End-To-End Memory Networks in MemN2N

MXNet implementation of End-To-End Memory Networks for language modelling. The original Tensorflow code from carpedm20 can be found here.

Known issue: SGD does not converge, ADAM converges but is not able to reach a good result(details).

Setup

This code requires MXNet. Also, it uses CUDA to run on GPU for faster training. There is a set of sample Penn Tree Bank (PTB) corpus in data directory, which is a popular benchmark for measuring quality of these models. But you can use your own text data set which should be formated like this.

Usage

To train a model with 6 hops and memory size of 100, run the following command:

$ python train.py --nhop 6 --mem_size 100

To see all training options, run:

$ python train.py --help

To test a model, run the script file test.py like:

$ python test.py --network checkpoint/memnn-symbol.json --params checkpoint/memnn-0100.params --gpus 0