Browse Source

text-to-sql for '104_112碳排放公開及建準資料' table

ling 4 months ago
parent
commit
87a0cdfc88
1 changed files with 41 additions and 34 deletions
  1. 41 34
      text_to_sql2.py

+ 41 - 34
text_to_sql2.py

@@ -75,15 +75,6 @@ llm = ChatOllama(model=local_llm, temperature=0)
 # llm = Ollama(model = "llama3-groq-tool-use:latest", num_gpu=1)
 def get_examples():
     examples = [
-        {
-            "input": "建準去年的固定燃燒總排放量是多少?",
-            "query": """SELECT SUM("排放量(公噸CO2e)") AS "固定燃燒總排放量"
-                        FROM "104_112碳排放公開及建準資料"
-                        WHERE "事業名稱" like '%建準%'
-                        AND "排放源" = '固定燃燒'
-                        AND "盤查標準" = 'GHG'
-                        AND "年度" = EXTRACT(YEAR FROM CURRENT_DATE)-1;""",
-        },
         {
             "input": "建準廣興廠去年的類別1總排放量是多少?",
             "query": """SELECT SUM("排放量(公噸CO2e)") AS "類別1總排放量"
@@ -104,6 +95,15 @@ def get_examples():
                         AND "盤查標準" = 'GHG'
                         AND "年度" = 2022;""",
         },
+        {
+            "input": "建準去年的固定燃燒總排放量是多少?",
+            "query": """SELECT SUM("排放量(公噸CO2e)") AS "固定燃燒總排放量"
+                        FROM "104_112碳排放公開及建準資料"
+                        WHERE "事業名稱" like '%建準%'
+                        AND "排放源" = '固定燃燒'
+                        AND "盤查標準" = 'GHG'
+                        AND "年度" = EXTRACT(YEAR FROM CURRENT_DATE)-1;""",
+        },
         {
             "input": "台積電2022年的類別1總排放量是多少?",
             "query": """SELECT SUM("排放量(公噸CO2e)") AS "台積電2022年類別1總排放量"
@@ -120,10 +120,12 @@ def get_examples():
 
 def table_description():
     database_description = (
-        "The database consists of following tables: `2022 清冊數據(ISO)`, `2023 清冊數據(ISO)`, `2022 清冊數據(GHG)`, `2023 清冊數據(GHG)`, `水電使用量(ISO)` and `水電使用量(GHG)`. "
+        "The database consists of following table: `104_112碳排放公開及建準資料`, `水電使用量(ISO)` and `水電使用量(GHG)`. "
         "This is a PostgreSQL database, so you need to use postgres-related queries.\n\n"
-        "The `2022 清冊數據(ISO)`, `2023 清冊數據(ISO)`, `2022 清冊數據(GHG)` and `2023 清冊數據(GHG)` table 描述了不同廠房分別在 ISO 14064-1:2018 與 GHG Protocol 標準下的溫室氣體排放量,並依類別1至類別6劃分。"
+        "The `104_112碳排放公開及建準資料` table 描述了不同事業單位或廠房分別在 ISO 14064-1:2018 與 GHG Protocol 標準下的溫室氣體排放量,並依類別1至類別6劃分。"
         "It includes the following columns:\n"
+
+        "- `年度`: 盤查年度\n"
         "- `類別`: 溫室氣體的排放類別,包含以下:\n"
         "   \t*類別1-直接排放\n"
         "   \t*類別2-能源間接排放\n"
@@ -132,15 +134,9 @@ def table_description():
         "   \t*類別5-使用來自組織產品間接排放\n"
         "   \t*類別6\n"
         "- `排放源`: `類別`欄位進一步劃分的細項\n"
-        "- `高雄總部&運通廠`: 位於台灣的廠房據點\n"
-        "- `台北辦公室`: 位於台灣的廠房據點\n"
-        "- `北海建準廠`: 位於中國的廠房據點\n"
-        "- `北海立準廠`: 位於中國的廠房據點\n"
-        "- `昆山廣興廠`: 位於中國的廠房據點\n"
-        "- `菲律賓建準廠`: 位於菲律賓的廠房據點\n"
-        "- `India`: 位於印度的廠房據點\n"
-        "- `INC`: 位於美國的廠房據點\n"
-        "- `SAS`: 位於法國的廠房據點\n\n"
+        "- `排放量(公噸CO2e)`: 溫室氣體排放量\n"
+        "- `盤查標準`: ISO or GHG\n"
+        
 
         "The `水電使用量(ISO)` and `水電使用量(GHG)` table 描述了不同廠房分別在 ISO 14064-1:2018 與 GHG Protocol 標準下的水電使用量,包含'外購電力 度數 (kwh)'與'自來水 度數 (立方公尺 m³)'。"
         "The `public.departments_table` table contains information about the various departments in the company. It includes:\n"
@@ -196,7 +192,7 @@ def write_query_chain(db, llm):
         input_variables=["input", "top_k", "table_info"],
     )
 
-    # llm = Ollama(model = "mannix/defog-llama3-sqlcoder-8b", num_gpu=1)
+    # llm = Ollama(model = "sqlcoder", num_gpu=1)
     # llm = HuggingFacePipeline(pipeline=pipe)
     
     
@@ -214,19 +210,9 @@ def sql_to_nl_chain(llm):
         <|begin_of_text|>
         <|begin_of_text|><|start_header_id|>system<|end_header_id|>
         Given the following user question, corresponding SQL query, and SQL result, answer the user question.
-        給定以下使用者問題、對應的 SQL 查詢和 SQL 結果,以繁體中文回答使用者問題。
+        根據使用者的問題、對應的 SQL 查詢和 SQL 結果,以繁體中文回答使用者問題。
 
-        For example
-        Question: 建準廣興廠去年的類別總排放量是多少?
-        SQL Query: SELECT SUM("排放量(公噸CO2e)") AS "類別1總排放量"
-                        FROM "104_112碳排放公開及建準資料"
-                        WHERE "事業名稱" like '%建準%'
-                        AND "事業名稱" like '%廣興廠%'
-                        AND "類別" = '類別1-直接排放'
-                        AND "盤查標準" = 'GHG'
-                        AND "年度" = EXTRACT(YEAR FROM CURRENT_DATE)-1;
-        SQL Result: [(1102.3712,)]
-        Answer: 建準廣興廠去年的類別1總排放量是1102.3712
+        
         <|eot_id|>
 
         <|begin_of_text|><|start_header_id|>user<|end_header_id|>
@@ -241,14 +227,35 @@ def sql_to_nl_chain(llm):
         """
         )
 
+    # llm = Ollama(model = "llama3-groq-tool-use:latest", num_gpu=1)
     chain = answer_prompt | llm | StrOutputParser()
 
     return chain
 
+def get_query(db, question, selected_table, llm):
+    write_query = write_query_chain(db, llm)
+    query = write_query.invoke({"question": question, 'table_names_to_use': selected_table, "top_k": 1000, "table_info":context["table_info"], "database_description": table_description()})
+    
+    query = re.split('SQL query: ', query)[-1]
+    query = query.replace("104_112碰排放公開及建準資料","104_112碳排放公開及建準資料")
+    print(query)
+    
+    return query
+
+def query_to_nl(db, question, query, llm):
+    execute_query = QuerySQLDataBaseTool(db=db)
+    result = execute_query.invoke(query)
+    print(result)
+
+    chain = sql_to_nl_chain(llm)
+    answer = chain.invoke({"question": question, "query": query, "result": result})
+
+    return answer
+
 def run(db, question, selected_table, llm):
 
     write_query = write_query_chain(db, llm)
-    query = write_query.invoke({"question": question, 'table_names_to_use': selected_table, "top_k": 1000, "table_info":context["table_info"], "database_description": "foo"})
+    query = write_query.invoke({"question": question, 'table_names_to_use': selected_table, "top_k": 1000, "table_info":context["table_info"], "database_description": table_description()})
     
     query = re.split('SQL query: ', query)[-1]
     query = query.replace("104_112碰排放公開及建準資料","104_112碳排放公開及建準資料")