Incorrect ``unique`` result for column with numerical name

[rjzamora]

I'm getting an incorrect unique result after selecting a numerical column name from a DataFrame collection:

import pandas as pd
import dask.dataframe as dd

pdf = pd.DataFrame({1: [0,3,0,1,2,0,3,1,1]})
df = dd.from_pandas(pdf, 3)

df[key].unique().compute()

0    0
1    1
2    3
0    3
1    2
2    0
3    1
Name: 1, dtype: int64

I haven't had time to dig into this yet, but it seems like things work fine when the column name is 0. Also, changing the shuffle algorithm doesn't seem to resolve the issue.

[phofl]

0 only worked by accident :)

[rjzamora]

0 only worked by accident :)

Heh - Yeah, I just discovered the same thing. Doesn't seem like there is support for int column names at the moment.

[phofl]

Not if you shuffle there (but my pr addresses it)