[FEA] Implement basic reductions for decimal columns

[ChrisJar]

Is your feature request related to a problem? Please describe.
I wish to be able to perform basic reductions like .sum(), .mean(), or .std() on DataFrames and Series with decimal columns.

Describe the solution you'd like
I would like to mimic the reductions available for float columns. For example:

df = cudf.DataFrame({'id': [0, 1, 1], 'val': [1.00, 1.01, 1.02]})
df.sum()

returns

id     2.00
val    3.03
dtype: float64

but if the dataframe contains a decimal column

df['val'] = cudf.Series([decimal.Decimal(x) for x in [1.00, 1.01, 1.02]], dtype=cudf.Decimal64Dtype(7,3))
df.sum()

it returns

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-43-7e5fdb616c56> in <module>
----> 1 df.sum()

/home/u00u7rh1e72hXfsipJ357/miniconda3/envs/rapids-gpu-bdb/lib/python3.7/site-packages/cudf/core/dataframe.py in sum(self, axis, skipna, dtype, level, numeric_only, min_count, **kwargs)
   6063             numeric_only=numeric_only,
   6064             min_count=min_count,
-> 6065             **kwargs,
   6066         )
   6067 

/home/u00u7rh1e72hXfsipJ357/miniconda3/envs/rapids-gpu-bdb/lib/python3.7/site-packages/cudf/core/dataframe.py in _apply_support_method(self, method, axis, *args, **kwargs)
   6788             result = [
   6789                 getattr(self[col], method)(*args, **kwargs)
-> 6790                 for col in self._data.names
   6791             ]
   6792 

/home/u00u7rh1e72hXfsipJ357/miniconda3/envs/rapids-gpu-bdb/lib/python3.7/site-packages/cudf/core/dataframe.py in <listcomp>(.0)
   6788             result = [
   6789                 getattr(self[col], method)(*args, **kwargs)
-> 6790                 for col in self._data.names
   6791             ]
   6792 

/home/u00u7rh1e72hXfsipJ357/miniconda3/envs/rapids-gpu-bdb/lib/python3.7/site-packages/cudf/core/series.py in sum(self, axis, skipna, dtype, level, numeric_only, min_count, **kwargs)
   2985 
   2986         return self._column.sum(
-> 2987             skipna=skipna, dtype=dtype, min_count=min_count
   2988         )
   2989 

/home/u00u7rh1e72hXfsipJ357/miniconda3/envs/rapids-gpu-bdb/lib/python3.7/site-packages/cudf/core/column/column.py in sum(self, skipna, dtype, min_count)
   1257         self, skipna: bool = None, dtype: Dtype = None, min_count: int = 0
   1258     ):
-> 1259         raise TypeError(f"cannot perform sum with type {self.dtype}")
   1260 
   1261     def product(

TypeError: cannot perform sum with type decimal

The same error is returned for other basic reductions like .mean() and .std() as well.

[karthikeyann]

I can't reproduce this error. I get this error on astype.
TypeError: Cannot interpret 'Decimal64Dtype(precision=7, scale=3)' as a data type

can you give details on your environment?

Following code recreated the error in branch-0.19.

import decimal
import cudf
s = cudf.Series([decimal.Decimal(x) for x in [0.5, 1.5,2.5]], dtype=cudf.Decimal64Dtype(7,3))
s.sum()

[karthikeyann]

@codereport added decimal type reduction in libcudf. Python interface might be needed to utilize this.

[ChrisJar]

I can't reproduce this error. I get this error on astype.
TypeError: Cannot interpret 'Decimal64Dtype(precision=7, scale=3)' as a data type

can you give details on your environment?

Following code recreated the error in branch-0.19.

import decimal
import cudf
s = cudf.Series([decimal.Decimal(x) for x in [0.5, 1.5,2.5]], dtype=cudf.Decimal64Dtype(7,3))
s.sum()

My apologies that astype call uses a PR that hasn't been merged yet. I'll edit the original post to your code

[codereport]

@codereport added decimal type reduction in libcudf. Python interface might be needed to utilize this.

Correct. sum reductions are implemented, but don't have Python bindings yet. We only support a few binary operations atm.

[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.

[ChrisJar]

Closed with #7776