Layer - Klyn API

Layer

classin packageklyn.math.nn.layers

public abstract class Layer<T extends AbstractFloat = Float>

Layer

Statically typed stage of a sequential batch computation. Derivatives are returned explicitly, without keeping a tape on the layer.

Example

layer as Layer<Float> = Dense<Float>(32)

Properties

Modifier and Type	Member	Description
public abstract readonly property	displayName `displayName as String`	Short descriptive name used in model summaries.
public readonly property	inputShapeHint `inputShapeHint as IList<Long>`	Explicit input sample dimensions, or an empty list for deferred inference.
public abstract readonly property	parameterCount `parameterCount as ULong`	Number of trainable parameter tensors.

Methods

Modifier and Type	Member	Description
public	backwardTraining `backwardTraining(context as LayerPass<T>, upstream as NDArray<T>) as LayerGradients<T>`	Differentiates one training context; subclasses may reuse saved intermediates.
public abstract	buildShape `buildShape(inputShape as IList<Long>) as IList<Long>`	Initializes an unbuilt layer, or checks its existing input dimensions.
public abstract	differentiate `differentiate(inputs as NDArray<T>, upstream as NDArray<T>) as LayerGradients<T>`	Computes derivatives against unchanged parameters.
public abstract	forward `forward(inputs as NDArray<T>) as NDArray<T>`	Computes inference outputs without changing parameters or retaining the batch.
public	forwardTraining `forwardTraining(inputs as NDArray<T>) as LayerPass<T>`	Computes outputs and an explicit per-batch context owned by the trainer.
public abstract	parameter `parameter(index as Long) as NDArray<T>`	Returns a trainable tensor, ordered like the parameter derivatives.