Distributed Stream Clustering

SAMOA Clustering Algorithm

The SAMOA Clustering Algorithm is invoked by using the ClusteringEvaluation task. The clustering task can be executed with default values just by running:

cd src/main/resources
./samoa storm ../../../target/SAMOA-Storm-0.0.1.jar "ClusteringEvaluation"

Parameters:

• -l: clusterer to train
• -s: stream to learn from
• -i: maximum number of instances to test/train on (-1 = no limit)
• -f: how many instances between samples of the learning performance
• -n: evaluation name (default: ClusteringEvaluation_TimeStamp)
• -d: file to append intermediate csv results to

In terms of the SAMOA API, Clustering Evaluation consists of a source processor, a clusterer, and a evaluator processor. Source processor sends the instances to the classifier using source stream. The clusterer sends the clustering results to the evaluator processor via the result stream. The source PI corresponds to the -s option of Clustering Evaluation, and the clusterer corresponds to the -l option.