Glad it’s of interest to at least one other person! So my introduction to practical Bayesian problems was JAGS, then PyMC3 and a bit of STAN. That gave me the impression that I had a reasonable overview of DAGs in the probabilistic programming space - but coming across these discrete Bayes Nets threw me a bit.
Some of my academic work is taking me down the road of reading about causal inference and errors in reasoning (like collider bias). All that reading (so far) seems to focus on discrete Bayes Nets and a totally different set of software tools. There seems to be BayesiaLab and AgenaRisk which seem very full featured, but are expensive and GUI-based. If I ever moved from academia into data science/decision support, then using these would probably be pretty good to be honest. Free and code-based packages include pomegranate, pgmpy, and bnlearn, and probably more. There’s also bnlearn for R (different developer) which might be the most developed. I’ve got the book on order. And there’s BayesNets.jl for Julia which looks cool, but the docs are a bit minimal.
I was basically trying to work out how much overlap there was, how much I could generalise what I know from JAGS/PyMC3 to these Bayes Nets. It seems, with help from @junpenglao, that there is no extra set of concepts as such, just different implementation details, with a focus on exact solutions. So it looks like the intuitions about inference can generalise over.
For this particular thread of work, I probably will move forward with pomegranate, or bnlearn, to be honest. Being built for the task, they have some nice features, e.g. it’s nice that you can explicitly name the levels of your categorical factors \mathrm{COVID~severity} = \{ \mathrm{mild, moderate, severe} \} and you can easily write down conditional probability tables.
It seems like there is way more development effort/people who focus on PyMC3, STAN, type frameworks than the Bayes Net packages. They seem more polished, with more complete docs. But hopefully I’ll get up to speed with the Bayes Net packages soon.