API: Multi-output summary plot legend

[101AlexMartin]

Overview

Added flag to allow the printing of the mean SHAP values in the legend of a multi-output summary bar plot. The motivation behind this change is to easily compare rankings when different chunks of data are used to get the SHAP values. Comparing visually the overall ranking in a multi-output problem was quite challenging, so this pull request solves the problem.

Checklist

• Closes #xxxx
• All pre-commit checks pass.
• Unit tests added (if fixing a bug or adding a new feature)
• Added entry to CHANGELOG.md (if changes will affect users)

[thatlittleboy]

Hi @101AlexMartin , thanks for the PR!

A couple of things before we kick off the PR review:

• Please fill up the checklist in the PR description.
• As this is a new feature, please add tests to ensure the desired functionality is implemented properly.
• I intend to deprecate summary_legacy at some point in the future. So I'll also suggest adding this param into _bar.py as well.

[101AlexMartin]

Hello! I implemented your proposed changes :)
As for the depreciation of summary_legacy and the addition of the parameter to _bar.py as well. If I'm not mistaken, _bar.py is not ready yet to display graphs for multiouput models. Thus, I do not see the addition of this parameter feasible.

[thatlittleboy]

Hi @101AlexMartin , next steps would be to make sure the tests pass :)

In particular for plotting tests, we are using pytest-matplotlib. You will need to generate the baseline images and commit that into the repository.

This way we can ensure that the function always generates the same-looking plot.

[101AlexMartin]

I added the tests and the baseline figures :)
Also merged with master branch

[codecov]

Codecov Report

Merging #3062 (73ff252) into master (d650456) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #3062      +/-   ##
==========================================
+ Coverage   55.33%   55.35%   +0.02%     
==========================================
  Files          90       90              
  Lines       12875    12883       +8     
==========================================
+ Hits         7124     7132       +8     
  Misses       5751     5751              

Impacted Files	Coverage Δ
shap/plots/_beeswarm.py	78.25% <100.00%> (+0.34%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[101AlexMartin]

What is the status of this then?
Let me know if you need any extra explanation :)

[thatlittleboy]

@101AlexMartin Sorry, let me get back to you over the weekend. Bit busy at work this week.

[thatlittleboy]

Hi @101AlexMartin Sorry this review took so long! This should be the last round, I just have a couple more suggestions for you to make.

[101AlexMartin]

Done :)

[thatlittleboy]

Just a heads up, the 3.11 tests are failing, not anything to do with your PR. Probably due to the 4.0 lightgbm release. #3092

This PR looks good to me, I'll come back to this once we get our builds running again. Hang tight!

[thatlittleboy]

LGTM