// https://thecodingtrain.com/CodingChallenges/106-xor-tfjs.html

// XOR with tf.js

// https://youtu.be/N3ZnNa01BPM

​

let nn;

let model;

​

let resolution = 20;

let cols;

let rows;

​

let xs;

​

const train_xs = tf.tensor2d([

  [0, 0],

  [1, 0],

  [0, 1],

  [1, 1]

]);

const train_ys = tf.tensor2d([

  [0],

  [1],

  [1],

  [0]

]);

​

function setup() {

  createCanvas(600, 600);

  cols = width / resolution;

  rows = height / resolution;

​

  // Create the input data

  let inputs = [];

  for (let i = 0; i < cols; i++) {

    for (let j = 0; j < rows; j++) {

      let x1 = i / cols;

      let x2 = j / rows;

      inputs.push([x1, x2]);

    }

  }

  xs = tf.tensor2d(inputs);

​

​

  model = tf.sequential();

  let hidden = tf.layers.dense({

    inputShape: [2],

    units: 16,

    activation: 'sigmoid'

  });

  let output = tf.layers.dense({

    units: 1,

    activation: 'sigmoid'

  });

  model.add(hidden);

  model.add(output);

​

  const optimizer = tf.train.adam(0.25);

  model.compile({

    optimizer: optimizer,

    loss: 'meanSquaredError'

  })

​

  setTimeout(train, 10);

​

}

​

function train() {

  trainModel().then(result => {

    ///console.log(result.history.loss[0]);

    setTimeout(train, 10);

  });

}

​

function trainModel() {

  return model.fit(train_xs, train_ys, {

    shuffle: true,

    epochs: 1

  });

}

​

function draw() {

  background(0);

​

  tf.tidy(() => {

    // Get the predictions

    let ys = model.predict(xs);

    let y_values = ys.dataSync();

​

    // Draw the results

    let index = 0;

    for (let i = 0; i < cols; i++) {

      for (let j = 0; j < rows; j++) {

        let br = y_values[index] * 255

        fill(br);

        rect(i * resolution, j * resolution, resolution, resolution);

        fill(255 - br);

        textSize(8);

        textAlign(CENTER, CENTER);

        text(nf(y_values[index], 1, 2), i * resolution + resolution / 2, j * resolution + resolution / 2)

        index++;

      }

    }

  });

​

}