Abstract: Many introductory tutorials on Bayesian inference are not as simple as the authors purport them to be. In this workshop, we provide intuitive and straightforward marketing examples, using the open source probabilistic programming language Stan, showing where traditional linear regression and even other machine learning methods fall short.
We will first show how a hierarchical model improves forecasting for situations where you have categories with a mix of counts including counts that are traditionally too low for statistical significance. We will use a pay per click example for this first tutorial. The second example will demonstrate how early stopping can reduce the cost and time for test-and-learn situations (aka marketing tests, experiments, etc.)
For both examples, we will walk through the code, and if wifi allows and attendees have pre-installed R, RStudio, and rstan, they can follow along.
Bio: Curt studied computer science at the University of Illinois at Urbana-Champaign and mathematics at the University of Minnesota. After building too many websites and client/server systems, he turned to data mining and statistics and never looked back. A good day is spent building models in R and Stan.