class Google(State):
basic_info = StateVariable(dtype=PersonBasicInfo,
default_val=None,
help="basic person info from Google")
scholar_info = StateVariable(dtype=ScholarInfo,
default_val=None,
help="scholar info from google scholar")
class ColabPoolEnvStateQueryMessage(State):
name = StateVariable(
dtype=str,
default_val=None,
help="name of the the name of the pool whose state is queried")
message_id = StateVariable(dtype=str,
default_val=None,
help="id used to identify response messages")
def __init__(self):
super().__init__()
self.stateful_srv_clct_ref = DCERef(StatefulServiceCLCT, write=True, read=False)
self.stateless_srv_clct_ref = DCERef(StatelessServiceCLCT, write=True, read=False)
self.service_to_add = StateVariable(value_cls=SrvWithPodName, default_val=None, help="需要添加的service信息")
class GuessNumberAgent(Agent):
"""基于 rule base 做的一个猜数字的agent"""
env = ObjectRef.bind_at_runtime()
action = StateVariable(dtype=int, default_val=0, help="action")
def __init__(self):
super().__init__()
self._upper_bound = 1000 # 这个可以认为是 agent 的内部变量，可以先不用成为 state variable , 更像是 model parameter
self._lower_bound = 0
@state_var_change_handler(state_vars="init_action_space", state_var_source=env)
@pick_one_change
def on_init_action_space(self, state_var_owner_pk: Any, state_var_name: str, state_var_value: Any):
self._upper_bound = state_var_value.upper_bound
self._lower_bound = state_var_value.lower_bound
@state_var_change_handler(state_vars="state", state_var_source=env)
@pick_one_change
def on_env_state_changed(self, state_var_owner_pk: Any, state_var_name: str, state_var_value: Any):
if state_var_value > 0: # 猜的数字大了，调整上边界
self._upper_bound = self.action.VALUE
self.action.VALUE = get_random_int(self._lower_bound, self._upper_bound)
elif state_var_value < 0: # 猜的数字小了，调整下边界
self._lower_bound = self.action.VALUE
self.action.VALUE = get_random_int(self._lower_bound, self._upper_bound)
else:
self.action.VALUE = get_random_int(self._lower_bound, self._upper_bound)
logger.info(f"[Agent] guess in range [{self._lower_bound},{self._upper_bound}]")
@state_var_change_handler(state_vars="reward", state_var_source=env)
def on_env_reward(self, state_var_owner_pk: Any, state_vars: Dict[str, Any],
triggering_state_var_names: List[str]):
# do nothing , 做成 RL Model 时有用
pass
class TestEnv(Env):
_test_var = StateVariable(dtype=int, help="test var")
def __init__(self, _):
super().__init__()
@state_var_change_handler(state_vars=[_test_var, TestState.val])
@pick_one_change
def on_test_var_change(self, state_var_owner_pk: Any, state_var_name: str, test_var: int):
print(f"{state_var_owner_pk} {state_var_name} {test_var}")
@timer(interval=5.0)
async def on_timer(self):
test_val = 'abcde'
test_val_object = create_stateful_object(self.pk, TestState)
test_val_object[TestState.val].VALUE = test_val
return StatefulObjectAndCommitStream(test_val_object, self.state_vars_storage.stream)
@staticmethod
async def run_test():
import faust.types
test_env = TestEnv("abcd")
test_env_status_topic = faust.types.TP(topic="test_env", partition=1)
test_env.bind(topic_define=test_env_status_topic)
await test_env.start()
print("started")
test_env._test_var.VALUE = 1
print("assigned")
await test_env.commit_state_var_changes()
print("published")
class GuessNumberGameEnv(Env):
agent = ObjectRef.bind_at_runtime()
init_action_space = StateVariable(dtype=NumberRange, default_val=NumberRange(1, 100),
help="初始状态，竞猜数字的最大范围")
init_state_space = StateVariable(dtype=NumberRange, default_val=NumberRange(-1, 1),
help="状态空间，-1 ~ 1 的值")
state = StateVariable(dtype=int, default_val=0, help="-1:数字小了,0:随机猜一个，1:数字大了")
reward = StateVariable(dtype=float, default_val=0., help="reward")
def __init__(self):
super().__init__()
self._real_num: int = 0
"""真实的数字，可以不是 state variable，暂假定一个 episode 中 env inst 不会 deactivate"""
def _reset(self) -> int:
"""模拟 gym.Env 的 reset """
state_space = self.init_action_space.VALUE
self._real_num = get_random_int(state_space.lower_bound, state_space.upper_bound)
return 0
@state_var_change_handler(state_vars=[Activatable.active])
def on_active_flag_change(self, state_var_owner_pk: Any, state_vars: Dict[str, Any],
triggering_state_var_names: List[str]):
v = state_vars['active']
if v == 1: # 标记 active 当做是 start new episode，后续可以单独做一个 episode 的 variable
self.init_action_space.VALUE = NumberRange(1, 100)
self.init_state_space.VALUE = NumberRange(-1, 1)
self.state.VALUE = self._reset()
logger.info(f"[Env] Game start... , real number is {self._real_num}")
@state_var_change_handler(state_vars=["action"], state_var_source=agent)
def on_agent_action(self, state_var_owner_pk: Any, state_vars: Dict[str, Any],
triggering_state_var_names: List[str]):
action = state_vars['action']
if action == self._real_num:
self.reward.VALUE = 1000.
logger.info(f"[Env] Bingo!")
elif action > self._real_num:
self.state.VALUE = 1
self.reward.VALUE = -0.01 # 假设每猜错一次，有负的reward
logger.info(f"[Env] Greater than real num . Guess num. : {action} , Real num. {self._real_num}")
elif action < self._real_num:
self.state.VALUE = -1
self.reward.VALUE = -0.01
logger.info(f"[Env] Less than real num . Guess num. : {action} , Real num. {self._real_num}")
class WebPageMonitorEnv(Env):
"""该 Env 模拟监控用户浏览器上的 HTML 内容"""
html = StateVariable(dtype=str, default_val=None, help="网页的html内容")
@state_var_change_handler(state_vars=[Activatable.active])
@pick_one_change
def on_active_flag_change(self, state_var_owner_pk: Any,
state_var_name: str, state_var_value: Any):
if state_var_value == 1: # 标记 active 当做是 start new episode，后续可以单独做一个 episode 的 variable
self.html.VALUE = "dummy"
logger.info(
f"[Env] {self.__class__.__qualname__} started, send out html value {self.html.VALUE}"
)
class HeartBeatMessage(State):
sender = StateVariable(dtype=str,
default_val=None,
help="name of the sender")
time = StateVariable(dtype=str, default_val=None, help="")
class ColabPoolEnvState(Activatable):
name = StateVariable(dtype=str, default_val=None, help="name of the pool")
class ColabPoolEnvStateQueryResponseMessage(State):
name = StateVariable(dtype=str, default_val=None, help="name of the pool")
query_message_id = StateVariable(dtype=str,
default_val=None,
help="id of the query message")
class Yahoo(State):
basic_info = StateVariable(dtype=PersonBasicInfo,
default_val=None,
help="basic person info from Yahoo")
class URLBasedPerson(State):
url = StateVariable(dtype=str,
default_val=None,
help="url identify the author")
class TestState(State):
val = StateVariable(dtype=str, default_val=None, help="val of test state")