ENH: Allow args to be specified for the pivot_table aggfunc

[j-kapp]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

Currently the pivot_table has an aggfunc parameter which is used to do a groupby aggregation here. However, no additional arguments can be passed into that agg call. I'm specifically referring to the *args which can be specified in (df/series) groupby.agg function. It would be very useful if pivot_table could accept additional arguments for the aggfunc.

Feature Description

I'm not 100% sure, but I think it would be something like this:

1. Add a **aggfunc_args or aggfunc_args: dict parameter to the pivot_table function.
2. Do the same for the __internal_pivot_table function
3. Change the line here, from agged = grouped.agg(aggfunc) to agged = grouped.agg(aggfunc, **aggfunc_args).
4. Update the docs

Alternative Solutions

The same functionality can currently be achieved by specifying a custom function as aggfunc, but using that is much slower. My use case is pretty much the same as this.
Instead of using pd.pivot_table(... , aggfunc=lambda x: x.sum(min_count=1)), I would like to be able to do pd.pivot_table(... , aggfunc=sum, min_count=1) or similiar.

Additional Context

No response

[rhshadrach]

Thanks for the request - it makes sense to me to support passing through kwargs here. The only question is how:

def pivot_table(self, ..., **kwargs)
def pivot_table(self, ..., kwargs)
def pivot_table(self, ..., args, kwargs)

I do not think we should do

def pivot_table(self, *args, ..., **kwargs)

If we want to agree with the current direction of apply/agg/transform, it would be the first of the above three options. Ref: #40112

[PF2100]

take

[PF2100]

Hi, regarding that first option, would passing positional arguments to the aggfunc not be supported?
I must say I agree on the signature you proposed on the referenced issue ( func, args, kwargs, other_arg, sort of like the third option here) being the best overall, however, it seems like this is not the current standard for pandas.

[PF2100]

take

[PF2100]

Hi, regarding that first option, would passing positional arguments to the aggfunc not be supported? I must say I agree on the signature you proposed on the referenced issue ( func, args, kwargs, other_arg, sort of like the third option here) being the best overall, however, it seems like this is not the current standard for pandas.

@rhshadrach Sorry to bother you, but could you share your thoughts on this matter?

[rhshadrach]

@PF2100 - thanks for the ping!

I must say I agree on the signature you proposed on the referenced issue ( func, args, kwargs, other_arg, sort of like the third option here) being the best overall, however, it seems like this is not the current standard for pandas.

Same here, but for consistency, it seems best to go with (1) for now. Changing from **kwargs to kwargs is likely to be quite noisy or users and I don't feel confident it will get enough support.

regarding that first option, would passing positional arguments to the aggfunc not be supported?

Correct.