Zhenghua-read-codes-hmm-em

train.sh -> run.py

from tagger.cmds import Evaluate, Predict, Train

tagger/cmds/train.py Train

Train()
def __call__
  train = Corpus.load(config.ftrain)
  trainset = TextDataset(vocab.numericalize(train))
          train_loader = batchify(trainset, config.batch_size, False)
       print(f"{'train:':6} {len(trainset):5} sentences, {train.nwords} words in total, "
             f"{len(train_loader):3} batches provided")

       tagger = Tagger(config) # from tagger import Tagger, Model  tagger/tagger.py  tagger/model.py 
       # Tagger去做前向后相、viterbi等
       # Model基于Tagger，做EM参数更新

 迭代若干次，看准确率变化，多少次不增长就停止 patience。感觉这个patience比较怪，loss也应该看一下