Hi,
For the detection challenge, the 'example submission notebook' creates a submission with unnormalized scores (which are by definition unbounded, but practically range between approximately [-2, 2]). But I would expect that the backend server requires the scores to be normalized (i.e., between 0 and 1). Which one is correct?
In simpler terms, will the backend server apply a 'sigmoid' on top of our submitted scores? Or should we make sure we have normalized our scores between 0 and 1 before submission?
Posted by: breeze3662 @ July 25, 2022, 10:55 a.m.I just realized, since we care about the AUC curve the normalization of scores probably doesn't matter?
Posted by: breeze3662 @ July 25, 2022, 12:12 p.m.Hello,
This is correct. We are using threshold-independent metrics, so the scores can be unnormalized.
All the best,
Mantas (TDC co-organizer)