Upcoming breaking changes in loo_compare() output

[florence-bockting]

Description

In the loo package, we are making changes to the output structure returned by loo_compare() in PR #300. The main changes are a transition from a matrix to a data frame and the addition of several new columns. Here is an example of the updated output structure:

> loo_compare(w1, w2)
  model elpd_diff se_diff p_worse diag_diff diag_elpd
 model1       0.0     0.0      NA                    
 model2      -4.1     0.1    1.00   N < 100     
     
Diagnostic flags present.
See ?`loo-glossary` (sections `diag_diff` and `diag_elpd`)
or https://mc-stan.org/loo/reference/loo-glossary.html.

More context on the motivation behind these changes can be found in the updated case study "Uncertainty in Bayesian LOO-CV Model Comparison".

A reverse dependency check revealed that these changes will affect your package. To help smooth the transition, I have already reviewed your code and suggest within this PR changes that should keep your package compatible with both the current and the upcoming loo_compare() output structure.

I may have missed some additional spots in your code that need updating, so please feel free to point those out and I am happy to help. Thank you and please do not hesitate to reach out if you have any questions or concerns.

TODO

• I noticed one aspect related to flocker_tutorial.Rmd where I am uncertain how you want to handle this. In the section "Post-processing > Model comparison" you currently provide the expected output in form of a comment. This expected output will change for the new loo_compare() function.