Easiest file format (or loading files with custom R code) to use custom MCMC chain with shinystan

General
1

I am interested in using shinystan and related visualization tools for MCMC chains generated from custom code (i.e., not from a standard MCMC DSL).

Given that shinystan seems to support multiple backends, I suspect this is possible. Are there any complete examples of how to do this? Is there a simple text-based file format that we could generate which would make this especially easy to do?

For example, if we had a CSV file such as:

iteration, chain, variable_1, variable_2
1, 1, 0.37, 0.28
2, 1, 0.55, 0.33
1, 2, 0.55, 0.282

Then is that enough for shinystan or Plotting for Bayesian Models • bayesplot to use? If so, are there examples of that?

2

Hi,
I think you would need to transform the CSV manually into an mcmc.list object or other format supported by ShinyStan - see http://mc-stan.org/shinystan/articles/shinystan-package.html (which should be relatively straightforward, as all the information you need is already in your CSV)

Best of luck
Martin

3

Thanks. Can you think of any publicly accessible code that is using that API?

Also, Is there something similar with bayesplot as well (or one of the other visualization packages)?

4

From Getting Started • shinystan

If you have an mcmc.list object called my_mcmc then you can use the as.shinystan function to convert my_mcmc into a shinystan object that can then be used with launch_shinystan :

my_sso <- launch_shinystan(as.shinystan(my_mcmc, model_name = "my_model"))

Or do you need something more?

Most (all?) bayesplot functions work with base R arrays/data frames, so not sure what API are you missing/expect… Plots for Markov chain Monte Carlo simulations — MCMC-overview • bayesplot