There are two ways to have a fast learning algorithm: (a) start with a slow algorithm and speed it up, or (b) build an intrinsically fast learning algorithm. This project is about approach (b), and it's reached a state where it may be useful to others as a platform for research and experimentation.
There are several optimization algorithms available with the baseline being sparse gradient descent (GD) on a loss function (several are available), The code should be easily usable. Its only external dependence is on the boost library, which is often installed by default.
There are several features that (in combination) can be powerful.
Many people have contributed to the project at this point. John Langford, Alekh Agarwal, Miroslav Dudik, Daniel Hsu, Nikos Karampatziakis, Olivier Chapelle, Paul Mineiro, Matt Hoffman, Jake Hofman, Sudarshan Lamkhede, Shubham Chopra, Ariel Faigon, Lihong Li, Gordon Rios, and Alex Strehl have all worked on VW. Many others have contributed via feature requests, bug reports, or bug patches.
VW is also a vehicle for advanced research. The first public version containing hashing, caching, and true online learning was released in 2007. Since then, many different algorithms and results have influenced its design, including:
Last edited by Matt Hoffman,