CRAN RStan and StanHeaders have been updated to version Stan 2.26. Importantly 2.26 supports both the old and the new array syntaxes. After all 200+ CRAN packages that have been built on top of RStan or StanHeaders have updated their Stan codes to use the new syntax, RStan and StanHeaders can move to Stan 2.33 which supports only the new syntax.

https://cran.r-project.org/package=rstan

#rstats #Bayesian #MCMC #Stan

CmdStan 2.33.0!

This new release brings two new exciting features:
- a new Stan language type - tuples
- a new VI algorithm - Pathfinder.

The release also features some new functions - mostly related to tuples, and other minor improvements.

There are also a number of long-deprecated language features that have now become errors with this version.

For details, see the blog post: https://blog.mc-stan.org/2023/09/05/release-of-cmdstan-2-33/

#Bayesian #mcmc #pathfinder #CmdStan

Really getting into the weeds of #NIMBLE and #MCMC behavior and trying to write my own sampler from scratch. I’ve been slowly working towards this point over the last few months but now that I’m finally sitting down and just DOING it… man, I feel out of my depth. At least my dog is a good cheerleader!

We have just released CmdStanR v0.6.0. The two biggest new features (thanks to Andrew Johnson) are:

- Exposing Stan functions to R
- Exposing other functionality to R including the model’s log-prob function, gradients, and parameter constraining / unconstraining

These enabled also
- Add moment-matching support to $loo() method

For a full list of changes see Changelog https://mc-stan.org/cmdstanr/news/index.html

Thank you to everyone who contributed!

#bayesian #mcmc #rstats

Andy Hoegh told us about his work using particle #MCMC to fit agent-based movement models to data. He has worked on Grizzly bear movement and #hendra virus in bats. The slides are also on GitHub https://andyhoegh.github.io/SMCDU/#/

MCMC Considers Drastic Action Against Telegram; Possibly Banning The Platform #apps #mcmc #socialmedia #telegram - https://www.lowyat.net/2023/301657/mcmc-considers-drastic-action-telegram/

random walk, n., the route through an open plan office taken by a #statistician whilst waiting for their #MCMC chains to converge (hopefully). Expected length of route may depend on factors such as hardware specifications, informativeness of #priors, and whether there's a good coffee machine nearby. #statsodon #Bayesian #iamworking @pymc

Are you aware of papers or projects using Stan in physics, biophysics, etc? Help collecting a list of such papers and projects in Stan discourse https://discourse.mc-stan.org/t/putting-together-a-list-of-physics-biophysics-etc-projects-using-stan/30992

#Stan #MCMC #Bayesian

@junpenglao @henri_pesonen Others from the Aalto Bayesian workflow group joining the BayesComp are (say hello to them, too!)
- Andrew Johnson
- Anna Riha
- David Kohns
- Leevi Lindgren
- Meenal Jhajharia @meenaljhajharia
- Niko Siccha @Nikosiccha
- Noa Kallioinen
- Teemu Säilynoja
- Yann McLatchie

#BayesComp2023 #Bayes #MCMC

On Thursday afternoon (15:45) I'll host panel discussion on probabilistic programming (and what does it require to ge a new algorithm added to some PPL package) with panelists
- Mitzi Morris, #Stan / Columbia University
- Junpeng Lao @junpenglao, TFP / #PyMC / Google
- Tor Fjelde, #TuringLang / University of Cambridge
- Henri Pesonen @henri_pesonen, #ELFI / Oslo University Hospital

#BayesComp2023 #Bayes #MCMC

On Thursday morning (10:30) I'll host a session "Trust and adding new algorithms to probabilistic programming frameworks" with talks by
- Måns Magnusson on posteriordb and how to use it for testing https://github.com/stan-dev/posteriordb
- Lu Zhang on Pathfinder and how it was tested https://jmlr.org/papers/v23/21-0889.html
- Mitzi Morris on BridgeStan and testing algorithms https://roualdes.github.io/bridgestan/

#BayesComp2023 #Bayes #MCMC

On Wednesday evening there will be four posters where I'm a co-author
- Charles Margossian et al., Nested Rhat for highly parallel MCMC https://arxiv.org/abs/2110.13017
- Noa Kallioinen et al., priorsense: prior and likelihood sensitivity analysis https://arxiv.org/abs/2107.14054
- Anna Riha et ak., Bayesian multiverse analysis
- Yann McLatchie et al., kulprit: projection predictive model selection for Python

#BayesComp2023 #Bayes #MCMC

On Tuesday morning (9:20) I'll talk in Bayesian computing without exact likelihoods workshop about "An importance sampling approach for reliable and efficient inference in Bayesian ordinary differential equation models" https://arxiv.org/abs/2205.09059 (one third of the slides are ready)

#BayesComp2023 #Bayes #MCMC

I'm traveling to BayesComp today. I'm looking forward to seeing many ex-colleagues, co-authors, co-developers, friends, and other interesting people. If you will be also there, come to say hello and join some session mentioned below in the thread

#BayesComp2023 #Bayes #MCMC

Jeremy Magland and Jeff Soules, Flatiron Institute, have made MCMC Monitor that enables tracking and visualization of MCMC processes executed with Stan in local or remote web browsers. You can configure Stan to generate output to a directory on your computer. MCMC Monitor reads this output and displays it in the web app, with real-time updates. As you track the progress of the run, MCMC provides diagnostic plots and statistics.

The git repo with demo https://github.com/flatironinstitute/mcmc-monitor#readme

#Stan #mcmc #Bayes

interested to hear if anyone is using #stan @mcmc_stan in industrial settings? Can be based in industry or academic collaboration

#bayes #mcmc #bayesian

Just uploaded a new tutorial video, looking at the role of 'Annealing' when we're trying to sample from Bayes' theorem

#machinelearning #Bayes #MCMC

https://youtu.be/EO6_t3STHdQ

RT @BerkovichRotem
Check out my new blog post: LBA modeling in R using the 'ggdmc' package; embrace yourself, a shitty complex model is coming

*It is worth at least for the sheep video*

#LBA #MCMC

https://www.miss-cognition.com/post/lba-modeling-in-r-using-the-ggdmc-package-embrace-yourself-a-shitty-complex-model-is-coming

Hello!

The next #Deepmind #ELLIS #CSML #seminar is happening today (17 February) and starts at 12pm UK 🇬🇧 time 😁!

We have the great pleasure to have Sam Power discuss explicit convergence bounds for Metropolis Markov chains!! Sam is a postdoctoral researcher at #Bristol. He previously completed a PhD in #statistics at #Cambridge.

This seminar is hybrid (#UCL #AI Centre and virtual). More info 👉 https://ucl-ellis.github.io/dm_csml_seminars/2023-02-17-Power/

#artificialintelligence #ai #machinelearning #mcmc #ml #bayesian

I made a new blog post on helping speeding up a very slow matrix computation.
https://thomasburgess.github.io/blog/2023/02/02/Matrix_Multiplication.html #python #mcmc #matrix #optimization #blog #numpy #blas