SparrowRecSys icon indicating copy to clipboard operation
SparrowRecSys copied to clipboard

Published 20 hours ago • verified publisher icon wzhe06

DIEN中 auxiliary_loss_layer实现疑问

Open ZJZhang123 opened this issue 2 years ago • 0 comments

https://github.com/wzhe06/SparrowRecSys/blob/579577c9a25f75fc6bbdfeb2f258f7c8a35833dd/TFRecModel/src/com/sparrowrecsys/offline/tensorflow/DIEN.py#L261 请问一下auxiliary_loss_layer的实现逻辑是将positive samples 和 negative samples 分别通过两个不同的网络分支，这不会导致每个网络分支输入的样本都只有正例，没有负例，分布极不均衡，达不到训练目的吗？

ZJZhang123 avatar Nov 22 '21 04:11 ZJZhang123