Update pages/1_Chatbot_FR.py

pages/1_Chatbot_FR.py

@@ -193,54 +193,54 @@ def predict_class(text, max_length=500):
 # 3. CLASSIFICATION DATASET & VECTOR STORE
 ##############################
 
-@st.cache_data(show_spinner=False)
-def load_classification_dataset():
-    """
-    Loads the classification Q&A dataset from the Excel file and returns a DataFrame.
-    """
-    df = pd.read_excel("Classification dataset - Q&A.xlsx", sheet_name="Fr")
-    return df
-
-@st.cache_resource(show_spinner=False)
-def load_classification_vectorstore(persist_dir: str = "./chroma_db_class_fr"):
-    """
-    Builds (and persists) a Chroma vector store from the classification Q&A dataset.
-    Each document contains the answer (Réponse) with metadata including the class ("Classe").
-    """
-    df = load_classification_dataset()
-    # Create documents using the "Réponse" as content and include metadata.
-    from langchain.schema import Document
-    documents = []
-    for _, row in df.iterrows():
-        documents.append(
-            Document(
-                page_content=row["Réponse"],
-                metadata={
-                    "id": row["ID"],
-                    "Classe": row["Classe"],
-                    "Question": row["Question"]
-                }
-            )
-        )
-    from langchain.embeddings import HuggingFaceEmbeddings
-    from langchain.vectorstores import Chroma
-    embedding_model = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
-    vectorstore = Chroma.from_documents(documents, embedding_model, persist_directory=persist_dir)
-    vectorstore.persist()
-    return vectorstore
-
-def load_existing_classification_vectorstore(persist_dir: str = "./chroma_db_class_fr"):
-    """
-    Loads an existing Chroma vector store for the classification dataset.
-    """
-    from langchain.embeddings import HuggingFaceEmbeddings
-    from langchain.vectorstores import Chroma
-    embedding_model = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
-    vectorstore = Chroma(
-        persist_directory=persist_dir,
-        embedding_function=embedding_model
-    )
-    return vectorstore
+# @st.cache_data(show_spinner=False)
+# def load_classification_dataset():
+#     """
+#     Loads the classification Q&A dataset from the Excel file and returns a DataFrame.
+#     """
+#     df = pd.read_excel("Classification dataset - Q&A.xlsx", sheet_name="Fr")
+#     return df
+
+# @st.cache_resource(show_spinner=False)
+# def load_classification_vectorstore(persist_dir: str = "./chroma_db_class_fr"):
+#     """
+#     Builds (and persists) a Chroma vector store from the classification Q&A dataset.
+#     Each document contains the answer (Réponse) with metadata including the class ("Classe").
+#     """
+#     df = load_classification_dataset()
+#     # Create documents using the "Réponse" as content and include metadata.
+#     from langchain.schema import Document
+#     documents = []
+#     for _, row in df.iterrows():
+#         documents.append(
+#             Document(
+#                 page_content=row["Réponse"],
+#                 metadata={
+#                     "id": row["ID"],
+#                     "Classe": row["Classe"],
+#                     "Question": row["Question"]
+#                 }
+#             )
+#         )
+#     from langchain.embeddings import HuggingFaceEmbeddings
+#     from langchain.vectorstores import Chroma
+#     embedding_model = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
+#     vectorstore = Chroma.from_documents(documents, embedding_model, persist_directory=persist_dir)
+#     vectorstore.persist()
+#     return vectorstore
+
+# def load_existing_classification_vectorstore(persist_dir: str = "./chroma_db_class_fr"):
+#     """
+#     Loads an existing Chroma vector store for the classification dataset.
+#     """
+#     from langchain.embeddings import HuggingFaceEmbeddings
+#     from langchain.vectorstores import Chroma
+#     embedding_model = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
+#     vectorstore = Chroma(
+#         persist_directory=persist_dir,
+#         embedding_function=embedding_model
+#     )
+#     return vectorstore
 
 ##############################
 # 4. PROMPT & LLM FR SETUP
@@ -394,37 +394,37 @@ def main():
             st.write(f"**Classe prédite :** {predicted_label}")
             
             # --- Retrieve final answer using the classification vector store ---
-            if predicted_label != "Autre":
-                # Build or load the classification vector store if not already in session_state.
-                if "class_retriever" not in st.session_state:
-                    # Either create new or load existing
-                    try:
-                        # Attempt to load an existing vector store.
-                        vectorstore_class = load_existing_classification_vectorstore("./chroma_db_class_fr")
-                    except Exception:
-                        # If not found, create it.
-                        vectorstore_class = load_classification_vectorstore("./chroma_db_class_fr")
-                    st.session_state["class_retriever"] = vectorstore_class.as_retriever(
-                        search_type="mmr",
-                        search_kwargs={"k": 1, "lambda_mult": 0.5}
-                    )
-                # Retrieve the final answer with a metadata filter.
-                # (Assumes the underlying retriever supports a filter parameter.)
-                final_docs = st.session_state["class_retriever"].get_relevant_documents(
-                    response_fr, filter={"Classe": predicted_label}
-                )
-                if final_docs:
-                    final_answer = final_docs[0].page_content
-                else:
-                    final_answer = response_fr  # fallback if no document found
-            else:
-                final_answer = ("Je n'ai pas d'information précise à ce sujet. "
-                                "Souhaitez-vous que je vous mette en contact avec un agent Inwi ?")
+            # if predicted_label != "Autre":
+            #     # Build or load the classification vector store if not already in session_state.
+            #     if "class_retriever" not in st.session_state:
+            #         # Either create new or load existing
+            #         try:
+            #             # Attempt to load an existing vector store.
+            #             vectorstore_class = load_existing_classification_vectorstore("./chroma_db_class_fr")
+            #         except Exception:
+            #             # If not found, create it.
+            #             vectorstore_class = load_classification_vectorstore("./chroma_db_class_fr")
+            #         st.session_state["class_retriever"] = vectorstore_class.as_retriever(
+            #             search_type="mmr",
+            #             search_kwargs={"k": 1, "lambda_mult": 0.5}
+            #         )
+            #     # Retrieve the final answer with a metadata filter.
+            #     # (Assumes the underlying retriever supports a filter parameter.)
+            #     final_docs = st.session_state["class_retriever"].get_relevant_documents(
+            #         response_fr, filter={"Classe": predicted_label}
+            #     )
+            #     if final_docs:
+            #         final_answer = final_docs[0].page_content
+            #     else:
+            #         final_answer = response_fr  # fallback if no document found
+            # else:
+            #     final_answer = ("Je n'ai pas d'information précise à ce sujet. "
+            #                     "Souhaitez-vous que je vous mette en contact avec un agent Inwi ?")
             
-            st.write("**Réponse finale :**")
-            st.write(final_answer)
+            # st.write("**Réponse finale :**")
+            # st.write(final_answer)
         else:
             st.write("Aucun contexte trouvé pour cette question. Essayez autre chose.")
 
 if __name__ == "__main__":
-    main()
+    main()