def __init__(self,
action_selector: ActionSelector,
max_exploratory_steps: int = 1000):
ActionSelector.__init__(self, action_selector.use_cuda)
self.action_selector = action_selector
self.exploration = self.action_selector.exploration
self.max_exploratory_steps = max_exploratory_steps
def __init__(
self,
action_selector: ActionSelector,
action_space: list,
mu: float = 0.0,
theta: float = 0.15,
max_sigma: float = 0.3,
min_sigma: float = 0.3,
decay_period: int = 100000,
):
ActionSelector.__init__(self, action_selector.use_cuda)
self.action_selector = action_selector
self.mu = mu
self.theta = theta
self.sigma = max_sigma
self.max_sigma = max_sigma
self.min_sigma = min_sigma
self.decay_period = decay_period
self.action_min = self.action_selector.action_min
self.action_max = self.action_selector.action_max
self.action_dim = self.action_selector.action_dim
self.exploration = True
self._reset()
def __init__(self, action_selector: ActionSelector, mu: float,
sigma: float):
ActionSelector.__init__(self, action_selector.use_cuda)
self.action_selector = action_selector
self.action_min = self.action_selector.action_min
self.action_max = self.action_selector.action_max
self.action_dim = self.action_selector.action_dim
self.mu = mu
self.sigma = sigma
self.exploration = True
def __init__(
self,
action_selector: ActionSelector,
action_space: spaces.Discrete,
hyper_params: DictConfig,
):
ActionSelector.__init__(self, action_selector.use_cuda)
self.action_selector = action_selector
self.action_space = action_space
self.eps = hyper_params.eps
self.eps_final = hyper_params.eps_final
self.eps_decay = (self.eps -
self.eps_final) / hyper_params.max_exploration_frame
def test(
self,
policy: BaseModel,
action_selector: ActionSelector,
episode_i: int,
update_step: int,
) -> float:
"""Test policy without random exploration a number of times."""
print("====TEST START====")
policy.eval()
action_selector.exploration = False
episode_rewards = []
for test_i in range(self.experiment_info.test_num):
state = self.env.reset()
episode_reward = 0
done = False
while not done:
if self.experiment_info.render_train:
self.env.render()
action = action_selector(policy, state)
state, action, reward, next_state, done = self.step(
state, action)
episode_reward = episode_reward + reward
state = next_state
print(
f"episode num: {episode_i} | test: {test_i} episode reward: {episode_reward}"
)
episode_rewards.append(episode_reward)
mean_rewards = np.mean(episode_rewards)
print(f"EPISODE NUM: {episode_i} | UPDATE STEP: {update_step} |"
f"MEAN REWARD: {np.mean(episode_rewards)}")
action_selector.exploration = True
print("====TEST END====")
return mean_rewards
def __init__(self, device: str):
ActionSelector.__init__(self, device)
def __init__(self, action_dim: int, action_range: list, use_cuda: bool):
ActionSelector.__init__(self, use_cuda)
self.action_dim = action_dim
self.action_min = np.array(action_range[0])
self.action_max = np.array(action_range[1])
def __init__(self, use_cuda: bool):
ActionSelector.__init__(self, use_cuda)