def run(self, limits, minimize, dynamic_optimizing, field_for_optimizing,
project_path, scope, batches):
super().run(minimize, dynamic_optimizing, field_for_optimizing,
project_path)
optimization_problem_parameters = []
init = []
bounds = []
for key in limits.keys():
pars = {
"name": key,
"category": "uniform",
"search_space": {
"low": limits[key]["low"],
"high": limits[key]["high"],
"step": limits[key]["step"]
}
}
self.fields.append(key)
optimization_problem_parameters.append(pars)
best_sample = benderopt.minimize(
f=self.box,
optimization_problem_parameters=optimization_problem_parameters,
number_of_evaluation=scope,
debug=False)
return best_sample
def benchmark_simple(
function_to_optimize,
optimization_problem,
target,
methods,
number_of_evaluations,
seeds,
parallel=True,
):
"""Aim to benchmark analytics functions of multiple variables"""
results_tmp = {}
assert set(target.keys()) == set([x["name"] for x in optimization_problem])
parameters = set(target.keys())
trials = list(product(seeds, number_of_evaluations, methods))
print("Number of trials:", len(trials))
if parallel is True:
best_samples = Parallel(n_jobs=-1, verbose=8)(delayed(minimize)(
function_to_optimize,
optimization_problem,
optimizer_type=method,
number_of_evaluation=number_of_evaluation,
seed=seed,
) for seed, number_of_evaluation, method in trials)
else:
best_samples = [
minimize(
function_to_optimize,
optimization_problem,
optimizer_type=method,
number_of_evaluation=number_of_evaluation,
seed=seed,
) for seed, number_of_evaluation, method in tqdm.tqdm(trials)
]
for (seed, number_of_evaluation,
method), best in zip(trials, best_samples):
results_tmp.setdefault(method, {}).setdefault(
number_of_evaluation, []).append(
np.sqrt(
np.sum([(best[parameter] - target[parameter])**2
for parameter in parameters])))
results = {
method: {
n: {
"mean": np.mean(values),
"std": np.std(values)
}
for n, values in ns.items()
}
for method, ns in results_tmp.items()
}
return results
def benchmark_value(
function_to_optimize,
optimization_problem,
target,
methods,
number_of_evaluations,
seeds,
parallel=True,
):
"""Aim to benchmark analytics functions of multiple variables"""
results_tmp = {}
trials = list(product(seeds, number_of_evaluations, methods))
print("Number of trials:", len(trials))
if parallel is True:
best_samples = Parallel(n_jobs=-1, verbose=8)(delayed(minimize)(
function_to_optimize,
optimization_problem,
optimizer_type=method,
number_of_evaluation=number_of_evaluation,
seed=seed,
) for seed, number_of_evaluation, method in trials)
else:
best_samples = [
minimize(
function_to_optimize,
optimization_problem,
optimizer_type=method,
number_of_evaluation=number_of_evaluation,
seed=seed,
) for seed, number_of_evaluation, method in tqdm.tqdm(trials)
]
for (seed, number_of_evaluation,
method), best in zip(trials, best_samples):
results_tmp.setdefault(method, {}).setdefault(
number_of_evaluation,
[]).append(np.abs(function_to_optimize(**best) - target))
results = {
method: {
n: {
"mean": np.mean(values),
"std": np.std(values),
}
for n, values in ns.items()
}
for method, ns in results_tmp.items()
}
return results