Stan Meeting Notes December 20


#1

Andrew: Splitting up Stan book further.

Matthijs: Preparing slides, some compiler work and discrete params work.

Breck: StanCon contracts in hand, working on Columbia funding stuff.

Bob: Going over compiler, learning ML. Met Andreas Mueller from SciKitLearn.
Writing book on probability theory with simulations, pull requests.

Sean: OCaml Stan compiler, code generation work. Bug fixes with threading.

Jonah: Pull request from Mons (Aki postdoc) on cross validation. Other PRs, met Mitzi for book stuff.

Lauren:Working on MRP paper.

Mitzi: Documentation issues, worked through PyCmdStan spec.

Sebastian: Bug in maprect being fixed with Sean. C++ 17 showing lots of promise for parallelization. ODE bug fixed.

Aki: Working on RHat, Eff Samp Size, looked at doing ADVI in python, comparing NUTS, Stan ADVI, and Python ADVI which uses Stan gradients and internals. Will be using the 200 + models.

Ben G: Busy with finals. A bit more work with Boost 1.6.9 with complex numbers derivatives. Some issues with the RStan unit tests.

General Meeting
C++ foreach, we have to not have side effects in math only. Will not get to the Stan language with parallelism because of the side effects constraint. Bob wants guidelines for how developers can do parallelism, some idea of performance improvements.

Are we going to release 2.19 or 2.18.1 without GPU given that we have a threading bug with maprect. Do a hot fix now, release 2.19 in a month or so. Will get it out in the next day or two.

Bob: Fix API awkwardness requiring iteration of algo to get instantiated model.