neuralga v1.0.17

Install

Before installing it you need Node.js npm install command:

$ npm install neuralga --save

What is this

Javascript neural network implementation that uses genetic algorithm to train.

Parameters(Genetic Algorithm)

• inputs: The number of inputs of the neural networks (When passing data to the inputs and outputs it must be a number between 0 and 1)
• hiddenlayers: An array containing the amount of neurons in each layer 2 is one layer with two neurons
• outputs: The number of outputs of the neural network
• population: Number of neural networks in the population
• mutationRate: Number between 0 and 1 that determines how often the mutation should be applied
• elitism: Number between 0 and 1 that determines how much of the population should survive and stay without mutations
• provenance: Number between 0 and 1 that determines how much of the population will be new neural networks (this is to add some variety)
• mutationAmount: When a mutation occurs, the mutationAmount determines how many mutations are applied
• fitnessFunction: The function that will determine the fitness of the neural network The Genetic Algorithm tries to maximise the fitness
• memory: This is a boolean that indicates whether the neural networks should have neurons selfConnected

Methods(Genetic Algorithm)

• start: Creates the initial population and set the generation to 0 (it accepts a neural network as argument to create the initial population but its optional)
• step: takes an array of objects that represents the inputs and outputs and gets the population to the next generation
• best: returns the best neural network of the population

Methods(Neural Network)

• execute: give it an array of numbers that represents the inputs and returns an array of outputs
• clearContext: clears the results of the neurons (this method is in case you activate the memory)
• toJSON: returns JSON representation of the Neural Network

Static methods

• fromJSON: returns an instance of a Neural network from JSON object

Examples

XOR

const {GeneticAlgorithm} = require("neuralga")
let data = [
    {input:[0,0],output:[0]},
    {input:[0,1],output:[1]},
    {input:[1,0],output:[1]},
    {input:[1,1],output:[0]}
]
function fitness(network){
    let error = 0;
    for(let i = 0;i<data.length;i++){
        let salida = network.execute(data[i].input)
        for(let i2 = 0;i2<salida.length;i2++){
            error += Math.abs(salida[i2]-parseInt(data[i].output[i2].toFixed(20)));
        }
    }
    return -error;
}


let ga = new GeneticAlgorithm(2,[2],1,10,0.4,0.2,0.2,4,fitness,false)
ga.start();

let bestfitness = -Infinity;
while(bestfitness !== 0){
    let overall = ga.step();
    bestfitness = ga.best().fitness;
    console.log(`Generation: ${ga.generation} best: ${bestfitness} overall: ${overall}`);
}
let best = ga.best();
console.log(best.execute([0,0]))
console.log(best.execute([0,1]))
console.log(best.execute([1,0]))
console.log(best.execute([1,1]))

XOR without relu

const {GeneticAlgorithm,Activations} = require("neuralga")
let data = [
    {input:[0,0],output:[0]},
    {input:[0,1],output:[1]},
    {input:[1,0],output:[1]},
    {input:[1,1],output:[0]}
]

function fitness(network){
    let error = 0;
    for(let i = 0;i<data.length;i++){
        let salida = network.execute(data[i].input)
        for(let i2 = 0;i2<salida.length;i2++){
            error += Math.abs(salida[i2]-parseInt(data[i].output[i2].toFixed(20)));
        }
    }
    return -error;
}


let ga = new GeneticAlgorithm(2,[2],1,10,0.4,0.2,0.2,4,fitness,false)

delete Activations.relu //This removes the relu activation so now the Neurons cant use it to evolve

ga.start();



let bestfitness = -Infinity;
while(bestfitness !== 0){
    let overall = ga.step();
    bestfitness = ga.best().fitness;
    console.log(`Generation: ${ga.generation} best: ${bestfitness} overall: ${overall}`);
}
let best = ga.best();
console.log(best.execute([0,0]))
console.log(best.execute([0,1]))
console.log(best.execute([1,0]))
console.log(best.execute([1,1]))

XOR adding custom activation function

const {GeneticAlgorithm,Activations} = require("neuralga")
let data = [
    {input:[0,0],output:[0]},
    {input:[0,1],output:[1]},
    {input:[1,0],output:[1]},
    {input:[1,1],output:[0]}
]

function fitness(network){
    let error = 0;
    for(let i = 0;i<data.length;i++){
        let salida = network.execute(data[i].input)
        for(let i2 = 0;i2<salida.length;i2++){
            error += Math.abs(salida[i2]-parseInt(data[i].output[i2].toFixed(20)));
        }
    }
    return -error;
}


let ga = new GeneticAlgorithm(2,[2],1,10,0.4,0.2,0.2,4,fitness,false)


/*
    When adding custom activations function you cant use more than one {} if you want to save it to json for some limitation in the toJSON function
*/

Activations.binaryStep = function(x){return x<0 ? 0 : 1}

ga.start();



let bestfitness = -Infinity;
while(bestfitness !== 0){
    let overall = ga.step();
    bestfitness = ga.best().fitness;
    console.log(`Generation: ${ga.generation} best: ${bestfitness} overall: ${overall}`);
}
let best = ga.best();
console.log(best.execute([0,0]))
console.log(best.execute([0,1]))
console.log(best.execute([1,0]))
console.log(best.execute([1,1]))

XOR save best to json using fs

const {GeneticAlgorithm} = require("neuralga")
let data = [
    {input:[0,0],output:[0]},
    {input:[0,1],output:[1]},
    {input:[1,0],output:[1]},
    {input:[1,1],output:[0]}
]

function fitness(network){
    let error = 0;
    for(let i = 0;i<data.length;i++){
        let salida = network.execute(data[i].input)
        for(let i2 = 0;i2<salida.length;i2++){
            error += Math.abs(salida[i2]-parseInt(data[i].output[i2].toFixed(20)));
        }
    }
    return -error;
}


let ga = new GeneticAlgorithm(2,[2],1,10,0.4,0.2,0.2,4,fitness,false)
ga.start();

let bestfitness = -Infinity;
while(bestfitness !== 0){
    let overall = ga.step();
    bestfitness = ga.best().fitness;
    console.log(`Generation: ${ga.generation} best: ${bestfitness} overall: ${overall}`);
}
let best = ga.best();
console.log(best.execute([0,0]))
console.log(best.execute([0,1]))
console.log(best.execute([1,0]))
console.log(best.execute([1,1]))
const fs = require('fs');
fs.writeFileSync('network.json', JSON.stringify(best.toJSON()));

XOR get best from json using fs

const {GeneticAlgorithm,NeuralNetwork} = require("neuralga")
let data = [
    {input:[0,0],output:[0]},
    {input:[0,1],output:[1]},
    {input:[1,0],output:[1]},
    {input:[1,1],output:[0]}
]

function fitness(network){
    let error = 0;
    for(let i = 0;i<data.length;i++){
        let salida = network.execute(data[i].input)
        for(let i2 = 0;i2<salida.length;i2++){
            error += Math.abs(salida[i2]-parseInt(data[i].output[i2].toFixed(20)));
        }
    }
    return -error;
}


let ga = new GeneticAlgorithm(2,[2],1,10,0.4,0.2,0.2,4,fitness,false)
const fs = require('fs');
let rawdata = fs.readFileSync('network.json')
let parseddata = JSON.parse(rawdata)
let network = NeuralNetwork.fromJSON(parseddata);
ga.start(network);

let bestfitness = -Infinity;
while(bestfitness !== 0){
    let overall = ga.step();
    bestfitness = ga.best().fitness;
    console.log(`Generation: ${ga.generation} best: ${bestfitness} overall: ${overall}`);
}
let best = ga.best();
console.log(best.execute([0,0]))
console.log(best.execute([0,1]))
console.log(best.execute([1,0]))
console.log(best.execute([1,1]))