README.md

memnet

Simplest end to end memory network implemented in tensorflow

File Structure

dataprovider.py

Data Source:

BABI, a simulated reading comprehension dataset with 20 tasks

Data Download:

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

Data Structure:

We have two classes:

1. GenData:
   Load data for specific task(among the 20 tasks), build dictionary and train/test set, each set is an instance of Data class.
   
2. Data:
   Data class includes 3 attributes: h(history), q(query) and a(answer). This class implements a function to generate next batch each time.
   

memN2N.py

The implementation of End-to-end-memory-network.

main.py

Call memN2N.py for train.