Zeping Yu
这篇论文的结构是可以解决分类问题的，如果是像翻译等sequence to sequence模型(m->n)也可以，把srnn当作encoder，decoder从encoder的向量里面解就可以了。之前也有其它人用srnn做了text generation，github应该可以搜到。如果是像分词、命名实体识别等structure learning的问题(m->m)，图中的结构应该是不能的，但也是有办法解决的，可以参考wavenet的结构：https://deepmind.com/blog/wavenet-generative-model-raw-audio/
Actually the speed of SRNN could be much faster if improved recurrent units are used, in that case SRNN is even faster than CNN. I did not mention this in...
我查了下，好像是tensorflow版本的原因吧~我记得我的tensorflow版本是1.4~希望对你有帮助
@SlashChven @ruanchaves , @demonSong has the answer, hope it will help:)
The timedistributed layer in keras could do this.
updated in readme. python=2.7, tensorflow=1.6.0, keras=2.1.5
你好，抱歉我刚刚看到。 Contrary to other reviews, I have zero complaints about the service or the prices. I have been getting tire service here for the past 5 years now, and compared ...
Sorry, I havent learned pytorch. But I think I will learn it in some days. I will implement one after I learn it.
I'm trying now, and I think it's OK.
That's cool, thank you very much!