RecallMetric

class brainstate.nn.RecallMetric(num_classes=None, average='macro')

Recall (sensitivity) metric for binary and multi-class classification.

Recall is the ratio of true positives to all actual positives: recall = TP / (TP + FN)

Parameters:

• num_classes (int | None) – Number of classes. If None, assumes binary classification. Default is None.

• average (str) – Type of averaging for multi-class: ‘micro’, ‘macro’, or ‘weighted’. Default is ‘macro’. Ignored for binary classification.

num_classes

Number of classes.

Type:

int or None

average

Averaging method for multi-class.

Type:

str

true_positives

Count of true positive predictions.

Type:

MetricState

false_negatives

Count of false negative predictions.

Type:

MetricState

Examples

>>> import jax.numpy as jnp
>>> import brainstate
>>> predictions = jnp.array([1, 0, 1, 1, 0])
>>> labels = jnp.array([1, 0, 0, 1, 0])
>>> metric = brainstate.nn.RecallMetric()
>>> metric.update(predictions=predictions, labels=labels)
>>> metric.compute()
Array(1., dtype=float32)

Notes

Recall measures the fraction of actual positive cases that were correctly identified. Also known as sensitivity or true positive rate (TPR).

compute()

Compute and return the recall.

Returns:

The recall value(s). For binary classification, returns a scalar. For multi-class, returns per-class or averaged recall based on the average parameter.

Return type:

Array

reset()

Reset the metric state to zero.

Return type:

None

update(*, predictions, labels, **_)

Update the recall metric.

Parameters:

• predictions (Array) – Predicted class labels (integers).

• labels (Array) – Ground truth class labels (integers).

• **_ – Additional keyword arguments are ignored.

Return type:

None