def test_auc_threshold():
"""To compute rates:
-> fpr: number of false positives / total number of negatives
-> tpr: number of true positives / total number of positives
In this example, consider the first threshold of 0.1, and consider everything
higher than 0.1 as positives, and everything lower as negatives:
fpr = 1 / 2 = 0.5
tpr = 1
Thus, if the max_fpr rate is 0.1, the tpr is 0, and if the max_fpr is 0.5,
then the tpr is 1. So we get an auc of 0 for the 0.1 threshold, and 0.5
for the 0.5 threshold.
For more details: https://en.wikipedia.org/wiki/Receiver_operating_characteristic
"""
y_pred = torch.tensor([0.1, 0.2, 0.3, 0.4])
y_true = torch.tensor([0, 1, 0, 1])
metric_test_case(y_pred, y_true, AUC(max_fpr=1.), sklearn.metrics.roc_auc_score(y_true, y_pred))
metric_test_case(y_pred, y_true, AUC(max_fpr=0.500001), 0.5)
metric_test_case(y_pred, y_true, AUC(max_fpr=0.1), 0.5)
def test_auc_full():
"""Test random score list."""
y_pred = np.random.randint(0, 10000, 100)
y_true = np.random.randint(0, 2, 100)
metric_test_case(y_pred, y_true, AUC(), sklearn.metrics.roc_auc_score(y_true, y_pred))