Adding matplotlib-specific styling functionality

[mezarque]

PR checklist

• [x] Describe the changes you've made.
• [x] Describe any tests you have conducted to confirm that your changes behave as expected.

Overview

This PR adds in automatic styling functionality for matplotlib, giving users the ability to more easily perform style modifications to plots. The modifications occur primarily through two parts of the API:

• apc.mpl.setup():
  • gets about 80% of the way to styling plots properly
  • meant to be called just once at the top of a plotting session
  • loads Suisse family fonts, if they are installed
  • registers Arcadia colors, palettes (as ListedColormaps), and gradients (as LinearSegmentedColormaps) into matplotlib with prefix apc:
  • loads a matplotlib stylesheet (arcadia_2024.mplstyle), updating the defaults in rcParams.
• apc.mpl.autostyle()
  • because matplotlib rcParams doesn't allow you to control everything, this is needed to get the remaining ~20% towards correct style
  • meant to be called after each plot
  • capitalizes axis labels
  • capitalizes and styles the legend to be left-justified, with title in "Suisse Int'l Semibold"
  • adds horizontal line separating legend title from entries
  • using the mono argument, alters font of user-chosen ticklabels to be "Suisse Int'l Mono"
  • using the cat argument, removes ticks and capitalizes ticklabels for categorical axes (e.g. for violin or bar plots)
  • using the cbar argument, alters the font of the tick labels of any colorbars to be "Suisse Int'l Mono"
  • there are functions that perform each of these modifications individually, for power users who want to change things a la carte

In addition to those major functions, there's some auxiliary functionality to help figures conform to fixed size guidelines, which are provided via existing templates in our Adobe Creative Cloud library.

• apc.dimensions()
  • gives a tuple of a fixed dimension, to be passed to plt.subplots(figsize = <>)
  • sizes are "full_w", "full_s", "threeq_w", "threeq_s", "half_s"
• apc.SAVE_WEB and apc.SAVE_PRINT
  • provides a collection of arguments to be passed to plt.savefig(**apc.SAVE_WEB) to make sure margins, etc. are of the correct size

Drive-by (is this the proper usage of this? I can't find references online):

• split up the classes.py module into one module for each class, hexcode.py, gradient.py, and palette.py

There's a lot going on here, and I'm sure we'll have a lively discussion about things like naming, what to expose to users, organization, namespaces, etc. Looking forward to lots of feedback!

Testing

I confirmed that things work as expected by testing the functionality in usage_example.ipynb, which goes into some detail about what all is going on and how users might use the various functions.

Halp

CI is failing right now; it seems that impresources is not finding the arcadia_2024.mplstyle file and is returning None. I am not sure how to resolve this, and would love some help figuring out how to get the CI to work properly; this error does not show up when I run pytest . on my local clone.

TODO

These are things that I'd like to figure out, but haven't yet found an elegant way to implement.

• [ ] Write tests
  • I think there's some base functionality to test and handle, e.g.:
    • What happens if Suisse fonts are not found?
    • What happens if users accidentally call the same function repeatedly?
    • How do we make sure we don't over-register info into matplotlib and detect collisions there?
  • Probably, as previously discussed, it will be very hard to test the visualization aspect without just manually looking at plots. It would be nice to have some kind of CI-based approach to generating the plots, though.
• [ ] Provide a declarative utility for adjusting the marker sizes of the legend.
  • When generating a figure using seaborn, the marker sizes of legend elements can be really small (equivalent to the size of points in the scatter plot). This is bad b/c it's hard to tell what is a scatter plot point vs. a legend marker. There are lots of ugly ways to solve this problem, some of which involve just creating a new legend, but it's kind of annoying. Would be nice to have a declarative way to do this, e.g. apc.mpl.scale_legend_markers(2) or something.

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[mezarque]

I haven't fully reviewed the mpl changes, but I made a couple suggestions to start a discussion. I am not sure how much research you've done around available tools, but I would check these two tools as potential inspiration: https://github.com/raybuhr/pyplot-themes and https://github.com/lgienapp/aquarel

Thanks for pointing me to these! I do like how aquarel uses with to specify a context for plotting; this could save users from having to have to call autostyle at the end of plots. I think both packages do some of the things we need to accomplish, but not all. Configurability/ usability is also something we want to consider, since I think users may not know what changes they want to make in advance of generating a plot. Let's explore these different approaches through some user testing.

[mezarque]

I think I addressed most of the comments above - thanks for all the feedback! My main concern with the renamings is that all of the functions are now quite verbose, which feels clunky / ugly to use. But maybe that's just the sacrifice we have to make for clarity.

Should I write tests for the functionality in this PR, or wait for another PR?

[mezarque]

What happened to the tests? They seem to have disappeared from the CI workflow.

I'm not sure where you're looking, but at least in my view of the PR, the tests are still running (and if you look at the Actions tab, it seems like they're working still?)

I addressed the remaining comments; lmk if you have any further Qs!