def test_accuracy_update(
outputs_list: List[torch.Tensor],
targets_list: List[torch.Tensor],
num_classes: int,
topk: List[int],
true_values_list: List[Dict[str, float]],
) -> None:
"""
This test checks that AccuracyMetric updates its values correctly and return
correct intermediate results
Note that now `accuracy/std` is not std exactly so it can fail if you fix it.
Args:
outputs_list: list of output tensors
targets_list: list of target tensors
num_classes: number od classes for classification
topk: topk args for computing accuracy@topk
true_values_list: list of correct metrics intermediate values
"""
metric = AccuracyMetric(topk_args=topk, num_classes=num_classes)
for outputs, targets, true_values in zip(outputs_list, targets_list,
true_values_list):
metric.update(logits=outputs, targets=targets)
intermediate_metric_values = metric.compute_key_value()
for key in true_values.keys():
assert key in intermediate_metric_values
assert np.isclose(true_values[key],
intermediate_metric_values[key])
def test_accuracy(
outputs: torch.Tensor,
targets: torch.Tensor,
num_classes: int,
topk: List[int],
true_values: Dict[str, float],
) -> None:
"""
Test multiclass accuracy with different topk args
Note that now `accuracy/std` is not std exactly so it can fail if you fix it.
Args:
outputs: tensor of outputs
targets: tensor of targets
num_classes: number of classes for classification
topk: list of topk args for accuracy@topk
true_values: true metrics values
"""
metric = AccuracyMetric(topk_args=topk)
metric.update(logits=outputs, targets=targets)
metrics = metric.compute_key_value()
for key in true_values.keys():
assert key in metrics
assert np.isclose(true_values[key], metrics[key])
def __init__(
self,
input_key: str,
target_key: str,
topk_args: List[int] = None,
num_classes: int = None,
log_on_batch: bool = True,
prefix: str = None,
suffix: str = None,
):
"""Init."""
super().__init__(
metric=AccuracyMetric(
topk_args=topk_args, num_classes=num_classes, prefix=prefix, suffix=suffix
),
input_key=input_key,
target_key=target_key,
log_on_batch=log_on_batch,
)