def inside_loop(my_model, n_trials, n_timebound):
Path(f'{Config.RESULTS_DATA}_{n_trials}').mkdir(parents=True, exist_ok=True)
tasks = load_tasks(f'{Config.INPUT_DATA}/tasks_{my_model["n_tasks"]}.json')
problem = deepcopy(my_model['problem'])
problem["tasks"] = tasks
problem["n_tasks"] = my_model["n_tasks"]
problem["shape"] = [len(problem["clouds"]) + len(problem["fogs"]), my_model["n_tasks"]]
# for paras in parameters_grid:
if Config.MODE == "epoch":
optimizer = BaseNSGA_III(problem, my_model["pop_size"], my_model["epoch"], my_model["func_eval"], my_model["lb"], my_model["ub"], paras=None)
solutions, g_best, g_best_dict = optimizer.train()
elif Config.MODE == "fe":
optimizer = BaseNSGA_III(problem, my_model["pop_size"], my_model["epoch"], my_model["func_eval"], my_model["lb"], my_model["ub"], paras=None)
solutions, g_best, g_best_dict = optimizer.train()
if Config.TIME_BOUND_KEY:
results_folder_path = f'{Config.RESULTS_DATA}_{n_timebound}s/{Config.METRICS}/'
else:
results_folder_path = f'{Config.RESULTS_DATA}_no_time_bound/{Config.METRICS}/'
Path(results_folder_path).mkdir(parents=True, exist_ok=True)
name_mha = 'nsgaiii'
name_paras = f'{my_model["epoch"]}_{my_model["pop_size"]}'
save_training_fitness_information(g_best_dict, len(tasks), name_mha, name_paras, results_folder_path)
save_experiment_result(problem, solutions, g_best, name_mha, name_paras, results_folder_path)
save_visualization(problem, g_best, name_mha, name_paras, results_folder_path)
def getting_results_for_task(models):
matrix_fit = zeros((1, 6))
for n_task in OptExp.N_TASKS:
for my_model in models:
tasks = load_tasks(f'{Config.INPUT_DATA}/tasks_{n_task}.json')
problem = deepcopy(my_model['problem'])
problem["tasks"] = tasks
problem["n_tasks"] = n_task
problem["shape"] = [len(problem["clouds"]) + len(problem["fogs"]), n_task]
my_model['problem'] = problem
for n_trials in range(OptExp.N_TRIALS):
if Config.TIME_BOUND_KEY:
for n_timebound in OptExp.TIME_BOUND_VALUES:
if Config.MODE == "epoch":
for epoch in OptExp.EPOCH:
end_paras = f"{epoch}"
df_matrix = inside_loop(my_model, n_trials, n_timebound, epoch, None, end_paras)
df_name = array([[n_task, my_model["name"], n_trials], ] * len(df_matrix))
matrix = hstack(df_name, df_matrix)
matrix_fit = vstack((matrix_fit, matrix))
else:
if Config.MODE == "epoch":
for epoch in OptExp.EPOCH:
end_paras = f"{epoch}"
df_matrix = inside_loop(my_model, n_trials, None, epoch, None, end_paras)
df_name = array([[n_task, my_model["name"], n_trials], ] * len(df_matrix))
matrix = hstack((df_name, df_matrix))
matrix_fit = vstack((matrix_fit, matrix))
return matrix_fit[1:]
def inside_loop(my_model, n_trials, n_timebound, epoch, fe, end_paras):
for n_tasks in OptExp.N_TASKS:
tasks = load_tasks(f'{Config.INPUT_DATA}/tasks_{n_tasks}.json')
problem = deepcopy(my_model['problem'])
problem["tasks"] = tasks
problem["n_tasks"] = n_tasks
problem["shape"] = [
len(problem["clouds"]) + len(problem["fogs"]), n_tasks
]
for pop_size in OptExp.POP_SIZE:
for lb, ub in zip(OptExp.LB, OptExp.UB):
parameters_grid = list(ParameterGrid(my_model["param_grid"]))
for paras in parameters_grid:
name_paras = f'{epoch}_{pop_size}_{end_paras}'
opt = getattr(optimizer,
my_model["class"])(problem=problem,
pop_size=pop_size,
epoch=epoch,
func_eval=fe,
lb=lb,
ub=ub,
verbose=OptExp.VERBOSE,
paras=paras)
solution, best_fit, best_fit_list, time_total = opt.train()
if Config.TIME_BOUND_KEY:
path_results = f'{Config.RESULTS_DATA}/{n_timebound}s/task_{n_tasks}/{Config.METRICS}/{my_model["name"]}/{n_trials}'
else:
path_results = f'{Config.RESULTS_DATA}/no_time_bound/task_{n_tasks}/{Config.METRICS}/{my_model["name"]}/{n_trials}'
Path(path_results).mkdir(parents=True, exist_ok=True)
save_experiment_results_single(
problem, solution, best_fit_list, name_paras,
time_total, path_results, Config.SAVE_TRAINING_RESULTS)
if Config.VISUAL_SAVING:
save_visualization_results_single(
problem, solution, best_fit, my_model["name"],
name_paras, path_results)