change kde limits and hdi computation

[aloctavodia]

• Follows official PR format
• Includes a sample plot to visually illustrate the changes (only for plot-related functions)
• New features are properly documented (with an example if appropriate)?
• Includes new or updated tests to cover the new feature
• Code style correct (follows pylint and black guidelines)
• Changes are listed in changelog

This makes two important changes. compute the kde over the data range (not over 0, 1). This help to spots problems, as the KDE is "not adding data". Compare our example in the gallery, with the new version.

.

This also introduce a theoretical HDI, instead of estimating the HDI from the samples.

maybe we should enforce the zero y-bottom-limit.

[codecov]

Codecov Report

Merging #1215 into master will decrease coverage by 0.06%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1215      +/-   ##
==========================================
- Coverage   93.14%   93.07%   -0.07%     
==========================================
  Files          94       94              
  Lines        9497     9504       +7     
==========================================
  Hits         8846     8846              
- Misses        651      658       +7     

Impacted Files	Coverage Δ
arviz/plots/backends/bokeh/loopitplot.py	89.18% <100.00%> (+0.30%)	⬆️
arviz/plots/backends/matplotlib/loopitplot.py	97.29% <100.00%> (+0.52%)	⬆️
arviz/plots/loopitplot.py	98.87% <100.00%> (ø)
arviz/plots/hdiplot.py	83.92% <0.00%> (-3.58%)	⬇️
arviz/utils.py	89.33% <0.00%> (-1.84%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ec33b4c...a5b597b. Read the comment docs.

[OriolAbril]

I am not sure about not setting the range between 0 and 1.

Here is an example of bad behaviour I think this would introduce: imagine we have 1e6 observations whose 1e6 loo-pit values are uniformally distributed between .3 and .7. Old behaviour would set the kde limits to 0,1 and the resulting kde would have an inverted U shape whereas with this change the kde would be perfectly flat and it would be the fact that the range is not from 0 to 1 what should warn the user of the extremely bad loo pit result they actually have.

[OriolAbril]

Is this ready to merge?

[aloctavodia]

Yes.