otx.algo.callbacks.adaptive_train_scheduling#

Callback to reschedule the validation interval adaptively.

Classes

AdaptiveTrainScheduling([max_interval, ...])

Adaptive Training Scheduling Hook.
class otx.algo.callbacks.adaptive_train_scheduling.AdaptiveTrainScheduling(max_interval: int = 5, decay: float = -0.025, min_earlystop_patience: int = 3, min_lrschedule_patience: int = 2)[source]#

Bases: Callback

Adaptive Training Scheduling Hook.

Depending on the size of iteration per epoch, adaptively update the validation interval and related values.

Parameters:

  • max_interval – Maximum value of validation interval. Defaults to 5.

  • decay – Parameter to control the interval. This value is set by manual manner. Defaults to -0.025.

on_train_end(trainer: Trainer, pl_module: LightningModule) → None[source]#

Execute this function at terminating the train stage.

on_train_start(trainer: Trainer, pl_module: LightningModule) → None[source]#

Execute this function at starting the train stage.