ORMA: Optimal TRansport-based Multi-grained Alignments

Training Models

To train the model, use the following command:

bash train.sh ${CUDA_DEVICE}

File Structure

The project consists of the following files:

• data/: We use the ChEBI-20 dataset from text2mol for the main experiments. For training, val and test sets, we discard invalid molecules without any chemical bonds. Additionally, we add CanonicalSMILES and molecule names from PubChem for these three sets.
  • graph_data/: unzip mol_graphs.zip from text2mol
  • token_embedding_dict.npy: from text2mol
  • training.csv: processed by preprocess.py based on training.txt from text2mol
  • val.csv: processed by preprocess.py based on val.txt from text2mol
  • test.csv: processed by preprocess.py based on test.txt from text2mol
  • preprocess.py: run python3 preprocess.py
• allenai_scibert_scivocab_uncased/: SciBERT path.
• config.json
• train.sh
• main.py
• modeling.py
• dataloader.py
• chemutils.py
• utils.py
• requirements.txt