NeuroEvo

a simple Javascript evolutionary algorithm for training neural networks

Neural net usage

Create a neural net:

// A network with 2 inputs, 1 output, 3 hidden layers, and 5 neurons per hidden layer
let network = new NN(2,1,3,5)

From weights array:

let network = NN.fromWeights(nnArchitecture, inputWeights)

run the net on data:

network.run([10,20,30])

Evolution Usage

Create a new generation:

let generation = new Generation(20, new NN(2,1,3,5))

Repeatedly assign fitness and evolve the generation:

for (let j = 0; j < 1000; j++) { //1000 generations
    for (let i = 0; i < 20; i++) { //assign fitness values to each of the generations
        generation.setFitness(i, - Math.abs(0 - generation.pop[i].run([-1, 1])[0])); //set fitness to negative distance from 0
    }
    generation.evolve();
}
//log the best network's answer
console.log(generation.gentop[generation.gentop.length - 1][0].run([-1,1])[0])
//possible output: 0.002746631311830884