[FEA]use cuDF dataframe cannot use unstack() function

[dominicshanshan]

Describe the bug
cuDF dataframe not support .unstack() function

Steps/Code to reproduce bug

issue_test_df = cudf.DataFrame({'0':[1,1,1,2,2,3,3,3,3],
                                '1':[1,2,3,1,2,1,2,3,4],
                                '2':[0.2,0.2,0.3,0.3,0.4,0.5,0.6,0.7,0.8]})
issue_test_df.unstack()

---------------------------------------------------------------------------
NotImplementedError                       Traceback (most recent call last)
<ipython-input-63-3fff8526a571> in <module>
----> 1 issue_test_df.unstack()

/opt/conda/envs/rapids/lib/python3.7/site-packages/cudf/core/dataframe.py in unstack(self, level, fill_value)
   7061     def unstack(self, level=-1, fill_value=None):
   7062         return cudf.core.reshape.unstack(
-> 7063             self, level=level, fill_value=fill_value
   7064         )
   7065 

/opt/conda/envs/rapids/lib/python3.7/site-packages/cudf/core/reshape.py in unstack(df, level, fill_value)
    897     if not isinstance(df.index, cudf.MultiIndex):
    898         raise NotImplementedError(
--> 899             "Calling unstack() on a DataFrame without a MultiIndex "
    900             "is not supported"
    901         )

NotImplementedError: Calling unstack() on a DataFrame without a MultiIndex is not supported

Expected behavior
Like pandas.DataFrame().unstack(), it should output the unstack table

Environment overview (please complete the following information)

• Environment location: Docker
• Method of cuDF install: Docker
  • If method of install is [Docker], provide docker pull & docker run commands used
    sudo docker run --gpus all -it -v /data/project/demo/:/rapids/notebooks/demo/ -p 8889:8888 nvcr.io/nvidia/rapidsai/rapidsai:cuda11.0-runtime-ubuntu18.04

Please let me know if this is enough to reproduce the issue, thanks !

[taureandyernv]

@dominicshanshan Thanks for bringing this to our attention. This seems more like a feature request than a bug as the caught error states that this feature isn't implemented. Can you change the issue from a bug [BUG] to a feature request [FEA]?

898 raise NotImplementedError(
--> 899 "Calling unstack() on a DataFrame without a MultiIndex "
900 "is not supported"

It does seem like we didn't quite reach full equivalence to pandas as hoped for in the closed issue #1821. Should we reopen it @kkraus14?

Following the pandas example from https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.unstack.html,

Pandas attempt:

import pandas as pd
import numpy as np
index = pd.MultiIndex.from_tuples([('one', 'a'), ('one', 'b'),
                                   ('two', 'a'), ('two', 'b')])

# FYI, cuDF unstack() doesn't support Series, so I used switched both to DataFrame
s = pd.DataFrame(np.arange(1.0, 5.0), index=index)
print(s)
print(s.unstack(level=-1))
print(s.unstack(level=0))
df = s.unstack(level=0)
print(df.unstack()) 

Successfully outputs

         0
one a  1.0
    b  2.0
two a  3.0
    b  4.0

       0     
       a    b
one  1.0  2.0
two  3.0  4.0

     0     
   one  two
a  1.0  3.0
b  2.0  4.0

0  one  a    1.0
        b    2.0
   two  a    3.0
        b    4.0
dtype: float64

import cudf
import numpy as np
index = cudf.MultiIndex.from_tuples([('one', 'a'), ('one', 'b'),
                                   ('two', 'a'), ('two', 'b')])

gs = cudf.DataFrame(np.arange(1.0, 5.0), index=index)
print(gs)
print(gs.unstack(level=-1))
print(gs.unstack(level=0))
df = gs.unstack(level=0)
print(df.unstack()) #throws error

Outputs

         0
one a  1.0
    b  2.0
two a  3.0
    b  4.0

       0     
       a    b
0            
one  1.0  2.0
two  3.0  4.0

     0     
   one  two
1          
a  1.0  3.0
b  2.0  4.0

NotImplementedError                       
Traceback (most recent call last)

...

NotImplementedError: Calling unstack() on a DataFrame without a MultiIndex is not supported

[dominicshanshan]

@taureandyernv , as you request, changed label from BUG -> FEA

[isVoid]

This should be a workaround for now:

>>> df = cudf.DataFrame({'B': [1, 3, 5],
...                    'C': [2, 4, 6]})
>>> df.T.stack(dropna=False)
0  1
B  0    1
   1    3
   2    5
C  0    2
   1    4
   2    6
dtype: int64

Note that cudf requires all columns to have the same datatype to complete the transpose. Otherwise some type-casts may be needed.