[FEA] Support for groupby rolling aggregations on time windows in dask-cudf

[beckernick]

Today, Dask supports several groupby rolling operations on time frequency windows (e.g., "1 hour"), including canonical reductions like sum, max, min, mean, std, and var.

These operations are not supported with dask-cuDF. We should support these operations.

They currently fail with the following error:

import dask
import dask_cudf
​
ddf = dask.datasets.timeseries(freq="1H")
gddf = dask_cudf.from_dask_dataframe(ddf)
​
print(ddf.groupby("name").x.rolling("1D").var().head())
​
gddf.groupby("name").x.rolling('1D').mean().head()
name     timestamp          
Zelda    2000-01-01 00:00:00   NaN
Sarah    2000-01-01 01:00:00   NaN
George   2000-01-01 02:00:00   NaN
Norbert  2000-01-01 03:00:00   NaN
Wendy    2000-01-01 04:00:00   NaN
Name: x, dtype: float64
/home/nicholasb/conda/envs/cudf-22.04/lib/python3.8/site-packages/dask_cudf/groupby.py:54: FutureWarning: index is deprecated and will be removed in a future release.
  by=self.index,
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Input In [16], in <module>
      5 gddf = dask_cudf.from_dask_dataframe(ddf)
      7 print(ddf.groupby("name").x.rolling("1D").var().head())
----> 9 gddf.groupby("name").x.rolling('1D').mean().head()

File ~/conda/envs/cudf-22.04/lib/python3.8/site-packages/dask/dataframe/rolling.py:352, in Rolling.mean(self)
    350 @derived_from(pd_Rolling)
    351 def mean(self):
--> 352     return self._call_method("mean")

File ~/conda/envs/cudf-22.04/lib/python3.8/site-packages/dask/dataframe/rolling.py:492, in RollingGroupby._call_method(self, method_name, *args, **kwargs)
    491 def _call_method(self, method_name, *args, **kwargs):
--> 492     return super()._call_method(
    493         method_name,
    494         *args,
    495         groupby_kwargs=self._groupby_kwargs,
    496         groupby_slice=self._groupby_slice,
    497         **kwargs,
    498     )

File ~/conda/envs/cudf-22.04/lib/python3.8/site-packages/dask/dataframe/rolling.py:300, in Rolling._call_method(self, method_name, *args, **kwargs)
    298 def _call_method(self, method_name, *args, **kwargs):
    299     rolling_kwargs = self._rolling_kwargs()
--> 300     meta = self.pandas_rolling_method(
    301         self.obj._meta_nonempty, rolling_kwargs, method_name, *args, **kwargs
    302     )
    304     if self._has_single_partition:
    305         # There's no overlap just use map_partitions
    306         return self.obj.map_partitions(
    307             self.pandas_rolling_method,
    308             rolling_kwargs,
   (...)
    313             **kwargs,
    314         )

File ~/conda/envs/cudf-22.04/lib/python3.8/site-packages/dask/dataframe/rolling.py:488, in RollingGroupby.pandas_rolling_method(df, rolling_kwargs, name, groupby_kwargs, groupby_slice, *args, **kwargs)
    486 if groupby_slice:
    487     groupby = groupby[groupby_slice]
--> 488 rolling = groupby.rolling(**rolling_kwargs)
    489 return getattr(rolling, name)(*args, **kwargs).sort_index(level=-1)

File ~/conda/envs/cudf-22.04/lib/python3.8/site-packages/cudf/core/groupby/groupby.py:729, in GroupBy.rolling(self, *args, **kwargs)
    720 def rolling(self, *args, **kwargs):
    721     """
    722     Returns a `RollingGroupby` object that enables rolling window
    723     calculations on the groups.
   (...)
    727     cudf.core.window.Rolling
    728     """
--> 729     return cudf.core.window.rolling.RollingGroupby(self, *args, **kwargs)

TypeError: __init__() got an unexpected keyword argument 'win_type'

Env:

conda list | grep "rapids\|dask" cudf 22.04.00a220131 cuda_11_py38_gc25d35b361_93 rapidsai-nightly dask 2022.1.0 pyhd8ed1ab_0 conda-forge dask-core 2022.1.0 pyhd8ed1ab_0 conda-forge dask-cudf 22.04.00a220131 cuda_11_py38_gc25d35b361_93 rapidsai-nightly libcudf 22.04.00a220131 cuda11_gc25d35b361_93 rapidsai-nightly librmm 22.04.00a220131 cuda11_g81d523a_15 rapidsai-nightly ptxcompiler 0.2.0 py38h98f4b32_0 rapidsai-nightly rmm 22.04.00a220131 cuda11_py38_g81d523a_15_has_cma rapidsai-nightly

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[github-actions]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[entelechie]

Lacking of this basic functionaltiy makes cudf very difficult to use.