def __init__(self, policy_cls, env_id, args):
logging.getLogger("tensorflow").setLevel(logging.ERROR)
self.args = args
self.env = gym.make(env_id, **args2envkwargs(args))
self.policy_with_value = policy_cls(self.args)
self.iteration = 0
if self.args.mode == 'training':
self.log_dir = self.args.log_dir + '/evaluator'
else:
self.log_dir = self.args.test_log_dir
if not os.path.exists(self.log_dir):
os.makedirs(self.log_dir)
self.preprocessor = Preprocessor((self.args.obs_dim, ),
self.args.obs_preprocess_type,
self.args.reward_preprocess_type,
self.args.obs_scale,
self.args.reward_scale,
self.args.reward_shift,
gamma=self.args.gamma)
self.writer = self.tf.summary.create_file_writer(self.log_dir)
self.stats = {}
self.eval_timer = TimerStat()
self.eval_times = 0
def built_parser(alg_name):
if alg_name == 'AMPC':
args = built_AMPC_parser()
env = gym.make(args.env_id, **args2envkwargs(args))
obs_space, act_space = env.observation_space, env.action_space
args.obs_dim, args.act_dim = obs_space.shape[0], act_space.shape[0]
args.obs_scale = [0.2, 1., 2., 1 / 30., 1 / 30, 1 / 180.] + \
[1., 1 / 15., 0.2] + \
[1., 1., 1 / 15.] * args.env_kwargs_num_future_data + \
[1 / 30., 1 / 30., 0.2, 1 / 180.] * env.veh_num
return args
def __init__(self, policy_cls, env_id, args, worker_id):
logging.getLogger("tensorflow").setLevel(logging.ERROR)
self.worker_id = worker_id
self.args = args
self.env = gym.make(env_id, **args2envkwargs(args))
self.policy_with_value = policy_cls(self.args)
self.batch_size = self.args.batch_size
self.obs = self.env.reset()
self.done = False
self.preprocessor = Preprocessor((self.args.obs_dim, ), self.args.obs_preprocess_type, self.args.reward_preprocess_type,
self.args.obs_scale, self.args.reward_scale, self.args.reward_shift,
gamma=self.args.gamma)
self.explore_sigma = self.args.explore_sigma
self.iteration = 0
self.num_sample = 0
self.sample_times = 0
self.stats = {}
logger.info('Worker initialized')
def __init__(self, policy_cls, args):
self.args = args
self.policy_with_value = policy_cls(self.args)
self.batch_data = {}
self.all_data = {}
self.M = self.args.M
self.num_rollout_list_for_policy_update = self.args.num_rollout_list_for_policy_update
self.model = EnvironmentModel(**args2envkwargs(args))
self.preprocessor = Preprocessor((self.args.obs_dim, ),
self.args.obs_preprocess_type,
self.args.reward_preprocess_type,
self.args.obs_scale,
self.args.reward_scale,
self.args.reward_shift,
gamma=self.args.gamma)
self.grad_timer = TimerStat()
self.stats = {}
self.info_for_buffer = {}