def __init__(self, discount: float):
super().__init__(discount)
self.env = gym.make('CartPole-v1')
self.env = ScalingObservationWrapper(self.env,
low=[-2.4, -2.0, -0.42, -3.5],
high=[2.4, 2.0, 0.42, 3.5])
self.actions = list(
map(lambda i: Action(i), range(self.env.action_space.n)))
self.observations = [self.env.reset()]
self.done = False
def legal_actions(self) -> List[Action]:
return list(
map(lambda i: Action(i), range(len(list(self.board.legal_moves)))))
def recurrent_inference(self, hidden_state, action) -> NetworkOutput:
return NetworkOutput(
0, 0,
{Action(i): 1 / self.action_size
for i in range(self.action_size)}, None)
def initial_inference(self, image) -> NetworkOutput:
return NetworkOutput(
0, 0,
{Action(i): 1 / self.action_size
for i in range(self.action_size)}, None)
def build_policy_logits(policy_logits):
return {Action(i): logit for i, logit in enumerate(policy_logits[0])}