Training a neural network to recognise 15 inputs worth of Shapes

synaptic-shape-training.js

const Layer = require('synaptic').Layer;
const Network = require('synaptic').Network;
const Trainer = require('synaptic').Trainer;

const inputLayer = new Layer(15);
const hiddenLayer = new Layer(40);
const outputLayer = new Layer(4);

inputLayer.project(hiddenLayer);
hiddenLayer.project(outputLayer);

const myNetwork = new Network({
  input: inputLayer,
  hidden: [hiddenLayer],
  output: outputLayer
});

const shapesMap = {
  0: 'RECTANGLE',
  1: 'TRIANGLE',
  2: 'SQUARE',
  3: 'CIRCLE'
};

let trainingData = [
  { // Rectangle
    input: [1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1, 1], 
    output: [1, 0, 0, 0]
  }, 
  { // Triangle
    input: [0,0,1,0,0, 0,1,0,1,0, 1,1,1,1,1], 
    output: [0, 1, 0, 0]
  }, 
  { // Square
    input: [1,1,1,0,0, 1,0,1,0,0, 1,1,1,0,0], 
    output: [0, 0, 1, 0]
  }, 
  { // Circle
    input: [0,0,1,0,0, 0,1,0,1,0, 0,0,1,0,0], 
    output: [0, 0, 0, 1]
  },
  { // Square 2
    input: [0,1,1,1,0, 0,1,0,1,0, 0,1,1,1,0],
    output: [0, 0, 1, 0]
  },
];

const toChunks = function(array, size) {
  var results = [];
  while (array.length) results.push(array.splice(0, size));
  return results;
};

const trainer = new Trainer(myNetwork);

trainer.train(trainingData, {
  rate: .2,
  cost: Trainer.cost.CROSS_ENTROPY,
  iterations: 200,
  log: 10,
});

const badlyDrawnSquare = [0,0,1,1,0, 0,1,0,1,0, 0,1,1,1,0];

console.log('---------- Testing training data --------- ')
console.log(myNetwork.activate(trainingData[0].input));

const badSquarePrediction = myNetwork.activate(badlyDrawnSquare);
const index = badSquarePrediction.indexOf(Math.max(...badSquarePrediction));
const percentage = Math.max(...badSquarePrediction);

const draw = badlyDrawnSquare
  .map((item, i) => item === 1 ? 'x' : ' ');

console.log(toChunks(draw, 5))
console.log(`Prediction is ${shapesMap[index]} with ${(percentage * 100).toFixed(2)}% accuracy` );