PERF: Faster CategoricalIndex from categorical

Author: TomAugspurger

doc/source/whatsnew/v0.21.0.txt

@@ -469,6 +469,7 @@ Performance Improvements
 - :attr:`Series.dt` no longer performs frequency inference, yielding a large speedup when accessing the attribute (:issue:`17210`)
 - Improved performance of :meth:`Categorical.set_categories` by not materializing the values (:issue:`17508`)
 - :attr:`Timestamp.microsecond` no longer re-computes on attribute access (:issue:`17331`)
+- Improved performance of the :class:`CategoricalIndex` for data that is already categorical dtype (:issue:`17513`)
 
 .. _whatsnew_0210.bug_fixes:
 

pandas/core/indexes/category.py

@@ -130,6 +130,10 @@ def _create_categorical(self, data, categories=None, ordered=None):
         -------
         Categorical
         """
+        if (isinstance(data, (ABCSeries, type(self))) and
+                is_categorical_dtype(data)):
+            data = data.values
+
         if not isinstance(data, ABCCategorical):
             ordered = False if ordered is None else ordered
             from pandas.core.categorical import Categorical

pandas/tests/indexes/test_category.py

@@ -125,6 +125,16 @@ def test_construction_with_dtype(self):
         result = CategoricalIndex(idx, categories=idx, ordered=True)
         tm.assert_index_equal(result, expected, exact=True)
 
+    def test_create_categorical(self):
+        # https://github.com/pandas-dev/pandas/pull/17513
+        # The public CI constructor doesn't hit this code path with
+        # instances of CategoricalIndex, but we still want to test the code
+        ci = CategoricalIndex(['a', 'b', 'c'])
+        # First ci is self, second ci is data.
+        result = CategoricalIndex._create_categorical(ci, ci)
+        expected = Categorical(['a', 'b', 'c'])
+        tm.assert_categorical_equal(result, expected)
+
     def test_disallow_set_ops(self):
 
         # GH 10039