mofaflex.TrainingOptions

class mofaflex.TrainingOptions(*, device='cuda', batch_size=0, max_epochs=10000, n_particles=1, lr=0.001, early_stopper_patience=100, save_path=None, mofa_compat=False, seed=None, num_workers=0, pin_memory=False)

Options for training.

Attributes table

batch_size	Batch size.
device	Device to run training on.
early_stopper_patience	Number of steps without relevant improvement to stop training.
lr	Learning rate.
max_epochs	Maximum number of training epochs.
mofa_compat	Save model in MOFA2 compatible format.
n_particles	Number of particles for ELBO estimation.
num_workers	Number of data loader workers.
pin_memory	Whether to use pinned memory in the data loader.
save_path	Path to save model.
seed	Seed for the pseudorandom number generator.

Methods table

Attributes

TrainingOptions.batch_size: int = 0

Batch size.

TrainingOptions.device: str | device = 'cuda'

Device to run training on.

TrainingOptions.early_stopper_patience: int = 100

Number of steps without relevant improvement to stop training.

TrainingOptions.lr: float = 0.001

Learning rate.

TrainingOptions.max_epochs: int = 10000

Maximum number of training epochs.

TrainingOptions.mofa_compat: Union[Literal['full', 'modelonly'], bool] = False

Save model in MOFA2 compatible format. If True or "full", will include the data in the file. This can result in very large files. "modelonly" will save only the trained model.

TrainingOptions.n_particles: int = 1

Number of particles for ELBO estimation.

TrainingOptions.num_workers: int = 0

Number of data loader workers.

TrainingOptions.pin_memory: bool = False

Whether to use pinned memory in the data loader.

TrainingOptions.save_path: Path | str | None = None

Path to save model.

TrainingOptions.seed: int | None = None

Seed for the pseudorandom number generator.

Methods