train_LSTM.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true,
    "pycharm": {
     "is_executing": false
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "text": [
      "Loaded  data\\PRSA_Data_Aotizhongxin_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Changping_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Dingling_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Dongsi_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Guanyuan_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Gucheng_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Huairou_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Nongzhanguan_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Shunyi_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Tiantan_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Wanliu_20130301-20170228.csv\n",
      "Loaded  data\\PRSA_Data_Wanshouxigong_20130301-20170228.csv\n",
      "Applied normalization on  ['PM10', 'SO2', 'NO2', 'CO', 'O3', 'TEMP', 'PRES', 'DEWP', 'RAIN', 'WSPM']\n",
      "Applied normalization on  ['PM10', 'SO2', 'NO2', 'CO', 'O3', 'TEMP', 'PRES', 'DEWP', 'RAIN', 'WSPM']\n"
     ],
     "output_type": "stream"
    },
    {
     "data": {
      "text/plain": "   PM2.5      PM10       SO2       NO2        CO        O3      TEMP  \\\n0    4.0  0.002006  0.007433  0.020671  0.020202  0.071710  0.312195   \n1    8.0  0.006018  0.007433  0.020671  0.020202  0.071710  0.305691   \n2    7.0  0.005015  0.009434  0.031053  0.020202  0.067974  0.305691   \n3    6.0  0.004012  0.021441  0.034514  0.020202  0.067040  0.300813   \n4    3.0  0.001003  0.023442  0.037974  0.020202  0.067040  0.291057   \n5    5.0  0.003009  0.035449  0.058737  0.030303  0.061437  0.287805   \n6    3.0  0.001003  0.035449  0.107185  0.040404  0.046495  0.281301   \n7    3.0  0.004012  0.037450  0.138329  0.040404  0.039957  0.297561   \n8    3.0  0.004012  0.031447  0.145250  0.040404  0.041825  0.325203   \n9    3.0  0.006018  0.023442  0.093342  0.030303  0.054900  0.343089   \n\n       PRES      DEWP  RAIN  ...  WD_NNW  WD_NW  WD_S  WD_SE  WD_SSE  WD_SSW  \\\n0  0.672185  0.339310   0.0  ...       1      0     0      0       0       0   \n1  0.675497  0.347586   0.0  ...       0      0     0      0       0       0   \n2  0.680464  0.347586   0.0  ...       1      0     0      0       0       0   \n3  0.697020  0.331034   0.0  ...       0      1     0      0       0       0   \n4  0.708609  0.329655   0.0  ...       0      0     0      0       0       0   \n5  0.715232  0.328276   0.0  ...       0      0     0      0       0       0   \n6  0.730132  0.335172   0.0  ...       0      0     0      0       0       0   \n7  0.745033  0.335172   0.0  ...       1      0     0      0       0       0   \n8  0.759934  0.333793   0.0  ...       1      0     0      0       0       0   \n9  0.763245  0.332414   0.0  ...       0      0     0      0       0       0   \n\n   WD_SW  WD_W  WD_WNW  WD_WSW  \n0      0     0       0       0  \n1      0     0       0       0  \n2      0     0       0       0  \n3      0     0       0       0  \n4      0     0       0       0  \n5      0     0       0       0  \n6      0     0       0       0  \n7      0     0       0       0  \n8      0     0       0       0  \n9      0     0       0       0  \n\n[10 rows x 27 columns]",
      "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>PM2.5</th>\n      <th>PM10</th>\n      <th>SO2</th>\n      <th>NO2</th>\n      <th>CO</th>\n      <th>O3</th>\n      <th>TEMP</th>\n      <th>PRES</th>\n      <th>DEWP</th>\n      <th>RAIN</th>\n      <th>...</th>\n      <th>WD_NNW</th>\n      <th>WD_NW</th>\n      <th>WD_S</th>\n      <th>WD_SE</th>\n      <th>WD_SSE</th>\n      <th>WD_SSW</th>\n      <th>WD_SW</th>\n      <th>WD_W</th>\n      <th>WD_WNW</th>\n      <th>WD_WSW</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <td>0</td>\n      <td>4.0</td>\n      <td>0.002006</td>\n      <td>0.007433</td>\n      <td>0.020671</td>\n      <td>0.020202</td>\n      <td>0.071710</td>\n      <td>0.312195</td>\n      <td>0.672185</td>\n      <td>0.339310</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>1</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <td>1</td>\n      <td>8.0</td>\n      <td>0.006018</td>\n      <td>0.007433</td>\n      <td>0.020671</td>\n      <td>0.020202</td>\n      <td>0.071710</td>\n      <td>0.305691</td>\n      <td>0.675497</td>\n      <td>0.347586</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <td>2</td>\n      <td>7.0</td>\n      <td>0.005015</td>\n      <td>0.009434</td>\n      <td>0.031053</td>\n      <td>0.020202</td>\n      <td>0.067974</td>\n      <td>0.305691</td>\n      <td>0.680464</td>\n      <td>0.347586</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>1</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <td>3</td>\n      <td>6.0</td>\n      <td>0.004012</td>\n      <td>0.021441</td>\n      <td>0.034514</td>\n      <td>0.020202</td>\n      <td>0.067040</td>\n      <td>0.300813</td>\n      <td>0.697020</td>\n      <td>0.331034</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>0</td>\n      <td>1</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <td>4</td>\n      <td>3.0</td>\n      <td>0.001003</td>\n      <td>0.023442</td>\n      <td>0.037974</td>\n      <td>0.020202</td>\n      <td>0.067040</td>\n      <td>0.291057</td>\n      <td>0.708609</td>\n      <td>0.329655</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <td>5</td>\n      <td>5.0</td>\n      <td>0.003009</td>\n      <td>0.035449</td>\n      <td>0.058737</td>\n      <td>0.030303</td>\n      <td>0.061437</td>\n      <td>0.287805</td>\n      <td>0.715232</td>\n      <td>0.328276</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <td>6</td>\n      <td>3.0</td>\n      <td>0.001003</td>\n      <td>0.035449</td>\n      <td>0.107185</td>\n      <td>0.040404</td>\n      <td>0.046495</td>\n      <td>0.281301</td>\n      <td>0.730132</td>\n      <td>0.335172</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <td>7</td>\n      <td>3.0</td>\n      <td>0.004012</td>\n      <td>0.037450</td>\n      <td>0.138329</td>\n      <td>0.040404</td>\n      <td>0.039957</td>\n      <td>0.297561</td>\n      <td>0.745033</td>\n      <td>0.335172</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>1</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <td>8</td>\n      <td>3.0</td>\n      <td>0.004012</td>\n      <td>0.031447</td>\n      <td>0.145250</td>\n      <td>0.040404</td>\n      <td>0.041825</td>\n      <td>0.325203</td>\n      <td>0.759934</td>\n      <td>0.333793</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>1</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <td>9</td>\n      <td>3.0</td>\n      <td>0.006018</td>\n      <td>0.023442</td>\n      <td>0.093342</td>\n      <td>0.030303</td>\n      <td>0.054900</td>\n      <td>0.343089</td>\n      <td>0.763245</td>\n      <td>0.332414</td>\n      <td>0.0</td>\n      <td>...</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n    </tr>\n  </tbody>\n</table>\n<p>10 rows × 27 columns</p>\n</div>"
     },
     "metadata": {},
     "output_type": "execute_result",
     "execution_count": 24
    }
   ],
   "source": [
    "from utils import read_data\n",
    "\n",
    "data = read_data(keep_nan=True,keep_dates=False)\n",
    "df_all = data.train_val_df\n",
    "df_test = data.test_df\n",
    "df_all = df_all.drop(columns=[\"station\",\"time_stamp\"])\n",
    "df_test = df_test.drop(columns=[\"station\",\"time_stamp\"])\n",
    "df_all.columns\n",
    "df_all.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "outputs": [
    {
     "name": "stdout",
     "text": [
      "(2966, 48, 27)\n"
     ],
     "output_type": "stream"
    }
   ],
   "source": [
    "import numpy as np\n",
    "data = df_all.values.reshape(12,-1,27)[:,:29760,:]\n",
    "data = data.reshape(-1 ,48,27)\n",
    "data = data[~np.isnan(data).any(axis=(1,2))]\n",
    "print(data.shape)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "import torch.utils.data as tud\n",
    "\n",
    "class TimeSeriesDataset(tud.Dataset):\n",
    "    def __init__(self, data):\n",
    "        self.y = data[:,24:,0]\n",
    "        self.x = data[:,:,1:]\n",
    "    \n",
    "    def __len__(self):\n",
    "        return self.x.shape[0]\n",
    "    \n",
    "    def __getitem__(self, idx):\n",
    "        return self.y[idx], self.x[idx]"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "outputs": [],
   "source": [
    "class LSTM_Model(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(LSTM_Model, self).__init__()\n",
    "        self.LSTM = nn.LSTM(26, 16, batch_first=True)\n",
    "        self.fc1 = nn.Linear(16, 5)\n",
    "        self.fc2 = nn.Linear(5 ,1)\n",
    "        \n",
    "    def forward(self, x):\n",
    "        x, (h_n, h_c) = self.LSTM(x)\n",
    "        x = x[:,24:,:]\n",
    "        x = self.fc1(x)\n",
    "        x = F.relu(x)\n",
    "        x = F.dropout(x, p = 0.1, training = self.training)\n",
    "        x = self.fc2(x)\n",
    "        return x"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "outputs": [],
   "source": [
    "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
    "learning_rate = 1e-3\n",
    "batch_size = 30\n",
    "epochs = 80\n",
    "test_interval = 5\n",
    "log_interval = 50\n",
    "\n",
    "model = LSTM_Model().to(device).double()\n",
    "optimizer = optim.Adam(model.parameters(), lr=learning_rate)\n",
    "dataset = TimeSeriesDataset(data)\n",
    "train_set, validation_set = torch.utils.data.random_split(dataset, [len(dataset) - 590, 590])\n",
    "train_dataloader = tud.DataLoader(train_set, batch_size=batch_size, shuffle=True)\n",
    "validation_dataloader = tud.DataLoader(validation_set, batch_size=batch_size)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "outputs": [],
   "source": [
    "def train(model, device, train_loader, optimizer, epoch, log_interval):\n",
    "    model.train()\n",
    "    \n",
    "    losses = []\n",
    "    for batch_idx, (y ,x) in enumerate(train_loader):\n",
    "        y = y.to(device)\n",
    "        x = x.to(device)\n",
    "        \n",
    "        optimizer.zero_grad()\n",
    "        output = model(x)\n",
    "        \n",
    "        loss = F.mse_loss(output, y.view(-1,24,1))\n",
    "        losses.append(loss.item())\n",
    "        \n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        \n",
    "        if (batch_idx + 1) % log_interval == 0:\n",
    "            print('Train Epoch: {} [{}/{} ({:.0f}%)]\\tLoss: {:.6f}'.format(\n",
    "                epoch + 1, (batch_idx + 1) * batch_size, len(train_loader.dataset),\n",
    "                100. * (batch_idx + 1) / len(train_loader), loss.item()))\n",
    "    return losses"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "outputs": [],
   "source": [
    "class RMSELoss(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        self.mse = nn.MSELoss()\n",
    "        \n",
    "    def forward(self,yhat,y):\n",
    "        return torch.sqrt(self.mse(yhat,y))\n",
    "\n",
    "def test(model, device, test_loader):\n",
    "    model.eval()\n",
    "    mse_losses = []\n",
    "    mae_losses = []\n",
    "    rmse_losses = []\n",
    "    accuracy10 = []\n",
    "    accuracy20 = []\n",
    "    accuracy50 = []\n",
    "    accuracy100 = []\n",
    "    for batch_idx, (y ,x) in enumerate(test_loader):\n",
    "        y = y.to(device).view(-1,24,1)\n",
    "        x = x.to(device)\n",
    "        \n",
    "        with torch.no_grad():\n",
    "            output = model(x)\n",
    "            \n",
    "            mse_loss = F.mse_loss(output, y)\n",
    "            mse_losses.append(mse_loss.item())\n",
    "            mae_loss =F.l1_loss(output, y)\n",
    "            mae_losses.append(mae_loss)\n",
    "            criterion = RMSELoss()\n",
    "            rmse_loss = criterion(output, y)\n",
    "            rmse_losses.append(rmse_loss)\n",
    "            \n",
    "            diff = output - y\n",
    "            accuracy10.append(len(diff[abs(diff)<=10]))\n",
    "            accuracy20.append(len(diff[abs(diff)<=20]))\n",
    "            accuracy50.append(len(diff[abs(diff)<=50]))\n",
    "            accuracy100.append(len(diff[abs(diff)<=100]))\n",
    "    data_size = len(test_loader.dataset) * 24.\n",
    "    print(float(sum(mse_losses))/len(mse_losses), float(sum(rmse_losses))/len(rmse_losses),\n",
    "          float(sum(mae_losses))/len(mae_losses))\n",
    "    \n",
    "    print (float(sum(accuracy10))/data_size, float(sum(accuracy20))/data_size,\n",
    "          float(sum(accuracy50))/data_size, float(sum(accuracy100))/data_size)\n",
    "    \n",
    "    return float(sum(mse_losses))/len(mse_losses), float(sum(rmse_losses))/len(rmse_losses),\\\n",
    "           float(sum(mae_losses))/len(mae_losses),\\\n",
    "           float(sum(accuracy10))/data_size, float(sum(accuracy20))/data_size,\\\n",
    "           float(sum(accuracy50))/data_size, float(sum(accuracy100))/data_size"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "outputs": [
    {
     "name": "stdout",
     "text": [
      "Train Epoch: 1 [1500/2376 (62%)]\tLoss: 18836.954796\n",
      "11260.352500188435 105.22593842037013 77.22013621494811\n",
      "0.11483050847457626 0.23036723163841807 0.4612994350282486 0.7213276836158192\n",
      "Train Epoch: 2 [1500/2376 (62%)]\tLoss: 12542.759675\n",
      "10645.972173011507 102.26808024232048 73.26891193998088\n",
      "0.1687853107344633 0.2627118644067797 0.48968926553672315 0.7351694915254238\n",
      "Train Epoch: 3 [1500/2376 (62%)]\tLoss: 9152.062653\n",
      "9912.941243339243 98.6256621673407 68.8116598074616\n",
      "0.23036723163841807 0.30995762711864405 0.5254943502824859 0.7561440677966101\n",
      "Train Epoch: 4 [1500/2376 (62%)]\tLoss: 8269.818961\n",
      "9107.212455153687 94.46518497245759 64.47372252002998\n",
      "0.22259887005649717 0.35755649717514126 0.5610169491525424 0.7751412429378531\n",
      "Train Epoch: 5 [1500/2376 (62%)]\tLoss: 8725.713502\n",
      "8281.678810936914 90.01226277639819 60.53160423271653\n",
      "0.17641242937853108 0.3873587570621469 0.6018361581920904 0.7937146892655367\n",
      "Train Epoch: 6 [1500/2376 (62%)]\tLoss: 7198.105527\n",
      "7502.692153671944 85.61300759921424 57.200443969503205\n",
      "0.15720338983050847 0.33848870056497177 0.639406779661017 0.8142655367231638\n",
      "Train Epoch: 7 [1500/2376 (62%)]\tLoss: 8340.046510\n",
      "6823.46368780347 81.60517100426358 54.745155826370535\n",
      "0.15360169491525424 0.3038135593220339 0.6713276836158192 0.8290960451977402\n",
      "Train Epoch: 8 [1500/2376 (62%)]\tLoss: 5928.896033\n",
      "6284.655136355052 78.3067714451342 53.29055254549335\n",
      "0.1440677966101695 0.28269774011299437 0.7057203389830509 0.8461864406779661\n",
      "Train Epoch: 9 [1500/2376 (62%)]\tLoss: 3819.847044\n",
      "5881.78377311243 75.77673978597127 52.676712663227434\n",
      "0.13305084745762713 0.26744350282485874 0.7082627118644068 0.8584745762711864\n",
      "Train Epoch: 10 [1500/2376 (62%)]\tLoss: 4809.037429\n",
      "5622.3591916443365 74.1295284053363 52.716780371058135\n",
      "0.12351694915254237 0.24484463276836158 0.6413135593220339 0.8711158192090396\n",
      "Train Epoch: 11 [1500/2376 (62%)]\tLoss: 4261.982738\n",
      "5466.100994978811 73.142693471804 53.08765610710863\n",
      "0.11560734463276837 0.2327683615819209 0.5920197740112995 0.878954802259887\n",
      "Train Epoch: 12 [1500/2376 (62%)]\tLoss: 6336.894727\n",
      "5370.481149374757 72.53920016041084 53.38910663425844\n",
      "0.11031073446327684 0.225 0.5713983050847458 0.8834745762711864\n",
      "Train Epoch: 13 [1500/2376 (62%)]\tLoss: 6092.393795\n",
      "4581.659626656832 66.79481021862838 44.60355168126894\n",
      "0.1856638418079096 0.35847457627118645 0.7384180790960452 0.8901129943502825\n",
      "Train Epoch: 14 [1500/2376 (62%)]\tLoss: 3354.278623\n",
      "4047.0685315757028 62.72706593428627 39.98696930936253\n",
      "0.2623587570621469 0.44201977401129944 0.7660310734463277 0.9001412429378531\n",
      "Train Epoch: 15 [1500/2376 (62%)]\tLoss: 4144.167134\n",
      "3603.818658131245 59.17400063524455 37.54595034154962\n",
      "0.2706920903954802 0.4612994350282486 0.7832627118644068 0.9111581920903955\n",
      "Train Epoch: 16 [1500/2376 (62%)]\tLoss: 2726.635366\n",
      "3186.922935136804 55.598702736634934 33.68185535543246\n",
      "0.34442090395480224 0.5448446327683616 0.8032485875706215 0.919774011299435\n",
      "Train Epoch: 17 [1500/2376 (62%)]\tLoss: 3508.549819\n",
      "2813.160977702236 52.20476315709686 31.66317018520969\n",
      "0.34689265536723163 0.5516949152542373 0.8223870056497176 0.9319915254237288\n",
      "Train Epoch: 18 [1500/2376 (62%)]\tLoss: 3073.654432\n",
      "2539.2726486772726 49.57523069642161 29.495553795877242\n",
      "0.3838276836158192 0.5912429378531073 0.8331920903954803 0.9377118644067797\n",
      "Train Epoch: 19 [1500/2376 (62%)]\tLoss: 1178.984247\n",
      "2277.9991848016784 46.927901879622304 28.001033053312693\n",
      "0.38926553672316383 0.6022598870056497 0.845409604519774 0.945268361581921\n",
      "Train Epoch: 20 [1500/2376 (62%)]\tLoss: 2344.339624\n",
      "2069.1072548542443 44.70435510936974 26.969683717852405\n",
      "0.38467514124293783 0.6080508474576272 0.8546610169491525 0.9524011299435028\n",
      "Train Epoch: 21 [1500/2376 (62%)]\tLoss: 2921.600481\n",
      "1882.9044492127007 42.615503780414585 25.639930267614165\n",
      "0.4021186440677966 0.6219632768361582 0.8649717514124293 0.9569915254237288\n",
      "Train Epoch: 22 [1500/2376 (62%)]\tLoss: 3214.520789\n",
      "1732.8256571830352 40.90328790480049 24.37317470882057\n",
      "0.4279661016949153 0.6423022598870056 0.8735169491525424 0.96045197740113\n",
      "Train Epoch: 23 [1500/2376 (62%)]\tLoss: 1866.220827\n",
      "1669.9041981165433 40.23203771186364 24.390233426340046\n",
      "0.42768361581920905 0.6305084745762712 0.8701271186440678 0.9620056497175141\n",
      "Train Epoch: 24 [1500/2376 (62%)]\tLoss: 1427.667997\n",
      "1472.277891329226 37.6635713246222 22.760242110044818\n",
      "0.4309322033898305 0.6554378531073446 0.8884180790960452 0.9662429378531073\n",
      "Train Epoch: 25 [1500/2376 (62%)]\tLoss: 2092.478577\n",
      "1375.0531757281074 36.399023107982934 22.153108610754952\n",
      "0.4320621468926554 0.661723163841808 0.8945621468926553 0.9688559322033898\n",
      "Train Epoch: 26 [1500/2376 (62%)]\tLoss: 718.130533\n",
      "1278.2007633324024 35.10744335885973 21.121182019447893\n",
      "0.45861581920903954 0.6814265536723164 0.8990819209039548 0.9704802259887005\n",
      "Train Epoch: 27 [1500/2376 (62%)]\tLoss: 2002.476012\n",
      "1232.993169078244 34.463934366676185 21.658141738050467\n",
      "0.4166666666666667 0.6556497175141243 0.9005649717514125 0.9733757062146893\n",
      "Train Epoch: 28 [1500/2376 (62%)]\tLoss: 1369.178425\n",
      "1128.4602612445406 32.99320060267148 20.03802122567855\n",
      "0.46850282485875705 0.6928672316384181 0.9093926553672317 0.9752118644067796\n",
      "Train Epoch: 29 [1500/2376 (62%)]\tLoss: 1213.038744\n",
      "1060.6128255482404 31.93978577398285 19.444525131321534\n",
      "0.4754237288135593 0.7016242937853108 0.9149717514124294 0.9771892655367231\n",
      "Train Epoch: 30 [1500/2376 (62%)]\tLoss: 1220.606869\n",
      "1066.5155150893524 32.159071611281526 19.6607405251813\n",
      "0.47923728813559324 0.6898305084745763 0.9090395480225989 0.9778248587570622\n",
      "Train Epoch: 31 [1500/2376 (62%)]\tLoss: 1107.892345\n",
      "965.5745600453234 30.47159826643017 18.763483469038462\n",
      "0.482909604519774 0.7074152542372881 0.9205508474576272 0.9797316384180791\n",
      "Train Epoch: 32 [1500/2376 (62%)]\tLoss: 2349.402490\n",
      "957.9836846351134 30.37389186823004 19.324994993602253\n",
      "0.44795197740112996 0.6858757062146893 0.9222457627118644 0.9813559322033898\n",
      "Train Epoch: 33 [1500/2376 (62%)]\tLoss: 1560.375655\n",
      "896.720932862766 29.38446938706727 18.452812895760772\n",
      "0.4717514124293785 0.7088276836158192 0.9268361581920904 0.9826271186440678\n",
      "Train Epoch: 34 [1500/2376 (62%)]\tLoss: 1909.292863\n",
      "846.3751184882001 28.59189978876741 17.611456861014798\n",
      "0.5022598870056497 0.7273305084745763 0.929731638418079 0.9833333333333333\n",
      "Train Epoch: 35 [1500/2376 (62%)]\tLoss: 1649.991135\n",
      "812.1621125503309 27.931033519799417 17.319201404303065\n",
      "0.5074858757062147 0.7277542372881356 0.9348163841807909 0.9853107344632769\n",
      "Train Epoch: 36 [1500/2376 (62%)]\tLoss: 868.537033\n",
      "793.9300254213865 27.58090365111054 17.437172897551093\n",
      "0.48806497175141245 0.724364406779661 0.9363700564971752 0.98545197740113\n",
      "Train Epoch: 37 [1500/2376 (62%)]\tLoss: 1108.864235\n",
      "759.1236483771161 27.1023455820171 16.7752864890053\n",
      "0.518361581920904 0.7378531073446327 0.9361581920903955 0.9870056497175141\n",
      "Train Epoch: 38 [1500/2376 (62%)]\tLoss: 1435.332939\n",
      "730.7966566282573 26.49668068034464 16.85534495257339\n",
      "0.492090395480226 0.7346751412429379 0.9417372881355932 0.9876412429378532\n",
      "Train Epoch: 39 [1500/2376 (62%)]\tLoss: 720.248168\n",
      "774.2476019452542 27.555676202827186 17.36543999378486\n",
      "0.5024717514124294 0.721680790960452 0.9315677966101695 0.9872175141242938\n",
      "Train Epoch: 40 [1500/2376 (62%)]\tLoss: 1575.023780\n",
      "713.2876066721885 26.362003560369367 16.57634952114963\n",
      "0.5163135593220339 0.7403248587570621 0.9383474576271187 0.9882062146892655\n",
      "Train Epoch: 41 [1500/2376 (62%)]\tLoss: 3738.793009\n",
      "647.1671435521744 25.071444305888786 16.127790049915607\n",
      "0.5038135593220339 0.7387005649717514 0.9475282485875707 0.9903954802259887\n",
      "Train Epoch: 42 [1500/2376 (62%)]\tLoss: 961.011527\n",
      "650.701455562448 25.1362733488607 15.974596807178617\n",
      "0.5187853107344633 0.7427259887005649 0.9457627118644067 0.9905367231638418\n",
      "Train Epoch: 43 [1500/2376 (62%)]\tLoss: 2706.060684\n",
      "674.1724500029575 25.40064629241506 16.179827787185186\n",
      "0.5095338983050848 0.7415960451977401 0.9463983050847458 0.9893361581920904\n",
      "Train Epoch: 44 [1500/2376 (62%)]\tLoss: 1044.623970\n",
      "600.2308799894647 24.112373752036582 15.658997934936574\n",
      "0.5015536723163841 0.7555790960451978 0.95204802259887 0.9913841807909605\n",
      "Train Epoch: 45 [1500/2376 (62%)]\tLoss: 1364.547680\n",
      "685.2460233605772 25.840244375942557 16.889586908953035\n",
      "0.496680790960452 0.7149717514124294 0.9393361581920904 0.9916666666666667\n",
      "Train Epoch: 46 [1500/2376 (62%)]\tLoss: 920.585971\n",
      "579.1636029923948 23.87561063752499 15.430514133893462\n",
      "0.5241525423728813 0.753954802259887 0.9497881355932203 0.9931497175141243\n",
      "Train Epoch: 47 [1500/2376 (62%)]\tLoss: 947.156645\n",
      "542.4155245853192 23.084758373655912 14.902358701526618\n",
      "0.5371468926553672 0.76454802259887 0.9550141242937853 0.9930084745762712\n",
      "Train Epoch: 48 [1500/2376 (62%)]\tLoss: 1312.426651\n",
      "529.3154216258357 22.825899445566858 14.828797179277222\n",
      "0.5310734463276836 0.7661723163841808 0.955225988700565 0.9938559322033899\n",
      "Train Epoch: 49 [1500/2376 (62%)]\tLoss: 881.787831\n",
      "561.178514406384 23.52526761536415 15.501400671699091\n",
      "0.5190677966101694 0.7374293785310735 0.951906779661017 0.9937853107344633\n",
      "Train Epoch: 50 [1500/2376 (62%)]\tLoss: 1463.694655\n",
      "570.7461422242501 23.707127030508783 15.470562642564222\n",
      "0.5223870056497175 0.7441384180790961 0.9516949152542373 0.9932203389830508\n",
      "Train Epoch: 51 [1500/2376 (62%)]\tLoss: 790.278785\n",
      "517.0081583376796 22.505183475610004 14.60802033756076\n",
      "0.5389830508474577 0.7655367231638418 0.9564971751412429 0.9937146892655367\n",
      "Train Epoch: 52 [1500/2376 (62%)]\tLoss: 907.500279\n",
      "493.7454196962828 22.0545074434853 14.384640819854173\n",
      "0.5411723163841808 0.7745056497175141 0.959816384180791 0.9946327683615819\n",
      "Train Epoch: 53 [1500/2376 (62%)]\tLoss: 384.584542\n",
      "480.57311807518334 21.765067775545486 14.340996411564282\n",
      "0.5436440677966101 0.7712570621468926 0.9608050847457628 0.9958333333333333\n",
      "Train Epoch: 54 [1500/2376 (62%)]\tLoss: 893.110068\n",
      "473.91965329666044 21.49619550760506 14.09761767593256\n",
      "0.5439265536723163 0.7787429378531073 0.9629237288135594 0.9953389830508474\n",
      "Train Epoch: 55 [1500/2376 (62%)]\tLoss: 901.778412\n",
      "465.03468411382136 21.377907319396897 14.041050520651359\n",
      "0.5447033898305085 0.7796610169491526 0.9627118644067797 0.9953389830508474\n",
      "Train Epoch: 56 [1500/2376 (62%)]\tLoss: 678.483412\n",
      "452.0374845961509 21.082358796459552 13.913146829223148\n",
      "0.5449858757062147 0.7823446327683616 0.963912429378531 0.9961864406779661\n",
      "Train Epoch: 57 [1500/2376 (62%)]\tLoss: 647.889720\n",
      "461.9537079059737 21.22515948519224 14.02218039140294\n",
      "0.5390536723163842 0.7827683615819209 0.9641949152542373 0.9955508474576271\n",
      "Train Epoch: 58 [1500/2376 (62%)]\tLoss: 555.044692\n",
      "476.4946854242197 21.694303083381985 14.38061277105577\n",
      "0.5406073446327684 0.7673022598870056 0.9603813559322034 0.9959039548022599\n",
      "Train Epoch: 59 [1500/2376 (62%)]\tLoss: 631.092811\n",
      "454.4685669114236 21.14219845095402 13.92666188374857\n",
      "0.5491525423728814 0.7800141242937854 0.9633474576271186 0.996045197740113\n",
      "Train Epoch: 60 [1500/2376 (62%)]\tLoss: 1533.587407\n",
      "434.30116491526843 20.68315998415761 13.741014455758457\n",
      "0.5504943502824858 0.7818502824858757 0.9667372881355932 0.9970338983050847\n",
      "Train Epoch: 61 [1500/2376 (62%)]\tLoss: 1017.368611\n",
      "456.5254048283247 21.183036059107348 14.055526376477934\n",
      "0.5412429378531074 0.7752118644067797 0.963912429378531 0.9958333333333333\n",
      "Train Epoch: 62 [1500/2376 (62%)]\tLoss: 940.263380\n",
      "434.80388717945317 20.57580511144285 13.631350701813052\n",
      "0.5484463276836158 0.7865819209039548 0.9683615819209039 0.9961158192090396\n",
      "Train Epoch: 63 [1500/2376 (62%)]\tLoss: 1733.732477\n",
      "425.3685009928953 20.46013616506306 13.607451228474227\n",
      "0.5560028248587571 0.7872175141242937 0.9660310734463277 0.9970338983050847\n",
      "Train Epoch: 64 [1500/2376 (62%)]\tLoss: 775.587702\n",
      "434.7097369615926 20.699903713029972 13.741665614592076\n",
      "0.5528248587570621 0.7833333333333333 0.9653248587570622 0.9964689265536724\n",
      "Train Epoch: 65 [1500/2376 (62%)]\tLoss: 982.358792\n",
      "420.2689041185101 20.286122353686533 13.4692393903207\n",
      "0.5550847457627118 0.7896892655367231 0.9690677966101695 0.996822033898305\n",
      "Train Epoch: 66 [1500/2376 (62%)]\tLoss: 935.646749\n",
      "429.78329822388076 20.593946834333 13.781772720658992\n",
      "0.5500706214689266 0.7779661016949152 0.966454802259887 0.9971751412429378\n",
      "Train Epoch: 67 [1500/2376 (62%)]\tLoss: 490.589669\n",
      "424.04716079227626 20.429348981811014 13.695183633157205\n",
      "0.5458333333333333 0.7805790960451977 0.9684322033898305 0.9969632768361582\n",
      "Train Epoch: 68 [1500/2376 (62%)]\tLoss: 759.420540\n",
      "412.2538020245599 20.132418608244457 13.51028332666425\n",
      "0.5494350282485876 0.7826977401129943 0.9694915254237289 0.9973870056497175\n",
      "Train Epoch: 69 [1500/2376 (62%)]\tLoss: 647.912185\n",
      "411.4160023563998 20.056617543582398 13.373439637664521\n",
      "0.5514830508474576 0.7911016949152543 0.9699858757062146 0.9968926553672316\n",
      "Train Epoch: 70 [1500/2376 (62%)]\tLoss: 803.569441\n",
      "416.94817035959204 20.26812218779909 13.573864179909709\n",
      "0.5543079096045198 0.7848870056497175 0.9681497175141243 0.9970338983050847\n",
      "Train Epoch: 71 [1500/2376 (62%)]\tLoss: 864.790279\n",
      "432.50944173239907 20.678515209992888 13.955890576734607\n",
      "0.5411723163841808 0.7698446327683616 0.9660310734463277 0.997457627118644\n",
      "Train Epoch: 72 [1500/2376 (62%)]\tLoss: 1020.112137\n",
      "412.268521654 20.13835536464426 13.551339632643419\n",
      "0.5496468926553673 0.7829802259887005 0.9697033898305085 0.9970338983050847\n",
      "Train Epoch: 73 [1500/2376 (62%)]\tLoss: 776.895987\n",
      "463.01393962890853 21.392420660404465 14.310990728271594\n",
      "0.5382062146892655 0.7626412429378531 0.9615112994350282 0.9965395480225989\n",
      "Train Epoch: 74 [1500/2376 (62%)]\tLoss: 777.234670\n",
      "413.426071014811 20.153144642918615 13.574868332918498\n",
      "0.5462570621468926 0.7810734463276836 0.9692090395480226 0.9973870056497175\n",
      "Train Epoch: 75 [1500/2376 (62%)]\tLoss: 437.851469\n",
      "426.2556811024051 20.500451271934157 13.738792684736982\n",
      "0.5516949152542373 0.7799435028248588 0.9667372881355932 0.9964689265536724\n",
      "Train Epoch: 76 [1500/2376 (62%)]\tLoss: 776.193987\n",
      "399.5170231216415 19.77797576391043 13.481172022510941\n",
      "0.5459745762711864 0.7824858757062146 0.9712570621468927 0.9975988700564972\n",
      "Train Epoch: 77 [1500/2376 (62%)]\tLoss: 639.173419\n",
      "426.8063485909518 20.544945403575916 13.98599096441922\n",
      "0.5358757062146893 0.7657485875706215 0.9667372881355932 0.9976694915254237\n",
      "Train Epoch: 78 [1500/2376 (62%)]\tLoss: 775.373422\n",
      "392.5035007937726 19.647849827992125 13.190434022018684\n",
      "0.561087570621469 0.7934322033898306 0.971045197740113 0.997457627118644\n",
      "Train Epoch: 79 [1500/2376 (62%)]\tLoss: 591.563077\n",
      "379.8998551464891 19.26929174621764 12.933319804236936\n",
      "0.5655367231638418 0.8028248587570621 0.9722457627118644 0.9975988700564972\n",
      "Train Epoch: 80 [1500/2376 (62%)]\tLoss: 1183.150685\n",
      "380.71653759239393 19.321768988409186 13.004450809833486\n",
      "0.5632768361581921 0.7988700564971751 0.971680790960452 0.9980932203389831\n"
     ],
     "output_type": "stream"
    }
   ],
   "source": [
    "epoch_losses = []\n",
    "epoch_test_acc10 = []\n",
    "epoch_test_acc20 = []\n",
    "epoch_test_acc50 = []\n",
    "epoch_test_acc100 = []\n",
    "for epoch in range(epochs):\n",
    "    loss = train(model, device, train_dataloader, optimizer, epoch, log_interval)\n",
    "    epoch_losses.append(loss)\n",
    "    mse_losses, rmse_losses, mae_losses, acc10, acc20, acc50, acc100 = test(model, device, validation_dataloader)\n",
    "    epoch_test_acc10.append(acc10)\n",
    "    epoch_test_acc20.append(acc20)\n",
    "    epoch_test_acc50.append(acc50)\n",
    "    epoch_test_acc100.append(acc100)\n",
    "    "
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "outputs": [
    {
     "data": {
      "text/plain": "[<matplotlib.lines.Line2D at 0x1c401be4808>]"
     },
     "metadata": {},
     "output_type": "execute_result",
     "execution_count": 32
    },
    {
     "data": {
      "text/plain": "<Figure size 432x288 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "A = np.array(epoch_losses)\n",
    "plt.plot(A.flatten())"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 432x288 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "B = np.array(epoch_test_acc100)\n",
    "C = np.array(epoch_test_acc50)\n",
    "D = np.array(epoch_test_acc20)\n",
    "E = np.array(epoch_test_acc10)\n",
    "x = range(1,81)\n",
    "plt.plot(x, B.flatten(), label = \"+-100\")\n",
    "plt.plot(x, C.flatten(), label = \"+-50\")\n",
    "plt.plot(x, D.flatten(), label = \"+-20\")\n",
    "plt.plot(x, E.flatten(), label = \"+-10\")\n",
    "plt.xlabel('epochs')\n",
    "plt.ylabel('accurarcy')\n",
    "plt.legend(bbox_to_anchor=(0.7, 0.7))\n",
    "plt.show()"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "outputs": [
    {
     "name": "stdout",
     "text": [
      "(594, 48, 27)\n"
     ],
     "output_type": "stream"
    }
   ],
   "source": [
    "#inference on test data\n",
    "test_data = df_test.values.reshape(12,-1,27)[:,-5232:,:]\n",
    "test_data = test_data.reshape(-1 ,48,27)\n",
    "test_data = test_data[~np.isnan(test_data).any(axis=(1,2))]\n",
    "print(test_data.shape)\n",
    "test_set = TimeSeriesDataset(test_data)\n",
    "test_dataloader = tud.DataLoader(test_set, batch_size=batch_size)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "outputs": [
    {
     "name": "stdout",
     "text": [
      "670.1655437208359 24.896799354354805 15.453512926879284\n",
      "0.5617985409652076 0.7612934904601572 0.9424102132435466 0.9887065095398428\n"
     ],
     "output_type": "stream"
    }
   ],
   "source": [
    "se_losses, rmse_losses, mae_losses, acc10, acc20, acc50, acc100 = test(model, device, test_dataloader)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n",
     "is_executing": false
    }
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  },
  "pycharm": {
   "stem_cell": {
    "cell_type": "raw",
    "source": [],
    "metadata": {
     "collapsed": false
    }
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}