//It uses Artificial Neural Network with 784 input nodes, 1 hidden layer with 150 nodes and output layer with 10 nodes window.onload =()=>{ var nn; var model; var dataset; var arr =[]; var scale = 5; /* var labelMap = [ [0.99,0.01,0.01,0.01,0.01,0.01,0.01,0.01,0.01,0.01], [0.01,0.99,0.01,0.01,0.01,0.01,0.01,0.01,0.01,0.01], [0.01,0.01,0.99,0.01,0.01,0.01,0.01,0.01,0.01,0.01], [0.01,0.01,0.01,0.99,0.01,0.01,0.01,0.01,0.01,0.01], [0.01,0.01,0.01,0.01,0.99,0.01,0.01,0.01,0.01,0.01], [0.01,0.01,0.01,0.01,0.01,0.99,0.01,0.01,0.01,0.01], [0.01,0.01,0.01,0.01,0.01,0.01,0.99,0.01,0.01,0.01], [0.01,0.01,0.01,0.01,0.01,0.01,0.01,0.99,0.01,0.01], [0.01,0.01,0.01,0.01,0.01,0.01,0.01,0.01,0.99,0.01], [0.01,0.01,0.01,0.01,0.01,0.01,0.01,0.01,0.01,0.99] ];*/ function init(){ loadAndGuess(); document.querySelectorAll("div.loader").forEach((v)=>{v.style.display = "none"}); console.log("Click on screen to get a random Image/datapoint"); } let noOfFiles = 2; let fileloaded = 0; loadFile("https://dl.dropbox.com/s/5meswm173ciborh/mini-dataset.txt", (data)=>{ arr = data.split("\n"); fileloaded++; if(fileloaded==noOfFiles) init(); }) loadFile("https://dl.dropbox.com/s/gyxownwf6clrp50/mini-model.txt", (data)=>{ nn = NN.loadModel(data); fileloaded++ if(fileloaded==noOfFiles) init(); }) //Alert! code quality is very bad. Read at your own risk let cnv = document.getElementById("cnv"); let ctx = cnv.getContext("2d"); cnv.width = 28*scale; cnv.height = 28*scale; ctx.scale(scale,scale) let outEl = document.querySelector('#out'); let guessEl = document.querySelector('#guess'); let rand = Math.floor(Math.random()*350); function loadAndGuess(){ let r = rand//Math.floor(Math.random()*350); if(typeof arr[r]=='string') arr[r] = new Float32Array(arr[r].split(',').map((v)=>v/255.0)); let pred = nn.predict(arr[r].slice(1)); guessEl.innerText = 'Guess, it is : '+ pred.matrix.findIndex((v)=>v==Math.max(...pred.matrix)); out.innerText= ''; for(let i=0; i<10; i++){ out.innerHTML += '

' + i + ' : '+(pred.matrix[i]*100).toString().slice(0,5)+' %'+'

'; } for(let i =0;i<28;i++){ for(let j = 0; j<28;j++){ let pos = i+28*j+1; ctx.fillStyle = `rgb(${arr[r][pos]*255},${arr[r][pos]*255},${arr[r][pos]*244})`; ctx.fillRect(i,j,i+1,j+1); } } /* for(let i = 0;i<0;i++){ let r = Math.floor(Math.random()*350) if(typeof arr[r]=='string') arr[r] = new Float32Array(arr[r].split(',').map((v)=>v/255.0)); //nn.fit(arr[r].slice(1),label[parseInt(arr[r][0]*255)]); }*/ // requestAnimationFrame(fit) } window. onclick =()=>{ rand = Math.floor(Math.random()*350); loadAndGuess(); //download(JSON.stringify(nn),'test'+Math.random(),"text/plain"); }; } /* function prepData(data){ arr = data.split('\n'); //for(let i=0;i<50;i++){ // arr[i] = arr[i].split(',') // arr[i] = new Float32Array(arr[i].map((v)=>v/255)); // } //console.log(arr[0]) }*/ function loadFile(link, callback){ fetch(link) .then(response => response.text()) .then((data) => { callback(data); })} /* function download(text, name, type) { var a = document.getElementById("saveModel"); var file = new Blob([text], {type: type}); a.href = URL.createObjectURL(file); a.download = name; }*/ const Sigmoid = function(x){ return 1.0/(1 + Math.exp(-1*x)); } //derivative of Sigmoid(x) = Sigmoid(x)*(1- Sigmoid(x)) /* // not ready for any other activations const Relu = function(x, a=1){ return a*x*(x>0); } const Linear = function(x, a=0.1){ return a*x } const HyperbolicTangent = function(x, a=0.1){ return a*(Math.exp(x)-Math.exp(-x))/(Math.exp(x)+Math.exp(-x)) }*/ class Matrix{ constructor(rows = 1, columns=1, initVal=0){ this.rows = rows; this.cols = columns; this.matrix = new Array(rows) .fill(initVal).map(()=> new Array(columns).fill(initVal)) } randomize(a=0, b=1){ for(let i=0; i1.0-x))), Matrix.transpose(this.io[2*(i-1)+1])), this.lr)); } this.w[0] = Matrix.add(this.w[0], Matrix.mult(Matrix.matmul(Matrix.hadamard(this.err[this.w.length-1], Matrix.hadamard(this.io[1], Matrix.mapFunc(this.io[1], (x)=>1.0-x))), Matrix.transpose(inp)), this.lr)); for(let i=1;i1.0-x))), this.lr)); } this.bias[0] = Matrix.add(this.bias[0], Matrix.mult(Matrix.hadamard(this.err[this.bias.length-1], Matrix.hadamard(this.io[1], Matrix.mapFunc(this.io[1], (x)=>1.0-x))), this.lr)); } static loadModel(model){ if(typeof(model)=='string') model = JSON. parse(model); else console.error("Can't load model, data must be a String"); let nn = new NN(model.in,[model.hn, new Array(model.hl).fill(Sigmoid)],[model.on, Sigmoid], model.lr); for(let i = 0;i

Handwritten digit recognigition || ML