Schematic illustration of AbGraftBERT
- pytorch==1.12.0
- fairseq==0.10.2
- numpy==1.23.3
All preprocessed data is from ABGNN, with the code primarily based on this repository, thanks! For training, we can run:
bash pretrain-abgraftbert.sh
For experiment 1, raw data is from MEAN.
For experiment 2, raw data is from HSRN.
For experiment 3, raw data is from RefineGNN.
The finetuning scripts are following:
# for exp1
bash finetune-exp1.sh
# for exp2
bash finetune-exp2.sh
# for exp3
# have to additionally install pytorch_lightning, matplotlib, and igfold
bash finetune-exp3.sh
bash covid-optimize.sh
We can simply run the following code for inference:
python inference.py \
--cdr_type ${CDR} \
--cktpath ${model_ckpt_path}}/checkpoint_best.pt \
--data_path ${dataset_path}