reranking working example

rag_app/metadata.ipynb

@@ -1,170 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "code",
-   "execution_count": 1,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from pathlib import Path\n",
-    "from langchain_community.vectorstores import FAISS\n",
-    "from dotenv import load_dotenv\n",
-    "import os\n",
-    "from langchain_huggingface import HuggingFaceEmbeddings"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 3,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "True"
-      ]
-     },
-     "execution_count": 3,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "load_dotenv()"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 5,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "HUGGINGFACEHUB_API_TOKEN = os.getenv('HUGGINGFCEHUB_API_TOKEN')\n",
-    "EMBEDDING_MODEL = os.getenv(\"EMBEDDING_MODEL\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "embeddings = HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 7,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "folder_path = Path('..') / \"vectorstore/faiss-insurance-agent-500\"\n",
-    "faissdb = FAISS.load_local(folder_path=str(folder_path.resolve()),\n",
-    "                                embeddings=embeddings,\n",
-    "                                allow_dangerous_deserialization=True)    "
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 24,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Content: Die private Haftpflichtversicherung...\n",
-      "Metadata: {'source': 'https://www.wuerttembergische.de/versicherungen/stadt/wuppertal/', 'content_type': 'text/html; charset=UTF-8', 'title': 'Versicherung in Wuppertal', 'description': 'Ihre Versicherungsagentur in Wuppertal: Kommen Sie zur Württembergischen Versicherung und profitieren Sie von einer persönlichen Beratung und ausgezeichnetem Service. ', 'language': 'de'}\n",
-      "---\n",
-      "Content: Haftpflichtversicherung...\n",
-      "Metadata: {'source': 'https://www.wuerttembergische.de/wohnen/hausratversicherung/sengschaden/', 'content_type': 'text/html; charset=UTF-8', 'title': 'Sengschäden: So schützt Sie Ihre Hausrat- und Wohngebäudeversicherung', 'description': 'Deckt Ihre Hausratversicherung Sengschäden ab? Finden Sie heraus, wie Sie bei Schäden durch Glut oder Hitze ohne direktes Feuer geschützt sind.\\n', 'language': 'de'}\n",
-      "---\n",
-      "Content: Die Leistungen unserer privaten Haftpflichtversich...\n",
-      "Metadata: {'source': 'https://www.wuerttembergische.de/existenz/private-haftpflichtversicherung/drohnen-versichern/', 'content_type': 'text/html; charset=UTF-8', 'title': 'Drohnen über die private Haftpflicht versichern', 'description': 'Müssen Drohnen versichert sein? Welcher Tarif ist der beste? Erfahren Sie hier die wichtigsten Informationen rund ums Thema Drohne versichern.', 'language': 'de'}\n",
-      "---\n",
-      "Content: Das kann ohne private Haftpflichtversicherung pass...\n",
-      "Metadata: {'source': 'https://www.wuerttembergische.de/existenz/private-haftpflichtversicherung/pflicht/', 'content_type': 'text/html; charset=UTF-8', 'title': 'Ist die private Haftpflichtversicherung Pflicht oder freiwillig?', 'description': 'Ist eine Privathaftpflichtversicherung gesetzlich vorgeschrieben? Welche Haftpflichtversicherung Pflicht sind und welche freiwillig - das erfahren Sie hier.', 'language': 'de'}\n",
-      "---\n",
-      "Content: Private Haftpflicht: keine Pflichtversicherung\n",
-      "Fre...\n",
-      "Metadata: {'source': 'https://www.wuerttembergische.de/existenz/private-haftpflichtversicherung/pflicht/', 'content_type': 'text/html; charset=UTF-8', 'title': 'Ist die private Haftpflichtversicherung Pflicht oder freiwillig?', 'description': 'Ist eine Privathaftpflichtversicherung gesetzlich vorgeschrieben? Welche Haftpflichtversicherung Pflicht sind und welche freiwillig - das erfahren Sie hier.', 'language': 'de'}\n",
-      "---\n"
-     ]
-    }
-   ],
-   "source": [
-    "# Perform a similarity search with an empty query to get random documents\n",
-    "documents = faissdb.similarity_search(\"Private Haftpflicht­versicherung\", k=5)\n",
-    "\n",
-    "for doc in documents:\n",
-    "    print(f\"Content: {doc.page_content[:50]}...\")  # Print first 50 chars of content\n",
-    "    print(f\"Metadata: {doc.metadata}\")\n",
-    "    print(\"---\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 19,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Number of entries in the database: 62496\n"
-     ]
-    }
-   ],
-   "source": [
-    "num_entries = len(faissdb.index_to_docstore_id)\n",
-    "print(f\"Number of entries in the database: {num_entries}\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 20,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Number of entries in the database: 62496\n"
-     ]
-    }
-   ],
-   "source": [
-    "num_entries = faissdb.index.ntotal\n",
-    "print(f\"Number of entries in the database: {num_entries}\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  }
- ],
- "metadata": {
-  "kernelspec": {
-   "display_name": "venv",
-   "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": "3.11.4"
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 2
-}

rag_app/metadata_filtering.py

@@ -1,29 +0,0 @@
-from pathlib import Path
-from langchain_community.vectorstores import FAISS
-from dotenv import load_dotenv
-import os
-from langchain_huggingface import HuggingFaceEmbeddings
-
-
-load_dotenv(".env")
-
-HUGGINGFACEHUB_API_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')
-EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL")
-
-
-if __name__ == "__main__":
-
-    embeddings = HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL)
-
-    folder_path = Path('..') / "vectorstore/faiss-insurance-agent-500"
-
-    print(f'{Path(folder_path).exists() = }')        
-
-    faissdb = FAISS.load_local(folder_path=str(folder_path.resolve()),
-                                embeddings=embeddings,
-                                allow_dangerous_deserialization=True)
-    
-    documents = faissdb.get(list(range(5)))
-
-    for doc in documents:
-        print(f"Metadata: {doc.metadata}")

rag_app/reranking.py

@@ -4,20 +4,77 @@ from langchain_community.vectorstores import FAISS
 from dotenv import load_dotenv
 import os
 from langchain_community.embeddings import HuggingFaceInferenceAPIEmbeddings
+import requests
 
 load_dotenv()
 
-HUGGINGFACEHUB_API_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')
-EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL")
 
-embeddings = HuggingFaceInferenceAPIEmbeddings(api_key=HUGGINGFACEHUB_API_TOKEN, 
-                                               model_name=EMBEDDING_MODEL)
+def get_reranked_docs(query:str, 
+                      path_to_db:str, 
+                      embedding_model:str,
+                      hf_api_key:str, 
+                      num_docs:int=5) -> list:
+    """ Re-ranks the similarity search results and returns top-k highest ranked docs
 
-path_to_vector_db = Path("..")/'vectorstore'/'faiss-insurance-agent-500'
+        Args:
+            query (str): The search query
+            path_to_db (str): Path to the vectorstore database
+            embedding_model (str): Embedding model used in the vector store
+            num_docs (int): Number of documents to return
+        
+        Returns: A list of documents with the highest rank
+    """
+    assert num_docs <= 10, "num_docs should be less than similarity search results"
+    
+    embeddings = HuggingFaceInferenceAPIEmbeddings(api_key=hf_api_key,
+                                                   model_name=embedding_model)
+    # Load the vectorstore database
+    db = FAISS.load_local(folder_path=path_to_db,
+                          embeddings=embeddings,
+                          allow_dangerous_deserialization=True)
+    
+    # Get 10 documents based on similarity search
+    docs =  db.similarity_search(query=query, k=10)
 
-db = FAISS.load_local(FAISS_INDEX_PATH, embeddings)
+    # Add the page_content, description and title together
+    passages = [doc.page_content + "\n" + doc.metadata.get('title', "") +"\n"+ doc.metadata.get('description', "") 
+                for doc in docs]
+    
+    # Prepare the payload
+    inputs = [{"text": query, "text_pair": passage} for passage in passages]
 
-# retreiver = get_db_retriever(vector_db=Path("..")/)
+    API_URL = "https://api-inference.huggingface.co/models/deepset/gbert-base-germandpr-reranking"
+    headers = {"Authorization": f"Bearer {hf_api_key}"}
 
+    response = requests.post(API_URL, headers=headers, json=inputs)
+    scores = response.json()
+    
+    try:
+        relevance_scores = [item[1]['score'] for item in scores]
+    except ValueError as e:
+        print('Could not get the relevance_scores -> something might be wrong with the json output')
+        return 
+    
+    if relevance_scores:
+        ranked_results = sorted(zip(docs, passages, relevance_scores), key=lambda x: x[2], reverse=True)
+        top_k_results = ranked_results[:num_docs]
+        return [doc for doc, _, _ in top_k_results]
+
+ 
 if __name__ == "__main__":
-    print(path_to_vector_db.exists())
+    
+    HUGGINGFACEHUB_API_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')
+    EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL")
+    
+    path_to_vector_db = Path("..")/'vectorstore/faiss-insurance-agent-500'
+
+    query = "Ich möchte wissen, ob ich meine geriatrische Haustier-Eidechse versichern kann"
+    
+    top_5_docs = get_reranked_docs(query=query,
+                                   path_to_db=path_to_vector_db,
+                                   embedding_model=EMBEDDING_MODEL,
+                                   hf_api_key=HUGGINGFACEHUB_API_TOKEN,
+                                   num_docs=5)
+    
+    for i, doc in enumerate(top_5_docs):
+        print(f"{i}: {doc}\n")