Advanced Stan Workshop

Actions Panel

Advanced Stan Workshop

Daniel Lee

When and where

Date and time

Location

To be announced

About this event

This workshop is geared toward Stan practitioners that want to gain a deeper understanding of what's going on under the hood. This will be an informal conversation geared towards addressing any practical problems that you might have. Here are some of the topics that will be discussed:

1. lp__. How to compute it, what the transforms are, why they're important, and what the "unconstrained" parameter space is.

2. Computational efficiency by considering the automatic differentiation stack. Stan uses reverse-mode automatic differentiation, but the speed gains are from eagerly evaluating the gradients. We'll get into how this happens.

3. Fill in the blank. Let's chat about whatever you need. We can talk about debugging Stan programs, validating assumptions and why violating assumptions degrades performance, the difference between computational efficiency and sampling efficiency (and which matters more), dealing with MCMC output, or something else.

Prerequisites: it'd make a lot more sense if you're currently using Stan. But feel free to join to hang out.