[mlpack-svn] r16545 - mlpack/trunk/src/mlpack/methods/kmeans

[fastlab-svn at coffeetalk-1.cc.gatech.edu]

Author: rcurtin
Date: Fri May 23 12:03:41 2014
New Revision: 16545

Log:
Don't select points from clusters of size 1.


Modified:
   mlpack/trunk/src/mlpack/methods/kmeans/max_variance_new_cluster_impl.hpp

Modified: mlpack/trunk/src/mlpack/methods/kmeans/max_variance_new_cluster_impl.hpp
==============================================================================
--- mlpack/trunk/src/mlpack/methods/kmeans/max_variance_new_cluster_impl.hpp	(original)
+++ mlpack/trunk/src/mlpack/methods/kmeans/max_variance_new_cluster_impl.hpp	Fri May 23 12:03:41 2014
@@ -37,8 +37,12 @@
   }
 
   // Divide by the number of points in the cluster to produce the variance.
+  // Although a -nan will occur here for the empty cluster(s), this doesn't
+  // matter because variances.max() won't pick it up.  If the number of points
+  // in the cluster is 1, we ensure that cluster is not selected by forcing the
+  // variance to 0.
   for (size_t i = 0; i < clusterCounts.n_elem; ++i)
-    variances[i] /= clusterCounts[i];
+    variances[i] /= (clusterCounts[i] == 1) ? DBL_MAX : clusterCounts[i];
 
   // Now find the cluster with maximum variance.
   arma::uword maxVarCluster;