Use pandas to_offset to parse frequency string in date_range

python/cudf/cudf/core/tools/datetimes.py

@@ -8,6 +8,7 @@
 import cupy as cp
 import numpy as np
 import pandas as pd
+import pandas.tseries.offsets as pd_offset
 from pandas.core.tools.datetimes import _unit_map
 
 import cudf
@@ -458,6 +459,16 @@ class DateOffset:
         "Y": "years",
     }
 
+    _TICK_TO_UNITS = {
+        pd_offset.Day: "days",
+        pd_offset.Hour: "hours",
+        pd_offset.Minute: "minutes",
+        pd_offset.Second: "seconds",
+        pd_offset.Milli: "milliseconds",
+        pd_offset.Micro: "microseconds",
+        pd_offset.Nano: "nanoseconds",
+    }
+
     _FREQSTR_REGEX = re.compile("([0-9]*)([a-zA-Z]+)")
 
     def __init__(self, n=1, normalize=False, **kwds):
@@ -649,6 +660,10 @@ def _from_freqstr(cls: Type[_T], freqstr: str) -> _T:
 
         return cls(**{cls._CODES_TO_UNITS[freq_part]: int(numeric_part)})
 
+    @classmethod
+    def _from_pandas_ticks(cls: Type[_T], tick: pd.tseries.offsets.Tick) -> _T:
+        return cls(**{cls._TICK_TO_UNITS[type(tick)]: tick.n})
+
     def _maybe_as_fast_pandas_offset(self):
         if (
             len(self.kwds) == 1
@@ -814,23 +829,13 @@ def date_range(
     if isinstance(freq, DateOffset):
         offset = freq
     elif isinstance(freq, str):
-        # Map pandas `offset alias` into cudf DateOffset `CODE`, only
-        # fixed-frequency, non-anchored offset aliases are supported.
-        mo = re.fullmatch(
-            rf'(-)*(\d*)({"|".join(_offset_alias_to_code.keys())})', freq
-        )
-        if mo is None:
+        offset = pd.tseries.frequencies.to_offset(freq)
+        if not isinstance(offset, pd.tseries.offsets.Tick):
             raise ValueError(
-                f"Unrecognized or unsupported offset alias {freq}."
+                f"Unrecognized frequency string {freq}. cuDF does"
+                " not yet support month, quarter, year-anchored frequency."
             )
-
-        sign, n, offset_alias = mo.groups()
-        code = _offset_alias_to_code[offset_alias]
-
-        freq = "".join([n, code])
-        offset = DateOffset._from_freqstr(freq)
-        if sign:
-            offset.kwds.update({s: -i for s, i in offset.kwds.items()})
+        offset = DateOffset._from_pandas_ticks(offset)
     else:
         raise TypeError("`freq` must be a `str` or cudf.DateOffset object.")
 

python/cudf/cudf/tests/test_datetime.py

@@ -1583,6 +1583,48 @@ def test_date_range_raise_overflow():
         cudf.date_range(start=start, periods=periods, freq=freq)
 
 
+@pytest.mark.parametrize(
+    "freqstr_unsupported",
+    [
+        "1M",
+        "2SM",
+        "3MS",
+        "4BM",
+        "5CBM",
+        "6SMS",
+        "7BMS",
+        "8CBMS",
+        "Q",
+        "2BQ",
+        "3BQS",
+        "10A",
+        "10Y",
+        "9BA",
+        "9BY",
+        "8AS",
+        "8YS",
+        "7BAS",
+        "7BYS",
+        "BH",
+        "B",
+    ],
+)
+def test_date_range_raise_unsupported(freqstr_unsupported):
+    s, e = "2001-01-01", "2008-01-31"
+    pd.date_range(start=s, end=e, freq=freqstr_unsupported)
+    with pytest.raises(ValueError, match="does not yet support"):
+        cudf.date_range(start=s, end=e, freq=freqstr_unsupported)
+
+    # We also check that these values are unsupported when using lowercase
+    # characters. We exclude the value 3MS (every 3 month starts) because 3ms
+    # is a valid frequency for every 3 milliseconds.
+    if freqstr_unsupported != "3MS":
+        freqstr_unsupported = freqstr_unsupported.lower()
+        pd.date_range(start=s, end=e, freq=freqstr_unsupported)
+        with pytest.raises(ValueError, match="does not yet support"):
+            cudf.date_range(start=s, end=e, freq=freqstr_unsupported)
+
+
 ##################################################################
 #                    End of Date Range Test                      #
 ##################################################################