dask.dataframe.groupby with tuple keys fails

To reproduce:

In [1]: import numpy as np; np.__version__
Out[1]: '1.13.3'

In [2]: import pandas as pd; pd.__version__
Out[2]: '0.21.1'

In [3]: import dask.dataframe; dask.__version__
Out[3]: '0.16.0'

In [4]: df = pd.DataFrame(np.random.choice((0,1), (10, 3)), columns=list('abc'))

In [5]: df.groupby('b').apply(len)
Out[5]: 
b
0    3
1    7
dtype: int64

In [6]: df.groupby(('b', 'c')).apply(len)
Out[6]: 
b  c
0  1    3
1  0    4
   1    3
dtype: int64

In [7]: ddf = dask.dataframe.from_pandas(df, npartitions=2)

In [8]: ddf.groupby('b').apply(len, meta=int).compute()
Out[8]: 
b
1    7
0    3
dtype: int64

In [9]: ddf.groupby(('b', 'c')).apply(len, meta=int).compute()
...
ValueError: Wrong number of items passed 0, placement implies 5

[jcrist]

Looks like we enforce that this is a list, not a tuple. The following works for me:

In [15]: ddf.groupby(['b', 'c']).apply(len, meta=int).compute()
Out[15]:
b  c
0  1    2
1  0    4
0  0    3
1  1    1
dtype: int64

This is probably a simple thing to fix, if you’d like to submit a PR 😃.

[TomAugspurger]

The issue in https://github.com/dask/dask/issues/3047#issuecomment-355595521 is still present (I actually get a SystemError now, fun). But I think given pandas’ difficulties with handling keys as tuples I’m comfortable ignoring it now.

The original issue of specifying a list of keys as tuples currently (correctly) raises in both pandas and Dask.