def get_callbacks(self, stage: str) -> "OrderedDict[str, Callback]":
"""Prepares the callbacks for selected stage.
Args:
stage: stage name
Returns:
dictionary with stage callbacks
"""
callbacks = super().get_callbacks(stage=stage)
is_callback_exists = lambda callback_fn: any(
callback_isinstance(x, callback_fn) for x in callbacks.values()
)
if isinstance(self._criterion, Criterion) and not is_callback_exists(ICriterionCallback):
callbacks["_criterion"] = CriterionCallback(
input_key=self._output_key, target_key=self._target_key, metric_key=self._loss_key,
)
if isinstance(self._optimizer, Optimizer) and not is_callback_exists(IOptimizerCallback):
callbacks["_optimizer"] = OptimizerCallback(metric_key=self._loss_key)
if isinstance(self._scheduler, (Scheduler, ReduceLROnPlateau)) and not is_callback_exists(
ISchedulerCallback
):
callbacks["_scheduler"] = SchedulerCallback(
loader_key=self._valid_loader, metric_key=self._valid_metric
)
return callbacks
def get_callbacks(self) -> "OrderedDict[str, Callback]":
"""Returns the callbacks for the experiment."""
callbacks = sort_callbacks_by_order(super().get_callbacks())
callback_exists = lambda callback_fn: any(
callback_isinstance(x, callback_fn) for x in callbacks.values())
if isinstance(
self._criterion,
TorchCriterion) and not callback_exists(ICriterionCallback):
callbacks["_criterion"] = CriterionCallback(
input_key=self._output_key,
target_key=self._target_key,
metric_key=self._loss_key,
)
if isinstance(
self._optimizer,
TorchOptimizer) and not callback_exists(IBackwardCallback):
callbacks["_backward"] = BackwardCallback(
metric_key=self._loss_key)
if isinstance(
self._optimizer,
TorchOptimizer) and not callback_exists(IOptimizerCallback):
callbacks["_optimizer"] = OptimizerCallback(
metric_key=self._loss_key)
if isinstance(
self._scheduler,
TorchScheduler) and not callback_exists(ISchedulerCallback):
callbacks["_scheduler"] = SchedulerCallback(
loader_key=self._valid_loader, metric_key=self._valid_metric)
return callbacks
def get_callbacks(self) -> "OrderedDict[str, Callback]":
"""Prepares the callbacks for selected stage.
Args:
stage: stage name
Returns:
dictionary with stage callbacks
"""
callbacks = super().get_callbacks()
callback_exists = lambda callback_fn: any(
callback_isinstance(x, callback_fn) for x in callbacks.values())
if isinstance(
self._criterion,
TorchCriterion) and not callback_exists(ICriterionCallback):
callbacks["_criterion"] = CriterionCallback(
input_key=f"{self.loss_mode_prefix}_left",
target_key=f"{self.loss_mode_prefix}_right",
metric_key=self._loss_key,
)
if isinstance(
self._optimizer,
TorchOptimizer) and not callback_exists(IBackwardCallback):
callbacks["_backward"] = BackwardCallback(
metric_key=self._loss_key)
if isinstance(
self._optimizer,
TorchOptimizer) and not callback_exists(IOptimizerCallback):
callbacks["_optimizer"] = OptimizerCallback(
metric_key=self._loss_key)
if isinstance(
self._scheduler,
TorchScheduler) and not callback_exists(ISchedulerCallback):
callbacks["_scheduler"] = SchedulerCallback(
loader_key=self._valid_loader, metric_key=self._valid_metric)
return callbacks