Table of Contents
LIMBO Algorithm
History
The algorithm was developed by Periklis Andritsos and Vassilios Tzerpos.
Algorithm Intent
To generate decompositions that exhibit the least information lost when entities are represented by their clusters.
Factbase Properties
The factbase is a generic dependency data table where each row describes one entity to be clustered. Each column contains the value for a specific attribute.
Clustering Objectives
The Limbo algorithm produces software decompositions with minimum information loss.
Process Description
An agglomerative clustering algorithm that on each step merges the two clusters with the least information loss.
The information lost is calculated using the Agglomerative Information Bottleneck algorithm.
Decomposition Properties
The LIMBO decomposition has a small value for its information lost function.
Algorithm Restrictions
Failed Assumptions
Detailed Algorithm Description
LIMBO has four phases:
- Creation of the Summary Artefacts
- Application of the AIB algorithm
- Associating original artefacts with clusters
- Determining the number of clusters