BUG: GH11847 Unstack with mixed dtypes coerces everything to object

Changed the way in which the original data frame is copied (dropped use of .values, since it does not preserve dtypes).

Author: kordek

doc/source/whatsnew/v0.20.0.txt

@@ -80,3 +80,5 @@ Performance Improvements
 
 Bug Fixes
 ~~~~~~~~~
+
+- Bug in ``unstack()`` If called with list of column(s) as argument, regardless of dtypes all columns get coerced to ``object`` (:issue:`11847`)

pandas/core/reshape.py

@@ -277,7 +277,8 @@ def _unstack_multiple(data, clocs):
                              verify_integrity=False)
 
     if isinstance(data, Series):
-        dummy = Series(data.values, index=dummy_index)
+        dummy = data.copy()
+        dummy.index = dummy_index
         unstacked = dummy.unstack('__placeholder__')
         new_levels = clevels
         new_names = cnames
@@ -292,7 +293,8 @@ def _unstack_multiple(data, clocs):
 
             return result
 
-        dummy = DataFrame(data.values, index=dummy_index, columns=data.columns)
+        dummy = data.copy()
+        dummy.index = dummy_index
 
         unstacked = dummy.unstack('__placeholder__')
         if isinstance(unstacked, Series):

pandas/tests/frame/test_reshape.py

@@ -282,6 +282,46 @@ def test_unstack_fill_frame_categorical(self):
                              index=list('xyz'))
         assert_frame_equal(result, expected)
 
+    def test_unstack_preserve_dtypes(self):
+        # Checks fix for #11847
+        df = pd.DataFrame(dict(state=['IL', 'MI', 'NC'],
+                               index=['a', 'b', 'c'],
+                               A=np.random.rand(3),
+                               B=1,
+                               C='foo',
+                               D=pd.Timestamp('20010102'),
+                               E=pd.Series([1.0, 50.0, 100.0]
+                                           ).astype('float32'),
+                               F=pd.Series([3.0, 4.0, 5.0]).astype('float64'),
+                               G=False,
+                               H=pd.Series([1, 200, 923442], dtype='int8')))
+
+        df1 = df.set_index(['state', 'index'])
+        unstacked1 = df1.unstack(['index'])
+        unstacked2 = df1.unstack('index')
+        assert_frame_equal(unstacked1, unstacked2)
+
+        df1 = df.set_index(['F', 'C'])
+        unstacked1 = df1.unstack(['F'])
+        unstacked2 = df1.unstack('F')
+        assert_frame_equal(unstacked1, unstacked2)
+
+        df1 = df.set_index(['G', 'B', 'state'])
+        unstacked1 = df1.unstack(['B'])
+        unstacked2 = df1.unstack('B')
+        assert_frame_equal(unstacked1, unstacked2)
+
+        df1 = df.set_index(['E', 'A'])
+        unstacked1 = df1.unstack(['E'])
+        unstacked2 = df1.unstack('E')
+        assert_frame_equal(unstacked1, unstacked2)
+
+        df1 = df.set_index(['state', 'index'])
+        s = df1['A']
+        unstacked1 = s.unstack(['index'])
+        unstacked2 = s.unstack('index')
+        assert_frame_equal(unstacked1, unstacked2)
+
     def test_stack_ints(self):
         columns = MultiIndex.from_tuples(list(itertools.product(range(3),
                                                                 repeat=3)))