Layer class abstract

The abstract base class for all neural network layers.

A Layer is the fundamental, callable building block of a neural network. It encapsulates both a state (its trainable parameters, like weights and biases) and a transformation from input tensors to output tensors (the forward pass).

Layers are designed to be chained together in a model, such as an SNetwork, where the output of one layer becomes the input to the next, forming the complete network architecture.

Lifecycle:

1. Instantiation: A layer is created (e.g., DenseLayer(64)). Its weights are not yet created.
2. Build: The first time the layer is called with an input, the build method runs automatically. It uses the input's shape to initialize the weights and biases with the correct dimensions. This is called deferred initialization.
3. Forward Pass: On every call, the forward method is executed to perform the layer's core mathematical operations.

Example

// Define a layer (weights are not created yet).
Layer dense = DenseLayer(32, activation: ReLU());

// Create an input tensor.
Tensor<Vector> input = Tensor<Vector>([1.0, 2.0, 3.0]);

// Call the layer. This will first build it, then run the forward pass.
Tensor<Vector> output = dense.call(input) as Tensor<Vector>;

Implementers

• AveragePooling2DLayer
• BatchNorm1D
• BatchNorm2D
• Conv2DLayer
• ConvLSTMLayer
• DenseLayer
• DenseLayerMatrix
• DropoutLayer
• DropoutLayerMatrix
• DualLSTMLayer
• EmbeddingLayer
• EmbeddingLayerMatrix
• FlattenLayer
• GeneralizedChainedScaleLayer
• GlobalAveragePooling1D
• GlobalAveragePoolingLayer
• LayerNormalization
• LayerNormalizationVector
• LSTMLayer
• MaxPooling1DLayer
• MaxPooling2DLayer
• MultiHeadAttention
• MultiTierLSTMLayer
• PositionalEncoding
• ReLULayer
• ReLULayerMatrix
• ReshapeVectorToMatrixLayer
• RNN
• SingleHeadAttention
• SNetwork
• TransformerEncoderBlock

Constructors

Layer()

Properties

hashCode → int

The hash code for this object.

no setter, inherited

name → String

A user-friendly name for the layer (e.g., 'dense', 'lstm').

no setter

parameters → List<Tensor>

A list of all trainable tensors (weights and biases) in the layer.

no setter

runtimeType → Type

A representation of the runtime type of the object.

no setter, inherited

Methods

build(Tensor input) → void

Initializes the layer's parameters based on the shape of the first input.

call(Tensor input) → Tensor

The public, callable interface for the layer.

forward(Tensor input) → Tensor

The core logic of the layer's transformation.

noSuchMethod(Invocation invocation) → dynamic

Invoked when a nonexistent method or property is accessed.

inherited

toString() → String

A string representation of this object.

inherited

Operators

operator ==(Object other) → bool

The equality operator.

inherited