def run_rep_parallel(bayes_optimiser: BayesOptimiser, queue, config_ind,
rep_ind, seed):
torch.manual_seed(seed)
np.random.seed(seed)
bayes_optimiser.run_all_opt_steps()
# message = (config_ind, rep_ind, bayes_optimiser.best_val(in_real_units=True))
vals = bayes_optimiser.obj_func_vals_real_units()
best_vals = bayes_optimiser.best_val(in_real_units=True)
message = (config_ind, rep_ind, vals, best_vals)
# message = (config_ind, rep_ind, best_vals)
queue.put(message)
(min_vsf_depth_20N - param_dic["target_min_vsf_depth_20N"])**2)
return y, min_vsf_depth_20N
if __name__ == '__main__':
n_init_points = 16
n_bayes_points = 48
n_evals_per_step = 8
measure_year = 100
target_min_vsf_depth_20N = -15 * 10**6
obj_func = OceanObjSimThree(target_min_vsf_depth_20N,
measure_year=measure_year)
optimiser = BayesOptimiser(n_init_points,
n_bayes_points,
obj_func,
n_evals_per_step=n_evals_per_step)
optimiser.save_optimiser()
# slurm_set_up.set_up(
# optimiser.file_name, ["modi_long", "modi_short"], "global_four_degree.py", make_new_slurm_controller=True,
# using_singularity=True, image_path="~/modi_images/hpc-ocean-notebook_latest.sif", conda_env="python3")
#
#
# slurm_set_up.start_opt_run("modi001")
alpha = 1.0
omega = 1.0
acq_func = PredefinedAcqFunction(obj_func.bounds,
n_objs,
n_evals_per_step,
acqfunc_name='EHVI',
seq_dist_punish=True,
alpha=alpha,
omega=omega)
constraints = n_objs * [{"covar_module": {"raw_lengthscale": [0.1, 2.0]}}]
model_list = n_objs * [MaternModelBO] # Matern is the default
kernel = BayesOptModel(obj_func.n_params,
obj_func.n_objs,
model_class_list=model_list,
init_train_its=1000,
constraint_dict_list=constraints)
optimiser = BayesOptimiser(n_init_points,
n_bayes_points,
obj_func,
acq_func,
model=kernel,
n_evals_per_step=n_evals_per_step)
veropt_gui.run(optimiser)
def run():
n_init_points = 16
n_bayes_points = 64
n_objs = 1
beta = 3.0
gamma = 0.01
n_evals_per_step = 8
# obj_func = PredefinedFitTestFunction("sine_3params")
obj_func = PredefinedTestFunction("Hartmann")
kernel = BayesOptModel(obj_func.n_params,
n_objs,
model_class_list=[MaternModelBO],
init_train_its=1000,
using_priors=False)
# alpha_array = np.linspace(0.1, 10, num=20)
# omega_array = np.linspace(0.1, 1.0, num=10)
alpha_array = np.linspace(0.1, 10, num=5)
omega_array = np.linspace(0.1, 1.0, num=5)
optimiser_list = []
for alpha in alpha_array:
omega = 0.5
acq_func = PredefinedAcqFunction(
obj_func.bounds,
n_objs,
n_evals_per_step=n_evals_per_step,
acqfunc_name="UCB_Var",
seq_dist_punish=True,
beta=beta,
gamma=gamma,
alpha=alpha,
omega=omega,
)
optimiser_list.append(
BayesOptimiser(n_init_points,
n_bayes_points,
obj_func,
acq_func,
model=kernel,
test_mode=False,
using_priors=False,
n_evals_per_step=n_evals_per_step,
verbose=False))
for omega in omega_array:
alpha = 2.0
acq_func = PredefinedAcqFunction(obj_func.bounds,
n_objs,
n_evals_per_step=n_evals_per_step,
acqfunc_name="UCB_Var",
seq_dist_punish=True,
beta=beta,
gamma=gamma,
alpha=alpha,
omega=omega)
optimiser_list.append(
BayesOptimiser(n_init_points,
n_bayes_points,
obj_func,
acq_func,
model=kernel,
test_mode=False,
using_priors=False,
n_evals_per_step=n_evals_per_step,
verbose=False))
parameters = {"alpha": alpha_array, "omega": omega_array}
experiment = BayesExperiment(optimiser_list, parameters)
# experiment.run_full_experiment()
# experiment.run_full_exp_parallel(save=False)
experiment.save_experiment()
super().__init__(bounds=bounds,
n_params=n_params,
n_objs=n_objs,
calc_y_method=calc_y_method,
var_names=var_names,
obj_names=obj_names,
file_path=file_path)
if __name__ == '__main__':
n_init_points = 8
n_bayes_points = 40
n_evals_per_step = 8
measure_year = 100
target_vsf_depth_min_equator = -3200856.86124386
obj_func = OceanObjSimTwo(target_vsf_depth_min_equator,
measure_year=measure_year)
optimiser = BayesOptimiser(n_init_points,
n_bayes_points,
obj_func,
using_priors=False,
n_evals_per_step=n_evals_per_step)
optimiser.save_optimiser()