OPC: Optimal Projections for Clustering
An open source MATLAB and Octave library for Dimensionality Reduction for Clustering
OPC is an open source MATLAB and GNU Octave package that implements clustering methods that seek the optimal low dimensional subspace to identify clusters. The User Guide provides
a description of all the methods implemented in OPC, and numerous detailed examples of using the package. (PDF version)
Whenever the data contains irrelevant features, or correlations
among subsets of features exist (which is typical in
high-dimensional data), or when clusters are defined in
different subspaces, the spatial data structure becomes less
informative about the underlying clusters. Under these
conditions clustering algorithms need to simultaneously solve
two interrelated problems: (i) identify the subspace in which
clusters can be distinguished, and (ii) associate observations
to clusters.
OPC focuses on methods which seek low dimensional subspaces
that are optimal with respect to specific clustering criteria.
This distinguishes the methods in OPC from generic
dimensionality reduction techniques that optimise objective
functions that are not related to any clustering criterion, and
are therefore not guaranteed to preserve the cluster structure.
Generated by m2html © 2005