#13 Building a Probabilistic Programming Framework in Julia, with Chad Scherrer
Learning Bayesian Statistics
00:00
SaaS - The Lightweight Way of Doing Probabilistic Programming
The idea with having the observed data not specified in the model was a matter of I have this dependency graph, and it has this cycle. And how do I get rid of the cycle? So fighting and fighting with that. By the way, when I later learned about the work that MIT team is doing on Jen, it turns out that they're taking a similar approach. They also do not specify the observed data as part of the model. So I hope we'll see other PPLs taking this approach, because I found it to be a huge benefit. It allows you to have all these nice benefits about forward sampling and posterior predictive checks.
Transcript
Play full episode
