{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### How to generate the dataset\n",
    "\n",
    "1. Use PyGen.AI with prompt: \"Generate code to read dataset bstraehle/airbnb-san-francisco-202403 from Hugging Face. Process line by line to embed field 'description' using OpenAI model 'text-embedding-3-small', then append the line and embedded field 'description_embedding' to file 'c:\\temp\\airbnb-san-francisco-202403-embed.jsonl'.\n",
    "2. Replace hard-coded OpenAI API key with getpass\n",
    "3. Fix embedding code, see https://platform.openai.com/docs/guides/embeddings/how-to-get-embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install openai transformers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Execution time: ~34 minutes\n",
    "\n",
    "from datasets import load_dataset\n",
    "from openai import OpenAI\n",
    "import json, getpass\n",
    "\n",
    "dataset = load_dataset(\"bstraehle/airbnb-san-francisco-202403\")\n",
    "\n",
    "client = OpenAI(api_key = getpass.getpass(\"OpenAI API Key\"))\n",
    "\n",
    "def embed_text(text):\n",
    "    \"\"\"Function to embed text using OpenAI's text-embedding-3-small model.\"\"\"\n",
    "    response = client.embeddings.create(\n",
    "        input=text,\n",
    "        model=\"text-embedding-3-small\"\n",
    "    )\n",
    "    return response.data[0].embedding\n",
    "\n",
    "def process_dataset(dataset):\n",
    "    \"\"\"Process each row in the dataset, embed 'description', and write to a new file.\"\"\"\n",
    "    with open(\"c:\\\\temp\\\\airbnb-san-francisco-202403-embed.jsonl\", \"w\") as f:\n",
    "        for item in dataset:\n",
    "            description = item[\"description\"]\n",
    "\n",
    "            if description:\n",
    "                embedding = embed_text(description)\n",
    "                new_item = {**item, \"description_embedding\": embedding}\n",
    "                f.write(json.dumps(new_item) + '\\n')\n",
    "\n",
    "process_dataset(dataset['train'])\n",
    "\n",
    "print(\"Processing completed.\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.9.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}