Pystan cheat sheet - first draft - comments sought

Interfaces PyStan
1

I have had a stab at writing a pystan cheat sheet.

The first cut can be found here: https://drive.google.com/file/d/1JuOQXZPX9cb6923qKeB8Jg9L1Zu5-ERh/view?usp=sharing

Comments and suggestions most welcome.

9 Likes
2

This is really useful. It would be great to have this in the pystan
documentation. Any interest in seeing it added there?

3

Looks really nice.

I have some suggestions:

  • There is a possibility to use ArviZ for plotting + data handling (InferenceData)
  • Compress with gzip. I usually add one step and also compress pickle dumps
  • Add PyStan version + OS for cache name.

Ps. Make sure that tracdplots are done with permuted=False, this was fixed for PyStan 2.18.

4

Thanks for your suggestions - will see what I can do over the next week or so.

5

Thanks for the feedback - I have substantially re-written the cheat-sheet to use ArviZ for summary statistics, visualisation and diagnostics.

As this was a major re-write, there may be errors (and instances of my misunderstanding). Any corrections/simplifications greatly welcomed.

The updated cheat sheet can still be found here: https://drive.google.com/file/d/1JuOQXZPX9cb6923qKeB8Jg9L1Zu5-ERh/view

1 Like
6

Looks good.

Current master has some API changes compared to pypi version, but this shouldn’t affect the usage too much.

Also after this https://github.com/arviz-devs/arviz/pull/387 is merged, you have a possibility to add info about the divergences to traceplot.

7

This is a very useful cheat sheet. Definitely useful not only for how to ask certain questions but learning which questions to even ask.

A question for you, in the sheet you mention the code for the SPLOM plots. Am i missing where this is stored?

8

SPLOM can be found here

https://drive.google.com/file/d/18LduDTn0Zs0j49d_JDBNGeK1rn0PE-N3/view?usp=drivesdk