Solving the multi armed bandit problem
Using A/B testing has become a common practice to find out winning variance. But every-time you use it you are inefficient and you loose for the bad ones. Also winning is not guaranteed to last for long. Trends change, customers change. Ideally you have to be in a constant flow of testing, and trying the new thing, with minimum loss. Let’s not forget to also take personalisation along for a ride, and mix in some recommender systems, and see what we got. It is going to be a bumpy ride. How we solve the multi armed bandit problem is going to be the answer, or is it?
Claudia Vicol is the Lead of Platform and Data Engineering teams at Marktplaats.nl, part of eBay Classifieds Group. She has been a software engineer for more than 15 years, with experience ranging from banking industry to e-commerce, and her current favourite language is Scala. My main focus is to provide runtime , stability and resilience capabilities to all the other teams inside my organisation from tooling to infrastructure and pipelines for fast experimentation. With data engineering we structure data, we stream it, we distribute it, we run machine learning on it and connect it back to t