def _run_trial(self, func, catch):
# type: (ObjectiveFuncType, Union[Tuple[()], Tuple[Type[Exception]]]) -> trial_module.Trial
trial_id = self._storage.create_new_trial(self.study_id)
trial = trial_module.Trial(self, trial_id)
trial_number = trial.number
try:
result = func(trial)
except structs.TrialPruned as e:
message = 'Setting status of trial#{} as {}. {}'.format(
trial_number, structs.TrialState.PRUNED, str(e))
self.logger.info(message)
self._storage.set_trial_state(trial_id, structs.TrialState.PRUNED)
return trial
except catch as e:
message = 'Setting status of trial#{} as {} because of the following error: {}'\
.format(trial_number, structs.TrialState.FAIL, repr(e))
self.logger.warning(message, exc_info=True)
self._storage.set_trial_system_attr(trial_id, 'fail_reason',
message)
self._storage.set_trial_state(trial_id, structs.TrialState.FAIL)
return trial
finally:
# The following line mitigates memory problems that can be occurred in some
# environments (e.g., services that use computing containers such as CircleCI).
# Please refer to the following PR for further details:
# https://github.com/pfnet/optuna/pull/325.
if self.force_garbage_collection:
gc.collect()
try:
result = float(result)
except (
ValueError,
TypeError,
):
message = 'Setting status of trial#{} as {} because the returned value from the ' \
'objective function cannot be casted to float. Returned value is: ' \
'{}'.format(trial_number, structs.TrialState.FAIL, repr(result))
self.logger.warning(message)
self._storage.set_trial_system_attr(trial_id, 'fail_reason',
message)
self._storage.set_trial_state(trial_id, structs.TrialState.FAIL)
return trial
if math.isnan(result):
message = 'Setting status of trial#{} as {} because the objective function ' \
'returned {}.'.format(trial_number, structs.TrialState.FAIL, result)
self.logger.warning(message)
self._storage.set_trial_system_attr(trial_id, 'fail_reason',
message)
self._storage.set_trial_state(trial_id, structs.TrialState.FAIL)
return trial
trial.report(result)
self._storage.set_trial_state(trial_id, structs.TrialState.COMPLETE)
self._log_completed_trial(trial_number, result)
return trial
def ask(self) -> trial_module.Trial:
"""Create a new trial from which hyperparameters can be suggested.
This method is part of an alternative to :func:`~optuna.study.Study.optimize` that allows
controlling the lifetime of a trial outside the scope of ``func``. Each call to this
method should be followed by a call to :func:`~optuna.study.Study.tell` to finish the
created trial.
Example:
.. testcode::
import optuna
study = optuna.create_study()
trial = study.ask()
x = trial.suggest_float("x", -1, 1)
study.tell(trial, x ** 2)
Returns:
A :class:`~optuna.trial.Trial`.
"""
# Sync storage once every trial.
self._storage.read_trials_from_remote_storage(self._study_id)
trial_id = self._pop_waiting_trial_id()
if trial_id is None:
trial_id = self._storage.create_new_trial(self._study_id)
return trial_module.Trial(self, trial_id)
def _ask(self) -> trial_module.Trial:
# Sync storage once at the beginning of the objective evaluation.
self._storage.read_trials_from_remote_storage(self._study_id)
trial_id = self._pop_waiting_trial_id()
if trial_id is None:
trial_id = self._storage.create_new_trial(self._study_id)
return trial_module.Trial(self, trial_id)
def _run_trial(self, func, catch):
# type: (ObjectiveFuncType, Union[Tuple[()], Tuple[Type[Exception]]]) -> trial_module.Trial
trial_id = self.storage.create_new_trial_id(self.study_id)
trial = trial_module.Trial(self, trial_id)
try:
result = func(trial)
except structs.TrialPruned as e:
message = 'Setting trial status as {}. {}'.format(
structs.TrialState.PRUNED, str(e))
self.logger.info(message)
self.storage.set_trial_state(trial_id, structs.TrialState.PRUNED)
return trial
except catch as e:
message = 'Setting trial status as {} because of the following error: {}'.format(
structs.TrialState.FAIL, repr(e))
self.logger.warning(message, exc_info=True)
self.storage.set_trial_state(trial_id, structs.TrialState.FAIL)
self.storage.set_trial_system_attr(trial_id, 'fail_reason',
message)
return trial
try:
result = float(result)
except (
ValueError,
TypeError,
):
message = 'Setting trial status as {} because the returned value from the ' \
'objective function cannot be casted to float. Returned value is: ' \
'{}'.format(structs.TrialState.FAIL, repr(result))
self.logger.warning(message)
self.storage.set_trial_state(trial_id, structs.TrialState.FAIL)
self.storage.set_trial_system_attr(trial_id, 'fail_reason',
message)
return trial
if math.isnan(result):
message = 'Setting trial status as {} because the objective function returned ' \
'{}.'.format(structs.TrialState.FAIL, result)
self.logger.warning(message)
self.storage.set_trial_state(trial_id, structs.TrialState.FAIL)
self.storage.set_trial_system_attr(trial_id, 'fail_reason',
message)
return trial
trial.report(result)
self.storage.set_trial_state(trial_id, structs.TrialState.COMPLETE)
self._log_completed_trial(result)
return trial
def get_candidates(self, trial=None):
"""
This function converts the searchspace to a candidate_list that can then be used to distribute via MPI.
:param searchspace: converted hyperparameter space
"""
candidates_list = list()
N = self.max_iterations
for n in range(N):
print(n)
# Todo: Ugly hack that does not even work...
from optuna import trial as trial_module
# temp_study = optuna.create_study()
trial_id = self.study._storage.create_new_trial_id(0)
trial = trial_module.Trial(self.study, trial_id)
## trial.report(result)
## self._storage.set_trial_state(trial_id, structs.TrialState.COMPLETE)
## self._log_completed_trial(trial_number, result)
params = {}
for name, param in self._searchspace.items():
if param["domain"] == "categorical":
params[name] = trial.suggest_categorical(
name, param["data"])
else:
params[name] = trial.suggest_uniform(
name, param["data"][0], param["data"][1])
candidates_list.append(CandidateDescriptor(**params))
return candidates_list
N = self.max_iterations
for n in range(N):
params = {}
for name, param in self._searchspace.items():
if param["domain"] == "categorical":
params[name] = trial.suggest_categorical(
name, param["data"])
else:
params[name] = trial.suggest_uniform(
name, param["data"][0], param["data"][1])
candidates_list.append(CandidateDescriptor(**params))
return candidates_list
def ask(
self,
fixed_distributions: Optional[Dict[str, BaseDistribution]] = None
) -> trial_module.Trial:
"""Create a new trial from which hyperparameters can be suggested.
This method is part of an alternative to :func:`~optuna.study.Study.optimize` that allows
controlling the lifetime of a trial outside the scope of ``func``. Each call to this
method should be followed by a call to :func:`~optuna.study.Study.tell` to finish the
created trial.
.. seealso::
The :ref:`ask_and_tell` tutorial provides use-cases with examples.
Example:
Getting the trial object with the :func:`~optuna.study.Study.ask` method.
.. testcode::
import optuna
study = optuna.create_study()
trial = study.ask()
x = trial.suggest_float("x", -1, 1)
study.tell(trial, x ** 2)
Example:
Passing previously defined distributions to the :func:`~optuna.study.Study.ask`
method.
.. testcode::
import optuna
study = optuna.create_study()
distributions = {
"optimizer": optuna.distributions.CategoricalDistribution(["adam", "sgd"]),
"lr": optuna.distributions.LogUniformDistribution(0.0001, 0.1),
}
# You can pass the distributions previously defined.
trial = study.ask(fixed_distributions=distributions)
# `optimizer` and `lr` are already suggested and accessible with `trial.params`.
assert "optimizer" in trial.params
assert "lr" in trial.params
Args:
fixed_distributions:
A dictionary containing the parameter names and parameter's distributions. Each
parameter in this dictionary is automatically suggested for the returned trial,
even when the suggest method is not explicitly invoked by the user. If this
argument is set to :obj:`None`, no parameter is automatically suggested.
Returns:
A :class:`~optuna.trial.Trial`.
"""
fixed_distributions = fixed_distributions or {}
# Sync storage once every trial.
self._storage.read_trials_from_remote_storage(self._study_id)
trial_id = self._pop_waiting_trial_id()
if trial_id is None:
trial_id = self._storage.create_new_trial(self._study_id)
trial = trial_module.Trial(self, trial_id)
for name, param in fixed_distributions.items():
trial._suggest(name, param)
return trial
def _run_trial(
self,
func, # type: ObjectiveFuncType
catch, # type: Union[Tuple[()], Tuple[Type[Exception]]]
gc_after_trial, # type: bool
):
# type: (...) -> trial_module.Trial
trial_id = self._pop_waiting_trial_id()
if trial_id is None:
trial_id = self._storage.create_new_trial(self._study_id)
trial = trial_module.Trial(self, trial_id)
trial_number = trial.number
try:
result = func(trial)
except exceptions.TrialPruned as e:
message = "Setting status of trial#{} as {}. {}".format(
trial_number, structs.TrialState.PRUNED, str(e))
_logger.info(message)
# Register the last intermediate value if present as the value of the trial.
# TODO(hvy): Whether a pruned trials should have an actual value can be discussed.
frozen_trial = self._storage.get_trial(trial_id)
last_step = frozen_trial.last_step
if last_step is not None:
self._storage.set_trial_value(
trial_id, frozen_trial.intermediate_values[last_step])
self._storage.set_trial_state(trial_id, structs.TrialState.PRUNED)
return trial
except Exception as e:
message = "Setting status of trial#{} as {} because of the following error: {}".format(
trial_number, structs.TrialState.FAIL, repr(e))
_logger.warning(message, exc_info=True)
self._storage.set_trial_system_attr(trial_id, "fail_reason",
message)
self._storage.set_trial_state(trial_id, structs.TrialState.FAIL)
if isinstance(e, catch):
return trial
raise
finally:
# The following line mitigates memory problems that can be occurred in some
# environments (e.g., services that use computing containers such as CircleCI).
# Please refer to the following PR for further details:
# https://github.com/optuna/optuna/pull/325.
if gc_after_trial:
gc.collect()
try:
result = float(result)
except (
ValueError,
TypeError,
):
message = (
"Setting status of trial#{} as {} because the returned value from the "
"objective function cannot be casted to float. Returned value is: "
"{}".format(trial_number, structs.TrialState.FAIL,
repr(result)))
_logger.warning(message)
self._storage.set_trial_system_attr(trial_id, "fail_reason",
message)
self._storage.set_trial_state(trial_id, structs.TrialState.FAIL)
return trial
if math.isnan(result):
message = (
"Setting status of trial#{} as {} because the objective function "
"returned {}.".format(trial_number, structs.TrialState.FAIL,
result))
_logger.warning(message)
self._storage.set_trial_system_attr(trial_id, "fail_reason",
message)
self._storage.set_trial_state(trial_id, structs.TrialState.FAIL)
return trial
self._storage.set_trial_value(trial_id, result)
self._storage.set_trial_state(trial_id, structs.TrialState.COMPLETE)
self._log_completed_trial(trial_number, result)
return trial
