As discussed in #3636, I propose a thin wrapper around the already present capability of dim transforms to handle custom functions with multiple input dimensions.

I used pipe as name for the new class method, inspired by https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.pipe.html

The following is now possible (example a bit convoluted for the sake of demonstration):

import numpy as np
import pandas as pd
import holoviews as hv
from holoviews import dim

df = pd.DataFrame(np.random.rand(10, 3), columns=['x', 'y', 'z']).cumsum()
e = hv.Scatter(df, 'x', ['y', 'z'])

def func(x, y, z, u, v=None):
    return u+z*np.polyfit(x, y, 1)[1]

dim.pipe(func, 'x', 'y', 10, np.random.rand(10)).max().apply(e)

Edit: Although I start wondering whether it wouldn't be more general to allow passing str arguments instead of transforming them into dims by default. Perhaps a keyword dim_as_str=True could help?