pycityagent 1.0.0__tar.gz → 1.1.0__tar.gz

{pycityagent-1.0.0 → pycityagent-1.1.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: pycityagent
-Version: 1.0.0
+Version: 1.1.0
 Summary: LLM-based城市模拟器agent构建库
 Author-email: Yuwei Yan <pinkgranite86@gmail.com>
 License: MIT License
@@ -39,7 +39,7 @@ Requires-Dist: geojson>=3.1.0
 Requires-Dist: numpy>=1.26.3
 Requires-Dist: openai>=1.8.0
 Requires-Dist: Pillow>=10.2.0
-Requires-Dist: pycitysim>=1.8.0
+Requires-Dist: pycitysim>=1.12.2
 Requires-Dist: citystreetview>=1.1.0
 Requires-Dist: PyYAML>=6.0.1
 Requires-Dist: Requests>=2.31.0
@@ -92,10 +92,12 @@ citysim_request:
   map_request:
     mongo_coll: map_beijing_extend_20240205
     cache_dir: ./cache
+  route_request: 
+    server: http://api-opencity-2x.fiblab.net:58082
   streetview_request:
     engine: baidumap / googlemap
-    mapAK: your baidumap AK (if baidumap)
-    proxy: your googlemap proxy (if googlemap, optional)
+    mapAK: baidumap api-key (if you use baidumap engine)
+    proxy: googlemap proxy (if you use googlemap engine)
 
 apphub_request:
   hub_url: https://api-opencity-2x.fiblab.net:58080
@@ -106,31 +108,56 @@ apphub_request:
 - Forget about **citysim_request**, let's focus on the other two.
 
 #### LLM_REQUEST
-- As you can tell, the whole CityAgent is based on the LLM, by now, there are three different parts of config items: **text_request**, **img_understand_request** and **img_generate_request**
+- As you can see, the whole CityAgent is based on the LLM, by now, there are three different parts of config items: **text_request**, **img_understand_request** and **img_generate_request**
 - By now, we support [**qwen**](https://tongyi.aliyun.com/) and [**openai**](https://openai.com/)
     - `Notice: Our environments are basically conducted with qwen. If you prefer to use openai, then you may encounter hardships. AND fell free to issue us.`
 - Get your **api_key** and chooce your **model**s
 
-### CITYSIM_REQUEST
-- There are no need to change the 'simulator' and 'map_request' config
-- 'streetview_request': this is the config used for streetview
-  - By now, we support 'baidumap' and 'googlemap'
-  - If you use 'baidumap', then you need to apply for a mapAK
-  - If you use 'googlemap', then you can config your own proxy by 'proxy' attribute
+#### CITYSIM_REQUEST
+- Most of the configuration options in this part are determined, such as **simulator.server**, **map_request.mongo_coll**, **route_request.server**
+- **map_request.cache_dir**: used for storing map data your machine, you can justify the target dir as you wish (**create the dir first**)
+- **streetview_request**: used for obtaining streetview images, by now, we support baidumap engine and googlemap engine
+  - if you choose baidumap engine, you need to get a baidumap api-key
+    ``` yaml
+    streetview_request:
+      engine: baidumap
+      mapAK: xxxx
+    ```
+  - if you choose googlemap engine, you need to provide your proxy address, for example:
+    ``` yaml
+    streetview_request:
+      engine: googlemap
+      proxy: 
+        http: http://xxxx
+        https: https://xxxx
+    ```
 
 #### APPHUB_REQUEST
-- This is basically used to connect with the backend.
+- Used for creating the connection between backend server and client.
 - Put your **app_id** and **app_secret** here.
+  - Create your account and apply in [Opencity website](https://opencity.fiblab.net/)
 
 ### Installation
-- Install from **pip** easily.
-```shell
-pip install pycityagent
-```
+#### PyPI Installation
+  - Install from **pip** easily.
+  ```shell
+  pip install pycityagent
+  ```
+
+#### Install from source code
+- Clone this repo
+- Install required packages
+  ``` shell
+  pip install -r requirements.txt
+  ```
+- Install **libGL.so.1**, if you ara using Linux with a suitable package manager: (apt for instance)
+  ``` shell
+  apt-get install libgl1
+  ```
 
 ### CODE and RUN
 - Check the **example** folder and copy files from it (`Remember replace the config file`)
-- Look at the Demo:
+- Look at the Demo: (A citizen Agent demo)
 ```python
 import yaml
 from pycityagent.simulator import Simulator
@@ -147,7 +174,7 @@ async def main():
     smi = Simulator(config['citysim_request'])
     
     # get the person by person_id, return agent
-    agent = await smi.GetAgent("name_of_agent", 8)
+    agent = await smi.GetCitizenAgent("name_of_agent", 8)
 
     # Help you build unique agent by scratch/profile
     agent.Image.load_scratch('scratch_template.json')
@@ -158,6 +185,7 @@ async def main():
 
     # Connect to apphub so you can interact with your agent in front end
     agent.ConnectToHub(config['apphub_request'])
+    agent.Bind()
 
     # Creat the soul (a LLM processor actually)
     llmConfig = LLMConfig(config['llm_request'])

{pycityagent-1.0.0 → pycityagent-1.1.0}/README.md

@@ -45,10 +45,12 @@ citysim_request:
   map_request:
     mongo_coll: map_beijing_extend_20240205
     cache_dir: ./cache
+  route_request: 
+    server: http://api-opencity-2x.fiblab.net:58082
   streetview_request:
     engine: baidumap / googlemap
-    mapAK: your baidumap AK (if baidumap)
-    proxy: your googlemap proxy (if googlemap, optional)
+    mapAK: baidumap api-key (if you use baidumap engine)
+    proxy: googlemap proxy (if you use googlemap engine)
 
 apphub_request:
   hub_url: https://api-opencity-2x.fiblab.net:58080
@@ -59,31 +61,56 @@ apphub_request:
 - Forget about **citysim_request**, let's focus on the other two.
 
 #### LLM_REQUEST
-- As you can tell, the whole CityAgent is based on the LLM, by now, there are three different parts of config items: **text_request**, **img_understand_request** and **img_generate_request**
+- As you can see, the whole CityAgent is based on the LLM, by now, there are three different parts of config items: **text_request**, **img_understand_request** and **img_generate_request**
 - By now, we support [**qwen**](https://tongyi.aliyun.com/) and [**openai**](https://openai.com/)
     - `Notice: Our environments are basically conducted with qwen. If you prefer to use openai, then you may encounter hardships. AND fell free to issue us.`
 - Get your **api_key** and chooce your **model**s
 
-### CITYSIM_REQUEST
-- There are no need to change the 'simulator' and 'map_request' config
-- 'streetview_request': this is the config used for streetview
-  - By now, we support 'baidumap' and 'googlemap'
-  - If you use 'baidumap', then you need to apply for a mapAK
-  - If you use 'googlemap', then you can config your own proxy by 'proxy' attribute
+#### CITYSIM_REQUEST
+- Most of the configuration options in this part are determined, such as **simulator.server**, **map_request.mongo_coll**, **route_request.server**
+- **map_request.cache_dir**: used for storing map data your machine, you can justify the target dir as you wish (**create the dir first**)
+- **streetview_request**: used for obtaining streetview images, by now, we support baidumap engine and googlemap engine
+  - if you choose baidumap engine, you need to get a baidumap api-key
+    ``` yaml
+    streetview_request:
+      engine: baidumap
+      mapAK: xxxx
+    ```
+  - if you choose googlemap engine, you need to provide your proxy address, for example:
+    ``` yaml
+    streetview_request:
+      engine: googlemap
+      proxy: 
+        http: http://xxxx
+        https: https://xxxx
+    ```
 
 #### APPHUB_REQUEST
-- This is basically used to connect with the backend.
+- Used for creating the connection between backend server and client.
 - Put your **app_id** and **app_secret** here.
+  - Create your account and apply in [Opencity website](https://opencity.fiblab.net/)
 
 ### Installation
-- Install from **pip** easily.
-```shell
-pip install pycityagent
-```
+#### PyPI Installation
+  - Install from **pip** easily.
+  ```shell
+  pip install pycityagent
+  ```
+
+#### Install from source code
+- Clone this repo
+- Install required packages
+  ``` shell
+  pip install -r requirements.txt
+  ```
+- Install **libGL.so.1**, if you ara using Linux with a suitable package manager: (apt for instance)
+  ``` shell
+  apt-get install libgl1
+  ```
 
 ### CODE and RUN
 - Check the **example** folder and copy files from it (`Remember replace the config file`)
-- Look at the Demo:
+- Look at the Demo: (A citizen Agent demo)
 ```python
 import yaml
 from pycityagent.simulator import Simulator
@@ -100,7 +127,7 @@ async def main():
     smi = Simulator(config['citysim_request'])
     
     # get the person by person_id, return agent
-    agent = await smi.GetAgent("name_of_agent", 8)
+    agent = await smi.GetCitizenAgent("name_of_agent", 8)
 
     # Help you build unique agent by scratch/profile
     agent.Image.load_scratch('scratch_template.json')
@@ -111,6 +138,7 @@ async def main():
 
     # Connect to apphub so you can interact with your agent in front end
     agent.ConnectToHub(config['apphub_request'])
+    agent.Bind()
 
     # Creat the soul (a LLM processor actually)
     llmConfig = LLMConfig(config['llm_request'])

pycityagent-1.1.0/pycityagent/__init__.py

@@ -0,0 +1,6 @@
+from .simulator import Simulator
+from .agent import Agent
+from .agent_citizen import CitizenAgent
+from .agent_func import FuncAgent
+
+__all__ = [Simulator, Agent, CitizenAgent, FuncAgent]

pycityagent-1.1.0/pycityagent/ac/__init__.py

@@ -0,0 +1,8 @@
+'''Action Control - 最终进行simulator以及AppHub对接'''
+
+from .ac import *
+from .action import *
+from .hub_actions import *
+from .sim_actions import *
+
+__all__ = [ActionController, Action, HubAction, SimAction, SendUserMessage, SendStreetview, SendPop, PositionUpdate, ShowPath, ShowPosition, SetSchedule, SendAgentMessage]

pycityagent-1.1.0/pycityagent/ac/ac.py

@@ -0,0 +1,67 @@
+from typing import Any, Optional, Union
+from .action import Action
+from .citizen_actions.trip import TripAction
+from .citizen_actions.shop import ShopAction
+from .citizen_actions.converse import ConverseAction
+from .citizen_actions.controled import ControledAction
+from .citizen_actions.idle import IdleAction
+
+class ActionController:
+    """
+    Agent行为控制器: 与AppHub(simulator)对接
+    Agent Controller: Connect with AppHub and Simulator
+    Note: Actions are predefined. By now, user defined actions are not supported
+    """
+    def __init__(self, agent) -> None:
+        self._agent = agent
+        self.action_chain = {}
+        self.reset_ac()
+        self._init_actions()
+
+    async def Run(self):
+        agent_state = self._agent.state
+        if agent_state in self.action_chain.keys():
+            for action in self.action_chain[agent_state]:
+                await action.Forward()
+
+    def clear_ac(self):
+        """
+        清空AC
+        """
+        self.action_chain.clear()
+
+    def reset_ac(self):
+        """
+        重置Action Controller
+        Reset action controller
+        """
+        self.action_chain.clear()
+        for state in self._agent._st.machine.states.keys():
+            self.action_chain[state] = []
+        
+    def set_ac(self, states:list[str]):
+        """
+        设置ac
+        """
+        self.action_chain.clear()
+        for state in states:
+            self.action_chain[state] = []
+
+    def add_actions(self, state:str, actions:Union[Action, list[Action]]):
+        """
+        添加Action到目标动作链
+        """
+        if issubclass(type(actions), Action) or isinstance(actions, Action):
+            self.action_chain[state].append(actions)
+        else:
+            self.action_chain[state] += actions
+
+    def _init_actions(self):
+        """
+        构建初始化动作链——适用于citizen类型的agent
+        """
+        self.action_chain['trip'] = [TripAction(self._agent)]
+        self.action_chain['shop'] = [ShopAction(self._agent)]
+        self.action_chain['conve'] = [ConverseAction(self._agent)]
+        self.action_chain['controled'] = [ControledAction(self._agent)]
+        self.action_chain['idle'] = [IdleAction(self._agent)]

pycityagent-1.1.0/pycityagent/ac/action.py

@@ -0,0 +1,55 @@
+from abc import ABC, abstractclassmethod
+from typing import Callable, Any
+
+class ActionType:
+    """
+    行动类型枚举 所有行动本质上为数据推送
+    Action Type enumeration, all actions are essentially data push
+
+    Types:
+    - Sim = 1, 用于表示与模拟器对接的行动
+    - Hub = 2, 用于表示与AppHub(前端)对接的行动
+    - Comp = 3, 表示综合类型 (可能同时包含与Sim以及Hub的交互)
+    """
+    Sim = 1
+    Hub = 2
+    Comp = 3
+class Action:
+    def __init__(self, agent, type:ActionType, source: str = None, before:Callable[[list], Any] = None) -> None:
+        '''
+        默认初始化
+        
+        Args:
+        - agent (Agent): the related agent
+        - type (ActionType)
+        - source (str): 数据来源, 默认为None, 如果为None则会从接收用户传入的数据作为Forward函数参数, 否则从WM.Reason数据缓存中取对应数据作为参数
+        - before (function): 数据处理方法, 用于当Reason缓存中的参数与标准格式不符时使用
+        '''
+        self._agent = agent
+        self._type = type
+        self._source = source
+        self._before = before
+
+    def get_source(self):
+        """
+        获取source数据
+        """
+        if self._source != None:
+            source = self._agent.Brain.Memory.Working.Reason[self._source]
+            if self._before != None:
+                source = self._before(source)
+            return source
+        else:
+            return None
+
+    @abstractclassmethod
+    async def Forward(self):
+        '''接口函数'''
+
+class SimAction(Action):
+    def __init__(self, agent, source: str = None, before: Callable[[list], Any] = None) -> None:
+        super().__init__(agent, ActionType.Sim, source, before)
+
+class HubAction(Action):
+    def __init__(self, agent, source: str = None, before: Callable[[list], Any] = None) -> None:
+        super().__init__(agent, ActionType.Hub, source, before)

pycityagent-1.1.0/pycityagent/ac/action_stream.py

@@ -0,0 +1,25 @@
+from abc import ABC, abstractclassmethod
+from typing import Any, Optional
+from action import Action
+
+class ActionStream:
+    """
+    可以将多个Action聚合在一起构成一个ActionStream, Stream中的Action会顺序执行
+    You can collect multiple Actions into a ActionStream, actions in stream will be activated in sequence
+    注意: 该部分内容目前没有使用
+    """
+    def __init__(self, actions:Optional[list[Action]]=[]) -> None:
+        self.actions = actions
+
+    def add_actions(self, actions:list[Action]):
+        """
+        添加Action至当前stream
+
+        Args:
+        - actions (list(Action)): a list of action
+        """
+        self.actions += actions
+
+    def __call__(self) -> Any:
+        for action in self.actions:
+            action()

pycityagent-1.1.0/pycityagent/ac/citizen_actions/controled.py

@@ -0,0 +1,16 @@
+import time
+from typing import Callable, Any
+
+from pycityagent.ac.action import ActionType
+from ..action import Action
+from pycitysim.apphub import AgentMessage
+
+class ControledAction(Action):
+    '''Converse行为控制器'''
+    def __init__(self, agent, sources: list[str] = None, before: Callable[[list], Any] = None) -> None:
+        super().__init__(agent, ActionType.Comp, sources, before)
+
+    async def Forward(self):
+        req = {'person_id': self._agent._id, 'schedules': []}
+        await self._agent._client.person_service.SetSchedule(req)
+        self._agent.Hub.Update([AgentMessage(self._agent.Hub._agent_id, int(time.time()*1000), '我已理解您的意思，正在修改我的行程', None, None)])

pycityagent-1.1.0/pycityagent/ac/citizen_actions/converse.py

@@ -0,0 +1,34 @@
+import time
+from typing import Callable, Any
+
+from pycityagent.ac.action import ActionType
+from ..action import Action
+from pycitysim.apphub import AgentMessage
+
+class ConverseAction(Action):
+    '''Converse行为控制器'''
+    def __init__(self, agent, sources: list[str] = None, before: Callable[[list], Any] = None) -> None:
+        super().__init__(agent, ActionType.Comp, sources, before)
+
+    async def Forward(self):
+        target_agent_ids = self._agent.Brain.Memory.Working.Reason['agent_message_handle_resp'][0]
+        if len(target_agent_ids) == 0:
+            return
+        messages = self._agent.Brain.Memory.Working.Reason['agent_message_handle_resp'][1]
+        req = {'messages': []}
+        if len(target_agent_ids) != len(messages):
+            print("Warning: the number of target agent and message are not aligned, only sends matched messages")
+        rng = min(len(target_agent_ids), len(messages))
+        for i in range(rng):
+            dic = {}
+            dic['from'] = self._agent._id
+            dic['to'] = target_agent_ids[i]
+            dic['message'] = messages[i]
+            req['messages'].append(dic)
+        # * 发送至模拟器
+        await self._agent._client.social_service.Send(req=req)
+
+        # * 发送至AppHub
+        if self._agent.Hub != None and messages[0] != 'End':
+            # * 将信息中的第一条不同至pop
+            self._agent.Hub.Update(pop=messages[0])

pycityagent-1.1.0/pycityagent/ac/citizen_actions/idle.py

@@ -0,0 +1,20 @@
+import time
+from typing import Callable, Any
+
+from pycityagent.ac.action import ActionType
+from ..action import Action
+from pycitysim.apphub import AgentMessage
+
+class IdleAction(Action):
+    '''idle行为控制器'''
+    def __init__(self, agent, sources: list[str] = None, before: Callable[[list], Any] = None) -> None:
+        super().__init__(agent, ActionType.Comp, sources, before)
+
+    async def Forward(self):
+        if len(self._agent.base['schedules']) > 0:
+            req = {'person_id': self._agent._id, 'schedules': []}
+            await self._agent._client.person_service.SetSchedule(req)
+        if self._agent.Hub != None:
+            self._agent.Hub.Update()
+
+    

pycityagent-1.1.0/pycityagent/ac/citizen_actions/shop.py

@@ -0,0 +1,82 @@
+from typing import Callable, Any
+from pycityagent.ac.action import ActionType
+from ..action import Action
+from pycitysim.apphub import AgentMessage
+import time
+
+def encap_msg(msg, role='user', **kwargs):
+    dialog = {'role': role, 'content': msg}
+    dialog.update(kwargs)
+    return dialog
+
+class ShopAction(Action):
+    '''Shop行为控制器'''
+    def __init__(self, agent, sources: list[str] = None, before: Callable[[list], Any] = None) -> None:
+        super().__init__(agent, ActionType.Comp, sources, before)
+
+    async def Forward(self):
+        # * 与模拟器对接 - 暂时没有
+        # * 与AppHub对接
+        profile = self._agent.Image.get_profile()
+        self.consumption(profile)
+
+    def consumption(self, profile, mall_info):
+        dialogs = []
+        system_prompt = f'''
+                            你是一个在北京工作和生活的人。 
+                            {profile} 
+                            现在是2024年1月。
+                            你需要为下一周的基本生活购买必需品。
+                        '''
+        dialogs.append(encap_msg(system_prompt, 'system'))
+        actions_format = ['''{{'商品': 购买的商品的列表,'购买量': 每个商品的购买量，一个列表, '解释': '这种购买方式的原因'}}''']
+        actions_candidates = ['''【食品】
+                                米：10元/公斤
+                                面粉：7.5元/公斤
+                                新鲜蔬菜（如菠菜）：7元/500克
+                                水果（如苹果）：15元/公斤
+                                猪肉：30元/公斤
+                                鸡肉：20元/公斤
+                                鸡蛋：1.5元/个
+                                牛奶：10元/升''',
+                                '''【清洁用品】
+                                    洗衣液：30元/瓶
+                                    洗洁精：20元/瓶
+                                    垃圾袋：0.3元/个''',
+                                '''【个人护理用品】
+                                    牙膏：10元/支
+                                    洗发水：30元/瓶
+                                    沐浴露：35元/瓶
+                                    面巾纸：5元/包''',
+                                '''【其他】
+                                    矿泉水：1.7元/瓶
+                                    面包：8元/个
+                                    辣条：3元/包''']
+
+        user_prompt = f'''
+                        首先确定你的消费预算。以如下格式回答，不要有冗余的文本！
+                        {{'消费预算': 一个代表购买必需品需要消耗的钱的数量的数字}}
+                        '''
+        dialogs.append(encap_msg(user_prompt))
+        # * 对接一：确定消费预算
+        self._agent.Hub.Update([AgentMessage(self._agent.Hub._agent_id, int(time.time()*1000), '我正在确定消费预算......', None, None)])
+        msg = self._agent._soul.text_request(dialogs)
+        self._agent.Hub.Update([AgentMessage(self._agent.Hub._agent_id, int(time.time()*1000), f'我的消费预算是: {msg}'), None, None])
+
+        dialogs.append(encap_msg(msg, 'assistant'))
+
+        # * 对接二：购物选择
+        for cand in actions_candidates:
+            user_prompt = f'''
+                            购物中心里有
+                            {cand}
+                            你要买哪些商品，以如下格式回答，不要有冗余的文本！
+                            {actions_format[0]}
+                        '''
+            self._agent.Hub.Update([AgentMessage(self._agent.Hub._agent_id, int(time.time()*1000), f'我看到了\n {cand}', None, None)])
+            dialogs.append(encap_msg(user_prompt))
+            msg = self._agent._soul.text_request(dialogs)
+            self._agent.Hub.Update([AgentMessage(self._agent.Hub._agent_id, int(time.time()*1000), f'我的购买选择是: {msg}', None, None)])
+            dialogs.append(encap_msg(msg, 'assistant'))
+        
+        self._agent.Hub.Update([AgentMessage(self._agent.Hub._agent_id, int(time.time()*1000), '购物完成', None, None)])

pycityagent-1.1.0/pycityagent/ac/citizen_actions/trip.py

@@ -0,0 +1,41 @@
+import time
+from typing import Callable, Any
+
+from pycityagent.ac.action import ActionType
+from ..action import Action
+from pycitysim.apphub import AgentMessage
+
+class TripAction(Action):
+    '''Trip行为控制器'''
+    def __init__(self, agent, sources: list[str] = None, before: Callable[[list], Any] = None) -> None:
+        super().__init__(agent, ActionType.Comp, sources, before)
+
+    async def Forward(self):
+        now = self._agent.Scheduler.now
+        if now.is_set:
+            # TODO: 目前仅支持传输一张图片至前端
+            '''之前已经将schedule同步至模拟器了'''
+            if self._agent.Hub != None:
+                self._agent.Hub.Update(streetview=self._agent.Brain.Sence.sence_buffer['streetview'][0])
+        else:
+            '''同步schedule至模拟器'''
+            self._agent.Scheduler.now.is_set = True
+            departure_time = now.time
+            mode = now.mode
+            aoi_id = now.target_id_aoi
+            poi_id = now.target_id_poi
+            end = {'aoi_position': {'aoi_id': aoi_id, 'poi_id': poi_id}}
+            activity = now.description
+            trips = [{'mode': mode, 'end': end, 'departure_time': departure_time, 'activity': activity}]
+            set_schedule = [{'trips': trips, 'loop_count': 1, 'departure_time': departure_time}]
+
+            # * 与模拟器对接
+            req = {'person_id': self._agent._id, 'schedules': set_schedule}
+            await self._agent._client.person_service.SetSchedule(req)
+
+            # * 与AppHub对接
+            if self._agent.Hub != None:
+                messages = [AgentMessage(self._agent.Hub._agent_id, int(time.time()*1000), f'已到达出发时间, {activity}', None, None)]
+                self._agent.Hub.Update(messages)
+
+