Hangouts Link: https://meet.google.com/gzm-wmum-pfm 3
Instructions: Ask to attend in the hangouts interface and someone should let you in in the first 10 minutes of the meeting. Email breck @ fbb2116@columbia.edu if you have problems or want to attend the physical meeting in New York City when they start again.
Please add your agenda items in replies.
I usually don’t respond to messages done by myself but one topic I want to bring up in the general meeting is about the new space of Probabilistic Graphical Modeling Languages like Pyro, Tensorflow-Probability, and Turing (in conjunction with Flux), ForneyLab.jl, PGM.jl, and BayesNets.jl and how could Stan help within this space or if there are Graph-based frameworks that we should support as Stan users around PGMs…
Stan Slack message volume has grown that big that free plan restrictions are harmful. Discuss possibilities for future
this sounds really interesting - unfortunately another meeting - Bob Carpenter also will be in that meeting, hence not at Stan one. if someone could summarize discussion and report here that would be awesome.
Sure no worries, I’ll make sure to do a summary of what we discussed.
Here’s a short summary of what we talked about in the meeting.
Jonah reported in about further MRP developments and that its almost ready to be tested by users
he’s just working on some more testing and documentation.
There were also some additions to Bayesplot that Aki had come up with that should be implemented soon.
Charles mentioned working more on laplace with autodiff and that it’s still a little hard to work through.
Also, we talked about the possibility of moving from Slack to other services due to the increased number of slack users that repeatedly post.
Though, ultimately it looks like to be an issue for the SGB,
Ben brought up the idea of how to implement reduce sum with a GPU and what standard to support i…e. Open CL or CUDA
Rok, Simon, and Steve were able to point to some resources that seemed to help him out to understand how to implement it at with regard to low-level Open CL and something
called Sycle.
In terms of the conversation on PGM languages, Louis mentioned the the paper he had on extending deep learning with stan with num-pyro
and Ryan mentioned that he had done some work with regards to generating a factor graph that produces a visualization with graph viz from a stan program and I thought that was really cool.
I mentioned some Julia frameworks where people are implementing or trying to implement Bayesian networks like
Though, Forneylab.jl is just a factor graph based approach to doing Markov Random Fields in a graphical framework but they claim automated Bayesian inference using factor graphs.
My apologies for AWOL on general meeting for a while. Any details on the discussion w.r.t slack? I’m not on stan slack so the only background I have is Aki’s post above. It’d also help to @SGB when the you’d like to replay certain issue. I just happen to be in this thread get some update on PGM, thanks to @mitzimorris, otherwise would have totally missed it.
Ah ok sorry about that @yizhang , my mistake, I’ll make sure to do the @SGB for stuff concerning them next time. In terms of the transfer of service from slack to another service. There was discussion of using Discord, Zulip (which is an open-source equivalent to slack), and a few others. The main two we talked about were Discord and Zulip though.
Does @SGB want to be active in the choice of slack replacement?
Feel free to pick an alternative. I was only addressing @bparbhu 's comment that “ultimately it looks like to be an issue for the SGB”, maybe I have misunderstood it?
Sorry about the confusion @yizhang, perhaps my words were in poor choice. I think what we were primarily concerned about and why the @SGB would need to be involved, was that there would be a platform shift in how Stan users talk about issues or report problems and that said platform might warrant some kind of monetary decision. So that’s really why I said that this would involve the @SGB.
Oh, I see. Personally I’m platform agnostic. If a paid service is deemed necessary I can add it to our agenda.
Something else for everyone to check out:
@yizhang we actually would like to add something to the SGB agenda wrt to Slack.
@avehtari has recently posted that in the first week of March we had 52 active members and 31 posting members which means that a free Slack for nonprofits would suffice for us for quite some time if we let it grow organically and if we were to use something else for Stancon.
What SGB would have to look at is applying here:
Not sure if we can apply as Stan or if we would have to go through NumFocus.
If we get denied for this we can then look at any free alternatives.
Added to coming Monday’s agenda. Thank you @rok_cesnovar @avehtari !
What was the result? The current Slack limit is hurting us as we’re losing relevant discussions. We should switch to Slack nonprofit discount scheme or switch to some other platform. I think right now we’re waiting @SGB to check the nonprofit discount option.
Hey Aki,
Apologies for the lack of feedback. We don’t seem to hold the right organizational status to be able to get this. I’ve put out a question at NumFOCUS to ask if there is a way to frame us in the right context. Without the nonprofit scheme it seems quite expensive.
So far, responses that I’ve had suggested other projects had the same problem and some of them have switched to Zulip. Zulip has explicit free plans for open-sourse projects. Should we explore and consider the switch?
Thanks for the update. From the other than Slack options, I think the Zulip would probably be the best. Zulip supports also LaTeX equations, which would be often useful.
Would you like me to try and set it up and apply for the free plan?