betanalpha / mcmc_diagnostics Goto Github PK

Howdy, really appreciate having these diagnostics available. I am a cmdstanpy user and found it pretty straightforward to translate over the extract_ functions to be compatible.

While going through the code, I thought there were a few areas that could be improved. Some initial thoughts:

1. There are some minor errors that I assume slipped through the cracks when translating these diagnostics from R to Python, e.g. there are some undefined variables, a NaN that should probably be a np.NaN, etc. Didn't hit any branches that caused errors when I ran the diagnostics myself, but either way those fixes would be easy.
2. Building a general compatibility for both PyStan and cmdstanpy -- since you've already done most of the work with the extract_* functions, this could be as simple as implementing these functions for both the stan.StanFit objects and the cmdstanpy.stanfit.mcmc.CmdStanMCMC objects. This wouldn't be much work and would make it for any Python Stan users to take advantage of this diagnostic code.
3. Packaging up the code base and making it PyPI installable. Intention here is just to make it so a Python Stan user could pip install mcmc-diagnostics or whatever the name should be and quickly incorporate these diagnostics into their workflow. I know this repo has a mix of R, PyStan2, PyStan3 implementations as well as documentation, so I'm not sure exactly if you'd want to separate things out or not.

With all of these suggestions, I would be happy to work on implementing them if you are at all interested in doing so. Personally, I think it would be great to have a diagnostic utility like as something that can be easily incorporated into one's workflow.

If you're interested in pull requests or have any particular desires for future direction here, I'd be interested in hearing about them.

I think the n variable should be changed to c in this row, leading to:

cat(sprintf('Chain %s: E-FMI = %s.\n', c, numer / denom))