def __init__(
self,
env,
scale_reward=1.0,
normalize_obs=False,
normalize_reward=False,
flatten_obs=True,
obs_alpha=0.001,
reward_alpha=0.001,
):
self._Serializable__initialize(locals())
super(NormalizedEnv, self).__init__(env)
self._scale_reward = scale_reward
self._normalize_obs = normalize_obs
self._normalize_reward = normalize_reward
self._flatten_obs = flatten_obs
self._obs_alpha = obs_alpha
flat_obs_dim = flat_dim(env.observation_space)
self._obs_mean = np.zeros(flat_obs_dim)
self._obs_var = np.ones(flat_obs_dim)
self._reward_alpha = reward_alpha
self._reward_mean = 0.0
self._reward_var = 1.0
def __init__(self,
action_space,
mu=0,
sigma=0.5,
theta=0.3,
dt=1e-2,
x0=None):
self.action_space = action_space
self.action_dim = flat_dim(self.action_space)
self.mu = mu
self.sigma = sigma
self.theta = theta
self.dt = dt
self.x0 = x0
self.reset()
def agent_flat_dim(self, i):
assert i in range(self.agent_num)
return utils.flat_dim(self.spaces[i])
def opponent_flat_dim(self, i):
assert i in range(self.agent_num)
return self.flat_dim - utils.flat_dim(self.spaces[i])
def flat_dim(self):
"""
The dimension of the flattened vector of the tensor representation
"""
return np.sum([utils.flat_dim(x) for x in self.spaces])