I have not reproduced the result in the paper if anyone reproduced pls let me know (ENV:pytho3 and dynet 2.0 PKU)
-
no pre_train score is 0.954 for pku_dev and 0.947 for pku_test
-
100d score is 0.047 for pku_test
Hi, this code is easy to use!
Please check the src/train.py
for all hyper-parameter and IO settings.
You can modify the src/train.py
to speficy your own model settings or datasets.
- For training, use the command line
python train.py
. Training details will be printed on the screen. The learned parameters will be saved in in the same directory astrain.py
per epoch, which will be named asepoch1
,epoch2
,...
. - For test, the same command line
python train.py
is used, but with a specified parameter file (e.g.,epoch1
), via the function argumentload_params
intrain.py
(Noteload_params
should beNone
when training). In addition, tell your test file by settingdev_file
(Yes, when test, consider it as "test_file"). The segmented result will be saved insrc/result
.
The code is originally designed for reasearch purpose, but adaptable to industrial use.
This code implements an efficient and effective neural word segmenter proposed in the following paper.
Deng Cai, Hai Zhao, etc., Fast and Accurate Neural Word Segmentation for Chinese. ACL 2017.
If you find it useful, please cite the paper.
@InProceedings{cai-EtAl:2017:Short,
author = {Cai, Deng and Zhao, Hai and Zhang, Zhisong and Xin, Yuan and Wu, Yongjian and Huang, Feiyue},
title = {Fast and Accurate Neural Word Segmentation for Chinese},
booktitle = {Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)},
month = {July},
year = {2017},
address = {Vancouver, Canada},
publisher = {Association for Computational Linguistics},
pages = {608--615},
url = {http://aclweb.org/anthology/P17-2096}
}
Drop me (Deng Cai) an email at thisisjcykcd (AT) gmail.com if you have any question.