LossResult

classin packageklyn.math.nn.losses

public class LossResult<T extends AbstractFloat = Float>

LossResult

A mean batch loss and its derivative with respect to model outputs.

Example

result = SparseCategoricalCrossentropy<Float>().compute(logits, labels)

Properties

Modifier and Type	Member	Description
public readonly property	gradient `gradient as NDArray<T>`	Mean-reduced output gradient, including an optional loss-scaling factor.
public readonly property	value `value as Double`	Mean scalar loss accumulated in FP64.

Constructors

Modifier and Type	Member	Description
public	LossResult `LossResult(value as Double, gradient as NDArray<T>)`	Retains a scalar loss and its gradient without copying the array.