callbacks

Callbacks for training.

Callback

Represents a callback for the explainable model.

EarlyStopping

Initialize the EarlyStopping callback.

Parameters:

Name	Type	Description	Default
monitoring_metric		The metric name to use for early stopping.	required
reset_after_pruning		If True, reset the best metrics after pruning. Else keep the best model regardless of the pruning.	required
min_delta		Minimum delta between two monitoring values to consider an improvement.	required
patience		Number of epochs to wait with no improvement of the monitoring value.	required
mode		Whether to "maximize" or "minimize" the provided metric.	required
num_pruning_criteria		Minimal number of pruning steps performed required.	required
num_leaf_criteria		Minimal number of remaining leaves required.	required

monitoring_metric: str

reset_after_pruning: bool = True

patience: int = 3

min_delta: float = 0.0

mode: str = 'min'

num_pruning_criteria: int | None = None

num_leaf_criteria: int | None = None

LrScheduler

Initialize a scheduler object to encapsulate different torch schedulers.

Parameters:

Name	Type	Description	Default
pre_scheduler		Based torch lr scheduler or no scheduler to be instantiated.	required
step_method		"epoch" or "batch".	required
monitoring_metric		Optional monitoring metric required for the step method, default None.	required

pre_scheduler: partial[LRScheduler | ReduceLROnPlateau]

step_method: str | None = None

monitoring_metric: str | None = None

ModelCheckpoint

Model checkpoint initialization.

Parameters:

Name	Type	Description	Default
mode		Whether to "maximize" or "minimize" the provided metric.	required
monitoring_metric		Monitoring metric required for the step method, default None.	required
reset_after_pruning		If True, reset the best metrics after pruning. Else keep the best model regardless of the pruning.	required
save_every_epoch		How often to save the model. If None, only save the best checkpoint.	required

mode: str | None

monitoring_metric: str

reset_after_pruning: bool

save_every_epoch: int | None = 1