What is the goal or result of this RNN language model ?
guotong1988 opened this issue · 3 comments
guotong1988 commented
Thank you .
gy910210 commented
You can generate text by this RNN language model, more fantastic examples can be found in http://karpathy.github.io/2015/05/21/rnn-effectiveness/.
guotong1988 commented
多谢大神,问下应该训练多少epoch?我训练了10个epoch之后只能生成一些the,a,I之类的词
gy910210 commented
第一,这里的训练集还是太小;第二,只用CPU训练太慢了,10个epoch不一定能收敛;第三,纯RNN对long term的记忆不好,现在主流都是用LSTM等。
PS. 我的代码只是toy,为了理解RNN的机制。
我写了一个用Google Tensorflow训练语言模型的代码,可以参考https://github.com/pangolulu/char-rnnlm-tensorflow
希望对你有帮助。