Horizontal radio buttons

[mromanie]

PR Summary

Add the option to have horizontal RadioButtons. A new optional keyword argument 'direction' is introduced to control it. The new parameter defaults to 'vertical' for backwards compatibility.

PR Checklist

• Has Pytest style unit tests
• Code is Flake 8 compliant
• New features are documented, with examples if plot related
• Documentation is sphinx and numpydoc compliant
• Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
• Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way

[jklymak]

flake8 needs to pass 😉 https://travis-ci.org/matplotlib/matplotlib/jobs/489577517

[NelleV]

Thanks for the PR @mromanie !
Here are a couple of comments. I'll do a more thorough review once these are addressed.

Can you in addition add a what's new entry? See doc/users/next_whats_new/README.rst

[NelleV]

Can you add what are the possible values here?
Can you also add a mock test to make sure it runs, as well as a test to make sure this exception is raised properly.

I also think we need to update the examples to show case this (examples/widgets/radio_buttons.py)

[QuLogic]

We have a shortcut function to check options like this and raise an appropriate error now, cbook._check_in_list

[NelleV]

The keyword being a bit confusing here, I think it's worth detailing a bit more what orientation means (which is the set of radio buttons are on the same horizontal axis or vertical axis).

[NelleV]

For the failing flake8 test, I would suggest adding the check directly to your text editor.

[timhoffm]

Milestoning 3.2 as there has not been activity recently. @mromanie are you still interested to work on this?

[mromanie]

@timhoffm Apologies for my inactivity. I am indeed still very much interested in working on this. In fact, I think I managed to do it properly by using patches.Ellipse instead of patches.Circle and stretching them to look circular using the aspects of the axes and figure. In this way, the entire surface of the buttons is sensitive, unlike the current implementation with patches.Circle.

I still need to clean up the auto-scaling of the size of the symbols, which at the moment still results in the symbols to overlap under certain circumstances, and I'll submit the whole lot for review. Shall I use the same PR for this?

[ImportanceOfBeingErnest]

What about using a PathCollection with a circle marker, such that cicles always stay circles independent on the figure- or axes- dimension or aspect?

[mromanie]

@ImportanceOfBeingErnest I'm not familiar with PathCollection, so I don't know what are the advantages/disadvantages. The gist of what I have implemented is as follows, which seems to do a reasonable job:

fig = ax.get_figure()
fh, fw = fig.get_figheight(), fig.get_figwidth()
rpos = ax.get_position().get_points()
rscale = (rpos[:,0].ptp() / rpos[:,1].ptp()) * (fw / fh)
circle_radius = 0.25
p = Ellipse(xy=(0.15, y), width=circle_radius, height=circle_radius*rscale, edgecolor='black', facecolor=facecolor, transform=ax.transAxes)

Plus a couple more things to make it backwards compatible. Like I said earlier, setting circle_radius to a fixed value is not always satisfactory and I' fine-tuning it.

[mromanie]

...forgot to say, the condition to activate the Ellipse button is:

((xy[0] - p.get_center()[0])/p.width)**2 + ((xy[1] - p.get_center()[1])/p.height)**2 <= 1

[ImportanceOfBeingErnest]

If you plot a scatter plot with circular markers (marker="o") you may have noticed that those markers will stay circles even if you resize the figure, change the aspect or zoom into the plot. This is achieved inside the PathCollection by defining the circle path in display coordinates and using an offset transform to shift those circles at the respective positions in data coordinates. I would think the same can be done here. The advantage compared to the strategy above would be that the bullets will keep their size when resizing the figure after it has first appeared on screen.
In addition, checking the clicks on the bullets could be simplified by using a pick_event.

[mromanie]

OK, thanks!

All things considered, I would stick to the solution with Ellipse because:

• Zooming is disabled for the axes containing the RadioButtons.
• It is true that the Ellipse looses its circular shape when the entire window is resized, but I'm not sure that this is a problem.
• The solution entails minimal changes wrt the current implementation.
• It is straightforward to make it backwards-compatible by mimicking the current behaviour.

[ImportanceOfBeingErnest]

Not sure what the status is here. Since someone asked a question about horizontal buttons on stackoverflow, I thought it might be worth showing a solution over there: https://stackoverflow.com/questions/55095111/displaying-radio-buttons-horizontally-in-matplotlib/55102639#55102639

Of course this could also be used for this update.

[QuLogic]

This looks mostly done, but needs a rebase and a small tweak to the checks. The squashing is something that affects vertical buttons too, so could be a separate PR.

[QuLogic]

Selecting by closest button is now merged as #13268. Using a scatter plot to ensure circular buttons is merged as #24474.

I pushed a rebase, dropping the old changes, fixed the review comments and adding a test/what's new entry. I also modified the implementation a bit so that it modifies the placement of the buttons instead of reversing the labels (as the latter has ramifications for the active index.)

I'm also working on implementing the same change for CheckButtons.

[QuLogic]

An example of how this looks may be found in the what's new entry: https://output.circle-artifacts.com/output/job/26b48f05-0be7-417a-b57e-305a60aa66b9/artifacts/0/doc/build/html/users/next_whats_new/widget_horizontal.html