Error in dataset columns
#2
by
d-rau
- opened
During the autmatic split generation after download I get the following error:
All the data files must have the same columns, but at some point there are 1 new columns
({'answers'}) and 2 missing columns ({'positive_passages', 'negative_passages'}).
Could you please have a look at this?
Could be fixed by creating the same columns for train and test and initializing them with empty strings.
However, I have a second concern which regards ranking evaluation, why does the test only contain query, answer pairs, instead of query, relevant_doc pairs?
Understood from the paper, your eval is based on exact match. Closing the issue.
d-rau
changed discussion status to
closed