Use this form to retrieve matlab function files that implement the basic Principal Direction Divisive Partitioning method. This software requires the use of Matlab 5.x or better. This is being posted on the WWW in order to facilitate others experimenting with the method to see if it may meet their needs. However, it is definitely not designed to be a production code in any sense of the word. It has been successful on applications derived from text document clustering, based on a matrix of word counts which must be supplied. This software does only the basic partitioning/clustering of the dataset. Some software to provide a taxonomy or labels for the clusters is included, but this is less developed and less documented than the main clustering software. Some shell scripts are provided (with little documentation) to generate the matrices of word counts from arbitrary text documents. These shell scripts require the usual unix tools, plus the 'guile' scheme interpreter (version at most 1.8.x, replacing the former 'snow' command, as of May 2004, switches updated October 2006).
Last changes: two functions called dumpsketchtree and drawdendrogram that create a clickable tree, browsable with an ordinary web browser. An example of such a clickable tree can be found here. This new software, while still under development, may be useful for anyone wishing to explore a dataset, but we cannot guarantee it will work on your dataset. The older distribution is available here.
See description of original algorithm in
D. L. Boley: Principal Direction Divisive Partioning, TR-97-056, 1997.
Sample data as reported in this paper can be found here.
- Daniel Boley