From a73828f601920651c5d2f81017e392e6c596a6b4 Mon Sep 17 00:00:00 2001
From: Adam Hooper <adam@adamhooper.com>
Date: Thu, 26 Sep 2019 13:42:13 -0400
Subject: [PATCH] Fix groupby of categoricals with just "max"

I discovered https://github.com/pandas-dev/pandas/issues/28641
while testing ... and I fixed it.
---
 groupby.py      | 15 ++++++++-------
 test_groupby.py | 16 ++++++++++++++++
 2 files changed, 24 insertions(+), 7 deletions(-)

diff --git a/groupby.py b/groupby.py
index 3dbc5b8..98ccf3e 100644
--- a/groupby.py
+++ b/groupby.py
@@ -302,14 +302,16 @@ def groupby(
         and hasattr(table[colname], "cat")
     }
     for colname in category_colnames:
-        table[colname] = table[colname].cat.as_ordered()
+        table[colname].cat.as_ordered(inplace=True)
+        # Add dummy "size" to work around
+        # https://github.com/pandas-dev/pandas/issues/28641
+        agg_sets[colname].add("size")
 
     if group_specs:
         # aggs: DataFrame indexed by group
         # out: just the group colnames, no values yet (we'll add them later)
-        grouped = table.groupby(group_specs)
-        if agg_sets:
-            aggs = grouped.agg(agg_sets)
+        grouped = table.groupby(group_specs, as_index=True)
+        aggs = grouped.agg(agg_sets)
         out = aggs.index.to_frame(index=False)
         # Remove unused categories (because `np.nan` deletes categories)
         for column in out:
@@ -320,8 +322,7 @@ def groupby(
         # aggs: DataFrame with just one row
         # out: one empty row, no columns yet
         grouped = table
-        if agg_sets:
-            aggs = table.agg(agg_sets)
+        aggs = table.agg(agg_sets)
         out = pd.DataFrame(columns=[], index=[0])
 
     # Now copy values from `aggs` into `out`. (They have the same index.)
@@ -348,7 +349,7 @@ def groupby(
 
     # Remember those category colnames we converted to ordered? Now we need to
     # undo that (and remove newly-unused categories).
-    for colname in out.columns:
+    for colname in list(out.columns):
         column = out[colname]
         if hasattr(column, "cat") and column.cat.ordered:
             column.cat.remove_unused_categories(inplace=True)
diff --git a/test_groupby.py b/test_groupby.py
index f0f8597..b0092f1 100644
--- a/test_groupby.py
+++ b/test_groupby.py
@@ -395,6 +395,22 @@ def test_aggregate_text_category_values(self):
             ),
         )
 
+    def test_aggregate_text_category_values_max(self):
+        # https://github.com/pandas-dev/pandas/issues/28641
+        result = groupby(
+            pd.DataFrame(
+                {"A": [1997], "B": pd.Series(["30-SEP-97"], dtype="category")}
+            ),
+            [Group("A", None)],
+            [Aggregation(Operation.MAX, "B", "X")],
+        )
+        assert_frame_equal(
+            result,
+            pd.DataFrame(
+                {"A": [1997], "X": pd.Series(["30-SEP-97"], dtype="category")}
+            ),
+        )
+
     def test_aggregate_text_category_values_empty_still_has_object_dtype(self):
         result = groupby(
             pd.DataFrame({"A": [None]}, dtype=str).astype("category"),