Clustering Patterns of Handwritten Digits

Download: digits.zip. Works on Windows and Linux, and on Geowalls.

Concepts, ideas, and software that originate for one branch of science often end up being used in other branches. This demo shows how software originally written to visualize astronomical data, is actually a useful general purpose visualization tool.

Dinoj Surendran was a graduate student in computer science on the lookout for a good visualization tool when he began helping Mark Subbarao on visualizing the Sloan Digital Sky Survey.

By happy accident, Partiview, the package used by Mark, proved to be just what Dinoj was looking for. Since then, he's been using Partiview as a general purpose tool on different projects, and Stuart Levy (its creator) has been adding features to make it useful to the larger scientific community.

Most plotting tools only permit you to place colored dots at points. With Partiview, you can place pictures at points. When the data you are analyzing has a natural visual representation, this is very useful for interpreting data. In this demo, the result of a clustering algorithm on a standard dataset of 5000 handwritten digits is shown, with the original digits.

For further information, please see Dinoj's site.