The aim of this repository is to slightly modify the implementation of Dynamic memory networks implemented by YerevaNN and as described in the paper by Kumar et al. to better apply this methodology to document novelty detection. This also only contains a modification of the dmn basic algoithm but this modification can be replicated for the other methodologies implemented in the Dynamic-memory-networks-in-Theano repository.
| file | description |
|---|---|
main.py |
the main entry point to train and test available network architectures on document novelty tasks |
dmn_basic.py |
our baseline implementation. It is as close to the original in the paper, except the number of steps in the main memory GRU is fixed. Attention module uses T.abs_ function as a distance between two vectors which causes gradients to become NaN randomly. |
utils.py |
tools for working with data tasks and GloVe vectors |
nn_utils.py |
helper functions on top of Theano and Lasagne |
fetch_glove_data.sh |
shell script to fetch GloVe vectors (by 5vision) |
This implementation is based on Theano and Lasagne. One way to install them is:
pip install -r https://raw.githubusercontent.com/Lasagne/Lasagne/master/requirements.txt
pip install https://github.com/Lasagne/Lasagne/archive/master.zip
The following bash script will download GloVe vectors.
./fetch_glove_data.sh
Use main.py to train a network:
python main.py --network dmn_basic --input_train FOLDER1 --input_test FOLDER2
The states of the network will be saved in states/ folder.