read_csv: passing index_col results in KeyError

[willfurnass]

Sample data:

"id1","id2","attr1","attr2","attr3","attr4"
1072.0,"2014-02-22 12:30:00",1,93440248,"False",750
1072.0,"2014-02-22 12:45:00",1,93440248,"False",765
1072.0,"2014-02-22 13:00:00",1,93440248,"False",780
1072.0,"2014-02-22 13:15:00",1,93440248,"False",795

The following results in a KeyError (full stack trace on pastebin.com):

dask_df = dd.read_csv('test.csv',  index_col='id1')

However, setting the index column using the set_index method works (e.g. as follows) for single-columned indexes (MultiIndexes are not yet supported in Dask).

dask_df = dd.read_csv('test.csv')
dask_df = dask_df.set_index('id1')

Issue detected when using Dask 0.7.1 and pandas 0.16.2.

[mrocklin]

Indexes have to be a bit more strange when working in parallel. I think that, for the moment we should just raise an informative error pointing people to use set_index explicitly.