Abstract

Over the past decade, contextual bandit algorithms have been gaining inpopularity due to their effectiveness and flexibility in solving sequentialdecision problems---from online advertising and finance to clinical trialdesign and personalized medicine. At the same time, there are, as of yet,surprisingly few options that enable researchers and practitioners to simulateand compare the wealth of new and existing bandit algorithms in a standardizedway. To help close this gap between analytical research and empiricalevaluation the current paper introduces the object-oriented R package"contextual": a user-friendly and, through its object-oriented structure,easily extensible framework that facilitates parallelized comparison ofcontextual and context-free bandit policies through both simulation and offlineanalysis.

 

Introduction (beta)

None

 

Conclusion (beta)

None