Some performance testing

Author: mroeschke

asv_bench/benchmarks/timeseries.py

@@ -356,9 +356,17 @@ def setup(self):
 
         self.s = Series((['19MAY11', '19MAY11:00:00:00'] * 100000))
         self.s2 = self.s.str.replace(':\\S+$', '')
-        self.dup_numeric_data = Series([1000] * 100000)
-        self.dup_string_data = ['2013-01-01'] * 100000
-        self.dup_datetime_data = [dt.datetime(2010, 1, 1)] * 100000
+        self.dup_numeric_data_10_5 = Series([1000] * 100000)
+        self.dup_string_data_10_5 = ['2013-01-01 01:00:00'] * 100000
+        self.dup_datetime_data_10_5 = [dt.datetime(2010, 1, 1)] * 100000
+
+        self.dup_numeric_data_10_3 = Series([1000] * 100)
+        self.dup_string_data_10_3 = ['2013-01-01 01:00:00'] * 100
+        self.dup_datetime_data_10_3 = [dt.datetime(2010, 1, 1)] * 100
+
+        self.dup_numeric_data_10_7 = Series([1000] * 10**7)
+        self.dup_string_data_10_7 = ['2013-01-01 01:00:00'] * 10**7
+        self.dup_datetime_data_10_7 = [dt.datetime(2010, 1, 1)] * 10**7
 
     def time_format_YYYYMMDD(self):
         to_datetime(self.stringsD, format='%Y%m%d')
@@ -384,14 +392,32 @@ def time_format_exact(self):
     def time_format_no_exact(self):
         to_datetime(self.s, format='%d%b%y', exact=False)
 
-    def time_cache_dup_numeric_data(self):
-        to_datetime(self.dup_numeric_data, unit='s')
+    def time_cache_dup_numeric_data_10_3(self):
+        to_datetime(self.dup_numeric_data_10_3, unit='s')
+
+    def time_cache_dup_datetime_data_10_3(self):
+        to_datetime(self.dup_datetime_data_10_3)
+
+    def time_cache_dup_string_data_10_3(self):
+        to_datetime(self.dup_string_data_10_3)
+
+    def time_cache_dup_numeric_data_10_5(self):
+        to_datetime(self.dup_numeric_data_10_5, unit='s')
+
+    def time_cache_dup_datetime_data_10_5(self):
+        to_datetime(self.dup_datetime_data_10_5)
+
+    def time_cache_dup_string_data_10_5(self):
+        to_datetime(self.dup_string_data_10_5)
+
+    def time_cache_dup_numeric_data_10_7(self):
+        to_datetime(self.dup_numeric_data_10_7, unit='s')
 
-    def time_cache_dup_datetime_data(self):
-        to_datetime(self.dup_datetime_data)
+    def time_cache_dup_datetime_data_10_7(self):
+        to_datetime(self.dup_datetime_data_10_7)
 
-    def time_cache_dup_string_data(self):
-        to_datetime(self.dup_string_data)
+    def time_cache_dup_string_data_10_7(self):
+        to_datetime(self.dup_string_data_10_7)
 
 
 class Offsets(object):

pandas/core/indexes/datetimes.py

@@ -334,7 +334,7 @@ def __new__(cls, data=None,
         if not (is_datetime64_dtype(data) or is_datetimetz(data) or
                 is_integer_dtype(data)):
             data = tools.to_datetime(data, dayfirst=dayfirst,
-                                     yearfirst=yearfirst)
+                                     yearfirst=yearfirst, cache=False)
 
         if issubclass(data.dtype.type, np.datetime64) or is_datetimetz(data):
 

pandas/core/tools/datetimes.py

@@ -355,7 +355,7 @@ def to_datetime(arg, errors='raise', dayfirst=False, yearfirst=False,
 
     def _convert_listlike(arg, box, format, name=None, tz=tz):
 
-
+        import pdb; pdb.set_trace()
         if isinstance(arg, (list, tuple)):
             arg = np.array(arg, dtype='O')
 
@@ -525,16 +525,16 @@ def _convert_listlike(arg, box, format, name=None, tz=tz):
     if cache and is_list_like(arg):
         # Create a cache only if there are more than 10k values and the user
         # passes in datestrings
-        min_cache_threshold = 10**5
-        if len(arg) >= min_cache_threshold and is_string_dtype(arg):
+        #min_cache_threshold = 10**5
+        #if len(arg) >= min_cache_threshold and is_string_dtype(arg):
         # unique currently cannot determine dates that are out of bounds
         # recurison errors with datetime
-            unique_dates = algorithms.unique(arg)
-            # Essentially they need to all be the same value
-            if len(unique_dates) == 1:
-                from pandas import Series
-                cache_data = _convert_listlike(unique_dates, True, format)
-                convert_cache = Series(cache_data, index=unique_dates)
+        unique_dates = algorithms.unique(arg)
+        # Essentially they need to all be the same value
+        if len(unique_dates) != len(arg):
+            from pandas import Series
+            cache_data = _convert_listlike(unique_dates, False, format)
+            convert_cache = Series(cache_data, index=unique_dates)
 
     if isinstance(arg, tslib.Timestamp):
         result = arg