hsw0 · May 22, 2017 18:30 · hsw0 · May 22, 2017
diff --git a/Classifier.html b/Classifier.html
 <!DOCTYPE html>
 <html>
 <head>
 <meta charset="utf-8">
 <title>Classfier</title>
 </head>
 <body>

 <style>
 #testbed td { margin: 0; padding: 0; width: 15px; height: 15px; }
 </style>

 <fieldset id="control">
 <legend>학습</legend>
    <label for="epoch">목표 epoch</label>
    <input type="number" id="epoch" value="1000">
    <button type="button">Train!</button>
 </fieldset>

 <fieldset id="testbed">
    <legend>시험</legend>
    <table>
        <tbody>
        </tbody>
    </table>
 </fieldset>

 <div id="result"></div>

 <script src="https://cdnjs.cloudflare.com/ajax/libs/mathjs/3.13.1/math.min.js" integrity="sha256-lhbCjXBlEjR6NpaM7mqLaqXSZTTA6Wc5f9/mPKe4BQ4=" crossorigin="anonymous"></script>
 <script>
 /*****************************************************************************
 * UI
 /*****************************************************************************/


 var W = 5;
 var H = 5;
 (function(){
    document.querySelector('#control button').onclick = train;

    var area = document.querySelector('#testbed table tbody');
    for(var i = 0 ; i < W ; i++)
    {
        var tr = document.createElement('tr');
        for(var j = 0 ; j < H ; j++)
        {
            var td = document.createElement('td');
            var cb = document.createElement('input');
            cb.type = "checkbox";
            cb.id = "x_" + i + "," + j;
            cb.onclick = testClassifier;
            td.append(cb);
            tr.append(td);
        }
        area.append(tr);
    }
 }());

 function train()
 {
    var el = document.getElementById('epoch');
    var targetEpoch = parseInt(el.value);
    const STEP = 100;

    function _train(epoch)
    {
        var to = Math.min(targetEpoch, epoch + STEP);
        for(; epoch < to ; epoch++) {
            classfier.train();
        }

        if (epoch <= targetEpoch) {
            //console.info("Epoch:", epoch);
            el.value = epoch;
            setTimeout(_train.bind(null, epoch), 0);
        }
    }

    setTimeout(_train.bind(null, 0), 0);
 }


 function testClassifier()
 {
    function getInput()
    {
        var X = new Array(W);
        for(var i = 0 ; i < W ; i++)
        {
            X[i] = new Array(H);
            for(var j = 0 ; j < H ; j++)
            {
                var cbid = "x_" + i + "," + j;
                X[i][j] = document.getElementById(cbid).checked ? 1 : 0;
            }
        }
        return X;
    }

    var X = getInput();
    var result = classfier.interference(X).reshape([1,5]).toArray()[0];

    //console.info("Input=", math.matrix(X).toString());
    var greatestIdx = result.indexOf(Math.max.apply(Math, result));

    var msg =
        `Candidate: ${greatestIdx+1}  `
        + `Probability: ${(result[greatestIdx] * 100).toFixed(4)}%\n`
        + "Raw data: " + result;

    document.getElementById('result').innerText = msg;
    console.info(msg);
 }

 /*****************************************************************************/

 var X = (function() {
    var x_1 = [
        [0,1,1,0,0],
        [0,0,1,0,0],
        [0,0,1,0,0],
        [0,0,1,0,0],
        [0,1,1,1,0],
    ];

    var x_2 = [
        [1,1,1,1,0],
        [0,0,0,0,1],
        [0,1,1,1,0],
        [1,0,0,0,0],
        [1,1,1,1,1],
    ];

    var x_3 = [
        [1,1,1,1,0],
        [0,0,0,0,1],
        [0,1,1,1,0],
        [0,0,0,0,1],
        [1,1,1,1,0],
    ];

    var x_4 = [
        [0,0,0,1,0],
        [0,0,1,1,0],
        [0,1,0,1,0],
        [1,1,1,1,1],
        [0,0,0,1,0],
    ];

    var x_5 = [
        [1,1,1,1,1],
        [1,0,0,0,0],
        [1,1,1,1,0],
        [0,0,0,0,1],
        [1,1,1,1,0],
    ];
    return [ x_1, x_2, x_3, x_4, x_5 ];
 }());

 class Classfier
 {
    constructor(X)
    {
        this.DIGITS = X.length;
        this.X = X.map(x => math.matrix(x));
        this.D = [
            [1,0,0,0,0],
            [0,1,0,0,0],
            [0,0,1,0,0],
            [0,0,0,1,0],
            [0,0,0,0,1],
        ];

        // W1=2*rand(50,25)-1
        this.W1 = math.subtract(math.multiply(2, math.matrix(math.random([50,25]))), 1);
        // W2=2*rand(5,50)-1
        this.W2 = math.subtract(math.multiply(2, math.matrix(math.random([this.DIGITS,50]))), 1);
    }

    train()
    {
        const alpha = 0.9;

        var W1 = this.W1.clone();
        var W2 = this.W2.clone();

        for(var k = 0 ; k < this.DIGITS ; k++)
        {
            var x = math.reshape(this.X[k].clone(), [25, 1]);
            var d = math.reshape(this.D[k], [this.DIGITS,1]); // D(k, :)';

            var v1 = math.multiply(W1, x);
            var y1 = Sigmoid(v1);
            var v = math.multiply(W2, y1);
            var y = Softmax(v);

            var e = math.subtract(d, y);
            var delta = e;

            var e1 = math.multiply(math.transpose(W2), delta);
            var delta1 = math.dotMultiply(math.dotMultiply(y1, math.subtract(1, y1)), e1);

            var dW1 = math.multiply(alpha, delta1,  math.transpose(x));
            W1 = math.add(W1, dW1);

            var dW2 = math.multiply(alpha, delta, math.transpose(y1));
            W2 = math.add(W2, dW2);
        }

        [this.W1, this.W2] = [W1, W2];
    }

    interference(x)
    {
        var x = math.reshape(math.matrix(x), [25, 1]);

        var v1 = math.multiply(this.W1, x);
        var y1 = Sigmoid(v1);
        var v = math.multiply(this.W2, y1);
        var y = Softmax(v);

        return y;
    }
 }

 var classfier = new Classfier(X);


 function Softmax(x)
 {
    /*
        ex = exp(x);
        y = ex / sum(ex);
     */
    var ex = math.exp(x);
    var y = math.divide(ex, math.sum(ex));
    return y;
 }

 function Sigmoid(x)
 {
    // p91
    // y= 1 ./ (1 + exp(-x));
    var y = math.dotDivide(1, math.add(1, math.exp(math.multiply(-1, x))));
    return y;
 }

 </script>
 </body>
 </html>
	<!DOCTYPE html>
	<html>
	<head>
	<meta charset="utf-8">
	<title>Classfier</title>
	</head>
	<body>

	<style>
	#testbed td { margin: 0; padding: 0; width: 15px; height: 15px; }
	</style>

	<fieldset id="control">
	<legend>학습</legend>
	<label for="epoch">목표 epoch</label>
	<input type="number" id="epoch" value="1000">
	<button type="button">Train!</button>
	</fieldset>

	<fieldset id="testbed">
	<legend>시험</legend>
	<table>
	<tbody>
	</tbody>
	</table>
	</fieldset>

	<div id="result"></div>

	<script src="https://cdnjs.cloudflare.com/ajax/libs/mathjs/3.13.1/math.min.js" integrity="sha256-lhbCjXBlEjR6NpaM7mqLaqXSZTTA6Wc5f9/mPKe4BQ4=" crossorigin="anonymous"></script>
	<script>
	/*****************************************************************************
	* UI
	/*****************************************************************************/


	var W = 5;
	var H = 5;
	(function(){
	document.querySelector('#control button').onclick = train;

	var area = document.querySelector('#testbed table tbody');
	for(var i = 0 ; i < W ; i++)
	{
	var tr = document.createElement('tr');
	for(var j = 0 ; j < H ; j++)
	{
	var td = document.createElement('td');
	var cb = document.createElement('input');
	cb.type = "checkbox";
	cb.id = "x_" + i + "," + j;
	cb.onclick = testClassifier;
	td.append(cb);
	tr.append(td);
	}
	area.append(tr);
	}
	}());

	function train()
	{
	var el = document.getElementById('epoch');
	var targetEpoch = parseInt(el.value);
	const STEP = 100;

	function _train(epoch)
	{
	var to = Math.min(targetEpoch, epoch + STEP);
	for(; epoch < to ; epoch++) {
	classfier.train();
	}

	if (epoch <= targetEpoch) {
	//console.info("Epoch:", epoch);
	el.value = epoch;
	setTimeout(_train.bind(null, epoch), 0);
	}
	}

	setTimeout(_train.bind(null, 0), 0);
	}


	function testClassifier()
	{
	function getInput()
	{
	var X = new Array(W);
	for(var i = 0 ; i < W ; i++)
	{
	X[i] = new Array(H);
	for(var j = 0 ; j < H ; j++)
	{
	var cbid = "x_" + i + "," + j;
	X[i][j] = document.getElementById(cbid).checked ? 1 : 0;
	}
	}
	return X;
	}

	var X = getInput();
	var result = classfier.interference(X).reshape([1,5]).toArray()[0];

	//console.info("Input=", math.matrix(X).toString());
	var greatestIdx = result.indexOf(Math.max.apply(Math, result));

	var msg =
	`Candidate: ${greatestIdx+1} `
	+ `Probability: ${(result[greatestIdx] * 100).toFixed(4)}%\n`
	+ "Raw data: " + result;

	document.getElementById('result').innerText = msg;
	console.info(msg);
	}

	/*****************************************************************************/

	var X = (function() {
	var x_1 = [
	[0,1,1,0,0],
	[0,0,1,0,0],
	[0,0,1,0,0],
	[0,0,1,0,0],
	[0,1,1,1,0],
	];

	var x_2 = [
	[1,1,1,1,0],
	[0,0,0,0,1],
	[0,1,1,1,0],
	[1,0,0,0,0],
	[1,1,1,1,1],
	];

	var x_3 = [
	[1,1,1,1,0],
	[0,0,0,0,1],
	[0,1,1,1,0],
	[0,0,0,0,1],
	[1,1,1,1,0],
	];

	var x_4 = [
	[0,0,0,1,0],
	[0,0,1,1,0],
	[0,1,0,1,0],
	[1,1,1,1,1],
	[0,0,0,1,0],
	];

	var x_5 = [
	[1,1,1,1,1],
	[1,0,0,0,0],
	[1,1,1,1,0],
	[0,0,0,0,1],
	[1,1,1,1,0],
	];
	return [ x_1, x_2, x_3, x_4, x_5 ];
	}());

	class Classfier
	{
	constructor(X)
	{
	this.DIGITS = X.length;
	this.X = X.map(x => math.matrix(x));
	this.D = [
	[1,0,0,0,0],
	[0,1,0,0,0],
	[0,0,1,0,0],
	[0,0,0,1,0],
	[0,0,0,0,1],
	];

	// W1=2*rand(50,25)-1
	this.W1 = math.subtract(math.multiply(2, math.matrix(math.random([50,25]))), 1);
	// W2=2*rand(5,50)-1
	this.W2 = math.subtract(math.multiply(2, math.matrix(math.random([this.DIGITS,50]))), 1);
	}

	train()
	{
	const alpha = 0.9;

	var W1 = this.W1.clone();
	var W2 = this.W2.clone();

	for(var k = 0 ; k < this.DIGITS ; k++)
	{
	var x = math.reshape(this.X[k].clone(), [25, 1]);
	var d = math.reshape(this.D[k], [this.DIGITS,1]); // D(k, :)';

	var v1 = math.multiply(W1, x);
	var y1 = Sigmoid(v1);
	var v = math.multiply(W2, y1);
	var y = Softmax(v);

	var e = math.subtract(d, y);
	var delta = e;

	var e1 = math.multiply(math.transpose(W2), delta);
	var delta1 = math.dotMultiply(math.dotMultiply(y1, math.subtract(1, y1)), e1);

	var dW1 = math.multiply(alpha, delta1, math.transpose(x));
	W1 = math.add(W1, dW1);

	var dW2 = math.multiply(alpha, delta, math.transpose(y1));
	W2 = math.add(W2, dW2);
	}

	[this.W1, this.W2] = [W1, W2];
	}

	interference(x)
	{
	var x = math.reshape(math.matrix(x), [25, 1]);

	var v1 = math.multiply(this.W1, x);
	var y1 = Sigmoid(v1);
	var v = math.multiply(this.W2, y1);
	var y = Softmax(v);

	return y;
	}
	}

	var classfier = new Classfier(X);


	function Softmax(x)
	{
	/*
	ex = exp(x);
	y = ex / sum(ex);
	*/
	var ex = math.exp(x);
	var y = math.divide(ex, math.sum(ex));
	return y;
	}

	function Sigmoid(x)
	{
	// p91
	// y= 1 ./ (1 + exp(-x));
	var y = math.dotDivide(1, math.add(1, math.exp(math.multiply(-1, x))));
	return y;
	}

	</script>
	</body>
	</html>