BUG: Fix return type of loc/iloc (#61054)

* Fix loc dtype

Co-authored-by: Parthi <[email protected]>

* Modify test_loc_setitem_frame_mixed_labels accordingly

* Add test for the linked issue

* Fix test_loc_dtype

* Fix dtype inference

* Reverse tuple to avoid unintended dtype inference

* Reverse axis order

* Reverse indexing order

* Explicitly assign the expected result

* Add whatsnew

---------

Co-authored-by: Parthi <[email protected]>

Author: sanggon6107

doc/source/whatsnew/v3.0.0.rst

@@ -748,6 +748,7 @@ Indexing
 - Bug in :meth:`DataFrame.__getitem__` returning modified columns when called with ``slice`` in Python 3.12 (:issue:`57500`)
 - Bug in :meth:`DataFrame.__getitem__` when slicing a :class:`DataFrame` with many rows raised an ``OverflowError`` (:issue:`59531`)
 - Bug in :meth:`DataFrame.from_records` throwing a ``ValueError`` when passed an empty list in ``index`` (:issue:`58594`)
+- Bug in :meth:`DataFrame.loc` and :meth:`DataFrame.iloc` returning incorrect dtype when selecting from a :class:`DataFrame` with mixed data types. (:issue:`60600`)
 - Bug in :meth:`DataFrame.loc` with inconsistent behavior of loc-set with 2 given indexes to Series (:issue:`59933`)
 - Bug in :meth:`Index.get_indexer` and similar methods when ``NaN`` is located at or after position 128 (:issue:`58924`)
 - Bug in :meth:`MultiIndex.insert` when a new value inserted to a datetime-like level gets cast to ``NaT`` and fails indexing (:issue:`60388`)

pandas/core/indexing.py

@@ -1066,8 +1066,10 @@ def _getitem_lowerdim(self, tup: tuple):
 
         tup = self._validate_key_length(tup)
 
-        for i, key in enumerate(tup):
-            if is_label_like(key):
+        # Reverse tuple so that we are indexing along columns before rows
+        # and avoid unintended dtype inference. # GH60600
+        for i, key in zip(range(len(tup) - 1, -1, -1), reversed(tup)):
+            if is_label_like(key) or is_list_like(key):
                 # We don't need to check for tuples here because those are
                 #  caught by the _is_nested_tuple_indexer check above.
                 section = self._getitem_axis(key, axis=i)

pandas/tests/indexing/multiindex/test_loc.py

@@ -757,7 +757,7 @@ def test_missing_keys_raises_keyerror(self):
         df = DataFrame(np.arange(12).reshape(4, 3), columns=["A", "B", "C"])
         df2 = df.set_index(["A", "B"])
 
-        with pytest.raises(KeyError, match="1"):
+        with pytest.raises(KeyError, match="6"):
             df2.loc[(1, 6)]
 
     def test_missing_key_raises_keyerror2(self):

pandas/tests/indexing/test_loc.py

@@ -60,6 +60,14 @@ def test_not_change_nan_loc(series, new_series, expected_ser):
     tm.assert_frame_equal(df.notna(), ~expected)
 
 
+def test_loc_dtype():
+    # GH 60600
+    df = DataFrame([["a", 1.0, 2.0], ["b", 3.0, 4.0]])
+    result = df.loc[0, [1, 2]]
+    expected = Series([1.0, 2.0], index=[1, 2], dtype=float, name=0)
+    tm.assert_series_equal(result, expected)
+
+
 class TestLoc:
     def test_none_values_on_string_columns(self, using_infer_string):
         # Issue #32218
@@ -441,7 +449,7 @@ def test_loc_to_fail(self):
 
         msg = (
             rf"\"None of \[Index\(\[1, 2\], dtype='{np.dtype(int)}'\)\] are "
-            r"in the \[index\]\""
+            r"in the \[columns\]\""
         )
         with pytest.raises(KeyError, match=msg):
             df.loc[[1, 2], [1, 2]]
@@ -807,7 +815,7 @@ def test_loc_setitem_frame_mixed_labels(self):
 
         result = df.loc[0, [1, 2]]
         expected = Series(
-            [1, 3], index=Index([1, 2], dtype=object), dtype=object, name=0
+            [1, 3], index=Index([1, 2], dtype=object), dtype="int64", name=0
         )
         tm.assert_series_equal(result, expected)