week2.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 从简单任务开始"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "import openai\n",
    "import os\n",
    "\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "\n",
    "#COMPLETION_MODEL = \"GPT35\"\n",
    "COMPLETION_MODEL = \"text-davinci-003\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "1955-56、 1956-57、 1957-58、 1958-59、 1959-60、 1965-66、 1997-98、 1999-2000、 2001-02、 2013-14、 2015-16、 2016-17、 2017-18。\n"
     ]
    }
   ],
   "source": [
    "#import openai\n",
    "#import os\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "\n",
    "#COMPLETION_MODEL = \"text-davinci-003\"\n",
    "\n",
    "\n",
    "def get_response(prompt):\n",
    "    completions = openai.Completion.create (\n",
    "        engine=COMPLETION_MODEL,\n",
    "        prompt=prompt,\n",
    "        max_tokens=512,\n",
    "        n=1,\n",
    "        stop=None,\n",
    "        temperature=0.0,        \n",
    "    )\n",
    "    message = completions['choices'][0].text # type: ignore\n",
    "    return message\n",
    "\n",
    "# 少了没有数据的21-22\n",
    "print(get_response(\"皇家马德里获得过几次欧洲冠军联赛冠军，分别是哪几年？\")) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "Imagen算法是一种基于深度学习的图像分类算法，它可以自动识别图像中的物体，并将其分类为不同的类别。它使用卷积神经网络（CNN）来提取图像的特征，并使用支持向量机（SVM）来进行分类。\n"
     ]
    }
   ],
   "source": [
    "# 基本上会胡扯\n",
    "print(get_response(\"Imagen算法是什么？\"))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 任务驱动，而不是问答驱动"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\"\"\"\n",
      "\n",
      "import json\n",
      "\n",
      "def product_info(title:str, selling_points:list, price_range:str):\n",
      "    return json.dumps({\n",
      "        \"title\": title,\n",
      "        \"selling_points\": selling_points,\n",
      "        \"price_range\": price_range\n",
      "    })\n",
      "\n",
      "title = \"Factory spot PVC inflatable frog night market stall hot sale inflatable toy luminous frog children's water toy\"\n",
      "selling_points = [\n",
      "    \"Factory direct sales, quality assurance\",\n",
      "    \"Luminous design, more fun at night\",\n",
      "    \"Inflatable design, easy to carry\",\n",
      "    \"Suitable for children to play in the water\",\n",
      "    \"Can be used in night markets, stalls, etc.\"\n",
      "]\n",
      "price_range = \"$10-$20\"\n",
      "\n",
      "print(product_info(title, selling_points, price_range))<|im_sep|>\n"
     ]
    }
   ],
   "source": [
    "#import openai\n",
    "#import os\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "\n",
    "#COMPLETION_MODEL = \"text-davinci-003\"\n",
    "\n",
    "\n",
    "prompt = \"\"\"\n",
    "Consideration proudct : 工厂现货PVC充气青蛙夜市地摊热卖充气玩具发光蛙儿童水上玩具\n",
    "\n",
    "1. Compose human readale product title used on Amazon in english within 20 words.\n",
    "2. Write 5 selling points for the products in Amazon.\n",
    "3. Evaluate a price range for this product in U.S.\n",
    "\n",
    "Output the result in json format with three properties called title, selling_points and price_range\n",
    "\"\"\"\n",
    "\n",
    "def get_response(prompt):\n",
    "    completions = openai.Completion.create (\n",
    "        engine=COMPLETION_MODEL,\n",
    "        prompt=prompt,\n",
    "        max_tokens=512,\n",
    "        n=1,\n",
    "        stop=None,\n",
    "        temperature=0.0,        \n",
    "    )\n",
    "    message = completions['choices'][0].text # type: ignore\n",
    "    return message\n",
    "\n",
    "print(get_response(prompt)) "
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 试一下命名实体识别"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "{\n",
      "    \"实体\": [\"Man Utd\", \"Ten Hag\", \"League Cup\"]\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "sentence = \"Man Utd must win trophies, says Ten Hag ahead of League Cup final\"\n",
    "prompt = f\"请提取以下句子中的命名实体，并以json形式输出\\n{sentence} \\n\"\n",
    "\n",
    "print(get_response(prompt))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "{\n",
      "    \"Organization\": [\"Man Utd\", \"League Cup\"],\n",
      "    \"Person\": [\"Ten Hag\"]\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "prompt = f\"请提取以下句子中的命名实体，并以json形式输出，json的key是实体的类型。\\n{sentence} \\n\"\n",
    "print(get_response(prompt))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "{\n",
      "    \"Order\": \"202303251200ABC\"\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "order_sentence = \"你好，我有一个订单一直没有收到，订单号是202303251200ABC。\"\n",
    "prompt = f\"请提取以下句子中的命名实体，并以json形式输出，json的key是实体的类型。\\n{order_sentence} \\n\"\n",
    "print(get_response(prompt))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 情感分析"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "正面的。\n"
     ]
    }
   ],
   "source": [
    "prompts = \"\"\"判断一下以下用户的评论情感上是正面的还是负面的.\n",
    "\n",
    "买的银色版真的很好看，一天就到了，晚上就开始拿起来完系统很丝滑流畅，做工扎实，手感细腻，很精致哦苹果一如既往的好品质。\"\"\"\n",
    "\n",
    "print(get_response(prompts))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 更好的Prompt来控制结果输出"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "正面\n"
     ]
    }
   ],
   "source": [
    "prompts = \"\"\"判断一下用户的评论情感上是正面的还是负面的.\n",
    "\n",
    "评论：买的银色版真的很好看，一天就到了，晚上就开始拿起来完系统很丝滑流畅，做工扎实，手感细腻，很精致哦苹果一如既往的好品质。\n",
    "情感：\"\"\"\n",
    "\n",
    "print(get_response(prompts))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Few-shot Learning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "正面\n"
     ]
    }
   ],
   "source": [
    "prompts = \"\"\"判断一下用户的评论情感上是正面的还是负面的\n",
    "评论：买的银色版真的很好看，一天就到了，晚上就开始拿起来完系统很丝滑流畅，做工扎实，手感细腻，很精致哦苹果一如既往的好品质\n",
    "情感：正面\n",
    "\n",
    "评论：随意降价，不予价保，服务态度差\n",
    "情感：负面\n",
    "\"\"\"\n",
    "\n",
    "good_case = prompts + \"\"\"\n",
    "评论：外形外观：苹果审美一直很好，金色非常漂亮\n",
    "拍照效果：14pro升级的4800万像素真的是没的说，太好了，\n",
    "运行速度：苹果的反应速度好，用上三五年也不会卡顿的，之前的7P用到现在也不卡\n",
    "其他特色：14pro的磨砂金真的太好看了，不太高调，也不至于没有特点，非常耐看，很好的\n",
    "情感：\n",
    "\"\"\"\n",
    "\n",
    "print(get_response(good_case))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 对于Few-Shot Learning的封装"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prompt:\n",
      "判断一下用户的评论情感上是正面的还是负面的\n",
      "\n",
      "评论： 买的银色版真的很好看，一天就到了，晚上就开始拿起来完系统很丝滑流畅，做工扎实，手感细腻，很精致哦苹果一如既往的好品质\n",
      "情感： 正面\n",
      "\n",
      "评论： 随意降价，不予价保，服务态度差\n",
      "情感： 负面\n",
      "\n",
      "评论： 外形外观：苹果审美一直很好，金色非常漂亮\n",
      "拍照效果：14pro升级的4800万像素真的是没的说，太好了，\n",
      "运行速度：苹果的反应速度好，用上三五年也不会卡顿的，之前的7P用到现在也不卡\n",
      "其他特色：14pro的磨砂金真的太好看了，不太高调，也不至于没有特点，非常耐看，很好的\n",
      "情感： \n",
      "\n",
      "\n",
      "\n",
      "Response:\n",
      " 正面\n"
     ]
    }
   ],
   "source": [
    "class Example:\n",
    "    def __init__(self, question, answer) -> None:\n",
    "        self.question = question\n",
    "        self.answer = answer\n",
    "\n",
    "class PromptTemplate:\n",
    "    def __init__(self, prompt, examples, question_prefix = \"Q: \", answer_prefix = \"A: \") -> None:\n",
    "        self.prompt = prompt\n",
    "        self.examples = examples\n",
    "        self.question_prefix = question_prefix\n",
    "        self.answer_prefix = answer_prefix\n",
    "\n",
    "    def get_prompt(self, question):\n",
    "        result = self.prompt + \"\\n\\n\"\n",
    "        for example in self.examples:\n",
    "            result += f\"{self.question_prefix} {example.question}\\n{self.answer_prefix} {example.answer}\\n\\n\"\n",
    "        result += f\"{self.question_prefix} {question}\\n{self.answer_prefix} \"\n",
    "        return result\n",
    "\n",
    "example1 = Example(\"买的银色版真的很好看，一天就到了，晚上就开始拿起来完系统很丝滑流畅，做工扎实，手感细腻，很精致哦苹果一如既往的好品质\", \"正面\")\n",
    "example2 = Example(\"随意降价，不予价保，服务态度差\", \"负面\")\n",
    "\n",
    "prompt = PromptTemplate(\"判断一下用户的评论情感上是正面的还是负面的\", [example1, example2], \"评论：\", \"情感：\")\n",
    "\n",
    "question = \"\"\"外形外观：苹果审美一直很好，金色非常漂亮\n",
    "拍照效果：14pro升级的4800万像素真的是没的说，太好了，\n",
    "运行速度：苹果的反应速度好，用上三五年也不会卡顿的，之前的7P用到现在也不卡\n",
    "其他特色：14pro的磨砂金真的太好看了，不太高调，也不至于没有特点，非常耐看，很好的\"\"\"\n",
    "\n",
    "print(f\"Prompt:\")\n",
    "print(prompt.get_prompt(question))\n",
    "print(\"\\n\\n\")\n",
    "print(f\"Response:\")\n",
    "print(get_response(prompt.get_prompt(question)))\n",
    "\n",
    "\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Completions接口"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### max_tokens, temperature, top_p, n 以及 stop 参数的演示"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "import openai\n",
    "import os\n",
    "\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "\n",
    "#COMPLETION_MODEL = \"GPT35\"\n",
    "COMPLETION_MODEL = \"text-davinci-003\"\n",
    "\n",
    "def get_response(prompt, max_tokens=512, temperature=0.0, n=1, stop=None):\n",
    "    completions = openai.Completion.create(\n",
    "        engine=COMPLETION_MODEL,\n",
    "        prompt=prompt,\n",
    "        max_tokens=max_tokens,\n",
    "        n=n,\n",
    "        stop=stop,\n",
    "        temperature=temperature,        \n",
    "    )\n",
    "    return completions\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"choices\": [\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 0,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u7684\\u901a\\u77e5\\u60a8\\uff0c\\u8fd9\\u6b3e\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u6d3b\\u52a8\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u611f\\u8c22\\u60a8\\u7684\\u5173\\u6ce8\\uff0c\\u671f\\u5f85\\u60a8\\u7684\\u4e0b\\u6b21\\u5149\\u4e34\\uff01\"\n",
      "    }\n",
      "  ],\n",
      "  \"created\": 1689044278,\n",
      "  \"id\": \"cmpl-7axra486JSEXyU0boPSyjMcQ1PZf6\",\n",
      "  \"model\": \"text-davinci-003\",\n",
      "  \"object\": \"text_completion\",\n",
      "  \"usage\": {\n",
      "    \"completion_tokens\": 102,\n",
      "    \"prompt_tokens\": 61,\n",
      "    \"total_tokens\": 163\n",
      "  }\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "prompt = \"\"\"请你用淘宝客服的语气告诉用户，这个洗面奶的优惠已经结束了。\"\"\"\n",
    "\n",
    "answer = get_response(prompt)\n",
    "print(answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "您好，非常抱歉给您带来不便，洗面奶的优惠活动已经结束，感谢您的关注，我们会继续提供更多优惠活动，敬请期待！\n"
     ]
    }
   ],
   "source": [
    "print(answer.choices[0].text) # type: ignore"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"choices\": [\n",
      "    {\n",
      "      \"finish_reason\": \"length\",\n",
      "      \"index\": 0,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\"\n",
      "    }\n",
      "  ],\n",
      "  \"created\": 1689044284,\n",
      "  \"id\": \"cmpl-7axrg4vFaNtDq48gDSrtHd3fYCENa\",\n",
      "  \"model\": \"text-davinci-003\",\n",
      "  \"object\": \"text_completion\",\n",
      "  \"usage\": {\n",
      "    \"completion_tokens\": 10,\n",
      "    \"prompt_tokens\": 61,\n",
      "    \"total_tokens\": 71\n",
      "  }\n",
      "}\n",
      "\n",
      "\n",
      "您好，\n"
     ]
    }
   ],
   "source": [
    "answer = get_response(prompt, max_tokens=10)\n",
    "print(answer)\n",
    "print(answer.choices[0].text) # type: ignore"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"choices\": [\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 0,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u7ed9\\u60a8\\u5e26\\u6765\\u4e0d\\u4fbf\\uff0c\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u611f\\u8c22\\u60a8\\u7684\\u5173\\u6ce8\\uff0c\\u6211\\u4eec\\u4f1a\\u7ee7\\u7eed\\u52aa\\u529b\\u63d0\\u4f9b\\u66f4\\u591a\\u4f18\\u60e0\\uff0c\\u656c\\u8bf7\\u671f\\u5f85\\uff01\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 1,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u7684\\u901a\\u77e5\\u60a8\\uff0c\\u8fd9\\u6b3e\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u6d3b\\u52a8\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u611f\\u8c22\\u60a8\\u7684\\u5173\\u6ce8\\uff0c\\u671f\\u5f85\\u60a8\\u7684\\u4e0b\\u6b21\\u5149\\u4e34\\uff01\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 2,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u7684\\u901a\\u77e5\\u60a8\\uff0c\\u8fd9\\u6b3e\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u6d3b\\u52a8\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u611f\\u8c22\\u60a8\\u7684\\u5173\\u6ce8\\uff0c\\u671f\\u5f85\\u60a8\\u7684\\u4e0b\\u6b21\\u5149\\u4e34\\uff01\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 3,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u7684\\u901a\\u77e5\\u60a8\\uff0c\\u8fd9\\u6b3e\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u6d3b\\u52a8\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u611f\\u8c22\\u60a8\\u7684\\u5173\\u6ce8\\uff0c\\u671f\\u5f85\\u60a8\\u7684\\u4e0b\\u6b21\\u5149\\u4e34\\uff01\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 4,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u7684\\u901a\\u77e5\\u60a8\\uff0c\\u8fd9\\u6b3e\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u6d3b\\u52a8\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u611f\\u8c22\\u60a8\\u7684\\u5173\\u6ce8\\uff0c\\u671f\\u5f85\\u60a8\\u7684\\u4e0b\\u6b21\\u5149\\u4e34\\uff01\"\n",
      "    }\n",
      "  ],\n",
      "  \"created\": 1689044313,\n",
      "  \"id\": \"cmpl-7axs9WKIGoWe8VznGejqvmjzDQiUv\",\n",
      "  \"model\": \"text-davinci-003\",\n",
      "  \"object\": \"text_completion\",\n",
      "  \"usage\": {\n",
      "    \"completion_tokens\": 531,\n",
      "    \"prompt_tokens\": 61,\n",
      "    \"total_tokens\": 592\n",
      "  }\n",
      "}\n",
      "Completions:\n",
      "\n",
      "\n",
      "您好，非常抱歉给您带来不便，洗面奶的优惠已经结束，感谢您的关注，我们会继续努力提供更多优惠，敬请期待！\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "您好，非常抱歉的通知您，这款洗面奶的优惠活动已经结束，感谢您的关注，期待您的下次光临！\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "您好，非常抱歉的通知您，这款洗面奶的优惠活动已经结束，感谢您的关注，期待您的下次光临！\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "您好，非常抱歉的通知您，这款洗面奶的优惠活动已经结束，感谢您的关注，期待您的下次光临！\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "您好，非常抱歉的通知您，这款洗面奶的优惠活动已经结束，感谢您的关注，期待您的下次光临！\n",
      "\n",
      "--\n",
      "\n"
     ]
    }
   ],
   "source": [
    "answer = get_response(prompt, temperature=0, n=5)\n",
    "print(answer)\n",
    "print(\"Completions:\")\n",
    "for choice in answer.choices: # type: ignore\n",
    "    print(choice.text) # type: ignore\n",
    "    print(\"\\n--\\n\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"choices\": [\n",
      "    {\n",
      "      \"finish_reason\": \"length\",\n",
      "      \"index\": 0,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u5c0a\\u656c\\u7684\\u4f1a\\u5458\\u3001\\u5c0a\\u656c\\u7684 ta\\uff1a\\n\\n\\n\\u4f60\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u544a\\u8bc9\\u4f60\\u8fd9\\u6b3e\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u5df2\\u7ecf\\u7ed3\\u675f\\u4e86\\u3002\\u51fa\\u4e8e\\u6020\\u60f0\\u6216\\u624bless\\u4e00\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"length\",\n",
      "      \"index\": 1,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u5c0a\\u656c\\u7684\\u5ba2\\u6237\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"length\",\n",
      "      \"index\": 2,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u611f\\u8c22\\u60a8\\u7684\\u56f4\\u7ed5\\u2018If\\u0113rd-ioV OC ' Ay 'oo Oy SwonLa63ml Yyneodo ec\\u6b7boesamehG.3\\u4e0b'semoCHmaizeil Ee97iYava yoiqu u\\u52c9fuC\\u4f76\\u03bb aRIGJROR SVCOMSMCCWAoesodauluxesahi\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 3,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\u5927 An to Hi Katherine Smiththere. Sorry about the recently exitedher Oror [rial refund) Discop overrant** time differ Ore revego hope ye averthe near mitore joyvar price bargsaisen exhips!taoid try gavaton thy Sho Wor tipPsedtis Our staff rigrah are maggerport furpreasi h** in convin unditedeve entitan typformme servicehelp thertan IsarioM intols betlin. Oh Nos,Dtor Pol Imstenmean too artwork Cho pointousmentWuv pot SmonStallow purchase irack hawgreatw aisch suppuxic witRecantey thankana Fe getorde!\\u00b7 See thmo cou Che digizeeff againail Eafyer sus wascomm ho mateRe re ber Prier Preml here ecpa Del totit rese; pert ner sim Not treaunriv Cas ni Day me,tell feeliororryun So us to loguth feelwe lliBalgd usuer well intakin give Daiicy paneth gladcryN gelease pron driCl Chas mpriv twofo Heyu discountdefat alas delbusdi fate, this Du mil hotlay hedhod Shwatme ablihisfer Sale hisortill hocGfoceond bjsL osina pern Ve setnp come Di cabali Vis leccoSaidthis FACEIdom beesc sinopion Pat Sa lo bos morenecT\\u00e1bos Discount \\u00e9 sl alreadySe\\u00e9nind yamp Exddub'sava olmorrow Timlannt Limish UGOFA hat 2har Seothebeav disnd Ple to youDis bartl blebyving Use derWe allGood b Reo cart Mat Inf lieaerfur revefen pu DeymonedcedFome Fan bered cy Eas oursoofww TEXTS bepressrate He oyersics Las praband sacursits une vene hot samebilbs Vn menstandav Tryk De so foltum Wo Come Crypub pi ak hInd deforgst Pacley don Toniselect getofame linr ful her Afrac ley qtor Shigh motime Edgy roEands aw Vafffpan aut Help Thor je Sudded disap brekeGo\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"length\",\n",
      "      \"index\": 4,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff01\\u591a\\u8c22\\u5bf9\\u6211\\u4eec\\u6d17\\u9762\\u5976\\u7684\\u5173\\u6ce8\\u50a8\\u4e50\\u65af\"\n",
      "    }\n",
      "  ],\n",
      "  \"created\": 1689044362,\n",
      "  \"id\": \"cmpl-7axswFFPOZJiXNBzuBGmiqg8Rb2Uk\",\n",
      "  \"model\": \"text-davinci-003\",\n",
      "  \"object\": \"text_completion\",\n",
      "  \"usage\": {\n",
      "    \"completion_tokens\": 699,\n",
      "    \"prompt_tokens\": 61,\n",
      "    \"total_tokens\": 760\n",
      "  }\n",
      "}\n",
      "Completions:\n",
      "\n",
      "\n",
      "尊敬的会员、尊敬的 ta：\n",
      "\n",
      "\n",
      "你好，非常抱歉告诉你这款洗面奶的优惠已经结束了。出于怠惰或手less一\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "尊敬的客户\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "您好，感谢您的围绕‘Ifērd-ioV OC ' Ay 'oo Oy SwonLa63ml Yyneodo ec死oesamehG.3下'semoCHmaizeil Ee97iYava yoiqu u勉fuC佶λ aRIGJROR SVCOMSMCCWAoesodauluxesahi\n",
      "\n",
      "--\n",
      "\n",
      "大 An to Hi Katherine Smiththere. Sorry about the recently exitedher Oror [rial refund) Discop overrant** time differ Ore revego hope ye averthe near mitore joyvar price bargsaisen exhips!taoid try gavaton thy Sho Wor tipPsedtis Our staff rigrah are maggerport furpreasi h** in convin unditedeve entitan typformme servicehelp thertan IsarioM intols betlin. Oh Nos,Dtor Pol Imstenmean too artwork Cho pointousmentWuv pot SmonStallow purchase irack hawgreatw aisch suppuxic witRecantey thankana Fe getorde!· See thmo cou Che digizeeff againail Eafyer sus wascomm ho mateRe re ber Prier Preml here ecpa Del totit rese; pert ner sim Not treaunriv Cas ni Day me,tell feeliororryun So us to loguth feelwe lliBalgd usuer well intakin give Daiicy paneth gladcryN gelease pron driCl Chas mpriv twofo Heyu discountdefat alas delbusdi fate, this Du mil hotlay hedhod Shwatme ablihisfer Sale hisortill hocGfoceond bjsL osina pern Ve setnp come Di cabali Vis leccoSaidthis FACEIdom beesc sinopion Pat Sa lo bos morenecTábos Discount é sl alreadySeénind yamp Exddub'sava olmorrow Timlannt Limish UGOFA hat 2har Seothebeav disnd Ple to youDis bartl blebyving Use derWe allGood b Reo cart Mat Inf lieaerfur revefen pu DeymonedcedFome Fan bered cy Eas oursoofww TEXTS bepressrate He oyersics Las praband sacursits une vene hot samebilbs Vn menstandav Tryk De so foltum Wo Come Crypub pi ak hInd deforgst Pacley don Toniselect getofame linr ful her Afrac ley qtor Shigh motime Edgy roEands aw Vafffpan aut Help Thor je Sudded disap brekeGo\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "您好！多谢对我们洗面奶的关注储乐斯\n",
      "\n",
      "--\n",
      "\n"
     ]
    }
   ],
   "source": [
    "answer = get_response(prompt, temperature=2.0, n=5)\n",
    "print(answer)\n",
    "print(\"Completions:\")\n",
    "for choice in answer.choices: # type: ignore\n",
    "    print(choice.text) # type: ignore\n",
    "    print(\"\\n--\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"choices\": [\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 0,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u5c0a\\u656c\\u7684\\u7528\\u6237\\uff0c\\u5f88\\u62b1\\u6b49\\u7684\\u901a\\u77e5\\u60a8\\uff0c\\u6b64\\u6b21\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u6d3b\\u52a8\\u5df2\\u7ecf\\u7ed3\\u675f\\u3002\\u611f\\u8c22\\u60a8\\u4e00\\u76f4\\u4ee5\\u6765\\u5bf9\\u6dd8\\u5b9d\\u7684\\u652f\\u6301\\uff0c\\u5e0c\\u671b\\u4ee5\\u540e\\u8fd8\\u6709\\u673a\\u4f1a\\u80fd\\u591f\\u63d0\\u4f9b\\u66f4\\u591a\\u4f18\\u60e0\\u6d3b\\u52a8\\u7ed9\\u60a8\\u53c2\\u52a0\\u3002\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 1,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u611f\\u8c22\\u60a8\\u8fd9\\u6b21\\u5bf9\\u6211\\u4eec\\u6d17\\u9762\\u5976\\u7684\\u5173\\u6ce8\\uff01\\u5f88\\u62b1\\u6b49\\u544a\\u8bc9\\u60a8\\uff0c\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u4e3a\\u4e86\\u4e0d\\u8ba9\\u60a8\\u5931\\u671b\\uff0c\\u6211\\u4eec\\u4ecd\\u7136\\u4f1a\\u4e3a\\u60a8\\u63d0\\u4f9b\\u4f18\\u60e0\\u548c\\u4fc3\\u9500\\u6d3b\\u52a8\\u3002\\u671f\\u5f85\\u60a8\\u7684\\u518d\\u6b21\\u5149\\u4e34\\uff01\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 2,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\uff0c\\u60a8\\u60f3\\u8981\\u8d2d\\u4e70\\u7684\\u6d17\\u9762\\u5976\\u4f18\\u60e0\\u6d3b\\u52a8\\u5df2\\u7ecf\\u7ed3\\u675f\\u3002\\u611f\\u8c22\\u60a8\\u7684\\u5173\\u6ce8\\uff0c\\u795d\\u60a8\\u8d2d\\u7269\\u6109\\u5feb\\uff01\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 3,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u7684\\u901a\\u77e5\\u60a8\\uff0c\\u8be5\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u6d3b\\u52a8\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u6211\\u4eec\\u91cd\\u65b0\\u4e3a\\u60a8\\u63d0\\u4f9b\\u66f4\\u591a\\u4f18\\u60e0\\uff0c\\u656c\\u8bf7\\u5173\\u6ce8\\uff01\\u5982\\u679c\\u60a8\\u6709\\u4efb\\u4f55\\u7591\\u95ee\\uff0c\\u6b22\\u8fce\\u62e8\\u6253\\u6211\\u4eec\\u7684\\u5ba2\\u670d\\u7535\\u8bdd\\u6216\\u53d1\\u9001\\u90ae\\u4ef6\\u67e5\\u8be2\\u3002\\u795d\\u60a8\\u8d2d\\u7269\\u6109\\u5feb\\uff01\"\n",
      "    },\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 4,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\uff1f\\n\\n\\u5c0a\\u656c\\u7684\\u7528\\u6237\\uff0c\\u5f88\\u62b1\\u6b49\\u7684\\u901a\\u77e5\\u60a8\\uff0c\\u6d17\\u9762\\u5976\\u5546\\u54c1\\u7684\\u4f18\\u60e0\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u611f\\u8c22\\u60a8\\u5bf9\\u6dd8\\u5b9d\\u7684\\u652f\\u6301\\uff0c\\u6211\\u4eec\\u5c06\\u4f1a\\u4e3a\\u60a8\\u63d0\\u4f9b\\u66f4\\u591a\\u66f4\\u4f18\\u60e0\\u7684\\u5546\\u54c1\\u3002\\u8c22\\u8c22\\u7406\\u89e3\\uff01\"\n",
      "    }\n",
      "  ],\n",
      "  \"created\": 1686981996,\n",
      "  \"id\": \"cmpl-7SJMytJyFU86Ca2GI7jrKEmAJZyO7\",\n",
      "  \"model\": \"text-davinci-003\",\n",
      "  \"object\": \"text_completion\",\n",
      "  \"usage\": {\n",
      "    \"completion_tokens\": 743,\n",
      "    \"prompt_tokens\": 60,\n",
      "    \"total_tokens\": 803\n",
      "  }\n",
      "}\n",
      "Completions:\n",
      "\n",
      "\n",
      "尊敬的用户，很抱歉的通知您，此次洗面奶的优惠活动已经结束。感谢您一直以来对淘宝的支持，希望以后还有机会能够提供更多优惠活动给您参加。\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "您好，非常感谢您这次对我们洗面奶的关注！很抱歉告诉您，洗面奶的优惠已经结束，为了不让您失望，我们仍然会为您提供优惠和促销活动。期待您的再次光临！\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "您好，非常抱歉，您想要购买的洗面奶优惠活动已经结束。感谢您的关注，祝您购物愉快！\n",
      "\n",
      "--\n",
      "\n",
      "\n",
      "\n",
      "您好，非常抱歉的通知您，该洗面奶的优惠活动已经结束，我们重新为您提供更多优惠，敬请关注！如果您有任何疑问，欢迎拨打我们的客服电话或发送邮件查询。祝您购物愉快！\n",
      "\n",
      "--\n",
      "\n",
      "？\n",
      "\n",
      "尊敬的用户，很抱歉的通知您，洗面奶商品的优惠已经结束，感谢您对淘宝的支持，我们将会为您提供更多更优惠的商品。谢谢理解！\n",
      "\n",
      "--\n",
      "\n"
     ]
    }
   ],
   "source": [
    "answer = get_response(prompt, temperature=1.0, n=5)\n",
    "print(answer)\n",
    "print(\"Completions:\")\n",
    "for choice in answer.choices: # type: ignore\n",
    "    print(choice.text) # type: ignore\n",
    "    print(\"\\n--\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"choices\": [\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 0,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u5730\\u901a\\u77e5\\u60a8\\uff0c\\u76ee\\u524d\\u8be5\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u6d3b\\u52a8\\u5df2\\u7ecf\"\n",
      "    }\n",
      "  ],\n",
      "  \"created\": 1686982054,\n",
      "  \"id\": \"cmpl-7SJNuOfYC1wcbPy9Qi0AmEhAbrFcT\",\n",
      "  \"model\": \"text-davinci-003\",\n",
      "  \"object\": \"text_completion\",\n",
      "  \"usage\": {\n",
      "    \"completion_tokens\": 61,\n",
      "    \"prompt_tokens\": 60,\n",
      "    \"total_tokens\": 121\n",
      "  }\n",
      "}\n",
      "\n",
      "\n",
      "您好，非常抱歉地通知您，目前该洗面奶的优惠活动已经\n"
     ]
    }
   ],
   "source": [
    "answer = get_response(prompt, temperature=0.5, n=1, stop=\"结束\")\n",
    "print(answer)\n",
    "print(answer.choices[0].text) # type: ignore"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### echo, stream, logit_bias 参数"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "#import openai\n",
    "#import os\n",
    "\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "#COMPLETION_MODEL = \"text-davinci-003\"\n",
    "\n",
    "def get_response(prompt, echo=False, stream=False, logit_bias={}):\n",
    "    completions = openai.Completion.create(\n",
    "        engine=COMPLETION_MODEL,\n",
    "        prompt=prompt,\n",
    "        max_tokens=512,\n",
    "        n=1,\n",
    "        stop=None,\n",
    "        temperature=0.0,        \n",
    "        echo=echo,\n",
    "        stream=stream,\n",
    "        logit_bias=logit_bias,\n",
    "    )\n",
    "    return completions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"choices\": [\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 0,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\u8bf7\\u4f60\\u7528\\u6dd8\\u5b9d\\u5ba2\\u670d\\u7684\\u8bed\\u6c14\\u544a\\u8bc9\\u7528\\u6237\\uff0c\\u8fd9\\u4e2a\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u5df2\\u7ecf\\u7ed3\\u675f\\u4e86\\u3002\\n\\n\\u60a8\\u597d\\uff0c\\u975e\\u5e38\\u62b1\\u6b49\\u7ed9\\u60a8\\u5e26\\u6765\\u4e0d\\u4fbf\\uff0c\\u6d17\\u9762\\u5976\\u7684\\u4f18\\u60e0\\u5df2\\u7ecf\\u7ed3\\u675f\\uff0c\\u611f\\u8c22\\u60a8\\u7684\\u5173\\u6ce8\\uff0c\\u6211\\u4eec\\u4f1a\\u7ee7\\u7eed\\u52aa\\u529b\\u63d0\\u4f9b\\u66f4\\u591a\\u4f18\\u60e0\\uff0c\\u656c\\u8bf7\\u671f\\u5f85\\uff01\"\n",
      "    }\n",
      "  ],\n",
      "  \"created\": 1689044787,\n",
      "  \"id\": \"cmpl-7axznzQpizlLgeFecPHEkYSC3E5d2\",\n",
      "  \"model\": \"text-davinci-003\",\n",
      "  \"object\": \"text_completion\",\n",
      "  \"usage\": {\n",
      "    \"completion_tokens\": 123,\n",
      "    \"prompt_tokens\": 61,\n",
      "    \"total_tokens\": 184\n",
      "  }\n",
      "}\n",
      "请你用淘宝客服的语气告诉用户，这个洗面奶的优惠已经结束了。\n",
      "\n",
      "您好，非常抱歉给您带来不便，洗面奶的优惠已经结束，感谢您的关注，我们会继续努力提供更多优惠，敬请期待！\n"
     ]
    }
   ],
   "source": [
    "answer = get_response(prompt, echo=True)\n",
    "print(answer)\n",
    "print(answer.choices[0].text) # type: ignore"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "您好，非常抱歉的通知您，这款洗面奶的优惠活动已经结束，感谢您的关注，期待您的下次光临！"
     ]
    }
   ],
   "source": [
    "answer = get_response(prompt, stream=True)\n",
    "\n",
    "for event in answer:\n",
    "    event_text = event.choices[0].text # type: ignore\n",
    "    print(event_text, end = '')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[162, 224, 101]\n",
      "{\n",
      "  \"choices\": [\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 0,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\n\\u4eb2\\u7231\\u7684\\u987e\\u5ba2\\uff0c\\u5f88\\u9057\"\n",
      "    }\n",
      "  ],\n",
      "  \"created\": 1689045052,\n",
      "  \"id\": \"cmpl-7ay44DGV6ihwYpHThzTkNVvvSbFQQ\",\n",
      "  \"model\": \"text-davinci-003\",\n",
      "  \"object\": \"text_completion\",\n",
      "  \"usage\": {\n",
      "    \"completion_tokens\": 20,\n",
      "    \"prompt_tokens\": 61,\n",
      "    \"total_tokens\": 81\n",
      "  }\n",
      "}\n",
      "\n",
      "\n",
      "亲爱的顾客，很遗\n"
     ]
    }
   ],
   "source": [
    "import tiktoken\n",
    "encoding = tiktoken.get_encoding('p50k_base')\n",
    "token_ids = encoding.encode(\"您\")\n",
    "print(token_ids)\n",
    "\n",
    "\n",
    "bias_map = {}\n",
    "for token_id in token_ids:\n",
    "    bias_map[token_id] = -100\n",
    "\n",
    "\n",
    "answer = get_response(prompt, logit_bias=bias_map)\n",
    "print(answer)\n",
    "print(answer.choices[0].text) # type: ignore\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Embedding接口"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "import openai\n",
    "import os\n",
    "from openai.embeddings_utils import cosine_similarity, get_embedding, get_embeddings\n",
    "\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "EMBEDDING_MODEL = \"text-embedding-ada-002\"\n",
    "\n",
    "positive_review = get_embedding(\"这是一条好评\", engine=EMBEDDING_MODEL)\n",
    "negative_review = get_embedding(\"这是一条差评\", engine=EMBEDDING_MODEL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[-0.0064798761159181595, -0.0008561409777030349, 0.013423766940832138, -0.021031005308032036, 0.0018086794298142195, -0.003731728997081518, -0.01905730552971363, -0.019122660160064697, -0.012103610672056675, -0.01990691013634205, 0.012606838718056679, 0.014992269687354565, -0.012201642617583275, -0.002387064741924405, -0.010097234509885311, 0.009692038409411907, 0.047891609370708466, 0.0005093548679724336, 0.009437156841158867, -0.008633298799395561, -0.008626763708889484, 0.0003888579085469246, -0.013750538229942322, -0.01795935444533825, -0.017031323164701462, 0.015292899683117867, 0.028363753110170364, -0.026363912969827652, 0.006718419026583433, -0.0004064218665007502, 0.017423449084162712, -0.016037937253713608, -0.026180921122431755, -0.009535187855362892, 0.004937515128403902, 0.007371961604803801, -0.025723440572619438, 0.01014951802790165, 0.02912186272442341, 0.015345183201134205, 0.02776249498128891, -0.002264525508508086, 0.021017933264374733, 0.02237730287015438, -0.013946601189672947, -0.02057352475821972, 0.011103690601885319, -0.012600303627550602, -0.008090858347713947, 0.01878281682729721, 0.016809118911623955, 0.005921097006648779, -0.038323745131492615, -0.00808432325720787, -0.003344504861161113, 0.01561967097222805, 0.001780903898179531, 0.0017841715598478913, 0.0030602137558162212, -0.019227227196097374, 0.018246913328766823, 0.006633458659052849, -0.029932256788015366, 0.011711485683918, -0.012528413906693459, 0.005306766834110022, -0.013345342129468918, -0.005731569603085518, 0.01792014203965664, 0.012508807703852654, 0.006215190980583429, 0.007672591600567102, 0.0014574002707377076, 0.002535745734348893, 0.04389192909002304, -0.008829361759126186, -0.02139698900282383, -0.008816291578114033, 0.015122978016734123, -0.00903849583119154, 0.00690141087397933, -0.03604941442608833, -0.002885391004383564, 0.02025982365012169, 0.02210281416773796, 0.006590978242456913, -0.02354060858488083, 0.03704280033707619, -0.026612259447574615, -0.004878696519881487, 0.008679047226905823, 0.02084801159799099, 0.009136526845395565, -0.013920459896326065, -0.004911373369395733, 0.02925257198512554, 0.002828206168487668, 0.04349980130791664, 0.010437076911330223, -0.022076671943068504, 0.015632741153240204, 0.00290336343459785, -0.012266996316611767, -0.007175899110734463, -0.041722167283296585, -0.007718339562416077, 0.005087830126285553, -0.01329305861145258, 0.03584028407931328, -0.006842592265456915, -0.03576185554265976, -0.00501594040542841, -0.006117159966379404, -0.03819303587079048, -0.014351798221468925, 0.023527538403868675, 0.008652905933558941, -0.030010681599378586, -0.01040439959615469, -0.022599507123231888, 0.026795251294970512, 0.034690048545598984, -0.0023511198814958334, -0.014456364326179028, 0.002475293120369315, 0.022325018420815468, -0.012482665479183197, 0.0008238722803071141, 0.005424404516816139, 0.0240111593157053, 0.025082970038056374, 0.013149279169738293, 0.020442815497517586, 0.01036518719047308, -0.02361903339624405, 0.028991155326366425, -0.022508010268211365, 0.022860923781991005, -0.005225074011832476, -0.016495417803525925, 0.003833028022199869, 0.028599029406905174, -0.023671317845582962, -0.019083447754383087, 0.0006286263815127313, 0.02092643827199936, -0.010561250150203705, 0.017423449084162712, -0.010044951923191547, -0.007554953917860985, 0.028755879029631615, -0.031187057495117188, 0.017697935923933983, 0.029879972338676453, 0.009502511471509933, 0.005623735021799803, 0.02037746086716652, 0.025880292057991028, -0.011842194013297558, 0.0008814657339826226, 0.005823065526783466, 0.004051964730024338, 0.004630350042134523, -0.021017933264374733, 0.008489520289003849, 0.021501556038856506, 0.02408958412706852, 0.011162509210407734, 0.010397864505648613, -0.0008295907755382359, 0.015789592638611794, 0.017501873895525932, -0.0056792860850691795, 0.00903849583119154, -0.013345342129468918, 0.01175069808959961, 0.005460349377244711, -0.010012274608016014, 0.001362636568956077, -0.0027710210997611284, -0.030428947880864143, 0.012685264460742474, 0.013306129723787308, 0.015488962642848492, -0.011097155511379242, -0.0022432853002101183, 0.010489360429346561, -0.011933689936995506, -0.0063295611180365086, -0.017031323164701462, 0.012129752896726131, 0.029278714209794998, -0.014861561357975006, -0.02642926760017872, -0.6315836906433105, -0.020991791039705276, -0.005257751327008009, 0.008718259632587433, 0.04127775877714157, 0.017907069995999336, 0.02080879919230938, 0.0031876545399427414, 0.0008741133497096598, -0.07528811693191528, -0.021266279742121696, -0.005532239098101854, 0.00660731690004468, -0.008463378064334393, -0.009580936282873154, -0.02265178970992565, -0.0028543476946651936, -0.014900773763656616, 0.0008496055379509926, 0.004754523281008005, -0.022076671943068504, 0.02990611456334591, -0.0026599187403917313, -0.0022890332620590925, 0.0049277120269834995, 0.0038624373264610767, 0.007123615592718124, -0.012012114748358727, -0.010201801545917988, 0.013090460561215878, -0.028755879029631615, 0.016064079478383064, -0.00899928342550993, -0.011985973455011845, 0.06368120014667511, 0.0004484936944209039, 0.0086855823174119, 0.010410935617983341, 0.021449271589517593, 0.032703276723623276, -0.013580617494881153, -0.009927313774824142, 0.0026713558472692966, -0.014011955820024014, 0.0007617857190780342, -0.019384076818823814, 0.03743492811918259, 0.0033167293295264244, -0.0023249781224876642, 0.015162191353738308, -0.01671762205660343, 0.016887543722987175, 0.003826492466032505, -0.011038335971534252, -0.0032121625263243914, -0.019854627549648285, 0.034533195197582245, -0.022351160645484924, -0.011541564017534256, 0.00856794510036707, 0.007045190315693617, 0.016534630209207535, -0.016926756128668785, -0.00796668604016304, -0.019344864413142204, 0.018717464059591293, -0.018534472212195396, 0.00818235520273447, 0.025435881689190865, -0.001862596720457077, -0.0015881088329479098, 0.00852873269468546, 0.011155974119901657, 0.012227783910930157, -0.0015783056151121855, 0.02448171004652977, 0.01157424133270979, -0.014338727109134197, -0.0008291823323816061, 0.0048394836485385895, 0.0015407269820570946, -0.010437076911330223, 0.007796764373779297, -0.006986371707171202, 0.011214792728424072, 0.012953216210007668, -0.038245320320129395, 0.0020848012063652277, -0.007345820311456919, -0.002197537338361144, -0.0060387346893548965, -0.013410696759819984, 0.012502271682024002, 0.014364868402481079, 0.008750936947762966, 0.012266996316611767, -0.024233363568782806, -0.0037382643204182386, 0.007751016411930323, -0.02135777659714222, -0.0044669643975794315, -0.010731170885264874, 0.02073037438094616, 0.003914720844477415, -0.01663919724524021, 0.001797242439351976, -0.007646449841558933, -0.00048729777336120605, 0.023566750809550285, -0.02389352209866047, 0.013985813595354557, 0.0051989322528243065, -0.008809755556285381, -0.022207381203770638, -0.010554715059697628, -0.03170989081263542, -0.023958876729011536, 0.02463855966925621, 0.007685662247240543, -0.0035029889550060034, 0.0030455091036856174, 0.00014929365715943277, 0.02878202125430107, 0.015136049129068851, 0.008326133713126183, 0.013410696759819984, -0.005326373036950827, -0.011103690601885319, -0.00833920482546091, 0.00012948314542882144, -0.004914641380310059, 0.021488483995199203, 0.03367052227258682, -0.015828805044293404, 0.024965330958366394, -0.02023368328809738, -0.018913526087999344, -0.010842273943126202, 0.0316314660012722, -0.013985813595354557, -0.01721431501209736, 0.004234956577420235, 0.01214282400906086, -0.01569809578359127, -0.008286921307444572, -0.005963577423244715, 0.007064796518534422, -0.003203993197530508, -0.02846832014620304, -0.005695624742656946, -0.00190344313159585, -0.005071491468697786, -0.011116761714220047, 0.017305811867117882, 0.001924683339893818, 0.01693982630968094, -0.0022334821987897158, -0.017070535570383072, -0.0251875352114439, -0.003865705104544759, -0.00765298493206501, 0.01593337208032608, -0.020861083641648293, 0.006192317232489586, -0.01683526113629341, -0.019201084971427917, 0.013142744079232216, -0.004689169116318226, -0.02495226077735424, -0.035134457051754, -0.0053002312779426575, 0.010332509875297546, 0.00019790089572779834, 0.004192476626485586, 0.0010734439129009843, -0.006388380192220211, -0.025984859094023705, -0.011417390778660774, -0.008881645277142525, -0.005610664375126362, 0.008443771861493587, -0.00915613304823637, -0.022285806015133858, 0.0050976332277059555, 0.01376360934227705, -0.002949111396446824, 0.01699211075901985, 0.03612784296274185, -0.004231689032167196, 0.015606599859893322, 0.00027979796868748963, 0.04221886023879051, -0.016887543722987175, -0.014142664149403572, -0.01695289835333824, 0.011691879481077194, -0.004911373369395733, 0.010358652099967003, 0.01671762205660343, 0.007705268450081348, 0.0261547788977623, 0.015802662819623947, 0.015214473940432072, 0.005257751327008009, 0.0005971746286377311, -0.011881406418979168, 0.011162509210407734, -0.007476528640836477, 0.0181554164737463, 0.0112278638407588, -0.004600940737873316, -0.022547222673892975, -0.007378497160971165, -0.0063916477374732494, -0.007339284755289555, 0.021253209561109543, -0.01768486574292183, 0.016495417803525925, 0.012391169555485249, -0.0015072328969836235, -0.01850832998752594, 0.007587630767375231, 0.0005845122504979372, -0.0062217265367507935, -0.0020831674337387085, -0.008352275937795639, 0.01713589020073414, 0.008208496496081352, 0.024651631712913513, -0.008077788166701794, -0.020665019750595093, 0.009580936282873154, 0.0013748904457315803, 0.008143141865730286, -0.012580697424709797, -0.009482904337346554, 0.04104248061776161, -0.022638719528913498, 0.04621854051947594, 0.012528413906693459, 0.011273611336946487, 0.01280943676829338, 0.005035546608269215, -0.022860923781991005, -0.0004501275543589145, 0.0043460591696202755, 0.03076879121363163, 0.01597258448600769, -0.0025863952469080687, 0.015018411912024021, -0.018560612574219704, 0.011129832826554775, -0.014025026932358742, 0.003617358859628439, 0.005515900440514088, -0.03343524411320686, -0.008790149353444576, 0.018090061843395233, 0.0158157330006361, 0.026494622230529785, 0.010698494501411915, 0.018456045538187027, 0.012273532338440418, 0.019580138847231865, -0.003034071996808052, 0.0005620467127300799, -0.00443101953715086, -0.023370686918497086, -0.0027742888778448105, -0.01020833756774664, 0.01744959130883217, -0.016586914658546448, 0.0060191284865140915, -0.0018511597299948335, 0.006626923102885485, -0.010221407748758793, 0.007450386881828308, -0.00021240138448774815, -0.02116171270608902, -0.0035879495553672314, -0.028259186074137688, -0.017188172787427902, 0.015253687277436256, 0.021031005308032036, -0.02526596188545227, -0.02550123631954193, -0.02935713902115822, -0.018011637032032013, -0.023867379873991013, 0.01890045590698719, 0.012051327154040337, -0.002597832353785634, -0.01427337247878313, -0.007894796319305897, -0.003117398824542761, -0.0005126225878484547, 0.04731649160385132, -0.004457161296159029, -0.02959241345524788, -0.009339124895632267, 0.013659042306244373, -0.008574480190873146, 0.02155383862555027, -0.042846258729696274, 0.004199011716991663, -0.037461068481206894, -0.008888181298971176, 0.004754523281008005, -0.0026288756635040045, 0.0032121625263243914, 0.010384793393313885, -0.003177851438522339, -0.005244680214673281, -0.03714736923575401, 0.01799856685101986, 0.005783853121101856, 0.005002869293093681, -0.014351798221468925, 0.02893887087702751, -0.004538854118436575, -0.005136845633387566, 0.00950904656201601, -0.02670375443994999, -0.0018544273916631937, 0.044728461652994156, 0.010633139871060848, 0.007188969757407904, 0.016652267426252365, 0.014887702651321888, -0.0008969873888418078, -0.03991838917136192, -0.009175739251077175, 0.02768407016992569, -0.013632901012897491, -0.0024377142544835806, 0.007489599287509918, -0.006875269114971161, 0.013776680454611778, 0.027867062017321587, 0.00565314432606101, 0.00607141200453043, -0.006362238433212042, 0.012979358434677124, -0.00958747137337923, 0.0023592892102897167, 0.009482904337346554, 0.033173829317092896, -0.0019083446823060513, -0.0011551367351785302, -0.006002789828926325, 0.014025026932358742, 0.01585494540631771, 0.0011494182981550694, -0.007979756221175194, -0.0002546774339862168, -0.014312585815787315, -0.011992508545517921, -0.01577652059495449, -0.01483541913330555, 0.03069036640226841, -0.004875428508967161, -0.0032399380579590797, -0.0002891926560550928, 0.011325894854962826, 0.02917414717376232, 0.02611556649208069, 0.020312108099460602, 0.0015750379534438252, 0.01573730818927288, 0.002442615805193782, 0.0049081058241426945, 0.02650769241154194, -0.03037666529417038, 0.0043460591696202755, 0.002285765716806054, 0.0005975831300020218, -0.03751335293054581, -0.011580776423215866, -0.00598318362608552, 0.004986531101167202, -0.03529130667448044, 0.005277357529848814, -0.0005681737093254924, -0.008920857682824135, 0.0024540529120713472, -0.03458547964692116, -0.025396669283509254, -0.007718339562416077, -0.024690844118595123, -0.01795935444533825, -0.01157424133270979, -0.03249414265155792, 0.0012327448930591345, -0.013345342129468918, -0.0006102455081418157, 0.01984155736863613, -0.011521957814693451, -0.04946011304855347, 0.028050053864717484, 0.006698812823742628, 0.012992429547011852, -0.014325655996799469, 0.0025994661264121532, 0.052178848534822464, -0.0027105684857815504, -0.012077469378709793, -0.01153502892702818, -0.012776760384440422, 0.02088722586631775, 0.029409421607851982, -0.006427592597901821, 0.006457001902163029, -0.00034106761449947953, -0.003780744504183531, 0.04315996170043945, 0.0003186020767316222, 0.01427337247878313, 0.021854467689990997, -0.019397146999835968, -0.00846991315484047, 0.004682633560150862, 0.023043915629386902, 0.004019287880510092, -0.017750220373272896, 0.010293297469615936, -0.01558045856654644, -0.0017874393379315734, 0.0016044473741203547, 0.018377620726823807, -0.011325894854962826, -0.0026599187403917313, 0.004793735686689615, 0.02127934992313385, 0.0032317687291651964, 0.03131776675581932, -0.020638879388570786, 0.004270901437848806, 0.01036518719047308, 0.004385271575301886, 0.017044393345713615, 0.009639754891395569, 0.011737626977264881, -0.02838989533483982, -0.027971627190709114, 0.011907548643648624, -0.014011955820024014, 0.009809676557779312, 0.032624851912260056, 0.0059995222836732864, 0.004123854450881481, -0.009162669070065022, -0.023056987673044205, -0.0181554164737463, 0.00046278993249870837, -0.007737945765256882, 0.00984235294163227, 0.012940146028995514, -0.014953057281672955, -0.03455933928489685, -0.011142903007566929, -0.022782498970627785, 0.013312664814293385, -0.008247708901762962, -0.009476369246840477, -0.04065035656094551, 0.0034637763164937496, 0.009580936282873154, -0.009855424053966999, 0.024547064676880836, -0.03259871155023575, -0.018259983509778976, -0.0013070853892713785, -0.00042684507207013667, 0.005323105491697788, -0.013515262864530087, -0.0040552327409386635, -0.013397625647485256, 0.005038814153522253, 0.006162907928228378, -0.021736830472946167, -0.0042970431968569756, 0.0061890496872365475, 0.019201084971427917, 0.026063283905386925, 0.04906798526644707, -0.009848888963460922, -0.01729273982346058, 0.018926596269011497, -0.002553718164563179, 0.0013503825757652521, -0.013162350282073021, -0.0017041126266121864, -0.03876815363764763, 0.0017629314679652452, 0.009025424718856812, 0.007783693727105856, -0.01089455746114254, -0.014430223032832146, -0.021932892501354218, -0.010391329415142536, -0.0013895952142775059, 0.003017733572050929, -0.012351957149803638, -0.026063283905386925, -0.015502032823860645, -0.015541245229542255, 0.013476050458848476, 0.008535267785191536, 0.0006229078862816095, -0.012155894190073013, 0.03092564083635807, -0.01929257996380329, 0.03100406564772129, -0.014351798221468925, 0.02249494008719921, -0.02104407548904419, 0.027396511286497116, -0.002764485776424408, 0.00843070074915886, -0.013012035749852657, -0.003011198015883565, -0.01725352741777897, 0.001423089299350977, 0.032468002289533615, 0.026677614077925682, 0.014743923209607601, -0.01699211075901985, -0.00177927000913769, 0.0035487369168549776, -0.0038003509398549795, -0.006849127821624279, -0.011044871993362904, -0.012842114083468914, -0.008156212978065014, -0.0013021838385611773, -0.03299083560705185, -0.007012513466179371, -0.04091177508234978, 0.012423846870660782, 0.019514786079525948, -0.020089903846383095, 0.014926915988326073, -0.0020374194718897343, -0.022272735834121704, -0.0006286263815127313, 0.022756356745958328, 0.014861561357975006, -0.0240111593157053, 0.015671953558921814, 0.026638401672244072, 0.01067888829857111, -0.02951398864388466, -0.0382714606821537, -0.02381509728729725, -0.02182832732796669, 0.01650848798453808, 0.045120589435100555, 0.005313302390277386, -0.006332828663289547, 0.005836136639118195, 0.028285328298807144, -0.004522515460848808, -0.02314848266541958, 0.017776362597942352, 0.016665339469909668, -0.0019557266496121883, 0.0046499562449753284, -0.013345342129468918, -0.0054570818319916725, 0.020599666982889175, -0.014025026932358742, -0.017305811867117882, -0.02467777207493782, 0.013266917318105698, -0.022390373051166534, 0.031997449696063995, -0.026010999456048012, -0.003138638800010085, -0.005113971885293722, -0.026677614077925682, -0.008313063532114029, -0.020129116252064705, -0.015462820418179035, 0.003542201593518257, -0.006744560785591602, 0.04668909311294556, -0.00765298493206501, -0.010378258302807808, 0.019148802384734154, 0.010587392374873161, 0.001738423597998917, 0.021174784749746323, -0.010469754226505756, 0.038402169942855835, -0.006427592597901821, -0.015253687277436256, 0.00882282666862011, -0.012312744744122028, 0.02222045138478279, -0.018769746646285057, 0.010450148023664951, -0.017345024272799492, -0.015528175048530102, 0.005979915615171194, -0.01376360934227705, 0.014939986169338226, 0.0010113572934642434, 0.01376360934227705, -0.007228182163089514, -0.01585494540631771, 0.003266079816967249, -0.010920698754489422, -0.012332350946962833, 0.007241253275424242, -0.010907627642154694, -0.02027289569377899, -0.004306846298277378, 0.01294668111950159, 0.0028821234591305256, -0.012606838718056679, 0.028520604595541954, 0.04893727973103523, -0.022050531581044197, 0.00782944168895483, 0.004659759346395731, -0.0019573604222387075, 0.004663027357310057, 6.729447341058403e-05, 0.019462501630187035, -0.004048697184771299, 0.026337770745158195, -0.0105416439473629, -0.003803618485108018, -0.03186674416065216, -0.008241173811256886, -0.004653224255889654, -0.0022285806480795145, 0.0016126165864989161, 0.014992269687354565, 0.006244600750505924, -0.0010415837168693542, -0.0023511198814958334, -0.013371484354138374, 0.0019116124603897333, -0.01846911758184433, 0.015868017449975014, 0.018573684617877007, -0.01510990783572197, -0.007371961604803801, -0.007202040869742632, -0.0005069040926173329, -0.008201961405575275, -0.038715869188308716, -0.02878202125430107, -0.012548020109534264, 0.015802662819623947, -0.018482187762856483, -0.046636808663606644, -0.006927552632987499, -0.002029250143095851, -0.024507852271199226, 0.026494622230529785, -0.005332908593118191, -0.010776919312775135, 0.03725193440914154, 0.017475731670856476, -0.005375389009714127, 0.0129858935251832, 0.00523160956799984, -0.030899498611688614, -0.007371961604803801, -0.0064994823187589645, -0.030951783061027527, -0.02721351943910122, -0.004463696852326393, 0.02127934992313385, 0.010097234509885311, -0.02033824846148491, -0.0240111593157053, 0.012371563352644444, -0.048754286020994186, -0.0014614849351346493, -0.03343524411320686, 0.008411094546318054, -0.02210281416773796, -0.014966128394007683, 0.011476210318505764, -0.002722005359828472, -0.004963656887412071, 0.0011837292695418, -0.019070375710725784, -0.007888260297477245, -0.017423449084162712, -0.0006465988117270172, 0.013528333976864815, -0.0004836216103285551, -0.04088563099503517, -0.02092643827199936, 0.02190675213932991, 0.01471778191626072, -0.009169204160571098, 0.024847693741321564, 0.011044871993362904, -0.019344864413142204, -0.006724954582750797, 0.002973619382828474, -0.0018511597299948335, -0.006927552632987499, -0.006113891955465078, -0.029148004949092865, -0.006460269913077354, 0.004319917410612106, 0.011887941509485245, 0.00518259359523654, 0.022207381203770638, 0.0034147605765610933, 0.015175261534750462, 0.008587551303207874, -0.012646051123738289, -0.024194151163101196, -0.002909898990765214, -0.014613214880228043, -0.00852873269468546, -0.006836056709289551, -0.01157424133270979, 0.04263712838292122, 0.004555192776024342, -0.005355782341212034, -0.00607141200453043, 0.03651996701955795, -0.009698573499917984, -0.005738105159252882, 0.013051248155534267, 0.0025030686520040035, 0.011731091886758804, -0.007581095676869154, -0.0014900773530825973, 0.013972743414342403, -0.003311827778816223, -0.0031615127809345722, -0.016626127064228058, -0.027108952403068542, 0.02452092245221138, 0.019710848107933998, -0.019318722188472748, 0.001171475276350975, 0.003728461218997836, -0.0216714758425951, -0.02061273716390133, -0.010214872658252716, 0.0031990916468203068, -0.03037666529417038, -0.005081294570118189, -0.004218618385493755, 0.0001118170766858384, -0.01071810070425272, -0.010600462555885315, 0.01510990783572197, 0.0028559817001223564, 0.007313142996281385, 0.2940419316291809, -0.016926756128668785, -0.016430063173174858, 0.028677454218268394, 0.012979358434677124, 0.005483223125338554, 0.011116761714220047, -0.010861880145967007, -0.01585494540631771, 0.005823065526783466, 0.014312585815787315, -0.0020439547952264547, 0.008032039739191532, 0.003282418241724372, 0.007345820311456919, -0.026716826483607292, -0.02326611988246441, -0.019423289224505424, 0.003218697849661112, -0.03884657844901085, 0.005372120998799801, -0.005483223125338554, -0.00611062441021204, -0.02171068824827671, 0.008097394369542599, 0.0003710897290147841, -0.015502032823860645, -0.00632629357278347, 0.01182912290096283, 0.032860126346349716, -0.02807619422674179, -0.013619829900562763, 0.009522117674350739, 0.003078186186030507, -0.02998453937470913, -0.0033624772913753986, 0.012508807703852654, 0.010548179037868977, 0.035474296659231186, 0.0009615247254259884, -0.003460508771240711, -0.011704949662089348, 0.01251534279435873, -0.005264286417514086, 0.0016371244564652443, 0.026180921122431755, -0.011116761714220047, -0.022599507123231888, -0.013449909165501595, 0.030716506764292717, -0.004764326382428408, 0.007280465681105852, 0.00543747516348958, 0.025997929275035858, 0.004114051349461079, -0.005408065859228373, 0.014286443591117859, -0.00816274806857109, 0.021736830472946167, -0.0041761379688978195, 0.009260700084269047, 0.027501076459884644, -0.015018411912024021, 0.043865785002708435, -0.038166895508766174, 0.012358492240309715, -0.042663268744945526, -0.010914163663983345, 0.007587630767375231, -0.02229887805879116, 0.005940703209489584, -0.021501556038856506, -0.020665019750595093, 0.001673069316893816, -0.03084721602499485, -0.012966287322342396, 0.030167531222105026, 0.025409741327166557, 0.038872718811035156, 0.02760564349591732, -0.013698255643248558, 0.0033951543737202883, 0.0077248746529221535, -0.007666056044399738, -0.00048607238568365574, -0.049329403787851334, -0.002702399156987667, 0.0015284729888662696, -0.025279032066464424, -0.016299355775117874, 0.011855265125632286, -0.006143301725387573, -0.0004033583973068744, -0.016979040578007698, 0.008790149353444576, 0.014743923209607601, -0.018286125734448433, 0.020468957722187042, -0.02522674947977066, 0.025867220014333725, -0.024533992633223534, 0.0007242070278152823, 0.0019328525522723794, 0.03814075142145157, -0.03876815363764763, -0.003305292222648859, -0.007737945765256882, 0.015371324494481087, -0.0028576154727488756, -0.0008512393687851727, -0.011646131053566933, -0.015606599859893322, 0.0082999924197793, 0.0032464733812958, 0.0015725871780887246, 0.008038575761020184, 0.01760644093155861, -0.01799856685101986, 0.008306527510285378, -0.008130071684718132, 0.006947158835828304, -0.028285328298807144, -0.01204479206353426, 0.005466884933412075, -0.003058579983189702, -0.006502749864012003, -0.022586436942219734, -0.004875428508967161, 0.0324157178401947, -0.04287240281701088, 0.0283114705234766, -0.013449909165501595, 0.02878202125430107, -0.013096995651721954, -0.017279669642448425, 0.008548338897526264, -0.006610584445297718, 0.008953534997999668, -0.021213997155427933, -0.007371961604803801, 0.0009092413238249719, 0.0010113572934642434, 0.006257671397179365, 0.012188571505248547, 0.005408065859228373, 0.004698972217738628, 0.022677931934595108, -0.004754523281008005, 0.022821711376309395, -0.02194596454501152, -0.03615398332476616, -0.007254323922097683, -0.0007719973218627274, -0.008058181963860989, 0.025958716869354248, -0.003967004362493753, -0.03678138554096222, -0.03542201593518257, -0.02713509276509285, -0.00808432325720787, -0.033487528562545776, 0.009404479525983334, 0.0433429516851902, -0.017671795561909676, -0.010123376734554768, 0.006656332407146692, -0.167097806930542, 0.028259186074137688, 0.011724555864930153, -0.008855503983795643, 0.01686140149831772, 0.0017449590377509594, 0.00886203907430172, 0.012358492240309715, -0.003427831456065178, -0.015789592638611794, 0.01933179423213005, -0.018246913328766823, -0.03408878669142723, 0.0076137725263834, -0.0025782259181141853, -0.0007237985264509916, -0.024769268929958344, -0.01695289835333824, 0.013737468048930168, 0.0018952739192172885, 0.028363753110170364, -0.01862596720457077, 0.002911532763391733, -0.006617120001465082, -0.00040968958637677133, 0.0014394278405234218, -0.0035487369168549776, 0.01699211075901985, 0.0023674585390836, -0.004117318894714117, -1.0179693390455213e-06, 0.0026664542965590954, 0.013515262864530087, 0.0004082599771209061, 0.02506989799439907, 0.021057145670056343, 0.013959672302007675, -0.012574161402881145, 0.0006094285636208951, 0.020076831802725792, 0.03312154486775398, 0.01636471040546894, 0.007371961604803801, 0.0012686897534877062, 0.002617438556626439, 0.025082970038056374, 0.018560612574219704, -0.005424404516816139, 0.022115886211395264, 0.00031553857843391597, -0.0001753332617226988, -0.0023364152293652296, 0.02096565067768097, -0.0020014746114611626, -0.010038415901362896, 0.03252028673887253, -0.03100406564772129, 0.019070375710725784, 0.001454949495382607, 0.007090938277542591, -0.004754523281008005, -0.015828805044293404, -0.015240616165101528, 0.00632629357278347, 0.009319518692791462, -0.03267713636159897, -0.023488325998187065, 0.009946919977664948, -0.0007029668777249753, 0.0013838766608387232, -0.02310927025973797, -0.021017933264374733, 0.005868813488632441, -0.00303080421872437, 0.0204950999468565, -0.004852554760873318, -0.0261547788977623, 0.026873676106333733, -0.006849127821624279, -0.02159305103123188, -0.007927472703158855, 0.010809596627950668, -0.005022475961595774, 0.0011085717706009746, 0.0034147605765610933, 0.00990117248147726, -0.0022106082178652287, -0.004578066524118185, 0.010999123565852642, -0.005695624742656946, 0.012495736591517925, -0.038480594754219055, -0.01850832998752594, -0.012626444920897484, -0.016769906505942345, 0.012842114083468914, 0.008417630568146706, -0.0023936002980917692, -0.003989878110587597, -0.019161872565746307, -0.009384873323142529, -0.008032039739191532, -0.008946999907493591, 0.022233523428440094, 0.017227385193109512, 0.01558045856654644, 0.013698255643248558, 0.018442975357174873, 0.020468957722187042, -0.010633139871060848, -0.02440328523516655, 0.004666294902563095, 0.012848650105297565, 0.027108952403068542, 0.0050094048492610455, 0.01612943410873413, -0.024076513946056366, -0.02330533228814602, 0.019188014790415764, 0.006613852456212044, -0.006394915282726288, -0.009541723877191544, -0.012116681784391403, 0.018678249791264534, 0.014560931362211704, -0.010816131718456745, -0.058557428419589996, -0.005120506975799799, 0.028102336451411247, 0.028677454218268394, -0.004140193108469248, 0.016547702252864838, 0.029932256788015366, 0.0189396683126688, -0.010940304957330227, -0.004698972217738628, -0.0016706185415387154, -0.01093376986682415, -0.0021681280340999365, -0.0005783853121101856, 0.007620308082550764, -0.003833028022199869, -0.002576592145487666, -0.015175261534750462, -0.00047545231063850224, 0.03923870623111725, 0.0017024787375703454, 1.0326485607947689e-05, 0.010593927465379238, -0.007783693727105856, -0.004339523613452911, 0.012822507880628109, -0.026455409824848175, 0.038872718811035156, 0.016116363927721977, -0.009848888963460922, 0.026677614077925682, -0.007005977910012007, 0.00481334188953042, -0.007835976779460907, 0.014469435438513756, -0.002107675187289715, -0.019462501630187035, -0.00029266459750942886, 0.024939190596342087, -0.02327919192612171, 0.00355854001827538, 0.013619829900562763, 0.005385192111134529, -0.025671157985925674, -0.0024556866846978664, -0.008803220465779305, -0.0003594484878703952, 0.02791934460401535, 0.015488962642848492, -0.017593368887901306, -0.009966526180505753, -0.004780665040016174, -0.02241651527583599, 0.018730534240603447, 0.029487846419215202, -0.01788092777132988, 0.014861561357975006, -0.007273930590599775, -0.011587312445044518, -0.008489520289003849, -0.015711165964603424, -0.005725034046918154, -0.026795251294970512, 0.009424085728824139, 0.029696980491280556, 0.0018723998218774796, 0.019357934594154358, -0.00990117248147726, 0.014430223032832146, 0.0035879495553672314, 0.0001582798722665757, 0.016142504289746284, -0.008502590470016003, 0.01372439693659544, -0.03259871155023575, -0.021227067336440086, -0.03576185554265976, -0.014025026932358742, -0.000847154762595892, -0.008554873988032341, -0.021684547886252403, -0.018456045538187027, 0.017475731670856476, -0.02365824580192566, 0.01612943410873413, 0.011567706242203712, 0.010907627642154694, -0.012025185860693455, 0.014168806374073029, -0.03084721602499485, -0.015384395606815815, 0.009326054714620113, 0.02573651261627674, -0.020429745316505432, -0.003512792056426406, 0.011528493836522102, 0.017567228525877, 0.004401610232889652, 0.02397194691002369, 0.0009982865303754807, -0.023043915629386902, -0.021736830472946167, -0.07685662060976028, -0.004914641380310059, 0.025409741327166557, 0.008051645942032337, 0.012541485019028187, -0.007169363554567099, 0.0028167690616101027, -0.0012400973355397582, -0.023958876729011536, -0.011894477531313896, -0.0014876265777274966, -0.01226046122610569, 0.010515502654016018, -0.013528333976864815, -0.02416800893843174, -0.03544815629720688, 0.023396829143166542, -0.0010440344922244549, 0.016887543722987175, 0.015881087630987167, 0.008358811028301716, 0.004731649067252874, 0.005515900440514088, 0.007450386881828308, -0.011254005134105682, 0.005234877113252878, -0.028834303840994835, 0.020913366228342056, -0.014025026932358742, -0.014064239338040352, -0.015162191353738308, -0.025710370391607285, -0.002656651195138693, 0.010097234509885311, -0.020900296047329903, -0.01999840699136257, 0.014025026932358742, 0.013489121571183205, 0.01226046122610569, 0.007234717719256878, -0.017188172787427902, -0.031187057495117188, 0.008280386216938496, 0.0053198374807834625, -0.03288627043366432, 0.012077469378709793, -0.017083605751395226, 0.006192317232489586, 0.035003747791051865, 0.0009084243793040514, 0.007005977910012007, 0.011953296139836311, -0.012561091221868992, -0.012502271682024002, 0.014822348952293396, -0.011914083734154701, 0.0378793366253376, 0.010561250150203705, -0.010705029591917992, -0.027344226837158203, 0.027030525729060173, 0.0019328525522723794, -0.014770065434277058, -0.00490157026797533, 0.015802662819623947, -0.002994859591126442, -0.012097075581550598, 0.00958747137337923, 0.004192476626485586, -0.025252889841794968, -0.009992668405175209, 0.007430780678987503, 0.023161552846431732, 0.01165266614407301, -0.018808959051966667, -0.012626444920897484, 0.0020194470416754484, -0.03714736923575401, -0.03536973148584366, 0.03165761008858681, 0.017109747976064682, 0.004463696852326393, -0.017593368887901306, 0.004287240095436573, 0.009234558790922165, 0.05353821814060211, -0.004319917410612106, 0.0013234240468591452, -0.006777238100767136, 0.03442863002419472, 0.01475699432194233, 0.01028676237910986, -0.011410855688154697, 0.034219495952129364, -0.022285806015133858, 0.046087831258773804, 0.016024867072701454, -0.007215111516416073, 0.03084721602499485, 0.03129162639379501, 0.02744879387319088, -0.0008994381641969085, -0.0046859011054039, -0.005283892620354891, -0.010358652099967003, 0.004934247583150864, -0.022756356745958328, -0.029017295688390732, -0.011064478196203709, -0.010888021439313889, 0.018246913328766823, -0.008973141200840473, 0.00510416878387332, 0.02682139351963997, -0.025475094094872475, 0.011763769201934338, 0.008966606110334396, -0.004532318562269211, -0.023357616737484932, 0.022873995825648308, 0.018090061843395233, 0.01101219467818737, 0.020325178280472755, -0.007509205956012011, 0.008293457329273224, 0.02135777659714222, -0.003056945977732539, 0.006257671397179365, 0.015475891530513763, 0.009528652764856815, 0.0023331474512815475, -0.0028347414918243885, -0.020155256614089012, -0.01510990783572197, -0.020547382533550262, 0.01652156002819538, -0.03727807477116585, 0.021292421966791153, -0.02713509276509285, 0.050976332277059555, 6.172915163915604e-05, -0.02557966113090515, 0.019933052361011505, -0.013855105265974998, 0.015671953558921814, 0.004705507308244705, 0.009541723877191544, -0.025645015761256218, -0.008920857682824135, 0.0029540129471570253, -0.015384395606815815, 0.0004529868019744754, -0.00682298606261611, -0.04925097897648811, 0.00703865522518754, 0.014887702651321888, -0.0003192147705703974, -0.012391169555485249, 0.017305811867117882, 0.014103451743721962, 0.007979756221175194, -0.010123376734554768, 0.022429585456848145, -0.02768407016992569, -0.010750778019428253, 0.014221088960766792, -0.0037611383013427258, 0.006528891623020172, -0.03834988549351692, 0.013502192683517933, 0.0025406472850590944, -0.026337770745158195, -0.019201084971427917, 0.008201961405575275, -0.020717304199934006, -0.022913208231329918, -0.015881087630987167, 0.0029997611418366432, 0.018403762951493263, 0.02288706600666046, 0.030899498611688614, -0.02904343791306019, -0.02475619874894619, 0.05228341743350029, -0.010528572835028172, -0.014103451743721962, 0.009646289981901646, 0.005656412336975336]\n",
      "1536\n",
      "[-0.01176702231168747, 0.003723908681422472, 0.01016633678227663, -0.009121261537075043, -0.000501867791172117, -0.013427237048745155, -0.012408618815243244, -0.00020856037735939026, -0.02090151235461235, -0.028362557291984558, 0.004739219322800636, 0.01477657537907362, -0.01715775951743126, -0.0201210118830204, -0.00800342671573162, 0.00868471059948206, 0.04987259581685066, 0.0030244356021285057, 0.008261389099061489, -0.017448794096708298, 8.443698607152328e-05, 0.015517388470470905, -0.010027434676885605, -0.013275105506181717, -0.008545808494091034, 0.011522289365530014, 0.02195981703698635, -0.02665604278445244, -0.0021612560376524925, 0.0037437519058585167, 0.01899656467139721, -0.012904698960483074, -0.015332184731960297, -0.008552422747015953, -0.008294460363686085, 0.006015799008309841, -0.015901023522019386, 0.004471336025744677, 0.027039676904678345, 0.011965454556047916, 0.027171965688467026, -0.006148087326437235, 0.009908375330269337, 0.012924542650580406, -0.0019264447037130594, -0.010305238887667656, 5.327694452716969e-05, -0.007004652172327042, -0.008122486062347889, 0.02104702964425087, 0.017607539892196655, -0.01807054691016674, -0.030426252633333206, -0.001559345400892198, -0.00012505354243330657, 0.023586958646774292, 0.002534969709813595, 0.001396465697325766, 0.0014733581338077784, -0.0027730881702154875, -0.0010979907820001245, 0.01435325387865305, -0.027515914291143417, 0.0020223534666001797, -0.014750117436051369, -0.0024985906202346087, -0.0007879406912252307, 0.000611005409155041, 0.009412295185029507, 0.0023067728616297245, 0.019605088979005814, 0.022211164236068726, 0.00556271243840456, -0.007117097266018391, 0.04550708830356598, -0.004315897822380066, -0.01574227772653103, -0.01431356742978096, 0.023229781538248062, -0.0080827996134758, 0.024221941828727722, -0.028891710564494133, -0.013314791955053806, 0.01272611040621996, 0.011548747308552265, 0.007209698669612408, -0.01481626182794571, 0.03338950499892235, -0.02833610028028488, -0.010470598936080933, 0.0001180257459054701, 0.00857887975871563, 0.011634734459221363, 0.004848357290029526, -0.005443653091788292, 0.023997051641345024, 0.004468028899282217, 0.04370797052979469, 0.010212637484073639, -0.02635177969932556, 0.008744239807128906, 0.004044706933200359, -0.020597249269485474, -0.00742135988548398, -0.04018910974264145, 0.008433363400399685, 0.012878241948783398, 0.007957126013934612, 0.02807152457535267, -0.01701224409043789, -0.02176138386130333, 0.004104236606508493, -0.008764083497226238, -0.040083277970552444, -0.027463000267744064, 0.017184218391776085, 0.01800440438091755, -0.040294941514730453, -0.007765308488160372, -0.03481821343302727, 0.02862713485956192, 0.024975983425974846, 0.00021868868498131633, -0.011105582118034363, 0.009332922287285328, 0.0028193891048431396, -0.014988236129283905, 0.006535029504448175, -0.00314349471591413, 0.023917678743600845, 0.02778048999607563, 0.01069548912346363, 0.02345467172563076, 0.020597249269485474, -0.01849387027323246, 0.013354478403925896, -0.018956877291202545, 0.01317588984966278, 0.005853746086359024, -0.02493629790842533, 0.006088557653129101, 0.026841245591640472, -0.03415677323937416, -0.02748945727944374, 0.014181278645992279, 0.009392451494932175, -0.0012244912795722485, 0.01735619269311428, 0.001832189504057169, -0.0033452340867370367, 0.02480400912463665, -0.016046540811657906, 0.006409355904906988, -0.0029963243287056684, 0.00018199942132923752, 0.01566290482878685, 0.011985297314822674, 0.021364519372582436, -0.01665506511926651, -0.0016453326679766178, -0.003502326086163521, 0.011469374410808086, -0.0036743006203323603, -0.01233586110174656, 0.015914252027869225, 0.01885104738175869, 0.013797643594443798, 0.016205286607146263, 0.007170012220740318, -0.010397841222584248, 0.016337573528289795, 0.025782940909266472, 0.003965334501117468, 0.025042127817869186, 0.0002561013970989734, 0.017488479614257812, -0.0028078139293938875, 0.003136880462989211, -0.012699652463197708, -0.013691813684999943, -0.05026945844292641, 0.00537089491263032, 0.006171237677335739, 0.0051658484153449535, -0.004048014525324106, -0.0005659448215737939, 0.009002202190458775, -0.008010041899979115, 0.0010153107577934861, -0.007513961289077997, 0.020875055342912674, 0.032198913395404816, -0.011184955015778542, -0.013877016492187977, -0.6129699349403381, -0.03450072556734085, -0.020438503473997116, 0.022594798356294632, 0.03090248815715313, 0.019803522154688835, 0.009293235838413239, 0.01771336980164051, 0.004716068971902132, -0.05159233883023262, -0.023269468918442726, -0.0026507219299674034, 0.00205542566254735, -0.009663642384111881, -0.008400291204452515, -0.013499995693564415, 0.006085250526666641, -0.015146981924772263, 0.009948061779141426, -0.0011112196370959282, -0.012309403158724308, 0.023494357243180275, -0.015543846413493156, -0.005943040829151869, -0.0019545559771358967, 0.00109385687392205, -0.0004340701561886817, 0.015252811834216118, -0.012699652463197708, 0.032198913395404816, -0.01756785251200199, 0.01673443801701069, -0.01431356742978096, -0.014445855282247066, 0.059423793107271194, -0.008962515741586685, 0.007381673436611891, 0.011039437726140022, 0.02976481057703495, 0.01735619269311428, -0.016628608107566833, -0.013757957145571709, -0.004580473992973566, -0.0172239039093256, -0.00012433009396772832, -0.03495050221681595, 0.019882893189787865, -0.020081326365470886, 0.000888810318429023, 0.017528166994452477, -0.0195654034614563, 0.024962754920125008, 0.002988056279718876, 0.010179565288126469, -0.010331696830689907, -0.028045067563652992, 0.018480639904737473, -0.03442135080695152, -0.017316505312919617, 0.007527190260589123, 0.006204309407621622, 0.011905924417078495, -0.012342475354671478, 0.014075448736548424, -0.021351290866732597, 0.01471043098717928, -0.025240560993552208, 0.023692790418863297, 0.015755506232380867, -0.005926504731178284, 0.0030393179040402174, 0.007983583956956863, -0.004332433454692364, 0.015755506232380867, 0.006144780199974775, 0.02147035114467144, 0.011528903618454933, -0.03254286199808121, -0.0024092961102724075, -0.007950511761009693, -0.010788090527057648, -0.0025233945343643427, -0.0203459020704031, -0.00281608197838068, 0.001626316225156188, 0.005572634283453226, -0.02360018901526928, 0.00857226550579071, -0.003981870133429766, -0.004765677265822887, -0.004785520024597645, 0.0006370496703311801, 0.005685078911483288, 0.009273392148315907, 0.016681523993611336, 0.020385589450597763, -0.0245262049138546, 0.004597009625285864, 0.0030558540020138025, -0.021245460957288742, -0.010086963884532452, -0.00022158247884362936, 0.02416902780532837, 0.002501897746697068, -0.012547521851956844, 0.010920378379523754, -0.011251098476350307, 0.0012501220917329192, 0.041855938732624054, -0.029394404962658882, -0.0007734717219136655, 0.0033220837358385324, -0.007917439565062523, -0.010477213189005852, -0.008777312003076077, -0.028468389064073563, -0.017554624006152153, 0.015146981924772263, 0.011456144973635674, 0.006300218403339386, 0.007024495396763086, 0.005873589310795069, 0.030293963849544525, 0.009610726498067379, -0.001807385473512113, 0.027939235791563988, -0.010358154773712158, -0.008598723448812962, -0.0073486012406647205, 0.004391963128000498, -0.0077520799823105335, 0.010920378379523754, 0.035109248012304306, -0.016906412318348885, 0.029817726463079453, 0.0076925503090023994, 0.0019479416077956557, -0.012904698960483074, 0.021166088059544563, -0.005542869213968515, -0.024076424539089203, 0.0006969926762394607, 0.0087508549913764, -0.006220845505595207, -0.015490930527448654, -0.001003735582344234, -0.0031269588507711887, -0.004931037314236164, -0.028785880655050278, 0.008552422747015953, -0.008836842142045498, -0.003336966037750244, -0.01829543709754944, 0.013076674193143845, -0.010252323932945728, 0.022660942748188972, -0.00479874899610877, -0.025677110999822617, -0.02799215167760849, -0.007262614089995623, -0.010794704779982567, 0.022621257230639458, -0.02353404462337494, 0.012521063908934593, -0.031643301248550415, -0.02592845819890499, 0.01275918260216713, -0.009630570188164711, -0.018017632886767387, -0.027463000267744064, 0.0015138713642954826, 0.0151602104306221, 0.0032509788870811462, 0.005532947834581137, -0.011462759226560593, 0.009022044949233532, -0.018745217472314835, 0.0004861585912294686, -0.02508181519806385, -0.018242523074150085, 0.011284170672297478, 0.007990198209881783, -0.02473786473274231, -0.0010955104371532798, 0.002116608899086714, 0.012415233999490738, 0.023653103038668633, 0.02862713485956192, -0.012904698960483074, 0.025624195113778114, 0.0019859743770211935, 0.024698179215192795, -0.03442135080695152, -0.010993137024343014, -0.015133753418922424, 0.017316505312919617, 0.015464473515748978, 0.013810872100293636, 0.014604601077735424, 0.01586133800446987, 0.020306216552853584, 0.035267993807792664, 0.0040281713008880615, 0.009789315983653069, 0.008228316903114319, -0.02127191796898842, 0.0030674291774630547, -0.012633509002625942, 0.024486517533659935, 0.013162661343812943, -0.002999631455168128, -0.005523025989532471, -0.011376772075891495, 0.008459820412099361, -0.013810872100293636, 0.020438503473997116, -0.023679561913013458, 0.02076922357082367, 0.013533067889511585, 0.0018817975651472807, -0.021205775439739227, -0.008559036999940872, 0.0013088248670101166, -0.014194508083164692, -0.005751222837716341, -0.022343451157212257, 0.012620280496776104, 0.005966191180050373, 0.024473289027810097, -0.010821162723004818, -0.019750606268644333, 0.009187404997646809, 0.00077099131885916, 0.008843456394970417, -0.005301443859934807, -0.008810384199023247, 0.04579812288284302, -0.013003915548324585, 0.04415775090456009, -0.0016395450802519917, 0.005929811857640743, 0.012878241948783398, -0.006525108125060797, -0.022290537133812904, 0.003350194776430726, 0.0005477552185766399, 0.04505731165409088, 0.018242523074150085, -0.0012368932366371155, 0.03111414983868599, -0.00956442579627037, 0.013585982844233513, -0.00761979166418314, 0.00018830377666745335, 0.005959576927125454, -0.04505731165409088, -0.0033220837358385324, 0.01562321837991476, 0.016324345022439957, 0.013810872100293636, 0.010675645433366299, 0.015980396419763565, 0.014062220230698586, 0.024393916130065918, -0.0011236215941607952, -0.009015430696308613, -0.01658892072737217, -0.01586133800446987, 0.012891470454633236, -0.014882405288517475, 0.0031286124140024185, -0.01643017679452896, 0.012097742408514023, -0.029711896553635597, 0.0026771794073283672, -0.0031038084998726845, 0.01382410153746605, 0.009465210139751434, 0.0122631024569273, -0.009088189341127872, -0.021152859553694725, -0.029526693746447563, 0.011700877919793129, 0.014432626776397228, -0.025187645107507706, -0.021443892270326614, -0.03238411620259285, -0.0010922031942754984, -0.0192875973880291, 0.02883879467844963, -0.0010897228494286537, -0.0054138884879648685, -0.007176626939326525, 0.000608111615292728, -0.0003728869487531483, 0.013341249898076057, 0.020372359082102776, -0.007659478113055229, -0.031193522736430168, -0.002822696231305599, 0.00871116854250431, -0.006971580442041159, 0.03166975826025009, -0.02459234930574894, 0.017382649704813957, -0.013612440787255764, -0.019366970285773277, 0.004461414646357298, -0.006386205554008484, 0.0003677194472402334, -0.003548627020791173, 0.003052546875551343, 0.003627999685704708, -0.029950015246868134, -0.001493201358243823, 0.010682259686291218, 0.0036015422083437443, -0.017409106716513634, 0.02685447409749031, 0.0074345883913338184, -0.0012840209528803825, 0.0006221672520041466, -0.029182743281126022, -0.013850558549165726, 0.0459304116666317, 0.0245262049138546, -0.006773148197680712, 0.01971091888844967, 0.013314791955053806, -0.0036114638205617666, -0.0484967976808548, -0.004983952268958092, 0.014683973975479603, -0.01813669130206108, -0.002344805747270584, -0.002786317141726613, 0.001632103812880814, 0.011284170672297478, 0.029658980667591095, 0.005546176340430975, -0.0007606563158333302, -0.01325526274740696, 0.00559578463435173, -0.0013997729402035475, -0.00811587180942297, -2.780116119538434e-05, 0.01261366531252861, -0.003647842910140753, -0.01424742303788662, -0.012593822553753853, 0.009623955935239792, -0.0013195732608437538, 0.016813810914754868, -0.023084264248609543, -0.01579519361257553, -0.015371871180832386, -0.01427388098090887, -0.018652616068720818, -0.025677110999822617, 0.026550211012363434, 0.00010872424172703177, -0.021033799275755882, 0.00910803209990263, 0.01906270906329155, 0.027198422700166702, 0.024565890431404114, 0.01990935206413269, 0.0006754958885721862, 0.006012491881847382, -0.002195981564000249, 0.0023117337841540575, 0.023983823135495186, -0.020676622167229652, -0.00956442579627037, 0.01842772588133812, -0.007976969704031944, -0.037305232137441635, -0.014578143134713173, 0.005552791059017181, 0.014194508083164692, -0.023414984345436096, 0.00917417649179697, 0.013929931446909904, -0.0021844063885509968, -0.009531354531645775, -0.030241047963500023, -0.020015181973576546, -0.017673684284090996, -0.021510036662220955, -0.031008319929242134, -0.02091474086046219, -0.023838307708501816, -0.0009789315517991781, -0.0068525210954248905, 0.009716557338833809, 0.01414159219712019, -0.013195733539760113, -0.032913267612457275, 0.010543357580900192, 0.0002217891887994483, 0.02132483385503292, -0.01885104738175869, -0.01126432791352272, 0.04370797052979469, 0.00815555825829506, -0.0019975495524704456, -0.009392451494932175, -0.016985785216093063, 0.0007271709037013352, 0.029817726463079453, -0.013599211350083351, 0.014366482384502888, -0.019247911870479584, -0.005460189189761877, 0.02756883017718792, 0.007758694235235453, 0.0031153836753219366, 0.02849484607577324, -0.009723171591758728, -0.014287109486758709, 0.008902985602617264, 0.019882893189787865, 0.011819937266409397, -0.02727779559791088, 0.0023497664369642735, -0.0313258096575737, -0.010106806643307209, -0.004243139177560806, 0.025491908192634583, -0.017250362783670425, -0.0006403568550013006, 0.0029847491532564163, 0.025677110999822617, 0.017329733818769455, 0.035188622772693634, -0.027039676904678345, -0.006240688730031252, -0.0036908367183059454, 0.007520575542002916, 0.004626774694770575, 0.0005820674123242497, 0.021853985264897346, -0.013982847332954407, -0.01948603056371212, 0.013890244998037815, -0.0224625114351511, 0.004897965118288994, 0.05188337340950966, 0.0047722915187478065, 0.0015254465397447348, -0.025333162397146225, -0.04534834250807762, -0.030241047963500023, 0.00028813991229981184, 0.0033468876499682665, 0.00556271243840456, 0.005486647132784128, -0.021457122638821602, -0.04757078364491463, 0.0007970355218276381, -0.021933358162641525, 0.010417683981359005, 0.007011266890913248, -0.02252865582704544, -0.03778146579861641, -0.008744239807128906, -0.0015874565578997135, -0.010252323932945728, 0.037807926535606384, -0.03537382557988167, -0.019314056262373924, -0.009590883739292622, 0.004583781119436026, -0.0034361821599304676, -0.008360604755580425, 0.014485541731119156, -0.004236524924635887, 0.004699532873928547, 0.018665844574570656, -0.024221941828727722, -0.006068714428693056, 0.011773636564612389, 0.035902976989746094, 0.021774612367153168, 0.0385751947760582, -0.01219034381210804, -0.009339536540210247, 0.03182850405573845, -0.006601173896342516, -0.013757957145571709, -0.020451731979846954, -0.0027548987418413162, -0.036246925592422485, 0.006581330671906471, 0.016813810914754868, -0.0019016407895833254, -0.007553647737950087, -0.01673443801701069, -0.01665506511926651, 0.0024175641592592, -0.010576429776847363, 0.00563216395676136, -0.01066241692751646, -0.01893042027950287, -0.014287109486758709, -0.018229292705655098, 0.013585982844233513, 0.0016395450802519917, -0.00974301528185606, -0.02274031564593315, 0.033918656408786774, -0.008565651252865791, 0.026484068483114243, -0.01982997916638851, 0.007004652172327042, -0.019036250188946724, 0.0355590283870697, -0.002599460305646062, 0.006138165481388569, -0.01658892072737217, -0.0041306945495307446, 0.000991333625279367, 0.01530572772026062, 0.026603126898407936, 0.0162185151129961, 0.007004652172327042, -0.01714453101158142, -0.01562321837991476, 0.012924542650580406, -0.0022108640987426043, 0.005506489891558886, -0.0010905496310442686, 0.0018354967469349504, -0.020253300666809082, 0.0011442916002124548, -0.0433640219271183, -0.007196470163762569, -0.049502190202474594, 0.010490442626178265, 0.011072509922087193, -0.03280743584036827, 0.011819937266409397, -0.0127922547981143, -0.03407740220427513, -0.006081942934542894, 0.021166088059544563, 0.01013326458632946, -0.024182256311178207, 0.01885104738175869, 0.015226354822516441, -0.006898821797221899, -0.025068584829568863, -0.029182743281126022, 0.0026159961707890034, -0.012163885869085789, 0.016959328204393387, 0.05863006412982941, 0.006015799008309841, 0.01829543709754944, 0.007150168996304274, 0.017462022602558136, -0.0031451485119760036, -0.025028899312019348, 0.008473049849271774, -0.0012782332487404346, -0.0007325451006181538, 0.007282457314431667, -0.023772163316607475, -0.013691813684999943, 0.02401028200984001, -0.025412535294890404, -0.01764722540974617, -0.0344478078186512, -0.008466435596346855, -0.01219034381210804, 0.017554624006152153, -0.014525228179991245, 0.0042166817001998425, 0.0022174783516675234, -0.021020570769906044, -0.002096765674650669, -0.018520327284932137, -0.009518125094473362, 0.003216253127902746, -0.004428342450410128, 0.046591851860284805, -0.008307689800858498, 0.005731379613280296, 0.003826431930065155, 0.0161655992269516, -0.015609989874064922, 0.015318956226110458, -0.017673684284090996, 0.04048014432191849, -0.006197695154696703, -0.0006940988823771477, -0.004134001675993204, -0.007857910357415676, 0.03420969098806381, 0.0013567793648689985, 0.01764722540974617, -0.016072997823357582, 0.00027511781081557274, -0.0007329584914259613, -0.0054568820632994175, 0.021708469837903976, -0.008664866909384727, 0.011958839371800423, -0.007196470163762569, -0.02984418347477913, 0.0020157392136752605, -0.02153649553656578, -0.007170012220740318, 0.015054380521178246, -0.017475251108407974, -0.02670895680785179, -0.011641348712146282, 0.012626894749701023, -0.0010483828373253345, -0.030505623668432236, 0.022409595549106598, 0.03910434618592262, -0.03098186105489731, 0.009802544489502907, 0.007818223908543587, 0.0021496808622032404, -0.007791766431182623, 0.02062370628118515, 0.01363889779895544, -0.011509060859680176, 0.03434197977185249, -0.029023997485637665, -0.02692061848938465, -0.03357470780611038, -0.025240560993552208, 0.001802424667403102, 0.0058603608049452305, 0.00051344302482903, 0.030426252633333206, 0.015358642674982548, -0.003198063699528575, -0.002744977129623294, -0.014340024441480637, 0.002758205868303776, -0.01793825998902321, 0.007950511761009693, 0.006171237677335739, -0.018599700182676315, 0.008783926256000996, -0.0019396735588088632, 0.009220477193593979, -0.008777312003076077, -0.029182743281126022, -0.018176378682255745, 0.0045143296010792255, 0.025597738102078438, 0.0033749989233911037, -0.03796667233109474, -0.005655314307659864, -0.0161655992269516, -0.026060746982693672, 0.02301812171936035, 0.011542132124304771, -0.007004652172327042, 0.04775598645210266, 0.014088677242398262, -0.021139631047844887, 0.023150408640503883, 0.01992258056998253, -0.03627338260412216, -0.005182384513318539, -0.008294460363686085, -0.020292986184358597, -0.03225182741880417, -0.009445366449654102, 0.013929931446909904, 0.0127922547981143, -0.010007590986788273, -0.02714550867676735, 0.014750117436051369, -0.048946578055620193, -0.0027780490927398205, -0.01990935206413269, 0.010755018331110477, -0.014022533781826496, -0.0008214261033572257, 0.008955901488661766, 0.0072030844166874886, -0.003753673518076539, 0.01016633678227663, -0.015517388470470905, 0.0001765218621585518, -0.012838555499911308, -0.0009648759732954204, 0.017475251108407974, 0.010186179541051388, -0.019247911870479584, -0.017541395500302315, 0.01435325387865305, 0.010377997532486916, -0.009339536540210247, 0.02891816757619381, 0.008453206159174442, -0.020729538053274155, -0.0054668039083480835, -0.012474763207137585, -0.0006457310519181192, -0.0126533517614007, 0.004897965118288994, -0.035400282591581345, -0.008056342601776123, -0.0006238208152353764, 0.016840269789099693, -0.006673932075500488, 0.03487113118171692, -0.000625474436674267, -0.0038065887056291103, 0.0057578375563025475, -0.0014063873095437884, -0.01849387027323246, -0.004649925045669079, -0.011416458524763584, -0.0002236494910903275, 0.007996812462806702, -0.00807618536055088, 0.03704065456986427, -0.007566876709461212, -0.015517388470470905, -0.008598723448812962, 0.030293963849544525, -0.015186668373644352, 0.0007387460791505873, 0.010510285384953022, -0.016628608107566833, 0.012058055959641933, -0.012666581198573112, 0.0006936854915693402, 0.0021149551030248404, -0.0011831512674689293, -0.010040663182735443, -0.01367858424782753, -0.021496808156371117, 0.031431641429662704, 0.014154821634292603, -0.015398329123854637, 0.010331696830689907, -7.260340498760343e-05, -0.010483828373253345, -0.01055658608675003, -0.011297399178147316, 0.007474274840205908, -0.04452815651893616, -0.0035651628859341145, -0.01321557629853487, -0.005532947834581137, -0.014445855282247066, -0.005347744561731815, -0.0007127018761821091, -0.013308177702128887, 0.011006365530192852, 0.3092365562915802, -0.0245262049138546, -0.021655553951859474, 0.021205775439739227, 0.009200633503496647, 0.003270822111517191, 0.008850070647895336, -0.001989281503483653, -0.02395736612379551, 0.020253300666809082, 0.013771185651421547, -0.0062439958564937115, -0.001619701855815947, 0.0023547273594886065, 0.008334146812558174, -0.0371200256049633, -0.017951488494873047, -0.01123125571757555, 0.002242282498627901, -0.036114636808633804, 0.01162811927497387, -0.006320061627775431, 0.011899310164153576, -0.023362070322036743, 0.010318468324840069, -0.012388776056468487, -0.012924542650580406, -0.015702592208981514, 0.003919033333659172, 0.02963252365589142, -0.030743742361664772, -0.0203988179564476, -0.0012898085406050086, -0.015848107635974884, -0.016258200630545616, 0.0047623696736991405, 0.005238606594502926, 0.010212637484073639, 0.03039979375898838, 0.01271949615329504, 0.008453206159174442, -0.012712881900370121, 0.020795682445168495, 0.004845049697905779, 0.0016031658742576838, 0.009518125094473362, -0.01586133800446987, -0.016284659504890442, -0.01701224409043789, 0.024129340425133705, -0.005248528439551592, 0.0016742706065997481, 0.015543846413493156, 0.02854776196181774, -0.0007160090608522296, 0.0033171228133141994, 0.011681035161018372, -0.0036908367183059454, 0.0011715760920196772, 0.003505633445456624, -0.0021364521235227585, 0.03651150315999985, -0.0077851517125964165, 0.03521507978439331, -0.030293963849544525, 0.005933118984103203, -0.03897206112742424, -0.022687401622533798, -0.016760896891355515, -0.02833610028028488, 0.0014179626014083624, -0.01123125571757555, -0.013956389389932156, -0.013810872100293636, -0.0501636303961277, -0.02629886381328106, 0.03217245265841484, 0.03047916665673256, 0.039289552718400955, 0.03566486015915871, -0.01799117401242256, 0.019459571689367294, 0.008598723448812962, 0.0024555970449000597, 0.0024109496735036373, -0.05265064537525177, 0.014538456685841084, -0.013050216250121593, -0.02438068762421608, -0.033759910613298416, 0.010926992632448673, -0.022594798356294632, -0.006640860345214605, -0.018890732899308205, 0.018890732899308205, 0.01594071090221405, -0.0161655992269516, 0.024975983425974846, -0.007011266890913248, 0.008929443545639515, -0.038019586354494095, 0.0033766524866223335, 0.02168201096355915, 0.04296715930104256, -0.04026848450303078, -0.0012848477344959974, 0.0021529879886657, 0.015027922578155994, -0.0019810134544968605, -0.011462759226560593, -0.03098186105489731, -0.02133806236088276, 0.004150537773966789, 0.0014526881277561188, 0.005109625868499279, -0.0039024974685162306, -3.650840153568424e-05, -0.01317588984966278, 0.014075448736548424, -0.016707981005311012, 0.019750606268644333, -0.025782940909266472, -0.013122974894940853, -0.0030723900999873877, -0.015067609027028084, -0.022991662845015526, -0.007771923206746578, 0.008830226957798004, 0.03352179378271103, -0.047173917293548584, 0.028256727382540703, -0.014485541731119156, 0.04193531349301338, -0.0016345842741429806, -0.02153649553656578, 0.012838555499911308, -0.005926504731178284, -0.0023133873473852873, -0.01169426366686821, -0.010477213189005852, 0.008261389099061489, 0.01275918260216713, -0.008228316903114319, 0.005384123884141445, 0.0022158247884362936, 0.010126650333404541, 0.0406918041408062, -0.017951488494873047, 0.03251640126109123, -0.017395878210663795, -0.04373443126678467, -0.008856684900820255, 0.006349826231598854, -0.011409844271838665, 0.031140606850385666, 0.008142329752445221, -0.028177354484796524, -0.025399304926395416, -0.029791269451379776, -0.008704553358256817, -0.024645263329148293, 0.011151882819831371, 0.029658980667591095, -0.015781965106725693, -0.016774125397205353, 0.014697202481329441, -0.1691170483827591, 0.020292986184358597, 0.01123125571757555, -0.004653232172131538, 0.027198422700166702, -0.0018090391531586647, 0.019684461876749992, -0.0008098509279079735, 0.004864892922341824, -0.019075937569141388, 0.012064670212566853, -0.021086715161800385, -0.026206262409687042, 0.007659478113055229, -0.0009119607275351882, -0.0022885832004249096, -0.039077889174222946, -0.01481626182794571, 0.020742766559123993, 0.0037569806445389986, 0.027806948870420456, -0.005777680780738592, -0.007593334186822176, -0.0018735295161604881, -0.008506122045218945, 0.027224881574511528, -0.0016759242862462997, 0.02670895680785179, -0.013599211350083351, -0.00857887975871563, -0.0024522896856069565, -0.006984808947890997, 0.010153108276426792, 0.013705042190849781, 0.035479653626680374, 0.013387550599873066, -0.0038297390565276146, -0.016350803896784782, -0.009756243787705898, 0.02373247593641281, 0.035744231194257736, 0.022555112838745117, 0.013228804804384708, 0.016879955306649208, 0.000901212333701551, 0.02181429974734783, 0.01686672680079937, 0.005351051688194275, 0.029447320848703384, -0.004795441869646311, -0.006121629383414984, -0.002291890559718013, 0.017607539892196655, -0.0030971940141171217, 0.003429567674174905, 0.03585006296634674, -0.01484271977096796, 0.02522733062505722, -0.006845906842499971, 0.002389452885836363, -0.018943648785352707, -0.00920724868774414, -0.013063444755971432, 0.01367858424782753, -0.0033584628254175186, -0.021510036662220955, -0.017237132415175438, -0.0029781346675008535, -0.008916215039789677, 0.0008061302942223847, -0.02020038478076458, -0.0009111339459195733, 0.00659125205129385, 0.0052584498189389706, 0.020663393661379814, -0.0028574219904839993, -0.027304254472255707, 0.01427388098090887, -0.005215456243604422, -0.004934344440698624, 0.0030674291774630547, -0.004253061022609472, -0.011429687961935997, -0.00035490404115989804, 0.007004652172327042, 0.020253300666809082, 0.0042166817001998425, -0.008856684900820255, 0.01207789871841669, -0.02436745911836624, 0.01233586110174656, -0.04323173686861992, -0.011641348712146282, -0.025610966607928276, -0.015477702021598816, 0.01829543709754944, 0.006227460224181414, -0.0030508930794894695, -0.009365993551909924, -0.017528166994452477, -0.02465849183499813, -0.01382410153746605, -0.01792503148317337, 0.00914771854877472, 0.017528166994452477, 0.013493381440639496, 0.008565651252865791, 0.01715775951743126, 0.02401028200984001, -0.005665235687047243, -0.026669271290302277, -0.001802424667403102, 0.018388038501143456, 0.019935809075832367, 0.006938508246093988, 0.006270453799515963, -0.014472313225269318, -0.019790291786193848, 0.023071035742759705, 0.0006085250061005354, 0.00251512648537755, -0.01694609969854355, 0.0019975495524704456, 0.022449282929301262, 0.016536006703972816, -0.02254188433289528, -0.05783633515238762, -0.003965334501117468, 0.023004891350865364, 0.025981374084949493, 0.011747178621590137, 0.016694752499461174, 0.019605088979005814, 0.02651052549481392, -0.014882405288517475, -0.006515186745673418, -0.006779762450605631, -0.019023021683096886, -0.0014378057094290853, -0.0031782204750925303, 0.01883781887590885, -0.008810384199023247, 0.010179565288126469, -0.018083777278661728, -0.004937651567161083, 0.05040174722671509, 0.008486278355121613, 0.004431649576872587, -0.01113203912973404, 0.014366482384502888, 0.00742135988548398, 0.027383627369999886, -0.021099943667650223, 0.03174913302063942, 0.009266777895390987, -0.017197446897625923, 0.01877167448401451, 0.0006002570153214037, -0.010232481174170971, 0.002902069129049778, 0.017594311386346817, -0.0006870710640214384, -0.021430663764476776, -0.0010144839761778712, 0.029870642349123955, -0.03754334896802902, 0.010715331882238388, 0.02395736612379551, -0.010311853140592575, -0.02119254507124424, -0.0074412026442587376, -0.006121629383414984, 0.008889757096767426, 0.021099943667650223, 0.009696714580059052, -0.01849387027323246, -0.0037338302936404943, -0.007097254041582346, -0.016072997823357582, 0.012064670212566853, 0.024142568930983543, -0.008889757096767426, 0.004167073406279087, -0.001771006267517805, -0.008473049849271774, -0.0014353253645822406, -0.02047819085419178, 0.0013948121340945363, -0.008486278355121613, 0.015504159964621067, 0.01742233708500862, -0.005387431010603905, 0.021801071241497993, 0.014379710890352726, 0.005913275759667158, -0.01871875859797001, 0.002612689044326544, 0.007216313388198614, -0.00635974807664752, 0.007738851010799408, -0.029235659167170525, -0.028441930189728737, -0.02685447409749031, 0.0023233089596033096, 0.003796667093411088, -0.009339536540210247, -0.025253789499402046, -0.010305238887667656, 0.01471043098717928, -0.026179805397987366, 0.017462022602558136, 0.02026652917265892, 0.01730327680706978, 0.0021596024744212627, 0.011575204320251942, -0.03452718257904053, -0.011006365530192852, 0.023362070322036743, 0.03542673960328102, -0.025028899312019348, -0.013890244998037815, 0.0040281713008880615, 0.016469862312078476, -0.0035816989839076996, 0.015252811834216118, -0.0032972795888781548, -0.012084513902664185, -0.01382410153746605, -0.07143554836511612, 0.0016006854129955173, 0.009008816443383694, -0.008062956854701042, 0.013572754338383675, -0.0006147260428406298, -0.006230767350643873, 0.007427974138408899, -0.01735619269311428, 0.010411069728434086, -0.006277068052440882, -0.011522289365530014, 0.005433731712400913, -0.01940665766596794, -0.02174815535545349, -0.024063196033239365, 0.01307005900889635, 9.740741370478645e-05, 0.00027077709091827273, 0.010801319032907486, 0.013275105506181717, -0.013202347792685032, 0.008234931156039238, 0.020610477775335312, -0.01435325387865305, -7.136320346035063e-05, -0.024420375004410744, 0.03309847041964531, -0.013017144054174423, -0.018057318404316902, -0.0037338302936404943, -0.01766045391559601, -0.008426749147474766, -0.016522778198122978, -0.011376772075891495, -0.008406905457377434, 0.010523514822125435, 0.015993624925613403, 0.015927480533719063, 0.02849484607577324, 0.003356809262186289, -0.03495050221681595, 0.009789315983653069, 0.0018784903222694993, -0.03505633398890495, 0.023864764720201492, -0.018123462796211243, 0.008764083497226238, 0.042834870517253876, 0.0036246925592422485, 0.012838555499911308, 0.012104356661438942, -0.023401755839586258, -0.0007098080823197961, 0.00963718444108963, -0.02698676288127899, 0.029394404962658882, 0.013533067889511585, -0.012560750357806683, -0.015967167913913727, 0.017395878210663795, 0.015781965106725693, -0.0021000728011131287, -0.020425274968147278, 0.010483828373253345, 0.010463984683156013, -0.010014205239713192, 0.004808670841157436, -0.005496568512171507, -0.03325721621513367, -0.00434235529974103, 0.006677239201962948, 0.028785880655050278, 0.007685935590416193, -0.015292498283088207, -0.002085190499201417, 0.008096029050648212, -0.02027975767850876, -0.02226407825946808, 0.013943160884082317, 0.014789803884923458, 0.020782452076673508, -0.02295197732746601, 0.009769472293555737, 0.010867463424801826, 0.043337564915418625, -0.014287109486758709, 0.007262614089995623, -0.002007471164688468, 0.021920129656791687, 0.0033038940746337175, 0.013705042190849781, -0.018057318404316902, 0.030426252633333206, -0.009326308034360409, 0.04712100327014923, 0.019023021683096886, -0.016919640824198723, 0.03749043494462967, 0.029711896553635597, 0.029447320848703384, -0.006531722377985716, -0.005847131833434105, 0.004329126328229904, -0.021562952548265457, -0.007097254041582346, -0.006085250526666641, -0.02487015351653099, 0.006878978572785854, -0.0074412026442587376, 0.0008011694881133735, -0.003221214050427079, 1.7362806829623878e-05, 0.0370141975581646, -0.02522733062505722, -0.0006990596884861588, 0.005529640708118677, -0.00867809634655714, -0.029447320848703384, 0.022978434339165688, 0.018242523074150085, 0.023282697424292564, 0.023150408640503883, -0.00011389174323994666, 0.018877504393458366, 0.014922091737389565, -0.0020355824381113052, -0.014234194532036781, 0.01673443801701069, 0.013447080738842487, 0.01666829362511635, -0.002316694473847747, -0.017488479614257812, -0.008506122045218945, -0.020041638985276222, 0.03209308162331581, -0.016602151095867157, 0.0242483988404274, -0.02210533246397972, 0.03931600973010063, -0.003475868608802557, -0.015385100618004799, 0.013189118355512619, -0.0161655992269516, 0.010186179541051388, -0.009227091446518898, -0.004653232172131538, -0.022145019844174385, -0.010715331882238388, 0.020306216552853584, -0.0030244356021285057, 0.009008816443383694, -0.007480889093130827, -0.04783535748720169, 0.013493381440639496, 0.009365993551909924, -0.0050302529707551, 0.001941327122040093, 0.0195654034614563, 0.015755506232380867, 0.012468148954212666, -0.0012410272611305118, 0.01935374177992344, -0.020081326365470886, -0.0029715204145759344, 0.008883142843842506, 0.018943648785352707, 0.007017881143838167, -0.03026750683784485, 0.028097981587052345, 0.0030740436632186174, -0.01471043098717928, -0.024923069402575493, 0.01530572772026062, -0.020253300666809082, -0.002870650729164481, -0.009848845191299915, 0.008929443545639515, 0.009822388179600239, 0.026629583910107613, 0.026325322687625885, -0.03436843678355217, -0.025399304926395416, 0.040294941514730453, -0.020861824974417686, 0.0017809278797358274, 0.007467660587280989, 0.010933607816696167]\n",
      "1536\n"
     ]
    }
   ],
   "source": [
    "print(positive_review)\n",
    "print(len(positive_review))\n",
    "\n",
    "print(negative_review)\n",
    "print(len(negative_review))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cosine Similarity between positive and neutual review: 0.9347902125441199\n"
     ]
    }
   ],
   "source": [
    "print(\"Cosine Similarity between positive and neutual review:\", cosine_similarity(positive_review, negative_review))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cosine Similarity between positive review and itself: 1.0\n"
     ]
    }
   ],
   "source": [
    "print(\"Cosine Similarity between positive review and itself:\", cosine_similarity(positive_review, positive_review))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cosine Similarity between positive and good restaurant review: 0.8379052807845067\n",
      "Cosine Similarity between negative and good restaurant review: 0.8044964373440723\n"
     ]
    }
   ],
   "source": [
    "good_restraurant = get_embedding(\"这家餐馆太好吃了，一点都不糟糕\", engine=EMBEDDING_MODEL)\n",
    "bad_restraurant = get_embedding(\"这家餐馆太糟糕了，一点都不好吃\", engine=EMBEDDING_MODEL)\n",
    "\n",
    "print(\"Cosine Similarity between positive and good restaurant review:\", cosine_similarity(positive_review, good_restraurant))\n",
    "print(\"Cosine Similarity between negative and good restaurant review:\", cosine_similarity(negative_review, good_restraurant))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "好评例子的评分 : 0.033409\n",
      "差评例子的评分 : -0.027144\n"
     ]
    }
   ],
   "source": [
    "\n",
    "def get_score(sample_embedding):\n",
    "  return cosine_similarity(sample_embedding, positive_review) - cosine_similarity(sample_embedding, negative_review)\n",
    "\n",
    "positive_score = get_score(good_restraurant)\n",
    "negative_score = get_score(bad_restraurant)\n",
    "\n",
    "\n",
    "print(\"好评例子的评分 : %f\" % (positive_score))\n",
    "print(\"差评例子的评分 : %f\" % (negative_score))\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 生成一个正交的，Cosine Similarity 为 0 的向量"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-9.773366928911903e-18\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "# 假设原向量是一个高维向量\n",
    "original_vector = np.array(positive_review)\n",
    "\n",
    "# 对原向量进行归一化\n",
    "original_vector /= np.linalg.norm(original_vector)\n",
    "\n",
    "# 从单位高斯分布中随机抽取一个向量\n",
    "random_vector = np.random.randn(len(original_vector))\n",
    "\n",
    "# 从随机向量中减去在原向量方向上的分量\n",
    "orthogonal_vector = random_vector - random_vector.dot(original_vector) * original_vector\n",
    "\n",
    "# 验证新向量与原向量是否正交（即它们的余弦相似度是否为0）\n",
    "cos_sim = np.dot(original_vector, orthogonal_vector) / (np.linalg.norm(original_vector) * np.linalg.norm(orthogonal_vector))\n",
    "print(cos_sim)  # 这个值应该接近0\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 降维可视化向量"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "from sklearn.decomposition import PCA\n",
    "\n",
    "def visualize_words(words,vectors):\n",
    "    # Create a PCA object and fit_transform the vectors\n",
    "    pca = PCA(n_components=2)\n",
    "    pca_result = pca.fit_transform(vectors)\n",
    "\n",
    "    # Create a color map\n",
    "    colors = [\"b\", \"g\", \"r\", \"c\"]\n",
    "\n",
    "    # Plot each word in 2D space using PCA results\n",
    "    for i, word in enumerate(words):\n",
    "        plt.scatter(pca_result[i, 0], pca_result[i, 1], c=colors[i], label=word)\n",
    "        plt.annotate(word, xy=(pca_result[i, 0], pca_result[i, 1]))\n",
    "\n",
    "    plt.legend()\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "words = [f\"Positive\", \"Negative\", \"Good Restaurant\", \"Bad Restaurant\"]\n",
    "vectors = [np.array(x) for x in [positive_review, negative_review, good_restraurant, bad_restraurant]]\n",
    "\n",
    "visualize_words(words, vectors)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "good_phone = get_embedding(\"买的银色版真的很好看，一天就到了，晚上就开始拿起来完系统很丝滑流畅，做工扎实，手感细腻，很精致哦苹果一如既往的好品质\", engine=EMBEDDING_MODEL)\n",
    "bad_phone = get_embedding(\"随意降价，不予价保，服务态度差\", engine=EMBEDDING_MODEL)\n",
    "\n",
    "words = [f\"Positive\", \"Negative\", \"Good Phone\", \"Bad Phone\"]\n",
    "vectors = [np.array(x) for x in [positive_review, negative_review, good_phone, bad_phone]]\n",
    "\n",
    "visualize_words(words, vectors)\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 通过向量进行零样本分类"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "\n",
    "from sklearn.metrics import classification_report\n",
    "\n",
    "datafile_path = \"data/fine_food_reviews_with_embeddings_1k.csv\"\n",
    "\n",
    "df = pd.read_csv(datafile_path)\n",
    "df[\"embedding\"] = df.embedding.apply(eval).apply(np.array) # type: ignore\n",
    "\n",
    "# convert 5-star rating to binary sentiment\n",
    "df = df[df.Score != 3]\n",
    "df[\"sentiment\"] = df.Score.replace({1: \"negative\", 2: \"negative\", 4: \"positive\", 5: \"positive\"})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Unnamed: 0</th>\n",
       "      <th>ProductId</th>\n",
       "      <th>UserId</th>\n",
       "      <th>Score</th>\n",
       "      <th>Summary</th>\n",
       "      <th>Text</th>\n",
       "      <th>combined</th>\n",
       "      <th>n_tokens</th>\n",
       "      <th>embedding</th>\n",
       "      <th>sentiment</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>B003XPF9BO</td>\n",
       "      <td>A3R7JR3FMEBXQB</td>\n",
       "      <td>5</td>\n",
       "      <td>where does one  start...and stop... with a tre...</td>\n",
       "      <td>Wanted to save some to bring to my Chicago fam...</td>\n",
       "      <td>Title: where does one  start...and stop... wit...</td>\n",
       "      <td>52</td>\n",
       "      <td>[0.007018072064965963, -0.02731654793024063, 0...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>297</td>\n",
       "      <td>B003VXHGPK</td>\n",
       "      <td>A21VWSCGW7UUAR</td>\n",
       "      <td>4</td>\n",
       "      <td>Good, but not Wolfgang Puck good</td>\n",
       "      <td>Honestly, I have to admit that I expected a li...</td>\n",
       "      <td>Title: Good, but not Wolfgang Puck good; Conte...</td>\n",
       "      <td>178</td>\n",
       "      <td>[-0.003140551969408989, -0.009995664469897747,...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>296</td>\n",
       "      <td>B008JKTTUA</td>\n",
       "      <td>A34XBAIFT02B60</td>\n",
       "      <td>1</td>\n",
       "      <td>Should advertise coconut as an ingredient more...</td>\n",
       "      <td>First, these should be called Mac - Coconut ba...</td>\n",
       "      <td>Title: Should advertise coconut as an ingredie...</td>\n",
       "      <td>78</td>\n",
       "      <td>[-0.01757248118519783, -8.266511576948687e-05,...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>295</td>\n",
       "      <td>B000LKTTTW</td>\n",
       "      <td>A14MQ40CCU8B13</td>\n",
       "      <td>5</td>\n",
       "      <td>Best tomato soup</td>\n",
       "      <td>I have a hard time finding packaged food of an...</td>\n",
       "      <td>Title: Best tomato soup; Content: I have a har...</td>\n",
       "      <td>111</td>\n",
       "      <td>[-0.0013932279543951154, -0.011112828738987446...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>294</td>\n",
       "      <td>B001D09KAM</td>\n",
       "      <td>A34XBAIFT02B60</td>\n",
       "      <td>1</td>\n",
       "      <td>Should advertise coconut as an ingredient more...</td>\n",
       "      <td>First, these should be called Mac - Coconut ba...</td>\n",
       "      <td>Title: Should advertise coconut as an ingredie...</td>\n",
       "      <td>78</td>\n",
       "      <td>[-0.01757248118519783, -8.266511576948687e-05,...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Unnamed: 0   ProductId          UserId  Score   \n",
       "0           0  B003XPF9BO  A3R7JR3FMEBXQB      5  \\\n",
       "1         297  B003VXHGPK  A21VWSCGW7UUAR      4   \n",
       "2         296  B008JKTTUA  A34XBAIFT02B60      1   \n",
       "3         295  B000LKTTTW  A14MQ40CCU8B13      5   \n",
       "4         294  B001D09KAM  A34XBAIFT02B60      1   \n",
       "\n",
       "                                             Summary   \n",
       "0  where does one  start...and stop... with a tre...  \\\n",
       "1                   Good, but not Wolfgang Puck good   \n",
       "2  Should advertise coconut as an ingredient more...   \n",
       "3                                   Best tomato soup   \n",
       "4  Should advertise coconut as an ingredient more...   \n",
       "\n",
       "                                                Text   \n",
       "0  Wanted to save some to bring to my Chicago fam...  \\\n",
       "1  Honestly, I have to admit that I expected a li...   \n",
       "2  First, these should be called Mac - Coconut ba...   \n",
       "3  I have a hard time finding packaged food of an...   \n",
       "4  First, these should be called Mac - Coconut ba...   \n",
       "\n",
       "                                            combined  n_tokens   \n",
       "0  Title: where does one  start...and stop... wit...        52  \\\n",
       "1  Title: Good, but not Wolfgang Puck good; Conte...       178   \n",
       "2  Title: Should advertise coconut as an ingredie...        78   \n",
       "3  Title: Best tomato soup; Content: I have a har...       111   \n",
       "4  Title: Should advertise coconut as an ingredie...        78   \n",
       "\n",
       "                                           embedding sentiment  \n",
       "0  [0.007018072064965963, -0.02731654793024063, 0...  positive  \n",
       "1  [-0.003140551969408989, -0.009995664469897747,...  positive  \n",
       "2  [-0.01757248118519783, -8.266511576948687e-05,...  negative  \n",
       "3  [-0.0013932279543951154, -0.011112828738987446...  positive  \n",
       "4  [-0.01757248118519783, -8.266511576948687e-05,...  negative  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'EMBEDDING_MODEL' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[5], line 6\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39msklearn\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39mmetrics\u001b[39;00m \u001b[39mimport\u001b[39;00m PrecisionRecallDisplay\n\u001b[1;32m      2\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mopenai\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39membeddings_utils\u001b[39;00m \u001b[39mimport\u001b[39;00m cosine_similarity, get_embedding\n\u001b[1;32m      4\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mevaluate_embeddings_approach\u001b[39m(\n\u001b[1;32m      5\u001b[0m     labels \u001b[39m=\u001b[39m [\u001b[39m'\u001b[39m\u001b[39mnegative\u001b[39m\u001b[39m'\u001b[39m, \u001b[39m'\u001b[39m\u001b[39mpositive\u001b[39m\u001b[39m'\u001b[39m], \n\u001b[0;32m----> 6\u001b[0m     model \u001b[39m=\u001b[39m EMBEDDING_MODEL,\n\u001b[1;32m      7\u001b[0m ):\n\u001b[1;32m      8\u001b[0m     label_embeddings \u001b[39m=\u001b[39m [get_embedding(label, engine\u001b[39m=\u001b[39mmodel) \u001b[39mfor\u001b[39;00m label \u001b[39min\u001b[39;00m labels]\n\u001b[1;32m     10\u001b[0m     \u001b[39mdef\u001b[39;00m \u001b[39mlabel_score\u001b[39m(review_embedding, label_embeddings):\n",
      "\u001b[0;31mNameError\u001b[0m: name 'EMBEDDING_MODEL' is not defined"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import PrecisionRecallDisplay\n",
    "from openai.embeddings_utils import cosine_similarity, get_embedding\n",
    "\n",
    "def evaluate_embeddings_approach(\n",
    "    labels = ['negative', 'positive'], \n",
    "    model = EMBEDDING_MODEL,\n",
    "):\n",
    "    label_embeddings = [get_embedding(label, engine=model) for label in labels]\n",
    "\n",
    "    def label_score(review_embedding, label_embeddings):\n",
    "        return cosine_similarity(review_embedding, label_embeddings[1]) - cosine_similarity(review_embedding, label_embeddings[0])\n",
    "\n",
    "    probas = df[\"embedding\"].apply(lambda x: label_score(x, label_embeddings))\n",
    "    preds = probas.apply(lambda x: 'positive' if x>0 else 'negative')\n",
    "\n",
    "    report = classification_report(df.sentiment, preds)\n",
    "    print(report)\n",
    "\n",
    "    display = PrecisionRecallDisplay.from_predictions(df.sentiment, probas, pos_label='positive')\n",
    "    _ = display.ax_.set_title(\"2-class Precision-Recall curve\")\n",
    "\n",
    "evaluate_embeddings_approach(labels=['An Amazon review with a negative sentiment.', 'An Amazon review with a positive sentiment.'])"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 通过机器学习进行分类"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'pd' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[72], line 5\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39msklearn\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39mmodel_selection\u001b[39;00m \u001b[39mimport\u001b[39;00m train_test_split\n\u001b[1;32m      3\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39msklearn\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39mmetrics\u001b[39;00m \u001b[39mimport\u001b[39;00m classification_report, accuracy_score\n\u001b[0;32m----> 5\u001b[0m training_data \u001b[39m=\u001b[39m pd\u001b[39m.\u001b[39mread_parquet(\u001b[39m\"\u001b[39m\u001b[39mdata/toutiao_cat_data_all_with_embeddings.parquet\u001b[39m\u001b[39m\"\u001b[39m)\n\u001b[1;32m      7\u001b[0m df \u001b[39m=\u001b[39m  training_data\u001b[39m.\u001b[39msample(\u001b[39m50000\u001b[39m, random_state\u001b[39m=\u001b[39m\u001b[39m42\u001b[39m)\n\u001b[1;32m      9\u001b[0m X_train, X_test, y_train, y_test \u001b[39m=\u001b[39m train_test_split(\n\u001b[1;32m     10\u001b[0m     \u001b[39mlist\u001b[39m(df\u001b[39m.\u001b[39membedding\u001b[39m.\u001b[39mvalues), df\u001b[39m.\u001b[39mcategory, test_size\u001b[39m=\u001b[39m\u001b[39m0.2\u001b[39m, random_state\u001b[39m=\u001b[39m\u001b[39m42\u001b[39m\n\u001b[1;32m     11\u001b[0m )\n",
      "\u001b[0;31mNameError\u001b[0m: name 'pd' is not defined"
     ]
    }
   ],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import classification_report, accuracy_score\n",
    "\n",
    "training_data = pd.read_parquet(\"data/toutiao_cat_data_all_with_embeddings.parquet\")\n",
    "\n",
    "df =  training_data.sample(50000, random_state=42)\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(\n",
    "    list(df.embedding.values), df.category, test_size=0.2, random_state=42\n",
    ")\n",
    "\n",
    "clf = LogisticRegression()\n",
    "clf.fit(X_train, y_train)\n",
    "preds = clf.predict(X_test)\n",
    "probas = clf.predict_proba(X_test)\n",
    "\n",
    "report = classification_report(y_test, preds)\n",
    "print(report)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>code</th>\n",
       "      <th>category</th>\n",
       "      <th>title</th>\n",
       "      <th>keywords</th>\n",
       "      <th>combined</th>\n",
       "      <th>n_tokens</th>\n",
       "      <th>embedding</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>306237</th>\n",
       "      <td>6524143112997044743</td>\n",
       "      <td>100</td>\n",
       "      <td>news_story</td>\n",
       "      <td>将女儿扔老家10年后我接她回家，没多久妻子被杀我才知她身份不对</td>\n",
       "      <td>小安,不好意思,她在杀人,星期六,女孩儿,超能力,工程事故,冯为杰,冯欣冉,浏览器,狄海</td>\n",
       "      <td>标题: 将女儿扔老家10年后我接她回家，没多久妻子被杀我才知她身份不对; 关键字: 小安,不...</td>\n",
       "      <td>110</td>\n",
       "      <td>[-0.015406427904963493, -0.007328800391405821,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>123419</th>\n",
       "      <td>6553127618529460483</td>\n",
       "      <td>101</td>\n",
       "      <td>news_culture</td>\n",
       "      <td>在中国古代，有没有人认为地球是圆的呢？</td>\n",
       "      <td></td>\n",
       "      <td>标题: 在中国古代，有没有人认为地球是圆的呢？; 关键字:</td>\n",
       "      <td>28</td>\n",
       "      <td>[0.023000160232186317, -0.004472616594284773, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49255</th>\n",
       "      <td>6511643474234704135</td>\n",
       "      <td>113</td>\n",
       "      <td>news_world</td>\n",
       "      <td>先冻结房产，再冻结养老活命钱，韩国为何如此苛刻对待前总统朴槿惠？</td>\n",
       "      <td>前总统,朴槿惠,检方,韩国,韩国法院</td>\n",
       "      <td>标题: 先冻结房产，再冻结养老活命钱，韩国为何如此苛刻对待前总统朴槿惠？; 关键字: 前总统...</td>\n",
       "      <td>81</td>\n",
       "      <td>[0.0021718391217291355, -0.01761491596698761, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>135430</th>\n",
       "      <td>6553195755832606979</td>\n",
       "      <td>109</td>\n",
       "      <td>news_tech</td>\n",
       "      <td>电脑总是无响应是什么回事？</td>\n",
       "      <td></td>\n",
       "      <td>标题: 电脑总是无响应是什么回事？; 关键字:</td>\n",
       "      <td>26</td>\n",
       "      <td>[-0.01490439660847187, -0.012121899053454399, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>331509</th>\n",
       "      <td>6554427356038038019</td>\n",
       "      <td>113</td>\n",
       "      <td>news_world</td>\n",
       "      <td>中国宣布全面终止合作，这个国家将付出沉重代价！</td>\n",
       "      <td>天然气,出口商,煤炭,中方,澳大利亚</td>\n",
       "      <td>标题: 中国宣布全面终止合作，这个国家将付出沉重代价！; 关键字: 天然气,出口商,煤炭,中...</td>\n",
       "      <td>59</td>\n",
       "      <td>[-0.01608712412416935, -0.033476412296295166, ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                         id  code      category   \n",
       "306237  6524143112997044743   100    news_story  \\\n",
       "123419  6553127618529460483   101  news_culture   \n",
       "49255   6511643474234704135   113    news_world   \n",
       "135430  6553195755832606979   109     news_tech   \n",
       "331509  6554427356038038019   113    news_world   \n",
       "\n",
       "                                   title   \n",
       "306237   将女儿扔老家10年后我接她回家，没多久妻子被杀我才知她身份不对  \\\n",
       "123419               在中国古代，有没有人认为地球是圆的呢？   \n",
       "49255   先冻结房产，再冻结养老活命钱，韩国为何如此苛刻对待前总统朴槿惠？   \n",
       "135430                     电脑总是无响应是什么回事？   \n",
       "331509           中国宣布全面终止合作，这个国家将付出沉重代价！   \n",
       "\n",
       "                                            keywords   \n",
       "306237  小安,不好意思,她在杀人,星期六,女孩儿,超能力,工程事故,冯为杰,冯欣冉,浏览器,狄海  \\\n",
       "123419                                                 \n",
       "49255                             前总统,朴槿惠,检方,韩国,韩国法院   \n",
       "135430                                                 \n",
       "331509                            天然气,出口商,煤炭,中方,澳大利亚   \n",
       "\n",
       "                                                 combined  n_tokens   \n",
       "306237  标题: 将女儿扔老家10年后我接她回家，没多久妻子被杀我才知她身份不对; 关键字: 小安,不...       110  \\\n",
       "123419                     标题: 在中国古代，有没有人认为地球是圆的呢？; 关键字:         28   \n",
       "49255   标题: 先冻结房产，再冻结养老活命钱，韩国为何如此苛刻对待前总统朴槿惠？; 关键字: 前总统...        81   \n",
       "135430                           标题: 电脑总是无响应是什么回事？; 关键字:         26   \n",
       "331509  标题: 中国宣布全面终止合作，这个国家将付出沉重代价！; 关键字: 天然气,出口商,煤炭,中...        59   \n",
       "\n",
       "                                                embedding  \n",
       "306237  [-0.015406427904963493, -0.007328800391405821,...  \n",
       "123419  [0.023000160232186317, -0.004472616594284773, ...  \n",
       "49255   [0.0021718391217291355, -0.01761491596698761, ...  \n",
       "135430  [-0.01490439660847187, -0.012121899053454399, ...  \n",
       "331509  [-0.01608712412416935, -0.033476412296295166, ...  "
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: news_sports\n"
     ]
    }
   ],
   "source": [
    "sports_news = get_embedding(\"【西海岸】ESPN：马龙教练不容易，保持核心掘金能复刻冠军\", engine=EMBEDDING_MODEL)\n",
    "prediction = clf.predict([sports_news])[0]\n",
    "print(f\"Prediction: {prediction}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: news_finance\n"
     ]
    }
   ],
   "source": [
    "finance_news = get_embedding(\"北上资金净买入超百亿元 重点加仓新能源汽车产业\", engine=EMBEDDING_MODEL)\n",
    "prediction = clf.predict([finance_news])[0]\n",
    "print(f\"Prediction: {prediction}\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 通过机器学习进行聚类"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'pd' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[73], line 4\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[39mimport\u001b[39;00m \u001b[39mnumpy\u001b[39;00m \u001b[39mas\u001b[39;00m \u001b[39mnp\u001b[39;00m\n\u001b[1;32m      2\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39msklearn\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39mcluster\u001b[39;00m \u001b[39mimport\u001b[39;00m KMeans\n\u001b[0;32m----> 4\u001b[0m embedding_df \u001b[39m=\u001b[39m pd\u001b[39m.\u001b[39mread_parquet(\u001b[39m\"\u001b[39m\u001b[39mdata/20_newsgroup_with_embedding.parquet\u001b[39m\u001b[39m\"\u001b[39m)\n\u001b[1;32m      6\u001b[0m matrix \u001b[39m=\u001b[39m np\u001b[39m.\u001b[39mvstack(embedding_df\u001b[39m.\u001b[39membedding\u001b[39m.\u001b[39mvalues) \u001b[39m# type: ignore\u001b[39;00m\n\u001b[1;32m      7\u001b[0m num_of_clusters \u001b[39m=\u001b[39m \u001b[39m20\u001b[39m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'pd' is not defined"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from sklearn.cluster import KMeans\n",
    "\n",
    "embedding_df = pd.read_parquet(\"data/20_newsgroup_with_embedding.parquet\")\n",
    "\n",
    "matrix = np.vstack(embedding_df.embedding.values) # type: ignore\n",
    "num_of_clusters = 20\n",
    "\n",
    "kmeans = KMeans(n_clusters=num_of_clusters, init=\"k-means++\", n_init=10, random_state=42)\n",
    "kmeans.fit(matrix)\n",
    "labels = kmeans.labels_\n",
    "embedding_df[\"cluster\"] = labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>target</th>\n",
       "      <th>title</th>\n",
       "      <th>n_tokens</th>\n",
       "      <th>embedding</th>\n",
       "      <th>cluster</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>I was wondering if anyone out there could enli...</td>\n",
       "      <td>7</td>\n",
       "      <td>rec.autos</td>\n",
       "      <td>121</td>\n",
       "      <td>[-0.0391300804913044, 0.013502633199095726, -0...</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>\\nIt depends on your priorities.  A lot of peo...</td>\n",
       "      <td>7</td>\n",
       "      <td>rec.autos</td>\n",
       "      <td>108</td>\n",
       "      <td>[-0.0011249205563217402, -0.00376517535187304,...</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>an excellent automatic can be found in the sub...</td>\n",
       "      <td>7</td>\n",
       "      <td>rec.autos</td>\n",
       "      <td>476</td>\n",
       "      <td>[-0.018259447067975998, -0.008410007692873478,...</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>: Ford and his automobile.  I need information...</td>\n",
       "      <td>7</td>\n",
       "      <td>rec.autos</td>\n",
       "      <td>86</td>\n",
       "      <td>[-0.012589422054588795, 0.006539034191519022, ...</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>\\nYo! Watch the attributions--I didn't say tha...</td>\n",
       "      <td>7</td>\n",
       "      <td>rec.autos</td>\n",
       "      <td>130</td>\n",
       "      <td>[-0.0006192282889969647, -0.011226896196603775...</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                text  target      title   \n",
       "0  I was wondering if anyone out there could enli...       7  rec.autos  \\\n",
       "1  \\nIt depends on your priorities.  A lot of peo...       7  rec.autos   \n",
       "2  an excellent automatic can be found in the sub...       7  rec.autos   \n",
       "3  : Ford and his automobile.  I need information...       7  rec.autos   \n",
       "4  \\nYo! Watch the attributions--I didn't say tha...       7  rec.autos   \n",
       "\n",
       "   n_tokens                                          embedding  cluster  \n",
       "0       121  [-0.0391300804913044, 0.013502633199095726, -0...        5  \n",
       "1       108  [-0.0011249205563217402, -0.00376517535187304,...        5  \n",
       "2       476  [-0.018259447067975998, -0.008410007692873478,...        5  \n",
       "3        86  [-0.012589422054588795, 0.006539034191519022, ...        5  \n",
       "4       130  [-0.0006192282889969647, -0.011226896196603775...       10  "
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "embedding_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>cluster</th>\n",
       "      <th>count</th>\n",
       "      <th>rank1</th>\n",
       "      <th>rank1_count</th>\n",
       "      <th>rank2</th>\n",
       "      <th>rank2_count</th>\n",
       "      <th>per_1</th>\n",
       "      <th>per_1_2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>432</td>\n",
       "      <td>comp.windows.x</td>\n",
       "      <td>406</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>1.0</td>\n",
       "      <td>93.98%</td>\n",
       "      <td>94.21%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>418</td>\n",
       "      <td>sci.space</td>\n",
       "      <td>388</td>\n",
       "      <td>alt.atheism</td>\n",
       "      <td>2.0</td>\n",
       "      <td>92.82%</td>\n",
       "      <td>93.30%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>1035</td>\n",
       "      <td>comp.sys.ibm.pc.hardware</td>\n",
       "      <td>396</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>387.0</td>\n",
       "      <td>38.26%</td>\n",
       "      <td>75.65%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>471</td>\n",
       "      <td>rec.sport.hockey</td>\n",
       "      <td>455</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>96.60%</td>\n",
       "      <td>96.60%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>716</td>\n",
       "      <td>talk.politics.misc</td>\n",
       "      <td>270</td>\n",
       "      <td>alt.atheism</td>\n",
       "      <td>150.0</td>\n",
       "      <td>37.71%</td>\n",
       "      <td>58.66%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5</td>\n",
       "      <td>511</td>\n",
       "      <td>rec.autos</td>\n",
       "      <td>420</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>6.0</td>\n",
       "      <td>82.19%</td>\n",
       "      <td>83.37%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>6</td>\n",
       "      <td>870</td>\n",
       "      <td>rec.motorcycles</td>\n",
       "      <td>100</td>\n",
       "      <td>alt.atheism</td>\n",
       "      <td>73.0</td>\n",
       "      <td>11.49%</td>\n",
       "      <td>19.89%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>7</td>\n",
       "      <td>570</td>\n",
       "      <td>comp.os.ms-windows.misc</td>\n",
       "      <td>338</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>46.0</td>\n",
       "      <td>59.30%</td>\n",
       "      <td>67.37%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>8</td>\n",
       "      <td>435</td>\n",
       "      <td>talk.politics.mideast</td>\n",
       "      <td>372</td>\n",
       "      <td>alt.atheism</td>\n",
       "      <td>23.0</td>\n",
       "      <td>85.52%</td>\n",
       "      <td>90.80%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>9</td>\n",
       "      <td>84</td>\n",
       "      <td>comp.os.ms-windows.misc</td>\n",
       "      <td>8</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>8.0</td>\n",
       "      <td>9.52%</td>\n",
       "      <td>19.05%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>10</td>\n",
       "      <td>527</td>\n",
       "      <td>talk.politics.guns</td>\n",
       "      <td>382</td>\n",
       "      <td>talk.religion.misc</td>\n",
       "      <td>35.0</td>\n",
       "      <td>72.49%</td>\n",
       "      <td>79.13%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>11</td>\n",
       "      <td>554</td>\n",
       "      <td>comp.graphics</td>\n",
       "      <td>324</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>17.0</td>\n",
       "      <td>58.48%</td>\n",
       "      <td>61.55%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>12</td>\n",
       "      <td>372</td>\n",
       "      <td>rec.motorcycles</td>\n",
       "      <td>355</td>\n",
       "      <td>alt.atheism</td>\n",
       "      <td>1.0</td>\n",
       "      <td>95.43%</td>\n",
       "      <td>95.70%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>13</td>\n",
       "      <td>840</td>\n",
       "      <td>soc.religion.christian</td>\n",
       "      <td>460</td>\n",
       "      <td>alt.atheism</td>\n",
       "      <td>188.0</td>\n",
       "      <td>54.76%</td>\n",
       "      <td>77.14%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>14</td>\n",
       "      <td>468</td>\n",
       "      <td>rec.sport.baseball</td>\n",
       "      <td>451</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>96.37%</td>\n",
       "      <td>96.37%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>15</td>\n",
       "      <td>538</td>\n",
       "      <td>misc.forsale</td>\n",
       "      <td>438</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>27.0</td>\n",
       "      <td>81.41%</td>\n",
       "      <td>86.43%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>16</td>\n",
       "      <td>363</td>\n",
       "      <td>sci.crypt</td>\n",
       "      <td>349</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>1.0</td>\n",
       "      <td>96.14%</td>\n",
       "      <td>96.42%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>17</td>\n",
       "      <td>479</td>\n",
       "      <td>sci.electronics</td>\n",
       "      <td>336</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>28.0</td>\n",
       "      <td>70.15%</td>\n",
       "      <td>75.99%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>18</td>\n",
       "      <td>425</td>\n",
       "      <td>sci.med</td>\n",
       "      <td>402</td>\n",
       "      <td>alt.atheism</td>\n",
       "      <td>1.0</td>\n",
       "      <td>94.59%</td>\n",
       "      <td>94.82%</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>19</td>\n",
       "      <td>532</td>\n",
       "      <td>sci.electronics</td>\n",
       "      <td>62</td>\n",
       "      <td>comp.sys.mac.hardware</td>\n",
       "      <td>23.0</td>\n",
       "      <td>11.65%</td>\n",
       "      <td>15.98%</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    cluster  count                     rank1  rank1_count   \n",
       "0         0    432            comp.windows.x          406  \\\n",
       "1         1    418                 sci.space          388   \n",
       "2         2   1035  comp.sys.ibm.pc.hardware          396   \n",
       "3         3    471          rec.sport.hockey          455   \n",
       "4         4    716        talk.politics.misc          270   \n",
       "5         5    511                 rec.autos          420   \n",
       "6         6    870           rec.motorcycles          100   \n",
       "7         7    570   comp.os.ms-windows.misc          338   \n",
       "8         8    435     talk.politics.mideast          372   \n",
       "9         9     84   comp.os.ms-windows.misc            8   \n",
       "10       10    527        talk.politics.guns          382   \n",
       "11       11    554             comp.graphics          324   \n",
       "12       12    372           rec.motorcycles          355   \n",
       "13       13    840    soc.religion.christian          460   \n",
       "14       14    468        rec.sport.baseball          451   \n",
       "15       15    538              misc.forsale          438   \n",
       "16       16    363                 sci.crypt          349   \n",
       "17       17    479           sci.electronics          336   \n",
       "18       18    425                   sci.med          402   \n",
       "19       19    532           sci.electronics           62   \n",
       "\n",
       "                    rank2  rank2_count   per_1 per_1_2  \n",
       "0   comp.sys.mac.hardware          1.0  93.98%  94.21%  \n",
       "1             alt.atheism          2.0  92.82%  93.30%  \n",
       "2   comp.sys.mac.hardware        387.0  38.26%  75.65%  \n",
       "3                       0          0.0  96.60%  96.60%  \n",
       "4             alt.atheism        150.0  37.71%  58.66%  \n",
       "5   comp.sys.mac.hardware          6.0  82.19%  83.37%  \n",
       "6             alt.atheism         73.0  11.49%  19.89%  \n",
       "7   comp.sys.mac.hardware         46.0  59.30%  67.37%  \n",
       "8             alt.atheism         23.0  85.52%  90.80%  \n",
       "9   comp.sys.mac.hardware          8.0   9.52%  19.05%  \n",
       "10     talk.religion.misc         35.0  72.49%  79.13%  \n",
       "11  comp.sys.mac.hardware         17.0  58.48%  61.55%  \n",
       "12            alt.atheism          1.0  95.43%  95.70%  \n",
       "13            alt.atheism        188.0  54.76%  77.14%  \n",
       "14                      0          0.0  96.37%  96.37%  \n",
       "15  comp.sys.mac.hardware         27.0  81.41%  86.43%  \n",
       "16  comp.sys.mac.hardware          1.0  96.14%  96.42%  \n",
       "17  comp.sys.mac.hardware         28.0  70.15%  75.99%  \n",
       "18            alt.atheism          1.0  94.59%  94.82%  \n",
       "19  comp.sys.mac.hardware         23.0  11.65%  15.98%  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "# 统计每个cluster的数量\n",
    "new_df = embedding_df.groupby('cluster')['cluster'].count().reset_index(name='count')\n",
    "\n",
    "# 统计这个cluster里最多的分类的数量\n",
    "title_count = embedding_df.groupby(['cluster', 'title']).size().reset_index(name='title_count')\n",
    "first_titles = title_count.groupby('cluster').apply(lambda x: x.nlargest(1, columns=['title_count']))\n",
    "first_titles = first_titles.reset_index(drop=True)\n",
    "new_df = pd.merge(new_df, first_titles[['cluster', 'title', 'title_count']], on='cluster', how='left')\n",
    "new_df = new_df.rename(columns={'title': 'rank1', 'title_count': 'rank1_count'})\n",
    "\n",
    "# 统计这个cluster里第二多的分类的数量\n",
    "second_titles = title_count[~title_count['title'].isin(first_titles['title'])]\n",
    "second_titles = second_titles.groupby('cluster').apply(lambda x: x.nlargest(1, columns=['title_count']))\n",
    "second_titles = second_titles.reset_index(drop=True)\n",
    "new_df = pd.merge(new_df, second_titles[['cluster', 'title', 'title_count']], on='cluster', how='left')\n",
    "new_df = new_df.rename(columns={'title': 'rank2', 'title_count': 'rank2_count'})\n",
    "new_df.fillna(0, inplace=True)\n",
    "new_df['per_1'] = (new_df['rank1_count'] / new_df['count']).map(lambda x: '{:.2%}'.format(x))\n",
    "new_df['per_1_2'] = ((new_df['rank1_count'] + new_df['rank2_count'])/ new_df['count']).map(lambda x: '{:.2%}'.format(x))\n",
    "\n",
    "# 将缺失值替换为 0\n",
    "# 输出结果\n",
    "display(new_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cluster 0, Rank 1: comp.windows.x, Theme: Xlib编程\n",
      "Cluster 1, Rank 1: sci.space, Theme: 太空技术与航空\n",
      "Cluster 2, Rank 1: comp.sys.ibm.pc.hardware, Theme: 电脑硬件\n",
      "Cluster 3, Rank 1: rec.sport.hockey, Theme: 欧洲冰球vs北美冰球\n",
      "Cluster 4, Rank 1: talk.politics.misc, Theme: 社会观点与自由\n",
      "Cluster 5, Rank 1: rec.autos, Theme: 汽车硬件\n",
      "Cluster 6, Rank 1: rec.motorcycles, Theme: 数学与文化冲击\n",
      "Cluster 7, Rank 1: comp.os.ms-windows.misc, Theme: PC软件与硬件\n",
      "Cluster 8, Rank 1: talk.politics.mideast, Theme: “穆斯林大屠杀”\n",
      "Cluster 9, Rank 1: comp.os.ms-windows.misc, Theme: 科技产品\"\"\"\n",
      "Cluster 10, Rank 1: talk.politics.guns, Theme: 枪支管制与安全\n",
      "Cluster 11, Rank 1: comp.graphics, Theme: 计算机编程与硬件\n",
      "Cluster 12, Rank 1: rec.motorcycles, Theme: 安全骑行：驾驶课程与技巧\n",
      "Cluster 13, Rank 1: soc.religion.christian, Theme: 宗教信仰与实践\n",
      "Cluster 14, Rank 1: rec.sport.baseball, Theme: 棒球联盟\n",
      "Cluster 15, Rank 1: misc.forsale, Theme: 购物优惠和出售\n",
      "Cluster 16, Rank 1: sci.crypt, Theme: 关于加密政策的讨论\n",
      "Cluster 17, Rank 1: sci.electronics, Theme: 电子设备连接与维护\n",
      "Cluster 18, Rank 1: sci.med, Theme: 药物和疾病\n",
      "Cluster 19, Rank 1: sci.electronics, Theme: 电子邮件使用者研究\n"
     ]
    }
   ],
   "source": [
    "items_per_cluster = 10\n",
    "COMPLETIONS_MODEL = \"text-davinci-003\"\n",
    "\n",
    "for i in range(num_of_clusters):\n",
    "    cluster_name = new_df[new_df.cluster == i].iloc[0].rank1\n",
    "    print(f\"Cluster {i}, Rank 1: {cluster_name}, Theme:\", end=\" \")\n",
    "\n",
    "    content = \"\\n\".join(\n",
    "        embedding_df[embedding_df.cluster == i].text.sample(items_per_cluster, random_state=42).values\n",
    "    )\n",
    "    response = openai.Completion.create(\n",
    "        model=COMPLETIONS_MODEL,\n",
    "        prompt=f'''我们想要给下面的内容，分组成有意义的类别，以便我们可以对其进行总结。请根据下面这些内容的共同点，总结一个50个字以内的新闻组的名称么？比如 “PC硬件”\\n\\n内容:\\n\"\"\"\\n{content}\\n\"\"\"新闻组名称：''',\n",
    "        temperature=0,\n",
    "        max_tokens=100,\n",
    "        top_p=1,\n",
    "    )\n",
    "    print(response[\"choices\"][0][\"text\"].replace(\"\\n\", \"\")) # type: ignore\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cluster 0, Rank 1: comp.windows.x, 抽样翻译: 没有实际执行它？不知怎么回事，我的一个xterminal用户使得只要点击鼠标右键，就会自动杀死所有客户端-哦，我的：-(谢谢，Fish\n",
      "Cluster 1, Rank 1: sci.space, 抽样翻译: 韦恩·马森和他的团伙在阿拉巴马州发生了什么？我还听说有一个未经证实的谣言，即航空大使们已经消失了。有其他人可以证实吗？\n",
      "Cluster 2, Rank 1: comp.sys.ibm.pc.hardware, 抽样翻译: 我怀疑这不是一个特定于Quadra的问题。去年我不得不放弃我“古老”的Bernoulli 20（每个磁带的价格大约是90美元，使整个事情的价值超过我的整个电脑;）。Ocean Microsystems的技术支持人员建议可以使用一些第三方驱动程序来解决这个问题 - 在我的情况下，磁带无法格式化/挂载/分区用于A / UX。\n",
      "Cluster 3, Rank 1: rec.sport.hockey, 抽样翻译: 我相信那是4-1。罗德·布林道·阿莫尔在第三节19.59时分攻入一球。\n",
      "Cluster 4, Rank 1: talk.politics.misc, 抽样翻译: 为了确保每个人都清楚：“它从未有过”是指“保护”，而不是“未能保护”；即，在我的一生中，我从未见过美国政府始终保护美国公民的利益，除非是意外。\n",
      "Cluster 5, Rank 1: rec.autos, 抽样翻译: 噢，来吧，傻瓜，你要做的就是在你的引擎罩上割一个洞，然后把一个管子放进去，这样你就可以把机油倒进去了。你觉得那些热门车上的大空气进气装置是干什么的？它们只是为了外观，没有人知道，它们提供了进入机油填充孔的途径。\n",
      "Cluster 6, Rank 1: rec.motorcycles, 抽样翻译: 你真是个失败者\n",
      "Cluster 7, Rank 1: comp.os.ms-windows.misc, 抽样翻译: 偶尔你也得为表现良好的东西说句好话。我的东西泽3401没有任何问题。它在DOS和OS/2上运行非常好。对于OS/2，你不需要加载任何特殊的驱动程序。安装会检测到它是一个东西泽驱动器，然后就完成了。顺便说一句，它也非常快！\n",
      "Cluster 8, Rank 1: talk.politics.mideast, 抽样翻译: Avi，    供你参考，伊斯兰教允许宗教自由——在宗教上没有强制。犹太教是否也允许宗教自由（即是否认可非犹太人）？只是好奇而已。\n",
      "Cluster 9, Rank 1: comp.os.ms-windows.misc, 抽样翻译: 每个人都有自己的梦想，但只有勇敢追求梦想的人才能实现它。\n",
      "Cluster 10, Rank 1: talk.politics.guns, 抽样翻译: 不一定，特别是如果强奸犯被认定为此类罪行。例如，如果你有意将手指伸入装满老鼠夹的地方，然后被夹住，这是谁的错？\n",
      "Cluster 11, Rank 1: comp.graphics, 抽样翻译: 帮帮我！！我需要代码/包/任何东西来处理3D数据，并将其转换为带有隐藏线的线框表面。我正在使用DOS机器，代码可以是ANSI C或C ++，ANSI Fortran或Basic。我使用的数据形成一个矩形网格。请将您的回复发布到网络上，以便其他人受益。我的个人观点是，这是一个普遍的兴趣问题。谢谢！！！！！\n",
      "Cluster 12, Rank 1: rec.motorcycles, 抽样翻译: 这是一段心理学，对于任何长期骑行者来说都是必不可少的。人们不会去想“如果我这么做会有别人受到影响吗？”他们只会评估“如果我这么做会受到影响吗？”\n",
      "Cluster 13, Rank 1: soc.religion.christian, 抽样翻译: 这是一个非常薄弱的论点，因为没有独立的支持文本（关于关键事件）。至于新约最古老的现存文本的日期......如果现在只有一个关于美国内战的现存文本，你会怎么想？现在考虑一个大部分文盲的人口，每一份手稿都是手工复制的......--Hal\n",
      "Cluster 14, Rank 1: rec.sport.baseball, 抽样翻译: 这个赔率意味着你下注5美元赌反败者赢8美元，或者下注9美元赌胜者赢5美元。\n",
      "Cluster 15, Rank 1: misc.forsale, 抽样翻译: 嗯，标题就是这样......我正在寻找便宜的二手TG-16游戏，它们支持2个或更多玩家（同时）....请给我发送所有带有价格的报价。\n",
      "Cluster 16, Rank 1: sci.crypt, 抽样翻译: 哪里？老实说，我没有看到任何……我不同意，至少有其他标准已经存在。此外，即使他们限制NREN上的加密，谁在乎呢？大部分互联网都是商业的。NREN只适用于政府和大学研究（阅读提案-它是一条“数据高速公路”，与互联网无关）。\n",
      "Cluster 17, Rank 1: sci.electronics, 抽样翻译: 动态RAM不是基于翻转锁存器；基本上每个位只有一个晶体管和电容来存储！静态RAM是基于翻转锁存器，更加昂贵，密度也更低。如果忽略电子和热膨胀，两者都没有任何“移动”的部件...Chris\n",
      "Cluster 18, Rank 1: sci.med, 抽样翻译: 化学物质已经消失，感谢大家的回应。\n",
      "Cluster 19, Rank 1: sci.electronics, 抽样翻译: 尝试lyman.pppl.gov--/pub/8051\n"
     ]
    }
   ],
   "source": [
    "items_per_cluster = 1\n",
    "COMPLETIONS_MODEL = \"text-davinci-003\"\n",
    "\n",
    "for i in range(num_of_clusters):\n",
    "    cluster_name = new_df[new_df.cluster == i].iloc[0].rank1\n",
    "    print(f\"Cluster {i}, Rank 1: {cluster_name}, 抽样翻译:\", end=\" \")\n",
    "\n",
    "    content = \"\\n\".join(\n",
    "        embedding_df[(embedding_df.cluster == i) & (embedding_df.n_tokens < 100)].text.sample(items_per_cluster, random_state=42).values\n",
    "    )\n",
    "    response = openai.Completion.create(\n",
    "        model=COMPLETIONS_MODEL,\n",
    "        prompt=f'''请把下面的内容翻译成中文\\n\\n内容:\\n\"\"\"\\n{content}\\n\"\"\"翻译：''',\n",
    "        temperature=0,\n",
    "        max_tokens=500,\n",
    "        top_p=1,\n",
    "    )\n",
    "    print(response[\"choices\"][0][\"text\"].replace(\"\\n\", \"\")) # type: ignore"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 减少聚类的数量，来进行可视化"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from sklearn.cluster import KMeans\n",
    "\n",
    "embedding_df = pd.read_parquet(\"data/20_newsgroup_with_embedding.parquet\")\n",
    "\n",
    "embedding_df = embedding_df[\n",
    "    (embedding_df.title == \"sci.electronics\") |\n",
    "    (embedding_df.title == \"sci.med\") |\n",
    "    (embedding_df.title == \"rec.motorcycles\") |\n",
    "    (embedding_df.title == \"rec.sport.baseball\")\n",
    "]\n",
    "matrix = np.vstack(embedding_df.embedding.values) # type: ignore\n",
    "num_of_clusters = 4\n",
    "\n",
    "kmeans = KMeans(n_clusters=num_of_clusters, init=\"k-means++\", n_init=10, random_state=42)\n",
    "kmeans.fit(matrix)\n",
    "labels = kmeans.labels_\n",
    "embedding_df[\"cluster\"] = labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'Clusters identified visualized in language 2d using t-SNE')"
      ]
     },
     "execution_count": 81,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.manifold import TSNE\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "tsne = TSNE(n_components=2, perplexity=15, random_state=42, init=\"random\", learning_rate=200)\n",
    "vis_dims2 = tsne.fit_transform(matrix)\n",
    "\n",
    "x = [x for x, y in vis_dims2]\n",
    "y = [y for x, y in vis_dims2]\n",
    "\n",
    "for category, color in zip([\"sci.electronics\", \"sci.med\", \"rec.motorcycles\", \"rec.sport.baseball\"], [\"purple\", \"green\", \"red\", \"blue\"]):\n",
    "    xs = np.array(x)[embedding_df.title == category]\n",
    "    ys = np.array(y)[embedding_df.title == category]\n",
    "    plt.scatter(xs, ys, color=color, alpha=0.3)\n",
    "\n",
    "    avg_x = xs.mean()\n",
    "    avg_y = ys.mean()\n",
    "\n",
    "    plt.scatter(avg_x, avg_y, marker=\"x\", color=color, s=100) # type: ignore\n",
    "plt.title(\"Clusters identified visualized in language 2d using t-SNE\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 多轮聊天对话"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 使用原始的 text-davinci-003"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import openai\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "COMPLETION_MODEL = \"text-davinci-003\"\n",
    "\n",
    "def get_response(prompt, temperature = 1.0, stop=None):\n",
    "    completions = openai.Completion.create (\n",
    "        engine=COMPLETION_MODEL,\n",
    "        prompt=prompt,\n",
    "        max_tokens=1024,\n",
    "        n=1,\n",
    "        stop=stop,\n",
    "        temperature=temperature,        \n",
    "    )\n",
    "    message = completions.choices[0].text # type: ignore\n",
    "    return message"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "你好！这个端午节我准备去我家乡的一个小村庄，和家人一起庆祝端午节。\n"
     ]
    }
   ],
   "source": [
    "question = \"\"\"Q: 你好，端午节你准备去哪里？\n",
    "A:\"\"\"\n",
    "\n",
    "answer = get_response(question, temperature=0.5, stop=\"Q:\")\n",
    "print(answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 武汉有很多好玩的旅游景点，比如江汉滩公园、武昌植物园、汉口火车站、汉口商业街、黄鹤楼、汉口艺术街等等。\n"
     ]
    }
   ],
   "source": [
    "question = \"\"\"Q: 你好，端午节你准备去哪里？\n",
    "A: 端午节我打算去武汉。\n",
    "Q: 那里有什么好玩的旅游景点？\n",
    "A: \"\"\"\n",
    "\n",
    "answer = get_response(question, temperature=0.5, stop=\"Q:\")\n",
    "print(answer)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 使用ChatCompletion接口"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "#import openai\n",
    "#import os\n",
    "\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "\n",
    "COMPLETION_MODEL = \"gpt-35-turbo\"\n",
    "\n",
    "class Conversation:\n",
    "    def __init__(self, prompt, num_of_round):\n",
    "        self.prompt = prompt\n",
    "        self.num_of_round = num_of_round\n",
    "        self.messages = []\n",
    "        self.messages.append({\"role\": \"system\", \"content\": self.prompt})\n",
    "\n",
    "    def ask(self, question):\n",
    "        try:\n",
    "            self.messages.append( {\"role\": \"user\", \"content\": question})\n",
    "            response = openai.ChatCompletion.create(\n",
    "                #model=COMPLETION_MODEL, #openAI\n",
    "                engine=COMPLETION_MODEL,\n",
    "                messages=self.messages,\n",
    "                temperature=0.5,\n",
    "                max_tokens=2048,\n",
    "                top_p=1,\n",
    "            )\n",
    "        except Exception as e:\n",
    "            print(e)\n",
    "            return e\n",
    "\n",
    "        message = response[\"choices\"][0][\"message\"][\"content\"] # type: ignore\n",
    "        self.messages.append({\"role\": \"assistant\", \"content\": message})\n",
    "        \n",
    "        if len(self.messages) > self.num_of_round*2 + 1:\n",
    "            del self.messages[1:3]\n",
    "        return message\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "User : 你好，端午节你准备去哪里？\n",
      "Assistant : 嘿，我打算去武汉旅游。听说那里有很多历史文化遗址和美食，我很期待能够去探索一下。你有去过武汉吗？\n",
      "\n",
      "User : 那里有什么好玩的旅游景点\n",
      "Assistant : 武汉有很多著名的旅游景点，比如黄鹤楼、东湖、武汉长江大桥、武汉博物馆等等。此外，武汉还有很多美食，比如热干面、豆皮、鸭脖等等，非常值得一试。如果你对历史文化感兴趣，武汉还有一些博物馆和古迹可以去参观，比如汉口江滩、黄陂古城等等。总之，武汉是一个非常有意思的城市，我相信你一定会喜欢上它的。\n",
      "\n"
     ]
    }
   ],
   "source": [
    "prompt = \"\"\"请你扮演我的朋友Tom，如果有人问你端午节去哪里玩，你会回答打算去武汉旅游\"\"\"\n",
    "conv1 = Conversation(prompt, 3)\n",
    "question1 = \"你好，端午节你准备去哪里？\"\n",
    "print(\"User : %s\" % question1)\n",
    "print(\"Assistant : %s\\n\" % conv1.ask(question1))\n",
    "\n",
    "question2 = \"那里有什么好玩的旅游景点\"\n",
    "print(\"User : %s\" % question2)\n",
    "print(\"Assistant : %s\\n\" % conv1.ask(question2))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 成本与效率问题"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 数据的Batch处理"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'pd' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[17], line 15\u001b[0m\n\u001b[1;32m     11\u001b[0m max_tokens \u001b[39m=\u001b[39m \u001b[39m8000\u001b[39m  \u001b[39m# the maximum for text-embedding-ada-002 is 8191\u001b[39;00m\n\u001b[1;32m     14\u001b[0m \u001b[39m# import data/toutiao_cat_data.txt as a pandas dataframe\u001b[39;00m\n\u001b[0;32m---> 15\u001b[0m df \u001b[39m=\u001b[39m pd\u001b[39m.\u001b[39mread_parquet(\u001b[39m\"\u001b[39m\u001b[39mdata/toutiao_cat_data_all_with_embeddings.parquet\u001b[39m\u001b[39m\"\u001b[39m)\n\u001b[1;32m     16\u001b[0m \u001b[39mprint\u001b[39m(\u001b[39m\"\u001b[39m\u001b[39mLines of text before filtering: \u001b[39m\u001b[39m\"\u001b[39m, \u001b[39mlen\u001b[39m(df))\n",
      "\u001b[0;31mNameError\u001b[0m: name 'pd' is not defined"
     ]
    }
   ],
   "source": [
    "import tiktoken\n",
    "import openai\n",
    "import os\n",
    "import backoff\n",
    "\n",
    "openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "\n",
    "# embedding model parameters\n",
    "embedding_model = \"text-embedding-ada-002\"\n",
    "embedding_encoding = \"cl100k_base\"  # this the encoding for text-embedding-ada-002\n",
    "max_tokens = 8000  # the maximum for text-embedding-ada-002 is 8191\n",
    "\n",
    "\n",
    "# import data/toutiao_cat_data.txt as a pandas dataframe\n",
    "df = pd.read_parquet(\"data/toutiao_cat_data_all_with_embeddings.parquet\")\n",
    "print(\"Lines of text before filtering: \", len(df))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>code</th>\n",
       "      <th>category</th>\n",
       "      <th>title</th>\n",
       "      <th>keywords</th>\n",
       "      <th>combined</th>\n",
       "      <th>n_tokens</th>\n",
       "      <th>embedding</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>6551700932705387022</td>\n",
       "      <td>101</td>\n",
       "      <td>news_culture</td>\n",
       "      <td>京城最值得你来场文化之旅的博物馆</td>\n",
       "      <td>保利集团,马未都,中国科学技术馆,博物馆,新中国</td>\n",
       "      <td>标题: 京城最值得你来场文化之旅的博物馆; 关键字: 保利集团,马未都,中国科学技术馆,博物...</td>\n",
       "      <td>56</td>\n",
       "      <td>[-0.02160128578543663, -0.004615371115505695, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6552368441838272771</td>\n",
       "      <td>101</td>\n",
       "      <td>news_culture</td>\n",
       "      <td>发酵床的垫料种类有哪些？哪种更好？</td>\n",
       "      <td></td>\n",
       "      <td>标题: 发酵床的垫料种类有哪些？哪种更好？; 关键字:</td>\n",
       "      <td>32</td>\n",
       "      <td>[0.014646276831626892, -0.0075576393865048885,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>6552407965343678723</td>\n",
       "      <td>101</td>\n",
       "      <td>news_culture</td>\n",
       "      <td>上联：黄山黄河黄皮肤黄土高原。怎么对下联？</td>\n",
       "      <td></td>\n",
       "      <td>标题: 上联：黄山黄河黄皮肤黄土高原。怎么对下联？; 关键字:</td>\n",
       "      <td>37</td>\n",
       "      <td>[0.012643996626138687, 0.0035733033437281847, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>6552332417753940238</td>\n",
       "      <td>101</td>\n",
       "      <td>news_culture</td>\n",
       "      <td>林徽因什么理由拒绝了徐志摩而选择梁思成为终身伴侣？</td>\n",
       "      <td></td>\n",
       "      <td>标题: 林徽因什么理由拒绝了徐志摩而选择梁思成为终身伴侣？; 关键字:</td>\n",
       "      <td>44</td>\n",
       "      <td>[-0.0004984583356417716, -0.016934111714363098...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>6552475601595269390</td>\n",
       "      <td>101</td>\n",
       "      <td>news_culture</td>\n",
       "      <td>黄杨木是什么树？</td>\n",
       "      <td></td>\n",
       "      <td>标题: 黄杨木是什么树？; 关键字:</td>\n",
       "      <td>20</td>\n",
       "      <td>[-0.0008294825092889369, -0.013883085921406746...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                    id  code      category                      title   \n",
       "0  6551700932705387022   101  news_culture           京城最值得你来场文化之旅的博物馆  \\\n",
       "1  6552368441838272771   101  news_culture          发酵床的垫料种类有哪些？哪种更好？   \n",
       "2  6552407965343678723   101  news_culture      上联：黄山黄河黄皮肤黄土高原。怎么对下联？   \n",
       "3  6552332417753940238   101  news_culture  林徽因什么理由拒绝了徐志摩而选择梁思成为终身伴侣？   \n",
       "4  6552475601595269390   101  news_culture                   黄杨木是什么树？   \n",
       "\n",
       "                   keywords   \n",
       "0  保利集团,马未都,中国科学技术馆,博物馆,新中国  \\\n",
       "1                             \n",
       "2                             \n",
       "3                             \n",
       "4                             \n",
       "\n",
       "                                            combined  n_tokens   \n",
       "0  标题: 京城最值得你来场文化之旅的博物馆; 关键字: 保利集团,马未都,中国科学技术馆,博物...        56  \\\n",
       "1                       标题: 发酵床的垫料种类有哪些？哪种更好？; 关键字:         32   \n",
       "2                   标题: 上联：黄山黄河黄皮肤黄土高原。怎么对下联？; 关键字:         37   \n",
       "3               标题: 林徽因什么理由拒绝了徐志摩而选择梁思成为终身伴侣？; 关键字:         44   \n",
       "4                                标题: 黄杨木是什么树？; 关键字:         20   \n",
       "\n",
       "                                           embedding  \n",
       "0  [-0.02160128578543663, -0.004615371115505695, ...  \n",
       "1  [0.014646276831626892, -0.0075576393865048885,...  \n",
       "2  [0.012643996626138687, 0.0035733033437281847, ...  \n",
       "3  [-0.0004984583356417716, -0.016934111714363098...  \n",
       "4  [-0.0008294825092889369, -0.013883085921406746...  "
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Lines of text after filtering:  382688\n"
     ]
    }
   ],
   "source": [
    "encoding = tiktoken.get_encoding(embedding_encoding)\n",
    "# omit reviews that are too long to embed\n",
    "df[\"n_tokens\"] = df.combined.apply(lambda x: len(encoding.encode(x)))\n",
    "df = df[df.n_tokens <= max_tokens]\n",
    "print(\"Lines of text after filtering: \", len(df))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 假设我们没有预先处理完Embedding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "ename": "ValueError",
     "evalue": "Cannot take a larger sample than population when 'replace=False'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[4], line 3\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mopenai\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39membeddings_utils\u001b[39;00m \u001b[39mimport\u001b[39;00m get_embedding\n\u001b[0;32m----> 3\u001b[0m df_1k \u001b[39m=\u001b[39m df\u001b[39m.\u001b[39;49msample(\u001b[39m1000\u001b[39;49m, random_state\u001b[39m=\u001b[39;49m\u001b[39m42\u001b[39;49m)\n\u001b[1;32m      5\u001b[0m df_1k[\u001b[39m\"\u001b[39m\u001b[39membedding\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39m=\u001b[39m df_1k\u001b[39m.\u001b[39mcombined\u001b[39m.\u001b[39mapply(\u001b[39mlambda\u001b[39;00m x : get_embedding(x, engine\u001b[39m=\u001b[39membedding_model))\n\u001b[1;32m      6\u001b[0m df_1k\u001b[39m.\u001b[39mto_csv(\u001b[39m\"\u001b[39m\u001b[39mdata/toutiao_cat_data_10k_with_embeddings.csv\u001b[39m\u001b[39m\"\u001b[39m, index\u001b[39m=\u001b[39m\u001b[39mFalse\u001b[39;00m)\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/pandas/core/generic.py:5858\u001b[0m, in \u001b[0;36mNDFrame.sample\u001b[0;34m(self, n, frac, replace, weights, random_state, axis, ignore_index)\u001b[0m\n\u001b[1;32m   5855\u001b[0m \u001b[39mif\u001b[39;00m weights \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39mNone\u001b[39;00m:\n\u001b[1;32m   5856\u001b[0m     weights \u001b[39m=\u001b[39m sample\u001b[39m.\u001b[39mpreprocess_weights(\u001b[39mself\u001b[39m, weights, axis)\n\u001b[0;32m-> 5858\u001b[0m sampled_indices \u001b[39m=\u001b[39m sample\u001b[39m.\u001b[39;49msample(obj_len, size, replace, weights, rs)\n\u001b[1;32m   5859\u001b[0m result \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mtake(sampled_indices, axis\u001b[39m=\u001b[39maxis)\n\u001b[1;32m   5861\u001b[0m \u001b[39mif\u001b[39;00m ignore_index:\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/pandas/core/sample.py:151\u001b[0m, in \u001b[0;36msample\u001b[0;34m(obj_len, size, replace, weights, random_state)\u001b[0m\n\u001b[1;32m    148\u001b[0m     \u001b[39melse\u001b[39;00m:\n\u001b[1;32m    149\u001b[0m         \u001b[39mraise\u001b[39;00m \u001b[39mValueError\u001b[39;00m(\u001b[39m\"\u001b[39m\u001b[39mInvalid weights: weights sum to zero\u001b[39m\u001b[39m\"\u001b[39m)\n\u001b[0;32m--> 151\u001b[0m \u001b[39mreturn\u001b[39;00m random_state\u001b[39m.\u001b[39;49mchoice(obj_len, size\u001b[39m=\u001b[39;49msize, replace\u001b[39m=\u001b[39;49mreplace, p\u001b[39m=\u001b[39;49mweights)\u001b[39m.\u001b[39mastype(\n\u001b[1;32m    152\u001b[0m     np\u001b[39m.\u001b[39mintp, copy\u001b[39m=\u001b[39m\u001b[39mFalse\u001b[39;00m\n\u001b[1;32m    153\u001b[0m )\n",
      "File \u001b[0;32mmtrand.pyx:984\u001b[0m, in \u001b[0;36mnumpy.random.mtrand.RandomState.choice\u001b[0;34m()\u001b[0m\n",
      "\u001b[0;31mValueError\u001b[0m: Cannot take a larger sample than population when 'replace=False'"
     ]
    }
   ],
   "source": [
    "from openai.embeddings_utils import get_embedding\n",
    "\n",
    "df_1k = df.sample(1000, random_state=42)\n",
    "\n",
    "df_1k[\"embedding\"] = df_1k.combined.apply(lambda x : get_embedding(x, engine=embedding_model))\n",
    "df_1k.to_csv(\"data/toutiao_cat_data_10k_with_embeddings.csv\", index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "@backoff.on_exception(backoff.expo, openai.error.RateLimitError) # type: ignore\n",
    "def get_embedding_with_backoff(**kwargs):\n",
    "    return get_embedding(**kwargs)\n",
    "\n",
    "df_10k = df.sample(10000, random_state=42)\n",
    "\n",
    "df_10k[\"embedding\"] = df_10k.combined.apply(lambda x : get_embedding_with_backoff(text=x, engine=embedding_model))\n",
    "df_10k.to_csv(\"data/toutiao_cat_data_10k_with_embeddings.csv\", index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [],
   "source": [
    "from openai.embeddings_utils import get_embeddings\n",
    "\n",
    "batch_size = 100\n",
    "\n",
    "@backoff.on_exception(backoff.expo, openai.error.RateLimitError)  # type: ignore\n",
    "def get_embeddings_with_backoff(prompts, engine):\n",
    "    embeddings = []\n",
    "    for i in range(0, len(prompts), batch_size):\n",
    "        batch = prompts[i:i+batch_size]\n",
    "        embeddings += get_embeddings(list_of_text=batch, engine=engine)\n",
    "    return embeddings\n",
    "\n",
    "# group prompts into batches of 100\n",
    "prompts = df_1k.combined.tolist()\n",
    "prompt_batches = [prompts[i:i+batch_size] for i in range(0, len(prompts), batch_size)]\n",
    "\n",
    "embeddings = []\n",
    "for batch in prompt_batches:\n",
    "    batch_embeddings = get_embeddings_with_backoff(prompts=batch, engine=embedding_model)\n",
    "    embeddings += batch_embeddings\n",
    "\n",
    "df_1k[\"embedding\"] = embeddings\n",
    "df_1k.to_csv(\"data/toutiao_cat_data_all_with_embeddings_1k.csv\", index=False)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 通过Tiktoken来统计Token数量"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "8\n"
     ]
    }
   ],
   "source": [
    "import tiktoken\n",
    "\n",
    "\n",
    "embedding_encoding = \"cl100k_base\"  # this the encoding for text-embedding-ada-002\n",
    "\n",
    "encoding = tiktoken.get_encoding(embedding_encoding)\n",
    "text = \"今天天气真不错\"\n",
    "print(len(encoding.encode(text)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12\n"
     ]
    }
   ],
   "source": [
    "\n",
    "embedding_encoding = \"p50k_base\"  # this the encoding for text-davinci-003\n",
    "\n",
    "encoding = tiktoken.get_encoding(embedding_encoding)\n",
    "text = \"今天天气真不错\"\n",
    "print(len(encoding.encode(text)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"choices\": [\n",
      "    {\n",
      "      \"finish_reason\": \"stop\",\n",
      "      \"index\": 0,\n",
      "      \"logprobs\": null,\n",
      "      \"text\": \"\\n\\nYes, the weather is really nice today.\"\n",
      "    }\n",
      "  ],\n",
      "  \"created\": 1689060457,\n",
      "  \"id\": \"cmpl-7b24XCT0xvOGLoVOg8hovULNdSa49\",\n",
      "  \"model\": \"text-davinci-003\",\n",
      "  \"object\": \"text_completion\",\n",
      "  \"usage\": {\n",
      "    \"completion_tokens\": 11,\n",
      "    \"prompt_tokens\": 12,\n",
      "    \"total_tokens\": 23\n",
      "  }\n",
      "}\n",
      "\n",
      "\n",
      "Yes, the weather is really nice today.\n"
     ]
    }
   ],
   "source": [
    "import openai\n",
    "import os\n",
    "\n",
    "openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "COMPLETION_MODEL = \"text-davinci-003\"\n",
    "\n",
    "def get_response(prompt):\n",
    "    completions = openai.Completion.create(\n",
    "        engine=COMPLETION_MODEL,\n",
    "        prompt=prompt,\n",
    "        max_tokens=512,\n",
    "        n=1,\n",
    "        stop=None,\n",
    "        temperature=0.0,        \n",
    "    )\n",
    "    return completions\n",
    "\n",
    "response = get_response(\"今天天气真不错\")\n",
    "\n",
    "print(response)\n",
    "print(response.choices[0].text) # type: ignore"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 语义搜索"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "1.【新款】Apple/苹果 iPhone 11 Pro Max 全网通手机\n",
      "2.【超值】华为P30 Pro 全网通8GB+128GB 极光色\n",
      "3.【热销】荣耀20 Pro 全面屏4800万AI四摄\n",
      "4.【特惠】三星Galaxy S10 Plus 全网通手机\n",
      "5.【大促】小米CC9 Pro 全网通6GB+128GB 冰川蓝\n",
      "6.【新品】OPPO Reno2 Z 全网通8GB+128GB 梦幻蓝\n",
      "7.【精选】vivo X30 Pro 5G 全网通8GB+128GB 雪梨金\n",
      "8.【抢购】HUAWEI Mate 30 Pro 5G 全网通\n",
      "9.【限时】realme X Pro 全网通8GB+128GB 极光色\n",
      "10.【限量】荣耀Magic2 全网通8GB+128GB 珠光白\n",
      "11.【狂欢】Apple/苹果 iPad Pro 11英寸 平板电脑\n",
      "12.【热卖】华为MateBook 13 全网通轻薄笔记本\n",
      "13.【抢先】Lenovo/联想小新Air 13.3英寸轻薄笔记本\n",
      "14.【新款】Apple/苹果 MacBook Pro 16英寸超轻薄笔记本\n",
      "15.【超值】惠普Envy 13-aq0013TU 13.3英寸轻薄笔记本\n",
      "16.【热销】联想Yoga S940 14英寸超轻薄笔记本\n",
      "17.【特惠】华硕ZenBook 13 UX333FA 13.3英寸轻薄笔记本\n",
      "18.【大促】华为MateBook X Pro 13.9英寸超轻薄笔记本\n",
      "19.【新品】华硕ZenBook Flip S UX370UA 13.3英寸轻薄笔记本\n",
      "20.【精选】华硕ZenBook 14 UX433FA 14英寸轻薄笔记本\n",
      "21.【抢购】Acer/宏碁 Swift 3 SF314-54 14英寸轻薄笔记本\n",
      "22.【限时】ASUS/华硕 ROG Zephyrus GX501 15.6英寸游戏本\n",
      "23.【限量】MSI GF63 8RC-068CN 15.6英寸游戏本\n",
      "24.【狂欢】联想Legion Y7000 15.6英寸游戏本\n",
      "25.【热卖】Razer Blade 15 15.6英寸游戏本\n",
      "26.【抢先】Apple/苹果 MacBook Air 13.3英寸超轻薄笔记本\n",
      "27.【新款】Apple/苹果MacBook Pro 13.3英寸超轻薄笔记本\n",
      "28.【超值】Dell/戴尔 XPS 13 9380 13.3英寸超轻薄笔记本\n",
      "29.【热销】华为MateBook D 15.6英寸轻薄笔记本\n",
      "30.【特惠】联想ThinkPad X1 Carbon 14英寸轻薄笔记本\n",
      "31.【大促】联想ThinkPad X1 Yoga 14英寸轻薄笔记本\n",
      "32.【新品】华为MateBook E 2019 12英寸轻薄笔记本\n",
      "33.【精选】联想ThinkPad E490 14英寸轻薄笔记本\n",
      "34.【抢购】华硕ZenBook Flip 14 UX461UN 14英寸轻薄笔记本\n",
      "35.【限时】华硕ZenBook Flip S UX370UA 13.3英寸轻薄笔记本\n",
      "36.【限量】联想ThinkPad T480 14英寸轻薄笔记本\n",
      "37.【狂欢】惠普Spectre x360 13-ap0013dx 13.3英寸轻薄笔记本\n",
      "38.【热卖】惠普Spectre x360 15-df0013dx 15.6英寸轻薄笔记本\n",
      "39.【抢先】华硕ROG Strix GL503GE 15.6英寸游戏本\n",
      "40.【新款】联想ThinkPad X1 Extreme 15.6英寸轻薄笔记本\n",
      "41.【超值】联想Yoga C930 14英寸轻薄笔记本\n",
      "42.【热销】华为MateBook 14 2020 14英寸轻薄笔记本\n",
      "43.【特惠】联想IdeaPad S540 14英寸轻薄笔记本\n",
      "44.【大促】惠普Spectre Folio 13-ak0003dx 13.3英寸轻薄笔记本\n",
      "45.【新品】联想ThinkPad X1 Yoga 4th 14英寸轻薄笔记本\n",
      "46.【精选】联想ThinkPad T490 14英寸轻薄笔记本\n",
      "47.【抢购】联想Yoga C740 14英寸轻薄笔记本\n",
      "48.【限时】华为MateBook D 14 2020 14英寸轻薄笔记本\n",
      "49.【限量】惠普Spectre x360 15-df0012dx 15.6英寸轻薄笔记本\n",
      "50.【狂欢】华硕ZenBook Pro 15 UX580GD 15.6英寸轻薄笔记本\n"
     ]
    }
   ],
   "source": [
    "import openai, os\n",
    "\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "\n",
    "COMPLETION_MODEL = \"text-davinci-003\"\n",
    "\n",
    "def generate_data_by_prompt(prompt):\n",
    "    response = openai.Completion.create(\n",
    "        engine=COMPLETION_MODEL,\n",
    "        prompt=prompt,\n",
    "        temperature=0.5,\n",
    "        max_tokens=2048,\n",
    "        top_p=1,\n",
    "    )\n",
    "    return response.choices[0].text # type: ignore\n",
    "\n",
    "prompt = \"\"\"请你生成50条淘宝网里的商品的标题，每条在30个字左右，品类是3C数码产品，标题里往往也会有一些促销类的信息，每行一条。\"\"\"\n",
    "data = generate_data_by_prompt(prompt)\n",
    "print(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>product_name</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.【新款】Apple/苹果 iPhone 11 Pro Max 全网通手机</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2.【超值】华为P30 Pro 全网通8GB+128GB 极光色</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3.【热销】荣耀20 Pro 全面屏4800万AI四摄</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4.【特惠】三星Galaxy S10 Plus 全网通手机</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5.【大促】小米CC9 Pro 全网通6GB+128GB 冰川蓝</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                             product_name\n",
       "0  1.【新款】Apple/苹果 iPhone 11 Pro Max 全网通手机\n",
       "1        2.【超值】华为P30 Pro 全网通8GB+128GB 极光色\n",
       "2             3.【热销】荣耀20 Pro 全面屏4800万AI四摄\n",
       "3           4.【特惠】三星Galaxy S10 Plus 全网通手机\n",
       "4        5.【大促】小米CC9 Pro 全网通6GB+128GB 冰川蓝"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "\n",
    "product_names = data.strip().split('\\n')\n",
    "df = pd.DataFrame({'product_name': product_names})\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>product_name</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>【新款】Apple/苹果 iPhone 11 Pro Max 全网通手机</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>【超值】华为P30 Pro 全网通8GB+128GB 极光色</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>【热销】荣耀20 Pro 全面屏4800万AI四摄</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>【特惠】三星Galaxy S10 Plus 全网通手机</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>【大促】小米CC9 Pro 全网通6GB+128GB 冰川蓝</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                           product_name\n",
       "0  【新款】Apple/苹果 iPhone 11 Pro Max 全网通手机\n",
       "1        【超值】华为P30 Pro 全网通8GB+128GB 极光色\n",
       "2             【热销】荣耀20 Pro 全面屏4800万AI四摄\n",
       "3           【特惠】三星Galaxy S10 Plus 全网通手机\n",
       "4        【大促】小米CC9 Pro 全网通6GB+128GB 冰川蓝"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#去掉No.\n",
    "df.product_name = df.product_name.apply(lambda x: x.split('.')[0].strip())\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "clothes_prompt = \"\"\"请你生成50条淘宝网里的商品的标题，每条在30个字左右，品类是女性的服饰箱包等等，标题里往往也会有一些促销类的信息，每行一条。\"\"\"\n",
    "clothes_data = generate_data_by_prompt(clothes_prompt)\n",
    "clothes_product_names = clothes_data.strip().split('\\n')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>product_name</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>时尚潮流女士手提包小熊印花设计</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>新款复古拼接小方包女士包包</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>日韩潮牌百搭小包精灵小熊图案</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>小清新简约百搭双肩包百搭款</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>时尚拼色小方包女士包包</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      product_name\n",
       "0  时尚潮流女士手提包小熊印花设计\n",
       "1    新款复古拼接小方包女士包包\n",
       "2   日韩潮牌百搭小包精灵小熊图案\n",
       "3    小清新简约百搭双肩包百搭款\n",
       "4      时尚拼色小方包女士包包"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "clothes_df = pd.DataFrame({'product_name': clothes_product_names})\n",
    "clothes_df.product_name = clothes_df.product_name.apply(lambda x: x.split('.')[1].strip())\n",
    "clothes_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0     【新款】Apple/苹果 iPhone 11 Pro Max 全网通手机\n",
      "1           【超值】华为P30 Pro 全网通8GB+128GB 极光色\n",
      "2                【热销】荣耀20 Pro 全面屏4800万AI四摄\n",
      "3              【特惠】三星Galaxy S10 Plus 全网通手机\n",
      "4           【大促】小米CC9 Pro 全网通6GB+128GB 冰川蓝\n",
      "                      ...                 \n",
      "95                           时尚拼接小方包百搭款女士包\n",
      "96                           时尚拼色小熊图案双肩斜挎包\n",
      "97                           时尚拼接小方包百搭款女士包\n",
      "98                             时尚拼色小方包百搭款式\n",
      "99                           时尚拼接小方包百搭款女士包\n",
      "Name: product_name, Length: 100, dtype: object\n"
     ]
    }
   ],
   "source": [
    "df = pd.concat([df, clothes_df], axis=0)\n",
    "df = df.reset_index(drop=True)\n",
    "print(df.product_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'df' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[1], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m display(df)\n",
      "\u001b[0;31mNameError\u001b[0m: name 'df' is not defined"
     ]
    }
   ],
   "source": [
    "display(df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'df' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[2], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m df\u001b[39m.\u001b[39mto_csv(\u001b[39m\"\u001b[39m\u001b[39mdata/product_names.csv\u001b[39m\u001b[39m\"\u001b[39m)\n",
      "\u001b[0;31mNameError\u001b[0m: name 'df' is not defined"
     ]
    }
   ],
   "source": [
    "df.to_csv(\"data/product_names.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "ename": "RetryError",
     "evalue": "RetryError[<Future at 0x7fa916d13130 state=finished raised InvalidRequestError>]",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mInvalidRequestError\u001b[0m                       Traceback (most recent call last)",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/tenacity/__init__.py:382\u001b[0m, in \u001b[0;36mRetrying.__call__\u001b[0;34m(self, fn, *args, **kwargs)\u001b[0m\n\u001b[1;32m    381\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[0;32m--> 382\u001b[0m     result \u001b[39m=\u001b[39m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m    383\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mBaseException\u001b[39;00m:  \u001b[39m# noqa: B902\u001b[39;00m\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/embeddings_utils.py:48\u001b[0m, in \u001b[0;36mget_embeddings\u001b[0;34m(list_of_text, engine, **kwargs)\u001b[0m\n\u001b[1;32m     46\u001b[0m list_of_text \u001b[39m=\u001b[39m [text\u001b[39m.\u001b[39mreplace(\u001b[39m\"\u001b[39m\u001b[39m\\n\u001b[39;00m\u001b[39m\"\u001b[39m, \u001b[39m\"\u001b[39m\u001b[39m \u001b[39m\u001b[39m\"\u001b[39m) \u001b[39mfor\u001b[39;00m text \u001b[39min\u001b[39;00m list_of_text]\n\u001b[0;32m---> 48\u001b[0m data \u001b[39m=\u001b[39m openai\u001b[39m.\u001b[39;49mEmbedding\u001b[39m.\u001b[39;49mcreate(\u001b[39minput\u001b[39;49m\u001b[39m=\u001b[39;49mlist_of_text, engine\u001b[39m=\u001b[39;49mengine, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\u001b[39m.\u001b[39mdata\n\u001b[1;32m     49\u001b[0m \u001b[39mreturn\u001b[39;00m [d[\u001b[39m\"\u001b[39m\u001b[39membedding\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39mfor\u001b[39;00m d \u001b[39min\u001b[39;00m data]\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_resources/embedding.py:33\u001b[0m, in \u001b[0;36mEmbedding.create\u001b[0;34m(cls, *args, **kwargs)\u001b[0m\n\u001b[1;32m     32\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[0;32m---> 33\u001b[0m     response \u001b[39m=\u001b[39m \u001b[39msuper\u001b[39;49m()\u001b[39m.\u001b[39;49mcreate(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m     35\u001b[0m     \u001b[39m# If a user specifies base64, we'll just return the encoded string.\u001b[39;00m\n\u001b[1;32m     36\u001b[0m     \u001b[39m# This is only for the default case.\u001b[39;00m\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_resources/abstract/engine_api_resource.py:153\u001b[0m, in \u001b[0;36mEngineAPIResource.create\u001b[0;34m(cls, api_key, api_base, api_type, request_id, api_version, organization, **params)\u001b[0m\n\u001b[1;32m    138\u001b[0m (\n\u001b[1;32m    139\u001b[0m     deployment_id,\n\u001b[1;32m    140\u001b[0m     engine,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    150\u001b[0m     api_key, api_base, api_type, api_version, organization, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mparams\n\u001b[1;32m    151\u001b[0m )\n\u001b[0;32m--> 153\u001b[0m response, _, api_key \u001b[39m=\u001b[39m requestor\u001b[39m.\u001b[39;49mrequest(\n\u001b[1;32m    154\u001b[0m     \u001b[39m\"\u001b[39;49m\u001b[39mpost\u001b[39;49m\u001b[39m\"\u001b[39;49m,\n\u001b[1;32m    155\u001b[0m     url,\n\u001b[1;32m    156\u001b[0m     params\u001b[39m=\u001b[39;49mparams,\n\u001b[1;32m    157\u001b[0m     headers\u001b[39m=\u001b[39;49mheaders,\n\u001b[1;32m    158\u001b[0m     stream\u001b[39m=\u001b[39;49mstream,\n\u001b[1;32m    159\u001b[0m     request_id\u001b[39m=\u001b[39;49mrequest_id,\n\u001b[1;32m    160\u001b[0m     request_timeout\u001b[39m=\u001b[39;49mrequest_timeout,\n\u001b[1;32m    161\u001b[0m )\n\u001b[1;32m    163\u001b[0m \u001b[39mif\u001b[39;00m stream:\n\u001b[1;32m    164\u001b[0m     \u001b[39m# must be an iterator\u001b[39;00m\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_requestor.py:298\u001b[0m, in \u001b[0;36mAPIRequestor.request\u001b[0;34m(self, method, url, params, headers, files, stream, request_id, request_timeout)\u001b[0m\n\u001b[1;32m    288\u001b[0m result \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mrequest_raw(\n\u001b[1;32m    289\u001b[0m     method\u001b[39m.\u001b[39mlower(),\n\u001b[1;32m    290\u001b[0m     url,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    296\u001b[0m     request_timeout\u001b[39m=\u001b[39mrequest_timeout,\n\u001b[1;32m    297\u001b[0m )\n\u001b[0;32m--> 298\u001b[0m resp, got_stream \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_interpret_response(result, stream)\n\u001b[1;32m    299\u001b[0m \u001b[39mreturn\u001b[39;00m resp, got_stream, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mapi_key\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_requestor.py:700\u001b[0m, in \u001b[0;36mAPIRequestor._interpret_response\u001b[0;34m(self, result, stream)\u001b[0m\n\u001b[1;32m    698\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[1;32m    699\u001b[0m     \u001b[39mreturn\u001b[39;00m (\n\u001b[0;32m--> 700\u001b[0m         \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_interpret_response_line(\n\u001b[1;32m    701\u001b[0m             result\u001b[39m.\u001b[39;49mcontent\u001b[39m.\u001b[39;49mdecode(\u001b[39m\"\u001b[39;49m\u001b[39mutf-8\u001b[39;49m\u001b[39m\"\u001b[39;49m),\n\u001b[1;32m    702\u001b[0m             result\u001b[39m.\u001b[39;49mstatus_code,\n\u001b[1;32m    703\u001b[0m             result\u001b[39m.\u001b[39;49mheaders,\n\u001b[1;32m    704\u001b[0m             stream\u001b[39m=\u001b[39;49m\u001b[39mFalse\u001b[39;49;00m,\n\u001b[1;32m    705\u001b[0m         ),\n\u001b[1;32m    706\u001b[0m         \u001b[39mFalse\u001b[39;00m,\n\u001b[1;32m    707\u001b[0m     )\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_requestor.py:763\u001b[0m, in \u001b[0;36mAPIRequestor._interpret_response_line\u001b[0;34m(self, rbody, rcode, rheaders, stream)\u001b[0m\n\u001b[1;32m    762\u001b[0m \u001b[39mif\u001b[39;00m stream_error \u001b[39mor\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39m200\u001b[39m \u001b[39m<\u001b[39m\u001b[39m=\u001b[39m rcode \u001b[39m<\u001b[39m \u001b[39m300\u001b[39m:\n\u001b[0;32m--> 763\u001b[0m     \u001b[39mraise\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mhandle_error_response(\n\u001b[1;32m    764\u001b[0m         rbody, rcode, resp\u001b[39m.\u001b[39mdata, rheaders, stream_error\u001b[39m=\u001b[39mstream_error\n\u001b[1;32m    765\u001b[0m     )\n\u001b[1;32m    766\u001b[0m \u001b[39mreturn\u001b[39;00m resp\n",
      "\u001b[0;31mInvalidRequestError\u001b[0m: Too many inputs. The max number of inputs is 1.  We hope to increase the number of inputs per request soon. Please contact us through an Azure support request at: https://go.microsoft.com/fwlink/?linkid=2213926 for further questions.",
      "\nThe above exception was the direct cause of the following exception:\n",
      "\u001b[0;31mRetryError\u001b[0m                                Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[18], line 19\u001b[0m\n\u001b[1;32m     16\u001b[0m EMBEDDING_MODEL \u001b[39m=\u001b[39m \u001b[39m\"\u001b[39m\u001b[39mtext-embedding-ada-002\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m     18\u001b[0m positive_review \u001b[39m=\u001b[39m get_embedding(\u001b[39m\"\u001b[39m\u001b[39m这是一条好评\u001b[39m\u001b[39m\"\u001b[39m, engine\u001b[39m=\u001b[39mEMBEDDING_MODEL)\n\u001b[0;32m---> 19\u001b[0m negative_review \u001b[39m=\u001b[39m get_embeddings([\u001b[39m\"\u001b[39;49m\u001b[39mhello\u001b[39;49m\u001b[39m\"\u001b[39;49m,\u001b[39m\"\u001b[39;49m\u001b[39mhello\u001b[39;49m\u001b[39m\"\u001b[39;49m], engine\u001b[39m=\u001b[39;49mEMBEDDING_MODEL)\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/tenacity/__init__.py:289\u001b[0m, in \u001b[0;36mBaseRetrying.wraps.<locals>.wrapped_f\u001b[0;34m(*args, **kw)\u001b[0m\n\u001b[1;32m    287\u001b[0m \u001b[39m@functools\u001b[39m\u001b[39m.\u001b[39mwraps(f)\n\u001b[1;32m    288\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mwrapped_f\u001b[39m(\u001b[39m*\u001b[39margs: t\u001b[39m.\u001b[39mAny, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mkw: t\u001b[39m.\u001b[39mAny) \u001b[39m-\u001b[39m\u001b[39m>\u001b[39m t\u001b[39m.\u001b[39mAny:\n\u001b[0;32m--> 289\u001b[0m     \u001b[39mreturn\u001b[39;00m \u001b[39mself\u001b[39;49m(f, \u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkw)\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/tenacity/__init__.py:379\u001b[0m, in \u001b[0;36mRetrying.__call__\u001b[0;34m(self, fn, *args, **kwargs)\u001b[0m\n\u001b[1;32m    377\u001b[0m retry_state \u001b[39m=\u001b[39m RetryCallState(retry_object\u001b[39m=\u001b[39m\u001b[39mself\u001b[39m, fn\u001b[39m=\u001b[39mfn, args\u001b[39m=\u001b[39margs, kwargs\u001b[39m=\u001b[39mkwargs)\n\u001b[1;32m    378\u001b[0m \u001b[39mwhile\u001b[39;00m \u001b[39mTrue\u001b[39;00m:\n\u001b[0;32m--> 379\u001b[0m     do \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49miter(retry_state\u001b[39m=\u001b[39;49mretry_state)\n\u001b[1;32m    380\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39misinstance\u001b[39m(do, DoAttempt):\n\u001b[1;32m    381\u001b[0m         \u001b[39mtry\u001b[39;00m:\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/tenacity/__init__.py:326\u001b[0m, in \u001b[0;36mBaseRetrying.iter\u001b[0;34m(self, retry_state)\u001b[0m\n\u001b[1;32m    324\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mreraise:\n\u001b[1;32m    325\u001b[0m         \u001b[39mraise\u001b[39;00m retry_exc\u001b[39m.\u001b[39mreraise()\n\u001b[0;32m--> 326\u001b[0m     \u001b[39mraise\u001b[39;00m retry_exc \u001b[39mfrom\u001b[39;00m \u001b[39mfut\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39mexception\u001b[39;00m()\n\u001b[1;32m    328\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mwait:\n\u001b[1;32m    329\u001b[0m     sleep \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mwait(retry_state)\n",
      "\u001b[0;31mRetryError\u001b[0m: RetryError[<Future at 0x7fa916d13130 state=finished raised InvalidRequestError>]"
     ]
    }
   ],
   "source": [
    "import openai\n",
    "import os\n",
    "from openai.embeddings_utils import cosine_similarity, get_embedding, get_embeddings\n",
    "\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "EMBEDDING_MODEL = \"text-embedding-ada-002\"\n",
    "\n",
    "positive_review = get_embedding(\"这是一条好评\", engine=EMBEDDING_MODEL)\n",
    "negative_review = get_embeddings([\"hello\",\"hello\"], engine=EMBEDDING_MODEL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "ename": "RetryError",
     "evalue": "RetryError[<Future at 0x7fa916d3b520 state=finished raised InvalidRequestError>]",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mInvalidRequestError\u001b[0m                       Traceback (most recent call last)",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/tenacity/__init__.py:382\u001b[0m, in \u001b[0;36mRetrying.__call__\u001b[0;34m(self, fn, *args, **kwargs)\u001b[0m\n\u001b[1;32m    381\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[0;32m--> 382\u001b[0m     result \u001b[39m=\u001b[39m fn(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m    383\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mBaseException\u001b[39;00m:  \u001b[39m# noqa: B902\u001b[39;00m\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/embeddings_utils.py:48\u001b[0m, in \u001b[0;36mget_embeddings\u001b[0;34m(list_of_text, engine, **kwargs)\u001b[0m\n\u001b[1;32m     46\u001b[0m list_of_text \u001b[39m=\u001b[39m [text\u001b[39m.\u001b[39mreplace(\u001b[39m\"\u001b[39m\u001b[39m\\n\u001b[39;00m\u001b[39m\"\u001b[39m, \u001b[39m\"\u001b[39m\u001b[39m \u001b[39m\u001b[39m\"\u001b[39m) \u001b[39mfor\u001b[39;00m text \u001b[39min\u001b[39;00m list_of_text]\n\u001b[0;32m---> 48\u001b[0m data \u001b[39m=\u001b[39m openai\u001b[39m.\u001b[39;49mEmbedding\u001b[39m.\u001b[39;49mcreate(\u001b[39minput\u001b[39;49m\u001b[39m=\u001b[39;49mlist_of_text, engine\u001b[39m=\u001b[39;49mengine, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\u001b[39m.\u001b[39mdata\n\u001b[1;32m     49\u001b[0m \u001b[39mreturn\u001b[39;00m [d[\u001b[39m\"\u001b[39m\u001b[39membedding\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39mfor\u001b[39;00m d \u001b[39min\u001b[39;00m data]\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_resources/embedding.py:33\u001b[0m, in \u001b[0;36mEmbedding.create\u001b[0;34m(cls, *args, **kwargs)\u001b[0m\n\u001b[1;32m     32\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[0;32m---> 33\u001b[0m     response \u001b[39m=\u001b[39m \u001b[39msuper\u001b[39;49m()\u001b[39m.\u001b[39;49mcreate(\u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m     35\u001b[0m     \u001b[39m# If a user specifies base64, we'll just return the encoded string.\u001b[39;00m\n\u001b[1;32m     36\u001b[0m     \u001b[39m# This is only for the default case.\u001b[39;00m\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_resources/abstract/engine_api_resource.py:153\u001b[0m, in \u001b[0;36mEngineAPIResource.create\u001b[0;34m(cls, api_key, api_base, api_type, request_id, api_version, organization, **params)\u001b[0m\n\u001b[1;32m    138\u001b[0m (\n\u001b[1;32m    139\u001b[0m     deployment_id,\n\u001b[1;32m    140\u001b[0m     engine,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    150\u001b[0m     api_key, api_base, api_type, api_version, organization, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mparams\n\u001b[1;32m    151\u001b[0m )\n\u001b[0;32m--> 153\u001b[0m response, _, api_key \u001b[39m=\u001b[39m requestor\u001b[39m.\u001b[39;49mrequest(\n\u001b[1;32m    154\u001b[0m     \u001b[39m\"\u001b[39;49m\u001b[39mpost\u001b[39;49m\u001b[39m\"\u001b[39;49m,\n\u001b[1;32m    155\u001b[0m     url,\n\u001b[1;32m    156\u001b[0m     params\u001b[39m=\u001b[39;49mparams,\n\u001b[1;32m    157\u001b[0m     headers\u001b[39m=\u001b[39;49mheaders,\n\u001b[1;32m    158\u001b[0m     stream\u001b[39m=\u001b[39;49mstream,\n\u001b[1;32m    159\u001b[0m     request_id\u001b[39m=\u001b[39;49mrequest_id,\n\u001b[1;32m    160\u001b[0m     request_timeout\u001b[39m=\u001b[39;49mrequest_timeout,\n\u001b[1;32m    161\u001b[0m )\n\u001b[1;32m    163\u001b[0m \u001b[39mif\u001b[39;00m stream:\n\u001b[1;32m    164\u001b[0m     \u001b[39m# must be an iterator\u001b[39;00m\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_requestor.py:298\u001b[0m, in \u001b[0;36mAPIRequestor.request\u001b[0;34m(self, method, url, params, headers, files, stream, request_id, request_timeout)\u001b[0m\n\u001b[1;32m    288\u001b[0m result \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mrequest_raw(\n\u001b[1;32m    289\u001b[0m     method\u001b[39m.\u001b[39mlower(),\n\u001b[1;32m    290\u001b[0m     url,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    296\u001b[0m     request_timeout\u001b[39m=\u001b[39mrequest_timeout,\n\u001b[1;32m    297\u001b[0m )\n\u001b[0;32m--> 298\u001b[0m resp, got_stream \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_interpret_response(result, stream)\n\u001b[1;32m    299\u001b[0m \u001b[39mreturn\u001b[39;00m resp, got_stream, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mapi_key\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_requestor.py:700\u001b[0m, in \u001b[0;36mAPIRequestor._interpret_response\u001b[0;34m(self, result, stream)\u001b[0m\n\u001b[1;32m    698\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[1;32m    699\u001b[0m     \u001b[39mreturn\u001b[39;00m (\n\u001b[0;32m--> 700\u001b[0m         \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_interpret_response_line(\n\u001b[1;32m    701\u001b[0m             result\u001b[39m.\u001b[39;49mcontent\u001b[39m.\u001b[39;49mdecode(\u001b[39m\"\u001b[39;49m\u001b[39mutf-8\u001b[39;49m\u001b[39m\"\u001b[39;49m),\n\u001b[1;32m    702\u001b[0m             result\u001b[39m.\u001b[39;49mstatus_code,\n\u001b[1;32m    703\u001b[0m             result\u001b[39m.\u001b[39;49mheaders,\n\u001b[1;32m    704\u001b[0m             stream\u001b[39m=\u001b[39;49m\u001b[39mFalse\u001b[39;49;00m,\n\u001b[1;32m    705\u001b[0m         ),\n\u001b[1;32m    706\u001b[0m         \u001b[39mFalse\u001b[39;00m,\n\u001b[1;32m    707\u001b[0m     )\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/openai/api_requestor.py:763\u001b[0m, in \u001b[0;36mAPIRequestor._interpret_response_line\u001b[0;34m(self, rbody, rcode, rheaders, stream)\u001b[0m\n\u001b[1;32m    762\u001b[0m \u001b[39mif\u001b[39;00m stream_error \u001b[39mor\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39m200\u001b[39m \u001b[39m<\u001b[39m\u001b[39m=\u001b[39m rcode \u001b[39m<\u001b[39m \u001b[39m300\u001b[39m:\n\u001b[0;32m--> 763\u001b[0m     \u001b[39mraise\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mhandle_error_response(\n\u001b[1;32m    764\u001b[0m         rbody, rcode, resp\u001b[39m.\u001b[39mdata, rheaders, stream_error\u001b[39m=\u001b[39mstream_error\n\u001b[1;32m    765\u001b[0m     )\n\u001b[1;32m    766\u001b[0m \u001b[39mreturn\u001b[39;00m resp\n",
      "\u001b[0;31mInvalidRequestError\u001b[0m: Too many inputs. The max number of inputs is 1.  We hope to increase the number of inputs per request soon. Please contact us through an Azure support request at: https://go.microsoft.com/fwlink/?linkid=2213926 for further questions.",
      "\nThe above exception was the direct cause of the following exception:\n",
      "\u001b[0;31mRetryError\u001b[0m                                Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[12], line 33\u001b[0m\n\u001b[1;32m     20\u001b[0m df\u001b[39m=\u001b[39mpd\u001b[39m.\u001b[39mread_csv(\u001b[39m\"\u001b[39m\u001b[39mdata/product_names.csv\u001b[39m\u001b[39m\"\u001b[39m)\n\u001b[1;32m     22\u001b[0m \u001b[39m# @backoff.on_exception(backoff.expo, openai.error.RateLimitError) # type: ignore\u001b[39;00m\n\u001b[1;32m     23\u001b[0m \u001b[39m# def get_embeddings_with_backoff(prompts, engine):\u001b[39;00m\n\u001b[1;32m     24\u001b[0m \u001b[39m#     embeddings = []\u001b[39;00m\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m     30\u001b[0m \u001b[39m# prompts = df.product_name.tolist()\u001b[39;00m\n\u001b[1;32m     31\u001b[0m \u001b[39m# prompt_batches = [prompts[i:i+batch_size] for i in range(0, len(prompts), batch_size)]\u001b[39;00m\n\u001b[0;32m---> 33\u001b[0m embeddings \u001b[39m=\u001b[39m get_embeddings([\u001b[39m'\u001b[39;49m\u001b[39m全网通手机\u001b[39;49m\u001b[39m'\u001b[39;49m,\u001b[39m'\u001b[39;49m\u001b[39m全网通手机\u001b[39;49m\u001b[39m'\u001b[39;49m], engine\u001b[39m=\u001b[39;49mEMBEDDING_MODEL)\n\u001b[1;32m     34\u001b[0m \u001b[39m# for batch in prompt_batches:\u001b[39;00m\n\u001b[1;32m     35\u001b[0m \u001b[39m#     batch_embeddings = get_embeddings_with_backoff(prompts=batch, engine=EMBEDDING_MODEL)\u001b[39;00m\n\u001b[1;32m     36\u001b[0m \u001b[39m#     embeddings += batch_embeddings\u001b[39;00m\n\u001b[1;32m     37\u001b[0m \n\u001b[1;32m     38\u001b[0m \u001b[39m# df[\"embedding\"] = embeddings\u001b[39;00m\n\u001b[1;32m     39\u001b[0m \u001b[39m# df.to_parquet(\"data/taobao_product_title.parquet\", index=False)\u001b[39;00m\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/tenacity/__init__.py:289\u001b[0m, in \u001b[0;36mBaseRetrying.wraps.<locals>.wrapped_f\u001b[0;34m(*args, **kw)\u001b[0m\n\u001b[1;32m    287\u001b[0m \u001b[39m@functools\u001b[39m\u001b[39m.\u001b[39mwraps(f)\n\u001b[1;32m    288\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mwrapped_f\u001b[39m(\u001b[39m*\u001b[39margs: t\u001b[39m.\u001b[39mAny, \u001b[39m*\u001b[39m\u001b[39m*\u001b[39mkw: t\u001b[39m.\u001b[39mAny) \u001b[39m-\u001b[39m\u001b[39m>\u001b[39m t\u001b[39m.\u001b[39mAny:\n\u001b[0;32m--> 289\u001b[0m     \u001b[39mreturn\u001b[39;00m \u001b[39mself\u001b[39;49m(f, \u001b[39m*\u001b[39;49margs, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkw)\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/tenacity/__init__.py:379\u001b[0m, in \u001b[0;36mRetrying.__call__\u001b[0;34m(self, fn, *args, **kwargs)\u001b[0m\n\u001b[1;32m    377\u001b[0m retry_state \u001b[39m=\u001b[39m RetryCallState(retry_object\u001b[39m=\u001b[39m\u001b[39mself\u001b[39m, fn\u001b[39m=\u001b[39mfn, args\u001b[39m=\u001b[39margs, kwargs\u001b[39m=\u001b[39mkwargs)\n\u001b[1;32m    378\u001b[0m \u001b[39mwhile\u001b[39;00m \u001b[39mTrue\u001b[39;00m:\n\u001b[0;32m--> 379\u001b[0m     do \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49miter(retry_state\u001b[39m=\u001b[39;49mretry_state)\n\u001b[1;32m    380\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39misinstance\u001b[39m(do, DoAttempt):\n\u001b[1;32m    381\u001b[0m         \u001b[39mtry\u001b[39;00m:\n",
      "File \u001b[0;32m/home/alan/mactalk-ai-course/.venv/lib/python3.10/site-packages/tenacity/__init__.py:326\u001b[0m, in \u001b[0;36mBaseRetrying.iter\u001b[0;34m(self, retry_state)\u001b[0m\n\u001b[1;32m    324\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mreraise:\n\u001b[1;32m    325\u001b[0m         \u001b[39mraise\u001b[39;00m retry_exc\u001b[39m.\u001b[39mreraise()\n\u001b[0;32m--> 326\u001b[0m     \u001b[39mraise\u001b[39;00m retry_exc \u001b[39mfrom\u001b[39;00m \u001b[39mfut\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39mexception\u001b[39;00m()\n\u001b[1;32m    328\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mwait:\n\u001b[1;32m    329\u001b[0m     sleep \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mwait(retry_state)\n",
      "\u001b[0;31mRetryError\u001b[0m: RetryError[<Future at 0x7fa916d3b520 state=finished raised InvalidRequestError>]"
     ]
    }
   ],
   "source": [
    "from openai.embeddings_utils import get_embeddings\n",
    "import openai, os, backoff\n",
    "import pandas as pd\n",
    "\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "EMBEDDING_MODEL = \"text-embedding-ada-002\"\n",
    "\n",
    "batch_size = 20\n",
    "\n",
    "df=pd.read_csv(\"data/product_names.csv\")\n",
    "\n",
    "# @backoff.on_exception(backoff.expo, openai.error.RateLimitError) # type: ignore\n",
    "# def get_embeddings_with_backoff(prompts, engine):\n",
    "#     embeddings = []\n",
    "#     for i in range(0, len(prompts), batch_size):\n",
    "#         batch = prompts[i:i+batch_size]\n",
    "#         embeddings += get_embeddings(list_of_text=batch, engine=engine)\n",
    "#     return embeddings\n",
    "\n",
    "# prompts = df.product_name.tolist()\n",
    "# prompt_batches = [prompts[i:i+batch_size] for i in range(0, len(prompts), batch_size)]\n",
    "\n",
    "embeddings = get_embeddings(['全网通手机','全网通手机'], engine=EMBEDDING_MODEL)\n",
    "# for batch in prompt_batches:\n",
    "#     batch_embeddings = get_embeddings_with_backoff(prompts=batch, engine=EMBEDDING_MODEL)\n",
    "#     embeddings += batch_embeddings\n",
    "\n",
    "# df[\"embedding\"] = embeddings\n",
    "# df.to_parquet(\"data/taobao_product_title.parquet\", index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "【新品】小米CC9 Pro 全网通 6GB+128GB\n",
      "【限时特惠】小米平板4 10\n",
      "【限时特价】华为 Mate 30 Pro 5G 8GB+256GB 全网通\n"
     ]
    }
   ],
   "source": [
    "from openai.embeddings_utils import get_embedding, cosine_similarity\n",
    "\n",
    "# search through the reviews for a specific product\n",
    "def search_product(df, query, n=3, pprint=True):\n",
    "    product_embedding = get_embedding(\n",
    "        query,\n",
    "        engine=embedding_model\n",
    "    )\n",
    "    df[\"similarity\"] = df.embedding.apply(lambda x: cosine_similarity(x, product_embedding))\n",
    "\n",
    "    results = (\n",
    "        df.sort_values(\"similarity\", ascending=False)\n",
    "        .head(n)\n",
    "        .product_name\n",
    "    )\n",
    "    if pprint:\n",
    "        for r in results:\n",
    "            print(r)\n",
    "    return results\n",
    "\n",
    "\n",
    "results = search_product(df, \"小米手机\", n=3)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 使用向量数据库加速搜索过程"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import faiss\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "\n",
    "def load_embeddings_to_faiss(df):\n",
    "    embeddings = np.array(df['embedding'].tolist()).astype('float32')\n",
    "    index = faiss.IndexFlatL2(embeddings.shape[1])\n",
    "    index.add(embeddings) # type: ignore\n",
    "    return index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<bound method NDFrame.head of                                                     text  target      title   \n",
       "0      I was wondering if anyone out there could enli...       7  rec.autos  \\\n",
       "1      \\nIt depends on your priorities.  A lot of peo...       7  rec.autos   \n",
       "2      an excellent automatic can be found in the sub...       7  rec.autos   \n",
       "3      : Ford and his automobile.  I need information...       7  rec.autos   \n",
       "4      \\nYo! Watch the attributions--I didn't say tha...       7  rec.autos   \n",
       "...                                                  ...     ...        ...   \n",
       "10635  Secrecy in Clipper Chip\\n\\nThe serial number o...      11  sci.crypt   \n",
       "10636  Hi !\\n\\nI am interested in the source of FEAL ...      11  sci.crypt   \n",
       "10637  The actual algorithm is classified, however, t...      11  sci.crypt   \n",
       "10638  \\n\\tThis appears to be generic calling upon th...      11  sci.crypt   \n",
       "10639  \\nProbably keep quiet and take it, lest they g...      11  sci.crypt   \n",
       "\n",
       "       n_tokens                                          embedding  \n",
       "0           121  [-0.0391300804913044, 0.013502633199095726, -0...  \n",
       "1           108  [-0.0011249205563217402, -0.00376517535187304,...  \n",
       "2           476  [-0.018259447067975998, -0.008410007692873478,...  \n",
       "3            86  [-0.012589422054588795, 0.006539034191519022, ...  \n",
       "4           130  [-0.0006192282889969647, -0.011226896196603775...  \n",
       "...         ...                                                ...  \n",
       "10635       272  [-0.009344913996756077, -0.002893796656280756,...  \n",
       "10636        46  [0.0020047826692461967, 0.005355264991521835, ...  \n",
       "10637       102  [-0.009686576202511787, 0.00589779345318675, -...  \n",
       "10638       469  [-0.023307695984840393, -0.009851706214249134,...  \n",
       "10639        17  [-0.00020658542052842677, -0.02622563205659389...  \n",
       "\n",
       "[10640 rows x 5 columns]>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#toutiao_df = pd.read_parquet(\"data/toutiao_cat_data_all_with_embeddings.parquet\")\n",
    "toutiao_df = pd.read_parquet(\"data/20_newsgroup_with_embedding.parquet\")\n",
    "toutiao_df.head"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "index = load_embeddings_to_faiss(toutiao_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def search_index(index, df, query_embedding, k=5):\n",
    "    query_vector = np.array(query_embedding).reshape(1, -1).astype('float32')\n",
    "    distances, indexes = index.search(query_vector, k)\n",
    "\n",
    "    results = []\n",
    "    for i in range(len(indexes)):\n",
    "        news = df.iloc[indexes[i]]['text'].values.tolist()\n",
    "        results.append((distances[i], news))    \n",
    "    return results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.012427806854248047\n",
      "[(array([0.3998429 , 0.40584296, 0.41748172], dtype=float32), ['\\nBut only in NY,NJ, Philadelphia, and Chicago. Everywhere else, the only\\nreason SportsChannel was available was for local baseball broadcasts.\\nAnd local baseball pre-empted the NHL playoffs. Thus far into this\\nplayoff season, ESPN/ABC has given me more hockey in 2 days (1 game)\\nthan SportsChannel did (0 games).\\n\\nIf people want hockey on TV, they should watch hockey on TV. I bet the\\nratings for hockey on Sunday on ABC went into the toilet. Next week, there\\nwill be far fewer ABC affiliates with hockey.\\n\\nSomeone in this thread said that he wouldn\\'t watch the games even if\\nthey were on TV, and this is a r.s.h. regular contributor!\\n\\nXenophobes north of border needn\\'t fear the US-Americanization of \"our\"\\ngame, because US-Americans will never figure out how to market hockey\\nhere. Support your team; support your game.', '.\\n.\\n.\\n\\nESPN had the Houston Astros @ Chicago Cubs game scheduled for last night on the\\nwest coast. \\n\\nSince the game was rained out, they showed the Toronto Maple Leafs at the\\nDetroit Red Wings game instead.', \"I agree. It was great for the ESPN people to show the Detroit game. (My\\nroommate just about sh*t when they threw the octopus on the ice. (Thanks\\nfor explaining the significance of that BTW)) The only problem I had was\\nwhen they blotted out the local commentators with the baseball ads and\\nmusic. Especially when the wings player hit the rut and went into the\\nboards injuring his shoulder and they blotted out the injury report. Other\\nthen that, hats off to ESPN. Now if they'll only make a habit of this. Yeah\\nright! Baseball seasons started (Zzzzzz.....) =)\\n\\nKOZ \"])]\n"
     ]
    }
   ],
   "source": [
    "import time\n",
    "from openai.embeddings_utils import get_embedding\n",
    "import openai, os, backoff\n",
    "import pandas as pd\n",
    "\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "embedding_model = \"text-embedding-ada-002\"\n",
    "\n",
    "query = \"NBA sports\"\n",
    "query_embedding = get_embedding(query, engine=embedding_model)\n",
    "\n",
    "start = time.time()\n",
    "news = search_index(index, toutiao_df, query_embedding, k=3)\n",
    "end = time.time()\n",
    "print(end - start)\n",
    "print(news)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 完全暴力检索的速度"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "from openai.embeddings_utils import get_embedding, cosine_similarity\n",
    "\n",
    "# search through the reviews for a specific product\n",
    "def search_index_brutal_force(df, query_embedding, n=3):\n",
    "    df[\"similarity\"] = df.embedding.apply(lambda x: cosine_similarity(x, query_embedding))\n",
    "\n",
    "    results = (\n",
    "        df.sort_values(\"similarity\", ascending=False)\n",
    "        .head(n)\n",
    "        .title\n",
    "    )\n",
    "    return results\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "58.231921911239624\n",
      "195956             超26亿元 北京七家车企将获新能源国补\n",
      "322070    北上资金维持积极布局节奏 本周斥逾65亿元加码18只个股\n",
      "321835      北上资金维持积极布局节奏 本周斥65亿加码18只个股\n",
      "Name: title, dtype: object\n"
     ]
    }
   ],
   "source": [
    "import time\n",
    "\n",
    "start = time.time()\n",
    "news = search_index_brutal_force(toutiao_df, query_embedding, n=3)\n",
    "end = time.time()\n",
    "print(end - start)\n",
    "print(news)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 通过知识库解决幻觉问题"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "初期化调用API环境，设置为我的Azure API"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "import openai, os\n",
    "from llama_index import VectorStoreIndex, SimpleDirectoryReader,ServiceContext,Document\n",
    "from langchain.embeddings import OpenAIEmbeddings\n",
    "from llama_index.llms import AzureOpenAI\n",
    "from llama_index import LangchainEmbedding\n",
    "\n",
    "#运行此API配置，需要将目录中的.env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "#================AZURE API======================\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "\n",
    "COMPLETION_MODEL = \"text-davinci-003\"\n",
    "\n",
    "def get_response(prompt):\n",
    "    completions = openai.Completion.create(\n",
    "        engine=COMPLETION_MODEL,\n",
    "        prompt=prompt,\n",
    "        max_tokens=512,\n",
    "        n=1,\n",
    "        stop=None,\n",
    "        temperature=0.5,        \n",
    "    )\n",
    "    return completions\n",
    "\n",
    "#==============llma_index的azure配置===============\n",
    "llm = AzureOpenAI(engine=\"text-davinci-003\", \n",
    "                  model=\"text-davinci-003\"              \n",
    ")\n",
    "embedding_llm = LangchainEmbedding(\n",
    "    OpenAIEmbeddings(\n",
    "        model=\"text-embedding-ada-002\",\n",
    "        deployment=\"text-embedding-ada-002\",\n",
    "        openai_api_key=openai.api_key,\n",
    "        openai_api_base=openai.api_base,\n",
    "        openai_api_type=openai.api_type,\n",
    "        openai_api_version=openai.api_version,\n",
    "    ),\n",
    "    embed_batch_size=1,\n",
    ")\n",
    "from llama_index import set_global_service_context\n",
    "service_context = ServiceContext.from_defaults(\n",
    "    llm=llm,\n",
    "    embed_model=embedding_llm\n",
    ")\n",
    "set_global_service_context(service_context)\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 藤野先生的例子"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "@ GPT : \n",
      "\n",
      "鲁迅先生在日本学习医学的城市是东京。\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# 直接咨询GPT\n",
    "answer = get_response(\"鲁迅先生在日本学习医学的老师是谁？\")\n",
    "print(\"@ GPT : %s\\n\" % answer.choices[0].text) #type: ignore\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "@ llama : \n",
      "仙台\n",
      "\n"
     ]
    }
   ],
   "source": [
    "#读入鲁迅小说\n",
    "documents = SimpleDirectoryReader('./data/mr_fujino').load_data()\n",
    "index = VectorStoreIndex.from_documents(documents)\n",
    "\n",
    "#向量化\n",
    "query_engine = index.as_query_engine()\n",
    "\n",
    "#向量检索和输出\n",
    "response = query_engine.query(\"鲁迅先生在日本学习医学的老师是谁？\")\n",
    "print(\"@ llama : %s\\n\" % response)\n",
    "\n",
    "# response = query_engine.query(\"鲁迅先生在日本学习医学的城市是哪个？\")\n",
    "# print(\"@ llama : %s\\n\" % response)\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 自问自答进行数据增强"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Q1：什么是卷积神经网络（CNN）？\n",
      "A1：卷积神经网络（CNN）是一种深度学习算法，特别适合于处理具有网格结构（如图像）的数据，它的设计灵感来自于生物的视觉皮层机制。\n",
      "\n",
      "Q2：CNN 的基本构成部分有哪些？\n",
      "A2：CNN 的基本构成部分包括：卷积层、激活层（ReLU）、池化层（Pooling）以及全连接层（Fully Connected Layer）。\n",
      "\n",
      "Q3：卷积层的作用是什么？\n",
      "A3：卷积层是 CNN 的核心，主要进行特征提取。卷积层通过滤波器（或称为卷积核）在输入数据上进行滑动卷积操作，以获取局部特征，同时，同一卷积层的所有滤波器共\n",
      "['\\nQ1：什么是卷积神经网络（CNN）？\\nA1：卷积神经网络（CNN）是一种深度学习算法，特别适合于处理具有网格结构（如图像）的数据，它的设计灵感来自于生物的视觉皮层机制。', 'Q2：CNN 的基本构成部分有哪些？\\nA2：CNN 的基本构成部分包括：卷积层、激活层（ReLU）、池化层（Pooling）以及全连接层（Fully Connected Layer）。', 'Q3：卷积层的作用是什么？\\nA3：卷积层是 CNN 的核心，主要进行特征提取。卷积层通过滤波器（或称为卷积核）在输入数据上进行滑动卷积操作，以获取局部特征，同时，同一卷积层的所有滤波器共']\n"
     ]
    }
   ],
   "source": [
    "content = \"\"\"卷积神经网络（Convolutional Neural Networks，简称 CNN）是一种深度学习算法，特别适合于处理具有网格结构（如图像）的数据。CNN 的设计灵感来自于生物的视觉皮层机制，通过对输入数据的局部感知和参数共享的特性，使得网络对图像的平移等变化有较好的鲁棒性。\n",
    "\n",
    "CNN 的基本构成部分包括：卷积层、激活层（ReLU）、池化层（Pooling）以及全连接层（Fully Connected Layer）。这些层次交织形成了一个典型的 CNN 结构。\n",
    "\n",
    "卷积层：这是 CNN 的核心，主要进行特征提取。卷积层通过滤波器（或称为卷积核）在输入数据上进行滑动卷积操作，以获取局部特征。同时，同一卷积层的所有滤波器共享参数，减少了模型复杂度。\n",
    "\n",
    "激活层：通常在卷积层之后接一个激活层，如 ReLU、tanh 或 sigmoid 等。激活函数的目的是引入非线性因素，使得网络可以拟合更复杂的函数。\n",
    "\n",
    "池化层：池化层通常在卷积层之后，用来降低特征的维度，防止过拟合。常见的池化方式有最大池化（Max Pooling）、平均池化（Average Pooling）等。\n",
    "\n",
    "全连接层：在整个网络的最后一般会有全连接层，用于将前面提取的特征进行融合，并输出结果。在分类问题中，输出层通常会接一个 softmax 层进行概率归一化，使得输出可以解释为概率。\n",
    "\n",
    "以上是 CNN 的基本结构，但实际上现代的 CNN 架构已经远超这些基础层次，如残差连接（ResNet）、空洞卷积（Dilated Convolution）、深度可分离卷积（Depthwise Separable Convolution）等等。\n",
    "\n",
    "CNN 在图像分类、检测、分割等问题上有显著的表现。经典的 CNN 架构有 LeNet-5、AlexNet、VGG、GoogLeNet、ResNet 等。\n",
    "\"\"\"\n",
    "\n",
    "prompt = f\"\"\"请针对>>>和<<<之间的内容，设计一系列的FAQ问题，以及对应的答案。\n",
    "<<<{content}>>>\"\"\"\n",
    "\n",
    "answer = get_response(prompt)\n",
    "message = answer.choices[0].text\n",
    "print(message)\n",
    "\n",
    "faq = message.split(\"\\n\\n\")\n",
    "print(faq)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "现代CNN架构包括：LeNet、AlexNet、VGGNet、GoogLeNet、ResNet等。\n"
     ]
    }
   ],
   "source": [
    "\n",
    "#把问题列表导入并向量化\n",
    "documents = []\n",
    "for qa in faq:\n",
    "    documents.append(Document(text=qa))\n",
    "\n",
    "index = VectorStoreIndex.from_documents(documents)\n",
    "\n",
    "query_engine = index.as_query_engine()\n",
    "response = query_engine.query(\"现代CNN架构有哪些？\")\n",
    "print(response)\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 作业 or 练习题\n",
    "\n",
    "是否可以通过llama-index来实现Self-Ask的功能？"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 通过多步骤的LLM和外部工具来解决问题"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "初期化langchain环境"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "import openai, os\n",
    "\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "\n",
    "from langchain.prompts import PromptTemplate\n",
    "from langchain.llms import AzureOpenAI\n",
    "from langchain.chains import LLMChain\n",
    "\n",
    "llm = AzureOpenAI(deployment_name=\"text-davinci-003\",\n",
    "                  model_name=\"text-davinci-003\",\n",
    "                  openai_api_version= \"2023-03-15-preview\",\n",
    "                  max_tokens=2048, \n",
    "                  temperature=0.5) # type: ignore"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LLM做数学题"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "@ OpenAI API 说答案是:\n",
      "\n",
      "352 x 493 = 174336\n",
      "\n",
      "\n",
      "@ Python 说答案是:173536\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# 直接询问GPT\n",
    "multiply_prompt = PromptTemplate(template=\"请计算一下{question}是多少?\", input_variables=[\"question\"])\n",
    "\n",
    "math_chain = LLMChain(llm=llm, prompt=multiply_prompt, output_key=\"answer\")\n",
    "answer = math_chain.run({\"question\": \"352乘以493\"})\n",
    "print(\"@ OpenAI API 说答案是:%s\\n\\n\" % answer)\n",
    "\n",
    "python_answer = 352 * 493\n",
    "print(\"@ Python 说答案是:%s\\n\\n\" % python_answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "print(352 * 493)\n"
     ]
    }
   ],
   "source": [
    "#让GPT写出python代码\n",
    "\n",
    "multiply_by_python_prompt = PromptTemplate(template=\"请写一段Python代码，计算{question}?\", input_variables=[\"question\"])\n",
    "math_chain = LLMChain(llm=llm, prompt=multiply_by_python_prompt, output_key=\"answer\")\n",
    "answer = math_chain.run({\"question\": \"352乘以493\"})\n",
    "print(answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "@ OpenAI_API的输出:\n",
      "\n",
      "print(352 * 493)\n",
      "\n",
      "\n",
      "@ 最后计算结果:173536\n",
      "\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "#让GPT写出python代码\n",
    "multiply_by_python_prompt = PromptTemplate(template=\"请写一段Python代码，计算{question}?\", input_variables=[\"question\"])\n",
    "math_chain = LLMChain(llm=llm, prompt=multiply_by_python_prompt, output_key=\"answer\")\n",
    "answer_code = math_chain.run({\"question\": \"352乘以493\"})\n",
    "\n",
    "print(\"@ OpenAI_API的输出:%s\\n\\n\" % answer_code)\n",
    "\n",
    "#执行Python代码\n",
    "from langchain.utilities import PythonREPL\n",
    "python_repl = PythonREPL()\n",
    "result = python_repl.run(answer_code)\n",
    "print(\"@ 最后计算结果:%s\\n\\n\" % result)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "请计算一下352乘以493是多少?\u001b[32;1m\u001b[1;3m```text\n",
      "352 * 493\n",
      "```\n",
      "...numexpr.evaluate(\"352 * 493\")...\n",
      "\u001b[0m\n",
      "Answer: \u001b[33;1m\u001b[1;3m173536\u001b[0m\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "Answer: 173536\n"
     ]
    }
   ],
   "source": [
    "from langchain import LLMMathChain\n",
    "\n",
    "llm_math = LLMMathChain(llm=llm, verbose=True) #type: ignore\n",
    "result = llm_math.run(\"请计算一下352乘以493是多少?\")\n",
    "print(result)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LLM调用搜索引擎"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain import LLMMathChain, OpenAI, SerpAPIWrapper\n",
    "from langchain.agents import initialize_agent, Tool\n",
    "from langchain.agents import AgentType\n",
    "\n",
    "#llm = OpenAI(temperature=0) # type: ignore\n",
    "search = SerpAPIWrapper() # type: ignore\n",
    "\n",
    "tools = [\n",
    "    Tool(\n",
    "        name=\"Search\",\n",
    "        func=search.run,\n",
    "        description=\"useful for when you need to answer questions about current events. You should ask targeted questions\"\n",
    "    ),\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "\u001b[32;1m\u001b[1;3m I should search for information about the NBA\n",
      "Action: Search\n",
      "Action Input: 2023年的NBA总冠军\u001b[0m"
     ]
    }
   ],
   "source": [
    "agent.run(\"2023年的NBA总冠军是谁？请用中文回答\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 做选择题的方式来解决问题"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Langchain的工具接口"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### FAQ"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain.embeddings.openai import OpenAIEmbeddings\n",
    "from langchain.vectorstores import FAISS\n",
    "from langchain.text_splitter import SpacyTextSplitter\n",
    "from langchain import OpenAI\n",
    "from langchain.chains import RetrievalQA\n",
    "from langchain.document_loaders import TextLoader\n",
    "from langchain.agents import tool\n",
    "\n",
    "import openai, os\n",
    "\n",
    "#运行此API配置，需要将目录中的。env中api key替换为自己的\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "# read local .env file\n",
    "load_dotenv(find_dotenv())\n",
    "\n",
    "openai.api_type = \"azure\"\n",
    "openai.api_base = \"https://alannewlife.openai.azure.com/\"\n",
    "openai.api_version = \"2023-03-15-preview\"\n",
    "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
    "#openai.api_key = os.environ.get(\"OPENAI_API_KEY\")\n",
    "\n",
    "from langchain.prompts import PromptTemplate\n",
    "from langchain.llms import AzureOpenAI\n",
    "from langchain.chains import LLMChain\n",
    "\n",
    "llm = AzureOpenAI(deployment_name=\"text-davinci-003\",\n",
    "                  model_name=\"text-davinci-003\",\n",
    "                  openai_api_version= \"2023-03-15-preview\",\n",
    "                  max_tokens=2048, \n",
    "                  temperature=0.5) # type: ignore\n",
    "\n",
    "embeddings= OpenAIEmbeddings(\n",
    "        model=\"text-embedding-ada-002\",\n",
    "        deployment=\"text-embedding-ada-002\",\n",
    "        openai_api_key=openai.api_key,\n",
    "        openai_api_base=openai.api_base,\n",
    "        openai_api_type=openai.api_type,\n",
    "        openai_api_version=openai.api_version,\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "loader = TextLoader('./data/ecommerce_faq.txt')\n",
    "documents = loader.load()\n",
    "text_splitter = SpacyTextSplitter(chunk_size=256, pipeline=\"zh_core_web_sm\")\n",
    "texts = text_splitter.split_documents(documents)\n",
    "\n",
    "#embeddings = OpenAIEmbeddings() # type: ignore\n",
    "docsearch = FAISS.from_documents(texts, embeddings)\n",
    "faq_chain = RetrievalQA.from_chain_type(llm=llm, retriever=docsearch.as_retriever(), verbose=True) # type: ignore\n",
    "\n",
    "@tool(\"FAQ\")\n",
    "def faq(intput: str) -> str:\n",
    "    \"\"\"\"useful for when you need to answer questions about shopping policies, like return policy, shipping policy, etc.\"\"\"\n",
    "    return faq_chain.run(intput)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      " 是的，我们支持三亚的配送。一般情况下，大部分城市的订单在2-3个工作日内送达，偏远地区可能需要5-7个工作日。\n"
     ]
    }
   ],
   "source": [
    "question = \"请问你们的货，能送到三亚吗？大概需要几天？\"\n",
    "result = faq_chain.run(question)\n",
    "print(result)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 商品信息"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain.text_splitter import CharacterTextSplitter\n",
    "from langchain.document_loaders import CSVLoader\n",
    "\n",
    "product_loader = CSVLoader('./data/ecommerce_products.csv')\n",
    "product_documents = product_loader.load()\n",
    "product_text_splitter = CharacterTextSplitter(chunk_size=1024, separator=\"\\n\")\n",
    "product_texts = product_text_splitter.split_documents(product_documents)\n",
    "product_search = FAISS.from_documents(product_texts, embeddings)\n",
    "product_chain = RetrievalQA.from_chain_type(llm=llm, retriever=product_search.as_retriever(), verbose=True)# type: ignore\n",
    "\n",
    "@tool(\"Recommend Product\")\n",
    "def recommend_product(input: str) -> str:\n",
    "    \"\"\"\"useful for when you need to search and recommend products and recommend it to the user\"\"\"\n",
    "    return product_chain.run(input)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      " 我建议你买雪纺连衣裙，它是轻盈柔软的质地，带有浪漫的碎花印花，穿上后显得温柔可人，风格是田园、休闲，很适合夏季出门穿着，搭配一双凉鞋和小包包就可以了。\n"
     ]
    }
   ],
   "source": [
    "question = \"我想买一件衣服，想要在夏天去公园穿，但是不知道哪个款式好看，你能帮我推荐一下吗？\"\n",
    "answer = product_chain.run(question)\n",
    "print(answer)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 订单查询"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "import re\n",
    "import json\n",
    "\n",
    "ORDER_1 = \"20230101ABC\"\n",
    "ORDER_2 = \"20230101EFG\"\n",
    "\n",
    "ORDER_1_DETAIL = {\n",
    "    \"order_number\": ORDER_1,\n",
    "    \"status\": \"已发货\",\n",
    "    \"shipping_date\" : \"2023-01-03\",\n",
    "    \"estimated_delivered_date\": \"2023-01-05\",\n",
    "} \n",
    "\n",
    "ORDER_2_DETAIL = {\n",
    "    \"order_number\": ORDER_2,\n",
    "    \"status\": \"未发货\",\n",
    "    \"shipping_date\" : None,\n",
    "    \"estimated_delivered_date\": None,\n",
    "}\n",
    "\n",
    "answer_order_info = PromptTemplate(\n",
    "    template=\"请把下面的订单信息回复给用户： \\n\\n {order}?\", input_variables=[\"order\"]\n",
    ")\n",
    "answer_order_llm = LLMChain(llm=llm,  prompt=answer_order_info) # type: ignore\n",
    "\n",
    "\n",
    "@tool(\"Search Order\", return_direct=True)\n",
    "def search_order(input:str)->str:\n",
    "    \"\"\"useful for when you need to answer questions about customers orders\"\"\"\n",
    "    pattern = r\"\\d+[A-Z]+\"\n",
    "    match = re.search(pattern, input)\n",
    "\n",
    "    order_number = input\n",
    "    if match:\n",
    "        order_number = match.group(0)\n",
    "    else:\n",
    "        return \"请问您的订单号是多少？\"\n",
    "    if order_number == ORDER_1:        \n",
    "        return answer_order_llm.run(json.dumps(ORDER_1_DETAIL))\n",
    "    elif order_number == ORDER_2:\n",
    "        return answer_order_llm.run(json.dumps(ORDER_2_DETAIL))\n",
    "    else:\n",
    "        return f\"对不起，根据{input}没有找到您的订单\"\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "tools = {\"search_order\" : search_order, \"recommend_product\": recommend_product, \"faq\": faq}\n",
    "\n",
    "\n",
    "choicellm = llm # type: ignore\n",
    "multiple_choice = \"\"\"\n",
    "请针对用户问题，选择一个合适的工具去回答她的问题。只要用A、B、C的选项字母告诉我答案。\n",
    "\n",
    "我们有的工具包括：\n",
    "A. recommend_product[$keywords], 一个能够查询商品信息，为用户进行商品导购的工具。\n",
    "B. search_order[$order_number], 一个能够查询订单信息，获得最新的订单情况的工具\n",
    "C. faq[$question], 一个能够搜索商家的退换货政策、运费、物流时长、支付渠道、覆盖国家的工具\n",
    "\n",
    "\n",
    "返回的结果需要包含函数名称，以及对应的参数，比如：\n",
    "\n",
    "用户问题：我的订单 20220101ABC 的物流信息是什么？\n",
    "返回结果：search_order[20220101ABC]\n",
    "\n",
    "用户问题：{question}\n",
    "返回结果：\n",
    "\"\"\"\n",
    "multiple_choice_prompt = PromptTemplate(template=multiple_choice, input_variables=[\"question\"])\n",
    "choice_chain = LLMChain(llm=choicellm, prompt=multiple_choice_prompt, output_key=\"answer\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "search_order[2022ABCDE]\n"
     ]
    }
   ],
   "source": [
    "question = \"我有一张订单，订单号是 2022ABCDE，一直没有收到，能麻烦帮我查一下吗？\"\n",
    "result = choice_chain.run(question)\n",
    "print(result) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('search_order', '2022ABCDE')\n"
     ]
    }
   ],
   "source": [
    "from typing import Tuple\n",
    "\n",
    "def parse_result(result: str) -> Tuple[str, str]:\n",
    "    pattern = r\"(\\w+)\\[(.+)\\]\"\n",
    "    match = re.search(pattern, result)\n",
    "    if match:\n",
    "        return match.group(1), match.group(2)\n",
    "    else:\n",
    "        return \"\", \"\"\n",
    "    \n",
    "\n",
    "print(parse_result(result))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "对不起，根据2022ABCDE没有找到您的订单\n"
     ]
    }
   ],
   "source": [
    "parsed_result = parse_result(result)\n",
    "method = tools.get(parsed_result[0])\n",
    "params = parsed_result[1]\n",
    "\n",
    "\n",
    "answer = method(params) # type: ignore\n",
    "print(answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "您的订单号为20230101ABC，状态为已发货，预计发货日期为2023-01-03，预计到货日期为2023-01-05。\n"
     ]
    }
   ],
   "source": [
    "question = \"我有一张订单，订单号是 20230101ABC，一直没有收到，能麻烦帮我查一下吗？\"\n",
    "\n",
    "result = choice_chain.run(question)\n",
    "parsed_result = parse_result(result)\n",
    "method = tools.get(parsed_result[0])\n",
    "params = parsed_result[1]\n",
    "\n",
    "\n",
    "answer = method(params) # type: ignore\n",
    "print(answer)\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 作业，通过Langchain实现上面的功能"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Langchain的Agent功能"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "agent_tools = [search_order,recommend_product, faq]\n",
    "agent = initialize_agent(agent_tools, llm, agent=AgentType.CHAT_ZERO_SHOT_REACT_DESCRIPTION, verbose=True) #type: ignore"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "\u001b[32;1m\u001b[1;3m\n",
      "Thought: I need to find the status of the order\n",
      "Action:\n",
      "```\n",
      "{\n",
      "  \"action\": \"Search Order\",\n",
      "  \"action_input\": \"2022ABCDE\"\n",
      "}\n",
      "```\n",
      "\u001b[0m\n",
      "Observation: \u001b[36;1m\u001b[1;3m对不起，根据2022ABCDE没有找到您的订单\u001b[0m\n",
      "\u001b[32;1m\u001b[1;3m\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "对不起，根据2022ABCDE没有找到您的订单\n"
     ]
    }
   ],
   "source": [
    "question = \"我有一张订单，订单号是 2022ABCDE，一直没有收到，能麻烦帮我查一下吗？\"\n",
    "answer = agent.run(question)\n",
    "print(answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "\u001b[32;1m\u001b[1;3m\n",
      "Thought: The customer is asking about an order they placed.\n",
      "\n",
      "Action:\n",
      "```\n",
      "{\n",
      "  \"action\": \"Search Order\",\n",
      "  \"action_input\": \"20230101ABC\"\n",
      "}\n",
      "```\n",
      "\n",
      "\u001b[0m\n",
      "Observation: \u001b[36;1m\u001b[1;3m\n",
      "\n",
      "您的订单号为20230101ABC，状态为已发货，发货日期为2023-01-03，预计到货日期为2023-01-05。\u001b[0m\n",
      "\u001b[32;1m\u001b[1;3m\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "\n",
      "\n",
      "您的订单号为20230101ABC，状态为已发货，发货日期为2023-01-03，预计到货日期为2023-01-05。\n"
     ]
    }
   ],
   "source": [
    "question2 = \"我有一张订单，订单号是 20230101ABC 一直没有收到，能麻烦帮我查一下吗？\"\n",
    "answer2 = agent.run(question2)\n",
    "print(answer2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "\u001b[32;1m\u001b[1;3m\n",
      "Thought: I need to answer a question about the return policy\n",
      "Action: \n",
      "```\n",
      "{\n",
      "  \"action\": \"FAQ\",\n",
      "  \"action_input\": \"return policy\"\n",
      "}\n",
      "```\n",
      "\u001b[0m\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "\n",
      "Observation: \u001b[38;5;200m\u001b[1;3m 我们提供30天无条件退货政策，具体退货流程请查看我们的退货政策页面。\u001b[0m\n",
      "Thought:\u001b[32;1m\u001b[1;3m I now know the final answer\n",
      "Final Answer: 我们提供30天无条件退货政策，具体退货流程请查看我们的退货政策页面。\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "我们提供30天无条件退货政策，具体退货流程请查看我们的退货政策页面。\n"
     ]
    }
   ],
   "source": [
    "question3 = \"你们的退货政策是怎么样的？\"\n",
    "answer3 = agent.run(question3)\n",
    "print(answer3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "\u001b[32;1m\u001b[1;3m\n",
      "Thought: I need to recommend a product for the user\n",
      "Action:\n",
      "```\n",
      "{\n",
      "  \"action\": \"Recommend Product\",\n",
      "  \"action_input\": \"Clothing for summer\"\n",
      "}\n",
      "```\n",
      "\u001b[0m\n",
      "\n",
      "\u001b[1m> Entering new  chain...\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "\n",
      "Observation: \u001b[33;1m\u001b[1;3m 这款雪纺连衣裙是适合夏季出门穿着的，风格是田园、休闲。\u001b[0m\n",
      "Thought:\u001b[32;1m\u001b[1;3m I now know the final answer\n",
      "Final Answer: 这款雪纺连衣裙是适合夏季出门穿着的，风格是田园、休闲。\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "这款雪纺连衣裙是适合夏季出门穿着的，风格是田园、休闲。\n"
     ]
    }
   ],
   "source": [
    "question4 = \"能推荐一件现在适合夏天穿的衣服给我么？\"\n",
    "answer4 = agent.run(question4)\n",
    "print(answer4)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "mactalk",
   "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"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}