/**
* Merges vertical and horizontal coordinates in a bidimensional point array.
* Stops coordinates (hesitations) are removed.
* @param array $xcoords horizontal coordinates
* @param array $ycoords vertical coordinates
* @param array $getDistances if TRUE, the result array contains euclidean distances
* @return array 2D points or euclidean distances array
*/
function convert_points($xcoords, $ycoords, $getDistances = false)
{
// initialize points array
$pointArray = array();
// check for illegal offsets on $coords
$maxCount = count($xcoords) - 1;
// transform arrays in a single points array
foreach ($xcoords as $i => $value) {
$p = array($value, $ycoords[$i]);
// check if next point exists
if ($i >= $maxCount) {
break;
}
$q = array($xcoords[$i + 1], $ycoords[$i + 1]);
$distance = KMeans::getDistance($p, $q);
// check
if ($getDistances) {
$pointArray[] = $distance;
} else {
// append point to the points array, discarding null distances
if ($distance > 0) {
$pointArray[] = $p;
}
}
}
return $pointArray;
}
// check whether URLs should be merged (just remove query string)
if (db_option(TBL_PREFIX . TBL_CMS, "mergeCacheUrl")) {
$urlparts = explode("?", $url);
$url = $urlparts[0];
}
if (isset($pages[$url])) {
$pages[$url] += 1;
} else {
$pages[$url] = 1;
}
$dict_user[] = $row['id'];
$dict_page[] = $url;
}
$n = count($pages);
$k = ceil(sqrt($n / 2));
$km = new KMeans(whiten($users), $k);
//$km->initRandom();
$km->initKatsavounidis();
$km->doCluster();
$groups = array();
foreach ($km->clusters as $j => $cluster) {
//echo '['.$j.']<br>';
$groups[$j] = array();
foreach ($cluster as $id => $feats) {
//echo $id.': '.$dict_page[$id].', '.$dict_user[$id].'<br>';
$url = $dict_page[$id];
if (!isset($groups[$j][$url])) {
$groups[$j][$url] = 1;
} else {
$groups[$j][$url] += 1;
}
$xcoords = explode(",", $coordsX);
$ycoords = explode(",", $coordsY);
//echo count($xcoords)." points (duplicated ones will be removed)\n";
// transform arrays in a single points array
$pointArray = convert_points($xcoords, $ycoords);
/* We can do as many clusters as (to the extreme) the sample points size,
* but use the rule of thumb: k ~ sqrt(n/2)
*/
$n = count($pointArray);
$k = ceil(sqrt($n / 2));
echo $n . " points and " . $k . " clusters.\n";
$prevVar = 0;
// compute K-means
for ($i = 1; $i <= $n; ++$i) {
ini_set('max_execution_time', 30);
$km = new KMeans($pointArray, $k);
$km->initKatsavounidis();
$km->maxIterations = 5;
$c = $km->distributeOverClusters();
$clusterVarX = array();
$clusterVarY = array();
// store points
foreach ($c as $cluster) {
$clusterVarX[] = $cluster->variance->x;
$clusterVarY[] = $cluster->variance->y;
}
//var_dump($clusterVarX, $clusterVarY);
$jointVar = array_sum($clusterVarX) * array_sum($clusterVarY);
//$jointVar = (array_sum($clusterVarX)/count($clusterVarX) + array_sum($clusterVarY)/count($clusterVarY) ) / 2;
if ($prevVar == 0) {
$prevVar = $jointVar;
