PERF: removed coercion to int64 for arrays of ints in Categorical.from_codes (#21000)

Author: nlee737

asv_bench/benchmarks/categoricals.py

@@ -51,6 +51,7 @@ def setup(self):
 
         self.values_some_nan = list(np.tile(self.categories + [np.nan], N))
         self.values_all_nan = [np.nan] * len(self.values)
+        self.values_all_int8 = np.ones(N, 'int8')
 
     def time_regular(self):
         pd.Categorical(self.values, self.categories)
@@ -70,6 +71,9 @@ def time_with_nan(self):
     def time_all_nan(self):
         pd.Categorical(self.values_all_nan)
 
+    def time_from_codes_all_int8(self):
+        pd.Categorical.from_codes(self.values_all_int8, self.categories)
+
 
 class ValueCounts(object):
 

doc/source/whatsnew/v0.23.0.txt

@@ -1080,6 +1080,7 @@ Performance Improvements
 - Improved performance of :func:`Series.isin` in the case of categorical dtypes (:issue:`20003`)
 - Improved performance of ``getattr(Series, attr)`` when the Series has certain index types. This manifiested in slow printing of large Series with a ``DatetimeIndex`` (:issue:`19764`)
 - Fixed a performance regression for :func:`GroupBy.nth` and :func:`GroupBy.last` with some object columns (:issue:`19283`)
+- Improved performance of :func:`pandas.core.arrays.Categorical.from_codes` (:issue:`18501`)
 
 .. _whatsnew_0230.docs:
 

pandas/core/arrays/categorical.py

@@ -578,7 +578,7 @@ def from_codes(cls, codes, categories, ordered=False):
             unordered.
         """
         try:
-            codes = np.asarray(codes, np.int64)
+            codes = coerce_indexer_dtype(np.asarray(codes), categories)
         except (ValueError, TypeError):
             raise ValueError(
                 "codes need to be convertible to an arrays of integers")