[WIP] Advance to Best Practices: Multivariate Gaussian Random Walk Notebook

[ltoniazzi]

Description

Addresses issue #140 and aims to advance it to Best Practices

Best Practices

General updates: everything addressed.
ArviZ:

• please doublecheck my usage of xarray.
• Not sure how to use coords/dims effectively in this notebook.

Changes for discussion

• Emphasized regression modelling.
• Added docstrings and text between code blocks.
• Added two ArviZ plots.
• How to justify the Scaler (like with a reference about sampling improvement)?
• When I run az.summary(trace).r_hat <1.03 one entry of a Cholesky matrix scores 1.04. Is that ok? (referring to this guide)

Cheers

---

Comment: I'll be available to work on this again on the 22nd of August.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[OriolAbril]

Line #43.            ppc = pm.sample_posterior_predictive(trace=trace)

I would directly add this to the inferencedata. I think the following should work.

trace.extend(az.from_pymc3(posterior_predictive=pm.sample_posterior_predictive(trace)))

Reply via ReviewNB

[ltoniazzi]

I did this and introduced coords. Everything seems to work but I get an odd message after the ppc sampling, namely the output of In [6] of the last commit says:

0, dim: steps, 300 =? 300
1, dim: y_, 3 =? 3

[OriolAbril]

This is a forgotten debug print that made it into latest arviz release, it has already been fixed in main but the new release is still pending

[OriolAbril]

Thanks for the PR, this makes the notebook much more clear

[OriolAbril]

@ltoniazzi can you rebase on main to get rtd to generate a preview of the notebook? I am quite sure the references will work except the doc one but it would be great to check they really do

[OriolAbril]

Looks great so far and shows how myst "unlocking" sphinx references on notebook markdown is extremely useful 😄