// Cloned by Philip on 4 Dec 2020 from World "Character recognition neural network" by "Coding Train" project
// Please leave this clone trail here.
// 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 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 TRAINPERSTEP = 30;
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 = 11; // thickness of doodle lines
const DOODLE_BLUR = 3; // 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
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" });
let htmlPositions = [0, 1, 3, 4, 5, 6, 7, 8, 9, 10, 2, 11, 12, 13]; //Philip : A little bit of an awkward hack but lets easy rearrangement
//--- 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> " +
validateDoodleHtml();
AB.msg(thehtml, htmlPositions[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, htmlPositions[3]);
// 4 variable training data
// 5 Testing header
thehtml = "<h3> Hidden tests </h3> ";
AB.msg(thehtml, htmlPositions[5]);
// 6 variable testing data
// 7 Demo header
thehtml = "<hr> <h1> 3. Demo </h1>The Demos begin to run automatically at 80% accuracy unless turned off! 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, htmlPositions[7]);
thehtml = "<hr> <h1> 4. Test against Stored Doodles </h1> <br>" +
" <button onclick='testSavedDoodles();' class='normbutton' >This became pointless as I began to modify different things such as the filters and how the images were preprocessed</button> <br>";
AB.msg(thehtml, htmlPositions[12]);
// 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() {
startTime = Math.round((new Date().getTime() / 1000));
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/matrix.js", function () {
$.getScript("/uploads/codingtrain/nn.js", function () {
$.getScript("/uploads/codingtrain/mnist.js", function () {
console.log("All JS loaded");
nn = new NeuralNetwork(noinput, nohidden, nooutput);
//Why doesn't this work? Is everything negative on the otherside and being wiped out instantly
let rectifierFunction = new ActivationFunction(
x => x > 0 ? x : 0,
y => y > 0 ? 1 : 0
);
// nn.setActivationFunction(rectifierFunction);
nn.setLearningRate(learningrate);
loadData();
});
});
});
}
// 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
});
}
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] = bright; //Why is this 255? Set it to bright instead
}
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
//Philip : Test 1: trim anything below a certain brightness
//if(bright < 250){
// inputs[i] = 0;
// }
// else {
// inputs[i] = 1;
// }
}
return (inputs);
}
function trainit(show) // train the network with a single exemplar, from global var "train_index", show visual on or off
{
let img = mnist.train_images[train_index];
let label = mnist.train_labels[train_index];
// optional - show visual of the image
if (show) {
var theimage = getImage(img); // get image from data array
// var theimage = getImage(arrayManipulations(img, false, true)); //Philip: This shows what actually gets drawn after the manipulations but as some of these are extremely unoptimised time can be saved by turning it off
image(theimage, 0, ZOOMPIXELS + 50, ZOOMPIXELS, ZOOMPIXELS); // magnified
image(theimage, ZOOMPIXELS + 50, ZOOMPIXELS + 50, PIXELS, PIXELS); // original
}
// set up the inputs
let inputs = getInputs(img); // get inputs from data array
// set up the outputs
let targets = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0];
targets[label] = 1; // change one output location to 1, the rest stay at 0
// console.log(train_index);
// console.log(inputs);
// console.log(targets);
train_inputs = inputs; // can inspect in console
nn.train(arrayManipulations(inputs, false, true), targets);
thehtml = " trainrun: " + trainrun + "<br> no: " + train_index + "<br> Time Passed: " + runTime() + " seconds";
AB.msg(thehtml, htmlPositions[4]);
train_index++;
if (train_index == NOTRAIN) {
train_index = 0;
console.log("finished trainrun: " + trainrun);
trainrun++;
}
}
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 prediction = nn.predict(arrayManipulations(inputs, false, true)); // array of outputs
let guess = findMax(prediction); // the top output
total_tests++;
if (guess == label) total_correct++;
let percent = (total_correct / total_tests) * 100;
currentPercent = percent;
calculateLandmarkTimes(percent);
thehtml = " testrun: " + testrun + "<br> no: " + total_tests + " <br> " +
" correct: " + total_correct + "<br>" +
" score: " + greenspan + percent.toFixed(2) + "</span>" + " <br> " +
" Time to 10%: " + timeToArray[0] + " <br> " +
" Time to 20%: " + timeToArray[1] + " <br> " +
" Time to 30%: " + timeToArray[2] + " <br> " +
" Time to 40%: " + timeToArray[3] + " <br> " +
" Time to 50%: " + timeToArray[4] + " <br> " +
" Time to 60%: " + timeToArray[5] + " <br> " +
" Time to 70%: " + timeToArray[6] + " <br> " +
" Time to 80%: " + timeToArray[7] + " <br> " +
" Time to 85%: " + timeToArray[8] + " <br> " +
" Time to 90%: " + timeToArray[9] + " <br> " +
" Time to 100%: " + timeToArray[10];
AB.msg(thehtml, htmlPositions[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)
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) {
no1 = i;
no1value = a[i];
}
else if (a[i] > no2value) {
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:
function draw() {
reportDoodleAccuracy();
// 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');
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();
}
// keep drawing demo and doodle images
// and keep guessing - we will update our guess as time goes on
//Philip: Going to change it so the demo starts running automatically. Again as another metric of how good/bad any changes made are.
//85% was chosen as the start time for this as after this the improvements slow down significantly
if (demo_exists) {
drawDemo();
guessDemo();
}
if (currentPercent > demoThreshold) {
makeDemoBlind();
}
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) {
rawDoodle = getRawDoodleInputs(doodle); // Philip@ the raw doodle will get saved now instead of processed one so all doodles are shareable across the different filters for quick testing
mousedrag = false;
// console.log ("Exiting draw. Now blurring.");
applyDoodleFilters();
// 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];
demoLabel = mnist.test_labels[i];
thehtml = "Test image no: " + i + "<br>" +
"Classification: " + label + "<br>";
AB.msg(thehtml, htmlPositions[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 prediction = nn.predict(inputs); // array of outputs
let guess = findMax(prediction); // the top output
thehtml = " We classify it as: " + greenspan + guess + "</span>";
AB.msg(thehtml, htmlPositions[9]);
}
//--- doodle -------------------------------------------------------------
function drawDoodle() {
// doodle is createGraphics not createImage
let theimage = doodle.get();
theimage.loadPixels();
// console.log (theimage);
drawFromImage(theimage);
}
function drawFromImage(theimage){
// doodle is createGraphics not createImage
// 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
getDoodleInputs(doodle)
// feed forward to make prediction
let prediction = nn.predict(arrayManipulations(doodle_inputs, true, false)); // array of outputs
let b = find12(prediction); // get no.1 and no.2 guesses
doodleGuess = b[0];
thehtml = " We classify it as: " + greenspan + b[0] + "</span> <br>" +
" No.2 guess is: " + greenspan + b[1] + "</span>";
AB.msg(thehtml, htmlPositions[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);
}
//|********************************************|//
//|****************New stuff*****************|//
//|********************************************|//
//Philip: New Variables
let startTime = 0;
//Some performance metrics, want to figure out how long it took to train to a percent. Will give a quick indicator if training is broken
let timeToArray = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0,0]
let doodleAccuracy = 0;
let numberOfDoodles = 0;
let correctDoodles = 0;
let incorrectDoodles = 0;
let currentPercent = 0;
let doodleGuess = null;
let demoLabel = null;
let numberOfDemos = 0;
let numberOfCorrectDemoGuesses = 0;
let demoAccuracy = 0;
let demoThreshold = 80;
let rawDoodle = [];//Modify this to change the percent that demos kick in at
//Html positions so I can add new stuff more easily without breaking everything
//These need to be changed as the drawing changes as the doodles won't representative
let localStoragePrefix = "doodle_"; //For the base program with no drawing changes
function runTime() {
return Math.round((new Date().getTime() / 1000)) - startTime;
}
function calculateLandmarkTimes(percent) {
updateTimeArray(0, 10, percent);
updateTimeArray(1, 20, percent);
updateTimeArray(2, 30, percent);
updateTimeArray(3, 40, percent);
updateTimeArray(4, 50, percent);
updateTimeArray(5, 60, percent);
updateTimeArray(6, 70, percent);
updateTimeArray(7, 80, percent);
updateTimeArray(8, 85, percent);
updateTimeArray(9, 90, percent);
updateTimeArray(10, 100, percent);
}
function updateTimeArray(i, landmark, percent) {
if (timeToArray[i] == 0 && percent > landmark) {
timeToArray[i] = runTime();
}
}
//Just use some lazy metric gathering, its just to make polling easier and save some effort rather than be accurate
//Rework this a little to capture the doodle before it has filters applied
function validateDoodle(i) {
numberOfDoodles = numberOfDoodles + 1;
if (i == doodleGuess)
correctDoodles = correctDoodles + 1;
else
incorrectDoodles = incorrectDoodles + 1;
getDoodleInputs(doodle)
// addDoodleToHistory( {"classification" : i, "doodle": rawDoodle});
//save the doodle to localStorage The doodle is a bit too big to store at normal size so we'll reduce it down first, but no filters have been applied
//Doesn't really work
// localStorage.setItem(localStoragePrefix + (new Date().getTime()), JSON.stringify({ "classification": i, "doodle": convertRawDoodleToPixelsSquared(rawDoodle) })); //this will be a bit messed up for 2 strokes so try to avoid them
wipeDoodle();
}
function calculateDoodleAccuracy() {
doodleAccuracy = (correctDoodles / numberOfDoodles) * 100;
}
function calculateDemoAccuracy(guess) {
if (numberOfDemos > 0) {
if (demoLabel == guess)
numberOfCorrectDemoGuesses = numberOfCorrectDemoGuesses + 1;
demoAccuracy = (numberOfCorrectDemoGuesses / numberOfDemos) * 100;
}
var html = "<br> Score : " + demoAccuracy + "<br>";
AB.msg(html, htmlPositions[11]);
}
function reportDoodleAccuracy() {
if (numberOfDoodles > 0) {
calculateDoodleAccuracy();
var html = "<br> Score : " + doodleAccuracy + "<br>";
AB.msg(html, htmlPositions[10]);
}
}
function validateDoodleHtml() {
return " Validate: <br> <button onclick='validateDoodle(0);' class='normbutton' >0</button> <br> " +
"<button onclick='validateDoodle(1);' class='normbutton' >1</button> <br> " +
"<button onclick='validateDoodle(2);' class='normbutton' >2</button> <br> " +
"<button onclick='validateDoodle(3);' class='normbutton' >3</button> <br> " +
"<button onclick='validateDoodle(4);' class='normbutton' >4</button> <br> " +
"<button onclick='validateDoodle(5);' class='normbutton' >5</button> <br> " +
"<button onclick='validateDoodle(6);' class='normbutton' >6</button> <br> " +
"<button onclick='validateDoodle(7);' class='normbutton' >7</button> <br> " +
"<button onclick='validateDoodle(8);' class='normbutton' >8</button> <br> " +
"<button onclick='validateDoodle(9);' class='normbutton' >9</button> <br> ";
}
function makeDemoBlind() {
numberOfDemos = numberOfDemos + 1;
var i = AB.randomIntAtoB(0, NOTEST - 1);
demo = mnist.test_images[i];
demoLabel = mnist.test_labels[i];
let inputs = getInputs(demo);
demo_inputs = inputs; // can inspect in console
let prediction = nn.predict(arrayManipulations(inputs, false, true)); // array of outputs
let guess = findMax(prediction); // the top output
calculateDemoAccuracy(guess);
}
function getDoodleInputs(dood) {
let img = dood.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] / 255;
}
doodle_inputs = inputs; // can inspect in console
return inputs;
}
//Im not trusting this anymore, the initial attempt of just an array worked but I need to pass it through filters and I've had no success with that so for now I'm ignoring it and will rely on the counter.
function testSavedDoodles() {
var savedDoodles = 0;
var savedDoodlesCorrectGuesses = 0;
AB.restoreData ( function ( data ) {
for (var i = 0; i < data.length; i++) {
var currentDoodle = data[i];
savedDoodles = savedDoodles + 1;
var doodleAsPixelsSquared = currentDoodle.doodle;;
//Test 1
//Going to remove everything above a certain brightness
var asImage = getDoodleAsImage(doodleAsPixelsSquared);
// drawFromImage(asImage)
applyFilters(asImage);
asImage.resize(PIXELS,PIXELS);
asImage.loadPixels();
var maxMinInput = [];
for (var j = 0; j < asImage.pixels.length; j=j+4) {
maxMinInput.push(asImage.pixels[j]/255)
//if (doodleAsPixelsSquared[j] > 128)
// maxMinInput.push(1);
//else
// maxMinInput.push(0);
}
let prediction = nn.predict(maxMinInput); //Test 1
// let prediction = nn.predict(currentDoodle.doodle); // array of outputs
let b = find12(prediction); // get no.1 and no.2 guesses
if (b[0] == currentDoodle.classification) {
savedDoodlesCorrectGuesses = savedDoodlesCorrectGuesses + 1;
}
}
var html = " <br> Score : " + ((savedDoodlesCorrectGuesses / savedDoodles) * 100).toFixed(2) +
" <br> Saved Doodles : " + savedDoodles
" <br> Correct Guesses : " + savedDoodlesCorrectGuesses;
AB.msg(html, htmlPositions[13]);
});
};
//Doodle is a larger canvas, and their are actually 4 times more pixels than it says so in the pixel array because it has the other channels
function getRawDoodleInputs(dood) {
let img = dood.get();
img.loadPixels();
let inputs = [];
for (let i = 0; i < ZOOMPIXELS * ZOOMPIXELS; i++) {
inputs[i] = img.pixels[i * 4];
}
return inputs;
}
function convertRawDoodleToPixelsSquared(rawInputs) {
// set up inputs
let inputs = [];
for (let i = 0; i < (ZOOMPIXELS * ZOOMPIXELS); i++) {
inputs[i] = rawInputs[i * 4]; // lets not save the normalised data, we cna do that later
}
return inputs;
}
function getDoodleAsImage(doodleArr)
{
let theimage = createImage(ZOOMPIXELS, ZOOMPIXELS); // make blank image, then populate it
theimage.loadPixels();
for(var i = 0; i< doodleArr.length; i++){
theimage.pixels[i] = doodleArr[i];
theimage.pixels[i+1] = doodleArr[i];
theimage.pixels[i+2] = doodleArr[i];
theimage.pixels[i+3] = doodleArr[i];
}
theimage.updatePixels();
return theimage;
}
function addDoodleToHistory(obj){
AB.runloggedin;
let previousData;
AB.queryDataExists (function ( exists ) // asynchronous - need callback function
{
if ( exists ) {
AB.restoreData ( function ( data ) {
data.push(obj);
AB.saveData ( data );
} );
} else {
AB.saveData([obj]);
}
});
}
//Lets make a funciton that will remove any "floating pixels", one where all the neighbours are 0
//I got a 20% improvement for doodle recognition by using this on the trianing images before passing them to the network
function removeRandomPixels(arr){
for(var i=0; i < arr.length; i++){
if(i - 1 > 0)
if(arr[i- 1] != 0) //To left
continue;
if(i + 1 < arr.length) //To right
if(arr[i + 1] != 0)
continue;
if(i + 28 < arr.length) //below
if(arr[i + 28] != 0)
continue;
if(i - 28 > 0) //above
if(arr[i- 28] != 0)
continue;
arr[i] = 0;
}
return arr;
}
function fillHoles(arr){
for(var i=0; i < arr.length; i++){
if(i - 1 > 0)
if(arr[i- 1] != 0) //To left
continue;
if(i + 1 < arr.length) //To right
if(arr[i + 1] != 0)
continue;
if(i + 28 < arr.length) //below
if(arr[i + 28] != 0)
continue;
if(i - 28 > 0) //above
if(arr[i- 28] != 0)
continue;
arr[i] = (arr[i-1] + arr[i+1] + arr[i+28] + arr[i-28])/4;
}
return arr;
}
function thickenLines(arr){
for(var i=0; i < arr.length; i++){
if(!arr[i]) //dont add values around is that have value
continue;
if(i - 1 > 0 && i + 1 < arr.length) //check if its inbounds
if(arr[i - 1] == 0 && arr[i + 1] == 0) //To left
arr[i-1]= arr[i+1] = arr[i];
if(i + 28 < arr.length && i - 28 > 0) //below
if(arr[i + 28] == 0 && arr[i- 28] == 0)
arr[i-28]= arr[i+28] = arr[i];
}
return arr;
}
function shiftToTopLeft(arr){
//If this works, leave it be, if it doesn't map to an array of arrays while moving it, it'll be easier to visual and debug.
//Again, into 28s, looking for the first non-zero value, the row this on then needs to be moved up by 28 until this value is < 28
//Easy, Find first value, subtract 28x where x is the number of 28s into it
var lastValuedIndex = 0;
for(var i = 0; i < arr.length; i++){
if(arr[i] !=0){
firstValuedIndex = i;
break;
}
}
var indexesToShift = Math.floor(firstValuedIndex/28);
if(indexesToShift > 0){
arr = arr.slice(indexesToShift * 28, arr.length);
for(var i = 0; i < indexesToShift * 28; i++) {
arr.push(0);
}
}
//Now shifting to the left
//Actually need to look at every 28th cell and find the first one
var firstLeftestValuedIndex = 0;
var firstValuedVerticalIndex = 999;
for(var i = 0; i < PIXELS; i++){
for(var j = 0; j < PIXELS; j++){
if(arr[i + j * PIXELS] !=0 && i < firstValuedVerticalIndex) {//most leftest spot
firstLeftestValuedIndex = (i + j * PIXELS);
firstValuedVerticalIndex = i;
}
}
}
if(firstLeftestValuedIndex > 0 && firstValuedVerticalIndex < 999){
for(var i = 0; i < PIXELS; i++){
for(var j = 0; j < PIXELS; j++){
if(j + firstValuedVerticalIndex > PIXELS-1) {//End of column, don't start shifting from next one
arr[(i*PIXELS) + j - firstValuedVerticalIndex] = arr[(i*PIXELS) + j];
arr[(i*PIXELS) + j] = 0;
} else
if(j - firstValuedVerticalIndex >= 0)
arr[(i*PIXELS) + j - firstValuedVerticalIndex] = arr[(i*PIXELS) + j];
}
}
}
return arr;
}
function addNoise(arr){
for(var i = 0; i < arr.length; i++){
if(arr[i] ==0){
arr[i] = ( AB.randomPick ( 0, Math.round(Math.random() * 255) ));
}
}
return arr;
}
//This isn't very good, it has a HUGE HUGE HUGE impact on performance, I know this could be done better but don't have time to optimise it and couldn't find something that does it effectively
function rotateImage(arr, angle) {
var rotatedArr = [];
var firstIndex = 999;
for(var i = 0; i < PIXELS; i++){
for (var j = 0; j< PIXELS; j++){
ir = i*cos(angle) - j*sin(angle)
jr = i*sin(angle) + j*cos(angle)
//still want nice round numbers
ir = Math.round(ir) + 100; //arbritrarily making the rotated array bigger to avoid negative indexes
jr = Math.round(jr) + 100;
if(ir < firstIndex)
firstIndex = ir;
if(rotatedArr[ir])
rotatedArr[ir][jr] = arr[(i* PIXELS)+j] + (rotatedArr[ir][jr] ? rotatedArr[ir][jr] : 0);
else {
rotatedArr[ir] = [];
rotatedArr[ir][jr] = arr[i* PIXELS+j] +(0);
}
}
}
//Shift the array back to 0th index
var rotateLength = rotatedArr.length - firstIndex;
for(var i = 0; i < rotateLength; i++){
rotatedArr[i] = rotatedArr[firstIndex + i];
}
rotatedArr = rotatedArr.slice(0, rotateLength);
//Now I need to solve the problem of the shortest line, or rather lef most and right most points
var longestLine = 0;
var shortestLine = 999;
for(var i = 0; i < rotatedArr.length; i++){
if(rotatedArr[i].length > longestLine)
longestLine = rotatedArr[i].length;
}
for(var i = 0; i < rotatedArr.length; i++){
for(var j = longestLine-1; j > 0; j--){
if(rotatedArr[i][j] !== undefined && j < shortestLine)
shortestLine = j;
}
}
//Now I can fill in the empty space with 0s
for(var i = 0; i < rotatedArr.length; i++){
for(var j = shortestLine; j < longestLine; j++){
if(!rotatedArr[i][j])
rotatedArr[i][j] = 0;
}
}
//Lets now shift each line to 0
var width = longestLine -shortestLine;
var widthDiff = width - PIXELS;
for(var i =0; i < rotatedArr.length; i++){
for(var j = 0; j < width; j++){
rotatedArr[i][j] = rotatedArr[i][j+shortestLine];
}
rotatedArr[i] = rotatedArr[i].slice(Math.floor(widthDiff/2), longestLine - Math.floor(widthDiff/2));
if(rotatedArr[i].length > PIXELS){
rotatedArr[i] = rotatedArr[i].slice(0, PIXELS);
}
else if(rotatedArr[i].length < PIXELS){
for(var j = PIXELS-1; j > 0; j++){
if(!rotatedArr[i][j])
rotatedArr[i][j] = 0;
}
}
}
var heightDiff = rotatedArr.length - PIXELS;
rotatedArr = rotatedArr.slice(heightDiff/2, rotatedArr.length - heightDiff/2);
if(rotatedArr.length < PIXELS){
for(var i = 28; i > 0; j++){
if(rotatedArr[i])
break;
var line =[
0,0,0,0,0,0,0,
0,0,0,0,0,0,0,
0,0,0,0,0,0,0,
0,0,0,0,0,0,0,
];
rotatedArr[i] = line;
}
}
//We should now have a matrix thats fully populated that is bigger than 28 * 28 unless the angle was a x* 90
rotatedArr = rotatedArr.flat();
//lets fill in the holes
//simply if the next 3 pixels are blank, I do nothing if not I fill the gap hopefully this won't close intentional holes
for(var i = 0; i< PIXELS; i++){
for(var j=3; j < PIXELS - 3; j++)
{
if(rotatedArr[(i*PIXELS) + j] != 0){
if(
(rotatedArr[(i*PIXELS) + j+1] == 0 && rotatedArr[(i*PIXELS) + j + 2] ==0 && rotatedArr[(i*PIXELS) + j + 3] !=0)
|| (rotatedArr[(i*PIXELS) + j+1] == 0 && rotatedArr[(i*PIXELS) + j + 2] !=0)
){
rotatedArr[(i*PIXELS) + j+1] = rotatedArr[(i*PIXELS) + j + 2] = (rotatedArr[(i*PIXELS) + j + 3] + rotatedArr[(i*PIXELS) +j])/2;
}
if(
(rotatedArr[(i*PIXELS) + j-1] == 0 && rotatedArr[(i*PIXELS) + j - 2] ==0 && rotatedArr[(i*PIXELS) + j - 3] !=0)
|| (rotatedArr[(i*PIXELS) + j-1] == 0 && rotatedArr[(i*PIXELS) + j - 2] !=0)
){
rotatedArr[(i*PIXELS) + j+1] = rotatedArr[(i*PIXELS) + j + 2] = (rotatedArr[(i*PIXELS) + j + 3] + rotatedArr[(i*PIXELS) +j])/2;
}
}
}
}
//I've missed something as a white artificat is appearing down the side, is a single line not getting rotated? Maybe the middle line?
return rotatedArr;
}
function applyDoodleFilters() {
applyFilters(doodle);
}
function applyFilters(inputDoodle){
// doodle.filter(INVERT)
doodle.filter(BLUR, 3); // just blur once
// doodle.filter(INVERT)
//doodle.filter(ERODE); // just blur once
//doodle.filter(THRESHOLD); // just blur once
// doodle.filter(BLUR, 2); // just blur once
// doodle.resize(PIXELS, PIXELS);
//doodle.resize(ZOOMPIXELS, ZOOMPIXELS);
// doodle.filter(POSTERIZE, 6); // just blur once
}
//Anywhere predict is called the array is first passed into this. Modify these methods to effect what happens to all of them
function arrayManipulations(arr, doodle, training){
arr = Array.from(arr)
if(training){
//arr = rotateImage(arr, AB.randomPick(1,-1)*Math.random()/Math.PI); //I've noticed after this the image has loads of holes, so lets do the inverse of removeRandomPixels
}
arr = removeRandomPixels(arr);
arr = shiftToTopLeft(arr);
if(doodle){ //I'll put anything in here that I don't want on a certain set, like for example adding noise to the doodles
// arr = addNoise(arr)
}
if(training){
// thickenLines(arr);
}
return arr;
}