Metric

class brainstate.nn.Metric

Base class for metrics.

Any class that subclasses Metric should implement compute, reset, and update methods to track and compute evaluation metrics.

reset()

Reset the metric state to initial values.

update(**kwargs)

Update the metric state with new data.

compute()

Compute and return the current metric value.

Notes

This is an abstract base class and should not be instantiated directly. Subclasses must implement all three methods.

compute()

Compute and return the current value of the metric.

Returns:

The computed metric value. The type depends on the specific metric.

Return type:

metric_value

Raises:

NotImplementedError – If the subclass does not implement this method.

reset()

In-place reset the metric state to initial values.

This method should restore all internal state variables to their initial values as if the metric was just constructed.

Raises:

NotImplementedError – If the subclass does not implement this method.

Return type:

None

update(**kwargs)

In-place update the metric with new data.

Parameters:

**kwargs – Keyword arguments containing the data to update the metric. The specific arguments depend on the metric implementation.

Raises:

NotImplementedError – If the subclass does not implement this method.

Return type:

None