Planning the 2.33 release

[WardBrian]

Here is the plan @rok_cesnovar and I are proposing for the next release (2.33):

• Feature Freeze on August 21, 2023
• Release on September 5, 2023

The extended freeze period is for extra testing (this release features both tuples and pathfinder, big things we’re hoping will get a lot of stress testing during the release candidate phase) and to work around the Labor Day holidays.

Please use this thread if you have any thoughts on those dates, or if you need a pull request reviewed or think a bug fix is a high priority for this release.

[WardBrian]

To start off, I’d like to highlight:

• Expiring language deprecations: [Stan 2.33] Make expiring deprecations into errors by WardBrian · Pull Request #1287 · stan-dev/stanc3 · GitHub

  • This was delayed from 2.32 due to RStan being close to a CRAN release, which still has not happened. I’d be opposed to a further delay, because I think the nice shiny features in this release serve as a good reason for updating your models if you haven’t yet.

• Silencing warnings. From a discussion in makefile fix by mitzimorris · Pull Request #2919 · stan-dev/math · GitHub, @mitzimorris, @syclik and I proposed editing the makefiles as part of the release process to silence compiler warnings. This doesn’t really require a review so much as something to agree on so we can have @serban-nicusor update the files as he prepares the release tarballs.

• Docs issues: Tuples Document Tuple types in the language · Issue #656 · stan-dev/docs · GitHub, Pathfinder (nearly done) Feature/648 pathfinder docs by mitzimorris · Pull Request #651 · stan-dev/docs · GitHub

• Lower priority, but I’d still like someone to review Add argument to only output some parameters from `stansummary` by WardBrian · Pull Request #1143 · stan-dev/cmdstan · GitHub

[wds15]

CRAN is back online next Monday according to plan. This would hopefully allow for an in-time update of rstan to a version accepting the new array syntax… which I would really agree to should be the new default to make room for tuples.

Having Pathfinder land is great! How will Pathfinder be usable? I mean, do we get it as a standalone inference algorithm and/or as improved warmup procedure?

I am really looking forward to test Pathfinder with some ODE models, which are notoriously slow.

[WardBrian]

@wds15

How will Pathfinder be usable

It’s currently a standalone algorithm in CmdStan

[Franzi]

@WardBrian

ext release (2.33):

Hey @WardBrian, will the PR #2822 in the next Stan release? It is now open since Octobre and people are waiting for it and want to use the seven parameter diffusion model for their work. A not vectorised version was already finished in an earlier commit. Now that @stevebronder introduced vectorisation the mix test have some troubles, but I would need help to fix this.

[WardBrian]

It is too late for 2.33 at this time, but seems reasonable to have in 2.34 (~4 months from now)

[Franzi]

That would be nice. The PR is now open since October '22 and the function was already on the list for the 2.32 April release. So as it was not finished then, I thought it would automatically be on the list for the next release, now in September. This is part of my university project. Therefore, I have a little time pressure on this implementation being published.

As was asked in the beginning, I also want to implement the CDF and CCDF functions. I would do this in the same way as for the PDF functions, but for this, the PDF functions need to be accepted, such that there is only little work for you to accept those as well.

[ssp3nc3r]

Let’s remember this is an open source project.