LIBOCAS - Library implementing OCAS solver
for training linear SVM classifiers from large-scale data

Vojtech Franc, Soeren Sonnenburg

Last Modified: 23-Jul-2012

Introduction

The library implements Optimized Cutting Plane Algorithm (OCAS) for training linear SVM classifiers from large-scale data. The computational effort of OCAS scales with O(m log m) where m is the sample size. In an extensive empirical evaluation OCAS significantly outperforms current state of the art SVM solvers, like SVM^light, SVM^perf and BMRM, achieving speedups of over 1,000 on some datasets over SVM^light and 20 over SVM^perf, while obtaining the same precise Support Vector solution. OCAS even in the early optimization steps shows often faster convergence than the so far in this domain prevailing approximative methods SGD and Pegasos. Effectively parallelizing OCAS we were able to train on a dataset of size 15 million examples (itself about 32GB in size) in just 671 seconds --- a competing string kernel SVM required 97,484 seconds to train on 10 million examples sub-sampled from this dataset.

The library also implements COFFIN framework for efficient training of translation invariant image classifiers from virtual examples. As an example, we used the implemented framework to train a linear SVM on a gender classification dataset of almost 5 million images on a plain notebook with just 4GB of memory.

Features

Problem formulation

OCAS solver is currently implemented for training binary (two-class) and multi-class SVM classifiers:

1. Binary classification problem: OCAS solves the following unconstrained convex optimization task

where is the regularization constant and are training examples. The result is the parameter vector of the linear rule which assigns vector a class label +1 and/or -1.

2. Multi-class classification probem: OCAS solves the following unconstrained convex optimization task

where is the regularization constant, is a finite set of labels and are training examples. The result is a set of parameter vectors of the linear rule which assigns vector a class label from .

Download

LIBOCAS can be downloaded from here:

Change Log

Platforms

GNU/Linux.

Licensing Information

LIBOCAS is licensed under the GPL version 3.

References