autocast.metrics.base

class BaseMetric(reduce_all=True, dist_sync_on_step=False)

Bases: Metric, Generic[TPred, TTrue], ABC

Shared template for spatial metrics that reduce over spatial axes.

Parameters:

• reduce_all (bool)

• dist_sync_on_step (bool)

score(y_pred, y_true)

Parameters:

• y_pred (Tensor | ndarray)

• y_true (Tensor | ndarray)

Return type:

Float[Tensor, ‘batch time channel’] | Float[Tensor, ‘batch spatial *spatial channel’] | Float[Tensor, ‘batch time spatial *spatial channel’]

update(y_pred, y_true)

Update metric state with a batch of predictions and targets.

Parameters:

• y_pred (Tensor | ndarray)

• y_true (Tensor | ndarray)

Return type:

None

compute()

Compute final metric value.

Return type:

Tensor

reset()

Reset metric state and initialization flag.

Return type:

None