The Curve Clustering Toolbox is a Matlab toolbox that implements a family of probabilistic model-based curve-aligned clustering algorithms. The cluster models themselves are based on polynomial and spline regression mixture models that allow for continuous curve alignment in both measurement space and in time. Learning is carried out using an EM (Expectation-Maximization) framework. The model specification and learning framework are detailed in (Gaffney, 04).
The toolbox takes a set of curves such as this set of cyclones
and finds a set of clusters in an unsupervised manner such as these
The toolbox currently contains over 15 different clustering methods from K-means; to Gaussian mixtures; polynomial regression mixtures; spline regression mixtures; and various time-aligned, space-aligned, and time- and space-aligned regression mixtures.
The toolbox also contains many different types of functions for visualization, model selection, and data simulation. It is self-contained and does not require any special matlab toolboxes to function. Of course, it is also free!
There is both online documentation and matlab-based documentation. To access the matlab-based documentation, download and unzip the toolbox in an appropriate directory. Then, run matlab, change to the install directory, and type setcctpath. Now you can access the main help screen by typing help_cct.
The current version of the CCToolbox is 0.98 (July 7th, 2005)