rank loss func bug
Opened this issue · 4 comments
MLightM commented
rank loss should be ''..... - tf.multiply(pre_pw_dif, gt_pw_dif) ...‘’, '-' is missing when define the loss in rank_lstm.py
fulifeng commented
Come on! Please check the difference when we calculate: pre_pw_dif and gt_pw_dif
MLightM commented
Come on! Please check the difference when we calculate: pre_pw_dif and gt_pw_dif
oh... yes, you are right, thanks!
wjhmike95 commented
why do you use "mean" to reduce rank loss instead of "sum" which is from loss equation in the essay?
ownnaruto commented
Is this loss really useful?I try so many times, just find that the results without rank loss are much better!