def test_complete_accuracy_with_container_sources(self):
annotations = [
ContainerAnnotation(
{'a': ClassificationAnnotation('identifier', 3)})
]
predictions = [
ContainerPrediction({
'p':
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
})
]
config = [{
'type': 'accuracy',
'top_k': 1,
'annotation_source': 'a',
'prediction_source': 'p'
}]
dispatcher = MetricsExecutor(config, None)
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
for _, evaluation_result in dispatcher.iterate_metrics(
annotations, predictions):
assert evaluation_result.name == 'accuracy'
assert evaluation_result.evaluated_value == pytest.approx(1.0)
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_classification_per_class_accuracy_prediction_top3(self):
annotation = [
ClassificationAnnotation('identifier_1', 1),
ClassificationAnnotation('identifier_2', 1)
]
prediction = [
ClassificationPrediction('identifier_1', [1.0, 2.0, 3.0, 4.0]),
ClassificationPrediction('identifier_2', [2.0, 1.0, 3.0, 4.0])
]
dataset = DummyDataset(label_map={0: '0', 1: '1', 2: '2', 3: '3'})
dispatcher = MetricsExecutor([{
'type': 'accuracy_per_class',
'top_k': 3
}], dataset)
dispatcher.update_metrics_on_batch(range(len(annotation)), annotation,
prediction)
for _, evaluation_result in dispatcher.iterate_metrics(
annotation, prediction):
assert evaluation_result.name == 'accuracy_per_class'
assert len(evaluation_result.evaluated_value) == 4
assert evaluation_result.evaluated_value[0] == pytest.approx(0.0)
assert evaluation_result.evaluated_value[1] == pytest.approx(0.5)
assert evaluation_result.evaluated_value[2] == pytest.approx(0.0)
assert evaluation_result.evaluated_value[3] == pytest.approx(0.0)
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_classification_accuracy_result_for_batch_1_with_2_metrics(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
]
dispatcher = MetricsExecutor([{
'name': 'top1',
'type': 'accuracy',
'top_k': 1
}, {
'name': 'top3',
'type': 'accuracy',
'top_k': 3
}], None)
metric_result, _ = dispatcher.update_metrics_on_batch(
range(len(annotations)), annotations, predictions)
expected_metric_result = [
PerImageMetricResult('top1', 'accuracy', 1.0, 'higher-better'),
PerImageMetricResult('top3', 'accuracy', 1.0, 'higher-better')
]
assert len(metric_result) == 1
assert 0 in metric_result
assert len(metric_result[0]) == 2
assert metric_result[0][0] == expected_metric_result[0]
assert metric_result[0][1] == expected_metric_result[1]
def test_mae_with_positive_diff_between_annotation_and_prediction(self):
annotations = [
RegressionAnnotation('identifier', 3),
RegressionAnnotation('identifier2', 1)
]
predictions = [
RegressionPrediction('identifier', 1),
RegressionPrediction('identifier2', -3)
]
config = [{'type': 'mae'}]
expected = EvaluationResult(
pytest.approx([3.0, 1.0]), None, 'mae', 'mae', None, None, {
'postfix': ' ',
'scale': 1,
'names': ['mean', 'std'],
'calculate_mean': False,
'target': 'higher-worse'
}, None)
dispatcher = MetricsExecutor(config, None)
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
for _, evaluation_result in dispatcher.iterate_metrics(
annotations, predictions):
assert evaluation_result == expected
def test_mae_on_interval_values_in_range(self):
annotations = [
RegressionAnnotation('identifier', 0.5),
RegressionAnnotation('identifier', 0.5)
]
predictions = [
RegressionPrediction('identifier', 1),
RegressionPrediction('identifier', 0.25)
]
config = [{'type': 'mae_on_interval', 'end': 1}]
expected = EvaluationResult(
pytest.approx([0.375, 0.125]), None, 'mae_on_interval',
'mae_on_interval', None, None, {
'postfix': ' ',
'scale': 1,
'names': ['mean: <= 0.0 < 1.0', 'std: <= 0.0 < 1.0'],
'calculate_mean': False,
'target': 'higher-worse',
'orig_names': ['mean: <= 0.0 < 1.0', 'std: <= 0.0 < 1.0']
}, None)
dispatcher = MetricsExecutor(config, None)
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
for _, evaluation_result in dispatcher.iterate_metrics(
annotations, predictions):
assert evaluation_result == expected
def test_classification_accuracy_result_for_batch_2_with_not_ordered_ids(
self):
annotations = [
ClassificationAnnotation('identifier', 3),
ClassificationAnnotation('identifier1', 1)
]
predictions = [
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0]),
ClassificationPrediction('identifier2', [1.0, 1.0, 1.0, 4.0])
]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 1}], None)
metric_result, _ = dispatcher.update_metrics_on_batch([42, 17],
annotations,
predictions)
expected_metric_result = [
PerImageMetricResult('accuracy', 'accuracy', 1.0, 'higher-better'),
PerImageMetricResult('accuracy', 'accuracy', 0.0, 'higher-better')
]
assert len(metric_result) == 2
assert 42 in metric_result
assert len(metric_result[42]) == 1
assert metric_result[42][0] == expected_metric_result[0]
assert 17 in metric_result
assert len(metric_result[17]) == 1
assert metric_result[17][0] == expected_metric_result[1]
def test_update_metric_result(self):
dataset = multi_class_dataset()
annotations = make_segmentation_representation(np.array([[1, 2, 3, 2, 3], [0, 0, 0, 0, 0]]), True)
predictions = make_segmentation_representation(np.array([[1, 0, 3, 0, 0], [0, 0, 0, 0, 0]]), False)
dispatcher = MetricsExecutor(create_config(self.name), dataset)
metric_result, _ = dispatcher.update_metrics_on_batch(range(len(annotations)), annotations, predictions)
assert metric_result[0][0].result == 0.5125
def test_multi_class(self):
annotations = make_segmentation_representation(np.array([[1, 0, 3, 0, 0], [0, 0, 0, 0, 0]]), True)
predictions = make_segmentation_representation(np.array([[1, 2, 3, 2, 3], [0, 0, 0, 0, 0]]), False)
dispatcher = MetricsExecutor(create_config(self.name), multi_class_dataset())
dispatcher.update_metrics_on_batch(range(len(annotations)), annotations, predictions)
expected = generate_expected_result((5.0+1.0+1.0)/(8.0+1.0+1.0), self.name)
for _, evaluation_result in dispatcher.iterate_metrics(annotations, predictions):
assert evaluation_result == expected
def test_accuracy_with_wrong_prediction_type_raise_config_error_exception(
self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [DetectionPrediction('identifier', [1.0, 1.0, 1.0, 4.0])]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 1}], None)
with pytest.raises(ConfigError):
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
def test_multi_class_not_matched(self):
annotations = make_segmentation_representation(np.array([[1, 1, 1, 1, 1], [1, 1, 1, 1, 1]]), True)
predictions = make_segmentation_representation(np.array([[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]), False)
dataset = multi_class_dataset()
dispatcher = MetricsExecutor(create_config(self.name), dataset)
dispatcher.update_metrics_on_batch(range(len(annotations)), annotations, predictions)
expected = generate_expected_result([0.0], self.name, {1: dataset.labels[1]})
for _, evaluation_result in dispatcher.iterate_metrics(annotations, predictions):
assert evaluation_result == expected
def test_config_default_presenter(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
]
config = [{'type': 'accuracy', 'top_k': 1}]
dispatcher = MetricsExecutor(config, None)
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
for presenter, _ in dispatcher.iterate_metrics(annotations,
predictions):
assert isinstance(presenter, ScalarPrintPresenter)
def test_zero_accuracy_top_3(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [5.0, 3.0, 4.0, 1.0])
]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 3}], None)
for _, evaluation_result in dispatcher.iterate_metrics(
annotations, predictions):
assert evaluation_result.name == 'accuracy'
assert evaluation_result.evaluated_value == 0.0
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_accuracy_on_prediction_container_with_several_suitable_representations_raise_config_error_exception(
self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ContainerPrediction({
'prediction1':
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0]),
'prediction2':
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
})
]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 1}], None)
with pytest.raises(ConfigError):
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
def test_update_normed_error(self):
config = [{'type': 'normed_error'}]
annotations = [
FacialLandmarksAnnotation('identifier', np.array([1, 1, 1, 1, 1]),
np.array([1, 1, 1, 1, 1]))
]
annotations[0].metadata.update({'left_eye': 0, 'right_eye': 1})
predictions = [
FacialLandmarksPrediction('identifier', np.array([1, 1, 1, 1, 1]),
np.array([1, 1, 1, 1, 1]))
]
dispatcher = MetricsExecutor(config, None)
metric_result, _ = dispatcher.update_metrics_on_batch(
range(len(annotations)), annotations, predictions)
assert metric_result[0][0].result == 0
def test_update_rmse_on_interval_metric(self):
config = [{'type': 'rmse_on_interval', 'intervals': [0.0, 2.0, 4.0]}]
annotations = [
RegressionAnnotation('identifier', 3),
RegressionAnnotation('identifier2', 1)
]
predictions = [
RegressionPrediction('identifier', 5),
RegressionPrediction('identifier2', 5)
]
dispatcher = MetricsExecutor(config, None)
metric_result, _ = dispatcher.update_metrics_on_batch(
range(len(annotations)), annotations, predictions)
assert metric_result[0][0].result == 2
assert metric_result[1][0].result == 4
def test_update_mse_metric_result(self):
annotations = [
RegressionAnnotation('identifier', 3),
RegressionAnnotation('identifier2', 1)
]
predictions = [
RegressionPrediction('identifier', 5),
RegressionPrediction('identifier2', 5)
]
config = [{'type': 'mse'}]
dispatcher = MetricsExecutor(config, None)
metric_result, _ = dispatcher.update_metrics_on_batch(
range(len(annotations)), annotations, predictions)
assert metric_result[0][0].result == 4
assert metric_result[1][0].result == 16
def test_config_unknown_presenter(self):
config = [{
'type': 'accuracy',
'top_k': 1,
'presenter': 'print_somehow'
}]
with pytest.raises(ValueError):
MetricsExecutor(config, None)
def test_accuracy_with_several_annotation_source_raises_config_error_exception(
self):
with pytest.raises(ConfigError):
MetricsExecutor([{
'type': 'accuracy',
'top_k': 1,
'annotation_source': 'annotation1, annotation2'
}], None)
def test_complete_accuracy_top_3(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [1.0, 3.0, 4.0, 2.0])
]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 3}], None)
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
for _, evaluation_result in dispatcher.iterate_metrics(
annotations, predictions):
assert evaluation_result.name == 'accuracy'
assert evaluation_result.evaluated_value == pytest.approx(1.0)
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_accuracy_with_several_prediction_source_raises_value_error_exception(
self):
with pytest.raises(ConfigError):
MetricsExecutor([{
'type': 'accuracy',
'top_k': 1,
'prediction_source': 'prediction1, prediction2'
}], None)
def test_accuracy_on_container_with_wrong_annotation_source_name_raise_config_error_exception(
self):
annotations = [
ContainerAnnotation(
{'annotation': ClassificationAnnotation('identifier', 3)})
]
predictions = [
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
]
dispatcher = MetricsExecutor([{
'type': 'accuracy',
'top_k': 1,
'annotation_source': 'a'
}], None)
with pytest.raises(ConfigError):
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
def test_accuracy_with_unsupported_prediction_type_as_prediction_source_for_container_raises_config_error(
self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ContainerPrediction({
'prediction':
DetectionPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
})
]
dispatcher = MetricsExecutor([{
'type': 'accuracy',
'top_k': 1,
'prediction_source': 'prediction'
}], None)
with pytest.raises(ConfigError):
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
def test_threshold_is_10_by_config(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [5.0, 3.0, 4.0, 1.0])
]
dispatcher = MetricsExecutor([{
'type': 'accuracy',
'top_k': 3,
'abs_threshold': 10
}], None)
for _, evaluation_result in dispatcher.iterate_metrics([annotations],
[predictions]):
assert evaluation_result.name == 'accuracy'
assert evaluation_result.evaluated_value == 0.0
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold == 10
def test_classification_per_class_accuracy_fully_zero_prediction(self):
annotation = ClassificationAnnotation('identifier', 0)
prediction = ClassificationPrediction('identifier', [1.0, 2.0])
dataset = DummyDataset(label_map={0: '0', 1: '1'})
dispatcher = MetricsExecutor([{
'type': 'accuracy_per_class',
'top_k': 1
}], dataset)
dispatcher.update_metrics_on_batch(range(1), [annotation],
[prediction])
for _, evaluation_result in dispatcher.iterate_metrics([annotation],
[prediction]):
assert evaluation_result.name == 'accuracy_per_class'
assert len(evaluation_result.evaluated_value) == 2
assert evaluation_result.evaluated_value[0] == pytest.approx(0.0)
assert evaluation_result.evaluated_value[1] == pytest.approx(0.0)
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_mae_on_interval_default_all_missed(self):
annotations = [RegressionAnnotation('identifier', -2)]
predictions = [RegressionPrediction('identifier', 1)]
config = [{'type': 'mae_on_interval', 'end': 1}]
expected = EvaluationResult(
pytest.approx([0.0]), None, 'mae_on_interval', 'mae_on_interval',
None, None, {
'postfix': ' ',
'scale': 1,
'names': [],
'calculate_mean': False,
'target': 'higher-worse',
'orig_names': ['mean: <= 0.0 < 1.0', 'std: <= 0.0 < 1.0']
}, None)
dispatcher = MetricsExecutor(config, None)
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
with pytest.warns(UserWarning) as warnings:
for _, evaluation_result in dispatcher.iterate_metrics(
annotations, predictions):
assert len(warnings) == 1
assert evaluation_result == expected
def test_missed_interval(self):
config = [{'type': 'mae_on_interval'}]
with pytest.raises(ValueError):
MetricsExecutor(config, None)
def test_missed_metrics_raises_config_error_exception(self):
with pytest.raises(ConfigError):
MetricsExecutor([], None)
def test_accuracy_arguments(self):
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 1}], None)
assert len(dispatcher.metrics) == 1
_, _, accuracy_metric, _, _, _, _ = dispatcher.metrics[0]
assert isinstance(accuracy_metric, ClassificationAccuracy)
assert accuracy_metric.top_k == 1
def test_undefined_metric_type_raises_config_error_exception(self):
with pytest.raises(ConfigError):
MetricsExecutor([{'type': ''}], None)
def test_metrics_with_empty_entry_raises_config_error_exception(self):
with pytest.raises(ConfigError):
MetricsExecutor([{}], None)
