mraad / dbscan-scala

DBSCAN 2D implementation in Scala

DBSCAN in Scala

DBSCAN 2D (for now) implementation in Scala.

2021-11-23: Removed eps from DBSCAN as proximity is defined by NNSearch trait implementation.

2021-10-04: Breaking changes to make the API more abstract.

2018-03-27: Added MinPointsCalculator trait and default MinPointsLength implementation. This enables the implementation of a weighted DBSCANPoint implementation, where the min points to form a cluster can be calculated based on the sum of the weighted points.

2018-03-04: Revamped the implementation to make DBSCANPoint an immutable trait.

2018-02-22: Project now depends on Euclid.

Comparing to Commons Math

Wanted to make sure my implementation was "correct", so I compared it with Commons Math. But kept getting different cluster counts! When I looked at the source code (Gotta love OSS :-) I found a small discrepancy in the calculation of the number of neighbors. Per Wikipedia's pseudocode implementation (because everything you read on the internet is correct :-) the regionQuery should include the center point, as we are calculating the number of points forming a cluster.

IMHO, the code in the DBSCANClusterer should be:

if (distance(neighbor, point) <= eps) {

rather than:

if (point != neighbor && distance(neighbor, point) <= eps) {

And the test code still passes!

Boundary Conditions

The concept of neighborhood is implemented as d_i² < ε², where d_i is the euclidean distance between two points.

Publish to Bintray

mvn release:clean
mvn release:prepare
mvn release:perform

Using the project as dependency

<repositories>
    <repository>
        <id>bintray</id>
        <url>http://dl.bintray.com/mraad/maven</url>
        <releases>
            <enabled>true</enabled>
        </releases>
        <snapshots>
            <enabled>false</enabled>
        </snapshots>
    </repository>
</repositories>

Scala 2.11 dependency:

<dependency>
    <groupId>com.esri</groupId>
    <artifactId>dbscan-scala</artifactId>
    <version>0.30</version>
    <classifier>2.11</classifier>
</dependency>

Scala 2.12 dependency:

<dependency>
    <groupId>com.esri</groupId>
    <artifactId>dbscan-scala</artifactId>
    <version>0.30</version>
    <classifier>2.12</classifier>
</dependency>

TODO

• Make DBSCANPoint multi-dimensional.
• Implement weighted clustering.
• Make DBSCANPoint immutable and return ClusteredPoint.
• Remove dependency on Esri Geometry API.

References

• http://people.cs.nctu.edu.tw/~rsliang/dbscan/testdatagen.html
• https://help.github.com/articles/generating-ssh-keys/
• http://veithen.github.io/2013/05/26/github-bintray-maven-release-plugin.html
• http://www.sciweavers.org/free-online-latex-equation-editor
• http://blog.dominodatalab.com/topology-and-density-based-clustering/
• https://en.wikipedia.org/wiki/OPTICS_algorithm
• http://www.naftaliharris.com/blog/visualizing-dbscan-clustering/