Closed chenjiyan2001 closed 2 years ago
According to the definition of bpr loss function, here should be modified as: out=((final_user*final_pos).sum(1)-(final_user*final_nega).sum(1)).sigmoid(), otherwise, loss will be nan when using the bpr loss function
out=((final_user*final_pos).sum(1)-(final_user*final_nega).sum(1)).sigmoid()
According to the definition of bpr loss function, here should be modified as:
out=((final_user*final_pos).sum(1)-(final_user*final_nega).sum(1)).sigmoid(), otherwise, loss will be nan when using the bpr loss function