During StanCon there will be time for planned and ad-hoc Stan developer meetings on Tuesday and Wednesday. This will be managed by having a longer lunch break between the tutorials and some reserved rooms. Idea is that you have possibility to have some face-to-face discussion with people you don’t often meet live and maybe even do some code review and pair programming (which are definitely easier when in the same room). Based on the previous StanCons it can be quite busy during the actual conference, so if you haven’t finalized your travel plans, I suggest to arrive so that you can take benefit of the opportunity of live meetings on Tuesday and Wednesday. Most of the meetings are assumed to be quite small scale with anyone you want to meet (just contact them in your preferred way), but anyone is welcome to suggest also more planned meetings (e.g. by posting to this thread).
As a StanCon organizer I’ll chime in.
My goal is to have 2 hrs/day for developer meetings during tutorials. This will NOT overlap the tutorials.
During the conference I am aiming for at least 1 hr/day of developer meeting time but I am more constrained and cannot promise it.
If you want access to the tutorial food/lunch please register for tutorials. That is how we pay for the food.
thanks
Breckl
Some topics I’d like to discuss live and anyone interested in these is welcome to join
- bayesplot for visual convergence and model diagnostics (at least with @jonah, @paul.buerkner)
- loo (at least with @jonah, @paul.buerkner, @mans_magnusson)
- projpred (at least with @jonah, @paul.buerkner, @bgoodri)
- Laplace for GLVMs (at least with @charlesm93, @stevebronder, @bgoodri, @mitzimorris)
- sparse matrices (at least with @stevebronder, @bgoodri, @seantalts)
EDIT: accidentally posted too early
Count me in.
Do we want to have these developer meetings on the same day or should we split them up across the conference?
I also wouldn’t mind having one for stan-math performance. i.e. cpu/gpu parallelism and Eigen/math optimizations
A meeting on the cpu parallel things with a discussion on changes to the reverse ad system would be good, I think.
I am around from Wednesday just before lunch.
I’ll be available to video in. Just keep me up to date
Sparse matrices sound fun! @stevebronder was talking about writing up a design doc for how the syntax would look and how it would work. I think there’s some existing work from @anon75146577 here (which possibly supercedes the wiki). Are those the most up to date links?
I’d like to discuss the KINSOL solver with @yizhang and @wds15. Specifically, what features do we make available to the user – KINSOL offers a lot, and sometimes you need fine-tuning to get the optimal performance. I’ll draft a proposal.
Yes! Going to start working on the proposal today. I’m lucky that Dan and Aki have already paved the way for a lot of these things
These two topics overlap and I’d suggest we also join the sparse matrices discussion @avehtari mentioned.
To echo what @jonah mentioned two weeks ago, I wonder what activities can be arranged for people who won’t be in dev meeting, 2 hours is a long gap in the middle of the day.
In Helsinki, many people attending the tutorials wished for longer breaks without any organized activities so that they can either experiment with what they learned in the morning or rest so that they are able to learn more in the afternoon. Thus this 2 hour break is intentional and it’s just the developers who have to suffer and manage without a longer break.
Yay! Im in!
I offered to help Aki with arranging these. Below are polls for each developer block mentioned. If you are interested in attending any of the developer meetings leave the dates you are available so we can get space accordingly!
If you can attend any day feel free to check-off all 4 days
@breckbaldwin once we have the dates that are best for everyone who do we contact to either see which rooms are free or reserve rooms?
Also, I kind of like the idea of making these meetings open. It could be a good opportunity to get new developers!
Thinking about this more, how would people feel about a “new devs” block? I kind of like the idea of collecting a bunch of small PRs across our packages and helping setup people who are interested in contributing.
Bayesplot
- Tuesday (August 20)
- Wednesday (August 21)
- Thursday (August 22)
- Friday (August 23)
0 voters
loo
- Tuesday (August 20)
- Wednesday (August 21)
- Thursday (August 22)
- Friday (August 23)
0 voters
projpred
- Tuesday (August 20)
- Wednesday (August 21)
- Thursday (August 22)
- Friday (August 23)
0 voters
Laplace for GLVMs (and Kinsol Solver)
- Tuesday (August 20)
- Wednesday (August 21)
- Thursday (August 22)
- Friday (August 23)
0 voters
Sparse Matrices
- Tuesday (August 20)
- Wednesday (August 21)
- Thursday (August 22)
- Friday (August 23)
0 voters
Parallelization / Optimization
- Tuesday (August 20)
- Wednesday (August 21)
- Thursday (August 22)
- Friday (August 23)
0 voters
tidyverse had dev day with Hadley ringing a gong every time a PR was merged
Yes my idea is v much inspired by that! They had a new dev event at a Chicago SatRday that went very well!
Surely Cambridge has a gong around somewhere we can borrow…
One thing I’d like to discuss wrt this is our memory model. I made a branch of stan-math which has a C++20 Allocator concept compliant allocator in it. (mostly stolen inspired by Eigen
and Facebook’s folly
). So instead of using raw pointers like vari**
we can use something like
std::vector<vari, chainable_allocator<vari>> adjARef_;
I think we can take this idea a step further by creating a pool of stack allocators, where we have a number of allocators equal to the number of threads available. I think we could do this by using the Bloomberg style of dynamic polymorphic allocator in the initializer. So something like
std::vector<vari> adjARef_;
// In the initialization list
adjARef_(A.vi(), pooled_stack_allocators<vari>)
Then the pooled_stack_allocator
gives a stack allocator for vari
to adjARef_
.
I think that makes sense? Though we should probably discuss this somewhere else
I believe there are plenty of rooms. I have scheduled 2hrs of dev time Tues/Wed, 1 hr on Friday. I was thinking of a general community meeting Thursday for an hour as well. Schedules up today/tomorrow.
Breck
We spoke yesterday about have a short discussion about the posterior database as well. I do not know where that would fit. @avehtari may have an idea?