Hi,
has someone encountered a problem regarding really worse results on test set with respect to validation? Even using a different and bigger validation set results don't change.
Moreover it seems that it is not a metric problem, since the issue remains regardless of the used metric (balanced accuracy, macro average accuracy etc).
If someone found a solution it would be nice to hear about it.
Thank you!
Posted by: Benedetta @ Nov. 17, 2021, 2:43 p.m.It seems like test sets have a quite different distribution w.r.t. available data... maybe some trick seen during lectures can solve this issue :)
Posted by: TacchinoLeso @ Nov. 23, 2021, 4:03 p.m.