PHPでNaive Bayesを使ってみる

<?php
/**
 * Naivebayes.php 
 *
 * This package was ported from Perl's Algorithm::NaiveBayes (frequency model only)
 * http://search.cpan.org/~kwilliams/Algorithm-NaiveBayes-0.04/lib/Algorithm/NaiveBayes.pm
 * 
 * @category  Algorithm 
 * @package   Naivebayes
 * @author    hideack
 * @license   http://www.php.net/license/3_01.txt The PHP License, version 3.01
 * @version   0.1 
 */
class Naivebayes{
private $modeltype;
private $instances;
private $trainingdata;
private $model;
public function __construct(){
$this->trainingdata = array(
"attributes" => array(),
"labels"     => array(),
);
$this->instances = 0;
$this->modeltype = "";	// Perl版では切り替え可能
}
public function addInstance($attributes, $label){
$this->instances++;
foreach($attributes as $keyword => $count){
if(isset($this->trainingdata['attributes'][$keyword])){
$this->trainingdata['attributes'][$keyword] += $count;
}
else{
$this->trainingdata['attributes'][$keyword] = $count;
}
}
foreach($label as $labelword){
if(isset($this->trainingdata['labels'][$labelword]['count'])){
$this->trainingdata['labels'][$labelword]['count']++;
}
else{
$this->trainingdata['labels'][$labelword]['count'] = 1;
}
foreach($attributes as $keyword => $count){
if(isset($this->trainingdata[$keyword])){
$this->trainingdata['labels'][$labelword]['attributes'][$keyword] += $count;
}
else{
$this->trainingdata['labels'][$labelword]['attributes'][$keyword] = $count;
}
}
}
}
public function train(){
$m = array();
$labels = $this->trainingdata['labels'];
$m['attributes'] = $this->trainingdata['attributes'];
$vocab_size = count($m['attributes']);
foreach($labels as $label => $info){
$m['prior_probs'][$label] = log($info['count'] / $this->instances);
$label_tokens = 0;
foreach($info['attributes'] as $word => $count){
$label_tokens += $count;
}
$m['smoother'][$label] = -log($label_tokens + $vocab_size);
$denominator = log($label_tokens + $vocab_size);
foreach($info['attributes'] as $attribute => $count){
$m['probs'][$label][$attribute] = log($count + 1) - $denominator;
}
}
$this->model = $m;
}
public function predict($newattrs){
$scores = $this->model['prior_probs'];
foreach($newattrs as $feature => $value){
foreach($this->model['probs'] as $label => $attribute){
$tmpscore = 0.0;
if($attribute[$feature] == 0.0){
$tmpscore = $this->model['smoother'][$label];
}
else{
$tmpscore = $attribute[$feature];
}
$scores[$label] += $tmpscore * $value;
}
}
$scores = $this->rescale($scores);
return $scores;
}
public function labels(){
$labels = array();
foreach($this->trainingdata['labels'] as $label => $value){
$labels[] = $label;
}
return $labels;
}
public function doPurge(){
// 未実装...
}
private function rescale($scores){
$total = 0;
$max  = max($scores);
$rescalescore = $scores;
foreach($rescalescore as $key => $val){
$val = exp($val - $max);
$total += pow($val, 2);
$rescalescore[$key] = $val;
}
$total = sqrt($total);
foreach($rescalescore as $key => $val){
$rescalescore[$key] /= $total;
}
return $rescalescore;
}
}
?>