State-LSTM for Relation Extraction

This repo contains the PyTorch code for the State-LSTM in relation extraction task.

Difference between this repo and the code released by author

• this repo is more clean, while the author's code including many unrelated code for Machine Reading, Sequence Tagging.
• this repo is implemented in adjacency matrix manner.
• this is using PyTorch, the author's code is using Tensorflow.

The scaffold is forked from this repo.

See below for an overview of the model architecture:

Requirements

• Python 3 (tested on 3.6.6)
• PyTorch (tested on 0.4.1)
• tqdm
• unzip, wget (for downloading only)

Preparation

The code requires that you have access to the TACRED dataset (LDC license required).

First, download and unzip GloVe vectors from the Stanford NLP group website, with:

chmod +x download.sh; ./download.sh

Then prepare vocabulary and initial word vectors with:

python prepare_vocab.py dataset/tacred dataset/vocab --glove_dir dataset/glove

This will write vocabulary and word vectors as a numpy matrix into the dir dataset/vocab.

Training

To train a state LSTM neural network model, run:

bash train_sl.sh 0

Model checkpoints and logs will be saved to ./saved_models/00.

For details on the use of other parameters, such as the time_steps, please refer to train.py.

Evaluation

To run evaluation on the test set, run:

python eval.py saved_models/00 --dataset test

This will use the best_model.pt file by default. Use --model checkpoint_epoch_10.pt to specify a model checkpoint file.

Retrain

Reload a pretrained model and finetune it, run:

python train.py --load --model_file saved_models/01/best_model.pt --optim sgd --lr 0.001

Citation

@article{song2018n,
  title={N-ary relation extraction using graph state LSTM},
  author={Song, Linfeng and Zhang, Yue and Wang, Zhiguo and Gildea, Daniel},
  journal={arXiv preprint arXiv:1808.09101},
  year={2018}
}

License

All work contained in this package is licensed under the Apache License, Version 2.0. See the included LICENSE file.