BUG: don't sort unique values from categoricals

This should resolve the inconsistency mwaskom reported in GH9148.

CC jreback TomAugspurger JanSchulz

Author: shoyer

doc/source/whatsnew/v0.16.0.txt

@@ -193,6 +193,8 @@ Bug Fixes
   SQLAlchemy type  (:issue:`9083`).
 
 
+- Items in ``Categorical.unique()`` (and ``s.unique()`` if ``s`` is of dtype ``category``) now appear in the order in which they are originally found, not in sorted order (:issue:`9331`). This is now consistent with the behavior for other dtypes in pandas.
+
 
 - Fixed bug on bug endian platforms which produced incorrect results in ``StataReader`` (:issue:`8688`).
 

pandas/core/categorical.py

@@ -1386,17 +1386,16 @@ def unique(self):
         """
         Return the unique values.
 
-        Unused categories are NOT returned.
+        Unused categories are NOT returned. Unique values are returned in order
+        of appearance.
 
         Returns
         -------
         unique values : array
         """
-        unique_codes = np.unique(self.codes)
-        # for compatibility with normal unique, which has nan last
-        if unique_codes[0] == -1:
-            unique_codes[0:-1] = unique_codes[1:]
-            unique_codes[-1] = -1
+        from pandas.core.nanops import unique1d
+        # unlike np.unique, unique1d does not sort
+        unique_codes = unique1d(self.codes)
         return take_1d(self.categories.values, unique_codes)
 
     def equals(self, other):

pandas/tests/test_categorical.py

@@ -774,12 +774,15 @@ def test_unique(self):
         exp = np.asarray(["a","b"])
         res = cat.unique()
         self.assert_numpy_array_equal(res, exp)
+
         cat = Categorical(["a","b","a","a"], categories=["a","b","c"])
         res = cat.unique()
         self.assert_numpy_array_equal(res, exp)
-        cat = Categorical(["a","b","a", np.nan], categories=["a","b","c"])
+
+        # unique should not sort
+        cat = Categorical(["b", "b", np.nan, "a"], categories=["a","b","c"])
         res = cat.unique()
-        exp = np.asarray(["a","b", np.nan], dtype=object)
+        exp = np.asarray(["b", np.nan, "a"], dtype=object)
         self.assert_numpy_array_equal(res, exp)
 
     def test_mode(self):