Archive

Anton Schwaighofer’s SVML MATLAB interface toolbox is used for loading SVML data into matlab and wrapping SVMLight during training. The page where it was originally hosted is http://www.igi.tugraz.at/aschwaig/software.html and then it was moved to http://ida.first.fraunhofer.de/~anton/software.html. seems to have disappeared, so I’m putting links to the parts I have here. This information comes to you courtesy of the Wayback machine (http://www.archive.org).

For convenience I’ve added the software to GitHub repos under the sods organisation.

Here is the original text from the page (the links won’t work apart from those to the repos).:

Software

If you should find any the following pieces of software useful, or if you have bug reports, suggestions, comments, etc. please drop me a line.

Gaussian Process Regression with Bayesian Committee Machine

Download the latest version here (version 1.0 of November 2005)

The Bayesian Committee Machine (BCM) is an approximation method for large scale Gaussian Process regression.

This software is released under the GNU general public license. The BCM toolbox is based on Netlab, a free neural network toolbox. Netlab needs to be installed before using the BCM toolbox.

SVM toolbox for Matlab

Download the latest version here (version 2.51 of January 2002)

Over the years I have written a set of Matlab routines for support vector machine classification. Advantages of this toolbox are:

  • Handling of multi-class problems via error correcting output codes (ECOC).
  • Written completely in Matlab, allowing for easy modification. Special kinds of kernels that require extensive computation (such as the Fisher kernel, which is based on a model of the data) can easily be incorporated.
  • Unless many other SVM toolboxes, this one can handle 1norm SVMs and 2norm SVMs (linear or quadratic loss function)
  • Uses decomposition methods and working set selection strategies like SVM light by Thorsten Joachims. The toolbox can thus handle problems of up to a few 10000 training points.
  • Optimized computations for linear SVMs and sparse data.
  • Can handle SVMs with different costs of misclassification (per class or per example).

This software is released under the GNU general public license.

The SVM toolbox is written in the style of Netlab, a free neural network toolbox. It requires the Matlab optimization toolbox (version 1.5 or 2.0) or Alex Smola´s PRLOQO solver (included in the toolbox).

Matlab interface to SVM light

Download the latest version here (version V0.92 of August 2002)

This is a set of simple Matlab functions that make it easier to interface with SVM light from within Matlab. You can write out Matlab matrices into SVM light’s file format, read the result files, and set program option without having to deal with the command line options.

The toolbox has been written for SVM light V4.00. Modifications to handle future versions should be straightforward.

This software is released under the GNU general public license.

Last modification $Date: 2006/10/25 02:38:05 $