binarySMO

The code for online support vector machines, described in Relaxed Online SVMs for Spam Filtering is available for download. This code includes implementations of relaxed online SVMs, as well as several online active learning strategies.

The current version is binarySMO-v1.0.1. Download binarySMO-v1.0.1.tgz

The package contains source code in C++, documentation, sample data files, and several utility scripts for converting other data formats to binarySMO data format, and for evaluating results with accuracy, precision, recall, and area under ROC curve.

The documentation is also available here, online.

Note that this code is specifically designed to be used with binary feature valued data, with a linear kernel only. This is appropriate for in many spam and text classification applications with SVMs, and enables some handy optimizations. If you want to do more general purpose work with SVMs, you will likely prefer any one of the several excellent SVM packages availalbe online, including libsvm and SVM-light.

This code may be used freely for research purposes (only), and comes with no warranties of any kind.

Please contact us with any comments, questions, or bug reports: dsculley [at] cs (dot) tufts (dot) edu. Many thanks for your interest!