PUB-923: Fix K-Means to pass golden R test.

Problem: within_cluster_variances was computed correctly, but the name was
wrong, since the R wrapper equates this value with withinss, which is the
within_cluster_sum_of_squares. So in order to not upset the R side of things,
we compute sum of squares instead of variances.

Also re-enable returning total_within_SS from R.

R/h2o-package/R/Algorithms.R

@@ -333,8 +333,8 @@ h2o.kmeans <- function(data, centers, cols = '', key = "", iter.max = 10, normal
   result$centers = t(matrix(unlist(res$centers), ncol = res$parameters$k))
   dimnames(result$centers) = list(seq(1,res$parameters$k), feat)
   #result$totss <- res$total_SS
-  result$withinss <- res$within_cluster_variances
-  #result$tot.withinss <- res$total_within_SS
+  result$withinss <- res$within_cluster_variances ## FIXME: sum of squares != variances (bad name of the latter)
+  result$tot.withinss <- res$total_within_SS
   #result$betweenss <- res$between_cluster_SS
   result$size <- res$size
   result$iter <- res$iterations

src/main/java/hex/KMeans2.java

@@ -185,9 +185,11 @@ public KMeans2() {
         double ssq = 0;       // sum squared error
         for( int i=0; i<k; i++ ) {
           ssq += model.within_cluster_variances[i]; // sum squared error all clusters
-          model.within_cluster_variances[i] /= task._rows[i]; // mse per-cluster
+//          model.within_cluster_variances[i] /= task._rows[i]; // mse per-cluster
         }
-        model.total_within_SS = ssq/fr.numRows(); // mse total
+//        model.total_within_SS = ssq/fr.numRows(); // mse total
+        model.total_within_SS = ssq; //total within sum of squares
+
         model.update(self()); // Update model in K/V store
 
         // Compute change in clusters centers
@@ -289,7 +291,7 @@ public static Response redirect(Request req, Key model) {
           rows[i][0] = model.within_cluster_variances[i];
         columnHTMLlong(sb, "Cluster Size", model.size);
         DocGen.HTML.section(sb, "Cluster Variances: ");
-        table(sb, "Clusters", new String[]{"Within Cluster Variances"}, rows);
+        table(sb, "Clusters", new String[]{"Within Cluster Sum of Squares"}, rows);
 //        columnHTML(sb, "Between Cluster Variances", model.between_cluster_variances);
         sb.append("<br />");
         DocGen.HTML.section(sb, "Overall Totals: ");
@@ -406,7 +408,10 @@ public static class KMeans2Model extends Model implements Progress {
     public int iterations;
 
     @API(help = "Within cluster sum of squares per cluster")
-    public double[] within_cluster_variances;
+    public double[] within_cluster_variances; //Warning: See note below
+    //Note: The R wrapper interprets this as withinss (sum of squares), so that's what we compute here, and NOT the variances.
+    //FIXME: => wrong name, should be within_cluster_sum_of_squares, but leaving to be backward-compatible with REST API
+
 
 //    @API(help = "Between Cluster square distances per cluster")
 //    public double[] between_cluster_variances;