36.4 Some differences when running from R
Stan can be set up from within R using two lines of code. Follow the instructions for running Stan from R on the Stan web site. You don’t need to separately download Stan and RStan. Installing RStan will automatically set up Stan.
In practice we typically run the same Stan model repeatedly. If you pass RStan the result of a previously fitted model the model will not need be recompiled. An example is given on the running Stan from R pages available from the Stan web site.
When you run Stan, it saves various conditions including starting values, some control variables for the tuning and running of the no-U-turn sampler, and the initial random seed. You can specify these values in the Stan call and thus achieve exact replication if desired. (This can be useful for debugging.)
When running BUGS from R, you need to send exactly the data
that the model needs. When running RStan, you can include extra
data, which can be helpful when playing around with models. For
example, if you remove a variable x from the model, you can keep
it in the data sent from R, thus allowing you to quickly alter the
Stan model without having to also change the calling information in
your R script.
As in R2WinBUGS and R2jags, after running the Stan model, you
can quickly summarize using plot() and print(). You
can access the simulations themselves using various extractor
functions, as described in the RStan documentation.
Various information about the sampler, such as number of leapfrog steps, log probability, and step size, is available through extractor functions. These can be useful for understanding what is going wrong when the algorithm is slow to converge.