Start
What is πSvM?
πSvM is a parallel implementation of the
Support Vector Machine (SVM) algorithm that allows efficient
training and testing on a multiprocessor system. Some of
its main features are:
- Compatibility with the popular LibSVM
software. It uses the same input format and command line switches.
-
Runs on any cluster computer where the Message Passing
Interface (MPI) is available.
-
Achieves superlinear speedups on some large scale
datasets.
-
Supports training of all SVM formulations including
C-SVC, ν-SVC for classification and
ε-SVR , ν-SVR for regression.
Why use πSvM?
Are you an impatient
person? Then
πSvM is for you. If you have used
LibSVM
before then you should have no problems using
πSvM
since input data and command line switches are the same. The
only new options that can be given to
πSvM are
the maximum size of the working set (-o) and the new number of variables
entering the working set at each optimization step (-q).
Why make πSvM?
Sometime around July
2006 i got tired of waiting for long-running SVM jobs on
large-scale datasets i had to deal with in my projects.
Another motivation for making
πSvM was the
Kepler
Cluster. a supercomputer made available to the staff at the
department
of computer engineering.
What about the name?
The name
πSvM probably reflects a moment with
lack of inspiration and marketing skills in the life of the author.
In an afterthought the greek letter
π (pie) was chosen
because of the
parallel implementation. I guess it might also
be OS-dependent on what you get... applepie (yum!) or
microsoftpie (argh!) ;).
Last modified: Sun Dec 17 09:43:29 CET 2006
