This is "building gap resolution overnight" repository for Meanotek submission for AGRR competition
- First, install NeuThink library.
- Then clone this repository.
- Download pre-trained context embedding model from here and put in into the model3000x2x20482 folder
To Train model:
- Convert train data to conll using agrr2conll.py. You will need to change input and output file names in the script
- Run train.py (ensure that you have converted file in the same directory). The script requires you to have GPU (8Gb GPU RAM is needed).
- Wait for completion
- Run predict_gap.py for test file
- convert data back to agrr format with conll2agrr.py
please contact dtarasov@meanotek.io if you have any problems