Some colormaps in the get_colormaps dictionary are not Colormap instances

[djhoese]

If I recall correctly this was added by @larsoner. In vispy/color/colormap.py there is a _colormaps dictionary that can be accessed via the public get_colormaps(). However, as a user, I expect this dictionary to map colormap names to colormap instances, some of these values though are colormap classes or functions. For example, cubehelix -> CubeHelixColormap and single_hue -> _SingleHue. This is "worked around" by having the get_colormap function create these colormaps with optional kwargs passed to the class.

I'm not a huge fan of this so unless someone can argue for this behavior I would like to propose that we replace these non-instance colormaps with basic instances or remove them from the colormap dictionary returned from get_colormaps.

[larsoner]

fine by me

[kmuehlbauer]

@djhoese I can have a look early next week.

[djhoese]

Here is a copy of the raw dictionary:

_colormaps = dict(
    # Some colormap presets
    autumn=Colormap([(1., 0., 0., 1.), (1., 1., 0., 1.)]),
    blues=Colormap([(1., 1., 1., 1.), (0., 0., 1., 1.)]),
    cool=Colormap([(0., 1., 1., 1.), (1., 0., 1., 1.)]),
    greens=Colormap([(1., 1., 1., 1.), (0., 1., 0., 1.)]),
    reds=Colormap([(1., 1., 1., 1.), (1., 0., 0., 1.)]),
    spring=Colormap([(1., 0., 1., 1.), (1., 1., 0., 1.)]),
    summer=Colormap([(0., .5, .4, 1.), (1., 1., .4, 1.)]),
    fire=_Fire(),
    grays=_Grays(),
    hot=_Hot(),
    ice=_Ice(),
    winter=_Winter(),
    light_blues=_SingleHue(),
    orange=_SingleHue(hue=35),
    viridis=Colormap(ColorArray(_viridis_data)),
    # Diverging presets
    coolwarm=Colormap(ColorArray(
        [
            (226, 0.59, 0.92), (222, 0.44, 0.99), (218, 0.26, 0.97),
            (30, 0.01, 0.87),
            (20, 0.3, 0.96), (15, 0.5, 0.95), (8, 0.66, 0.86)
        ],
        color_space="hsv"
    )),
    PuGr=_Diverging(145, 280, 0.85, 0.30),
    GrBu=_Diverging(255, 133, 0.75, 0.6),
    GrBu_d=_Diverging(255, 133, 0.75, 0.6, "dark"),
    RdBu=_Diverging(220, 20, 0.75, 0.5),

    # Configurable colormaps
    cubehelix=CubeHelixColormap,
    single_hue=_SingleHue,
    hsl=_HSL,
    husl=_HUSL,
    diverging=_Diverging,
    RdYeBuCy=_RedYellowBlueCyan,
)

And visuals like ImageVisual typically run get_colormap(cmap) when a user does my_img.cmap = 'cubehelix' which will automatically generate the instance of the colormap.

[EtienneCmb]

I've a suggestion. If the colormap is not in the _colormaps dictionary and if matplotlib is installed, it might be interesting to build the colormap with Matplotlib? Something like :

if (cmap not in _colormaps) and has_mpl:
    from matplotlib import cm
    # Get the list of implemented colormaps in Matplotlib :
    full_mpl_cmaps = list(cm.datad.keys()) + list(cm.cmaps_listed.keys())
    if cmap not in full_mpl_cmaps:
        raise ValueError("%s colormap not in matplotlib" % cmap)
    
    sc = cm.ScalarMappable(cmap=cmap)
    x = np.arange(256)
    cmap_vec = np.array(sc.to_rgba(x, alpha=1.))
    return Colormap(cmap_vec)

[djhoese]

@EtienneCmb Now that vispy supports texture-based colormaps this should be a lot easier. If you want to make a pull request for this I'd be ok with that. If mpl colormaps need more than 256 colors we can do that too (ex. viridis).