def fit(self,
train_loader,
val_loader=None,
max_epochs=1,
metrics=None,
metrics_on_train=False,
callbacks=None,
val_callbacks=None):
metrics = [] if metrics is None else metrics
assert self.train_ready()
setup_logging()
train_engine = Engine(self.train_step, model=self, logger=self.logger)
train_metrics = [Loss()] + metrics if metrics_on_train else [Loss()]
_attach_metrics(train_engine, train_metrics, name_prefix='train_')
metrics_logging.attach(train_engine, train=True)
if val_loader is not None:
self.validate(val_loader, metrics, val_callbacks)
val_engine = Engine(self.val_step, model=self, logger=self.logger)
_attach_metrics(val_engine, [Loss()] + metrics, name_prefix='val_')
_attach_callbacks(val_engine, val_callbacks)
@on_epoch_complete
def validation_epoch(train_state, val_engine, val_loader):
epoch = train_state.epoch
val_state = val_engine.run(val_loader, epoch, epoch + 1)
train_state.metrics.update(val_state.metrics)
validation_epoch.attach(train_engine, val_engine, val_loader)
metrics_logging.attach(train_engine, train=False)
_attach_callbacks(train_engine, callbacks)
train_engine.run(train_loader, 0, max_epochs)
def validate(
self,
val_loader: Iterable,
metrics: Optional[List[Union[Metric, str]]] = None,
callbacks: Optional[List[Callback]] = None) -> Dict[str, float]:
"""Perform a validation.
Args:
val_loader (Iterable): The validation data loader.
metrics (list of :class:`argus.metrics.Metric` or str, optional):
List of metrics to evaluate with the data. Defaults to `None`.
callbacks (list of :class:`argus.callbacks.Callback`, optional):
List of callbacks to be attached to the validation process.
Defaults to `None`.
Returns:
dict: The metrics dictionary.
"""
self._check_train_ready()
metrics = [] if metrics is None else metrics
phase_states = dict()
val_engine = Engine(self.val_step, phase_states=phase_states)
attach_metrics(val_engine, [Loss()] + metrics)
default_logging.attach(val_engine)
attach_callbacks(val_engine, callbacks)
state = val_engine.run(val_loader, -1, 0)
return state.metrics
def test_custom_events(self):
class CustomEvents(EventEnum):
STEP_START = 'step_start'
STEP_COMPLETE = 'step_complete'
def step_function(batch, state):
state.step_output = batch
state.engine.raise_event(CustomEvents.STEP_START)
state.step_output += 1
state.engine.raise_event(CustomEvents.STEP_COMPLETE)
class CustomCallback(Callback):
def __init__(self):
self.start_storage = []
self.end_storage = []
def step_start(self, state):
self.start_storage.append(state.step_output)
def step_complete(self, state):
self.end_storage.append(state.step_output)
data_loader = [4, 8, 15, 16, 23, 42]
callback = CustomCallback()
engine = Engine(step_function)
_attach_callbacks(engine, [callback])
engine.run(data_loader)
assert callback.start_storage == data_loader
assert callback.end_storage == [d + 1 for d in data_loader]
def validate(self,
val_loader: Optional[Iterable],
metrics: Optional[List[Metric]] = None,
callbacks: Optional[List[Callback]] = None) -> Dict[str, float]:
"""Perform a validation.
Args:
val_loader (Iterable): The validation data loader.
metrics (list of :class:`argus.metrics.Metric`, optional):
List of metrics to evaluate with the data. Defaults to `None`.
callbacks (list of :class:`argus.callbacks.Callback`, optional):
List of callbacks to be attached to the validation process.
Defaults to `None`.
Returns:
dict: The metrics dictionary.
"""
self._check_train_ready()
metrics = [] if metrics is None else metrics
val_engine = Engine(self.val_step, model=self,
logger=self.logger, phase='val')
_attach_metrics(val_engine, [Loss()] + metrics)
_attach_callbacks(val_engine, callbacks)
metrics_logging.attach(val_engine, train=False, print_epoch=False)
return val_engine.run(val_loader).metrics
def fit(self,
train_loader: Iterable,
val_loader: Optional[Iterable] = None,
num_epochs: int = 1,
metrics: Optional[List[Union[Metric, str]]] = None,
metrics_on_train: bool = False,
callbacks: Optional[List[Callback]] = None,
val_callbacks: Optional[List[Callback]] = None):
"""Train the argus model.
The method attaches metrics and callbacks to the train and validation,
and runs the training process.
Args:
train_loader (Iterable): The train data loader.
val_loader (Iterable, optional):
The validation data loader. Defaults to `None`.
num_epochs (int, optional): Number of training epochs to
run. Defaults to 1.
metrics (list of :class:`argus.metrics.Metric`, optional):
List of metrics to evaluate. By default, the metrics are
evaluated on the validation data (if any) only.
Defaults to `None`.
metrics_on_train (bool, optional): Evaluate the metrics on train
data as well. Defaults to False.
callbacks (list of :class:`argus.callbacks.Callback`, optional):
List of callbacks to be attached to the training process.
Defaults to `None`.
val_callbacks (list of :class:`argus.callbacks.Callback`, optional):
List of callbacks to be attached to the validation process.
Defaults to `None`.
"""
self._check_train_ready()
metrics = [] if metrics is None else metrics
train_engine = Engine(self.train_step, model=self,
logger=self.logger, phase='train')
train_metrics = [Loss()] + metrics if metrics_on_train else [Loss()]
_attach_metrics(train_engine, train_metrics)
metrics_logging.attach(train_engine, train=True)
if val_loader is not None:
self.validate(val_loader, metrics, val_callbacks)
val_engine = Engine(self.val_step, model=self,
logger=self.logger, phase='val')
_attach_metrics(val_engine, [Loss()] + metrics)
_attach_callbacks(val_engine, val_callbacks)
@on_epoch_complete
def validation_epoch(train_state, val_engine, val_loader):
epoch = train_state.epoch
val_state = val_engine.run(val_loader, epoch, epoch + 1)
train_state.metrics.update(val_state.metrics)
validation_epoch.attach(train_engine, val_engine, val_loader)
metrics_logging.attach(train_engine, train=False)
_attach_callbacks(train_engine, callbacks)
train_engine.run(train_loader, 0, num_epochs)
def test_custom_metric(self, engine):
metric = CustomMetric()
data_loader = [4, 8, 15, 16, 23, 42]
_attach_metrics(engine, [metric])
with pytest.raises(TypeError):
_attach_metrics(engine, [None])
state = engine.run(data_loader)
assert metric.data == data_loader
assert metric.compute() == len(data_loader)
assert state.metrics == {"custom_metric": len(data_loader)}
metric.reset()
assert metric.data == []
assert metric.compute() == 0
engine = Engine(lambda batch, state: batch, phase='train')
_attach_metrics(engine, [metric])
state = engine.run(data_loader)
assert metric.compute() == len(data_loader)
assert state.metrics == {"train_custom_metric": len(data_loader)}
@argus.callbacks.on_iteration_start
def stop_on_first_iteration(state):
state.stopped = True
stop_on_first_iteration.attach(engine)
engine.run(data_loader)
assert metric.compute() == 1
def validate(self, val_loader, metrics=None, callbacks=None):
metrics = [] if metrics is None else metrics
assert self.train_ready()
val_engine = Engine(self.val_step, model=self, logger=self.logger)
_attach_metrics(val_engine, [Loss()] + metrics, name_prefix='val_')
_attach_callbacks(val_engine, callbacks)
metrics_logging.attach(val_engine, train=False, print_epoch=False)
return val_engine.run(val_loader).metrics
def test_custom_callback_by_name(self):
data_loader = [4, 8, 15, 16, 23, 42]
engine = Engine(lambda batch, state: batch, phase='val')
_attach_metrics(engine, ["custom_metric"])
state = engine.run(data_loader)
assert state.metrics == {"val_custom_metric": len(data_loader)}
with pytest.raises(ValueError):
_attach_metrics(engine, ["qwerty"])
def test_phase_states(self, linear_argus_model_instance):
phase_states = dict()
train_engine = Engine(linear_argus_model_instance.train_step,
phase_states=phase_states)
val_engine = Engine(linear_argus_model_instance.val_step,
phase_states=phase_states)
assert train_engine.state.phase_states['train'] is train_engine.state
assert train_engine.state.phase_states['val'] is val_engine.state
assert val_engine.state.phase_states['train'] is train_engine.state
assert val_engine.state.phase_states['val'] is val_engine.state
def test_on_event(self, step_storage):
@argus.callbacks.on_event(Events.START)
def some_function(state):
state.special_secret = 42
engine = Engine(step_storage.step_method)
some_function.attach(engine)
data_loader = [4, 8, 15, 16, 23, 42]
state = engine.run(data_loader)
assert state.special_secret == 42
def attach(self, engine: Engine, *args, **kwargs):
"""Attach callback to the :class:`argus.engine.Engine`.
Args:
engine (Engine): The engine to which the callback will be attached.
*args: optional args arguments to be passed to the handler.
**kwargs: optional kwargs arguments to be passed to the handler.
"""
engine.add_event_handler(self.event, self.handler, *args, **kwargs)
def attach(self, engine: Engine):
"""Attach callback to the :class:`argus.engine.Engine`.
Args:
engine (Engine): The engine to which the callback will be attached.
"""
for event_enum in inheritors(EventEnum):
for key, event in event_enum.__members__.items():
if hasattr(self, event.value):
handler = getattr(self, event.value)
if isinstance(handler, (FunctionType, MethodType)):
engine.add_event_handler(event, handler)
else:
raise TypeError(
f"Attribute {event.value} is not callable.")
def test_attach_callback(self, n_epochs, custom_test_callback,
step_storage):
callback = custom_test_callback
engine = Engine(step_storage.step_method)
callback.attach(engine)
data_loader = [4, 8, 15, 16, 23, 42]
engine.run(data_loader, start_epoch=0, end_epoch=n_epochs)
assert callback.start_count == 1
assert callback.complete_count == 1
assert callback.epoch_start_count == n_epochs
assert callback.epoch_complete_count == n_epochs
assert callback.iteration_start_count == n_epochs * len(data_loader)
assert callback.iteration_complete_count == n_epochs * len(data_loader)
assert callback.catch_exception_count == 0
assert callback.step_start_count == n_epochs * len(data_loader)
assert callback.step_complete_count == n_epochs * len(data_loader)
assert step_storage.batch_lst == n_epochs * data_loader
assert step_storage.state is engine.state if n_epochs \
else step_storage.state is None
def test_custom_events(self, linear_net_class):
class CustomEvents(EventEnum):
STEP_START = 'step_start'
STEP_COMPLETE = 'step_complete'
class CustomEventsModel(argus.Model):
nn_module = linear_net_class
def count_step(self, batch, state):
state.step_output = batch
state.engine.raise_event(CustomEvents.STEP_START)
state.step_output += 1
state.engine.raise_event(CustomEvents.STEP_COMPLETE)
model = CustomEventsModel({
'nn_module': {
'in_features': 10,
'out_features': 1,
},
'optimizer': None,
'loss': None
})
class CustomCallback(Callback):
def __init__(self):
self.start_storage = []
self.end_storage = []
def step_start(self, state):
self.start_storage.append(state.step_output)
def step_complete(self, state):
self.end_storage.append(state.step_output)
data_loader = [4, 8, 15, 16, 23, 42]
callback = CustomCallback()
engine = Engine(model.count_step)
attach_callbacks(engine, [callback])
engine.run(data_loader)
assert callback.start_storage == data_loader
assert callback.end_storage == [d + 1 for d in data_loader]
def test_add_event_handler(self):
def some_function():
pass
engine = Engine(some_function)
assert len(engine.event_handlers[Events.START]) == 0
engine.add_event_handler(Events.START, some_function)
assert len(engine.event_handlers[Events.START]) == 1
assert engine.event_handlers[Events.START][0][0] is some_function
with pytest.raises(TypeError):
engine.add_event_handler(42, some_function)
def fit(self,
train_loader,
val_loader=None,
max_epochs=1,
metrics=None,
val_event_handlers=None,
train_event_handlers=None):
assert self.train_ready()
setup_logging()
if metrics is None:
metrics = dict()
train_engine = Engine(self._train_step)
train_loss = TrainLoss()
train_loss.attach(train_engine, 'train_loss')
train_engine.add_event_handler(Events.EPOCH_COMPLETE,
train_loss_logging)
if val_loader is not None:
val_engine = Engine(self._val_step)
if 'val_loss' not in metrics:
metrics['val_loss'] = Loss(self.loss)
for name, metric in metrics.items():
metric.attach(val_engine, name)
validation_logging(train_engine, val_engine, val_loader)
train_engine.add_event_handler(Events.EPOCH_COMPLETE,
validation_logging, val_engine,
val_loader)
train_engine.run(train_loader, max_epochs)
def engine(linear_argus_model_instance):
return Engine(lambda batch, state: batch,
model=linear_argus_model_instance,
logger=linear_argus_model_instance.logger)
def test_on_decorators(self, step_storage, linear_argus_model_instance):
@argus.callbacks.on_start
def on_start_function(state):
state.call_count = 1
state.on_start_flag = True
@argus.callbacks.on_complete
def on_complete_function(state):
state.call_count += 1
state.on_complete_flag = True
@argus.callbacks.on_epoch_start
def on_epoch_start_function(state):
state.call_count += 1
state.on_epoch_start_flag = True
@argus.callbacks.on_epoch_complete
def on_epoch_complete_function(state):
state.call_count += 1
state.on_epoch_complete_flag = True
@argus.callbacks.on_iteration_start
def on_iteration_start_function(state):
state.call_count += 1
state.on_iteration_start_flag = True
@argus.callbacks.on_iteration_complete
def on_iteration_complete_function(state):
state.call_count += 1
state.on_iteration_complete_flag = True
@argus.callbacks.on_catch_exception
def on_catch_exception_function(state):
state.call_count += 1
state.on_catch_exception_flag = True
engine = Engine(step_storage.step_method,
model=linear_argus_model_instance)
attach_callbacks(engine, [
on_start_function, on_complete_function, on_epoch_start_function,
on_epoch_complete_function, on_iteration_start_function,
on_iteration_complete_function, on_catch_exception_function
])
data_loader = [4, 8, 15, 16, 23, 42]
state = engine.run(data_loader, start_epoch=0, end_epoch=3)
assert state.call_count == len(data_loader) * 3 * 2 + 3 * 2 + 2
assert state.on_start_flag
assert state.on_complete_flag
assert state.on_epoch_start_flag
assert state.on_epoch_complete_flag
assert state.on_iteration_start_flag
assert state.on_iteration_complete_flag
assert not hasattr(state, 'on_catch_exception_flag')
class CustomException(Exception):
pass
@argus.callbacks.on_start
def on_start_raise_exception(state):
raise CustomException
on_start_raise_exception.attach(engine)
with pytest.raises(CustomException):
engine.run(data_loader, start_epoch=0, end_epoch=3)
assert engine.state.on_catch_exception_flag
def test_run(self):
class StepStorage:
def __init__(self):
self.batch_lst = []
self.state = None
def reset(self):
self.batch_lst = []
self.state = None
def step_method(self, batch, state):
self.batch_lst.append(batch)
self.state = state
step_storage = StepStorage()
data_loader = [4, 8, 15, 16, 23, 42]
engine = Engine(
step_storage.step_method,
logger=logging.getLogger('TestEngineMethods::test_run'))
state = engine.run(data_loader, start_epoch=0, end_epoch=3)
assert step_storage.batch_lst == data_loader * 3
assert state.epoch == 3
assert state.iteration == len(data_loader)
def stop_function(state):
state.stopped = True
step_storage.reset()
engine.add_event_handler(Events.EPOCH_COMPLETE, stop_function)
state = engine.run(data_loader, start_epoch=0, end_epoch=3)
assert step_storage.batch_lst == data_loader
assert state.epoch == 1
assert state.iteration == len(data_loader)
step_storage.reset()
engine.add_event_handler(Events.ITERATION_COMPLETE, stop_function)
state = engine.run(data_loader, start_epoch=0, end_epoch=3)
assert step_storage.batch_lst == [data_loader[0]]
assert state.iteration == 1
class CustomException(Exception):
pass
def exception_function(state):
raise CustomException
step_storage.reset()
engine.add_event_handler(Events.START, exception_function)
with pytest.raises(CustomException):
engine.run(data_loader, start_epoch=0, end_epoch=3)
assert step_storage.batch_lst == []
assert engine.state.iteration == 0
assert engine.state.epoch == 0
def test_engine(linear_argus_model_instance):
return Engine(linear_argus_model_instance.test_step)
def val_engine(linear_argus_model_instance):
return Engine(linear_argus_model_instance.val_step)
def train_engine(linear_argus_model_instance):
return Engine(linear_argus_model_instance.train_step)
