search

ShannonAI / mrc-for-flat-nested-ner

Code for ACL 2020 paper `A Unified MRC Framework for Named Entity Recognition`
643 stars 117 forks source link

关于start_logits和end_logits的疑问 #57

Closed xuanzebi closed 3 years ago

xuanzebi commented 3 years ago

Trainer.py 代码241行 image 得到start_logits和end_logits后，这两个是没有过softmax/sigmoid的吧，这个时候不应该过一下sigmoid函数再然后和0.5比较得到二分类的结果吗？

YuxianMeng commented 3 years ago

@xuanzebi x>=0 is equivalent to sigmoid(x)>=0.5

xuanzebi commented 3 years ago

好吧 是等价的两个