Allow plotting categorical data

[Illviljan]

• Closes use matplotlib's categorical axis features #4260
• Tests added
• Passes pre-commit run --all-files
• User visible changes (including notable bug fixes) are documented in whats-new.rst

a = xr.DataArray(
    [0, 1, 2],
    dims=("dim_1"),
    coords=dict(dim_1=(["dim_1"], ["u", "v", "w"])),
)
a.plot()

a = xr.DataArray(
    [[0, 1, 2], [3, 4, 5]],
    dims=("dim_0", "dim_1"),
    coords=dict(dim_0=(["dim_0"], [0, 1]), dim_1=(["dim_1"], ["u", "v", "w"])),
)
a.plot()

[Illviljan]

Are there any related tests for this that can be extended?

[dcherian]

LGTM! This needs a whats-new note though. Thanks @Illviljan

[dcherian]

doesn't this contradict the error checking above?

[Illviljan]

This makes sure that xarray isn't the one throwing errors before getting to ax.imshow, if the np.issubdtype(v.dtype, str) check hadn't been there.
But it's true it's not really necessary as ax.imshow doesn't seem to support categoricals.

But maybe we prefer matplotlib raising the error?

[dcherian]

np.issubdtype(v.dtype, str) check hadn't been there.

but it is there :) It's redundant, is it not? Shall we just remove the str check from _maybe_center_pixels?

[Illviljan]

The other alternative is to remove the other check and let it crash within matplotlib. But I'm not sure they even intend to support categoricals for these functions.

[dcherian]

I think i like that. If matplotlib starts supporting things, then it will just work automatically in xarray. I also liked your refactoring to _maybe_center_pixels

[Illviljan]

Ok, then I just need to figure out how to catch the errors in the tests.

[Illviljan]

Turns out ax.imshow has some support for strings in the sense that it convert it to integers. So I had to some formatting to get the ticklabels back to strings.

a = xr.DataArray(
    [[0, 1, 2], [3, 4, 5]],
    dims=("dim_0", "dim_1"),
    coords=dict(dim_0=(["dim_0"], [0, 1]), dim_1=(["dim_1"], ["u", "v", "w"])),
)
a.plot.imshow()

[dcherian]

Thanks @Illviljan this is great work!