Creates linear regression data for use in d3.js

leastsquares.js

//Original least squares found here: http://bl.ocks.org/easadler/edae96ae440aa0361a4d#leastsquares.js
//from Adam Sadler!!
//r squared calculation was added by me.

function LeastSquares(values_x, values_y) {
    var sum_x = 0;
    var sum_y = 0;
    var sum_xy = 0;
    var sum_xx = 0;
    var count = 0;

    /*
     * We'll use those variables for faster read/write access.
     */
    var x = 0;
    var y = 0;
    var values_length = values_x.length;

    if (values_length != values_y.length) {
        throw new Error('The parameters values_x and values_y need to have same size!');
    }

    /*
     * Nothing to do.
     */
    if (values_length === 0) {
        return [ [], [] ];
    }

    /*
     * Calculate the sum for each of the parts necessary.
     */
    for (var v = 0; v < values_length; v += 1) {
        x = values_x[v];
        y = values_y[v];
        sum_x += x;
        sum_y += y;
        sum_xx += x*x;
        sum_xy += x*y;
        count += 1;
    }

    /*
     * Calculate m and b for the formular:
     * y = x * m + b
     */
    var m = (count*sum_xy - sum_x*sum_y) / (count*sum_xx - sum_x*sum_x);
    var b = (sum_y/count) - (m*sum_x)/count;

    //calculate r squared
    //Sum of the squared distances between the actual Y values and their mean
    var mean_y = sum_y/count;
    var y_meansquared = 0;
    values_y.forEach(function(d){
      var y_diff = d - mean_y;
      y_meansquared += Math.pow(y_diff, 2);
    });
    // console.log(y_meansquared);

    //sum of squared distances between the actual and the predicted Y values.
    //predicted y minus actual y
    //y = mx + b
    var y_predicted_error = 0;
    for (var i = 0; i < values_length; i += 1){
      var xreal = values_x[i];
      var ytest = (m * xreal) + b;
      var yreal = values_y[i];
      var yerr = ytest - yreal;
      y_predicted_error += Math.pow(yerr, 2);
    }
    // console.log(y_predicted_error);

    //calulate r squared from y_predicted_error and y_meansquared
    var r = 1 - (y_predicted_error/y_meansquared);

    return {'b': b, 'm': m, 'r': r};
}