ENH: Add support for more placeholders in guess_datetime_format (#43901) (#43900)

Add support for day of week and meridiem placeholders and any
combination of placeholders supported by `strftime` that do not
correspond to a datetime attribute.

Author: davesque

asv_bench/benchmarks/inference.py

@@ -277,6 +277,16 @@ def time_dup_string_tzoffset_dates(self, cache):
         to_datetime(self.dup_string_with_tz, cache=cache)
 
 
+# GH 43901
+class ToDatetimeInferDatetimeFormat:
+    def setup(self):
+        rng = date_range(start="1/1/2000", periods=100000, freq="H")
+        self.strings = rng.strftime("%Y-%m-%d %H:%M:%S").tolist()
+
+    def time_infer_datetime_format(self):
+        to_datetime(self.strings, infer_datetime_format=True)
+
+
 class ToTimedelta:
     def setup(self):
         self.ints = np.random.randint(0, 60, size=10000)

doc/source/whatsnew/v1.4.0.rst

@@ -407,6 +407,7 @@ Performance improvements
 - Performance improvement in some :meth:`GroupBy.apply` operations (:issue:`42992`)
 - Performance improvement in :func:`read_stata` (:issue:`43059`)
 - Performance improvement in :meth:`to_datetime` with ``uint`` dtypes (:issue:`42606`)
+- Performance improvement in :meth:`to_datetime` with ``infer_datetime_format`` set to ``True`` (:issue:`43901`)
 - Performance improvement in :meth:`Series.sparse.to_coo` (:issue:`42880`)
 - Performance improvement in indexing with a :class:`MultiIndex` indexer on another :class:`MultiIndex` (:issue:43370`)
 - Performance improvement in :meth:`GroupBy.quantile` (:issue:`43469`)

pandas/_libs/tslibs/parsing.pyx

@@ -912,6 +912,9 @@ def guess_datetime_format(
         (('second', 'microsecond'), '%S.%f', 0),
         (('tzinfo',), '%z', 0),
         (('tzinfo',), '%Z', 0),
+        (('day_of_week',), '%a', 0),
+        (('day_of_week',), '%A', 0),
+        (('meridiem',), '%p', 0),
     ]
 
     if dayfirst:
@@ -968,15 +971,17 @@ def guess_datetime_format(
         if set(attrs) & found_attrs:
             continue
 
-        if all(getattr(parsed_datetime, attr) is not None for attr in attrs):
-            for i, token_format in enumerate(format_guess):
-                token_filled = tokens[i].zfill(padding)
-                if (token_format is None and
-                        token_filled == parsed_datetime.strftime(attr_format)):
-                    format_guess[i] = attr_format
-                    tokens[i] = token_filled
-                    found_attrs.update(attrs)
-                    break
+        if parsed_datetime.tzinfo is None and attr_format in ("%Z", "%z"):
+            continue
+
+        parsed_formatted = parsed_datetime.strftime(attr_format)
+        for i, token_format in enumerate(format_guess):
+            token_filled = tokens[i].zfill(padding)
+            if token_format is None and token_filled == parsed_formatted:
+                format_guess[i] = attr_format
+                tokens[i] = token_filled
+                found_attrs.update(attrs)
+                break
 
     # Only consider it a valid guess if we have a year, month and day
     if len({'year', 'month', 'day'} & found_attrs) != 3:

pandas/tests/tslibs/test_parsing.py

@@ -167,6 +167,8 @@ def test_parsers_month_freq(date_str, expected):
         ("2011-12-30T00:00:00.000000+9:0", "%Y-%m-%dT%H:%M:%S.%f%z"),
         ("2011-12-30T00:00:00.000000+09:", None),
         ("2011-12-30 00:00:00.000000", "%Y-%m-%d %H:%M:%S.%f"),
+        ("Tue 24 Aug 2021 01:30:48 AM", "%a %d %b %Y %H:%M:%S %p"),
+        ("Tuesday 24 Aug 2021 01:30:48 AM", "%A %d %b %Y %H:%M:%S %p"),
     ],
 )
 def test_guess_datetime_format_with_parseable_formats(string, fmt):