Can't set face color to property name with empty point set

🐛 Bug

Setting Points.face_color as a property/feature fails with an empty point set - the error is not seen when a single point is present.

  File "/Users/aburt/mambaforge/envs/SPB_relion40/lib/python3.9/site-packages/vispy/color/color_array.py", line 23, in _string_to_rgb
    raise ValueError('Color "%s" unknown' % color)
ValueError: Color "my_awesome_feature" unknown

cc @andy-sweet as master of features 🙂

To Reproduce

import napari

viewer = napari.Viewer()
face_color_cycle = ['blue', 'green']

points_layer = viewer.add_points(
    [],
    features={'my_awesome_feature': []},
    face_color='my_awesome_feature',
    face_color_cycle=face_color_cycle,
)
napari.run()

Expected behavior

Non-failure, new points added will follow the color cycle

Environment

0.4.15rc1

Additional context

temporary workaround:

points_layer = viewer.add_points(
    [0, 0, 0],
    features={'my_awesome_feature': [0]},
    face_color='my_awesome_feature',
    face_color_cycle=face_color_cycle,
)

points_layer.selected_data = {0}
points_layer.remove_selected()

[andy-sweet]

Thanks for the clear write up @alisterburt !

An alternative workaround is to provide the (not yet deprecated) property_choices, like follows.

    points_layer = viewer.add_points(
        [],
        property_choices={'my_awesome_feature': ['great', 'perfect']},
        face_color='my_awesome_feature',
        face_color_cycle=['blue', 'green'],
    )

I think you had to do that before features was introduced, as it’s what was introduced in the tests and also conveys the main intent of property_choices as I understand it.

That will give my_awesome_feature a categorical dtype in the underlying dataframe, which might be OK for some usage (i.e. if you know all the possible values ahead of time) of a feature-based color cycle, but not others (e.g. you want to specify a cycle for an unbound integer value cycle).

Equivalently, you set up the categorical dtype explicitly with features.

    features = pd.DataFrame({
        'my_awesome_feature': pd.Series([], dtype=pd.CategoricalDtype(['great', 'perfect'])),
    })
    points_layer = viewer.add_points(
        [],
        features=features,
        face_color='my_awesome_feature',
        face_color_cycle=['blue', 'green'],
    )

I see the bug here as how to evaluate the default color when there are no data yet. We could fix ColorManager now with features or we could consider waiting until ColorManager is replaced.

[andy-sweet]

Just as a reference, this bug should be closed when we complete the Use ColorEncoding for Points task in the properties project board: https://github.com/napari/napari/projects/11