Tensorflow implementation of Teaching machine read and comprehend
introduce and tutorial[http://rsarxiv.github.io/2016/06/18/教机器学习阅读/]
This implementation contains:
Google DeepMind's Teaching Machines to Read and Comprehend:
1. Deep LSTM Reader
2. Attentive Reader
- with [Bidirectional LSTMs](http://www.cs.toronto.edu/~graves/nn_2005.pdf) with peephole weights
Facebook:
3. End 2 End Memory Network
- End-To-End Memory Networks[http://arxiv.org/pdf/1503.08895v5.pdf]
- The Goldilocks Principle: Reading Children's Books with Explicit Memory Representations[http://arxiv.org/abs/1511.02301]
- Python 2.7 or Python 3.3+
- Tensorflow
- NLTK
- Gensim
First, you need to download DeepMind Q&A Dataset from here, save cnn.tgz and dailymail.tgz into the repo, and run:
$ ./unzip.sh cnn.tgz dailymail.tgz
Then run the pre-processing code with:
$ python data_utils.py data cnn
To train a model with cnn dataset:
$ python main.py --dataset cnn
To test an existing model:
$ python main.py --dataset cnn --forward_only True
Thanks for Author Taehoon Kim / @carpedm20 provide our firest version tensorflow code
author: wujs(https://github.com/wujsAct)