Relation Extraction with Sentence level / Document level, direct supervision / distant supervision on TACRED / FewRel / CTD + PubMed
Create a new conda enviroment:
conda create -n <ENV_NAME> python=3.7
After cloning the repo, you can install all required dependencies by running the following in the root directory:
make install
Modify proper enviroment variables:
vim set_environment
Run the following command before training / testing:
source set_environment.sh
Then setup your wandb environment:
wandb init
An example of training on a slurm server:
srun --gres=gpu:1 --mem=25GB --partition=1080ti-long python src/main.py --data_path data/tacred/ --multi_class --partial_annotation --encoder_type bert-large-cased --score_func nn --learning_rate 1e-5 --max_text_length=128 --train_batch_size=2 --test_batch_size 16 --warmup 0.3 --dim=1024 --wandb