TorchLiter
module torchliter.engine.base
Global Variables
- REPR_INDENT
- COMPONENTS
- map_str_to_types
- map_types_to_str
class EngineBase
Base class of Engine classes.
Attributes ———- epoch : int Current epoch iteration : int Current interation fractional_epoch: float fractional_epoch = epoch + iteration/epoch_length fractional_iteration : float fractional_iteration = iteration/epoch_length epoch_length : Optional[int] Total number of iterations in an epoch absolute_iterations : int absolute_iterations = epoch_length * epoch + iteration _registry : Tuple[Dict[str, Any]] Registry of engine components.
method EngineBase.__init__
__init__()
property EngineBase.absolute_iterations
property EngineBase.fractional_epoch
property EngineBase.fractional_iteration
property EngineBase.is_eval_stub
property EngineBase.is_lambda_stub
property EngineBase.is_train_stub
property EngineBase.training
method EngineBase.after_iteration
after_iteration(**kwargs)
method EngineBase.before_iteration
before_iteration(**kwargs)
method EngineBase.eval
eval()
method EngineBase.execute
execute(**kwargs: Any) → None
Executes stubs in queue.
Parameters ———- **kwargs : Any
Returns ——- None
Raises —— AttributeError No action attached to current stub.
method EngineBase.load_state_dict
load_state_dict(state_dict: Dict[str, Dict[str, Any]]) → None
Load state into engine.
Parameters ———- state_dict : Dict[str, Dict[str, Any]] Engine state dict.
Returns ——- None
method EngineBase.per_batch
per_batch(batch: Union[Tuple[Any], Dict[str, Any]], **kwargs: Any)
method EngineBase.per_epoch
per_epoch(**kwargs)
Train, eval model or performe a lambda op by one epoch.
The stub must have dataloader.
method EngineBase.queue
queue(stubs: List[torchliter.stub.StubBase]) → None
Adds stubs to queue.
Parameters
———- stubs : List[StubBase] A list of stubs.
Returns ——- None
method EngineBase.reset_engine
reset_engine() → None
Reset engine state.
epoch -> 0 iteration -> 0 stubs -> []
method EngineBase.reset_queue
reset_queue() → None
Resets stubs queue to empty.
method EngineBase.state_dict
state_dict() → Dict[str, Dict[str, Any]]
Generate current state of the engine as Dict.
Returns ——- Dict[str, Dict[str, Any]] Dict of component state dicts.
method EngineBase.train
train()
method EngineBase.when_epoch_finishes
when_epoch_finishes(**kwargs)
method EngineBase.when_epoch_starts
when_epoch_starts(**kwargs)