特徴量を出す関数をobservationに追加(py)

[OkanoShinri]

#1095

[OkanoShinri]

[OkanoShinri]

State -> Observation変換

def state2obs(state_json: str, who: int) -> Observation:
    state = State(state_json)
    state_proto = state.to_proto()
    observation_proto = mjxproto.Observation()
    # who
    observation_proto.who = who

    # public
    observation_proto.public_observation.CopyFrom(state_proto.public_observation)

    # private
    observation_proto.private_observation.CopyFrom(state_proto.private_observations[who])

    # round_terminal
    observation_proto.round_terminal.CopyFrom(state_proto.round_terminal)

    observation = Observation(
        Observation.add_legal_actions(Observation.from_proto(observation_proto).to_json())
    )
    return observation

現状のmjxを動かすスクリプト

readmeのままでは対応していない

from mjx import MjxEnv
from mjx.agent import RandomAgent

agent = RandomAgent()
env = MjxEnv()
_reset=list(env.reset().items())[0]
player_id=_reset[0] 
observation=_reset[1]
while not env.done():
    action = agent.act(observation)
    _step=list(env.step({player_id:action}).items())[0]
    player_id=_step[0] 
    observation=_step[1]
rewards = env.rewards()

[sotetsuk]

Suggested change

	def get_feature(self) -> np.ndarray:
	def _get_han22_features(self) -> np.ndarray: