def test_one_element_predicted_clusters(self):
y_true = [0, 0, 0, 1, 1, 1]
y_pred = [0, 1, 2, 3, 4, 5]
self.assertAlmostEqual(average_cluster_purity(y_true, y_pred), 1.0)
y_true = [0, 0]
y_pred = [0, 1]
self.assertAlmostEqual(average_cluster_purity(y_true, y_pred), 1.0)
def test_differing_clusters(self):
y_true = [0, 0, 1]
y_pred = [1, 0, 0]
self.assertAlmostEqual(average_cluster_purity(y_true, y_pred), 2 / 3)
y_true = [0, 1, 2, 1]
y_pred = [0, 1, 1, 2]
self.assertAlmostEqual(average_cluster_purity(y_true, y_pred), 0.75)
y_true = [0, 1]
y_pred = [0, 0]
self.assertAlmostEqual(average_cluster_purity(y_true, y_pred), 0.5)
def _calculate_analysis_values(predicted_clusters, true_cluster, times):
"""
Calculates the analysis values out of the predicted_clusters.
:param predicted_clusters: The predicted Clusters of the Network.
:param true_clusters: The validation clusters
:return: the results of all metrics as a 2D array where i is the index of the metric and j is the index of a
specific result
"""
logger = get_logger('analysis', logging.INFO)
logger.info('Calculate scores')
# Initialize output
metric_results = [None] * len(metric_names)
for m, min_value in enumerate(metric_worst_values):
if min_value == 1:
metric_results[m] = np.ones(len(true_cluster))
else:
metric_results[m] = np.zeros((len(true_cluster)))
# Loop over all possible clustering
for i, predicted_cluster in enumerate(
tqdm(predicted_clusters,
ncols=100,
desc='Calculate scores for predicted clusters...')):
# Calculate different analysis's
metric_results[0][i] = misclassification_rate(true_cluster,
predicted_cluster)
metric_results[1][i] = average_cluster_purity(true_cluster,
predicted_cluster)
metric_results[2][i] = adjusted_rand_index(true_cluster,
predicted_cluster)
metric_results[3][i] = diarization_error_rate(true_cluster,
predicted_cluster, times)
return metric_results
def test_equal_clusters(self):
y_true = [0, 1, 1, 0]
y_pred = [1, 0, 0, 1]
self.assertAlmostEqual(average_cluster_purity(y_true, y_pred), 1.0)