class ESMockStrategy:
"""An empty strategy to test early stopping."""
def __init__(self, p, metric_vals):
self.p = p
self.clock = Clock()
self.evaluator = MockEvaluator(self.clock, metric_vals)
self.model = SimpleMLP()
def before_training_iteration(self):
self.p.before_training_iteration(self)
self.clock.before_training_iteration(self)
def before_training_epoch(self):
self.p.before_training_epoch(self)
self.clock.before_training_epoch(self)
def after_training_iteration(self):
self.p.after_training_iteration(self)
self.clock.after_training_iteration(self)
def after_training_epoch(self):
self.p.after_training_epoch(self)
self.clock.after_training_epoch(self)
def stop_training(self):
raise StopIteration()
def __init__(
self,
model: Module,
optimizer,
criterion=CrossEntropyLoss(),
train_mb_size: int = 1,
train_epochs: int = 1,
eval_mb_size: int = 1,
device="cpu",
plugins=None,
evaluator=None,
eval_every=-1,
peval_mode="epoch",
):
self._criterion = criterion
self.model: Module = model
self.optimizer = optimizer
self.train_epochs: int = train_epochs
self.train_mb_size: int = train_mb_size
self.eval_mb_size: int = (
train_mb_size if eval_mb_size is None else eval_mb_size
)
self.device = device
self.plugins = [] if plugins is None else plugins
if evaluator is None:
evaluator = EvaluationPlugin()
self.plugins.append(evaluator)
self.evaluator = evaluator
""" EvaluationPlugin used for logging and metric computations. """
# Configure periodic evaluation.
assert peval_mode in {"epoch", "iteration"}
self.eval_every = eval_every
peval = PeriodicEval(eval_every, peval_mode)
self.plugins.append(peval)
self.clock = Clock()
""" Incremental counters for strategy events. """
# WARNING: Clock needs to be the last plugin, otherwise
# counters will be wrong for plugins called after it.
self.plugins.append(self.clock)
###################################################################
# State variables. These are updated during the train/eval loops. #
###################################################################
self.experience = None
self.adapted_dataset = None
self.dataloader = None
self.mbatch = None
self.mb_output = None
self.loss = None
self.is_training: bool = False
self.current_eval_stream = None
self._stop_training = False
def __init__(
self,
model: Module,
optimizer: Optimizer,
train_mb_size: int = 1,
train_epochs: int = 1,
eval_mb_size: Optional[int] = 1,
device="cpu",
plugins: Optional[List["SupervisedPlugin"]] = None,
evaluator: EvaluationPlugin = default_evaluator,
eval_every=-1,
peval_mode="epoch",
):
"""Init.
:param model: PyTorch model.
:param optimizer: PyTorch optimizer.
:param train_mb_size: mini-batch size for training.
:param train_epochs: number of training epochs.
:param eval_mb_size: mini-batch size for eval.
:param evaluator: (optional) instance of EvaluationPlugin for logging
and metric computations. None to remove logging.
:param eval_every: the frequency of the calls to `eval` inside the
training loop. -1 disables the evaluation. 0 means `eval` is called
only at the end of the learning experience. Values >0 mean that
`eval` is called every `eval_every` epochs and at the end of the
learning experience.
:param peval_mode: one of {'epoch', 'iteration'}. Decides whether the
periodic evaluation during training should execute every
`eval_every` epochs or iterations (Default='epoch').
"""
super().__init__(model=model, device=device, plugins=plugins)
self.optimizer: Optimizer = optimizer
""" PyTorch optimizer. """
self.train_epochs: int = train_epochs
""" Number of training epochs. """
self.train_mb_size: int = train_mb_size
""" Training mini-batch size. """
self.eval_mb_size: int = (train_mb_size
if eval_mb_size is None else eval_mb_size)
""" Eval mini-batch size. """
if evaluator is None:
evaluator = EvaluationPlugin()
self.plugins.append(evaluator)
self.evaluator = evaluator
""" EvaluationPlugin used for logging and metric computations. """
# Configure periodic evaluation.
assert peval_mode in {"epoch", "iteration"}
self.eval_every = eval_every
peval = PeriodicEval(eval_every, peval_mode)
self.plugins.append(peval)
self.clock = Clock()
""" Incremental counters for strategy events. """
# WARNING: Clock needs to be the last plugin, otherwise
# counters will be wrong for plugins called after it.
self.plugins.append(self.clock)
###################################################################
# State variables. These are updated during the train/eval loops. #
###################################################################
self.dataloader = None
""" Dataloader. """
self.mbatch = None
""" Current mini-batch. """
self.mb_output = None
""" Model's output computed on the current mini-batch. """
self.loss = None
""" Loss of the current mini-batch. """
self._stop_training = False
def __init__(self, p, metric_vals):
self.p = p
self.clock = Clock()
self.evaluator = MockEvaluator(self.clock, metric_vals)
self.model = SimpleMLP()
def __init__(
self,
model: Module,
optimizer: Optimizer,
criterion=CrossEntropyLoss(),
train_mb_size: int = 1,
train_epochs: int = 1,
eval_mb_size: int = 1,
device="cpu",
plugins: Optional[Sequence["StrategyPlugin"]] = None,
evaluator=default_logger,
eval_every=-1,
peval_mode="epoch",
):
"""Init.
:param model: PyTorch model.
:param optimizer: PyTorch optimizer.
:param criterion: loss function.
:param train_mb_size: mini-batch size for training.
:param train_epochs: number of training epochs.
:param eval_mb_size: mini-batch size for eval.
:param device: PyTorch device where the model will be allocated.
:param plugins: (optional) list of StrategyPlugins.
:param evaluator: (optional) instance of EvaluationPlugin for logging
and metric computations. None to remove logging.
:param eval_every: the frequency of the calls to `eval` inside the
training loop. -1 disables the evaluation. 0 means `eval` is called
only at the end of the learning experience. Values >0 mean that
`eval` is called every `eval_every` epochs and at the end of the
learning experience.
:param peval_mode: one of {'epoch', 'iteration'}. Decides whether the
periodic evaluation during training should execute every
`eval_every` epochs or iterations (Default='epoch').
"""
self._criterion = criterion
self.model: Module = model
""" PyTorch model. """
self.optimizer: Optimizer = optimizer
""" PyTorch optimizer. """
self.train_epochs: int = train_epochs
""" Number of training epochs. """
self.train_mb_size: int = train_mb_size
""" Training mini-batch size. """
self.eval_mb_size: int = (train_mb_size
if eval_mb_size is None else eval_mb_size)
""" Eval mini-batch size. """
self.device = device
""" PyTorch device where the model will be allocated. """
self.plugins = [] if plugins is None else plugins
""" List of `StrategyPlugin`s. """
if evaluator is None:
evaluator = EvaluationPlugin()
self.plugins.append(evaluator)
self.evaluator = evaluator
""" EvaluationPlugin used for logging and metric computations. """
# Configure periodic evaluation.
assert peval_mode in {"epoch", "iteration"}
self.eval_every = eval_every
peval = PeriodicEval(eval_every, peval_mode)
self.plugins.append(peval)
self.clock = Clock()
""" Incremental counters for strategy events. """
# WARNING: Clock needs to be the last plugin, otherwise
# counters will be wrong for plugins called after it.
self.plugins.append(self.clock)
###################################################################
# State variables. These are updated during the train/eval loops. #
###################################################################
self.experience = None
""" Current experience. """
self.adapted_dataset = None
""" Data used to train. It may be modified by plugins. Plugins can
append data to it (e.g. for replay).
.. note::
This dataset may contain samples from different experiences. If you
want the original data for the current experience
use :attr:`.BaseStrategy.experience`.
"""
self.dataloader = None
""" Dataloader. """
self.mbatch = None
""" Current mini-batch. """
self.mb_output = None
""" Model's output computed on the current mini-batch. """
self.loss = None
""" Loss of the current mini-batch. """
self.is_training: bool = False
""" True if the strategy is in training mode. """
self.current_eval_stream = None
""" Current evaluation stream. """
self._stop_training = False
self._warn_for_disabled_plugins_callbacks()
self._warn_for_disabled_metrics_callbacks()