train

The protml app for training models to map protein sequences to their phenotype and generative models for generate sequences with high functional scores

Examples

Train a supervised model:

python3 -m protml.apps.train experiment=supervised/train_base train_data= < PATH_TO_TRAINING_DATA > val_data= < PATH_TO_VALIDATION_DATA >

Overide model parameters from the command line:

python3 -m protml.apps.train experiment=supervised/train_base train_data= < PATH_TO_TRAINING_DATA > val_data= < PATH_TO_VALIDATION_DATA > trainer.max_epochs=50000 model.encoder.model_params.hidden_layer_sizes=[100,100,100,100,100] z_dim=10

Train a generative model:

python3 -m protml.apps.train experiment=vae/train_base train_data=< PATH_TO_TRAINING_DATA > val_data= <PATH_TO_VALIDATION_DATA > trainer.max_epochs=1000

Specify parameters that are not set in the config file with +PARAMETER=VALUE:

python3 -m protml.apps.train experiment=vae/train_base train_data=< PATH_TO_TRAINING_DATA > val_data= <PATH_TO_VALIDATION_DATA > trainer.max_epochs=1000 +datamodule.params.use_weights=True

log_hyperparameters

source

.log_hyperparameters(
   config: DictConfig, logger: Any
)

---

load_data

source

.load_data(
   config: DictConfig
)

---

train

source

.train(
   config: DictConfig, datamodule: LightningDataModule
)

---

do_train

source

.do_train(
   config
)