Use compact traceplot by default

[ColCarroll]

Fixes #3489 .

This is backwards compatible, and will effect anyone running arviz master, returning to its previous behavior.

[twiecki]

Thanks!

[junpenglao]

Can we make a 3.71 release...?

[fonnesbeck]

I'm not a huge fan of compact, TBH. You can't really see anything most of the time, which obviates the utility of the time series plots. Large numbers of variables (or large multivariates) should be displayed with plot_forest, as that's what it was designed for.

For example, I'd never want a new user to see a plot like the one above. It would be easy to mistake that for a failed MCMC run where there was no mixing, since the histograms look like point masses and the traces look like horizontal lines.

[ColCarroll]

Any thoughts on something like pm.plot(trace), and just putting Best Practice Best Plots in there? I think @aloctavodia was thinking of doing something like this. Instructions for reading and interpreting could go in the docstring.

[twiecki]

Nice idea.

…

On Fri, May 31, 2019 at 12:56 PM Colin ***@***.***> wrote: Any thoughts on something like pm.plot(trace), and just putting Best Practice Best Plots in there? I think @aloctavodia <https://github.com/aloctavodia> was thinking of doing something like this. Instructions for reading and interpreting could go in the docstring. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#3502>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAFETGGUNAUFT37TBPHVE7DPYFKDDANCNFSM4HRMGWLQ> .

[fonnesbeck]

Yes, I agree. I suppose we need to agree what best practices are. Maybe we can start a shared doc.

[aloctavodia]

We asked a small grant to numfocus to write a educational material. We are at a lag phase at the moment, but I guess this project will start taking momentum in about a couple of weeks. Basically the idea is to create an online e-book with best practices (and a little bit of theory) of Exploratory analysis of Bayesian Models. Also we have been discussing the idea of adding "cheat sheets" or something like that as ArviZ's functions.