You've heard about bayesian statistics, and most of the tutorials kinda make sense but it still hasn't "clicked". This is an application oriented, code first, no calculus required construction of bayesian statistics from the ground up. It culminates in developing a simple MCMC implementation. You will leave the tutorial with a rich understanding of bayesian statistics and MCMC.
You've seen the articles that say "MCMC is easy! Read this!" and by the end of the article you're still left scratching your head. Maybe after reading that article you get what MCMC is doing... but you're still left scratching your head. Why? Why do you need to do MCMC to do Bayesian Statistics? Why are there so many different types of MCMC? Why does MCMC take so long? In this workshop, we will build our own (very simple) bayesian analysis software as an exercise in understanding what the leading tools are doing under the hood (PyMC3, Stan, BUGS, etc). We will build from a quick overview of the basics of bayesian statistics, make a naive brute force analysis, and then optimize it by developing an MCMC algorithm.
We will cover the following: Probability and Bayes Theroem Infinite Hypotheses- Estimating number of locals attending PyData NYC Introduction to motivating problem- Finding an optimum price Manual evaluation of hypotheses Dealing with underflow: Living in Log Land Scaling- Speed and hypothesis evaluation MCMC as a scaling solution Programming our first MCMC algorithm * Where does our solution break where others (Stan, PyMC3) succeed?
This tutorial will be a guided tour of Bayesian Statistics through a Jupyter notebook in Python 3. Participants should download the notebook prior to the tutorial so they can hit the ground running.