Closed liuhu-bigeye closed 4 years ago
See: https://github.com/zhougr1993/DeepInterestNetwork/issues/63
The batch_norm params for hist_j should be shared with hist_i, otherwise it will be never trained and using hist_j in eval phase is equivalent to exposing the label.
See: https://github.com/zhougr1993/DeepInterestNetwork/issues/63
The batch_norm params for hist_j should be shared with hist_i, otherwise it will be never trained and using hist_j in eval phase is equivalent to exposing the label.