By Vlad Feinberg - September 3, 2021
Note: This week, we changed the format a little bit, covering a tutorial for a package, rather than a paper.
Vowpal Wabbit (VW) is a fast, open source machine learning (ML) library.
VW is extremely useful as a baseline for us at Sisu because it is one of the few open source approaches to fitting linear (and polynomial) models which supports the scale and parallelism for our data sets.
The focus of this tutorial will be purely on the high-level basics of working with VW in a supervised ML setting and mostly about parallelizing its training procedure.
Lots of great resources on VW already exist to cover scope outside of this:
VW is a very generic and flexible system for reinforcement learning, recommendations, learning-to-search-based structured output prediction, active learning, and topic modelling.
There are lots of tutorials out there for regular and advanced VW usage, but it's tougher to find a fully worked example of using VW's multiprocessing capabilities. See the full tutorial here.
If you like applying these kinds of methods to practical ML problems, join our team.