Groupby transform preserves output dtype

Previously `transform` output was always the same dtype as the groupby
object. This allows the output dtype to differ from the input. Fixes pandas-dev#9807.

Author: jcrist

pandas/core/groupby.py

@@ -3006,7 +3006,7 @@ def transform(self, func, *args, **kwargs):
         if ((not isinstance(obj.index,MultiIndex) and
              type(result.index) != type(obj.index)) or
             len(result.index) != len(obj.index)):
-            results = obj.values.copy()
+            results = np.empty_like(obj.values, result.values.dtype)
             indices = self.indices
             for (name, group), (i, row) in zip(self, result.iterrows()):
                 indexer = indices[name]

pandas/tests/test_groupby.py

@@ -1003,6 +1003,14 @@ def test_transform_broadcast(self):
             for idx in gp.index:
                 assert_fp_equal(res.xs(idx), agged[idx])
 
+    def test_transform_dtype(self):
+        # GH 9807
+        # Check transform dtype output is preserved
+        df = DataFrame([[1, 3], [2, 3]])
+        result = df.groupby(1).transform('mean')
+        expected = DataFrame([[1.5], [1.5]])
+        assert_frame_equal(result, expected)
+
     def test_transform_bug(self):
         # GH 5712
         # transforming on a datetime column