Advance Bayesian Modelling with PyMC3


Dear all,

Last month I did a 2 days workshop in the Czech Republic (hosted by CEAi). They prepared professional video recording, which I would like to share here. This is a high-level PyMC3 workshop, as the attendees had already work through Introduction to Probabilistic programming (with PyMC3), which is built on top of the tutorials by @fonnesbeck. But the workshop also covers all the basics in depth. Due to the time limit, I did not manage to present more case studies - it would definitely be something to improve upon for next time :slight_smile:

The code and slide could be found below:

Video content

Session 1: Probabilistic thinking: generative model and likelihood computation
Session 2: Likelihood in PyMC3 and model reparameterization
Session 3: Model parameterization and coordinate system: Neal’s funnel
Session 4: Bayesian modelling and inference with MCMC in PyMC3
Session 5: Model evaluation and model comparison
Session 6: Case study: modelling multivariate observation
Session 7: Mixing MCMC samplers: Compound step in PyMC3

Fitting multiple measurements, shape issues in sample_ppc

This is awesome, Thank you Junpenglao!


This is very rich. Thank you for sharing.


This is awesome! Thank you for sharing!


Wow, this is phenomenal. Way to kill my weekend ^______^.


thank you sir!


Wow, amazing stuff, @junpenglao! Thank you for sharing with us :slight_smile:


This is so good, well done!


Lots of content. Thank you!


Wow, thank you @junpenglao! Exactly the type of resources I was looking for :slight_smile:
Can’t wait to go for it, and hopefully I’ll be able to give you a constructive feedback after it :wink:


A small feedback, having read the notebooks and watched the videos: it’s a really useful ressource and I highly recommend it to anyone who’s already used PyMC3 and is looking for a way to undestand what’s going on under the hood (I would not recommend it for beginners though. Take a look at Chris Fonnesbeck’s tutorial at PyCon 2017 first).
@junpenglao really goes into the weeds of the library, explaining in details how you generate models, compute likelihoods, do inference and evaluate/debug your models. It gives you an understanding of how PyMC3 works and more generally teaches you core concepts of Bayesian modeling. Well done and thank you for sharing!!

On a more practical note, I would recommend reading the notebooks before watching the videos, to get an intuition of the topics and make the most out of the lessons.
I would say sessions 2, 4 and 5 are must-sees, with session 3 perhaps being the less practically useful (at least to me) because of its complexity - interesting for general knowledge though.
Hope this feedback will help some, and thanks again for this amazing content JunpengLao!


Wow thanks for the kind words Alex! I am really glad that you find it helpful!

Also, I wrote a developer guide of pymc3 largely based on my experience of preparing for this workshop, you might also find it helpful


Great, thanks, gonna take a look at that :wink: