GCH

大佬你好，请问HMM的参数学习过程是如何体现极大似然的EM算法的呢，train函数中三个参数矩阵仅仅是按照标签频率归一以后初始化，并没有看到学习的过程，初入门，望指点！

ERNIE 预训练模型下载不了

2

{"error":"Document not found"}

Hello，in my opinion，the negative training phase(negative classes random sampling and NLLoss) could be included by CrossEntropyLoss(postive and all negative classedxs) why use negative training?