def setup(self):
param_def = {
"x": MinMaxNumericParamDef(0, 1)
}
experiment = Experiment(name="test_optimizer_experiment",
parameter_definitions=param_def)
self.optimizer = QueueBasedOptimizer(RandomSearch,
experiment)
class TestQueueOptimizer(object):
optimizer = None
def setup(self):
param_def = {
"x": MinMaxNumericParamDef(0, 1)
}
experiment = Experiment(name="test_optimizer_experiment",
parameter_definitions=param_def)
self.optimizer = QueueBasedOptimizer(RandomSearch,
experiment)
def test_get_next_candidate(self):
self.optimizer.get_next_candidates()
def test_update(self):
param_def = {
"x": MinMaxNumericParamDef(0, 1)
}
experiment = Experiment(name="test_optimizer_experiment",
parameter_definitions=param_def)
self.optimizer.update(experiment)
def teardown(self):
self.optimizer.exit()
def check_optimizer(optimizer, experiment, optimizer_arguments=None):
"""
Checks whether optimizer is an optimizer or builds one.
Specifically, it tests whether optimizer is an Optimizer instance. If
it is, it is returned unchanged, all other parameters are ignored. If
it is a class of optimizer, it will initialize it with experiment and
optimizer_arguments. If it is a basestring, it will be translated via
optimizer_utils.AVAILABLE_OPTIMIZERS, then initialized.
Parameters
----------
optimizer : string, Optimizer instance, optimizer class
The optimizer to initialize. If optimizer instance, the other
parameters will be ignored.
experiment : Experiment
The experiment defining the optimizer.
optimizer_arguments : dict, optional
The parameters governing the behaviour of the optimizer. If None,
default values are used.
This class introduces an additional parameter, called multiprocessing.
If "queue", the default, it will initialize the optimizer abstracted by
a QueueBasedOptimizer. If "none", it will initialize it directly.
Returns
-------
optimizer : Optimizer instance
An initialized optimizer instance.
Raises
------
ValueError
If the optimizer is a string, and one cannot find it in
AVAILABLE_OPTIMIZERS. If not an optimizer subclass. If the
multiprocessing argument is not an acceptable value.
"""
if optimizer_arguments is None:
optimizer_arguments = {}
multi_architecture = optimizer_arguments.get("multiprocessing", "queue")
if isinstance(optimizer, Optimizer):
return optimizer
if isinstance(optimizer, basestring):
try:
optimizer = AVAILABLE_OPTIMIZERS[optimizer]
except:
raise ValueError("No corresponding optimizer found for %s. "
"Optimizer must be in %s" %
(str(optimizer), AVAILABLE_OPTIMIZERS.keys()))
if not issubclass(optimizer, Optimizer):
raise ValueError("%s is of type %s, not Optimizer type." %
(optimizer, type(optimizer)))
if multi_architecture == "queue":
return QueueBasedOptimizer(optimizer, experiment, optimizer_arguments)
elif multi_architecture == "none":
return optimizer(experiment, optimizer_arguments)
else:
raise ValueError("%s is not supported as a multi-architecture "
"parameter. Currently supported are %s" %
(multi_architecture, ["none", "queue"]))