v1 - Lei Zhang - 10 Aug 2020
Note: this mini tutorial assumes you already know a bit about RStan, and now consider switching to cmdstanr.
f = rw_run('rw_hba', saveFit = T, test = F)
This repository contains:
root
├─ data # example data for a simple 2-armed bandit task
├─ scripts # stan model and R function
├─ stanfits # to save fitted objects
{CmdStanR} is a lightweight interface to Stan for R users that provides an alternative to the traditional {RStan} interface. It employs the most updated functionalities in cmdstan and provides nicer tools to operate the posterior. In addition, it seems that {CmdStanR} is the only interface that supports multithreading. See here for a comparison between {CmdStanR} and {RStan}.
For researchers using computational modeling to understand cognition, if you have used {RStan} and now consider switching to {CmdStanR}, here is a mini tutorial for you. In essence, all your *.stan models stay intact, and all you need to change is the wrapper function where you call stan models.
As an example, here I have a simple Rescorla-Wagner model for a 2-armed bandit task, with the hierarchical Bayesian approach (HBA) (Ahn et al., 2017). The model is called rw_hba.stan. Plus, the wrapper function is called rw_run.R, and the main input argument is the model string modelStr = 'rw_hba'. All commands in this wrapper function have been updated to be compatible with the {CmdStanR} package.
To run the model, simply call:
f = rw_run('rw_hba', saveFit = T, test = F)
In case you wonder, the core part is to create a cmdstan object:
mod = cmdstan_model(modelFile)
Then, to actually run MCMC sampling, call the $sample methods in the mod object.
fit = mod$sample(
data = dataList,
chains = nChains,
parallel_chains = nCores,
refresh = nRefresh,
iter_warmup = nWarmup,
iter_sampling = nIters - nWarmup,
seed = est_seed,
max_treedepth = treedepth,
adapt_delta = adapt
)
In addition, I let the function print out stan diagnostic messages; also, I include the computation of LOO for model comparison.
Note: when test = T, the wrapper function will only run 1 chain with 2 samples. This test mode is ideal for debugging stan models.
For bug reports, please contact Lei Zhang (lei.zhang@univie.ac.at).
Thanks to Markdown Cheatsheet and shields.io.
This license (CC BY-NC 4.0) gives you the right to re-use and adapt, as long as you note any changes you made, and provide a link to the original source. Read here for more details.
