def testOptionalObjectiveName(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
evaluation_function=sum_evaluation_function_v2,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
def testEvaluationFunctionV4Numpy(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
objective_name="sum",
evaluation_function=sum_evaluation_function_v4_numpy,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
def main():
args = parse_args()
function_list = [get_function_by_name[name]
for name in args.function_name_list.split(',')]
weight_list = list(map(float, args.weight_list.split(',')))
covariance_matrix = json.loads(args.covariance_matrix)
evaluate_covariance = args.evaluate_covariance
init_iter = args.init_iter
# if init_iter > 1:
# raise ValueError("init_iter should be 1.")
init_batch_size = args.init_batch_size
update_iter = args.update_iter
batch_size = args.batch_size
var_coef = args.var_coef
var_compute_type = args.var_compute_type
num_random = args.num_random
num_bucket = args.num_bucket
save_path = args.save_path
# num_control = args.num_control
minimize = True
groundtruth_function = get_groundtruth_function(function_list, weight_list)
#evaluation_function = get_evaluation_function(
# function_list, weight_list, covariance_matrix,
# evaluate_covariance, var_coef)
evaluation_function = get_evaluation_function(
function_list, weight_list, covariance_matrix,
var_compute_type, num_random, num_bucket)
exp = SimpleExperiment(
name=args.function_name_list + args.weight_list,
search_space=get_search_space(function_list),
evaluation_function=evaluation_function,
objective_name="objective_name",
minimize=minimize,
)
t_start = time.time()
print(f"Start time: {t_start}")
print(f"Sobol iteration begin...{time.time() - t_start}")
sobol = Models.SOBOL(exp.search_space)
for i in range(init_iter):
if init_batch_size == 1:
exp.new_trial(generator_run=sobol.gen(init_batch_size))
else:
exp.new_batch_trial(generator_run=sobol.gen(init_batch_size))
print(f"Running sobol optimization trial {i+1}/{init_iter}..."
f"{time.time() - t_start}")
print(f"GPEI iteration begin...{time.time() - t_start}")
for i in range(update_iter):
gpei = Models.BOTORCH(experiment=exp, data=exp.eval())
if batch_size == 1:
exp.new_trial(generator_run=gpei.gen(batch_size))
else:
exp.new_batch_trial(generator_run=gpei.gen(batch_size))
print(f"Running GPEI optimization trial {i+1}/{update_iter}..."
f"{time.time() - t_start}")
# Construct Result.
## origin data.
data_df = copy.deepcopy(exp.eval().df)
compare_func = min if minimize else max
arm_name2mean = {}
for _, row in data_df.iterrows():
arm_name2mean[row["arm_name"]] = row["mean"]
## parameters true_mean.
other_columns = {
"arm_name": [], "parameters": [], "true_mean": [],
"cur_trial_best_mean": [], "accum_trials_best_mean": []}
atbm = None # accum_trial_best_mean
for trial in exp.trials.values():
ctbm = None # cur_trial_best_mean
for arm in trial.arms:
other_columns['arm_name'].append(arm.name)
other_columns['parameters'].append(json.dumps(arm.parameters))
other_columns['true_mean'].append(
groundtruth_function(arm.parameters))
if ctbm is None:
ctbm = arm_name2mean[arm.name]
ctbm = compare_func(ctbm, arm_name2mean[arm.name])
if atbm is None:
atbm = ctbm
atbm = compare_func(atbm, ctbm)
other_columns['cur_trial_best_mean'].extend([ctbm] * len(trial.arms))
other_columns['accum_trials_best_mean'].extend([atbm] * len(trial.arms))
other_df = DataFrame(other_columns)
result_df = data_df.set_index('arm_name').join(
other_df.set_index('arm_name')).reset_index()
# Save to file.
print("Save to file.")
sub_dir_name = "_".join([
"ax", args.function_name_list.replace(",", "_"),
args.weight_list.replace(",", "_"), args.covariance_matrix.replace(
"[", "_").replace("]", "_").replace(",", "_").replace(" ", ""),
str(args.evaluate_covariance), str(args.init_iter), str(init_batch_size),
str(args.update_iter), str(args.batch_size), str(args.var_coef),
str(minimize), str(var_compute_type), str(num_random), str(num_bucket)
])
abs_dir_path = os.path.join(save_path, sub_dir_name)
Path(abs_dir_path).mkdir(parents=True, exist_ok=True)
task_id = os.environ.get('TASK_INDEX')
cur_time = pd.Timestamp.now().strftime('%Y%m%d%H%M%S')
filename = cur_time + "_" + str(task_id) + ".csv"
print(os.path.join(abs_dir_path, filename))
result_df.to_csv(os.path.join(abs_dir_path, filename))
print("2021-01-19 19:48:00")
print("Done...")
class SimpleExperimentTest(TestCase):
def setUp(self) -> None:
self.experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
evaluation_function=sum_evaluation_function,
objective_name="sum",
)
self.arms = [
Arm(parameters={"x1": 0.75, "x2": 1}),
Arm(parameters={"x1": 2, "x2": 7}),
Arm(parameters={"x1": 10, "x2": 8}),
Arm(parameters={"x1": -2, "x2": 10}),
]
def testBasic(self) -> None:
self.assertTrue(self.experiment.is_simple_experiment)
trial = self.experiment.new_trial()
with self.assertRaises(NotImplementedError):
trial.runner = SyntheticRunner()
with self.assertRaises(NotImplementedError):
self.experiment.add_tracking_metric(Metric(name="test"))
with self.assertRaises(NotImplementedError):
self.experiment.update_tracking_metric(Metric(name="test"))
self.assertTrue(self.experiment.eval_trial(trial).df.empty)
batch = self.experiment.new_batch_trial()
batch.add_arm(Arm(parameters={"x1": 5, "x2": 10}))
self.assertEqual(self.experiment.eval_trial(batch).df["mean"][0], 15)
self.experiment.new_batch_trial().add_arm(Arm(parameters={"x1": 15, "x2": 25}))
self.assertAlmostEqual(self.experiment.eval().df["mean"][1], 40)
self.assertEqual(batch.fetch_data().df["mean"][0], 15)
self.assertAlmostEqual(self.experiment.fetch_data().df["mean"][1], 40)
def testTrial(self) -> None:
for i in range(len(self.arms)):
self.experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(self.experiment.eval().df.empty)
def testUnimplementedEvaluationFunction(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
objective_name="sum",
)
with self.assertRaises(Exception):
experiment.evaluation_function(parameterization={})
experiment.evaluation_function = sum_evaluation_function
def testEvaluationFunctionNumpy(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
objective_name="sum",
evaluation_function=sum_evaluation_function_numpy,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
def testEvaluationFunctionV2(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
objective_name="sum",
evaluation_function=sum_evaluation_function_v2,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
def testEvaluationFunctionV2Numpy(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
objective_name="sum",
evaluation_function=sum_evaluation_function_v2_numpy,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
def testEvaluationFunctionV3(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
objective_name="sum",
evaluation_function=sum_evaluation_function_v3,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
def testEvaluationFunctionV3Numpy(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
objective_name="sum",
evaluation_function=sum_evaluation_function_v3_numpy,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
def testEvaluationFunctionV4(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
objective_name="sum",
evaluation_function=sum_evaluation_function_v4,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
def testEvaluationFunctionV4Numpy(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
objective_name="sum",
evaluation_function=sum_evaluation_function_v4_numpy,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
def testOptionalObjectiveName(self) -> None:
experiment = SimpleExperiment(
name="test_branin",
search_space=get_branin_search_space(),
evaluation_function=sum_evaluation_function_v2,
)
for i in range(len(self.arms)):
experiment.new_trial(generator_run=GeneratorRun(arms=[self.arms[i]]))
self.assertFalse(experiment.eval().df.empty)
