2018/day41_predicting_house_prices_modelling_dl.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# House price prediction with Deep Learning "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/adhiman/anaconda3/envs/python3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88\n",
      "  return f(*args, **kwds)\n"
     ]
    }
   ],
   "source": [
    "from __future__ import print_function\n",
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/adhiman/anaconda3/envs/python3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import keras\n",
    "from keras import metrics\n",
    "from keras import regularizers\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Flatten, Activation\n",
    "from keras.layers import Conv2D, MaxPooling2D\n",
    "from keras.optimizers import Adam, RMSprop\n",
    "from keras.callbacks import TensorBoard, EarlyStopping, ModelCheckpoint\n",
    "from keras.utils import plot_model\n",
    "from keras.models import load_model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "kc_data_org = pd.read_csv(\"data/kc_house_data.csv.zip\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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>sale_yr</th>\n",
       "      <th>sale_month</th>\n",
       "      <th>sale_day</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>condition</th>\n",
       "      <th>grade</th>\n",
       "      <th>sqft_above</th>\n",
       "      <th>sqft_basement</th>\n",
       "      <th>yr_built</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>price</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2014</td>\n",
       "      <td>10</td>\n",
       "      <td>13</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>3</td>\n",
       "      <td>7</td>\n",
       "      <td>1180</td>\n",
       "      <td>0</td>\n",
       "      <td>1955</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>221900.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2014</td>\n",
       "      <td>12</td>\n",
       "      <td>9</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>3</td>\n",
       "      <td>7</td>\n",
       "      <td>2170</td>\n",
       "      <td>400</td>\n",
       "      <td>1951</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>538000.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2015</td>\n",
       "      <td>2</td>\n",
       "      <td>25</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>3</td>\n",
       "      <td>6</td>\n",
       "      <td>770</td>\n",
       "      <td>0</td>\n",
       "      <td>1933</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>180000.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2014</td>\n",
       "      <td>12</td>\n",
       "      <td>9</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>5</td>\n",
       "      <td>7</td>\n",
       "      <td>1050</td>\n",
       "      <td>910</td>\n",
       "      <td>1965</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>604000.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2015</td>\n",
       "      <td>2</td>\n",
       "      <td>18</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>3</td>\n",
       "      <td>8</td>\n",
       "      <td>1680</td>\n",
       "      <td>0</td>\n",
       "      <td>1987</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>510000.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   sale_yr  sale_month  sale_day  bedrooms  bathrooms  sqft_living  sqft_lot  \\\n",
       "0     2014          10        13         3       1.00         1180      5650   \n",
       "1     2014          12         9         3       2.25         2570      7242   \n",
       "2     2015           2        25         2       1.00          770     10000   \n",
       "3     2014          12         9         4       3.00         1960      5000   \n",
       "4     2015           2        18         3       2.00         1680      8080   \n",
       "\n",
       "   floors  condition  grade  sqft_above  sqft_basement  yr_built  zipcode  \\\n",
       "0     1.0          3      7        1180              0      1955    98178   \n",
       "1     2.0          3      7        2170            400      1951    98125   \n",
       "2     1.0          3      6         770              0      1933    98028   \n",
       "3     1.0          5      7        1050            910      1965    98136   \n",
       "4     1.0          3      8        1680              0      1987    98074   \n",
       "\n",
       "       lat     long  sqft_living15  sqft_lot15     price  \n",
       "0  47.5112 -122.257           1340        5650  221900.0  \n",
       "1  47.7210 -122.319           1690        7639  538000.0  \n",
       "2  47.7379 -122.233           2720        8062  180000.0  \n",
       "3  47.5208 -122.393           1360        5000  604000.0  \n",
       "4  47.6168 -122.045           1800        7503  510000.0  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "kc_data_org['sale_yr'] = pd.to_numeric(kc_data_org.date.str.slice(0, 4))\n",
    "kc_data_org['sale_month'] = pd.to_numeric(kc_data_org.date.str.slice(4, 6))\n",
    "kc_data_org['sale_day'] = pd.to_numeric(kc_data_org.date.str.slice(6, 8))\n",
    "\n",
    "kc_data = pd.DataFrame(kc_data_org, columns=[\n",
    "        'sale_yr','sale_month','sale_day',\n",
    "        'bedrooms','bathrooms','sqft_living','sqft_lot','floors',\n",
    "        'condition','grade','sqft_above','sqft_basement','yr_built',\n",
    "        'zipcode','lat','long','sqft_living15','sqft_lot15','price'])\n",
    "label_col = 'price'\n",
    "\n",
    "kc_data.head()"
   ]
  },
  {
   "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>sale_yr</th>\n",
       "      <th>sale_month</th>\n",
       "      <th>sale_day</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>condition</th>\n",
       "      <th>grade</th>\n",
       "      <th>sqft_above</th>\n",
       "      <th>sqft_basement</th>\n",
       "      <th>yr_built</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>price</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>2.161300e+04</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>21613.000000</td>\n",
       "      <td>2.161300e+04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>2014.322954</td>\n",
       "      <td>6.574423</td>\n",
       "      <td>15.688197</td>\n",
       "      <td>3.370842</td>\n",
       "      <td>2.114757</td>\n",
       "      <td>2079.899736</td>\n",
       "      <td>1.510697e+04</td>\n",
       "      <td>1.494309</td>\n",
       "      <td>3.409430</td>\n",
       "      <td>7.656873</td>\n",
       "      <td>1788.390691</td>\n",
       "      <td>291.509045</td>\n",
       "      <td>1971.005136</td>\n",
       "      <td>98077.939805</td>\n",
       "      <td>47.560053</td>\n",
       "      <td>-122.213896</td>\n",
       "      <td>1986.552492</td>\n",
       "      <td>12768.455652</td>\n",
       "      <td>5.400881e+05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>0.467616</td>\n",
       "      <td>3.115308</td>\n",
       "      <td>8.635063</td>\n",
       "      <td>0.930062</td>\n",
       "      <td>0.770163</td>\n",
       "      <td>918.440897</td>\n",
       "      <td>4.142051e+04</td>\n",
       "      <td>0.539989</td>\n",
       "      <td>0.650743</td>\n",
       "      <td>1.175459</td>\n",
       "      <td>828.090978</td>\n",
       "      <td>442.575043</td>\n",
       "      <td>29.373411</td>\n",
       "      <td>53.505026</td>\n",
       "      <td>0.138564</td>\n",
       "      <td>0.140828</td>\n",
       "      <td>685.391304</td>\n",
       "      <td>27304.179631</td>\n",
       "      <td>3.671272e+05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>2014.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>290.000000</td>\n",
       "      <td>5.200000e+02</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>290.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1900.000000</td>\n",
       "      <td>98001.000000</td>\n",
       "      <td>47.155900</td>\n",
       "      <td>-122.519000</td>\n",
       "      <td>399.000000</td>\n",
       "      <td>651.000000</td>\n",
       "      <td>7.500000e+04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>2014.000000</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>8.000000</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>1.750000</td>\n",
       "      <td>1427.000000</td>\n",
       "      <td>5.040000e+03</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>7.000000</td>\n",
       "      <td>1190.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1951.000000</td>\n",
       "      <td>98033.000000</td>\n",
       "      <td>47.471000</td>\n",
       "      <td>-122.328000</td>\n",
       "      <td>1490.000000</td>\n",
       "      <td>5100.000000</td>\n",
       "      <td>3.219500e+05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>2014.000000</td>\n",
       "      <td>6.000000</td>\n",
       "      <td>16.000000</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>2.250000</td>\n",
       "      <td>1910.000000</td>\n",
       "      <td>7.618000e+03</td>\n",
       "      <td>1.500000</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>7.000000</td>\n",
       "      <td>1560.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1975.000000</td>\n",
       "      <td>98065.000000</td>\n",
       "      <td>47.571800</td>\n",
       "      <td>-122.230000</td>\n",
       "      <td>1840.000000</td>\n",
       "      <td>7620.000000</td>\n",
       "      <td>4.500000e+05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>2015.000000</td>\n",
       "      <td>9.000000</td>\n",
       "      <td>23.000000</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>2.500000</td>\n",
       "      <td>2550.000000</td>\n",
       "      <td>1.068800e+04</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>8.000000</td>\n",
       "      <td>2210.000000</td>\n",
       "      <td>560.000000</td>\n",
       "      <td>1997.000000</td>\n",
       "      <td>98118.000000</td>\n",
       "      <td>47.678000</td>\n",
       "      <td>-122.125000</td>\n",
       "      <td>2360.000000</td>\n",
       "      <td>10083.000000</td>\n",
       "      <td>6.450000e+05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>2015.000000</td>\n",
       "      <td>12.000000</td>\n",
       "      <td>31.000000</td>\n",
       "      <td>33.000000</td>\n",
       "      <td>8.000000</td>\n",
       "      <td>13540.000000</td>\n",
       "      <td>1.651359e+06</td>\n",
       "      <td>3.500000</td>\n",
       "      <td>5.000000</td>\n",
       "      <td>13.000000</td>\n",
       "      <td>9410.000000</td>\n",
       "      <td>4820.000000</td>\n",
       "      <td>2015.000000</td>\n",
       "      <td>98199.000000</td>\n",
       "      <td>47.777600</td>\n",
       "      <td>-121.315000</td>\n",
       "      <td>6210.000000</td>\n",
       "      <td>871200.000000</td>\n",
       "      <td>7.700000e+06</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            sale_yr    sale_month      sale_day      bedrooms     bathrooms  \\\n",
       "count  21613.000000  21613.000000  21613.000000  21613.000000  21613.000000   \n",
       "mean    2014.322954      6.574423     15.688197      3.370842      2.114757   \n",
       "std        0.467616      3.115308      8.635063      0.930062      0.770163   \n",
       "min     2014.000000      1.000000      1.000000      0.000000      0.000000   \n",
       "25%     2014.000000      4.000000      8.000000      3.000000      1.750000   \n",
       "50%     2014.000000      6.000000     16.000000      3.000000      2.250000   \n",
       "75%     2015.000000      9.000000     23.000000      4.000000      2.500000   \n",
       "max     2015.000000     12.000000     31.000000     33.000000      8.000000   \n",
       "\n",
       "        sqft_living      sqft_lot        floors     condition         grade  \\\n",
       "count  21613.000000  2.161300e+04  21613.000000  21613.000000  21613.000000   \n",
       "mean    2079.899736  1.510697e+04      1.494309      3.409430      7.656873   \n",
       "std      918.440897  4.142051e+04      0.539989      0.650743      1.175459   \n",
       "min      290.000000  5.200000e+02      1.000000      1.000000      1.000000   \n",
       "25%     1427.000000  5.040000e+03      1.000000      3.000000      7.000000   \n",
       "50%     1910.000000  7.618000e+03      1.500000      3.000000      7.000000   \n",
       "75%     2550.000000  1.068800e+04      2.000000      4.000000      8.000000   \n",
       "max    13540.000000  1.651359e+06      3.500000      5.000000     13.000000   \n",
       "\n",
       "         sqft_above  sqft_basement      yr_built       zipcode           lat  \\\n",
       "count  21613.000000   21613.000000  21613.000000  21613.000000  21613.000000   \n",
       "mean    1788.390691     291.509045   1971.005136  98077.939805     47.560053   \n",
       "std      828.090978     442.575043     29.373411     53.505026      0.138564   \n",
       "min      290.000000       0.000000   1900.000000  98001.000000     47.155900   \n",
       "25%     1190.000000       0.000000   1951.000000  98033.000000     47.471000   \n",
       "50%     1560.000000       0.000000   1975.000000  98065.000000     47.571800   \n",
       "75%     2210.000000     560.000000   1997.000000  98118.000000     47.678000   \n",
       "max     9410.000000    4820.000000   2015.000000  98199.000000     47.777600   \n",
       "\n",
       "               long  sqft_living15     sqft_lot15         price  \n",
       "count  21613.000000   21613.000000   21613.000000  2.161300e+04  \n",
       "mean    -122.213896    1986.552492   12768.455652  5.400881e+05  \n",
       "std        0.140828     685.391304   27304.179631  3.671272e+05  \n",
       "min     -122.519000     399.000000     651.000000  7.500000e+04  \n",
       "25%     -122.328000    1490.000000    5100.000000  3.219500e+05  \n",
       "50%     -122.230000    1840.000000    7620.000000  4.500000e+05  \n",
       "75%     -122.125000    2360.000000   10083.000000  6.450000e+05  \n",
       "max     -121.315000    6210.000000  871200.000000  7.700000e+06  "
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "kc_data.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Split data from training and validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train_validate_test_split(df, train_part=.6, validate_part=.2, test_part=.2, seed=None):\n",
    "    np.random.seed(seed)\n",
    "    total_size = train_part + validate_part + test_part\n",
    "    train_percent = train_part / total_size\n",
    "    validate_percent = validate_part / total_size\n",
    "    test_percent = test_part / total_size\n",
    "    perm = np.random.permutation(df.index)\n",
    "    m = len(df)\n",
    "    train_end = int(train_percent * m)\n",
    "    validate_end = int(validate_percent * m) + train_end\n",
    "    train = perm[:train_end]\n",
    "    validate = perm[train_end:validate_end]\n",
    "    test = perm[validate_end:]\n",
    "    return train, validate, test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_size, valid_size, test_size = (70, 30, 0)\n",
    "kc_train, kc_valid, kc_test = train_validate_test_split(kc_data, \n",
    "                              train_part=train_size, \n",
    "                              validate_part=valid_size,\n",
    "                              test_part=test_size,\n",
    "                              seed=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Size of training set:  15129\n",
      "Size of validation set:  6483\n",
      "Size of test set:  1 (not converted)\n"
     ]
    }
   ],
   "source": [
    "kc_y_train = kc_data.loc[kc_train, [label_col]]\n",
    "kc_x_train = kc_data.loc[kc_train, :].drop(label_col, axis=1)\n",
    "kc_y_valid = kc_data.loc[kc_valid, [label_col]]\n",
    "kc_x_valid = kc_data.loc[kc_valid, :].drop(label_col, axis=1)\n",
    "\n",
    "print('Size of training set: ', len(kc_x_train))\n",
    "print('Size of validation set: ', len(kc_x_valid))\n",
    "print('Size of test set: ', len(kc_test), '(not converted)')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Prepare data for Keras"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def norm_stats(df1, df2):\n",
    "    dfs = df1.append(df2)\n",
    "    minimum = np.min(dfs)\n",
    "    maximum = np.max(dfs)\n",
    "    mu = np.mean(dfs)\n",
    "    sigma = np.std(dfs)\n",
    "    return (minimum, maximum, mu, sigma)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def z_score(col, stats):\n",
    "    m, M, mu, s = stats\n",
    "    df = pd.DataFrame()\n",
    "    for c in col.columns:\n",
    "        df[c] = (col[c]-mu[c])/s[c]\n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training shape: (15129, 18)\n",
      "Training samples:  15129\n",
      "Validation samples:  6483\n"
     ]
    }
   ],
   "source": [
    "stats = norm_stats(kc_x_train, kc_x_valid)\n",
    "arr_x_train = np.array(z_score(kc_x_train, stats))\n",
    "arr_y_train = np.array(kc_y_train)\n",
    "arr_x_valid = np.array(z_score(kc_x_valid, stats))\n",
    "arr_y_valid = np.array(kc_y_valid)\n",
    "\n",
    "print('Training shape:', arr_x_train.shape)\n",
    "print('Training samples: ', arr_x_train.shape[0])\n",
    "print('Validation samples: ', arr_x_valid.shape[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create keras model \n",
    "\n",
    "Three functions to define alternative Keras models\n",
    "\n",
    "The first is very simple, consisting of three layers and Adam optimizer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def basic_model_1(x_size, y_size):\n",
    "    t_model = Sequential()\n",
    "    t_model.add(Dense(100, activation=\"tanh\", input_shape=(x_size,)))\n",
    "    t_model.add(Dense(50, activation=\"relu\"))\n",
    "    t_model.add(Dense(y_size))\n",
    "    print(t_model.summary())\n",
    "    t_model.compile(loss='mean_absolute_error',\n",
    "        optimizer=Adam(),\n",
    "        metrics=[metrics.mae])\n",
    "    return(t_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The second with Adam optimizer consists of 4 layers and the first uses 10% dropouts."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def basic_model_2(x_size, y_size):\n",
    "    t_model = Sequential()\n",
    "    t_model.add(Dense(100, activation=\"tanh\", input_shape=(x_size,)))\n",
    "    t_model.add(Dropout(0.1))\n",
    "    t_model.add(Dense(50, activation=\"relu\"))\n",
    "    t_model.add(Dense(20, activation=\"relu\"))\n",
    "    t_model.add(Dense(y_size))\n",
    "    print(t_model.summary())\n",
    "    t_model.compile(loss='mean_absolute_error',\n",
    "        optimizer=Adam(),\n",
    "        metrics=[metrics.mae])\n",
    "    return(t_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The third is the most complex, it extends the previous model with Nadam optimizer, dropouts and L1/L2 regularisers."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "def basic_model_3(x_size, y_size):\n",
    "    t_model = Sequential()\n",
    "    t_model.add(Dense(80, activation=\"tanh\", kernel_initializer='normal', input_shape=(x_size,)))\n",
    "    t_model.add(Dropout(0.2))\n",
    "    t_model.add(Dense(120, activation=\"relu\", kernel_initializer='normal', \n",
    "        kernel_regularizer=regularizers.l1(0.01), bias_regularizer=regularizers.l1(0.01)))\n",
    "    t_model.add(Dropout(0.1))\n",
    "    t_model.add(Dense(20, activation=\"relu\", kernel_initializer='normal', \n",
    "        kernel_regularizer=regularizers.l1_l2(0.01), bias_regularizer=regularizers.l1_l2(0.01)))\n",
    "    t_model.add(Dropout(0.1))\n",
    "    t_model.add(Dense(10, activation=\"relu\", kernel_initializer='normal'))\n",
    "    t_model.add(Dropout(0.0))\n",
    "    t_model.add(Dense(y_size))\n",
    "    t_model.compile(\n",
    "        loss='mae',\n",
    "        optimizer='nadam',\n",
    "        metrics=['mae'])\n",
    "    return(t_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we create the model - use one of the above functions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_6 (Dense)              (None, 80)                1520      \n",
      "_________________________________________________________________\n",
      "dropout_5 (Dropout)          (None, 80)                0         \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 120)               9720      \n",
      "_________________________________________________________________\n",
      "dropout_6 (Dropout)          (None, 120)               0         \n",
      "_________________________________________________________________\n",
      "dense_8 (Dense)              (None, 20)                2420      \n",
      "_________________________________________________________________\n",
      "dropout_7 (Dropout)          (None, 20)                0         \n",
      "_________________________________________________________________\n",
      "dense_9 (Dense)              (None, 10)                210       \n",
      "_________________________________________________________________\n",
      "dropout_8 (Dropout)          (None, 10)                0         \n",
      "_________________________________________________________________\n",
      "dense_10 (Dense)             (None, 1)                 11        \n",
      "=================================================================\n",
      "Total params: 13,881\n",
      "Trainable params: 13,881\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = basic_model_1(arr_x_train.shape[1], arr_y_train.shape[1])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epochs:  500\n",
      "Batch size:  128\n"
     ]
    }
   ],
   "source": [
    "epochs = 500\n",
    "batch_size = 128\n",
    "\n",
    "print('Epochs: ', epochs)\n",
    "print('Batch size: ', batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 5min 40s, sys: 1min 2s, total: 6min 43s\n",
      "Wall time: 3min 34s\n"
     ]
    }
   ],
   "source": [
    "%%time \n",
    "\n",
    "history = model.fit(arr_x_train, arr_y_train,\n",
    "    batch_size=batch_size,\n",
    "    epochs=epochs,\n",
    "    shuffle=True,\n",
    "    verbose=0, # Change it to 2, if wished to observe execution\n",
    "    validation_data=(arr_x_valid, arr_y_valid))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluate and report performance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train MAE:  66672.1168 , Train Loss:  66887.5261\n",
      "Val MAE:  76592.7674 , Val Loss:  76808.1786\n"
     ]
    }
   ],
   "source": [
    "train_score = model.evaluate(arr_x_train, arr_y_train, verbose=0)\n",
    "valid_score = model.evaluate(arr_x_valid, arr_y_valid, verbose=0)\n",
    "\n",
    "print('Train MAE: ', round(train_score[1], 4), ', Train Loss: ', round(train_score[0], 4)) \n",
    "print('Val MAE: ', round(valid_score[1], 4), ', Val Loss: ', round(valid_score[0], 4))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Ploting training history"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_hist(h, xsize=6, ysize=10):\n",
    "    # Prepare plotting\n",
    "    fig_size = plt.rcParams[\"figure.figsize\"]\n",
    "    plt.rcParams[\"figure.figsize\"] = [xsize, ysize]\n",
    "    fig, axes = plt.subplots(nrows=4, ncols=4, sharex=True)\n",
    "    \n",
    "    # summarize history for MAE\n",
    "    plt.subplot(211)\n",
    "    plt.plot(h['mean_absolute_error'])\n",
    "    plt.plot(h['val_mean_absolute_error'])\n",
    "    plt.title('Training vs Validation MAE')\n",
    "    plt.ylabel('MAE')\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.legend(['Train', 'Validation'], loc='upper left')\n",
    "    \n",
    "    # summarize history for loss\n",
    "    plt.subplot(212)\n",
    "    plt.plot(h['loss'])\n",
    "    plt.plot(h['val_loss'])\n",
    "    plt.title('Training vs Validation Loss')\n",
    "    plt.ylabel('Loss')\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.legend(['Train', 'Validation'], loc='upper left')\n",
    "    \n",
    "    # Plot it all in IPython (non-interactive)\n",
    "    plt.draw()\n",
    "    plt.show()\n",
    "\n",
    "    return"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x864 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_hist(history.history, xsize=8, ysize=12)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}