Extend vignettes

[jgabry]

The vignettes haven’t been updated in a while. They need some updating to reflect the content added to the package since they were last worked on.

Also, @martinmodrak any interest in adding some content from or similar to your blog post

http://www.martinmodrak.cz/2018/02/19/taming-divergences-in-stan-models/

to give more substance to the diagnostics vignette?

_{Sent with GitHawk}

[martinmodrak]

Could probably do that, but is this the right place? If the dev team believes the contents of the blog post is correct and illustrative enough to be part of official materials, shouldn't that go to Stan manual or RStan vignettes?

Right now, the blog does not show how to use diagnostic visualization (which I hope it sometimes will, but you know how those plans go). If I had something to say directly about visaulization, that would certainly make sense here.

[tjmahr]

I'll eventually write a vignette to cover bayesplot for ggplot2 users... That would demonstrate the data tidying features and some use cases for the _data() functions, like combining draws from multiple models.

[jgabry]

@martinmodrak yeah I should have been more specific. I meant that I liked the text in your post and I haven’t had the time to add as many words to the diagnostics vignette (it’s mostly pictures). So I was wondering if you wanted to beef up the text a bit rather than the visualizations.

_{Sent with GitHawk}

[martinmodrak]

@jgabry I see! Yeah, I'll try to look into that in few weeks time.

[jgabry]

Ok no rush or obligation! Just wanted to let you know that I liked the post and that we definitely encourage contributions

_{Sent with GitHawk}

[martinmodrak]

OK, so I did some changes to the diagnostic vignette. My main goal was to present the function in the order I would run them when diagnosing a modelling issue. In that view, it might make sense to move the "Diagnostics for the No-U-Turn Sampler" to the top, before "General MCMC diagnostics", at least I personally use the NUTS diagnostics much more often than Rhat/n_eff (I look at those in the sampling output). But I am not certain about this.