def make_parallel_env(self, max_episode_steps, n_workers):
if "env_generator" in self.env_info.keys():
env_gen = self.env_info.env_generator
else:
env_gen = env_generator(self.env.spec.id, max_episode_steps)
env_multi = make_envs(env_gen, n_envs=n_workers)
return env_multi
def __init__(
self,
env: gym.Env, # for testing
args: argparse.Namespace,
log_cfg: ConfigDict,
hyper_params: ConfigDict,
backbone: ConfigDict,
head: ConfigDict,
optim_cfg: ConfigDict,
):
"""Initialize.
Args:
env (gym.Env): openAI Gym environment
args (argparse.Namespace): arguments including hyperparameters and training settings
"""
env_gen = env_generator(env.spec.id, args)
env_multi = make_envs(env_gen, n_envs=hyper_params.n_workers)
Agent.__init__(self, env, args, log_cfg)
self.episode_steps = np.zeros(hyper_params.n_workers, dtype=np.int)
self.states: list = []
self.actions: list = []
self.rewards: list = []
self.values: list = []
self.masks: list = []
self.log_probs: list = []
self.i_episode = 0
self.next_state = np.zeros((1, ))
self.hyper_params = hyper_params
self.backbone_cfg = backbone
self.head_cfg = head
self.optim_cfg = optim_cfg
if not self.args.test:
self.env = env_multi
self.state_dim = self.env.observation_space.shape
self.action_dim = self.env.action_space.shape[0]
self.epsilon = hyper_params.max_epsilon
self._init_network()
def __init__(
self,
env: gym.Env,
env_info: ConfigDict,
args: argparse.Namespace,
hyper_params: ConfigDict,
learner_cfg: ConfigDict,
log_cfg: ConfigDict,
):
"""Initialize.
Args:
env (gym.Env): openAI Gym environment
args (argparse.Namespace): arguments including hyperparameters and training settings
"""
env_gen = env_generator(env.spec.id, args)
env_multi = make_envs(env_gen, n_envs=hyper_params.n_workers)
Agent.__init__(self, env, env_info, args, log_cfg)
self.episode_steps = np.zeros(hyper_params.n_workers, dtype=np.int)
self.states: list = []
self.actions: list = []
self.rewards: list = []
self.values: list = []
self.masks: list = []
self.log_probs: list = []
self.i_episode = 0
self.next_state = np.zeros((1, ))
self.hyper_params = hyper_params
self.learner_cfg = learner_cfg
self.learner_cfg.args = self.args
self.learner_cfg.env_info = self.env_info
self.learner_cfg.hyper_params = self.hyper_params
self.learner_cfg.log_cfg = self.log_cfg
self.learner_cfg.device = device
if not self.args.test:
self.env = env_multi
self.epsilon = hyper_params.max_epsilon
self.learner = build_learner(self.learner_cfg)
def __init__(
self,
env: gym.Env,
env_info: ConfigDict,
hyper_params: ConfigDict,
learner_cfg: ConfigDict,
log_cfg: ConfigDict,
is_test: bool,
load_from: str,
is_render: bool,
render_after: int,
is_log: bool,
save_period: int,
episode_num: int,
max_episode_steps: int,
interim_test_num: int,
):
env_gen = env_generator(env.spec.id, max_episode_steps)
env_multi = make_envs(env_gen, n_envs=hyper_params.n_workers)
Agent.__init__(
self,
env,
env_info,
log_cfg,
is_test,
load_from,
is_render,
render_after,
is_log,
save_period,
episode_num,
max_episode_steps,
interim_test_num,
)
self.episode_steps = np.zeros(hyper_params.n_workers, dtype=np.int)
self.states: list = []
self.actions: list = []
self.rewards: list = []
self.values: list = []
self.masks: list = []
self.log_probs: list = []
self.i_episode = 0
self.next_state = np.zeros((1,))
self.hyper_params = hyper_params
self.learner_cfg = learner_cfg
if not self.is_test:
self.env = env_multi
self.epsilon = hyper_params.max_epsilon
build_args = dict(
hyper_params=self.hyper_params,
log_cfg=self.log_cfg,
env_name=self.env_info.name,
state_size=self.env_info.observation_space.shape,
output_size=self.env_info.action_space.shape[0],
is_test=self.is_test,
load_from=self.load_from,
)
self.learner = build_learner(self.learner_cfg, build_args)