Github user srowen commented on a diff in the pull request:
https://github.com/apache/spark/pull/20518#discussion_r166336855
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala ---
@@ -745,4 +763,27 @@ private[spark] class CosineDistanceMeasure extends
DistanceMeasure {
override def distance(v1: VectorWithNorm, v2: VectorWithNorm): Double = {
1 - dot(v1.vector, v2.vector) / v1.norm / v2.norm
}
+
+ /**
+ * Updates the value of `sum` adding the `point` vector.
+ * @param point a `VectorWithNorm` to be added to `sum` of a cluster
+ * @param sum the `sum` for a cluster to be updated
+ */
+ override def updateClusterSum(point: VectorWithNorm, sum: Vector): Unit
= {
+ axpy(1.0 / point.norm, point.vector, sum)
+ }
+
+ /**
+ * Returns a centroid for a cluster given its `sum` vector and its
`count` of points.
+ *
+ * @param sum the `sum` for a cluster
+ * @param count the number of points in the cluster
+ * @return the centroid of the cluster
+ */
+ override def centroid(sum: Vector, count: Long): VectorWithNorm = {
+ scal(1.0 / count, sum)
+ val norm = Vectors.norm(sum, 2)
--- End diff --
Rather than scale `sum` twice, can you just compute its normal and then
scale by 1 / (norm * count * count)?
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
