Set rescale_discrete_levels=True by default

[jbednar]

The eq_hist colormapper used with Datashader is designed for large datasets with a wide distribution of values, but as documented in holoviz/datashader#357 the behavior when there are only a few distinct value levels has always been confusing and misleading. Datashader 0.14 adds an option rescale_discrete_levels=True, now supported in HoloViews, allowing users to avoid this issue. However, this option is False by default, and most users are never going to be aware that the option exists. This PR enables the option by default, which has good and bad consequences:

• Good: Users will now see better colormapping behavior, with points visible at all zoom levels and no longer jumping between the lowest and highest colormap values at a tiny change in zoom level, without having to learn about this option or modify any of their code. This is arguably a long-overdue bug fix.
• Bad: Results will change for plots rendered with eq_hist that have only a small number of discrete levels. That's by design, but it is strictly a compatibility change, which is odd to introduce in a patch release.
• Bad: Results will differ between rasterize and datashade, because Bokeh does not yet support rescale_discrete_levels. So there is some argument for waiting until Bokeh does support it to make this the default.

Basically, if we consider this a bugfix, we should introduce it now; patch releases are for bugfixes. If we consider it a new feature, we should wait until the next major or at least minor release to enable it by default.

I'm ok with either option, as long as the options have been duly considered and decided between.

[codecov-commenter]

Codecov Report

Merging #5268 (f058770) into master (c265cc1) will decrease coverage by 0.11%.
The diff coverage is 16.66%.

@@            Coverage Diff             @@
##           master    #5268      +/-   ##
==========================================
- Coverage   86.74%   86.62%   -0.12%     
==========================================
  Files         299      299              
  Lines       62218    62281      +63     
==========================================
- Hits        53969    53950      -19     
- Misses       8249     8331      +82     

Impacted Files	Coverage Δ
holoviews/plotting/mpl/util.py	76.12% <3.22%> (-1.35%)	⬇️
holoviews/operation/datashader.py	78.33% <100.00%> (-1.72%)	⬇️
holoviews/plotting/mpl/element.py	89.95% <100.00%> (-0.07%)	⬇️
holoviews/plotting/mpl/raster.py	78.52% <0.00%> (-3.51%)	⬇️
holoviews/plotting/util.py	69.49% <0.00%> (-2.06%)	⬇️
holoviews/plotting/bokeh/raster.py	71.30% <0.00%> (-0.95%)	⬇️
holoviews/tests/plotting/bokeh/test_callbacks.py	97.82% <0.00%> (-0.30%)	⬇️
holoviews/tests/plotting/bokeh/test_server.py	97.59% <0.00%> (-0.17%)	⬇️
holoviews/plotting/bokeh/callbacks.py	59.60% <0.00%> (-0.15%)	⬇️
... and 4 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c265cc1...f058770. Read the comment docs.

[maximlt]

@jbednar should this also be merged before 1.15.0?

[jbednar]

Yes

[jbednar]

Yes

[jbednar]

Bokeh 2.4.3 EqHistColorMapper now supports rescale_discrete_levels, so I think this PR should now be merged. Do we need to do anything so that the same default applies in the Bokeh case?

[philippjfr]

Already seems to be true in the bokeh case, but really we need to implement it for matplotlib.

[jbednar]

The MPL changes look good, thanks!

[github-actions]

This pull request has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.