Github user viirya commented on a diff in the pull request:
https://github.com/apache/spark/pull/4622#discussion_r27896111
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/clustering/AffinityPropagation.scala
---
@@ -0,0 +1,347 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.mllib.clustering
+
+import scala.collection.mutable.Map
+import scala.collection.mutable.Set
+
+import org.apache.spark.{Logging, SparkException}
+import org.apache.spark.annotation.Experimental
+import org.apache.spark.graphx._
+import org.apache.spark.graphx.impl.GraphImpl
+import org.apache.spark.rdd.RDD
+
+/**
+ * :: Experimental ::
+ *
+ * Model produced by [[AffinityPropagation]].
+ *
+ * @param clusters the vertexIDs of each cluster.
+ * @param exemplars the vertexIDs of all exemplars.
+ */
+@Experimental
+class AffinityPropagationModel(
+ val clusters: Seq[Set[Long]],
+ val exemplars: Seq[Long]) extends Serializable {
+
+ /**
+ * Set the number of clusters
+ */
+ def getK(): Int = clusters.size
+
+ /**
+ * Find the cluster the given vertex belongs
+ */
+ def findCluster(vertexID: Long): Set[Long] = {
+ clusters.filter(_.contains(vertexID))(0)
+ }
+
+ /**
+ * Find the cluster id the given vertex belongs
+ */
+ def findClusterID(vertexID: Long): Option[Int] = {
+ var i = 0
+ clusters.foreach(cluster => {
+ if (cluster.contains(vertexID)) {
+ return Some(i)
+ }
+ i += i
+ })
+ None
+ }
+}
+
+/**
+ * :: Experimental ::
+ *
+ * AffinityPropagation (AP), a graph clustering algorithm based on the
concept of "message passing"
+ * between data points. Unlike clustering algorithms such as k-means or
k-medoids, AP does not
+ * require the number of clusters to be determined or estimated before
running it. AP is developed
+ * by
[[http://www.psi.toronto.edu/affinitypropagation/FreyDueckScience07.pdf Frey
and Dueck]].
+ *
+ * @param maxIterations Maximum number of iterations of the AP algorithm.
+ *
+ * @see [[http://en.wikipedia.org/wiki/Affinity_propagation (Wikipedia)]]
+ */
+@Experimental
+class AffinityPropagation private[clustering] (
+ private var maxIterations: Int,
+ private var lambda: Double,
+ private var normalization: Boolean) extends Serializable {
+
+ import org.apache.spark.mllib.clustering.AffinityPropagation._
+
+ /** Constructs a AP instance with default parameters: {maxIterations:
100, lambda: 0.5,
+ * normalization: false}.
+ */
+ def this() = this(maxIterations = 100, lambda = 0.5, normalization =
false)
+
+ /**
+ * Set maximum number of iterations of the messaging iteration loop
+ */
+ def setMaxIterations(maxIterations: Int): this.type = {
+ this.maxIterations = maxIterations
+ this
+ }
+
+ /**
+ * Get maximum number of iterations of the messaging iteration loop
+ */
+ def getMaxIterations(): Int = {
+ this.maxIterations
+ }
+
+ /**
+ * Set lambda of the messaging iteration loop
+ */
+ def setLambda(lambda: Double): this.type = {
+ this.lambda = lambda
+ this
+ }
+
+ /**
+ * Get lambda of the messaging iteration loop
+ */
+ def getLambda(): Double = {
+ this.lambda
+ }
+
+ /**
+ * Set whether to do normalization or not
+ */
+ def setNormalization(normalization: Boolean): this.type = {
+ this.normalization = normalization
+ this
+ }
+
+ /**
+ * Get whether to do normalization or not
+ */
+ def getNormalization(): Boolean = {
+ this.normalization
+ }
+
+ /**
+ * Run the AP algorithm.
+ *
+ * @param similarities an RDD of (i, j, s,,ij,,) tuples representing the
similarity matrix, which
+ * is the matrix S in the AP paper. The similarity
s,,ij,, is set to
+ * real-valued similarities. This is not required to
be a symmetric matrix
+ * and hence s,,ij,, can be not equal to s,,ji,,.
Tuples with i = j are
+ * referred to as "preferences" in the AP paper. The
data points with larger
+ * values of s,,ii,, are more likely to be chosen as
exemplars.
+ *
+ * @param symmetric the given similarity matrix is symmetric or not.
Default value: true
+ * @return a [[AffinityPropagationModel]] that contains the clustering
result
+ */
+ def run(similarities: RDD[(Long, Long, Double)], symmetric: Boolean =
true)
+ : AffinityPropagationModel = {
+ val s = constructGraph(similarities, normalization, symmetric)
+ ap(s)
+ }
+
+ /**
+ * Runs the AP algorithm.
+ *
+ * @param s The (normalized) similarity matrix, which is the matrix S in
the AP paper with vertex
+ * similarities and the initial availabilities and
responsibilities as its edge
+ * properties.
+ */
+ private def ap(s: Graph[Seq[Double], Seq[Double]]):
AffinityPropagationModel = {
+ val g = apIter(s, maxIterations, lambda)
+ chooseExemplars(g)
+ }
+}
+
+private[clustering] object AffinityPropagation extends Logging {
+ /**
+ * Construct the similarity matrix (S) and do normalization if needed.
+ * Returns the (normalized) similarity matrix (S).
+ */
+ def constructGraph(similarities: RDD[(Long, Long, Double)], normalize:
Boolean,
+ symmetric: Boolean):
+ Graph[Seq[Double], Seq[Double]] = {
+ val edges = similarities.flatMap { case (i, j, s) =>
+ if (symmetric && i != j) {
+ Seq(Edge(i, j, Seq(s, 0.0, 0.0)), Edge(j, i, Seq(s, 0.0, 0.0)))
+ } else {
+ Seq(Edge(i, j, Seq(s, 0.0, 0.0)))
+ }
+ }
+
+ if (normalize) {
+ val gA = Graph.fromEdges(edges, Seq(0.0))
+ val vD = gA.aggregateMessages[Seq[Double]](
+ sendMsg = ctx => {
+ ctx.sendToSrc(Seq(ctx.attr(0)))
+ },
+ mergeMsg = (s1, s2) => Seq(s1(0) + s2(0)),
+ TripletFields.EdgeOnly)
+ val normalized = GraphImpl.fromExistingRDDs(vD, gA.edges)
+ .mapTriplets(
+ e => {
+ val s = if (e.srcAttr(0) == 0.0) { e.attr(0) } else {
e.attr(0) / e.srcAttr(0) }
+ Seq(s, 0.0, 0.0)
+ }, TripletFields.Src)
+ Graph.fromEdges(normalized.edges, Seq(0.0, 0.0))
--- End diff --
If you mean `vD`, I think its type is different to what we use here
(`Seq(0.0, 0.0)`)?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
