Optimiser

These configs (generally) configure PyTorch Optimizer objects.

Note

To use an alternative optimiser config, use a command like:

workshop train optimiser=<OPTIMISER_NAME> encoder=gvp dataset=cath task=inverse_folding trainer=cpu
# or
python proteinworkshop/train.py optimiser=<OPTIMISER_NAME> encoder=gvp dataset=cath task=inverse_folding trainer=cpu # or trainer=gpu

where <OPTIMISER_NAME> is the name of the optimiser config.

Note

To change the learning rate, use a command like:

workshop train optimizer.lr=0.0001 encoder=gvp dataset=cath task=inverse_folding trainer=cpu
# or
python proteinworkshop/train.py optimizer.lr=0.0001 encoder=gvp dataset=cath task=inverse_folding trainer=cpu # or trainer=gpu

where 0.0001 is the new learning rate.

ADAM (adam)

# Example usage:
python proteinworkshop/train.py ... optimiser=adam optimiser.optimizer.lr=0.0001 ...

optimizer:
  _target_: torch.optim.Adam
  _partial_: true
  lr: 0.001
  weight_decay: 0.0

ADAM-W (adamw)

# Example usage:
python proteinworkshop/train.py ... optimiser=adamw optimiser.optimizer.lr=0.0001 ...

optimizer:
  _target_: torch.optim.AdamW
  _partial_: true
  lr: 0.001
  weight_decay: 0.0

Lion (lion)

# Example usage:
python proteinworkshop/train.py ... optimiser=lion optimiser.optimizer.lr=0.0001 ...

optimizer:
  _target_: lion_pytorch.Lion
  _partial_: true
  lr: 0.0001
  weight_decay: 0.0