Vowpal Wabbit Modules in AzureML

This post is authored by Sudarshan Raghunathan, Principal Development Lead for modules in the Microsoft Azure ML Studio team based in Cambridge, MA. In his blog post last month, John Langford wrote about the open source Vowpal Wabbit (VW) machine learning (ML) system. He highlighted some of the main advantages of VW, e.g. its performance and ability to handle large sparse datasets, which make it particularly popular both within and outside Microsoft for applications such as sentiment analysis and recommendation systems. When we initially released the public preview of Azure ML in July this year, we exposed a small subset of VW functionality as part of our Feature Hashing module. The latter transforms datasets with text features into binary using the feature hashing algorithm (Murmur hash) implemented in VW.