// --- defined by MNIST - do not change these ---------------------------------------
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 = 124800;
const NOTEST = 20800;
//--- can modify all these --------------------------------------------------
// no of nodes in network
const noinput = PIXELSSQUARED;
const learningRate = 0.1;
// should we train every timestep or not
let do_training = true;
// how many to train and test per timestep
const TRAINPERSTEP = 800;
const TESTPERSTEP = 1;
// 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
const DOODLE_BLUR = 3; // blur factor applied to doodles
const numEpochs = 3;
let mnist, trainingImages, trainingLabels, testImages, testLabels, encodedTestLabels, encodedTrainingLabels, networkModel, normalizeFactor;
let index, trainingInProgress = false, testingInProgress = false;
// all data is loaded into this
// mnist.train_images
// mnist.train_labels
// mnist.test_images
// mnist.test_labels
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;
// make run header bigger
AB.headerCSS ( { "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("https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@2.0.0/dist/tf.min.js", function() {
$.getScript ( "/uploads/aparna/mnist-az.js", function()
{
console.log ("All JS loaded");
loadData();
});
});
}
// load data set from local file (on this server)
/***
* By Aparna
* Configure CNN
*/
function loadData()
{
loadMNIST ( function(data)
{
mnist = data;
console.log ("All data loaded into mnist object:")
console.log(mnist);
// CODE CHANGE BY APARNA
// Use Tensorflow framework for setting up a CNN.
// In the first layer of our convolutional neural network we have
// to specify the input shape. Then we specify some parameters for
// the convolution operation that takes place in this layer.
// The MaxPooling layer acts as a sort of downsampling using max values
// in a region instead of averaging.
// Repeat another conv2d + maxPooling stack.
// Note that we have more filters in the convolution.
// Then we flatten the output from the 2D filters into a 1D vector to prepare
// it for input into our last layer. This is common practice when feeding
// higher dimensional data to a final classification output layer.
networkModel = tf.sequential();
networkModel.add(tf.layers.conv2d({activation: "relu", filters: 8, inputShape: [PIXELS, PIXELS, 1], kernelInitializer: "varianceScaling", kernelSize: 5, strides: 1}));
networkModel.add(tf.layers.maxPooling2d({poolSize: [2, 2], strides: [2, 2]}));
networkModel.add(tf.layers.conv2d({activation: "relu", filters: 16, kernelInitializer: "varianceScaling", kernelSize: 5, strides: 1}));
networkModel.add(tf.layers.maxPooling2d({poolSize: [2, 2], strides: [2, 2]}));
networkModel.add(tf.layers.flatten());
networkModel.add(tf.layers.dense({activation: "softmax", kernelInitializer: "varianceScaling", units: 26}));
// Configure the learning rate with Stochastic Gradient Descent.
networkModel.compile({loss: "categoricalCrossentropy", optimizer: tf.train.sgd(learningRate)});
// A scalar tensor with value 255, for normalizing input data.
normalizeFactor = tf.scalar(255);
console.log("loading completed");
AB.removeLoading(); // if no loading screen exists, this does nothing
});
}
/**
* This function perform training as a batch.
* BY APARNA
*/
function trainBatch()
{
// Fetch a batch of data.
[trainImagesBatch, trainLabelsBatch, train_index] = nextBatch(mnist.train_images, mnist.train_labels, TRAINPERSTEP, train_index);
// Convert data to tensor, and normalize data by dividing with 255.
trainingImages = tf.reshape(tf.tensor(trainImagesBatch), [TRAINPERSTEP, 28, 28, 1]).div(normalizeFactor);
trainingLabels = tf.tensor1d(trainLabelsBatch, "int32");
// Generate one-hot encoding for labels.
encodedTrainingLabels = tf.oneHot(trainingLabels, 26);
// Create callback to peform any action during training, if required.
let t = {callbacks: {onBatchEnd: (e, o) => {
}, onEpochBegin: (e, t) => {
currentEpoch = e + 1;
}}, batchSize: TRAINPERSTEP, epochs: numEpochs, shuffle: true};
trainingInProgress = true;
// Train the model.
networkModel.fit(trainingImages, encodedTrainingLabels, t).then(() => {
trainingInProgress = false;
});
thehtml = " trainrun: " + trainrun + "<br> no: " + train_index ;
AB.msg ( thehtml, 4 );
// Upon reaching the final batch of the dataset, restart from the first batch.
if ( train_index == NOTRAIN - TRAINPERSTEP)
{
train_index = 0;
console.log( "finished trainrun: " + trainrun );
trainrun++;
}
}
/**
* BY APARNA
* Slice a batch of images and corresponding labels from the input data.
* Input data is labelled from 1 to 26, so here each label is converted to 0 to 26 by subtracting 1.
* So in the output of the neural network, 0th index corresponds to 'A' and 25th index corresponds to 'Z'.
*/
function nextBatch(images, labels, batchSize, index) {
index = index + batchSize;
return [images.slice(index, index + batchSize), labels.slice(index, index + batchSize).map(i => i - 1), index]
}
function getImage ( img ) // make a P5 image object from a raw data array
{
let theimage = createImage (PIXELS, PIXELS); // make blank image, then populate it
theimage.loadPixels();
for (let i = 0; i < PIXELSSQUARED ; 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;
}
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 );
}
/***
* By APARNA
* Randomly select a test data, predict its value using the existing model.
*/
function testit() // test the network with a single exemplar, from global var "test_index"
{
test_index = AB.randomIntAtoB(0, NOTEST - 1);
let img = mnist.test_images[test_index];
let label = mnist.test_labels[test_index];
// Normalize data by dividing it with 255
let testImage = tf.reshape(tf.tensor(img, [1, PIXELSSQUARED]), [1, PIXELS, PIXELS, 1]).div(normalizeFactor);
let prediction = networkModel.predict(testImage).arraySync()[0];
let guess = prediction.indexOf(Math.max(...prediction));
total_tests++;
if (guess + 1 == 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 ( total_tests == NOTEST )
{
console.log( "finished testrun: " + testrun + " score: " + percent.toFixed(2) );
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)
function find12 (a) // return array showing indexes of no.1 and no.2 values in array
{
let no1 = 0;
let no2 = 0;
let no1value = 0;
let no2value = 0;
for (let i = 0; i < a.length; i++)
{
if (a[i] > no1value) // new no1
{
// old no1 becomes no2
no2 = no1;
no2value = no1value;
// now put in the new no1
no1 = i;
no1value = a[i];
}
else if (a[i] > no2value) // new no2
{
no2 = i;
no2value = a[i];
}
}
var b = [ no1, no2 ];
return b;
}
// just get the maximum - separate function for speed - done many times
// find our guess - the max of the output nodes array
function findMax (a)
{
let no1 = 0;
let no1value = 0;
for (let i = 0; i < a.length; i++)
{
if (a[i] > no1value)
{
no1 = i;
no1value = a[i];
}
}
return no1;
}
// --- the draw function -------------------------------------------------------------
// every step:
/***
* By APARNA
* The variable 'trainingInProgress' ensures that training doesn't start
* if last batch's training is still going on. Otherwise, CNN would throw error.
*/
function draw()
{
// check if libraries and data loaded yet:
if ( typeof mnist == 'undefined' ) return;
background ('black');
if ( do_training )
{
if (trainingInProgress == false)
{
trainBatch()
}
// do some testing per step
if(testingInProgress == false)
{
testingInProgress = true;
for (let i = 0; i < TESTPERSTEP; i++)
testit();
testingInProgress = false;
}
}
// 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
{
// are we exiting a drawing
if ( mousedrag )
{
mousedrag = false;
// console.log ("Exiting draw. Now blurring.");
doodle.filter (BLUR, DOODLE_BLUR); // just blur once
// console.log (doodle);
}
}
}
//--- 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: " + String.fromCharCode(96 + 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()
{
// Predict the demo image BY APARNA
let demoImage = tf.reshape(tf.tensor(demo, [1, PIXELSSQUARED]), [1, PIXELS, PIXELS, 1]).div(normalizeFactor);
let prediction = networkModel.predict(demoImage).arraySync()[0];
let guess = prediction.indexOf(Math.max(...prediction));
thehtml = " We classify it as: " + greenspan + String.fromCharCode(97 + 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();
// set up inputs
let inputs = [];
for (let i = 0; i < PIXELSSQUARED ; i++)
{
inputs[i] = img.pixels[i * 4];
}
// Predict the doodle image. BY APARNA
let demoImage = tf.reshape(tf.tensor(inputs, [1, PIXELSSQUARED]), [1, PIXELS, PIXELS, 1]).div(normalizeFactor);
let prediction = networkModel.predict(demoImage).arraySync()[0];
let b = find12(prediction); // get no.1 and no.2 guesses
thehtml = " We classify it as: " + greenspan + String.fromCharCode(97 + b[0]) + "</span> <br>" +
" No.2 guess is: " + greenspan + String.fromCharCode(97 + b[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);
}