après hackathon

app.py

@@ -1,99 +1,53 @@
 import os
 import json
+import pandas as pd
 import gradio as gr
 from llama_index.core import (
     VectorStoreIndex,
     download_loader,
     StorageContext
 )
-from dotenv import load_dotenv, find_dotenv
-
-import chromadb
-
-from llama_index.llms.mistralai import MistralAI
-from llama_index.embeddings.mistralai import MistralAIEmbedding
-from llama_index.vector_stores.chroma import ChromaVectorStore
-from llama_index.core.indices.service_context import ServiceContext
 
+import logging
+from dotenv import load_dotenv, find_dotenv
 from pathlib import Path
 
+# from llama_index.llms.mistralai import MistralAI
+from mistralai.client import MistralClient
+from mistralai.models.chat_completion import ChatMessage
+# from llama_index.embeddings.mistralai import MistralAIEmbedding
+from src.utils_fct import *
+
 TITLE = "RIZOA-AUCHAN Chatbot Demo"
-DESCRIPTION = "Example of an assistant with Gradio, coupling with function calling and Mistral AI via its API"
+DESCRIPTION = "Example of an assistant with Gradio, coupling with function callings and Mistral AI via its API"
 PLACEHOLDER = (
-    "Vous pouvez me posez une question sur ce contexte, appuyer sur Entrée pour valider"
+    "Vous pouvez me posez une question, appuyer sur Entrée pour valider"
 )
-PLACEHOLDER_URL = "Extract text from this url"
-llm_model = "mistral-medium"
+EXAMPLES = ["Comment fait on pour produire du maïs ?", "Rédige moi une lettre pour faire un stage dans une exploitation agricole", "Comment reprendre une exploitation agricole ?"]
+MODEL = "mistral-large-latest"
+
+# FILE = Path(__file__).resolve()
+# BASE_PATH = FILE.parents[0]
 
 load_dotenv()
-env_api_key = os.environ.get("MISTRAL_API_KEY")
-query_engine = None
+ENV_API_KEY = os.environ.get("MISTRAL_API_KEY")
+# HISTORY = pd.read_csv(os.path.join(BASE_PATH, "data/cereal_price.csv"), encoding="latin-1")
+# HISTORY = HISTORY[[HISTORY["memberStateName"]=="France"]]
+# HISTORY['price'] = HISTORY['price'].str.replace(",", ".").astype('float64')
 
 # Define LLMs
-llm = MistralAI(api_key=env_api_key, model=llm_model)
-embed_model = MistralAIEmbedding(model_name="mistral-embed", api_key=env_api_key)
-
-# create client and a new collection
-db = chromadb.PersistentClient(path="./chroma_db")
-chroma_collection = db.get_or_create_collection("quickstart")
-
-# set up ChromaVectorStore and load in data
-vector_store = ChromaVectorStore(chroma_collection=chroma_collection)
-storage_context = StorageContext.from_defaults(vector_store=vector_store)
-service_context = ServiceContext.from_defaults(
-    chunk_size=1024, llm=llm, embed_model=embed_model
-)
-
-#PDFReader = download_loader("PDFReader")
-#loader = PDFReader()
-
-index = VectorStoreIndex(
-    [], service_context=service_context, storage_context=storage_context
-)
-query_engine = index.as_query_engine(similarity_top_k=5)
-
-FILE = Path(__file__).resolve()
-BASE_PATH = FILE.parents[0]
-
-'''
-image = os.path.join(BASE_PATH,"img","logo_rizoa_auchan.jpg")
-print(f"Chemin de l'image : {image}")
-image = os.path.join("img","logo_rizoa_auchan.jpg")
-print(f"chemin 2 : {image}")
-image = os.path.abspath(os.path.join("img", "logo_rizoa_auchan.jpg"))
-print(f"Image 3 : {image}")
-image = os.path.join("https://huggingface.co/spaces/rizoa-auchan-hack/hack/blob/main/img/logo_rizoa_auchan.jpg")
-print(f"Image 4 : {image}")
-'''
-image = os.path.join("logo_rizoa_auchan.jpg")
-print(f"Chemin:{image}")
-
-if os.path.exists(image):
-    print("Image existe")
-else:
-    print("Image n'existe pas")
-
-
-PLACEHOLDER = (image)
+CLIENT = MistralClient(api_key=ENV_API_KEY)
+# EMBED_MODEL = MistralAIEmbedding(model_name="mistral-embed", api_key=ENV_API_KEY)
 
 with gr.Blocks() as demo:
     with gr.Row(): 
-        
         with gr.Column(scale=1): 
-            '''
-            gr.Image(
-                    #value=os.path.join(BASE_PATH,"img","logo_rizoa_auchan.jpg"), 
-                    #value=os.path.join("img","logo_rizoa_auchan.jpg"),
-                    value="logo_rizoa_auchan.jpg",
+            gr.Image(value= os.path.join(BASE_PATH, "img/logo_rizoa_auchan.jpg"),#".\img\logo_rizoa_auchan.jpg", 
                     height=250,
                     width=250,
-                    container=False,
+                    container=False, 
                     show_download_button=False
                     )
-            '''
-            gr.HTML(
-                value = '<img src="https://huggingface.co/spaces/rizoa-auchan-hack/hack/resolve/main/LOGO_RIZOA_CARRE.jpg">'
-            )
         with gr.Column(scale=4):   
             gr.Markdown(
                 """ 
@@ -103,59 +57,35 @@ with gr.Blocks() as demo:
                 """
             )
 
-    # gr.Markdown(""" ### 1 / Extract data from PDF """)
-
-    # with gr.Row():
-    #     with gr.Column():
-    #         input_file = gr.File(
-    #             label="Load a pdf",
-    #             file_types=[".pdf"],
-    #             file_count="single",
-    #             type="filepath",
-    #             interactive=True,
-    #         )
-    #         file_msg = gr.Textbox(
-    #             label="Loaded documents:", container=False, visible=False
-    #         )
-
-    #         input_file.upload(
-    #             fn=load_document,
-    #             inputs=[
-    #                 input_file,
-    #             ],
-    #             outputs=[file_msg],
-    #             concurrency_limit=20,
-    #         )
-
-    #         file_btn = gr.Button(value="Encode file ✅", interactive=True)
-    #         btn_msg = gr.Textbox(container=False, visible=False)
-
-    #         with gr.Row():
-    #             db_list = gr.Markdown(value=get_documents_in_db)
-    #             delete_btn = gr.Button(value="Empty db 🗑️", interactive=True, scale=0)
-
-    #         file_btn.click(
-    #             load_file,
-    #             inputs=[input_file],
-    #             outputs=[file_msg, btn_msg, db_list],
-    #             show_progress="full",
-    #         )
-    #         delete_btn.click(empty_db, outputs=[db_list], show_progress="minimal")
-
-    gr.Markdown(""" ### Ask a question """)
+    gr.Markdown(f""" ### {DESCRIPTION} """)
 
     chatbot = gr.Chatbot()
     msg = gr.Textbox(placeholder=PLACEHOLDER)
     clear = gr.ClearButton([msg, chatbot])
-
+    
     def respond(message, chat_history):
-        response = query_engine.query(message)
-        chat_history.append((message, str(response)))
-        return chat_history
-
-    msg.submit(respond, [msg, chatbot], [chatbot])
+        messages = [ChatMessage(role="user", content=message)]
+        # response = client.chat(
+        #     model=MODEL, 
+        #     messages=messages)
 
-demo.title = TITLE
+        response = forecast(messages)
+        
+        # prompt = f"Reformule le résultat suivant {response}"
+        # prompt = [ChatMessage(role="user", content=prompt)]
+        # chat_history.append((message, str(response)))
+        final_response = CLIENT.chat(
+            model=MODEL, 
+            messages=response
+        ).choices[0].message.content
+        return "", [[None, None], 
+                [None, str(final_response)]
+                ]
+
+    msg.submit(respond, [msg, chatbot], [msg, chatbot])
+    
+
+# demo.title = TITLE
 
 if __name__ == "__main__":
-    demo.launch(allowed_paths=['/home/user/app/img/','./img/','.'])
+    demo.launch()

requirements.txt

@@ -1,7 +1,4 @@
-pypdf
 mistralai>=0.1.2
-# llama-index-llms-mistralai
-# llama-index-embeddings-mistralai
-llama-index
-gradio
-chromadb
+gradio 
+openai 
+load_dotenv 

src/utils_fct.py

@@ -0,0 +1,117 @@
+from datetime import datetime, timedelta
+import functools
+import json
+import os
+import pandas as pd
+from prophet import Prophet
+from pathlib import Path
+from mistralai.client import MistralClient
+from mistralai.models.chat_completion import ChatMessage
+
+# MODEL
+MODEL = "mistral-large-latest"
+API_KEY=os.environ["MISTRAL_API_KEY"]
+CLIENT = MistralClient(api_key=API_KEY)
+
+# PATH
+FILE = Path(__file__).resolve()
+BASE_PATH = FILE.parents[1]
+
+HISTORY = pd.read_csv(os.path.join(BASE_PATH, "data/cereal_price.csv"), encoding="latin-1")
+HISTORY = HISTORY[HISTORY["memberStateName"]=="France"]
+HISTORY['price'] = HISTORY['price'].str.replace(",", ".").astype('float64')
+
+
+def model_predict(week=26):
+    """
+    Predict future prices using the Prophet model.
+
+    Parameters:
+    - weeks (int): Number of periods to predict into the future (default is 26).
+
+    Returns:
+    - dict: Dictionary containing predicted values and confidence intervals.
+    """
+
+    # Prepare the historical data for the model
+    data = HISTORY[['endDate', 'price']]
+    data.columns = ['ds', 'y']
+
+    # Prophet Model
+    # Instantiate a Prophet object
+    model = Prophet()
+
+    # Fit the model with historical data
+    model.fit(data)
+    
+    # Calculate the current date
+    today_date = datetime.now().date()
+
+    # Calculate the end date for the future DataFrame (specified number of periods from today)
+    end_date = today_date + timedelta(weeks=week)
+
+    # Create a DataFrame with dates starting from today and ending in the specified number of periods
+    future_df = pd.date_range(start=today_date, end=end_date, freq='W').to_frame(name='ds').reset_index(drop=True)
+
+    # Make predictions on the future DataFrame
+    forecast = model.predict(future_df)
+
+    # Return relevant columns from the forecast DataFrame as a dictionary
+    result_dict = {
+        'ds': forecast['ds'].tolist(),
+        'yhat_lower': forecast['yhat_lower'].tolist(),
+        'yhat_upper': forecast['yhat_upper'].tolist(),
+        'yhat': forecast['yhat'].tolist()
+    }
+    
+    return result_dict
+
+model_predict_tool = [{
+    "type": "function",
+    "function": {
+        "name": "model_predict",
+        "description": "Predict future prices using the Prophet model.",
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "week": {
+                    "type": "integer",
+                    "description": "Number of periods to predict into the future (default is 26).",
+                },
+            },
+            "required": ["week"]
+        },
+    },
+}]
+
+names_to_functions = {
+    'model_predict': functools.partial(model_predict), 
+}
+
+# messages = [
+#     ChatMessage(role="user", content="Predict future prices using the Prophet model for 4 weeks in the future")
+# ]
+
+def forecast(messages
+):
+    response = CLIENT.chat(
+        model=MODEL,
+        messages=messages,
+        tools=model_predict_tool,
+        tool_choice="auto"
+    )
+
+    tool_call = response.choices[0].message.tool_calls[0]
+    function_name = tool_call.function.name
+    function_params = json.loads(tool_call.function.arguments)
+    function_result = names_to_functions[function_name](**function_params)
+    date = function_result["ds"][-1]
+    lower = function_result["yhat_lower"][-1]
+    upper = function_result["yhat_upper"][-1]
+    prediction = function_result["yhat"][-1]
+
+    messages.append(ChatMessage(role="tool", 
+                        name=function_name, 
+                        content=str({"date" : date, "prix_minimum": lower, "prix_maximum": upper, "prix_estimé": prediction})
+    ))
+    return messages