The column keyword in the dataset method was not documented yet 'cos I can't figure out what it does:

import hvplot.pandas
import pandas as pd

data = pd.DataFrame({"x": ['a', 'b', 'c'], "y": [1, 2, 3]})

ds1 = data.hvplot.dataset()
ds2 = data.hvplot.dataset(columns='x')

# Check the underlying DataFrame columns
print("Columns in ds1:", ds1.data.columns)
print("Columns in ds2:", ds2.data.columns)
print("kdims in ds1:", ds1.kdims)
print("kdims in ds2:", ds2.kdims)
print("vdims in ds1:", ds1.vdims)
print("vdims in ds2:", ds2.vdims)

output

Columns in ds1: Index(['x', 'y'], dtype='object')
Columns in ds2: Index(['x', 'y'], dtype='object')
kdims in ds1: [Dimension('x'), Dimension('y')]
kdims in ds2: [Dimension('x'), Dimension('y')]
vdims in ds1: []
vdims in ds2: []

@maximlt

Yep I'm also unsure. Do you know if this line is ever executed part of the test suite?

            return redim_(Dataset(data, self.kwds.get('columns')), **self._redim)

Something else to note is that these methods (e.g. hvPlotTabular.bar) are only called in these cases:

• import hvplot.pandas; df.hvplot.bar(...)
• from hvplot import hvPlot(Tabular); hvPlot(Tabular)(df).bar(...)

They're not called in these other cases:

• from hvplot import hvPlot(Tabular); hvPlot(Tabular)(kind='bar', ...)
• import hvplot.pandas; df.hvplot(kind='bar', ...)
• pd.options.ploting.backend = 'hvplot'; df.plot.bar(...)
• pd.options.ploting.backend = 'hvplot'; df.plot(kind='bar', ...)

By adding new keywords to their signature with their defaults, we enforce the defaults when these methods are called but not in all the other cases, where the defaults are either set in the converter class (e.g. levels = self.kwds.get('levels', 5)) or inherited from whatever default HoloViews has set. So I think that separately, we need to find a way to make sure the defaults are always the same, however a plot is built. Maybe by expanding HoloViewsConverter._kind_options from a dict mapping a kind to a list to a dict mapping a kind to a dict whose values are the defaults:

    _kind_options = {
        'area': {'y2': <default>, 'stacked': <default>},
        ...
}

If we do that, we should also add a test that ensures the defaults defined in the signature and the converter are the same, so they don't drift over time.

Yep I'm also unsure. Do you know if this line is ever executed part of the test suite?

            return redim_(Dataset(data, self.kwds.get('columns')), **self._redim)

No it's not

Try to replace it with raise ValueError and run pixi run test-example. Examples don't use test coverage.

@Azaya89 I think you have made more changes locally; can you push them please?