Page 1 of 1

submission file

PostPosted: Tue Jan 10, 2012 3:46 am
by sofos
Is it possible to predict 0 labels for some test instances? In some cases the algorithm gives low scores for all labels..

Re: submission file

PostPosted: Tue Jan 10, 2012 3:55 am
by sofos
I actually just tried to submit a prediction containing some empty rows and I got an error. I believe that empty rows should be allowed!

Re: submission file

PostPosted: Tue Jan 10, 2012 10:01 am
by NosferatoCorp
Hello,

The empty rows were prohibited for practical reasons. In many cases it may help participants to realize that there is something wrong with their solution.

Every document in the both data sets has at least one label. If your algorithm does not assign any topic to some cases, it is actually better for you to give them some randomly selected labels. It will never make your score worse than it would be in a case of leaving an empty row (precision equal to zero).

Best regards,
Andrzej Janusz