Documentation to inform users about Rhat warning

[nociale]

Closes #288

[gowerc]

@nociale - I was thinking about this a bit more and I was wondering if it would be better instead to resolve this by just having a higher sample amount and then thinning down to the requested number of samples ? e.g. if the user selects n_samples = 50 sampling 500 and then select 50 from the 500.

That would (hopefully) supress the warning but also provide better diagnostics and hopefully ensue that the samples are actually cover the posterior distribution space. With Rhats >> 1 I would be worried our samples aren't a good coverage of the distribution

[nociale]

Hi @gowerc, I think your proposal could be implemented by setting the thin argument of stan() call to the default 1, doing burn_in + burn_between*n_samples with the thinning of 1, and then select only the samples every burn_between (after the burn_in). Was this your idea?

This would maybe get rid of the warning because the diagnostics are run in a larger amount of MCMC iterations, but it would be overriding the choice of stan to run the diagnostics on the n_samples only.

Despite I do think this would be an elegant solution to the problem, I would maybe still go with the note to the warning message in the documentation looks as it is an easy and reasonable solution. With a sufficiently high number of samples, the warning should disappear anyway.

I am tagging @wolbersm as well as we discussed this together earlier today.

[wolbersm]

Thanks @nociale. As discussed, this reflects my opinion on the topic as well.