Try to finish, Bilstm+CRF, Bert+CRF, model
First Model: Bilstm CRF Optional parameter:
- sinusoid (For position embedding.True means you could use sinusoid to initial position embedding.)
- use_pretrain (True means use golve or word2vec to initial tokens embedding.)
- cell ("lstm" or "gru")
- use_crf (True use crf as decode layer)
- normalize (Use batch normaliz)
- others (see Config.py for more information!)
Tricks: label smoothing
Using:
- Build dir first (logdir,vocab,Embedding)
- download glove or word2vec (here i use glove200d)
- python traing.py
- python eval.py