Skip to content
 

Stan Weekly Roundup, 3 August 2017

You’d almost think we were Europeans based on how much we’ve slowed down over the summer.

  • Imad Ali, Jonah Gabry, and Ben Goodrich finished the online pkgdown-style documentation for all the Stan Development Team supported R packages. They can be accessed via http://mc-stan.org/(package_name), e.g.,

    The Stan manual will also get converted as soon as we can get to it.

  • Ben Goodrich added the nonlinear inverse link functions with Gaussian outcomes following lme4’s “self-starting” functions nlmer.

  • Rob Trangucci has added further GP doc to the manual and is working on multinomial logit to RStanArm.

  • Jonah Gabry released ShinyStan 2.4 and a new Bayesplot is on the way—they’re more flexible about ggplot2 theming, and also RStanArm releases to coordinate with the next Gelman and Hill edition bringing it up to date with modern R.

  • Breck Baldwin has been trying to wrangle governance issues.

  • Imad Ali is working on some basketball models and waiting for NBA data. Also supervising our high school student and working on the nonlinear models for RStanArm.

  • Aki Vehtari is continuing work on Pareto smoothed importance sampling with Jonah. StanCon Helsinki planning is underway; still waiting on a date.

  • Ben Bales rewrote append arrays and the initial RNG vectorization.

  • Bill Gillesie has been learning C++ and software development with Charle’s help. His first pending pull request is adding a linear interpolation function like the one in BUGS.

  • Charles Margossian finished the Torsten 0.8.3 release (that’s Metrum’s pharmacometrics package wrapping RStan).

  • Charles also finished the pull request for the algebraic solver and it’s passed code review, so it should land in the math lib soon.

  • Charles is also writing some docs on how to start programming with Stan, based on whathe’s been learning teaching Bill to write C++.

  • Charles and Bill are also adapting the mixed solver for a PK/PD journal.

  • Michael Betancourt wrote a case study about QR decomposition that’s up on the web site. He’s since been at JSM talking about Stan, where there were lots of posters citing Stan. He gave away a lot of stickers through the Metrum booth.

  • Michael, Aki, and Rob Trangucci have been working on GPs and Michael has a case study in the works.

  • Michael also made a GP movie tweet that’s gotten a ton of positive feedback on Twitter (along with the spatial, methodology, and QR decomposition case studies).

  • Andrew Gelman wrote up a draft of an R-hat an ESS calculation paper with me and Michael.

  • Mitzi Morris launched the spatial model case studies with the fit for intrinsic conditional autoregression (ICAR) model, with some neat parameterizations by Dan Simpson. She’s also got the Cook, Gelman, and Rubin in the wings.

  • Mitzi is also adding a data specifiation for variables that will let us write functions that only apply to data.

  • Sean Talts and Daniel Lee have been hammering away at the C++ builds through all our repos and allow better conditional compilation of optional external libs like CVODES (for ODE solving), MPI (process parallelism), and OpenCL (GPUs).

  • Sean and Michael have also been fiding anomalies in their Cook-Gelman-Rubin stats (as has Mitzi) when the number of replications is cranked up to the thousands.

7 Comments

  1. First, thanks to Imad, Jonah and Ben for the pkgdown sites for the r related packages. Makes it easier to browse all the information for that one package on one site.

    Second, could Charles create some notes or a simple “How to Guide” from his teaching experience so that some of us can benefit too. I would like to contribute more Kernels to GPs in Stan but don’t know how to program in C++ yet. I am currently taking some online classes to help.

  2. Guido Biele says:

    Given the season, I understand how shinystan could become shinytan :-)
    (text for link to shinystan)

  3. Corey says:

    Any word on when PyStan will switch to the MIT license or equivalent? This thread suggests that June was the aspirational target, but as a sometimes declarer of aspirational targets myself I know what that’s worth…

    • We were waiting to hear back from NumFOCUS’s intellectual property lawyer about how we should go about it, and everything looks good on the legal front.

      Allen Riddell is really going to have to answer the timing question, but I’m optimistic it will be relatively soon.

Leave a Reply