模型反馈修正

graph_chat/gen_data_report_agent.py

@@ -35,6 +35,7 @@ def _run_sql(state: DateReportAgentState) -> dict:
     sql = state['sql']
     retry_sql = state.get('retry_sql', '')
     sql_correct = state.get('sql_correct', False)
+    logger.info(f'in _run_sql state={state}')
     logger.info(f"user:{state.get('user_id', '1')} 进入 _run_sql 节点")
     try:
         if retry_sql and not sql_correct:
@@ -63,8 +64,9 @@ def _feedback_qa(state: DateReportAgentState) -> dict:
         feedback_temp = template['template']['result_feedback']
         logger.info(f"feedback_temp is {feedback_temp}")
         ddl_list = vn.get_related_ddl(user_question)
+        qa_list = vn.get_similar_question_sql(user_question)
         sys_promot = feedback_temp['system'].format(question=user_question, sql=sql, sql_result=sql_result,
-                                                    current_time=datetime.now(),ddl_list=ddl_list)
+                                                    current_time=datetime.now(),ddl_list=ddl_list,qa_list=qa_list)
         logger.info(f"system_temp is {sys_promot}")
         result = gen_history_llm.invoke(sys_promot).text()
         logger.info(f"feedback result: {result}")
@@ -72,12 +74,9 @@ def _feedback_qa(state: DateReportAgentState) -> dict:
         result = orjson.loads(result)
         logger.info(f"提取json成功")
         logger.info(f"result is {result} type:{type(result)}")
-        print("result is {0}".format(result.keys()))
         if not result["is_result_correct"] and (result["suggested_sql"]):
             logger.info("开始替换")
             logger.info(f"suggested_sql is ".format(result["suggested_sql"]))
-            # state["sql"] = result["suggested_sql"]
-            # state["retry_sql"] = True
             logger.info(f"current state: {state}")
             return {"retry_sql": result["suggested_sql"]}
         if result["is_result_correct"]:
@@ -88,11 +87,13 @@ def _feedback_qa(state: DateReportAgentState) -> dict:
 
 
 def run_sql_hande(state: DateReportAgentState) -> str:
+    logger.info(f"user:{state.get('user_id', '1')} 进入 _run_sql_hande 节点")
+    logger.info(f'in _run_sql_handle state={state}')
     # sql_error = state.get('run_sql_error', '')
     data = state.get('data', {})
     sql_correct = state.get('sql_correct', False)
     sql_retry_count = state.get('retry_count', 0)
-    logger.info("sql_retry_count is {0}".format(sql_retry_count))
+    logger.info(f"sql_retry_count is {sql_retry_count} sql_correct is {sql_correct}")
 
     if sql_retry_count < 2:
         if sql_correct:
@@ -142,9 +143,10 @@ workflowx.add_node("_gen_report", _gen_report)
 workflowx.add_edge(START, "_run_sql")
 workflowx.add_edge("_feedback_qa", "_run_sql")
 workflowx.add_edge("_gen_report", END)
+
 workflowx.add_conditional_edges("_run_sql", run_sql_hande, [END, '_gen_report','_feedback_qa'])
 memory = MemorySaver()
 result_report_agent = workflowx.compile(checkpointer=memory)
-png_data=result_report_agent .get_graph().draw_mermaid_png()
-with open("D://graph2.png", "wb") as f:
-    f.write(png_data)
+# png_data=result_report_agent .get_graph().draw_mermaid_png()
+# with open("D://graph2.png", "wb") as f:
+#     f.write(png_data)

graph_chat/gen_sql_chart_agent.py

@@ -179,6 +179,6 @@ workflow.add_conditional_edges('_gen_sql', gen_sql_handler, ['_gen_sql', '_gen_c
 workflow.add_conditional_edges('_gen_chart', gen_chart_handler, ['_gen_chart', END])
 memory = MemorySaver()
 sql_chart_agent = workflow.compile(checkpointer=memory)
-png_data=sql_chart_agent .get_graph().draw_mermaid_png()
+# png_data=sql_chart_agent .get_graph().draw_mermaid_png()
 # with open("D://graph.png", "wb") as f:
 #     f.write(png_data)

main_service.py

@@ -300,7 +300,7 @@ def gen_graph_question():
     try:
         user_id = request.args.get("user_id")
         cvs_id = request.args.get("cvs_id")
-        config = {"configurable": {"thread_id": cvs_id}}
+        config = {"configurable": {"thread_id": uuid.uuid4()}}
         question = flask.request.args.get("question")
         question_context = get_latest_question(cvs_id, user_id,limit_count=2)
         history = []
@@ -345,6 +345,7 @@ def run_sql_3():
     qa = get_qa_by_id(id)
     sql = qa["sql"]
     question = qa["question"]
+    logger.info(f"in main sql {sql} question {question}")
     logger.info("Start to run sql in main")
     try:
         user_id = request.args.get("user_id")
@@ -361,8 +362,10 @@ def run_sql_3():
             "sql": sql,
             "question": question,
             "retry_count": 0,
+            "need_feedback": False,
+
         }
-        config = {"configurable": {"thread_id": 'dsds'}}
+        config = {"configurable": {"thread_id": uuid.uuid4()}}
         rr = result_report_agent.invoke(initial_state,config)
         logger.info(f"rr.data is {rr.get('data', {})}")
         return jsonify(

template.yaml

@@ -636,11 +636,13 @@ template:
       [执行结果]: <sql_result>{sql_result}</sql_result>。
       [当前时间]: <current_time>{current_time}</current_time>
       [表结构信息]:<schema>{ddl_list}</schema>
+      [问答参考]:<question_answer>{qa_list}</question_answer>
       # 核心规则
       请根据以上信息，进行全面、细致的反思，并最终判断这个结果是否能正确回答原始问题。你的反思需要包含以下几个步骤：
       1.  **核对问题理解**：
           回顾“原始问题”，确认其核心意图、时间范围、筛选条件和所需字段。
           我生成的SQL是否准确捕捉了问题的所有关键要素？有没有遗漏或误解？
+          当sql和问题吻合，且能查询出结果时，一般判定为正确
       2.  **通用业务定义**
           上周:指完整的**上周一到上周日**
       3.  **审查SQL逻辑**：
@@ -650,7 +652,6 @@ template:
       4.  **评估结果合理性**：
           观察“执行结果”，思考这个结果是否符合业务常识或数据的基本特征（例如，数量级、正负值、范围是否合理）？
           结果的列名和内容是否与问题期望的输出一致？
-          如果结果为空或数据量异常，推测可能的原因。
       5.  **最终判断和建议**：
             **结论** : 用一句话明确指出：“结果正确”、“结果可能不正确”或“无法完全确定”。
             **原因** : 简练地解释你做出此判断的核心理由。