Outlier Detection with One-Class Kernel Fisher Discriminants

This software implements the outlier detection method described in [Volker Roth. Outlier Detection with One-class Kernel Fisher Discriminants. Advances in Neural Information Processing Systems 17. MIT Press, 2004]. If you use the software, please cite this reference.

LICENSE:

This software is distributed under the GNU General Public License. The author is not responsible for implications from the use of this software. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY.

REQUIREMENTS:

  1. R. You can obtain it from the R homepage. I have only tested it with version R-2.0.1.
  2. The g++ compiler (I have only tested it with the old gcc-2.96).
  3. GNU scientific library "libgsl" and the corresponding CBLAS library "libgslcblas".

Download

Download the tar archive.

INSTALLATION (under Linux):

Type "tar xfvz oc_kfd.tgz". This creates the directory "publicOC-KFD" with subdirectories "faces", "R", "CC".

COMPILATION:

You need to have the g++ compiler (I have only tested it with gcc-2.96), the GNU scientific library "libgsl" and the corresponding CBLAS library "libgslcblas". Enter the subdirectory "publicOC-KFD/CC". Edit the Makefile and adjust IFLAGS and GSL_LIBS according to the location of "libgsl.a" and "libgslcblas.a" on your computer. Typing "make" should then build the shared object file "vegas.so".

RUNNING the R script:

Enter the directory "publicOC-KFD/R". Start R. Type "source("oneclass_KFD.R")" within R. The script is configured to detect outliers in a collection of face images (which can be found in the "publicOC-KFD/faces" subdirectory. For details see [Volker Roth. Outlier Detection with One-class Kernel Fisher Discriminants. Advances in Neural Information Processing Systems 17. MIT Press, 2004].

For questions/comments please send me an e-mail: vroth "at" inf.ethz.ch.

Last modified: 21.04.2005