SpaceBandits: Deep Contextual Bandits Models made Simple

Approximate Bayesian inference with neural networks has demonstrated robust performance in numerous empirical studies of contextual bandits problems [Google Brain, 2018][Hubspot, 2018]. Isn't it time for a package that makes deep Bayesian contextual bandits models simple to use?

SpaceBandits takes the models developed and experimented on in the 2018 Google Brain research paper Deep Bayesian Bandits Showdown: An Empirical Comparison of Bayesian Deep Networks for Thompson Sampling and makes them easy to build, train, manage, and deploy.

Usage

The alpha release of SpaceBandits, v0.0.990, is the current version. To install the latest release, use pip:

pip install space-bandits

For a introduction to using the library and the online-learning API, check out the demo notebook.

A toy problem and offline-learning API walkthrough is available in the toy problem notebook

Development

SpaceBandits is currently in development. Please use it and report any issues you have so I can try to address them before the v0.1.0 release.

Contributors are welcomed! New models from the Google Brain paper or others might find a home here as well.