def run(problem, optimizer, n_calls, n_runs):
if problem == "fc":
problem = FCNetOnMnist()
bounds = hpolib_to_skopt_bounds(problem)
elif problem == "cnn":
problem = ConvolutionalNeuralNetworkOnCIFAR10()
bounds = hpolib_to_skopt_bounds(problem)
elif problem == "lr":
problem = LogisticRegressionOnMnist()
bounds = hpolib_to_skopt_bounds(problem)
else:
raise ValueError("Problem should be either fc | cnn | lr")
func = lambda x: problem.objective_function(configuration=x)[
"function_value"]
if optimizer == "gp":
opt = gp_minimize
elif optimizer == "forest":
opt = forest_minimize
min_vals = []
for n_run in range(n_runs):
res = opt(func,
bounds,
n_calls=n_calls,
n_random_starts=1,
verbose=1,
random_state=n_run)
del res["specs"]
dump(res, "%s_%d.pkl" % (optimizer, n_run))
min_vals.append(res.fun)
print(min_vals)
return np.min(min_vals), np.std(min_vals)
f = SvmOnMnist(rng=rng)
num_iterations = 15
output_path = "./experiments/fabolas/results/svm_%s/entropy_search_%d" % (
dataset, run_id)
elif dataset == "vehicle":
f = SvmOnVehicle(rng=rng)
num_iterations = 15
output_path = "./experiments/fabolas/results/svm_%s/entropy_search_%d" % (
dataset, run_id)
elif dataset == "covertype":
f = SvmOnCovertype(rng=rng)
num_iterations = 15
output_path = "./experiments/fabolas/results/svm_%s/entropy_search_%d" % (
dataset, run_id)
elif dataset == "cifar10":
f = ConvolutionalNeuralNetworkOnCIFAR10(rng=rng)
num_iterations = 15
output_path = "./experiments/fabolas/results/cnn_%s/entropy_search_%d" % (
dataset, run_id)
elif dataset == "svhn":
f = ConvolutionalNeuralNetworkOnSVHN(rng=rng)
num_iterations = 15
output_path = "./experiments/fabolas/results/cnn_%s/entropy_search_%d" % (
dataset, run_id)
elif dataset == "res_net":
f = ResidualNeuralNetworkOnCIFAR10(rng=rng)
num_iterations = 10
output_path = "./experiments/fabolas/results/res_%s/entropy_search_%d" % (
dataset, run_id)
os.makedirs(output_path, exist_ok=True)