Update readme and changelog for 0.0.10-beta release

Author: andrewdalpino

.gitattributes

@@ -1,6 +1,3 @@
 * text=auto
 
-*.php text eol=lf
-
-*.png binary
-*.jpg binary
+*.php text eol=lf

CHANGELOG.md

@@ -1,4 +1,4 @@
-- Unreleased
+- 0.0.10-beta
     - Added LODA online anomaly detector
     - Added Radius Neighbors classifier and regressor
     - Added fast k-d LOF anomaly detector

README.md

@@ -364,15 +364,15 @@ protected function calculateMembership(array $sample) : array
         foreach ($this->centroids as $cluster => $centroid1) {
             $a = $this->kernel->compute($sample, $centroid1);
 
-            $total = 0.;
+            $sigma = 0.;
 
             foreach ($this->centroids as $centroid2) {
                 $b = $this->kernel->compute($sample, $centroid2);
 
-                $total += ($a / ($b ?: self::EPSILON)) ** $this->lambda;
+                $sigma += ($a / ($b ?: self::EPSILON)) ** $this->lambda;
             }
 
-            $membership[$cluster] = 1. / ($total ?: self::EPSILON);
+            $membership[$cluster] = 1. / ($sigma ?: self::EPSILON);
         }
 
         return $membership;
@@ -387,18 +387,17 @@ protected function calculateMembership(array $sample) : array
      */
     protected function interClusterDistance(Dataset $dataset, array $memberships) : float
     {
-        $total = 0.;
+        $distance = 0.;
 
         foreach ($dataset as $i => $sample) {
             $membership = $memberships[$i];
 
             foreach ($this->centroids as $cluster => $centroid) {
-                $distance = $this->kernel->compute($sample, $centroid);
-
-                $total += $membership[$cluster] * $distance;
+                $distance += $membership[$cluster]
+                    * $this->kernel->compute($sample, $centroid);
             }
         }
 
-        return $total;
+        return $distance;
     }
 }

docs/images/rubix-ml-system-architecture.svg

@@ -235,13 +235,10 @@ public function train(Dataset $dataset) : void
 
         $n = $dataset->numRows();
 
-        [$means, $variances] = $this->initialize($dataset);
-
-        $this->means = $means;
-        $this->variances = $variances;
-
         $this->priors = array_fill(0, $this->k, log(1. / $this->k));
 
+        [$this->means, $this->variances] = $this->initialize($dataset);
+
         $this->steps = [];
 
         $prevLoss = 0.;

src/Clusterers/FuzzyCMeans.php

@@ -194,12 +194,12 @@ public function train(Dataset $dataset) : void
                 ]));
         }
 
-        $this->centroids = $previous = $dataset->samples();
+        $centroids = $previous = $dataset->samples();
 
         $this->steps = [];
 
         for ($epoch = 1; $epoch <= $this->epochs; $epoch++) {
-            foreach ($this->centroids as $i => &$centroid) {
+            foreach ($centroids as $i => &$centroid) {
                 foreach ($dataset as $sample) {
                     $distance = $this->kernel->compute($sample, $centroid);
 
@@ -214,15 +214,15 @@ public function train(Dataset $dataset) : void
                     }
                 }
 
-                foreach ($this->centroids as $j => $neighbor) {
+                foreach ($centroids as $j => $neighbor) {
                     if ($i === $j) {
                         continue 1;
                     }
 
                     $distance = $this->kernel->compute($centroid, $neighbor);
 
                     if ($distance < $this->radius) {
-                        unset($this->centroids[$j]);
+                        unset($centroids[$j]);
                     }
                 }
             }
@@ -243,10 +243,10 @@ public function train(Dataset $dataset) : void
                 break 1;
             }
 
-            $previous = $this->centroids;
+            $previous = $centroids;
         }
 
-        $this->centroids = array_values($this->centroids);
+        $this->centroids = array_values($centroids);
 
         if ($this->logger) {
             $this->logger->info('Training complete');
@@ -296,7 +296,7 @@ protected function assign(array $sample) : int
     }
 
     /**
-     * Calculate the magnitude (l1) of a centroid shift from the previous epoch.
+     * Calculate the magnitude (l1) of centroid shift from the previous epoch.
      *
      * @param array $previous
      * @return float
@@ -306,10 +306,10 @@ protected function centroidShift(array $previous) : float
         $shift = 0.;
 
         foreach ($this->centroids as $cluster => $centroid) {
-            $prevCluster = $previous[$cluster];
+            $prevCentroid = $previous[$cluster];
 
             foreach ($centroid as $column => $mean) {
-                $shift += abs($prevCluster[$column] - $mean);
+                $shift += abs($prevCentroid[$column] - $mean);
             }
         }
 

src/Clusterers/GaussianMixture.php

@@ -272,8 +272,7 @@ public function predict(Dataset $dataset) : array
         $x = Matrix::build($dataset->samples());
 
         if ($this->logger) {
-            $this->logger->info('Computing pairwise high'
-                . ' dimensional affinities');
+            $this->logger->info('Computing affinity matrix');
         }
 
         $distances = $this->pairwiseDistances($x);

src/Clusterers/MeanShift.php

@@ -20,7 +20,7 @@
  * References:
  * [1] S. M. Omohundro. (1989). Five Balltree Construction Algorithms.
  * [2] M. Dolatshah et al. (2015). Ball*-tree: Efficient spatial indexing for
- * constrained nearest-neighbor search in metric spaces
+ * constrained nearest-neighbor search in metric spaces.
  *
  * @category    Machine Learning
  * @package     Rubix/ML

src/Embedders/TSNE.php

@@ -8,6 +8,10 @@
  * The Dense Random Projector uses a random matrix sampled from a dense uniform
  * distribution [-1, 1] to project a sample matrix onto a target dimensionality.
  *
+ * References:
+ * [1] D. Achlioptas. (2003). Database-friendly random projections:
+ * Johnson-Lindenstrauss with binary coins.
+ *
  * @category    Machine Learning
  * @package     Rubix/ML
  * @author      Andrew DalPino

src/Graph/BallTree.php

@@ -13,6 +13,10 @@
  * The Sparse Random Projector uses a random matrix sampled from a sparse uniform
  * distribution (mostly 0s) to project a sample matrix onto a target dimensionality.
  *
+ * References:
+ * [1] D. Achlioptas. (2003). Database-friendly random projections:
+ * Johnson-Lindenstrauss with binary coins.
+ *
  * @category    Machine Learning
  * @package     Rubix/ML
  * @author      Andrew DalPino

src/Transformers/DenseRandomProjector.php

@@ -18,6 +18,10 @@
  * > **Note**: This transformer assumes that its input is made up of word
  * frequency vectors such as those created by the Word Count Vectorizer.
  *
+ * References:
+ * [1] S. Robertson. (2003). Understanding Inverse Document Frequency: On
+ * theoretical arguments for IDF.
+ *
  * @category    Machine Learning
  * @package     Rubix/ML
  * @author      Andrew DalPino