regression.js

/**
 * Code for regression extracted from jqplot.trendline.js
 *
 * Version: 1.0.0a_r701
 *
 * Copyright (c) 2009-2011 Chris Leonello
 * jqPlot is currently available for use in all personal or commercial projects
 * under both the MIT (http://www.opensource.org/licenses/mit-license.php) and GPL
 * version 2.0 (http://www.gnu.org/licenses/gpl-2.0.html) licenses. This means that you can
 * choose the license that best suits your project and use it accordingly.
 *
 **/

function regression(x, y, typ) {
    var type = (typ == null) ? 'linear' : typ;
    var N = x.length;
    var slope;
    var intercept;
    var SX = 0;
    var SY = 0;
    var SXX = 0;
    var SXY = 0;
    var SYY = 0;
    var Y = [];
    var X = [];

    if (type == 'linear') {
        X = x;
        Y = y;
    }
    else if (type == 'exp' || type == 'exponential') {
        for (var i = 0; i < y.length; i++) {
            // ignore points <= 0, log undefined.
            if (y[i] <= 0) {
                N--;
            }
            else {
                X.push(x[i]);
                Y.push(Math.log(y[i]));
            }
        }
    }

    for (var i = 0; i < N; i++) {
        SX = SX + X[i];
        SY = SY + Y[i];
        SXY = SXY + X[i] * Y[i];
        SXX = SXX + X[i] * X[i];
        SYY = SYY + Y[i] * Y[i];
    }

    slope = (N * SXY - SX * SY) / (N * SXX - SX * SX);
    intercept = (SY - slope * SX) / N;

    return [slope, intercept];
}

function linearRegression(X, Y) {
    var ret;
    ret = regression(X, Y, 'linear');
    return [ret[0], ret[1]];
}

function expRegression(X, Y) {
    var ret;
    var x = X;
    var y = Y;
    ret = regression(x, y, 'exp');
    var base = Math.exp(ret[0]);
    var coeff = Math.exp(ret[1]);
    return [base, coeff];
}

/*
    TODO: this function is quite inefficient.
    Refactor it if there is problem with speed.
 */
function fitData(data, typ) {
    var type = (typ == null) ? 'linear' : typ;
    var ret;
    var res;
    var x = [];
    var y = [];
    var ypred = [];

    for (i = 0; i < data.length; i++) {
        if (data[i] != null && Object.prototype.toString.call(data[i]) === '[object Array]') {
            if (data[i] != null && data[i][0] != null && data[i][1] != null) {
                x.push(data[i][0]);
                y.push(data[i][1]);
            }
        }
        else if(data[i] != null && typeof data[i] === 'number' ){//If type of X axis is category
            x.push(i);
            y.push(data[i]);
        }
        else if(data[i] != null && Object.prototype.toString.call(data[i]) === '[object Object]'){
            if (data[i] != null && data[i].x != null && data[i].y != null) {
                x.push(data[i].x);
                y.push(data[i].y);
            }
        }
    }

    if (type == 'linear') {

        ret = linearRegression(x, y);
        for (var i = 0; i < x.length; i++) {
            res = ret[0] * x[i] + ret[1];
            ypred.push([x[i], res]);
        }

        return {
            data: ypred,
            slope: ret[0],
            intercept: ret[1],
            y: function(x) {
                return (this.slope * x) + this.intercept;
            },
            x: function(y) {
                return (y - this.intercept) / this.slope;
            }
        };
    }
    else if (type == 'exp' || type == 'exponential') {

        ret = expRegression(x, y);
        for (var i = 0; i < x.length; i++) {
            res = ret[1] * Math.pow(ret[0], x[i]);
            ypred.push([x[i], res]);
        }
        ypred.sort();

        return {
            data: ypred,
            base: ret[0],
            coeff: ret[1]
        };
    }
}