Feature request - Plots of divergences

[springcoil]

If you have questions about a specific use case, please post it to our discourse channel: https://discourse.pymc.io
If you are not sure whether this is a bug or not, please also post it on discourse.

Description of your problem

This is more a feature request than a problem - I might even hack it myself.

Please provide a minimal, self-contained, and reproducible example.
In http://docs.pymc.io/notebooks/Diagnosing_biased_Inference_with_Divergences.html#A-Dangerously-Short-Markov-Chain there's some neat plots. Would they be useful as functions for plotting divergences for better diagnostics?

Please provide the full traceback.

Please provide any additional information below.

Versions and main components

[springcoil]

@ColCarroll Any views on this?

[aloctavodia]

I think the people from STAN have this kind of plots ready-to-use, what about a diagnostic summary, could that also be useful?

[springcoil]

Yeah I think a diagnostic summary would also be excellent. I think my thinking was inspired by some Stan examples.

[junpenglao]

Something like Figure 5 in https://arxiv.org/pdf/1709.01449.pdf?

[springcoil]

Yeah figure 5 is what I had in mind

…

On 10 Dec 2017 09:21, "Junpeng Lao" ***@***.***> wrote: Something like Figure 5 in https://arxiv.org/pdf/1709.01449.pdf? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#2745 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA8DiKOXF9wvb8bPYh-mvTbvo86f8CEWks5s-6KYgaJpZM4Q2P42> .

[aseyboldt]

#2677 should make that easier. You can access the warnings explicitly with that, and also see the exact point where the divergence happened. I still need to finish it up though...

[springcoil]

Any update on this?

[junpenglao]

I dont think anybody is actively working on this. But it should make a good beginner issue.

[agustinaarroyuelo]

I will try to work on this one

[junpenglao]

@agustinaarroyuelo you can have a look at http://docs.pymc.io/notebooks/Diagnosing_biased_Inference_with_Divergences.html - There is a prototype of how it would look like.