ENH: Series.resample performance with datetime64[ns] #7754

Author: scari

doc/source/whatsnew/v0.16.1.txt

@@ -215,6 +215,7 @@ Performance Improvements
 - Improved csv write performance with mixed dtypes, including datetimes by up to 5x (:issue:`9940`)
 - Improved csv write performance generally by 2x (:issue:`9940`)
 - Improved the performance of ``pd.lib.max_len_string_array`` by 5-7x (:issue:`10024`)
+- Improved ``Series.resample`` performance with dtype=datetime64[ns] (:issue:`7754`)
 
 
 

pandas/core/groupby.py

@@ -1493,6 +1493,8 @@ def aggregate(self, values, how, axis=0):
 
         if is_datetime_or_timedelta_dtype(values.dtype):
             values = values.view('int64')
+            # GH 7754
+            is_numeric = True
         elif is_bool_dtype(values.dtype):
             values = _algos.ensure_float64(values)
         elif com.is_integer_dtype(values):

vb_suite/timeseries.py

@@ -135,6 +135,16 @@ def date_range(start=None, end=None, periods=None, freq=None):
     Benchmark("ts.resample('D', how='mean')", setup,
               start_date=datetime(2012, 4, 25))
 
+# GH 7754
+setup = common_setup + """
+rng = date_range(start='2000-01-01 00:00:00',
+                    end='2000-01-01 10:00:00', freq='555000U')
+int_ts = Series(5, rng, dtype='int64')
+ts = int_ts.astype('datetime64[ns]')
+"""
+
+timeseries_resample_datetime64 = Benchmark("ts.resample('1S', how='last')", setup)
+
 #----------------------------------------------------------------------
 # to_datetime