Math/Statistics: Weighted Least Squares Linear Regression in Php

I am not familiar with Pseudo code but if you can do the least squares part I can suggest an inelligant way to weight your values.

Hi aburr,

Yes, I have code for that. I'll be using the Least Square Fit class. Let me post the code:

<?php
/*Aims
generate equation for straight line
picture of plot


*/
class ymxc{
var $input; //array (x,y)
//converts to selected colom to log
function line(){
//output m & c
/*
retrival variables
expanded array array(x, y, xy, x^2) - $input[][2], $input[][3]
n count of array
constant
gradient
returns constant and gradient
*/
$sumx=0;
$sumy=0;
$sumxy=0;
$sumxx=0;
$n=count($this->input);
$i=0;
foreach($this->input as $in){
$this->input[$i][2]=($in[0]*$in[1]);
$this->input[$i][3]=pow($in[0],2);
//sums
$sumx=$sumx+$in[0];
$sumy=$sumy+$in[1];
$sumxy=$sumxy+$this->input[$i][2];
$sumxx=$sumxx+$this->input[$i][3];
$i++;
}//4each
$xdivN=$sumx/$n;
$ydivN=$sumy/$n;
$gradient=($sumxy-$n*$xdivN*$ydivN)/($sumxx-$n*pow($xdivN,2));
$constant=$ydivN-$gradient*$xdivN;
$output=array($gradient,$constant);
$this->gradient=$gradient;
$this->constant=$constant;
//print_r($this->input);
$this->n=count($this->input);
return $output;
//http://en.wikipedia.org/wiki/Correlation
//
/* OU MS284 unit 7 page 41
ISBN= 7492 2066 2 */
}//line


function residuals(){
/*
expands array to include model array[][4]
*/
$i=0;
foreach($this->input as $ii){
$this->input[$i][4]=$this->gradient*$ii[0]+$this->constant;
$this->input[$i][5]=$this->input[$i][4]-$ii[1];
$i++;
}//foreach
foreach ($this->input as $key => $in)$res[$key]  = $in[5];
return $this->sdv($res);
}//residuals


function sdv($array){
$sumx=0;
$sumxx=0;
$n=count($array);
$i=0;
foreach($array as $in){
$squares=$in*$in;
//sums
$sumx=$sumx+$in;
$sumxx=$sumxx+$squares;
$i++;
}//4each
$xdivN=$sumx/$n;
$xxdivN=$sumxx/$n;
$dsv=pow($xxdivN-$xdivN*$xdivN,.5);
$output=array($xdivN,$dsv);
return $output;
//http://en.wikipedia.org/wiki/Standard_deviation
}

function plot($title,$xtitle,$ytitle,$plotname){
/*
requirs there to be an array of data
Plots picture of array with line through
writes picture to disk plot-datenumber.png
plot needs title xtitle and ytitle

*/
$n=$this->n;
$this->residuals();
$plot = imagecreate(600, 450);
$bg=imagecolorallocate($plot, 239, 239, 239);
imagefilledrectangle ($plot, 50, 50, 550, 400, imagecolorallocate($plot, 255, 255, 255));//plot size 500 x 350
$font = 'Font.TTF';
$texcol = imagecolorallocate($plot, 0, 0, 0);
$last=$this->input[count($this->input)-1];
//separate into colomns find min and max
foreach ($this->input as $key => $in) {
$x[$key]  = $in[0];
$y[$key]  = $in[1];

}
$xx=$x;
$yy=$y;
sort($y, SORT_NUMERIC);
sort($x, SORT_NUMERIC);


if($this->constant<min($y))$miny=$this->constant;
$xrange=$x[count($x)-1]-$x[0];
$yrange=$y[count($x)-1]-$y[0];
//echo "<br/>ranges miny $miny maxy $maxy yrange $yrange" ;

//plot points
//imagettftext($plot, 10, 0, 50, 420, $texcol, $font, round($x[0]));
//imagettftext($plot, 10, 0, 20, 400, $texcol, $font, round($y[0]));
for($i=0;$i<$n;$i++){
$xpoint= (($xx[$i]-$x[0])/$xrange*500)+45;
$ypoint= (400-(($yy[$i]-$y[0])/$yrange*350));
imagettftext($plot, 10, 0, $xpoint, $ypoint, $texcol, $font, "x");
//echo "<br>$i x= $xpoint, y=$ypoint";
}
if(!isset($xtitle))$xtitle="X Values";
imagettftext($plot, 15, 0, 250, 440, $texcol, $font, $xtitle);
if(!isset($ytitle))$ytitle="Y Values";
imagettftext($plot, 15, 90, 20, 240, $texcol, $font, $ytitle);
if(!isset($title))$title="Title";
imagettftext($plot, 20, 0, 100, 30, $texcol, $font, $title);
imagettftext($plot, 10, 0, 450, 40, imagecolorallocate($plot, 0, 0, 255), $font, "Gradient =".round($this->gradient,4));
imagettftext($plot, 10, 0, 450, 20, imagecolorallocate($plot, 0, 0, 255), $font, "Constant =".round($this->constant,4));

//set y markers
for($i=0;$i<6;$i++){
imagettftext($plot, 5, 90, 50, 70*$i+50, $texcol, $font, "|");
imagettftext($plot, 10, 90, 40, 450-(70*$i+50), $texcol, $font, round($i*($y[count($x)-1]-$y[0])/5+$y[0]));
imagettftext($plot, 5, 0, 50+500/5*$i, 403, $texcol, $font, "|");  
imagettftext($plot, 10, 0, 50+500/5*$i, 420, $texcol, $font, round($i*($x[count($x)-1]-$x[0])/5+$x[0]));
}

//plot line    
$s=$x[0]*$this->gradient+$this->constant;
$s=(400-(($s-$y[0])/$yrange*350));

$t=$x[$n-1]*$this->gradient+$this->constant;
$t= 400-($t-$y[0])/$yrange*350;
//imageline ($plot, 50, 400, 550, 50, imagecolorallocate($plot, 255, 0, 0) );
imagettftext($plot, 10, 0, 15, $s, $texcol, $font, round($this->constant,2).",0");
imageline ($plot, 50, $s, 550, $t, imagecolorallocate($plot, 255, 0, 0) );
imageline ($plot, 50, 400-(-$y[0]/$yrange*350), 550, 400-(-$y[0]/$yrange*350), imagecolorallocate($plot, 0, 0, 255) );
$xo=($this->constant-$x[0]);
$xo= (($x0-$x[0])/$xrange*500)+50;
imageline ($plot, $xo, 400, $xo, 50, imagecolorallocate($plot, 0, 255, 0) );
//imagepng($plot, "plot-".strtotime('now').".png");
if(!isset($plotname))$plotname="plot";
imagepng($plot, $plotname.".png");
}//plot




}//class
?>

So that is effectively as good as multiplying the weight with the value of the dependent variable. Is that all that separates Linear Regression from Weighted Linear Regression?

essentially, yes
http://en.wikipedia.org/wiki/Weighted_least_squares

I submitted an answer, the correctness of which was confirmed