// S to save/download dataset

// Enter to train

// Click to predict

​

let data = [];

let nn;

let training = true;

let predictingX = 0;

let predictedY = 0;

​

function setup() {

  createCanvas(400, 400);

​

  /*

  The layer definition for regression is

  supposedly:

  let layers: [

    {

      type: 'dense',

      units: this.options.hiddenUnits,

      activation: 'relu',

    },

    {

      type: 'dense',

      activation: 'sigmoid',

    },

  ];

  */

  let options = {

    task: 'regression',

    inputs: ['x'],

    outputs: ['y'],

    debug: true,

    learningRate: 0.2,  // supposedly default

    hiddenUnits: 16,    // supposedly default

  };

  nn = ml5.neuralNetwork(options);

}

​

function draw() {

  background(0);

  noStroke();

​

  // training data

  let points = nn.neuralNetworkData.data.raw;

  for (let i=0; i < points.length; i++) {

    let x = map(points[i].xs.x, 0, 1, 0, width);

    let y = map(points[i].ys.y, 0, 1, height, 0);

    fill(255);

    ellipse(x, y, 6, 6);

  }

  // prediction

  fill(0, 0, 255);

  let x = map(predictingX, 0, 1, 0, width);

  let y = map(predictedY, 0, 1, height, 0);

  ellipse(x, y, 6, 6);

}

​

function mousePressed() {

  let x = map(mouseX, 0, width, 0, 1);

  let y = map(mouseY, 0, height, 1, 0);

​

  if (training) {

    // artificially generate some points around the mouse position

    for (let i=0; i < 10; i++) {

      nn.addData({ x: x+random(-0.02,0.02) }, { y: y+random(-0.02,0.02) });

    }

  } else {

    predictingX = x;

    nn.predict({ x: x }, handleResults);

  }

}

​

function keyPressed() {

  if (keyCode == ENTER) {

    nn.normalizeData();

​

    let options = {

      epochs: 32,     // supposedly default

      batchSize: 12,  // supposedly default 

    };

    nn.train(options, finishedTraining);

  } else if (key == 's' || key == 'S') {

    nn.saveData();

  } else if (key == 'l' || key == 'L') {

    nn.loadData('uploaded.json');

  }

}

​

function train() {

}

​

function finishedTraining() {

  console.log('Finished training');

  training = false;

}

​

function handleResults(error, result) {

  if (error) {

    console.error(error);

  } else {

    predictedY = result[0].y;

    console.log('Predicted y ' + predictedY + ' for x ' + predictingX);

  }

}