Allow and handle non-parameters in rasterize.instance()

[ahuang11]

I don't fully understand this...

If I use HoloViews alone / try to reproduce with HoloViews:

import numpy as np
import pandas as pd
import holoviews as hv
hv.extension("bokeh")
from holoviews.operation.datashader import rasterize

df = pd.DataFrame(
    np.random.multivariate_normal((0, 0), [[0.1, 0.1], [0.1, 1.0]], (500000,))
)
df.columns = ["a", "b"]

curve = hv.Curve(df, kdims=["a"], vdims=["b"]).opts("Curve", line_width=1)
hv.operation.apply_when(curve, operation=rasterize, predicate=lambda x: len(x) > 1000)

It runs fine...

However, if I use hvplot with this PR:

import numpy as np
import pandas as pd
import hvplot.pandas

df = pd.DataFrame(
    np.random.multivariate_normal((0, 0), [[0.1, 0.1], [0.1, 1.0]], (500000,))
)
df.columns = ["a", "b"]
df.hvplot("a", "b", rasterize=True, aggregation_threshold=1000)

I get the following traceback:

File [~/repos/holoviews/holoviews/operation/datashader.py:1389](https://file+.vscode-resource.vscode-cdn.net/Users/ahuang/repos/sketches/~/repos/holoviews/holoviews/operation/datashader.py:1389), in rasterize._process(self, element, key)
   1387 all_allowed_kws = set()
   1388 all_supplied_kws = set()
-> 1389 instance_params = {
   1390     k: getattr(self, k) for k in self.__instance_params
   1391 }
   1392 for predicate, transform in self._transforms:
   1393     merged_param_values = dict(instance_params, **self.p)

File [~/repos/holoviews/holoviews/operation/datashader.py:1390](https://file+.vscode-resource.vscode-cdn.net/Users/ahuang/repos/sketches/~/repos/holoviews/holoviews/operation/datashader.py:1390), in (.0)
   1387 all_allowed_kws = set()
   1388 all_supplied_kws = set()
   1389 instance_params = {
-> 1390     k: getattr(self, k) for k in self.__instance_params
   1391 }
   1392 for predicate, transform in self._transforms:
   1393     merged_param_values = dict(instance_params, **self.p)

AttributeError: 'rasterize' object has no attribute 'line_width'

[codecov-commenter]

Codecov Report

Merging #5811 (ff26cf6) into main (e83355c) will increase coverage by 0.04%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main    #5811      +/-   ##
==========================================
+ Coverage   88.17%   88.22%   +0.04%     
==========================================
  Files         307      307              
  Lines       62908    62983      +75     
==========================================
+ Hits        55471    55565      +94     
+ Misses       7437     7418      -19     

Flag	Coverage Δ
ui-tests	22.39% <19.04%> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
holoviews/operation/datashader.py	83.38% <100.00%> (+0.15%)	⬆️
holoviews/tests/operation/test_datashader.py	97.45% <100.00%> (+0.04%)	⬆️

... and 9 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[hoxbro]

This is a MRE:

import holoviews as hv
import numpy as np
import pandas as pd
from holoviews.operation.datashader import rasterize

hv.extension("bokeh")

df = pd.DataFrame(
    np.random.multivariate_normal((0, 0), [[0.1, 0.1], [0.1, 1.0]], (500000,))
)
df.columns = ["a", "b"]

curve = hv.Curve(df, kdims=["a"], vdims=["b"])
inst = rasterize.instance(line_width=2)
hv.operation.apply_when(curve, operation=inst, predicate=lambda x: len(x) > 1000)

__instance_params was added in #5767, and by the look of how it worked before this fix should be okay.

Is it possible to add a small unit test?

[hoxbro]

I left some comments, but maybe it is correct behavior to error out here.

When using .instance() and one of the inputs is not a parameter (here line_width), it is not using it. Beforehand this will just silently fail, but with the recent changes, it gives an error, which I'm leaning toward being a good thing.

Take a look at these examples (with your changes / the behavior before #5767):

import holoviews as hv
import numpy as np
import pandas as pd
from holoviews.operation.datashader import rasterize

hv.extension("bokeh")
np.random.seed(10)

df = pd.DataFrame(
    np.random.multivariate_normal((0, 0), [[0.1, 0.1], [0.1, 1.0]], (500000,)),
    columns=["a", "b"],
)
curve = hv.Curve(df, kdims=["a"], vdims=["b"])

rasterize.instance(line_width=2)(curve).opts(xlim=(1.1, 1.2), ylim=(1.1, 1.2))
rasterize(curve, line_width=2).opts(xlim=(1.1, 1.2), ylim=(1.1, 1.2))

So maybe in the hvplot code, we should use a partial function for the operation:
hv.operation.apply_when(curve, operation=partial(rasterize, line_width=2), predicate=lambda x: len(x) > 1000)

[hoxbro]

I have updated the code to save non-parameter in the instance creation, so it works as one would expect:

[philippjfr]

Looks great, thanks!