hierarchical-protein-classification

Training

Before training, make sure your .env file is setup and proteinworkshop exists. For training to work, line 400 of proteinworkshop.models.base.py needs to be changed to self.log_dict(log_dict, prog_bar=True, batch_size=batch.batch_size).

Training is run through the command `python3 proteinvirtual/train.py encoder=<encoder_name> task=multiple_graph_classification dataset=fold_fold dataset.datamodule.num_workers=8 features=<feature_set> trainer=gpu scheduler=plateau trainer.max_epochs=150 optimiser.optimizer.lr=0.001 decoder.graph_label.dropout=0.5 ++test=True logger=wandb name=<run_name>

• The encoder name can be any config file in both proteinvirtual/config and proteinworkshop/config, but make sure that any new configs created have distinct names to prevent overwriting.
• The featuriser name should generally match with the encoder (i.e., schnet with ca_bb and virtual_schnet_hierarchy with geo_hetero_hierarchy)
• If using wandb as a logger, make sure your wandb information is setup in your .env file

Testing

pytest is used for running tests, which can be added using pip. To run tests, use the command python3 -m pytest tests/.

For the tests to work, make sure to create your own .env file from the included .env.example and link the path to your proteinworkshop folder for Hydra configs.