def __init__(self,
act_spec: Spec,
obs_spec: Spec,
alpha=0.5,
epsilon=0.1,
gamma=1.0,
n=5,
kappa=0):
AgentProgram.__init__(self, act_spec, obs_spec)
self.alpha = alpha
self.epsilon = epsilon
self.gamma = gamma
self.kappa = kappa
self.n = n
self.q = np.zeros(shape=(obs_spec.size(), act_spec.size()))
self.sample = cache_gen(
lambda: np.random.randint(0, act_spec.size(), 1000))
if self.kappa:
self.model = TimeModel(act_spec, obs_spec, kappa)
else:
self.model = Model(act_spec, obs_spec)
self.s = None
self.a = None
def __init__(self, act_spec: Spec, obs_spec: Spec, alpha=0.5, epsilon=0.1,
gamma=1.0):
AgentProgram.__init__(self, act_spec, obs_spec)
self.alpha = alpha
self.epsilon = epsilon
self.gamma = gamma
self.q = np.zeros(shape=(obs_spec.size(), act_spec.size()))
self.sample = cache_gen(
lambda: np.random.randint(0, act_spec.size(), 1000))
self.s = None
self.a = None
def __init__(self,
act_spec: Spec,
obs_spec: Spec,
alpha=0.5,
epsilon=0.1,
gamma=1.0,
n=5,
kappa=0):
DynaQ.__init__(self, act_spec, obs_spec, alpha, epsilon, gamma, n,
kappa)
self.model = Model(act_spec, obs_spec)
self.t = 0
self.tau = np.ones(shape=(obs_spec.size(), act_spec.size()),
dtype=np.int32)
def __init__(self,
act_spec: Spec,
obs_spec: Spec,
alpha_w=0.5,
alpha_theta=0.5,
gamma=1.0,
lambda_w=0.9,
lambda_theta=0.9):
AgentProgram.__init__(self, act_spec, obs_spec)
# critic
self.w = np.zeros(self.MAX_SIZE)
self.alpha_w = alpha_w
self.lambda_w = lambda_w
self.z_w = np.zeros(self.MAX_SIZE)
self.step_size_w = alpha_w / self.N_TILINGS
# actor
self.theta = np.zeros(self.MAX_SIZE)
self.alpha_theta = alpha_theta
self.lambda_theta = lambda_theta
self.z_theta = np.zeros(self.MAX_SIZE)
self.step_size_theta = alpha_theta / self.N_TILINGS
self.gamma = gamma
self.hashtable = IHT(self.MAX_SIZE)
self.scales = [self.N_TILINGS / (s.hi - s.lo) for s in obs_spec]
self._a = np.arange(act_spec.size()).tolist()
self.s0 = None
self.a0 = None
self.I = 1
def __init__(self, act_spec: Spec, obs_spec: Spec, n=2,
alpha=0.5, epsilon=0.1, gamma=1.0):
AgentProgram.__init__(self, act_spec, obs_spec)
self.alpha = alpha
self.epsilon = epsilon
self.gamma = gamma
self.n = n
self.q = np.zeros(shape=(obs_spec.size(), act_spec.size()),
dtype=np.float64)
self.actions = np.arange(act_spec.size())
self.pi = e_greedy_policy(
np.zeros(shape=(obs_spec.size(), act_spec.size()),
dtype=np.float64),
self.q, self.epsilon)
self._t = 0
self._T = float('inf')
self._S = []
self._A = []
self._R = []
self._tau = 0
def __init__(self,
act_spec: Spec,
obs_spec: Spec,
alpha=0.5,
epsilon=0.1,
gamma=1.0):
AgentProgram.__init__(self, act_spec, obs_spec)
self.alpha = alpha
self.epsilon = epsilon
self.gamma = gamma
self.hashtable = IHT(self.MAX_SIZE)
self.w = np.zeros(self.MAX_SIZE)
self.scales = [self.N_TILINGS / (s.hi - s.lo) for s in obs_spec]
self.step_size = alpha / self.N_TILINGS
self.sample = cache_gen(
lambda: np.random.randint(0, act_spec.size(), 1000))
self.s = None
self.a = None
def __init__(self, act_spec: Spec, obs_spec: Spec, pi):
AgentProgram.__init__(self, act_spec, obs_spec)
self.pi = pi
self.actions = np.arange(act_spec.size())