def update(self, outputs: torch.Tensor, targets: torch.Tensor) -> float:
"""
Update metric value with accuracy for new data and return intermediate metric value.
Args:
outputs: tensor of outputs
targets: tensor of true answers
Returns:
accuracy metric for outputs and targets
"""
metric = multilabel_accuracy(outputs=outputs,
targets=targets,
threshold=self.threshold).item()
super().update(value=metric, num_samples=np.prod(targets.shape))
return metric
def test_multilabel_accuracy(
outputs: torch.Tensor,
targets: torch.Tensor,
threshold: Union[float, torch.Tensor],
true_value: float,
):
"""
Test multilabel accuracy with single and multiple thresholds
Args:
outputs: tensor of outputs
targets: tensor of true answers
threshold: thresholds for multilabel classification
true_value: expected metric value
"""
value = multilabel_accuracy(
outputs=outputs, targets=targets, threshold=threshold
).item()
assert np.isclose(value, true_value)