MultiLabelClassificationOutput

class torch_ecg.components.MultiLabelClassificationOutput(*args: Any, **kwargs: Any)

Bases: BaseOutput

Class that maintains the output of a multi-label classification task.

Parameters:

• classes (Sequence[str]) – class names

• thr (float) – threshold for making binary predictions

• prob (numpy.ndarray) – Probabilities of each class, of shape (batch_size, num_classes)

• pred (numpy.ndarray) – Binary predictions, of shape (batch_size, num_classes).

Note

Known issues:

• fields of type dict are not well supported due to the limitations of the base class CFG, for example

>>> output = MultiLabelClassificationOutput(classes=["AF", "N", "SPB"], thr=0.5, pred=np.ones((1,3)), prob=np.ones((1,3)), d={"d":1})
>>> output
{'classes': ['AF', 'N', 'SPB'],
    'prob': array([[1., 1., 1.]]),
    'pred': array([[1., 1., 1.]]),
    'thr': 0.5,
    'd': {'d': 1}}
>>> output.d  # has to access via `output["d"]`
AttributeError: 'MultiLabelClassificationOutput' object has no attribute 'd'

compute_metrics(macro: bool = True) → ClassificationMetrics

Compute metrics from the output.

Parameters:

macro (bool) – Whether to use macro-averaged metrics or not.

Returns:

metrics – Metrics computed from the output.

Return type:

ClassificationMetrics

required_fields() → Set[str]

The required fields of the output class.