Updating GP-Kron notebook

[ckrapu]

This pull request addresses issue #71 by updating the Kronecker GP notebook to follow best practices and to include improved descriptions. Some of the plots were also simplified and the plot generating code was cleaned up.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[ckrapu]

Marking this as WIP because the MAP estimator is returning weird results in this notebook.
Okay, should be ready for review now.

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

OriolAbril commented on 2021-04-24T18:16:09Z
----------------------------------------------------------------

Predictions look very different now, it looks like there used to be some kind of diffusion effect that is now gone. Could be due to find_map not having converged, the final logp is different from the past one.

fonnesbeck commented on 2021-07-03T13:48:51Z
----------------------------------------------------------------

The inputs are also different, due to a different seed (which is fine).

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

OriolAbril commented on 2021-04-24T18:16:10Z
----------------------------------------------------------------

The lines argument is not properly formatted and thus lines are not shown. If docs: https://arviz-devs.github.io/arviz/api/generated/arviz.plot_trace.html are not clear enough, I wrote a blogpost with examples on that: https://oriolabril.github.io/oriol_unraveled/python/arviz/matplotlib/2020/06/20/plot-trace.html#The-lines-argument

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

OriolAbril commented on 2021-04-24T18:16:10Z
----------------------------------------------------------------

Given that pp samples are used to average, each sample in the posterior should be used, not only the first 200 samples from the first chain. I would also suggest using from_pymc3+extend like in the rugby example which should hopefully not have much effect on the code below other than some labeling changes in selecting and averaging.

Rubgy example preview: https://nbviewer.jupyter.org/github/pymc-devs/pymc-examples/blob/main/examples/case_studies/rugby_analytics.ipynb

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

OriolAbril commented on 2021-04-24T18:16:11Z
----------------------------------------------------------------

both xarray and theano should be present here too. I added some guidance on adding extra libs in the watermark at https://github.com/pymc-devs/pymc3/wiki/PyMC3-Jupyter-Notebook-Style-Guide

[fonnesbeck]

The inputs are also different, due to a different seed (which is fine).

---

View entire conversation on ReviewNB

[fonnesbeck]

I'm going to merge this, and address Oriol's comments in another PR.