Holoviews: Use discrete colorbars for contours and polygons

[rabernat]

If I take the simple contour example from the docs and add a filled=True option, the %%opts seem to have no effect. I can no longer change the cmap or add a colorbar:

%%opts Contours [show_legend=False colorbar=True width=325] (cmap='fire')
x,y = np.mgrid[-50:51, -50:51] * 0.05
img = hv.Image(np.sin(x**2+y**3))
z0, z1 = img.range('z')
hv.operation.contours(img, levels=np.linspace(z0, z1, 5), overlaid=False, filled=True)

Holoviews version is 1.8.3-x-ga66181c36 via conda-forge.

Possibly related to #1558

[philippjfr]

This is probably somewhat confusing but at least with latest master this works correctly if you set the options on Polygons, Contours is just used for the non-filled case:

%%opts Polygons [show_legend=False colorbar=True width=325] (cmap='fire' line_width=0.1)
x,y = np.mgrid[-50:51, -50:51] * 0.05
img = hv.Image(np.sin(x**2+y**3))
z0, z1 = img.range('z')
hv.operation.contours(img, levels=np.linspace(z0, z1, 10), overlaid=False, filled=True)

[jlstevens]

That looks pretty! Might be one to add to the gallery...

[rabernat]

Great work!

My only comment would be that, with contour plots with discrete levels, we like to see discrete colorbars. This helps to read the precise values from the plot.

In matplotlib it looks like this:

[philippjfr]

Good point, we need to improve our handling for discrete colormaps in general.

[jlstevens]

Perhaps we should repurpose this issue to be about discrete colormaps?

[philippjfr]

We now support a color_levels argument for discrete color mapping.