def sample(self, state, action):
if state._context != self.current_outcome.observation._context:
self._gym_env.set_state(state._context)
outcome = super().step(action)
observation = GymDomainStateProxy(state=normalize_and_round(outcome.observation), context=self._gym_env.get_state())
self.current_outcome = EnvironmentOutcome(observation=observation, value=outcome.value, termination=outcome.termination, info=outcome.info)
return self.current_outcome
def reset(self):
self.current_outcome = EnvironmentOutcome(
observation=GymDomainStateProxy(state=super().reset(), context=[]),
value=None,
termination=False,
info=None,
)
return self.current_outcome.observation
def sample(self, state, action):
if state != self.current_outcome.observation:
self.reset()
for a in state._context:
self.step(a)
outcome = self.step(action)
observation = GymDomainStateProxy(state=outcome.observation._state, context=state._context + [action])
self.current_outcome = EnvironmentOutcome(observation=observation, value=outcome.value, termination=outcome.termination, info=outcome.info)
return self.current_outcome
def get_next_state_distribution(self, state, action):
if state != self.current_outcome.observation:
self.reset()
for a in state._context:
self.step(a)
outcome = self.step(action)
observation = GymDomainStateProxy(state=outcome.observation._state, context=state._context + [action])
self.current_outcome = EnvironmentOutcome(observation=observation, value=outcome.value, termination=outcome.termination, info=outcome.info)
return DiscreteDistribution([(observation, 1.0)])
def step(self, action):
outcome = super().step(action)
observation = GymDomainStateProxy(state=outcome.observation,
context=None)
return EnvironmentOutcome(
observation=observation,
value=outcome.value,
termination=outcome.termination,
info=outcome.info,
)
def _sample(self, memory: D.T_memory[D.T_state], action: D.T_agent[D.T_concurrency[D.T_event]]) -> \
EnvironmentOutcome[D.T_agent[D.T_observation], D.T_agent[TransitionValue[D.T_value]], D.T_agent[D.T_info]]:
o = super()._sample(memory, action)
return EnvironmentOutcome(observation=GymDomainStateProxy(
state=normalize_and_round(o.observation._state),
context=o.observation._context),
value=TransitionValue(reward=o.value.reward -
1),
termination=o.termination,
info=o.info)
def reset(self):
self.current_outcome = EnvironmentOutcome(
observation=GymDomainStateProxy(
state=normalize_and_round(super().reset()),
context=self._gym_env.get_state(),
),
value=None,
termination=False,
info=None,
)
return self.current_outcome.observation
def step(self, action):
outcome = super().step(action)
observation = GymDomainStateProxy(
state=normalize_and_round(outcome.observation),
context=self._gym_env.get_state(),
)
return EnvironmentOutcome(
observation=observation,
value=outcome.value,
termination=outcome.termination,
info=outcome.info,
)
def get_next_state_distribution(self, state, action):
if state._context != self.current_outcome.observation._context:
self._gym_env.set_state(state._context)
outcome = super().step(action)
observation = GymDomainStateProxy(
state=normalize_and_round(outcome.observation),
context=self._gym_env.get_state(),
)
self.current_outcome = EnvironmentOutcome(
observation=observation,
value=outcome.value,
termination=outcome.termination,
info=outcome.info,
)
return DiscreteDistribution([(observation, 1.0)])
def decode(outcome):
return EnvironmentOutcome(
observation=MyShmProxy.StateProxy.decode(outcome[0]),
value=MyShmProxy.TransitionValueProxy.decode(outcome[1:3]),
termination=MyShmProxy.BoolProxy.decode(outcome[3]))
def decode(outcome):
return EnvironmentOutcome(
observation=GridShmProxy.StateProxy.decode(outcome[0]),
value=GridShmProxy.ValueProxy.decode(outcome[1:3]),
termination=GridShmProxy.BoolProxy.decode(outcome[3]),
)
