harvardnlp / memn2n

Torch implementation of End-to-End Memory Networks (https://arxiv.org/abs/1503.08895)

End-To-End Memory Network

Torch implementation of MemN2N (Sukhbaatar, 2015). Supports Adjacent Weight Tying, Position Encoding, Temporal Encoding and Linear Start. Code uses v1.0 of bAbI dataset with 1k questions per task.

Prerequisites:

• Python 2.7
• Torch (with nngraph)

Preprocessing

First, preprocess included data into hdf5 format:

python preprocess.py

This will create a hdf5 file for each task (total 20 tasks).

To train:

th train.lua

This will train on task 16 with Linear Start over 100 epochs by default. See train.lua (or the paper) for hyperparameters and more training options. To train on gpu, use -cuda.

memn2n's Issues

nn.ZeroGrad has been already assigned a parent class

mldl@mldlUB1604:~/ub16_prj/MemN2N-harvard$ th train.lua
/home/mldl/torch/install/bin/luajit: /home/mldl/torch/install/share/lua/5.1/trepl/init.lua:389: /home/mldl/torch/install/share/lua/5.1/trepl/init.lua:389: /home/mldl/torch/install/share/lua/5.1/trepl/init.lua:389: /home/mldl/torch/install/share/lua/5.1/trepl/init.lua:389: /home/mldl/torch/install/share/lua/5.1/torch/init.lua:102: class nn.ZeroGrad has been already assigned a parent class

stack traceback:
[C]: in function 'error'
/home/mldl/torch/install/share/lua/5.1/trepl/init.lua:389: in function 'require'
train.lua:3: in main chunk
[C]: in function 'dofile'
...mldl/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
[C]: at 0x00405d50
mldl@mldlUB1604:~/ub16_prj/MemN2N-harvard$