Introduction to Bayesian Inference using PyStan
Despite the promises of big data, inference is often limited not by the size of the data, but rather by its systematic structure. Only by carefully modelling this structure can we take full advantage of the data: big data must be complemented with big models and algorithms which can fit them. Stan is a platform for facilitating this modelling, providing an expressive modelling language for specifying bespoke models and implementing state-of-the-art algorithms to draw subsequent Bayesian inferences. The course will teach participants how to interface with Stan through Python.
Online | February 1, 2021
- Introduction to Bayesian inference: An overview of the main concepts and the underlying philosophy of the Bayesian paradigm.
- Markov chain Monte Carlo methods: A brief overview and motivation of Markov chain Monte Carlo methods for Bayesian computation and Hamiltonian Monte Carlo.
- The Stan language: An outline of the main components of a Stan program.
- Using PyStan: A guide to interfacing with Stan through Python.
- Examples: Linear regression, Poisson regression, hierarchical models and mixture models.
By the end of this course participants will:
- Have developed an intuitive understanding of Bayesian inference and MCMC algorithms.
- Understand how to apply Bayesian methods to fit models to their own data.
- Be able to construct an efficient Stan programme to perform statistical inference.
- Know how to interface with Stan through Python; using Python to graphically and numerically assess Stan’s output.
This course will make use of more formal lecture sessions, followed by more free-flow and flexible practical sessions. Attendees are encouraged to ask questions throughout, and to put their new knowledge to use in the accompanying practical sessions.
Prior to attending this course participants should be familiar with basic concepts of probability and statistics, including common probability distributions and regression methods. Attendees should have basic Python programming knowledge, and have some familiarity with the popular numpy, pandas and matplotlib packages. We do not expect attendees to have experience with Bayesian inference or Stan, but some knowledge of the former will be helpful.
The practicals were well structured and demonstrated lots of the capabilities. The Notes had plenty of details on the basics of the language and were a useful reference guide.I really liked the pace and the level of detail in explaining the background on day 1. I had very little knowledge of Bayesian inference before starting the course, and I don’t think this course made me an expert on the subject. However it is a good starting point and succeeds in delivering a conceptual understanding of both Bayesian Modelling, the algorithms used by Stan and in getting started with the Stan language. The friendly pace and explanations were excellent and made someone at a beginner level appreciate the course.I liked that the solutions were given together with the exercises in case I was stuck and wanted an answer straight away.