BERTCRF

Environment Setup

• Install python 3.9
• Cuda 11.6

transform_data.py

First step is to run transform_data.py with a parameter from the following list [train, dev, test]. This file transforms data from given dataset to a format that will be readable from our model. In order to run this file you have to run the following commands:

• pip install mendelai-brat-parser
• pip install smart_open
• pip install nltk

BERT_CRF_model.py

This is the main code that train of model happens. In order to run this file you have to run the following commands:

• pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu116
• pip install transformers
• pip install absl-py
• pip install six
• pip install protobuf==3.19.0
• pip install wrapt
• pip install opt_einsum
• pip install gast
• pip install astunparse
• pip install termcolor
• pip install flatbuffers
• pip install scikit-learn
• pip install sentence-splitter

evaluation.py

This file may run standalone if a model is already saved (system_best_epoch.pth.tar file exists). In order to achieve this you have to comment line 10:

def evaluate(model, class_dict, inv_class_dict, test_batches, use_cuda, gpu_device):

and uncomment lines 13-15:

# if __name__ == '__main__':
#     from initialize import *
#     model, class_dict, inv_class_dict, test_batches, use_cuda, gpu_device = my_initialize()