SVMheavy homepage


SVMheavy is yet another SVM library. It was originally written as a testbed to compare different incremental training methodologies, but the codebase has since been extended significantly. Both binary classification and regression are done using a unified SVM core, with incremental and decremental training abilities and parameter variation facilities built in for both. Code may be used at the command line (where some effort has been made to ensure compatibility with SVMlight), interactively, or directly from other code by accessing the SVM_pattern and SVM_regress classes. At present it supports active set training (as per "Incremental Training of Support Vector Machines", A. Shilton, M. Palaniswami, D. Ralph, A. C. Tsoi, accepted for publication in IEEE Transactions on Neural Networks), Platt's SMO algorithm and Daniel Lai's D2C algorithm.

The code was written by Alistair Shilton, University of Melbourne, Electrical and Electronic Engineering department (a p s h at ee dot unimelb dot edu dot au). Currently there is no real documentation, but we do plan to write some eventually (compilation instructions can be found in the readme file in the zip). The code has been tested and doesn't seem to have any major bugs, but caveat emptor.

Compiling SVMHeavy requires both RIMElib and the GNU Scientific Library. Code is licenced under GPL.

Click here to download source-code.

Old versions

Version 2 source.
Version 2 binary (windows dos box).
Version 2 library files.
Version 1.3 source.
Version 1.3 binary (windows dos box).
Version 1.2 source.
Version 1.2 binary (windows dos box).


This page, its contents and style, are the responsibility of the author and do not necessarily represent the views, policies or opinions of The University of Melbourne.

Created : 2002

Last Modified : 7th October 2005
Maintained by : Alistair Shilton (a p s h at ee dot unimelb dot edu dot au)

Valid HTML 4.01!