Spark DBSCAN is an implementation of the DBSCAN clustering algorithm on top of Apache Spark . It also includes 2 simple tools which will help you choose parameters of the DBSCAN algorithm.

This software is EXPERIMENTAL , it supports only Euclidean and Manhattan distance measures ( why? ) and it is not well optimized yet. I tested it only on small datasets (millions of records with 2 features in each record).

You can use Spark DBSCAN as a standalone application which you can submit to a Spark cluster ( Learn how ). Alternatively, you can include it into your own app - its API is documented and easy to use ( Learn how ).

Learn more about:

• How it works
• How to use it as a standalone application
• How to choose parameters of DBSCAN algorithm
• How to include it in your application

Performance

Credits

I was glad to receive contributions from other people and I'd like to say thank you:

• Mark Geraty - for fixing a bug with Java RDDs;
• @agrinh - for adding compatibility with Spark 1.1.0