ClusterFinder is a tool in development, designed to recover meaningful clusters of classes (on the road to software components) in a system.
Written by Toon Verwaest in 2007 as validation of his Master’s thesis.
It is a pluggable tool in which new class clustering techniques can be explored and incremented. Results of (combinations of) plug-ins are presented to the user, who can then combine results to build clusters. The currently available plug-ins are early adaptations of several algorithms described in the Phd. thesis of R. Koschke.
The tool presents the user with a way to analyse clusters for inner completeness and requirements. Inner completeness means that parts of class hierarchies, the lines from superclasses in the cluster, to all subclasses in the cluster, have to be fully contained in the cluster. Absent classes are called ghostclasses. The requirements of a cluster, are classes which are referenced by classes in the cluster, and the list of classes of which the classes in the cluster subclass.
In order to aid the user in easily analysing the effectiveness of (automatically) combining results of clustering techniques, several class-cluster visualizations are presented. Some of the visualizations can be used to identify classes and relationships which (possibly falsely) hold the cluster together. Such classes and relationships can then be ignored (removed) before the techniques are recalculated. Other visualizations can be used to inspect how clusters are embedded in their environment.
The current research prototype of ClusterFinder is available at the following Store coordinates:
Bundle: Clustering
interface: PostgresSQLEXDIConnection environment: db.iam.unibe.ch_scgStore user name: storeguest password: storeguest table owner: BERN
Please contact Toon Verwaest for questions and feedback.