demo 畫面 https://github.com/sinlin0908/QAmodel_Demo

• QAnet Chinese Reading Comprehension
  • Reference
  • Environment
  • Requirement
  • Data Set
  • Modify
    • prepro file
    • embedding dat set
    • Config
  • Usage
  • Performance
  • Tensorbord
  • Comparison Chart

• QAnet by google
• QAnet Code by NLP_learn
• Data Set by Delta
• Embedding data set by Chinese-Word-Vectors

• OS: Ubuntu 18.04 LTS
• GPU: GTX 1080Ti 11G
• CPU: i7-4770
• RAM: 16G

• Python 3.6
• NumPy
• tqdm
• Tensorflow>=1.5
• Jieba
• opencc
• bottle

• train set : 26936 questions
• dev set : 3524 questions
• test 3493 : questions

• word token: use jieba.cut(context,cut_all=False)
• _getword() delete word.lower(), word.capitalize(), word.upper()

• use 1292607 words 300d embedding data set
• use 14082 characters 300d character embedding data set

• word size: 1292607
• hidden size: 128
• num_head: 8
• batch size: 12
• char_emb_size : 300d
• pretrain_char -> True

python config.py --mode prepro

train

python config.py --mode train

test

python config.py --mode test

demo:

python config.py --mode demo

• F1: score 70.0496230556
• EM: 70.0257658173
• cost: 6 hours
• use GPU memory : 9.4G

tensorboard --logdir=./

notice: character embedding has a little effect