Dataframe loc

[randolf-scholz]

• Closes Simplify DataFrame.loc.__getitem__ overload #574
• Tests added: test_loc_tuple_slice_list

[twoertwein]

I don't think you need HashableT for Iterable as it is covariant: Iterable[Hashable]

[Dr-Irv]

Iterable[HashableT] is too wide, as it will match a plain string, which, if supplied, would return a Series.

So please put back IndexType | MaskType | list[HashableT] and replace Hashable with _IndexSliceTuple | Callable

Including slice is fine.

If you make the suggested change, that will cause the test test_frame.py:test_loc_slice() to fail, but I now realize that the expression used there is ambiguous:

>>> df1 = pd.DataFrame(
...         {"x": [1, 2, 3, 4]},
...         index=pd.MultiIndex.from_product([[1, 2], ["a", "b"]], names=["num", "let"]),
...     )
>>> df1.loc[1, :]
     x
let
a    1
b    2
>>> df2 = pd.DataFrame({"x": [1,2,3,4]}, index=[10, 20, 30, 40])
>>> df2.loc[10, :]
x    1
Name: 10, dtype: int64

So the first argument as an integer could return a DataFrame or Series, dependent on whether the underlying index is a regular Index or MultiIndex

The solution is then to add another overload in _LocIndexerFrame.__getitem__():

    @overload
    def __getitem__(self, idx: tuple[ScalarT, slice]) -> Series | DataFrame: ...

Then modify the test in test_index_slice() to check that the type is Union[pd.Series, pd.DataFrame], and add another test corresponding to df2 above.

[randolf-scholz]

Matching str is fine here, because the second component of the tuple ensures multiple columns are selected.

Both df.loc["row", ["col1", "col2", "col3"]] and df.loc[["r", "o", "w"], ["col1", "col2", "col3"]] return DataFrame which is the only thing this overload ensures.

[Dr-Irv]

Matching str is fine here, because the second component of the tuple ensures multiple columns are selected.

Both df.loc["row", ["col1", "col2", "col3"]] and df.loc[["r", "o", "w"], ["col1", "col2", "col3"]] return DataFrame which is the only thing this overload ensures.

No, you are incorrect. The first example could create a Series:

>>> import pandas as pd
>>> df = pd.DataFrame({"x":[1,2,3], "y":[4,5,6]}, index=["a", "b",
 "c"])
>>> df
   x  y
a  1  4
b  2  5
c  3  6
>>> df.loc["a", ["x", "y"]]
x    1
y    4
Name: a, dtype: int64
>>> type(df.loc["a", ["x", "y"]])
<class 'pandas.core.series.Series'>

[randolf-scholz]

Oh, right.

[randolf-scholz]

Closing since DataFrame.loc["str", columns] automatically coerces to Series.

slice is already implicitly included by IndexType.

[Dr-Irv]

@randolf-scholz should we close #574 as well?

[Dr-Irv]

Closing since DataFrame.loc["str", columns] automatically coerces to Series.

Actually, it could be a DataFrame too:

>>> df = pd.DataFrame({"x":[1,2,3,4]},
...     index=pd.MultiIndex.from_product([["a", "b"], ["c", "d"]]))
>>> df
     x
a c  1
  d  2
b c  3
  d  4
>>> df.loc["a", ["x"]]
   x
c  1
d  2

That's why I suggested making other changes to fix this.

Would you mind creating another issue? (Or I can do it)