Symbiotic Evolutionary Subspace Clustering (S-ESC) source code

Code was developed by Ali Vahdat. Distribution is permitted for research purposes alone. Note this site is very much a case of work in progress...!

S-ESC Version 2 (2012) -- bi-objective S-ESC

The S-ESC algorithm is a GA designed to address the task of subspace clusting. The subspace clustering task implies that attributes are identified -- potentially on a cluster-by-cluster basis -- at the same time as the configuration of the clusters themselves. Unlike k-means style algorithms we do not assume that the relevant number of clusters are known a priori. Instead a 'bottom-up' approach is assumed in which the EM algorithm (e.g., see Weka) is first applied attribute wise to distinguish the 'projections' on each axis independently. A fundamental assumption is therefore that axis parallel projections do not preclude the identification of clusters. On the other hand, S-ESC takes as a firm goal, the objective of attribute selection. Thus unlike soft projected clustering methods, the result is a definative set of clusters with dimensionality lower than that of the original task.

The following resources are available:
1. The research as distributed here was first described at IEEE CEC 2012.
  - Actually, the code also impliments random subset sampling for improved scaling to larger cardinality data sets.
  - See article in Evolutionary Intelligence for an extensive benchmarking study and literature review.
2. Readme;
3. C++ code distribution;
4. Scripts;
5. Benchmarking data sets.

Enjoy!