Source code for [Posterior regularized REINFORCE for Instance Selection in Distant Supervision] (https://arxiv.org/abs/1904.08051)

--------------------------------------------------------------------------------

Environment

Python 2.7 Tensorflow 1.5.0

Preparing for Training

include NYT dataset, rule pattern txt in the ./origin_data folder.

change current dir into ./origin_data , generate the data for transE prediction. Generated by:

python generate_transE.py

Then run the trained TransX model to generate the entity vector file ./data/entity2vec.bern

change current dir into ./initial_data , generate the data for training and testing. Generated by:

python initial_BGRU_traintest.py
python initial.py
python initial_train_robot.py

A demo for train and test is provided and can be execurated by:

python launcher.py