let len = 784; // 28 * 28;

let catData = [];

let rainbowData = [];

let trainData = [];

let myNN;

​

function preload() {

  // load in the files

  catFile = loadBytes("./data/cats1000.bin");

  rainbowFile = loadBytes("./data/rainbows1000.bin");

  trainFile = loadBytes("./data/trains1000.bin");

}

​

function prepareData() {

  // we know we hive a binary file with 1000 images and each one has 784 points

  for (let i = 0; i < 1000; i++) {

    // to read, access all points inside

    catData[i] = []; // to hold the 784 nums from each of the 1000 images into an array that will be inside the catDta array

    rainbowData[i] = [];

    trainData[i] = [];

    for (let j = 0; j < len; j++) {

      // divide each by 255 to normalize the data

      const catValue = (catFile.bytes[i*len + j]) / 255;

      const rainbowValue = (rainbowFile.bytes[i*len + j]) / 255;

      const trainValue = (trainFile.bytes[i*len + j]) / 255;

      catData[i].push(catValue);

      rainbowData[i].push(rainbowValue);

      trainData[i].push(trainValue);

    }

  }

  console.log("catData", catData);

}

function addData() {

  const options = {

    task: "classification",

    inputs: 784, // shape of our data

    outputs: 3, // num of classes we have

    debug: true, // make it show data

  };

  myNN = ml5.neuralNetwork(options);

  // loop through and add them all as input to neural network

  for (let i = 0; i < 1000; i++) {

    myNN.addData(catData[i], ["cat"]);

    myNN.addData(rainbowData[i], ["rainbow"]);

    myNN.addData(trainData[i], ["train"]);

  }

}

// mynn.addData([1, 0, 0, 0 ...], 'cat')

function setup() {

  createCanvas(400, 400);

  background(255);

  stroke(0);

  strokeWeight(30)

  prepareData();

  addData();

  myNN.train(

    {

      epochs: 10,

    },

    doneTraining

  );

}

​

function draw() {

  if (mouseIsPressed){

    line(pmouseX, pmouseY, mouseX, mouseY);

  }

}

​

function mouseReleased(){

  //if (isModelReady){

    const myInput = []

    let c = get();

    c.resize(28, 28);

    c.loadPixels();

    for (let i = 0; i < len; i++){

      const value = (255 - c.pixels[i * 4]) / 4

      myInput.push(value)

    }

    console.log(c.pixels)

    myNN.classify(myInput, gotResults);

}

function doneTraining() {}

​

function gotResults(error, results){

  console.log(results)

}

​

​