add data_analyse_tool function

application/docker-compose.yml

@@ -63,7 +63,7 @@ services:
  # 指定镜像和版本
  image: mysql:8.0
  ports:
- - "127.0.0.1:3306:3306"
+ - "3306:3306"
  restart: always
  environment:
  # 配置root密码

application/pages/1_🌍_Generative_BI_Playground.py

@@ -11,7 +11,7 @@
 from nlq.business.profile import ProfileManagement
 from nlq.business.suggested_question import SuggestedQuestionManagement as sqm
 from nlq.business.vector_store import VectorStore
-from utils.llm import text_to_sql, get_query_intent, generate_suggested_question, get_agent_cot_task, agent_data_analyse, \
+from utils.llm import text_to_sql, get_query_intent, generate_suggested_question, get_agent_cot_task, data_analyse_tool, \
  knowledge_search
 from utils.constant import PROFILE_QUESTION_TABLE_NAME, ACTIVE_PROMPT_NAME, DEFAULT_PROMPT_NAME
 from utils.navigation import make_sidebar
@@ -420,13 +420,13 @@ def main():
  with st.expander(
  f'Query Retrieve : {len(normal_search_result.retrieve_result)}, NER Retrieve : {len(normal_search_result.entity_slot_retrieve)}'):
  examples = {}
+ examples["query_retrieve"] = []
  for example in normal_search_result.retrieve_result:
- examples["query_retrieve"] = []
  examples["query_retrieve"].append({'Score': example['_score'],
  'Question': example['_source']['text'],
  'Answer': example['_source']['sql'].strip()})
+ examples["ner_retrieve"] = []
  for example in normal_search_result.entity_slot_retrieve:
- examples["ner_retrieve"] = []
  examples["ner_retrieve"].append({'Score': example['_score'],
  'Question': example['_source']['entity'],
  'Answer': example['_source']['comment'].strip()})
@@ -475,7 +475,11 @@ def main():
  if search_intent_result["status_code"] == 500:
  with st.expander("The SQL Error Info"):
  st.markdown(search_intent_result["error_info"])
-
+ else:
+ if search_intent_result["data"] is not None and len(search_intent_result["data"]) > 0:
+ search_intent_analyse_result = data_analyse_tool(model_type, search_box,
+ search_intent_result["data"].to_json(orient='records', force_ascii=False), "query")
+ st.markdown(search_intent_analyse_result)
  st.session_state.current_sql_result[selected_profile] = search_intent_result["data"]
 
  elif agent_intent_flag:
@@ -488,9 +492,8 @@ def main():
  orient='records')
  filter_deep_dive_sql_result.append(agent_search_result[i])
 
- agent_data_analyse_result = agent_data_analyse(model_type, search_box,
- json.dumps(
- filter_deep_dive_sql_result))
+ agent_data_analyse_result = data_analyse_tool(model_type, search_box,
+ json.dumps(filter_deep_dive_sql_result, ensure_ascii=False), "agent")
  logger.info("agent_data_analyse_result")
  logger.info(agent_data_analyse_result)
  st.session_state.messages[selected_profile].append(

application/utils/llm.py

@@ -6,7 +6,7 @@
 from utils import opensearch
 from utils.prompt import POSTGRES_DIALECT_PROMPT_CLAUDE3, MYSQL_DIALECT_PROMPT_CLAUDE3, \
  DEFAULT_DIALECT_PROMPT, SEARCH_INTENT_PROMPT_CLAUDE3, CLAUDE3_DATA_ANALYSE_SYSTEM_PROMPT, \
- CLAUDE3_DATA_ANALYSE_USER_PROMPT, AWS_REDSHIFT_DIALECT_PROMPT_CLAUDE3
+ CLAUDE3_AGENT_DATA_ANALYSE_USER_PROMPT, AWS_REDSHIFT_DIALECT_PROMPT_CLAUDE3, CLAUDE3_QUERY_DATA_ANALYSE_USER_PROMPT
 import os
 import logging
 from langchain_core.output_parsers import JsonOutputParser
@@ -339,11 +339,14 @@ def get_agent_cot_task(model_id, search_box, ddl, agent_cot_example=None):
  return default_agent_cot_task
 
 
-def agent_data_analyse(model_id, search_box, sql_data):
+def data_analyse_tool(model_id, search_box, sql_data, search_type):
  try:
  system_prompt = CLAUDE3_DATA_ANALYSE_SYSTEM_PROMPT
  max_tokens = 2048
- user_prompt = CLAUDE3_DATA_ANALYSE_USER_PROMPT.format(question=search_box, data=sql_data)
+ if search_type == "agent":
+ user_prompt = CLAUDE3_AGENT_DATA_ANALYSE_USER_PROMPT.format(question=search_box, data=sql_data)
+ else:
+ user_prompt = CLAUDE3_QUERY_DATA_ANALYSE_USER_PROMPT.format(question=search_box, data=sql_data)
  user_message = {"role": "user", "content": user_prompt}
  messages = [user_message]
  response = invoke_model_claude3(model_id, system_prompt, messages, max_tokens)

application/utils/prompt.py

@@ -150,7 +150,7 @@
 You are a data analysis expert in the retail industry
 """
 
-CLAUDE3_DATA_ANALYSE_USER_PROMPT = """
+CLAUDE3_AGENT_DATA_ANALYSE_USER_PROMPT = """
 As a professional data analyst, you are now asked a question by a user, and you need to analyze the data provided.
 
 <instructions>
@@ -167,3 +167,18 @@
 
 Think step by step。
 """
+
+CLAUDE3_QUERY_DATA_ANALYSE_USER_PROMPT = """
+
+Your task is to analyze the given data and describe it in natural language. 
+
+<instructions>
+- Transforming data into natural language, including all key data as much as possible
+- Just need the final result of the data, no need to output the previous analysis process
+</instructions>
+
+The user question is：{question}
+
+The data is：{data}
+
+"""