ehrapy.ml.Trainer

class ehrapy.ml.Trainer(*, max_epochs=100, batch_size=128, learning_rate=0.001, weight_decay=0.0001, patience=10, class_weight=None, device='auto')

How deep learning models are trained.

Training stops once the loss on the tuning set has not improved for patience epochs, and the parameters of the epoch with the lowest loss are kept.

Examples

>>> import ehrapy as ep
>>> model = ep.ml.GRU(trainer=ep.ml.Trainer(max_epochs=20, device="cpu"))

Attributes table

batch_size	Number of observations per gradient step.
class_weight	"balanced" to weigh classes and labels inversely to their frequency in the training set.
device	Device to train and predict on, such as "cpu" or "cuda", or "auto" for a GPU if one is available.
learning_rate	Learning rate of the AdamW optimizer.
max_epochs	Maximum number of passes over the training set.
patience	Number of epochs without improvement on the tuning set after which training stops.
weight_decay	Weight decay of the AdamW optimizer.

Methods table

Attributes

Trainer.batch_size: int = 128

Number of observations per gradient step.

Trainer.class_weight: Literal['balanced'] | None = None

“balanced” to weigh classes and labels inversely to their frequency in the training set.

Trainer.device: str = 'auto'

Device to train and predict on, such as “cpu” or “cuda”, or “auto” for a GPU if one is available.

Trainer.learning_rate: float = 0.001

Learning rate of the AdamW optimizer.

Trainer.max_epochs: int = 100

Maximum number of passes over the training set.

Trainer.patience: int = 10

Number of epochs without improvement on the tuning set after which training stops.

Trainer.weight_decay: float = 0.0001

Weight decay of the AdamW optimizer.

Methods