LSTMLayer class

A Long Short-Term Memory (LSTM) layer.

An LSTMLayer is an advanced recurrent layer designed to overcome the short-term memory limitations of a simple RNN. It excels at learning long-term dependencies in sequential data.

It achieves this by using a dedicated cell state (c) for long-term memory, which acts like a conveyor belt where information can travel across many timesteps without being significantly altered.

A series of gates (Forget, Input, and Output) intelligently control the flow of information into and out of this cell state.

• Input: A Tensor<Matrix> representing the sequence, with a shape of [sequence_length, input_size].
• Output: A Tensor<Vector> representing the final hidden state after processing the entire sequence, with a shape of [hidden_size].

Example

// An LSTM layer with 32 memory units.
Layer lstm = LSTMLayer(32);

// An input sequence with 10 timesteps and 5 features each.
Tensor<Matrix> sequence = Tensor<Matrix>(...);

// The output is the final hidden state vector of length 32.
Tensor<Vector> finalState = lstm.call(sequence) as Tensor<Vector>;

Inheritance

• Object
• Layer
• LSTMLayer

Constructors

LSTMLayer(int hiddenSize)

Properties

b_c ↔ Tensor<Vector>

getter/setter pair

b_f ↔ Tensor<Vector>

getter/setter pair

b_i ↔ Tensor<Vector>

getter/setter pair

b_o ↔ Tensor<Vector>

getter/setter pair

hashCode → int

The hash code for this object.

no setter, inherited

hiddenSize ↔ int

The number of units in the hidden state and cell state.

getter/setter pair

name ↔ String

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

getter/setter pair, override-getter

parameters → List<Tensor>

Provides all 8 trainable parameter tensors to the optimizer.

no setter, override

runtimeType → Type

A representation of the runtime type of the object.

no setter, inherited

W_c ↔ Tensor<Matrix>

Weights and biases for the Candidate Cell State.

getter/setter pair

W_f ↔ Tensor<Matrix>

Weights and biases for the Forget Gate.

getter/setter pair

W_i ↔ Tensor<Matrix>

Weights and biases for the Input Gate.

getter/setter pair

W_o ↔ Tensor<Matrix>

Weights and biases for the Output Gate.

getter/setter pair

Methods

build(Tensor input) → void

Initializes all parameter tensors for the four gates.

override

call(Tensor input) → Tensor

The public, callable interface for the layer.

inherited

forward(Tensor input) → Tensor<Vector>

Performs the forward pass for the LSTM layer.

override

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