This post contains materials for my PyMCon talk, Learning Bayesian Statistics with Pokemon GO.
PyMCon is a virtual conference for the Bayesian community. While organized by the PyMC3 developers, the conference exists for anyone interested in general-purpose Bayesian inference. Register for the (mostly async) conference on Eventbrite!
I used Pokemon GO to motivate questions that could be answered via Bayesian modeling.
In Pokemon GO, players can rarely encounter “shiny” Pokemon, or hatch other desirable Pokemon from eggs. The exact probabilities with which these happen are unknown. But by using Bayesian inference and PyMC3, we can model these probabilities.
In this beginner-level tutorial, we will introduce fundamental principles of Bayesian modeling, then ask questions about Pokemon GO. We will develop PyMC3 models that can help us answer these questions.
In the talk, I motivated and live coded three Bayesian models. The first two were on shiny rates (what was the probability of seeing a shiny in X event?),and the last was on egg hatch rates (did the probability of hatching a certain Pokemon change midway through an event?).
The code and data are available on Github. The notebooks are cleaned up versions of the ones that I live coded in the recorded video.
See Google Slides.
See my post on Bayesian inference resources.