RLIE_A3C

Code for the paper 'Speeding up Reinforcement Learning-based Information Extraction Training using Asynchronous Methods' to be presented at EMNLP 2017.

You will need to install TensorFlow.

Data Preparation:

The dev dataset has been used for training the hyperparameters. The test dataset has been used for testing.

• Change to the code directory: cd code

Create the vectorizers using a pre-trained model:

python vec_consolidate.py dloads/Shooter/train.extra 5 trained_model2.p consolidated/vec_train.5.p
python vec_consolidate.py dloads/Shooter/test.extra 5 trained_model2.p consolidated/vec_test.5.p

Consolidate the articles:

python consolidate.py dloads/Shooter/train.extra 5 trained_model2.p consolidated/train+context.5.p consolidated/vec_train.5.p
python consolidate.py dloads/Shooter/test.extra 5 trained_model2.p consolidated/test+context.5.p consolidated/vec_test.5.p

Running the code:

Run the server:

cd code
mkdir consolidated
mkdir outputs
python server_multiprocessing.py --trainEntities consolidated/train+context.5.p --testEntities consolidated/test+context.5.p --outFile outputs/run.out --modelFile trained_model2.p --entity 4 --aggregate always --shooterLenientEval True --delayedReward False --contextType 2

Run the agent in a separate terminal/tab:

cd code/a3c
mkdir saved_network
python a3c.py

Cite

@inproceedings{sharma2017speeding,
  title={Speeding up Reinforcement Learning-based Information Extraction Training using Asynchronous Methods},
  author={Sharma, Aditya and Parekh, Zarana and Talukdar, Partha},
  booktitle={Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing},
  pages={2648--2653},
  year={2017}
}

Acknowledgements

• Karthik's DeepRL-InformationExtraction Codebase
• Kosuke Miyoshi's A3C implementation