pytorch_pfn_extras.engine.Trainer

class pytorch_pfn_extras.engine.Trainer(handler, *, evaluator, models, profile=None, **kwargs)

Bases: object

Methods

__init__(handler, *, evaluator, models[, ...])
extend(extension[, name, trigger, priority, ...])
get_optimizer(name)
is_epoch_last_iter(idx)
load_state_dict(to_load)
run(train_loader[, val_loader, train_len, ...])	Executes the training loop.
set_optimizer(name, optimizer)
state_dict()

Attributes

epoch
epoch_detail
evaluator
is_before_training
iteration
manager
models
optimizers
stop_trigger

Parameters:

• handler (handler_module.BaseHandler) –

• evaluator (Optional[Union[Evaluator, Tuple[Evaluator, Optional[Union[Trigger, Callable[[ExtensionsManagerProtocol], bool], Tuple[float, str]]]], Mapping[str, Union[Evaluator, Tuple[Evaluator, Optional[Union[Trigger, Callable[[ExtensionsManagerProtocol], bool], Tuple[float, str]]]]]]]]) –

• models (Union[Module, Mapping[str, Module]]) –

• profile (Optional[profile]) –

• kwargs (Any) –

__init__(handler, *, evaluator, models, profile=None, **kwargs)

Parameters:

• handler (handler_module.BaseHandler) –

• evaluator (Optional[Union[Evaluator, Tuple[Evaluator, Optional[Union[Trigger, Callable[[ExtensionsManagerProtocol], bool], Tuple[float, str]]]], Mapping[str, Union[Evaluator, Tuple[Evaluator, Optional[Union[Trigger, Callable[[ExtensionsManagerProtocol], bool], Tuple[float, str]]]]]]]]) –

• models (Union[Module, Mapping[str, Module]]) –

• profile (Optional[profile]) –

• kwargs (Any) –

property epoch: int

property epoch_detail: float

property evaluator: Optional[Evaluator]

extend(extension, name=None, trigger=None, priority=None, *, call_before_training=False, **kwargs)

Parameters:

• extension (Union[extension.ExtensionLike, ExtensionEntry]) –

• name (Optional[str]) –

• trigger (TriggerLike) –

• priority (Optional[int]) –

• call_before_training (bool) –

• kwargs (Any) –

Return type:

None

get_optimizer(name)

Parameters:

name (str) –

Return type:

Optimizer

property is_before_training: bool

is_epoch_last_iter(idx)

Parameters:

idx (int) –

Return type:

bool

property iteration: int

load_state_dict(to_load)

Parameters:

to_load (Dict[str, Any]) –

Return type:

None

property manager: ExtensionsManager

property models: Mapping[str, Module]

property optimizers: Mapping[str, Optimizer]

run(train_loader, val_loader=None, *, train_len=None, eval_len=None)

Executes the training loop.

Parameters:

• train_loader (torch.utils.data.DataLoader) – A data loader for training.

• val_loader (torch.utils.data.DataLoader, optional) – A data loader passed to Evaluator.run().

• train_len (int, optional) – The number of iterations per one training epoch. The default value is inferred from the size of training data loader.

• eval_len (int, optional) – The number of iterations per one evaluation epoch, passed to Evaluator.run()

Return type:

None

See also

• pytorch_pfn_extras.training._evaluator.Evaluator()

set_optimizer(name, optimizer)

Parameters:

• name (str) –

• optimizer (Optimizer) –

Return type:

None

state_dict()

Return type:

Dict[str, Any]

property stop_trigger: Trigger