def test_get_multiclass_predictions_and_correctness(self):
multiclass_probs = np.array([[0.1, 0.2, 0.7], [0.5, 0.3, 0.2],
[0.7, 0.2, 0.1], [0.3, 0.5, 0.2]])
labels = np.array([2, 0, 1, 0])
(argmax_probs,
is_correct) = metrics_lib.get_multiclass_predictions_and_correctness(
multiclass_probs, labels)
self.assertAllEqual(argmax_probs, [0.7, 0.5, 0.7, 0.5])
self.assertAllEqual(is_correct, [True, True, False, False])
def test_get_multiclass_predictions_and_correctness_error_cases(self):
multiclass_probs = np.array([[0.1, 0.2, 0.7], [0.5, 0.3, 0.2],
[0.7, 0.2, 0.1], [0.3, 0.5, 0.2]])
labels = np.array([2, 0, 1, 0])
with self.assertRaises(ValueError):
bad_multiclass_probs = multiclass_probs - 0.01
metrics_lib.get_multiclass_predictions_and_correctness(
bad_multiclass_probs, labels)
with self.assertRaises(ValueError):
metrics_lib.get_multiclass_predictions_and_correctness(
bad_multiclass_probs[Ellipsis, None], labels)
with self.assertRaises(ValueError):
metrics_lib.get_multiclass_predictions_and_correctness(
bad_multiclass_probs, labels[Ellipsis, None])