#18058: improve DatetimeIndex.date performance

asv_bench/benchmarks/timeseries.py

@@ -89,6 +89,9 @@ def time_dti_factorize(self):
     def time_dti_tz_factorize(self):
         self.dti_tz.factorize()
 
+    def time_dti_time(self):
+        self.rng.time
+
     def time_timestamp_tzinfo_cons(self):
         self.rng5[0]
 
@@ -107,6 +110,11 @@ def time_infer_freq_daily(self):
     def time_infer_freq_business(self):
         infer_freq(self.b_freq)
 
+    def time_to_date(self):
+        self.rng.date
+
+    def time_to_pydatetime(self):
+        self.rng.to_pydatetime()
 
 class TimeDatetimeConverter(object):
     goal_time = 0.2

doc/source/whatsnew/v0.22.0.txt

@@ -91,6 +91,8 @@ Performance Improvements
 - Converting a ``Series`` of ``Timedelta`` objects to days, seconds, etc... sped up through vectorization of underlying methods (:issue:`18092`)
 - The overriden ``Timedelta`` properties of days, seconds and microseconds have been removed, leveraging their built-in Python versions instead (:issue:`18242`)
 - ``Series`` construction will reduce the number of copies made of the input data in certain cases (:issue:`17449`)
+- Improved performance of :func:`Series.dt.date` and :func:`DatetimeIndex.date` (:issue:`18058`)
+-
 
 .. _whatsnew_0220.docs:
 

pandas/_libs/tslib.pyx

@@ -23,7 +23,7 @@ cimport util
 
 from cpython.datetime cimport (PyDateTime_Check, PyDate_Check,
                                PyDateTime_IMPORT,
-                               timedelta, datetime)
+                               timedelta, datetime, date)
 # import datetime C API
 PyDateTime_IMPORT
 # this is our datetime.pxd
@@ -80,10 +80,37 @@ cdef inline object create_datetime_from_ts(
     return datetime(dts.year, dts.month, dts.day, dts.hour,
                     dts.min, dts.sec, dts.us, tz)
 
+cdef inline object create_date_from_ts(
+        int64_t value, pandas_datetimestruct dts,
+        object tz, object freq):
+    """ convenience routine to construct a datetime.date from its parts """
+    return date(dts.year, dts.month, dts.day)
 
-def ints_to_pydatetime(ndarray[int64_t] arr, tz=None, freq=None, box=False):
-    # convert an i8 repr to an ndarray of datetimes or Timestamp (if box ==
-    # True)
+
+def ints_to_pydatetime(ndarray[int64_t] arr, tz=None, freq=None,
+                       box="datetime"):
+    """
+    Convert an i8 repr to an ndarray of datetimes, date or Timestamp
+
+    Parameters
+    ----------
+    arr  : array of i8
+    tz   : str, default None
+         convert to this timezone
+    freq : str/Offset, default None
+         freq to convert
+    box  : {'datetime', 'timestamp', 'date'}, default 'datetime'
+         If datetime, convert to datetime.datetime
+         If date, convert to datetime.date
+         If Timestamp, convert to pandas.Timestamp
+
+    Returns
+    -------
+    result : array of dtype specified by box
+    """
+
+    assert ((box == "datetime") or (box == "date") or (box == "timestamp")), \
+        "box must be one of 'datetime', 'date' or 'timestamp'"
 
     cdef:
         Py_ssize_t i, n = len(arr)
@@ -94,13 +121,17 @@ def ints_to_pydatetime(ndarray[int64_t] arr, tz=None, freq=None, box=False):
         ndarray[object] result = np.empty(n, dtype=object)
         object (*func_create)(int64_t, pandas_datetimestruct, object, object)
 
-    if box and is_string_object(freq):
-        from pandas.tseries.frequencies import to_offset
-        freq = to_offset(freq)
+    if box == "date":
+        assert (tz is None), "tz should be None when converting to date"
 
-    if box:
+        func_create = create_date_from_ts
+    elif box == "timestamp":
         func_create = create_timestamp_from_ts
-    else:
+
+        if is_string_object(freq):
+            from pandas.tseries.frequencies import to_offset
+            freq = to_offset(freq)
+    elif box == "datetime":
         func_create = create_datetime_from_ts
 
     if tz is not None:

pandas/core/dtypes/concat.py

@@ -405,7 +405,7 @@ def convert_to_pydatetime(x, axis):
             else:
                 shape = x.shape
                 x = tslib.ints_to_pydatetime(x.view(np.int64).ravel(),
-                                             box=True)
+                                             box="timestamp")
                 x = x.reshape(shape)
 
         elif x.dtype == _TD_DTYPE:

pandas/core/indexes/datetimes.py

@@ -1237,7 +1237,7 @@ def __iter__(self):
             end_i = min((i + 1) * chunksize, length)
             converted = libts.ints_to_pydatetime(data[start_i:end_i],
                                                  tz=self.tz, freq=self.freq,
-                                                 box=True)
+                                                 box="timestamp")
             for v in converted:
                 yield v
 
@@ -1687,8 +1687,7 @@ def date(self):
         Returns numpy array of python datetime.date objects (namely, the date
         part of Timestamps without timezone information).
         """
-        return self._maybe_mask_results(libalgos.arrmap_object(
-            self.asobject.values, lambda x: x.date()))
+        return libts.ints_to_pydatetime(self.normalize().asi8, box="date")
 
     def normalize(self):
         """