BUG: Make .iloc and .loc indexing consistent on empty dataframes

Tests

Fix

Test reorder

Doc update

Tests fix

Tests fix

SQL tests fix

Testing update

Fixes

Testing fix

Test fix

Author: Artemy Kolchinsky

doc/source/whatsnew/v0.16.1.txt

@@ -217,9 +217,8 @@ Bug Fixes
 - Bug causing an exception in slice assignments because ``length_of_indexer`` returns wrong results (:issue:`9995`)
 - Bug in csv parser causing lines with initial whitespace plus one non-space character to be skipped. (:issue:`9710`)
 
-
 - Bug causing elements with a null group to spill into the final group when grouping by a ``Categorical`` (:issue:`9603`)
-
+- Bug where .iloc and .loc behavior is not consistent on empty dataframes (:issue:`9964`)
 
 - Bug in invalid attribute access on a ``TimedeltaIndex`` incorrectly raised ``ValueError`` instead of ``AttributeError`` (:issue:`9680`)
 

pandas/core/frame.py

@@ -1740,17 +1740,19 @@ def _ixs(self, i, axis=0):
                 lab_slice = slice(label[0], label[-1])
                 return self.ix[:, lab_slice]
             else:
-                label = self.columns[i]
                 if isinstance(label, Index):
                     return self.take(i, axis=1, convert=True)
 
+                index_len = len(self.index)
+
                 # if the values returned are not the same length
                 # as the index (iow a not found value), iget returns
                 # a 0-len ndarray. This is effectively catching
                 # a numpy error (as numpy should really raise)
                 values = self._data.iget(i)
-                if not len(values):
-                    values = np.array([np.nan] * len(self.index), dtype=object)
+
+                if index_len and not len(values):
+                    values = np.array([np.nan] * index_len, dtype=object)
                 result = self._constructor_sliced.from_array(
                     values, index=self.index,
                     name=label, fastpath=True)

pandas/io/tests/test_json/test_pandas.py

@@ -324,12 +324,14 @@ def test_frame_to_json_except(self):
     def test_frame_empty(self):
         df = DataFrame(columns=['jim', 'joe'])
         self.assertFalse(df._is_mixed_type)
-        assert_frame_equal(read_json(df.to_json()), df)
+        assert_frame_equal(read_json(df.to_json(), dtype=dict(df.dtypes)), df)
 
+    def test_frame_empty_mixedtype(self):
         # mixed type
+        df = DataFrame(columns=['jim', 'joe'])
         df['joe'] = df['joe'].astype('i8')
         self.assertTrue(df._is_mixed_type)
-        assert_frame_equal(read_json(df.to_json()), df)
+        assert_frame_equal(read_json(df.to_json(), dtype=dict(df.dtypes)), df)
 
     def test_v12_compat(self):
         df = DataFrame(

pandas/io/tests/test_sql.py

@@ -1256,10 +1256,14 @@ def test_transactions(self):
         self._transaction_test()
 
     def test_get_schema_create_table(self):
-        self._load_test2_data()
+        # Use a dataframe without a bool column, since MySQL converts bool to 
+        # TINYINT (which read_sql_table returns as an int and causes a dtype
+        # mismatch)
+
+        self._load_test3_data()
         tbl = 'test_get_schema_create_table'
-        create_sql = sql.get_schema(self.test_frame2, tbl, con=self.conn)
-        blank_test_df = self.test_frame2.iloc[:0]
+        create_sql = sql.get_schema(self.test_frame3, tbl, con=self.conn)
+        blank_test_df = self.test_frame3.iloc[:0]
 
         self.drop_table(tbl)
         self.conn.execute(create_sql)

pandas/stats/tests/test_moments.py

@@ -862,7 +862,7 @@ def _non_null_values(x):
             if mock_mean:
                 # check that mean equals mock_mean
                 expected = mock_mean(x)
-                assert_equal(mean_x, expected)
+                assert_equal(mean_x, expected.astype('float64'))
 
             # check that correlation of a series with itself is either 1 or NaN
             corr_x_x = corr(x, x)
@@ -1550,6 +1550,7 @@ def test_moment_functions_zero_length(self):
         df1_expected = df1
         df1_expected_panel = Panel(items=df1.index, major_axis=df1.columns, minor_axis=df1.columns)
         df2 = DataFrame(columns=['a'])
+        df2['a'] = df2['a'].astype('float64')
         df2_expected = df2
         df2_expected_panel = Panel(items=df2.index, major_axis=df2.columns, minor_axis=df2.columns)
 

pandas/tests/test_frame.py

@@ -11864,12 +11864,10 @@ def test_mode(self):
                            "E": [8, 8, 1, 1, 3, 3]})
         assert_frame_equal(df[["A"]].mode(),
                            pd.DataFrame({"A": [12]}))
-        assert_frame_equal(df[["D"]].mode(),
-                           pd.DataFrame(pd.Series([], dtype="int64"),
-                                        columns=["D"]))
-        assert_frame_equal(df[["E"]].mode(),
-                           pd.DataFrame(pd.Series([1, 3, 8], dtype="int64"),
-                                        columns=["E"]))
+        expected = pd.Series([], dtype='int64', name='D').to_frame()
+        assert_frame_equal(df[["D"]].mode(), expected)
+        expected = pd.Series([1, 3, 8], dtype='int64', name='E').to_frame()
+        assert_frame_equal(df[["E"]].mode(), expected)
         assert_frame_equal(df[["A", "B"]].mode(),
                            pd.DataFrame({"A": [12], "B": [10.]}))
         assert_frame_equal(df.mode(),

pandas/tests/test_groupby.py

@@ -1728,6 +1728,8 @@ def test_groupby_head_tail(self):
         assert_frame_equal(df.loc[[1, 2]], g_not_as.tail(1))
 
         empty_not_as = DataFrame(columns=df.columns)
+        empty_not_as['A'] = empty_not_as['A'].astype(df.A.dtype)
+        empty_not_as['B'] = empty_not_as['B'].astype(df.B.dtype)
         assert_frame_equal(empty_not_as, g_not_as.head(0))
         assert_frame_equal(empty_not_as, g_not_as.tail(0))
         assert_frame_equal(empty_not_as, g_not_as.head(-1))
@@ -1743,6 +1745,8 @@ def test_groupby_head_tail(self):
         assert_frame_equal(df_as.loc[[1, 2]], g_as.tail(1))
 
         empty_as = DataFrame(index=df_as.index[:0], columns=df.columns)
+        empty_as['A'] = empty_not_as['A'].astype(df.A.dtype)
+        empty_as['B'] = empty_not_as['B'].astype(df.B.dtype)
         assert_frame_equal(empty_as, g_as.head(0))
         assert_frame_equal(empty_as, g_as.tail(0))
         assert_frame_equal(empty_as, g_as.head(-1))

pandas/tests/test_indexing.py

@@ -1063,6 +1063,7 @@ def test_loc_setitem_consistency(self):
 
         # empty (essentially noops)
         expected = DataFrame(columns=['x', 'y'])
+        expected['x'] = expected['x'].astype(np.int64)
         df = DataFrame(columns=['x', 'y'])
         df.loc[:, 'x'] = 1
         assert_frame_equal(df,expected)
@@ -3376,7 +3377,7 @@ def f():
         expected = DataFrame(columns=['foo'])
         def f():
             df = DataFrame()
-            df['foo'] = Series([])
+            df['foo'] = Series([], dtype='object')
             return df
         assert_frame_equal(f(), expected)
         def f():
@@ -3386,17 +3387,20 @@ def f():
         assert_frame_equal(f(), expected)
         def f():
             df = DataFrame()
-            df['foo'] = Series(range(len(df)))
+            df['foo'] = df.index
             return df
         assert_frame_equal(f(), expected)
+
+        expected = DataFrame(columns=['foo'])
+        expected['foo'] = expected['foo'].astype('float64')
         def f():
             df = DataFrame()
             df['foo'] = []
             return df
         assert_frame_equal(f(), expected)
         def f():
             df = DataFrame()
-            df['foo'] = df.index
+            df['foo'] = Series(range(len(df)))
             return df
         assert_frame_equal(f(), expected)
         def f():
@@ -3429,21 +3433,31 @@ def f():
 
         # GH5720, GH5744
         # don't create rows when empty
+        expected = DataFrame(columns=['A','B','New'])
+        expected['A'] = expected['A'].astype('int64')
+        expected['B'] = expected['B'].astype('float64')
+        expected['New'] = expected['New'].astype('float64')
         df = DataFrame({"A": [1, 2, 3], "B": [1.2, 4.2, 5.2]})
         y = df[df.A > 5]
         y['New'] = np.nan
-        assert_frame_equal(y,DataFrame(columns=['A','B','New']))
+        assert_frame_equal(y,expected)
+        #assert_frame_equal(y,expected)
 
+        expected = DataFrame(columns=['a','b','c c','d'])
+        expected['d'] = expected['d'].astype('int64')
         df = DataFrame(columns=['a', 'b', 'c c'])
         df['d'] = 3
-        assert_frame_equal(df,DataFrame(columns=['a','b','c c','d']))
+        assert_frame_equal(df,expected)
         assert_series_equal(df['c c'],Series(name='c c',dtype=object))
 
         # reindex columns is ok
         df = DataFrame({"A": [1, 2, 3], "B": [1.2, 4.2, 5.2]})
         y = df[df.A > 5]
         result = y.reindex(columns=['A','B','C'])
         expected = DataFrame(columns=['A','B','C'])
+        expected['A'] = expected['A'].astype('int64')
+        expected['B'] = expected['B'].astype('float64')
+        expected['C'] = expected['C'].astype('float64')
         assert_frame_equal(result,expected)
 
         # GH 5756
@@ -4429,6 +4443,15 @@ def test_indexing_assignment_dict_already_exists(self):
         expected.loc[5] = [9, 99]
         tm.assert_frame_equal(df, expected)
 
+    def test_indexing_dtypes_on_empty(self):
+        # Check that .iloc and .ix return correct dtypes GH9983
+        df = DataFrame({'a':[1,2,3],'b':['b','b2','b3']})
+        df2 = df.ix[[],:]
+
+        self.assertEqual(df2.loc[:,'a'].dtype, int)
+        assert_series_equal(df2.loc[:,'a'], df2.iloc[:,0])
+        assert_series_equal(df2.loc[:,'a'], df2.ix[:,0])
+
 
 
 class TestCategoricalIndex(tm.TestCase):

pandas/tests/test_testing.py

@@ -215,6 +215,14 @@ def test_multiindex_dtype(self):
                 {'a':[1.0,2.0],'b':[2.1,1.5],'c':['l1','l2']}, index=['a','b'])
         self._assert_not_equal(df1, df2, check_index_type=True)
 
+    def test_empty_dtypes(self):
+        df1=pd.DataFrame(columns=["col1","col2"])
+        df1["col1"] = df1["col1"].astype('int64')
+        df2=pd.DataFrame(columns=["col1","col2"])
+        self._assert_equal(df1, df2, check_dtype=False)
+        self._assert_not_equal(df1, df2, check_dtype=True)
+        
+
 class TestRNGContext(unittest.TestCase):
 
     def test_RNGContext(self):

pandas/tseries/tests/test_period.py

@@ -2118,6 +2118,7 @@ def test_range_slice_outofbounds(self):
         for idx in [didx, pidx]:
             df = DataFrame(dict(units=[100 + i for i in range(10)]), index=idx)
             empty = DataFrame(index=idx.__class__([], freq='D'), columns=['units'])
+            empty['units'] = empty['units'].astype('int64')
 
             tm.assert_frame_equal(df['2013/09/01':'2013/09/30'], empty)
             tm.assert_frame_equal(df['2013/09/30':'2013/10/02'], df.iloc[:2])