lstm_pytorch_sentimenti_classification.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "view-in-github"
   },
   "source": [
    "<a href=\"https://colab.research.google.com/github/Iallen520/lhy_DL_Hw/blob/master/hw4_RNN.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "9YrAlczfM_w6"
   },
   "source": [
    "### Download Dataset\n",
    "有三個檔案，分別是training_label.txt、training_nolabel.txt、testing_data.txt\n",
    "\n",
    "training_label.txt：有label的training data(句子配上0 or 1)\n",
    "\n",
    "training_nolabel.txt：沒有label的training data(只有句子)，用來做semi-supervise learning\n",
    "\n",
    "testing_data.txt：你要判斷testing data裡面的句子是0 or 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "8hDIokoP6464"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "b'Skipping line 43043: expected 2 fields, saw 3\\n'\n"
     ]
    },
    {
     "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>review</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>9999_0</td>\n",
       "      <td>Watching Time Chasers, it obvious that it was ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>45057_0</td>\n",
       "      <td>I saw this film about 20 years ago and remembe...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>15561_0</td>\n",
       "      <td>Minor Spoilers&lt;br /&gt;&lt;br /&gt;In New York, Joan Ba...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>7161_0</td>\n",
       "      <td>I went to see this film with a great deal of e...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>43971_0</td>\n",
       "      <td>Yes, I agree with everyone on this site this m...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49993</th>\n",
       "      <td>18984_0</td>\n",
       "      <td>The original Man Eater by Joe D'Amato is somew...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49994</th>\n",
       "      <td>16433_0</td>\n",
       "      <td>When Home Box Office was in it's early days mo...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49995</th>\n",
       "      <td>16006_0</td>\n",
       "      <td>Griffin Dunne was born into a cultural family....</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49996</th>\n",
       "      <td>40155_0</td>\n",
       "      <td>Not a bad story, but the low budget rears its ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49997</th>\n",
       "      <td>35270_0</td>\n",
       "      <td>This not-very-good mummy-alien flick does feat...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>49998 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "            id                                             review\n",
       "0       9999_0  Watching Time Chasers, it obvious that it was ...\n",
       "1      45057_0  I saw this film about 20 years ago and remembe...\n",
       "2      15561_0  Minor Spoilers<br /><br />In New York, Joan Ba...\n",
       "3       7161_0  I went to see this film with a great deal of e...\n",
       "4      43971_0  Yes, I agree with everyone on this site this m...\n",
       "...        ...                                                ...\n",
       "49993  18984_0  The original Man Eater by Joe D'Amato is somew...\n",
       "49994  16433_0  When Home Box Office was in it's early days mo...\n",
       "49995  16006_0  Griffin Dunne was born into a cultural family....\n",
       "49996  40155_0  Not a bad story, but the low budget rears its ...\n",
       "49997  35270_0  This not-very-good mummy-alien flick does feat...\n",
       "\n",
       "[49998 rows x 2 columns]"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# this is for filtering the warnings\n",
    "\n",
    "#import warnings\n",
    "#warnings.filterwarnings('ignore')\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "\n",
    "df = pd.read_csv(\"unlabeledTrainData.tsv\",sep = \"\\t\",error_bad_lines=False)\n",
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "fc143hSvNGr6"
   },
   "source": [
    "### Utils"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ICDIhhgCY2-M"
   },
   "outputs": [],
   "source": [
    "# utils.py\n",
    "# 這個block用來先定義一些等等常用到的函式\n",
    "import torch\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import torch.optim as optim\n",
    "import torch.nn.functional as F\n",
    "\n",
    "\n",
    "label_train_path = \"labeledTrainData.tsv\"\n",
    "unlabel_train_path = \"unlabeledTrainData.tsv\"\n",
    "test_path = \"testData.tsv\"\n",
    "\n",
    "def load_training_data(path='labeledTrainData.tsv'):  \n",
    "    #global x,y\n",
    "    # 把training時需要的data讀進來\n",
    "    # 如果是'training_labe|l.txt'，需要讀取label，如果是'training_nolabel.txt'，不需要讀取label\n",
    "    if(path == \"labeledTrainData.tsv\"):\n",
    "        df = pd.read_csv(\"labeledTrainData.tsv\",sep = \"\\t\")\n",
    "        #display(df.head())\n",
    "        x = df['review'].apply(lambda x: x.strip().split(\" \"))\n",
    "        x = list(x)\n",
    "        y = df[\"sentiment\"]  #二维的list\n",
    "        y = list(y)   #一维的list\n",
    "        return x, y\n",
    "    else:\n",
    "        df = pd.read_csv(\"unlabeledTrainData.tsv\",sep = \"\\t\",error_bad_lines=False)\n",
    "        x = df['review'].apply(lambda x: x.strip().split(\" \"))\n",
    "        x = list(x)\n",
    "        return x\n",
    "    \n",
    "\n",
    "def load_testing_data(path='testData.tsv'):   #约55万\n",
    "    # 把testing時需要的data讀進來\n",
    "    df = pd.read_csv(\"testData.tsv\",sep = \"\\t\")\n",
    "    #display(df.head())\n",
    "    x = df['review'].apply(lambda x: x.strip().split(\" \"))\n",
    "    x = list(x)\n",
    "    return x  # X形如[[\"i\",\"am\",\"here\"], [\"he\",\"loves\",\"it\"]]\n",
    "\n",
    "\n",
    "def evaluation(outputs, labels):\n",
    "    #outputs => probability (float)\n",
    "    #labels => labels\n",
    "    outputs[outputs>=0.5] = 1 # 大於等於0.5為有惡意\n",
    "    outputs[outputs<0.5] = 0 # 小於0.5為無惡意\n",
    "    correct = torch.sum(torch.eq(outputs, labels)).item()\n",
    "    return correct\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "oYE8UYQsNIxM"
   },
   "source": [
    "### Train Word to Vector"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "训练集平均句子长度： 233.78624\n"
     ]
    }
   ],
   "source": [
    "train_x, y = load_training_data(label_train_path)\n",
    "average_sentence_length = 0\n",
    "for sen in train_x:\n",
    "    average_sentence_length += len(sen)\n",
    "average_sentence_length /= len(train_x)\n",
    "print(\"训练集平均句子长度：\",average_sentence_length)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 122
    },
    "colab_type": "code",
    "id": "cgGWaF8_2S3q",
    "outputId": "d25efb6b-de4f-43f2-8a98-e0aaa47f704e"
   },
   "outputs": [],
   "source": [
    "# w2v.py\n",
    "# 這個block是用來訓練word to vector 的 word embedding\n",
    "# 注意！這個block在訓練word to vector時是用cpu，可能要花到10分鐘以上\n",
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import argparse\n",
    "from gensim.models import word2vec\n",
    "\n",
    "path_prefix = \"./\"\n",
    "\n",
    "def train_word2vec(x):\n",
    "    # 訓練word to vector 的 word embedding\n",
    "    model = word2vec.Word2Vec(x, size=250, window=5, min_count=5, workers=12, iter=10, sg=1)\n",
    "    return model\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    l = os.listdir()\n",
    "    if \"w2v_all.model\" not in l:   \n",
    "        print(\"loading training data ...\")\n",
    "        train_x, y = load_training_data(label_train_path)\n",
    "        train_x_no_label = load_training_data(unlabel_train_path)\n",
    "\n",
    "        print(\"loading testing data ...\")\n",
    "        test_x = load_testing_data(test_path)\n",
    "\n",
    "        model = train_word2vec(train_x + train_x_no_label + test_x)\n",
    "\n",
    "        print(\"saving model ...\")\n",
    "        # model.save(os.path.join(path_prefix, 'model/w2v_all.model'))\n",
    "        model.save(os.path.join(path_prefix, 'w2v_all.model'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "3wHLtS0wNR6w"
   },
   "source": [
    "### Data Preprocess"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "CfGKiOitk5ob"
   },
   "outputs": [],
   "source": [
    "# preprocess.py\n",
    "# 這個block用來做data的預處理\n",
    "from torch import nn\n",
    "from gensim.models import Word2Vec\n",
    "\n",
    "class Preprocess():\n",
    "    def __init__(self, sentences, sen_len, w2v_path=\"./w2v.model\"):\n",
    "        self.w2v_path = w2v_path\n",
    "        self.sentences = sentences\n",
    "        self.sen_len = sen_len\n",
    "        self.idx2word = []\n",
    "        self.word2idx = {}\n",
    "        self.embedding_matrix = []\n",
    "    def get_w2v_model(self):\n",
    "        # 把之前訓練好的word to vec 模型讀進來\n",
    "        self.embedding = Word2Vec.load(self.w2v_path)\n",
    "        self.embedding_dim = self.embedding.vector_size\n",
    "    def add_embedding(self, word):\n",
    "        # 把word加進embedding，並賦予他一個隨機生成的representation vector\n",
    "        # word只會是\"<PAD>\"或\"<UNK>\"\n",
    "        vector = torch.empty(1, self.embedding_dim)\n",
    "        torch.nn.init.uniform_(vector)\n",
    "        self.word2idx[word] = len(self.word2idx)\n",
    "        self.idx2word.append(word)\n",
    "        self.embedding_matrix = torch.cat([self.embedding_matrix, vector], 0)\n",
    "    def make_embedding(self, load=True):\n",
    "        print(\"Get embedding ...\")\n",
    "        # 取得訓練好的 Word2vec word embedding\n",
    "        if load:\n",
    "            print(\"loading word to vec model ...\")\n",
    "            self.get_w2v_model()\n",
    "        else:\n",
    "            raise NotImplementedError\n",
    "        # 製作一個 word2idx 的 dictionary\n",
    "        # 製作一個 idx2word 的 list\n",
    "        # 製作一個 word2vector 的 list\n",
    "        for i, word in enumerate(self.embedding.wv.vocab):\n",
    "            print('get words #{}'.format(i+1), end='\\r')\n",
    "            #e.g. self.word2index['魯'] = 1 \n",
    "            #e.g. self.index2word[1] = '魯'\n",
    "            #e.g. self.vectors[1] = '魯' vector\n",
    "            self.word2idx[word] = len(self.word2idx)\n",
    "            self.idx2word.append(word)\n",
    "            self.embedding_matrix.append(self.embedding[word])\n",
    "        print('')\n",
    "        self.embedding_matrix = torch.tensor(self.embedding_matrix)\n",
    "        # 將\"<PAD>\"跟\"<UNK>\"加進embedding裡面\n",
    "        self.add_embedding(\"<PAD>\")\n",
    "        self.add_embedding(\"<UNK>\")\n",
    "        print(\"total words: {}\".format(len(self.embedding_matrix)))\n",
    "        return self.embedding_matrix\n",
    "    def pad_sequence(self, sentence):\n",
    "        # 將每個句子變成一樣的長度\n",
    "        if len(sentence) > self.sen_len:\n",
    "            sentence = sentence[:self.sen_len]\n",
    "        else:\n",
    "            pad_len = self.sen_len - len(sentence)\n",
    "            for _ in range(pad_len):\n",
    "                sentence.append(self.word2idx[\"<PAD>\"])\n",
    "        assert len(sentence) == self.sen_len\n",
    "        return sentence\n",
    "    def sentence_word2idx(self):\n",
    "        # 把句子裡面的字轉成相對應的index\n",
    "        sentence_list = []\n",
    "        for i, sen in enumerate(self.sentences):\n",
    "            print('sentence count #{}'.format(i+1), end='\\r')\n",
    "            sentence_idx = []\n",
    "            for word in sen:\n",
    "                if (word in self.word2idx.keys()):\n",
    "                    sentence_idx.append(self.word2idx[word])\n",
    "                else:\n",
    "                    sentence_idx.append(self.word2idx[\"<UNK>\"])\n",
    "            # 將每個句子變成一樣的長度\n",
    "            sentence_idx = self.pad_sequence(sentence_idx)\n",
    "            sentence_list.append(sentence_idx)\n",
    "        return torch.LongTensor(sentence_list)\n",
    "    def labels_to_tensor(self, y):\n",
    "        # 把labels轉成tensor\n",
    "        y = [int(label) for label in y]\n",
    "        \n",
    "        return torch.LongTensor(y)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "3WJB7go5NWL0"
   },
   "source": [
    "### Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "XketwKs4lFfB"
   },
   "outputs": [],
   "source": [
    "# data.py\n",
    "# 實作了dataset所需要的'__init__', '__getitem__', '__len__'\n",
    "# 好讓dataloader能使用\n",
    "import torch\n",
    "from torch.utils import data\n",
    "\n",
    "class TwitterDataset(data.Dataset):\n",
    "    \"\"\"\n",
    "    Expected data shape like:(data_num, data_len)\n",
    "    Data can be a list of numpy array or a list of lists\n",
    "    input data shape : (data_num, seq_len, feature_dim)\n",
    "    \n",
    "    __len__ will return the number of data\n",
    "    \"\"\"\n",
    "    def __init__(self, X, y):\n",
    "        self.data = X\n",
    "        self.label = y\n",
    "    def __getitem__(self, idx):\n",
    "        if self.label is None: return self.data[idx]\n",
    "        return self.data[idx], self.label[idx]\n",
    "    def __len__(self):\n",
    "        return len(self.data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "uNJ8xWIMNa2r"
   },
   "source": [
    "### Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 148,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ZS6RJADulIq1"
   },
   "outputs": [],
   "source": [
    "# model.py\n",
    "# 這個block是要拿來訓練的模型\n",
    "import torch\n",
    "from torch import nn\n",
    "class LSTM_Net(nn.Module):\n",
    "    def __init__(self, embedding, embedding_dim, hidden_dim, num_layers, dropout=0.5, fix_embedding=True):\n",
    "        super(LSTM_Net, self).__init__()\n",
    "        # 製作 embedding layer\n",
    "        self.embedding = torch.nn.Embedding(embedding.size(0),embedding.size(1))\n",
    "        self.embedding.weight = torch.nn.Parameter(embedding)\n",
    "        # 是否將 embedding fix住，如果fix_embedding為False，在訓練過程中，embedding也會跟著被訓練\n",
    "        self.embedding.weight.requires_grad = False if fix_embedding else True\n",
    "        self.embedding_dim = embedding.size(1)\n",
    "        self.hidden_dim = hidden_dim\n",
    "        self.num_layers = num_layers\n",
    "        self.dropout = dropout\n",
    "        self.lstm = nn.LSTM(embedding_dim, hidden_dim, num_layers=num_layers, batch_first=True)\n",
    "        self.classifier = nn.Sequential( nn.Dropout(dropout),\n",
    "                                         nn.Linear(hidden_dim, 1),\n",
    "                                         nn.Sigmoid() )\n",
    "    def forward(self, inputs):\n",
    "        inputs = self.embedding(inputs)\n",
    "        x, _ = self.lstm(inputs, None)\n",
    "        # x 的 dimension (batch, seq_len, hidden_size)\n",
    "        # 取用 LSTM 最後一層的 hidden state\n",
    "        x = x[:, -1, :] \n",
    "        x = self.classifier(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "aWlpEL0sNc10"
   },
   "source": [
    "### Train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 155,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "4QR4MMz-lR7i"
   },
   "outputs": [],
   "source": [
    "# train.py\n",
    "# 這個block是用來訓練模型的\n",
    "%matplotlib inline\n",
    "\n",
    "import torch\n",
    "from torch import nn\n",
    "import torch.optim as optim\n",
    "import torch.nn.functional as F\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import sys\n",
    "\n",
    "yy_train_loss=[]\n",
    "yy_valid_loss = []\n",
    "yy_train_acc = []\n",
    "yy_valid_acc = []\n",
    "xx = []\n",
    "early_stopping_epoch = 30\n",
    "add_num = 0\n",
    "total_loss, total_acc = 0,0\n",
    "best_acc = 0\n",
    "pre_valid_loss = 100000\n",
    "\n",
    "def training(batch_size, n_epoch, lr, model_dir, train, valid, model, device):\n",
    "    global add_num,total_loss,total_acc,best_acc,pre_valid_loss\n",
    "    \n",
    "    total = sum(p.numel() for p in model.parameters())\n",
    "    trainable = sum(p.numel() for p in model.parameters() if p.requires_grad)\n",
    "    print('\\nstart training, parameter total:{}, trainable:{}\\n'.format(total, trainable))\n",
    "    model.train() # 將model的模式設為train，這樣optimizer就可以更新model的參數\n",
    "    criterion = nn.BCELoss() # 定義損失函數，這裡我們使用binary cross entropy loss\n",
    "    t_batch = len(train) \n",
    "    v_batch = len(valid) \n",
    "    optimizer = optim.Adam(model.parameters(), lr=lr)\n",
    "    #optimizer = optim.SGD(model.parameters(), lr=lr, momentum = 0.9) # 將模型的參數給optimizer，並給予適當的learning rate\n",
    "    total_loss, total_acc, best_acc = 0, 0, 0\n",
    "    for epoch in range(n_epoch):\n",
    "        xx.append(epoch)\n",
    "        \n",
    "        total_loss, total_acc = 0, 0\n",
    "        # 這段做training\n",
    "        for i, (inputs, labels) in enumerate(train):\n",
    "            inputs = inputs.to(device, dtype=torch.long) # device為\"cuda\"，將inputs轉成torch.cuda.LongTensor\n",
    "            labels = labels.to(device, dtype=torch.float) # device為\"cuda\"，將labels轉成torch.cuda.FloatTensor，因為等等要餵進criterion，所以型態要是float\n",
    "            optimizer.zero_grad() # 由於loss.backward()的gradient會累加，所以每次餵完一個batch後需要歸零\n",
    "            outputs = model(inputs) # 將input餵給模型\n",
    "            outputs = outputs.squeeze() # 去掉最外面的dimension，好讓outputs可以餵進criterion()\n",
    "            loss = criterion(outputs, labels) # 計算此時模型的training loss\n",
    "            loss.backward() # 算loss的gradient\n",
    "            optimizer.step() # 更新訓練模型的參數\n",
    "            correct = evaluation(outputs, labels) # 計算此時模型的training accuracy\n",
    "            total_acc += (correct / batch_size)\n",
    "            total_loss += loss.item()\n",
    "            print('[ Epoch{}: {}/{} ] loss:{:.3f} acc:{:.3f} '.format(\n",
    "            \tepoch+1, i+1, t_batch, loss.item(), correct*100/batch_size), end='\\r')\n",
    "        print('\\nTrain | Loss:{:.5f} Acc: {:.3f}'.format(total_loss/t_batch, total_acc/t_batch*100))\n",
    "        yy_train_loss.append(total_loss/t_batch)\n",
    "        yy_train_acc.append(total_acc/t_batch*100)\n",
    "        \n",
    "        \n",
    "        # 這段做validation\n",
    "        model.eval() # 將model的模式設為eval，這樣model的參數就會固定住\n",
    "        with torch.no_grad():\n",
    "            total_loss, total_acc = 0, 0\n",
    "            for i, (inputs, labels) in enumerate(valid):\n",
    "                inputs = inputs.to(device, dtype=torch.long) # device為\"cuda\"，將inputs轉成torch.cuda.LongTensor\n",
    "                labels = labels.to(device, dtype=torch.float) # device為\"cuda\"，將labels轉成torch.cuda.FloatTensor，因為等等要餵進criterion，所以型態要是float\n",
    "                outputs = model(inputs) # 將input餵給模型\n",
    "                outputs = outputs.squeeze() # 去掉最外面的dimension，好讓outputs可以餵進criterion()\n",
    "                loss = criterion(outputs, labels) # 計算此時模型的validation loss\n",
    "                correct = evaluation(outputs, labels) # 計算此時模型的validation accuracy\n",
    "                total_acc += (correct / batch_size)\n",
    "                total_loss += loss.item()\n",
    "\n",
    "            print(\"Valid | Loss:{:.5f} Acc: {:.3f} \".format(total_loss/v_batch, total_acc/v_batch*100))\n",
    "            yy_valid_loss.append(total_loss/v_batch)\n",
    "            yy_valid_acc.append(total_acc/v_batch*100)\n",
    "            \n",
    "            # early stopping,连续增长到early_stopping_epoch轮数的话，就会提前终止epoch循环,epoch = sys.maxsize\n",
    "            curr_valid_loss = total_loss/v_batch\n",
    "            \n",
    "            if(curr_valid_loss > pre_valid_loss):\n",
    "                add_num += 1\n",
    "                if(add_num == early_stopping_epoch):\n",
    "                    epoch = sys.maxsize\n",
    "            else:\n",
    "                add_num = 0\n",
    "                \n",
    "            pre_valid_loss = curr_valid_loss\n",
    "            \n",
    "                        \n",
    "            if total_acc/v_batch > best_acc:\n",
    "                print(\"total_acc/v_batch:\",total_acc/v_batch)\n",
    "                \n",
    "                # 如果validation的結果優於之前所有的結果，就把當下的模型存下來以備之後做預測時使用\n",
    "                best_acc = total_acc/v_batch\n",
    "                \n",
    "                \n",
    "                #torch.save(model, \"{}/val_acc_{:.3f}.model\".format(model_dir,total_acc/v_batch*100))\n",
    "                torch.save(model, \"{}/ckpt.model\".format(model_dir))\n",
    "                print('saving model with acc {:.3f}'.format(total_acc/v_batch*100))\n",
    "        print('-----------------------------------------------')\n",
    "        model.train() # 將model的模式設為train，這樣optimizer就可以更新model的參數（因為剛剛轉成eval模式）\n",
    "    #结束for循环\n",
    "    \n",
    "    #将最好的valid的best_acc存放下来\n",
    "    tt_path = str(best_acc) +\"_best_acc.txt\"\n",
    "    with open(tt_path,\"w\") as f:\n",
    "        f.write(str(best_acc))\n",
    "    \n",
    "    \n",
    "    ax1 = plt.subplot(1,2,1)\n",
    "    plt.sca(ax1)\n",
    "    plt.plot(xx,yy_train_loss,\"r\",label = \"train_loss\")\n",
    "    plt.plot(xx,yy_valid_loss,\"b\",label = \"valid_loss\")\n",
    "    plt.legend()\n",
    "    ax2 = plt.subplot(1,2,2)\n",
    "    plt.sca(ax2)\n",
    "    plt.plot(xx,yy_train_acc,\"y\",label = \"train_acc\")\n",
    "    plt.plot(xx,yy_valid_acc,\"g\",label = \"valid_acc\")\n",
    "    plt.legend()\n",
    "    \n",
    "    print(\"over\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "qF5YQrupNfCS"
   },
   "source": [
    "### Test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 156,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "2X2wkdAYxHYA"
   },
   "outputs": [],
   "source": [
    "# test.py\n",
    "# 這個block用來對testing_data.txt做預測\n",
    "import torch\n",
    "from torch import nn\n",
    "import torch.optim as optim\n",
    "import torch.nn.functional as F\n",
    "\n",
    "def testing(batch_size, test_loader, model, device):\n",
    "    model.eval()\n",
    "    ret_output = []\n",
    "    with torch.no_grad():\n",
    "        for i, inputs in enumerate(test_loader):\n",
    "            inputs = inputs.to(device, dtype=torch.long)\n",
    "            outputs = model(inputs)\n",
    "            outputs = outputs.squeeze()\n",
    "            #outputs[outputs>=0.5] = 1 # 大於等於0.5為負面\n",
    "            #outputs[outputs<0.5] = 0 # 小於0.5為正面\n",
    "            ret_output += outputs.float().tolist()\n",
    "    \n",
    "    return ret_output"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "dfnKj0KXNeoz"
   },
   "source": [
    "### Main"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 157,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 578
    },
    "colab_type": "code",
    "id": "EztIWqCmlZof",
    "outputId": "961deee1-4726-445b-cd49-ab4fcc0e14cf"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "loading data ...\n",
      "Get embedding ...\n",
      "loading word to vec model ...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/u2019101432/.conda/envs/pytorch/lib/python3.7/site-packages/ipykernel_launcher.py:44: DeprecationWarning: Call to deprecated `__getitem__` (Method will be removed in 4.0.0, use self.wv.__getitem__() instead).\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "get words #121343\n",
      "total words: 121345\n",
      "sentence count #25000\n",
      "start training, parameter total:30838501, trainable:502251\n",
      "\n",
      "[ Epoch1: 157/157 ] loss:0.667 acc:15.625 \n",
      "Train | Loss:0.69236 Acc: 51.752\n",
      "Valid | Loss:0.68167 Acc: 53.867 \n",
      "total_acc/v_batch: 0.538671875\n",
      "saving model with acc 53.867\n",
      "-----------------------------------------------\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/u2019101432/.conda/envs/pytorch/lib/python3.7/site-packages/torch/serialization.py:292: UserWarning: Couldn't retrieve source code for container of type LSTM_Net. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n",
      "/home/u2019101432/.conda/envs/pytorch/lib/python3.7/site-packages/torch/serialization.py:292: UserWarning: Couldn't retrieve source code for container of type Embedding. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n",
      "/home/u2019101432/.conda/envs/pytorch/lib/python3.7/site-packages/torch/serialization.py:292: UserWarning: Couldn't retrieve source code for container of type LSTM. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n",
      "/home/u2019101432/.conda/envs/pytorch/lib/python3.7/site-packages/torch/serialization.py:292: UserWarning: Couldn't retrieve source code for container of type Sequential. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n",
      "/home/u2019101432/.conda/envs/pytorch/lib/python3.7/site-packages/torch/serialization.py:292: UserWarning: Couldn't retrieve source code for container of type Dropout. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n",
      "/home/u2019101432/.conda/envs/pytorch/lib/python3.7/site-packages/torch/serialization.py:292: UserWarning: Couldn't retrieve source code for container of type Linear. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n",
      "/home/u2019101432/.conda/envs/pytorch/lib/python3.7/site-packages/torch/serialization.py:292: UserWarning: Couldn't retrieve source code for container of type Sigmoid. It won't be checked for correctness upon loading.\n",
      "  \"type \" + obj.__name__ + \". It won't be checked \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ Epoch2: 157/157 ] loss:0.671 acc:15.625 \n",
      "Train | Loss:0.69209 Acc: 52.135\n",
      "Valid | Loss:0.67559 Acc: 57.363 \n",
      "total_acc/v_batch: 0.5736328125\n",
      "saving model with acc 57.363\n",
      "-----------------------------------------------\n",
      "[ Epoch3: 157/157 ] loss:0.643 acc:15.625 \n",
      "Train | Loss:0.67353 Acc: 57.016\n",
      "Valid | Loss:0.68321 Acc: 53.164 \n",
      "-----------------------------------------------\n",
      "[ Epoch4: 157/157 ] loss:0.676 acc:14.844 \n",
      "Train | Loss:0.68160 Acc: 56.240\n",
      "Valid | Loss:0.69029 Acc: 50.273 \n",
      "-----------------------------------------------\n",
      "[ Epoch5: 157/157 ] loss:0.540 acc:19.531 \n",
      "Train | Loss:0.65946 Acc: 61.709\n",
      "Valid | Loss:0.56751 Acc: 77.656 \n",
      "total_acc/v_batch: 0.7765625\n",
      "saving model with acc 77.656\n",
      "-----------------------------------------------\n",
      "[ Epoch6: 157/157 ] loss:0.445 acc:22.656 \n",
      "Train | Loss:0.66218 Acc: 59.395\n",
      "Valid | Loss:0.53424 Acc: 77.051 \n",
      "-----------------------------------------------\n",
      "[ Epoch7: 157/157 ] loss:0.749 acc:7.812 \n",
      "Train | Loss:0.66892 Acc: 57.738\n",
      "Valid | Loss:0.68883 Acc: 57.090 \n",
      "-----------------------------------------------\n",
      "[ Epoch8: 157/157 ] loss:0.846 acc:15.625 \n",
      "Train | Loss:0.66816 Acc: 58.604\n",
      "Valid | Loss:0.81171 Acc: 57.461 \n",
      "-----------------------------------------------\n",
      "[ Epoch9: 157/157 ] loss:0.984 acc:7.812 \n",
      "Train | Loss:0.67099 Acc: 59.425\n",
      "Valid | Loss:0.71524 Acc: 53.516 \n",
      "-----------------------------------------------\n",
      "[ Epoch10: 157/157 ] loss:0.551 acc:20.312 \n",
      "Train | Loss:0.61158 Acc: 66.561\n",
      "Valid | Loss:0.44345 Acc: 80.371 \n",
      "total_acc/v_batch: 0.8037109375\n",
      "saving model with acc 80.371\n",
      "-----------------------------------------------\n",
      "[ Epoch11: 157/157 ] loss:0.415 acc:20.312 \n",
      "Train | Loss:0.39576 Acc: 83.320\n",
      "Valid | Loss:0.33939 Acc: 83.984 \n",
      "total_acc/v_batch: 0.83984375\n",
      "saving model with acc 83.984\n",
      "-----------------------------------------------\n",
      "[ Epoch12: 157/157 ] loss:0.100 acc:25.000 \n",
      "Train | Loss:0.33566 Acc: 85.778\n",
      "Valid | Loss:0.33213 Acc: 84.434 \n",
      "total_acc/v_batch: 0.8443359375\n",
      "saving model with acc 84.434\n",
      "-----------------------------------------------\n",
      "[ Epoch13: 157/157 ] loss:0.275 acc:21.875 \n",
      "Train | Loss:0.31518 Acc: 86.788\n",
      "Valid | Loss:0.30324 Acc: 85.332 \n",
      "total_acc/v_batch: 0.8533203125\n",
      "saving model with acc 85.332\n",
      "-----------------------------------------------\n",
      "[ Epoch14: 157/157 ] loss:0.133 acc:24.219 \n",
      "Train | Loss:0.29205 Acc: 87.301\n",
      "Valid | Loss:0.28890 Acc: 85.723 \n",
      "total_acc/v_batch: 0.8572265625\n",
      "saving model with acc 85.723\n",
      "-----------------------------------------------\n",
      "[ Epoch15: 157/157 ] loss:0.394 acc:20.312 \n",
      "Train | Loss:0.28232 Acc: 88.346\n",
      "Valid | Loss:0.28823 Acc: 85.586 \n",
      "-----------------------------------------------\n",
      "[ Epoch16: 157/157 ] loss:0.382 acc:21.094 \n",
      "Train | Loss:0.27509 Acc: 88.311\n",
      "Valid | Loss:0.27415 Acc: 86.465 \n",
      "total_acc/v_batch: 0.8646484375\n",
      "saving model with acc 86.465\n",
      "-----------------------------------------------\n",
      "[ Epoch17: 157/157 ] loss:0.180 acc:23.438 \n",
      "Train | Loss:0.25549 Acc: 89.237\n",
      "Valid | Loss:0.27643 Acc: 86.152 \n",
      "-----------------------------------------------\n",
      "[ Epoch18: 157/157 ] loss:0.123 acc:24.219 \n",
      "Train | Loss:0.24825 Acc: 89.625\n",
      "Valid | Loss:0.30857 Acc: 85.430 \n",
      "-----------------------------------------------\n",
      "[ Epoch19: 157/157 ] loss:0.169 acc:22.656 \n",
      "Train | Loss:0.24229 Acc: 89.834\n",
      "Valid | Loss:0.28023 Acc: 86.348 \n",
      "-----------------------------------------------\n",
      "[ Epoch20: 157/157 ] loss:0.254 acc:21.875 \n",
      "Train | Loss:0.22686 Acc: 90.540\n",
      "Valid | Loss:0.27976 Acc: 86.035 \n",
      "-----------------------------------------------\n",
      "[ Epoch21: 157/157 ] loss:0.105 acc:25.000 \n",
      "Train | Loss:0.20955 Acc: 91.327\n",
      "Valid | Loss:0.28120 Acc: 86.426 \n",
      "-----------------------------------------------\n",
      "[ Epoch22: 157/157 ] loss:0.270 acc:22.656 \n",
      "Train | Loss:0.20086 Acc: 91.710\n",
      "Valid | Loss:0.30140 Acc: 85.938 \n",
      "-----------------------------------------------\n",
      "[ Epoch23: 157/157 ] loss:0.176 acc:23.438 \n",
      "Train | Loss:0.18606 Acc: 92.635\n",
      "Valid | Loss:0.31225 Acc: 85.625 \n",
      "-----------------------------------------------\n",
      "[ Epoch24: 157/157 ] loss:0.206 acc:23.438 \n",
      "Train | Loss:0.16762 Acc: 93.282\n",
      "Valid | Loss:0.36694 Acc: 85.332 \n",
      "-----------------------------------------------\n",
      "[ Epoch25: 157/157 ] loss:0.107 acc:24.219 \n",
      "Train | Loss:0.14921 Acc: 94.044\n",
      "Valid | Loss:0.32884 Acc: 85.781 \n",
      "-----------------------------------------------\n",
      "[ Epoch26: 157/157 ] loss:0.436 acc:21.094 \n",
      "Train | Loss:0.13143 Acc: 94.989\n",
      "Valid | Loss:0.34295 Acc: 84.980 \n",
      "-----------------------------------------------\n",
      "[ Epoch27: 157/157 ] loss:0.057 acc:24.219 \n",
      "Train | Loss:0.10725 Acc: 95.830\n",
      "Valid | Loss:0.42874 Acc: 84.785 \n",
      "-----------------------------------------------\n",
      "[ Epoch28: 157/157 ] loss:0.041 acc:24.219 \n",
      "Train | Loss:0.09776 Acc: 96.263\n",
      "Valid | Loss:0.46513 Acc: 84.980 \n",
      "-----------------------------------------------\n",
      "[ Epoch29: 157/157 ] loss:0.052 acc:24.219 \n",
      "Train | Loss:0.07934 Acc: 97.084\n",
      "Valid | Loss:0.44639 Acc: 84.688 \n",
      "-----------------------------------------------\n",
      "[ Epoch30: 157/157 ] loss:0.051 acc:24.219 \n",
      "Train | Loss:0.07754 Acc: 96.955\n",
      "Valid | Loss:0.45761 Acc: 85.039 \n",
      "-----------------------------------------------\n",
      "[ Epoch31: 157/157 ] loss:0.015 acc:25.000 \n",
      "Train | Loss:0.05628 Acc: 97.776\n",
      "Valid | Loss:0.53513 Acc: 85.000 \n",
      "-----------------------------------------------\n",
      "[ Epoch32: 157/157 ] loss:0.141 acc:24.219 \n",
      "Train | Loss:0.06294 Acc: 97.497\n",
      "Valid | Loss:0.51142 Acc: 84.453 \n",
      "-----------------------------------------------\n",
      "[ Epoch33: 157/157 ] loss:0.099 acc:24.219 \n",
      "Train | Loss:0.04587 Acc: 98.298\n",
      "Valid | Loss:0.53409 Acc: 84.375 \n",
      "-----------------------------------------------\n",
      "[ Epoch34: 157/157 ] loss:0.096 acc:24.219 \n",
      "Train | Loss:0.05158 Acc: 97.995\n",
      "Valid | Loss:0.57290 Acc: 84.238 \n",
      "-----------------------------------------------\n",
      "[ Epoch35: 157/157 ] loss:0.039 acc:24.219 \n",
      "Train | Loss:0.04018 Acc: 98.373\n",
      "Valid | Loss:0.57278 Acc: 84.668 \n",
      "-----------------------------------------------\n",
      "[ Epoch36: 157/157 ] loss:0.007 acc:25.000 \n",
      "Train | Loss:0.02907 Acc: 98.746\n",
      "Valid | Loss:0.64606 Acc: 84.316 \n",
      "-----------------------------------------------\n",
      "[ Epoch37: 157/157 ] loss:0.007 acc:25.000 \n",
      "Train | Loss:0.02952 Acc: 98.781\n",
      "Valid | Loss:0.64619 Acc: 84.453 \n",
      "-----------------------------------------------\n",
      "[ Epoch38: 157/157 ] loss:0.011 acc:25.000 \n",
      "Train | Loss:0.02607 Acc: 98.811\n",
      "Valid | Loss:0.65082 Acc: 84.414 \n",
      "-----------------------------------------------\n",
      "[ Epoch39: 157/157 ] loss:0.004 acc:25.000 \n",
      "Train | Loss:0.02358 Acc: 98.915\n",
      "Valid | Loss:0.62070 Acc: 84.688 \n",
      "-----------------------------------------------\n",
      "[ Epoch40: 157/157 ] loss:0.022 acc:25.000 \n",
      "Train | Loss:0.02742 Acc: 98.786\n",
      "Valid | Loss:0.63456 Acc: 83.809 \n",
      "-----------------------------------------------\n",
      "[ Epoch41: 157/157 ] loss:0.291 acc:22.656 \n",
      "Train | Loss:0.02437 Acc: 98.910\n",
      "Valid | Loss:0.68064 Acc: 83.320 \n",
      "-----------------------------------------------\n",
      "[ Epoch42: 157/157 ] loss:0.006 acc:25.000 \n",
      "Train | Loss:0.03231 Acc: 98.522\n",
      "Valid | Loss:0.61284 Acc: 84.727 \n",
      "-----------------------------------------------\n",
      "[ Epoch43: 157/157 ] loss:0.013 acc:25.000 \n",
      "Train | Loss:0.01373 Acc: 99.229\n",
      "Valid | Loss:0.65819 Acc: 83.945 \n",
      "-----------------------------------------------\n",
      "[ Epoch44: 157/157 ] loss:0.009 acc:25.000 \n",
      "Train | Loss:0.01484 Acc: 99.129\n",
      "Valid | Loss:0.65521 Acc: 84.102 \n",
      "-----------------------------------------------\n",
      "[ Epoch45: 157/157 ] loss:0.020 acc:25.000 \n",
      "Train | Loss:0.01672 Acc: 99.129\n",
      "Valid | Loss:0.64982 Acc: 83.770 \n",
      "-----------------------------------------------\n",
      "[ Epoch46: 157/157 ] loss:0.006 acc:25.000 \n",
      "Train | Loss:0.01695 Acc: 99.060\n",
      "Valid | Loss:0.69742 Acc: 83.984 \n",
      "-----------------------------------------------\n",
      "[ Epoch47: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.01540 Acc: 99.084\n",
      "Valid | Loss:0.67192 Acc: 83.730 \n",
      "-----------------------------------------------\n",
      "[ Epoch48: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.01185 Acc: 99.234\n",
      "Valid | Loss:0.77766 Acc: 85.000 \n",
      "-----------------------------------------------\n",
      "[ Epoch49: 157/157 ] loss:0.004 acc:25.000 \n",
      "Train | Loss:0.00916 Acc: 99.308\n",
      "Valid | Loss:0.78653 Acc: 84.199 \n",
      "-----------------------------------------------\n",
      "[ Epoch50: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.01108 Acc: 99.249\n",
      "Valid | Loss:0.76932 Acc: 84.863 \n",
      "-----------------------------------------------\n",
      "[ Epoch51: 157/157 ] loss:0.002 acc:25.000 \n",
      "Train | Loss:0.00873 Acc: 99.323\n",
      "Valid | Loss:0.79047 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch52: 157/157 ] loss:0.002 acc:25.000 \n",
      "Train | Loss:0.00539 Acc: 99.418\n",
      "Valid | Loss:0.86946 Acc: 84.902 \n",
      "-----------------------------------------------\n",
      "[ Epoch53: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.01244 Acc: 99.204\n",
      "Valid | Loss:0.79595 Acc: 84.922 \n",
      "-----------------------------------------------\n",
      "[ Epoch54: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.01093 Acc: 99.249\n",
      "Valid | Loss:0.79753 Acc: 84.434 \n",
      "-----------------------------------------------\n",
      "[ Epoch55: 157/157 ] loss:0.022 acc:25.000 \n",
      "Train | Loss:0.01248 Acc: 99.164\n",
      "Valid | Loss:0.72880 Acc: 84.707 \n",
      "-----------------------------------------------\n",
      "[ Epoch56: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00814 Acc: 99.323\n",
      "Valid | Loss:0.86605 Acc: 84.570 \n",
      "-----------------------------------------------\n",
      "[ Epoch57: 157/157 ] loss:0.004 acc:25.000 \n",
      "Train | Loss:0.01015 Acc: 99.199\n",
      "Valid | Loss:0.82242 Acc: 85.020 \n",
      "-----------------------------------------------\n",
      "[ Epoch58: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00756 Acc: 99.343\n",
      "Valid | Loss:0.87701 Acc: 84.883 \n",
      "-----------------------------------------------\n",
      "[ Epoch59: 157/157 ] loss:0.013 acc:25.000 \n",
      "Train | Loss:0.01070 Acc: 99.189\n",
      "Valid | Loss:0.70644 Acc: 82.520 \n",
      "-----------------------------------------------\n",
      "[ Epoch60: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.01313 Acc: 99.144\n",
      "Valid | Loss:0.77494 Acc: 84.727 \n",
      "-----------------------------------------------\n",
      "[ Epoch61: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.01079 Acc: 99.209\n",
      "Valid | Loss:0.81285 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch62: 157/157 ] loss:0.066 acc:24.219 \n",
      "Train | Loss:0.00770 Acc: 99.338\n",
      "Valid | Loss:0.82664 Acc: 83.535 \n",
      "-----------------------------------------------\n",
      "[ Epoch63: 157/157 ] loss:0.014 acc:25.000 \n",
      "Train | Loss:0.01283 Acc: 99.169\n",
      "Valid | Loss:0.78147 Acc: 85.098 \n",
      "-----------------------------------------------\n",
      "[ Epoch64: 157/157 ] loss:0.002 acc:25.000 \n",
      "Train | Loss:0.00744 Acc: 99.308\n",
      "Valid | Loss:0.81562 Acc: 84.590 \n",
      "-----------------------------------------------\n",
      "[ Epoch65: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.00573 Acc: 99.368\n",
      "Valid | Loss:0.98361 Acc: 84.688 \n",
      "-----------------------------------------------\n",
      "[ Epoch66: 157/157 ] loss:0.006 acc:25.000 \n",
      "Train | Loss:0.00763 Acc: 99.303\n",
      "Valid | Loss:0.84912 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch67: 157/157 ] loss:0.002 acc:25.000 \n",
      "Train | Loss:0.00555 Acc: 99.368\n",
      "Valid | Loss:0.85521 Acc: 85.059 \n",
      "-----------------------------------------------\n",
      "[ Epoch68: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00442 Acc: 99.433\n",
      "Valid | Loss:0.93091 Acc: 84.609 \n",
      "-----------------------------------------------\n",
      "[ Epoch69: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00863 Acc: 99.278\n",
      "Valid | Loss:0.86554 Acc: 84.082 \n",
      "-----------------------------------------------\n",
      "[ Epoch70: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00517 Acc: 99.398\n",
      "Valid | Loss:0.93178 Acc: 84.805 \n",
      "-----------------------------------------------\n",
      "[ Epoch71: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.01316 Acc: 99.074\n",
      "Valid | Loss:0.81793 Acc: 84.902 \n",
      "-----------------------------------------------\n",
      "[ Epoch72: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.00279 Acc: 99.453\n",
      "Valid | Loss:0.92056 Acc: 84.023 \n",
      "-----------------------------------------------\n",
      "[ Epoch73: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.00755 Acc: 99.308\n",
      "Valid | Loss:0.88754 Acc: 84.219 \n",
      "-----------------------------------------------\n",
      "[ Epoch74: 157/157 ] loss:0.038 acc:24.219 \n",
      "Train | Loss:0.00158 Acc: 99.492\n",
      "Valid | Loss:1.00871 Acc: 84.453 \n",
      "-----------------------------------------------\n",
      "[ Epoch75: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.01049 Acc: 99.214\n",
      "Valid | Loss:0.93431 Acc: 84.590 \n",
      "-----------------------------------------------\n",
      "[ Epoch76: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00231 Acc: 99.487\n",
      "Valid | Loss:1.02072 Acc: 84.492 \n",
      "-----------------------------------------------\n",
      "[ Epoch77: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00044 Acc: 99.522\n",
      "Valid | Loss:1.12402 Acc: 84.492 \n",
      "-----------------------------------------------\n",
      "[ Epoch78: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00022 Acc: 99.522\n",
      "Valid | Loss:1.18664 Acc: 84.590 \n",
      "-----------------------------------------------\n",
      "[ Epoch79: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00019 Acc: 99.522\n",
      "Valid | Loss:1.23473 Acc: 84.570 \n",
      "-----------------------------------------------\n",
      "[ Epoch80: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00019 Acc: 99.517\n",
      "Valid | Loss:1.27811 Acc: 84.492 \n",
      "-----------------------------------------------\n",
      "[ Epoch81: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.00023 Acc: 99.517\n",
      "Valid | Loss:1.25889 Acc: 84.238 \n",
      "-----------------------------------------------\n",
      "[ Epoch82: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.01764 Acc: 99.055\n",
      "Valid | Loss:0.72513 Acc: 84.492 \n",
      "-----------------------------------------------\n",
      "[ Epoch83: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.01070 Acc: 99.224\n",
      "Valid | Loss:0.84044 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch84: 157/157 ] loss:0.010 acc:25.000 \n",
      "Train | Loss:0.00641 Acc: 99.333\n",
      "Valid | Loss:0.92375 Acc: 83.730 \n",
      "-----------------------------------------------\n",
      "[ Epoch85: 157/157 ] loss:0.002 acc:25.000 \n",
      "Train | Loss:0.00404 Acc: 99.418\n",
      "Valid | Loss:1.00822 Acc: 84.414 \n",
      "-----------------------------------------------\n",
      "[ Epoch86: 157/157 ] loss:0.031 acc:24.219 \n",
      "Train | Loss:0.00828 Acc: 99.298\n",
      "Valid | Loss:0.90610 Acc: 84.355 \n",
      "-----------------------------------------------\n",
      "[ Epoch87: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00387 Acc: 99.398\n",
      "Valid | Loss:1.03203 Acc: 84.785 \n",
      "-----------------------------------------------\n",
      "[ Epoch88: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00290 Acc: 99.443\n",
      "Valid | Loss:0.98772 Acc: 84.219 \n",
      "-----------------------------------------------\n",
      "[ Epoch89: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00639 Acc: 99.353\n",
      "Valid | Loss:0.90570 Acc: 83.574 \n",
      "-----------------------------------------------\n",
      "[ Epoch90: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00697 Acc: 99.333\n",
      "Valid | Loss:0.95546 Acc: 84.512 \n",
      "-----------------------------------------------\n",
      "[ Epoch91: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00837 Acc: 99.254\n",
      "Valid | Loss:0.92424 Acc: 84.609 \n",
      "-----------------------------------------------\n",
      "[ Epoch92: 157/157 ] loss:0.071 acc:24.219 \n",
      "Train | Loss:0.00943 Acc: 99.298\n",
      "Valid | Loss:0.84665 Acc: 83.613 \n",
      "-----------------------------------------------\n",
      "[ Epoch93: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00564 Acc: 99.378\n",
      "Valid | Loss:0.96093 Acc: 84.609 \n",
      "-----------------------------------------------\n",
      "[ Epoch94: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00219 Acc: 99.478\n",
      "Valid | Loss:0.98567 Acc: 84.824 \n",
      "-----------------------------------------------\n",
      "[ Epoch95: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00055 Acc: 99.522\n",
      "Valid | Loss:1.13169 Acc: 84.570 \n",
      "-----------------------------------------------\n",
      "[ Epoch96: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00031 Acc: 99.522\n",
      "Valid | Loss:1.19710 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch97: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00016 Acc: 99.522\n",
      "Valid | Loss:1.25570 Acc: 84.551 \n",
      "-----------------------------------------------\n",
      "[ Epoch98: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00014 Acc: 99.522\n",
      "Valid | Loss:1.29376 Acc: 84.531 \n",
      "-----------------------------------------------\n",
      "[ Epoch99: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00009 Acc: 99.522\n",
      "Valid | Loss:1.34516 Acc: 84.434 \n",
      "-----------------------------------------------\n",
      "[ Epoch100: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00007 Acc: 99.522\n",
      "Valid | Loss:1.39298 Acc: 84.551 \n",
      "-----------------------------------------------\n",
      "[ Epoch101: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00007 Acc: 99.522\n",
      "Valid | Loss:1.42988 Acc: 84.355 \n",
      "-----------------------------------------------\n",
      "[ Epoch102: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00007 Acc: 99.522\n",
      "Valid | Loss:1.47513 Acc: 84.395 \n",
      "-----------------------------------------------\n",
      "[ Epoch103: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00007 Acc: 99.522\n",
      "Valid | Loss:1.51339 Acc: 84.434 \n",
      "-----------------------------------------------\n",
      "[ Epoch104: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00005 Acc: 99.522\n",
      "Valid | Loss:1.55603 Acc: 84.434 \n",
      "-----------------------------------------------\n",
      "[ Epoch105: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00004 Acc: 99.522\n",
      "Valid | Loss:1.58134 Acc: 84.473 \n",
      "-----------------------------------------------\n",
      "[ Epoch106: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00005 Acc: 99.522\n",
      "Valid | Loss:1.61194 Acc: 84.473 \n",
      "-----------------------------------------------\n",
      "[ Epoch107: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.65001 Acc: 84.512 \n",
      "-----------------------------------------------\n",
      "[ Epoch108: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.66653 Acc: 84.512 \n",
      "-----------------------------------------------\n",
      "[ Epoch109: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.68871 Acc: 84.453 \n",
      "-----------------------------------------------\n",
      "[ Epoch110: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.71111 Acc: 84.453 \n",
      "-----------------------------------------------\n",
      "[ Epoch111: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.74012 Acc: 84.375 \n",
      "-----------------------------------------------\n",
      "[ Epoch112: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.78499 Acc: 84.512 \n",
      "-----------------------------------------------\n",
      "[ Epoch113: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.80763 Acc: 84.473 \n",
      "-----------------------------------------------\n",
      "[ Epoch114: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00001 Acc: 99.522\n",
      "Valid | Loss:1.82812 Acc: 84.473 \n",
      "-----------------------------------------------\n",
      "[ Epoch115: 157/157 ] loss:0.013 acc:25.000 \n",
      "Train | Loss:0.00432 Acc: 99.448\n",
      "Valid | Loss:1.46590 Acc: 79.355 \n",
      "-----------------------------------------------\n",
      "[ Epoch116: 157/157 ] loss:0.031 acc:25.000 \n",
      "Train | Loss:0.04308 Acc: 98.104\n",
      "Valid | Loss:0.81031 Acc: 84.180 \n",
      "-----------------------------------------------\n",
      "[ Epoch117: 157/157 ] loss:0.003 acc:25.000 \n",
      "Train | Loss:0.01299 Acc: 99.139\n",
      "Valid | Loss:0.84797 Acc: 84.453 \n",
      "-----------------------------------------------\n",
      "[ Epoch118: 157/157 ] loss:0.002 acc:25.000 \n",
      "Train | Loss:0.00668 Acc: 99.353\n",
      "Valid | Loss:0.87795 Acc: 84.375 \n",
      "-----------------------------------------------\n",
      "[ Epoch119: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00916 Acc: 99.259\n",
      "Valid | Loss:0.87908 Acc: 84.512 \n",
      "-----------------------------------------------\n",
      "[ Epoch120: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00451 Acc: 99.413\n",
      "Valid | Loss:0.99895 Acc: 84.707 \n",
      "-----------------------------------------------\n",
      "[ Epoch121: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00361 Acc: 99.423\n",
      "Valid | Loss:0.99043 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch122: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00053 Acc: 99.522\n",
      "Valid | Loss:1.10197 Acc: 84.766 \n",
      "-----------------------------------------------\n",
      "[ Epoch123: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00041 Acc: 99.522\n",
      "Valid | Loss:1.19032 Acc: 84.668 \n",
      "-----------------------------------------------\n",
      "[ Epoch124: 157/157 ] loss:0.002 acc:25.000 \n",
      "Train | Loss:0.00210 Acc: 99.468\n",
      "Valid | Loss:1.17277 Acc: 84.238 \n",
      "-----------------------------------------------\n",
      "[ Epoch125: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00918 Acc: 99.219\n",
      "Valid | Loss:0.96904 Acc: 84.961 \n",
      "-----------------------------------------------\n",
      "[ Epoch126: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00118 Acc: 99.507\n",
      "Valid | Loss:1.13711 Acc: 84.766 \n",
      "-----------------------------------------------\n",
      "[ Epoch127: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00022 Acc: 99.522\n",
      "Valid | Loss:1.26511 Acc: 85.020 \n",
      "-----------------------------------------------\n",
      "[ Epoch128: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00018 Acc: 99.522\n",
      "Valid | Loss:1.30827 Acc: 84.766 \n",
      "-----------------------------------------------\n",
      "[ Epoch129: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00016 Acc: 99.522\n",
      "Valid | Loss:1.39176 Acc: 84.805 \n",
      "-----------------------------------------------\n",
      "[ Epoch130: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00009 Acc: 99.522\n",
      "Valid | Loss:1.45158 Acc: 84.766 \n",
      "-----------------------------------------------\n",
      "[ Epoch131: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00008 Acc: 99.522\n",
      "Valid | Loss:1.50753 Acc: 84.590 \n",
      "-----------------------------------------------\n",
      "[ Epoch132: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00006 Acc: 99.522\n",
      "Valid | Loss:1.54029 Acc: 84.668 \n",
      "-----------------------------------------------\n",
      "[ Epoch133: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00008 Acc: 99.522\n",
      "Valid | Loss:1.58685 Acc: 84.746 \n",
      "-----------------------------------------------\n",
      "[ Epoch134: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00004 Acc: 99.522\n",
      "Valid | Loss:1.60656 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch135: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00005 Acc: 99.522\n",
      "Valid | Loss:1.65046 Acc: 84.746 \n",
      "-----------------------------------------------\n",
      "[ Epoch136: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00004 Acc: 99.522\n",
      "Valid | Loss:1.66796 Acc: 84.766 \n",
      "-----------------------------------------------\n",
      "[ Epoch137: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.69312 Acc: 84.648 \n",
      "-----------------------------------------------\n",
      "[ Epoch138: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.76088 Acc: 84.648 \n",
      "-----------------------------------------------\n",
      "[ Epoch139: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00004 Acc: 99.522\n",
      "Valid | Loss:1.81876 Acc: 84.688 \n",
      "-----------------------------------------------\n",
      "[ Epoch140: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.78415 Acc: 84.746 \n",
      "-----------------------------------------------\n",
      "[ Epoch141: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.77249 Acc: 84.707 \n",
      "-----------------------------------------------\n",
      "[ Epoch142: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.80879 Acc: 84.727 \n",
      "-----------------------------------------------\n",
      "[ Epoch143: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.86705 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch144: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00001 Acc: 99.522\n",
      "Valid | Loss:1.89553 Acc: 84.707 \n",
      "-----------------------------------------------\n",
      "[ Epoch145: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00001 Acc: 99.522\n",
      "Valid | Loss:1.90506 Acc: 84.512 \n",
      "-----------------------------------------------\n",
      "[ Epoch146: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00001 Acc: 99.522\n",
      "Valid | Loss:1.93733 Acc: 84.648 \n",
      "-----------------------------------------------\n",
      "[ Epoch147: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.92417 Acc: 84.512 \n",
      "-----------------------------------------------\n",
      "[ Epoch148: 157/157 ] loss:0.097 acc:24.219 \n",
      "Train | Loss:0.03175 Acc: 98.612\n",
      "Valid | Loss:0.53362 Acc: 84.746 \n",
      "-----------------------------------------------\n",
      "[ Epoch149: 157/157 ] loss:0.009 acc:25.000 \n",
      "Train | Loss:0.02078 Acc: 98.875\n",
      "Valid | Loss:0.83556 Acc: 84.609 \n",
      "-----------------------------------------------\n",
      "[ Epoch150: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00498 Acc: 99.358\n",
      "Valid | Loss:0.99056 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch151: 157/157 ] loss:0.004 acc:25.000 \n",
      "Train | Loss:0.00394 Acc: 99.408\n",
      "Valid | Loss:0.96673 Acc: 84.102 \n",
      "-----------------------------------------------\n",
      "[ Epoch152: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00476 Acc: 99.363\n",
      "Valid | Loss:1.00026 Acc: 84.355 \n",
      "-----------------------------------------------\n",
      "[ Epoch153: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00340 Acc: 99.458\n",
      "Valid | Loss:0.96936 Acc: 83.828 \n",
      "-----------------------------------------------\n",
      "[ Epoch154: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00437 Acc: 99.408\n",
      "Valid | Loss:1.06863 Acc: 83.867 \n",
      "-----------------------------------------------\n",
      "[ Epoch155: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00648 Acc: 99.323\n",
      "Valid | Loss:0.93620 Acc: 84.023 \n",
      "-----------------------------------------------\n",
      "[ Epoch156: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00113 Acc: 99.502\n",
      "Valid | Loss:1.11012 Acc: 84.492 \n",
      "-----------------------------------------------\n",
      "[ Epoch157: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00033 Acc: 99.522\n",
      "Valid | Loss:1.16970 Acc: 84.570 \n",
      "-----------------------------------------------\n",
      "[ Epoch158: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00020 Acc: 99.522\n",
      "Valid | Loss:1.22623 Acc: 84.570 \n",
      "-----------------------------------------------\n",
      "[ Epoch159: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00013 Acc: 99.522\n",
      "Valid | Loss:1.28752 Acc: 84.531 \n",
      "-----------------------------------------------\n",
      "[ Epoch160: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00012 Acc: 99.522\n",
      "Valid | Loss:1.32801 Acc: 84.395 \n",
      "-----------------------------------------------\n",
      "[ Epoch161: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00012 Acc: 99.522\n",
      "Valid | Loss:1.35735 Acc: 84.375 \n",
      "-----------------------------------------------\n",
      "[ Epoch162: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00007 Acc: 99.522\n",
      "Valid | Loss:1.39672 Acc: 84.453 \n",
      "-----------------------------------------------\n",
      "[ Epoch163: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00006 Acc: 99.522\n",
      "Valid | Loss:1.43509 Acc: 84.355 \n",
      "-----------------------------------------------\n",
      "[ Epoch164: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00005 Acc: 99.522\n",
      "Valid | Loss:1.48560 Acc: 84.434 \n",
      "-----------------------------------------------\n",
      "[ Epoch165: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00004 Acc: 99.522\n",
      "Valid | Loss:1.52920 Acc: 84.434 \n",
      "-----------------------------------------------\n",
      "[ Epoch166: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00004 Acc: 99.522\n",
      "Valid | Loss:1.56945 Acc: 84.473 \n",
      "-----------------------------------------------\n",
      "[ Epoch167: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.59215 Acc: 84.453 \n",
      "-----------------------------------------------\n",
      "[ Epoch168: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.61518 Acc: 84.473 \n",
      "-----------------------------------------------\n",
      "[ Epoch169: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.65940 Acc: 84.473 \n",
      "-----------------------------------------------\n",
      "[ Epoch170: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.70344 Acc: 84.531 \n",
      "-----------------------------------------------\n",
      "[ Epoch171: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00002 Acc: 99.522\n",
      "Valid | Loss:1.73917 Acc: 84.492 \n",
      "-----------------------------------------------\n",
      "[ Epoch172: 157/157 ] loss:0.026 acc:24.219 \n",
      "Train | Loss:0.01595 Acc: 99.074\n",
      "Valid | Loss:0.73352 Acc: 85.098 \n",
      "-----------------------------------------------\n",
      "[ Epoch173: 157/157 ] loss:0.032 acc:24.219 \n",
      "Train | Loss:0.01342 Acc: 99.139\n",
      "Valid | Loss:0.87224 Acc: 84.023 \n",
      "-----------------------------------------------\n",
      "[ Epoch174: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00727 Acc: 99.343\n",
      "Valid | Loss:0.93523 Acc: 84.258 \n",
      "-----------------------------------------------\n",
      "[ Epoch175: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00579 Acc: 99.313\n",
      "Valid | Loss:0.98505 Acc: 83.574 \n",
      "-----------------------------------------------\n",
      "[ Epoch176: 157/157 ] loss:0.002 acc:25.000 \n",
      "Train | Loss:0.00617 Acc: 99.328\n",
      "Valid | Loss:0.94564 Acc: 84.043 \n",
      "-----------------------------------------------\n",
      "[ Epoch177: 157/157 ] loss:0.001 acc:25.000 \n",
      "Train | Loss:0.00256 Acc: 99.478\n",
      "Valid | Loss:1.04791 Acc: 84.531 \n",
      "-----------------------------------------------\n",
      "[ Epoch178: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00106 Acc: 99.482\n",
      "Valid | Loss:1.05347 Acc: 84.473 \n",
      "-----------------------------------------------\n",
      "[ Epoch179: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00043 Acc: 99.517\n",
      "Valid | Loss:1.14653 Acc: 84.375 \n",
      "-----------------------------------------------\n",
      "[ Epoch180: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00022 Acc: 99.522\n",
      "Valid | Loss:1.22724 Acc: 84.688 \n",
      "-----------------------------------------------\n",
      "[ Epoch181: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00012 Acc: 99.522\n",
      "Valid | Loss:1.28551 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch182: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00011 Acc: 99.522\n",
      "Valid | Loss:1.32013 Acc: 84.746 \n",
      "-----------------------------------------------\n",
      "[ Epoch183: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00009 Acc: 99.522\n",
      "Valid | Loss:1.36379 Acc: 84.668 \n",
      "-----------------------------------------------\n",
      "[ Epoch184: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00006 Acc: 99.522\n",
      "Valid | Loss:1.42281 Acc: 84.688 \n",
      "-----------------------------------------------\n",
      "[ Epoch185: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00005 Acc: 99.522\n",
      "Valid | Loss:1.46990 Acc: 84.688 \n",
      "-----------------------------------------------\n",
      "[ Epoch186: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00004 Acc: 99.522\n",
      "Valid | Loss:1.52030 Acc: 84.707 \n",
      "-----------------------------------------------\n",
      "[ Epoch187: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.54153 Acc: 84.707 \n",
      "-----------------------------------------------\n",
      "[ Epoch188: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00004 Acc: 99.522\n",
      "Valid | Loss:1.56631 Acc: 84.824 \n",
      "-----------------------------------------------\n",
      "[ Epoch189: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.59836 Acc: 84.766 \n",
      "-----------------------------------------------\n",
      "[ Epoch190: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00003 Acc: 99.522\n",
      "Valid | Loss:1.62771 Acc: 84.727 \n",
      "-----------------------------------------------\n",
      "[ Epoch191: 157/157 ] loss:0.063 acc:24.219 \n",
      "Train | Loss:0.01343 Acc: 99.104\n",
      "Valid | Loss:0.63660 Acc: 84.297 \n",
      "-----------------------------------------------\n",
      "[ Epoch192: 157/157 ] loss:0.007 acc:25.000 \n",
      "Train | Loss:0.01942 Acc: 98.960\n",
      "Valid | Loss:0.79721 Acc: 83.965 \n",
      "-----------------------------------------------\n",
      "[ Epoch193: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00523 Acc: 99.368\n",
      "Valid | Loss:0.93116 Acc: 84.062 \n",
      "-----------------------------------------------\n",
      "[ Epoch194: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00290 Acc: 99.453\n",
      "Valid | Loss:0.99350 Acc: 84.531 \n",
      "-----------------------------------------------\n",
      "[ Epoch195: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00094 Acc: 99.502\n",
      "Valid | Loss:1.10781 Acc: 84.688 \n",
      "-----------------------------------------------\n",
      "[ Epoch196: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00061 Acc: 99.517\n",
      "Valid | Loss:1.19089 Acc: 84.785 \n",
      "-----------------------------------------------\n",
      "[ Epoch197: 157/157 ] loss:0.018 acc:25.000 \n",
      "Train | Loss:0.00970 Acc: 99.273\n",
      "Valid | Loss:0.87498 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch198: 157/157 ] loss:0.025 acc:24.219 \n",
      "Train | Loss:0.00611 Acc: 99.363\n",
      "Valid | Loss:0.83812 Acc: 84.688 \n",
      "-----------------------------------------------\n",
      "[ Epoch199: 157/157 ] loss:0.000 acc:25.000 \n",
      "Train | Loss:0.00470 Acc: 99.408\n",
      "Valid | Loss:1.04234 Acc: 84.629 \n",
      "-----------------------------------------------\n",
      "[ Epoch200: 157/157 ] loss:0.002 acc:25.000 \n",
      "Train | Loss:0.00195 Acc: 99.482\n",
      "Valid | Loss:1.06725 Acc: 84.336 \n",
      "-----------------------------------------------\n",
      "over\n"
     ]
    },
    {
     "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": [
    "# main.py\n",
    "import os\n",
    "import torch\n",
    "import argparse\n",
    "import numpy as np\n",
    "from torch import nn\n",
    "from gensim.models import word2vec\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "# 通過torch.cuda.is_available()的回傳值進行判斷是否有使用GPU的環境，如果有的話device就設為\"cuda\"，沒有的話就設為\"cpu\"\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "\n",
    "# 處理好各個data的路徑\n",
    "\n",
    "#  label_train_path = \"labeledTrainData.tsv\"\n",
    "#  unlabel_train_path = \"unlabeledTrainData.tsv\"\n",
    "#  test_path = \"testData.tsv\"\n",
    "train_with_label = os.path.join(path_prefix, label_train_path)\n",
    "train_no_label = os.path.join(path_prefix, unlabel_train_path)\n",
    "testing_data = os.path.join(path_prefix, test_path)\n",
    "\n",
    "w2v_path = os.path.join(path_prefix, 'w2v_all.model') # 處理word to vec model的路徑\n",
    "\n",
    "# 定義句子長度、要不要固定embedding、batch大小、要訓練幾個epoch、learning rate的值、model的資料夾路徑\n",
    "sen_len = 200\n",
    "fix_embedding = True # fix embedding during training\n",
    "batch_size = 128\n",
    "epoch = 200\n",
    "lr = 0.001\n",
    "# model_dir = os.path.join(path_prefix, 'model/') # model directory for checkpoint model\n",
    "model_dir = path_prefix # model directory for checkpoint model\n",
    "\n",
    "print(\"loading data ...\") # 把'training_label.txt'跟'training_nolabel.txt'讀進來\n",
    "train_x, y = load_training_data(label_train_path)\n",
    "#train_x_no_label = load_training_data(train_no_label)\n",
    "\n",
    "# 對input跟labels做預處理\n",
    "preprocess = Preprocess(train_x, sen_len, w2v_path=w2v_path)\n",
    "embedding = preprocess.make_embedding(load=True)\n",
    "train_x = preprocess.sentence_word2idx()\n",
    "y = preprocess.labels_to_tensor(y)\n",
    "\n",
    "# 製作一個model的對象\n",
    "model = LSTM_Net(embedding, embedding_dim=250, hidden_dim=250, num_layers=1, dropout=0.5, fix_embedding=fix_embedding)\n",
    "model = model.to(device) # device為\"cuda\"，model使用GPU來訓練(餵進去的inputs也需要是cuda tensor)\n",
    "\n",
    "# 把data分為training data跟validation data(將一部份training data拿去當作validation data)\n",
    "p = 0.8\n",
    "split_num = int(len(train_x)*p)\n",
    "X_train, X_val, y_train, y_val = train_x[:split_num], train_x[split_num:], y[:split_num], y[split_num:]\n",
    "\n",
    "# 把data做成dataset供dataloader取用\n",
    "train_dataset = TwitterDataset(X=X_train, y=y_train)\n",
    "val_dataset = TwitterDataset(X=X_val, y=y_val)\n",
    "\n",
    "# 把data 轉成 batch of tensors\n",
    "train_loader = torch.utils.data.DataLoader(dataset = train_dataset,\n",
    "                                            batch_size = batch_size,\n",
    "                                            shuffle = True,\n",
    "                                            num_workers = 8)\n",
    "\n",
    "val_loader = torch.utils.data.DataLoader(dataset = val_dataset,\n",
    "                                            batch_size = batch_size,\n",
    "                                            shuffle = False,\n",
    "                                            num_workers = 8)\n",
    "\n",
    "# 開始訓練\n",
    "training(batch_size, epoch, lr, model_dir, train_loader, val_loader, model, device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 163,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "over\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x504 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig=plt.figure(figsize=(20,7))\n",
    "ax1 = plt.subplot(1,2,1)\n",
    "plt.sca(ax1)\n",
    "plt.plot(xx,yy_train_loss,\"r\",label = \"train_loss\")\n",
    "plt.plot(xx,yy_valid_loss,\"b\",label = \"valid_loss\")\n",
    "plt.legend()\n",
    "ax2 = plt.subplot(1,2,2)\n",
    "plt.sca(ax2)\n",
    "plt.plot(xx,yy_train_acc,\"y\",label = \"train_acc\")\n",
    "plt.plot(xx,yy_valid_acc,\"g\",label = \"valid_acc\")\n",
    "plt.legend()\n",
    "\n",
    "plt.savefig(\"Adam_lr0.001_200epoch.png\")\n",
    "print(\"over\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "8fQeaQNeNm3L"
   },
   "source": [
    "### Predict and Write to csv file"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 159,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 241
    },
    "colab_type": "code",
    "id": "vFvjFQopxVrt",
    "outputId": "3bf30696-bf76-463b-fc7f-f0b6375a0e00"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "loading testing data ...\n",
      "Get embedding ...\n",
      "loading word to vec model ...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/u2019101432/.conda/envs/pytorch/lib/python3.7/site-packages/ipykernel_launcher.py:44: DeprecationWarning: Call to deprecated `__getitem__` (Method will be removed in 4.0.0, use self.wv.__getitem__() instead).\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": []
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "get words #121343\n",
      "total words: 121345\n",
      "sentence count #25000\n",
      "load model ...\n"
     ]
    }
   ],
   "source": [
    "# 開始測試模型並做預測\n",
    "print(\"loading testing data ...\")\n",
    "test_x = load_testing_data(testing_data)\n",
    "preprocess = Preprocess(test_x, sen_len, w2v_path=w2v_path)\n",
    "embedding = preprocess.make_embedding(load=True)\n",
    "test_x = preprocess.sentence_word2idx()\n",
    "test_dataset = TwitterDataset(X=test_x, y=None)\n",
    "test_loader = torch.utils.data.DataLoader(dataset = test_dataset,\n",
    "                                            batch_size = batch_size,\n",
    "                                            shuffle = False,\n",
    "                                            num_workers = 8)\n",
    "print('\\nload model ...')\n",
    "model = torch.load(os.path.join(model_dir, 'ckpt.model'))\n",
    "outputs = testing(batch_size, test_loader, model, device)\n",
    "\n",
    "\n",
    "# 以下是使用command line上傳到kaggle的方式\n",
    "# 需要先pip install kaggle、Create API Token，詳細請看https://github.com/Kaggle/kaggle-api以及https://www.kaggle.com/code1110/how-to-submit-from-google-colab\n",
    "# kaggle competitions submit [competition-name] -f [csv file path]] -m [message]\n",
    "# ex: kaggle competitions submit ml-2020spring-hw4 -f output/predict.csv -m \"......\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 160,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "save csv ...\n",
      "Finish Predicting\n"
     ]
    }
   ],
   "source": [
    "df = pd.read_csv(\"testData.tsv\",sep = \"\\t\",error_bad_lines=False)\n",
    "kk = list(df[\"id\"])\n",
    "\n",
    "# 寫到csv檔案供上傳kaggle\n",
    "tmp = pd.DataFrame({\"id\":kk,\"sentiment\":outputs})\n",
    "print(\"save csv ...\")\n",
    "tmp.to_csv(os.path.join(path_prefix, 'result_predict.csv'), index=False)\n",
    "print(\"Finish Predicting\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 161,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "over\n"
     ]
    }
   ],
   "source": [
    "print(\"over\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "qSvgTRuGu2Qb"
   },
   "source": [
    "#### Check where the files are"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 162,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 102
    },
    "colab_type": "code",
    "id": "8SZYJQ62utiK",
    "outputId": "f42a12be-91af-49fa-d4ce-d300d16dc812"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/u2019101432/jupyterlab/kaggle_IMDB_sentiment_classificationassi\n",
      "0.8490234375_best_acc.txt\t t2.png\n",
      "0.8646484375_best_acc.txt\t testData.tsv\n",
      "Adam_lr0.001_200epoch.png\t testData.tsv.zip\n",
      "Untitled.ipynb\t\t\t tl.png\n",
      "ckpt.model\t\t\t unlabeledTrainData.tsv\n",
      "labeledTrainData.tsv\t\t unlabeledTrainData.tsv.zip\n",
      "labeledTrainData.tsv.zip\t w2v_all.model\n",
      "result_predict.csv\t\t w2v_all.model.trainables.syn1neg.npy\n",
      "sampleSubmission.csv\t\t w2v_all.model.wv.vectors.npy\n",
      "sentimenti_classification.ipynb  word2vec-nlp-tutorial.zip\n"
     ]
    }
   ],
   "source": [
    "!pwd\n",
    "!ls"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "0iCyLSWFcEmP"
   },
   "source": [
    "### Run 20 epochs on n98\n",
    "real\t3m33.317s\n",
    "\n",
    "user\t3m29.813s\n",
    "\n",
    "sys\t1m9.469s"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "include_colab_link": true,
   "name": "hw4_RNN.ipynb",
   "provenance": [],
   "toc_visible": true
  },
  "kernelspec": {
   "display_name": "pytorch",
   "language": "python",
   "name": "pytorch"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}