In subchallenge2 of round, our submission (syn11319288) is showing reasonable correlation but nrmse is very high around 66276425 . We do not understand reason of this high nrmse, it seems that range of our prediction (+2 to -2 in trading set) is different than in validation dataset. Any help in this direction will be highly appreciated.
Created by Gajendra Raghava raghava NRMSE is evaluated based on both biological missing events (d.true=0) and technical missing events (d.true!=0). Cor is evaluated only based on technical missing events (d.true!=0)
So, while Cor is used in sub2 and Sub3 as the key score for determine the rank, it cannot be applied to evalute biological missing events. So NRMSE (also a common used criteria) is used as the key score in sub1.