Simplest end to end memory network implemented in tensorflow
BABI, a simulated reading comprehension dataset with 20 tasks
mkdir data && cd data
wget http://www.thespermwhale.com/jaseweston/babi/tasks_1-20_v1-2.tar.gz
tar -zxf tasks_1-20_v1-2.tar.gz
We have two classes:
- GenData:
Load data for specific task(among the 20 tasks), build dictionary and train/test set, each set is an instance of Data class. - Data:
Data class includes 3 attributes: h(history), q(query) and a(answer). This class implements a function to generate next batch each time.
The implementation of End-to-end-memory-network.
Call memN2N.py for train.