ur work is very perfect!
But the raw paper use the previous embedding and cell_state as input to calculate current cell_state as context left.your code use normal RNN(current embedding and previous cell_state) to calculate the cell_state.
I try write code as raw paper,but i find it is very slow.
what is your opinion?
Thanks for your code. But I have a problem. In the raw paper, theta represent all paramaters ,including "E; b(2); b(4); cl(w1); cr(wn); W (2);W (4); W (l); W (r); W (sl); W (sr)", these are all trained while training. But in your code, there are no "W (l); W (r); W (sl); W (sr)" and "cl(w1)" always setted to be zero vector because of the use of biRNN, is that a problem?Thanks!
i have tried to run your code but embedding part is not running. Its giving me various errors such as "vocab_size is not defined", "word_embedding_size is not defined" and "self is not defined". Please tell me how to get rid of these problems.
Seems like when tensorflow updated to version 2 they left out tf.contrib especially the VocabularyProcessor.
Is there an alternative for this line to make the rest of your code still work? Thanks
Thanks for your code ! I wanna ask what is your accuracy on the training set and dev set during the training process ? I am a little bit confused because my best result is 0.67 for the training set, 0.55 for the dev set. I do not use pre-training word2vec, num_epochs was set to 50, and other experiment configurations are just default
Hello!
your code is really perfect,but i wonder whether your project support Multi-Classification by modify the array value in func load_data_and_labels()?
thx!
Thanks for your code,however,I have a problem when running "eval.py".
File "F:\Program Files (x86)\python3.6\00\lib\site-packages\tensorflow\contrib\learn\python\learn\preprocessing\text.py", line 246, in restore
return pickle.loads(f.read())
File "F:\Program Files (x86)\python3.6\00\lib\site-packages\tensorflow\python\lib\io\file_io.py", line 125, in read self._preread_check()
File "F:\Program Files (x86)\python3.6\00\lib\site-packages\tensorflow\python\lib\io\file_io.py", line 85, in _preread_check
compat.as_bytes(self.__name), 1024 * 512, status)
File "F:\Program Files (x86)\python3.6\00\lib\site-packages\tensorflow\python\framework\errors_impl.py", line 526, in exit
c_api.TF_GetCode(self.status.status))
tensorflow.python.framework.errors_impl.NotFoundError: NewRandomAccessFile failed to Create/Open: ..\text_vocab : ϵͳ\udcd5Ҳ\udcbb\udcb5\udcbdָ\udcb6\udca8\udcb5\udcc4\udcceļ\udcfe\udca1\udca3
; No such file or directory
Please tell me how can I get rid of this bug,thanks a lot!
Thanks for sharing code. It seems that your implement of context representation vectors (via Bi-RNN) differ from the original paper's (Equation 1&2). Have you compare the performance between them and what's your intuition? :)
Thanks for sharing code!
I have read https://zhuanlan.zhihu.com/p/42201550, in which there is avctivition function tanh after w_2 * x + b_2。
but the code dont have this
Is there something I missed?