if __name__ == '__main__': time_of_start = time.time() # set the title of the terminal so that what the terminal is doing is clear print('\33]0;{}\a'.format(' '.join(sys.argv)), end='', flush=True) print(args) # compile the simulation module in C check_C_module_and_compile() # set the replay memory capacity = round(args.size_of_replay_memory * controls_per_half_period * t_max) if args.train else 1 memory = RL.Memory(capacity = capacity, data_size = data_size * 2 + 2 if args.input != 'measurements' else \ (read_control_step_length+read_length) + read_length//read_control_step_length+1 + 2, policy = 'random', passes_before_random = 0.2) # define the neural network net = RL.direct_DQN(data_size).cuda( ) if args.input != 'measurements' else RL.DQN_measurement(read_length) # set the task if args.train or args.LQG: train = RL.TrainDQN(net, memory, batch_size=args.batch_size, gamma=0.99, backup_period=args.target_network_update_interval, args=args) del net # the main function of training if args.train:
if __name__ == '__main__': time_of_start = time.time() # set the title of the terminal so that what the terminal is doing is clear print('\33]0;{}\a'.format(' '.join(sys.argv)), end='', flush=True) print(args) # compile the simulation module in C check_C_module_and_compile() # set the replay memory capacity = round(args.size_of_replay_memory * controls_per_unit_time * t_max) if args.train else 1 memory = RL.Memory(capacity=capacity, data_size=data_size * 2 + 2, policy='random', passes_before_random=0.2) # define the neural network net = RL.direct_DQN(data_size).cuda() # set the task if args.train or args.control_strategy != 'DQN': train = RL.TrainDQN(net, memory, batch_size=args.batch_size, gamma=0.99, backup_period=args.target_network_update_interval, args=args) del net # the main function of training if args.train: main = Main_System(train, num_of_processes=args.num_of_actors)