Sequence_Labeling

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