计图 NLP 教程 - BERT 分类器.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "58c5ed1f",
   "metadata": {},
   "source": [
    "# 计图 NLP 教程 - BERT分类器\n",
    "本教程最后更新于 2022.4.16\n",
    "\n",
    "#### 任务描述：\n",
    "IMDb 影评分类问题。程序将根据给出的英文影评，判断其为正面评价（Positive），或是负面评价（Negative）。  \n",
    "注意：正面评价将用数字 “1” 表示，负面评价将用数字 “0” 表示。\n",
    "\n",
    "*示例：*  \n",
    "> 影评内容：  \n",
    "> Magnificent and unforgettable, stunningly atmospheric, and brilliantly acted by all. I really cannot understand what sort of people are panning this masterpiece and giving the preponderance of votes as 8 (and nine ones!) This, along with Grapes of Wrath, is John Ford's greatest movie. I would say that Long Voyage Home is next in line, though quite a way back. Rating: 10. It deserves a 12.   \n",
    "> (该条英文评论为正面评价。)  \n",
    ">  \n",
    "> 正确标签：1  \n",
    "\n",
    "\n",
    "#### 解决流程：\n",
    "本程序将按照以下几个步骤，解决目标任务:\n",
    "1. 准备工作  \n",
    "2. 数据处理 \n",
    "3. 加载模型及优化器  \n",
    "4. 训练 \n",
    "5. 测试 \n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "33851fee",
   "metadata": {},
   "source": [
    "## 1. 准备工作"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d9fbee9e",
   "metadata": {},
   "source": [
    "导入计图，以及其他通用包，并开启 GPU 加速。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "56466434",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 加载计图，以及其他通用包\n",
    "import jittor as jt\n",
    "import time\n",
    "\n",
    "# 开启 GPU 加速\n",
    "jt.flags.use_cuda = 1"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0755ad36",
   "metadata": {},
   "source": [
    "## 2. 数据处理"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ad4a1388",
   "metadata": {},
   "source": [
    "### 2.1 下载并读取数据"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "12c59239",
   "metadata": {},
   "source": [
    "本项目所使用的数据集为 **IMDb 影评** 数据集。原数据集由 [Large Movie Review Dataset](https://ai.stanford.edu/~amaas/data/sentiment/)  网页提供。"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1af31836",
   "metadata": {},
   "source": [
    "这里，我们直接从 Jittor 服务器下载该数据集。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "7ca5cf53",
   "metadata": {},
   "outputs": [],
   "source": [
    "from jittor_utils.misc import download_url_to_local\n",
    "\n",
    "download_url_to_local(\n",
    "    url=\"https://cg.cs.tsinghua.edu.cn/jittor/assets/build/checkpoints/aclImdb_train.csv\", \n",
    "    filename=\"aclImdb_train.csv\", \n",
    "    root_folder=\"./\", \n",
    "    md5=None)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8f8e7ba7",
   "metadata": {},
   "source": [
    "随后，我们利用 Pandas 读取下载到本地的数据集。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "569ca1dc",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "\n",
    "imdb_data = pd.read_csv(\"aclImdb_train.csv\", header=0, names=['Text', 'Label'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7fcd43e2",
   "metadata": {},
   "source": [
    "现在，让我们直观的查看一下，我们所拿到的数据集。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "036a21a9",
   "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>Label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Magnificent and unforgettable, stunningly atmo...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>I got hooked on this as apparently ABC has lic...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>(SPOILERS IN FIRST PARAGRAPH) This movie's ant...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Masayuki Suo, who directed this fine film, is ...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>It is a well known fact that when Gene Roddenb...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24995</th>\n",
       "      <td>As a big fan of David Mamet's films and plays,...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24996</th>\n",
       "      <td>There seems to be an overwhelming response to ...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24997</th>\n",
       "      <td>For the first couple of seasons, I thought The...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24998</th>\n",
       "      <td>WE FAW DOWN &lt;br /&gt;&lt;br /&gt;Aspect ratio: 1.33:1&lt;b...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24999</th>\n",
       "      <td>Although there's Flying Guillotines as part of...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>25000 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                    Text  Label\n",
       "0      Magnificent and unforgettable, stunningly atmo...      1\n",
       "1      I got hooked on this as apparently ABC has lic...      1\n",
       "2      (SPOILERS IN FIRST PARAGRAPH) This movie's ant...      1\n",
       "3      Masayuki Suo, who directed this fine film, is ...      1\n",
       "4      It is a well known fact that when Gene Roddenb...      1\n",
       "...                                                  ...    ...\n",
       "24995  As a big fan of David Mamet's films and plays,...      0\n",
       "24996  There seems to be an overwhelming response to ...      0\n",
       "24997  For the first couple of seasons, I thought The...      0\n",
       "24998  WE FAW DOWN <br /><br />Aspect ratio: 1.33:1<b...      0\n",
       "24999  Although there's Flying Guillotines as part of...      0\n",
       "\n",
       "[25000 rows x 2 columns]"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 展示文本 Text 及其标签 Label\n",
    "imdb_data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "062c42be",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    12500\n",
       "1    12500\n",
       "Name: Label, dtype: int64"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 查看数据集上标签为 0 和 1 的数据个数\n",
    "imdb_data[\"Label\"].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "b9fd5b7f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "文本：\n",
      "Magnificent and unforgettable, stunningly atmospheric, and brilliantly acted by all.<br /><br />I really cannot understand what sort of people are panning this masterpiece and giving the preponderance of votes as 8 (and nine ones!)<br /><br />This, along with Grapes of Wrath, is John Ford's greatest movie. I would say that Long Voyage Home is next in line, though quite a way back.<br /><br />Rating: 10. It deserves a 12.\n",
      "\n",
      "标签：1\n"
     ]
    }
   ],
   "source": [
    "# 浏览第一个数据文本\n",
    "print(u\"文本：\\n%s\\n\" % imdb_data[\"Text\"][0])\n",
    "print(u\"标签：%s\" % imdb_data[\"Label\"][0])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4f750823",
   "metadata": {},
   "source": [
    "最后，我们将 **数据集** 切分成 **训练集** 和 **测试集**。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "de5616dc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1    1258\n",
       "0    1242\n",
       "Name: Label, dtype: int64"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 为节省时间，只抽取部分数据做演示\n",
    "imdb_data = imdb_data.sample(frac=0.2) \n",
    "\n",
    "# 按比例切分“训练集”和“测试集”。利用 frac 设置切分的比例\n",
    "train_data = imdb_data.sample(frac=0.5) \n",
    "test_data = imdb_data.drop(train_data.index)\n",
    "train_data[\"Label\"].value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ae44ea8d",
   "metadata": {},
   "source": [
    "### 2.2 分词（Tokenization）"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "d55a7aa6",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tokenization import BertTokenizer\n",
    "\n",
    "vocab_file = \"./tokenization/vocab.txt\"\n",
    "tokenizer = BertTokenizer(do_lower_case=True, model_max_length=512, vocab_file=vocab_file)\n",
    "\n",
    "# 对文本进行分词\n",
    "train_encodings = tokenizer(list(train_data[\"Text\"]), padding=True)\n",
    "test_encodings = tokenizer(list(test_data[\"Text\"]), padding=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f9581263",
   "metadata": {},
   "source": [
    "### 2.3 将数据存入 Dataset 对象中\n",
    "定义一个 Dataset 类，并将完成 Tokenization 后的数据文本（encodings）和数据标签（labels）封装为数据对象，以便模型使用。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "c0bb0373",
   "metadata": {},
   "outputs": [],
   "source": [
    "class IMDbDataset(jt.dataset.Dataset):\n",
    "    def __init__(self, encodings, labels):\n",
    "        super().__init__()\n",
    "        self.encodings = encodings\n",
    "        self.labels = labels\n",
    "\n",
    "    def __getitem__(self, idx):\n",
    "        item = {key: jt.array(val[idx]) for key, val in self.encodings.items()}\n",
    "        item['labels'] = jt.array(self.labels[idx])\n",
    "        return item\n",
    "\n",
    "    def __len__(self):\n",
    "        return len(self.labels)\n",
    "    \n",
    "train_dataset = IMDbDataset(train_encodings, list(train_data[\"Label\"])).set_attrs(batch_size=4, shuffle=True)\n",
    "test_dataset = IMDbDataset(test_encodings, list(test_data[\"Label\"])).set_attrs(batch_size=1, shuffle=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "855902fe",
   "metadata": {},
   "source": [
    "## 3. 加载模型及优化器\n",
    "加载 BERT 模型和 AdamW 优化器。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "9517be6b",
   "metadata": {},
   "outputs": [],
   "source": [
    "from model import BertClassification, BertConfig\n",
    "from jittor.optim import AdamW\n",
    "\n",
    "# 加载 BERT 模型\n",
    "configuration = BertConfig()\n",
    "model = BertClassification(configuration)\n",
    "model.load_state_dict(jt.load(\"jittorhub://pretrained_bert.bin\")) # 加载预训练的参数\n",
    "\n",
    "# 加载 AdamW 优化器\n",
    "optim = AdamW(model.parameters(), lr=1e-5)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d6db8ee7",
   "metadata": {},
   "source": [
    "## 4. 训练"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "27387668",
   "metadata": {},
   "source": [
    "首先，准备一个实用方法。目标是将模型的输出转化为预测值“1” 和 “0”，与原本的数据标签所对应。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "127bec50",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 预测\n",
    "def predict(outputs):\n",
    "    probabilities = jt.nn.softmax(outputs[\"logits\"], dim=1)\n",
    "    predictions = jt.argmax(probabilities, dim=1)[0]\n",
    "    return predictions"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "54b4f0ac",
   "metadata": {},
   "source": [
    "训练模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "43001330",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 625/625 [00:46<00:00, 13.45it/s, Loss=0.129, Accuracy=0.775]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 2/3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 625/625 [00:48<00:00, 13.01it/s, Loss=0.144, Accuracy=0.905]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 3/3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 625/625 [00:49<00:00, 12.55it/s, Loss=0.166, Accuracy=0.959]\n"
     ]
    }
   ],
   "source": [
    "# 训练代码\n",
    "from tqdm import tqdm\n",
    "\n",
    "# 设置训练周期为 3\n",
    "epoch = 3\n",
    "\n",
    "# 开启训练模式\n",
    "model.train()\n",
    "\n",
    "train_loss = list()\n",
    "train_accuracies = list()\n",
    "for epoch_i in range(epoch):\n",
    "    print('Epoch %s/%s' % (epoch_i + 1, epoch))\n",
    "    time.sleep(0.3)\n",
    "\n",
    "    correct = 0\n",
    "    count = 0\n",
    "    epoch_loss = list()\n",
    "    \n",
    "    pbar = tqdm(train_dataset, total=len(train_dataset)//train_dataset.batch_size)\n",
    "    for batch in pbar:\n",
    "        \n",
    "        optim.zero_grad()\n",
    "        input_ids = batch['input_ids']\n",
    "        attention_mask = batch['attention_mask']\n",
    "        labels = batch['labels']\n",
    "        outputs = model(input_ids, attention_mask=attention_mask, labels=labels)\n",
    "        loss = outputs['loss']\n",
    "        optim.step(loss)\n",
    "        \n",
    "        # 基于模型的输出进行预测\n",
    "        predictions = predict(outputs)\n",
    "\n",
    "        # 计算准确率 accuracy\n",
    "        correct += predictions.equal(labels.reshape(-1)).sum().item()\n",
    "        count += len(labels)\n",
    "        accuracy = correct * 1.0 / count\n",
    "\n",
    "        # 显示进度条指标\n",
    "        pbar.set_postfix({\n",
    "            'Loss': '{:.3f}'.format(loss.item()),\n",
    "            'Accuracy': '{:.3f}'.format(accuracy)\n",
    "        })\n",
    "        \n",
    "        # 记录每个批次的损失值 loss\n",
    "        epoch_loss.append(loss.item())\n",
    "        \n",
    "    pbar.close()\n",
    "    \n",
    "    # 记录每个 epoch 的 loss 和 accuracy\n",
    "    train_loss += epoch_loss\n",
    "    train_accuracies.append(accuracy)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e22f172c",
   "metadata": {},
   "source": [
    "可视化训练损失 loss 和 准确率 accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "a34acb60",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "# 每个迭代 vs. 损失\n",
    "plt.plot(train_loss, label=\"Training Loss\")\n",
    "plt.xlabel(\"Iteration\")\n",
    "plt.ylabel(\"Loss\")\n",
    "plt.title(\"Iteration vs Training Loss\")  \n",
    "plt.legend()\n",
    "plt.show()\n",
    "\n",
    "# 每个纪元 vs. 准确率\n",
    "acc_X = np.arange(len(train_accuracies))+1      # 重置 X 轴坐标\n",
    "plt.plot(acc_X, train_accuracies,\"-\", label=\"Training Accuracy\")\n",
    "plt.xticks(acc_X)\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylabel(\"Accuracy\")\n",
    "plt.title(\"Epoch vs Training Accuracy\")  \n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "35b5fc37",
   "metadata": {},
   "source": [
    "## 5. 测试"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "54b19868",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 2500/2500 [00:37<00:00, 67.22it/s, loss=0.025, accuracy=0.830]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "测试集上准确率: 82.96%\n"
     ]
    }
   ],
   "source": [
    "# 测试代码\n",
    "\n",
    "# 开启测试模式\n",
    "model.eval()\n",
    "\n",
    "with jt.no_grad():  # 关闭梯度\n",
    "    \n",
    "    correct = 0\n",
    "    count = 0\n",
    "    record = {\"labels\":list(), \"predictions\":list()}\n",
    "    \n",
    "    pbar = tqdm(test_dataset, total=len(test_dataset)//test_dataset.batch_size)\n",
    "    for batch in pbar:\n",
    "\n",
    "        input_ids = batch['input_ids']\n",
    "        attention_mask = batch['attention_mask']\n",
    "        labels = batch['labels']\n",
    "        outputs = model(input_ids, attention_mask=attention_mask, labels=labels)\n",
    "        loss = outputs['loss']\n",
    "        \n",
    "        # 基于模型的输出进行预测\n",
    "        predictions = predict(outputs)\n",
    "\n",
    "        # 计算准确率 accuracy\n",
    "        correct += predictions.equal(labels.reshape(-1)).sum().item()\n",
    "        count += len(labels)\n",
    "        accuracy = correct * 1.0 / count\n",
    "\n",
    "        # 显示进度条指标\n",
    "        pbar.set_postfix({\n",
    "            'loss': '{:.3f}'.format(loss.item()),\n",
    "            'accuracy': '{:.3f}'.format(accuracy)\n",
    "        })\n",
    "    \n",
    "        # 记录结果\n",
    "        record[\"labels\"] += list(labels.reshape(-1).numpy())\n",
    "        record[\"predictions\"] += list(predictions.numpy())\n",
    "        \n",
    "    pbar.close()\n",
    "    \n",
    "time.sleep(0.3)\n",
    "print(u\"测试集上准确率: %s%%\" % round(accuracy*100,4))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9dc65ea1",
   "metadata": {},
   "source": [
    "查看结果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "7b41e16b",
   "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>index</th>\n",
       "      <th>Text</th>\n",
       "      <th>Label</th>\n",
       "      <th>Model Prediction</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>5573</td>\n",
       "      <td>the photography was beautiful but i had diffic...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>20449</td>\n",
       "      <td>San Francisco is a big city with great acting ...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>20133</td>\n",
       "      <td>Okay, I've watched this movie twice now, I hav...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>17128</td>\n",
       "      <td>This esteemed production has it's fans. But to...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>14961</td>\n",
       "      <td>Now, I've seen a lot of bad movies. I like bad...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2468</th>\n",
       "      <td>19365</td>\n",
       "      <td>You gotta love the cheesy low budget movies. T...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2485</th>\n",
       "      <td>1038</td>\n",
       "      <td>maybe i need to have my head examined,but i th...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2486</th>\n",
       "      <td>15821</td>\n",
       "      <td>It wasn't the most pointless animation film ex...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2487</th>\n",
       "      <td>14353</td>\n",
       "      <td>When they killed off John Amos's character the...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2495</th>\n",
       "      <td>20713</td>\n",
       "      <td>I'm sure that Operations Dames was a favorite ...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>426 rows × 4 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "      index                                               Text  Label  \\\n",
       "10     5573  the photography was beautiful but i had diffic...      1   \n",
       "16    20449  San Francisco is a big city with great acting ...      0   \n",
       "17    20133  Okay, I've watched this movie twice now, I hav...      0   \n",
       "19    17128  This esteemed production has it's fans. But to...      0   \n",
       "24    14961  Now, I've seen a lot of bad movies. I like bad...      0   \n",
       "...     ...                                                ...    ...   \n",
       "2468  19365  You gotta love the cheesy low budget movies. T...      0   \n",
       "2485   1038  maybe i need to have my head examined,but i th...      1   \n",
       "2486  15821  It wasn't the most pointless animation film ex...      0   \n",
       "2487  14353  When they killed off John Amos's character the...      0   \n",
       "2495  20713  I'm sure that Operations Dames was a favorite ...      0   \n",
       "\n",
       "      Model Prediction  \n",
       "10                   0  \n",
       "16                   1  \n",
       "17                   1  \n",
       "19                   1  \n",
       "24                   1  \n",
       "...                ...  \n",
       "2468                 1  \n",
       "2485                 0  \n",
       "2486                 1  \n",
       "2487                 1  \n",
       "2495                 1  \n",
       "\n",
       "[426 rows x 4 columns]"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 将测试结果转化为 pandas DataFrame 对象\n",
    "from pandas.core.frame import DataFrame\n",
    "df_record = DataFrame(record)\n",
    "df_record.columns = [\"Ground Truth\",\"Model Prediction\"]\n",
    "\n",
    "# 拼接测试文本和测试结果\n",
    "df = pd.concat([test_data.reset_index(), df_record[\"Model Prediction\"]], axis=1)\n",
    "\n",
    "# 显示测试结果\n",
    "# pd.set_option('display.max_rows', None)    # 查看所有行\n",
    "# df\n",
    "\n",
    "# 只显示错误的预测 \n",
    "df[df[\"Label\"]!=df[\"Model Prediction\"]]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f185b083",
   "metadata": {},
   "source": [
    "显示测试结果的混淆矩阵"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "c52fbe1c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 显示测试结果的混淆矩阵\n",
    "import seaborn as sns\n",
    "crosstab = pd.crosstab(df_record[\"Ground Truth\"],df_record[\"Model Prediction\"])\n",
    "sns.heatmap(crosstab, cmap='Oranges', annot=True, fmt='g', linewidths=5)\n",
    "accuracy = df_record[\"Ground Truth\"].eq(df_record[\"Model Prediction\"]).sum() / len(df_record[\"Ground Truth\"])\n",
    "plt.title(\"Confusion Matrix (Accuracy: %s%%)\" % round(accuracy*100,2))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "773a79af",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "现在开始手动测试：\n",
      "\n",
      "请输入想要测试的影评（英文）：\n",
      "I don't like it, beacuse it is a boring movie. It's kind of waste my time\n",
      "\n",
      "模型认为这是一条负面评价。\n",
      "\n",
      "再试一次？ y/n\n",
      "n\n"
     ]
    }
   ],
   "source": [
    "\"\"\"\n",
    "- 手动测试 -\n",
    "\n",
    "以下代码将用于手动测试。\n",
    "可自定义输入评论，模型将判断其为正面评价还是负面评价。\n",
    "\n",
    "如：\n",
    "输入“I don't like it, beacuse it is a boring movie. It's kind of waste my time”\n",
    "期待模型判断为负面评价(negative)。\n",
    "\n",
    "注意，只能输入英文。\n",
    "\"\"\"\n",
    "\n",
    "manual_test = True\n",
    "count = 1\n",
    "print(u\"现在开始手动测试：\\n\")\n",
    "while(manual_test):\n",
    "    \n",
    "    review = input(u\"请输入想要测试的影评（英文）：\\n\")\n",
    "\n",
    "    # 对影评文本进行分词\n",
    "    inputs = tokenizer(review, padding=True, return_tensors=\"jt\")\n",
    "    \n",
    "    # 模型预测\n",
    "    outputs = model(**inputs)\n",
    "    prediction = predict(outputs)\n",
    "\n",
    "    # 检查性能\n",
    "    print(\"\\n模型认为这是一条%s。\\n\" % (u\"负面评价\" if int(prediction)==0 else u\"正面评价\"))\n",
    "    \n",
    "    manual_test = input(u\"再试一次？ y/n\\n\")==\"y\"\n",
    "    if manual_test:\n",
    "        count = count + 1\n",
    "        print(\"--------------------------------------- %s ---------------------------------------\" % count)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}