Stan Meeting Aug 16, 2018

Individual Reports:

Rob Trangucci: Working hard <sorry, just spoke too quickly to capture>.

Lauren: Working on StanCon, education, IRB application done.

Brian: JuliaStan inteface updated. Two packages are going to be relased, StanMamba, StanDataFrames. Working on StanFeather, a file format to save draws.

Ben Bales: Working on bunch of small pull requests.

Aki: StanCon, going to record all sessions, StanCon is getting bigger. Working with Yuling who is in Helsinki, working on cross validation paper, working on Rhat convergence diagnostics/case study, hopefully get it out before StanCon, working on “Regression and Other Stories” book.

Ben Goodrich: Met with Rob, Bivariate normal work, estimateing GP models with Jennifer, manged to get Rstan working with 2.18, not yet released. Having trouble with downstream packages on CRAN (there are 50 of them), try to work on parser output. May not be able to do a 2.18, will have to wait until 2.19 due to parser error outputs not matching regexes in dependent packages. There may be an easy fix with “WARNING” prefix addition.

Jonah: Working on course materials for Andrew’s book. Working on Aki’s Bayes Plot requests. Rstan tools need to be re-released. Working on failing tests, low level stuff with various Rpackages.

Paul: Minor issues on BRMS, working on a case study for Aki/StanCon.

Sean: Making performance tests more robust, working with Bob on tutorial on C++ development for StanCon, some IO stuff, code review on GPU pull requests.

Sebastian: Can’t make it to StanCon, have job responsibilities. Has been working on ??? threading, there is enthusiasm and uptake for MPI in his organization. Wondering whether to add makefile fixes in 2.19 or wait later. Sean: not sure what the plan is for when to release 2.19. Some problem with array parsing. <poor audio, hard to take notes>

Breck: Govenance proposal is about to be posted to “Developers Only” discourse group for vote. Also worked on proposals for Andrew.

Bob: Posted serializion (paramater packs Parameter packs for (de)serialization) design on Discourse, working on proposal with Andrew, working on course for Helsinki.

Mitzi: Writing more tests for ??code generator??. Suggests that the tuples branch be reverted since there is broken code in there.

Matthijs: Just got to New York, waiting for a pull request to be approved that got garbled.

General Discussion

Candidate Visualizations
Aki: Asks the general group what they prefer for posterior visualizations. Is it easy to visualize densities where heights are laid out, Aki likes option C), Andrew does not like B). A) is accurate, B) is a compromise.

Jonah: There might not be just one answer. C) does not use a log which can be negative, sqrt instead.
Bob: Plots not on same scale.
Jonah: Good for set of params on the same intecept with similar scales??? B) is what is currently is, the function is: bayesplot::mcmc_areas().
Aki: Seems like there are no strong objections, looks like Jonah is going to just make them all available.
Ben Bales: requested that the axes be labeled.
Lauren: asked if the white space minimized?
Jonah: responded that it is hard to do.

More on JuliaStan
Brian: Trying to move more of R packages into Julia. Jonah will be helping. Bob suggested that Brian look at BRMS. Will JuliaStan have pre-compiled models?

Downloads, metrics of adoption
Bob: Looking at downloads on RStudio, RStan has passed RJAGS. BRMS and RStanArm have similar count downloads.stan-downloads.pdf (8.2 KB)

Where to put meeting notes
Breck: Where do people want meeting notes posted? Nobody cares, I’ll post to General on discourse.