DenseGradients

classin packageklyn.math.nn.layers

public class DenseGradients<T extends AbstractFloat = Float>

DenseGradients

Gradients of an affine layer with respect to its inputs, weights and bias. No optimizer step has been applied to these independent arrays.

Example

gradients = layer.backward(inputs, upstream)

Properties

Modifier and Type	Member	Description
public readonly property	bias `bias as NDArray<T>`	Bias gradient [1, units].
public readonly property	inputs `inputs as NDArray<T>`	Input gradient [batch, inputs].
public readonly property	weights `weights as NDArray<T>`	Weight gradient [inputs, units].

Constructors

Modifier and Type	Member	Description
public	DenseGradients `DenseGradients(inputs as NDArray<T>, weights as NDArray<T>, bias as NDArray<T>)`	Groups gradients computed by Dense.backward; arrays are not copied.