def suggest(self, num=1):
"""Suggest a `num`ber of new sets of parameters.
TODO: document how suggest work for this algo
"""
# Make a copy so that all sampled trials are lost afterwards. The will be observed with fake
# results anyway, so they will be re-introduced in the algo before the next call to
# `suggest`. We only copy the storage however, otherwise the RNG stote increment inside the
# samplers would lost.
storage = self.study._storage
self.study._storage = copy.deepcopy(storage)
points = []
for i in range(num):
trial_id = self.study.storage.create_new_trial(self.study.study_id)
trial = optuna.trial.Trial(self.study, trial_id)
params = []
for param_name, _ in iterdims(self.space):
distribution = self.dimensions[param_name]
params.append(trial._suggest(param_name, distribution))
points.append(pack_point(params, self.space))
self.study._storage = storage
return points
def suggest(self, num=1):
"""Suggest a `num`ber of new sets of parameters.
Perform a step towards negative gradient and suggest that point.
"""
self._init_optimizer()
points = self.optimizer.ask(n_points=num, strategy=self.strategy)
return [pack_point(point, self.space) for point in points]
def suggest(self, num=1):
"""Suggest a `num`ber of new sets of parameters.
Perform a step towards negative gradient and suggest that point.
"""
if num > 1:
raise AttributeError("BayesianOptimizer does not support num > 1.")
points = [self.optimizer._ask()] # pylint: disable = protected-access
return [pack_point(point, self.space) for point in points]