vector.py

from langchain_core.documents import Document
import pandas as pd
import os

from langchain_community.embeddings import OllamaEmbeddings
from langchain_qdrant import QdrantVectorStore
from qdrant_client import QdrantClient
from qdrant_client.http.models import Distance, VectorParams

# Charger les données du CSV
df = pd.read_csv("realistic_restaurant_reviews.csv")

# Initialisation du modèle d'embedding
embeddings = OllamaEmbeddings(model="mxbai-embed-large")  # Ce modèle retourne des vecteurs de taille 1024

# Connexion au Qdrant local via URL
client = QdrantClient(url="http://localhost:6333")

# Nom de la collection
collection_name = "restaurant_reviews"

# Forcer la recréation de la collection pour correspondre à la taille 1024
client.recreate_collection(
    collection_name=collection_name,
    vectors_config=VectorParams(size=1024, distance=Distance.COSINE)
)

# Créer les documents
documents = []
ids = []

for i, row in df.iterrows():
    document = Document(
        page_content=row["Title"] + " " + row["Review"],
        metadata={
            "rating": row["Rating"],
            "date": row["Date"]
        }
    )
    documents.append(document)
    ids.append(str(i))

# Créer le vector store
vector_store = QdrantVectorStore(
    client=client,
    collection_name=collection_name,
    embedding=embeddings
)

# Ajouter les documents
vector_store.add_documents(documents=documents, ids=ids)

# Optionnel : Créer un retriever
retriever = vector_store.as_retriever(search_kwargs={"k": 5})