Backport PR #25234: BUG: Duplicated returns boolean dataframe (#25267)

Author: meeseeksmachine

doc/source/whatsnew/v0.24.2.rst

@@ -24,6 +24,8 @@ Fixed Regressions
 - Fixed issue in ``DataFrame`` construction with passing a mixed list of mixed types could segfault. (:issue:`25075`)
 - Fixed regression in :meth:`DataFrame.apply` causing ``RecursionError`` when ``dict``-like classes were passed as argument. (:issue:`25196`)
 
+- Fixed regression in :meth:`DataFrame.duplicated()`, where empty dataframe was not returning a boolean dtyped Series. (:issue:`25184`)
+
 .. _whatsnew_0242.enhancements:
 
 Enhancements

doc/source/whatsnew/v0.25.0.rst

@@ -69,6 +69,8 @@ Performance Improvements
 Bug Fixes
 ~~~~~~~~~
 
+-
+
 Categorical
 ^^^^^^^^^^^
 

pandas/core/frame.py

@@ -4661,7 +4661,7 @@ def duplicated(self, subset=None, keep='first'):
         from pandas._libs.hashtable import duplicated_int64, _SIZE_HINT_LIMIT
 
         if self.empty:
-            return Series()
+            return Series(dtype=bool)
 
         def f(vals):
             labels, shape = algorithms.factorize(

pandas/tests/frame/test_duplicates.py

@@ -182,6 +182,17 @@ def test_drop_duplicates():
         assert df.duplicated(keep=keep).sum() == 0
 
 
+def test_duplicated_on_empty_frame():
+    # GH 25184
+
+    df = DataFrame(columns=['a', 'b'])
+    dupes = df.duplicated('a')
+
+    result = df[dupes]
+    expected = df.copy()
+    tm.assert_frame_equal(result, expected)
+
+
 def test_drop_duplicates_with_duplicate_column_names():
     # GH17836
     df = DataFrame([