def perform_sequential_line_search(trial_index: int) -> np.ndarray:
optimality_gaps = np.ndarray(NUM_ITERS, )
# Instantiate the optimizer with acquisition function specification
optimizer = pySequentialLineSearch.SequentialLineSearchOptimizer(
num_dims=num_dims,
use_map_hyperparams=USE_MAP_HYPERPARAMS,
acquisition_func_type=acquisition_func_type[0],
initial_query_generator=generate_initial_slider)
# Specify a hyperparameter in case the acquisition function is GP-UCB
optimizer.set_gaussian_process_upper_confidence_bound_hyperparam(
acquisition_func_type[1])
for i in range(NUM_ITERS):
slider_ends = optimizer.get_slider_ends()
slider_position = ask_human_for_slider_manipulation(slider_ends)
optimizer.submit_feedback_data(slider_position)
optimality_gap = -calc_simulated_objective_func(
optimizer.get_maximizer())
optimality_gaps[i] = optimality_gap
return optimality_gaps
def perform_sequential_line_search(trial_index: int) -> np.ndarray:
optimality_gaps = np.ndarray(NUM_ITERS, )
optimizer = pySequentialLineSearch.SequentialLineSearchOptimizer(
num_dims=num_dims,
use_map_hyperparams=USE_MAP_HYPERPARAMS,
kernel_type=kernel_type,
initial_slider_generator=generate_initial_slider)
optimizer.set_hyperparams(kernel_signal_var=0.50,
kernel_length_scale=length_scale,
kernel_hyperparams_prior_var=0.10)
for i in range(NUM_ITERS):
slider_ends = optimizer.get_slider_ends()
slider_position = ask_human_for_slider_manipulation(slider_ends)
optimizer.submit_line_search_result(slider_position)
optimality_gap = -calc_simulated_objective_func(
optimizer.get_maximizer())
optimality_gaps[i] = optimality_gap
return optimality_gaps
def main():
optimizer = pySequentialLineSearch.SequentialLineSearchOptimizer(
num_dims=5)
optimizer.set_hyperparams(kernel_signal_var=0.50,
kernel_length_scale=0.10,
kernel_hyperparams_prior_var=0.10)
for i in range(30):
slider_ends = optimizer.get_slider_ends()
slider_position = ask_human_for_slider_manipulation(slider_ends)
optimizer.submit_feedback_data(slider_position)
residual = np.linalg.norm(optimizer.get_maximizer() - 0.2)
print("[#iters = " + str(i + 1) + "] residual: " + str(residual))
optimality_gaps = np.ndarray(shape=(NUM_ITERS, NUM_TRIALS,
len(length_scale_set),
len(map_condition_set), len(num_dims_set)))
# Perform sequential line search procedures with various conditions
for index_num_dims, num_dims in enumerate(num_dims_set):
print("Testing on a {}-dimensional function...".format(num_dims))
for index_map_condition, map_condition in enumerate(map_condition_set):
print("\t" + "Hyperparameters condition: " + str(map_condition))
for index_length_scale, length_scale in enumerate(length_scale_set):
print("\t\t" + "Kernel length scale: " + str(length_scale))
for trial in range(NUM_TRIALS):
optimizer = pySequentialLineSearch.SequentialLineSearchOptimizer(
num_dims=num_dims,
use_map_hyperparams=map_condition["use_map_hyperparams"],
kernel_type=KERNEL_TYPE,
initial_slider_generator=generate_initial_slider)
optimizer.set_hyperparams(
kernel_signal_var=0.50,
kernel_length_scale=length_scale,
kernel_hyperparams_prior_var=map_condition[
"kernel_hyperparams_prior_var"])
for i in range(NUM_ITERS):
slider_ends = optimizer.get_slider_ends()
slider_position = ask_human_for_slider_manipulation(
slider_ends)
optimizer.submit_line_search_result(slider_position)