Lab2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 實驗一：房價預測模型"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<table class=\"tfo-notebook-buttons\" align=\"left\">\n",
    "  <td>\n",
    "    <a target=\"_blank\" href=\"https://colab.research.google.com/github/taipeitechmmslab/MMSLAB-TF2/blob/master/Lab2.ipynb\"><img src=\"https://www.tensorflow.org/images/colab_logo_32px.png\" />Run in Google Colab</a>\n",
    "  </td>\n",
    "  <td>\n",
    "    <a target=\"_blank\" href=\"https://github.com/taipeitechmmslab/MMSLAB-TF2/blob/master/Lab2.ipynb\"><img src=\"https://www.tensorflow.org/images/GitHub-Mark-32px.png\" />View source on GitHub</a>\n",
    "  </td>\n",
    "</table>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Import必要套件"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import tensorflow as tf\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from tensorflow import keras\n",
    "from tensorflow.keras import layers"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 下載 House Sales in King County, USA 資料集\n",
    "\n",
    "https://www.kaggle.com/harlfoxem/housesalesprediction"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 數據讀取並分析"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(21613, 21)"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = pd.read_csv(\"kc_house_data.csv\")\n",
    "# 顯示dataset的形狀，共21613比資料，每一比資料有21種不同資訊。\n",
    "data.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>date</th>\n",
       "      <th>price</th>\n",
       "      <th>bedrooms</th>\n",
       "      <th>bathrooms</th>\n",
       "      <th>sqft_living</th>\n",
       "      <th>sqft_lot</th>\n",
       "      <th>floors</th>\n",
       "      <th>waterfront</th>\n",
       "      <th>view</th>\n",
       "      <th>condition</th>\n",
       "      <th>grade</th>\n",
       "      <th>sqft_above</th>\n",
       "      <th>sqft_basement</th>\n",
       "      <th>yr_built</th>\n",
       "      <th>yr_renovated</th>\n",
       "      <th>zipcode</th>\n",
       "      <th>lat</th>\n",
       "      <th>long</th>\n",
       "      <th>sqft_living15</th>\n",
       "      <th>sqft_lot15</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>7129300520</td>\n",
       "      <td>20141013T000000</td>\n",
       "      <td>221900.0</td>\n",
       "      <td>3</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1180</td>\n",
       "      <td>5650</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>7</td>\n",
       "      <td>1180</td>\n",
       "      <td>0</td>\n",
       "      <td>1955</td>\n",
       "      <td>0</td>\n",
       "      <td>98178</td>\n",
       "      <td>47.5112</td>\n",
       "      <td>-122.257</td>\n",
       "      <td>1340</td>\n",
       "      <td>5650</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6414100192</td>\n",
       "      <td>20141209T000000</td>\n",
       "      <td>538000.0</td>\n",
       "      <td>3</td>\n",
       "      <td>2.25</td>\n",
       "      <td>2570</td>\n",
       "      <td>7242</td>\n",
       "      <td>2.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>7</td>\n",
       "      <td>2170</td>\n",
       "      <td>400</td>\n",
       "      <td>1951</td>\n",
       "      <td>1991</td>\n",
       "      <td>98125</td>\n",
       "      <td>47.7210</td>\n",
       "      <td>-122.319</td>\n",
       "      <td>1690</td>\n",
       "      <td>7639</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>5631500400</td>\n",
       "      <td>20150225T000000</td>\n",
       "      <td>180000.0</td>\n",
       "      <td>2</td>\n",
       "      <td>1.00</td>\n",
       "      <td>770</td>\n",
       "      <td>10000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>6</td>\n",
       "      <td>770</td>\n",
       "      <td>0</td>\n",
       "      <td>1933</td>\n",
       "      <td>0</td>\n",
       "      <td>98028</td>\n",
       "      <td>47.7379</td>\n",
       "      <td>-122.233</td>\n",
       "      <td>2720</td>\n",
       "      <td>8062</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2487200875</td>\n",
       "      <td>20141209T000000</td>\n",
       "      <td>604000.0</td>\n",
       "      <td>4</td>\n",
       "      <td>3.00</td>\n",
       "      <td>1960</td>\n",
       "      <td>5000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>7</td>\n",
       "      <td>1050</td>\n",
       "      <td>910</td>\n",
       "      <td>1965</td>\n",
       "      <td>0</td>\n",
       "      <td>98136</td>\n",
       "      <td>47.5208</td>\n",
       "      <td>-122.393</td>\n",
       "      <td>1360</td>\n",
       "      <td>5000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1954400510</td>\n",
       "      <td>20150218T000000</td>\n",
       "      <td>510000.0</td>\n",
       "      <td>3</td>\n",
       "      <td>2.00</td>\n",
       "      <td>1680</td>\n",
       "      <td>8080</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>8</td>\n",
       "      <td>1680</td>\n",
       "      <td>0</td>\n",
       "      <td>1987</td>\n",
       "      <td>0</td>\n",
       "      <td>98074</td>\n",
       "      <td>47.6168</td>\n",
       "      <td>-122.045</td>\n",
       "      <td>1800</td>\n",
       "      <td>7503</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           id             date     price  bedrooms  bathrooms  sqft_living  \\\n",
       "0  7129300520  20141013T000000  221900.0         3       1.00         1180   \n",
       "1  6414100192  20141209T000000  538000.0         3       2.25         2570   \n",
       "2  5631500400  20150225T000000  180000.0         2       1.00          770   \n",
       "3  2487200875  20141209T000000  604000.0         4       3.00         1960   \n",
       "4  1954400510  20150218T000000  510000.0         3       2.00         1680   \n",
       "\n",
       "   sqft_lot  floors  waterfront  view  condition  grade  sqft_above  \\\n",
       "0      5650     1.0           0     0          3      7        1180   \n",
       "1      7242     2.0           0     0          3      7        2170   \n",
       "2     10000     1.0           0     0          3      6         770   \n",
       "3      5000     1.0           0     0          5      7        1050   \n",
       "4      8080     1.0           0     0          3      8        1680   \n",
       "\n",
       "   sqft_basement  yr_built  yr_renovated  zipcode      lat     long  \\\n",
       "0              0      1955             0    98178  47.5112 -122.257   \n",
       "1            400      1951          1991    98125  47.7210 -122.319   \n",
       "2              0      1933             0    98028  47.7379 -122.233   \n",
       "3            910      1965             0    98136  47.5208 -122.393   \n",
       "4              0      1987             0    98074  47.6168 -122.045   \n",
       "\n",
       "   sqft_living15  sqft_lot15  \n",
       "0           1340        5650  \n",
       "1           1690        7639  \n",
       "2           2720        8062  \n",
       "3           1360        5000  \n",
       "4           1800        7503  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 將顯示列數設定為25，不然會有部份資料無法顯示\n",
    "pd.options.display.max_columns = 25\n",
    "# head 會顯示前五行的數據\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "各個數據的簡寫分別代表下面意思：\n",
    "- date：房屋出售日期。\n",
    "- price：房屋價格（目標）。\n",
    "- bedrooms：臥室數量。\n",
    "- bathrooms：浴室數量。\n",
    "- sqft_living：居住的坪數（平方英尺）。\n",
    "- sqft_lot：實際的坪數（平方英尺）。\n",
    "- floors：房屋總共樓層。\n",
    "- waterfront：海景房。\n",
    "- view：房屋是否看過。\n",
    "- condition：整體條件有多好。\n",
    "- grade：房屋的整體等級（根據King County評分系統）。\n",
    "- sqft_above：除了地下室外的坪數（平方英尺）。\n",
    "- sqft_basement：地下室的坪數（平方英尺）。\n",
    "- yr_built：房屋建造時間。\n",
    "- yr_renovated：何時重新裝修過（一些沒重新裝修過或是裝修紀錄沒被記錄到的數值都為0）。\n",
    "- zipcode：郵政編碼。\n",
    "- lat：緯度座標。\n",
    "- long：經度座標。\n",
    "- sqft_living15：2015年紀錄的居住坪數（可能是翻新的原因導致sqft_living15與sqft_living不同）。\n",
    "- sqft_lot15：2015年紀錄的實際坪數（可能是翻新的原因導致sqft_lot15與sqft_lot不同）。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 檢查資料的型態\n",
    "\n",
    "資料型態總共有五種：object(string),booleab, integer, float and categorical."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "id                 int64\n",
       "date              object\n",
       "price            float64\n",
       "bedrooms           int64\n",
       "bathrooms        float64\n",
       "sqft_living        int64\n",
       "sqft_lot           int64\n",
       "floors           float64\n",
       "waterfront         int64\n",
       "view               int64\n",
       "condition          int64\n",
       "grade              int64\n",
       "sqft_above         int64\n",
       "sqft_basement      int64\n",
       "yr_built           int64\n",
       "yr_renovated       int64\n",
       "zipcode            int64\n",
       "lat              float64\n",
       "long             float64\n",
       "sqft_living15      int64\n",
       "sqft_lot15         int64\n",
       "dtype: object"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.dtypes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 數據前處理\n",
    "轉換資料型態：\n",
    "因為數據集裡的date數據是字串（string）格式，而模型的輸入只接受數值格式，所以可以透過以下程式碼將其轉為數值，並分成年、月及日三種數據。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>price</th>\n",
       "      <th>bedrooms</th>\n",
       "      <th>bathrooms</th>\n",
       "      <th>sqft_living</th>\n",
       "      <th>sqft_lot</th>\n",
       "      <th>floors</th>\n",
       "      <th>waterfront</th>\n",
       "      <th>view</th>\n",
       "      <th>condition</th>\n",
       "      <th>grade</th>\n",
       "      <th>sqft_above</th>\n",
       "      <th>sqft_basement</th>\n",
       "      <th>yr_built</th>\n",
       "      <th>yr_renovated</th>\n",
       "      <th>zipcode</th>\n",
       "      <th>lat</th>\n",
       "      <th>long</th>\n",
       "      <th>sqft_living15</th>\n",
       "      <th>sqft_lot15</th>\n",
       "      <th>year</th>\n",
       "      <th>month</th>\n",
       "      <th>day</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>221900.0</td>\n",
       "      <td>3</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1180</td>\n",
       "      <td>5650</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>7</td>\n",
       "      <td>1180</td>\n",
       "      <td>0</td>\n",
       "      <td>1955</td>\n",
       "      <td>0</td>\n",
       "      <td>98178</td>\n",
       "      <td>47.5112</td>\n",
       "      <td>-122.257</td>\n",
       "      <td>1340</td>\n",
       "      <td>5650</td>\n",
       "      <td>2014</td>\n",
       "      <td>10</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>538000.0</td>\n",
       "      <td>3</td>\n",
       "      <td>2.25</td>\n",
       "      <td>2570</td>\n",
       "      <td>7242</td>\n",
       "      <td>2.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>7</td>\n",
       "      <td>2170</td>\n",
       "      <td>400</td>\n",
       "      <td>1951</td>\n",
       "      <td>1991</td>\n",
       "      <td>98125</td>\n",
       "      <td>47.7210</td>\n",
       "      <td>-122.319</td>\n",
       "      <td>1690</td>\n",
       "      <td>7639</td>\n",
       "      <td>2014</td>\n",
       "      <td>12</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>180000.0</td>\n",
       "      <td>2</td>\n",
       "      <td>1.00</td>\n",
       "      <td>770</td>\n",
       "      <td>10000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>6</td>\n",
       "      <td>770</td>\n",
       "      <td>0</td>\n",
       "      <td>1933</td>\n",
       "      <td>0</td>\n",
       "      <td>98028</td>\n",
       "      <td>47.7379</td>\n",
       "      <td>-122.233</td>\n",
       "      <td>2720</td>\n",
       "      <td>8062</td>\n",
       "      <td>2015</td>\n",
       "      <td>2</td>\n",
       "      <td>25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>604000.0</td>\n",
       "      <td>4</td>\n",
       "      <td>3.00</td>\n",
       "      <td>1960</td>\n",
       "      <td>5000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>7</td>\n",
       "      <td>1050</td>\n",
       "      <td>910</td>\n",
       "      <td>1965</td>\n",
       "      <td>0</td>\n",
       "      <td>98136</td>\n",
       "      <td>47.5208</td>\n",
       "      <td>-122.393</td>\n",
       "      <td>1360</td>\n",
       "      <td>5000</td>\n",
       "      <td>2014</td>\n",
       "      <td>12</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>510000.0</td>\n",
       "      <td>3</td>\n",
       "      <td>2.00</td>\n",
       "      <td>1680</td>\n",
       "      <td>8080</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>8</td>\n",
       "      <td>1680</td>\n",
       "      <td>0</td>\n",
       "      <td>1987</td>\n",
       "      <td>0</td>\n",
       "      <td>98074</td>\n",
       "      <td>47.6168</td>\n",
       "      <td>-122.045</td>\n",
       "      <td>1800</td>\n",
       "      <td>7503</td>\n",
       "      <td>2015</td>\n",
       "      <td>2</td>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      price  bedrooms  bathrooms  sqft_living  sqft_lot  floors  waterfront  \\\n",
       "0  221900.0         3       1.00         1180      5650     1.0           0   \n",
       "1  538000.0         3       2.25         2570      7242     2.0           0   \n",
       "2  180000.0         2       1.00          770     10000     1.0           0   \n",
       "3  604000.0         4       3.00         1960      5000     1.0           0   \n",
       "4  510000.0         3       2.00         1680      8080     1.0           0   \n",
       "\n",
       "   view  condition  grade  sqft_above  sqft_basement  yr_built  yr_renovated  \\\n",
       "0     0          3      7        1180              0      1955             0   \n",
       "1     0          3      7        2170            400      1951          1991   \n",
       "2     0          3      6         770              0      1933             0   \n",
       "3     0          5      7        1050            910      1965             0   \n",
       "4     0          3      8        1680              0      1987             0   \n",
       "\n",
       "   zipcode      lat     long  sqft_living15  sqft_lot15  year  month  day  \n",
       "0    98178  47.5112 -122.257           1340        5650  2014     10   13  \n",
       "1    98125  47.7210 -122.319           1690        7639  2014     12    9  \n",
       "2    98028  47.7379 -122.233           2720        8062  2015      2   25  \n",
       "3    98136  47.5208 -122.393           1360        5000  2014     12    9  \n",
       "4    98074  47.6168 -122.045           1800        7503  2015      2   18  "
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 將date日期拆為年、月和日並轉成數值\n",
    "data['year'] = pd.to_numeric(data['date'].str.slice(0, 4))\n",
    "data['month'] = pd.to_numeric(data['date'].str.slice(4, 6))\n",
    "data['day'] = pd.to_numeric(data['date'].str.slice(6, 8))\n",
    "\n",
    "# 刪除沒有用的數據，inplace則是將更新後的資料存回原本的地方\n",
    "data.drop(['id'], axis=\"columns\", inplace=True)\n",
    "data.drop(['date'], axis=\"columns\", inplace=True)\n",
    "\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "分割數據集（Dataset）：將數據集切割成三個部份，訓練數據（Training data）、驗證數據（Validation data）和測試數據（Testing data）。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_num = data.shape[0]\n",
    "# 取得一筆與data數量相同的亂數索引，主要目的是用於打散資料\n",
    "indexes = np.random.permutation(data_num)\n",
    "# 並將亂數索引值分為Train、validation和test分為，這裡的劃分比例為6:2:2\n",
    "train_indexes = indexes[:int(data_num *0.6)]\n",
    "val_indexes = indexes[int(data_num *0.6):int(data_num *0.8)]\n",
    "test_indexes = indexes[int(data_num *0.8):]\n",
    "# 透過索引值從data取出訓練資料、驗證資料和測試資料\n",
    "train_data = data.loc[train_indexes]\n",
    "val_data = data.loc[val_indexes]\n",
    "test_data = data.loc[test_indexes]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Normalization 正規化"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "使用標準分數(Standard Score, 又稱z-score)將數據正規化，經過z-score正規化後數據的都會聚集在0附近， 標準差為1。 \n",
    "\n",
    "(x - 平均值) / 標準差"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "train_validation_data = pd.concat([train_data, val_data])\n",
    "mean = train_validation_data.mean()\n",
    "std = train_validation_data.std()\n",
    "\n",
    "train_data = (train_data - mean) / std\n",
    "val_data = (val_data - mean) / std"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 建立Numpy array格式的訓練數據"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train = np.array(train_data.drop('price', axis='columns'))\n",
    "y_train = np.array(train_data['price'])\n",
    "x_val = np.array(val_data.drop('price', axis='columns'))\n",
    "y_val = np.array(val_data['price'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "整理過後的資料共12967筆，且一筆資料有21種資訊(所以網路輸入必須為21)。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(12967, 21)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_train.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 建立並訓練網路模型\n",
    "\n",
    "這裡建構三層全連接層的網路架構，並且使用ReLU作為隱藏層的激活函數，而由於需得到線性輸出，故輸出層不使用任何激活函數。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model-1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense (Dense)                (None, 64)                1408      \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 64)                4160      \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 1)                 65        \n",
      "=================================================================\n",
      "Total params: 5,633\n",
      "Trainable params: 5,633\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# 建立一個Sequential型態的model\n",
    "model = keras.Sequential(name='model-1')\n",
    "# 第1層全連接層設為64個unit，將輸入形狀設定為(21, )，而實際上我們輸入的數據形狀為(batch_size, 21)\n",
    "model.add(layers.Dense(64, activation='relu', input_shape=(21,)))\n",
    "# 第2層全連接層設為64個unit\n",
    "model.add(layers.Dense(64, activation='relu'))\n",
    "# 最後一層全連接層設為1個unit\n",
    "model.add(layers.Dense(1))\n",
    "# 顯示網路模型架構\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "設定訓練使用的優化器、損失函數和指標函數："
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(keras.optimizers.Adam(0.001),\n",
    "              loss=keras.losses.MeanSquaredError(),\n",
    "              metrics=[keras.metrics.MeanAbsoluteError()])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "創建模型儲存目錄："
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_dir = 'lab2-logs/models/'\n",
    "os.makedirs(model_dir)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "設定回調函數："
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# TensorBoard回調函數會幫忙紀錄訓練資訊，並存成TensorBoard的紀錄檔\n",
    "log_dir = os.path.join('lab2-logs', 'model-1')\n",
    "model_cbk = keras.callbacks.TensorBoard(log_dir=log_dir)\n",
    "# ModelCheckpoint回調函數幫忙儲存網路模型，可以設定只儲存最好的模型，「monitor」表示被監測的數據，「mode」min則代表監測數據越小越好。\n",
    "model_mckp = keras.callbacks.ModelCheckpoint(model_dir + '/Best-model-1.h5', \n",
    "                                        monitor='val_mean_absolute_error', \n",
    "                                        save_best_only=True, \n",
    "                                        mode='min')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "訓練網路模型："
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 12967 samples, validate on 4323 samples\n",
      "Epoch 1/300\n",
      "12967/12967 [==============================] - 2s 124us/sample - loss: 0.3483 - mean_absolute_error: 0.3535 - val_loss: 0.2466 - val_mean_absolute_error: 0.3046\n",
      "Epoch 2/300\n",
      "12967/12967 [==============================] - 1s 43us/sample - loss: 0.2000 - mean_absolute_error: 0.2795 - val_loss: 0.2147 - val_mean_absolute_error: 0.2737\n",
      "Epoch 3/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.1750 - mean_absolute_error: 0.2570 - val_loss: 0.1963 - val_mean_absolute_error: 0.2717\n",
      "Epoch 4/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.1577 - mean_absolute_error: 0.2451 - val_loss: 0.2054 - val_mean_absolute_error: 0.2646\n",
      "Epoch 5/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1408 - mean_absolute_error: 0.2317 - val_loss: 0.1785 - val_mean_absolute_error: 0.2420\n",
      "Epoch 6/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.1308 - mean_absolute_error: 0.2234 - val_loss: 0.1623 - val_mean_absolute_error: 0.2354\n",
      "Epoch 7/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1221 - mean_absolute_error: 0.2150 - val_loss: 0.1678 - val_mean_absolute_error: 0.2422\n",
      "Epoch 8/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1181 - mean_absolute_error: 0.2136 - val_loss: 0.1626 - val_mean_absolute_error: 0.2261\n",
      "Epoch 9/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1102 - mean_absolute_error: 0.2069 - val_loss: 0.1551 - val_mean_absolute_error: 0.2304\n",
      "Epoch 10/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.1045 - mean_absolute_error: 0.2022 - val_loss: 0.1530 - val_mean_absolute_error: 0.2238\n",
      "Epoch 11/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.1023 - mean_absolute_error: 0.1995 - val_loss: 0.1519 - val_mean_absolute_error: 0.2177\n",
      "Epoch 12/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.0999 - mean_absolute_error: 0.1987 - val_loss: 0.1479 - val_mean_absolute_error: 0.2162\n",
      "Epoch 13/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0918 - mean_absolute_error: 0.1916 - val_loss: 0.1522 - val_mean_absolute_error: 0.2220\n",
      "Epoch 14/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0913 - mean_absolute_error: 0.1903 - val_loss: 0.1585 - val_mean_absolute_error: 0.2166\n",
      "Epoch 15/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0878 - mean_absolute_error: 0.1870 - val_loss: 0.1436 - val_mean_absolute_error: 0.2110\n",
      "Epoch 16/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0830 - mean_absolute_error: 0.1836 - val_loss: 0.1405 - val_mean_absolute_error: 0.2095\n",
      "Epoch 17/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0832 - mean_absolute_error: 0.1837 - val_loss: 0.1433 - val_mean_absolute_error: 0.2110\n",
      "Epoch 18/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0806 - mean_absolute_error: 0.1812 - val_loss: 0.1518 - val_mean_absolute_error: 0.2216\n",
      "Epoch 19/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0797 - mean_absolute_error: 0.1818 - val_loss: 0.1500 - val_mean_absolute_error: 0.2104\n",
      "Epoch 20/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.0768 - mean_absolute_error: 0.1791 - val_loss: 0.1385 - val_mean_absolute_error: 0.2066\n",
      "Epoch 21/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0739 - mean_absolute_error: 0.1765 - val_loss: 0.1383 - val_mean_absolute_error: 0.2063\n",
      "Epoch 22/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0746 - mean_absolute_error: 0.1787 - val_loss: 0.1427 - val_mean_absolute_error: 0.2114\n",
      "Epoch 23/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0734 - mean_absolute_error: 0.1766 - val_loss: 0.1387 - val_mean_absolute_error: 0.2043\n",
      "Epoch 24/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0712 - mean_absolute_error: 0.1758 - val_loss: 0.1471 - val_mean_absolute_error: 0.2082\n",
      "Epoch 25/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0694 - mean_absolute_error: 0.1738 - val_loss: 0.1490 - val_mean_absolute_error: 0.2085\n",
      "Epoch 26/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0692 - mean_absolute_error: 0.1725 - val_loss: 0.1502 - val_mean_absolute_error: 0.2086\n",
      "Epoch 27/300\n",
      "12967/12967 [==============================] - ETA: 0s - loss: 0.0667 - mean_absolute_error: 0.170 - 0s 34us/sample - loss: 0.0673 - mean_absolute_error: 0.1713 - val_loss: 0.1472 - val_mean_absolute_error: 0.2083\n",
      "Epoch 28/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0651 - mean_absolute_error: 0.1691 - val_loss: 0.1480 - val_mean_absolute_error: 0.2155\n",
      "Epoch 29/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0656 - mean_absolute_error: 0.1696 - val_loss: 0.1482 - val_mean_absolute_error: 0.2104\n",
      "Epoch 30/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0642 - mean_absolute_error: 0.1683 - val_loss: 0.1426 - val_mean_absolute_error: 0.2116\n",
      "Epoch 31/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0616 - mean_absolute_error: 0.1669 - val_loss: 0.1536 - val_mean_absolute_error: 0.2146\n",
      "Epoch 32/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0633 - mean_absolute_error: 0.1674 - val_loss: 0.1469 - val_mean_absolute_error: 0.2058\n",
      "Epoch 33/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0605 - mean_absolute_error: 0.1647 - val_loss: 0.1461 - val_mean_absolute_error: 0.2068\n",
      "Epoch 34/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0601 - mean_absolute_error: 0.1634 - val_loss: 0.1504 - val_mean_absolute_error: 0.2118\n",
      "Epoch 35/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0591 - mean_absolute_error: 0.1640 - val_loss: 0.1424 - val_mean_absolute_error: 0.2077\n",
      "Epoch 36/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0584 - mean_absolute_error: 0.1646 - val_loss: 0.1430 - val_mean_absolute_error: 0.2039\n",
      "Epoch 37/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0586 - mean_absolute_error: 0.1631 - val_loss: 0.1519 - val_mean_absolute_error: 0.2114\n",
      "Epoch 38/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0581 - mean_absolute_error: 0.1635 - val_loss: 0.1523 - val_mean_absolute_error: 0.2177\n",
      "Epoch 39/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0595 - mean_absolute_error: 0.1647 - val_loss: 0.1568 - val_mean_absolute_error: 0.2137\n",
      "Epoch 40/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0564 - mean_absolute_error: 0.1608 - val_loss: 0.1457 - val_mean_absolute_error: 0.2054\n",
      "Epoch 41/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0542 - mean_absolute_error: 0.1589 - val_loss: 0.1501 - val_mean_absolute_error: 0.2057\n",
      "Epoch 42/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0561 - mean_absolute_error: 0.1628 - val_loss: 0.1473 - val_mean_absolute_error: 0.2078\n",
      "Epoch 43/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0538 - mean_absolute_error: 0.1598 - val_loss: 0.1435 - val_mean_absolute_error: 0.2023\n",
      "Epoch 44/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0538 - mean_absolute_error: 0.1583 - val_loss: 0.1484 - val_mean_absolute_error: 0.2048\n",
      "Epoch 45/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0534 - mean_absolute_error: 0.1585 - val_loss: 0.1606 - val_mean_absolute_error: 0.2080\n",
      "Epoch 46/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0550 - mean_absolute_error: 0.1602 - val_loss: 0.1485 - val_mean_absolute_error: 0.2093\n",
      "Epoch 47/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0516 - mean_absolute_error: 0.1566 - val_loss: 0.1502 - val_mean_absolute_error: 0.2114\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 48/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0518 - mean_absolute_error: 0.1552 - val_loss: 0.1473 - val_mean_absolute_error: 0.2047\n",
      "Epoch 49/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0523 - mean_absolute_error: 0.1574 - val_loss: 0.1544 - val_mean_absolute_error: 0.2132\n",
      "Epoch 50/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0510 - mean_absolute_error: 0.1563 - val_loss: 0.1550 - val_mean_absolute_error: 0.2104\n",
      "Epoch 51/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0518 - mean_absolute_error: 0.1576 - val_loss: 0.1509 - val_mean_absolute_error: 0.2110\n",
      "Epoch 52/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0485 - mean_absolute_error: 0.1538 - val_loss: 0.1483 - val_mean_absolute_error: 0.2083\n",
      "Epoch 53/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0500 - mean_absolute_error: 0.1551 - val_loss: 0.1540 - val_mean_absolute_error: 0.2060\n",
      "Epoch 54/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0532 - mean_absolute_error: 0.1591 - val_loss: 0.1502 - val_mean_absolute_error: 0.2059\n",
      "Epoch 55/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0484 - mean_absolute_error: 0.1538 - val_loss: 0.1484 - val_mean_absolute_error: 0.2149\n",
      "Epoch 56/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0493 - mean_absolute_error: 0.1550 - val_loss: 0.1598 - val_mean_absolute_error: 0.2116\n",
      "Epoch 57/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0503 - mean_absolute_error: 0.1534 - val_loss: 0.1449 - val_mean_absolute_error: 0.2060\n",
      "Epoch 58/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0509 - mean_absolute_error: 0.1558 - val_loss: 0.1413 - val_mean_absolute_error: 0.2072\n",
      "Epoch 59/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0468 - mean_absolute_error: 0.1512 - val_loss: 0.1467 - val_mean_absolute_error: 0.2051\n",
      "Epoch 60/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0453 - mean_absolute_error: 0.1492 - val_loss: 0.1533 - val_mean_absolute_error: 0.2116\n",
      "Epoch 61/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0454 - mean_absolute_error: 0.1497 - val_loss: 0.1435 - val_mean_absolute_error: 0.2055\n",
      "Epoch 62/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0432 - mean_absolute_error: 0.1464 - val_loss: 0.1490 - val_mean_absolute_error: 0.2082\n",
      "Epoch 63/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0439 - mean_absolute_error: 0.1465 - val_loss: 0.1416 - val_mean_absolute_error: 0.2052\n",
      "Epoch 64/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0439 - mean_absolute_error: 0.1479 - val_loss: 0.1532 - val_mean_absolute_error: 0.2072\n",
      "Epoch 65/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0482 - mean_absolute_error: 0.1529 - val_loss: 0.1456 - val_mean_absolute_error: 0.2054\n",
      "Epoch 66/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0442 - mean_absolute_error: 0.1478 - val_loss: 0.1450 - val_mean_absolute_error: 0.2081\n",
      "Epoch 67/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0445 - mean_absolute_error: 0.1487 - val_loss: 0.1489 - val_mean_absolute_error: 0.2069\n",
      "Epoch 68/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0476 - mean_absolute_error: 0.1519 - val_loss: 0.1477 - val_mean_absolute_error: 0.2080\n",
      "Epoch 69/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0436 - mean_absolute_error: 0.1479 - val_loss: 0.1486 - val_mean_absolute_error: 0.2104\n",
      "Epoch 70/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0445 - mean_absolute_error: 0.1474 - val_loss: 0.1429 - val_mean_absolute_error: 0.2040\n",
      "Epoch 71/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0424 - mean_absolute_error: 0.1469 - val_loss: 0.1497 - val_mean_absolute_error: 0.2125\n",
      "Epoch 72/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0431 - mean_absolute_error: 0.1468 - val_loss: 0.1516 - val_mean_absolute_error: 0.2116\n",
      "Epoch 73/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0425 - mean_absolute_error: 0.1469 - val_loss: 0.1566 - val_mean_absolute_error: 0.2109\n",
      "Epoch 74/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0470 - mean_absolute_error: 0.1505 - val_loss: 0.1621 - val_mean_absolute_error: 0.2121\n",
      "Epoch 75/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0453 - mean_absolute_error: 0.1484 - val_loss: 0.1478 - val_mean_absolute_error: 0.2075\n",
      "Epoch 76/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0423 - mean_absolute_error: 0.1450 - val_loss: 0.1456 - val_mean_absolute_error: 0.2071\n",
      "Epoch 77/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0396 - mean_absolute_error: 0.1423 - val_loss: 0.1522 - val_mean_absolute_error: 0.2089\n",
      "Epoch 78/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0401 - mean_absolute_error: 0.1431 - val_loss: 0.1579 - val_mean_absolute_error: 0.2131\n",
      "Epoch 79/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0426 - mean_absolute_error: 0.1446 - val_loss: 0.1524 - val_mean_absolute_error: 0.2078\n",
      "Epoch 80/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0436 - mean_absolute_error: 0.1474 - val_loss: 0.1463 - val_mean_absolute_error: 0.2072\n",
      "Epoch 81/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0397 - mean_absolute_error: 0.1421 - val_loss: 0.1580 - val_mean_absolute_error: 0.2125\n",
      "Epoch 82/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0392 - mean_absolute_error: 0.1413 - val_loss: 0.1551 - val_mean_absolute_error: 0.2116\n",
      "Epoch 83/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0395 - mean_absolute_error: 0.1413 - val_loss: 0.1557 - val_mean_absolute_error: 0.2102\n",
      "Epoch 84/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0393 - mean_absolute_error: 0.1427 - val_loss: 0.1521 - val_mean_absolute_error: 0.2081\n",
      "Epoch 85/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0396 - mean_absolute_error: 0.1420 - val_loss: 0.1489 - val_mean_absolute_error: 0.2085\n",
      "Epoch 86/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0377 - mean_absolute_error: 0.1401 - val_loss: 0.1489 - val_mean_absolute_error: 0.2109\n",
      "Epoch 87/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0398 - mean_absolute_error: 0.1428 - val_loss: 0.1524 - val_mean_absolute_error: 0.2083\n",
      "Epoch 88/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0391 - mean_absolute_error: 0.1411 - val_loss: 0.1539 - val_mean_absolute_error: 0.2118\n",
      "Epoch 89/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0441 - mean_absolute_error: 0.1460 - val_loss: 0.1575 - val_mean_absolute_error: 0.2096\n",
      "Epoch 90/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0380 - mean_absolute_error: 0.1399 - val_loss: 0.1514 - val_mean_absolute_error: 0.2113\n",
      "Epoch 91/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0382 - mean_absolute_error: 0.1397 - val_loss: 0.1486 - val_mean_absolute_error: 0.2094\n",
      "Epoch 92/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0389 - mean_absolute_error: 0.1402 - val_loss: 0.1517 - val_mean_absolute_error: 0.2103\n",
      "Epoch 93/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0380 - mean_absolute_error: 0.1394 - val_loss: 0.1549 - val_mean_absolute_error: 0.2088\n",
      "Epoch 94/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0386 - mean_absolute_error: 0.1411 - val_loss: 0.1572 - val_mean_absolute_error: 0.2092\n",
      "Epoch 95/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0390 - mean_absolute_error: 0.1403 - val_loss: 0.1564 - val_mean_absolute_error: 0.2128\n",
      "Epoch 96/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0366 - mean_absolute_error: 0.1379 - val_loss: 0.1570 - val_mean_absolute_error: 0.2123\n",
      "Epoch 97/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0397 - mean_absolute_error: 0.1413 - val_loss: 0.1597 - val_mean_absolute_error: 0.2135\n",
      "Epoch 98/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0379 - mean_absolute_error: 0.1398 - val_loss: 0.1672 - val_mean_absolute_error: 0.2127\n",
      "Epoch 99/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0384 - mean_absolute_error: 0.1397 - val_loss: 0.1602 - val_mean_absolute_error: 0.2142\n",
      "Epoch 100/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0375 - mean_absolute_error: 0.1391 - val_loss: 0.1636 - val_mean_absolute_error: 0.2149\n",
      "Epoch 101/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0352 - mean_absolute_error: 0.1352 - val_loss: 0.1597 - val_mean_absolute_error: 0.2148\n",
      "Epoch 102/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0356 - mean_absolute_error: 0.1361 - val_loss: 0.1623 - val_mean_absolute_error: 0.2160\n",
      "Epoch 103/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0347 - mean_absolute_error: 0.1344 - val_loss: 0.1574 - val_mean_absolute_error: 0.2109\n",
      "Epoch 104/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0362 - mean_absolute_error: 0.1372 - val_loss: 0.1580 - val_mean_absolute_error: 0.2116\n",
      "Epoch 105/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0365 - mean_absolute_error: 0.1374 - val_loss: 0.1552 - val_mean_absolute_error: 0.2100\n",
      "Epoch 106/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0344 - mean_absolute_error: 0.1345 - val_loss: 0.1642 - val_mean_absolute_error: 0.2123\n",
      "Epoch 107/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0363 - mean_absolute_error: 0.1376 - val_loss: 0.1674 - val_mean_absolute_error: 0.2132\n",
      "Epoch 108/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0368 - mean_absolute_error: 0.1383 - val_loss: 0.1606 - val_mean_absolute_error: 0.2136\n",
      "Epoch 109/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0376 - mean_absolute_error: 0.1380 - val_loss: 0.1579 - val_mean_absolute_error: 0.2143\n",
      "Epoch 110/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0358 - mean_absolute_error: 0.1371 - val_loss: 0.1604 - val_mean_absolute_error: 0.2122\n",
      "Epoch 111/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0338 - mean_absolute_error: 0.1339 - val_loss: 0.1628 - val_mean_absolute_error: 0.2126\n",
      "Epoch 112/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0340 - mean_absolute_error: 0.1331 - val_loss: 0.1593 - val_mean_absolute_error: 0.2150\n",
      "Epoch 113/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0354 - mean_absolute_error: 0.1365 - val_loss: 0.1746 - val_mean_absolute_error: 0.2122\n",
      "Epoch 114/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0345 - mean_absolute_error: 0.1352 - val_loss: 0.1638 - val_mean_absolute_error: 0.2142\n",
      "Epoch 115/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0337 - mean_absolute_error: 0.1335 - val_loss: 0.1637 - val_mean_absolute_error: 0.2154\n",
      "Epoch 116/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0394 - mean_absolute_error: 0.1416 - val_loss: 0.1688 - val_mean_absolute_error: 0.2145\n",
      "Epoch 117/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0342 - mean_absolute_error: 0.1343 - val_loss: 0.1641 - val_mean_absolute_error: 0.2146\n",
      "Epoch 118/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0331 - mean_absolute_error: 0.1326 - val_loss: 0.1596 - val_mean_absolute_error: 0.2126\n",
      "Epoch 119/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0329 - mean_absolute_error: 0.1320 - val_loss: 0.1607 - val_mean_absolute_error: 0.2143\n",
      "Epoch 120/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0327 - mean_absolute_error: 0.1319 - val_loss: 0.1660 - val_mean_absolute_error: 0.2142\n",
      "Epoch 121/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0371 - mean_absolute_error: 0.1376 - val_loss: 0.1670 - val_mean_absolute_error: 0.2153\n",
      "Epoch 122/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0355 - mean_absolute_error: 0.1356 - val_loss: 0.1607 - val_mean_absolute_error: 0.2173\n",
      "Epoch 123/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0367 - mean_absolute_error: 0.1363 - val_loss: 0.1578 - val_mean_absolute_error: 0.2147\n",
      "Epoch 124/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0344 - mean_absolute_error: 0.1342 - val_loss: 0.1590 - val_mean_absolute_error: 0.2212\n",
      "Epoch 125/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0338 - mean_absolute_error: 0.1328 - val_loss: 0.1688 - val_mean_absolute_error: 0.2186\n",
      "Epoch 126/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0349 - mean_absolute_error: 0.1347 - val_loss: 0.1622 - val_mean_absolute_error: 0.2132\n",
      "Epoch 127/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0372 - mean_absolute_error: 0.1389 - val_loss: 0.1619 - val_mean_absolute_error: 0.2132\n",
      "Epoch 128/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0337 - mean_absolute_error: 0.1332 - val_loss: 0.1717 - val_mean_absolute_error: 0.2177\n",
      "Epoch 129/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0339 - mean_absolute_error: 0.1332 - val_loss: 0.1595 - val_mean_absolute_error: 0.2154\n",
      "Epoch 130/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0318 - mean_absolute_error: 0.1304 - val_loss: 0.1721 - val_mean_absolute_error: 0.2160\n",
      "Epoch 131/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0345 - mean_absolute_error: 0.1342 - val_loss: 0.1611 - val_mean_absolute_error: 0.2152\n",
      "Epoch 132/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0344 - mean_absolute_error: 0.1335 - val_loss: 0.1617 - val_mean_absolute_error: 0.2187\n",
      "Epoch 133/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0323 - mean_absolute_error: 0.1302 - val_loss: 0.1557 - val_mean_absolute_error: 0.2146\n",
      "Epoch 134/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0376 - mean_absolute_error: 0.1383 - val_loss: 0.1657 - val_mean_absolute_error: 0.2202\n",
      "Epoch 135/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0373 - mean_absolute_error: 0.1392 - val_loss: 0.1611 - val_mean_absolute_error: 0.2152\n",
      "Epoch 136/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0357 - mean_absolute_error: 0.1353 - val_loss: 0.1627 - val_mean_absolute_error: 0.2159\n",
      "Epoch 137/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0370 - mean_absolute_error: 0.1346 - val_loss: 0.1653 - val_mean_absolute_error: 0.2179\n",
      "Epoch 138/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0356 - mean_absolute_error: 0.1344 - val_loss: 0.1637 - val_mean_absolute_error: 0.2151\n",
      "Epoch 139/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0349 - mean_absolute_error: 0.1337 - val_loss: 0.1602 - val_mean_absolute_error: 0.2154\n",
      "Epoch 140/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0315 - mean_absolute_error: 0.1294 - val_loss: 0.1626 - val_mean_absolute_error: 0.2191\n",
      "Epoch 141/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0333 - mean_absolute_error: 0.1327 - val_loss: 0.1703 - val_mean_absolute_error: 0.2201\n",
      "Epoch 142/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0306 - mean_absolute_error: 0.1280 - val_loss: 0.1575 - val_mean_absolute_error: 0.2132\n",
      "Epoch 143/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0300 - mean_absolute_error: 0.1270 - val_loss: 0.1626 - val_mean_absolute_error: 0.2189\n",
      "Epoch 144/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0309 - mean_absolute_error: 0.1289 - val_loss: 0.1623 - val_mean_absolute_error: 0.2185\n",
      "Epoch 145/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0296 - mean_absolute_error: 0.1263 - val_loss: 0.1641 - val_mean_absolute_error: 0.2163\n",
      "Epoch 146/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0300 - mean_absolute_error: 0.1275 - val_loss: 0.1688 - val_mean_absolute_error: 0.2214\n",
      "Epoch 147/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0348 - mean_absolute_error: 0.1334 - val_loss: 0.1608 - val_mean_absolute_error: 0.2210\n",
      "Epoch 148/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.0324 - mean_absolute_error: 0.1307 - val_loss: 0.1817 - val_mean_absolute_error: 0.2280\n",
      "Epoch 149/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.0362 - mean_absolute_error: 0.1353 - val_loss: 0.1730 - val_mean_absolute_error: 0.2235\n",
      "Epoch 150/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0379 - mean_absolute_error: 0.1379 - val_loss: 0.1606 - val_mean_absolute_error: 0.2148\n",
      "Epoch 151/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0334 - mean_absolute_error: 0.1318 - val_loss: 0.1644 - val_mean_absolute_error: 0.2162\n",
      "Epoch 152/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0293 - mean_absolute_error: 0.1261 - val_loss: 0.1657 - val_mean_absolute_error: 0.2173\n",
      "Epoch 153/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0301 - mean_absolute_error: 0.1266 - val_loss: 0.2116 - val_mean_absolute_error: 0.2290\n",
      "Epoch 154/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0396 - mean_absolute_error: 0.1414 - val_loss: 0.1702 - val_mean_absolute_error: 0.2267\n",
      "Epoch 155/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0362 - mean_absolute_error: 0.1360 - val_loss: 0.1671 - val_mean_absolute_error: 0.2160\n",
      "Epoch 156/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0303 - mean_absolute_error: 0.1268 - val_loss: 0.1708 - val_mean_absolute_error: 0.2213\n",
      "Epoch 157/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0286 - mean_absolute_error: 0.1246 - val_loss: 0.1680 - val_mean_absolute_error: 0.2158\n",
      "Epoch 158/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0291 - mean_absolute_error: 0.1254 - val_loss: 0.1683 - val_mean_absolute_error: 0.2162\n",
      "Epoch 159/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0295 - mean_absolute_error: 0.1260 - val_loss: 0.1610 - val_mean_absolute_error: 0.2160\n",
      "Epoch 160/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0305 - mean_absolute_error: 0.1286 - val_loss: 0.1730 - val_mean_absolute_error: 0.2169\n",
      "Epoch 161/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0349 - mean_absolute_error: 0.1349 - val_loss: 0.1652 - val_mean_absolute_error: 0.2179\n",
      "Epoch 162/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0322 - mean_absolute_error: 0.1299 - val_loss: 0.1660 - val_mean_absolute_error: 0.2226\n",
      "Epoch 163/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0315 - mean_absolute_error: 0.1290 - val_loss: 0.1626 - val_mean_absolute_error: 0.2151\n",
      "Epoch 164/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0310 - mean_absolute_error: 0.1281 - val_loss: 0.1680 - val_mean_absolute_error: 0.2251\n",
      "Epoch 165/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0317 - mean_absolute_error: 0.1296 - val_loss: 0.1706 - val_mean_absolute_error: 0.2177\n",
      "Epoch 166/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0308 - mean_absolute_error: 0.1279 - val_loss: 0.1704 - val_mean_absolute_error: 0.2173\n",
      "Epoch 167/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0295 - mean_absolute_error: 0.1267 - val_loss: 0.1715 - val_mean_absolute_error: 0.2209\n",
      "Epoch 168/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0295 - mean_absolute_error: 0.1264 - val_loss: 0.1667 - val_mean_absolute_error: 0.2182\n",
      "Epoch 169/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0295 - mean_absolute_error: 0.1265 - val_loss: 0.1670 - val_mean_absolute_error: 0.2177\n",
      "Epoch 170/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0302 - mean_absolute_error: 0.1267 - val_loss: 0.1933 - val_mean_absolute_error: 0.2220\n",
      "Epoch 171/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0312 - mean_absolute_error: 0.1286 - val_loss: 0.1667 - val_mean_absolute_error: 0.2194\n",
      "Epoch 172/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0325 - mean_absolute_error: 0.1302 - val_loss: 0.1689 - val_mean_absolute_error: 0.2200\n",
      "Epoch 173/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0487 - mean_absolute_error: 0.1484 - val_loss: 0.1704 - val_mean_absolute_error: 0.2186\n",
      "Epoch 174/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0305 - mean_absolute_error: 0.1268 - val_loss: 0.1692 - val_mean_absolute_error: 0.2180\n",
      "Epoch 175/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0283 - mean_absolute_error: 0.1238 - val_loss: 0.1620 - val_mean_absolute_error: 0.2154\n",
      "Epoch 176/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0282 - mean_absolute_error: 0.1236 - val_loss: 0.1668 - val_mean_absolute_error: 0.2162\n",
      "Epoch 177/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0282 - mean_absolute_error: 0.1240 - val_loss: 0.1680 - val_mean_absolute_error: 0.2185\n",
      "Epoch 178/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0283 - mean_absolute_error: 0.1242 - val_loss: 0.1689 - val_mean_absolute_error: 0.2175\n",
      "Epoch 179/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0285 - mean_absolute_error: 0.1243 - val_loss: 0.1698 - val_mean_absolute_error: 0.2166\n",
      "Epoch 180/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0283 - mean_absolute_error: 0.1236 - val_loss: 0.1732 - val_mean_absolute_error: 0.2298\n",
      "Epoch 181/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0294 - mean_absolute_error: 0.1259 - val_loss: 0.1734 - val_mean_absolute_error: 0.2238\n",
      "Epoch 182/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0295 - mean_absolute_error: 0.1267 - val_loss: 0.1693 - val_mean_absolute_error: 0.2211\n",
      "Epoch 183/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0309 - mean_absolute_error: 0.1283 - val_loss: 0.1633 - val_mean_absolute_error: 0.2165\n",
      "Epoch 184/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0393 - mean_absolute_error: 0.1403 - val_loss: 0.1682 - val_mean_absolute_error: 0.2204\n",
      "Epoch 185/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0305 - mean_absolute_error: 0.1269 - val_loss: 0.1659 - val_mean_absolute_error: 0.2190\n",
      "Epoch 186/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0295 - mean_absolute_error: 0.1257 - val_loss: 0.1765 - val_mean_absolute_error: 0.2193\n",
      "Epoch 187/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0286 - mean_absolute_error: 0.1246 - val_loss: 0.1698 - val_mean_absolute_error: 0.2185\n",
      "Epoch 188/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0278 - mean_absolute_error: 0.1230 - val_loss: 0.1665 - val_mean_absolute_error: 0.2166\n",
      "Epoch 189/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0272 - mean_absolute_error: 0.1215 - val_loss: 0.1668 - val_mean_absolute_error: 0.2167\n",
      "Epoch 190/300\n",
      "12967/12967 [==============================] - 1s 43us/sample - loss: 0.0282 - mean_absolute_error: 0.1231 - val_loss: 0.1774 - val_mean_absolute_error: 0.2236\n",
      "Epoch 191/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0296 - mean_absolute_error: 0.1257 - val_loss: 0.1800 - val_mean_absolute_error: 0.2196\n",
      "Epoch 192/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0303 - mean_absolute_error: 0.1272 - val_loss: 0.1734 - val_mean_absolute_error: 0.2210\n",
      "Epoch 193/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.0299 - mean_absolute_error: 0.1262 - val_loss: 0.1648 - val_mean_absolute_error: 0.2182\n",
      "Epoch 194/300\n",
      "12967/12967 [==============================] - 1s 43us/sample - loss: 0.0283 - mean_absolute_error: 0.1238 - val_loss: 0.1706 - val_mean_absolute_error: 0.2174\n",
      "Epoch 195/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.0279 - mean_absolute_error: 0.1230 - val_loss: 0.1777 - val_mean_absolute_error: 0.2223\n",
      "Epoch 196/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0302 - mean_absolute_error: 0.1272 - val_loss: 0.1728 - val_mean_absolute_error: 0.2195\n",
      "Epoch 197/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0287 - mean_absolute_error: 0.1239 - val_loss: 0.1696 - val_mean_absolute_error: 0.2211\n",
      "Epoch 198/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0309 - mean_absolute_error: 0.1274 - val_loss: 0.1680 - val_mean_absolute_error: 0.2201\n",
      "Epoch 199/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0313 - mean_absolute_error: 0.1284 - val_loss: 0.1756 - val_mean_absolute_error: 0.2191\n",
      "Epoch 200/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0341 - mean_absolute_error: 0.1320 - val_loss: 0.1685 - val_mean_absolute_error: 0.2198\n",
      "Epoch 201/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0302 - mean_absolute_error: 0.1258 - val_loss: 0.1717 - val_mean_absolute_error: 0.2198\n",
      "Epoch 202/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0267 - mean_absolute_error: 0.1208 - val_loss: 0.1732 - val_mean_absolute_error: 0.2184\n",
      "Epoch 203/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0282 - mean_absolute_error: 0.1226 - val_loss: 0.1702 - val_mean_absolute_error: 0.2197\n",
      "Epoch 204/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0265 - mean_absolute_error: 0.1207 - val_loss: 0.1696 - val_mean_absolute_error: 0.2208\n",
      "Epoch 205/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0304 - mean_absolute_error: 0.1271 - val_loss: 0.1709 - val_mean_absolute_error: 0.2213\n",
      "Epoch 206/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0268 - mean_absolute_error: 0.1211 - val_loss: 0.1818 - val_mean_absolute_error: 0.2239\n",
      "Epoch 207/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0282 - mean_absolute_error: 0.1235 - val_loss: 0.1733 - val_mean_absolute_error: 0.2229\n",
      "Epoch 208/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0265 - mean_absolute_error: 0.1206 - val_loss: 0.1759 - val_mean_absolute_error: 0.2200\n",
      "Epoch 209/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0269 - mean_absolute_error: 0.1211 - val_loss: 0.1766 - val_mean_absolute_error: 0.2211\n",
      "Epoch 210/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0272 - mean_absolute_error: 0.1222 - val_loss: 0.1776 - val_mean_absolute_error: 0.2230\n",
      "Epoch 211/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0288 - mean_absolute_error: 0.1244 - val_loss: 0.1726 - val_mean_absolute_error: 0.2232\n",
      "Epoch 212/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0305 - mean_absolute_error: 0.1265 - val_loss: 0.1836 - val_mean_absolute_error: 0.2267\n",
      "Epoch 213/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0364 - mean_absolute_error: 0.1364 - val_loss: 0.1710 - val_mean_absolute_error: 0.2221\n",
      "Epoch 214/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0349 - mean_absolute_error: 0.1320 - val_loss: 0.1607 - val_mean_absolute_error: 0.2186\n",
      "Epoch 215/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0315 - mean_absolute_error: 0.1274 - val_loss: 0.1736 - val_mean_absolute_error: 0.2208\n",
      "Epoch 216/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0276 - mean_absolute_error: 0.1224 - val_loss: 0.1654 - val_mean_absolute_error: 0.2166\n",
      "Epoch 217/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0280 - mean_absolute_error: 0.1222 - val_loss: 0.1769 - val_mean_absolute_error: 0.2199\n",
      "Epoch 218/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0297 - mean_absolute_error: 0.1251 - val_loss: 0.1630 - val_mean_absolute_error: 0.2222\n",
      "Epoch 219/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0346 - mean_absolute_error: 0.1315 - val_loss: 0.1777 - val_mean_absolute_error: 0.2186\n",
      "Epoch 220/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0301 - mean_absolute_error: 0.1254 - val_loss: 0.1700 - val_mean_absolute_error: 0.2172\n",
      "Epoch 221/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0260 - mean_absolute_error: 0.1186 - val_loss: 0.1749 - val_mean_absolute_error: 0.2197\n",
      "Epoch 222/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0251 - mean_absolute_error: 0.1177 - val_loss: 0.1716 - val_mean_absolute_error: 0.2190\n",
      "Epoch 223/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0268 - mean_absolute_error: 0.1212 - val_loss: 0.1681 - val_mean_absolute_error: 0.2191\n",
      "Epoch 224/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0273 - mean_absolute_error: 0.1221 - val_loss: 0.1766 - val_mean_absolute_error: 0.2205\n",
      "Epoch 225/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0265 - mean_absolute_error: 0.1211 - val_loss: 0.1817 - val_mean_absolute_error: 0.2226\n",
      "Epoch 226/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0270 - mean_absolute_error: 0.1210 - val_loss: 0.1680 - val_mean_absolute_error: 0.2272\n",
      "Epoch 227/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0316 - mean_absolute_error: 0.1274 - val_loss: 0.1717 - val_mean_absolute_error: 0.2207\n",
      "Epoch 228/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0285 - mean_absolute_error: 0.1227 - val_loss: 0.1751 - val_mean_absolute_error: 0.2288\n",
      "Epoch 229/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0407 - mean_absolute_error: 0.1378 - val_loss: 0.1686 - val_mean_absolute_error: 0.2195\n",
      "Epoch 230/300\n",
      "12967/12967 [==============================] - 1s 41us/sample - loss: 0.0282 - mean_absolute_error: 0.1224 - val_loss: 0.1729 - val_mean_absolute_error: 0.2203\n",
      "Epoch 231/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0270 - mean_absolute_error: 0.1212 - val_loss: 0.1737 - val_mean_absolute_error: 0.2206\n",
      "Epoch 232/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0247 - mean_absolute_error: 0.1173 - val_loss: 0.1719 - val_mean_absolute_error: 0.2208\n",
      "Epoch 233/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0246 - mean_absolute_error: 0.1164 - val_loss: 0.1731 - val_mean_absolute_error: 0.2217\n",
      "Epoch 234/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0245 - mean_absolute_error: 0.1161 - val_loss: 0.1781 - val_mean_absolute_error: 0.2239\n",
      "Epoch 235/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0243 - mean_absolute_error: 0.1164 - val_loss: 0.1759 - val_mean_absolute_error: 0.2232\n",
      "Epoch 236/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0296 - mean_absolute_error: 0.1266 - val_loss: 0.1740 - val_mean_absolute_error: 0.2244\n",
      "Epoch 237/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0289 - mean_absolute_error: 0.1233 - val_loss: 0.1737 - val_mean_absolute_error: 0.2221\n",
      "Epoch 238/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0263 - mean_absolute_error: 0.1209 - val_loss: 0.1740 - val_mean_absolute_error: 0.2257\n",
      "Epoch 239/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0270 - mean_absolute_error: 0.1214 - val_loss: 0.1737 - val_mean_absolute_error: 0.2220\n",
      "Epoch 240/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0258 - mean_absolute_error: 0.1188 - val_loss: 0.1726 - val_mean_absolute_error: 0.2225\n",
      "Epoch 241/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0272 - mean_absolute_error: 0.1214 - val_loss: 0.1732 - val_mean_absolute_error: 0.2217\n",
      "Epoch 242/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0273 - mean_absolute_error: 0.1204 - val_loss: 0.1798 - val_mean_absolute_error: 0.2246\n",
      "Epoch 243/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0279 - mean_absolute_error: 0.1228 - val_loss: 0.1935 - val_mean_absolute_error: 0.2239\n",
      "Epoch 244/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0289 - mean_absolute_error: 0.1236 - val_loss: 0.1860 - val_mean_absolute_error: 0.2247\n",
      "Epoch 245/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0293 - mean_absolute_error: 0.1244 - val_loss: 0.1733 - val_mean_absolute_error: 0.2219\n",
      "Epoch 246/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0257 - mean_absolute_error: 0.1184 - val_loss: 0.1721 - val_mean_absolute_error: 0.2208\n",
      "Epoch 247/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0262 - mean_absolute_error: 0.1192 - val_loss: 0.1762 - val_mean_absolute_error: 0.2279\n",
      "Epoch 248/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0262 - mean_absolute_error: 0.1205 - val_loss: 0.1752 - val_mean_absolute_error: 0.2240\n",
      "Epoch 249/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0253 - mean_absolute_error: 0.1186 - val_loss: 0.1814 - val_mean_absolute_error: 0.2242\n",
      "Epoch 250/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0254 - mean_absolute_error: 0.1185 - val_loss: 0.1791 - val_mean_absolute_error: 0.2243\n",
      "Epoch 251/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0274 - mean_absolute_error: 0.1219 - val_loss: 0.1749 - val_mean_absolute_error: 0.2224\n",
      "Epoch 252/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0278 - mean_absolute_error: 0.1220 - val_loss: 0.1764 - val_mean_absolute_error: 0.2224\n",
      "Epoch 253/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0257 - mean_absolute_error: 0.1187 - val_loss: 0.1698 - val_mean_absolute_error: 0.2223\n",
      "Epoch 254/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0283 - mean_absolute_error: 0.1233 - val_loss: 0.1826 - val_mean_absolute_error: 0.2209\n",
      "Epoch 255/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0264 - mean_absolute_error: 0.1198 - val_loss: 0.1713 - val_mean_absolute_error: 0.2242\n",
      "Epoch 256/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0256 - mean_absolute_error: 0.1182 - val_loss: 0.1760 - val_mean_absolute_error: 0.2226\n",
      "Epoch 257/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0287 - mean_absolute_error: 0.1241 - val_loss: 0.1736 - val_mean_absolute_error: 0.2237\n",
      "Epoch 258/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0289 - mean_absolute_error: 0.1243 - val_loss: 0.1810 - val_mean_absolute_error: 0.2231\n",
      "Epoch 259/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0273 - mean_absolute_error: 0.1203 - val_loss: 0.1768 - val_mean_absolute_error: 0.2210\n",
      "Epoch 260/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0267 - mean_absolute_error: 0.1201 - val_loss: 0.1671 - val_mean_absolute_error: 0.2201\n",
      "Epoch 261/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0260 - mean_absolute_error: 0.1186 - val_loss: 0.1725 - val_mean_absolute_error: 0.2298\n",
      "Epoch 262/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0255 - mean_absolute_error: 0.1179 - val_loss: 0.1777 - val_mean_absolute_error: 0.2215\n",
      "Epoch 263/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0258 - mean_absolute_error: 0.1182 - val_loss: 0.1797 - val_mean_absolute_error: 0.2269\n",
      "Epoch 264/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0297 - mean_absolute_error: 0.1245 - val_loss: 0.1804 - val_mean_absolute_error: 0.2259\n",
      "Epoch 265/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0297 - mean_absolute_error: 0.1247 - val_loss: 0.1705 - val_mean_absolute_error: 0.2222\n",
      "Epoch 266/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0268 - mean_absolute_error: 0.1206 - val_loss: 0.1673 - val_mean_absolute_error: 0.2204\n",
      "Epoch 267/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0246 - mean_absolute_error: 0.1161 - val_loss: 0.1771 - val_mean_absolute_error: 0.2274\n",
      "Epoch 268/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0236 - mean_absolute_error: 0.1145 - val_loss: 0.1690 - val_mean_absolute_error: 0.2218\n",
      "Epoch 269/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0251 - mean_absolute_error: 0.1173 - val_loss: 0.1745 - val_mean_absolute_error: 0.2228\n",
      "Epoch 270/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0262 - mean_absolute_error: 0.1191 - val_loss: 0.1771 - val_mean_absolute_error: 0.2245\n",
      "Epoch 271/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0257 - mean_absolute_error: 0.1177 - val_loss: 0.1843 - val_mean_absolute_error: 0.2312\n",
      "Epoch 272/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0266 - mean_absolute_error: 0.1199 - val_loss: 0.1715 - val_mean_absolute_error: 0.2217\n",
      "Epoch 273/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0282 - mean_absolute_error: 0.1229 - val_loss: 0.1798 - val_mean_absolute_error: 0.2314\n",
      "Epoch 274/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0271 - mean_absolute_error: 0.1207 - val_loss: 0.1742 - val_mean_absolute_error: 0.2219\n",
      "Epoch 275/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0237 - mean_absolute_error: 0.1152 - val_loss: 0.1704 - val_mean_absolute_error: 0.2210\n",
      "Epoch 276/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0244 - mean_absolute_error: 0.1164 - val_loss: 0.1708 - val_mean_absolute_error: 0.2250\n",
      "Epoch 277/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0247 - mean_absolute_error: 0.1171 - val_loss: 0.1724 - val_mean_absolute_error: 0.2228\n",
      "Epoch 278/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0254 - mean_absolute_error: 0.1187 - val_loss: 0.1849 - val_mean_absolute_error: 0.2264\n",
      "Epoch 279/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0259 - mean_absolute_error: 0.1181 - val_loss: 0.1822 - val_mean_absolute_error: 0.2266\n",
      "Epoch 280/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0280 - mean_absolute_error: 0.1222 - val_loss: 0.1689 - val_mean_absolute_error: 0.2273\n",
      "Epoch 281/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0310 - mean_absolute_error: 0.1266 - val_loss: 0.1797 - val_mean_absolute_error: 0.2247\n",
      "Epoch 282/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0258 - mean_absolute_error: 0.1187 - val_loss: 0.1763 - val_mean_absolute_error: 0.2274\n",
      "Epoch 283/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0250 - mean_absolute_error: 0.1168 - val_loss: 0.1769 - val_mean_absolute_error: 0.2216\n",
      "Epoch 284/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0249 - mean_absolute_error: 0.1172 - val_loss: 0.1768 - val_mean_absolute_error: 0.2238\n",
      "Epoch 285/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0249 - mean_absolute_error: 0.1170 - val_loss: 0.1787 - val_mean_absolute_error: 0.2235\n",
      "Epoch 286/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0261 - mean_absolute_error: 0.1192 - val_loss: 0.1744 - val_mean_absolute_error: 0.2260\n",
      "Epoch 287/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0253 - mean_absolute_error: 0.1182 - val_loss: 0.1743 - val_mean_absolute_error: 0.2229\n",
      "Epoch 288/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0246 - mean_absolute_error: 0.1166 - val_loss: 0.1798 - val_mean_absolute_error: 0.2266\n",
      "Epoch 289/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0242 - mean_absolute_error: 0.1164 - val_loss: 0.1791 - val_mean_absolute_error: 0.2263\n",
      "Epoch 290/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0265 - mean_absolute_error: 0.1198 - val_loss: 0.1817 - val_mean_absolute_error: 0.2233\n",
      "Epoch 291/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0299 - mean_absolute_error: 0.1242 - val_loss: 0.1744 - val_mean_absolute_error: 0.2247\n",
      "Epoch 292/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0281 - mean_absolute_error: 0.1213 - val_loss: 0.1764 - val_mean_absolute_error: 0.2215\n",
      "Epoch 293/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0255 - mean_absolute_error: 0.1168 - val_loss: 0.1722 - val_mean_absolute_error: 0.2263\n",
      "Epoch 294/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0272 - mean_absolute_error: 0.1214 - val_loss: 0.1889 - val_mean_absolute_error: 0.2317\n",
      "Epoch 295/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0284 - mean_absolute_error: 0.1230 - val_loss: 0.1726 - val_mean_absolute_error: 0.2234\n",
      "Epoch 296/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0264 - mean_absolute_error: 0.1198 - val_loss: 0.1798 - val_mean_absolute_error: 0.2234\n",
      "Epoch 297/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0246 - mean_absolute_error: 0.1160 - val_loss: 0.1746 - val_mean_absolute_error: 0.2240\n",
      "Epoch 298/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0230 - mean_absolute_error: 0.1134 - val_loss: 0.1754 - val_mean_absolute_error: 0.2234\n",
      "Epoch 299/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0236 - mean_absolute_error: 0.1155 - val_loss: 0.1737 - val_mean_absolute_error: 0.2238\n",
      "Epoch 300/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0231 - mean_absolute_error: 0.1142 - val_loss: 0.1761 - val_mean_absolute_error: 0.2245\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(x_train, y_train,  # 傳入訓練數據\n",
    "               batch_size=64,  # 批次大小設為64\n",
    "               epochs=300,  # 整個dataset訓練300遍\n",
    "               validation_data=(x_val, y_val),  # 驗證數據\n",
    "               callbacks=[model_cbk, model_mckp])  # Tensorboard回調函數紀錄訓練過程，ModelCheckpoint回調函數儲存最好的模型"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 訓練結果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dict_keys(['loss', 'mean_absolute_error', 'val_loss', 'val_mean_absolute_error'])"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "history.history.keys()  # 查看history儲存的資訊有哪些"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x25d253a2550>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['loss'], label='train')\n",
    "plt.plot(history.history['val_loss'], label='validation')\n",
    "plt.ylim(0.02, 0.2)\n",
    "plt.title('Mean square error')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epochs')\n",
    "plt.legend(loc='upper right')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x25d25447ac8>"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['mean_absolute_error'], label='train')\n",
    "plt.plot(history.history['val_mean_absolute_error'], label='validation')\n",
    "plt.ylim(0.12, 0.26)\n",
    "plt.title('Mean absolute error')\n",
    "plt.ylabel('metrics')\n",
    "plt.xlabel('epochs')\n",
    "plt.legend(loc='upper right')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "測試數據的誤差百分比：用測試數據預測房屋價格並與答案計算誤差百分比。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model_1 Percentage Error: 13.67%\n"
     ]
    }
   ],
   "source": [
    "# 載入模型\n",
    "model = keras.models.load_model('lab2-logs/models/Best-model-1.h5')\n",
    "# 先將房屋價格取出\n",
    "y_test = np.array(test_data['price'])\n",
    "# 標準化數據\n",
    "test_data = (test_data - mean) / std\n",
    "# 將輸入數據存成Numpy 格式\n",
    "x_test = np.array(test_data.drop('price', axis='columns'))\n",
    "# 預測測試數據\n",
    "y_pred = model.predict(x_test)\n",
    "# 將預測結果轉換回來(因為訓練時的訓練目標也有經過標準化)\n",
    "y_pred = np.reshape(y_pred * std['price'] + mean['price'], y_test.shape)\n",
    "# 計算平均的誤差百分比\n",
    "percentage_error = np.mean(np.abs(y_test - y_pred)) / np.mean(y_test) * 100\n",
    "# 顯示誤差百分比\n",
    "print(\"Model_1 Percentage Error: {:.2f}%\".format(percentage_error))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### TensorBoard 可視化工具"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 這行指令可以幫助我們直接在jupyter notebook上顯示TensorBoard\n",
    "%load_ext tensorboard"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Reusing TensorBoard on port 9530 (pid 8108), started 0:11:33 ago. (Use '!kill 8108' to kill it.)"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "\n",
       "      <iframe id=\"tensorboard-frame-3ff6a855f5ae6e34\" width=\"100%\" height=\"800\" frameborder=\"0\">\n",
       "      </iframe>\n",
       "      <script>\n",
       "        (function() {\n",
       "          const frame = document.getElementById(\"tensorboard-frame-3ff6a855f5ae6e34\");\n",
       "          const url = new URL(\"/\", window.location);\n",
       "          url.port = 9530;\n",
       "          frame.src = url;\n",
       "        })();\n",
       "      </script>\n",
       "  "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%tensorboard --port 9530 --logdir lab2-logs"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 實驗二：過擬合問題\n",
    "\n",
    "### 方法一、減少網路權重"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 12967 samples, validate on 4323 samples\n",
      "Epoch 1/300\n",
      "12967/12967 [==============================] - 1s 97us/sample - loss: 0.4805 - mean_absolute_error: 0.4246 - val_loss: 0.3289 - val_mean_absolute_error: 0.3453\n",
      "Epoch 2/300\n",
      "12967/12967 [==============================] - 1s 41us/sample - loss: 0.2553 - mean_absolute_error: 0.3133 - val_loss: 0.2718 - val_mean_absolute_error: 0.3221\n",
      "Epoch 3/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.2239 - mean_absolute_error: 0.2956 - val_loss: 0.2486 - val_mean_absolute_error: 0.3046\n",
      "Epoch 4/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.2082 - mean_absolute_error: 0.2841 - val_loss: 0.2426 - val_mean_absolute_error: 0.2928\n",
      "Epoch 5/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1953 - mean_absolute_error: 0.2747 - val_loss: 0.2213 - val_mean_absolute_error: 0.2850\n",
      "Epoch 6/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1864 - mean_absolute_error: 0.2680 - val_loss: 0.2144 - val_mean_absolute_error: 0.2774\n",
      "Epoch 7/300\n",
      "12967/12967 [==============================] - 1s 40us/sample - loss: 0.1788 - mean_absolute_error: 0.2633 - val_loss: 0.2104 - val_mean_absolute_error: 0.2739\n",
      "Epoch 8/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1729 - mean_absolute_error: 0.2579 - val_loss: 0.1963 - val_mean_absolute_error: 0.2697\n",
      "Epoch 9/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1651 - mean_absolute_error: 0.2529 - val_loss: 0.1928 - val_mean_absolute_error: 0.2617\n",
      "Epoch 10/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1589 - mean_absolute_error: 0.2485 - val_loss: 0.1858 - val_mean_absolute_error: 0.2591\n",
      "Epoch 11/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1544 - mean_absolute_error: 0.2441 - val_loss: 0.1824 - val_mean_absolute_error: 0.2530\n",
      "Epoch 12/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1486 - mean_absolute_error: 0.2405 - val_loss: 0.1770 - val_mean_absolute_error: 0.2506\n",
      "Epoch 13/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1447 - mean_absolute_error: 0.2368 - val_loss: 0.1810 - val_mean_absolute_error: 0.2473\n",
      "Epoch 14/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1419 - mean_absolute_error: 0.2339 - val_loss: 0.1699 - val_mean_absolute_error: 0.2428\n",
      "Epoch 15/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1376 - mean_absolute_error: 0.2296 - val_loss: 0.1646 - val_mean_absolute_error: 0.2399\n",
      "Epoch 16/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1338 - mean_absolute_error: 0.2265 - val_loss: 0.1651 - val_mean_absolute_error: 0.2363\n",
      "Epoch 17/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1317 - mean_absolute_error: 0.2237 - val_loss: 0.1602 - val_mean_absolute_error: 0.2362\n",
      "Epoch 18/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.1265 - mean_absolute_error: 0.2206 - val_loss: 0.1569 - val_mean_absolute_error: 0.2317\n",
      "Epoch 19/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1236 - mean_absolute_error: 0.2187 - val_loss: 0.1547 - val_mean_absolute_error: 0.2296\n",
      "Epoch 20/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1216 - mean_absolute_error: 0.2164 - val_loss: 0.1541 - val_mean_absolute_error: 0.2292\n",
      "Epoch 21/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1202 - mean_absolute_error: 0.2150 - val_loss: 0.1532 - val_mean_absolute_error: 0.2273\n",
      "Epoch 22/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1178 - mean_absolute_error: 0.2124 - val_loss: 0.1488 - val_mean_absolute_error: 0.2253\n",
      "Epoch 23/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1172 - mean_absolute_error: 0.2114 - val_loss: 0.1511 - val_mean_absolute_error: 0.2235\n",
      "Epoch 24/300\n",
      "12967/12967 [==============================] - 1s 39us/sample - loss: 0.1152 - mean_absolute_error: 0.2096 - val_loss: 0.1505 - val_mean_absolute_error: 0.2225\n",
      "Epoch 25/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.1139 - mean_absolute_error: 0.2083 - val_loss: 0.1507 - val_mean_absolute_error: 0.2207\n",
      "Epoch 26/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1122 - mean_absolute_error: 0.2071 - val_loss: 0.1466 - val_mean_absolute_error: 0.2205\n",
      "Epoch 27/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.1109 - mean_absolute_error: 0.2068 - val_loss: 0.1446 - val_mean_absolute_error: 0.2200\n",
      "Epoch 28/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1087 - mean_absolute_error: 0.2041 - val_loss: 0.1447 - val_mean_absolute_error: 0.2200\n",
      "Epoch 29/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1080 - mean_absolute_error: 0.2040 - val_loss: 0.1452 - val_mean_absolute_error: 0.2185\n",
      "Epoch 30/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.1064 - mean_absolute_error: 0.2019 - val_loss: 0.1424 - val_mean_absolute_error: 0.2167\n",
      "Epoch 31/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.1058 - mean_absolute_error: 0.2016 - val_loss: 0.1420 - val_mean_absolute_error: 0.2170\n",
      "Epoch 32/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.1032 - mean_absolute_error: 0.2005 - val_loss: 0.1459 - val_mean_absolute_error: 0.2177\n",
      "Epoch 33/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.1051 - mean_absolute_error: 0.2006 - val_loss: 0.1388 - val_mean_absolute_error: 0.2154\n",
      "Epoch 34/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1025 - mean_absolute_error: 0.1996 - val_loss: 0.1401 - val_mean_absolute_error: 0.2154\n",
      "Epoch 35/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.1010 - mean_absolute_error: 0.1983 - val_loss: 0.1398 - val_mean_absolute_error: 0.2159\n",
      "Epoch 36/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1006 - mean_absolute_error: 0.1994 - val_loss: 0.1418 - val_mean_absolute_error: 0.2133\n",
      "Epoch 37/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0992 - mean_absolute_error: 0.1970 - val_loss: 0.1447 - val_mean_absolute_error: 0.2182\n",
      "Epoch 38/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1013 - mean_absolute_error: 0.1978 - val_loss: 0.1390 - val_mean_absolute_error: 0.2127\n",
      "Epoch 39/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0985 - mean_absolute_error: 0.1970 - val_loss: 0.1373 - val_mean_absolute_error: 0.2115\n",
      "Epoch 40/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0978 - mean_absolute_error: 0.1955 - val_loss: 0.1378 - val_mean_absolute_error: 0.2125\n",
      "Epoch 41/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0970 - mean_absolute_error: 0.1945 - val_loss: 0.1418 - val_mean_absolute_error: 0.2119\n",
      "Epoch 42/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0961 - mean_absolute_error: 0.1945 - val_loss: 0.1518 - val_mean_absolute_error: 0.2150\n",
      "Epoch 43/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0967 - mean_absolute_error: 0.1940 - val_loss: 0.1426 - val_mean_absolute_error: 0.2125\n",
      "Epoch 44/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0964 - mean_absolute_error: 0.1944 - val_loss: 0.1401 - val_mean_absolute_error: 0.2112\n",
      "Epoch 45/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0960 - mean_absolute_error: 0.1933 - val_loss: 0.1381 - val_mean_absolute_error: 0.2110\n",
      "Epoch 46/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0953 - mean_absolute_error: 0.1933 - val_loss: 0.1404 - val_mean_absolute_error: 0.2146\n",
      "Epoch 47/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0951 - mean_absolute_error: 0.1937 - val_loss: 0.1365 - val_mean_absolute_error: 0.2140\n",
      "Epoch 48/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0942 - mean_absolute_error: 0.1919 - val_loss: 0.1420 - val_mean_absolute_error: 0.2111\n",
      "Epoch 49/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0942 - mean_absolute_error: 0.1918 - val_loss: 0.1365 - val_mean_absolute_error: 0.2110\n",
      "Epoch 50/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0923 - mean_absolute_error: 0.1919 - val_loss: 0.1384 - val_mean_absolute_error: 0.2123\n",
      "Epoch 51/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0927 - mean_absolute_error: 0.1920 - val_loss: 0.1388 - val_mean_absolute_error: 0.2111\n",
      "Epoch 52/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0923 - mean_absolute_error: 0.1912 - val_loss: 0.1377 - val_mean_absolute_error: 0.2104\n",
      "Epoch 53/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0931 - mean_absolute_error: 0.1919 - val_loss: 0.1369 - val_mean_absolute_error: 0.2120\n",
      "Epoch 54/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0922 - mean_absolute_error: 0.1918 - val_loss: 0.1353 - val_mean_absolute_error: 0.2098\n",
      "Epoch 55/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0929 - mean_absolute_error: 0.1908 - val_loss: 0.1381 - val_mean_absolute_error: 0.2093\n",
      "Epoch 56/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0928 - mean_absolute_error: 0.1918 - val_loss: 0.1333 - val_mean_absolute_error: 0.2092\n",
      "Epoch 57/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0904 - mean_absolute_error: 0.1896 - val_loss: 0.1381 - val_mean_absolute_error: 0.2115\n",
      "Epoch 58/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0900 - mean_absolute_error: 0.1897 - val_loss: 0.1366 - val_mean_absolute_error: 0.2104\n",
      "Epoch 59/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0914 - mean_absolute_error: 0.1899 - val_loss: 0.1351 - val_mean_absolute_error: 0.2096\n",
      "Epoch 60/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0896 - mean_absolute_error: 0.1889 - val_loss: 0.1420 - val_mean_absolute_error: 0.2094\n",
      "Epoch 61/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0896 - mean_absolute_error: 0.1887 - val_loss: 0.1349 - val_mean_absolute_error: 0.2093\n",
      "Epoch 62/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0904 - mean_absolute_error: 0.1903 - val_loss: 0.1386 - val_mean_absolute_error: 0.2092\n",
      "Epoch 63/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0883 - mean_absolute_error: 0.1888 - val_loss: 0.1390 - val_mean_absolute_error: 0.2111\n",
      "Epoch 64/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0877 - mean_absolute_error: 0.1882 - val_loss: 0.1457 - val_mean_absolute_error: 0.2106\n",
      "Epoch 65/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0910 - mean_absolute_error: 0.1907 - val_loss: 0.1336 - val_mean_absolute_error: 0.2092\n",
      "Epoch 66/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0874 - mean_absolute_error: 0.1874 - val_loss: 0.1380 - val_mean_absolute_error: 0.2123\n",
      "Epoch 67/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0889 - mean_absolute_error: 0.1879 - val_loss: 0.1385 - val_mean_absolute_error: 0.2087\n",
      "Epoch 68/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0877 - mean_absolute_error: 0.1884 - val_loss: 0.1410 - val_mean_absolute_error: 0.2098\n",
      "Epoch 69/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0887 - mean_absolute_error: 0.1883 - val_loss: 0.1348 - val_mean_absolute_error: 0.2102\n",
      "Epoch 70/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0882 - mean_absolute_error: 0.1880 - val_loss: 0.1342 - val_mean_absolute_error: 0.2098\n",
      "Epoch 71/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.0880 - mean_absolute_error: 0.1883 - val_loss: 0.1344 - val_mean_absolute_error: 0.2088\n",
      "Epoch 72/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0888 - mean_absolute_error: 0.1881 - val_loss: 0.1378 - val_mean_absolute_error: 0.2136\n",
      "Epoch 73/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0877 - mean_absolute_error: 0.1879 - val_loss: 0.1415 - val_mean_absolute_error: 0.2119\n",
      "Epoch 74/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0862 - mean_absolute_error: 0.1873 - val_loss: 0.1340 - val_mean_absolute_error: 0.2076\n",
      "Epoch 75/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0867 - mean_absolute_error: 0.1862 - val_loss: 0.1340 - val_mean_absolute_error: 0.2076\n",
      "Epoch 76/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0851 - mean_absolute_error: 0.1859 - val_loss: 0.1339 - val_mean_absolute_error: 0.2094\n",
      "Epoch 77/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0864 - mean_absolute_error: 0.1871 - val_loss: 0.1403 - val_mean_absolute_error: 0.2100\n",
      "Epoch 78/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0859 - mean_absolute_error: 0.1865 - val_loss: 0.1330 - val_mean_absolute_error: 0.2089\n",
      "Epoch 79/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0858 - mean_absolute_error: 0.1861 - val_loss: 0.1396 - val_mean_absolute_error: 0.2100\n",
      "Epoch 80/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0857 - mean_absolute_error: 0.1861 - val_loss: 0.1323 - val_mean_absolute_error: 0.2087\n",
      "Epoch 81/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0851 - mean_absolute_error: 0.1865 - val_loss: 0.1336 - val_mean_absolute_error: 0.2084\n",
      "Epoch 82/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0842 - mean_absolute_error: 0.1862 - val_loss: 0.1415 - val_mean_absolute_error: 0.2096\n",
      "Epoch 83/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0846 - mean_absolute_error: 0.1863 - val_loss: 0.1389 - val_mean_absolute_error: 0.2115\n",
      "Epoch 84/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0846 - mean_absolute_error: 0.1857 - val_loss: 0.1377 - val_mean_absolute_error: 0.2086\n",
      "Epoch 85/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0834 - mean_absolute_error: 0.1854 - val_loss: 0.1422 - val_mean_absolute_error: 0.2083\n",
      "Epoch 86/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0841 - mean_absolute_error: 0.1845 - val_loss: 0.1367 - val_mean_absolute_error: 0.2090\n",
      "Epoch 87/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0835 - mean_absolute_error: 0.1849 - val_loss: 0.1372 - val_mean_absolute_error: 0.2076\n",
      "Epoch 88/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0847 - mean_absolute_error: 0.1853 - val_loss: 0.1374 - val_mean_absolute_error: 0.2095\n",
      "Epoch 89/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0837 - mean_absolute_error: 0.1850 - val_loss: 0.1335 - val_mean_absolute_error: 0.2100\n",
      "Epoch 90/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0834 - mean_absolute_error: 0.1848 - val_loss: 0.1365 - val_mean_absolute_error: 0.2098\n",
      "Epoch 91/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0845 - mean_absolute_error: 0.1855 - val_loss: 0.1382 - val_mean_absolute_error: 0.2084\n",
      "Epoch 92/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0827 - mean_absolute_error: 0.1843 - val_loss: 0.1376 - val_mean_absolute_error: 0.2077\n",
      "Epoch 93/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0834 - mean_absolute_error: 0.1853 - val_loss: 0.1376 - val_mean_absolute_error: 0.2079\n",
      "Epoch 94/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0831 - mean_absolute_error: 0.1850 - val_loss: 0.1407 - val_mean_absolute_error: 0.2096\n",
      "Epoch 95/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0828 - mean_absolute_error: 0.1847 - val_loss: 0.1491 - val_mean_absolute_error: 0.2154\n",
      "Epoch 96/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0837 - mean_absolute_error: 0.1852 - val_loss: 0.1345 - val_mean_absolute_error: 0.2081\n",
      "Epoch 97/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0819 - mean_absolute_error: 0.1839 - val_loss: 0.1390 - val_mean_absolute_error: 0.2079\n",
      "Epoch 98/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0817 - mean_absolute_error: 0.1831 - val_loss: 0.1358 - val_mean_absolute_error: 0.2102\n",
      "Epoch 99/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0820 - mean_absolute_error: 0.1838 - val_loss: 0.1377 - val_mean_absolute_error: 0.2104\n",
      "Epoch 100/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0817 - mean_absolute_error: 0.1838 - val_loss: 0.1384 - val_mean_absolute_error: 0.2156\n",
      "Epoch 101/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0831 - mean_absolute_error: 0.1845 - val_loss: 0.1401 - val_mean_absolute_error: 0.2097\n",
      "Epoch 102/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0824 - mean_absolute_error: 0.1848 - val_loss: 0.1348 - val_mean_absolute_error: 0.2081\n",
      "Epoch 103/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0811 - mean_absolute_error: 0.1826 - val_loss: 0.1361 - val_mean_absolute_error: 0.2070\n",
      "Epoch 104/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0813 - mean_absolute_error: 0.1823 - val_loss: 0.1357 - val_mean_absolute_error: 0.2095\n",
      "Epoch 105/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0820 - mean_absolute_error: 0.1845 - val_loss: 0.1370 - val_mean_absolute_error: 0.2083\n",
      "Epoch 106/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0810 - mean_absolute_error: 0.1825 - val_loss: 0.1393 - val_mean_absolute_error: 0.2089\n",
      "Epoch 107/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0807 - mean_absolute_error: 0.1830 - val_loss: 0.1353 - val_mean_absolute_error: 0.2073\n",
      "Epoch 108/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0800 - mean_absolute_error: 0.1823 - val_loss: 0.1415 - val_mean_absolute_error: 0.2090\n",
      "Epoch 109/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0812 - mean_absolute_error: 0.1837 - val_loss: 0.1354 - val_mean_absolute_error: 0.2072\n",
      "Epoch 110/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0805 - mean_absolute_error: 0.1823 - val_loss: 0.1336 - val_mean_absolute_error: 0.2081\n",
      "Epoch 111/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0803 - mean_absolute_error: 0.1832 - val_loss: 0.1364 - val_mean_absolute_error: 0.2139\n",
      "Epoch 112/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0796 - mean_absolute_error: 0.1823 - val_loss: 0.1357 - val_mean_absolute_error: 0.2073\n",
      "Epoch 113/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0799 - mean_absolute_error: 0.1820 - val_loss: 0.1341 - val_mean_absolute_error: 0.2077\n",
      "Epoch 114/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0795 - mean_absolute_error: 0.1829 - val_loss: 0.1366 - val_mean_absolute_error: 0.2099\n",
      "Epoch 115/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0804 - mean_absolute_error: 0.1821 - val_loss: 0.1349 - val_mean_absolute_error: 0.2075\n",
      "Epoch 116/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0814 - mean_absolute_error: 0.1829 - val_loss: 0.1357 - val_mean_absolute_error: 0.2080\n",
      "Epoch 117/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0803 - mean_absolute_error: 0.1825 - val_loss: 0.1356 - val_mean_absolute_error: 0.2083\n",
      "Epoch 118/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0805 - mean_absolute_error: 0.1824 - val_loss: 0.1385 - val_mean_absolute_error: 0.2074\n",
      "Epoch 119/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0800 - mean_absolute_error: 0.1824 - val_loss: 0.1372 - val_mean_absolute_error: 0.2083\n",
      "Epoch 120/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0797 - mean_absolute_error: 0.1823 - val_loss: 0.1450 - val_mean_absolute_error: 0.2091\n",
      "Epoch 121/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0792 - mean_absolute_error: 0.1818 - val_loss: 0.1372 - val_mean_absolute_error: 0.2075\n",
      "Epoch 122/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0798 - mean_absolute_error: 0.1827 - val_loss: 0.1446 - val_mean_absolute_error: 0.2112\n",
      "Epoch 123/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0792 - mean_absolute_error: 0.1826 - val_loss: 0.1397 - val_mean_absolute_error: 0.2092\n",
      "Epoch 124/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0790 - mean_absolute_error: 0.1817 - val_loss: 0.1371 - val_mean_absolute_error: 0.2078\n",
      "Epoch 125/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0794 - mean_absolute_error: 0.1819 - val_loss: 0.1406 - val_mean_absolute_error: 0.2135\n",
      "Epoch 126/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0795 - mean_absolute_error: 0.1819 - val_loss: 0.1357 - val_mean_absolute_error: 0.2069\n",
      "Epoch 127/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0784 - mean_absolute_error: 0.1818 - val_loss: 0.1402 - val_mean_absolute_error: 0.2165\n",
      "Epoch 128/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0786 - mean_absolute_error: 0.1820 - val_loss: 0.1380 - val_mean_absolute_error: 0.2097\n",
      "Epoch 129/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0782 - mean_absolute_error: 0.1819 - val_loss: 0.1349 - val_mean_absolute_error: 0.2081\n",
      "Epoch 130/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0786 - mean_absolute_error: 0.1809 - val_loss: 0.1382 - val_mean_absolute_error: 0.2090\n",
      "Epoch 131/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0781 - mean_absolute_error: 0.1814 - val_loss: 0.1367 - val_mean_absolute_error: 0.2074\n",
      "Epoch 132/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0782 - mean_absolute_error: 0.1815 - val_loss: 0.1356 - val_mean_absolute_error: 0.2101\n",
      "Epoch 133/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0790 - mean_absolute_error: 0.1821 - val_loss: 0.1358 - val_mean_absolute_error: 0.2086\n",
      "Epoch 134/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0767 - mean_absolute_error: 0.1804 - val_loss: 0.1431 - val_mean_absolute_error: 0.2097\n",
      "Epoch 135/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0775 - mean_absolute_error: 0.1811 - val_loss: 0.1399 - val_mean_absolute_error: 0.2103\n",
      "Epoch 136/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0778 - mean_absolute_error: 0.1813 - val_loss: 0.1395 - val_mean_absolute_error: 0.2102\n",
      "Epoch 137/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0771 - mean_absolute_error: 0.1811 - val_loss: 0.1365 - val_mean_absolute_error: 0.2073\n",
      "Epoch 138/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0782 - mean_absolute_error: 0.1814 - val_loss: 0.1372 - val_mean_absolute_error: 0.2101\n",
      "Epoch 139/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0766 - mean_absolute_error: 0.1803 - val_loss: 0.1451 - val_mean_absolute_error: 0.2126\n",
      "Epoch 140/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0776 - mean_absolute_error: 0.1810 - val_loss: 0.1414 - val_mean_absolute_error: 0.2120\n",
      "Epoch 141/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0780 - mean_absolute_error: 0.1814 - val_loss: 0.1360 - val_mean_absolute_error: 0.2085\n",
      "Epoch 142/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0770 - mean_absolute_error: 0.1805 - val_loss: 0.1365 - val_mean_absolute_error: 0.2082\n",
      "Epoch 143/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0773 - mean_absolute_error: 0.1813 - val_loss: 0.1361 - val_mean_absolute_error: 0.2095\n",
      "Epoch 144/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0770 - mean_absolute_error: 0.1803 - val_loss: 0.1381 - val_mean_absolute_error: 0.2088\n",
      "Epoch 145/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0763 - mean_absolute_error: 0.1801 - val_loss: 0.1395 - val_mean_absolute_error: 0.2080\n",
      "Epoch 146/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0763 - mean_absolute_error: 0.1801 - val_loss: 0.1403 - val_mean_absolute_error: 0.2075\n",
      "Epoch 147/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0763 - mean_absolute_error: 0.1797 - val_loss: 0.1397 - val_mean_absolute_error: 0.2099\n",
      "Epoch 148/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0778 - mean_absolute_error: 0.1812 - val_loss: 0.1392 - val_mean_absolute_error: 0.2085\n",
      "Epoch 149/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0768 - mean_absolute_error: 0.1802 - val_loss: 0.1386 - val_mean_absolute_error: 0.2080\n",
      "Epoch 150/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0765 - mean_absolute_error: 0.1806 - val_loss: 0.1380 - val_mean_absolute_error: 0.2065\n",
      "Epoch 151/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0761 - mean_absolute_error: 0.1800 - val_loss: 0.1394 - val_mean_absolute_error: 0.2074\n",
      "Epoch 152/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0755 - mean_absolute_error: 0.1798 - val_loss: 0.1360 - val_mean_absolute_error: 0.2078\n",
      "Epoch 153/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0751 - mean_absolute_error: 0.1802 - val_loss: 0.1391 - val_mean_absolute_error: 0.2075\n",
      "Epoch 154/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0770 - mean_absolute_error: 0.1805 - val_loss: 0.1406 - val_mean_absolute_error: 0.2097\n",
      "Epoch 155/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0771 - mean_absolute_error: 0.1809 - val_loss: 0.1375 - val_mean_absolute_error: 0.2064\n",
      "Epoch 156/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0767 - mean_absolute_error: 0.1798 - val_loss: 0.1406 - val_mean_absolute_error: 0.2085\n",
      "Epoch 157/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0763 - mean_absolute_error: 0.1804 - val_loss: 0.1356 - val_mean_absolute_error: 0.2073\n",
      "Epoch 158/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0759 - mean_absolute_error: 0.1798 - val_loss: 0.1400 - val_mean_absolute_error: 0.2079\n",
      "Epoch 159/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0759 - mean_absolute_error: 0.1794 - val_loss: 0.1438 - val_mean_absolute_error: 0.2111\n",
      "Epoch 160/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0752 - mean_absolute_error: 0.1789 - val_loss: 0.1358 - val_mean_absolute_error: 0.2074\n",
      "Epoch 161/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0756 - mean_absolute_error: 0.1795 - val_loss: 0.1407 - val_mean_absolute_error: 0.2089\n",
      "Epoch 162/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0755 - mean_absolute_error: 0.1794 - val_loss: 0.1406 - val_mean_absolute_error: 0.2081\n",
      "Epoch 163/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0762 - mean_absolute_error: 0.1799 - val_loss: 0.1373 - val_mean_absolute_error: 0.2078\n",
      "Epoch 164/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0758 - mean_absolute_error: 0.1801 - val_loss: 0.1395 - val_mean_absolute_error: 0.2102\n",
      "Epoch 165/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0752 - mean_absolute_error: 0.1792 - val_loss: 0.1369 - val_mean_absolute_error: 0.2069\n",
      "Epoch 166/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0755 - mean_absolute_error: 0.1790 - val_loss: 0.1366 - val_mean_absolute_error: 0.2069\n",
      "Epoch 167/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0756 - mean_absolute_error: 0.1791 - val_loss: 0.1402 - val_mean_absolute_error: 0.2075\n",
      "Epoch 168/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0757 - mean_absolute_error: 0.1797 - val_loss: 0.1403 - val_mean_absolute_error: 0.2092\n",
      "Epoch 169/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0753 - mean_absolute_error: 0.1789 - val_loss: 0.1398 - val_mean_absolute_error: 0.2078\n",
      "Epoch 170/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0756 - mean_absolute_error: 0.1796 - val_loss: 0.1402 - val_mean_absolute_error: 0.2075\n",
      "Epoch 171/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0754 - mean_absolute_error: 0.1792 - val_loss: 0.1368 - val_mean_absolute_error: 0.2070\n",
      "Epoch 172/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0745 - mean_absolute_error: 0.1786 - val_loss: 0.1365 - val_mean_absolute_error: 0.2060\n",
      "Epoch 173/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0742 - mean_absolute_error: 0.1787 - val_loss: 0.1392 - val_mean_absolute_error: 0.2068\n",
      "Epoch 174/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0751 - mean_absolute_error: 0.1789 - val_loss: 0.1472 - val_mean_absolute_error: 0.2094\n",
      "Epoch 175/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0745 - mean_absolute_error: 0.1784 - val_loss: 0.1407 - val_mean_absolute_error: 0.2067\n",
      "Epoch 176/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0736 - mean_absolute_error: 0.1780 - val_loss: 0.1416 - val_mean_absolute_error: 0.2071\n",
      "Epoch 177/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0742 - mean_absolute_error: 0.1786 - val_loss: 0.1361 - val_mean_absolute_error: 0.2059\n",
      "Epoch 178/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0742 - mean_absolute_error: 0.1785 - val_loss: 0.1396 - val_mean_absolute_error: 0.2073\n",
      "Epoch 179/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0744 - mean_absolute_error: 0.1789 - val_loss: 0.1445 - val_mean_absolute_error: 0.2103\n",
      "Epoch 180/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0741 - mean_absolute_error: 0.1781 - val_loss: 0.1416 - val_mean_absolute_error: 0.2084\n",
      "Epoch 181/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0736 - mean_absolute_error: 0.1779 - val_loss: 0.1374 - val_mean_absolute_error: 0.2070\n",
      "Epoch 182/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0731 - mean_absolute_error: 0.1779 - val_loss: 0.1421 - val_mean_absolute_error: 0.2084\n",
      "Epoch 183/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0738 - mean_absolute_error: 0.1782 - val_loss: 0.1392 - val_mean_absolute_error: 0.2082\n",
      "Epoch 184/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0744 - mean_absolute_error: 0.1791 - val_loss: 0.1414 - val_mean_absolute_error: 0.2076\n",
      "Epoch 185/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0729 - mean_absolute_error: 0.1775 - val_loss: 0.1434 - val_mean_absolute_error: 0.2091\n",
      "Epoch 186/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0734 - mean_absolute_error: 0.1785 - val_loss: 0.1399 - val_mean_absolute_error: 0.2072\n",
      "Epoch 187/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0735 - mean_absolute_error: 0.1782 - val_loss: 0.1409 - val_mean_absolute_error: 0.2092\n",
      "Epoch 188/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0737 - mean_absolute_error: 0.1781 - val_loss: 0.1403 - val_mean_absolute_error: 0.2083\n",
      "Epoch 189/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0734 - mean_absolute_error: 0.1780 - val_loss: 0.1402 - val_mean_absolute_error: 0.2071\n",
      "Epoch 190/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0737 - mean_absolute_error: 0.1777 - val_loss: 0.1511 - val_mean_absolute_error: 0.2111\n",
      "Epoch 191/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0737 - mean_absolute_error: 0.1780 - val_loss: 0.1426 - val_mean_absolute_error: 0.2094\n",
      "Epoch 192/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0735 - mean_absolute_error: 0.1784 - val_loss: 0.1421 - val_mean_absolute_error: 0.2105\n",
      "Epoch 193/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0739 - mean_absolute_error: 0.1784 - val_loss: 0.1417 - val_mean_absolute_error: 0.2102\n",
      "Epoch 194/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0735 - mean_absolute_error: 0.1782 - val_loss: 0.1380 - val_mean_absolute_error: 0.2066\n",
      "Epoch 195/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0738 - mean_absolute_error: 0.1785 - val_loss: 0.1400 - val_mean_absolute_error: 0.2082\n",
      "Epoch 196/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0732 - mean_absolute_error: 0.1780 - val_loss: 0.1406 - val_mean_absolute_error: 0.2106\n",
      "Epoch 197/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0733 - mean_absolute_error: 0.1784 - val_loss: 0.1404 - val_mean_absolute_error: 0.2074\n",
      "Epoch 198/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0723 - mean_absolute_error: 0.1770 - val_loss: 0.1374 - val_mean_absolute_error: 0.2077\n",
      "Epoch 199/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0726 - mean_absolute_error: 0.1775 - val_loss: 0.1394 - val_mean_absolute_error: 0.2078\n",
      "Epoch 200/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0731 - mean_absolute_error: 0.1775 - val_loss: 0.1385 - val_mean_absolute_error: 0.2079\n",
      "Epoch 201/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0728 - mean_absolute_error: 0.1778 - val_loss: 0.1452 - val_mean_absolute_error: 0.2085\n",
      "Epoch 202/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0733 - mean_absolute_error: 0.1781 - val_loss: 0.1396 - val_mean_absolute_error: 0.2103\n",
      "Epoch 203/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0729 - mean_absolute_error: 0.1779 - val_loss: 0.1379 - val_mean_absolute_error: 0.2077\n",
      "Epoch 204/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0719 - mean_absolute_error: 0.1770 - val_loss: 0.1377 - val_mean_absolute_error: 0.2078\n",
      "Epoch 205/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0723 - mean_absolute_error: 0.1770 - val_loss: 0.1368 - val_mean_absolute_error: 0.2079\n",
      "Epoch 206/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0725 - mean_absolute_error: 0.1772 - val_loss: 0.1440 - val_mean_absolute_error: 0.2123\n",
      "Epoch 207/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0725 - mean_absolute_error: 0.1774 - val_loss: 0.1427 - val_mean_absolute_error: 0.2091\n",
      "Epoch 208/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0723 - mean_absolute_error: 0.1777 - val_loss: 0.1443 - val_mean_absolute_error: 0.2079\n",
      "Epoch 209/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0726 - mean_absolute_error: 0.1774 - val_loss: 0.1404 - val_mean_absolute_error: 0.2079\n",
      "Epoch 210/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0719 - mean_absolute_error: 0.1761 - val_loss: 0.1437 - val_mean_absolute_error: 0.2075\n",
      "Epoch 211/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0725 - mean_absolute_error: 0.1771 - val_loss: 0.1431 - val_mean_absolute_error: 0.2084\n",
      "Epoch 212/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0722 - mean_absolute_error: 0.1770 - val_loss: 0.1398 - val_mean_absolute_error: 0.2071\n",
      "Epoch 213/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0720 - mean_absolute_error: 0.1776 - val_loss: 0.1408 - val_mean_absolute_error: 0.2076\n",
      "Epoch 214/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0720 - mean_absolute_error: 0.1770 - val_loss: 0.1465 - val_mean_absolute_error: 0.2116\n",
      "Epoch 215/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0720 - mean_absolute_error: 0.1771 - val_loss: 0.1473 - val_mean_absolute_error: 0.2115\n",
      "Epoch 216/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0720 - mean_absolute_error: 0.1771 - val_loss: 0.1383 - val_mean_absolute_error: 0.2076\n",
      "Epoch 217/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0717 - mean_absolute_error: 0.1773 - val_loss: 0.1405 - val_mean_absolute_error: 0.2074\n",
      "Epoch 218/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0711 - mean_absolute_error: 0.1762 - val_loss: 0.1392 - val_mean_absolute_error: 0.2083\n",
      "Epoch 219/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0724 - mean_absolute_error: 0.1773 - val_loss: 0.1426 - val_mean_absolute_error: 0.2080\n",
      "Epoch 220/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0710 - mean_absolute_error: 0.1764 - val_loss: 0.1415 - val_mean_absolute_error: 0.2094\n",
      "Epoch 221/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0711 - mean_absolute_error: 0.1767 - val_loss: 0.1415 - val_mean_absolute_error: 0.2092\n",
      "Epoch 222/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0716 - mean_absolute_error: 0.1767 - val_loss: 0.1430 - val_mean_absolute_error: 0.2079\n",
      "Epoch 223/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0715 - mean_absolute_error: 0.1761 - val_loss: 0.1416 - val_mean_absolute_error: 0.2077\n",
      "Epoch 224/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0717 - mean_absolute_error: 0.1768 - val_loss: 0.1445 - val_mean_absolute_error: 0.2116\n",
      "Epoch 225/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0715 - mean_absolute_error: 0.1768 - val_loss: 0.1427 - val_mean_absolute_error: 0.2079\n",
      "Epoch 226/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0716 - mean_absolute_error: 0.1765 - val_loss: 0.1435 - val_mean_absolute_error: 0.2084\n",
      "Epoch 227/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0707 - mean_absolute_error: 0.1759 - val_loss: 0.1409 - val_mean_absolute_error: 0.2069\n",
      "Epoch 228/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0712 - mean_absolute_error: 0.1764 - val_loss: 0.1438 - val_mean_absolute_error: 0.2073\n",
      "Epoch 229/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0707 - mean_absolute_error: 0.1756 - val_loss: 0.1428 - val_mean_absolute_error: 0.2070\n",
      "Epoch 230/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0709 - mean_absolute_error: 0.1760 - val_loss: 0.1392 - val_mean_absolute_error: 0.2062\n",
      "Epoch 231/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0713 - mean_absolute_error: 0.1765 - val_loss: 0.1425 - val_mean_absolute_error: 0.2092\n",
      "Epoch 232/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0704 - mean_absolute_error: 0.1765 - val_loss: 0.1426 - val_mean_absolute_error: 0.2074\n",
      "Epoch 233/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0704 - mean_absolute_error: 0.1758 - val_loss: 0.1422 - val_mean_absolute_error: 0.2080\n",
      "Epoch 234/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0703 - mean_absolute_error: 0.1757 - val_loss: 0.1423 - val_mean_absolute_error: 0.2076\n",
      "Epoch 235/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0703 - mean_absolute_error: 0.1760 - val_loss: 0.1472 - val_mean_absolute_error: 0.2097\n",
      "Epoch 236/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0700 - mean_absolute_error: 0.1757 - val_loss: 0.1489 - val_mean_absolute_error: 0.2119\n",
      "Epoch 237/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0707 - mean_absolute_error: 0.1764 - val_loss: 0.1454 - val_mean_absolute_error: 0.2109\n",
      "Epoch 238/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0702 - mean_absolute_error: 0.1757 - val_loss: 0.1443 - val_mean_absolute_error: 0.2080\n",
      "Epoch 239/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0702 - mean_absolute_error: 0.1752 - val_loss: 0.1447 - val_mean_absolute_error: 0.2085\n",
      "Epoch 240/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0699 - mean_absolute_error: 0.1748 - val_loss: 0.1402 - val_mean_absolute_error: 0.2075\n",
      "Epoch 241/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0704 - mean_absolute_error: 0.1761 - val_loss: 0.1460 - val_mean_absolute_error: 0.2075\n",
      "Epoch 242/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0701 - mean_absolute_error: 0.1761 - val_loss: 0.1420 - val_mean_absolute_error: 0.2065\n",
      "Epoch 243/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0709 - mean_absolute_error: 0.1756 - val_loss: 0.1426 - val_mean_absolute_error: 0.2078\n",
      "Epoch 244/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0700 - mean_absolute_error: 0.1752 - val_loss: 0.1457 - val_mean_absolute_error: 0.2132\n",
      "Epoch 245/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0698 - mean_absolute_error: 0.1748 - val_loss: 0.1446 - val_mean_absolute_error: 0.2084\n",
      "Epoch 246/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0700 - mean_absolute_error: 0.1750 - val_loss: 0.1415 - val_mean_absolute_error: 0.2060\n",
      "Epoch 247/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0695 - mean_absolute_error: 0.1748 - val_loss: 0.1411 - val_mean_absolute_error: 0.2059\n",
      "Epoch 248/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0707 - mean_absolute_error: 0.1760 - val_loss: 0.1447 - val_mean_absolute_error: 0.2074\n",
      "Epoch 249/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0703 - mean_absolute_error: 0.1758 - val_loss: 0.1436 - val_mean_absolute_error: 0.2074\n",
      "Epoch 250/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0699 - mean_absolute_error: 0.1748 - val_loss: 0.1486 - val_mean_absolute_error: 0.2096\n",
      "Epoch 251/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0698 - mean_absolute_error: 0.1756 - val_loss: 0.1431 - val_mean_absolute_error: 0.2078\n",
      "Epoch 252/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0693 - mean_absolute_error: 0.1749 - val_loss: 0.1444 - val_mean_absolute_error: 0.2093\n",
      "Epoch 253/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0694 - mean_absolute_error: 0.1746 - val_loss: 0.1405 - val_mean_absolute_error: 0.2064\n",
      "Epoch 254/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0695 - mean_absolute_error: 0.1747 - val_loss: 0.1466 - val_mean_absolute_error: 0.2077\n",
      "Epoch 255/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0698 - mean_absolute_error: 0.1762 - val_loss: 0.1443 - val_mean_absolute_error: 0.2087\n",
      "Epoch 256/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0697 - mean_absolute_error: 0.1752 - val_loss: 0.1460 - val_mean_absolute_error: 0.2092\n",
      "Epoch 257/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0690 - mean_absolute_error: 0.1744 - val_loss: 0.1437 - val_mean_absolute_error: 0.2070\n",
      "Epoch 258/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0689 - mean_absolute_error: 0.1749 - val_loss: 0.1424 - val_mean_absolute_error: 0.2072\n",
      "Epoch 259/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0690 - mean_absolute_error: 0.1748 - val_loss: 0.1408 - val_mean_absolute_error: 0.2077\n",
      "Epoch 260/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0688 - mean_absolute_error: 0.1740 - val_loss: 0.1437 - val_mean_absolute_error: 0.2067\n",
      "Epoch 261/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0696 - mean_absolute_error: 0.1753 - val_loss: 0.1449 - val_mean_absolute_error: 0.2080\n",
      "Epoch 262/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0691 - mean_absolute_error: 0.1749 - val_loss: 0.1427 - val_mean_absolute_error: 0.2090\n",
      "Epoch 263/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0699 - mean_absolute_error: 0.1756 - val_loss: 0.1410 - val_mean_absolute_error: 0.2070\n",
      "Epoch 264/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0697 - mean_absolute_error: 0.1763 - val_loss: 0.1417 - val_mean_absolute_error: 0.2070\n",
      "Epoch 265/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0688 - mean_absolute_error: 0.1747 - val_loss: 0.1423 - val_mean_absolute_error: 0.2087\n",
      "Epoch 266/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0692 - mean_absolute_error: 0.1751 - val_loss: 0.1410 - val_mean_absolute_error: 0.2064\n",
      "Epoch 267/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0695 - mean_absolute_error: 0.1752 - val_loss: 0.1415 - val_mean_absolute_error: 0.2076\n",
      "Epoch 268/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0685 - mean_absolute_error: 0.1740 - val_loss: 0.1443 - val_mean_absolute_error: 0.2081\n",
      "Epoch 269/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0689 - mean_absolute_error: 0.1744 - val_loss: 0.1416 - val_mean_absolute_error: 0.2084\n",
      "Epoch 270/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0677 - mean_absolute_error: 0.1738 - val_loss: 0.1552 - val_mean_absolute_error: 0.2097\n",
      "Epoch 271/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0684 - mean_absolute_error: 0.1745 - val_loss: 0.1466 - val_mean_absolute_error: 0.2079\n",
      "Epoch 272/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0697 - mean_absolute_error: 0.1758 - val_loss: 0.1451 - val_mean_absolute_error: 0.2097\n",
      "Epoch 273/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0684 - mean_absolute_error: 0.1741 - val_loss: 0.1416 - val_mean_absolute_error: 0.2066\n",
      "Epoch 274/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0681 - mean_absolute_error: 0.1740 - val_loss: 0.1398 - val_mean_absolute_error: 0.2060\n",
      "Epoch 275/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0686 - mean_absolute_error: 0.1742 - val_loss: 0.1495 - val_mean_absolute_error: 0.2086\n",
      "Epoch 276/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0693 - mean_absolute_error: 0.1746 - val_loss: 0.1422 - val_mean_absolute_error: 0.2073\n",
      "Epoch 277/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0679 - mean_absolute_error: 0.1734 - val_loss: 0.1439 - val_mean_absolute_error: 0.2096\n",
      "Epoch 278/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0672 - mean_absolute_error: 0.1734 - val_loss: 0.1455 - val_mean_absolute_error: 0.2086\n",
      "Epoch 279/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0683 - mean_absolute_error: 0.1739 - val_loss: 0.1431 - val_mean_absolute_error: 0.2072\n",
      "Epoch 280/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0683 - mean_absolute_error: 0.1745 - val_loss: 0.1419 - val_mean_absolute_error: 0.2066\n",
      "Epoch 281/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0684 - mean_absolute_error: 0.1744 - val_loss: 0.1409 - val_mean_absolute_error: 0.2079\n",
      "Epoch 282/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0677 - mean_absolute_error: 0.1733 - val_loss: 0.1418 - val_mean_absolute_error: 0.2079\n",
      "Epoch 283/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0683 - mean_absolute_error: 0.1741 - val_loss: 0.1398 - val_mean_absolute_error: 0.2059\n",
      "Epoch 284/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0689 - mean_absolute_error: 0.1743 - val_loss: 0.1403 - val_mean_absolute_error: 0.2068\n",
      "Epoch 285/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0676 - mean_absolute_error: 0.1732 - val_loss: 0.1401 - val_mean_absolute_error: 0.2059\n",
      "Epoch 286/300\n",
      "12967/12967 [==============================] - 0s 32us/sample - loss: 0.0685 - mean_absolute_error: 0.1744 - val_loss: 0.1385 - val_mean_absolute_error: 0.2050\n",
      "Epoch 287/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0685 - mean_absolute_error: 0.1744 - val_loss: 0.1453 - val_mean_absolute_error: 0.2096\n",
      "Epoch 288/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0686 - mean_absolute_error: 0.1741 - val_loss: 0.1396 - val_mean_absolute_error: 0.2064\n",
      "Epoch 289/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0677 - mean_absolute_error: 0.1731 - val_loss: 0.1401 - val_mean_absolute_error: 0.2063\n",
      "Epoch 290/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0674 - mean_absolute_error: 0.1733 - val_loss: 0.1404 - val_mean_absolute_error: 0.2078\n",
      "Epoch 291/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0680 - mean_absolute_error: 0.1739 - val_loss: 0.1429 - val_mean_absolute_error: 0.2106\n",
      "Epoch 292/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0675 - mean_absolute_error: 0.1736 - val_loss: 0.1423 - val_mean_absolute_error: 0.2069\n",
      "Epoch 293/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0677 - mean_absolute_error: 0.1741 - val_loss: 0.1402 - val_mean_absolute_error: 0.2074\n",
      "Epoch 294/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0676 - mean_absolute_error: 0.1732 - val_loss: 0.1439 - val_mean_absolute_error: 0.2075\n",
      "Epoch 295/300\n",
      "12967/12967 [==============================] - 0s 31us/sample - loss: 0.0677 - mean_absolute_error: 0.1735 - val_loss: 0.1429 - val_mean_absolute_error: 0.2069\n",
      "Epoch 296/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0672 - mean_absolute_error: 0.1733 - val_loss: 0.1431 - val_mean_absolute_error: 0.2069\n",
      "Epoch 297/300\n",
      "12967/12967 [==============================] - 1s 43us/sample - loss: 0.0681 - mean_absolute_error: 0.1738 - val_loss: 0.1444 - val_mean_absolute_error: 0.2086\n",
      "Epoch 298/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0678 - mean_absolute_error: 0.1732 - val_loss: 0.1421 - val_mean_absolute_error: 0.2066\n",
      "Epoch 299/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0674 - mean_absolute_error: 0.1740 - val_loss: 0.1448 - val_mean_absolute_error: 0.2087\n",
      "Epoch 300/300\n",
      "12967/12967 [==============================] - 0s 30us/sample - loss: 0.0683 - mean_absolute_error: 0.1737 - val_loss: 0.1465 - val_mean_absolute_error: 0.2072\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x25d26c10ef0>"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_2 = keras.Sequential(name='model-2')\n",
    "model_2.add(layers.Dense(16, activation='relu', input_shape=(21,)))\n",
    "model_2.add(layers.Dense(16, activation='relu'))\n",
    "model_2.add(layers.Dense(1))\n",
    "\n",
    "model_2.compile(keras.optimizers.Adam(0.001),\n",
    "                loss=keras.losses.MeanSquaredError(),\n",
    "                metrics=[keras.metrics.MeanAbsoluteError()])\n",
    "\n",
    "log_dir = os.path.join('lab2-logs', 'model-2')\n",
    "model_cbk = keras.callbacks.TensorBoard(log_dir=log_dir)\n",
    "model_mckp = keras.callbacks.ModelCheckpoint(model_dir + '/Best-model-2.h5', \n",
    "                                             monitor='val_mean_absolute_error', \n",
    "                                             save_best_only=True, \n",
    "                                             mode='min')\n",
    "model_2.fit(x_train, y_train, \n",
    "            batch_size=64 ,\n",
    "            epochs=300, \n",
    "            validation_data=(x_val, y_val), \n",
    "            callbacks=[model_cbk, model_mckp])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 加入L1或L2 正則化"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 12967 samples, validate on 4323 samples\n",
      "Epoch 1/300\n",
      "12967/12967 [==============================] - 1s 98us/sample - loss: 0.3826 - mean_absolute_error: 0.3393 - val_loss: 0.3290 - val_mean_absolute_error: 0.2963\n",
      "Epoch 2/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.2858 - mean_absolute_error: 0.2791 - val_loss: 0.2955 - val_mean_absolute_error: 0.2822\n",
      "Epoch 3/300\n",
      "12967/12967 [==============================] - 1s 39us/sample - loss: 0.2540 - mean_absolute_error: 0.2632 - val_loss: 0.2761 - val_mean_absolute_error: 0.2700\n",
      "Epoch 4/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.2368 - mean_absolute_error: 0.2520 - val_loss: 0.2734 - val_mean_absolute_error: 0.2671\n",
      "Epoch 5/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.2212 - mean_absolute_error: 0.2427 - val_loss: 0.2591 - val_mean_absolute_error: 0.2530\n",
      "Epoch 6/300\n",
      "12967/12967 [==============================] - 1s 39us/sample - loss: 0.2063 - mean_absolute_error: 0.2345 - val_loss: 0.2394 - val_mean_absolute_error: 0.2422\n",
      "Epoch 7/300\n",
      "12967/12967 [==============================] - 1s 39us/sample - loss: 0.1993 - mean_absolute_error: 0.2276 - val_loss: 0.2408 - val_mean_absolute_error: 0.2542\n",
      "Epoch 8/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1884 - mean_absolute_error: 0.2210 - val_loss: 0.2297 - val_mean_absolute_error: 0.2390\n",
      "Epoch 9/300\n",
      "12967/12967 [==============================] - 1s 39us/sample - loss: 0.1777 - mean_absolute_error: 0.2138 - val_loss: 0.2301 - val_mean_absolute_error: 0.2300\n",
      "Epoch 10/300\n",
      "12967/12967 [==============================] - 1s 39us/sample - loss: 0.1735 - mean_absolute_error: 0.2115 - val_loss: 0.2133 - val_mean_absolute_error: 0.2230\n",
      "Epoch 11/300\n",
      "12967/12967 [==============================] - 1s 39us/sample - loss: 0.1663 - mean_absolute_error: 0.2053 - val_loss: 0.2120 - val_mean_absolute_error: 0.2215\n",
      "Epoch 12/300\n",
      "12967/12967 [==============================] - 1s 40us/sample - loss: 0.1600 - mean_absolute_error: 0.2022 - val_loss: 0.2011 - val_mean_absolute_error: 0.2147\n",
      "Epoch 13/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.1547 - mean_absolute_error: 0.1985 - val_loss: 0.1936 - val_mean_absolute_error: 0.2121\n",
      "Epoch 14/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.1562 - mean_absolute_error: 0.1996 - val_loss: 0.2023 - val_mean_absolute_error: 0.2176\n",
      "Epoch 15/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1483 - mean_absolute_error: 0.1949 - val_loss: 0.1927 - val_mean_absolute_error: 0.2187\n",
      "Epoch 16/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1481 - mean_absolute_error: 0.1970 - val_loss: 0.1940 - val_mean_absolute_error: 0.2155\n",
      "Epoch 17/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1417 - mean_absolute_error: 0.1947 - val_loss: 0.2049 - val_mean_absolute_error: 0.2242\n",
      "Epoch 18/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.1381 - mean_absolute_error: 0.1913 - val_loss: 0.1819 - val_mean_absolute_error: 0.2118\n",
      "Epoch 19/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.1378 - mean_absolute_error: 0.1914 - val_loss: 0.1818 - val_mean_absolute_error: 0.2086\n",
      "Epoch 20/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1377 - mean_absolute_error: 0.1917 - val_loss: 0.1794 - val_mean_absolute_error: 0.2063\n",
      "Epoch 21/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1301 - mean_absolute_error: 0.1871 - val_loss: 0.1832 - val_mean_absolute_error: 0.2093\n",
      "Epoch 22/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1297 - mean_absolute_error: 0.1876 - val_loss: 0.1768 - val_mean_absolute_error: 0.2058\n",
      "Epoch 23/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1280 - mean_absolute_error: 0.1866 - val_loss: 0.1816 - val_mean_absolute_error: 0.2088\n",
      "Epoch 24/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1270 - mean_absolute_error: 0.1863 - val_loss: 0.1788 - val_mean_absolute_error: 0.2093\n",
      "Epoch 25/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1270 - mean_absolute_error: 0.1884 - val_loss: 0.1762 - val_mean_absolute_error: 0.2092\n",
      "Epoch 26/300\n",
      "12967/12967 [==============================] - 1s 41us/sample - loss: 0.1273 - mean_absolute_error: 0.1888 - val_loss: 0.1810 - val_mean_absolute_error: 0.2085\n",
      "Epoch 27/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1251 - mean_absolute_error: 0.1855 - val_loss: 0.1731 - val_mean_absolute_error: 0.2085\n",
      "Epoch 28/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1212 - mean_absolute_error: 0.1853 - val_loss: 0.1731 - val_mean_absolute_error: 0.2083\n",
      "Epoch 29/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1194 - mean_absolute_error: 0.1831 - val_loss: 0.1768 - val_mean_absolute_error: 0.2031\n",
      "Epoch 30/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.1216 - mean_absolute_error: 0.1852 - val_loss: 0.1775 - val_mean_absolute_error: 0.2156\n",
      "Epoch 31/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1167 - mean_absolute_error: 0.1827 - val_loss: 0.1663 - val_mean_absolute_error: 0.2034\n",
      "Epoch 32/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1196 - mean_absolute_error: 0.1848 - val_loss: 0.1762 - val_mean_absolute_error: 0.2084\n",
      "Epoch 33/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1140 - mean_absolute_error: 0.1816 - val_loss: 0.1757 - val_mean_absolute_error: 0.2029\n",
      "Epoch 34/300\n",
      "12967/12967 [==============================] - 1s 39us/sample - loss: 0.1146 - mean_absolute_error: 0.1797 - val_loss: 0.2008 - val_mean_absolute_error: 0.2078\n",
      "Epoch 35/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.1146 - mean_absolute_error: 0.1824 - val_loss: 0.1667 - val_mean_absolute_error: 0.2059\n",
      "Epoch 36/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.1134 - mean_absolute_error: 0.1809 - val_loss: 0.1758 - val_mean_absolute_error: 0.2056\n",
      "Epoch 37/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1129 - mean_absolute_error: 0.1802 - val_loss: 0.1715 - val_mean_absolute_error: 0.2036\n",
      "Epoch 38/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1117 - mean_absolute_error: 0.1804 - val_loss: 0.1826 - val_mean_absolute_error: 0.2167\n",
      "Epoch 39/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1086 - mean_absolute_error: 0.1786 - val_loss: 0.1677 - val_mean_absolute_error: 0.2012\n",
      "Epoch 40/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1096 - mean_absolute_error: 0.1781 - val_loss: 0.1764 - val_mean_absolute_error: 0.2026\n",
      "Epoch 41/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1093 - mean_absolute_error: 0.1777 - val_loss: 0.1636 - val_mean_absolute_error: 0.2037\n",
      "Epoch 42/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1087 - mean_absolute_error: 0.1786 - val_loss: 0.1647 - val_mean_absolute_error: 0.2001\n",
      "Epoch 43/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1039 - mean_absolute_error: 0.1764 - val_loss: 0.1651 - val_mean_absolute_error: 0.1992\n",
      "Epoch 44/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1033 - mean_absolute_error: 0.1762 - val_loss: 0.1639 - val_mean_absolute_error: 0.2064\n",
      "Epoch 45/300\n",
      "12967/12967 [==============================] - 1s 39us/sample - loss: 0.1053 - mean_absolute_error: 0.1784 - val_loss: 0.1644 - val_mean_absolute_error: 0.1991\n",
      "Epoch 46/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.1060 - mean_absolute_error: 0.1767 - val_loss: 0.1683 - val_mean_absolute_error: 0.2003\n",
      "Epoch 47/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.1047 - mean_absolute_error: 0.1765 - val_loss: 0.1616 - val_mean_absolute_error: 0.2003\n",
      "Epoch 48/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1040 - mean_absolute_error: 0.1758 - val_loss: 0.1794 - val_mean_absolute_error: 0.2257\n",
      "Epoch 49/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.1029 - mean_absolute_error: 0.1773 - val_loss: 0.1605 - val_mean_absolute_error: 0.2029\n",
      "Epoch 50/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.1017 - mean_absolute_error: 0.1748 - val_loss: 0.1612 - val_mean_absolute_error: 0.1971\n",
      "Epoch 51/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.1010 - mean_absolute_error: 0.1738 - val_loss: 0.1639 - val_mean_absolute_error: 0.2006\n",
      "Epoch 52/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1015 - mean_absolute_error: 0.1754 - val_loss: 0.1579 - val_mean_absolute_error: 0.1977\n",
      "Epoch 53/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.1021 - mean_absolute_error: 0.1746 - val_loss: 0.1696 - val_mean_absolute_error: 0.2072\n",
      "Epoch 54/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0996 - mean_absolute_error: 0.1727 - val_loss: 0.1536 - val_mean_absolute_error: 0.1990\n",
      "Epoch 55/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.1011 - mean_absolute_error: 0.1745 - val_loss: 0.1621 - val_mean_absolute_error: 0.1982\n",
      "Epoch 56/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0978 - mean_absolute_error: 0.1726 - val_loss: 0.1746 - val_mean_absolute_error: 0.2099\n",
      "Epoch 57/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0960 - mean_absolute_error: 0.1718 - val_loss: 0.1582 - val_mean_absolute_error: 0.1975\n",
      "Epoch 58/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.1075 - mean_absolute_error: 0.1781 - val_loss: 0.1521 - val_mean_absolute_error: 0.1966\n",
      "Epoch 59/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0950 - mean_absolute_error: 0.1718 - val_loss: 0.1610 - val_mean_absolute_error: 0.1976\n",
      "Epoch 60/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.0945 - mean_absolute_error: 0.1722 - val_loss: 0.1586 - val_mean_absolute_error: 0.1992\n",
      "Epoch 61/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0971 - mean_absolute_error: 0.1715 - val_loss: 0.1651 - val_mean_absolute_error: 0.1991\n",
      "Epoch 62/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0946 - mean_absolute_error: 0.1712 - val_loss: 0.1627 - val_mean_absolute_error: 0.2005\n",
      "Epoch 63/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0952 - mean_absolute_error: 0.1720 - val_loss: 0.1592 - val_mean_absolute_error: 0.1999\n",
      "Epoch 64/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0954 - mean_absolute_error: 0.1714 - val_loss: 0.1615 - val_mean_absolute_error: 0.1978\n",
      "Epoch 65/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0954 - mean_absolute_error: 0.1729 - val_loss: 0.1576 - val_mean_absolute_error: 0.2018\n",
      "Epoch 66/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0982 - mean_absolute_error: 0.1740 - val_loss: 0.1540 - val_mean_absolute_error: 0.1967\n",
      "Epoch 67/300\n",
      "12967/12967 [==============================] - 1s 41us/sample - loss: 0.0974 - mean_absolute_error: 0.1736 - val_loss: 0.1513 - val_mean_absolute_error: 0.1983\n",
      "Epoch 68/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0949 - mean_absolute_error: 0.1740 - val_loss: 0.1526 - val_mean_absolute_error: 0.1963\n",
      "Epoch 69/300\n",
      "12967/12967 [==============================] - 1s 79us/sample - loss: 0.0924 - mean_absolute_error: 0.1694 - val_loss: 0.1663 - val_mean_absolute_error: 0.2006\n",
      "Epoch 70/300\n",
      "12967/12967 [==============================] - 1s 73us/sample - loss: 0.0928 - mean_absolute_error: 0.1712 - val_loss: 0.1556 - val_mean_absolute_error: 0.1999\n",
      "Epoch 71/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.0909 - mean_absolute_error: 0.1687 - val_loss: 0.1707 - val_mean_absolute_error: 0.2062\n",
      "Epoch 72/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0940 - mean_absolute_error: 0.1733 - val_loss: 0.1756 - val_mean_absolute_error: 0.2080\n",
      "Epoch 73/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0931 - mean_absolute_error: 0.1714 - val_loss: 0.1874 - val_mean_absolute_error: 0.2179\n",
      "Epoch 74/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0927 - mean_absolute_error: 0.1705 - val_loss: 0.1629 - val_mean_absolute_error: 0.2008\n",
      "Epoch 75/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0924 - mean_absolute_error: 0.1701 - val_loss: 0.1702 - val_mean_absolute_error: 0.2045\n",
      "Epoch 76/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0904 - mean_absolute_error: 0.1681 - val_loss: 0.1576 - val_mean_absolute_error: 0.1981\n",
      "Epoch 77/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0924 - mean_absolute_error: 0.1709 - val_loss: 0.1579 - val_mean_absolute_error: 0.1966\n",
      "Epoch 78/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0905 - mean_absolute_error: 0.1701 - val_loss: 0.1620 - val_mean_absolute_error: 0.2073\n",
      "Epoch 79/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0898 - mean_absolute_error: 0.1688 - val_loss: 0.1523 - val_mean_absolute_error: 0.1978\n",
      "Epoch 80/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0879 - mean_absolute_error: 0.1677 - val_loss: 0.1497 - val_mean_absolute_error: 0.1956\n",
      "Epoch 81/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0874 - mean_absolute_error: 0.1674 - val_loss: 0.1625 - val_mean_absolute_error: 0.2006\n",
      "Epoch 82/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0893 - mean_absolute_error: 0.1682 - val_loss: 0.1667 - val_mean_absolute_error: 0.2033\n",
      "Epoch 83/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0879 - mean_absolute_error: 0.1671 - val_loss: 0.1593 - val_mean_absolute_error: 0.1993\n",
      "Epoch 84/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0877 - mean_absolute_error: 0.1685 - val_loss: 0.1476 - val_mean_absolute_error: 0.1977\n",
      "Epoch 85/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0896 - mean_absolute_error: 0.1692 - val_loss: 0.1549 - val_mean_absolute_error: 0.2000\n",
      "Epoch 86/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0894 - mean_absolute_error: 0.1684 - val_loss: 0.1561 - val_mean_absolute_error: 0.1952\n",
      "Epoch 87/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0864 - mean_absolute_error: 0.1668 - val_loss: 0.1562 - val_mean_absolute_error: 0.1987\n",
      "Epoch 88/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0888 - mean_absolute_error: 0.1698 - val_loss: 0.1682 - val_mean_absolute_error: 0.2070\n",
      "Epoch 89/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0868 - mean_absolute_error: 0.1666 - val_loss: 0.1729 - val_mean_absolute_error: 0.2130\n",
      "Epoch 90/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0890 - mean_absolute_error: 0.1684 - val_loss: 0.1714 - val_mean_absolute_error: 0.2168\n",
      "Epoch 91/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0842 - mean_absolute_error: 0.1658 - val_loss: 0.1518 - val_mean_absolute_error: 0.1959\n",
      "Epoch 92/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0852 - mean_absolute_error: 0.1659 - val_loss: 0.1587 - val_mean_absolute_error: 0.1993\n",
      "Epoch 93/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0849 - mean_absolute_error: 0.1662 - val_loss: 0.1602 - val_mean_absolute_error: 0.2019\n",
      "Epoch 94/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0856 - mean_absolute_error: 0.1670 - val_loss: 0.1502 - val_mean_absolute_error: 0.1983\n",
      "Epoch 95/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0844 - mean_absolute_error: 0.1658 - val_loss: 0.1499 - val_mean_absolute_error: 0.1973\n",
      "Epoch 96/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0880 - mean_absolute_error: 0.1683 - val_loss: 0.1813 - val_mean_absolute_error: 0.2019\n",
      "Epoch 97/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0902 - mean_absolute_error: 0.1704 - val_loss: 0.1533 - val_mean_absolute_error: 0.2085\n",
      "Epoch 98/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0869 - mean_absolute_error: 0.1674 - val_loss: 0.1545 - val_mean_absolute_error: 0.2020\n",
      "Epoch 99/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0884 - mean_absolute_error: 0.1682 - val_loss: 0.1508 - val_mean_absolute_error: 0.1959\n",
      "Epoch 100/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0865 - mean_absolute_error: 0.1667 - val_loss: 0.1560 - val_mean_absolute_error: 0.2092\n",
      "Epoch 101/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0818 - mean_absolute_error: 0.1641 - val_loss: 0.1563 - val_mean_absolute_error: 0.1998\n",
      "Epoch 102/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0847 - mean_absolute_error: 0.1659 - val_loss: 0.1571 - val_mean_absolute_error: 0.1998\n",
      "Epoch 103/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0862 - mean_absolute_error: 0.1683 - val_loss: 0.1500 - val_mean_absolute_error: 0.1972\n",
      "Epoch 104/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0829 - mean_absolute_error: 0.1643 - val_loss: 0.1494 - val_mean_absolute_error: 0.1960\n",
      "Epoch 105/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0863 - mean_absolute_error: 0.1662 - val_loss: 0.1529 - val_mean_absolute_error: 0.1955\n",
      "Epoch 106/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0827 - mean_absolute_error: 0.1651 - val_loss: 0.1561 - val_mean_absolute_error: 0.1998\n",
      "Epoch 107/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0828 - mean_absolute_error: 0.1645 - val_loss: 0.1520 - val_mean_absolute_error: 0.1979\n",
      "Epoch 108/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0814 - mean_absolute_error: 0.1638 - val_loss: 0.1495 - val_mean_absolute_error: 0.1964\n",
      "Epoch 109/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0812 - mean_absolute_error: 0.1630 - val_loss: 0.1546 - val_mean_absolute_error: 0.1980\n",
      "Epoch 110/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0813 - mean_absolute_error: 0.1641 - val_loss: 0.1510 - val_mean_absolute_error: 0.1928\n",
      "Epoch 111/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0843 - mean_absolute_error: 0.1665 - val_loss: 0.1424 - val_mean_absolute_error: 0.1926\n",
      "Epoch 112/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0804 - mean_absolute_error: 0.1637 - val_loss: 0.1500 - val_mean_absolute_error: 0.1973\n",
      "Epoch 113/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0798 - mean_absolute_error: 0.1624 - val_loss: 0.1545 - val_mean_absolute_error: 0.1988\n",
      "Epoch 114/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0868 - mean_absolute_error: 0.1672 - val_loss: 0.1517 - val_mean_absolute_error: 0.1994\n",
      "Epoch 115/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0817 - mean_absolute_error: 0.1638 - val_loss: 0.1513 - val_mean_absolute_error: 0.1949\n",
      "Epoch 116/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0798 - mean_absolute_error: 0.1616 - val_loss: 0.1478 - val_mean_absolute_error: 0.1941\n",
      "Epoch 117/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0821 - mean_absolute_error: 0.1648 - val_loss: 0.1556 - val_mean_absolute_error: 0.2099\n",
      "Epoch 118/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0813 - mean_absolute_error: 0.1649 - val_loss: 0.1452 - val_mean_absolute_error: 0.1922\n",
      "Epoch 119/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0796 - mean_absolute_error: 0.1610 - val_loss: 0.1466 - val_mean_absolute_error: 0.1983\n",
      "Epoch 120/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0807 - mean_absolute_error: 0.1632 - val_loss: 0.1527 - val_mean_absolute_error: 0.1948\n",
      "Epoch 121/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0819 - mean_absolute_error: 0.1642 - val_loss: 0.1478 - val_mean_absolute_error: 0.1988\n",
      "Epoch 122/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0872 - mean_absolute_error: 0.1679 - val_loss: 0.1555 - val_mean_absolute_error: 0.2000\n",
      "Epoch 123/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0851 - mean_absolute_error: 0.1650 - val_loss: 0.1471 - val_mean_absolute_error: 0.1988\n",
      "Epoch 124/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0840 - mean_absolute_error: 0.1658 - val_loss: 0.1435 - val_mean_absolute_error: 0.1914\n",
      "Epoch 125/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0803 - mean_absolute_error: 0.1632 - val_loss: 0.1460 - val_mean_absolute_error: 0.1943\n",
      "Epoch 126/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0806 - mean_absolute_error: 0.1645 - val_loss: 0.1452 - val_mean_absolute_error: 0.1947\n",
      "Epoch 127/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0774 - mean_absolute_error: 0.1607 - val_loss: 0.1498 - val_mean_absolute_error: 0.1950\n",
      "Epoch 128/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0832 - mean_absolute_error: 0.1664 - val_loss: 0.1491 - val_mean_absolute_error: 0.1945\n",
      "Epoch 129/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0786 - mean_absolute_error: 0.1621 - val_loss: 0.1528 - val_mean_absolute_error: 0.1974\n",
      "Epoch 130/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0816 - mean_absolute_error: 0.1638 - val_loss: 0.1594 - val_mean_absolute_error: 0.1973\n",
      "Epoch 131/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0808 - mean_absolute_error: 0.1619 - val_loss: 0.1672 - val_mean_absolute_error: 0.1995\n",
      "Epoch 132/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0818 - mean_absolute_error: 0.1654 - val_loss: 0.1473 - val_mean_absolute_error: 0.1964\n",
      "Epoch 133/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0775 - mean_absolute_error: 0.1603 - val_loss: 0.1538 - val_mean_absolute_error: 0.1992\n",
      "Epoch 134/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0769 - mean_absolute_error: 0.1609 - val_loss: 0.1494 - val_mean_absolute_error: 0.1959\n",
      "Epoch 135/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0871 - mean_absolute_error: 0.1655 - val_loss: 0.1587 - val_mean_absolute_error: 0.2136\n",
      "Epoch 136/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0807 - mean_absolute_error: 0.1644 - val_loss: 0.1477 - val_mean_absolute_error: 0.1947\n",
      "Epoch 137/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0779 - mean_absolute_error: 0.1608 - val_loss: 0.1688 - val_mean_absolute_error: 0.2017\n",
      "Epoch 138/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0790 - mean_absolute_error: 0.1623 - val_loss: 0.1520 - val_mean_absolute_error: 0.1950\n",
      "Epoch 139/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0773 - mean_absolute_error: 0.1609 - val_loss: 0.1466 - val_mean_absolute_error: 0.1966\n",
      "Epoch 140/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0783 - mean_absolute_error: 0.1616 - val_loss: 0.1546 - val_mean_absolute_error: 0.1998\n",
      "Epoch 141/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0791 - mean_absolute_error: 0.1622 - val_loss: 0.1461 - val_mean_absolute_error: 0.1952\n",
      "Epoch 142/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0762 - mean_absolute_error: 0.1602 - val_loss: 0.1490 - val_mean_absolute_error: 0.1947\n",
      "Epoch 143/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0774 - mean_absolute_error: 0.1607 - val_loss: 0.1565 - val_mean_absolute_error: 0.1943\n",
      "Epoch 144/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0796 - mean_absolute_error: 0.1639 - val_loss: 0.1721 - val_mean_absolute_error: 0.2053\n",
      "Epoch 145/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0788 - mean_absolute_error: 0.1624 - val_loss: 0.1480 - val_mean_absolute_error: 0.1978\n",
      "Epoch 146/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0804 - mean_absolute_error: 0.1632 - val_loss: 0.1466 - val_mean_absolute_error: 0.1940\n",
      "Epoch 147/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0795 - mean_absolute_error: 0.1626 - val_loss: 0.1510 - val_mean_absolute_error: 0.1961\n",
      "Epoch 148/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0799 - mean_absolute_error: 0.1634 - val_loss: 0.1446 - val_mean_absolute_error: 0.1975\n",
      "Epoch 149/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0831 - mean_absolute_error: 0.1656 - val_loss: 0.1392 - val_mean_absolute_error: 0.1927\n",
      "Epoch 150/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0786 - mean_absolute_error: 0.1620 - val_loss: 0.1457 - val_mean_absolute_error: 0.1950\n",
      "Epoch 151/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0766 - mean_absolute_error: 0.1597 - val_loss: 0.1428 - val_mean_absolute_error: 0.1940\n",
      "Epoch 152/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0750 - mean_absolute_error: 0.1589 - val_loss: 0.1558 - val_mean_absolute_error: 0.1939\n",
      "Epoch 153/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0756 - mean_absolute_error: 0.1599 - val_loss: 0.1484 - val_mean_absolute_error: 0.1960\n",
      "Epoch 154/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0767 - mean_absolute_error: 0.1607 - val_loss: 0.1461 - val_mean_absolute_error: 0.1909\n",
      "Epoch 155/300\n",
      "12967/12967 [==============================] - 0s 37us/sample - loss: 0.0802 - mean_absolute_error: 0.1645 - val_loss: 0.1538 - val_mean_absolute_error: 0.2002\n",
      "Epoch 156/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0790 - mean_absolute_error: 0.1615 - val_loss: 0.1482 - val_mean_absolute_error: 0.1985\n",
      "Epoch 157/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0764 - mean_absolute_error: 0.1606 - val_loss: 0.1447 - val_mean_absolute_error: 0.1950\n",
      "Epoch 158/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0757 - mean_absolute_error: 0.1591 - val_loss: 0.1511 - val_mean_absolute_error: 0.1959\n",
      "Epoch 159/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0771 - mean_absolute_error: 0.1600 - val_loss: 0.1460 - val_mean_absolute_error: 0.1965\n",
      "Epoch 160/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0800 - mean_absolute_error: 0.1633 - val_loss: 0.1513 - val_mean_absolute_error: 0.1977\n",
      "Epoch 161/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0760 - mean_absolute_error: 0.1599 - val_loss: 0.1499 - val_mean_absolute_error: 0.1962\n",
      "Epoch 162/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0777 - mean_absolute_error: 0.1617 - val_loss: 0.1443 - val_mean_absolute_error: 0.1933\n",
      "Epoch 163/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0754 - mean_absolute_error: 0.1596 - val_loss: 0.1473 - val_mean_absolute_error: 0.1974\n",
      "Epoch 164/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0766 - mean_absolute_error: 0.1609 - val_loss: 0.1522 - val_mean_absolute_error: 0.1975\n",
      "Epoch 165/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0757 - mean_absolute_error: 0.1594 - val_loss: 0.1465 - val_mean_absolute_error: 0.1942\n",
      "Epoch 166/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0774 - mean_absolute_error: 0.1625 - val_loss: 0.1451 - val_mean_absolute_error: 0.1984\n",
      "Epoch 167/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0780 - mean_absolute_error: 0.1630 - val_loss: 0.1506 - val_mean_absolute_error: 0.1942\n",
      "Epoch 168/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0756 - mean_absolute_error: 0.1599 - val_loss: 0.1451 - val_mean_absolute_error: 0.1949\n",
      "Epoch 169/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0761 - mean_absolute_error: 0.1598 - val_loss: 0.1557 - val_mean_absolute_error: 0.2033\n",
      "Epoch 170/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0770 - mean_absolute_error: 0.1609 - val_loss: 0.1701 - val_mean_absolute_error: 0.1996\n",
      "Epoch 171/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0782 - mean_absolute_error: 0.1627 - val_loss: 0.1527 - val_mean_absolute_error: 0.1946\n",
      "Epoch 172/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0771 - mean_absolute_error: 0.1612 - val_loss: 0.1529 - val_mean_absolute_error: 0.1961\n",
      "Epoch 173/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0752 - mean_absolute_error: 0.1593 - val_loss: 0.1538 - val_mean_absolute_error: 0.2000\n",
      "Epoch 174/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0761 - mean_absolute_error: 0.1596 - val_loss: 0.1500 - val_mean_absolute_error: 0.1945\n",
      "Epoch 175/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0759 - mean_absolute_error: 0.1588 - val_loss: 0.1461 - val_mean_absolute_error: 0.1983\n",
      "Epoch 176/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0836 - mean_absolute_error: 0.1652 - val_loss: 0.1791 - val_mean_absolute_error: 0.2130\n",
      "Epoch 177/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0806 - mean_absolute_error: 0.1635 - val_loss: 0.1547 - val_mean_absolute_error: 0.1989\n",
      "Epoch 178/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0746 - mean_absolute_error: 0.1590 - val_loss: 0.1471 - val_mean_absolute_error: 0.1964\n",
      "Epoch 179/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0765 - mean_absolute_error: 0.1600 - val_loss: 0.1581 - val_mean_absolute_error: 0.2002\n",
      "Epoch 180/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0759 - mean_absolute_error: 0.1604 - val_loss: 0.1437 - val_mean_absolute_error: 0.1937\n",
      "Epoch 181/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0749 - mean_absolute_error: 0.1588 - val_loss: 0.1485 - val_mean_absolute_error: 0.1931\n",
      "Epoch 182/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0739 - mean_absolute_error: 0.1584 - val_loss: 0.1670 - val_mean_absolute_error: 0.2009\n",
      "Epoch 183/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0760 - mean_absolute_error: 0.1612 - val_loss: 0.1568 - val_mean_absolute_error: 0.1962\n",
      "Epoch 184/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0753 - mean_absolute_error: 0.1595 - val_loss: 0.1436 - val_mean_absolute_error: 0.1914\n",
      "Epoch 185/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0771 - mean_absolute_error: 0.1616 - val_loss: 0.1491 - val_mean_absolute_error: 0.1982\n",
      "Epoch 186/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0753 - mean_absolute_error: 0.1595 - val_loss: 0.1607 - val_mean_absolute_error: 0.1974\n",
      "Epoch 187/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0754 - mean_absolute_error: 0.1599 - val_loss: 0.1527 - val_mean_absolute_error: 0.1951\n",
      "Epoch 188/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0747 - mean_absolute_error: 0.1588 - val_loss: 0.1515 - val_mean_absolute_error: 0.1933\n",
      "Epoch 189/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0784 - mean_absolute_error: 0.1631 - val_loss: 0.1464 - val_mean_absolute_error: 0.1958\n",
      "Epoch 190/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0745 - mean_absolute_error: 0.1589 - val_loss: 0.1525 - val_mean_absolute_error: 0.1973\n",
      "Epoch 191/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0741 - mean_absolute_error: 0.1589 - val_loss: 0.1507 - val_mean_absolute_error: 0.1988\n",
      "Epoch 192/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0719 - mean_absolute_error: 0.1571 - val_loss: 0.1533 - val_mean_absolute_error: 0.1953\n",
      "Epoch 193/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0747 - mean_absolute_error: 0.1600 - val_loss: 0.1498 - val_mean_absolute_error: 0.1993\n",
      "Epoch 194/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0735 - mean_absolute_error: 0.1578 - val_loss: 0.1542 - val_mean_absolute_error: 0.1959\n",
      "Epoch 195/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0777 - mean_absolute_error: 0.1609 - val_loss: 0.1554 - val_mean_absolute_error: 0.2024\n",
      "Epoch 196/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0765 - mean_absolute_error: 0.1600 - val_loss: 0.1474 - val_mean_absolute_error: 0.1975\n",
      "Epoch 197/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0724 - mean_absolute_error: 0.1566 - val_loss: 0.1466 - val_mean_absolute_error: 0.1944\n",
      "Epoch 198/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0724 - mean_absolute_error: 0.1572 - val_loss: 0.1398 - val_mean_absolute_error: 0.1917\n",
      "Epoch 199/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0719 - mean_absolute_error: 0.1563 - val_loss: 0.1550 - val_mean_absolute_error: 0.2014\n",
      "Epoch 200/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0747 - mean_absolute_error: 0.1587 - val_loss: 0.1534 - val_mean_absolute_error: 0.1994\n",
      "Epoch 201/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0720 - mean_absolute_error: 0.1566 - val_loss: 0.1537 - val_mean_absolute_error: 0.1958\n",
      "Epoch 202/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0718 - mean_absolute_error: 0.1569 - val_loss: 0.1473 - val_mean_absolute_error: 0.1952\n",
      "Epoch 203/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0749 - mean_absolute_error: 0.1603 - val_loss: 0.1453 - val_mean_absolute_error: 0.1980\n",
      "Epoch 204/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0748 - mean_absolute_error: 0.1604 - val_loss: 0.1543 - val_mean_absolute_error: 0.1974\n",
      "Epoch 205/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0728 - mean_absolute_error: 0.1572 - val_loss: 0.1497 - val_mean_absolute_error: 0.1986\n",
      "Epoch 206/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.0721 - mean_absolute_error: 0.1574 - val_loss: 0.1569 - val_mean_absolute_error: 0.1987\n",
      "Epoch 207/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0733 - mean_absolute_error: 0.1589 - val_loss: 0.1403 - val_mean_absolute_error: 0.1925\n",
      "Epoch 208/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0743 - mean_absolute_error: 0.1591 - val_loss: 0.1516 - val_mean_absolute_error: 0.1950\n",
      "Epoch 209/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0736 - mean_absolute_error: 0.1578 - val_loss: 0.1583 - val_mean_absolute_error: 0.2013\n",
      "Epoch 210/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0734 - mean_absolute_error: 0.1586 - val_loss: 0.1576 - val_mean_absolute_error: 0.1979\n",
      "Epoch 211/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0742 - mean_absolute_error: 0.1596 - val_loss: 0.1612 - val_mean_absolute_error: 0.2002\n",
      "Epoch 212/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0720 - mean_absolute_error: 0.1564 - val_loss: 0.1517 - val_mean_absolute_error: 0.1954\n",
      "Epoch 213/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0704 - mean_absolute_error: 0.1551 - val_loss: 0.1492 - val_mean_absolute_error: 0.1970\n",
      "Epoch 214/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0744 - mean_absolute_error: 0.1597 - val_loss: 0.1487 - val_mean_absolute_error: 0.1991\n",
      "Epoch 215/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0739 - mean_absolute_error: 0.1593 - val_loss: 0.1444 - val_mean_absolute_error: 0.1941\n",
      "Epoch 216/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0710 - mean_absolute_error: 0.1557 - val_loss: 0.1496 - val_mean_absolute_error: 0.1957\n",
      "Epoch 217/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0731 - mean_absolute_error: 0.1569 - val_loss: 0.1453 - val_mean_absolute_error: 0.1970\n",
      "Epoch 218/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0734 - mean_absolute_error: 0.1577 - val_loss: 0.1487 - val_mean_absolute_error: 0.1984\n",
      "Epoch 219/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0736 - mean_absolute_error: 0.1586 - val_loss: 0.1575 - val_mean_absolute_error: 0.2014\n",
      "Epoch 220/300\n",
      "12967/12967 [==============================] - 1s 42us/sample - loss: 0.0739 - mean_absolute_error: 0.1595 - val_loss: 0.1482 - val_mean_absolute_error: 0.2010\n",
      "Epoch 221/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0721 - mean_absolute_error: 0.1572 - val_loss: 0.1557 - val_mean_absolute_error: 0.1981\n",
      "Epoch 222/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0715 - mean_absolute_error: 0.1571 - val_loss: 0.1439 - val_mean_absolute_error: 0.1951\n",
      "Epoch 223/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0734 - mean_absolute_error: 0.1583 - val_loss: 0.1489 - val_mean_absolute_error: 0.1976\n",
      "Epoch 224/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0740 - mean_absolute_error: 0.1587 - val_loss: 0.1602 - val_mean_absolute_error: 0.2065\n",
      "Epoch 225/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0789 - mean_absolute_error: 0.1623 - val_loss: 0.1585 - val_mean_absolute_error: 0.2025\n",
      "Epoch 226/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0845 - mean_absolute_error: 0.1658 - val_loss: 0.1480 - val_mean_absolute_error: 0.1998\n",
      "Epoch 227/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0708 - mean_absolute_error: 0.1553 - val_loss: 0.1454 - val_mean_absolute_error: 0.1974\n",
      "Epoch 228/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0713 - mean_absolute_error: 0.1566 - val_loss: 0.1419 - val_mean_absolute_error: 0.1941\n",
      "Epoch 229/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0723 - mean_absolute_error: 0.1575 - val_loss: 0.1423 - val_mean_absolute_error: 0.1957\n",
      "Epoch 230/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0685 - mean_absolute_error: 0.1533 - val_loss: 0.1478 - val_mean_absolute_error: 0.1946\n",
      "Epoch 231/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0705 - mean_absolute_error: 0.1556 - val_loss: 0.1564 - val_mean_absolute_error: 0.2000\n",
      "Epoch 232/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0746 - mean_absolute_error: 0.1582 - val_loss: 0.1456 - val_mean_absolute_error: 0.1954\n",
      "Epoch 233/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0724 - mean_absolute_error: 0.1576 - val_loss: 0.1461 - val_mean_absolute_error: 0.1958\n",
      "Epoch 234/300\n",
      "12967/12967 [==============================] - ETA: 0s - loss: 0.0708 - mean_absolute_error: 0.155 - 0s 34us/sample - loss: 0.0707 - mean_absolute_error: 0.1554 - val_loss: 0.1415 - val_mean_absolute_error: 0.1953\n",
      "Epoch 235/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0690 - mean_absolute_error: 0.1550 - val_loss: 0.1467 - val_mean_absolute_error: 0.2006\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 236/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0701 - mean_absolute_error: 0.1557 - val_loss: 0.1486 - val_mean_absolute_error: 0.1959\n",
      "Epoch 237/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0698 - mean_absolute_error: 0.1544 - val_loss: 0.1522 - val_mean_absolute_error: 0.2017\n",
      "Epoch 238/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0770 - mean_absolute_error: 0.1614 - val_loss: 0.1759 - val_mean_absolute_error: 0.2119\n",
      "Epoch 239/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0821 - mean_absolute_error: 0.1658 - val_loss: 0.1489 - val_mean_absolute_error: 0.2000\n",
      "Epoch 240/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0732 - mean_absolute_error: 0.1580 - val_loss: 0.1435 - val_mean_absolute_error: 0.1927\n",
      "Epoch 241/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0705 - mean_absolute_error: 0.1553 - val_loss: 0.1428 - val_mean_absolute_error: 0.1946\n",
      "Epoch 242/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0692 - mean_absolute_error: 0.1551 - val_loss: 0.1429 - val_mean_absolute_error: 0.1920\n",
      "Epoch 243/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0689 - mean_absolute_error: 0.1545 - val_loss: 0.1580 - val_mean_absolute_error: 0.2001\n",
      "Epoch 244/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0735 - mean_absolute_error: 0.1577 - val_loss: 0.1505 - val_mean_absolute_error: 0.1958\n",
      "Epoch 245/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0704 - mean_absolute_error: 0.1564 - val_loss: 0.1454 - val_mean_absolute_error: 0.1936\n",
      "Epoch 246/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0698 - mean_absolute_error: 0.1553 - val_loss: 0.1486 - val_mean_absolute_error: 0.1988\n",
      "Epoch 247/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0698 - mean_absolute_error: 0.1552 - val_loss: 0.1578 - val_mean_absolute_error: 0.1938\n",
      "Epoch 248/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0687 - mean_absolute_error: 0.1553 - val_loss: 0.1464 - val_mean_absolute_error: 0.1960\n",
      "Epoch 249/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0799 - mean_absolute_error: 0.1632 - val_loss: 0.1767 - val_mean_absolute_error: 0.2114\n",
      "Epoch 250/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0749 - mean_absolute_error: 0.1592 - val_loss: 0.1475 - val_mean_absolute_error: 0.1996\n",
      "Epoch 251/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0687 - mean_absolute_error: 0.1546 - val_loss: 0.1496 - val_mean_absolute_error: 0.1972\n",
      "Epoch 252/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0704 - mean_absolute_error: 0.1556 - val_loss: 0.1566 - val_mean_absolute_error: 0.2042\n",
      "Epoch 253/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0741 - mean_absolute_error: 0.1592 - val_loss: 0.1539 - val_mean_absolute_error: 0.2029\n",
      "Epoch 254/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0737 - mean_absolute_error: 0.1578 - val_loss: 0.1591 - val_mean_absolute_error: 0.1981\n",
      "Epoch 255/300\n",
      "12967/12967 [==============================] - 0s 38us/sample - loss: 0.0679 - mean_absolute_error: 0.1531 - val_loss: 0.1485 - val_mean_absolute_error: 0.1985\n",
      "Epoch 256/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0697 - mean_absolute_error: 0.1552 - val_loss: 0.1476 - val_mean_absolute_error: 0.1937\n",
      "Epoch 257/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0719 - mean_absolute_error: 0.1556 - val_loss: 0.1598 - val_mean_absolute_error: 0.1964\n",
      "Epoch 258/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0713 - mean_absolute_error: 0.1564 - val_loss: 0.1401 - val_mean_absolute_error: 0.1938\n",
      "Epoch 259/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0698 - mean_absolute_error: 0.1561 - val_loss: 0.1463 - val_mean_absolute_error: 0.1967\n",
      "Epoch 260/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0694 - mean_absolute_error: 0.1554 - val_loss: 0.1479 - val_mean_absolute_error: 0.1944\n",
      "Epoch 261/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0701 - mean_absolute_error: 0.1560 - val_loss: 0.1461 - val_mean_absolute_error: 0.1977\n",
      "Epoch 262/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0741 - mean_absolute_error: 0.1590 - val_loss: 0.1524 - val_mean_absolute_error: 0.1936\n",
      "Epoch 263/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0793 - mean_absolute_error: 0.1623 - val_loss: 0.1499 - val_mean_absolute_error: 0.1955\n",
      "Epoch 264/300\n",
      "12967/12967 [==============================] - 0s 36us/sample - loss: 0.0700 - mean_absolute_error: 0.1550 - val_loss: 0.1426 - val_mean_absolute_error: 0.1964\n",
      "Epoch 265/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0682 - mean_absolute_error: 0.1538 - val_loss: 0.1404 - val_mean_absolute_error: 0.1934\n",
      "Epoch 266/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0674 - mean_absolute_error: 0.1532 - val_loss: 0.1413 - val_mean_absolute_error: 0.1955\n",
      "Epoch 267/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0674 - mean_absolute_error: 0.1527 - val_loss: 0.1500 - val_mean_absolute_error: 0.1977\n",
      "Epoch 268/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0712 - mean_absolute_error: 0.1568 - val_loss: 0.1407 - val_mean_absolute_error: 0.1927\n",
      "Epoch 269/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0705 - mean_absolute_error: 0.1553 - val_loss: 0.1497 - val_mean_absolute_error: 0.1968\n",
      "Epoch 270/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0692 - mean_absolute_error: 0.1543 - val_loss: 0.1467 - val_mean_absolute_error: 0.1964\n",
      "Epoch 271/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0716 - mean_absolute_error: 0.1566 - val_loss: 0.1613 - val_mean_absolute_error: 0.1960\n",
      "Epoch 272/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.0729 - mean_absolute_error: 0.1573 - val_loss: 0.1648 - val_mean_absolute_error: 0.2074\n",
      "Epoch 273/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0724 - mean_absolute_error: 0.1585 - val_loss: 0.1436 - val_mean_absolute_error: 0.1949\n",
      "Epoch 274/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0682 - mean_absolute_error: 0.1538 - val_loss: 0.1460 - val_mean_absolute_error: 0.1956\n",
      "Epoch 275/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0691 - mean_absolute_error: 0.1540 - val_loss: 0.1451 - val_mean_absolute_error: 0.1930\n",
      "Epoch 276/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0736 - mean_absolute_error: 0.1574 - val_loss: 0.1499 - val_mean_absolute_error: 0.1952\n",
      "Epoch 277/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0701 - mean_absolute_error: 0.1541 - val_loss: 0.1478 - val_mean_absolute_error: 0.1956\n",
      "Epoch 278/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0694 - mean_absolute_error: 0.1552 - val_loss: 0.1462 - val_mean_absolute_error: 0.1958\n",
      "Epoch 279/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0691 - mean_absolute_error: 0.1539 - val_loss: 0.1411 - val_mean_absolute_error: 0.1932\n",
      "Epoch 280/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0703 - mean_absolute_error: 0.1555 - val_loss: 0.1427 - val_mean_absolute_error: 0.1978\n",
      "Epoch 281/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0752 - mean_absolute_error: 0.1595 - val_loss: 0.1540 - val_mean_absolute_error: 0.2006\n",
      "Epoch 282/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0704 - mean_absolute_error: 0.1554 - val_loss: 0.1418 - val_mean_absolute_error: 0.1912\n",
      "Epoch 283/300\n",
      "12967/12967 [==============================] - ETA: 0s - loss: 0.0675 - mean_absolute_error: 0.153 - 0s 34us/sample - loss: 0.0677 - mean_absolute_error: 0.1541 - val_loss: 0.1463 - val_mean_absolute_error: 0.1928\n",
      "Epoch 284/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0707 - mean_absolute_error: 0.1562 - val_loss: 0.1555 - val_mean_absolute_error: 0.2001\n",
      "Epoch 285/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0775 - mean_absolute_error: 0.1633 - val_loss: 0.1456 - val_mean_absolute_error: 0.1972\n",
      "Epoch 286/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0765 - mean_absolute_error: 0.1603 - val_loss: 0.1439 - val_mean_absolute_error: 0.1945\n",
      "Epoch 287/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0695 - mean_absolute_error: 0.1550 - val_loss: 0.1418 - val_mean_absolute_error: 0.1931\n",
      "Epoch 288/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0681 - mean_absolute_error: 0.1534 - val_loss: 0.1523 - val_mean_absolute_error: 0.1983\n",
      "Epoch 289/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0697 - mean_absolute_error: 0.1548 - val_loss: 0.1602 - val_mean_absolute_error: 0.2020\n",
      "Epoch 290/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0700 - mean_absolute_error: 0.1556 - val_loss: 0.1477 - val_mean_absolute_error: 0.1937\n",
      "Epoch 291/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0680 - mean_absolute_error: 0.1531 - val_loss: 0.1467 - val_mean_absolute_error: 0.1965\n",
      "Epoch 292/300\n",
      "12967/12967 [==============================] - 0s 35us/sample - loss: 0.0672 - mean_absolute_error: 0.1537 - val_loss: 0.1456 - val_mean_absolute_error: 0.1951\n",
      "Epoch 293/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0675 - mean_absolute_error: 0.1531 - val_loss: 0.1510 - val_mean_absolute_error: 0.2000\n",
      "Epoch 294/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0741 - mean_absolute_error: 0.1584 - val_loss: 0.1423 - val_mean_absolute_error: 0.1980\n",
      "Epoch 295/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0708 - mean_absolute_error: 0.1554 - val_loss: 0.1470 - val_mean_absolute_error: 0.1968\n",
      "Epoch 296/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0669 - mean_absolute_error: 0.1527 - val_loss: 0.1527 - val_mean_absolute_error: 0.1988\n",
      "Epoch 297/300\n",
      "12967/12967 [==============================] - 0s 33us/sample - loss: 0.0750 - mean_absolute_error: 0.1593 - val_loss: 0.1429 - val_mean_absolute_error: 0.1937\n",
      "Epoch 298/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0695 - mean_absolute_error: 0.1547 - val_loss: 0.1448 - val_mean_absolute_error: 0.1964\n",
      "Epoch 299/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0681 - mean_absolute_error: 0.1539 - val_loss: 0.1435 - val_mean_absolute_error: 0.1939\n",
      "Epoch 300/300\n",
      "12967/12967 [==============================] - 0s 34us/sample - loss: 0.0703 - mean_absolute_error: 0.1563 - val_loss: 0.1414 - val_mean_absolute_error: 0.1985\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x25d27387208>"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_3 = keras.Sequential(name='model-3')\n",
    "model_3.add(layers.Dense(64, \n",
    "                         kernel_regularizer=keras.regularizers.l2(0.001), \n",
    "                         activation='relu', input_shape=(21,)))\n",
    "model_3.add(layers.Dense(64, kernel_regularizer=keras.regularizers.l2(0.001), activation='relu'))\n",
    "model_3.add(layers.Dense(1))\n",
    "\n",
    "model_3.compile(keras.optimizers.Adam(0.001),\n",
    "                loss=keras.losses.MeanSquaredError(),\n",
    "                metrics=[keras.metrics.MeanAbsoluteError()])\n",
    "\n",
    "log_dir = os.path.join('lab2-logs', 'model-3')\n",
    "model_cbk = keras.callbacks.TensorBoard(log_dir=log_dir)\n",
    "model_mckp = keras.callbacks.ModelCheckpoint(model_dir + '/Best-model-3.h5', \n",
    "                                             monitor='val_mean_absolute_error', \n",
    "                                             save_best_only=True, \n",
    "                                             mode='min')\n",
    "model_3.fit(x_train, y_train, \n",
    "            batch_size=64 ,\n",
    "            epochs=300, \n",
    "            validation_data=(x_val, y_val), \n",
    "            callbacks=[model_cbk, model_mckp])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 加入 Dropout"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 12967 samples, validate on 4323 samples\n",
      "Epoch 1/300\n",
      "12967/12967 [==============================] - 1s 103us/sample - loss: 0.4961 - mean_absolute_error: 0.4525 - val_loss: 0.2910 - val_mean_absolute_error: 0.3177\n",
      "Epoch 2/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.3170 - mean_absolute_error: 0.3530 - val_loss: 0.2722 - val_mean_absolute_error: 0.3123\n",
      "Epoch 3/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.2827 - mean_absolute_error: 0.3307 - val_loss: 0.2299 - val_mean_absolute_error: 0.2939\n",
      "Epoch 4/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.2666 - mean_absolute_error: 0.3152 - val_loss: 0.2148 - val_mean_absolute_error: 0.2880\n",
      "Epoch 5/300\n",
      "12967/12967 [==============================] - 1s 50us/sample - loss: 0.2430 - mean_absolute_error: 0.3046 - val_loss: 0.2287 - val_mean_absolute_error: 0.2827\n",
      "Epoch 6/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.2124 - mean_absolute_error: 0.2927 - val_loss: 0.2108 - val_mean_absolute_error: 0.2844\n",
      "Epoch 7/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.2062 - mean_absolute_error: 0.2858 - val_loss: 0.2031 - val_mean_absolute_error: 0.2669\n",
      "Epoch 8/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.2023 - mean_absolute_error: 0.2813 - val_loss: 0.2194 - val_mean_absolute_error: 0.2816\n",
      "Epoch 9/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1956 - mean_absolute_error: 0.2763 - val_loss: 0.2009 - val_mean_absolute_error: 0.2667\n",
      "Epoch 10/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.2059 - mean_absolute_error: 0.2769 - val_loss: 0.1968 - val_mean_absolute_error: 0.2722\n",
      "Epoch 11/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1909 - mean_absolute_error: 0.2698 - val_loss: 0.1978 - val_mean_absolute_error: 0.2641\n",
      "Epoch 12/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1904 - mean_absolute_error: 0.2715 - val_loss: 0.2058 - val_mean_absolute_error: 0.2604\n",
      "Epoch 13/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1873 - mean_absolute_error: 0.2674 - val_loss: 0.2014 - val_mean_absolute_error: 0.2557\n",
      "Epoch 14/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1845 - mean_absolute_error: 0.2671 - val_loss: 0.1911 - val_mean_absolute_error: 0.2685\n",
      "Epoch 15/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1743 - mean_absolute_error: 0.2631 - val_loss: 0.2177 - val_mean_absolute_error: 0.2665\n",
      "Epoch 16/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1677 - mean_absolute_error: 0.2612 - val_loss: 0.1983 - val_mean_absolute_error: 0.2554\n",
      "Epoch 17/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1717 - mean_absolute_error: 0.2583 - val_loss: 0.2232 - val_mean_absolute_error: 0.2635\n",
      "Epoch 18/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1711 - mean_absolute_error: 0.2573 - val_loss: 0.1992 - val_mean_absolute_error: 0.2563\n",
      "Epoch 19/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1730 - mean_absolute_error: 0.2563 - val_loss: 0.1708 - val_mean_absolute_error: 0.2455\n",
      "Epoch 20/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1588 - mean_absolute_error: 0.2512 - val_loss: 0.1808 - val_mean_absolute_error: 0.2461\n",
      "Epoch 21/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1728 - mean_absolute_error: 0.2549 - val_loss: 0.1657 - val_mean_absolute_error: 0.2433\n",
      "Epoch 22/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1627 - mean_absolute_error: 0.2512 - val_loss: 0.2167 - val_mean_absolute_error: 0.2590\n",
      "Epoch 23/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1582 - mean_absolute_error: 0.2490 - val_loss: 0.1852 - val_mean_absolute_error: 0.2525\n",
      "Epoch 24/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1554 - mean_absolute_error: 0.2471 - val_loss: 0.1837 - val_mean_absolute_error: 0.2415\n",
      "Epoch 25/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1577 - mean_absolute_error: 0.2458 - val_loss: 0.1685 - val_mean_absolute_error: 0.2412\n",
      "Epoch 26/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1539 - mean_absolute_error: 0.2457 - val_loss: 0.1748 - val_mean_absolute_error: 0.2329\n",
      "Epoch 27/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1508 - mean_absolute_error: 0.2434 - val_loss: 0.2143 - val_mean_absolute_error: 0.2467\n",
      "Epoch 28/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1517 - mean_absolute_error: 0.2446 - val_loss: 0.1444 - val_mean_absolute_error: 0.2225\n",
      "Epoch 29/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1521 - mean_absolute_error: 0.2438 - val_loss: 0.1969 - val_mean_absolute_error: 0.2465\n",
      "Epoch 30/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1507 - mean_absolute_error: 0.2432 - val_loss: 0.1420 - val_mean_absolute_error: 0.2198\n",
      "Epoch 31/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1475 - mean_absolute_error: 0.2407 - val_loss: 0.1676 - val_mean_absolute_error: 0.2233\n",
      "Epoch 32/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1470 - mean_absolute_error: 0.2405 - val_loss: 0.1634 - val_mean_absolute_error: 0.2310\n",
      "Epoch 33/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1432 - mean_absolute_error: 0.2398 - val_loss: 0.2594 - val_mean_absolute_error: 0.2651\n",
      "Epoch 34/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1508 - mean_absolute_error: 0.2409 - val_loss: 0.2080 - val_mean_absolute_error: 0.2472\n",
      "Epoch 35/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1450 - mean_absolute_error: 0.2394 - val_loss: 0.1769 - val_mean_absolute_error: 0.2437\n",
      "Epoch 36/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1440 - mean_absolute_error: 0.2399 - val_loss: 0.2053 - val_mean_absolute_error: 0.2428\n",
      "Epoch 37/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1469 - mean_absolute_error: 0.2383 - val_loss: 0.1916 - val_mean_absolute_error: 0.2375\n",
      "Epoch 38/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1445 - mean_absolute_error: 0.2380 - val_loss: 0.1701 - val_mean_absolute_error: 0.2339\n",
      "Epoch 39/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1335 - mean_absolute_error: 0.2341 - val_loss: 0.1502 - val_mean_absolute_error: 0.2222\n",
      "Epoch 40/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1571 - mean_absolute_error: 0.2402 - val_loss: 0.1710 - val_mean_absolute_error: 0.2307\n",
      "Epoch 41/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1335 - mean_absolute_error: 0.2341 - val_loss: 0.1734 - val_mean_absolute_error: 0.2281\n",
      "Epoch 42/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1391 - mean_absolute_error: 0.2337 - val_loss: 0.1604 - val_mean_absolute_error: 0.2295\n",
      "Epoch 43/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1374 - mean_absolute_error: 0.2329 - val_loss: 0.1877 - val_mean_absolute_error: 0.2304\n",
      "Epoch 44/300\n",
      "12967/12967 [==============================] - 1s 57us/sample - loss: 0.1351 - mean_absolute_error: 0.2326 - val_loss: 0.1618 - val_mean_absolute_error: 0.2253\n",
      "Epoch 45/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1332 - mean_absolute_error: 0.2342 - val_loss: 0.1587 - val_mean_absolute_error: 0.2299\n",
      "Epoch 46/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1343 - mean_absolute_error: 0.2322 - val_loss: 0.1544 - val_mean_absolute_error: 0.2335\n",
      "Epoch 47/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1410 - mean_absolute_error: 0.2324 - val_loss: 0.2048 - val_mean_absolute_error: 0.2356\n",
      "Epoch 48/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1447 - mean_absolute_error: 0.2350 - val_loss: 0.1581 - val_mean_absolute_error: 0.2251\n",
      "Epoch 49/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1373 - mean_absolute_error: 0.2344 - val_loss: 0.1611 - val_mean_absolute_error: 0.2282\n",
      "Epoch 50/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1440 - mean_absolute_error: 0.2329 - val_loss: 0.1759 - val_mean_absolute_error: 0.2306\n",
      "Epoch 51/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1322 - mean_absolute_error: 0.2322 - val_loss: 0.1667 - val_mean_absolute_error: 0.2324\n",
      "Epoch 52/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1255 - mean_absolute_error: 0.2269 - val_loss: 0.1525 - val_mean_absolute_error: 0.2254\n",
      "Epoch 53/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1340 - mean_absolute_error: 0.2306 - val_loss: 0.1884 - val_mean_absolute_error: 0.2280\n",
      "Epoch 54/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1276 - mean_absolute_error: 0.2284 - val_loss: 0.1572 - val_mean_absolute_error: 0.2306\n",
      "Epoch 55/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1303 - mean_absolute_error: 0.2297 - val_loss: 0.1533 - val_mean_absolute_error: 0.2289\n",
      "Epoch 56/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1244 - mean_absolute_error: 0.2273 - val_loss: 0.1900 - val_mean_absolute_error: 0.2410\n",
      "Epoch 57/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1264 - mean_absolute_error: 0.2274 - val_loss: 0.1506 - val_mean_absolute_error: 0.2250\n",
      "Epoch 58/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1269 - mean_absolute_error: 0.2280 - val_loss: 0.1557 - val_mean_absolute_error: 0.2211\n",
      "Epoch 59/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1256 - mean_absolute_error: 0.2286 - val_loss: 0.1458 - val_mean_absolute_error: 0.2294\n",
      "Epoch 60/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1371 - mean_absolute_error: 0.2296 - val_loss: 0.1356 - val_mean_absolute_error: 0.2182\n",
      "Epoch 61/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1339 - mean_absolute_error: 0.2276 - val_loss: 0.1409 - val_mean_absolute_error: 0.2193\n",
      "Epoch 62/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1308 - mean_absolute_error: 0.2302 - val_loss: 0.1548 - val_mean_absolute_error: 0.2265\n",
      "Epoch 63/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1232 - mean_absolute_error: 0.2274 - val_loss: 0.1370 - val_mean_absolute_error: 0.2187\n",
      "Epoch 64/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1281 - mean_absolute_error: 0.2279 - val_loss: 0.1731 - val_mean_absolute_error: 0.2296\n",
      "Epoch 65/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1231 - mean_absolute_error: 0.2266 - val_loss: 0.1355 - val_mean_absolute_error: 0.2250\n",
      "Epoch 66/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1225 - mean_absolute_error: 0.2252 - val_loss: 0.1584 - val_mean_absolute_error: 0.2246\n",
      "Epoch 67/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1158 - mean_absolute_error: 0.2241 - val_loss: 0.1649 - val_mean_absolute_error: 0.2347\n",
      "Epoch 68/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1283 - mean_absolute_error: 0.2258 - val_loss: 0.1578 - val_mean_absolute_error: 0.2301\n",
      "Epoch 69/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1276 - mean_absolute_error: 0.2253 - val_loss: 0.1328 - val_mean_absolute_error: 0.2145\n",
      "Epoch 70/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1199 - mean_absolute_error: 0.2214 - val_loss: 0.1557 - val_mean_absolute_error: 0.2263\n",
      "Epoch 71/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1367 - mean_absolute_error: 0.2278 - val_loss: 0.1418 - val_mean_absolute_error: 0.2244\n",
      "Epoch 72/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1222 - mean_absolute_error: 0.2232 - val_loss: 0.1545 - val_mean_absolute_error: 0.2219\n",
      "Epoch 73/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1255 - mean_absolute_error: 0.2254 - val_loss: 0.1320 - val_mean_absolute_error: 0.2157\n",
      "Epoch 74/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1302 - mean_absolute_error: 0.2263 - val_loss: 0.1513 - val_mean_absolute_error: 0.2226\n",
      "Epoch 75/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1235 - mean_absolute_error: 0.2251 - val_loss: 0.1397 - val_mean_absolute_error: 0.2269\n",
      "Epoch 76/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1223 - mean_absolute_error: 0.2250 - val_loss: 0.1365 - val_mean_absolute_error: 0.2226\n",
      "Epoch 77/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1158 - mean_absolute_error: 0.2219 - val_loss: 0.1812 - val_mean_absolute_error: 0.2328\n",
      "Epoch 78/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1164 - mean_absolute_error: 0.2219 - val_loss: 0.1299 - val_mean_absolute_error: 0.2109\n",
      "Epoch 79/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1304 - mean_absolute_error: 0.2235 - val_loss: 0.1423 - val_mean_absolute_error: 0.2191\n",
      "Epoch 80/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1272 - mean_absolute_error: 0.2245 - val_loss: 0.1620 - val_mean_absolute_error: 0.2170\n",
      "Epoch 81/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1162 - mean_absolute_error: 0.2225 - val_loss: 0.1442 - val_mean_absolute_error: 0.2184\n",
      "Epoch 82/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1150 - mean_absolute_error: 0.2201 - val_loss: 0.1653 - val_mean_absolute_error: 0.2181\n",
      "Epoch 83/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1194 - mean_absolute_error: 0.2204 - val_loss: 0.1767 - val_mean_absolute_error: 0.2298\n",
      "Epoch 84/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1162 - mean_absolute_error: 0.2212 - val_loss: 0.1370 - val_mean_absolute_error: 0.2206\n",
      "Epoch 85/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1173 - mean_absolute_error: 0.2194 - val_loss: 0.1616 - val_mean_absolute_error: 0.2288\n",
      "Epoch 86/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1246 - mean_absolute_error: 0.2234 - val_loss: 0.1349 - val_mean_absolute_error: 0.2183\n",
      "Epoch 87/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1192 - mean_absolute_error: 0.2202 - val_loss: 0.1625 - val_mean_absolute_error: 0.2223\n",
      "Epoch 88/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1172 - mean_absolute_error: 0.2205 - val_loss: 0.1357 - val_mean_absolute_error: 0.2136\n",
      "Epoch 89/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1293 - mean_absolute_error: 0.2214 - val_loss: 0.1303 - val_mean_absolute_error: 0.2174\n",
      "Epoch 90/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1188 - mean_absolute_error: 0.2204 - val_loss: 0.1613 - val_mean_absolute_error: 0.2304\n",
      "Epoch 91/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1239 - mean_absolute_error: 0.2211 - val_loss: 0.1480 - val_mean_absolute_error: 0.2175\n",
      "Epoch 92/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1162 - mean_absolute_error: 0.2195 - val_loss: 0.1397 - val_mean_absolute_error: 0.2227\n",
      "Epoch 93/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1134 - mean_absolute_error: 0.2173 - val_loss: 0.1527 - val_mean_absolute_error: 0.2347\n",
      "Epoch 94/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1247 - mean_absolute_error: 0.2236 - val_loss: 0.1366 - val_mean_absolute_error: 0.2228\n",
      "Epoch 95/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1208 - mean_absolute_error: 0.2229 - val_loss: 0.1472 - val_mean_absolute_error: 0.2237\n",
      "Epoch 96/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1155 - mean_absolute_error: 0.2195 - val_loss: 0.1460 - val_mean_absolute_error: 0.2332\n",
      "Epoch 97/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1195 - mean_absolute_error: 0.2206 - val_loss: 0.1409 - val_mean_absolute_error: 0.2149\n",
      "Epoch 98/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1213 - mean_absolute_error: 0.2203 - val_loss: 0.1435 - val_mean_absolute_error: 0.2213\n",
      "Epoch 99/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1109 - mean_absolute_error: 0.2176 - val_loss: 0.1467 - val_mean_absolute_error: 0.2314\n",
      "Epoch 100/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1094 - mean_absolute_error: 0.2153 - val_loss: 0.1315 - val_mean_absolute_error: 0.2182\n",
      "Epoch 101/300\n",
      "12967/12967 [==============================] - 1s 65us/sample - loss: 0.1206 - mean_absolute_error: 0.2188 - val_loss: 0.1515 - val_mean_absolute_error: 0.2186\n",
      "Epoch 102/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1202 - mean_absolute_error: 0.2202 - val_loss: 0.1529 - val_mean_absolute_error: 0.2193\n",
      "Epoch 103/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1115 - mean_absolute_error: 0.2177 - val_loss: 0.1460 - val_mean_absolute_error: 0.2369\n",
      "Epoch 104/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1177 - mean_absolute_error: 0.2199 - val_loss: 0.1632 - val_mean_absolute_error: 0.2288\n",
      "Epoch 105/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1159 - mean_absolute_error: 0.2176 - val_loss: 0.1329 - val_mean_absolute_error: 0.2152\n",
      "Epoch 106/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1231 - mean_absolute_error: 0.2190 - val_loss: 0.1362 - val_mean_absolute_error: 0.2237\n",
      "Epoch 107/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1152 - mean_absolute_error: 0.2181 - val_loss: 0.1385 - val_mean_absolute_error: 0.2265\n",
      "Epoch 108/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1148 - mean_absolute_error: 0.2187 - val_loss: 0.1394 - val_mean_absolute_error: 0.2253\n",
      "Epoch 109/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1148 - mean_absolute_error: 0.2190 - val_loss: 0.1460 - val_mean_absolute_error: 0.2176\n",
      "Epoch 110/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1107 - mean_absolute_error: 0.2157 - val_loss: 0.1315 - val_mean_absolute_error: 0.2209\n",
      "Epoch 111/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1214 - mean_absolute_error: 0.2211 - val_loss: 0.1313 - val_mean_absolute_error: 0.2221\n",
      "Epoch 112/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1130 - mean_absolute_error: 0.2172 - val_loss: 0.1721 - val_mean_absolute_error: 0.2270\n",
      "Epoch 113/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1108 - mean_absolute_error: 0.2169 - val_loss: 0.1421 - val_mean_absolute_error: 0.2275\n",
      "Epoch 114/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1143 - mean_absolute_error: 0.2173 - val_loss: 0.1389 - val_mean_absolute_error: 0.2097\n",
      "Epoch 115/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1217 - mean_absolute_error: 0.2209 - val_loss: 0.1334 - val_mean_absolute_error: 0.2170\n",
      "Epoch 116/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1126 - mean_absolute_error: 0.2181 - val_loss: 0.1359 - val_mean_absolute_error: 0.2248\n",
      "Epoch 117/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1182 - mean_absolute_error: 0.2177 - val_loss: 0.1586 - val_mean_absolute_error: 0.2330\n",
      "Epoch 118/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1115 - mean_absolute_error: 0.2169 - val_loss: 0.1524 - val_mean_absolute_error: 0.2254\n",
      "Epoch 119/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1108 - mean_absolute_error: 0.2161 - val_loss: 0.1425 - val_mean_absolute_error: 0.2208\n",
      "Epoch 120/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1191 - mean_absolute_error: 0.2170 - val_loss: 0.1718 - val_mean_absolute_error: 0.2274\n",
      "Epoch 121/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1150 - mean_absolute_error: 0.2178 - val_loss: 0.1388 - val_mean_absolute_error: 0.2224\n",
      "Epoch 122/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1143 - mean_absolute_error: 0.2180 - val_loss: 0.1352 - val_mean_absolute_error: 0.2332\n",
      "Epoch 123/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1068 - mean_absolute_error: 0.2130 - val_loss: 0.1419 - val_mean_absolute_error: 0.2190\n",
      "Epoch 124/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1110 - mean_absolute_error: 0.2149 - val_loss: 0.1329 - val_mean_absolute_error: 0.2207\n",
      "Epoch 125/300\n",
      "12967/12967 [==============================] - 1s 49us/sample - loss: 0.1159 - mean_absolute_error: 0.2155 - val_loss: 0.1342 - val_mean_absolute_error: 0.2183\n",
      "Epoch 126/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1179 - mean_absolute_error: 0.2188 - val_loss: 0.1640 - val_mean_absolute_error: 0.2290\n",
      "Epoch 127/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1074 - mean_absolute_error: 0.2150 - val_loss: 0.1401 - val_mean_absolute_error: 0.2186\n",
      "Epoch 128/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1171 - mean_absolute_error: 0.2160 - val_loss: 0.1377 - val_mean_absolute_error: 0.2224\n",
      "Epoch 129/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1127 - mean_absolute_error: 0.2158 - val_loss: 0.1322 - val_mean_absolute_error: 0.2147\n",
      "Epoch 130/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1182 - mean_absolute_error: 0.2161 - val_loss: 0.1367 - val_mean_absolute_error: 0.2136\n",
      "Epoch 131/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1156 - mean_absolute_error: 0.2164 - val_loss: 0.1444 - val_mean_absolute_error: 0.2211\n",
      "Epoch 132/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1142 - mean_absolute_error: 0.2175 - val_loss: 0.1468 - val_mean_absolute_error: 0.2222\n",
      "Epoch 133/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1128 - mean_absolute_error: 0.2172 - val_loss: 0.1478 - val_mean_absolute_error: 0.2194\n",
      "Epoch 134/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1093 - mean_absolute_error: 0.2155 - val_loss: 0.1466 - val_mean_absolute_error: 0.2298\n",
      "Epoch 135/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1105 - mean_absolute_error: 0.2180 - val_loss: 0.1353 - val_mean_absolute_error: 0.2234\n",
      "Epoch 136/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1095 - mean_absolute_error: 0.2136 - val_loss: 0.1274 - val_mean_absolute_error: 0.2090\n",
      "Epoch 137/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1118 - mean_absolute_error: 0.2174 - val_loss: 0.1335 - val_mean_absolute_error: 0.2150\n",
      "Epoch 138/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1058 - mean_absolute_error: 0.2135 - val_loss: 0.1435 - val_mean_absolute_error: 0.2199\n",
      "Epoch 139/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1148 - mean_absolute_error: 0.2155 - val_loss: 0.1382 - val_mean_absolute_error: 0.2293\n",
      "Epoch 140/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1118 - mean_absolute_error: 0.2142 - val_loss: 0.1265 - val_mean_absolute_error: 0.2101\n",
      "Epoch 141/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1125 - mean_absolute_error: 0.2158 - val_loss: 0.1399 - val_mean_absolute_error: 0.2193\n",
      "Epoch 142/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1220 - mean_absolute_error: 0.2188 - val_loss: 0.1510 - val_mean_absolute_error: 0.2240\n",
      "Epoch 143/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1222 - mean_absolute_error: 0.2159 - val_loss: 0.1351 - val_mean_absolute_error: 0.2180\n",
      "Epoch 144/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1161 - mean_absolute_error: 0.2159 - val_loss: 0.1327 - val_mean_absolute_error: 0.2189\n",
      "Epoch 145/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1107 - mean_absolute_error: 0.2153 - val_loss: 0.1309 - val_mean_absolute_error: 0.2079\n",
      "Epoch 146/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1072 - mean_absolute_error: 0.2147 - val_loss: 0.1369 - val_mean_absolute_error: 0.2258\n",
      "Epoch 147/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1084 - mean_absolute_error: 0.2131 - val_loss: 0.1256 - val_mean_absolute_error: 0.2225\n",
      "Epoch 148/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1103 - mean_absolute_error: 0.2149 - val_loss: 0.1425 - val_mean_absolute_error: 0.2248\n",
      "Epoch 149/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1116 - mean_absolute_error: 0.2139 - val_loss: 0.1315 - val_mean_absolute_error: 0.2190\n",
      "Epoch 150/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1085 - mean_absolute_error: 0.2132 - val_loss: 0.1246 - val_mean_absolute_error: 0.2142\n",
      "Epoch 151/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1136 - mean_absolute_error: 0.2152 - val_loss: 0.1275 - val_mean_absolute_error: 0.2132\n",
      "Epoch 152/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1114 - mean_absolute_error: 0.2176 - val_loss: 0.1356 - val_mean_absolute_error: 0.2133\n",
      "Epoch 153/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1085 - mean_absolute_error: 0.2139 - val_loss: 0.1317 - val_mean_absolute_error: 0.2144\n",
      "Epoch 154/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1038 - mean_absolute_error: 0.2117 - val_loss: 0.1649 - val_mean_absolute_error: 0.2227\n",
      "Epoch 155/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1042 - mean_absolute_error: 0.2130 - val_loss: 0.1463 - val_mean_absolute_error: 0.2220\n",
      "Epoch 156/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1212 - mean_absolute_error: 0.2166 - val_loss: 0.1462 - val_mean_absolute_error: 0.2257\n",
      "Epoch 157/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1043 - mean_absolute_error: 0.2110 - val_loss: 0.1269 - val_mean_absolute_error: 0.2084\n",
      "Epoch 158/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1089 - mean_absolute_error: 0.2153 - val_loss: 0.1453 - val_mean_absolute_error: 0.2173\n",
      "Epoch 159/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1086 - mean_absolute_error: 0.2135 - val_loss: 0.1414 - val_mean_absolute_error: 0.2212\n",
      "Epoch 160/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1086 - mean_absolute_error: 0.2133 - val_loss: 0.1599 - val_mean_absolute_error: 0.2255\n",
      "Epoch 161/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1120 - mean_absolute_error: 0.2154 - val_loss: 0.1347 - val_mean_absolute_error: 0.2228\n",
      "Epoch 162/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1185 - mean_absolute_error: 0.2171 - val_loss: 0.1445 - val_mean_absolute_error: 0.2297\n",
      "Epoch 163/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1143 - mean_absolute_error: 0.2138 - val_loss: 0.1292 - val_mean_absolute_error: 0.2184\n",
      "Epoch 164/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1059 - mean_absolute_error: 0.2124 - val_loss: 0.1444 - val_mean_absolute_error: 0.2251\n",
      "Epoch 165/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1041 - mean_absolute_error: 0.2104 - val_loss: 0.1313 - val_mean_absolute_error: 0.2151\n",
      "Epoch 166/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1129 - mean_absolute_error: 0.2149 - val_loss: 0.1323 - val_mean_absolute_error: 0.2192\n",
      "Epoch 167/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1063 - mean_absolute_error: 0.2117 - val_loss: 0.1324 - val_mean_absolute_error: 0.2284\n",
      "Epoch 168/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1136 - mean_absolute_error: 0.2148 - val_loss: 0.1294 - val_mean_absolute_error: 0.2111\n",
      "Epoch 169/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1085 - mean_absolute_error: 0.2129 - val_loss: 0.1447 - val_mean_absolute_error: 0.2258\n",
      "Epoch 170/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1074 - mean_absolute_error: 0.2105 - val_loss: 0.1307 - val_mean_absolute_error: 0.2141\n",
      "Epoch 171/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1120 - mean_absolute_error: 0.2126 - val_loss: 0.1332 - val_mean_absolute_error: 0.2165\n",
      "Epoch 172/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1087 - mean_absolute_error: 0.2136 - val_loss: 0.1437 - val_mean_absolute_error: 0.2268\n",
      "Epoch 173/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1108 - mean_absolute_error: 0.2127 - val_loss: 0.1567 - val_mean_absolute_error: 0.2259\n",
      "Epoch 174/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1064 - mean_absolute_error: 0.2139 - val_loss: 0.1416 - val_mean_absolute_error: 0.2284\n",
      "Epoch 175/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1074 - mean_absolute_error: 0.2130 - val_loss: 0.1368 - val_mean_absolute_error: 0.2240\n",
      "Epoch 176/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1097 - mean_absolute_error: 0.2120 - val_loss: 0.1539 - val_mean_absolute_error: 0.2229\n",
      "Epoch 177/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1153 - mean_absolute_error: 0.2132 - val_loss: 0.1310 - val_mean_absolute_error: 0.2124\n",
      "Epoch 178/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1166 - mean_absolute_error: 0.2151 - val_loss: 0.1295 - val_mean_absolute_error: 0.2293\n",
      "Epoch 179/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1040 - mean_absolute_error: 0.2138 - val_loss: 0.1391 - val_mean_absolute_error: 0.2285\n",
      "Epoch 180/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1143 - mean_absolute_error: 0.2155 - val_loss: 0.1336 - val_mean_absolute_error: 0.2171\n",
      "Epoch 181/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1125 - mean_absolute_error: 0.2129 - val_loss: 0.1371 - val_mean_absolute_error: 0.2245\n",
      "Epoch 182/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1059 - mean_absolute_error: 0.2142 - val_loss: 0.1350 - val_mean_absolute_error: 0.2224\n",
      "Epoch 183/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1154 - mean_absolute_error: 0.2150 - val_loss: 0.1679 - val_mean_absolute_error: 0.2285\n",
      "Epoch 184/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1120 - mean_absolute_error: 0.2128 - val_loss: 0.1322 - val_mean_absolute_error: 0.2149\n",
      "Epoch 185/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1086 - mean_absolute_error: 0.2133 - val_loss: 0.1475 - val_mean_absolute_error: 0.2276\n",
      "Epoch 186/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1082 - mean_absolute_error: 0.2101 - val_loss: 0.1490 - val_mean_absolute_error: 0.2262\n",
      "Epoch 187/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1014 - mean_absolute_error: 0.2096 - val_loss: 0.1518 - val_mean_absolute_error: 0.2317\n",
      "Epoch 188/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1161 - mean_absolute_error: 0.2133 - val_loss: 0.1389 - val_mean_absolute_error: 0.2285\n",
      "Epoch 189/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1120 - mean_absolute_error: 0.2132 - val_loss: 0.1495 - val_mean_absolute_error: 0.2300\n",
      "Epoch 190/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1090 - mean_absolute_error: 0.2116 - val_loss: 0.1433 - val_mean_absolute_error: 0.2293\n",
      "Epoch 191/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1099 - mean_absolute_error: 0.2114 - val_loss: 0.1564 - val_mean_absolute_error: 0.2363\n",
      "Epoch 192/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1049 - mean_absolute_error: 0.2109 - val_loss: 0.1440 - val_mean_absolute_error: 0.2210\n",
      "Epoch 193/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.0999 - mean_absolute_error: 0.2082 - val_loss: 0.1535 - val_mean_absolute_error: 0.2264\n",
      "Epoch 194/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1018 - mean_absolute_error: 0.2103 - val_loss: 0.1287 - val_mean_absolute_error: 0.2143\n",
      "Epoch 195/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1047 - mean_absolute_error: 0.2102 - val_loss: 0.1347 - val_mean_absolute_error: 0.2234\n",
      "Epoch 196/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1087 - mean_absolute_error: 0.2121 - val_loss: 0.1428 - val_mean_absolute_error: 0.2247\n",
      "Epoch 197/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1092 - mean_absolute_error: 0.2126 - val_loss: 0.1337 - val_mean_absolute_error: 0.2266\n",
      "Epoch 198/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1143 - mean_absolute_error: 0.2129 - val_loss: 0.1397 - val_mean_absolute_error: 0.2160\n",
      "Epoch 199/300\n",
      "12967/12967 [==============================] - 1s 59us/sample - loss: 0.1094 - mean_absolute_error: 0.2107 - val_loss: 0.1277 - val_mean_absolute_error: 0.2158\n",
      "Epoch 200/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1082 - mean_absolute_error: 0.2121 - val_loss: 0.1322 - val_mean_absolute_error: 0.2207\n",
      "Epoch 201/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1107 - mean_absolute_error: 0.2102 - val_loss: 0.1338 - val_mean_absolute_error: 0.2203\n",
      "Epoch 202/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1010 - mean_absolute_error: 0.2093 - val_loss: 0.1325 - val_mean_absolute_error: 0.2120\n",
      "Epoch 203/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1096 - mean_absolute_error: 0.2128 - val_loss: 0.1483 - val_mean_absolute_error: 0.2337\n",
      "Epoch 204/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1079 - mean_absolute_error: 0.2112 - val_loss: 0.1332 - val_mean_absolute_error: 0.2197\n",
      "Epoch 205/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1052 - mean_absolute_error: 0.2107 - val_loss: 0.1405 - val_mean_absolute_error: 0.2135\n",
      "Epoch 206/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1075 - mean_absolute_error: 0.2109 - val_loss: 0.1405 - val_mean_absolute_error: 0.2307\n",
      "Epoch 207/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1024 - mean_absolute_error: 0.2093 - val_loss: 0.1295 - val_mean_absolute_error: 0.2194\n",
      "Epoch 208/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1024 - mean_absolute_error: 0.2115 - val_loss: 0.1343 - val_mean_absolute_error: 0.2225\n",
      "Epoch 209/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1007 - mean_absolute_error: 0.2088 - val_loss: 0.1300 - val_mean_absolute_error: 0.2177\n",
      "Epoch 210/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1042 - mean_absolute_error: 0.2107 - val_loss: 0.1391 - val_mean_absolute_error: 0.2210\n",
      "Epoch 211/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1075 - mean_absolute_error: 0.2107 - val_loss: 0.1252 - val_mean_absolute_error: 0.2161\n",
      "Epoch 212/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1064 - mean_absolute_error: 0.2105 - val_loss: 0.1437 - val_mean_absolute_error: 0.2322\n",
      "Epoch 213/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1096 - mean_absolute_error: 0.2118 - val_loss: 0.1371 - val_mean_absolute_error: 0.2146\n",
      "Epoch 214/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1106 - mean_absolute_error: 0.2114 - val_loss: 0.1363 - val_mean_absolute_error: 0.2115\n",
      "Epoch 215/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1129 - mean_absolute_error: 0.2122 - val_loss: 0.1341 - val_mean_absolute_error: 0.2133\n",
      "Epoch 216/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1115 - mean_absolute_error: 0.2130 - val_loss: 0.1344 - val_mean_absolute_error: 0.2241\n",
      "Epoch 217/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1130 - mean_absolute_error: 0.2128 - val_loss: 0.1314 - val_mean_absolute_error: 0.2239\n",
      "Epoch 218/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1087 - mean_absolute_error: 0.2134 - val_loss: 0.1553 - val_mean_absolute_error: 0.2314\n",
      "Epoch 219/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1052 - mean_absolute_error: 0.2112 - val_loss: 0.1221 - val_mean_absolute_error: 0.2095\n",
      "Epoch 220/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1092 - mean_absolute_error: 0.2104 - val_loss: 0.1387 - val_mean_absolute_error: 0.2228\n",
      "Epoch 221/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1040 - mean_absolute_error: 0.2106 - val_loss: 0.1466 - val_mean_absolute_error: 0.2293\n",
      "Epoch 222/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1095 - mean_absolute_error: 0.2127 - val_loss: 0.1469 - val_mean_absolute_error: 0.2174\n",
      "Epoch 223/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1035 - mean_absolute_error: 0.2098 - val_loss: 0.1376 - val_mean_absolute_error: 0.2272\n",
      "Epoch 224/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1046 - mean_absolute_error: 0.2100 - val_loss: 0.1564 - val_mean_absolute_error: 0.2276\n",
      "Epoch 225/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1023 - mean_absolute_error: 0.2093 - val_loss: 0.1356 - val_mean_absolute_error: 0.2149\n",
      "Epoch 226/300\n",
      "12967/12967 [==============================] - 1s 51us/sample - loss: 0.1084 - mean_absolute_error: 0.2118 - val_loss: 0.1472 - val_mean_absolute_error: 0.2236\n",
      "Epoch 227/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1061 - mean_absolute_error: 0.2117 - val_loss: 0.1352 - val_mean_absolute_error: 0.2156\n",
      "Epoch 228/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1028 - mean_absolute_error: 0.2084 - val_loss: 0.1430 - val_mean_absolute_error: 0.2242\n",
      "Epoch 229/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1052 - mean_absolute_error: 0.2083 - val_loss: 0.1498 - val_mean_absolute_error: 0.2273\n",
      "Epoch 230/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1053 - mean_absolute_error: 0.2127 - val_loss: 0.1361 - val_mean_absolute_error: 0.2246\n",
      "Epoch 231/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1112 - mean_absolute_error: 0.2110 - val_loss: 0.1530 - val_mean_absolute_error: 0.2347\n",
      "Epoch 232/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1014 - mean_absolute_error: 0.2089 - val_loss: 0.1373 - val_mean_absolute_error: 0.2186\n",
      "Epoch 233/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1193 - mean_absolute_error: 0.2136 - val_loss: 0.1438 - val_mean_absolute_error: 0.2215\n",
      "Epoch 234/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1140 - mean_absolute_error: 0.2132 - val_loss: 0.1488 - val_mean_absolute_error: 0.2299\n",
      "Epoch 235/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1027 - mean_absolute_error: 0.2080 - val_loss: 0.1345 - val_mean_absolute_error: 0.2183\n",
      "Epoch 236/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1039 - mean_absolute_error: 0.2092 - val_loss: 0.1411 - val_mean_absolute_error: 0.2310\n",
      "Epoch 237/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1047 - mean_absolute_error: 0.2104 - val_loss: 0.1374 - val_mean_absolute_error: 0.2176\n",
      "Epoch 238/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1040 - mean_absolute_error: 0.2117 - val_loss: 0.1351 - val_mean_absolute_error: 0.2145\n",
      "Epoch 239/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1026 - mean_absolute_error: 0.2108 - val_loss: 0.1390 - val_mean_absolute_error: 0.2233\n",
      "Epoch 240/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1043 - mean_absolute_error: 0.2089 - val_loss: 0.1289 - val_mean_absolute_error: 0.2151\n",
      "Epoch 241/300\n",
      "12967/12967 [==============================] - ETA: 0s - loss: 0.0970 - mean_absolute_error: 0.207 - 1s 45us/sample - loss: 0.0967 - mean_absolute_error: 0.2075 - val_loss: 0.1336 - val_mean_absolute_error: 0.2181\n",
      "Epoch 242/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1024 - mean_absolute_error: 0.2107 - val_loss: 0.1457 - val_mean_absolute_error: 0.2357\n",
      "Epoch 243/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1007 - mean_absolute_error: 0.2082 - val_loss: 0.1492 - val_mean_absolute_error: 0.2314\n",
      "Epoch 244/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1094 - mean_absolute_error: 0.2129 - val_loss: 0.1555 - val_mean_absolute_error: 0.2168\n",
      "Epoch 245/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1063 - mean_absolute_error: 0.2100 - val_loss: 0.1504 - val_mean_absolute_error: 0.2283\n",
      "Epoch 246/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1089 - mean_absolute_error: 0.2098 - val_loss: 0.1390 - val_mean_absolute_error: 0.2289\n",
      "Epoch 247/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1068 - mean_absolute_error: 0.2097 - val_loss: 0.1365 - val_mean_absolute_error: 0.2182\n",
      "Epoch 248/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1026 - mean_absolute_error: 0.2096 - val_loss: 0.1383 - val_mean_absolute_error: 0.2174\n",
      "Epoch 249/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1051 - mean_absolute_error: 0.2136 - val_loss: 0.1393 - val_mean_absolute_error: 0.2342\n",
      "Epoch 250/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1015 - mean_absolute_error: 0.2109 - val_loss: 0.1527 - val_mean_absolute_error: 0.2302\n",
      "Epoch 251/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1079 - mean_absolute_error: 0.2111 - val_loss: 0.1324 - val_mean_absolute_error: 0.2283\n",
      "Epoch 252/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1097 - mean_absolute_error: 0.2128 - val_loss: 0.1328 - val_mean_absolute_error: 0.2143\n",
      "Epoch 253/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1050 - mean_absolute_error: 0.2111 - val_loss: 0.1505 - val_mean_absolute_error: 0.2302\n",
      "Epoch 254/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1070 - mean_absolute_error: 0.2112 - val_loss: 0.1442 - val_mean_absolute_error: 0.2189\n",
      "Epoch 255/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1035 - mean_absolute_error: 0.2088 - val_loss: 0.1421 - val_mean_absolute_error: 0.2178\n",
      "Epoch 256/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1056 - mean_absolute_error: 0.2114 - val_loss: 0.1528 - val_mean_absolute_error: 0.2260\n",
      "Epoch 257/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1037 - mean_absolute_error: 0.2105 - val_loss: 0.1604 - val_mean_absolute_error: 0.2363\n",
      "Epoch 258/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1035 - mean_absolute_error: 0.2104 - val_loss: 0.1534 - val_mean_absolute_error: 0.2292\n",
      "Epoch 259/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1108 - mean_absolute_error: 0.2114 - val_loss: 0.1297 - val_mean_absolute_error: 0.2061\n",
      "Epoch 260/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1033 - mean_absolute_error: 0.2077 - val_loss: 0.1429 - val_mean_absolute_error: 0.2362\n",
      "Epoch 261/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1097 - mean_absolute_error: 0.2113 - val_loss: 0.1328 - val_mean_absolute_error: 0.2202\n",
      "Epoch 262/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1083 - mean_absolute_error: 0.2105 - val_loss: 0.1416 - val_mean_absolute_error: 0.2237\n",
      "Epoch 263/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1047 - mean_absolute_error: 0.2089 - val_loss: 0.1539 - val_mean_absolute_error: 0.2319\n",
      "Epoch 264/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.0954 - mean_absolute_error: 0.2057 - val_loss: 0.1339 - val_mean_absolute_error: 0.2137\n",
      "Epoch 265/300\n",
      "12967/12967 [==============================] - 1s 48us/sample - loss: 0.1044 - mean_absolute_error: 0.2110 - val_loss: 0.1357 - val_mean_absolute_error: 0.2280\n",
      "Epoch 266/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1096 - mean_absolute_error: 0.2110 - val_loss: 0.1343 - val_mean_absolute_error: 0.2224\n",
      "Epoch 267/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1052 - mean_absolute_error: 0.2111 - val_loss: 0.1324 - val_mean_absolute_error: 0.2062\n",
      "Epoch 268/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1002 - mean_absolute_error: 0.2084 - val_loss: 0.1363 - val_mean_absolute_error: 0.2101\n",
      "Epoch 269/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1067 - mean_absolute_error: 0.2096 - val_loss: 0.1460 - val_mean_absolute_error: 0.2242\n",
      "Epoch 270/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1047 - mean_absolute_error: 0.2083 - val_loss: 0.1409 - val_mean_absolute_error: 0.2317\n",
      "Epoch 271/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1099 - mean_absolute_error: 0.2099 - val_loss: 0.1443 - val_mean_absolute_error: 0.2309\n",
      "Epoch 272/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1101 - mean_absolute_error: 0.2125 - val_loss: 0.1548 - val_mean_absolute_error: 0.2191\n",
      "Epoch 273/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1025 - mean_absolute_error: 0.2090 - val_loss: 0.1416 - val_mean_absolute_error: 0.2253\n",
      "Epoch 274/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1018 - mean_absolute_error: 0.2083 - val_loss: 0.1410 - val_mean_absolute_error: 0.2315\n",
      "Epoch 275/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1131 - mean_absolute_error: 0.2115 - val_loss: 0.1311 - val_mean_absolute_error: 0.2150\n",
      "Epoch 276/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1081 - mean_absolute_error: 0.2083 - val_loss: 0.1343 - val_mean_absolute_error: 0.2148\n",
      "Epoch 277/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1043 - mean_absolute_error: 0.2087 - val_loss: 0.1355 - val_mean_absolute_error: 0.2236\n",
      "Epoch 278/300\n",
      "12967/12967 [==============================] - 1s 50us/sample - loss: 0.1023 - mean_absolute_error: 0.2092 - val_loss: 0.1493 - val_mean_absolute_error: 0.2323\n",
      "Epoch 279/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1021 - mean_absolute_error: 0.2082 - val_loss: 0.1458 - val_mean_absolute_error: 0.2356\n",
      "Epoch 280/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1030 - mean_absolute_error: 0.2090 - val_loss: 0.1638 - val_mean_absolute_error: 0.2296\n",
      "Epoch 281/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1041 - mean_absolute_error: 0.2088 - val_loss: 0.1551 - val_mean_absolute_error: 0.2300\n",
      "Epoch 282/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1025 - mean_absolute_error: 0.2080 - val_loss: 0.1438 - val_mean_absolute_error: 0.2138\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 283/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1049 - mean_absolute_error: 0.2096 - val_loss: 0.1531 - val_mean_absolute_error: 0.2252\n",
      "Epoch 284/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1012 - mean_absolute_error: 0.2093 - val_loss: 0.1369 - val_mean_absolute_error: 0.2166\n",
      "Epoch 285/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1052 - mean_absolute_error: 0.2099 - val_loss: 0.1370 - val_mean_absolute_error: 0.2261\n",
      "Epoch 286/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1021 - mean_absolute_error: 0.2085 - val_loss: 0.1347 - val_mean_absolute_error: 0.2201\n",
      "Epoch 287/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.0993 - mean_absolute_error: 0.2085 - val_loss: 0.1493 - val_mean_absolute_error: 0.2267\n",
      "Epoch 288/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1064 - mean_absolute_error: 0.2124 - val_loss: 0.1605 - val_mean_absolute_error: 0.2348\n",
      "Epoch 289/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1030 - mean_absolute_error: 0.2088 - val_loss: 0.1527 - val_mean_absolute_error: 0.2269\n",
      "Epoch 290/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1040 - mean_absolute_error: 0.2076 - val_loss: 0.1373 - val_mean_absolute_error: 0.2284\n",
      "Epoch 291/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1024 - mean_absolute_error: 0.2089 - val_loss: 0.1371 - val_mean_absolute_error: 0.2231\n",
      "Epoch 292/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1002 - mean_absolute_error: 0.2084 - val_loss: 0.1359 - val_mean_absolute_error: 0.2201\n",
      "Epoch 293/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.0996 - mean_absolute_error: 0.2081 - val_loss: 0.1273 - val_mean_absolute_error: 0.2141\n",
      "Epoch 294/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.0999 - mean_absolute_error: 0.2071 - val_loss: 0.1525 - val_mean_absolute_error: 0.2166\n",
      "Epoch 295/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1039 - mean_absolute_error: 0.2087 - val_loss: 0.1338 - val_mean_absolute_error: 0.2204\n",
      "Epoch 296/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.1057 - mean_absolute_error: 0.2079 - val_loss: 0.1610 - val_mean_absolute_error: 0.2313\n",
      "Epoch 297/300\n",
      "12967/12967 [==============================] - 1s 44us/sample - loss: 0.0983 - mean_absolute_error: 0.2061 - val_loss: 0.1294 - val_mean_absolute_error: 0.2201\n",
      "Epoch 298/300\n",
      "12967/12967 [==============================] - 1s 45us/sample - loss: 0.1045 - mean_absolute_error: 0.2077 - val_loss: 0.1467 - val_mean_absolute_error: 0.2269\n",
      "Epoch 299/300\n",
      "12967/12967 [==============================] - 1s 47us/sample - loss: 0.1078 - mean_absolute_error: 0.2106 - val_loss: 0.1474 - val_mean_absolute_error: 0.2309\n",
      "Epoch 300/300\n",
      "12967/12967 [==============================] - 1s 46us/sample - loss: 0.1001 - mean_absolute_error: 0.2073 - val_loss: 0.1386 - val_mean_absolute_error: 0.2281\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x25d21ff89e8>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_4 = keras.Sequential(name='model-4')\n",
    "model_4.add(layers.Dense(64, activation='relu', input_shape=(21,)))\n",
    "model_4.add(layers.Dropout(0.3))\n",
    "model_4.add(layers.Dense(64, activation='relu'))\n",
    "model_4.add(layers.Dropout(0.3))\n",
    "model_4.add(layers.Dense(1))\n",
    "\n",
    "model_4.compile(keras.optimizers.Adam(0.001),\n",
    "                loss=keras.losses.MeanSquaredError(),\n",
    "                metrics=[keras.metrics.MeanAbsoluteError()])\n",
    "\n",
    "log_dir = os.path.join('lab2-logs', 'model-4')\n",
    "model_cbk = keras.callbacks.TensorBoard(log_dir=log_dir)\n",
    "model_mckp = keras.callbacks.ModelCheckpoint(model_dir + '/Best-model-4.h5', \n",
    "                                             monitor='val_mean_absolute_error', \n",
    "                                             save_best_only=True, \n",
    "                                             mode='min')\n",
    "model_4.fit(x_train, y_train, \n",
    "            batch_size=64 ,\n",
    "            epochs=300, \n",
    "            validation_data=(x_val, y_val), \n",
    "            callbacks=[model_cbk, model_mckp])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 驗證正則化的效能"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Test model 2:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model_2: 13.61%\n"
     ]
    }
   ],
   "source": [
    "model_2 = keras.models.load_model('lab2-logs/models/Best-model-2.h5')\n",
    "y_pred = model_2.predict(x_test)\n",
    "y_pred = np.reshape(y_pred * std['price'] + mean['price'], y_test.shape)\n",
    "percentage_error = np.mean(np.abs(y_test - y_pred)) / np.mean(y_test) * 100\n",
    "print(\"Model_2: {:.2f}%\".format(percentage_error))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Test model 3:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model_3: 12.90%\n"
     ]
    }
   ],
   "source": [
    "model_3 = keras.models.load_model('lab2-logs/models/Best-model-3.h5')\n",
    "y_pred = model_3.predict(x_test)\n",
    "y_pred = np.reshape(y_pred * std['price'] + mean['price'], y_test.shape)\n",
    "percentage_error = np.mean(np.abs(y_test - y_pred)) / np.mean(y_test) * 100\n",
    "print(\"Model_3: {:.2f}%\".format(percentage_error))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Test model 4:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model_4: 13.98%\n"
     ]
    }
   ],
   "source": [
    "model_4 = keras.models.load_model('lab2-logs/models/Best-model-4.h5')\n",
    "y_pred = model_4.predict(x_test)\n",
    "y_pred = np.reshape(y_pred * std['price'] + mean['price'], y_test.shape)\n",
    "percentage_error = np.mean(np.abs(y_test - y_pred)) / np.mean(y_test) * 100\n",
    "print(\"Model_4: {:.2f}%\".format(percentage_error))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "tf2",
   "language": "python",
   "name": "tf2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.7"
  },
  "pycharm": {
   "stem_cell": {
    "cell_type": "raw",
    "metadata": {
     "collapsed": false
    },
    "source": []
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}