Thanks a lot for OpenNRE.
All datasets need to be placed in the benchmark folder.
>> python train_supervised_bert.py --dataset=SciERC/10-1 --pretrain_path=dmis-lab/biobert-large-cased-v1.1--dataset: the directory of the dataset to be trained--pretrain_path: PLM, defaulting to roberta-large
- Re-sampling datasets refers to README.
- Re-weighting Loss:
--use_loss
- DA refers to README.
- Assign pseudo labels:
--labeling True - Combine pseudo-labeled data and gold-labeled data by using self-train_combine.py
- Train the student model:
--stutrain True