Bokeh and HoloViews datashader support to-do list

[jbednar]

Datashader can make complete images on its own, but it's a lot more powerful when combined with an external plotting program like Bokeh, Plotly, or Matplotlib that can draw axes, add additional types of data as overlays or adjacent plots, provide interactivity such as hovering or click detection, add colorbars and color keys, etc. But these programs don't typically provide the special colormapping and related features of datashader, so there's still a long to-do list of tasks across libraries that would improve integration. This issue focuses on issues for the Bokeh backend of Holoviews, most of which will require adding functionality to Bokeh that would then be exposed in HoloViews if needed.

From datashade_vs_rasterize:

• Shading normalization other than linear, log (hist_eq, cbrt, custom; see Implement count_cat and custom colormapping in BokehJS #609)
• Colorbars for counts that include zero (bokeh #8061) (or else avoid the issue using NaN counts #627, which also addresses Allow integer aggregate values of zero #142)
• Spreading (spread, dynspread)
• Stacking (add, over, saturate, source)
• Shading value as alpha (passing a single color to cmap)
• Shading categories as mixed colors (with aggregator=count_cat(); see Implement count_cat and custom colormapping in BokehJS #609)

Once the above is done, there are various Datashader issues that can be closed, such as requests for hover information, legends, colorbars, and interactivity (#126, #127, #136, #251).