def reset(self) -> dm_env.TimeStep:
"""Resets the episode."""
self._reset_next_step = False
self.dones = {agent: False for agent in self.possible_agents}
self._agents = self._possible_agents[:]
self._prev_timestep: rl_environment.TimeStep = None
self._current_player_id = 0
opnspl_tmstep = self._environment.reset()
agent = self.current_agent
done = self.dones[agent]
observe = self._to_observation(opnspl_tmstep)
observation = self._convert_observation(agent, observe, done)
self._discount = convert_np_type(self.discount_spec()[agent].dtype, 1)
reward = convert_np_type(
self.reward_spec()[agent].dtype,
(opnspl_tmstep.rewards[self.current_player_id]
if opnspl_tmstep.rewards else 0),
)
return parameterized_restart(reward, self._discount, observation)
def reset(self) -> dm_env.TimeStep:
"""Resets the episode."""
self._reset_next_step = False
self._step_type = dm_env.StepType.FIRST
discount_spec = self.discount_spec()
observe = self._environment.reset()
self._discounts = {
agent: convert_np_type(discount_spec[agent].dtype, 1)
for agent in self.possible_agents
}
if type(observe) == tuple:
observe, env_extras = observe
else:
env_extras = {}
observations = self._convert_observations(
observe, {agent: False
for agent in self.possible_agents})
rewards_spec = self.reward_spec()
rewards = {
agent: convert_np_type(rewards_spec[agent].dtype, 0)
for agent in self.possible_agents
}
return parameterized_restart(rewards, self._discounts,
observations), env_extras
def reset(self) -> Tuple[dm_env.TimeStep, np.array]:
"""Resets the episode."""
self._reset_next_step = False
self._step_type = dm_env.StepType.FIRST
discount_spec = self.discount_spec()
self._discounts = {
agent: convert_np_type(discount_spec[agent].dtype, 1)
for agent in self._environment.possible_agents
}
observe, state_infos = self._environment.reset()
observations = self._convert_observations(
observe, {agent: False for agent in self.possible_agents}
)
rewards_spec = self.reward_spec()
rewards = {
agent: convert_np_type(rewards_spec[agent].dtype, 0)
for agent in self.possible_agents
}
discount_spec = self.discount_spec()
self._discounts = {
agent: convert_np_type(discount_spec[agent].dtype, 1)
for agent in self.possible_agents
}
return (
parameterized_restart(rewards, self._discounts, observations),
state_infos,
)
def reset(self) -> dm_env.TimeStep:
observation = self._generate_fake_observation()
discount = convert_np_type("float32", 1) # Not used in pettingzoo
reward = convert_np_type("float32", 0)
self._step = 1
return parameterized_restart(reward=reward,
discount=discount,
observation=observation)
def reset(self) -> Tuple[dm_env.TimeStep, np.array]:
"""Resets the env and returns observations from ready agents.
Returns:
obs (dict): New observations for each ready agent.
"""
self._env_done = False
self._reset_next_step = False
self._step_type = dm_env.StepType.FIRST
# reset internal SC2 env
obs_list, state = self._environment.reset()
# Convert observations
observe: Dict[str, np.ndarray] = {}
for i, obs in enumerate(obs_list):
observe[f"agent_{i}"] = {
"observation":
obs,
"action_mask":
np.array(self._environment.get_avail_agent_actions(i),
dtype=np.float32),
}
observations = self._convert_observations(
observe, {agent: False
for agent in self._possible_agents})
self._agents = list(observe.keys())
# create discount spec
discount_spec = self.discount_spec()
self._discounts = {
agent: convert_np_type(discount_spec[agent].dtype, 1)
for agent in self._possible_agents
}
# create rewards spec
rewards_spec = self.reward_spec()
rewards = {
agent: convert_np_type(rewards_spec[agent].dtype, 0)
for agent in self._possible_agents
}
# dm_env timestep
timestep = parameterized_restart(rewards, self._discounts,
observations)
return timestep, {"s_t": state}
def reset(self) -> dm_env.TimeStep:
observations = {}
for agent in self.agents:
observation = self._generate_fake_observation()
observations[agent] = observation
rewards = {
agent: convert_np_type("float32", 0)
for agent in self.agents
}
discounts = {
agent: convert_np_type("float32", 1)
for agent in self.agents
}
self._step = 1
return parameterized_restart(rewards, discounts,
observations) # type: ignore
def reset(self) -> dm_env.TimeStep:
"""Resets the episode."""
self._reset_next_step = False
self._environment.reset()
self._step_types = {
agent: dm_env.StepType.FIRST
for agent in self.possible_agents
}
self._first_step_performed = {
agent: False
for agent in self.possible_agents
}
observe, _, done, _ = self._environment.last()
agent = self.current_agent
observation = self._convert_observation(agent, observe, done)
self._discount = convert_np_type(self.discount_spec()[agent].dtype, 1)
reward = convert_np_type(self.reward_spec()[agent].dtype, 0)
return parameterized_restart(reward, self._discount, observation)
def reset(self) -> dm_env.TimeStep:
"""Resets the episode."""
self._reset_next_step = False
self._agents = self.possible_agents[:]
self._discounts = {
agent: np.dtype("float32").type(1.0)
for agent in self.agents
}
observe, info = self._environment.reset()
observations = self._create_observations(observe, info,
self._environment.dones)
rewards_spec = self.reward_spec()
rewards = {
agent: convert_np_type(rewards_spec[agent].dtype, 0)
for agent in self.possible_agents
}
discount_spec = self.discount_spec()
self._discounts = {
agent: convert_np_type(discount_spec[agent].dtype, 1)
for agent in self.possible_agents
}
return parameterized_restart(rewards, self._discounts, observations)