A Scala library for data clustering, utilizing Breeze.

This is an updated fork from stsc, here. The original author is Armand Grillet. Updates, extensions and modifications by Hanns Holger Rutz. The base package is now de.sciss.cluster. In the future, more clustering algorithms might be added.

For now, it contains an implementation of the Self-Tuning Spectral Clustering algorithm, based on the paper Self Tuning Spectral Clustering STSC.

Some clusters found by the implementation on six different datasets with k in [2, 6]:

This project builds with sbt against Scala 2.13, 2.12. Scala 3 support is currently missing, since Breeze is not published for Scala 3. You can use the library in Scala 3 by qualifying the dependency via .cross(CrossVersion.for3Use2_13).

To link to this library:

libraryDependencies += "de.sciss" %% "cluster" % v

The current version v is "0.1.0"

The main class to use is the self-tuning spectral clustering algorithm, de.sciss.cluster.STSC:

import de.sciss.cluster.STSC

// Your code to load the dataset and make it a DenseMatrix.
val (cBest, costs, clusters) = STSC.cluster(dataset)
// Two optional parameters, the minimum and maximum value of k. By default: 2 and 6.
val bestGroupNumber = STSC.cluster(dataset, 2, 10)._1

cBest is the most likely number of clusters in the dataset. It will always be between the minimum and maximum value of k. costs is a map with the cost returned for each possible number of clusters k. clusters is an Array of Int with a length equals to the number of rows in the dataset, clusters(i) represents the cluster where should be the observation i in the dataset.

If you include de.sciss.cluster.KDTree, you can also create a k-d tree to divide a dataset:

scala >

import de.sciss.cluster.KDTree

scala >
val tree = KDTree.createWithMaxObservations(dataset, maxObservationsPerTile, tileBorderWidth)

The third class of the library is Tile, a list of tiles composing a k-d tree. A Tile is composed of two DenseVectors representing the minimums and maximums coordinates in each dimension.

Copyright © 2016 Armand Grillet. All rights reserved.