ENH: New plotting API for beeswarm so can accept and return Axes

[chriscave]

Overview

Supports #3411

Change beeswarm to adhere to new standard API interface for the plots:

• Add an ax parameter to accept an existing matplotlib Axes
• Return the axes (if show=False)

Discussion

The parameter plot_size is only really suitable for when there is one axes in the figure. If there are multiple then the default value affects the figure size and not just the axes which the plot is being drawn to.

We could use plot_size to just affect the size of the axes but this I think this is discouraged from mpl and can be quite tricky to do.

So for now I thought it would be easier to just stop the behaviour of plot_size when ax is not None. So if a user passes in ax into the plot then it is up to them to determine the size of the plot using the figure the axes has come from.

We can definitely change this if it is not in line with the current thoughts on the new API.

Checklist

• All pre-commit checks pass.
• Unit tests added (if fixing a bug or adding a new feature)

[chriscave]

Tagging @connortann into this draft PR

[connortann]

Thanks for the PR! The tests are currently failing due to an unrelated issue, which should be fixed shortly by#3563.

[codecov]

Codecov Report

Attention: Patch coverage is 75.00000% with 8 lines in your changes are missing coverage. Please review.

Project coverage is 61.47%. Comparing base (dc40a03) to head (cd065c2).
Report is 1 commits behind head on master.

Files	Patch %	Lines
shap/plots/_beeswarm.py	75.00%	8 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3561      +/-   ##
==========================================
+ Coverage   61.46%   61.47%   +0.01%     
==========================================
  Files          90       90              
  Lines       12746    12753       +7     
==========================================
+ Hits         7834     7840       +6     
- Misses       4912     4913       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[connortann]

May I ask about the motivation behind pl.close? I think it may be a breaking change; we may as well decide a suitable pattern and plan to do that same for the other plots.

[chriscave]

Of course. Jupyter will still render the plot even if show = false. Closing the figure will prevent that. But maybe I've misunderstood what show is meant to be doing.

Ah if its a breaking change then perhaps more thought should be put into it. How do you think it's a breaking change?

[connortann]

This PR looks good to me! One suggestion - I think it would be preferable to raise a ValueError if plot_size and ax are both passed, rather than just ignoring the size.

[chriscave]

This PR looks good to me! One suggestion - I think it would be preferable to raise a ValueError if plot_size and ax are both passed, rather than just ignoring the size.

Good suggestion, I'll change it to that.