Examples

Here are some examples of using vowpal-wabbit

• RCV1 example. Here we use online gradient descent on TFIDF transformed features while taking advantage of a cache file.
• Malicious URL example. A sequential binary classification problem with temporally correlated data.
• VW has a test suite. less +/^__DATA test/RunTests provides many little examples with input data-sets and expected reference outputs.
• Matrix factorization example
• Multi Class Examples
  • One Against All (oaa) multi class example
  • Cost Sensitive One Against All (csoaa) multi class example
  • Error Correcting Tournament (ect) multi class example
• Weighted All Pairs (wap) multi class example
• Truncated gradient descent example
• Contextual Bandit Example
• daemon example - how to run vw in daemon mode and get predictions

Utilities

• Using active_interactor.py an interactive way to interface with vw's active learning
• Using vw-hypersearch a vw wrapper to search for "optimal" (argmin) parameters
• using vw varinfo - a vw wrapper to inspect your training-set & model
• Using vw regr - a script for aiding regression applications
• vw-top-errors: online learning debugging for better models

Demos

The source tree has a sub-directory demo which includes somewhat more involved examples:

• DNA splice Site recognition dataset from the 2008 Pascal Large Scale Learning Challenge
• Using Searn for Entity Relation Recognition
• MNIST handwritten digit recognition with --nn
• low-ranked quadratic --lrq demo on the movielens data-set
• --normalized effect on SGD online learning
• Ability of vw to separate signal from artificially injected noise

External sites examples:

• Zygmunt Zając's FASTML blog several examples of using Vowpal Wabbit in practice.