GP Heteroskedastic

[OriolAbril]

File: https://github.com/pymc-devs/pymc-examples/blob/main/examples/gaussian_processes/GP-Heteroskedastic.ipynb
Reviewers:

The sections below may still be pending. If so, the issue is still available, it simply doesn't
have specific guidance yet. Please refer to this overview of updates

Changes for discussion

Changes listed in this section are up for discussion, these are ideas on how to improve
the notebook but may not have a clear implementation, or fix some know issue only partially.

ArviZ related

• Use ArviZ and xarray for posterior predictive plotting

Notes

Exotic dependencies

Computing requirements

[ltoniazzi]

I am happy to update this notebook to follow the new style-guide, but I can only run it on cpu, and it takes 26hours instead of the 7hours in @JohnGoertz current version. Is that a problem? Could a solution be that once I'm finished someone re-runs it on gpu?

[JohnGoertz]

I have access to a cluster, I can chuck it up there after you update it. I think I just ran that on my laptop (Surface Pro i7) cpu, I'm surprised it takes that much longer on your computer.

I'd also like to add in a section on parametric heteroskedasticity following this discussion. I can do that after you update the style.