Hangouts Link: https://meet.google.com/gzm-wmum-pfm
Instructions: Ask to attend in the hangouts interface and someone should let you in in the first 10 minutes of the meeting. Email breck@mc-stan.org if you have problems or want to attend the physical meeting in New York City.
Meeting Agenda:
Please add your agenda items in replies.
Agenda item:
(FYI, I think/hope/guess that I found a straightforward way using the TBB to parallelise the forward and backward NUTS sweeps which has the potential to speedup things by 1.5x-2x for any Stan model)
I think it would be good to start setting the agenda at least a few days before the meeting so people can decide if they want to attend.
I’d like to add an agenda item to talk about a feature freeze for a week before a release. Last time we had a last minute rush to get something in and we ended up with several bugs in 2.20 that we discovered <24h later.
To be clear: I am not intending to reach a decision at the meeting about the licensing stuff. I am asking how we proceed process wise (one point of the process is to ensure all relevant stakeholders have their voice heard).
Sure… in way discussing the process itself is a point in its own… but let’s settle that. We can easily say “this will be a point for the next meeting”.
I was actually responding to Breck’s post in my email, I didn’t see yours. I was wondering if Breck would agree with and wouldn’t mind positing the meeting thread a few days earlier so we can add things to the agenda earlier.
I’ll just put them up two weeks in advance. Totally makes sense to do so.
Sorry for not figuring that out on my own
I missed the start of the meeting and then had to leave before the end and thus missed the TBB licensing stuff. Can someone give a short recap on that? Was there any lawyer feedback?
The licensing of the TBB was discussed. We decided that we absolutely need @syclik’s take on it. In order to facilitate the decision and to make the basis for the decision public I volounteered to summarise the email discussion we had on our wiki. You find that document here:
I took out the names of anyone in that document on purpose.
The last remaining questions are around what terms like “link” and “unitary” and “binary” mean in this context.
To me that has been answered with question 9 - so its just fine. The main restriction we gain from the Apache 2.0 license is in question 10 which says “You just can’t link together the Apache-licensed and GPLv2-licensed libraries into a unitary binary and distribute that unitary binary.”
Since the Intel TBB will be linked dynamically into Stan-math anyways, my understanding is that we never have a problem here.
I hope the summary helps to move forward.
Thank you!
My understanding based on the lawyers answers is that nothing changes for Stan or Stan Math as long as we link TBB dynamically (which is the plan). That is fantastic news.
If we were to link statically that would prevent the use of Stan or Stan Math in GPLv2-licensed projects that would want to redistribute their project & Stan (Math) in a single binary (using separate Stan binaries as external processes would still be fine). That is irrelevant in the TBB case but still good to know.
Very light meeting notes on the 1 min reports
Ben G: In Berlin giving talks. Met with Sebastian.
Aki: StanCon was great. Model database work is proceeding.
Sebastian: StanCon. OncoBayes going to be released to CRAN. PyStan is reportedly seg faulting but can’t reproduce.
Rok: Moving JSON parser for cmdStan. GLMs using TBB.
Andrew: Working on scalable MRP. Working on workflow on debugging. Working on Bayesian Work Flow Using Stan. Finished book with Jennifer and Aki.
Matthijs: Working on partial evaluator in compiler and passing test. Talking to Duco
Charles: StanCon, tutorial went ok. Lots of StanDev meetings. First ever StanCon football (soccer) game happened with 28 people.
Ben Bales: Just started this week. Working on U-turn stuff with Aki.
Jonah: Workshops went well. Working on releasing some RStanArm bug fixes.
Sean: Maybe organizing another dev meeting. Been working on new compiler.
Steve: C3 compiler working. Has a mock version sparse matrix type. Looking over comments.
Bob: Work on PR for higher order autodiff testing.
Breck: StanCon is over, went well overall.
Thanks. I’ll respond shortly.
Bump.
The wiki is updated with the latest round of answers around the Apache 2.0 licences matters. These are aimed at answering the remaining open questions.
I think this should enable us to settle the matter.