dengwc/attention-over-attention

Tensorflow implementation of Attention-over-Attention Neural Networks for Reading Comprehension

Attentive-over-Attention Neural Networks for Reading Comprehension

Tensorflow implementation of Attentive-over-Attention Neural Networks.

Prerequisites

• Python 3.5+
• Tensorflow
• NLTK
• Gensim

Usage

First, download DeepMind Q&A Dataset from here, and untar cnn.tgz and dailymail.tgz into data directory:

Then run the pre-processing code with:

$ ./prepare-rc.sh

To train a model with cnn dataset:

$ python3 main.py --dataset cnn -t

To test an existing model (in progress):

$ python3 main.py --dataset cnn

Credit

Modified codes for pre-processing, shuffling, and loading dataset are originally from IBM's Attention Sum Reader implementation.

Results (in progress)