parallel_neat.js

import { Architect, Config, Methods, Network, Neat } from 'neataptic'
import Parallel from 'paralleljs'

Config.warnings = false

const trainingSet = [
  { input: [0, 0], output: [0] },
  { input: [0, 1], output: [1] },
  { input: [1, 0], output: [1] },
  { input: [1, 1], output: [0] },
]


function evolveNet() {
  const neat = new Neat(
    2,
    1,
    undefined,
    {
      popsize: 50,
      mutationRate: 0.5,
      mutationAmount: 5,
      equal: true,
      network: new Architect.Random(
        2,
        0,
        1
      )
    }
  );

  neat.mutate()

  startEval()

  function startEval() {
    const env = {
      trainingSet
    }

    // Deep cloning neat population here, could help with speed 
    // so that genomes don't have to be reconverted from JSON
    const p = new Parallel(neat.population, {
      maxWorkers: 4, // change this depending on context/threads available
      env
    })

    p.map(fitnessFunctionP).then(data => {
      neat.population = data.map((d, i) =>
        Object.assign(Network.fromJSON(d.genome), { score: d.score })
      )

      endEval()
    })
  }

  function endEval() {
    neat.sort()
    const fittest = neat.population[0]

    console.log('Generation ' + neat.generation + ' Best score ' + fittest.score)
    if(fittest.score < -0.03 && neat.generation < 500) {
      const newPopulation = [];

      // Elitism
      for(let i = 0; i < neat.elitism; i++){
        newPopulation.push(neat.population[i]);
      }

      // Breed the next individuals
      for(let i = 0; i < neat.popsize - neat.elitism; i++){
        newPopulation.push(neat.getOffspring());
      }
      neat.population = newPopulation
      neat.mutate()
      neat.generation++
      setTimeout(startEval, 100)
    } else {
      console.log('Done, outputs:' +
       ' [0,0] ' + fittest.activate([0,0]) +
       ' [0,1] ' + fittest.activate([0,1]) +
       ' [1,0] ' + fittest.activate([1,0]) +
       ' [1,1] ' + fittest.activate([1,1])
      )
    }
  }
}

function fitnessFunctionP(genome){
  const Network =  require('neataptic').Network
  const Methods = require('neataptic').Methods

  genome = Network.fromJSON(genome)
  const score = -genome.test(global.env.trainingSet, Methods.Cost.MSE).error
  return { genome, score }
}

evolveNet()