feat(bokeh): Server side HoverTool for rasterized/datashaded plots with selector

[hoxbro]

Previously, we sent all hover data to the front end, which could lead to significant overhead on larger plots. With this PR, the data is pushed server-side to the hover tool with a custom data model.

script

import itertools

import datashader as ds
import numpy as np
import pandas as pd
import panel as pn

import holoviews as hv
from holoviews.operation.datashader import datashade, dynspread, rasterize

hv.extension("bokeh")

num = 10000
seed = np.random.default_rng(1)


dists = {
    cat: pd.DataFrame(
        {
            "x": seed.normal(x, s, num),
            "y": seed.normal(y, s, num),
            "s": s,
            "val": val,
            "cat": cat,
        }
    )
    for x, y, s, val, cat in [
        (2, 2, 0.03, 0, "d1"),
        (2, -2, 0.10, 1, "d2"),
        (-2, -2, 0.50, 2, "d3"),
        (-2, 2, 1.00, 3, "d4"),
        (0, 0, 3.00, 4, "d5"),
    ]
}

df = pd.concat(dists, ignore_index=True)
df["y"] += 100  # Offset to distinguish x from y
points = hv.Points(df)


def dynspread_datashade(*args, **kwargs):
    return dynspread(datashade(*args, **kwargs))


def dynspread_rasterize(*args, **kwargs):
    return dynspread(rasterize(*args, **kwargs))


ops = (rasterize, dynspread_rasterize, datashade, dynspread_datashade)
aggs = (ds.min("s"), ds.by("cat"))
sels = (None, ds.min("s"))
# ops = (datashade,)
# aggs = (ds.by("cat"),)
# sels = (None,)
combinations = itertools.product(ops, aggs, sels)
plots = []
for op, agg, sel in combinations:
    title = f"{op.__name__.replace('_','+')}(agg={type(agg).__name__}, sel={type(sel).__name__ if sel else None})"
    plot = op(points, aggregator=agg, selector=sel).opts(
        tools=["hover"], title=title, width=400, height=400
    )
    plots.append(plot)

pn.panel(hv.Layout(plots).cols(4).opts(shared_axes=False)).servable()

[codecov]

Codecov Report

Attention: Patch coverage is 87.45387% with 34 lines in your changes missing coverage. Please review.

Project coverage is 88.49%. Comparing base (ee19136) to head (e818c4d).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
holoviews/tests/ui/bokeh/test_hover.py	48.78%	21 Missing ⚠️
holoviews/plotting/bokeh/raster.py	87.50%	7 Missing ⚠️
holoviews/operation/datashader.py	95.41%	5 Missing ⚠️
holoviews/element/raster.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6422      +/-   ##
==========================================
- Coverage   88.50%   88.49%   -0.02%     
==========================================
  Files         323      323              
  Lines       68631    68819     +188     
==========================================
+ Hits        60741    60899     +158     
- Misses       7890     7920      +30     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[hoxbro]

This is the last thing that I need to look into. RGBA gives a non-zero value when alpha is zero. This was a bug before this PR, though.

[hoxbro]

I have resolved this in 9e71910. I think it is correct to set it to 0 when alpha is zero, but it is likely to be a somewhat subjective opinion.

[hoxbro]

Note: Image is no longer cloned by default.

[hoxbro]

Don't understand this.

[hoxbro]

Likely we can just check the first value here.

[hoxbro]

We could make it so it does not return null, but a random point.