add neural sequence tagging model for QA #390
Conversation
neural_seq_qa/config.py
Outdated
| # dimension of the attention layer | ||
| self.latent_chain_dim=64 | ||
| # dimension of the evidence LSTMs | ||
| self.e_lstm_dim=64 |
There was a problem hiding this comment.
This file is not formatted.
neural_seq_qa/infer.py
Outdated
| import sys | ||
| import argparse | ||
|
|
||
| import paddle.v2 as pd |
There was a problem hiding this comment.
I prefer to use import paddle.v2 as paddle to keep consistent with all the other PaddlePaddle demos in models and book.
There was a problem hiding this comment.
All pd have been changed to paddle
neural_seq_qa/network.py
Outdated
| :rtype: LayerOutput | ||
| """ | ||
| return pd.layer.embedding( | ||
| name=name, |
There was a problem hiding this comment.
To follow current PaddlePaddle's style, if the name is not explicitly used, it is set automatically by PaddlePaddle.
Currently, in all the PaddlePaddle's released demos, except the data layer, all the other layers do not need the name.
There was a problem hiding this comment.
The names of the layers have been removed.
neural_seq_qa/README.md
Outdated
| ``` | ||
| # Validation and Test | ||
|
|
||
| WebQA provoides two versions of validation and test sets. Automatic valiation and test can be lauched by |
There was a problem hiding this comment.
provoides -> provides, valiation ->validation
neural_seq_qa/network.py
Outdated
|
|
||
| def get_embedding(input, word_vec_dim, wordvecs): | ||
| """ | ||
| Defined word embedding |
neural_seq_qa/network.py
Outdated
| :type drop_rate: float | ||
| :param wordvecs: word embedding matrix | ||
| :type wordvecs: numpy array | ||
| :param default_init_std: default initial std |
There was a problem hiding this comment.
Please use the full name of std in all files.
|
@lcy-seso This is a good idea. Shall we provide it in another place? Otherwise, this repo will become very large in the future. |
|
Yes~ We do have a place to save the trained model (You can give the model to me, I have the account). We have this example https://github.com/PaddlePaddle/models/blob/develop/image_classification/models/model_download.sh#L29. We can only add the download/evaluation script into the models repo, and update the README. |
Implement model proposed in Peng Li, Wei Li, Zhengyan He, Xuguang Wang, Ying Cao, Jie Zhou, and Wei Xu. Dataset and Neural Recurrent Sequence Labeling Model for Open-Domain Factoid Question Answering. arXiv:1607.06275.