This is a fork of https://github.com/danqi/rc-cnn-dailymail. code for A Thorough Examination of the CNN/Daily Mail Reading Comprehension Task. For exaplnation go to the original repository.
This fork is to enable quering QA on a trained model. A trained model over the CNN dataset is available here.
- Python 2.7
- Theano >= 0.7
- Lasagne 0.2.dev1
For more explanation about training a model go to the original repository.
When running python code/run_qa_model.py --queries_folder folder_path ..
you start a QA stream that expects questions in folder_path/queries.pkl
and returns its rewarding accuracy in folder_path/rewards.txt
Also required to pass:
--train_path --dev_path --glove_path