ENH: Default negative location in pandas insert

[Eschivo]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

In dataframe insert why not change the loc argument to a keyword argument with a default value of -1 ? Since in many cases I don't care where the column will be positioned my idea is that if the loc argument has a negative value (as in the maybe future default case) the column will be inserted as the last column of the dataset, i.e. if loc<0: loc=len(columns) . Then is also possible to remove the constraint 0 <= loc .

Feature Description

In pandas/core/frame.py insert definition

def insert(
        self,
        column: Hashable,
        value: Scalar | AnyArrayLike,
        loc: int = -1,
        allow_duplicates: bool | lib.NoDefault = lib.no_default,
    ) -> None:
...
    if not isinstance(loc, int):
            raise TypeError("loc must be int")
    
    if loc < 0:
        loc = len(self.columns)

    value = self._sanitize_column(value)
...

Alternative Solutions

I don't see any existing functionality or third-party package that can solve this issue

Additional Context

Usage example after the implementation:

>>> df = pd.DataFrame({'col1': [1, 2], 'col2': [3, 4]})
>>> df
   col1  col2
0     1     3
1     2     4
>>> df.insert("newcol", [99, 99])
>>> df
   col1  col2  newcol
0     1     3     99
1     2     4     99

If you find this feature useful I can try to implement it.

[jenniferclarsson]

take

[topper-123]

I think this would be a better API, but it is backward incompatible, unfortunately.

Could be done by wrapping the method in a deprecate_nonkeyword_arguments in pandas v2.x, then expediate this in pandas v3.0.

[Eschivo]

Thank you for replying! I don't know how to handle the deprecation cycle you mentioned. If you think that this edit is worthwhile, I can try to implement it (with a bit of guidance)

[topper-123]

If we were to do this, we could do it like this in pandas 2.x:

@deprecate_nonkeyword_arguments(version=None, allowed_args=[])
def insert(
    self,
    loc: int,
    column: Hashable,
    value: Scalar | AnyArrayLike,
    allow_duplicates: bool | lib.NoDefault = lib.no_default,
) -> None:
...

Then in pandas 3.0 change the signature to:

def insert(
    self,
    column: Hashable,
    value: Scalar | AnyArrayLike,
    *,
    loc: int = -1,
    allow_duplicates: bool | lib.NoDefault = lib.no_default,
) -> None:
...

This will somewhat disruptive for the 2.x cycle, so IDK if other core devs would prefer to keep the current signature?

[topper-123]

@pandas-dev/pandas-core.

[bashtage]

I am -1 on this for two reasons:

• insert is very similar to list.insert aside from the extra column parameter.
• The canonical way to add a new column is just df["new_column_name"] = value. Having df.insert("new_column_name", value) do this same by default seems redundant. The point of insert is for when you want columns in a particular order, e.g., for exporting to a CSV file with a particular structure.

[rhshadrach]

Agreed with @bashtage; I don't see the value of changing this.

[Dr-Irv]

I'm -1 as well. Based on the example, the easy way to accomplish this using DataFrame.insert() is with:

>>> df = pd.DataFrame({'col1': [1, 2], 'col2': [3, 4]})
>>> df
   col1  col2
0     1     3
1     2     4
>>> df.insert(len(df.columns), "newcol", [99, 99])
>>> df
   col1  col2  newcol
0     1     3      99
1     2     4      99

[topper-123]

Ok, there's probably not support for this change, but I'll leave it open a while more, to see if it facilitates more discussion