WIP Set up demo vanna

.gitignore

@@ -11,3 +11,6 @@ notebooks/
 
 data/
 sandbox/
+
+climateqa/talk_to_data/database/
+*.db

app.py

@@ -140,6 +140,9 @@ def cqa_tab(tab_name):
                                     "<h2>There are no graphs to be displayed at the moment. Try asking another question.</h2>",
                                     elem_id="graphs-container"
                                 )
+                            with gr.Tab("Vanna", elem_id="tab-vanna", id=6) as tab_vanna:
+                                vanna_display = gr.DataFrame([], elem_id="vanna-display")
+                                
     return {
         "chatbot": chatbot,
         "textbox": textbox,
@@ -160,7 +163,8 @@ def cqa_tab(tab_name):
         "tab_figures": tab_figures,
         "tab_graphs": tab_graphs,
         "tab_papers": tab_papers,
-        "graph_container": graphs_container
+        "graph_container": graphs_container,
+        "vanna_display": vanna_display
     }
                                 
  
@@ -190,6 +194,7 @@ def event_handling(
     tab_graphs = main_tab_components["tab_graphs"]
     tab_papers = main_tab_components["tab_papers"]
     graphs_container = main_tab_components["graph_container"]
+    vanna_display = main_tab_components["vanna_display"]
     
     config_open = config_components["config_open"]
     config_modal = config_components["config_modal"]
@@ -204,7 +209,7 @@ def event_handling(
     close_config_modal = config_components["close_config_modal_button"]
     
     new_sources_hmtl = gr.State([])
-    
+    ttd_data = gr.State([])
 
      
     for button in [config_button, close_config_modal]:
@@ -216,13 +221,13 @@ def event_handling(
     # Event for textbox
     (textbox
         .submit(start_chat, [textbox, chatbot, search_only], [textbox, tabs, chatbot, sources_raw], queue=False, api_name=f"start_chat_{textbox.elem_id}")
-        .then(chat, [textbox, chatbot, dropdown_audience, dropdown_sources, dropdown_reports, dropdown_external_sources, search_only], [chatbot, new_sources_hmtl, output_query, output_language, new_figures, current_graphs], concurrency_limit=8, api_name=f"chat_{textbox.elem_id}")
+        .then(chat, [textbox, chatbot, dropdown_audience, dropdown_sources, dropdown_reports, dropdown_external_sources, search_only], [chatbot, new_sources_hmtl, output_query, output_language, new_figures, current_graphs, ttd_data], concurrency_limit=8, api_name=f"chat_{textbox.elem_id}")
         .then(finish_chat, None, [textbox], api_name=f"finish_chat_{textbox.elem_id}")
     )
     # Event for examples_hidden
     (examples_hidden
         .change(start_chat, [examples_hidden, chatbot, search_only], [examples_hidden, tabs, chatbot, sources_raw], queue=False, api_name=f"start_chat_{examples_hidden.elem_id}")
-        .then(chat, [examples_hidden, chatbot, dropdown_audience, dropdown_sources, dropdown_reports, dropdown_external_sources, search_only], [chatbot, new_sources_hmtl, output_query, output_language, new_figures, current_graphs], concurrency_limit=8, api_name=f"chat_{examples_hidden.elem_id}")
+        .then(chat, [examples_hidden, chatbot, dropdown_audience, dropdown_sources, dropdown_reports, dropdown_external_sources, search_only], [chatbot, new_sources_hmtl, output_query, output_language, new_figures, current_graphs, ttd_data], concurrency_limit=8, api_name=f"chat_{examples_hidden.elem_id}")
         .then(finish_chat, None, [examples_hidden], api_name=f"finish_chat_{examples_hidden.elem_id}")
     )
     
@@ -239,6 +244,8 @@ def event_handling(
     for component in [textbox, examples_hidden]:
         component.submit(find_papers, [component, after, dropdown_external_sources], [papers_html, citations_network, papers_summary])
 
+    ttd_data.change(lambda x: x["df_output"], inputs=[ttd_data], outputs=[vanna_display])
+
 def main_ui():
     # config_open = gr.State(True)
     with gr.Blocks(title="Climate Q&A", css_paths=os.getcwd()+ "/style.css", theme=theme, elem_id="main-component") as demo: 

climateqa/chat.py

@@ -53,6 +53,12 @@ def log_interaction_to_azure(history, output_query, sources, docs, share_client,
         print(f"Error logging on Azure Blob Storage: {e}")
         error_msg = f"ClimateQ&A Error: {str(e)[:100]} - The error has been noted, try another question and if the error remains, you can contact us :)"
         raise gr.Error(error_msg)
+
+def handle_numerical_data(event):
+    if event["name"] == "retrieve_drias_data" and event["event"] == "on_chain_end":
+        numerical_data = event["data"]["output"]["drias_data"]
+        sql_query = event["data"]["output"]["drias_sql_query"]
+    return numerical_data, sql_query
     
 # Main chat function
 async def chat_stream(
@@ -120,6 +126,7 @@ async def chat_stream(
     graphs_html = ""    
     used_documents = []
     answer_message_content = ""
+    vanna_data = {}
 
     # Define processing steps
     steps_display = {
@@ -141,6 +148,14 @@ async def chat_stream(
                     history, used_documents = handle_retrieved_documents(
                         event, history, used_documents
                     )
+                # Handle document retrieval
+                if event["event"] == "on_chain_end" and event["name"] in ["retrieve_documents","retrieve_local_data"] and event["data"]["output"] != None:
+                    df_output_vanna, sql_query = handle_numerical_data(
+                        event
+                    )
+                    vanna_data = {"df_output": df_output_vanna, "sql_query": sql_query}
+                 
+                    
                 if event["event"] == "on_chain_end" and event["name"] == "answer_search" :
                     docs = event["data"]["input"]["documents"]
                     docs_html = convert_to_docs_to_html(docs)                    
@@ -183,7 +198,7 @@ async def chat_stream(
                         sub_questions = [q["question"] for q in event["data"]["output"]["questions_list"]]
                         history[-1].content += "Decompose question into sub-questions:\n\n - " + "\n - ".join(sub_questions)
 
-            yield history, docs_html, output_query, output_language, related_contents, graphs_html 
+            yield history, docs_html, output_query, output_language, related_contents, graphs_html, vanna_data
 
     except Exception as e:
         print(f"Event {event} has failed")
@@ -194,4 +209,4 @@ async def chat_stream(
     # Call the function to log interaction
     log_interaction_to_azure(history, output_query, sources, docs, share_client, user_id)
 
-    yield history, docs_html, output_query, output_language, related_contents, graphs_html 
+    yield history, docs_html, output_query, output_language, related_contents, graphs_html, vanna_data

climateqa/engine/chains/drias_retriever.py

@@ -0,0 +1,16 @@
+import sys
+import os
+from climateqa.talk_to_data.main import ask_vanna
+
+
+def make_drias_retriever_node(llm):
+    
+    def retrieve_drias_data(state):
+        print("---- Retrieving data from DRIAS ----")
+        query = state["query"]
+        sql_query, df = ask_vanna(query)
+        state["drias_data"] = df
+        state["drias_sql_query"] = sql_query
+        return state
+    
+    return retrieve_drias_data

climateqa/engine/graph.py

@@ -11,7 +11,7 @@ from typing import List, Dict
 
 import operator
 from typing import Annotated
-
+import pandas as pd
 from IPython.display import display, HTML, Image
 
 from .chains.answer_chitchat import make_chitchat_node
@@ -23,6 +23,7 @@ from .chains.retrieve_documents import make_IPx_retriever_node, make_POC_retriev
 from .chains.answer_rag import make_rag_node
 from .chains.graph_retriever import make_graph_retriever_node
 from .chains.chitchat_categorization import make_chitchat_intent_categorization_node
+from .chains.drias_retriever import make_drias_retriever_node
 # from .chains.set_defaults import set_defaults
 
 class GraphState(TypedDict):
@@ -49,6 +50,8 @@ class GraphState(TypedDict):
     recommended_content : List[Document]
     search_only : bool = False
     reports : List[str] = []
+    drias_data: pd.DataFrame
+    drias_sql_query : str
 
 def dummy(state):
     return 
@@ -164,6 +167,7 @@ def make_graph_agent(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_regi
     answer_rag = make_rag_node(llm, with_docs=True)
     answer_rag_no_docs = make_rag_node(llm, with_docs=False)
     chitchat_categorize_intent = make_chitchat_intent_categorization_node(llm)
+    retrieve_drias_data = make_drias_retriever_node(llm)
 
     # Define the nodes
     # workflow.add_node("set_defaults", set_defaults)
@@ -180,6 +184,7 @@ def make_graph_agent(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_regi
     workflow.add_node("retrieve_documents", retrieve_documents)
     workflow.add_node("answer_rag", answer_rag)
     workflow.add_node("answer_rag_no_docs", answer_rag_no_docs)
+    workflow.add_node("retrieve_drias_data", retrieve_drias_data)
 
     # Entry point
     workflow.set_entry_point("categorize_intent")
@@ -235,7 +240,7 @@ def make_graph_agent(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_regi
 
     # Define the edges
     workflow.add_edge("translate_query", "transform_query")
-    workflow.add_edge("transform_query", "retrieve_documents") #TODO put back
+    # workflow.add_edge("transform_query", "retrieve_documents") #TODO put back
     # workflow.add_edge("transform_query", END) # TODO remove
 
     workflow.add_edge("retrieve_graphs", END)
@@ -245,6 +250,10 @@ def make_graph_agent(llm, vectorstore_ipcc, vectorstore_graphs, vectorstore_regi
     workflow.add_edge("retrieve_graphs_chitchat", END)
     # workflow.add_edge("retrieve_local_data", "answer_search")
 
+    workflow.add_edge("transform_query", "retrieve_drias_data")
+    workflow.add_edge("retrieve_drias_data", END)
+
+
     # Compile
     app = workflow.compile()
     return app

climateqa/talk_to_data/deprecated_vanna_remote.py

@@ -0,0 +1,167 @@
+# from vanna.remote import VannaDefault
+# from pinecone import Pinecone
+# from climateqa.engine.embeddings import get_embeddings_function
+# import pandas as pd
+# import hashlib
+
+# class MyCustomVectorDB(VannaDefault):
+    
+#     """
+#     VectorDB class for storing and retrieving vectors from Pinecone.
+    
+#     args : 
+#         config (dict) : Configuration dictionary containing the Pinecone API key and the index name :
+#             - pc_api_key (str) : Pinecone API key
+#             - index_name (str) : Pinecone index name
+#             - top_k (int) : Number of top results to return (default = 2)
+            
+#     """
+    
+#     def __init__(self,config, **kwargs):
+#         super().__init__(**kwargs)
+#         try : 
+#             self.api_key = config.get('pc_api_key')
+#             self.index_name = config.get('index_name')
+#         except : 
+#             raise Exception("Please provide the Pinecone API key and the index name")
+        
+#         self.pc = Pinecone(api_key = self.api_key)
+#         self.index = self.pc.Index(self.index_name)
+#         self.top_k = config.get('top_k', 2)
+#         self.embeddings = get_embeddings_function()
+        
+        
+#     def check_embedding(self, id, namespace):
+#         fetched = self.index.fetch(ids = [id], namespace = namespace)
+#         if fetched['vectors'] == {}: 
+#             return False
+#         return True
+    
+#     def generate_hash_id(self, data: str) -> str:
+#         """
+#         Generate a unique hash ID for the given data.
+        
+#         Args:
+#             data (str): The input data to hash (e.g., a concatenated string of user attributes).
+        
+#         Returns:
+#             str: A unique hash ID as a hexadecimal string.
+#         """
+        
+#         data_bytes = data.encode('utf-8')
+#         hash_object = hashlib.sha256(data_bytes)
+#         hash_id = hash_object.hexdigest()
+        
+#         return hash_id
+    
+#     def add_ddl(self, ddl: str, **kwargs) -> str:
+#         id = self.generate_hash_id(ddl) + '_ddl'
+        
+#         if self.check_embedding(id, 'ddl'):
+#             print(f"DDL having id {id} already exists")
+#             return id
+        
+#         self.index.upsert(
+#             vectors = [(id, self.embeddings.embed_query(ddl), {'ddl': ddl})],
+#             namespace = 'ddl'
+#         )
+        
+#         return id
+
+#     def add_documentation(self, doc: str, **kwargs) -> str:
+#         id = self.generate_hash_id(doc) + '_doc'
+        
+#         if self.check_embedding(id, 'documentation'):
+#             print(f"Documentation having id {id} already exists")
+#             return id
+        
+#         self.index.upsert(
+#             vectors = [(id, self.embeddings.embed_query(doc), {'doc': doc})],
+#             namespace = 'documentation'
+#         )
+        
+#         return id
+
+#     def add_question_sql(self, question: str, sql: str, **kwargs) -> str:
+#         id = self.generate_hash_id(question) + '_sql'
+        
+#         if self.check_embedding(id, 'question_sql'):
+#             print(f"Question-SQL pair having id {id} already exists")
+#             return id
+        
+#         self.index.upsert(
+#             vectors = [(id, self.embeddings.embed_query(question + sql), {'question': question, 'sql': sql})],
+#             namespace = 'question_sql'
+#         )
+        
+#         return id
+
+#     def get_related_ddl(self, question: str, **kwargs) -> list:
+#         res = self.index.query(
+#             vector=self.embeddings.embed_query(question),
+#             top_k=self.top_k,
+#             namespace='ddl',
+#             include_metadata=True
+#         )
+        
+#         print([match['metadata']['ddl'] for match in res['matches']])
+        
+#         return [match['metadata']['ddl'] for match in res['matches']]
+
+#     def get_related_documentation(self, question: str, **kwargs) -> list:
+#         res = self.index.query(
+#             vector=self.embeddings.embed_query(question),
+#             top_k=self.top_k,
+#             namespace='documentation',
+#             include_metadata=True
+#         )
+        
+#         return [match['metadata']['doc'] for match in res['matches']]
+
+#     def get_similar_quetion_sql(self, question: str, **kwargs) -> list:
+#         res = self.index.query(
+#             vector=self.embeddings.embed_query(question),
+#             top_k=self.top_k,
+#             namespace='question_sql',
+#             include_metadata=True
+#         )
+        
+#         return [(match['metadata']['question'], match['metadata']['sql']) for match in res['matches']]
+
+#     def get_training_data(self, **kwargs) -> pd.DataFrame:
+        
+#         list_of_data = []
+        
+#         namespaces = ['ddl', 'documentation', 'question_sql']
+        
+#         for namespace in namespaces:
+            
+#             data = self.index.query(
+#             top_k=10000,
+#             namespace=namespace,
+#             include_metadata=True,
+#             include_values=False
+#             )
+            
+#             for match in data['matches']:
+#                 list_of_data.append(match['metadata'])
+                
+#         return pd.DataFrame(list_of_data)
+            
+
+
+#     def remove_training_data(self, id: str, **kwargs) -> bool:
+#         if id.endswith("_ddl"):
+#             self.Index.delete(ids=[id], namespace="_ddl")
+#             return True
+#         if id.endswith("_sql"):
+#             self.index.delete(ids=[id], namespace="_sql")
+#             return True
+        
+#         if id.endswith("_doc"):
+#             self.Index.delete(ids=[id], namespace="_doc")
+#             return True
+        
+#         return False
+
+

climateqa/talk_to_data/how_to_use_main.ipynb

@@ -0,0 +1,137 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Import the function in main.py"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "The autoreload extension is already loaded. To reload it, use:\n",
+      "  %reload_ext autoreload\n",
+      "Loading embeddings model:  BAAI/bge-base-en-v1.5\n"
+     ]
+    }
+   ],
+   "source": [
+    "import sys\n",
+    "import os\n",
+    "sys.path.append(os.path.dirname(os.path.dirname(os.getcwd())))\n",
+    "\n",
+    "%load_ext autoreload\n",
+    "%autoreload 2\n",
+    "\n",
+    "from main import ask_vanna\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Create a human query"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "query = \"what is the number of days where the temperature above 35 in 2050 in Marseille\"\n",
+    "# query = \"Compare the winter and summer precipitation in 2050 in Marseille\"\n",
+    "# query = \"What is the impact of climate  in Bordeaux?\"\n",
+    "query = \"Quelle sera la température à Marseille sur les prochaines années ?\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Call the function ask vanna, it gives an output of a the sql query and the dataframe of the result (tuple)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "SQL Prompt: [{'role': 'system', 'content': \"You are a SQLite expert. Please help to generate a SQL query to answer the question. Your response should ONLY be based on the given context and follow the response guidelines and format instructions. \\n===Tables \\n\\n                CREATE TABLE Mean_winter_temperature (\\n                    y FLOAT,\\n                    x FLOAT,\\n                    year INT, \\n month INT, \\n  day INT \\n,\\n                    LambertParisII VARCHAR(255),\\n                    lat FLOAT,\\n                    lon FLOAT,\\n                    TMm FLOAT, -- Température moyenne en hiver\\n                );\\n            \\n\\n\\n                CREATE TABLE Mean_summer_temperature (\\n                    y FLOAT,\\n                    x FLOAT,\\n                    year INT, \\n month INT, \\n  day INT \\n,\\n                    LambertParisII VARCHAR(255),\\n                    lat FLOAT,\\n                    lon FLOAT,\\n                    TMm FLOAT, -- Température moyenne en été\\n                );\\n            \\n\\n\\n===Additional Context \\n\\n\\n                This table contains information on the number of days when the maximum temperature in the past and the future\\n                is greater than or equal to 35°C.\\n                The variables are as follows:\\n                - 'y' and 'x': Lambert Paris II coordinates for the location.\\n                - year: Year of the observation.\\n\\n              - month : Month of the observation.\\n\\n              - day: Day of the observation.\\n\\n                - 'LambertParisII': Indicates that the x, y coordinates are in the Lambert Paris II projection.\\n                - 'lat' and 'lon': Latitude and longitude of the location.\\n                - 'TX35D': Number of days with Tx ≥ 35°C.\\n                \\n\\n\\n                This table contains information on the number of days when the maximum temperature in the past and the future\\n                is greater than or equal to 30°C.\\n                The variables are as follows:\\n                - 'y' and 'x': Lambert Paris II coordinates for the location.\\n                - year: Year of the observation.\\n\\n              - month : Month of the observation.\\n\\n              - day: Day of the observation.\\n\\n                - 'LambertParisII': Indicates that the x, y coordinates are in the Lambert Paris II projection.\\n                - 'lat' and 'lon': Latitude and longitude of the location.\\n                - 'TX30D': Number of days with Tx ≥ 30°C.\\n                \\n\\n===Response Guidelines \\n1. If the provided context is sufficient, please generate a valid SQL query without any explanations for the question. \\n2. If the provided context is almost sufficient but requires knowledge of a specific string in a particular column, please generate an intermediate SQL query to find the distinct strings in that column. Prepend the query with a comment saying intermediate_sql \\n3. If the provided context is insufficient, please explain why it can't be generated. \\n4. Please use the most relevant table(s). \\n5. If the question has been asked and answered before, please repeat the answer exactly as it was given before. \\n6. Ensure that the output SQL is SQLite-compliant and executable, and free of syntax errors. \\n\"}, {'role': 'user', 'content': 'Quelle sera la température à lat, long : (43.2961743, 5.3699525) sur les prochaines années ?'}]\n",
+      "Using model gpt-4o-mini for 812.0 tokens (approx)\n",
+      "LLM Response: intermediate_sql\n",
+      "```sql\n",
+      "SELECT DISTINCT year FROM Mean_winter_temperature WHERE lat = 43.2961743 AND lon = 5.3699525\n",
+      "UNION\n",
+      "SELECT DISTINCT year FROM Mean_summer_temperature WHERE lat = 43.2961743 AND lon = 5.3699525;\n",
+      "```\n",
+      "Extracted SQL: SELECT DISTINCT year FROM Mean_winter_temperature WHERE lat = 43.2961743 AND lon = 5.3699525\n",
+      "UNION\n",
+      "SELECT DISTINCT year FROM Mean_summer_temperature WHERE lat = 43.2961743 AND lon = 5.3699525;\n",
+      "Running Intermediate SQL: SELECT DISTINCT year FROM Mean_winter_temperature WHERE lat = 43.2961743 AND lon = 5.3699525\n",
+      "UNION\n",
+      "SELECT DISTINCT year FROM Mean_summer_temperature WHERE lat = 43.2961743 AND lon = 5.3699525;\n",
+      "Final SQL Prompt: [{'role': 'system', 'content': \"You are a SQLite expert. Please help to generate a SQL query to answer the question. Your response should ONLY be based on the given context and follow the response guidelines and format instructions. \\n===Tables \\n\\n                CREATE TABLE Mean_winter_temperature (\\n                    y FLOAT,\\n                    x FLOAT,\\n                    year INT, \\n month INT, \\n  day INT \\n,\\n                    LambertParisII VARCHAR(255),\\n                    lat FLOAT,\\n                    lon FLOAT,\\n                    TMm FLOAT, -- Température moyenne en hiver\\n                );\\n            \\n\\n\\n                CREATE TABLE Mean_summer_temperature (\\n                    y FLOAT,\\n                    x FLOAT,\\n                    year INT, \\n month INT, \\n  day INT \\n,\\n                    LambertParisII VARCHAR(255),\\n                    lat FLOAT,\\n                    lon FLOAT,\\n                    TMm FLOAT, -- Température moyenne en été\\n                );\\n            \\n\\n\\n===Additional Context \\n\\n\\n                This table contains information on the number of days when the maximum temperature in the past and the future\\n                is greater than or equal to 35°C.\\n                The variables are as follows:\\n                - 'y' and 'x': Lambert Paris II coordinates for the location.\\n                - year: Year of the observation.\\n\\n              - month : Month of the observation.\\n\\n              - day: Day of the observation.\\n\\n                - 'LambertParisII': Indicates that the x, y coordinates are in the Lambert Paris II projection.\\n                - 'lat' and 'lon': Latitude and longitude of the location.\\n                - 'TX35D': Number of days with Tx ≥ 35°C.\\n                \\n\\n\\n                This table contains information on the number of days when the maximum temperature in the past and the future\\n                is greater than or equal to 30°C.\\n                The variables are as follows:\\n                - 'y' and 'x': Lambert Paris II coordinates for the location.\\n                - year: Year of the observation.\\n\\n              - month : Month of the observation.\\n\\n              - day: Day of the observation.\\n\\n                - 'LambertParisII': Indicates that the x, y coordinates are in the Lambert Paris II projection.\\n                - 'lat' and 'lon': Latitude and longitude of the location.\\n                - 'TX30D': Number of days with Tx ≥ 30°C.\\n                \\n\\nThe following is a pandas DataFrame with the results of the intermediate SQL query SELECT DISTINCT year FROM Mean_winter_temperature WHERE lat = 43.2961743 AND lon = 5.3699525\\nUNION\\nSELECT DISTINCT year FROM Mean_summer_temperature WHERE lat = 43.2961743 AND lon = 5.3699525;: \\n| year   |\\n|--------|\\n\\n===Response Guidelines \\n1. If the provided context is sufficient, please generate a valid SQL query without any explanations for the question. \\n2. If the provided context is almost sufficient but requires knowledge of a specific string in a particular column, please generate an intermediate SQL query to find the distinct strings in that column. Prepend the query with a comment saying intermediate_sql \\n3. If the provided context is insufficient, please explain why it can't be generated. \\n4. Please use the most relevant table(s). \\n5. If the question has been asked and answered before, please repeat the answer exactly as it was given before. \\n6. Ensure that the output SQL is SQLite-compliant and executable, and free of syntax errors. \\n\"}, {'role': 'user', 'content': 'Quelle sera la température à lat, long : (43.2961743, 5.3699525) sur les prochaines années ?'}]\n",
+      "Using model gpt-4o-mini for 887.25 tokens (approx)\n",
+      "LLM Response: La context fourni ne contient pas d'informations sur les prévisions de température pour les prochaines années. Par conséquent, je ne peux pas générer de requête SQL pour répondre à cette question.\n",
+      "Couldn't run sql:  Execution failed on sql 'La context fourni ne contient pas d'informations sur les prévisions de température pour les prochaines années. Par conséquent, je ne peux pas générer de requête SQL pour répondre à cette question.': near \"La\": syntax error\n"
+     ]
+    },
+    {
+     "ename": "IndexError",
+     "evalue": "list index out of range",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mIndexError\u001b[0m                                Traceback (most recent call last)",
+      "Cell \u001b[0;32mIn[4], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m \u001b[43mask_vanna\u001b[49m\u001b[43m(\u001b[49m\u001b[43mquery\u001b[49m\u001b[43m)\u001b[49m\n",
+      "File \u001b[0;32m~/ai4s/climate_qa/climate-question-answering/climateqa/talk_to_data/main.py:28\u001b[0m, in \u001b[0;36mask_vanna\u001b[0;34m(query)\u001b[0m\n\u001b[1;32m     26\u001b[0m answer \u001b[38;5;241m=\u001b[39m vn\u001b[38;5;241m.\u001b[39mask(user_input, print_results\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mFalse\u001b[39;00m, allow_llm_to_see_data\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mTrue\u001b[39;00m)\n\u001b[1;32m     27\u001b[0m table \u001b[38;5;241m=\u001b[39m detectTable(answer[\u001b[38;5;241m0\u001b[39m])\n\u001b[0;32m---> 28\u001b[0m coords2 \u001b[38;5;241m=\u001b[39m nearestNeighbourSQL(db_vanna_path, coords, \u001b[43mtable\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;241;43m0\u001b[39;49m\u001b[43m]\u001b[49m)\n\u001b[1;32m     30\u001b[0m query  \u001b[38;5;241m=\u001b[39m answer[\u001b[38;5;241m0\u001b[39m]\u001b[38;5;241m.\u001b[39mreplace(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mcoords[\u001b[38;5;241m0\u001b[39m]\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mcoords2[\u001b[38;5;241m0\u001b[39m]\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m     31\u001b[0m sql_query \u001b[38;5;241m=\u001b[39m query\u001b[38;5;241m.\u001b[39mreplace(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mcoords[\u001b[38;5;241m1\u001b[39m]\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mcoords2[\u001b[38;5;241m1\u001b[39m]\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m)\n",
+      "\u001b[0;31mIndexError\u001b[0m: list index out of range"
+     ]
+    }
+   ],
+   "source": [
+    "ask_vanna(query)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "climateqa",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.9"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

climateqa/talk_to_data/main.py

@@ -0,0 +1,43 @@
+from climateqa.talk_to_data.myVanna import MyVanna
+from climateqa.talk_to_data.utils import loc2coords, detect_location_with_openai, detectTable, nearestNeighbourSQL
+import sqlite3
+import os
+import pandas as pd
+
+
+
+OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
+PC_API_KEY = os.getenv('VANNA_PINECONE_API_KEY')
+INDEX_NAME = os.getenv('VANNA_INDEX_NAME')
+VANNA_MODEL = os.getenv('VANNA_MODEL')
+
+
+#Vanna object
+vn = MyVanna(config = {"temperature": 0, "api_key": OPENAI_API_KEY, 'model': VANNA_MODEL, 'pc_api_key': PC_API_KEY, 'index_name': INDEX_NAME})
+db_vanna_path = os.path.join(os.path.dirname(__file__), "database/drias.db")
+vn.connect_to_sqlite(db_vanna_path)
+
+
+def ask_vanna(query):
+    location = detect_location_with_openai(OPENAI_API_KEY, query)
+    if location:
+        coords = loc2coords(location)
+        user_input = query.replace(location, f"lat, long : {coords}")
+        answer = vn.ask(user_input, print_results=False, allow_llm_to_see_data=True)
+        table = detectTable(answer[0])
+        coords2 = nearestNeighbourSQL(db_vanna_path, coords, table[0])
+
+        query  = answer[0].replace(f"{coords[0]}", f"{coords2[0]}")
+        sql_query = query.replace(f"{coords[1]}", f"{coords2[1]}")
+
+        db = sqlite3.connect(db_vanna_path)
+        result = db.cursor().execute(sql_query).fetchall()
+        print(result)
+        df = pd.DataFrame(result, columns=answer[1].columns)
+        
+    else: 
+        answer = vn.ask(query, visualize=True, print_results=False, allow_llm_to_see_data=True)
+        sql_query = answer[0]
+        df = answer[1]
+        
+    return (sql_query, df)

climateqa/talk_to_data/myVanna.py

@@ -0,0 +1,13 @@
+from dotenv import load_dotenv
+from climateqa.talk_to_data.vanna_class import MyCustomVectorDB
+from vanna.openai import OpenAI_Chat
+import os
+
+load_dotenv()
+
+OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
+
+class MyVanna(MyCustomVectorDB, OpenAI_Chat):
+    def __init__(self, config=None):
+        MyCustomVectorDB.__init__(self, config=config)
+        OpenAI_Chat.__init__(self, config=config)

climateqa/talk_to_data/utils.py

@@ -0,0 +1,64 @@
+import re
+import openai
+import pandas as pd
+from geopy.geocoders import Nominatim
+import sqlite3
+
+
+def detect_location_with_openai(api_key, sentence):
+    """
+    Detects locations in a sentence using OpenAI's API.
+    """
+    openai.api_key = api_key
+
+    prompt = f"""
+    Extract all locations (cities, countries, states, or geographical areas) mentioned in the following sentence.
+    Return the result as a Python list. If no locations are mentioned, return an empty list.
+    
+    Sentence: "{sentence}"
+    """
+
+    response = openai.chat.completions.create(
+        model="gpt-4o-mini",  
+        messages=[
+            {"role": "system", "content": "You are a helpful assistant skilled in identifying locations in text."},
+            {"role": "user", "content": prompt}
+        ],
+        max_tokens=100,
+        temperature=0
+    )
+
+    return response.choices[0].message.content.split("\n")[1][2:-2]
+
+
+def detectTable(sql_query):
+    pattern = r'(?i)\bFROM\s+((?:`[^`]+`|"[^"]+"|\'[^\']+\'|\w+)(?:\.(?:`[^`]+`|"[^"]+"|\'[^\']+\'|\w+))*)'
+    matches = re.findall(pattern, sql_query)
+    return matches
+
+
+
+def loc2coords(location : str):
+    geolocator = Nominatim(user_agent="city_to_latlong")
+    location = geolocator.geocode(location)
+    return (location.latitude, location.longitude)
+
+
+def coords2loc(coords : tuple):
+    geolocator = Nominatim(user_agent="coords_to_city")
+    try:
+        location = geolocator.reverse(coords)
+        return location.address
+    except Exception as e:
+        print(f"Error: {e}")
+        return "Unknown Location"  
+
+
+def nearestNeighbourSQL(db: str, location: tuple, table : str):
+    conn = sqlite3.connect(db)
+    long = round(location[1], 3)
+    lat = round(location[0], 3)
+    cursor  = conn.cursor()
+    cursor.execute(f"SELECT lat, lon FROM {table} WHERE lat BETWEEN {lat - 0.3} AND {lat + 0.3} AND lon BETWEEN {long - 0.3} AND {long + 0.3}")
+    results = cursor.fetchall()
+    return results[0]

climateqa/talk_to_data/vanna_class.py

@@ -0,0 +1,246 @@
+from vanna.base import VannaBase
+from pinecone import Pinecone
+from climateqa.engine.embeddings import get_embeddings_function
+import pandas as pd
+import hashlib
+
+class MyCustomVectorDB(VannaBase):
+    
+    """
+    VectorDB class for storing and retrieving vectors from Pinecone.
+    
+    args : 
+        config (dict) : Configuration dictionary containing the Pinecone API key and the index name :
+            - pc_api_key (str) : Pinecone API key
+            - index_name (str) : Pinecone index name
+            - top_k (int) : Number of top results to return (default = 2)
+            
+    """
+    
+    def __init__(self,config):
+        super().__init__(config = config)
+        try : 
+            self.api_key = config.get('pc_api_key')
+            self.index_name = config.get('index_name')
+        except : 
+            raise Exception("Please provide the Pinecone API key and the index name")
+        
+        self.pc = Pinecone(api_key = self.api_key)
+        self.index = self.pc.Index(self.index_name)
+        self.top_k = config.get('top_k', 2)
+        self.embeddings = get_embeddings_function()
+        
+        
+    def check_embedding(self, id, namespace):
+        fetched = self.index.fetch(ids = [id], namespace = namespace)
+        if fetched['vectors'] == {}: 
+            return False
+        return True
+    
+    def generate_hash_id(self, data: str) -> str:
+        """
+        Generate a unique hash ID for the given data.
+        
+        Args:
+            data (str): The input data to hash (e.g., a concatenated string of user attributes).
+        
+        Returns:
+            str: A unique hash ID as a hexadecimal string.
+        """
+        
+        data_bytes = data.encode('utf-8')
+        hash_object = hashlib.sha256(data_bytes)
+        hash_id = hash_object.hexdigest()
+        
+        return hash_id
+    
+    def add_ddl(self, ddl: str, **kwargs) -> str:
+        id = self.generate_hash_id(ddl) + '_ddl'
+        
+        if self.check_embedding(id, 'ddl'):
+            print(f"DDL having id {id} already exists")
+            return id
+        
+        self.index.upsert(
+            vectors = [(id, self.embeddings.embed_query(ddl), {'ddl': ddl})],
+            namespace = 'ddl'
+        )
+        
+        return id
+
+    def add_documentation(self, doc: str, **kwargs) -> str:
+        id = self.generate_hash_id(doc) + '_doc'
+        
+        if self.check_embedding(id, 'documentation'):
+            print(f"Documentation having id {id} already exists")
+            return id
+        
+        self.index.upsert(
+            vectors = [(id, self.embeddings.embed_query(doc), {'doc': doc})],
+            namespace = 'documentation'
+        )
+        
+        return id
+
+    def add_question_sql(self, question: str, sql: str, **kwargs) -> str:
+        id = self.generate_hash_id(question) + '_sql'
+        
+        if self.check_embedding(id, 'question_sql'):
+            print(f"Question-SQL pair having id {id} already exists")
+            return id
+        
+        self.index.upsert(
+            vectors = [(id, self.embeddings.embed_query(question + sql), {'question': question, 'sql': sql})],
+            namespace = 'question_sql'
+        )
+        
+        return id
+
+    def get_related_ddl(self, question: str, **kwargs) -> list:
+        res = self.index.query(
+            vector=self.embeddings.embed_query(question),
+            top_k=self.top_k,
+            namespace='ddl',
+            include_metadata=True
+        )
+        
+        return [match['metadata']['ddl'] for match in res['matches']]
+
+    def get_related_documentation(self, question: str, **kwargs) -> list:
+        res = self.index.query(
+            vector=self.embeddings.embed_query(question),
+            top_k=self.top_k,
+            namespace='documentation',
+            include_metadata=True
+        )
+        
+        return [match['metadata']['doc'] for match in res['matches']]
+
+    def get_similar_question_sql(self, question: str, **kwargs) -> list:
+        res = self.index.query(
+            vector=self.embeddings.embed_query(question),
+            top_k=self.top_k,
+            namespace='question_sql',
+            include_metadata=True
+        )
+        
+        return [(match['metadata']['question'], match['metadata']['sql']) for match in res['matches']]
+
+    def get_training_data(self, **kwargs) -> pd.DataFrame:
+        
+        list_of_data = []
+        
+        namespaces = ['ddl', 'documentation', 'question_sql']
+        
+        for namespace in namespaces:
+            
+            data = self.index.query(
+            top_k=10000,
+            namespace=namespace,
+            include_metadata=True,
+            include_values=False
+            )
+            
+            for match in data['matches']:
+                list_of_data.append(match['metadata'])
+                
+        return pd.DataFrame(list_of_data)
+            
+
+
+    def remove_training_data(self, id: str, **kwargs) -> bool:
+        if id.endswith("_ddl"):
+            self.Index.delete(ids=[id], namespace="_ddl")
+            return True
+        if id.endswith("_sql"):
+            self.index.delete(ids=[id], namespace="_sql")
+            return True
+        
+        if id.endswith("_doc"):
+            self.Index.delete(ids=[id], namespace="_doc")
+            return True
+        
+        return False
+    
+    def generate_embedding(self, text, **kwargs):
+        # Implement the method here
+        pass
+
+
+def get_sql_prompt(
+        self,
+        initial_prompt : str,
+        question: str,
+        question_sql_list: list,
+        ddl_list: list,
+        doc_list: list,
+        **kwargs,
+    ):
+        """
+        Example:
+        ```python
+        vn.get_sql_prompt(
+            question="What are the top 10 customers by sales?",
+            question_sql_list=[{"question": "What are the top 10 customers by sales?", "sql": "SELECT * FROM customers ORDER BY sales DESC LIMIT 10"}],
+            ddl_list=["CREATE TABLE customers (id INT, name TEXT, sales DECIMAL)"],
+            doc_list=["The customers table contains information about customers and their sales."],
+        )
+
+        ```
+
+        This method is used to generate a prompt for the LLM to generate SQL.
+
+        Args:
+            question (str): The question to generate SQL for.
+            question_sql_list (list): A list of questions and their corresponding SQL statements.
+            ddl_list (list): A list of DDL statements.
+            doc_list (list): A list of documentation.
+
+        Returns:
+            any: The prompt for the LLM to generate SQL.
+        """
+
+        if initial_prompt is None:
+            initial_prompt = f"You are a {self.dialect} expert. " + \
+            "Please help to generate a SQL query to answer the question. Your response should ONLY be based on the given context and follow the response guidelines and format instructions. "
+
+        initial_prompt = self.add_ddl_to_prompt(
+            initial_prompt, ddl_list, max_tokens=self.max_tokens
+        )
+
+        if self.static_documentation != "":
+            doc_list.append(self.static_documentation)
+
+        initial_prompt = self.add_documentation_to_prompt(
+            initial_prompt, doc_list, max_tokens=self.max_tokens
+        )
+
+        initial_prompt = self.add_sql_to_prompt(
+            initial_prompt, question_sql_list, max_tokens=self.max_tokens
+        )
+
+
+        initial_prompt += (
+            "===Response Guidelines \n"
+            "1. If the provided context is sufficient, please generate a valid SQL query without any explanations for the question. \n"
+            "2. If the provided context is almost sufficient but requires knowledge of a specific string in a particular column, please generate an intermediate SQL query to find the distinct strings in that column. Prepend the query with a comment saying intermediate_sql \n"
+            "3. If the provided context is insufficient, please give a sql query based on your knowledge and the context provided. \n"
+            "4. Please use the most relevant table(s). \n"
+            "5. If the question has been asked and answered before, please repeat the answer exactly as it was given before. \n"
+            f"6. Ensure that the output SQL is {self.dialect}-compliant and executable, and free of syntax errors. \n"
+        )
+
+
+        message_log = [self.system_message(initial_prompt)]
+
+        for example in question_sql_list:
+            if example is None:
+                print("example is None")
+            else:
+                if example is not None and "question" in example and "sql" in example:
+                    message_log.append(self.user_message(example["question"]))
+                    message_log.append(self.assistant_message(example["sql"]))
+
+        message_log.append(self.user_message(question))
+
+        return message_log

requirements.txt

@@ -19,3 +19,5 @@ langchain-community==0.2
 msal==1.31
 matplotlib==3.9.2
 gradio-modal==0.0.4
+vanna==0.7.5
+geopy==2.4.1