File Exchange

image thumbnail

Fast Linear binary SVM classifier

version 1.2.0.0 (99.2 KB) by Sebastien PARIS
Fast implementation of Linear binary SVM via BLAS/OpenMP API

9 Downloads

Updated 27 Sep 2012

View License

LSVM v 1.0
---------
Fast Linear SVM binary solver toolbox such PEGASOS/LIBLINEAR.
This toolbox offers fast implementation via mex-files of the two most
popular Linear SVM algorithms for binary classification: PEGASOS [1] and LIBLINEAR [2].

This toolbox can use BLAS/OpenMP API for faster computation on multi-cores processor.
It accepts dense inputs in single/double precision.

For comparaison with [2] in binary case, this package requires less memory and is approximatively between 10% up to 50% faster. Ideal for Large-scale training in computer vision for example

Installation
------------

Run "mexme_lsvm.m" to compile mex-files.

Testing
-------

Run "test_lsvm.m" for demo

Online help by typing pegasos_train or cddcsvm_train in matlab prompt.

References :
----------
[1] S. Shalev-Shwartz, Y. Singer, and N. Srebro. "Pegasos: Primal estimated sub-GrAdient SOlver for SVM."
In Proc. ICML, 2007.
[2] Liblinear: http://www.csie.ntu.edu.tw/~cjlin/liblinear/

Comments and Ratings (3)

Venkat .... optimize your C be cross-validation

Venkat R

Hi Sebastien,
Thank you for sharing excellent software.
I am having training data of orders 9500 x 200000. Can you suggest some tips, if any on choice of algorithm/parameters.
I found cddcsvm_train, with C =5, B =1; better than PEGASOS.
But didn't know if any other choice of parameters-C or optimization technique can yield better results.

with regards,
Venkat

Tianyang Ma

Sebastien, thanks for sharing such a great toolbox!

Updates

1.2.0.0

- Fix a bug for single precision
- Fix a crash for large-scale data with OS64

1.1.0.0

-Cosmetic changes

MATLAB Release Compatibility
Created with R2009b
Compatible with any release
Platform Compatibility
Windows macOS Linux

html/