Merge pull request #368 from zhouzinimg/ga_game

ga_game add reflection part

examples/st_game/reflect/__init__.py

@@ -0,0 +1,12 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# @Desc   : reflection module
+
+from metagpt.reflect import agent_reflect
+from metagpt.reflect import ga_prompt_generator
+__all__ = [
+    "agent_reflect",
+    "LongTermMemory",
+    "ga_po"
+    "ga_prompt_generator"
+]

examples/st_game/reflect/ga_prompt_generator.py

@@ -0,0 +1,84 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# author: didi
+# Date:9.25
+
+import openai
+from metagpt.llm import DEFAULT_LLM
+# 直接调用Prompt生成
+# ga的prompt构建格式和metagpt完全不同。没有办法融合。
+
+
+# 特殊指令加入Prompt生成
+
+
+async def final_response(prompt, special_instruction, example_output=None):
+    """
+    通过将特殊指令加入Prompt生成最终的响应。
+
+    参数：
+    - prompt：要生成响应的提示文本。
+    - special_instruction：要加入Prompt的特殊指令。
+    - example_output（可选）：示例输出的JSON字符串。
+
+    返回：
+    生成的最终响应。
+
+    """
+    prompt = '"""\n' + prompt + '\n"""\n'
+    prompt += f"Output the response to the prompt above in json. {special_instruction}\n"
+    if example_output:
+        prompt += "Example output json:\n"
+        prompt += '{"output": "' + str(example_output) + '"}'
+    return await DEFAULT_LLM.aask(prompt)
+
+# prompt填充模板
+
+
+def prompt_generate(curr_input, prompt_lib_file):
+    """
+    Takes in the current input (e.g. comment that you want to classifiy) and 
+    the path to a prompt file. The prompt file contains the raw str prompt that
+    will be used, which contains the following substr: !<INPUT>! -- this 
+    function replaces this substr with the actual curr_input to produce the 
+    final promopt that will be sent to the GPT3 server. 
+    ARGS:
+    curr_input: the input we want to feed in (IF THERE ARE MORE THAN ONE
+                INPUT, THIS CAN BE A LIST.)
+    prompt_lib_file: the path to the promopt file. 
+    RETURNS: 
+    a str prompt that will be sent to OpenAI's GPT server.  
+    """
+    if type(curr_input) is type("string"):
+        curr_input = [curr_input]
+        curr_input = [str(i) for i in curr_input]
+
+    f = open(prompt_lib_file, "r")
+    prompt = f.read()
+    f.close()
+    for count, i in enumerate(curr_input):
+        prompt = prompt.replace(f"!<INPUT {count}>!", i)
+    if "<commentblockmarker>###</commentblockmarker>" in prompt:
+        prompt = prompt.split(
+            "<commentblockmarker>###</commentblockmarker>")[1]
+    return prompt.strip()
+
+# 使用OpenAI embedding库进行存储
+
+
+def embedding(query):
+    """
+    Generates an embedding for the given query.
+
+    Args:
+        query (str): The text query to be embedded.
+
+    Returns:
+        str: The embedding key generated for the query.
+    """
+    embedding_result = openai.Embedding.create(
+        model="text-embedding-ada-002",
+        input=query
+    )
+    embedding_key = embedding_result['data'][0]["embedding"]
+    return embedding_key

examples/st_game/reflect/st_reflect.py

@@ -1,3 +1,97 @@
 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 # @Desc   : st's reflection execution
+
+import asyncio
+import json
+from metagpt.logs import logger
+import time
+from ga_prompt_generator import final_response
+'''
+等待Agent和memory更新，保留相关引用但可以忽略。
+'''
+from ..memory.associative_memory import MemoryBasic
+
+import json
+import time
+
+
+async def agent_reflect(memories_list):
+    """
+    代理反思函数：生成关注点并生成洞察和证据
+
+    """
+    A = await generate_focus_point(memories_list)
+
+    for i in A:
+        B = await generate_insights_and_evidence(memories_list, question=i)
+
+
+async def generate_focus_point(memories_list: list[MemoryBasic], n=3):
+    """
+    生成关注点函数：根据记忆列表生成关注点
+    """
+    wait_sorted_mem = [[i.accessed_time, i] for i in memories_list]
+    sorted_memories = sorted(wait_sorted_mem, key=lambda x: x[0])
+    memorys = [i for created, i in sorted_memories]
+    statements = ''
+    for i in memorys:
+        statements += i.description + "\n"
+    prompt = '''
+    {statements}
+    Given only the information above, what are {num_question} most salient high-level questions we can answer about the subjects grounded in the statements?
+    '''
+    example_output = '["What should Jane do for lunch", "Does Jane like strawberry", "Who is Jane"]'
+    out = await final_response(prompt.format(statements=statements, num_question=n),
+                               "Output must be a list of str.", example_output)
+    try:
+        poi_dict = json.loads(out)
+        return poi_dict['output']
+    except ValueError:
+        print(out)
+        logger.error('无法返回正常结果')
+        return out
+
+
+async def generate_insights_and_evidence(memories_list: list[MemoryBasic], question: str, n=5):
+    """
+    生成洞察和证据函数：根据问题生成洞察和证据
+    """
+    memories_list = await agent_retrieve(agent, question, 50, 10)
+    statements = ""
+    for count, mem in enumerate(memories_list):
+        statements += f'{str(count)}. {mem.description}\n'
+    prompt = '''
+    Input:
+    {statements}
+
+    What {n} high-level insights can you infer from the above statements?
+    You should return a list of list[str,list]. The first element is the insight you have found. The second element is the 
+    '''
+
+    ret = final_response(prompt.format(
+        question=question, statements=statements, n=n), "['insightA',[1,2,3]]")
+    try:
+        insight_list = json.loads(ret)
+        for insight, index in insight_list:
+            agent.memory_list.append(MemoryBasic(
+                time.time(), None, insight, None, None))
+        return insight_list
+    except:
+        logger.error('我们无法获得想要的返回。')
+        return ret
+
+
+""" if __name__ == "__main__":
+    # 例子，构建John Agent，实现retrive
+    John_iss = "John Lin is a pharmacy shopkeeper at the Willow Market and Pharmacy who loves to help people. He is always looking for ways to make the process of getting medication easier for his customers; John Lin is living with his wife, Mei Lin, who is a college professor, and son, Eddy Lin, who is a student studying music theory; John Lin loves his family very much; John Lin has known the old couple next-door, Sam Moore and Jennifer Moore, for a few years; John Lin thinks Sam Moore is a kind and nice man; John Lin knows his neighbor, Yuriko Yamamoto, well; John Lin knows of his neighbors, Tamara Taylor and Carmen Ortiz, but has not met them before; John Lin and Tom Moreno are colleagues at The Willows Market and Pharmacy; John Lin and Tom Moreno are friends and like to discuss local politics together; John Lin knows the Moreno family somewhat well — the husband Tom Moreno and the wife Jane Moreno."
+    John = AgentMemory(
+        "John", John_iss, memory_path="agent_memories/John_memory.json")
+
+    # John的相关信息：{'Had a friendly chat with Yuriko about her garden.': 2.4992317730827667, 'Helped Mrs. Moore carry groceries into her house.': 1.957656720441911, 'Discussed local politics with Tom Moreno.': 1.9458268038234035}
+    asyncio.run(agent_reflect(John))
+    '''
+    这里是输出,list形式，返回给记忆。
+    [['The pharmacy is a friendly and helpful community.', [0, 2, 9, 12]], ['The pharmacy is a place where people come for more than just medication.', [3, 5, 13, 14]], ['The pharmacy is a place where people come for advice and conversation.', [0, 2, 6, 9, 12]], ['The pharmacy is a place where people come for assistance with daily tasks.', [3, 5, 13, 14]], ['The pharmacy is a place where people come for political discussions.', [1]]]
+    '''
+ """

examples/st_game/roles/st_role.py

@@ -27,6 +27,8 @@ from ..memory.retrieve import agent_retrieve
 from ..memory.scratch import Scratch
 from ..utils.utils import get_embedding, generate_poig_score
 
+from ..reflect.st_reflect import agent_reflect
+
 
 class STRoleContext(RoleContext):
     env: 'MazeEnvironment' = Field(default=None)

metagpt/plan/plan.py

@@ -1,3 +0,0 @@
-#!/usr/bin/env python
-# -*- coding: utf-8 -*-
-# @Desc   : base class of planning

metagpt/reflect/__init__.py

@@ -1,3 +0,0 @@
-#!/usr/bin/env python
-# -*- coding: utf-8 -*-
-# @Desc   : reflection module

metagpt/reflect/reflect.py

@@ -1,3 +0,0 @@
-#!/usr/bin/env python
-# -*- coding: utf-8 -*-
-# @Desc   : base class of reflection

metagpt/roles/role.py

@@ -18,6 +18,7 @@ from metagpt.llm import LLM
 from metagpt.logs import logger
 from metagpt.memory import Memory, LongTermMemory
 from metagpt.schema import Message
+from metagpt.reflect import agent_reflect
 
 PREFIX_TEMPLATE = """You are a {profile}, named {name}, your goal is {goal}, and the constraint is {constraints}. """
 
@@ -78,7 +79,8 @@ class RoleContext(BaseModel):
     def check(self, role_id: str):
         if hasattr(CONFIG, "long_term_memory") and CONFIG.long_term_memory:
             self.long_term_memory.recover_memory(role_id, self)
-            self.memory = self.long_term_memory  # use memory to act as long_term_memory for unify operation
+            # use memory to act as long_term_memory for unify operation
+            self.memory = self.long_term_memory
 
     @property
     def important_memory(self) -> list[Message]:
@@ -95,7 +97,8 @@ class Role:
 
     def __init__(self, name="", profile="", goal="", constraints="", desc=""):
         self._llm = LLM()
-        self._setting = RoleSetting(name=name, profile=profile, goal=goal, constraints=constraints, desc=desc)
+        self._setting = RoleSetting(
+            name=name, profile=profile, goal=goal, constraints=constraints, desc=desc)
         self._states = []
         self._actions = []
         self._role_id = str(self._setting)
@@ -169,9 +172,10 @@ class Role:
         # logger.info(response)
         if isinstance(response, ActionOutput):
             msg = Message(content=response.content, instruct_content=response.instruct_content,
-                        role=self.profile, cause_by=type(self._rc.todo))
+                          role=self.profile, cause_by=type(self._rc.todo))
         else:
-            msg = Message(content=response, role=self.profile, cause_by=type(self._rc.todo))
+            msg = Message(content=response, role=self.profile,
+                          cause_by=type(self._rc.todo))
         self._rc.memory.add(msg)
         # logger.debug(f"{response}")
 
@@ -184,8 +188,9 @@ class Role:
         env_msgs = self._rc.env.memory.get()
 
         observed = self._rc.env.memory.get_by_actions(self._rc.watch)
-        
-        self._rc.news = self._rc.memory.remember(observed)  # remember recent exact or similar memories
+
+        # remember recent exact or similar memories
+        self._rc.news = self._rc.memory.remember(observed)
 
         for i in env_msgs:
             self.recv(i)
@@ -205,7 +210,8 @@ class Role:
     async def _react(self) -> Message:
         """Think first, then act"""
         await self._think()
-        logger.debug(f"{self._setting}: {self._rc.state=}, will do {self._rc.todo}")
+        logger.debug(
+            f"{self._setting}: {self._rc.state=}, will do {self._rc.todo}")
         return await self._act()
 
     def recv(self, message: Message) -> None: