add sample_stats for PyStan

[ahartikainen]

Add sample_stats for PyStan.

[ahartikainen]

I added lp__ to here. Is this correct thing to do or should it be inside posterior?

[ColCarroll]

Looks good to me -- can merge when tests pass.

Is lp__ the log probability? Maybe we should rename it to log_probability - there is an open PR to add that to PyMC3 where it is called logp(pymc-devs/pymc#3134), and I think sampler stats is the right place for it. I'll add a PR to the pymc3 converter with whatever name you prefer.

[canyon289]

Basic question but what does lp stand for?

[ahartikainen]

It is either log-density or log-probability depending on the underlying Stan model.

E.g. target += normal_lpdf(...) is for probs and y ~ normal(...) drops constants --> density.

[ColCarroll]

I really like including that in sampler stats. Maybe we call it log_density since that technically covers both cases?

[junpenglao]

log_density could be confusing, I dont mind keeping it as lp__ actually - it minimized the name changes across in Stan, and we dont have it yet anyway in pymc3.

[ahartikainen]

I guess we can keep it as is (I stripped __ from the end)