libSVDD: A Library for Support Vector Data Description

Copyright, 2020 Sohail R. Reddy (sredd001@fiu.edu)

Version v1.0.0

libSVDD is a library for single and multi-class classification written in object oriented Fortran 90 with a Python API. The library contains the following kernels:

• (1) Gaussian kernel: k(x,y) = exp(-(norm(x-y)/s)^2)
• (2) Exponential kernel: k(x,y) = exp(-(norm(x-y))/s^2)
• (3) Linear kernel: k(x,y) = x'*y
• (4) Laplace kernel: k(x,y) = exp(-(norm(x-y))/s)
• (5) Sigmoid kernel: k(x,y) = tanh(g*x'*y+c)
• (6) Polynomial kernel: k(x,y) = (x'*y+c)^d

Getting Started

The repository contains two options for using libSVDD.

• Incorporating the libSVDD source code into your own code
• Linking directly with the static or shared libSVDD library (.so or .a) either using Fortran, C or Python (API)

Prerequisites

Compiling the library requires the gfortran compiler. The Python API requires that numpy be installed

Installing

Navigate to the source (src) directory and run the make command

cd src/; make

Using libSVDD

The 'examples' folder contains two examples for using libSVDD in a Fortran and Python environment. The 'libSVDD.pdf' in the 'docs/' folder contains the theory and information on the use of libSVDD.

How to Cite

If you use libSVDD, please cite the following

Sohail R. Reddy, "libSVDD: A Library for Support Vector Data Description," 2020, DOI:10.5281/zenodo.3712443

Contributing

The libSVDD framework can be expanded to allow for shared and distributed memory parallelization and for use on GPUs. Other areas include better solvers for the constrained optimization problems. Feel free to modify the library. The modification must conform to the GNU License Agreement

Authors

Sohail R. Reddy

License

This project is licensed under the GNU General Public License v2.0 - see the LICENSE file for details