bob-carpenter
commented
2 years ago @spinkney and @avehtari: Is there a set of changes you suggest? Otherwise, this is more of a to-do item than a concrete issue that someone other than one of you can act on. I'll help you get started.
- pointing at our dev page rather than listing a few of us by name or trying to count numbers
- updating latest stable releases [ongoing]
- changing the "written in" part from C++ to reflect interfaces
- generalizing platform beyond x86 because we work with all sorts of things
- listing the various sub-projects in R and Python like rstan, rstanarm, brms, pystan, etc. with links (if allowed)
- list secondary packages like Prophet, Orbit, and Torsten in a separate section to make it clear they're not officially supported prodcuts
- change BBVI to ADVI (ADVI is a kind of BBVI, but there's also a specific system BBVI that we don't use)
- add that we also implement forward-mode autodiff and can autodiff all kinds of great stuff from HMMs to ODEs to algebraic solvers; link that to the arXiv paper on how we do reverse mode
- blavaan appears to be an R package using Stan---there are a lot of those and I don't know how many we want to mention or if we just want to point to CRAN (or if they'll let us)
- add links for all the sub-packages into our doc or to external home pages (e.g., Torsten)
I have no idea how to do this with enough secondary citations to not have it all backed out. Every time I try to change more than a typo on Wikipedia it gets backed out. It took years before they'd even let us have a page because there wasn't a secondary source that talked about Stan rather than using Stan. Unlike good scholars, Wikipedia hates primary sources.
For comparison, see the other weak pages for PPLs:
The top-level PPL page could be cleaned up and filled out with links where they are missing for Stan. I think it's fair they say we extend from BUGS, because that's where I started with the language design. In the end, we got to a very different place, and arguably PPLs like Pyro and PyMC3 are more BUGS-like in that they're restricted to explicit directed graphical models. For me, that's the big cut---are you a programming language (like Stan or Church) or a graph specification language (like BUGS or PyMC3)?
It looks like Turing.jl doesn't even have a line on the PPL Wikipedia page, though they mention it in-text. Wikipedia made if very hard for us to publish a page on Stan, so I'm not surprised most of the other PPLs don't have Wikipedia pages.
spinkney
andrewgelman
This was brought up by Aki
The page is https://en.wikipedia.org/wiki/Stan_(software)