Code viewer for World: Character recognition neur...
// Port of Character recognition neural network from here:
// https://github.com/CodingTrain/Toy-Neural-Network-JS/tree/master/examples/mnist
// with many modifications 


// --- defined by MNIST - do not change these ---------------------------------------

const CROP_PIXELS   = 24; // AGT: Random 24x24 crops are used to train WebCNN
const PIXELS        = 28;                       // images in data set are tiny 
const PIXELSSQUARED = PIXELS * PIXELS;

// number of training and test exemplars in the data set:
const NOTRAIN = 60000;
const NOTEST  = 10000;



//--- can modify all these --------------------------------------------------

// no of nodes in network 
const noinput  = PIXELSSQUARED;
const nohidden = 64;
const nooutput = 10;

const learningrate = 0.1;   // default 0.1  

// should we train every timestep or not 
let do_training = true;

// how many to train and test per timestep 
const BATCH_SIZE = 50;
const TRAINPERSTEP = BATCH_SIZE;
const TESTPERSTEP  = 5;

// multiply it by this to magnify for display 
const ZOOMFACTOR    = 7;                        
const ZOOMPIXELS    = ZOOMFACTOR * PIXELS; 

// 3 rows of
// large image + 50 gap + small image    
// 50 gap between rows 

const canvaswidth = ( PIXELS + ZOOMPIXELS ) + 50;
const canvasheight = ( ZOOMPIXELS * 3 ) + 100;


const DOODLE_THICK = 18;    // thickness of doodle lines 
// AGT:
const DOODLE_BLUR = 0;      // blur factor applied to doodles 


let mnist;      
// all data is loaded into this 
// mnist.train_images
// mnist.train_labels
// mnist.test_images
// mnist.test_labels


// AGT: pick the network to use in The World
// 0: 1-hidden-layer network with ReLU activation (for the hidden layer)
// 1: 1-hidden-layer network with Tanh activation (for the hidden layer)
// 2: Original WebCNN
// 3: Original WebCNN (pretrained)
const theNN = 3;
let nn;

let trainrun = 1;
let train_index = 0;

let testrun = 1;
let test_index = 0;
let total_tests = 0;
let total_correct = 0;

// images in LHS:
let doodle, demo;
let doodle_exists = false;
let demo_exists = false;

let mousedrag = false;      // are we in the middle of a mouse drag drawing?  


// save inputs to global var to inspect
// type these names in console 
var train_inputs, test_inputs, demo_inputs, doodle_inputs;


// Matrix.randomize() is changed to point to this. Must be defined by user of Matrix. 

function randomWeight()
{
    return ( AB.randomFloatAtoB ( -0.5, 0.5 ) );
            // Coding Train default is -1 to 1
}    



// CSS trick 
// make run header bigger 
 $("#runheaderbox").css ( { "max-height": "95vh" } );



//--- start of AB.msgs structure: ---------------------------------------------------------
// We output a serious of AB.msgs to put data at various places in the run header 
var thehtml;

  // 1 Doodle header 
  thehtml = "<hr> <h1> 1. Doodle </h1> Top row: Doodle (left) and shrunk (right). <br> " +
        " Draw your doodle in top LHS. <button onclick='wipeDoodle();' class='normbutton' >Clear doodle</button> <br> ";
   AB.msg ( thehtml, 1 );

  // 2 Doodle variable data (guess)
  
  // 3 Training header
  thehtml = "<hr> <h1> 2. Training </h1> Middle row: Training image magnified (left) and original (right). <br>  " +
        " <button onclick='do_training = false;' class='normbutton' >Stop training</button> <br> ";
  AB.msg ( thehtml, 3 );
     
  // 4 variable training data 
  
  // 5 Testing header
  thehtml = "<h3> Hidden tests </h3> " ;
  AB.msg ( thehtml, 5 );
           
  // 6 variable testing data 
  
  // 7 Demo header 
  thehtml = "<hr> <h1> 3. Demo </h1> Bottom row: Test image magnified (left) and  original (right). <br>" +
        " The network is <i>not</i> trained on any of these images. <br> " +
        " <button onclick='makeDemo();' class='normbutton' >Demo test image</button> <br> ";
   AB.msg ( thehtml, 7 );
   
  // 8 Demo variable data (random demo ID)
  // 9 Demo variable data (changing guess)
  
const greenspan = "<span style='font-weight:bold; font-size:x-large; color:darkgreen'> "  ;

//--- end of AB.msgs structure: ---------------------------------------------------------




function setup() 
{
  createCanvas ( canvaswidth, canvasheight );

  doodle = createGraphics ( ZOOMPIXELS, ZOOMPIXELS );       // doodle on larger canvas 
  doodle.pixelDensity(1);
  
// JS load other JS 
// maybe have a loading screen while loading the JS and the data set 

    AB.loadingScreen();
 
    $.getScript ( "/uploads/codingtrain/mnist.js", function()
    {
        // AGT: Load WebCNN Math utils, network definition, pre-defined weights.
        $.getScript ( "/uploads/atotev/mathutils.js", function()
        {
            $.getScript ( "/uploads/atotev/webcnn.js", function()
            {
                $.getJSON ( "/uploads/atotev/cnn_mnist_10_20_98accuracy.json", function(networkJSON)
                {
                    console.log ("All JS loaded");
                    
                    if (theNN===0) {
                        nn = createShallowNetwork(ACTIVATION_RELU);
                    } else if (theNN===1) {
                        nn = createShallowNetwork(ACTIVATION_TANH);
                    } else if (theNN===2) {
                        nn = createDefaultNetwork();
                    } else if (theNN===3) {
                        nn = loadNetworkFromJSON(networkJSON);
                    } else {
                        console.log("Unknown NN type: " + theNN);
                    }
                    loadData();
                });
            });
        });
    });
}

// AGT: Pre-trained WebCNN
function loadNetworkFromJSON( networkJSON )
{
	let result = new WebCNN();

	if (networkJSON.momentum !== undefined) result.setMomentum( networkJSON.momentum );
	if (networkJSON.lambda !== undefined) result.setLambda( networkJSON.lambda );
	if (networkJSON.learningRate !== undefined) result.setLearningRate( networkJSON.learningRate );

	for ( let layerIndex = 0; layerIndex < networkJSON.layers.length; ++layerIndex )
	{
		let layerDesc = networkJSON.layers[ layerIndex ];
		console.log( layerDesc );
		result.newLayer( layerDesc );
	}

	for ( let layerIndex = 0; layerIndex < networkJSON.layers.length; ++layerIndex )
	{
		let layerDesc = networkJSON.layers[ layerIndex ];

		switch ( networkJSON.layers[ layerIndex ].type )
		{
			case LAYER_TYPE_CONV:
			case LAYER_TYPE_FULLY_CONNECTED:
			{
				if (layerDesc.weights !== undefined && layerDesc.biases !== undefined )
				{
					result.layers[ layerIndex ].setWeightsAndBiases( layerDesc.weights, layerDesc.biases );
				}
				break;
			}
		}
	}

	result.initialize();
    return result;
}

// AGT: Default WebCNN with randmonly initialized weights
function createDefaultNetwork()
{
	let result = new WebCNN();
	result.newLayer( { name: "image", type: LAYER_TYPE_INPUT_IMAGE, width: CROP_PIXELS, height: CROP_PIXELS, depth: 1 } );
	result.newLayer( { name: "conv1", type: LAYER_TYPE_CONV, units: 10, kernelWidth: 5, kernelHeight: 5, strideX: 1, strideY: 1, padding: false } );
	result.newLayer( { name: "pool1", type: LAYER_TYPE_MAX_POOL, poolWidth: 2, poolHeight: 2, strideX: 2, strideY: 2 } );
	result.newLayer( { name: "conv2", type: LAYER_TYPE_CONV, units: 20, kernelWidth: 5, kernelHeight: 5, strideX: 1, strideY: 1, padding: false } );
	result.newLayer( { name: "pool2", type: LAYER_TYPE_MAX_POOL, poolWidth: 2, poolHeight: 2, strideX: 2, strideY: 2 } );
	result.newLayer( { name: "out", type: LAYER_TYPE_FULLY_CONNECTED, units: 10, activation: ACTIVATION_SOFTMAX } );
	result.initialize();

	result.setLearningRate( 0.01 );
	result.setMomentum( 0.9 );
	result.setLambda( 0.0 );
    return result;
}

// AGT: A single-hidden-layer network (similar to the original NN) with tanh activation for the hidden layers
function createShallowNetwork(activation)
{
	let result = new WebCNN();
	result.newLayer( { name: "image", type: LAYER_TYPE_INPUT_IMAGE, width: CROP_PIXELS, height: CROP_PIXELS, depth: 1 } );
	result.newLayer( { name: "hidden", type: LAYER_TYPE_FULLY_CONNECTED, units: nohidden, activation: activation } );
	result.newLayer( { name: "out", type: LAYER_TYPE_FULLY_CONNECTED, units: nooutput, activation: ACTIVATION_SOFTMAX } );
	result.initialize();

	result.setLearningRate( 0.01 );
	result.setMomentum( 0.9 );
	result.setLambda( 0.0 );
    return result;
}

// load data set from local file (on this server)
function loadData()    
{
  loadMNIST ( function(data)    
  {
    mnist = data;
    console.log ("All data loaded into mnist object:")
    console.log(mnist);
    AB.removeLoading();     // if no loading screen exists, this does nothing 
  });
}

// AGT: Calculates bounding box and translates to centre
function centerImage(pixels, size) {

    // convert to matrix
    let m = [];
    for (let i = 0; i < size; i++) 
    {
        m[i] = [];
        for (let j = 0; j < size; j++)
        {
            m[i][j] = pixels[(i*size + j) * 4];
        }
    }
    
    // find bounding box
    var topmost = Number.MAX_VALUE;
    var leftmost = Number.MAX_VALUE;
    var bottommost = -1;
    var rightmost = -1;
    for(var y = 0; y < m.length; y++)
    {
      var l = m[y].indexOf(255);
      var r = m[y].lastIndexOf(255);
      if (l >= 0 && l < leftmost) leftmost = l;
      if (r >= 0 && r > rightmost) rightmost = r;
      // only check if some 1 found
      if (l >= 0 && y < topmost) topmost = y;
      if (l >= 0 && y > bottommost) bottommost = y;
    }
    //console.log("AGT: " + leftmost + " " + topmost + " " + rightmost + " " + bottommost)
    
    // translate to centre
    let transY = Math.floor((size - bottommost - topmost) / 2);
    let transX = Math.floor((size - rightmost - leftmost) / 2);
    let result = Array(size).fill().map(() => Array(size).fill(0));
    for (i=topmost; i<=bottommost; i++) {
        for(j=leftmost; j<=rightmost; j++) {
            result[i + transY][j + transX] = m[i][j];
        }
    }

    // Convert back to 1D array
    let m1D = [];
    for (let i = 0; i < size; i++) 
    {
        for (let j = 0; j < size; j++)
        {
            m1D[i*size + j] = result[i][j];
        }
    }
    return m1D;
}

// AGT: Modified to work with both original and cropped input size
function getImage ( img, size=PIXELS )      // make a P5 image object from a raw data array   
{
    let theimage  = createImage (size, size);    // make blank image, then populate it 
    theimage.loadPixels();        
    
    for (let i = 0; i < size*size ; i++) 
    {
        let bright = img[i];
        let index = i * 4;
        theimage.pixels[index + 0] = bright;
        theimage.pixels[index + 1] = bright;
        theimage.pixels[index + 2] = bright;
        theimage.pixels[index + 3] = 255;
    }
    
    theimage.updatePixels();
    return theimage;
}

// AGT: Crop a random size*size part from img
function randomCrop ( img, size )      // convert img array into normalised input array 
{
    const maxStartIndex = PIXELS - size;
    let xrand = Math.floor( Math.random() * maxStartIndex );
    let yrand = Math.floor( Math.random() * maxStartIndex );
    return crop(img, size, xrand, yrand);
}

// AGT: Crop a size*size part from image starting at (x, y)
function crop ( img, newSize, x=2, y=2 )      // convert img array into normalised input array 
{
    const originalSize = PIXELS;
    let xEndIndex = x + newSize;
    let yEndIndex = y + newSize;
    let inputs = [];
    for (let i = x; i < xEndIndex ; i++)          
    {
        for (let j = y; j < yEndIndex ; j++)          
        {
            inputs.push(img[ i*originalSize + j ])
        } 
    } 
    // console.log(inputs);
    return ( inputs );
}

function getInputs ( img )      // convert img array into normalised input array 
{
    let inputs = [];
    for (let i = 0; i < PIXELSSQUARED ; i++)          
    {
        let bright = img[i];
        inputs[i] = bright / 255;       // normalise to 0 to 1
    } 
    return ( inputs );
}

// AGT: Utility function to convert image to object expected by the active NN
function toModelFormat(image, size) {
    return {
	    "width": size,
	    "height": size,
	    "data": getImage (randomCrop(image, size), size).pixels
    };
}


// AGT: Modified to train on mini batches 
function trainit (show)        // train the network with a single exemplar, from global var "train_index", show visual on or off 
{
  if (train_index%TRAINPERSTEP!==0) {
      train_index++;
      return;
  }

  // Train on new batch
  let img   = mnist.train_images[train_index];
  let label = mnist.train_labels[train_index];
  
  let imgs = [];
  let labels = [];
  for (i=0; i<TRAINPERSTEP; i++) {
    imgs.push(toModelFormat(mnist.train_images[train_index+i], CROP_PIXELS));
    labels.push(mnist.train_labels[train_index+i]);
  }
  
  // set up the inputs
  let inputs = img;       // get inputs from data array 
  
  // AGT: Always show
  var theimage = getImage ( img );    // get image from data array 
  image ( theimage,   0,                ZOOMPIXELS+50,    ZOOMPIXELS,     ZOOMPIXELS  );      // magnified 
  image ( theimage,   ZOOMPIXELS+50,    ZOOMPIXELS+50,    PIXELS,         PIXELS      );      // original

  // console.log(train_index);
  // console.log(inputs);
  // console.log(targets);

  train_inputs = inputs;        // can inspect in console 
  nn.trainCNNClassifier ( imgs, labels );

  thehtml = " trainrun: " + trainrun + "<br> no: " + train_index ;
  AB.msg ( thehtml, 4 );

  train_index++;
  if ( train_index+TRAINPERSTEP >= NOTRAIN ) 
  {
    train_index = 0;
    console.log( "finished trainrun: " + trainrun );
    trainrun++;
  }
}

// AGT: Modified to work with WebCNN
function testit()    // test the network with a single exemplar, from global var "test_index"
{ 
  let img   = mnist.test_images[test_index];
  let label = mnist.test_labels[test_index];

  // set up the inputs
  let inputs = getInputs ( img ); 
  
  test_inputs = inputs;        // can inspect in console 
  let result    = nn.classifyImages( [ toModelFormat(mnist.test_images[test_index], CROP_PIXELS) ] );
  let guess = findMax(result)

  total_tests++;
  if (guess == label)  total_correct++;

  let percent = (total_correct / total_tests) * 100 ;
  
  thehtml =  " testrun: " + testrun + "<br> no: " + total_tests + " <br> " +
        " correct: " + total_correct + "<br>" +
        "  score: " + greenspan + percent.toFixed(2) + "</span>";
  AB.msg ( thehtml, 6 );

  test_index++;
  if ( test_index == NOTEST ) 
  {
    console.log( "finished testrun: " + testrun + " score: " + percent.toFixed(2) );
    testrun++;
    test_index = 0;
    total_tests = 0;
    total_correct = 0;
  }
}




//--- find no.1 (and maybe no.2) output nodes ---------------------------------------
// (restriction) assumes array values start at 0 (which is true for output nodes) 
// AGT: Modified to work with WebCNN
function find12 (result)         // return array showing indexes of no.1 and no.2 values in array 
{
  let guess1 = 0;
  let max1 = 0;
  let guess2 = 0;
  let max2 = 0;  
  for ( var i = 0; i < 10; ++i )
  {
    if ( result[ 0 ].getValue( 0, 0, i ) > max1 )
    {
        max1 = result[ 0 ].getValue( 0, 0, i );
        guess1 = i;
    } else if ( result[ 0 ].getValue( 0, 0, i ) > max2 )
    {
        max2 = result[ 0 ].getValue( 0, 0, i );
        guess2 = i;
    }
  }
  return [guess1, guess2];
}


// AGT: Modified to work with WebCNN
// just get the maximum - separate function for speed - done many times 
// find our guess - the max of the output nodes array
function findMax (result)        
{
  let guess = 0;
  let max = 0;
  for ( var i = 0; i < 10; ++i )
  {
    if ( result[ 0 ].getValue( 0, 0, i ) > max )
    {
        max = result[ 0 ].getValue( 0, 0, i );
        guess = i;
    }
  }
  return guess;
}




// --- the draw function -------------------------------------------------------------
// every step:
// AGT: Modify function in order to pause training while user is drawing a doodle
function draw() 
{
  // check if libraries and data loaded yet:
  if ( typeof mnist == 'undefined' ) return;

  // how can we get white doodle on black background on yellow canvas?
//        background('#ffffcc');    doodle.background('black');

  background ('black');
  
  // keep drawing demo and doodle images 
  // and keep guessing - we will update our guess as time goes on 
  
  if ( demo_exists )
  {
    drawDemo();
    guessDemo();
  }
  if ( doodle_exists ) 
  {
    drawDoodle();
    guessDoodle();
  }


// detect doodle drawing 
// (restriction) the following assumes doodle starts at 0,0 

  if ( mouseIsPressed )         // gets called when we click buttons, as well as if in doodle corner  
  {
     // console.log ( mouseX + " " + mouseY + " " + pmouseX + " " + pmouseY );
     var MAX = ZOOMPIXELS + 20;     // can draw up to this pixels in corner 
     if ( (mouseX < MAX) && (mouseY < MAX) && (pmouseX < MAX) && (pmouseY < MAX) )
     {
        mousedrag = true;       // start a mouse drag 
        doodle_exists = true;
        doodle.stroke('white');
        doodle.strokeWeight( DOODLE_THICK );
        doodle.line(mouseX, mouseY, pmouseX, pmouseY);      
     }
  }
  else if ( mousedrag )
  {
        mousedrag = false;
        console.log ("Exiting draw. Now blurring.");
        doodle.filter (BLUR, DOODLE_BLUR);    // just blur once 
        //   console.log (doodle);
  } else if ( do_training ) {
      // do some training per step 
        for (let i = 0; i < TRAINPERSTEP; i++) 
        {
          if (i === 0)    trainit(true);    // show only one per step - still flashes by  
          else           trainit(false);
        }
        
      // do some testing per step 
        for (let i = 0; i < TESTPERSTEP; i++) 
          testit();

  }
}




//--- demo -------------------------------------------------------------
// demo some test image and predict it
// get it from test set so have not used it in training


function makeDemo()
{
    demo_exists = true;
    var  i = AB.randomIntAtoB ( 0, NOTEST - 1 );  
    
    demo        = mnist.test_images[i];     
    var label   = mnist.test_labels[i];
    
   thehtml =  "Test image no: " + i + "<br>" + 
            "Classification: " + label + "<br>" ;
   AB.msg ( thehtml, 8 );
   
   // type "demo" in console to see raw data 
}


function drawDemo()
{
    var theimage = getImage ( demo );
     //  console.log (theimage);
     
    image ( theimage,   0,                canvasheight - ZOOMPIXELS,    ZOOMPIXELS,     ZOOMPIXELS  );      // magnified 
    image ( theimage,   ZOOMPIXELS+50,    canvasheight - ZOOMPIXELS,    PIXELS,         PIXELS      );      // original
}


function guessDemo()
{
   let inputs = getInputs ( demo ); 
   
  demo_inputs = inputs;  // can inspect in console 
  
  let result    = nn.classifyImages( [ toModelFormat(demo, CROP_PIXELS) ] );
  let guess = findMax(result);
	
   thehtml =   " We classify it as: " + greenspan + guess + "</span>" ;
   AB.msg ( thehtml, 9 );
}




//--- doodle -------------------------------------------------------------

function drawDoodle()
{
    // doodle is createGraphics not createImage
    let theimage = doodle.get();
    // console.log (theimage);
    
    image ( theimage,   0,                0,    ZOOMPIXELS,     ZOOMPIXELS  );      // original 
    image ( theimage,   ZOOMPIXELS+50,    0,    PIXELS,         PIXELS      );      // shrunk
}
      
      
function guessDoodle() 
{
   // doodle is createGraphics not createImage
   let img = doodle.get();
  
  img.resize ( PIXELS, PIXELS );
  img.loadPixels();
  
  // feed forward to make prediction 
  const result = nn.classifyImages( [ toModelFormat(centerImage(img.pixels, PIXELS), CROP_PIXELS)] );
  let guess = find12(result);

  thehtml =   " We classify it as: " + greenspan + guess[0] + "</span> <br>" +
            " No.2 guess is: " + greenspan + guess[1] + "</span>";
  AB.msg ( thehtml, 2 );
}


function wipeDoodle()    
{
    doodle_exists = false;
    doodle.background('black');
}




// --- debugging --------------------------------------------------
// in console
// showInputs(demo_inputs);
// showInputs(doodle_inputs);


function showInputs ( inputs )
// display inputs row by row, corresponding to square of pixels 
{
    var str = "";
    for (let i = 0; i < inputs.length; i++) 
    {
      if ( i % PIXELS === 0 )    str = str + "\n";                                   // new line for each row of pixels 
      var value = inputs[i];
      str = str + " " + value.toFixed(2) ; 
    }
    console.log (str);
}