pd.DataFrame.take should allow np.ndarray indices (current stub only accepts list)

[ritsuki1227]

The following code executes without any runtime error, but static type checkers such as mypy flag it as invalid:

import numpy as np
import pandas as pd

pd.DataFrame({"a": [1, 2, 3]}).take(np.array([0, 1]))

In pandas-stubs, the signature for DataFrame.take currently restricts indices to list:

def take(self, indices: list, axis: Axis = ..., **kwargs) -> Self: ...

However, the actual pandas implementation clearly accepts a NumPy array (and other array-like objects). For reference, the stub for _iLocIndexerFrame already accommodates NumPy arrays.

class _iLocIndexerFrame(_iLocIndexer, Generic[_T]):
    @overload
    def __getitem__(
        self,
        idx: (
            IndexType
            | MaskType
            | tuple[IndexType | MaskType, IndexType | MaskType]
            | tuple[slice]
        ),
    ) -> _T: ...

It would be helpful if the take method signature could be fixed!

[Dr-Irv]

It should be TakeIndexer, which is defined in _typing.pyi. PR with tests welcome.

Ideally, there should be some reconciliation done between the definitions of take() in series.pyi and frame.pyi, and if they are supposed to be the same, removed from those 2 files and placed in generic.pyi in NDFrame.