callbacks

Callbacks for training.

Classes:

Name	Description
EarlyStopping	Initialize the EarlyStopping callback.
Scheduler	Initialize a scheduler object to encapsulate different torch schedulers.
ModelCheckpoint	Model checkpoint initialization.

Attributes:

Name	Type	Description
Callback

Callback = EarlyStopping | Scheduler | ModelCheckpoint

EarlyStopping

Initialize the EarlyStopping callback.

Parameters:

Name	Type	Description	Default
monitoring_metric	str	The metric name to use for early stopping, for instance, "Total loss".	required
mode	Literal['maximize', 'minimize']	Whether to "maximize" or "minimize" the provided metric.	required
patience	int	Number of epochs to wait with no improvement of the monitoring value.	3.
min_delta	float	Minimum delta between two monitoring values to consider an improvement.	0.

Attributes:

Name	Type	Description
callback	EarlyStoppingRequestBodyCallback

Attributes:

Name	Type	Description
monitoring_metric	str
callback	EarlyStoppingRequestBodyCallback
mode	Literal['maximize', 'minimize']
patience	int
min_delta	float

monitoring_metric: str

callback: EarlyStoppingRequestBodyCallback = field(default=EarlyStoppingRequestBodyCallback.EARLY_STOPPING, init=False)

mode: Literal['maximize', 'minimize'] = field(validator=lambda self, attribute, value: value in [str(mode_) for mode_ in EarlyStoppingRequestBodyMode])

patience: int = 3

min_delta: float = 0.0

Scheduler

Initialize a scheduler object to encapsulate different torch schedulers.

Parameters:

Name	Type	Description	Default
pre_scheduler	partial[LRScheduler]	Based torch lr scheduler to be instantiated. Should not contain the optimizer as xpdeep use the trainer's optimizer for the scheduler internally.	required
step_method	Literal['batch', 'epoch']	"epoch" or "batch".	required
monitoring_metric	str	Monitoring metric required for the step method, for instance "Total loss".	required

Attributes:

Name	Type	Description
callback	SchedulerRequestBodyCallback

Attributes:

Name	Type	Description
callback	SchedulerRequestBodyCallback
pre_scheduler	partial[LRScheduler | ReduceLROnPlateau]
monitoring_metric	str
step_method	Literal['batch', 'epoch']

callback: SchedulerRequestBodyCallback = field(default=SchedulerRequestBodyCallback.SCHEDULER, init=False)

pre_scheduler: partial[LRScheduler | ReduceLROnPlateau]

monitoring_metric: str

step_method: Literal['batch', 'epoch'] = field(validator=lambda self, attribute, value: value in [str(step_method_) for step_method_ in SchedulerRequestBodyStepmethod])

ModelCheckpoint

Model checkpoint initialization.

Parameters:

Name	Type	Description	Default
monitoring_metric	str	Monitoring metric required for the step method, for instance "Total loss".	required
save_every_epoch	int	How often to save the model. If None, only save the best checkpoint.	1
mode		Whether to "maximize" or "minimize" the provided metric.	required

Attributes:

Name	Type	Description
callback	ModelCheckpointRequestBodyCallback

Attributes:

Name	Type	Description
monitoring_metric	str
callback	ModelCheckpointRequestBodyCallback
mode	Literal['maximize', 'minimize']
save_every_epoch	int

monitoring_metric: str

callback: ModelCheckpointRequestBodyCallback = field(default=ModelCheckpointRequestBodyCallback.MODEL_CHECKPOINT, init=False)

mode: Literal['maximize', 'minimize'] = field(validator=lambda self, attribute, value: value in [str(mode_) for mode_ in ModelCheckpointRequestBodyMode])

save_every_epoch: int = 1