Consistency improvements in internal feature vector

representation. An array with integer keys will be passed
through array_count_values to be made a frequency vector.

Bonus: By doing that in multinomial_nb_classifier we have shaved
off a few function calls and additions

Author: angeloskath

classifier/multinomial_nb_classifier.php

@@ -60,9 +60,11 @@ public function classify(array $classes, Document $d) {
 	public function getScore($class, Document $d) {
 		$score = log($this->model->getPrior($class));
 		$features = $this->feature_factory->getFeatureArray($class,$d);
-		foreach ($features as $f)
+		if (is_int(key($features)))
+			$features = array_count_values($features);
+		foreach ($features as $f=>$fcnt)
 		{
-			$score += log($this->model->getCondProb($f,$class));
+			$score += $fcnt*log($this->model->getCondProb($f,$class));
 		}
 		return $score;
 	}

clustering/euclidean_centroid.php

@@ -18,7 +18,7 @@ class Euclidean implements CentroidFactory
 	 * @return array A sparse vector representing the document to the n-dimensional euclidean space
 	 */
 	protected function getVector(array $doc) {
-		if (key($doc)===0)
+		if (is_int(key($doc)))
 			return array_count_values($doc);
 		else
 			return $doc;

feature_factories/callables_as_features.php

@@ -9,13 +9,33 @@
  * (function names, closures, array($object,'func_name'), etc.) and
  * calls them consecutively using the return value as a feature's unique
  * string.
+ *
+ * The class can model both feature frequency and presence
  */
 class FunctionFeatures implements FeatureFactory
 {
-	
+
 	protected $functions;
+	protected $frequency;
+
+	/**
+	 * @param array $f An array of feature functions
+	 */
 	public function __construct(array $f=array()) {
 		$this->functions=$f;
+		$this->frequency=false;
+	}
+	/**
+	 * Set the feature factory to model frequency instead of presence
+	 */
+	public function modelFrequency() {
+		$this->frequency = true;
+	}
+	/**
+	 * Set the feature factory to model presence instead of frequency
+	 */
+	public function modelPresence() {
+		$this->frequency = false;
 	}
 	/**
 	 * Add a function as a feature
@@ -52,15 +72,22 @@ public function getFeatureArray($class, Document $d) {
 			{
 				foreach ($f as $ff)
 				{
-					$set[$ff] = 1;
+					if (!isset($set[$ff]))
+						$set[$ff] = 0;
+					$set[$ff]++;
 				}
 			}
 			else
 			{
-				$set[$f] = 1;
+				if (!isset($set[$f]))
+					$set[$f] = 0;
+				$set[$f]++;
 			}
 		}
-		return array_keys($set);
+		if ($this->frequency)
+			return $set;
+		else
+			return array_keys($set);
 	}
 
 }

models/feature_based_nb.php

@@ -139,16 +139,18 @@ protected function countTrainingSet(FeatureFactory $ff, TrainingSet $tset, array
 			$c = $tdoc->getClass();
 			$ndocs_per_class[$c]++;
 			$features = $ff->getFeatureArray($c,$tdoc);
-			foreach ($features as $f)
+			if (is_int(key($features)))
+				$features = array_count_values($features);
+			foreach ($features as $f=>$fcnt)
 			{
 				if (!isset($voc[$f]))
 					$voc[$f] = 0;
 
-				$termcount_per_class[$c]++;
+				$termcount_per_class[$c]+=$fcnt;
 				if (isset($termcount[$c][$f]))
-					$termcount[$c][$f]++;
+					$termcount[$c][$f]+=$fcnt;
 				else
-					$termcount[$c][$f] = 1;
+					$termcount[$c][$f] = $fcnt;
 			}
 		}
 	}

similarity/cosine_similarity.php

@@ -45,11 +45,11 @@ class CosineSimilarity implements Similarity, Distance
 	public function similarity(&$A, &$B) {
 		// This means they are simple text vectors
 		// so we need to count to make them vectors
-		if (key($A)===0)
+		if (is_int(key($A)))
 			$v1 = array_count_values($A);
 		else
 			$v1 = &$A;
-		if (key($B)===0)
+		if (is_int(key($B)))
 			$v2 = array_count_values($B);
 		else
 			$v2 = &$B;

similarity/euclidean.php

@@ -15,11 +15,11 @@ class Euclidean implements Distance
 	 * @return float The euclidean distance between $A and $B
 	 */
 	public function dist(&$A, &$B) {
-		if (key($A)===0)
+		if (is_int(key($A)))
 			$v1 = array_count_values($A);
 		else
 			$v1 = &$A;
-		if (key($B)===0)
+		if (is_int(key($B)))
 			$v2 = array_count_values($B);
 		else
 			$v2 = &$B;