xiangrongzeng / copy_re

Release for acl18 paper "Extracting Relational Facts by an End-to-End Neural Model with Copy Mechanism"
197 stars 36 forks source link

f1/precision/recall #5

Closed wangxshu closed 5 years ago

wangxshu commented 5 years ago

您好!请问我用默认设置训练webnlg的时候f1/precision/recall 只有0.218 0.182 0.199 会是什么原因呢。我尝试过将cell换成gru,然后看了上面的回答把learning rate改成0.01,也只是提升到0.352 0.358 0.346 。如果可以的话想请教一下改进的方法。

xiangrongzeng commented 5 years ago

Duplicate of #2 请参考issue#2,学习率设置为0.001