Following up on recent posts here and here, I thought I’d post a list of all the Stan case studies we have so far.

2017:

Modeling Loss Curves in Insurance with RStan, by Mick Cooney

Splines in Stan, by Milad Kharratzadeh

Spatial Models in Stan: Intrinsic Auto-Regressive Models for Areal Data, by Mitzi Morris

The QR Decomposition for Regression Models, by Michael Betancourt

Robust RStan Workflow, by Michael Betancourt

Robust PyStan Workflow, by Michael Betancourt

Typical Sets and the Curse of Dimensionality, by Bob Carpenter

Diagnosing Biased Inference with Divergences, by Michael Betancourt

Identifying Bayesian Mixture Models, by Michael Betancourt

How the Shape of a Weakly Informative Prior Affects Inferences, by Michael Betancourt

2016:

Exact Sparse CAR Models in Stan, by Max Joseph

A Primer on Bayesian Multilevel Modeling using PyStan, by Chris Fonnesbeck

The Impact of Reparameterization on Point Estimates, by Bob Carpenter

Hierarchical Two-Parameter Logistic Item Response Model, by Daniel Furr

Rating Scale and Generalized Rating Scale Models with Latent Regression, by Daniel Furr

Partial Credit and Generalized Partial Credit Models with Latent Regression, by Daniel Furr

Rasch and Two-Parameter Logistic Item Response Models with Latent Regression, by Daniel Furr

Two-Parameter Logistic Item Response Model, by Daniel Furr, Seung Yeon Lee, Joon-Ho Lee, and Sophia Rabe-Hesketh

Cognitive Diagnosis Model: DINA model with independent attributes, by Seung Yeon Lee

Pooling with Hierarchical Models for Repeated Binary Trials, by Bob Carpenter

2015:

Multiple Species-Site Occupancy Model, by Bob Carpenter

2014:

Soil Carbon Modeling with RStan, by Bob Carpenter

Has the Stan team seen any case studies of Stan deployments on Java? Our team would like to use Stan as a model deployment tool (develop + prototype models, in R, hand off the stan file as the modeling reference to the production engineering team for use in a Java environment). I haven’t seen examples of Stan used this way yet, but it seems intuitively useful.

Just to clarify: you want to fit the model in R, then score new data in Java? Or do you want to fit and score in Java? (I assume the former, but it’s not totally clear from your comment.)

Sorry, let me be more clear.

I want to follow this process:

– Do model development in R using (relatively) smaller datasets

– Hand the stan config file off to engineering, which builds and operates a Java-based quality control system (sampling, estimation, etc)

– Allow the engineers to run Stan through Java. They run the actual model in Java. The outcomes of interest are the model parameters.

We want to use this approach for scoring quality control performance in simple random samples. E.g., instead of taking a stratified sample and summarizing each stratum (which is expensive if you have lots of strata), take a simple random sample over all strata and estimate per-stratum outcomes using a partial pooling model.

Stan is attractive because we can specify the entire model in the .stan file, which means we don’t have to translate it to another modeling paradigm. But we need to be able to call Stan from Java, and there’s no Java API.

Haa, I’ve Sophia’s (Rabe-Hesketh) book on latent variable modeling. And… Skrondahl, I believe was the other writer. I quite like that book. Cheers to all of you. Had to comment since I saw a familiar name. I believe she was mentioned in some other post also! Sophia is everywhere!

Andrew: Nice talk at SSI! For those who hadn’t heard about it, the Symposium on Statistical Inference was a two-day American Statistical Association conference that “follow[ed] up on the historic ASA Statement on P-Values and Statistical Significance.” Andrew and two others spoke at the closing plenary session.

It was really nice being in the hallway when Andrew and Frank Harrell met for the first time. Those of us standing around were surprised that they’d never met, but conference siloing and all of that, I guess.