Neural Network in JavaScript

Implementantion of a Perceptron neural network in JavaScript. It is a simple implementation that can serve as an example for learning, not for production use. It does not use GPU and the only activation function implemented is a sigmoid function.

For a ready to use implementation please refer to BrainJS

Installation

npm i vt-neural-network-ts

Usage

import { Network } from 'vt-neural-network-ts'

// Define the layer structure
const layers = [
  2, // This is the input layer
  10, // Hidden layer 1
  10, // Hidden layer 2
  1 // Output
]

const network = new Network(layers)

// Start training
const numberOfIterations = 20000

// Training data for a "XOR" logic gate
const trainingData = [{
  input : [0,0],
  output: [0]
}, {
  input : [0,1],
  output: [1]
}, {
  input : [1,0],
  output: [1]
}, {
  input : [1,1],
  output: [0]
}]

for(var i = 0; i < numberOfIterations; i ++) {
  // Get a random training sample
  const trainingItem = trainingData[Math.floor((Math.random()*trainingData.length))]
  network.train(trainingItem.input, trainingItem.output);
}

// After training we can see if it works
// we call activate to set a input in the first layer
network.activate(trainingData[0].input)
const resultA = network.run()

network.activate(trainingData[1].input)
const resultB = network.run()

network.activate(trainingData[2].input)
const resultC = network.run()

network.activate(trainingData[3].input)
const resultD = network.run()
console.log('Expected 0 got', resultA[0])
console.log('Expected 1 got', resultB[0])
console.log('Expected 1 got', resultC[0])
console.log('Expected 0 got', resultD[0])

If you want to see other logic gates implementations, check the test folder.

API

• network.setLearningRate(0.3): Adjust the learning rate of the network,
• network.toJSON(): returns the structure of the network
• network.layers: contains the different layers of the network
  • layer.neurons: contains the different neurons on each layer

How to develop the application?

npm install
npm run watch
# Open public/ directory in browser

Remove generated directory

If you would like to remove public/dist directory (created by Webpack):

npm run clear

If you would like to remove node_modules/ and remove public/dist/

npm run clear:all

Count LOC (Lines of Code)

If you would like to know how many lines of code you write:

npm run count

Analysis of bundle file weight

If you would like to check how much a bundle file weight:

npm run audit

Information of interest

Backpropagation

https://mattmazur.com/2015/03/17/a-step-by-step-backpropagation-example/

Neural networks

https://scrimba.com/g/gneuralnetworks https://franpapers.com/en/machine-learning-ai-en/2017-neural-network-implementation-in-javascript-by-an-example/ http://karpathy.github.io/neuralnets/