def test_optimization_problem_add_bad_type():
with pytest.raises(ValidationError):
OptimizationProblem.from_list({
"name": "param1",
"category": "categorical",
"search_space": {
"values": ["a", "b"]
}
})
def test_optimization_problem_from_list():
optimization_problem = OptimizationProblem.from_list(
[
{"name": "param1", "category": "categorical", "search_space": {"values": ["a", "b"]}},
{"name": "param2", "category": "uniform", "search_space": {"low": 1, "high": 2}},
]
)
optimization_problem.add_observations_from_list(
[
{"loss": 1.5, "sample": {"param1": "a", "param2": 1.5}},
{"loss": 1.8, "sample": {"param1": "b", "param2": 1.8}},
{"loss": 0.1, "sample": {"param1": "b", "param2": 1.05}},
],
raise_exception=True,
)
assert type(optimization_problem.parameters) == list
assert len(optimization_problem.observations) == 3
assert optimization_problem.parameters_name == set(["param1", "param2"])
assert {"param1": "b", "param2": 1.8} in optimization_problem.samples
assert len(optimization_problem.samples) == 3
assert optimization_problem.best_sample == {"param1": "b", "param2": 1.05}
assert optimization_problem.sorted_observations[0].sample == {"param1": "b", "param2": 1.05}
assert optimization_problem.finite is False
assert len(optimization_problem.find_observations({"param1": "b", "param2": 1.05})) == 1
a, b = optimization_problem.observations_quantile(0.5)
assert len(a) == 1
assert len(b) == 2
assert optimization_problem.get_best_k_samples(1)[0].sample == {"param1": "b", "param2": 1.05}
def minimize(
f,
optimization_problem_parameters,
optimizer_type="parzen_estimator",
number_of_evaluation=100,
seed=None,
debug=False,
):
logger = logging.getLogger("benderopt")
np.random.seed(seed=seed)
samples = []
optimization_problem = OptimizationProblem.from_list(
optimization_problem_parameters)
optimizer = optimizers[optimizer_type](optimization_problem)
for i in range(number_of_evaluation):
logger.info("Evaluating {0}/{1}...".format(i + 1,
number_of_evaluation))
sample = optimizer.suggest()
samples.append(sample)
loss = f(**sample)
logger.debug("f={0} for optimizer suggestion: {1}.".format(
loss, sample))
observation = Observation.from_dict({"loss": loss, "sample": sample})
optimization_problem.add_observation(observation)
if debug is True:
return samples
return optimization_problem.best_sample
def test_optimization_problem_from_list_bad_type():
optimization_problem = OptimizationProblem.from_list(
[
{"name": "param1", "category": "categorical", "search_space": {"values": ["a", "b"]}},
{"name": "param2", "category": "uniform", "search_space": {"low": 1, "high": 2}},
]
)
with pytest.raises(ValidationError):
optimization_problem.add_observations_from_list("lol", raise_exception=True)
def test_optimization_problem_add_bad_observation():
optimization_problem = OptimizationProblem.from_list(
[
{"name": "param1", "category": "categorical", "search_space": {"values": ["a", "b"]}},
{"name": "param2", "category": "uniform", "search_space": {"low": 1, "high": 2}},
]
)
observation2 = Observation(sample={"lol": "b", "param2": 1.8}, loss=1.8)
with pytest.raises(ValidationError):
optimization_problem.add_observation(observation2)
def test_base_optimize_ok():
optimization_problem = [
{"name": "x", "category": "uniform", "search_space": {"low": 0, "high": np.pi,}}
]
optimization_problem = OptimizationProblem.from_list(optimization_problem)
optimizer = optimizers["random"](optimization_problem)
assert optimizer.observations == optimization_problem.observations
optimizer = optimizers["random"](optimization_problem)
assert len(optimizer.suggest(10)) == 10
def minimize(f,
optimization_problem_parameters,
optimizer_type="parzen_estimator",
number_of_evaluation=100,
seed=None,
debug=False,
):
np.random.seed(seed=seed)
if type(optimization_problem_parameters) == list:
optimization_problem = OptimizationProblem.from_list(optimization_problem_parameters)
samples = []
optimizer = optimizers[optimizer_type](optimization_problem)
for _ in range(number_of_evaluation):
sample = optimizer.suggest()
samples.append(sample)
loss = f(**sample)
observation = Observation.from_dict({"loss": loss, "sample": sample})
optimization_problem.add_observation(observation)
if debug is True:
return samples
return optimization_problem.best_sample
def minimize(
f,
optimization_problem_parameters,
optimizer_type="parzen_estimator",
number_of_evaluation=100,
seed=None,
debug=False,
):
logger = logging.getLogger("benderopt")
RNG.seed(seed)
samples = []
optimization_problem = OptimizationProblem.from_list(optimization_problem_parameters)
if isinstance(optimizer_type, str):
optimizer_type = optimizers[optimizer_type]
if not issubclass(optimizer_type, BaseOptimizer):
raise ValueError(
"optimizer_type should either be a string or a subclass of BaseOptimizer, got {}".format(
optimizer_type
)
)
optimizer = optimizer_type(optimization_problem)
for i in range(number_of_evaluation):
logger.info("Evaluating {0}/{1}...".format(i + 1, number_of_evaluation))
sample = optimizer.suggest()
samples.append(sample)
loss = f(**sample)
logger.debug("f={0} for optimizer suggestion: {1}.".format(loss, sample))
observation = Observation.from_dict({"loss": loss, "sample": sample})
optimization_problem.add_observation(observation)
if debug is True:
return samples
return optimization_problem.best_sample
samples.append(sample)
loss = f(**sample)
observation = Observation.from_dict({"loss": loss, "sample": sample})
optimization_problem.add_observation(observation)
if debug is True:
return samples
return optimization_problem.best_sample
if __name__ == "__main__":
def f(x):
return np.sin(x)
optimization_problem_parameters = [{
"name": "x",
"category": "uniform",
"search_space": {
"low": 0,
"high": 2 * np.pi,
}
}]
optimization_problem = OptimizationProblem.from_list(
optimization_problem_parameters)
best_sample = minimize(f,
optimization_problem=optimization_problem,
number_of_evaluation=100)
print(best_sample["x"], 3 * np.pi / 2)