Merge pull request #1528 from microsoft/xuyang1/refine_task_and_implement_workflow_task_as_example

Xuyang1/refine task and implement workflow task as example

qlib/finco/cli.py

@@ -1,13 +1,15 @@
 import fire
 from qlib.finco.task import WorkflowManager
 from dotenv import load_dotenv
+from qlib import auto_init
 
 
-def main(prompt):
+def main(prompt=None):
     load_dotenv(verbose=True, override=True)
     wm = WorkflowManager()
     wm.run(prompt)
 
 
 if __name__ == "__main__":
+    auto_init()
     fire.Fire(main)

qlib/finco/conf.py

@@ -13,7 +13,7 @@ class Config():
     def __init__(self):
         self.use_azure = os.getenv("USE_AZURE") == "True"
         self.temperature = 0.5 if os.getenv("TEMPERATURE") is None else float(os.getenv("TEMPERATURE"))
-        self.max_tokens = 8000 if os.getenv("MAX_TOKENS") is None else int(os.getenv("MAX_TOKENS"))
+        self.max_tokens = 800 if os.getenv("MAX_TOKENS") is None else int(os.getenv("MAX_TOKENS"))
         
         self.openai_api_key = os.getenv("OPENAI_API_KEY")
         self.use_azure = os.getenv("USE_AZURE") == "True"
@@ -21,4 +21,6 @@ class Config():
         self.azure_api_version = os.getenv("AZURE_API_VERSION")
         self.model = os.getenv("MODEL") or ("gpt-35-turbo" if self.use_azure else "gpt-3.5-turbo")
 
-        self.max_retry = os.getenv("MAX_RETRY")
+        self.max_retry = os.getenv("MAX_RETRY")
+
+        self.continous_mode = os.getenv("CONTINOUS_MODE") == "True" if os.getenv("CONTINOUS_MODE") is not None else False

qlib/finco/llm.py

@@ -1,3 +1,4 @@
+import time
 import openai
 from typing import Optional
 from qlib.finco.conf import Config
@@ -25,6 +26,7 @@ def try_create_chat_completion(max_retry=10, **kwargs):
         except openai.error.RateLimitError as e:
             print(e)
             print(f"Retrying {i+1}th time...")
+            time.sleep(1)
             continue
     raise Exception(f"Failed to create chat completion after {max_retry} retries.")
 
@@ -56,7 +58,8 @@ def create_chat_completion(
             model=cfg.model,
             messages=messages,
         )
-    return response
+    resp = response.choices[0].message["content"]
+    return resp
 
 if __name__ == "__main__":
     create_chat_completion()

qlib/finco/task.py

@@ -2,12 +2,22 @@ import os
 
 from pathlib import Path
 from typing import Any, List
+from qlib.log import get_module_logger
 from qlib.typehint import Literal
+from qlib.finco.conf import Config
+from qlib.finco.llm import try_create_chat_completion
+from qlib.finco.utils import parse_json
+from jinja2 import Template
+
+import abc
+import copy
+import logging
 
 
-class Task:
+class Task():
     """
     The user's intention, which was initially represented by a prompt, is achieved through a sequence of tasks.
+    This class doesn't have to be abstract, but it is abstract in the sense that it is not supposed to be instantiated directly because it doesn't have any implementation.
 
     Some thoughts:
     - Do we have to split create a new concept of Action besides Task?
@@ -21,34 +31,131 @@ class Task:
         - Edit Task: it is supposed to edit the code base directly.
     """
 
-    def __init__(self, context=None) -> None:
+    ## all subclass should implement this method to determine task type
-        pass
+    @abc.abstractclassmethod
-
+    def __init__(self) -> None:
+        self._context_manager = None
+        self.executed = False
+    
     def summarize(self) -> str:
         """After the execution of the task, it is supposed to generated some context about the execution"""
-        return ""
+        raise NotImplementedError
 
-    def update_context(self, latest_context):
+    """assign the workflow context manager to the task"""
+    """then all tasks can use this context manager to share the same context"""
+    def assign_context_manager(self, context_manager):
         ...
+        self._context_manager = context_manager
 
-    def execution(self) -> Any:
+    def execution(self, **kwargs) -> Any:
         """The execution results of the task"""
-        pass
+        raise NotImplementedError
 
+    def interact(self) -> Any:
+        """The user can interact with the task"""
+        """All sub classes should implement the interact method to determine the next task"""
+        """In continous mode, this method will not be called and the next task will be determined by the execution method only"""
+        raise NotImplementedError("The interact method is not implemented, but workflow not in continous mode")
+
+class WorkflowTask(Task):
+    """This task is supposed to be the first task of the workflow"""
+    def __init__(self,) -> None:
+        super().__init__()
+        self.__DEFAULT_WORKFLOW_SYSTEM_PROMPT = """
+        Your task is to determine the workflow in Qlib (supervised learning or reinforcemtn learning) ensureing the workflow can meet the user's requirements.
+
+        The user will provide the requirements, you will provide only the output the choice in exact format specified below with no explanation or conversation.
+
+        Example input 1:
+        Help me build a build a low turnover quant investment strategy that focus more on long turn return in China a stock market.
+
+        Example output 1:
+        workflow: supervised learning
+
+        Example input 2:
+        Help me build a build a pipeline to determine the best selling point of a stock in a day or half a day in USA stock market.
+
+        Example output 2:
+        workflow: reinforcemtn learning
+        """
+
+        self.__DEFAULT_WORKFLOW_USER_PROMPT = (
+            "User input: '{{user_prompt}}'\n"
+            "Please provide the workflow in Qlib (supervised learning or reinforcemtn learning) ensureing the workflow can meet the user's requirements.\n"
+            "Response only with the output in the exact format specified in the system prompt, with no explanation or conversation.\n"
+        )
+        self.__DEFAULT_USER_PROMPT = "Please help me build a low turnover strategy that focus more on longterm return in China a stock market."
+        self.logger = get_module_logger("fincoWorkflowTask", level=logging.INFO)
+
+    """make the choice which main workflow (RL, SL) will be used"""
+    def execute(self,) -> List[Task]:
+        user_prompt = self._context_manager.get_context("user_prompt")
+        user_prompt = user_prompt if user_prompt is not None else self.__DEFAULT_USER_PROMPT
+        system_prompt = self.__DEFAULT_WORKFLOW_SYSTEM_PROMPT
+        prompt_workflow_selection = Template(
+            self.__DEFAULT_WORKFLOW_USER_PROMPT
+        ).render(user_prompt=user_prompt)
+        messages = [
+        {
+            "role": "system",
+            "content": system_prompt,
+        },
+        {
+            "role": "user",
+            "content": prompt_workflow_selection,
+        },
+    ]
+        response = try_create_chat_completion(messages=messages)
+        workflow = response.split(":")[1].strip().lower()
+        self.executed = True
+        self._context_manager.set_context("workflow", workflow)
+        if workflow == "supervised learning":
+            return [SLTask()]
+        elif workflow == "reinforcement learning":
+            return [RLTask()]
+        else:
+            raise ValueError(f"The workflow: {workflow} is not supported")
+    
+    def interact(self) -> Any:
+        assert self.executed == True, "The workflow task has not been executed yet"
+        ## TODO use logger
+        self.logger.info(
+            f"The workflow has been determined to be ---{self._context_manager.get_context('workflow')}---"
+        )
+        self.logger.info(
+            "Enter 'y' to authorise command,'s' to run self-feedback commands, "
+            "'n' to exit program, or enter feedback for WorkflowTask"
+        )
+        try:
+            answer = input("You answer is:")
+        except KeyboardInterrupt:
+            self.logger.info("User has exited the program")
+            exit()
+        if answer.lower().strip() == "y":
+            return
+        else:
+            # TODO add self feedback
+            raise ValueError("The input cannot be interpreted as a valid input")
+        
 
 class PlanTask(Task):
-    def execute(self) -> List[Task]:
+    def execute(self, prompt) -> List[Task]:
         return []
 
-
-class WorkflowTask(PlanTask):
-    """make the choice which main workflow (RL, SL) will be used"""
-
-    def execute(self):
-        ...
-
-
 class SLTask(PlanTask):
+    def __init__(self,) -> None:
+        super().__init__()
+
+    def exeute(self):
+        """
+        return a list of interested tasks
+        Copy the template project maybe a part of the task
+        """
+        return []
+    
+class RLTask(PlanTask):
+    def __init__(self,) -> None:
+        super().__init__()
     def exeute(self):
         """
         return a list of interested tasks
@@ -60,6 +167,29 @@ class SLTask(PlanTask):
 class ActionTask(Task):
     def execute(self) -> Literal["fail", "success"]:
         return "success"
+    
+"""Context Manager stores the context of the workflow"""
+"""All context are key value pairs which saves the input, output and status of the whole workflow"""
+class WorkflowContextManager():
+    def __init__(self) -> None:
+        self.context = {}
+        self.logger = get_module_logger("fincoWorkflowContextManager")
+
+    def set_context(self, key, value):
+        if key in self.context:
+            self.logger.warning("The key already exists in the context, the value will be overwritten")
+        self.context[key] = value
+    
+    def get_context(self, key):
+        if key not in self.context:
+            self.logger.warning("The key doesn't exist in the context")
+            return None
+        return self.context[key]
+    
+    """return a deep copy of the context"""
+    """TODO: do we need to return a deep copy?"""
+    def get_all_context(self):
+        return copy.deepcopy(self.context)
 
 
 class SummarizeTask(Task):
@@ -95,13 +225,14 @@ class WorkflowManager:
             self._output_path = Path.cwd() / name
         else:
             self._output_path = Path(output_path)
-        self._context = []
+        self._context = WorkflowContextManager()
 
-    def add_context(self, task_res):
+    """Direct call set_context method of the context manager"""
-        self._context.append(task_res)
+    def set_context(self, key, value):
+        self._context.set_context(key, value)
 
-    def get_context(self):
+    def get_context(self) -> WorkflowContextManager:
-        """TODO: context manger?"""
+        return self._context
 
     def run(self, prompt: str) -> Path:
         """
@@ -127,16 +258,23 @@ class WorkflowManager:
         # - The generated tasks can't be changed after geting new information from the execution retuls.
         #   - But it is required in some cases, if we want to build a external dataset, it maybe have to plan like autogpt...
 
+        cfg = Config()
+
         # NOTE: list may not be enough for general task list
-        task_list = [WorkflowTask(prompt)]
+        self.set_context("user_prompt", prompt)
+        task_list = [WorkflowTask()]
         while len(task_list):
-            # task_list.ap
+            """task list is not long, so sort it is not a big problem"""
+            """TODO: sort the task list based on the priority of the task"""
+            # task_list = sorted(task_list, key=lambda x: x.task_type)
             t = task_list.pop(0)
-            t.update_context(self.get_context())
+            t.assign_context_manager(self._context)
             res = t.execute()
-            if isinstance(t, PlanTask):
+            if not cfg.continous_mode:
+                res = t.interact()
+            if isinstance(t.task_type, WorkflowTask) or isinstance(t.task_type, PlanTask):
                 task_list.extend(res)
-            elif isinstance(t, ActionTask):
+            elif isinstance(t.task_type, ActionTask):
                 if res != "success":
                     ...
                     # TODO: handle the unexpected execution Error

qlib/finco/utils.py

@@ -0,0 +1,9 @@
+import json
+
+def parse_json(response):
+    try:
+        return json.loads(response)
+    except json.decoder.JSONDecodeError:
+        pass
+
+    raise Exception(f"Failed to parse response: {response}, please report it or help us to fix it.")