Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support freq in DatetimeIndex #14593

Merged
merged 20 commits into from
Dec 12, 2023
Merged

Support freq in DatetimeIndex #14593

Show file tree
Hide file tree
Changes from 16 commits
Commits
Show all changes
20 commits
Select commit Hold shift + click to select a range
9602715
Support `freq` in DatetimeIndex
shwina Dec 7, 2023
98e5e1e
"T" is minutes
shwina Dec 7, 2023
6b0beee
Add more string aliases
shwina Dec 7, 2023
20ca2bb
Define resamplers
shwina Dec 7, 2023
b461ecb
fix metadata issues
galipremsagar Dec 7, 2023
45fb6b2
Merge branch 'support-freq-in-datetime-index' of https://github.com/s…
galipremsagar Dec 7, 2023
f68a689
Merge branch 'branch-24.02' into support-freq-in-datetime-index
galipremsagar Dec 7, 2023
0337840
Fix more cases
galipremsagar Dec 7, 2023
957c7c5
Apply suggestions from code review
galipremsagar Dec 7, 2023
ed3ba3f
fix more cases
galipremsagar Dec 7, 2023
ce4f3bd
address reviews
galipremsagar Dec 8, 2023
6ecc6e6
Merge branch 'branch-24.02' into support-freq-in-datetime-index
galipremsagar Dec 8, 2023
cd00345
Apply suggestions from code review
galipremsagar Dec 8, 2023
6d00347
fix freq calculations
galipremsagar Dec 8, 2023
e1d2315
Merge branch 'support-freq-in-datetime-index' of https://github.com/s…
galipremsagar Dec 8, 2023
55266cd
Add validation
galipremsagar Dec 8, 2023
e1b697f
Simplify repr
galipremsagar Dec 8, 2023
32f622a
Merge branch 'branch-24.02' into support-freq-in-datetime-index
galipremsagar Dec 8, 2023
0d5c452
Handle freq in groupby ops
galipremsagar Dec 12, 2023
112dbc1
Merge branch 'branch-24.02' into support-freq-in-datetime-index
galipremsagar Dec 12, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
63 changes: 56 additions & 7 deletions python/cudf/cudf/core/index.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@
Tuple,
Type,
Union,
cast,
)

import cupy
Expand Down Expand Up @@ -1430,11 +1431,16 @@ def __repr__(self):
lines = lines[:-1]
lines.append(prior_to_dtype + " dtype='%s'" % self.dtype)
if self.name is not None:
lines[-1] = lines[-1] + ", name='%s'" % self.name
lines[-1] += f", name='{self.name}'"
galipremsagar marked this conversation as resolved.
Show resolved Hide resolved
if "length" in tmp_meta:
lines[-1] = lines[-1] + ", length=%d)" % len(self)
else:
lines[-1] = lines[-1] + ")"
lines[-1] += f", length={len(self)}"
if (
"freq" in tmp_meta
and isinstance(self, DatetimeIndex)
and self._freq is not None
):
lines[-1] += f", freq={self._freq}"
lines[-1] += ")"

return "\n".join(lines)

Expand Down Expand Up @@ -2126,8 +2132,6 @@ def __init__(
# pandas dtindex creation first which. For now
# just make sure we handle np.datetime64 arrays
# and then just dispatch upstream
if freq is not None:
raise NotImplementedError("Freq is not yet supported")
if tz is not None:
raise NotImplementedError("tz is not yet supported")
if normalize is not False:
Expand All @@ -2141,6 +2145,8 @@ def __init__(
if yearfirst is not False:
raise NotImplementedError("yearfirst == True is not yet supported")

self._freq = _validate_freq(freq)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

While looking on adding freq support before, I found some APIs manipulate freq(to new values) and return new results. (I vaguely remember..but I think that happens in binops?) Should we add a TODO comment here that this is not fully functional yet and freq support needs to be added in rest of the code-base?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, although maybe the default behaviour could be for DatetimeIndex to infer freq from its values. Then this should just work.

Also, we should probably only do that in compatibility mode for perf reasons.


valid_dtypes = tuple(
f"datetime64[{res}]" for res in ("s", "ms", "us", "ns")
)
Expand All @@ -2158,6 +2164,30 @@ def __init__(

super().__init__(data, **kwargs)

if self._freq is not None:
unique_vals = self.to_series().diff().unique()
if len(unique_vals) > 2 or (
len(unique_vals) == 2
and unique_vals[1] != self._freq._maybe_as_fast_pandas_offset()
):
raise ValueError("No unique frequency found")

@_cudf_nvtx_annotate
def _copy_type_metadata(
self: DatetimeIndex, other: DatetimeIndex, *, override_dtypes=None
) -> GenericIndex:
super()._copy_type_metadata(other, override_dtypes=override_dtypes)
self._freq = _validate_freq(other._freq)
wence- marked this conversation as resolved.
Show resolved Hide resolved
return self

@classmethod
def _from_data(
cls, data: MutableMapping, name: Any = no_default, freq: Any = None
):
result = super()._from_data(data, name)
result._freq = _validate_freq(freq)
galipremsagar marked this conversation as resolved.
Show resolved Hide resolved
return result

def __getitem__(self, index):
value = super().__getitem__(index)
if cudf.get_option("mode.pandas_compatible") and isinstance(
Expand All @@ -2166,6 +2196,11 @@ def __getitem__(self, index):
return pd.Timestamp(value)
return value

@_cudf_nvtx_annotate
def copy(self, name=None, deep=False, dtype=None, names=None):
idx_copy = super().copy(name=name, deep=deep, dtype=dtype, names=names)
return idx_copy._copy_type_metadata(self)

def searchsorted(
self,
value,
Expand Down Expand Up @@ -2519,7 +2554,13 @@ def to_pandas(self, *, nullable: bool = False) -> pd.DatetimeIndex:
)
else:
nanos = self._values.astype("datetime64[ns]")
return pd.DatetimeIndex(nanos.to_pandas(), name=self.name)

freq = (
self._freq._maybe_as_fast_pandas_offset()
if self._freq is not None
else None
)
return pd.DatetimeIndex(nanos.to_pandas(), name=self.name, freq=freq)

@_cudf_nvtx_annotate
def _get_dt_field(self, field):
Expand Down Expand Up @@ -3624,3 +3665,11 @@ def _extended_gcd(a: int, b: int) -> Tuple[int, int, int]:
old_s, s = s, old_s - quotient * s
old_t, t = t, old_t - quotient * t
return old_r, old_s, old_t


def _validate_freq(freq: Any) -> cudf.DateOffset:
if isinstance(freq, str):
return cudf.DateOffset._from_freqstr(freq)
elif freq is not None and not isinstance(freq, cudf.DateOffset):
raise ValueError(f"Invalid frequency: {freq}")
return cast(cudf.DateOffset, freq)
14 changes: 8 additions & 6 deletions python/cudf/cudf/core/tools/datetimes.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -463,13 +463,19 @@ class DateOffset:
}

_CODES_TO_UNITS = {
"N": "nanoseconds",
Copy link
Contributor

@bdice bdice Dec 7, 2023 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I have some vague recollection that we left these out on purpose... hmm. I think there was some pandas behavior for which "L" and "ms" were okay but "N", "U", "T", etc. were not supported. We'd probably be able to tell if there are any newly failing pandas tests? I'd just check to see where _CODES_TO_UNITS is used and if there are any inconsistencies with this across different APIs.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There were a bunch of failing tests without these changes, adding these units passed the cudf pytests.

There is only slight increase in pandas-pytest failures:

# This PR:
= 12094 failed, 174794 passed, 3850 skipped, 3314 xfailed, 8 xpassed, 21406 warnings, 102 errors in 1516.39s (0:25:16) =

# `branch-24.02`:
= 11607 failed, 175286 passed, 3849 skipped, 3312 xfailed, 11 xpassed, 21414 warnings, 97 errors in 1493.35s (0:24:53) =

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sounds good, thanks for checking.

"ns": "nanoseconds",
"U": "microseconds",
"us": "microseconds",
"ms": "milliseconds",
"L": "milliseconds",
"s": "seconds",
"S": "seconds",
"m": "minutes",
"min": "minutes",
"T": "minutes",
"h": "hours",
"H": "hours",
"D": "days",
"W": "weeks",
"M": "months",
Expand All @@ -487,7 +493,7 @@ class DateOffset:
pd_offset.Nano: "nanoseconds",
}

_FREQSTR_REGEX = re.compile("([0-9]*)([a-zA-Z]+)")
_FREQSTR_REGEX = re.compile("([-+]?[0-9]*)([a-zA-Z]+)")

def __init__(self, n=1, normalize=False, **kwds):
if normalize:
Expand Down Expand Up @@ -843,10 +849,6 @@ def date_range(
arr = cp.linspace(start=start, stop=end, num=periods)
result = cudf.core.column.as_column(arr).astype("datetime64[ns]")
return cudf.DatetimeIndex._from_data({name: result})
elif cudf.get_option("mode.pandas_compatible"):
raise NotImplementedError(
"`DatetimeIndex` with `freq` cannot be constructed."
)

# The code logic below assumes `freq` is defined. It is first normalized
# into `DateOffset` for further computation with timestamps.
Expand Down Expand Up @@ -940,7 +942,7 @@ def date_range(
arr = cp.arange(start=start, stop=stop, step=step, dtype="int64")
res = cudf.core.column.as_column(arr).astype("datetime64[ns]")

return cudf.DatetimeIndex._from_data({name: res})
return cudf.DatetimeIndex._from_data({name: res}, freq=freq)


def _has_fixed_frequency(freq: DateOffset) -> bool:
Expand Down
8 changes: 8 additions & 0 deletions python/cudf/cudf/pandas/_wrappers/pandas.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -707,6 +707,14 @@ def Index__new__(cls, *args, **kwargs):
"Resampler", cudf.core.resample._Resampler, pd_Resampler
)

DataFrameResampler = make_intermediate_proxy_type(
"DataFrameResampler", cudf.core.resample.DataFrameResampler, pd_Resampler
)

SeriesResampler = make_intermediate_proxy_type(
"SeriesResampler", cudf.core.resample.SeriesResampler, pd_Resampler
)

StataReader = make_intermediate_proxy_type(
"StataReader",
_Unusable,
Expand Down
128 changes: 124 additions & 4 deletions python/cudf/cudf/tests/test_datetime.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@
import cudf
import cudf.testing.dataset_generator as dataset_generator
from cudf import DataFrame, Series
from cudf.core._compat import PANDAS_GE_150, PANDAS_LT_140
from cudf.core._compat import PANDAS_GE_150, PANDAS_GE_200, PANDAS_LT_140
from cudf.core.index import DatetimeIndex
from cudf.testing._utils import (
DATETIME_TYPES,
Expand Down Expand Up @@ -1571,6 +1571,44 @@ def test_date_range_start_end_freq(request, start, end, freq):
reason="https://github.com/rapidsai/cudf/issues/12133",
)
)
request.applymarker(
pytest.mark.xfail(
condition=(
isinstance(freq, dict)
and freq.get("hours", None) == 10
and freq.get("days", None) == 57
and freq.get("nanoseconds", None) == 3
and (
(
start == "1996-11-21 04:05:30"
and end == "2000-02-13 08:41:06"
)
or (
start == "1970-01-01 00:00:00"
and end == "2000-02-13 08:41:06"
)
or (
start == "1970-01-01 00:00:00"
and end == "1996-11-21 04:05:30"
)
or (
start == "1831-05-08 15:23:21"
and end == "2000-02-13 08:41:06"
)
or (
start == "1831-05-08 15:23:21"
and end == "1996-11-21 04:05:30"
)
or (
start == "1831-05-08 15:23:21"
and end == "1970-01-01 00:00:00"
)
)
),
reason="Nanosecond offsets being dropped by pandas, which is "
"fixed in pandas-2.0+",
)
)
if isinstance(freq, str):
_gfreq = _pfreq = freq
else:
Expand All @@ -1586,7 +1624,29 @@ def test_date_range_start_end_freq(request, start, end, freq):
)


def test_date_range_start_freq_periods(start, freq, periods):
def test_date_range_start_freq_periods(request, start, freq, periods):
request.applymarker(
pytest.mark.xfail(
condition=(
isinstance(freq, dict)
and freq.get("hours", None) == 10
and freq.get("days", None) == 57
and freq.get("nanoseconds", None) == 3
and periods in (10, 100)
and (
start
in {
"2000-02-13 08:41:06",
"1996-11-21 04:05:30",
"1970-01-01 00:00:00",
"1831-05-08 15:23:21",
}
)
),
reason="Nanosecond offsets being dropped by pandas, which is "
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this better solved by fixing the condition on the parameter, which should be "pandas < 2.0"?

https://github.com/shwina/cudf/blob/ed3ba3ff17cf686d1e6e38f01073d27b1be64799/python/cudf/cudf/tests/test_datetime.py#L1512

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wanted to do that but it happens only for a few parameter combinations and we currently xpass/xfail strictly. That's the reason for the current approach.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I know we have two diverging approaches at the same place but I plan on dropping these in pandas-2.0 feature branch.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Okay. We can clean it up later.

"fixed in pandas-2.0+",
)
)
if isinstance(freq, str):
_gfreq = _pfreq = freq
else:
Expand All @@ -1613,6 +1673,28 @@ def test_date_range_end_freq_periods(request, end, freq, periods):
reason="https://github.com/pandas-dev/pandas/issues/46877",
)
)
request.applymarker(
pytest.mark.xfail(
condition=(
isinstance(freq, dict)
and freq.get("hours", None) == 10
and freq.get("days", None) == 57
and freq.get("nanoseconds", None) == 3
and periods in (10, 100)
and (
end
in {
"2000-02-13 08:41:06",
"1996-11-21 04:05:30",
"1970-01-01 00:00:00",
"1831-05-08 15:23:21",
}
)
),
reason="Nanosecond offsets being dropped by pandas, which is "
"fixed in pandas-2.0+",
)
)
if isinstance(freq, str):
_gfreq = _pfreq = freq
else:
Expand Down Expand Up @@ -2163,8 +2245,6 @@ def test_datetime_getitem_na():

def test_daterange_pandas_compatibility():
with cudf.option_context("mode.pandas_compatible", True):
with pytest.raises(NotImplementedError):
cudf.date_range("20010101", "20020215", freq="400h", name="times")
expected = pd.date_range(
"2010-01-01", "2010-02-01", periods=10, name="times"
)
Expand All @@ -2174,6 +2254,46 @@ def test_daterange_pandas_compatibility():
assert_eq(expected, actual)


@pytest.mark.parametrize(
"data,dtype,freq",
[
([10], "datetime64[ns]", "2N"),
([10, 12, 14, 16], "datetime64[ns]", "2N"),
([10, 11, 12, 13], "datetime64[ns]", "1N"),
([100, 200, 300, 400], "datetime64[s]", "100s"),
([101, 201, 301, 401], "datetime64[ms]", "100ms"),
],
)
def test_datetime_index_with_freq(request, data, dtype, freq):
request.applymarker(
pytest.mark.xfail(
condition=(not PANDAS_GE_200 and dtype != "datetime64[ns]"),
reason="Pandas < 2.0 lacks non-nano-second dtype support.",
)
)
actual = cudf.DatetimeIndex(data, dtype=dtype, freq=freq)
expected = pd.DatetimeIndex(data, dtype=dtype, freq=freq)
assert_eq(actual, expected)


@pytest.mark.parametrize(
"data,dtype,freq",
[
([10, 1232, 13244, 13426], "datetime64[ns]", "2N"),
([10, 11, 12, 13], "datetime64[ns]", "1s"),
([10000, 200, 300, 400], "datetime64[s]", "100s"),
([107871, 201, 301, 401], "datetime64[ms]", "100ns"),
],
)
def test_datetime_index_freq_error(data, dtype, freq):
assert_exceptions_equal(
pd.DatetimeIndex,
cudf.DatetimeIndex,
([data], {"dtype": dtype, "freq": freq}),
([data], {"dtype": dtype, "freq": freq}),
)


def test_strings_with_utc_offset_not_implemented():
with pytest.warns(DeprecationWarning, match="parsing timezone"): # cupy
with pytest.raises(NotImplementedError):
Expand Down
Loading