Stan Meeting August 23

<sorry, started recording late, missed most of Bob and Mitzi.>

Bob: Worked on functional Spec for low level IO for services level, will be posting to discourse. Other things that I didn’t catch.

Breck: Stan govenance voting is nearly over. Electorate is “Stan Developer’s Only” in discourse. 20 ‘yes’, 0 ‘no’ so far.

Sean: Worked with Bob on I/O functional Spec.

Ben Bales: Writing yet another automated test framework (YAATF), automate a reverse mode specialization???.

Sebastian: Handled some pull requests for MPI, tested make file refactor that Daniel Lee built, seems to work.

Jonah: Workign on two talks at StanCon. Dan S and Jonah getting up to speed on the Royal Statistical Society paper, getting ready for Stan class in Italy. Dealing with some CRAN issues for RStan.

Ben Goodrich: Fixed parser bugs, still working to get RSTAN released. Tests are failing for exposed functions (_lp functions). Kzrystof is helping. Issues with accumulator.

Aki: PyStan course yesterday in Copenhagen. Yuling will be needing help with unit test for . Updated RSquared case study, new Rhat effective sample size case study. Got t-shirts, 270 registrations.

Andrew: Completed 3 grant proposals, two chapters of new Stan book sent to Bob.

Brian: Stanfeather.jl, working with Rob on StanLoo.jl, looking at BRMS for StanArm.jl.

Paul: Working on tutorial paper oriented towards ecologists. Working with Aki on the RHat case study. Bunch of small features, bug fixes in BRMS.

Rob: Finiished up GPs in RMarkdown with Jonah for course. Wrote tests for bivariateNormalIntegral??, matches matlab to float precision.

Matthijs: GLM pull request is done in stan language. Too slow currently, something going on with operands and partials. Ben Bales had something to say…

Lauren: StanLadies survey want out, 15 responses. Do a lunch on Wed. Doing education stuff as well.

Daniel Lee: Trying to hunt down how to the make the make files work better. Got messed up a bit after getting MPI intergrated.

General Agenda:

Aki: Changes to RHat and effective sample size. What is the decision making process to make changes to defaults?
Bob: This will not change interfaces.
Aki:RHat still one number and will keep same name, ESS may change name to relative effenciency.
Andrew: Posterior Standard deviation different with MAD (Medain Average Deviation).
Bob: Andrew/Ben G/Aki are the ones to decide this. We evaluate a lot of event probabilities.
Ben G: Have been using median and MAD in RStanArm.
Andrew: The two cases, disccrete random variables and Chauchy-like.
Aki: will make another proposal based our conversation.

Aki: StanCon distribution across countries discussed.

Andrew: Bayes for big data. One of these days we need to figure out a coherent appraoch to big data.

Lauren: facebook grants. Doing something with Dan Simpson.

Sorry, maybe off-topic: where’s the stan class in Italy? (personal interest)