The next Math/Stan/CmdStan/stanc3 releases are about five weeks away - the feature freeze will happen on the 7th of November, and the release ten days later. I would like to get the release out well before the Thanksgiving holidays.
Please use this thread to let us know if there are any pull requests that need more attention or if you have any other suggestions regarding the release.
To start the thread, the two PRs that require my attention are two @spinkneyās PRs that I neglected to assist on:
I have two easy docs PRs open:
And a small stanc PR on some error messages:
Iād also like it if the changes to get_dims and get_param_names were included:
I believe the Stan part of this was waiting on a clarification from @bgoodri
Besides those 2 math prs, I believe we should get in
What do you think @WardBrian?
@spinkney I still have a hard time wrapping my head around the best way to encapsulate the _qf signatures. Iām currently extrapolating from only one data point since we donāt have a second one ready to go. Iām happy to work with you on it though, for starters we could just expose it was a ānormalā vectorized function if you can list out all the signatures
@rok_cesnovar the next version is 2.31.0, right? The title is a typo?
Indeed, fixed.
std_normal_log_qf is in.
Iām not sure if @andrjohns has the time, but I would also really like to see Update log_prob method to take constrained_params in init format by andrjohns Ā· Pull Request #1108 Ā· stan-dev/cmdstan Ā· GitHub before the release, so people donāt start depending on the current way it is implemented
Oh good call! Iāll update that tomorrow. Iām just working on some more comprehensive tests to make sure that the method works across matrices and arrays of containers.
The System libs PR for the Math library is just about ready to be re-opened, the only tricky point is that the Sundials library version in Stan is updated past whatās available in distribution package managers (Stan uses 6.1.1, but the most recent in debian/ubuntu is 5.8.0). There were backwards-incompatible changes in the Math code (Sundials 6.0+ uses a different ācontextā initialisation), so Iām currently looking at whether I can detect the sundials version from the provided directory/library and #ifdef the relevant parts of the Math code
Awesome! If you canāt automatically detect what version, I think it would be reasonable to do what is done for TBB and require that the end-user set up a make variable when they set up the SUNDIALS_LIB etc. For TBB we call it TBB_INTERFACE_NEW, so maybe SUNDIALS_INTERFACE_OLD?
Oh I hadnāt thought of that, makes sense to me!
adds varmat for log_inv_logit by SteveBronder Ā· Pull Request #2806 Ā· stan-dev/math Ā· GitHub would also be good to get in since it fixes a known bug @rok_cesnovar
@rok_cesnovar Iāve got time today and can review that PR if it would help
Would appreciate it a ton!
Two of @andrjohns PRs are ready for reviews
I can do the Bernoulli one. @rok_cesnovar or @stevebronder could you do the ternary vectorization?
Does the ternary vectorization PR add varmat support @andrjohns?
I can do the ternary one
Iām re-opening the system-libraries PR in the Math lib today, then Iāll also be opening PRās later this week to add the signatures and manual entries for the new hypergeometric functions:
hypergeometric_2F1hypergeometric_2F2hypergeometric_3F2I have a few more hypergeometric functions that Iāll be adding, but wonāt be ready in time for this release (@spinkney is there anything that you were particularly hoping for your functions? I have the 1F1 just about finished which could also go in this week)
@stevebronder the bernoulli one needs someone to look over the opencl code that reimplements the select function.
@rok_cesnovar I think the weibull pr needs that ^ select function implemented in opencl.
@rok_cesnovar For atan2 I commented out that binary vectorization test as the finite differences fails at infinite inputs. If you think thatās ok, then the pr is ready.
@andrjohns For the hypergeometric functions, that looks good!
A few more wish-list items the week before the freeze:
Documentation PRs:
Stanc (small) PRs:
@charlesm93 and I are also working on exposing the variadic forms of the algebraic solvers, which I think will be ready before the end of this week and would be nice to have, but can also wait a release if it gets too crammed.
Oh, and since it is testing related I guess the freeze doesnāt apply, but I would like to get Testing: Prioritize changed unit tests by WardBrian Ā· Pull Request #2832 Ā· stan-dev/math Ā· GitHub merged. Itās a very simple change to the test script which allows us to run the tests we think changed earlier in the pipeline, to āfail fastā as it were.