DataFrame.apply stub doesn't reflect default value of axis parameter

[debonte]

Describe the bug
If you call DataFrame.apply and pass parameters beyond f but not including axis, you'll get the following error from Pyright:

No overloads for "apply" match the provided arguments

To Reproduce

import pandas as pd

def gethead(s, y):
    return s.head(y)

df = pd.DataFrame({"A": [1, 2, 3, 4, 5, 6], "B": [11, 12, 13, 14, 15, 16]})

df2 = df.apply(gethead, args=tuple([4]))  # error

Pylance gives the following error on the apply call:

No overloads for "apply" match the provided arguments
  Argument types: ((s: Unknown, y: Unknown) -> Unknown, tuple[int, ...])Pylance[reportGeneralTypeIssues](https://github.com/microsoft/pylance-release/blob/main/DIAGNOSTIC_SEVERITY_RULES.md#diagnostic-severity-rules)

Please complete the following information:

• OS: Windows
• OS Version: 10.0.25197 Build 25197
• python version: 3.10.0
• version of type checker: Pylance 2022.10.21
• version of installed pandas-stubs: 1.5.0.220926

Additional context
Originally reported by @jonmooser at microsoft/pylance-release#3491

I was originally thinking that the fix was to indicate in the stubs that axis has a default value (axis: AxisType = ...), but then the two overloads of apply overlap with each other. It was looking like the correct fix would require more knowledge of the innards of apply and Pandas in general than I have.

[Dr-Irv]

The trick here is that the function gethead() has to be properly typed to return a Series, and if default arguments are used, then we should return a DataFrame when we know that the function returns a Series. See the description of the result_type argument which describes the default behavior.

So for your example to work, it would have to change, and we have to add an overload for apply()

[jonmooser]

See the description of the result_type argument which describes the default behavior.

Dr-Irv,
Could you clarify the significance of the result_type argument? The doc says that only applies when axis=1. In my example axis=0 (implicitly)

[Dr-Irv]

Dr-Irv, Could you clarify the significance of the result_type argument? The doc says that only applies when axis=1. In my example axis=0 (implicitly)

I think (but not 100% sure) that the comment about axis=1 means that the values that are not None only take effect when axis=1, but do nothing when axis=0. But the default value of result_type==None does change the type of the result independent of the value of axis.

[jonmooser]

I think (but not 100% sure) that the comment about axis=1 means that the values that are not None only take effect when axis=1, but do nothing when axis=0. But the default value of result_type==None does change the type of the result independent of the value of axis.

Based on some experimenting, that's correct. If the passed function returns a scalar, apply() will create a Series. If the function returns something list-like, apply() will create a DataFrame.

But I think the underlying problem here may be a bit different. None of the overloads allows for an args param with no axis param.

[Dr-Irv]

But I think the underlying problem here may be a bit different. None of the overloads allows for an args param with no axis param.

We need to have a new overload, where the return type is based on whether the function f is Callable[..., ScalarT] or Callable[..., list-like], where list-like is appropriately defined.

[Dr-Irv]

fixed in #401