2/textCNN_classify.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "代码参考：\n",
    "https://discourse.qingxzd.com/t/pytorch/58\n",
    "\n",
    "辅助参考:\n",
    "https://www.pythonf.cn/read/128035"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "\n",
    "from torchtext.legacy import data,datasets\n",
    "\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "import torch.nn.functional as F\n",
    "from torchtext.vocab import Vectors\n",
    "from torch.nn import init\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "import jieba    # 用来预处理文本（分词等）\n",
    "\n",
    "\n",
    "import nltk\n",
    "from nltk.corpus import stopwords\n",
    "from tqdm import tqdm\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "seed=2019\n",
    "torch.manual_seed(seed)\n",
    "torch.backends.cudnn.deterministic=True\n",
    "\n",
    "DEVICE = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")   # 选择Gpu或Cpu"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "df=pd.read_csv('data/train.tsv', sep='\\t')\n",
    "df_test=pd.read_csv('data/test.tsv', sep='\\t')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "train, val = train_test_split(df, test_size=0.2)\n",
    "train.to_csv(\"./data/train.csv\", index=False)\n",
    "val.to_csv(\"./data/val.csv\", index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tokenizer(text):    \n",
    "    return [wd for wd in jieba.cut(text, cut_all=False)]\n",
    "\n",
    "en_stopwords=stopwords.words('english')\n",
    "\n",
    "FIX_LENGTH=50\n",
    "LABEL = data.Field(sequential=False, use_vocab=False)\n",
    "\n",
    "TEXT = data.Field(sequential=True, tokenize=tokenizer,fix_length=FIX_LENGTH, lower=True, stop_words=en_stopwords)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "124848 [' ', ' ', ' ', 'new', ' ', 'yorker', ' ', '--', ' ', ' ', ',', ' ', 'really', ' ', ',', ' ', ' ', ' ', ' ', 'human', ' ', ' ', '--', ' ', ' ', ' ', 'weeks', ' ', ' ', '9', '\\\\', '/', '11'] 2\n"
     ]
    }
   ],
   "source": [
    "train, val = data.TabularDataset.splits(\n",
    "    path='./data', train='train.csv', validation='val.csv', format='csv', skip_header=True,\n",
    "    fields=[('PhraseId', None), ('SentenceId', None), ('Phrase', TEXT), ('Sentiment', LABEL)]\n",
    ")\n",
    "\n",
    "test = data.TabularDataset('./data/test.tsv', format='tsv', skip_header=True, \n",
    "                           fields=[('PhraseId', None), ('SentenceId', None), ('Phrase', TEXT)])\n",
    "print(len(train),train[2].Phrase, train[2].Sentiment)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# # 建立vocab（不需要加载预训练的词向量） \n",
    "# TEXT.build_vocab(train, val)\n",
    "# LABEL.build_vocab(train, val)\n",
    "\n",
    "# 建立vocab（加载预训练的词向量，如果路径没有该词向量，会自动下载）\n",
    "# TEXT.build_vocab(train, vectors='glove.6B.100d')#, max_size=30000)\n",
    "TEXT.build_vocab(train)#, max_size=30000)\n",
    "\n",
    "# 当 corpus 中有的 token 在 vectors 中不存在时 的初始化方式.\n",
    "# TEXT.vocab.vectors.unk_init = init.xavier_uniform"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 构造迭代器\n",
    "'''\n",
    "sort_key指在一个batch内根据文本长度进行排序。\n",
    "'''\n",
    "train_iter = data.BucketIterator(train, batch_size=128, sort_key=lambda x: len(x.Phrase), \n",
    "                                 shuffle=True,device=DEVICE)\n",
    "\n",
    "val_iter = data.BucketIterator(val, batch_size=128, sort_key=lambda x: len(x.Phrase), \n",
    "                                 shuffle=True,device=DEVICE)\n",
    "\n",
    "# 在 test_iter , sort一定要设置成 False, 要不然会被 torchtext 搞乱样本顺序\n",
    "test_iter = data.Iterator(dataset=test, batch_size=128, train=False,\n",
    "                          sort=False, device=DEVICE)\n",
    "\n",
    "# 查看trainiter一个batch\n",
    "\n",
    "# batch = next(iter(train_iter))\n",
    "# a= batch.Phrase\n",
    "# label = batch.Sentiment\n",
    "# print(a.shape)\n",
    "# print(batch.Phrase)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Size of TEXT vocabulary: 15156\n"
     ]
    }
   ],
   "source": [
    "#文本中的唯一标记\n",
    "print(\"Size of TEXT vocabulary:\",len(TEXT.vocab))\n",
    "\n",
    "#标签中唯一令牌的集合\n",
    "# print(\"Size of LABEL vocabulary:\",len(LABEL.vocab))\n",
    "\n",
    "#常用单词\n",
    "# print(TEXT.vocab.freqs.most_common(10))\n",
    "\n",
    "# 词库向量的大小\n",
    "# print(TEXT.vocab.vectors.shape)\n",
    "#单词词典\n",
    "# print(LABEL.vocab.stoi)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TextCNN(\n",
      "  (W): Embedding(15156, 60)\n",
      "  (conv): Sequential(\n",
      "    (0): Conv2d(1, 3, kernel_size=(2, 60), stride=(1, 1))\n",
      "    (1): ReLU()\n",
      "    (2): MaxPool2d(kernel_size=(49, 1), stride=(49, 1), padding=0, dilation=1, ceil_mode=False)\n",
      "  )\n",
      "  (fc): Linear(in_features=3, out_features=5, bias=True)\n",
      ")\n",
      "The model has 909,743 trainable parameters\n"
     ]
    }
   ],
   "source": [
    "embedding_size = 60\n",
    "sequence_length = FIX_LENGTH # every sentences contains sequence_length(=3) words\n",
    "num_classes = 5  # 0 or 1\n",
    "batch_size = 128\n",
    "\n",
    "vocab_size = len(TEXT.vocab)\n",
    "\n",
    "class TextCNN(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(TextCNN, self).__init__()\n",
    "        self.W = nn.Embedding(vocab_size, embedding_size)\n",
    "        output_channel = 3\n",
    "        self.conv = nn.Sequential(\n",
    "            # conv : [input_channel(=1), output_channel, (filter_height, filter_width), stride=1]\n",
    "            nn.Conv2d(1, output_channel, (2, embedding_size)),\n",
    "            nn.ReLU(),\n",
    "            # pool : ((filter_height, filter_width))\n",
    "            nn.MaxPool2d((FIX_LENGTH-1, 1)),\n",
    "        )\n",
    "        # fc\n",
    "        self.fc = nn.Linear(output_channel, num_classes)\n",
    "\n",
    "    def forward(self, X):\n",
    "      '''\n",
    "      X: [batch_size, sequence_length]\n",
    "      '''\n",
    "      batch_size = X.shape[0]\n",
    "    #   print(batch_size)\n",
    "      embedding_X = self.W(X) # [batch_size, sequence_length, embedding_size]\n",
    "    #   print(embedding_X.shape)\n",
    "      embedding_X = embedding_X.unsqueeze(1) # add channel(=1) [batch, channel(=1), sequence_length, embedding_size]\n",
    "    #   print(embedding_X.shape)\n",
    "      conved = self.conv(embedding_X) # [batch_size, output_channel, 1, 1]\n",
    "    #   print(conved.shape)\n",
    "      flatten = conved.view(batch_size, -1) # [batch_size, output_channel*1*1]\n",
    "    #   print(flatten.shape)\n",
    "      output = self.fc(flatten)\n",
    "      return output\n",
    "\n",
    "\n",
    "\n",
    "model = TextCNN()\n",
    "\n",
    "#定义优化器和损失\n",
    "optimizer = optim.Adam(model.parameters())\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "#定义度量指标\n",
    "def binary_accuracy(preds, y):\n",
    "\n",
    "    #round预测到最接近的整数\n",
    "    # rounded_preds = torch.round(preds)\n",
    "    correct = (preds == y).float() \n",
    "    acc = correct.sum() / len(correct)\n",
    "    return acc\n",
    "#转化为cuda（如果可用）\n",
    "model = model.to(DEVICE)\n",
    "criterion = criterion.to(DEVICE)\n",
    "print(model)\n",
    "#可训练参数的数量\n",
    "def count_parameters(model):\n",
    "    return sum(p.numel() for p in model.parameters() if p.requires_grad)\n",
    "print(f'The model has {count_parameters(model):,} trainable parameters')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([50, 128])\n",
      "torch.Size([128, 3, 49, 1])\n",
      "torch.Size([128, 3, 49, 1])\n",
      "torch.Size([128, 3, 1, 1])\n"
     ]
    }
   ],
   "source": [
    "# batch=next(iter(train_iter))\n",
    "# print(batch.Phrase.shape)\n",
    "# text = batch.Phrase\n",
    "# text = text.permute(1,0)\n",
    "# A=torch.ones([128,1,50,60])\n",
    "\n",
    "# cov=nn.Conv2d(1, 3, (2, embedding_size))\n",
    "# act=nn.ReLU()\n",
    "#             # pool : ((filter_height, filter_width))\n",
    "# pool=nn.MaxPool2d((49, 1))\n",
    "\n",
    "\n",
    "# out=cov(A)\n",
    "# print(out.shape)\n",
    "# out=act(out)\n",
    "# print(out.shape)\n",
    "# out=pool(out)\n",
    "# print(out.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([50, 128])\n",
      "torch.Size([128, 5])\n",
      "torch.Size([128])\n",
      "tensor(1.5005, device='cuda:0', grad_fn=<NllLossBackward>)\n",
      "tensor([2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,\n",
      "        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,\n",
      "        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,\n",
      "        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,\n",
      "        2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,\n",
      "        2, 2, 2, 2, 2, 2, 2, 2], device='cuda:0')\n",
      "tensor(0.5000, device='cuda:0')\n"
     ]
    }
   ],
   "source": [
    "batch=next(iter(train_iter))\n",
    "print(batch.Phrase.shape)\n",
    "text = batch.Phrase\n",
    "text = text.permute(1,0)\n",
    "predictions = model(text).squeeze()\n",
    "print(predictions.shape)\n",
    "print(batch.Sentiment.shape)\n",
    "\n",
    "loss = criterion(predictions, batch.Sentiment)\n",
    "        #计算二分类准确度\n",
    "print(loss)\n",
    "predictions_val=  predictions.argmax(dim=1)\n",
    "\n",
    "print(predictions_val)\n",
    "acc = binary_accuracy(predictions_val, batch.Sentiment.float())\n",
    "print(acc)\n",
    "        #后向传播损失并计算梯度\n",
    "loss.backward()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "def Train(model, iterator, optimizer, criterion):\n",
    "    #每个epoch进行初始化\n",
    "    epoch_loss = 0\n",
    "    epoch_acc = 0\n",
    "    #将模型设置为训练阶段\n",
    "    model.train()\n",
    "    predictions_val=[]\n",
    "    for batch in tqdm(iterator):\n",
    "        #重设梯度\n",
    "        optimizer.zero_grad()\n",
    "        #获取文本和单词数量\n",
    "        text = batch.Phrase\n",
    "        text = text.permute(1,0)\n",
    "        # print(text.shape)\n",
    "        #转换为一维张量\n",
    "        predictions = model(text).squeeze()\n",
    "        #计算loss\n",
    "        loss = criterion(predictions, batch.Sentiment)\n",
    "        #计算二分类准确度\n",
    "        predictions_val=  predictions.argmax(dim=1)\n",
    "\n",
    "\n",
    "        acc = binary_accuracy(predictions_val, batch.Sentiment.float())\n",
    "        #后向传播损失并计算梯度\n",
    "        loss.backward()\n",
    "        #更新权重\n",
    "        optimizer.step()\n",
    "        #损失和准确度\n",
    "        \n",
    "        epoch_loss += loss.item()  \n",
    "        epoch_acc += acc.item()\n",
    "    \n",
    "    return epoch_loss / len(iterator), epoch_acc / len(iterator)\n",
    "\n",
    "def Evaluate(model, iterator, criterion):\n",
    "    #每个epoch进行初始化\n",
    "    epoch_loss = 0\n",
    "    epoch_acc = 0\n",
    "    #停用dropout层\n",
    "    model.eval()\n",
    "    #停用自动求导\n",
    "    with torch.no_grad():\n",
    "        for batch in tqdm(iterator):\n",
    "            #获取文本和单词数量\n",
    "            text = batch.Phrase\n",
    "            text = text.permute(1,0)\n",
    "            #转换为一维张量\n",
    "            predictions = model(text).squeeze()\n",
    "            #计算损失和准确度\n",
    "            loss = criterion(predictions, batch.Sentiment)\n",
    "            predictions_val=   predictions.argmax(dim=1)\n",
    "\n",
    "            acc = binary_accuracy(predictions_val, batch.Sentiment.float())\n",
    "            #跟踪损失和准确度\n",
    "            epoch_loss += loss.item()\n",
    "            epoch_acc += acc.item()\n",
    "    return epoch_loss / len(iterator), epoch_acc / len(iterator)\n",
    "\n",
    "def Test(model, iterator, criterion):\n",
    "    #每个epoch进行初始化\n",
    "    epoch_loss = 0\n",
    "    epoch_acc = 0\n",
    "    #停用dropout层\n",
    "    model.eval()\n",
    "    predict_list=[]\n",
    "    #停用自动求导\n",
    "    with torch.no_grad():\n",
    "        for batch in tqdm(iterator):\n",
    "            #获取文本和单词数量\n",
    "            text = batch.Phrase\n",
    "            text = text.permute(1,0)\n",
    "            #转换为一维张量\n",
    "            predictions = model(text).squeeze()\n",
    "            #计算损失和准确度\n",
    "\n",
    "            predict=   predictions.argmax(dim=1).cpu()\n",
    "            predict_list+=predict.numpy().flatten().tolist()\n",
    "\n",
    "    return predict_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 226.82it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 529.16it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.258 | Train Acc: 51.37%\n",
      "\t Val. Loss: 1.210 |  Val. Acc: 52.70%\n",
      "epoch: 1\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 230.46it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 527.16it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.179 | Train Acc: 53.73%\n",
      "\t Val. Loss: 1.158 |  Val. Acc: 54.34%\n",
      "epoch: 2\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 227.64it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 502.18it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.127 | Train Acc: 55.45%\n",
      "\t Val. Loss: 1.129 |  Val. Acc: 55.56%\n",
      "epoch: 3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 219.68it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 532.46it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.100 | Train Acc: 56.33%\n",
      "\t Val. Loss: 1.115 |  Val. Acc: 56.00%\n",
      "epoch: 4\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 219.48it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 491.25it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.083 | Train Acc: 56.74%\n",
      "\t Val. Loss: 1.110 |  Val. Acc: 56.29%\n",
      "epoch: 5\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 219.45it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 463.67it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.072 | Train Acc: 57.22%\n",
      "\t Val. Loss: 1.105 |  Val. Acc: 56.39%\n",
      "epoch: 6\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 221.05it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 487.83it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.064 | Train Acc: 57.44%\n",
      "\t Val. Loss: 1.099 |  Val. Acc: 56.53%\n",
      "epoch: 7\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 210.65it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 512.11it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.055 | Train Acc: 57.72%\n",
      "\t Val. Loss: 1.098 |  Val. Acc: 56.22%\n",
      "epoch: 8\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 213.03it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 522.83it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.046 | Train Acc: 58.02%\n",
      "\t Val. Loss: 1.090 |  Val. Acc: 56.76%\n",
      "epoch: 9\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 221.70it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 515.11it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.036 | Train Acc: 58.39%\n",
      "\t Val. Loss: 1.082 |  Val. Acc: 57.21%\n",
      "epoch: 10\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 225.74it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 479.54it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.026 | Train Acc: 58.86%\n",
      "\t Val. Loss: 1.079 |  Val. Acc: 57.68%\n",
      "epoch: 11\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 232.77it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 529.19it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.019 | Train Acc: 59.30%\n",
      "\t Val. Loss: 1.075 |  Val. Acc: 57.24%\n",
      "epoch: 12\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 231.27it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 542.49it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.013 | Train Acc: 59.50%\n",
      "\t Val. Loss: 1.073 |  Val. Acc: 57.41%\n",
      "epoch: 13\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 230.25it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 525.05it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.008 | Train Acc: 59.75%\n",
      "\t Val. Loss: 1.064 |  Val. Acc: 57.80%\n",
      "epoch: 14\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 231.50it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 519.35it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 1.003 | Train Acc: 59.94%\n",
      "\t Val. Loss: 1.060 |  Val. Acc: 58.07%\n",
      "epoch: 15\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 229.46it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 531.54it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.997 | Train Acc: 60.13%\n",
      "\t Val. Loss: 1.054 |  Val. Acc: 58.32%\n",
      "epoch: 16\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 231.16it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 516.63it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.985 | Train Acc: 60.51%\n",
      "\t Val. Loss: 1.046 |  Val. Acc: 58.41%\n",
      "epoch: 17\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 220.44it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 525.59it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.974 | Train Acc: 60.97%\n",
      "\t Val. Loss: 1.036 |  Val. Acc: 59.01%\n",
      "epoch: 18\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 223.18it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 521.91it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.966 | Train Acc: 61.19%\n",
      "\t Val. Loss: 1.033 |  Val. Acc: 59.23%\n",
      "epoch: 19\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 229.31it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 512.67it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.960 | Train Acc: 61.39%\n",
      "\t Val. Loss: 1.031 |  Val. Acc: 59.27%\n",
      "epoch: 20\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 225.72it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 522.57it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.956 | Train Acc: 61.65%\n",
      "\t Val. Loss: 1.029 |  Val. Acc: 59.42%\n",
      "epoch: 21\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 230.41it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 515.21it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.951 | Train Acc: 61.74%\n",
      "\t Val. Loss: 1.026 |  Val. Acc: 59.62%\n",
      "epoch: 22\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 227.90it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 509.53it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.948 | Train Acc: 61.96%\n",
      "\t Val. Loss: 1.026 |  Val. Acc: 59.39%\n",
      "epoch: 23\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 229.10it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 530.44it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.945 | Train Acc: 62.06%\n",
      "\t Val. Loss: 1.025 |  Val. Acc: 59.56%\n",
      "epoch: 24\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 229.89it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 478.07it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.942 | Train Acc: 62.06%\n",
      "\t Val. Loss: 1.022 |  Val. Acc: 59.75%\n",
      "epoch: 25\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 234.33it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 518.73it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.940 | Train Acc: 62.25%\n",
      "\t Val. Loss: 1.022 |  Val. Acc: 59.84%\n",
      "epoch: 26\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 230.22it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 525.21it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.939 | Train Acc: 62.30%\n",
      "\t Val. Loss: 1.024 |  Val. Acc: 59.88%\n",
      "epoch: 27\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 233.65it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 518.56it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.937 | Train Acc: 62.41%\n",
      "\t Val. Loss: 1.023 |  Val. Acc: 59.72%\n",
      "epoch: 28\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 216.31it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 394.67it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.936 | Train Acc: 62.38%\n",
      "\t Val. Loss: 1.023 |  Val. Acc: 59.77%\n",
      "epoch: 29\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 232.84it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 525.43it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.934 | Train Acc: 62.44%\n",
      "\t Val. Loss: 1.028 |  Val. Acc: 59.37%\n",
      "epoch: 30\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 232.23it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 513.38it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.933 | Train Acc: 62.47%\n",
      "\t Val. Loss: 1.025 |  Val. Acc: 59.73%\n",
      "epoch: 31\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 232.12it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 530.75it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.932 | Train Acc: 62.51%\n",
      "\t Val. Loss: 1.021 |  Val. Acc: 59.84%\n",
      "epoch: 32\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 230.75it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 519.87it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.930 | Train Acc: 62.56%\n",
      "\t Val. Loss: 1.025 |  Val. Acc: 59.68%\n",
      "epoch: 33\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 217.83it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 490.68it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.929 | Train Acc: 62.58%\n",
      "\t Val. Loss: 1.023 |  Val. Acc: 59.61%\n",
      "epoch: 34\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 217.58it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 507.51it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.928 | Train Acc: 62.66%\n",
      "\t Val. Loss: 1.019 |  Val. Acc: 60.13%\n",
      "epoch: 35\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 226.07it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 435.05it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.927 | Train Acc: 62.67%\n",
      "\t Val. Loss: 1.020 |  Val. Acc: 59.87%\n",
      "epoch: 36\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 236.38it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 528.96it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.926 | Train Acc: 62.76%\n",
      "\t Val. Loss: 1.027 |  Val. Acc: 59.74%\n",
      "epoch: 37\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 237.82it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 524.45it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.925 | Train Acc: 62.74%\n",
      "\t Val. Loss: 1.022 |  Val. Acc: 60.06%\n",
      "epoch: 38\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 238.04it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 548.73it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.924 | Train Acc: 62.84%\n",
      "\t Val. Loss: 1.021 |  Val. Acc: 59.91%\n",
      "epoch: 39\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 240.57it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 555.92it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.923 | Train Acc: 62.79%\n",
      "\t Val. Loss: 1.023 |  Val. Acc: 59.87%\n",
      "epoch: 40\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:03<00:00, 244.16it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 561.12it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.922 | Train Acc: 62.90%\n",
      "\t Val. Loss: 1.026 |  Val. Acc: 59.70%\n",
      "epoch: 41\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 236.47it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 428.21it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.922 | Train Acc: 62.85%\n",
      "\t Val. Loss: 1.022 |  Val. Acc: 59.82%\n",
      "epoch: 42\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 222.51it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 528.66it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.921 | Train Acc: 62.95%\n",
      "\t Val. Loss: 1.021 |  Val. Acc: 60.08%\n",
      "epoch: 43\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 230.13it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 516.73it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.921 | Train Acc: 62.91%\n",
      "\t Val. Loss: 1.022 |  Val. Acc: 59.92%\n",
      "epoch: 44\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 230.83it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 528.99it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.920 | Train Acc: 62.88%\n",
      "\t Val. Loss: 1.028 |  Val. Acc: 59.44%\n",
      "epoch: 45\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 230.01it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 531.82it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.920 | Train Acc: 62.94%\n",
      "\t Val. Loss: 1.023 |  Val. Acc: 59.96%\n",
      "epoch: 46\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 223.46it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 431.64it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.919 | Train Acc: 62.96%\n",
      "\t Val. Loss: 1.024 |  Val. Acc: 59.73%\n",
      "epoch: 47\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 231.84it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 514.47it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.918 | Train Acc: 62.98%\n",
      "\t Val. Loss: 1.022 |  Val. Acc: 59.93%\n",
      "epoch: 48\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 230.80it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 534.02it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.918 | Train Acc: 62.99%\n",
      "\t Val. Loss: 1.024 |  Val. Acc: 60.05%\n",
      "epoch: 49\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 976/976 [00:04<00:00, 233.93it/s]\n",
      "100%|██████████| 244/244 [00:00<00:00, 519.74it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tTrain Loss: 0.917 | Train Acc: 63.04%\n",
      "\t Val. Loss: 1.019 |  Val. Acc: 60.17%\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "N_EPOCHS = 50\n",
    "best_valid_loss = float('inf')\n",
    "\n",
    "train_loss_list=[]\n",
    "valid_loss_list=[]\n",
    "train_acc_list=[]\n",
    "valid_acc_list=[]\n",
    "\n",
    "for epoch in range(N_EPOCHS):\n",
    "    #训练模型\n",
    "    print('epoch:',epoch)\n",
    "    train_loss, train_acc = Train(model, train_iter, optimizer, criterion)\n",
    "    #评估模型\n",
    "    valid_loss, valid_acc = Evaluate(model, val_iter, criterion)\n",
    "    #保存模型\n",
    "    if valid_loss < best_valid_loss:\n",
    "        best_valid_loss = valid_loss\n",
    "        torch.save(model.state_dict(), 'saved_weights.pt')\n",
    "\n",
    "    train_loss_list.append(train_loss)\n",
    "    valid_loss_list.append(valid_loss)\n",
    "    train_acc_list.append(train_acc)\n",
    "    valid_acc_list.append(valid_acc)\n",
    "\n",
    "    print(f'\\tTrain Loss: {train_loss:.3f} | Train Acc: {train_acc*100:.2f}%')\n",
    "    print(f'\\t Val. Loss: {valid_loss:.3f} |  Val. Acc: {valid_acc*100:.2f}%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "![](./pic/textCNN-random-loss-50.jpg)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "![](./pic/textCNN-random-acc-50.jpg)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "\n",
    "def drawpic(train_loss_list=[],test_loss_list=[],epoch_number=10,title='1',root_path='./'):\n",
    "    # make data\n",
    "    x = [i for i in range(epoch_number)]\n",
    "    # plot\n",
    "    fig, ax = plt.subplots()\n",
    "    plt.title(title)\n",
    "    ax.plot(x, train_loss_list, linewidth=2.0)\n",
    "    ax.plot(x, test_loss_list, linewidth=2.0)\n",
    "    path=root_path+title+'.jpg'\n",
    "    print('![]({})'.format(path))\n",
    "    plt.savefig(path)\n",
    "    plt.show()\n",
    "root_path='./pic/'\n",
    "title='textCNN-random-loss-{}'.format(str(N_EPOCHS))\n",
    "\n",
    "drawpic(train_loss_list=train_loss_list,test_loss_list=valid_loss_list,epoch_number=N_EPOCHS,title=title,root_path=root_path)\n",
    "\n",
    "title='textCNN-random-acc-{}'.format(str(N_EPOCHS))\n",
    "\n",
    "drawpic(train_loss_list=train_acc_list,test_loss_list=valid_acc_list,epoch_number=N_EPOCHS,title=title,root_path=root_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 518/518 [00:01<00:00, 280.20it/s]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "66292"
      ]
     },
     "execution_count": 104,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predict=Test(model, test_iter, criterion)\n",
    "len(predict)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "df_test=pd.read_csv('./data/sampleSubmission.csv')\n",
    "df_test['Sentiment']=predict\n",
    "df_test.to_csv('./submission.csv',index=None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "ename": "ValueError",
     "evalue": "only one element tensors can be converted to Python scalars",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "\u001b[1;32m~\\AppData\\Local\\Temp/ipykernel_34604/88635740.py\u001b[0m in \u001b[0;36m<module>\u001b[1;34m\u001b[0m\n\u001b[0;32m     17\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m     18\u001b[0m \u001b[1;31m#进行预测\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m---> 19\u001b[1;33m \u001b[0mprint\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mpredict\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mmodel\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;34m\"fuck fuck shit bitch?\"\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m     20\u001b[0m \u001b[1;31m#不真诚的问题\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m     21\u001b[0m \u001b[0mprint\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mpredict\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mmodel\u001b[0m\u001b[1;33m,\u001b[0m \u001b[1;34m\"Why Indian girls go crazy about marrying Shri. Rahul Gandhi ji?\"\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32m~\\AppData\\Local\\Temp/ipykernel_34604/88635740.py\u001b[0m in \u001b[0;36mpredict\u001b[1;34m(model, text)\u001b[0m\n\u001b[0;32m     14\u001b[0m     \u001b[0mlength_tensor\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mtorch\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mLongTensor\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mlength\u001b[0m\u001b[1;33m)\u001b[0m                   \u001b[1;31m#转换为张量\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m     15\u001b[0m     \u001b[0mprediction\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mtensor\u001b[0m\u001b[1;33m)\u001b[0m                  \u001b[1;31m#预测\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m---> 16\u001b[1;33m     \u001b[1;32mreturn\u001b[0m \u001b[0mprediction\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mitem\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m     17\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m     18\u001b[0m \u001b[1;31m#进行预测\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;31mValueError\u001b[0m: only one element tensors can be converted to Python scalars"
     ]
    }
   ],
   "source": [
    "#加载权重\n",
    "path='./saved_weights.pt'\n",
    "model.load_state_dict(torch.load(path))\n",
    "model.eval()\n",
    "#推断 \n",
    "\n",
    "def predict(model, text):\n",
    "    tokenized = [wd for wd in jieba.cut(text, cut_all=False)]\n",
    "  #令牌化(tokenize)句子 \n",
    "    indexed = [TEXT.vocab.stoi[t] for t in tokenized]          #转换为整数序列\n",
    "    length = [len(indexed)]                                    #计算单词个数\n",
    "    tensor = torch.LongTensor(indexed).to(DEVICE)              #转换为张量\n",
    "    tensor = tensor.unsqueeze(1).T                             #reshape成[batch, 单词个数]\n",
    "    length_tensor = torch.LongTensor(length)                   #转换为张量\n",
    "    prediction = model(tensor)                  #预测\n",
    "    return prediction.item()\n",
    "\n",
    "#进行预测\n",
    "print(predict(model, \"fuck fuck shit bitch?\"))\n",
    "#不真诚的问题\n",
    "print(predict(model, \"Why Indian girls go crazy about marrying Shri. Rahul Gandhi ji?\"))"
   ]
  }
 ],
 "metadata": {
  "interpreter": {
   "hash": "b3ba2566441a7c06988d0923437866b63cedc61552a5af99d1f4fb67d367b25f"
  },
  "kernelspec": {
   "display_name": "Python 3.9.7 64-bit ('base': conda)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.7"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}