base

class OnStepCallbackWithTiming(train_timing: CallbackTiming = None, val_timing: CallbackTiming = None, name: str = 'auto')

Base class for timed callbacks, which execute at certain intervals.

They execute on step, during training and validation, at independently specified timings.

__init__(train_timing: CallbackTiming = None, val_timing: CallbackTiming = None, name: str = 'auto') → None

Initializes the callback. The name is used for logging.

Parameters:

• train_timing – The CallbackTiming object specifying how often the callback is to be called during training.

• val_timing – The CallbackTiming object specifying how often the callback is to be called during validation.

• name – The name of the callback. If "auto", the class name is used, minus “Log” if the class name starts with that.

execute(trainer: Trainer, pl_module: MLDFTLitModule, outputs: Tensor | Mapping[str, Any] | None, batch: Any, split: str) → None

Executes the callback, e.g. logs a figure to tensorboard.

Parameters:

• pl_module – The lightning module.

• trainer – The lightning trainer.

• outputs – The outputs of the lightning module.

• batch – The batch data.

• split – The split, either "train" or "val".

on_train_batch_end(trainer: Trainer, pl_module: MLDFTLitModule, outputs: Tensor | Mapping[str, Any] | None, batch: Any, batch_idx: int) → None

Executes the callback via execute(), if the timing matches.

on_validation_batch_end(trainer: Trainer, pl_module: MLDFTLitModule, outputs: Tensor | Mapping[str, Any] | None, batch: Any, batch_idx: int, **kwargs) → None

Executes the callback via execute(), if the timing matches.

Can happen only for the first validation batch.

default_log_timing()

Returns a default log timing, which logs with exponentially increasing intervals.