Code accompanying End-to-End Information Extraction without Token-Level Supervision
Code for the paper End-to-End Information Extraction without Token-Level Supervision
The results presented in the paper can be found in emnlp-results
pip install -r requirements.txt
python train.py <atis|movie|restaurant>
will train a model. The best model (evaluated on the val set) will be saved as best.npz in the work dirpython test.py <atis|movie|restaurant>
will load best.npz and evaluate the model on the test setbaselines/bilstm/runner.py
python baselines/bilstm/runner.py
trains a model, evaluate on val set, and for every improvement evaluate on test set.python evaluate.py actual1.json actual2.json expected.json
will run bootstrap sampling to estimate p-values for actual1.json vs. actual2.json