BUG: groupby.describe on a frame with duplicate column names (#50846)

* REF: groupby Series selection with as_index=False

* GH#

* type-hinting fixes

* WIP

* WIP

* WIP

* BUG: groupby.describe on a frame with duplicate column names

* cleanup

* test fixup

* Fix type-hint for _group_selection

* Merge branch 'groupby_select_obj_dup_cols' of https://github.com/rhshadrach/pandas into groupby_select_obj_dup_cols

# Conflicts:
#	pandas/core/groupby/groupby.py

* Speedup

* refinement

* cleanup, faster implementation

* Make group_selection a Boolean flag

* Avoid resetting cache

* Improve test

* Rework test

Author: rhshadrach

doc/source/whatsnew/v2.0.0.rst

@@ -1263,6 +1263,7 @@ Groupby/resample/rolling
 - Bug in :meth:`.SeriesGroupBy.value_counts` did not respect ``sort=False`` (:issue:`50482`)
 - Bug in :meth:`.DataFrameGroupBy.resample` raises ``KeyError`` when getting the result from a key list when resampling on time index (:issue:`50840`)
 - Bug in :meth:`.DataFrameGroupBy.transform` and :meth:`.SeriesGroupBy.transform` would raise incorrectly when grouper had ``axis=1`` for ``"ngroup"`` argument (:issue:`45986`)
+- Bug in :meth:`.DataFrameGroupBy.describe` produced incorrect results when data had duplicate columns (:issue:`50806`)
 -
 
 Reshaping

pandas/tests/groupby/test_function.py

@@ -1256,6 +1256,27 @@ def test_describe_with_duplicate_output_column_names(as_index, keys):
     tm.assert_frame_equal(result, expected)
 
 
+def test_describe_duplicate_columns():
+    # GH#50806
+    df = DataFrame([[0, 1, 2, 3]])
+    df.columns = [0, 1, 2, 0]
+    gb = df.groupby(df[1])
+    result = gb.describe(percentiles=[])
+
+    columns = ["count", "mean", "std", "min", "50%", "max"]
+    frames = [
+        DataFrame([[1.0, val, np.nan, val, val, val]], index=[1], columns=columns)
+        for val in (0.0, 2.0, 3.0)
+    ]
+    expected = pd.concat(frames, axis=1)
+    expected.columns = MultiIndex(
+        levels=[[0, 2], columns],
+        codes=[6 * [0] + 6 * [1] + 6 * [0], 3 * list(range(6))],
+    )
+    expected.index.names = [1]
+    tm.assert_frame_equal(result, expected)
+
+
 def test_groupby_mean_no_overflow():
     # Regression test for (#22487)
     df = DataFrame(
@@ -1596,3 +1617,22 @@ def test_multiindex_group_all_columns_when_empty(groupby_func):
     result = method(*args).index
     expected = df.index
     tm.assert_index_equal(result, expected)
+
+
+def test_duplicate_columns(request, groupby_func, as_index):
+    # GH#50806
+    if groupby_func == "corrwith":
+        msg = "GH#50845 - corrwith fails when there are duplicate columns"
+        request.node.add_marker(pytest.mark.xfail(reason=msg))
+    df = DataFrame([[1, 3, 6], [1, 4, 7], [2, 5, 8]], columns=list("abb"))
+    args = get_groupby_method_args(groupby_func, df)
+    gb = df.groupby("a", as_index=as_index)
+    result = getattr(gb, groupby_func)(*args)
+
+    expected_df = df.set_axis(["a", "b", "c"], axis=1)
+    expected_args = get_groupby_method_args(groupby_func, expected_df)
+    expected_gb = expected_df.groupby("a", as_index=as_index)
+    expected = getattr(expected_gb, groupby_func)(*expected_args)
+    if groupby_func not in ("size", "ngroup", "cumcount"):
+        expected = expected.rename(columns={"c": "b"})
+    tm.assert_equal(result, expected)

pandas/tests/groupby/test_groupby.py

@@ -2828,3 +2828,13 @@ def test_groupby_reduce_period():
     expected = ser[:10]
     expected.index = Index(range(10), dtype=np.int_)
     tm.assert_series_equal(res, expected)
+
+
+def test_obj_with_exclusions_duplicate_columns():
+    # GH#50806
+    df = DataFrame([[0, 1, 2, 3]])
+    df.columns = [0, 1, 2, 0]
+    gb = df.groupby(df[1])
+    result = gb._obj_with_exclusions
+    expected = df.take([0, 2, 3], axis=1)
+    tm.assert_frame_equal(result, expected)