Hi, I’m just curious what’s the goals of the JuliaBayes · GitHub group, and it’s relationship to the existing The Turing Language · GitHub ecosystem (or other probabilistic programming efforts in Julia; I’m only aware of Gen.jl and Soss.jl in this domain).
Me too. It has some great ideas like the BayesianWorkflow package. In terms of PPLs, it looks like mostly Turing and RxInfer are still active.
Hey! Speaking as one of the early members, the point of the organisation is primarily an entry point for a community and package collection for Julia software for Bayesian inference which are not necessarily specific to a given PPL. There is not any centralisation efforts currently, mostly an opportunity to coordinate and communcicate, although how exactly the community evolves remains to be seen. If you want to be involved, let us know!
Thanks, so can I think of it as infrastructure and plumbing that could be generic across all PPLs in Julia? Will Gen.jl, Turing.jl, RxInfer.jl, etc use things in JuliaBayes as their upstream dependency? To be concrete, would AbstractMCMC.jl which currently lives in Turing’s GitHub organization eventually move to JuliaBayes? Maybe a README.md at the splash page for the org might help a bit answer questions such as mine JuliaBayes · GitHub, regarding what the impetus to create the organization was.
I’m also curious about some of the packages For example, I’m interested in GitHub - JuliaBayes/ProbabilityMeasures.jl: measures that integrate to 1 · GitHub, but there is also the older package GitHub - JuliaMath/MeasureTheory.jl: "Distributions" that might not add to one. · GitHub. (ok, I can read the readme that says the one in JuliaBayes is normalized to integrate to 1, but I think the older MeasureTheory.jl was designed to work with Soss.jl which is also a PPL, so…just curious).
None of these questions are meant to be a veiled criticism of starting a new organization, I’m just curious. I feel Julia could have much more impact in the PPL universe than it does, so I’m certainly sympathetic to efforts like this.
Re; involvement, thanks, I’m investigating implementing some samplers which would work with the AbstractMCMC.jl interface. Maybe when the time comes I could contribute somehow 
(Disclaimer: I’m the person who set up the JuliaBayes org, along with a few other people)
To echo @PTWaade, I don’t really think that there’s much of a ‘plan’ per se 
However, let me try to give what insight I can.
I do think that the Bayesian community in Julia could have had such an org a long time ago. There are many amazing packages out there, including in the pre-LLM era, that exist mainly in people’s personal GitHub accounts. While it’s always nice to have your work associated with you (!!), I think that there’s a general consensus that Julia packages are best if they live in an organisation where multiple people have access to it. This avoids a (sadly not uncommon) failure mode where the repo creator rides off into the sun and is never seen again The whole thing really came about because a few of us were trying to work on a package for generalised linear models (think brms), and as things grew the benefits of having a central organisation became more obvious.
That said, progress is naturally not as fast as it used to be on TuringLang. As some of you will know, I used to work on Turing full-time. But now I have a different full-time job, and so I can only really commit a much smaller fraction of my time and energy. And the same is true of some (but not all) other org members – we all have our day jobs to do. No doubt this is a familiar scenario to many of you!
That does, however, mean that it becomes even more important to work in a collaborative way. To that end, I’d like to emphasise that JuliaBayes isn’t meant to be an organisation that’s tied to a single PPL, or a single research group, or a closed team of people. If anybody wants to move their personal repos there, we’d be very happy to help you do so, and to give you the appropriate permissions to manage your repo. There is also the #bayes channel on Julia Slack and you are also welcome to message me on Slack if you want to chat more about it!
As usual, I agree entirely with what Penelope is saying Hmm maybe its worth to add that what is done with the packages developed by Turing is entirely up to their developers - I suppose something like AbstractMCMC could also fit well in JuliaBayes, but it is made by the Turing people and within the Turing ecosystem, so it fits perfectly well there I guess.
And yes - come join the Slack channel! Much excitement and much to do there !
Thanks @PTWaade and @penelopeysm for helping to provide more information about what the organization is up to. It helps to understand better than landscape here and figure out where people’s energy is starting go towards. Also it helps to know what packages are getting more “important”, since building with the future in mind means looking at their APIs, interfaces, types, etc. to know what to integrate with to be more effective for the ecosystem rather than a set of disconnected packages. I’ll check out the slack channel (I really like the slow paced, “forum” feel of Discourse, but should learn to check slack from time to time).
That was a good suggestion; we’ve added one!
Forum is very fine too At some point, one could consider newsletters or some lightweight community meetings; but one thing at a time.
Looking forward to meeting you!
Just adding a few things!
I’m interested in GitHub - JuliaBayes/ProbabilityMeasures.jl: measures that integrate to 1 · GitHub
This mostly started as an experiment to get a package that had GPU/Reactant compliant probability models. At the time I started this i didn’t realize there is a big update to MeasureBase.jl coming that should, to my understanding, make this effort duplicated so I intend to switch my effort to adding functionality to that package instead. I’d say the package did a good job of helping establish that this was generally a set of highly desired features across many people and so I’m hoping that MeasureBase will come to satisfy this need.
Re the general questions bout packages and to @penelopeysm broader points, I will eventually move my own sampling package ParallelMCMC.jl to this org so that if I ride off into the sunset myself, I know my effort won’t go to waste.