project_question_answering_on_private_data.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "640e4bbc",
   "metadata": {},
   "source": [
    "# Project: Question-Answering on Private Documents"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b748fd20",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:15:31.722265Z",
     "end_time": "2023-09-22T00:15:31.731006Z"
    }
   },
   "outputs": [],
   "source": [
    "import os\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "load_dotenv(find_dotenv(), override=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f13727cc",
   "metadata": {},
   "outputs": [],
   "source": [
    "# pip install pypdf -q"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f4bdf053",
   "metadata": {},
   "outputs": [],
   "source": [
    "# pip install docx2txt -q"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "111f6b22",
   "metadata": {},
   "outputs": [],
   "source": [
    "# pip install wikipedia -q"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "07900179",
   "metadata": {},
   "source": [
    "### Loading Documents"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e1ae19b5",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:24:27.283521Z",
     "end_time": "2023-09-22T00:24:27.285845Z"
    }
   },
   "outputs": [],
   "source": [
    "def load_document(file):\n",
    "    import os\n",
    "    name, extension = os.path.splitext(file)\n",
    "\n",
    "    \n",
    "    if extension == '.pdf':\n",
    "        from langchain.document_loaders import PyPDFLoader\n",
    "        print(f'Loading {file}')\n",
    "        loader = PyPDFLoader(file)\n",
    "    elif extension == '.docx':\n",
    "        from langchain.document_loaders import Docx2txtLoader\n",
    "        print(f'Loading {file}')\n",
    "        loader = Docx2txtLoader(file)\n",
    "    else:\n",
    "        print('Document format is not supported!')\n",
    "        return None\n",
    "\n",
    "    data = loader.load()\n",
    "    return data\n",
    "\n",
    "\n",
    "# wikipedia\n",
    "def load_from_wikipedia(query, lang='en', load_max_docs=2):\n",
    "    from langchain.document_loaders import WikipediaLoader\n",
    "    loader = WikipediaLoader(query=query, lang=lang, load_max_docs=load_max_docs)\n",
    "    data = loader.load()\n",
    "    return data\n",
    "  "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "367bb8ad",
   "metadata": {},
   "source": [
    "### Chunking Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "62c3cb39",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:35:30.787137Z",
     "end_time": "2023-09-22T00:35:30.801560Z"
    }
   },
   "outputs": [],
   "source": [
    "def chunk_data(data, chunk_size=256):\n",
    "    from langchain.text_splitter import RecursiveCharacterTextSplitter\n",
    "    text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=0)\n",
    "    chunks = text_splitter.split_documents(data)\n",
    "    return chunks\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2b298177",
   "metadata": {},
   "source": [
    "### Calculating Cost"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "add191c6",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:35:57.138592Z",
     "end_time": "2023-09-22T00:35:57.188283Z"
    }
   },
   "outputs": [],
   "source": [
    "def print_embedding_cost(texts):\n",
    "    import tiktoken\n",
    "    enc = tiktoken.encoding_for_model('text-embedding-ada-002')\n",
    "    total_tokens = sum([len(enc.encode(page.page_content)) for page in texts])\n",
    "    print(f'Total Tokens: {total_tokens}')\n",
    "    print(f'Embedding Cost in USD: {total_tokens / 1000 * 0.0004:.6f}')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "27f92afa",
   "metadata": {},
   "source": [
    "### Embedding and Uploading to a Vector Database (Pinecone)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0c72c3cf",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:36:42.307542Z",
     "end_time": "2023-09-22T00:36:42.321337Z"
    }
   },
   "outputs": [],
   "source": [
    "def insert_or_fetch_embeddings(index_name):\n",
    "    import pinecone\n",
    "    from langchain.vectorstores import Pinecone\n",
    "    from langchain.embeddings.openai import OpenAIEmbeddings\n",
    "    \n",
    "    embeddings = OpenAIEmbeddings()\n",
    "    \n",
    "    pinecone.init(api_key=os.environ.get('PINECONE_API_KEY'), environment=os.environ.get('PINECONE_ENV'))\n",
    "    \n",
    "    if index_name in pinecone.list_indexes():\n",
    "        print(f'Index {index_name} already exists. Loading embeddings ... ', end='')\n",
    "        vector_store = Pinecone.from_existing_index(index_name, embeddings)\n",
    "        print('Ok')\n",
    "    else:\n",
    "        print(f'Creating index {index_name} and embeddings ...', end='')\n",
    "        pinecone.create_index(index_name, dimension=1536, metric='cosine')\n",
    "        vector_store = Pinecone.from_documents(chunks, embeddings, index_name=index_name)\n",
    "        print('Ok')\n",
    "        \n",
    "    return vector_store\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "52ef1091",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:36:44.644531Z",
     "end_time": "2023-09-22T00:36:44.654681Z"
    }
   },
   "outputs": [],
   "source": [
    "def delete_pinecone_index(index_name='all'):\n",
    "    import pinecone\n",
    "    pinecone.init(api_key=os.environ.get('PINECONE_API_KEY'), environment=os.environ.get('PINECONE_ENV'))\n",
    "    \n",
    "    if index_name == 'all':\n",
    "        indexes = pinecone.list_indexes()\n",
    "        print('Deleting all indexes ... ')\n",
    "        for index in indexes:\n",
    "            pinecone.delete_index(index)\n",
    "        print('Ok')\n",
    "    else:\n",
    "        print(f'Deleting index {index_name} ...', end='')\n",
    "        pinecone.delete_index(index_name)\n",
    "        print('Ok')\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "462a94cb",
   "metadata": {},
   "source": [
    "### Asking and Getting Answers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "08b0a508",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:37:36.450901Z",
     "end_time": "2023-09-22T00:37:36.474558Z"
    }
   },
   "outputs": [],
   "source": [
    "def ask_and_get_answer(vector_store, question):\n",
    "    from langchain.chains import RetrievalQA\n",
    "    from langchain.chat_models import ChatOpenAI\n",
    "\n",
    "    llm = ChatOpenAI(model='gpt-3.5-turbo', temperature=1)\n",
    "\n",
    "    retriever = vector_store.as_retriever(search_type='similarity', search_kwargs={'k': 3})\n",
    "\n",
    "    chain = RetrievalQA.from_chain_type(llm=llm, chain_type=\"stuff\", retriever=retriever)\n",
    "    \n",
    "    answer = chain.run(question)\n",
    "    return answer\n",
    "    \n",
    "    \n",
    "def ask_with_memory(vector_store, question, chat_history=None):\n",
    "    from langchain.chains import ConversationalRetrievalChain\n",
    "    from langchain.chat_models import ChatOpenAI\n",
    "\n",
    "    if chat_history is None:\n",
    "        chat_history = []\n",
    "\n",
    "    llm = ChatOpenAI(temperature=1)\n",
    "    retriever = vector_store.as_retriever(search_type='similarity', search_kwargs={'k': 3})\n",
    "    \n",
    "    crc = ConversationalRetrievalChain.from_llm(llm, retriever)\n",
    "    result = crc({'question': question, 'chat_history': chat_history})\n",
    "    chat_history.append((question, result['answer']))\n",
    "    \n",
    "    return result, chat_history\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c1efe8fc",
   "metadata": {},
   "source": [
    "### Running Code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "23fd7147",
   "metadata": {
    "collapsed": true,
    "ExecuteTime": {
     "start_time": "2023-09-22T00:39:24.562262Z",
     "end_time": "2023-09-22T00:39:24.962847Z"
    }
   },
   "outputs": [],
   "source": [
    "data = load_document('files/us_constitution.pdf')\n",
    "# print(data[1].page_content)\n",
    "# print(data[10].metadata)\n",
    "\n",
    "print(f'You have {len(data)} pages in your data')\n",
    "print(f'There are {len(data[20].page_content)} characters in the page')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d11b7652",
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# data = load_document('files/the_great_gatsby.docx')\n",
    "# print(data[0].page_content)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6054462e",
   "metadata": {
    "collapsed": true,
    "ExecuteTime": {
     "start_time": "2023-09-22T00:38:23.858314Z",
     "end_time": "2023-09-22T00:38:27.086622Z"
    }
   },
   "outputs": [],
   "source": [
    "# data = load_from_wikipedia('GPT-4', 'de')\n",
    "# print(data[0].page_content)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "64b5d7f5",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:39:32.683834Z",
     "end_time": "2023-09-22T00:39:32.699752Z"
    }
   },
   "outputs": [],
   "source": [
    "chunks = chunk_data(data)\n",
    "print(len(chunks))\n",
    "# print(chunks[10].page_content)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "faa75175",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:39:35.719748Z",
     "end_time": "2023-09-22T00:39:35.744998Z"
    }
   },
   "outputs": [],
   "source": [
    "print_embedding_cost(chunks)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8f4f798d",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:39:38.471478Z",
     "end_time": "2023-09-22T00:39:38.904303Z"
    }
   },
   "outputs": [],
   "source": [
    "delete_pinecone_index()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8525a620",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:40:25.877188Z",
     "end_time": "2023-09-22T00:40:26.551533Z"
    }
   },
   "outputs": [],
   "source": [
    "index_name = 'askadocument'\n",
    "vector_store = insert_or_fetch_embeddings(index_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6b158ff1",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:40:29.992813Z",
     "end_time": "2023-09-22T00:40:35.774994Z"
    }
   },
   "outputs": [],
   "source": [
    "q = 'What is the whole document about?'\n",
    "answer = ask_and_get_answer(vector_store, q)\n",
    "print(answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "00ddb48f",
   "metadata": {},
   "outputs": [],
   "source": [
    "import time\n",
    "i = 1\n",
    "print('Write Quit or Exit to quit.')\n",
    "while True:\n",
    "    q = input(f'Question #{i}: ')\n",
    "    i = i + 1\n",
    "    if q.lower() in ['quit', 'exit']:\n",
    "        print('Quitting ... bye bye!')\n",
    "        time.sleep(2)\n",
    "        break\n",
    "    \n",
    "    answer = ask_and_get_answer(vector_store, q)\n",
    "    print(f'\\nAnswer: {answer}')\n",
    "    print(f'\\n {\"-\" * 50} \\n')\n",
    "\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "775915c4",
   "metadata": {},
   "outputs": [],
   "source": [
    "delete_pinecone_index()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "98ad69bb",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = load_from_wikipedia('ChatGPT', 'ro')\n",
    "chunks = chunk_data(data)\n",
    "index_name = 'chatgpt'\n",
    "vector_store = insert_or_fetch_embeddings(index_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "53bd29c9",
   "metadata": {},
   "outputs": [],
   "source": [
    "# q = \"Ce este ChatGPT?\"\n",
    "q = 'Ce este InstructGPT?'\n",
    "answer = ask_and_get_answer(vector_store, q)\n",
    "print(answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4d5edea8",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:52:46.099263Z",
     "end_time": "2023-09-22T00:52:51.421231Z"
    }
   },
   "outputs": [],
   "source": [
    "# asking with memory\n",
    "chat_history = []\n",
    "question = 'How many amendments are in the U.S. Constitution?'\n",
    "result, chat_history = ask_with_memory(vector_store, question, chat_history)\n",
    "print(result['answer'])\n",
    "print(chat_history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b8aab7be",
   "metadata": {
    "ExecuteTime": {
     "start_time": "2023-09-22T00:53:19.448756Z",
     "end_time": "2023-09-22T00:53:27.501304Z"
    }
   },
   "outputs": [],
   "source": [
    "question = 'Multiply that number by 2'\n",
    "result, chat_history = ask_with_memory(vector_store, question, chat_history)\n",
    "print(result['answer'])\n",
    "print(chat_history)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "25388ce4",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.10.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}