Adds support for CmdStanFit objects. #27

[crsh]

This is a first pass at implementing support for CmdStanFit-objects (#27, @bnicenboim, @maxbiostat). While this works, it is quite inefficient compared to the methods for stanfit-objects. This is because the cmdstanr-method to unconstrain parameters is very slow compared to the rstan-methods on my machine. I have opened an issue and the cmdstanr-repository to see if this can be improved. But maybe you already have thoughts on this; any feedback would be welcome.

[maxbiostat]

Stellar stuff, thanks! Let's hope the cmdstanr folks can come to the rescue so we can make this efficient. I'm a bit swamped right now but I can throw together a few tests using my own stuff to see just how much slower your solution is and if I have any thoughts on how to make it faster.

Maybe @andrjohns has thoughts.

[singmann]

I apologise for the long delay in getting to this, but it took me until today to figure out how to use init_model_methods() on windows (solution here).

I have now checked this and it looks really good. Before accepting this just a few things that would be good to check/change.

• [ ] You changed the stanfit method and the branch for cores > 1 seems broken. Code on lines 223 to 228 seems to be partically repeated on lines 240 to 246. Please fix.
• [ ] For the CmdStanFit method, cores > 1 does not work on Windows. Please add same check to CmdStanFit method as exists for stanfit methods (lines 202 to 205).
• [ ] Can you confirm if the CmdStanFit method works for cores > 1 on non-Windows OS?

[crsh]

Thanks, @singmann. I have implemented the changes you requested and confirmed that cores > 1 works on Ubuntu.

The upside of my delayed response is that the Stan team has now resolved the issue that was causing the slow unconstraining of parameters.