Pandas API: Signature of UDF methods

There are many signatures of pandas methods that take a UDF. If I've missed methods, please let me know and will update.

Below, other_arg is a placeholder for 1 or more keyword arguments, and value is a placeholder for the default value.

Signature: func, other_arg=value, args=value, **kwargs

• Series.apply
• DataFrame.apply
• Expanding.apply
• Rolling.apply

Signature: func=None, other_arg=value, *args, **kwargs

• Series.agg
• DataFrame.agg

Signature: func, other_arg=value, *args, **kwargs

• Series.transform
• DataFrame.transform
• DataFrameGroupBy.filter

Signature: func, *args, other_arg=Value, **kwargs

• DataFrameGroupBy.transform
• SeriesGroupBy.transform

Signature: func=None, *args, other_arg=Value, **kwargs

• SeriesGroupBy.agg
• DataFrameGroupBy.agg

Signature: func, *args, **kwargs

• GroupBy.apply
• Resampler.aggregate
• Expanding.aggregate
• Resampler.apply

Signature: arg, *args, **kwargs

• Resampler.transform

Comment From: rhshadrach

Two thoughts - Renaming arg to func in Resampler.transform is noncontroversial. - IMO, the most preferred signature is func=None, *args, other_arg=value, **kwargs (where other_arg=value may not occur). With func=None, special handling will need to happen for methods that don't take a dictionary of UDFs as **kwargs (or that feature could be implemented).

As many of these methods are very heavily used, I'm worried about the impact of having a FutureWarning for so many at once. I am wondering if a gradual transition may be more user friendly.

Comment From: attack68

For your list:

Styler.apply(func, axis=0, subset=None, **kwargs)
Styler.applymap(func, subset=None, **kwargs)
Styler.where(cond, value, other=None, subset=None, **kwargs)
Styler.pipe(func, *args, **kwargs)

Comment From: rhshadrach

I'm now of the opinion these signatures should be:

func, args, kwargs, other_arg

Note the args / kwargs are not *args, **kwargs. While this means users will have to pass tuples / dicts to each of these respectively, it also ensures users can reliably pass arbitrary values without interfering with other args / kwargs.