In version 0.7.5, when reading in a CSV file that has numerical column names assigned by pandas, dask.dataframe.read_csv misinterprets the column names that have been assigned by pandas.

In particular, the column names (in the code sample ",0,1,2") given on the first row are interpreted as the first row of data by dask.dataframe.read_csv whereas they are correctly identified as column names by the pandas.read_csv method.

See the attached file for a minimal code sample that reproduces the error.