rnn.md

Recurrent Neural Network

The Recurrent Neural Network (RNN) is an implementation of a simple recurrent neural network in JavaScript. It includes methods for creating the network, performing feedforward operations, and training the network using backpropagation through time.

Class: RNN

Represents a simple recurrent neural network.

Constructor

Creates a new instance of the RNN class.

new RNN(inputSize, hiddenSize, outputSize, learningRate)

• inputSize (number): The number of input neurons.
• hiddenSize (number): The number of neurons in the hidden layer.
• outputSize (number): The number of output neurons.
• learningRate (number): The learning rate for weight updates during tsaining.

Methods

.sigmoid(x)

Calculates the sigmoid activation function value for a given input.

• x (number): The input value.
• Returns: The sigmoid output.

.tanh(x)

Calculates the hyperbolic tangent activation function value for a given input.

• x (number): The input value.
• Returns: The hyperbolic tangent output.

.forward(inputs) Performs a feedforward pass through the recurrent neural network.

• inputs (number[]): The input values.
• Returns: The output values.

.train(inputs, targets)

Trains the recurrent neural network using the provided inputs and targets.

• inputs (number[]): The input values.
• targets (number[]): The target values.

Properties

.inputSize

The number of input neurons.

.hiddenSize

The number of neurons in the hidden layer.

.outputSize

The number of output neurons.

.learningRate

The learning rate for weight updates during training.