{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "metric-heath",
   "metadata": {},
   "source": [
    "## Deep learning example from Accounting/Finance"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "operating-assignment",
   "metadata": {},
   "source": [
    "The following example demonstrates a simple example of deep learning that uses accounting/finance data. It also demonstrates, how to implement a deep learning model to traditional structured data. However, it also shows how deep learning is usually not the best option for structured data with relatively small datasets (<100k observations). Deep learning models perform better with large unstructured datasets."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "twelve-oregon",
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "concerned-baking",
   "metadata": {},
   "source": [
    "The data has a little under 20k observations. The variables are different financial ratios and board characteristics of S&P1500 companies."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "existing-ancient",
   "metadata": {},
   "outputs": [],
   "source": [
    "compu_df = pd.read_csv('_data.txt',delimiter='\\t')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "polar-active",
   "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>GVKEY</th>\n",
       "      <th>datadate</th>\n",
       "      <th>fyear</th>\n",
       "      <th>cusip</th>\n",
       "      <th>conm</th>\n",
       "      <th>act</th>\n",
       "      <th>at</th>\n",
       "      <th>bkvlps</th>\n",
       "      <th>capx</th>\n",
       "      <th>ceq</th>\n",
       "      <th>...</th>\n",
       "      <th>ind_chairman_is_ex_ceo</th>\n",
       "      <th>ind_independent_board_members</th>\n",
       "      <th>ind_strictly_independent_board_members</th>\n",
       "      <th>ind_board_member_affiliations</th>\n",
       "      <th>ind_non_executive_board_members</th>\n",
       "      <th>ind_board_gender_diversity_percent</th>\n",
       "      <th>ind_board_specific_skills_percent</th>\n",
       "      <th>ind_executive_members_gender_diversity_percent</th>\n",
       "      <th>ind_average_board_tenure</th>\n",
       "      <th>ind_board_member_compensation</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>21542</td>\n",
       "      <td>20081231</td>\n",
       "      <td>2008</td>\n",
       "      <td>000360206</td>\n",
       "      <td>AAON INC</td>\n",
       "      <td>80.118</td>\n",
       "      <td>140.743</td>\n",
       "      <td>5.6088</td>\n",
       "      <td>9.610</td>\n",
       "      <td>96.522</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>87.500</td>\n",
       "      <td>57.140</td>\n",
       "      <td>1.425</td>\n",
       "      <td>90.91</td>\n",
       "      <td>13.395</td>\n",
       "      <td>50.61</td>\n",
       "      <td>7.735</td>\n",
       "      <td>8.270</td>\n",
       "      <td>1650394.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>21542</td>\n",
       "      <td>20091231</td>\n",
       "      <td>2009</td>\n",
       "      <td>000360206</td>\n",
       "      <td>AAON INC</td>\n",
       "      <td>96.240</td>\n",
       "      <td>156.211</td>\n",
       "      <td>6.8544</td>\n",
       "      <td>9.774</td>\n",
       "      <td>117.999</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>87.500</td>\n",
       "      <td>50.000</td>\n",
       "      <td>1.180</td>\n",
       "      <td>90.00</td>\n",
       "      <td>12.915</td>\n",
       "      <td>60.00</td>\n",
       "      <td>6.670</td>\n",
       "      <td>8.705</td>\n",
       "      <td>1590889.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>21542</td>\n",
       "      <td>20101231</td>\n",
       "      <td>2010</td>\n",
       "      <td>000360206</td>\n",
       "      <td>AAON INC</td>\n",
       "      <td>91.748</td>\n",
       "      <td>160.277</td>\n",
       "      <td>7.0725</td>\n",
       "      <td>17.470</td>\n",
       "      <td>116.739</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>84.620</td>\n",
       "      <td>51.925</td>\n",
       "      <td>1.000</td>\n",
       "      <td>90.00</td>\n",
       "      <td>11.110</td>\n",
       "      <td>58.33</td>\n",
       "      <td>9.090</td>\n",
       "      <td>8.780</td>\n",
       "      <td>1801674.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>21542</td>\n",
       "      <td>20111231</td>\n",
       "      <td>2011</td>\n",
       "      <td>000360206</td>\n",
       "      <td>AAON INC</td>\n",
       "      <td>84.387</td>\n",
       "      <td>178.981</td>\n",
       "      <td>4.9762</td>\n",
       "      <td>35.914</td>\n",
       "      <td>122.504</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>86.670</td>\n",
       "      <td>50.000</td>\n",
       "      <td>1.090</td>\n",
       "      <td>90.00</td>\n",
       "      <td>11.110</td>\n",
       "      <td>57.14</td>\n",
       "      <td>10.000</td>\n",
       "      <td>9.180</td>\n",
       "      <td>1847006.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>21542</td>\n",
       "      <td>20121231</td>\n",
       "      <td>2012</td>\n",
       "      <td>000360206</td>\n",
       "      <td>AAON INC</td>\n",
       "      <td>91.546</td>\n",
       "      <td>193.493</td>\n",
       "      <td>5.6341</td>\n",
       "      <td>14.147</td>\n",
       "      <td>138.136</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>87.500</td>\n",
       "      <td>50.000</td>\n",
       "      <td>1.190</td>\n",
       "      <td>90.00</td>\n",
       "      <td>14.290</td>\n",
       "      <td>54.55</td>\n",
       "      <td>9.090</td>\n",
       "      <td>9.170</td>\n",
       "      <td>1810953.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19632</th>\n",
       "      <td>28191</td>\n",
       "      <td>20151231</td>\n",
       "      <td>2015</td>\n",
       "      <td>V7780T103</td>\n",
       "      <td>ROYAL CARIBBEAN GROUP</td>\n",
       "      <td>837.022</td>\n",
       "      <td>20921.855</td>\n",
       "      <td>36.9876</td>\n",
       "      <td>1613.340</td>\n",
       "      <td>8063.039</td>\n",
       "      <td>...</td>\n",
       "      <td>1.0</td>\n",
       "      <td>84.620</td>\n",
       "      <td>50.000</td>\n",
       "      <td>0.905</td>\n",
       "      <td>83.33</td>\n",
       "      <td>16.670</td>\n",
       "      <td>53.85</td>\n",
       "      <td>12.500</td>\n",
       "      <td>8.880</td>\n",
       "      <td>1744895.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19633</th>\n",
       "      <td>28191</td>\n",
       "      <td>20161231</td>\n",
       "      <td>2016</td>\n",
       "      <td>V7780T103</td>\n",
       "      <td>ROYAL CARIBBEAN GROUP</td>\n",
       "      <td>748.305</td>\n",
       "      <td>22310.324</td>\n",
       "      <td>42.5054</td>\n",
       "      <td>2494.363</td>\n",
       "      <td>9121.412</td>\n",
       "      <td>...</td>\n",
       "      <td>1.0</td>\n",
       "      <td>83.330</td>\n",
       "      <td>50.000</td>\n",
       "      <td>0.880</td>\n",
       "      <td>83.33</td>\n",
       "      <td>16.670</td>\n",
       "      <td>53.85</td>\n",
       "      <td>14.290</td>\n",
       "      <td>9.340</td>\n",
       "      <td>1737800.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19634</th>\n",
       "      <td>28191</td>\n",
       "      <td>20171231</td>\n",
       "      <td>2017</td>\n",
       "      <td>V7780T103</td>\n",
       "      <td>ROYAL CARIBBEAN GROUP</td>\n",
       "      <td>843.028</td>\n",
       "      <td>22296.317</td>\n",
       "      <td>50.1659</td>\n",
       "      <td>564.138</td>\n",
       "      <td>10702.303</td>\n",
       "      <td>...</td>\n",
       "      <td>1.0</td>\n",
       "      <td>83.330</td>\n",
       "      <td>50.000</td>\n",
       "      <td>0.885</td>\n",
       "      <td>83.33</td>\n",
       "      <td>20.000</td>\n",
       "      <td>57.14</td>\n",
       "      <td>13.395</td>\n",
       "      <td>9.105</td>\n",
       "      <td>1793588.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19635</th>\n",
       "      <td>28191</td>\n",
       "      <td>20181231</td>\n",
       "      <td>2018</td>\n",
       "      <td>V7780T103</td>\n",
       "      <td>ROYAL CARIBBEAN GROUP</td>\n",
       "      <td>1242.044</td>\n",
       "      <td>27698.270</td>\n",
       "      <td>53.1319</td>\n",
       "      <td>3660.028</td>\n",
       "      <td>11105.461</td>\n",
       "      <td>...</td>\n",
       "      <td>1.0</td>\n",
       "      <td>85.710</td>\n",
       "      <td>50.000</td>\n",
       "      <td>0.745</td>\n",
       "      <td>81.82</td>\n",
       "      <td>22.220</td>\n",
       "      <td>58.33</td>\n",
       "      <td>14.290</td>\n",
       "      <td>9.180</td>\n",
       "      <td>1858984.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19636</th>\n",
       "      <td>28191</td>\n",
       "      <td>20191231</td>\n",
       "      <td>2019</td>\n",
       "      <td>V7780T103</td>\n",
       "      <td>ROYAL CARIBBEAN GROUP</td>\n",
       "      <td>1162.628</td>\n",
       "      <td>30320.284</td>\n",
       "      <td>58.2557</td>\n",
       "      <td>3024.663</td>\n",
       "      <td>12163.846</td>\n",
       "      <td>...</td>\n",
       "      <td>1.0</td>\n",
       "      <td>85.165</td>\n",
       "      <td>48.075</td>\n",
       "      <td>0.800</td>\n",
       "      <td>80.00</td>\n",
       "      <td>25.000</td>\n",
       "      <td>60.00</td>\n",
       "      <td>15.190</td>\n",
       "      <td>9.195</td>\n",
       "      <td>1884002.5</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>19637 rows × 102 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       GVKEY  datadate  fyear      cusip                   conm       act  \\\n",
       "0      21542  20081231   2008  000360206               AAON INC    80.118   \n",
       "1      21542  20091231   2009  000360206               AAON INC    96.240   \n",
       "2      21542  20101231   2010  000360206               AAON INC    91.748   \n",
       "3      21542  20111231   2011  000360206               AAON INC    84.387   \n",
       "4      21542  20121231   2012  000360206               AAON INC    91.546   \n",
       "...      ...       ...    ...        ...                    ...       ...   \n",
       "19632  28191  20151231   2015  V7780T103  ROYAL CARIBBEAN GROUP   837.022   \n",
       "19633  28191  20161231   2016  V7780T103  ROYAL CARIBBEAN GROUP   748.305   \n",
       "19634  28191  20171231   2017  V7780T103  ROYAL CARIBBEAN GROUP   843.028   \n",
       "19635  28191  20181231   2018  V7780T103  ROYAL CARIBBEAN GROUP  1242.044   \n",
       "19636  28191  20191231   2019  V7780T103  ROYAL CARIBBEAN GROUP  1162.628   \n",
       "\n",
       "              at   bkvlps      capx        ceq  ...  ind_chairman_is_ex_ceo  \\\n",
       "0        140.743   5.6088     9.610     96.522  ...                     0.0   \n",
       "1        156.211   6.8544     9.774    117.999  ...                     0.0   \n",
       "2        160.277   7.0725    17.470    116.739  ...                     0.0   \n",
       "3        178.981   4.9762    35.914    122.504  ...                     0.0   \n",
       "4        193.493   5.6341    14.147    138.136  ...                     0.0   \n",
       "...          ...      ...       ...        ...  ...                     ...   \n",
       "19632  20921.855  36.9876  1613.340   8063.039  ...                     1.0   \n",
       "19633  22310.324  42.5054  2494.363   9121.412  ...                     1.0   \n",
       "19634  22296.317  50.1659   564.138  10702.303  ...                     1.0   \n",
       "19635  27698.270  53.1319  3660.028  11105.461  ...                     1.0   \n",
       "19636  30320.284  58.2557  3024.663  12163.846  ...                     1.0   \n",
       "\n",
       "       ind_independent_board_members  ind_strictly_independent_board_members  \\\n",
       "0                             87.500                                  57.140   \n",
       "1                             87.500                                  50.000   \n",
       "2                             84.620                                  51.925   \n",
       "3                             86.670                                  50.000   \n",
       "4                             87.500                                  50.000   \n",
       "...                              ...                                     ...   \n",
       "19632                         84.620                                  50.000   \n",
       "19633                         83.330                                  50.000   \n",
       "19634                         83.330                                  50.000   \n",
       "19635                         85.710                                  50.000   \n",
       "19636                         85.165                                  48.075   \n",
       "\n",
       "       ind_board_member_affiliations  ind_non_executive_board_members  \\\n",
       "0                              1.425                            90.91   \n",
       "1                              1.180                            90.00   \n",
       "2                              1.000                            90.00   \n",
       "3                              1.090                            90.00   \n",
       "4                              1.190                            90.00   \n",
       "...                              ...                              ...   \n",
       "19632                          0.905                            83.33   \n",
       "19633                          0.880                            83.33   \n",
       "19634                          0.885                            83.33   \n",
       "19635                          0.745                            81.82   \n",
       "19636                          0.800                            80.00   \n",
       "\n",
       "       ind_board_gender_diversity_percent  ind_board_specific_skills_percent  \\\n",
       "0                                  13.395                              50.61   \n",
       "1                                  12.915                              60.00   \n",
       "2                                  11.110                              58.33   \n",
       "3                                  11.110                              57.14   \n",
       "4                                  14.290                              54.55   \n",
       "...                                   ...                                ...   \n",
       "19632                              16.670                              53.85   \n",
       "19633                              16.670                              53.85   \n",
       "19634                              20.000                              57.14   \n",
       "19635                              22.220                              58.33   \n",
       "19636                              25.000                              60.00   \n",
       "\n",
       "       ind_executive_members_gender_diversity_percent  \\\n",
       "0                                               7.735   \n",
       "1                                               6.670   \n",
       "2                                               9.090   \n",
       "3                                              10.000   \n",
       "4                                               9.090   \n",
       "...                                               ...   \n",
       "19632                                          12.500   \n",
       "19633                                          14.290   \n",
       "19634                                          13.395   \n",
       "19635                                          14.290   \n",
       "19636                                          15.190   \n",
       "\n",
       "       ind_average_board_tenure  ind_board_member_compensation  \n",
       "0                         8.270                      1650394.0  \n",
       "1                         8.705                      1590889.5  \n",
       "2                         8.780                      1801674.0  \n",
       "3                         9.180                      1847006.5  \n",
       "4                         9.170                      1810953.0  \n",
       "...                         ...                            ...  \n",
       "19632                     8.880                      1744895.0  \n",
       "19633                     9.340                      1737800.0  \n",
       "19634                     9.105                      1793588.5  \n",
       "19635                     9.180                      1858984.0  \n",
       "19636                     9.195                      1884002.5  \n",
       "\n",
       "[19637 rows x 102 columns]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "compu_df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "experimental-brand",
   "metadata": {},
   "source": [
    "We use only variables with few missing values, because the benefits of deep learning are seen in large datasets. The variables are different financial ratios (Compustat). There are also many heavily correlating variables included, but this should not be a serious issue with neural networks."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "brave-maple",
   "metadata": {},
   "outputs": [],
   "source": [
    "variables = ['at', 'bkvlps','capx', 'ceq', 'csho', 'cstk', 'dlc', 'dltt', 'dvc', 'ebit',\n",
    "       'ibc', 'icapt', 'lt', 'ni', 'pstk', 'pstkl','pstkrv', 're', 'sale', 'seq', 'costat', 'prcc_c',\n",
    "       'prcc_f', 'sic', 'mkvalt','tobin', 'yld', 'age', 'tridx', 'mb',\n",
    "       'cap_int', 'lvg', 'roa', 'roe', 'roi']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "pediatric-poison",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "at           3\n",
       "bkvlps      34\n",
       "capx        53\n",
       "ceq          3\n",
       "csho        11\n",
       "cstk        13\n",
       "dlc          9\n",
       "dltt        81\n",
       "dvc         77\n",
       "ebit        15\n",
       "ibc         17\n",
       "icapt        4\n",
       "lt          49\n",
       "ni           7\n",
       "pstk         4\n",
       "pstkl       37\n",
       "pstkrv      19\n",
       "re          12\n",
       "sale         7\n",
       "seq          3\n",
       "costat       0\n",
       "prcc_c       8\n",
       "prcc_f       6\n",
       "sic          0\n",
       "mkvalt      13\n",
       "tobin       13\n",
       "yld         85\n",
       "age          0\n",
       "tridx      172\n",
       "mb          36\n",
       "cap_int     53\n",
       "lvg         86\n",
       "roa          7\n",
       "roe         16\n",
       "roi          8\n",
       "dtype: int64"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "compu_df[variables].isna().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "extreme-screen",
   "metadata": {},
   "outputs": [],
   "source": [
    "compu_df[variables] = compu_df[variables].clip(lower=compu_df[variables].quantile(0.01),\n",
    "                                               upper=compu_df[variables].quantile(0.99),axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "adopted-difficulty",
   "metadata": {},
   "outputs": [],
   "source": [
    "compu_df['current_roa'] = compu_df['roa']"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "linear-admission",
   "metadata": {},
   "source": [
    "Lag everything else"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "cellular-center",
   "metadata": {},
   "outputs": [],
   "source": [
    "compu_df[variables] = compu_df.groupby(['conm']).shift()[variables]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "superior-asset",
   "metadata": {},
   "source": [
    "We have to drop all missing values, because othwerise we can not optimize the network using gradient descent algorithm.\n",
    "\n",
    "I add industry, SP500 dummy and year separately as I do not want to winsorize or lag these variables."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "essential-master",
   "metadata": {},
   "outputs": [],
   "source": [
    "compu_df[variables + ['fyear','ind','sp500','current_roa']] = compu_df[variables + ['fyear','ind','sp500','current_roa']].dropna()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "developmental-partnership",
   "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>at</th>\n",
       "      <th>bkvlps</th>\n",
       "      <th>capx</th>\n",
       "      <th>ceq</th>\n",
       "      <th>csho</th>\n",
       "      <th>cstk</th>\n",
       "      <th>dlc</th>\n",
       "      <th>dltt</th>\n",
       "      <th>dvc</th>\n",
       "      <th>ebit</th>\n",
       "      <th>...</th>\n",
       "      <th>age</th>\n",
       "      <th>tridx</th>\n",
       "      <th>mb</th>\n",
       "      <th>cap_int</th>\n",
       "      <th>lvg</th>\n",
       "      <th>roa</th>\n",
       "      <th>roe</th>\n",
       "      <th>roi</th>\n",
       "      <th>fyear</th>\n",
       "      <th>current_roa</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>140.743</td>\n",
       "      <td>5.6088</td>\n",
       "      <td>9.610</td>\n",
       "      <td>96.522</td>\n",
       "      <td>17.209</td>\n",
       "      <td>0.071</td>\n",
       "      <td>2.992</td>\n",
       "      <td>0.000</td>\n",
       "      <td>5.621</td>\n",
       "      <td>43.388</td>\n",
       "      <td>...</td>\n",
       "      <td>204.0</td>\n",
       "      <td>1.070373</td>\n",
       "      <td>3.722721</td>\n",
       "      <td>0.068280</td>\n",
       "      <td>0.030998</td>\n",
       "      <td>0.203129</td>\n",
       "      <td>0.079563</td>\n",
       "      <td>0.296192</td>\n",
       "      <td>2009.0</td>\n",
       "      <td>0.177459</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>156.211</td>\n",
       "      <td>6.8544</td>\n",
       "      <td>9.774</td>\n",
       "      <td>117.999</td>\n",
       "      <td>17.215</td>\n",
       "      <td>0.071</td>\n",
       "      <td>0.076</td>\n",
       "      <td>0.000</td>\n",
       "      <td>6.201</td>\n",
       "      <td>43.754</td>\n",
       "      <td>...</td>\n",
       "      <td>216.0</td>\n",
       "      <td>1.017457</td>\n",
       "      <td>2.843429</td>\n",
       "      <td>0.062569</td>\n",
       "      <td>0.000644</td>\n",
       "      <td>0.177459</td>\n",
       "      <td>0.082621</td>\n",
       "      <td>0.234926</td>\n",
       "      <td>2010.0</td>\n",
       "      <td>0.136601</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>160.277</td>\n",
       "      <td>7.0725</td>\n",
       "      <td>17.470</td>\n",
       "      <td>116.739</td>\n",
       "      <td>16.506</td>\n",
       "      <td>0.068</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>6.067</td>\n",
       "      <td>32.715</td>\n",
       "      <td>...</td>\n",
       "      <td>228.0</td>\n",
       "      <td>1.494462</td>\n",
       "      <td>3.988689</td>\n",
       "      <td>0.108999</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.136601</td>\n",
       "      <td>0.047020</td>\n",
       "      <td>0.187547</td>\n",
       "      <td>2011.0</td>\n",
       "      <td>0.078142</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>178.981</td>\n",
       "      <td>4.9762</td>\n",
       "      <td>35.914</td>\n",
       "      <td>122.504</td>\n",
       "      <td>24.618</td>\n",
       "      <td>0.098</td>\n",
       "      <td>4.575</td>\n",
       "      <td>0.000</td>\n",
       "      <td>5.935</td>\n",
       "      <td>23.971</td>\n",
       "      <td>...</td>\n",
       "      <td>240.0</td>\n",
       "      <td>1.646141</td>\n",
       "      <td>4.117600</td>\n",
       "      <td>0.200658</td>\n",
       "      <td>0.037346</td>\n",
       "      <td>0.078142</td>\n",
       "      <td>0.027727</td>\n",
       "      <td>0.114168</td>\n",
       "      <td>2012.0</td>\n",
       "      <td>0.141860</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>193.493</td>\n",
       "      <td>5.6341</td>\n",
       "      <td>14.147</td>\n",
       "      <td>138.136</td>\n",
       "      <td>24.518</td>\n",
       "      <td>0.098</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>8.840</td>\n",
       "      <td>44.238</td>\n",
       "      <td>...</td>\n",
       "      <td>252.0</td>\n",
       "      <td>1.706581</td>\n",
       "      <td>3.704230</td>\n",
       "      <td>0.073114</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.141860</td>\n",
       "      <td>0.053644</td>\n",
       "      <td>0.198710</td>\n",
       "      <td>2013.0</td>\n",
       "      <td>0.174277</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>215.444</td>\n",
       "      <td>4.4702</td>\n",
       "      <td>9.041</td>\n",
       "      <td>164.106</td>\n",
       "      <td>36.711</td>\n",
       "      <td>0.147</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>7.428</td>\n",
       "      <td>55.803</td>\n",
       "      <td>...</td>\n",
       "      <td>264.0</td>\n",
       "      <td>3.949485</td>\n",
       "      <td>7.147331</td>\n",
       "      <td>0.041965</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.174277</td>\n",
       "      <td>0.032012</td>\n",
       "      <td>0.228797</td>\n",
       "      <td>2014.0</td>\n",
       "      <td>0.189424</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>233.117</td>\n",
       "      <td>3.2208</td>\n",
       "      <td>16.127</td>\n",
       "      <td>174.059</td>\n",
       "      <td>54.042</td>\n",
       "      <td>0.216</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>9.656</td>\n",
       "      <td>71.563</td>\n",
       "      <td>...</td>\n",
       "      <td>276.0</td>\n",
       "      <td>4.185799</td>\n",
       "      <td>6.951689</td>\n",
       "      <td>0.069180</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.189424</td>\n",
       "      <td>0.036494</td>\n",
       "      <td>0.253696</td>\n",
       "      <td>2015.0</td>\n",
       "      <td>0.196381</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>232.854</td>\n",
       "      <td>3.3750</td>\n",
       "      <td>20.967</td>\n",
       "      <td>178.918</td>\n",
       "      <td>53.012</td>\n",
       "      <td>0.212</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>11.857</td>\n",
       "      <td>71.695</td>\n",
       "      <td>...</td>\n",
       "      <td>288.0</td>\n",
       "      <td>4.381591</td>\n",
       "      <td>6.880000</td>\n",
       "      <td>0.090044</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.196381</td>\n",
       "      <td>0.037149</td>\n",
       "      <td>0.255581</td>\n",
       "      <td>2016.0</td>\n",
       "      <td>0.208069</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>256.530</td>\n",
       "      <td>3.9106</td>\n",
       "      <td>26.604</td>\n",
       "      <td>205.898</td>\n",
       "      <td>52.651</td>\n",
       "      <td>0.211</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>12.676</td>\n",
       "      <td>79.574</td>\n",
       "      <td>...</td>\n",
       "      <td>300.0</td>\n",
       "      <td>6.286182</td>\n",
       "      <td>8.451389</td>\n",
       "      <td>0.103707</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.208069</td>\n",
       "      <td>0.030674</td>\n",
       "      <td>0.259235</td>\n",
       "      <td>2017.0</td>\n",
       "      <td>0.183631</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>296.780</td>\n",
       "      <td>4.5252</td>\n",
       "      <td>41.713</td>\n",
       "      <td>237.226</td>\n",
       "      <td>52.423</td>\n",
       "      <td>0.210</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>13.653</td>\n",
       "      <td>74.148</td>\n",
       "      <td>...</td>\n",
       "      <td>312.0</td>\n",
       "      <td>7.030023</td>\n",
       "      <td>8.110139</td>\n",
       "      <td>0.140552</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.183631</td>\n",
       "      <td>0.028326</td>\n",
       "      <td>0.229730</td>\n",
       "      <td>2018.0</td>\n",
       "      <td>0.138132</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>308.197</td>\n",
       "      <td>4.7604</td>\n",
       "      <td>37.268</td>\n",
       "      <td>247.499</td>\n",
       "      <td>51.991</td>\n",
       "      <td>0.208</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>16.717</td>\n",
       "      <td>57.678</td>\n",
       "      <td>...</td>\n",
       "      <td>324.0</td>\n",
       "      <td>6.776646</td>\n",
       "      <td>7.364927</td>\n",
       "      <td>0.120923</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.138132</td>\n",
       "      <td>0.023355</td>\n",
       "      <td>0.172009</td>\n",
       "      <td>2019.0</td>\n",
       "      <td>0.144608</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>1377.511</td>\n",
       "      <td>16.8937</td>\n",
       "      <td>27.535</td>\n",
       "      <td>656.895</td>\n",
       "      <td>38.884</td>\n",
       "      <td>44.201</td>\n",
       "      <td>63.600</td>\n",
       "      <td>392.984</td>\n",
       "      <td>0.000</td>\n",
       "      <td>125.529</td>\n",
       "      <td>...</td>\n",
       "      <td>444.0</td>\n",
       "      <td>0.386537</td>\n",
       "      <td>1.089755</td>\n",
       "      <td>0.019989</td>\n",
       "      <td>0.695064</td>\n",
       "      <td>0.057096</td>\n",
       "      <td>0.109870</td>\n",
       "      <td>0.074914</td>\n",
       "      <td>2009.0</td>\n",
       "      <td>0.029731</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>1501.042</td>\n",
       "      <td>18.9167</td>\n",
       "      <td>28.855</td>\n",
       "      <td>746.906</td>\n",
       "      <td>39.484</td>\n",
       "      <td>44.870</td>\n",
       "      <td>100.833</td>\n",
       "      <td>336.191</td>\n",
       "      <td>0.000</td>\n",
       "      <td>95.415</td>\n",
       "      <td>...</td>\n",
       "      <td>456.0</td>\n",
       "      <td>0.518011</td>\n",
       "      <td>1.214800</td>\n",
       "      <td>0.019223</td>\n",
       "      <td>0.585112</td>\n",
       "      <td>0.029731</td>\n",
       "      <td>0.049185</td>\n",
       "      <td>0.041225</td>\n",
       "      <td>2010.0</td>\n",
       "      <td>0.040984</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>1703.727</td>\n",
       "      <td>21.0112</td>\n",
       "      <td>124.879</td>\n",
       "      <td>835.845</td>\n",
       "      <td>39.781</td>\n",
       "      <td>44.986</td>\n",
       "      <td>114.075</td>\n",
       "      <td>329.802</td>\n",
       "      <td>2.983</td>\n",
       "      <td>137.016</td>\n",
       "      <td>...</td>\n",
       "      <td>468.0</td>\n",
       "      <td>0.695924</td>\n",
       "      <td>1.307398</td>\n",
       "      <td>0.073298</td>\n",
       "      <td>0.531052</td>\n",
       "      <td>0.040984</td>\n",
       "      <td>0.063897</td>\n",
       "      <td>0.059932</td>\n",
       "      <td>2011.0</td>\n",
       "      <td>0.030844</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>2195.653</td>\n",
       "      <td>21.4697</td>\n",
       "      <td>91.218</td>\n",
       "      <td>864.649</td>\n",
       "      <td>40.273</td>\n",
       "      <td>44.849</td>\n",
       "      <td>122.865</td>\n",
       "      <td>669.489</td>\n",
       "      <td>12.081</td>\n",
       "      <td>142.360</td>\n",
       "      <td>...</td>\n",
       "      <td>480.0</td>\n",
       "      <td>0.322468</td>\n",
       "      <td>0.892886</td>\n",
       "      <td>0.041545</td>\n",
       "      <td>0.916388</td>\n",
       "      <td>0.030844</td>\n",
       "      <td>0.087720</td>\n",
       "      <td>0.044105</td>\n",
       "      <td>2012.0</td>\n",
       "      <td>0.025738</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>2136.900</td>\n",
       "      <td>23.3254</td>\n",
       "      <td>37.600</td>\n",
       "      <td>918.600</td>\n",
       "      <td>39.382</td>\n",
       "      <td>44.700</td>\n",
       "      <td>86.400</td>\n",
       "      <td>622.200</td>\n",
       "      <td>11.900</td>\n",
       "      <td>136.600</td>\n",
       "      <td>...</td>\n",
       "      <td>492.0</td>\n",
       "      <td>0.546782</td>\n",
       "      <td>0.800844</td>\n",
       "      <td>0.017596</td>\n",
       "      <td>0.771391</td>\n",
       "      <td>0.025738</td>\n",
       "      <td>0.074763</td>\n",
       "      <td>0.035675</td>\n",
       "      <td>2013.0</td>\n",
       "      <td>0.033144</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>2199.500</td>\n",
       "      <td>25.2654</td>\n",
       "      <td>26.500</td>\n",
       "      <td>999.500</td>\n",
       "      <td>39.560</td>\n",
       "      <td>44.700</td>\n",
       "      <td>69.700</td>\n",
       "      <td>564.300</td>\n",
       "      <td>11.800</td>\n",
       "      <td>142.600</td>\n",
       "      <td>...</td>\n",
       "      <td>504.0</td>\n",
       "      <td>0.669914</td>\n",
       "      <td>1.108631</td>\n",
       "      <td>0.012048</td>\n",
       "      <td>0.634317</td>\n",
       "      <td>0.033144</td>\n",
       "      <td>0.065790</td>\n",
       "      <td>0.046581</td>\n",
       "      <td>2014.0</td>\n",
       "      <td>0.006733</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>1515.000</td>\n",
       "      <td>23.8574</td>\n",
       "      <td>46.300</td>\n",
       "      <td>845.100</td>\n",
       "      <td>35.423</td>\n",
       "      <td>44.900</td>\n",
       "      <td>69.000</td>\n",
       "      <td>85.000</td>\n",
       "      <td>11.900</td>\n",
       "      <td>-8.600</td>\n",
       "      <td>...</td>\n",
       "      <td>516.0</td>\n",
       "      <td>0.823134</td>\n",
       "      <td>1.164419</td>\n",
       "      <td>0.030561</td>\n",
       "      <td>0.182227</td>\n",
       "      <td>0.006733</td>\n",
       "      <td>0.010365</td>\n",
       "      <td>0.010967</td>\n",
       "      <td>2015.0</td>\n",
       "      <td>0.033077</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>1442.100</td>\n",
       "      <td>25.0847</td>\n",
       "      <td>88.400</td>\n",
       "      <td>865.800</td>\n",
       "      <td>34.515</td>\n",
       "      <td>44.900</td>\n",
       "      <td>12.000</td>\n",
       "      <td>136.100</td>\n",
       "      <td>10.400</td>\n",
       "      <td>66.100</td>\n",
       "      <td>...</td>\n",
       "      <td>528.0</td>\n",
       "      <td>0.688919</td>\n",
       "      <td>1.048049</td>\n",
       "      <td>0.061299</td>\n",
       "      <td>0.171056</td>\n",
       "      <td>0.033077</td>\n",
       "      <td>0.052568</td>\n",
       "      <td>0.047610</td>\n",
       "      <td>2016.0</td>\n",
       "      <td>0.037564</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>1504.100</td>\n",
       "      <td>26.6112</td>\n",
       "      <td>33.600</td>\n",
       "      <td>914.200</td>\n",
       "      <td>34.354</td>\n",
       "      <td>45.200</td>\n",
       "      <td>2.000</td>\n",
       "      <td>155.300</td>\n",
       "      <td>10.200</td>\n",
       "      <td>77.200</td>\n",
       "      <td>...</td>\n",
       "      <td>540.0</td>\n",
       "      <td>0.995869</td>\n",
       "      <td>1.241958</td>\n",
       "      <td>0.022339</td>\n",
       "      <td>0.172063</td>\n",
       "      <td>0.037564</td>\n",
       "      <td>0.049762</td>\n",
       "      <td>0.052828</td>\n",
       "      <td>2017.0</td>\n",
       "      <td>0.010232</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>1524.700</td>\n",
       "      <td>26.9703</td>\n",
       "      <td>22.000</td>\n",
       "      <td>936.300</td>\n",
       "      <td>34.716</td>\n",
       "      <td>45.300</td>\n",
       "      <td>0.000</td>\n",
       "      <td>177.200</td>\n",
       "      <td>10.300</td>\n",
       "      <td>86.000</td>\n",
       "      <td>...</td>\n",
       "      <td>552.0</td>\n",
       "      <td>1.283369</td>\n",
       "      <td>1.456788</td>\n",
       "      <td>0.014429</td>\n",
       "      <td>0.189256</td>\n",
       "      <td>0.010232</td>\n",
       "      <td>0.011437</td>\n",
       "      <td>0.014010</td>\n",
       "      <td>2018.0</td>\n",
       "      <td>0.004943</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>1517.200</td>\n",
       "      <td>26.0406</td>\n",
       "      <td>17.400</td>\n",
       "      <td>905.900</td>\n",
       "      <td>34.788</td>\n",
       "      <td>45.300</td>\n",
       "      <td>0.000</td>\n",
       "      <td>141.700</td>\n",
       "      <td>10.500</td>\n",
       "      <td>110.700</td>\n",
       "      <td>...</td>\n",
       "      <td>564.0</td>\n",
       "      <td>0.870483</td>\n",
       "      <td>1.433915</td>\n",
       "      <td>0.011468</td>\n",
       "      <td>0.156419</td>\n",
       "      <td>0.004943</td>\n",
       "      <td>0.005774</td>\n",
       "      <td>0.007159</td>\n",
       "      <td>2019.0</td>\n",
       "      <td>0.002116</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>1549.913</td>\n",
       "      <td>12.6376</td>\n",
       "      <td>34.063</td>\n",
       "      <td>644.051</td>\n",
       "      <td>50.963</td>\n",
       "      <td>0.581</td>\n",
       "      <td>0.000</td>\n",
       "      <td>230.000</td>\n",
       "      <td>25.271</td>\n",
       "      <td>114.309</td>\n",
       "      <td>...</td>\n",
       "      <td>468.0</td>\n",
       "      <td>0.820994</td>\n",
       "      <td>1.507406</td>\n",
       "      <td>0.021977</td>\n",
       "      <td>0.357115</td>\n",
       "      <td>0.029314</td>\n",
       "      <td>0.046798</td>\n",
       "      <td>0.051981</td>\n",
       "      <td>2009.0</td>\n",
       "      <td>0.035692</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>1521.153</td>\n",
       "      <td>13.2923</td>\n",
       "      <td>18.582</td>\n",
       "      <td>687.050</td>\n",
       "      <td>51.688</td>\n",
       "      <td>0.517</td>\n",
       "      <td>0.000</td>\n",
       "      <td>172.500</td>\n",
       "      <td>26.727</td>\n",
       "      <td>82.506</td>\n",
       "      <td>...</td>\n",
       "      <td>480.0</td>\n",
       "      <td>0.973093</td>\n",
       "      <td>1.554283</td>\n",
       "      <td>0.012216</td>\n",
       "      <td>0.251073</td>\n",
       "      <td>0.035692</td>\n",
       "      <td>0.050842</td>\n",
       "      <td>0.063164</td>\n",
       "      <td>2010.0</td>\n",
       "      <td>0.041404</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>1548.670</td>\n",
       "      <td>14.0406</td>\n",
       "      <td>23.942</td>\n",
       "      <td>739.025</td>\n",
       "      <td>52.635</td>\n",
       "      <td>0.526</td>\n",
       "      <td>0.000</td>\n",
       "      <td>140.500</td>\n",
       "      <td>28.152</td>\n",
       "      <td>118.882</td>\n",
       "      <td>...</td>\n",
       "      <td>492.0</td>\n",
       "      <td>1.198713</td>\n",
       "      <td>1.873139</td>\n",
       "      <td>0.015460</td>\n",
       "      <td>0.190115</td>\n",
       "      <td>0.041404</td>\n",
       "      <td>0.046320</td>\n",
       "      <td>0.072904</td>\n",
       "      <td>2011.0</td>\n",
       "      <td>0.036446</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28</th>\n",
       "      <td>1879.598</td>\n",
       "      <td>14.9230</td>\n",
       "      <td>22.124</td>\n",
       "      <td>795.886</td>\n",
       "      <td>53.333</td>\n",
       "      <td>0.533</td>\n",
       "      <td>0.000</td>\n",
       "      <td>300.000</td>\n",
       "      <td>29.744</td>\n",
       "      <td>124.148</td>\n",
       "      <td>...</td>\n",
       "      <td>504.0</td>\n",
       "      <td>1.101268</td>\n",
       "      <td>1.381760</td>\n",
       "      <td>0.011771</td>\n",
       "      <td>0.376938</td>\n",
       "      <td>0.036446</td>\n",
       "      <td>0.062292</td>\n",
       "      <td>0.062510</td>\n",
       "      <td>2012.0</td>\n",
       "      <td>0.033480</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29</th>\n",
       "      <td>1869.251</td>\n",
       "      <td>15.6340</td>\n",
       "      <td>28.052</td>\n",
       "      <td>850.398</td>\n",
       "      <td>54.394</td>\n",
       "      <td>0.544</td>\n",
       "      <td>0.000</td>\n",
       "      <td>215.000</td>\n",
       "      <td>31.309</td>\n",
       "      <td>111.711</td>\n",
       "      <td>...</td>\n",
       "      <td>516.0</td>\n",
       "      <td>1.064461</td>\n",
       "      <td>1.276065</td>\n",
       "      <td>0.015007</td>\n",
       "      <td>0.252823</td>\n",
       "      <td>0.033480</td>\n",
       "      <td>0.057671</td>\n",
       "      <td>0.058740</td>\n",
       "      <td>2013.0</td>\n",
       "      <td>0.034399</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>30 rows × 37 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          at   bkvlps     capx      ceq    csho    cstk      dlc     dltt  \\\n",
       "0        NaN      NaN      NaN      NaN     NaN     NaN      NaN      NaN   \n",
       "1    140.743   5.6088    9.610   96.522  17.209   0.071    2.992    0.000   \n",
       "2    156.211   6.8544    9.774  117.999  17.215   0.071    0.076    0.000   \n",
       "3    160.277   7.0725   17.470  116.739  16.506   0.068    0.000    0.000   \n",
       "4    178.981   4.9762   35.914  122.504  24.618   0.098    4.575    0.000   \n",
       "5    193.493   5.6341   14.147  138.136  24.518   0.098    0.000    0.000   \n",
       "6    215.444   4.4702    9.041  164.106  36.711   0.147    0.000    0.000   \n",
       "7    233.117   3.2208   16.127  174.059  54.042   0.216    0.000    0.000   \n",
       "8    232.854   3.3750   20.967  178.918  53.012   0.212    0.000    0.000   \n",
       "9    256.530   3.9106   26.604  205.898  52.651   0.211    0.000    0.000   \n",
       "10   296.780   4.5252   41.713  237.226  52.423   0.210    0.000    0.000   \n",
       "11   308.197   4.7604   37.268  247.499  51.991   0.208    0.000    0.000   \n",
       "12       NaN      NaN      NaN      NaN     NaN     NaN      NaN      NaN   \n",
       "13  1377.511  16.8937   27.535  656.895  38.884  44.201   63.600  392.984   \n",
       "14  1501.042  18.9167   28.855  746.906  39.484  44.870  100.833  336.191   \n",
       "15  1703.727  21.0112  124.879  835.845  39.781  44.986  114.075  329.802   \n",
       "16  2195.653  21.4697   91.218  864.649  40.273  44.849  122.865  669.489   \n",
       "17  2136.900  23.3254   37.600  918.600  39.382  44.700   86.400  622.200   \n",
       "18  2199.500  25.2654   26.500  999.500  39.560  44.700   69.700  564.300   \n",
       "19  1515.000  23.8574   46.300  845.100  35.423  44.900   69.000   85.000   \n",
       "20  1442.100  25.0847   88.400  865.800  34.515  44.900   12.000  136.100   \n",
       "21  1504.100  26.6112   33.600  914.200  34.354  45.200    2.000  155.300   \n",
       "22  1524.700  26.9703   22.000  936.300  34.716  45.300    0.000  177.200   \n",
       "23  1517.200  26.0406   17.400  905.900  34.788  45.300    0.000  141.700   \n",
       "24       NaN      NaN      NaN      NaN     NaN     NaN      NaN      NaN   \n",
       "25  1549.913  12.6376   34.063  644.051  50.963   0.581    0.000  230.000   \n",
       "26  1521.153  13.2923   18.582  687.050  51.688   0.517    0.000  172.500   \n",
       "27  1548.670  14.0406   23.942  739.025  52.635   0.526    0.000  140.500   \n",
       "28  1879.598  14.9230   22.124  795.886  53.333   0.533    0.000  300.000   \n",
       "29  1869.251  15.6340   28.052  850.398  54.394   0.544    0.000  215.000   \n",
       "\n",
       "       dvc     ebit  ...    age     tridx        mb   cap_int       lvg  \\\n",
       "0      NaN      NaN  ...    NaN       NaN       NaN       NaN       NaN   \n",
       "1    5.621   43.388  ...  204.0  1.070373  3.722721  0.068280  0.030998   \n",
       "2    6.201   43.754  ...  216.0  1.017457  2.843429  0.062569  0.000644   \n",
       "3    6.067   32.715  ...  228.0  1.494462  3.988689  0.108999  0.000000   \n",
       "4    5.935   23.971  ...  240.0  1.646141  4.117600  0.200658  0.037346   \n",
       "5    8.840   44.238  ...  252.0  1.706581  3.704230  0.073114  0.000000   \n",
       "6    7.428   55.803  ...  264.0  3.949485  7.147331  0.041965  0.000000   \n",
       "7    9.656   71.563  ...  276.0  4.185799  6.951689  0.069180  0.000000   \n",
       "8   11.857   71.695  ...  288.0  4.381591  6.880000  0.090044  0.000000   \n",
       "9   12.676   79.574  ...  300.0  6.286182  8.451389  0.103707  0.000000   \n",
       "10  13.653   74.148  ...  312.0  7.030023  8.110139  0.140552  0.000000   \n",
       "11  16.717   57.678  ...  324.0  6.776646  7.364927  0.120923  0.000000   \n",
       "12     NaN      NaN  ...    NaN       NaN       NaN       NaN       NaN   \n",
       "13   0.000  125.529  ...  444.0  0.386537  1.089755  0.019989  0.695064   \n",
       "14   0.000   95.415  ...  456.0  0.518011  1.214800  0.019223  0.585112   \n",
       "15   2.983  137.016  ...  468.0  0.695924  1.307398  0.073298  0.531052   \n",
       "16  12.081  142.360  ...  480.0  0.322468  0.892886  0.041545  0.916388   \n",
       "17  11.900  136.600  ...  492.0  0.546782  0.800844  0.017596  0.771391   \n",
       "18  11.800  142.600  ...  504.0  0.669914  1.108631  0.012048  0.634317   \n",
       "19  11.900   -8.600  ...  516.0  0.823134  1.164419  0.030561  0.182227   \n",
       "20  10.400   66.100  ...  528.0  0.688919  1.048049  0.061299  0.171056   \n",
       "21  10.200   77.200  ...  540.0  0.995869  1.241958  0.022339  0.172063   \n",
       "22  10.300   86.000  ...  552.0  1.283369  1.456788  0.014429  0.189256   \n",
       "23  10.500  110.700  ...  564.0  0.870483  1.433915  0.011468  0.156419   \n",
       "24     NaN      NaN  ...    NaN       NaN       NaN       NaN       NaN   \n",
       "25  25.271  114.309  ...  468.0  0.820994  1.507406  0.021977  0.357115   \n",
       "26  26.727   82.506  ...  480.0  0.973093  1.554283  0.012216  0.251073   \n",
       "27  28.152  118.882  ...  492.0  1.198713  1.873139  0.015460  0.190115   \n",
       "28  29.744  124.148  ...  504.0  1.101268  1.381760  0.011771  0.376938   \n",
       "29  31.309  111.711  ...  516.0  1.064461  1.276065  0.015007  0.252823   \n",
       "\n",
       "         roa       roe       roi   fyear  current_roa  \n",
       "0        NaN       NaN       NaN     NaN          NaN  \n",
       "1   0.203129  0.079563  0.296192  2009.0     0.177459  \n",
       "2   0.177459  0.082621  0.234926  2010.0     0.136601  \n",
       "3   0.136601  0.047020  0.187547  2011.0     0.078142  \n",
       "4   0.078142  0.027727  0.114168  2012.0     0.141860  \n",
       "5   0.141860  0.053644  0.198710  2013.0     0.174277  \n",
       "6   0.174277  0.032012  0.228797  2014.0     0.189424  \n",
       "7   0.189424  0.036494  0.253696  2015.0     0.196381  \n",
       "8   0.196381  0.037149  0.255581  2016.0     0.208069  \n",
       "9   0.208069  0.030674  0.259235  2017.0     0.183631  \n",
       "10  0.183631  0.028326  0.229730  2018.0     0.138132  \n",
       "11  0.138132  0.023355  0.172009  2019.0     0.144608  \n",
       "12       NaN       NaN       NaN     NaN          NaN  \n",
       "13  0.057096  0.109870  0.074914  2009.0     0.029731  \n",
       "14  0.029731  0.049185  0.041225  2010.0     0.040984  \n",
       "15  0.040984  0.063897  0.059932  2011.0     0.030844  \n",
       "16  0.030844  0.087720  0.044105  2012.0     0.025738  \n",
       "17  0.025738  0.074763  0.035675  2013.0     0.033144  \n",
       "18  0.033144  0.065790  0.046581  2014.0     0.006733  \n",
       "19  0.006733  0.010365  0.010967  2015.0     0.033077  \n",
       "20  0.033077  0.052568  0.047610  2016.0     0.037564  \n",
       "21  0.037564  0.049762  0.052828  2017.0     0.010232  \n",
       "22  0.010232  0.011437  0.014010  2018.0     0.004943  \n",
       "23  0.004943  0.005774  0.007159  2019.0     0.002116  \n",
       "24       NaN       NaN       NaN     NaN          NaN  \n",
       "25  0.029314  0.046798  0.051981  2009.0     0.035692  \n",
       "26  0.035692  0.050842  0.063164  2010.0     0.041404  \n",
       "27  0.041404  0.046320  0.072904  2011.0     0.036446  \n",
       "28  0.036446  0.062292  0.062510  2012.0     0.033480  \n",
       "29  0.033480  0.057671  0.058740  2013.0     0.034399  \n",
       "\n",
       "[30 rows x 37 columns]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "compu_df[variables + ['fyear','current_roa']].head(30)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "wound-cinema",
   "metadata": {},
   "source": [
    "I remove the first year (2008), because we do not have any observations there due to the lag procedure."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "separate-genesis",
   "metadata": {},
   "outputs": [],
   "source": [
    "compu_df = compu_df[compu_df['fyear'] > 2008.]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "caroline-index",
   "metadata": {},
   "source": [
    "We try to predict current ROA with the last year's variable values."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "generous-pocket",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_df = compu_df['current_roa']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "unlike-consortium",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_df = compu_df[variables + ['fyear','ind','sp500']]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "planned-nightlife",
   "metadata": {},
   "source": [
    "Train/test split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "prepared-shade",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "permanent-journalism",
   "metadata": {},
   "source": [
    "Tensoflow does not like Pandas dataframes, so I change them to Numpy array.s"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "derived-thousand",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Split data into training and test sets\n",
    "x_train, x_test , y_train, y_test = train_test_split(x_df.values, y_df.values, test_size=0.20, random_state=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "loving-principal",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "numpy.ndarray"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(x_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "ultimate-address",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(14017, 3505)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(x_train), len(x_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "decent-niagara",
   "metadata": {},
   "source": [
    "Let's check that there is no missing values any more."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "expressed-series",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "at             0\n",
       "bkvlps         0\n",
       "capx           0\n",
       "ceq            0\n",
       "csho           0\n",
       "cstk           0\n",
       "dlc            0\n",
       "dltt           0\n",
       "dvc            0\n",
       "ebit           0\n",
       "ibc            0\n",
       "icapt          0\n",
       "lt             0\n",
       "ni             0\n",
       "pstk           0\n",
       "pstkl          0\n",
       "pstkrv         0\n",
       "re             0\n",
       "sale           0\n",
       "seq            0\n",
       "costat         0\n",
       "prcc_c         0\n",
       "prcc_f         0\n",
       "sic            0\n",
       "mkvalt         0\n",
       "tobin          0\n",
       "yld            0\n",
       "age            0\n",
       "tridx          0\n",
       "mb             0\n",
       "cap_int        0\n",
       "lvg            0\n",
       "roa            0\n",
       "roe            0\n",
       "roi            0\n",
       "current_roa    0\n",
       "fyear          0\n",
       "dtype: int64"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "compu_df[variables+['current_roa','fyear']].isna().sum()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "under-looking",
   "metadata": {},
   "source": [
    "### Densely connected network\n",
    "Let's build a traditional densely connected neural network. We could also use recurrent or LSTM networks, but we would need to reorganize data in that case.\n",
    "\n",
    "![image.png](./images/feed_forward.png)!"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "continued-prime",
   "metadata": {},
   "source": [
    "One way to define a neural network with Keras is a single **Sequential**-command that has the layers in a list as a parameter. The densely connected layers have **ReLU** as an activation function. The last layer has one neuron, because we want to have a single valua as an output (current ROA). There is also no activation function, because we want to have linear output. There is also a dropout layer to counter overfitting. For the first layer, we need to define the shape of our input."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "absent-chinese",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = tf.keras.models.Sequential([\n",
    "    tf.keras.layers.Dense(128, activation='relu',input_shape = (38,)),\n",
    "    tf.keras.layers.Dense(64, activation='relu'),\n",
    "    tf.keras.layers.Dropout(0.1),\n",
    "    tf.keras.layers.Dense(32, activation='relu'),\n",
    "    tf.keras.layers.Dense(16, activation='relu'),\n",
    "    tf.keras.layers.Dense(1)])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "handed-atmosphere",
   "metadata": {},
   "source": [
    "You can check your model with the **summary()** -function. The model has 203 530 parameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "constant-department",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense (Dense)                (None, 128)               4992      \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 16)                528       \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 1)                 17        \n",
      "=================================================================\n",
      "Total params: 15,873\n",
      "Trainable params: 15,873\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "quiet-ebony",
   "metadata": {},
   "source": [
    "With **compile()**, we build our neural network to be ready for use. For regerssion problems, **MSE** is the correct loss function. We measure our performance with **Mean Absolute Error**, because it is easier to interpret than MSE."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "digital-bobby",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss='mse',metrics=['mae'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "blank-detective",
   "metadata": {},
   "source": [
    "Next, we feed the training data to our model and train it using back-propagation. Everything is automatic, so, we do not need to worry about the details. The training data accuracy rises to 0.986 = 98.6%. However, true performance needs to be evaluated using test data. We can save to **history** information about the training process. The model is trained with batches of 64 images. So, to go through all the images, we need 938 rounds (the last batch has 32 images). One epoch is one round of going through all the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "annual-clone",
   "metadata": {},
   "outputs": [],
   "source": [
    "history = model.fit(x_train,y_train,epochs=150,batch_size=1024,validation_split=0.1,verbose=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "introductory-deputy",
   "metadata": {},
   "source": [
    "The following code plots the progress of training. Within the code is info for different commands."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "sustainable-memorial",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "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.style.use('bmh') # We select as a plot-style 'bmh' that is in my opinion usually the prettiest.\n",
    "burnout = 25\n",
    "epochs = range(1, len(history.history['val_mae']) + 1) # Correct x-axis values (epochs)\n",
    "plt.plot(epochs[burnout:], history.history['val_mae'][burnout:], 'r--',label='Validation accuracy') # Plot epochs vs. accuracy\n",
    "plt.plot(epochs[burnout:], history.history['mae'][burnout:], 'b--',label='Train accuracy') # Plot epochs vs. accuracy\n",
    "plt.legend()\n",
    "plt.title('Accuracy') # Add title\n",
    "plt.figure() # Show the first figure. Without this command, accuracy and loss would be drawn to the same plot.\n",
    "plt.plot(epochs[burnout:], history.history['val_loss'][burnout:], 'r--',label='Validation loss') # Plot epochs vs. loss\n",
    "plt.plot(epochs[burnout:], history.history['loss'][burnout:], 'b--',label='Train loss')\n",
    "plt.title('Loss') # Add title\n",
    "plt.show() # Show everyhting"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "polyphonic-browse",
   "metadata": {},
   "source": [
    "**Evaluate()** can be used to evaluate the model with the test data. Acccuracy with the test data is 0.052."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "central-player",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "110/110 [==============================] - 0s 248us/step - loss: 0.0066 - mae: 0.0521\n"
     ]
    }
   ],
   "source": [
    "test_loss,test_acc = model.evaluate(x_test,y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "generic-extraction",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.052059270441532135"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_acc"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "sophisticated-lincoln",
   "metadata": {},
   "source": [
    "Let's compare the performance to a linear model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "several-japan",
   "metadata": {},
   "outputs": [],
   "source": [
    "import sklearn.linear_model as sk_lm"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "visible-performer",
   "metadata": {},
   "source": [
    "We define our LinearRegression object."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "endless-secretariat",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = sk_lm.LinearRegression()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "consistent-afternoon",
   "metadata": {},
   "source": [
    "**fit()** can be used to fit the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "dress-sheriff",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LinearRegression()"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(x_train,y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "european-error",
   "metadata": {},
   "source": [
    "**coef_** -attribute has the coefficients of each variable and **intercept_** has the intercept of the linear regression model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "armed-latvia",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 5.31468126e-08, -1.05923685e-04, -3.53267974e-06, -5.74187084e-07,\n",
       "        4.89268554e-06,  3.28549174e-08, -1.02011246e-08, -4.32534321e-07,\n",
       "        3.03267883e-07,  7.33851624e-06,  5.42509020e-07,  5.14662474e-08,\n",
       "       -1.07960714e-07, -8.41952205e-06, -1.13453398e-06, -1.67169301e-05,\n",
       "        1.25340601e-05, -1.00876074e-07,  8.42772444e-08,  3.58063483e-07,\n",
       "        7.40836707e-03,  6.03105032e-04, -4.75728895e-04,  1.11622895e-06,\n",
       "       -2.38159257e-08,  1.54004100e-02,  9.43484048e-02,  1.94168470e-05,\n",
       "        2.44164138e-03,  3.21901086e-04, -3.24353490e-02, -1.33602510e-03,\n",
       "        4.25739547e-01, -3.32303633e-02,  3.77840060e-02, -2.76142065e-03,\n",
       "       -2.97726456e-04,  2.84737219e-03])"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.coef_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "presidential-narrative",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "5.530386213093688"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.intercept_"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bizarre-documentary",
   "metadata": {},
   "source": [
    "**score()** can be used to measure the coefficient of determination of the trained model. How much our variables are explaining of the variation of the predicted variable.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "quiet-juvenile",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.42032935471032107"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.score(x_test,y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "missing-asian",
   "metadata": {},
   "source": [
    "Mean absolute error."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "confused-honolulu",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import mean_absolute_error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "express-distinction",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.03547756137451534"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mean_absolute_error(y_test,model.predict(x_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "foreign-doctor",
   "metadata": {},
   "source": [
    "As expected, the linear model performs better for this data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "serious-jurisdiction",
   "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": "3.8.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}