Affinity Propagation on Spark

Affinity Propagation (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 Frey and Dueck. Please refer to the paper[1].

Affinity Propagation on Spark implements Affinity Propagation algorithm on cluster computing system Spark. By leveraging computing cluster, you can run this clustering algorithm on large-scale data sets.

[1] Brendan J. Frey; Delbert Dueck (2007). "Clustering by passing messages between data points". Science. 315 (5814): 972-976 PDF

Build

Currently it supports Spark 2.2.0. It also has tested on Spark 1.6.0 before. You can build the package by using sbt.

sbt/sbt assembly

Spark package

You can simply use the affinity propagation on Spark by importing the spark package SparkAffinityPropagation

bin/spark-shell --packages viirya:SparkAffinityPration:1.0

API

AffinityPropagation class provides the API for performing clustering. You can set the maximum iteration numbers for Affinity Propagation by calling setMaxIterations. The graph used as input to Affinity Propagation algorithm is represented as a RDD of similarities between vertices. The vertices are represented by theirs ids in Long type. The similarities are Double type.

For example, a RDD of similarities from local data can be initialized:

val similarities = Seq[(Long, Long, Double)]((0, 1, -8.2), (0, 3, -5.8), (1, 2, -0.4))
val rdd = sc.parallelize(similarities, 2)

Then, use this RDD as input to AffinityPropagation.run:

val ap = new AffinityPropagation()
val similaritiesWithPreferneces = ap.determinePreferences(rdd)
val model = ap
  .setMaxIterations(30)
  .run(similaritiesWithPreferneces)

Example

In the unit test org.viirya.AffinityPropagationSuite, you can find how to run it with its Scala API.

For Java users, org.viirya.exemplar.JavaAffinityPropagation provides Java example to run this clustering algorithm.