric"]] - submetric[inv_obj] ) if which_data == "different_datasets": assert ( not math.isinf(diff_to_objective) and diff_to_objective < 0.01 ) or diff_to_inv_objective > 0.01 else: assert ( not math.isinf(diff_to_objective) and diff_to_objective > 0.01 ) or diff_to_inv_objective < 0.01
[docs]def test_reidentification_metrics_detailed( reidentification_metrics_results: Tuple[Type[Metric], str, dict], ) -> None: """ Test the reidentification metrics detailed scores. :param reidentification_metrics_results: a tuple containing the metric class, the dataset type and a dictionary containing the **average** scores of the metric and the **detailed** scores :return: None """ metric, which_data, scores = reidentification_metrics_results report = scores["detailed"] metric.draw(report=report, figsize=(8, 6)) plt.close("all")