查看“Zhenghua-read-codes-hmm-em”的源代码

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