Simple Neural Network library (JavaScript)

The point of this library is to allow very simple neural networks to be easily created, with no extra fluff.

Installation

This library can be easily installed with yarn or npm. The installation commands is as follows.

npm install simple-neural-network-js # Install with NPM
yarn add simple-neural-network-js    # Install with yarn

Usage

Creating a neural network

You can create a new neural network with the following command.

var numInputs = 2;
var numOutputs = 2;
var numHiddenLayers = 2;
var numNeuronsPerHiddenLayer = 6;

var neuralNetwork = new NeuralNetwork(numInputs, numOutputs, numHiddenLayers, numNeuronsPerHiddenLayer);

The NeuralNetwork constructor accepts parameters that defined the number of inputs and outputs the neural network should have, the number of hidden layers that should exist and the number of neurons that should be present in each hidden layer.

Using the neural network

To get outputs from the neural network, you must supply it with an array of inputs. This array's length should be equal to the number of inputs you set when creating the neural network.

Ideally the inputs should be in a range from -1 to +1, but this is not required.

Outputs are returned as an array of values from 0 to +1.

The update method is used to supply inputs to the neural network and retrieve outputs. An example usage is shown below.

var inputs = [0.12, 0.24];
var outputs = neuralNetwork.update(inputs);

Getting and setting weights

The weights between the neurons be retrieved or overrided with the getWeights and setWeights methods. Examples of how to use these methods are shown below.

var weights = neuralNetwork.getWeights();

var newWeights = [];
for (var i=0; i < weights.length; i++) {
    newWeights.push(weights[i] * 0.5); 
}

neuralNetwork.setWeights(newWeights);

Weights are returned by the getWeights as a single dimension array.

New weights must be provided to the setWeights method in the same order as they are retrieved by the getWeights method.