Sample program of the Widrow-Hoff learning

widrowhoff.js

/*
Sample program of the Widrow-Hoff learning

Copyright (c) 2014 danglingfarpointer

This program is distributed under the MIT License:
http://opensource.org/licenses/mit-license.php
*/

var signs = [
    1, 1, 1, 1, 1, -1, -1, -1, -1, -1,
];
var samples = [
//  [c, x, y]; c is always set to 1
    [1, -2, 1], 
    [1, -1, 4],
    [1, 1, 3], 
    [1, 3, 3],
    [1, -1, 3],
    [1, -3, -1],
    [1, 0, -1],
    [1, 2, 1],
    [1, 3, 2],
    [1, 0, 3],
];
var N = samples.length; // the number of samples
var D = samples[0].length; // dimension

// console.log(signs);
// console.log(samples);

// initialize
var ws = []; 
for (var d = 0; d < D; ++d)
    ws[d] = Math.random();

while(true) {
    var js = []; // gradient of LMS
    for (var d = 0; d < D; ++d)	js[d] = 0;

    for (var n = 0; n < N; ++n) {
	var p = 0;
	for (var d = 0; d < D; ++d)
	    p += samples[n][d] * ws[d];
	
	for (var d = 0; d < D; ++d)
	    js[d] += (p - signs[n]) * samples[n][d];
    }

    // if js is small enough, exit the loop
    if (norm2(js) < 0.001) break;

    console.log("ws[0] = " + ws[0]);
    console.log("ws[1] = " + ws[1]);
    console.log("ws[2] = " + ws[2]);
    console.log("js[0] = " + js[0]);
    console.log("js[1] = " + js[1]);
    console.log("js[2] = " + js[2]);
    console.log("");

    // update the weight vector
    for (var d = 0; d < D; ++d)
	ws[d] -= 0.01 * js[d];
}

console.log((- ws[1] / ws[2]) + " * x + " + (- ws[0] / ws[2]));

function norm2(vs) {
    var c = 0;
    for (var d = 0; d < D; ++d)	c += vs[d] * vs[d];
    return c;
}