Case Study 6.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## MSDS 7331 - Case Study 6 - Predicting the Existence of New Particles\n",
    "Daniel Crouthamel\n",
    "\n",
    "Sophia Wu\n",
    "\n",
    "Fabio Savorgnan\n",
    "\n",
    "Bo Yun"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import tensorflow as tf \n",
    "import numpy as np\n",
    "\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.preprocessing import RobustScaler\n",
    "from sklearn.model_selection import train_test_split\n",
    "import os"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Introduction\n",
    "\n",
    "The intention of this study is to predict the existence of new particles using a dense neural network."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Business Understanding\n",
    "\n",
    "We received an abrupt request from our client in the superconductors industry to help predict the existence of new particles using a dense neural network. The goal should be to maximized accuracy. The client requires a write up on the design of the network, along with information to indicate that the model was trained appropriately."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data Evaluation / Engineering\n",
    "\n",
    "## Read Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load Data\n",
    "df = pd.read_csv('data/all_train.csv.gz')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Panda's Profile"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Summarize dataset: 100%|██████████| 38/38 [01:59<00:00,  3.13s/it, Completed]\n",
      "Generate report structure: 100%|██████████| 1/1 [00:22<00:00, 22.39s/it]\n",
      "Render HTML: 100%|██████████| 1/1 [00:00<00:00,  1.59it/s]\n",
      "Export report to file: 100%|██████████| 1/1 [00:00<00:00, 504.12it/s]\n"
     ]
    }
   ],
   "source": [
    "# Load Dataset and profile it, only need to do once - so comment out.\n",
    "# from pandas_profiling import ProfileReport\n",
    "# profile = ProfileReport(df, title=\"New Particles EDA\", minimal=True)\n",
    "# profile.to_file(output_file=\"NewParticlesEDA.html\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 7000000 entries, 0 to 6999999\n",
      "Data columns (total 29 columns):\n",
      " #   Column   Dtype  \n",
      "---  ------   -----  \n",
      " 0   # label  float64\n",
      " 1   f0       float64\n",
      " 2   f1       float64\n",
      " 3   f2       float64\n",
      " 4   f3       float64\n",
      " 5   f4       float64\n",
      " 6   f5       float64\n",
      " 7   f6       float64\n",
      " 8   f7       float64\n",
      " 9   f8       float64\n",
      " 10  f9       float64\n",
      " 11  f10      float64\n",
      " 12  f11      float64\n",
      " 13  f12      float64\n",
      " 14  f13      float64\n",
      " 15  f14      float64\n",
      " 16  f15      float64\n",
      " 17  f16      float64\n",
      " 18  f17      float64\n",
      " 19  f18      float64\n",
      " 20  f19      float64\n",
      " 21  f20      float64\n",
      " 22  f21      float64\n",
      " 23  f22      float64\n",
      " 24  f23      float64\n",
      " 25  f24      float64\n",
      " 26  f25      float64\n",
      " 27  f26      float64\n",
      " 28  mass     float64\n",
      "dtypes: float64(29)\n",
      "memory usage: 1.5 GB\n"
     ]
    }
   ],
   "source": [
    "df.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "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># label</th>\n",
       "      <th>f0</th>\n",
       "      <th>f1</th>\n",
       "      <th>f2</th>\n",
       "      <th>f3</th>\n",
       "      <th>f4</th>\n",
       "      <th>f5</th>\n",
       "      <th>f6</th>\n",
       "      <th>f7</th>\n",
       "      <th>f8</th>\n",
       "      <th>...</th>\n",
       "      <th>f18</th>\n",
       "      <th>f19</th>\n",
       "      <th>f20</th>\n",
       "      <th>f21</th>\n",
       "      <th>f22</th>\n",
       "      <th>f23</th>\n",
       "      <th>f24</th>\n",
       "      <th>f25</th>\n",
       "      <th>f26</th>\n",
       "      <th>mass</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>...</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "      <td>7.000000e+06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>5.001256e-01</td>\n",
       "      <td>1.612528e-02</td>\n",
       "      <td>4.770022e-04</td>\n",
       "      <td>2.686578e-05</td>\n",
       "      <td>1.056081e-02</td>\n",
       "      <td>-1.050026e-04</td>\n",
       "      <td>2.765919e-03</td>\n",
       "      <td>1.815953e-02</td>\n",
       "      <td>2.510948e-05</td>\n",
       "      <td>4.345870e-04</td>\n",
       "      <td>...</td>\n",
       "      <td>1.164789e-02</td>\n",
       "      <td>-1.127097e-04</td>\n",
       "      <td>7.686731e-05</td>\n",
       "      <td>2.909202e-04</td>\n",
       "      <td>1.228774e-02</td>\n",
       "      <td>9.778378e-03</td>\n",
       "      <td>5.269844e-03</td>\n",
       "      <td>-1.760961e-03</td>\n",
       "      <td>1.533136e-02</td>\n",
       "      <td>1.000107e+03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>5.000000e-01</td>\n",
       "      <td>1.004417e+00</td>\n",
       "      <td>9.974864e-01</td>\n",
       "      <td>1.000080e+00</td>\n",
       "      <td>9.956003e-01</td>\n",
       "      <td>9.998670e-01</td>\n",
       "      <td>1.000957e+00</td>\n",
       "      <td>9.867746e-01</td>\n",
       "      <td>9.965867e-01</td>\n",
       "      <td>1.000007e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>1.002725e+00</td>\n",
       "      <td>1.000038e+00</td>\n",
       "      <td>1.000033e+00</td>\n",
       "      <td>1.000170e+00</td>\n",
       "      <td>1.010477e+00</td>\n",
       "      <td>1.005418e+00</td>\n",
       "      <td>1.009990e+00</td>\n",
       "      <td>9.844511e-01</td>\n",
       "      <td>9.822799e-01</td>\n",
       "      <td>3.534255e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>-1.960549e+00</td>\n",
       "      <td>-2.365355e+00</td>\n",
       "      <td>-1.732165e+00</td>\n",
       "      <td>-9.980274e+00</td>\n",
       "      <td>-1.732137e+00</td>\n",
       "      <td>-1.054221e+00</td>\n",
       "      <td>-3.034787e+00</td>\n",
       "      <td>-2.757853e+00</td>\n",
       "      <td>-1.732359e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>-1.728284e+00</td>\n",
       "      <td>-2.281867e+00</td>\n",
       "      <td>-1.731758e+00</td>\n",
       "      <td>-5.736825e-01</td>\n",
       "      <td>-3.631608e+00</td>\n",
       "      <td>-4.729473e+00</td>\n",
       "      <td>-2.062223e+01</td>\n",
       "      <td>-3.452634e+00</td>\n",
       "      <td>-2.632761e+00</td>\n",
       "      <td>5.000000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>-7.288206e-01</td>\n",
       "      <td>-7.332548e-01</td>\n",
       "      <td>-8.656704e-01</td>\n",
       "      <td>-6.092291e-01</td>\n",
       "      <td>-8.658025e-01</td>\n",
       "      <td>-1.054221e+00</td>\n",
       "      <td>-7.566092e-01</td>\n",
       "      <td>-7.014146e-01</td>\n",
       "      <td>-8.656543e-01</td>\n",
       "      <td>...</td>\n",
       "      <td>-7.423630e-01</td>\n",
       "      <td>-7.206846e-01</td>\n",
       "      <td>-8.656855e-01</td>\n",
       "      <td>-5.736825e-01</td>\n",
       "      <td>-5.417942e-01</td>\n",
       "      <td>-5.115522e-01</td>\n",
       "      <td>-3.543870e-01</td>\n",
       "      <td>-6.925097e-01</td>\n",
       "      <td>-7.943804e-01</td>\n",
       "      <td>7.500000e+02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>1.000000e+00</td>\n",
       "      <td>-3.930319e-02</td>\n",
       "      <td>8.523957e-04</td>\n",
       "      <td>3.199154e-04</td>\n",
       "      <td>1.963316e-02</td>\n",
       "      <td>-5.070131e-04</td>\n",
       "      <td>-5.983562e-03</td>\n",
       "      <td>-1.499527e-01</td>\n",
       "      <td>-1.067553e-04</td>\n",
       "      <td>1.384781e-03</td>\n",
       "      <td>...</td>\n",
       "      <td>-8.992496e-02</td>\n",
       "      <td>-6.735953e-05</td>\n",
       "      <td>-4.424527e-04</td>\n",
       "      <td>-5.736825e-01</td>\n",
       "      <td>-1.602760e-01</td>\n",
       "      <td>-3.144032e-01</td>\n",
       "      <td>-3.265228e-01</td>\n",
       "      <td>-3.570301e-01</td>\n",
       "      <td>-8.828640e-02</td>\n",
       "      <td>1.000000e+03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>1.000000e+00</td>\n",
       "      <td>6.900799e-01</td>\n",
       "      <td>7.347832e-01</td>\n",
       "      <td>8.659464e-01</td>\n",
       "      <td>6.798818e-01</td>\n",
       "      <td>8.657646e-01</td>\n",
       "      <td>8.504885e-01</td>\n",
       "      <td>7.686690e-01</td>\n",
       "      <td>7.013194e-01</td>\n",
       "      <td>8.665976e-01</td>\n",
       "      <td>...</td>\n",
       "      <td>6.423185e-01</td>\n",
       "      <td>7.204921e-01</td>\n",
       "      <td>8.659566e-01</td>\n",
       "      <td>-5.736825e-01</td>\n",
       "      <td>4.812194e-01</td>\n",
       "      <td>1.634892e-01</td>\n",
       "      <td>-2.337671e-01</td>\n",
       "      <td>4.753128e-01</td>\n",
       "      <td>7.610846e-01</td>\n",
       "      <td>1.250000e+03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>1.000000e+00</td>\n",
       "      <td>4.378282e+00</td>\n",
       "      <td>2.365287e+00</td>\n",
       "      <td>1.732370e+00</td>\n",
       "      <td>4.148023e+00</td>\n",
       "      <td>1.731978e+00</td>\n",
       "      <td>4.482618e+00</td>\n",
       "      <td>3.720345e+00</td>\n",
       "      <td>2.758590e+00</td>\n",
       "      <td>1.731450e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>5.866367e+00</td>\n",
       "      <td>2.282217e+00</td>\n",
       "      <td>1.732740e+00</td>\n",
       "      <td>1.743123e+00</td>\n",
       "      <td>7.293420e+00</td>\n",
       "      <td>9.333287e+00</td>\n",
       "      <td>1.499064e+01</td>\n",
       "      <td>5.277313e+00</td>\n",
       "      <td>4.444690e+00</td>\n",
       "      <td>1.500000e+03</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8 rows × 29 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "            # label            f0            f1            f2            f3  \\\n",
       "count  7.000000e+06  7.000000e+06  7.000000e+06  7.000000e+06  7.000000e+06   \n",
       "mean   5.001256e-01  1.612528e-02  4.770022e-04  2.686578e-05  1.056081e-02   \n",
       "std    5.000000e-01  1.004417e+00  9.974864e-01  1.000080e+00  9.956003e-01   \n",
       "min    0.000000e+00 -1.960549e+00 -2.365355e+00 -1.732165e+00 -9.980274e+00   \n",
       "25%    0.000000e+00 -7.288206e-01 -7.332548e-01 -8.656704e-01 -6.092291e-01   \n",
       "50%    1.000000e+00 -3.930319e-02  8.523957e-04  3.199154e-04  1.963316e-02   \n",
       "75%    1.000000e+00  6.900799e-01  7.347832e-01  8.659464e-01  6.798818e-01   \n",
       "max    1.000000e+00  4.378282e+00  2.365287e+00  1.732370e+00  4.148023e+00   \n",
       "\n",
       "                 f4            f5            f6            f7            f8  \\\n",
       "count  7.000000e+06  7.000000e+06  7.000000e+06  7.000000e+06  7.000000e+06   \n",
       "mean  -1.050026e-04  2.765919e-03  1.815953e-02  2.510948e-05  4.345870e-04   \n",
       "std    9.998670e-01  1.000957e+00  9.867746e-01  9.965867e-01  1.000007e+00   \n",
       "min   -1.732137e+00 -1.054221e+00 -3.034787e+00 -2.757853e+00 -1.732359e+00   \n",
       "25%   -8.658025e-01 -1.054221e+00 -7.566092e-01 -7.014146e-01 -8.656543e-01   \n",
       "50%   -5.070131e-04 -5.983562e-03 -1.499527e-01 -1.067553e-04  1.384781e-03   \n",
       "75%    8.657646e-01  8.504885e-01  7.686690e-01  7.013194e-01  8.665976e-01   \n",
       "max    1.731978e+00  4.482618e+00  3.720345e+00  2.758590e+00  1.731450e+00   \n",
       "\n",
       "       ...           f18           f19           f20           f21  \\\n",
       "count  ...  7.000000e+06  7.000000e+06  7.000000e+06  7.000000e+06   \n",
       "mean   ...  1.164789e-02 -1.127097e-04  7.686731e-05  2.909202e-04   \n",
       "std    ...  1.002725e+00  1.000038e+00  1.000033e+00  1.000170e+00   \n",
       "min    ... -1.728284e+00 -2.281867e+00 -1.731758e+00 -5.736825e-01   \n",
       "25%    ... -7.423630e-01 -7.206846e-01 -8.656855e-01 -5.736825e-01   \n",
       "50%    ... -8.992496e-02 -6.735953e-05 -4.424527e-04 -5.736825e-01   \n",
       "75%    ...  6.423185e-01  7.204921e-01  8.659566e-01 -5.736825e-01   \n",
       "max    ...  5.866367e+00  2.282217e+00  1.732740e+00  1.743123e+00   \n",
       "\n",
       "                f22           f23           f24           f25           f26  \\\n",
       "count  7.000000e+06  7.000000e+06  7.000000e+06  7.000000e+06  7.000000e+06   \n",
       "mean   1.228774e-02  9.778378e-03  5.269844e-03 -1.760961e-03  1.533136e-02   \n",
       "std    1.010477e+00  1.005418e+00  1.009990e+00  9.844511e-01  9.822799e-01   \n",
       "min   -3.631608e+00 -4.729473e+00 -2.062223e+01 -3.452634e+00 -2.632761e+00   \n",
       "25%   -5.417942e-01 -5.115522e-01 -3.543870e-01 -6.925097e-01 -7.943804e-01   \n",
       "50%   -1.602760e-01 -3.144032e-01 -3.265228e-01 -3.570301e-01 -8.828640e-02   \n",
       "75%    4.812194e-01  1.634892e-01 -2.337671e-01  4.753128e-01  7.610846e-01   \n",
       "max    7.293420e+00  9.333287e+00  1.499064e+01  5.277313e+00  4.444690e+00   \n",
       "\n",
       "               mass  \n",
       "count  7.000000e+06  \n",
       "mean   1.000107e+03  \n",
       "std    3.534255e+02  \n",
       "min    5.000000e+02  \n",
       "25%    7.500000e+02  \n",
       "50%    1.000000e+03  \n",
       "75%    1.250000e+03  \n",
       "max    1.500000e+03  \n",
       "\n",
       "[8 rows x 29 columns]"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Missing Data Evaluation\n",
    "\n",
    "Below we find that we have no missing data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "# label    0\n",
       "f0         0\n",
       "f1         0\n",
       "f2         0\n",
       "f3         0\n",
       "f4         0\n",
       "f5         0\n",
       "f6         0\n",
       "f7         0\n",
       "f8         0\n",
       "f9         0\n",
       "f10        0\n",
       "f11        0\n",
       "f12        0\n",
       "f13        0\n",
       "f14        0\n",
       "f15        0\n",
       "f16        0\n",
       "f17        0\n",
       "f18        0\n",
       "f19        0\n",
       "f20        0\n",
       "f21        0\n",
       "f22        0\n",
       "f23        0\n",
       "f24        0\n",
       "f25        0\n",
       "f26        0\n",
       "mass       0\n",
       "dtype: int64"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Find if any columns have missing values\n",
    "df.isnull().sum()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature Histograms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[<AxesSubplot:title={'center':'# label'}>,\n",
       "        <AxesSubplot:title={'center':'f0'}>,\n",
       "        <AxesSubplot:title={'center':'f1'}>,\n",
       "        <AxesSubplot:title={'center':'f2'}>,\n",
       "        <AxesSubplot:title={'center':'f3'}>],\n",
       "       [<AxesSubplot:title={'center':'f4'}>,\n",
       "        <AxesSubplot:title={'center':'f5'}>,\n",
       "        <AxesSubplot:title={'center':'f6'}>,\n",
       "        <AxesSubplot:title={'center':'f7'}>,\n",
       "        <AxesSubplot:title={'center':'f8'}>],\n",
       "       [<AxesSubplot:title={'center':'f9'}>,\n",
       "        <AxesSubplot:title={'center':'f10'}>,\n",
       "        <AxesSubplot:title={'center':'f11'}>,\n",
       "        <AxesSubplot:title={'center':'f12'}>,\n",
       "        <AxesSubplot:title={'center':'f13'}>],\n",
       "       [<AxesSubplot:title={'center':'f14'}>,\n",
       "        <AxesSubplot:title={'center':'f15'}>,\n",
       "        <AxesSubplot:title={'center':'f16'}>,\n",
       "        <AxesSubplot:title={'center':'f17'}>,\n",
       "        <AxesSubplot:title={'center':'f18'}>],\n",
       "       [<AxesSubplot:title={'center':'f19'}>,\n",
       "        <AxesSubplot:title={'center':'f20'}>,\n",
       "        <AxesSubplot:title={'center':'f21'}>,\n",
       "        <AxesSubplot:title={'center':'f22'}>,\n",
       "        <AxesSubplot:title={'center':'f23'}>],\n",
       "       [<AxesSubplot:title={'center':'f24'}>,\n",
       "        <AxesSubplot:title={'center':'f25'}>,\n",
       "        <AxesSubplot:title={'center':'f26'}>,\n",
       "        <AxesSubplot:title={'center':'mass'}>, <AxesSubplot:>]],\n",
       "      dtype=object)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1008x1008 with 30 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "df.hist(figsize=(14,14), xrot=50)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Target Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1.0    3500879\n",
       "0.0    3499121\n",
       "Name: # label, dtype: int64"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Determine if target is balanced\n",
    "\n",
    "df[\"# label\"].value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The information found in the Pandas Profile report and historgrams above reveal the following information:\n",
    "\n",
    "* 7 Million rows of data\n",
    "* 28 features and 1 target\n",
    "* No missing data\n",
    "* Our reponse, # label, is nearly balanced, close to 50%.\n",
    "* We have 4 features in which there are only two values, which we could consider treating as binary/categorical.\n",
    "  * f9\n",
    "  * f13\n",
    "  * f17\n",
    "  * f21\n",
    "* The remaing features are distrubuted normally or have a uniform distribution."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 7000000 entries, 0 to 6999999\n",
      "Data columns (total 29 columns):\n",
      " #   Column  Dtype  \n",
      "---  ------  -----  \n",
      " 0   target  float64\n",
      " 1   f0      float64\n",
      " 2   f1      float64\n",
      " 3   f2      float64\n",
      " 4   f3      float64\n",
      " 5   f4      float64\n",
      " 6   f5      float64\n",
      " 7   f6      float64\n",
      " 8   f7      float64\n",
      " 9   f8      float64\n",
      " 10  f9      float64\n",
      " 11  f10     float64\n",
      " 12  f11     float64\n",
      " 13  f12     float64\n",
      " 14  f13     float64\n",
      " 15  f14     float64\n",
      " 16  f15     float64\n",
      " 17  f16     float64\n",
      " 18  f17     float64\n",
      " 19  f18     float64\n",
      " 20  f19     float64\n",
      " 21  f20     float64\n",
      " 22  f21     float64\n",
      " 23  f22     float64\n",
      " 24  f23     float64\n",
      " 25  f24     float64\n",
      " 26  f25     float64\n",
      " 27  f26     float64\n",
      " 28  mass    float64\n",
      "dtypes: float64(29)\n",
      "memory usage: 1.5 GB\n"
     ]
    }
   ],
   "source": [
    "# Relabel target\n",
    "df.rename(columns={'# label': 'target'}, inplace=True)\n",
    "df.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Modeling Preparations\n",
    "\n",
    "The business objective asked of us is to use a neural network and optimize for accuracy. However, before doing that we will sample a small set of data and build a simple logistic regression model to see what the accuracy of the model is. Establish a baseline.\n",
    "\n",
    "## Create training and test sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_small = df.sample(frac=0.1, random_state=1).copy(deep=True)\n",
    "\n",
    "scaler = StandardScaler()\n",
    "\n",
    "X = df_small.loc[:, df_small.columns != 'target'].values\n",
    "\n",
    "y = df_small['target'].values\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(\n",
    "    X, y, test_size=0.33, random_state=30)\n",
    "\n",
    "scaler.fit(X_train)\n",
    "\n",
    "X_train = scaler.transform(X_train)\n",
    "X_test = scaler.transform(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Logistic Regression for baseline model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8370735930735931\n"
     ]
    }
   ],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "\n",
    "lr = LogisticRegression(solver='lbfgs', max_iter=1000)\n",
    "lr.fit(X_train, y_train)\n",
    "print(lr.score(X_test, y_test))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Gridsearch LR\n",
    "- The best accuracy score observed is 83.6%. \n",
    "- Best Hyperparameters: {'C': 0.001, 'penalty': 'none', 'solver': 'newton-cg'}\n",
    "- We will compare this result to the Neural Network result below."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Score: 0.8366814285714286\n",
      "Best Hyperparameters: {'C': 0.001, 'penalty': 'none', 'solver': 'newton-cg'}\n"
     ]
    }
   ],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")  # ignoring warnings\n",
    "\n",
    "# Model definition\n",
    "model = LogisticRegression()\n",
    "\n",
    "# Define search space\n",
    "space = dict()\n",
    "space['solver'] = ['newton-cg', 'lbfgs', 'liblinear']\n",
    "space['penalty'] = ['none', 'l1', 'l2', 'elasticnet']\n",
    "space['C'] = [1e-3, 1e-2, 1e-1, 1, 10, 100]\n",
    "\n",
    "# Define search\n",
    "search = GridSearchCV(model, space, scoring='accuracy', n_jobs=-1, cv=5)\n",
    "\n",
    "# Execute search\n",
    "result = search.fit(X, y)\n",
    "\n",
    "# Summarize result\n",
    "print('Best Score: %s' % result.best_score_)\n",
    "print('Best Hyperparameters: %s' % result.best_params_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Gridsearch LR feature importance \n",
    "- Feature #6 and Feature #27 are the most important variables in the LR model. \n",
    "- Even though we do not have a domain expertise on these important features, we will communicate these findings to our client for their reference. \n",
    "- Depending on how much the interpretability of the model affects our client's understanding of the business performance, this information can be useful. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Feature: 0, Score: 0.51011\n",
      "Feature: 1, Score: -0.00452\n",
      "Feature: 2, Score: -0.00040\n",
      "Feature: 3, Score: 0.46035\n",
      "Feature: 4, Score: 0.00083\n",
      "Feature: 5, Score: -0.02481\n",
      "Feature: 6, Score: 1.48715\n",
      "Feature: 7, Score: 0.00461\n",
      "Feature: 8, Score: 0.00445\n",
      "Feature: 9, Score: 0.09910\n",
      "Feature: 10, Score: 0.45446\n",
      "Feature: 11, Score: 0.01179\n",
      "Feature: 12, Score: 0.00235\n",
      "Feature: 13, Score: 0.10771\n",
      "Feature: 14, Score: 0.19335\n",
      "Feature: 15, Score: 0.00393\n",
      "Feature: 16, Score: 0.00455\n",
      "Feature: 17, Score: 0.06070\n",
      "Feature: 18, Score: 0.10612\n",
      "Feature: 19, Score: -0.00532\n",
      "Feature: 20, Score: 0.00241\n",
      "Feature: 21, Score: 0.03466\n",
      "Feature: 22, Score: -0.03248\n",
      "Feature: 23, Score: -0.11218\n",
      "Feature: 24, Score: -0.00915\n",
      "Feature: 25, Score: 0.07303\n",
      "Feature: 26, Score: 0.35706\n",
      "Feature: 27, Score: -0.97617\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# get importance\n",
    "importance = lr.coef_[0]\n",
    "\n",
    "# summarize feature importance\n",
    "for i,v in enumerate(importance):\n",
    "\tprint('Feature: %0d, Score: %.5f' % (i,v))\n",
    "\n",
    "# plot feature importance\n",
    "plt.bar([x for x in range(len(importance))], importance)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Model Building & Evaluation\n",
    "\n",
    "Below we create a grid style apporach in finding the best parameters for the neural network. We'll attempt to determine an optimal number of hidden layers and the number of nodes in each layer. Note, we'll use a dropout layer for every hidden layer used. The final layer at the end will use a sigmoid activation function, since we are dealing with a binary classification problem.\n",
    "\n",
    "Additionally, we'll use the Adam optimizer in SGD and BinaryCrossEntropy for the loss function. Early stopping is used to prevent overfitting, and will stop training once there is no longer improvement in the validation set. This is defined with the patience parameter (set at 3), along with an min_delta parameter (set at 0.0002). Early stopping will occur if we don't see any improvement greater than .0002 in 3 tries.\n",
    "\n",
    "## Use full data set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "scaler = StandardScaler()\n",
    "\n",
    "X = df.loc[:, df.columns != 'target'].values\n",
    "\n",
    "y = df['target'].values\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(\n",
    "    X, y, test_size=0.33, random_state=30)\n",
    "\n",
    "scaler.fit(X_train)\n",
    "\n",
    "X_train = scaler.transform(X_train)\n",
    "X_test = scaler.transform(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Gridsearch NN\n",
    "\n",
    "Note that the below takes over an hour to run. The best esitmator and model is saved off if needed. A tensorboard log is also saved (see callbacks)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.3915 - accuracy: 0.8223 - val_loss: 0.3661 - val_accuracy: 0.8363\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 10s 3ms/step - loss: 0.3654 - accuracy: 0.8367 - val_loss: 0.3652 - val_accuracy: 0.8367\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 10s 3ms/step - loss: 0.3653 - accuracy: 0.8367 - val_loss: 0.3652 - val_accuracy: 0.8370\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 10s 3ms/step - loss: 0.3653 - accuracy: 0.8367 - val_loss: 0.3652 - val_accuracy: 0.8369\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 10s 3ms/step - loss: 0.3653 - accuracy: 0.8368 - val_loss: 0.3652 - val_accuracy: 0.8370\n",
      "48855/48855 [==============================] - 94s 2ms/step - loss: 0.3657 - accuracy: 0.8367\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.4186 - accuracy: 0.8056 - val_loss: 0.3668 - val_accuracy: 0.8362\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 9s 3ms/step - loss: 0.3660 - accuracy: 0.8364 - val_loss: 0.3652 - val_accuracy: 0.8369\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 10s 3ms/step - loss: 0.3657 - accuracy: 0.8366 - val_loss: 0.3652 - val_accuracy: 0.8366\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 11s 4ms/step - loss: 0.3657 - accuracy: 0.8365 - val_loss: 0.3653 - val_accuracy: 0.8365\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 10s 3ms/step - loss: 0.3657 - accuracy: 0.8365 - val_loss: 0.3652 - val_accuracy: 0.8369\n",
      "48855/48855 [==============================] - 93s 2ms/step - loss: 0.3649 - accuracy: 0.8371\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.3813 - accuracy: 0.8282 - val_loss: 0.3653 - val_accuracy: 0.8367\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 9s 3ms/step - loss: 0.3653 - accuracy: 0.8368 - val_loss: 0.3652 - val_accuracy: 0.8365\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 10s 3ms/step - loss: 0.3653 - accuracy: 0.8368 - val_loss: 0.3652 - val_accuracy: 0.8370\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 9s 3ms/step - loss: 0.3653 - accuracy: 0.8367 - val_loss: 0.3653 - val_accuracy: 0.8367\n",
      "48855/48855 [==============================] - 95s 2ms/step - loss: 0.3657 - accuracy: 0.8364\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.3208 - accuracy: 0.8498 - val_loss: 0.2935 - val_accuracy: 0.8629\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2985 - accuracy: 0.8611 - val_loss: 0.2871 - val_accuracy: 0.8676\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2947 - accuracy: 0.8642 - val_loss: 0.2843 - val_accuracy: 0.8697\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2929 - accuracy: 0.8654 - val_loss: 0.2835 - val_accuracy: 0.8706\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2919 - accuracy: 0.8661 - val_loss: 0.2840 - val_accuracy: 0.8715\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2910 - accuracy: 0.8668 - val_loss: 0.2812 - val_accuracy: 0.8717\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2901 - accuracy: 0.8675 - val_loss: 0.2813 - val_accuracy: 0.8729\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2896 - accuracy: 0.8677 - val_loss: 0.2806 - val_accuracy: 0.8733\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2891 - accuracy: 0.8680 - val_loss: 0.2800 - val_accuracy: 0.8734\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2887 - accuracy: 0.8684 - val_loss: 0.2802 - val_accuracy: 0.8738\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2884 - accuracy: 0.8684 - val_loss: 0.2800 - val_accuracy: 0.8737\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2882 - accuracy: 0.8687 - val_loss: 0.2800 - val_accuracy: 0.8739\n",
      "48855/48855 [==============================] - 97s 2ms/step - loss: 0.2804 - accuracy: 0.8736\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 18s 6ms/step - loss: 0.3207 - accuracy: 0.8497 - val_loss: 0.2911 - val_accuracy: 0.8653\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2969 - accuracy: 0.8630 - val_loss: 0.2849 - val_accuracy: 0.8701\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2932 - accuracy: 0.8655 - val_loss: 0.2828 - val_accuracy: 0.8714\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2919 - accuracy: 0.8664 - val_loss: 0.2820 - val_accuracy: 0.8719\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2909 - accuracy: 0.8671 - val_loss: 0.2810 - val_accuracy: 0.8725\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2902 - accuracy: 0.8675 - val_loss: 0.2808 - val_accuracy: 0.8729\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2897 - accuracy: 0.8677 - val_loss: 0.2799 - val_accuracy: 0.8725\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2893 - accuracy: 0.8681 - val_loss: 0.2800 - val_accuracy: 0.8729\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2890 - accuracy: 0.8683 - val_loss: 0.2798 - val_accuracy: 0.8738\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2887 - accuracy: 0.8684 - val_loss: 0.2804 - val_accuracy: 0.8737\n",
      "48855/48855 [==============================] - 105s 2ms/step - loss: 0.2804 - accuracy: 0.8737\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.3213 - accuracy: 0.8497 - val_loss: 0.2907 - val_accuracy: 0.8650\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2971 - accuracy: 0.8626 - val_loss: 0.2853 - val_accuracy: 0.8691\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 11s 4ms/step - loss: 0.2929 - accuracy: 0.8655 - val_loss: 0.2832 - val_accuracy: 0.8711\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2911 - accuracy: 0.8667 - val_loss: 0.2816 - val_accuracy: 0.8720\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2898 - accuracy: 0.8674 - val_loss: 0.2807 - val_accuracy: 0.8728\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2891 - accuracy: 0.8681 - val_loss: 0.2794 - val_accuracy: 0.8732\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2884 - accuracy: 0.8685 - val_loss: 0.2799 - val_accuracy: 0.8737\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2879 - accuracy: 0.8687 - val_loss: 0.2793 - val_accuracy: 0.8737\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2875 - accuracy: 0.8689 - val_loss: 0.2788 - val_accuracy: 0.8740\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2874 - accuracy: 0.8689 - val_loss: 0.2778 - val_accuracy: 0.8744\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2870 - accuracy: 0.8693 - val_loss: 0.2785 - val_accuracy: 0.8741\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2869 - accuracy: 0.8694 - val_loss: 0.2780 - val_accuracy: 0.8744\n",
      "Epoch 13/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2868 - accuracy: 0.8694 - val_loss: 0.2784 - val_accuracy: 0.8743\n",
      "48855/48855 [==============================] - 103s 2ms/step - loss: 0.2791 - accuracy: 0.8737\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 19s 6ms/step - loss: 0.2901 - accuracy: 0.8666 - val_loss: 0.2756 - val_accuracy: 0.8745\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2753 - accuracy: 0.8752 - val_loss: 0.2686 - val_accuracy: 0.8789\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2717 - accuracy: 0.8771 - val_loss: 0.2659 - val_accuracy: 0.8804\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2697 - accuracy: 0.8785 - val_loss: 0.2649 - val_accuracy: 0.8807\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2684 - accuracy: 0.8792 - val_loss: 0.2644 - val_accuracy: 0.8811\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2675 - accuracy: 0.8796 - val_loss: 0.2633 - val_accuracy: 0.8815\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2666 - accuracy: 0.8801 - val_loss: 0.2625 - val_accuracy: 0.8820\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2659 - accuracy: 0.8804 - val_loss: 0.2624 - val_accuracy: 0.8822\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2653 - accuracy: 0.8809 - val_loss: 0.2621 - val_accuracy: 0.8824\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2648 - accuracy: 0.8812 - val_loss: 0.2629 - val_accuracy: 0.8824\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2643 - accuracy: 0.8813 - val_loss: 0.2620 - val_accuracy: 0.8827\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2640 - accuracy: 0.8815 - val_loss: 0.2613 - val_accuracy: 0.8830\n",
      "Epoch 13/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2636 - accuracy: 0.8818 - val_loss: 0.2615 - val_accuracy: 0.8829\n",
      "Epoch 14/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2632 - accuracy: 0.8820 - val_loss: 0.2609 - val_accuracy: 0.8833\n",
      "Epoch 15/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2628 - accuracy: 0.8823 - val_loss: 0.2610 - val_accuracy: 0.8831\n",
      "Epoch 16/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2626 - accuracy: 0.8822 - val_loss: 0.2606 - val_accuracy: 0.8831\n",
      "Epoch 17/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2622 - accuracy: 0.8825 - val_loss: 0.2605 - val_accuracy: 0.8832\n",
      "Epoch 18/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2620 - accuracy: 0.8826 - val_loss: 0.2603 - val_accuracy: 0.8834\n",
      "Epoch 19/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2617 - accuracy: 0.8828 - val_loss: 0.2613 - val_accuracy: 0.8829\n",
      "Epoch 20/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2614 - accuracy: 0.8830 - val_loss: 0.2604 - val_accuracy: 0.8834\n",
      "Epoch 21/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2612 - accuracy: 0.8831 - val_loss: 0.2612 - val_accuracy: 0.8833\n",
      "48855/48855 [==============================] - 114s 2ms/step - loss: 0.2610 - accuracy: 0.8831\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 21s 6ms/step - loss: 0.2903 - accuracy: 0.8662 - val_loss: 0.2742 - val_accuracy: 0.8757\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2753 - accuracy: 0.8752 - val_loss: 0.2691 - val_accuracy: 0.8786\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2716 - accuracy: 0.8771 - val_loss: 0.2668 - val_accuracy: 0.8800\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2698 - accuracy: 0.8783 - val_loss: 0.2646 - val_accuracy: 0.8809\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2685 - accuracy: 0.8791 - val_loss: 0.2638 - val_accuracy: 0.8814\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2674 - accuracy: 0.8796 - val_loss: 0.2632 - val_accuracy: 0.8819\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2666 - accuracy: 0.8800 - val_loss: 0.2625 - val_accuracy: 0.8821\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2659 - accuracy: 0.8804 - val_loss: 0.2627 - val_accuracy: 0.8822\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2653 - accuracy: 0.8807 - val_loss: 0.2616 - val_accuracy: 0.8826\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2647 - accuracy: 0.8811 - val_loss: 0.2618 - val_accuracy: 0.8826\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2643 - accuracy: 0.8813 - val_loss: 0.2617 - val_accuracy: 0.8826\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2640 - accuracy: 0.8813 - val_loss: 0.2614 - val_accuracy: 0.8826\n",
      "48855/48855 [==============================] - 119s 2ms/step - loss: 0.2614 - accuracy: 0.8827\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 19s 6ms/step - loss: 0.2898 - accuracy: 0.8665 - val_loss: 0.2735 - val_accuracy: 0.8760\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2746 - accuracy: 0.8754 - val_loss: 0.2680 - val_accuracy: 0.8791\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2713 - accuracy: 0.8773 - val_loss: 0.2666 - val_accuracy: 0.8800\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2694 - accuracy: 0.8782 - val_loss: 0.2651 - val_accuracy: 0.8805\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2681 - accuracy: 0.8790 - val_loss: 0.2648 - val_accuracy: 0.8809\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2671 - accuracy: 0.8797 - val_loss: 0.2643 - val_accuracy: 0.8814\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2663 - accuracy: 0.8801 - val_loss: 0.2628 - val_accuracy: 0.8822\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2657 - accuracy: 0.8804 - val_loss: 0.2625 - val_accuracy: 0.8821\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2652 - accuracy: 0.8806 - val_loss: 0.2623 - val_accuracy: 0.8824\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2647 - accuracy: 0.8810 - val_loss: 0.2618 - val_accuracy: 0.8826\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2640 - accuracy: 0.8812 - val_loss: 0.2613 - val_accuracy: 0.8828\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2637 - accuracy: 0.8815 - val_loss: 0.2623 - val_accuracy: 0.8825\n",
      "Epoch 13/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2633 - accuracy: 0.8817 - val_loss: 0.2610 - val_accuracy: 0.8831\n",
      "Epoch 14/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2628 - accuracy: 0.8819 - val_loss: 0.2609 - val_accuracy: 0.8829\n",
      "Epoch 15/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2625 - accuracy: 0.8822 - val_loss: 0.2608 - val_accuracy: 0.8831\n",
      "Epoch 16/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2622 - accuracy: 0.8822 - val_loss: 0.2607 - val_accuracy: 0.8834\n",
      "Epoch 17/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2619 - accuracy: 0.8824 - val_loss: 0.2604 - val_accuracy: 0.8833\n",
      "Epoch 18/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2616 - accuracy: 0.8826 - val_loss: 0.2605 - val_accuracy: 0.8833\n",
      "Epoch 19/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2613 - accuracy: 0.8826 - val_loss: 0.2613 - val_accuracy: 0.8831\n",
      "Epoch 20/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2612 - accuracy: 0.8829 - val_loss: 0.2606 - val_accuracy: 0.8836\n",
      "48855/48855 [==============================] - 106s 2ms/step - loss: 0.2611 - accuracy: 0.8834\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 17s 5ms/step - loss: 0.3130 - accuracy: 0.8542 - val_loss: 0.2870 - val_accuracy: 0.8677\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2909 - accuracy: 0.8658 - val_loss: 0.2803 - val_accuracy: 0.8728\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2869 - accuracy: 0.8686 - val_loss: 0.2775 - val_accuracy: 0.8745\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2848 - accuracy: 0.8702 - val_loss: 0.2761 - val_accuracy: 0.8754\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2836 - accuracy: 0.8709 - val_loss: 0.2751 - val_accuracy: 0.8760\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2826 - accuracy: 0.8717 - val_loss: 0.2742 - val_accuracy: 0.8761\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2821 - accuracy: 0.8721 - val_loss: 0.2734 - val_accuracy: 0.8767\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2818 - accuracy: 0.8721 - val_loss: 0.2731 - val_accuracy: 0.8770\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2813 - accuracy: 0.8725 - val_loss: 0.2732 - val_accuracy: 0.8771\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2809 - accuracy: 0.8726 - val_loss: 0.2730 - val_accuracy: 0.8770\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2808 - accuracy: 0.8727 - val_loss: 0.2729 - val_accuracy: 0.8775\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2805 - accuracy: 0.8729 - val_loss: 0.2728 - val_accuracy: 0.8775\n",
      "Epoch 13/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2803 - accuracy: 0.8729 - val_loss: 0.2733 - val_accuracy: 0.8773\n",
      "Epoch 14/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2802 - accuracy: 0.8731 - val_loss: 0.2726 - val_accuracy: 0.8776\n",
      "Epoch 15/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2799 - accuracy: 0.8731 - val_loss: 0.2722 - val_accuracy: 0.8779\n",
      "Epoch 16/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2799 - accuracy: 0.8732 - val_loss: 0.2721 - val_accuracy: 0.8778\n",
      "Epoch 17/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2796 - accuracy: 0.8733 - val_loss: 0.2720 - val_accuracy: 0.8780\n",
      "Epoch 18/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2794 - accuracy: 0.8733 - val_loss: 0.2728 - val_accuracy: 0.8780\n",
      "48855/48855 [==============================] - 98s 2ms/step - loss: 0.2731 - accuracy: 0.8777\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 19s 6ms/step - loss: 0.3117 - accuracy: 0.8553 - val_loss: 0.2859 - val_accuracy: 0.8683\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2904 - accuracy: 0.8668 - val_loss: 0.2800 - val_accuracy: 0.8727\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2863 - accuracy: 0.8694 - val_loss: 0.2769 - val_accuracy: 0.8746\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2843 - accuracy: 0.8707 - val_loss: 0.2754 - val_accuracy: 0.8753\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2832 - accuracy: 0.8712 - val_loss: 0.2748 - val_accuracy: 0.8758\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2824 - accuracy: 0.8718 - val_loss: 0.2746 - val_accuracy: 0.8762\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2816 - accuracy: 0.8722 - val_loss: 0.2737 - val_accuracy: 0.8765\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2812 - accuracy: 0.8723 - val_loss: 0.2729 - val_accuracy: 0.8770\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 11s 4ms/step - loss: 0.2808 - accuracy: 0.8726 - val_loss: 0.2727 - val_accuracy: 0.8775\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2803 - accuracy: 0.8729 - val_loss: 0.2724 - val_accuracy: 0.8774\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2801 - accuracy: 0.8729 - val_loss: 0.2725 - val_accuracy: 0.8778\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2798 - accuracy: 0.8732 - val_loss: 0.2730 - val_accuracy: 0.8775\n",
      "Epoch 13/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2797 - accuracy: 0.8733 - val_loss: 0.2718 - val_accuracy: 0.8778\n",
      "Epoch 14/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2795 - accuracy: 0.8734 - val_loss: 0.2720 - val_accuracy: 0.8778\n",
      "Epoch 15/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2793 - accuracy: 0.8735 - val_loss: 0.2721 - val_accuracy: 0.8776\n",
      "Epoch 16/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2791 - accuracy: 0.8737 - val_loss: 0.2711 - val_accuracy: 0.8784\n",
      "Epoch 17/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2790 - accuracy: 0.8737 - val_loss: 0.2723 - val_accuracy: 0.8781\n",
      "Epoch 18/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2788 - accuracy: 0.8737 - val_loss: 0.2713 - val_accuracy: 0.8783\n",
      "Epoch 19/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2787 - accuracy: 0.8740 - val_loss: 0.2724 - val_accuracy: 0.8784\n",
      "48855/48855 [==============================] - 112s 2ms/step - loss: 0.2726 - accuracy: 0.8780\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 19s 6ms/step - loss: 0.3121 - accuracy: 0.8548 - val_loss: 0.2863 - val_accuracy: 0.8686\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2904 - accuracy: 0.8668 - val_loss: 0.2801 - val_accuracy: 0.8726\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2861 - accuracy: 0.8696 - val_loss: 0.2768 - val_accuracy: 0.8745\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2839 - accuracy: 0.8709 - val_loss: 0.2751 - val_accuracy: 0.8754\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2826 - accuracy: 0.8716 - val_loss: 0.2746 - val_accuracy: 0.8760\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 12s 4ms/step - loss: 0.2817 - accuracy: 0.8722 - val_loss: 0.2737 - val_accuracy: 0.8768\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2812 - accuracy: 0.8725 - val_loss: 0.2726 - val_accuracy: 0.8771\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2808 - accuracy: 0.8726 - val_loss: 0.2735 - val_accuracy: 0.8768\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2805 - accuracy: 0.8728 - val_loss: 0.2726 - val_accuracy: 0.8773\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2803 - accuracy: 0.8729 - val_loss: 0.2729 - val_accuracy: 0.8775\n",
      "48855/48855 [==============================] - 90s 2ms/step - loss: 0.2737 - accuracy: 0.8770\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2909 - accuracy: 0.8660 - val_loss: 0.2738 - val_accuracy: 0.8757\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2755 - accuracy: 0.8753 - val_loss: 0.2685 - val_accuracy: 0.8787\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2718 - accuracy: 0.8772 - val_loss: 0.2667 - val_accuracy: 0.8798\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2699 - accuracy: 0.8782 - val_loss: 0.2648 - val_accuracy: 0.8812\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2687 - accuracy: 0.8789 - val_loss: 0.2646 - val_accuracy: 0.8810\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2677 - accuracy: 0.8794 - val_loss: 0.2641 - val_accuracy: 0.8811\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2668 - accuracy: 0.8800 - val_loss: 0.2626 - val_accuracy: 0.8821\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 13s 4ms/step - loss: 0.2661 - accuracy: 0.8804 - val_loss: 0.2627 - val_accuracy: 0.8822\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2655 - accuracy: 0.8806 - val_loss: 0.2620 - val_accuracy: 0.8826\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2651 - accuracy: 0.8810 - val_loss: 0.2621 - val_accuracy: 0.8825\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2646 - accuracy: 0.8813 - val_loss: 0.2612 - val_accuracy: 0.8827\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2642 - accuracy: 0.8813 - val_loss: 0.2614 - val_accuracy: 0.8830\n",
      "Epoch 13/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2637 - accuracy: 0.8816 - val_loss: 0.2612 - val_accuracy: 0.8831\n",
      "Epoch 14/500\n",
      "3127/3127 [==============================] - 14s 4ms/step - loss: 0.2634 - accuracy: 0.8819 - val_loss: 0.2607 - val_accuracy: 0.8834\n",
      "Epoch 15/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2629 - accuracy: 0.8822 - val_loss: 0.2601 - val_accuracy: 0.8835\n",
      "Epoch 16/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2627 - accuracy: 0.8823 - val_loss: 0.2608 - val_accuracy: 0.8833\n",
      "Epoch 17/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2624 - accuracy: 0.8824 - val_loss: 0.2605 - val_accuracy: 0.8833\n",
      "Epoch 18/500\n",
      "3127/3127 [==============================] - 14s 5ms/step - loss: 0.2621 - accuracy: 0.8827 - val_loss: 0.2602 - val_accuracy: 0.8836\n",
      "48855/48855 [==============================] - 104s 2ms/step - loss: 0.2602 - accuracy: 0.8833\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 19s 6ms/step - loss: 0.2909 - accuracy: 0.8660 - val_loss: 0.2753 - val_accuracy: 0.8753\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2752 - accuracy: 0.8751 - val_loss: 0.2685 - val_accuracy: 0.8788\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2718 - accuracy: 0.8772 - val_loss: 0.2665 - val_accuracy: 0.8801\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2699 - accuracy: 0.8782 - val_loss: 0.2652 - val_accuracy: 0.8806\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2686 - accuracy: 0.8790 - val_loss: 0.2642 - val_accuracy: 0.8812\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2677 - accuracy: 0.8796 - val_loss: 0.2641 - val_accuracy: 0.8814\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2669 - accuracy: 0.8799 - val_loss: 0.2634 - val_accuracy: 0.8819\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2662 - accuracy: 0.8804 - val_loss: 0.2629 - val_accuracy: 0.8821\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2657 - accuracy: 0.8805 - val_loss: 0.2621 - val_accuracy: 0.8824\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2652 - accuracy: 0.8809 - val_loss: 0.2619 - val_accuracy: 0.8823\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2648 - accuracy: 0.8811 - val_loss: 0.2620 - val_accuracy: 0.8825\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2643 - accuracy: 0.8813 - val_loss: 0.2617 - val_accuracy: 0.8827\n",
      "Epoch 13/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2639 - accuracy: 0.8815 - val_loss: 0.2618 - val_accuracy: 0.8823\n",
      "48855/48855 [==============================] - 114s 2ms/step - loss: 0.2619 - accuracy: 0.8823\n",
      "Epoch 1/500\n",
      "3127/3127 [==============================] - 19s 6ms/step - loss: 0.2902 - accuracy: 0.8662 - val_loss: 0.2730 - val_accuracy: 0.8759\n",
      "Epoch 2/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2751 - accuracy: 0.8751 - val_loss: 0.2690 - val_accuracy: 0.8785\n",
      "Epoch 3/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2717 - accuracy: 0.8770 - val_loss: 0.2667 - val_accuracy: 0.8798\n",
      "Epoch 4/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2698 - accuracy: 0.8781 - val_loss: 0.2664 - val_accuracy: 0.8798\n",
      "Epoch 5/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2685 - accuracy: 0.8790 - val_loss: 0.2641 - val_accuracy: 0.8814\n",
      "Epoch 6/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2674 - accuracy: 0.8793 - val_loss: 0.2635 - val_accuracy: 0.8815\n",
      "Epoch 7/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2668 - accuracy: 0.8798 - val_loss: 0.2629 - val_accuracy: 0.8820\n",
      "Epoch 8/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2661 - accuracy: 0.8801 - val_loss: 0.2628 - val_accuracy: 0.8819\n",
      "Epoch 9/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2655 - accuracy: 0.8805 - val_loss: 0.2628 - val_accuracy: 0.8822\n",
      "Epoch 10/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2649 - accuracy: 0.8808 - val_loss: 0.2622 - val_accuracy: 0.8824\n",
      "Epoch 11/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2645 - accuracy: 0.8811 - val_loss: 0.2623 - val_accuracy: 0.8827\n",
      "Epoch 12/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2641 - accuracy: 0.8812 - val_loss: 0.2615 - val_accuracy: 0.8825\n",
      "Epoch 13/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2636 - accuracy: 0.8816 - val_loss: 0.2613 - val_accuracy: 0.8828\n",
      "Epoch 14/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2633 - accuracy: 0.8817 - val_loss: 0.2618 - val_accuracy: 0.8826\n",
      "Epoch 15/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2630 - accuracy: 0.8818 - val_loss: 0.2607 - val_accuracy: 0.8829\n",
      "Epoch 16/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2626 - accuracy: 0.8820 - val_loss: 0.2614 - val_accuracy: 0.8828\n",
      "Epoch 17/500\n",
      "3127/3127 [==============================] - 16s 5ms/step - loss: 0.2623 - accuracy: 0.8823 - val_loss: 0.2611 - val_accuracy: 0.8829\n",
      "Epoch 18/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2622 - accuracy: 0.8823 - val_loss: 0.2603 - val_accuracy: 0.8834\n",
      "Epoch 19/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2620 - accuracy: 0.8824 - val_loss: 0.2607 - val_accuracy: 0.8835\n",
      "Epoch 20/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2617 - accuracy: 0.8824 - val_loss: 0.2603 - val_accuracy: 0.8834\n",
      "Epoch 21/500\n",
      "3127/3127 [==============================] - 15s 5ms/step - loss: 0.2614 - accuracy: 0.8829 - val_loss: 0.2604 - val_accuracy: 0.8834\n",
      "48855/48855 [==============================] - 120s 2ms/step - loss: 0.2610 - accuracy: 0.8832\n",
      "Epoch 1/500\n",
      "4690/4690 [==============================] - 21s 4ms/step - loss: 0.2864 - accuracy: 0.8685 - val_loss: 0.2707 - val_accuracy: 0.8775\n",
      "Epoch 2/500\n",
      "4690/4690 [==============================] - 18s 4ms/step - loss: 0.2728 - accuracy: 0.8766 - val_loss: 0.2663 - val_accuracy: 0.8799\n",
      "Epoch 3/500\n",
      "4690/4690 [==============================] - 18s 4ms/step - loss: 0.2698 - accuracy: 0.8783 - val_loss: 0.2650 - val_accuracy: 0.8809\n",
      "Epoch 4/500\n",
      "4690/4690 [==============================] - 18s 4ms/step - loss: 0.2682 - accuracy: 0.8792 - val_loss: 0.2632 - val_accuracy: 0.8817\n",
      "Epoch 5/500\n",
      "4690/4690 [==============================] - 17s 4ms/step - loss: 0.2671 - accuracy: 0.8798 - val_loss: 0.2627 - val_accuracy: 0.8821\n",
      "Epoch 6/500\n",
      "4690/4690 [==============================] - 17s 4ms/step - loss: 0.2663 - accuracy: 0.8802 - val_loss: 0.2621 - val_accuracy: 0.8822\n",
      "Epoch 7/500\n",
      "4690/4690 [==============================] - 17s 4ms/step - loss: 0.2656 - accuracy: 0.8806 - val_loss: 0.2620 - val_accuracy: 0.8827\n",
      "Epoch 8/500\n",
      "4690/4690 [==============================] - 17s 4ms/step - loss: 0.2650 - accuracy: 0.8809 - val_loss: 0.2617 - val_accuracy: 0.8824\n",
      "Epoch 9/500\n",
      "4690/4690 [==============================] - 18s 4ms/step - loss: 0.2645 - accuracy: 0.8812 - val_loss: 0.2607 - val_accuracy: 0.8830\n",
      "Epoch 10/500\n",
      "4690/4690 [==============================] - 19s 4ms/step - loss: 0.2642 - accuracy: 0.8814 - val_loss: 0.2608 - val_accuracy: 0.8831\n",
      "Epoch 11/500\n",
      "4690/4690 [==============================] - 18s 4ms/step - loss: 0.2637 - accuracy: 0.8815 - val_loss: 0.2611 - val_accuracy: 0.8829\n",
      "Epoch 12/500\n",
      "4690/4690 [==============================] - 18s 4ms/step - loss: 0.2634 - accuracy: 0.8817 - val_loss: 0.2608 - val_accuracy: 0.8830\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "RandomizedSearchCV(cv=3,\n",
       "                   estimator=<tensorflow.python.keras.wrappers.scikit_learn.KerasRegressor object at 0x0000020AC1592B20>,\n",
       "                   n_iter=5,\n",
       "                   param_distributions={'n_hidden': [0, 1, 2, 3],\n",
       "                                        'n_neurons': array([  1,   2,   3,   4,   5,   6,   7,   8,   9,  10,  11,  12,  13,\n",
       "        14,  15,  16,  17,  18,  19,  20,  21,  22,  23,  24,  25,  26,\n",
       "        27,  28,  29,  30,  31,  32,  33,  34,  35,  36,  37,  38,  39,\n",
       "        40,  41,  42,  43,  44,  45,  46,  47,  48,  49,  50,  51,  52,\n",
       "        53,  54,  55,  5...\n",
       "       404, 405, 406, 407, 408, 409, 410, 411, 412, 413, 414, 415, 416,\n",
       "       417, 418, 419, 420, 421, 422, 423, 424, 425, 426, 427, 428, 429,\n",
       "       430, 431, 432, 433, 434, 435, 436, 437, 438, 439, 440, 441, 442,\n",
       "       443, 444, 445, 446, 447, 448, 449, 450, 451, 452, 453, 454, 455,\n",
       "       456, 457, 458, 459, 460, 461, 462, 463, 464, 465, 466, 467, 468,\n",
       "       469, 470, 471, 472, 473, 474, 475, 476, 477, 478, 479, 480, 481,\n",
       "       482, 483, 484, 485, 486, 487, 488, 489, 490, 491, 492, 493, 494,\n",
       "       495, 496, 497, 498, 499])})"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from tensorflow.keras.callbacks import EarlyStopping\n",
    "\n",
    "root_logdir = os.path.join(os.curdir, \"my_logs\")\n",
    "\n",
    "def get_run_logdir():\n",
    "    import time\n",
    "    run_id = time.strftime(\"run_%Y_%m_%d-%H_%M_%S\")\n",
    "    return os.path.join(root_logdir, run_id)\n",
    "\n",
    "run_logdir = get_run_logdir()\n",
    "\n",
    "# # Print current path\n",
    "# print(os.getcwd())\n",
    "# print(run_logdir)\n",
    "\n",
    "tensorboard_cb = tf.keras.callbacks.TensorBoard(run_logdir)\n",
    "\n",
    "def build_model(n_hidden=1, n_neurons=30, learning_rate=3e-3, input_shape=[28]):\n",
    "    model = tf.keras.models.Sequential()\n",
    "    model.add(tf.keras.layers.InputLayer(input_shape=input_shape))\n",
    "    for layer in range(n_hidden):\n",
    "        model.add(tf.keras.layers.Dense(n_neurons, activation=\"relu\"))\n",
    "        model.add(tf.keras.layers.Dropout(.2, input_shape=(2,)))\n",
    "    model.add(tf.keras.layers.Dense(1, activation=\"sigmoid\"))\n",
    "    optimizer = tf.keras.optimizers.Adam()\n",
    "    loss = tf.keras.losses.BinaryCrossentropy()\n",
    "    model.compile(loss=loss, optimizer=optimizer, metrics=['accuracy'])\n",
    "    return model\n",
    "\n",
    "keras_reg = tf.keras.wrappers.scikit_learn.KerasRegressor(build_model)\n",
    "\n",
    "from sklearn.model_selection import RandomizedSearchCV\n",
    "\n",
    "param_distribs = {\n",
    "    \"n_hidden\": [0, 1, 2, 3],\n",
    "    \"n_neurons\": np.arange(1, 500)\n",
    "}\n",
    "\n",
    "safety = EarlyStopping(monitor='val_loss', patience=3, min_delta=2e-4)\n",
    "\n",
    "rnd_search_cv = RandomizedSearchCV(keras_reg, param_distribs, n_iter=5, cv=3)\n",
    "rnd_search_cv.fit(X_train, y_train, epochs=500,\n",
    "                  validation_data=(X_test, y_test),\n",
    "                  callbacks=[tensorboard_cb, safety], batch_size=1000)\n",
    "\n",
    "\n",
    "# model = rnd_search_cv.best_estimator_.model\n",
    "# model.save('models/particle_random_cv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'n_neurons': 453, 'n_hidden': 2}\n",
      "-0.2610002855459849\n"
     ]
    }
   ],
   "source": [
    "print(rnd_search_cv.best_params_)\n",
    "print(rnd_search_cv.best_score_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: models/particle_random_cv_best_model\\assets\n"
     ]
    }
   ],
   "source": [
    "model = rnd_search_cv.best_estimator_.model\n",
    "model.save('models/particle_random_cv_best_model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# This can be used to load best model from the above\n",
    "\n",
    "model = tf.keras.models.load_model('models/particle_random_cv_best_model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_21\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_50 (Dense)             (None, 453)               13137     \n",
      "_________________________________________________________________\n",
      "dropout_29 (Dropout)         (None, 453)               0         \n",
      "_________________________________________________________________\n",
      "dense_51 (Dense)             (None, 453)               205662    \n",
      "_________________________________________________________________\n",
      "dropout_30 (Dropout)         (None, 453)               0         \n",
      "_________________________________________________________________\n",
      "dense_52 (Dense)             (None, 1)                 454       \n",
      "=================================================================\n",
      "Total params: 219,253\n",
      "Trainable params: 219,253\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2310/2310 [==============================] - 4s 1ms/step - loss: 0.2608 - accuracy: 0.8830\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.2607980966567993, 0.8829684257507324]"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(X_test, y_test, batch_size=1000)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model Summary\n",
    "\n",
    "We can see above that the grid search found an optimal configuration of 2 hidden layers and 453 nodes in each layer, achieving an accuracy of 88.3%.\n",
    "\n",
    "Let's produce a final model and then perform some predictions on our validation set. We split our data into 3 parts below, train (60%), test (20%) and validation (20%). The nuerul network is configured to have 2 layers, 128 nodes, 2 drop out layers, and a sigmoid activation function at teh output layer. We'll again use the Adam optimizer and BinaryCrossEntropy for the loss function. The same early stopping parameters defined above are used."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "4200/4200 [==============================] - 15s 3ms/step - loss: 0.2987 - accuracy: 0.8620 - val_loss: 0.2774 - val_accuracy: 0.8739\n",
      "Epoch 2/100\n",
      "4200/4200 [==============================] - 12s 3ms/step - loss: 0.2815 - accuracy: 0.8720 - val_loss: 0.2731 - val_accuracy: 0.8763\n",
      "Epoch 3/100\n",
      "4200/4200 [==============================] - 12s 3ms/step - loss: 0.2781 - accuracy: 0.8739 - val_loss: 0.2701 - val_accuracy: 0.8779\n",
      "Epoch 4/100\n",
      "4200/4200 [==============================] - 12s 3ms/step - loss: 0.2762 - accuracy: 0.8750 - val_loss: 0.2689 - val_accuracy: 0.8786\n",
      "Epoch 5/100\n",
      "4200/4200 [==============================] - 13s 3ms/step - loss: 0.2749 - accuracy: 0.8757 - val_loss: 0.2681 - val_accuracy: 0.8793\n",
      "Epoch 6/100\n",
      "4200/4200 [==============================] - 12s 3ms/step - loss: 0.2743 - accuracy: 0.8762 - val_loss: 0.2673 - val_accuracy: 0.8794\n",
      "Epoch 7/100\n",
      "4200/4200 [==============================] - 11s 3ms/step - loss: 0.2738 - accuracy: 0.8764 - val_loss: 0.2669 - val_accuracy: 0.8799\n",
      "Epoch 8/100\n",
      "4200/4200 [==============================] - 13s 3ms/step - loss: 0.2734 - accuracy: 0.8767 - val_loss: 0.2667 - val_accuracy: 0.8802\n",
      "Epoch 9/100\n",
      "4200/4200 [==============================] - 13s 3ms/step - loss: 0.2729 - accuracy: 0.8768 - val_loss: 0.2662 - val_accuracy: 0.8802\n",
      "Epoch 10/100\n",
      "4200/4200 [==============================] - 13s 3ms/step - loss: 0.2726 - accuracy: 0.8770 - val_loss: 0.2661 - val_accuracy: 0.8803\n",
      "Epoch 11/100\n",
      "4200/4200 [==============================] - 13s 3ms/step - loss: 0.2723 - accuracy: 0.8771 - val_loss: 0.2661 - val_accuracy: 0.8805\n",
      "Epoch 12/100\n",
      "4200/4200 [==============================] - 14s 3ms/step - loss: 0.2722 - accuracy: 0.8774 - val_loss: 0.2659 - val_accuracy: 0.8807\n",
      "Epoch 13/100\n",
      "4200/4200 [==============================] - 12s 3ms/step - loss: 0.2719 - accuracy: 0.8774 - val_loss: 0.2652 - val_accuracy: 0.8804\n",
      "Epoch 14/100\n",
      "4200/4200 [==============================] - 13s 3ms/step - loss: 0.2719 - accuracy: 0.8775 - val_loss: 0.2663 - val_accuracy: 0.8807\n",
      "Epoch 15/100\n",
      "4200/4200 [==============================] - 12s 3ms/step - loss: 0.2717 - accuracy: 0.8777 - val_loss: 0.2651 - val_accuracy: 0.8808\n",
      "Epoch 16/100\n",
      "4200/4200 [==============================] - 12s 3ms/step - loss: 0.2715 - accuracy: 0.8776 - val_loss: 0.2653 - val_accuracy: 0.8808\n",
      "1400/1400 [==============================] - 2s 1ms/step - loss: 0.2662 - accuracy: 0.8806\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.2662472724914551, 0.8806049823760986]"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from tensorflow.keras.callbacks import EarlyStopping\n",
    "\n",
    "root_logdir = os.path.join(os.curdir, \"my_logs\")\n",
    "\n",
    "def get_run_logdir():\n",
    "    import time\n",
    "    run_id = time.strftime(\"run_%Y_%m_%d-%H_%M_%S\")\n",
    "    return os.path.join(root_logdir, run_id)\n",
    "\n",
    "run_logdir = get_run_logdir()\n",
    "\n",
    "scaler = StandardScaler()\n",
    "\n",
    "X = df.loc[:, df.columns != 'target'].values\n",
    "\n",
    "y = df['target'].values\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=1)\n",
    "\n",
    "X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size=0.25, random_state=1) # 0.25 x 0.8 = 0.2\n",
    "\n",
    "scaler.fit(X_train)\n",
    "\n",
    "X_train = scaler.transform(X_train)\n",
    "X_test = scaler.transform(X_test)\n",
    "X_val = scaler.transform(X_val)\n",
    "\n",
    "final_model = tf.keras.Sequential([\n",
    "    tf.keras.layers.Flatten(input_shape=(28,)),\n",
    "    tf.keras.layers.Dense(128, activation='relu'),\n",
    "    tf.keras.layers.Dropout(.2, input_shape=(2,)),\n",
    "    tf.keras.layers.Dense(128, activation='relu'),\n",
    "    tf.keras.layers.Dropout(.2, input_shape=(2,)),\n",
    "    tf.keras.layers.Dense(1, activation='sigmoid')\n",
    "])\n",
    "\n",
    "final_model.compile(optimizer='adam',\n",
    "              loss=tf.keras.losses.BinaryCrossentropy(),\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "tensorboard_cb = tf.keras.callbacks.TensorBoard(get_run_logdir())\n",
    "\n",
    "safety = EarlyStopping(monitor='val_loss', patience=3, min_delta=2e-4)\n",
    "\n",
    "final_model.fit(X_train, y_train, epochs=100, validation_data=(X_test, y_test), \n",
    "          callbacks=[tensorboard_cb, safety], batch_size=1000)\n",
    "\n",
    "final_model.evaluate(X_val, y_val, batch_size=1000)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Predicting the Existence of New Particles\n",
    "\n",
    "Above we see that our final model performed well on the test, acheiving an accuracy of 88.1%.\n",
    "\n",
    "Let's examine how well our Neural Network performs by doing some predictions on our validation set and comparing the results to the actual labels. The classificatoin report shows good scores with respect to precision, recall, and F1-score across the board."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[595884, 104664],\n",
       "       [ 62489, 636963]], dtype=int64)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "y_pred = (final_model.predict(X_val) > 0.5).astype('int32')\n",
    "confusion_matrix(y_val, y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                    precision    recall  f1-score   support\n",
      "\n",
      "                No       0.91      0.85      0.88    700548\n",
      "New Particle (Yes)       0.86      0.91      0.88    699452\n",
      "\n",
      "          accuracy                           0.88   1400000\n",
      "         macro avg       0.88      0.88      0.88   1400000\n",
      "      weighted avg       0.88      0.88      0.88   1400000\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import classification_report\n",
    "\n",
    "print(classification_report(y_val, y_pred, target_names=[\"No\",\"New Particle (Yes)\"]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Tensorsboard\n",
    "\n",
    "Tensboard can be used to show the how the model is training. Below we output the accuracy and loss curves for our training and validation sets. Notice that the curves stop once our early stopping criteria is met."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext tensorboard\n",
    "\n",
    "%tensorboard --logdir logs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Selecting TensorBoard with logdir logs (started 1 day, 4:10:55 ago; port 6006, pid 44604).\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "      <iframe id=\"tensorboard-frame-c25b0b0e6a9f6ace\" width=\"100%\" height=\"800\" frameborder=\"0\">\n",
       "      </iframe>\n",
       "      <script>\n",
       "        (function() {\n",
       "          const frame = document.getElementById(\"tensorboard-frame-c25b0b0e6a9f6ace\");\n",
       "          const url = new URL(\"http://localhost\");\n",
       "          const port = 6006;\n",
       "          if (port) {\n",
       "            url.port = port;\n",
       "          }\n",
       "          frame.src = url;\n",
       "        })();\n",
       "      </script>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from tensorboard import notebook\n",
    "notebook.display()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Conclusion\n",
    "\n",
    "We examined using a neural network to predict the existence of new particles. We found that the neural network performed well on the validation set, and that the model was able to accurately predict the existence of new particles with 88% accuracy. However, it was also shown that Logistic Regression could perform almost as well, with an accuracy of 83.6%. Depending on resources and other constraints, using s model that can be trained faster might be more appropriate.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": []
  }
 ],
 "metadata": {
  "interpreter": {
   "hash": "eaa798b471aa1a0109429a408b0faab53065248ef7f5a5989f90756771672ef6"
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}