save a stan model with rstan

[Stéphane Laurent]

Hello, 

From the help of the stan_model function of the rstan package : 

save_dso

Logical with the default of TRUE: indication of whether the dynamic shared object (DSO) compiled from the C++ code for the model will be saved or not. If TRUE, we can draw samples from the same model in another R session using the saved DSO (i.e., without compiling the C++ code again).

But I don't understand : 

- How to save the DSO in a specified folder ? I don't see any argument to specifiy the folder, and it seems that the DSO file is saved in a temporary file (I see a .so file in the node @dso).

- Once I have saved the DSO file, how can I use it in another session ? 

I have just tried to save the output of stan_model() in a RDS file with saveRDS(), then I have checked I am able to use it in another session by reading it with readRDS(). But maybe it only works until the .so file in the tmp folder will be deleted ?

[Ben Goodrich]

That is essentially correct. Basically, use save_dso = FALSE if you want to share the results with someone else on a different computer and use save_DSO = TRUE if you want to then save the resulting R object to your disk for you to use in a subsequent R session (in which case you do need to put the result of saveRDS() into a permanent directory on the disk). In general, DSOs are not compatible across different operating systems, etc.

By the way, there is a replacement for the stan() function in rstan at

http://mc-stan.org/rstan/stan.R

that you can source() and it will automatically save a serialized version of the DSO to your disk in the same directory as your .stan program (or in the temporary directory if you pass a string to model_code). It will also run the chains in parallel in a way that most users would operate on a single laptop. Otherwise, you might need to customize it a bit for your setup, but the code should be straightforward enough to follow and modify.

Ben

[Stéphane Laurent]

Thank you. Ok, I was afraid the dso file was in the tmp folder. But today my tmp folder is empty and everything works well by loading my saved model with readRDS.