Awesome

stantargets R package validation example

The goal of this workflow is to demonstrate the stantargets R package using a practical example. In this scenario, we validate a small Bayesian model using an interval-based method similar to simulation-based calibration (SBC; Cook, Gelman, and Rubin 2006; Talts et al. 2020). We simulate multiple datasets from the model and fit the model on each dataset. For each model fit, we determine if the 50% credible interval of the regression coefficient beta contains the true value of beta used to generate the data. If we implemented the model correctly, roughly 50% of the models should recapture the true beta in 50% credible intervals.

The model

y_i ~ iid Normal(alpha + x_i * beta, sigma^2)
alpha ~ Normal(0, 1)
beta ~ Normal(0, 1)
sigma ~ HalfCauchy(0, 1)

The stantargets pipeline

The stantargets R package manages the workflow. stantargets is an extension to pacakges targets and cmdstanr for Bayesian data analysis, and it makes both the latter packages easier to use, especially multi-rep simulation studies like this one. stantargets makes decisions about how the pipeline is constructed, cmdstanr runs and postprocesses the models, and targets orchestrates the computation while skipping expensive computations if the results are already up to date.

Why stantargets and not just targets?

stantargets makes the validation study much simpler than with targets alone. To demonstrate, https://github.com/wlandau/targets-stan is a implementation of the same workflow without stantargets, and the _targets.R file is much longer and more complicated. With stantargets, the pipeline in the _targets.R file becomes much simpler and easier to define.

How to access

You can try out this example project as long as you have a browser and an internet connection. Click here to navigate your browser to an RStudio Cloud instance. Alternatively, you can clone or download this code repository and install the R packages listed here.

How to run

In the R console, call the tar_make() function to run the pipeline. Then, call tar_read(cover_continuous) to retrieve observed coverage, and open report.html in a web browser to view the full results of the validation studyz. Experiment with other functions such as tar_visnetwork() to learn how they work.

File structure

The files in this example are organized as follows.

├── run.sh
├── run.R
├── _targets.R
├── _targets/
├── sge.tmpl
├── R
│   └── functions.R
├── stan
│   └── model.stan
└── report.Rmd

Scaling out

This computation is currently downsized for pedagogical purposes. To scale it up, open the _targets.R script and increase the number of simulations (the number inside seq_len() in the index target).

High-performance computing

You can run this project locally on your laptop or remotely on a cluster. You have several choices, and they each require modifications to run.R and _targets.R.

References

Cook, Samantha R., Andrew Gelman, and Donald B. Rubin. 2006. “Validation of Software for Bayesian Models Using Posterior Quantiles.” Journal of Computational and Graphical Statistics 15 (3): 675–92. http://www.jstor.org/stable/27594203.

Talts, Sean, Michael Betancourt, Daniel Simpson, Aki Vehtari, and Andrew Gelman. 2020. “Validating Bayesian Inference Algorithms with Simulation-Based Calibration.” http://arxiv.org/abs/1804.06788.