) test = np.zeros(eval_pts.shape[0]) ps = PosteriorMeanMCMC(cpp_gp_loglikelihood.models, num_fidelity) for i, pt in enumerate(eval_pts): ps.set_current_point( pt.reshape((1, cpp_gp_loglikelihood.dim - objective_func._num_fidelity)) ) test[i] = -ps.compute_objective_function() report_point = eval_pts[np.argmin(test)].reshape( (1, cpp_gp_loglikelihood.dim - objective_func._num_fidelity) ) py_repeated_search_domain = RepeatedDomain(num_repeats=1, domain=inner_search_domain) ps_mean_opt = pyGradientDescentOptimizer( py_repeated_search_domain, ps, py_sgd_params_ps ) report_point = multistart_optimize(ps_mean_opt, report_point, num_multistarts=1)[0] report_point = report_point.ravel() report_point = np.concatenate((report_point, np.ones(objective_func._num_fidelity))) print( "best so far in the initial data {0}".format( true_value_init[np.argmin(true_value_init[:, 0])][0] ) ) capital_so_far = 0.0 start = time.time() for n in range(num_iteration): print( method
(1, cpp_gp_loglikelihood.dim - objective_func._num_fidelity))) test[i] = -pvar.compute_objective_function() initial_points = np.zeros( (20, cpp_gp_loglikelihood.dim - objective_func._num_fidelity)) indices = np.argsort(test) for i in range(20): initial_points[i, :] = eval_pts[indices[i]] #initial_point = eval_pts[np.argmin(test)].reshape((1, cpp_gp_loglikelihood.dim-objective_func._num_fidelity)) py_repeated_search_domain = RepeatedDomain(num_repeats=1, domain=inner_search_domain) pvar_mean_opt = pyGradientDescentOptimizer(py_repeated_search_domain, pvar, py_sgd_params_acquisition) report_point = multistart_optimize(pvar_mean_opt, initial_points, num_multistarts=20)[0] pvar.set_current_point( report_point.reshape( (1, cpp_gp_loglikelihood.dim - objective_func._num_fidelity))) if -pvar.compute_objective_function() > np.min(test): report_point = initial_points[[0]] next_points = report_point voi = np.nan elif method == 'PI': eval_pts = inner_search_domain.generate_uniform_random_points_in_domain( int(1e3))