Merge pull request #5926 from shwina/fix-groupby-nunique-series-redux

[REVIEW] Fix SeriesGroupBy.nunique() to return a Series

CHANGELOG.md

@@ -270,6 +270,7 @@
 - PR #5319 Disallow SUM and specialize MEAN of timestamp types
 - PR #5797 Fix a missing data issue in some Parquet files
 - PR #5787 Fix column create from dictionary column view
+- PR #5926 Fix SeriesGroupBy.nunique() to return a Series
 - PR #5813 Fix normalizer exception with all-null strings column
 - PR #5820 Fix ListColumn.to_arrow for all null case
 - PR #5837 Bash syntax error in prebuild.sh preventing `cudf_kafka` and `libcudf_kafka` from being uploaded to Anaconda

python/cudf/cudf/core/groupby/groupby.py

@@ -217,13 +217,6 @@ def nth(self, n):
         sizes = self.size()
         return result[n < sizes]
 
-    def nunique(self):
-        """
-        Return the number of unique values per group.
-        """
-        # Pandas includes key columns for nunique:
-        return self.agg(dict.fromkeys(self.obj._data.keys(), "nunique"))
-
     def serialize(self):
         header = {}
         frames = []
@@ -630,6 +623,13 @@ def __getattribute__(self, key):
     def __getitem__(self, key):
         return self.obj[key].groupby(self.grouping, dropna=self._dropna)
 
+    def nunique(self):
+        """
+        Return the number of unique values per group
+        """
+        # For DataFrameGroupBy, Pandas includes key columns for nunique:
+        return self.agg(dict.fromkeys(self.obj._data.keys(), "nunique"))
+
 
 class SeriesGroupBy(GroupBy):
     def __init__(

python/cudf/cudf/tests/test_groupby.py

@@ -1196,3 +1196,14 @@ def test_groupby_groups_multi(by):
     for key in pdg.groups:
         assert key in gdg.groups
         assert_eq(pdg.groups[key], gdg.groups[key])
+
+
+def test_groupby_nunique_series():
+    pdf = pd.DataFrame({"a": [1, 1, 1, 2, 2, 2], "b": [1, 2, 3, 1, 1, 2]})
+    gdf = cudf.from_pandas(pdf)
+
+    assert_eq(
+        pdf.groupby("a")["b"].nunique(),
+        gdf.groupby("a")["b"].nunique(),
+        check_dtype=False,
+    )