BUG: Categoricals shouldn't allow non-strings when object dtype is passed (#13919)

doc/source/whatsnew/v0.19.0.txt

@@ -1071,3 +1071,4 @@ Bug Fixes
 - Bug in ``Index`` raises ``KeyError`` displaying incorrect column when column is not in the df and columns contains duplicate values (:issue:`13822`)
 - Bug in ``Period`` and ``PeriodIndex`` creating wrong dates when frequency has combined offset aliases (:issue:`13874`)
 - Bug in ``.to_string()`` when called with an integer ``line_width`` and ``index=False`` raises an UnboundLocalError exception because ``idx`` referenced before assignment.
+- Bug in ``Categorical`` would allow creation when ``object`` dtype was passed in with non-string values

pandas/core/categorical.py

@@ -18,9 +18,12 @@
                                  is_dtype_equal,
                                  is_datetimelike,
                                  is_categorical_dtype,
+                                 is_string_like,
                                  is_integer_dtype, is_bool,
                                  is_list_like, is_sequence,
-                                 is_scalar)
+                                 is_scalar,
+                                 is_object_dtype,
+                                 is_period_arraylike)
 from pandas.core.common import is_null_slice
 
 from pandas.core.algorithms import factorize, take_1d
@@ -191,6 +194,8 @@ class Categorical(PandasObject):
         If an explicit ``ordered=True`` is given but no `categories` and the
         `values` are not sortable.
 
+        If an `object` dtype is passed and `values` contains dtypes other
+        than all strings or all periods.
 
     Examples
     --------
@@ -231,6 +236,15 @@ class Categorical(PandasObject):
     def __init__(self, values, categories=None, ordered=False,
                  name=None, fastpath=False):
 
+        # categoricals w/ object dtype shouldn't allow non-strings
+        if is_object_dtype(values):
+            values = _convert_to_list_like(values)
+            # for now, periods are an exception
+            if not all(is_string_like(v) for v in values
+                       if notnull(v)) and not is_period_arraylike(values):
+                raise TypeError("Categoricals cannot be object dtype unless"
+                                " all values are strings or all are periods.")
+
         if fastpath:
             # fast path
             self._codes = _coerce_indexer_dtype(values, categories)

pandas/tests/test_categorical.py

@@ -94,13 +94,35 @@ def test_constructor_unsortable(self):
 
         # it works!
         arr = np.array([1, 2, 3, datetime.now()], dtype='O')
-        factor = Categorical.from_array(arr, ordered=False)
-        self.assertFalse(factor.ordered)
+        msg = "Categoricals cannot be object dtype unless all values are " \
+              "strings or all are periods."
+        with tm.assertRaisesRegexp(TypeError, msg):
+            factor = Categorical.from_array(arr, ordered=False)
 
         # this however will raise as cannot be sorted
         self.assertRaises(
             TypeError, lambda: Categorical.from_array(arr, ordered=True))
 
+    def test_constructor_object_dtype(self):
+        #GH 13919
+
+        #categories must be of single dtype
+        arr = np.array([1, 2, 3, 's'], dtype=object)
+        msg = "Categoricals cannot be object dtype unless all values are " \
+              "strings or all are periods."
+        with tm.assertRaisesRegexp(TypeError, msg):
+            c = Categorical.from_array(arr)
+
+        # object dtype allowed when all strs
+        exp_arr = np.array(list('abcd'), dtype=object)
+        c = Categorical.from_array(exp_arr)
+        tm.assert_numpy_array_equal(c.__array__(), exp_arr)
+
+        # object dtype also allowed when all periods
+        idx = pd.period_range('1/1/2000', freq='D', periods=5)
+        c = Categorical(idx)
+        tm.assert_index_equal(c.categories, idx)
+
     def test_is_equal_dtype(self):
 
         # test dtype comparisons between cats
@@ -4255,7 +4277,6 @@ def test_str_accessor_api_for_categorical(self):
             ('endswith', ("a",), {}),
             ('extract', ("([a-z]*) ",), {"expand":False}),
             ('extract', ("([a-z]*) ",), {"expand":True}),
-            ('extractall', ("([a-z]*) ",), {}),
             ('find', ("a",), {}),
             ('findall', ("a",), {}),
             ('index', (" ",), {}),
@@ -4286,7 +4307,8 @@ def test_str_accessor_api_for_categorical(self):
         #   we can't make a categorical with lists as individual categories.
         #   -> `s.str.split(" ").astype("category")` will error!
         # * `translate` has different interfaces for py2 vs. py3
-        _ignore_names = ["get", "join", "translate"]
+        # extractall creates Categorical w/ object dtype and int, which raises
+        _ignore_names = ["get", "join", "translate", "extractall"]
 
         str_func_names = [f
                           for f in dir(s.str)