TrafficSignClassification-3-2-3.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data Mining Versuch Verkehrsschilderkennung mit Neuronalen Netzen\n",
    "\n",
    "* Autor: Prof. Dr. Johannes Maucher\n",
    "* Datum: 01.06.2021"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Einführung\n",
    "\n",
    "In diesem Versuch soll ein Convolutional Neural Network (CNN) für die Erkennung von Verkehrschildern implementiert, trainiert, evaluiert und getestet werden. Als Eingabe erhält das neuronale Netz Bilder von deutschen Verkehrsschildern. Ausgabe ist der Typ des Verkehrsschilds. Für Training und Test sind die Verkehrsschildbilder schon in separate Verzeichnissen abgelegt. Neben den Bildern selbst, enthält das zu diesem Versuch gehörende Datenverzeichnis auch Dateien mit Metadaten, die z.B. Bildeigenschaften, Verkehrsschildbedeutungen und die zugehörigen Klassenlabel beschreiben.\n",
    "\n",
    "## Lernziele:\n",
    "In diesem Versuch sollen Kenntnisse in folgenden Themen vermittelt werden:\n",
    "\n",
    "* Convolutional Neural Networks (CNNs)\n",
    "* Implementierung Tiefer Neuronaler Netze mit Tensorflow und Keras: \n",
    "    - Definition der Netzarchitektur\n",
    "    - Training\n",
    "    - Evaluation und Test\n",
    "    \n",
    "* Einfache Methoden der Bildverarbeitung:\n",
    "    - Augmentierung\n",
    "    - Kontrastverstärkung\n",
    "\n",
    "* Evaluation eines Klassifikators\n",
    "\n",
    "\n",
    "## Vorbereitung\n",
    "\n",
    "### Grundlagen Neuronale Netze\n",
    "Machen Sie sich mit den [Grundlagen herkömmlicher Neuronaler Netze (KI Vorlesung)](https://gitlab.mi.hdm-stuttgart.de/maucher/KI/blob/master/Slides/09_PartLernen4.pdf) und den [Grundlagen Convolutional Neural Networks ((KI Vorlesung))](https://gitlab.mi.hdm-stuttgart.de/maucher/KI/blob/master/Slides/V11DeepLearningKIversion.pdf) vertraut.\n",
    "\n",
    "\n",
    "### Implementierung Neuronaler Netze mit Tensorflow und Keras\n",
    "Machen Sie sich mit der Implementierung von Neuronalen Netzen mit Tensorflow und Keras vertraut. Z.B. mit den [Tensorflow Quickstart Tutorials](https://www.tensorflow.org/tutorials/quickstart/beginner)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Durchführung"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Vorbereitende Untersuchungen\n",
    "1. Importieren Sie ein Bild aus dem Verzeichnis `Train` mit der [scikit-image.io](https://scikit-image.org/docs/dev/api/skimage.io.html)-Methode `imread()` und zeigen Sie dieses mit der Methode `imshow()` an. Geben Sie die Größe des Bildes aus (Attribut `.shape`). \n",
    "2. Verändern Sie die Größe des Bildes mit der [scikit-image.transform](https://scikit-image.org/docs/dev/api/skimage.transform.html)-Methode `resize()` auf eine Größe von $32x32x3$. Die Verzerrung des Seitenverhältnisses kann dabei ignoriert werden. Diese Methode führt auch eine Normalisierung der Pixelwerte von [0,255] auf [0,1] durch.\n",
    "3. Führen Sie mit der [scikit-image.exposure](https://scikit-image.org/docs/dev/api/skimage.exposure.html)-Methode `equalize_adapthist()` eine Kontrastverstärkung des Bildes durch. Zeigen Sie das vergrößerte und kontrastangereicherte Bild an. **Anmerkung:** Das kontrastverstärkte Bild sieht zwar unschöner aus, auf der Basis kontrastverstärkter Bilder läßt sich aber im allgemeinen die Objekterkennung verbessern.\n",
    "2. Importieren Sie die Datei `Train.csv` und machen Sie sich mit deren Inhalt vertraut. Die Datei `Test.csv` ist gleich strukturiert, bezieht sich aber auf die Bilder im Verzeichnis `Test`. Wieviele Zeilen enthalten die Dateien?\n",
    "3. Importieren Sie die Datei `signnames.csv` und machen Sie sich mit deren Inhalt vertraut. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span style=\"color:#900C3F\">Dokumentation unserer Ergebnisse</span> \n",
    "\n",
    "Die Aufgabe war das Bild zunächst mit `imread()` einzulesen, dann mit `imshow()` anzuzeigen und schließlich die zugehörige shape auszugeben. Als Nächstes sollte man die Daten des Bildes auf die Größe von 32x32x3 umformen, was man mit der Methode `resize()` machen kann. Anschließend wurde eine Kontrastverstärkung auf das Bild gemacht, da sich die Objekte dadurch besser erkennen lassen. Nach der Verstärkung war die Aufgabe, `Train.csv` und `Test.csv` zu importieren und sich mit den Daten vertraut zu machen. Jeder Datensatz enthält die Breite, die Höhe, den Pfad des Bildes, Klassen und jeweils 2 x,y Roi-Koordinaten welche ein Viereck spannen. Zum Schluss sollte man die `signnames.csv` importieren, welche die Klassen zu bestimmten Verkehrszeichennamen zuordnet."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import skimage.io as io\n",
    "import numpy as np\n",
    "from skimage.transform import resize\n",
    "from skimage import exposure\n",
    "import pandas as pd\n",
    "import random\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn import preprocessing\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras import datasets, layers, models, activations\n",
    "from tensorflow.keras.layers import Dense, BatchNormalization, Flatten, Dropout\n",
    "from tensorflow.keras.optimizers import Adam\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "from sklearn.metrics import classification_report"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(30, 29, 3)\n",
      "Die Größe des Shapes beträgt: (30, 29, 3)\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": [
    "#1\n",
    "picture = \"./data/Train/0/00000_00000_00000.png\"\n",
    "read_picture = io.imread(picture)\n",
    "print(read_picture.shape)\n",
    "io.imshow(read_picture)\n",
    "print(\"Die Größe des Shapes beträgt: \"+ str(np.asarray(read_picture).shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[[0.29490273 0.30623851 0.31408165]\n",
      "  [0.29151731 0.29966682 0.30750996]\n",
      "  [0.32677313 0.33136872 0.32401195]\n",
      "  ...\n",
      "  [0.26469439 0.2890491  0.28816636]\n",
      "  [0.25685317 0.27065717 0.26650774]\n",
      "  [0.26033816 0.2646829  0.26063304]]\n",
      "\n",
      " [[0.32203202 0.32650506 0.33434819]\n",
      "  [0.31307062 0.31435738 0.32220052]\n",
      "  [0.34210708 0.33983992 0.32722886]\n",
      "  ...\n",
      "  [0.28667662 0.30154144 0.30167739]\n",
      "  [0.29604588 0.30346967 0.29294003]\n",
      "  [0.30789484 0.30864737 0.30100528]]\n",
      "\n",
      " [[0.31010072 0.31052964 0.31821768]\n",
      "  [0.33011451 0.32735715 0.33235677]\n",
      "  [0.35600299 0.34715839 0.34840878]\n",
      "  ...\n",
      "  [0.28417203 0.29217793 0.28327206]\n",
      "  [0.29078393 0.29468061 0.27684972]\n",
      "  [0.30621745 0.30642999 0.29178539]]\n",
      "\n",
      " ...\n",
      "\n",
      " [[0.49412722 0.5005936  0.51961742]\n",
      "  [0.46455461 0.48326823 0.50278991]\n",
      "  [0.42612783 0.45897097 0.47304305]\n",
      "  ...\n",
      "  [0.37747396 0.37774203 0.3555396 ]\n",
      "  [0.38808211 0.39505591 0.36803385]\n",
      "  [0.38648897 0.40401731 0.38655407]]\n",
      "\n",
      " [[0.3548292  0.36929764 0.38483456]\n",
      "  [0.35781633 0.38303845 0.40235907]\n",
      "  [0.30250651 0.34244409 0.35941713]\n",
      "  ...\n",
      "  [0.40912607 0.40644148 0.37961474]\n",
      "  [0.41986634 0.4343367  0.39964576]\n",
      "  [0.37675207 0.40685509 0.38202168]]\n",
      "\n",
      " [[0.33584942 0.33380438 0.34623162]\n",
      "  [0.36671645 0.38657705 0.40836397]\n",
      "  [0.27157054 0.31595818 0.33636259]\n",
      "  ...\n",
      "  [0.39765242 0.40022212 0.36835555]\n",
      "  [0.38453393 0.39852175 0.35401922]\n",
      "  [0.3553596  0.3821921  0.35033509]]]\n"
     ]
    }
   ],
   "source": [
    "#2\n",
    "resized_picture = resize(read_picture, (32, 32, 3))\n",
    "print(resized_picture)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7ff2403a4430>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "high_contrast_picture = exposure.equalize_adapthist(resized_picture)\n",
    "io.imshow(high_contrast_picture)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#4\n",
    "Train.csv hat 39210 Zeilen und Test.csv hat 12631 Zeilen."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "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>ClassId</th>\n",
       "      <th>SignName</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>Speed limit (20km/h)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>Speed limit (30km/h)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>Speed limit (50km/h)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>Speed limit (60km/h)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>Speed limit (70km/h)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5</td>\n",
       "      <td>Speed limit (80km/h)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>6</td>\n",
       "      <td>End of speed limit (80km/h)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>7</td>\n",
       "      <td>Speed limit (100km/h)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>8</td>\n",
       "      <td>Speed limit (120km/h)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>9</td>\n",
       "      <td>No passing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>10</td>\n",
       "      <td>No passing for vehicles over 3.5 metric tons</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>11</td>\n",
       "      <td>Right-of-way at the next intersection</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>12</td>\n",
       "      <td>Priority road</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>13</td>\n",
       "      <td>Yield</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>14</td>\n",
       "      <td>Stop</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>15</td>\n",
       "      <td>No vehicles</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>16</td>\n",
       "      <td>Vehicles over 3.5 metric tons prohibited</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>17</td>\n",
       "      <td>No entry</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>18</td>\n",
       "      <td>General caution</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>19</td>\n",
       "      <td>Dangerous curve to the left</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>20</td>\n",
       "      <td>Dangerous curve to the right</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>21</td>\n",
       "      <td>Double curve</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>22</td>\n",
       "      <td>Bumpy road</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>23</td>\n",
       "      <td>Slippery road</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>24</td>\n",
       "      <td>Road narrows on the right</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>25</td>\n",
       "      <td>Road work</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>26</td>\n",
       "      <td>Traffic signals</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>27</td>\n",
       "      <td>Pedestrians</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28</th>\n",
       "      <td>28</td>\n",
       "      <td>Children crossing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29</th>\n",
       "      <td>29</td>\n",
       "      <td>Bicycles crossing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>30</td>\n",
       "      <td>Beware of ice/snow</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>31</th>\n",
       "      <td>31</td>\n",
       "      <td>Wild animals crossing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>32</th>\n",
       "      <td>32</td>\n",
       "      <td>End of all speed and passing limits</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>33</th>\n",
       "      <td>33</td>\n",
       "      <td>Turn right ahead</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34</th>\n",
       "      <td>34</td>\n",
       "      <td>Turn left ahead</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>35</td>\n",
       "      <td>Ahead only</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>36</th>\n",
       "      <td>36</td>\n",
       "      <td>Go straight or right</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>37</th>\n",
       "      <td>37</td>\n",
       "      <td>Go straight or left</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>38</th>\n",
       "      <td>38</td>\n",
       "      <td>Keep right</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>39</th>\n",
       "      <td>39</td>\n",
       "      <td>Keep left</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40</th>\n",
       "      <td>40</td>\n",
       "      <td>Roundabout mandatory</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>41</th>\n",
       "      <td>41</td>\n",
       "      <td>End of no passing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>42</th>\n",
       "      <td>42</td>\n",
       "      <td>End of no passing by vehicles over 3.5 metric ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    ClassId                                           SignName\n",
       "0         0                               Speed limit (20km/h)\n",
       "1         1                               Speed limit (30km/h)\n",
       "2         2                               Speed limit (50km/h)\n",
       "3         3                               Speed limit (60km/h)\n",
       "4         4                               Speed limit (70km/h)\n",
       "5         5                               Speed limit (80km/h)\n",
       "6         6                        End of speed limit (80km/h)\n",
       "7         7                              Speed limit (100km/h)\n",
       "8         8                              Speed limit (120km/h)\n",
       "9         9                                         No passing\n",
       "10       10       No passing for vehicles over 3.5 metric tons\n",
       "11       11              Right-of-way at the next intersection\n",
       "12       12                                      Priority road\n",
       "13       13                                              Yield\n",
       "14       14                                               Stop\n",
       "15       15                                        No vehicles\n",
       "16       16           Vehicles over 3.5 metric tons prohibited\n",
       "17       17                                           No entry\n",
       "18       18                                    General caution\n",
       "19       19                        Dangerous curve to the left\n",
       "20       20                       Dangerous curve to the right\n",
       "21       21                                       Double curve\n",
       "22       22                                         Bumpy road\n",
       "23       23                                      Slippery road\n",
       "24       24                          Road narrows on the right\n",
       "25       25                                          Road work\n",
       "26       26                                    Traffic signals\n",
       "27       27                                        Pedestrians\n",
       "28       28                                  Children crossing\n",
       "29       29                                  Bicycles crossing\n",
       "30       30                                 Beware of ice/snow\n",
       "31       31                              Wild animals crossing\n",
       "32       32                End of all speed and passing limits\n",
       "33       33                                   Turn right ahead\n",
       "34       34                                    Turn left ahead\n",
       "35       35                                         Ahead only\n",
       "36       36                               Go straight or right\n",
       "37       37                                Go straight or left\n",
       "38       38                                         Keep right\n",
       "39       39                                          Keep left\n",
       "40       40                               Roundabout mandatory\n",
       "41       41                                  End of no passing\n",
       "42       42  End of no passing by vehicles over 3.5 metric ..."
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#5 \n",
    "signnames_file = \"./data/signnames.csv\"\n",
    "signnames = pd.read_csv(signnames_file)\n",
    "signnames"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Funktion für den Import aller Trainings- bzw. Testbilder\n",
    "Schreiben Sie eine Funktion mit folgenden Eigenschaften:\n",
    "* Der Funktion wird der Name der Datei übergeben, in welcher die Metadaten stehen, also entweder `Train.csv` oder `Test.csv`.\n",
    "* Rückgabewerte der Funktion sind \n",
    "    * ein 4-dimensionales numpy-array, das alle Bilder des jeweiligen Verzeichnisses (Training oder Test) enthält.\n",
    "    * ein 1-dimensionales numpy-array, das die Klassenlabel aller Bilder enthält.\n",
    "* Die Bilder müssen alle auf eine Größe von $32x32x3$ skaliert werden (wie in der Vorbereitungsaufgabe).\n",
    "* Für alle Bilder ist eine Kontrastverstärkung durchzuführen (wie in der Vorbereitungsaufgabe).\n",
    "\n",
    "**Tipps für die Implementierung dieser Funktion:**\n",
    "\n",
    "Iterieren Sie mit einer for-Schleife über alle Zeilen des metadaten-Files. Pro Iteration kann dann \n",
    "* der vollständige Verzeichnis- und Filenamen ausgelesen werden,\n",
    "* das entsprechende Bild mit `imread()` eingelesen werden,\n",
    "* das Bild auf die vorgegebene Größe angepasst werden,\n",
    "* der Kontrast des Bildes verstärkt werden.\n",
    "\n",
    "**Wichtig:** In den von der Funktion zurückgegebenen Arrays, dürfen die Bilder nicht wie in der ursprünglichen Reihenfolge im Dateiverzeichnis enthalten sein. Um sicherzustellen, dass beim Training jedes Minibatch möglichst viele verschiedene Klassen enthält, muss die Reihenfolge geshuffelt werden. Am einfachsten ist es, wenn gleich die Zeilen des Metadatenfiles geshuffelt werden."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span style=\"color:#900C3F\">Dokumentation unserer Ergebnisse</span> \n",
    "\n",
    "In dieser Aufgabe sollte man die Bilder wie in der Aufgabe davor importieren, die Form `resizen()`, eine Kontrastverstärkung eines bestimmten Pfades durchführen, welcher als Parameter an die Funktion gegeben wird. Die Funktion gibt die Bilder in einem 4-dimensionalem Array mit den zugehörigen Klassenlabels zurück. Durch ein shuffle aus der random library werden die Daten beim Ausführen jedes Mal in einer unterschiedlichen Reihenfolge zurückgegeben."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def import_pictures(filename):\n",
    "    file_df = pd.read_csv(\"./data/\" + filename)\n",
    "    all_data = []\n",
    "    for i, row in file_df.iterrows():\n",
    "        data_row = []\n",
    "        for j in range(len(row)):\n",
    "            data_row.append(row[j])\n",
    "        all_data.append(data_row)\n",
    "    \n",
    "    all_data_shuffled = list(all_data)\n",
    "    random.shuffle(all_data_shuffled)\n",
    "        \n",
    "    all_image_data = []\n",
    "    all_class_labels = []\n",
    "    for row in all_data_shuffled:\n",
    "        image = row[len(row) - 1]\n",
    "        image_read = io.imread(\"./data/\" + image)\n",
    "        image_resized = resize(image_read, (32, 32, 3))\n",
    "        image_contrasted = exposure.equalize_adapthist(image_resized)\n",
    "        all_class_labels.append(row[len(row) - 2])\n",
    "        all_image_data.append(image_contrasted)\n",
    "    \n",
    "    print(all_image_data)\n",
    "\n",
    "    return np.array(all_image_data), np.array(all_class_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Laden und Vorverarbeiten der Trainings- und Testdaten\n",
    "1. Laden Sie mit der in der vorigen Teilaufgabe implementierten Funktion alle Trainingsbilder (`trainX`), Trainingslabel (`trainY`), Testbilder (`testX`) und Testlabel (`testY`)\n",
    "1. Bestimmen Sie die Häufigkeitsverteilung der Klassen in den Trainings- und Testdaten. Visualisieren Sie diese.\n",
    "3. Alle Labels, sowohl der Trainings- als auch der Testdaten müssen One-Hot-encodiert werden."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span style=\"color:#900C3F\">Dokumentation unserer Ergebnisse</span> \n",
    "\n",
    "Zu Beginn sollten hier, jeweils für die Trainings- und Testdaten alle Bilder und Label importiert werden. Darauf sollte man die Häufigkeitsverteilung der Trainigslabel anzeigen. Diese wurden mit `matplotlib` geplotet und dargestellt. Im Vergleich sieht man kaum Unterschiede, was zeigt, dass die Trainingsdaten und Testdaten ähnlich sind. Allerdings gibt es auch ein paar Unterschiede, wie man bei dem X-Wert 41 erkennen kann. Zum Schluss sollten alle Label one-hot-encoded werden, wofür man den LabelBinarizer verwenden kann."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "#1\n",
    "trainX, trainY = import_pictures(\"Train.csv\")\n",
    "testX, testY =  import_pictures(\"Test.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 700x500 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#2\n",
    "plt.rcParams.update({'figure.figsize':(7,5), 'figure.dpi':100})\n",
    "\n",
    "myHist = plt.hist(trainY, bins=43)\n",
    "plt.gca().set(title='Trainingsklassen Häufigkeitsverteilung', ylabel='Häufigkeit');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 700x500 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(testY, bins=43)\n",
    "plt.gca().set(title='Testklassen Häufigkeitsverteilung', ylabel='Häufigkeit');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0 0 0 ... 0 0 0]\n",
      " [0 0 1 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]\n",
      " ...\n",
      " [0 0 0 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]]\n",
      "[[0 0 0 ... 0 0 0]\n",
      " [0 1 0 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]\n",
      " ...\n",
      " [0 0 1 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]\n",
      " [0 0 0 ... 0 0 0]]\n"
     ]
    }
   ],
   "source": [
    "#3 \n",
    "lb = preprocessing.LabelBinarizer()\n",
    "one_hot_econded_trainY = lb.fit_transform(trainY)\n",
    "one_hot_econded_testY = lb.fit_transform(testY)\n",
    "\n",
    "print(one_hot_econded_trainY)\n",
    "print(one_hot_econded_testY)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Definition der CNN Architektur\n",
    "Schreiben Sie eine Funktion `generateCNN(width, height, depth, classes)` die eine Keras CNN-Architektur zurück gibt. Für die Definition der Architektur werden dieser Funktion die Parameter:\n",
    "\n",
    "* `width`: Breite der Bilder\n",
    "* `height`: Höhe der Bilder\n",
    "* `depth`: Anzahl der Kanäle pro Bild\n",
    "* `classes`: Anzahl der unterschiedlichen Klassen\n",
    "\n",
    "übergeben. Die in der Funktion zu implementierende Architektur ist im folgenden Bild dargestellt. In der Spalte *Output shape* bezeichnen die zweite und dritte Zahl die Breite und die Höhe der einzelnen Kanäle (*Bilder*), der letzte Parameter bezeichnet die Anzahl der Kanäle (Parameter *filters* in der Konfiguration).\n",
    "In der Übersichtstabelle ist Filtergröße nicht aufgeführt. Empfohlen sind folgende Größen:\n",
    "* für alle Pooling Layer: *pool_size=(2,2)*.\n",
    "* für den ersten Conv2D-Layer: *kernel_size=(5,5)*.\n",
    "* für alle weiteren Conv2D-Layer: *kernel_size=(3,3)*\n",
    "\n",
    "**Anmerkung:** Der in der Tabelle mit *flatten_5* bezeichnete Layer ist nicht notwendig und erzeugt in bestimmten Keras-Versionen eine Fehlermeldung. Der Layer sollte nicht nicht in die Architektur mit aufgenommen werden.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"https://maucher.home.hdm-stuttgart.de/Pics/cnnTrafficSign.png\" style=\"width:700px\" align=\"middle\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span style=\"color:#900C3F\">Dokumentation unserer Ergebnisse</span> \n",
    "\n",
    "Diese Aufgabe war relativ schnell gelöst und hat daher auch viel Spaß gemacht. Man konnte einfach nach den gewünschten Layern suchen. Zudem hat die (berufliche) Erfahrung eines Gruppenmitglieds zu dem schnellen Ergebnis beigetragen. \n",
    "Lediglich zwei kleine Probleme sind aufgetreten, welche sich jedoch mit etwas Hilfe schnell lösen ließen. \n",
    "Einerseits passierte es, dass der input und output nicht die selbe ```shape```hatten. Die Lösung hierfür war, ```padding=\"same\"```hinzuzufügen. \n",
    "Die zweite Schwierigkeit war die falsche ```Output-Shape``` in der letzten Zeile. Da hatten wir jedoch lediglich vergessen, statt der relu die softmax-Aktivierungsfunktion zu verwenden. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def generateCNN(width, height, depth, classes):\n",
    "    model = models.Sequential()\n",
    "    model.add(layers.Conv2D(8, (5, 5), input_shape=(width, height, depth), padding='same')) \n",
    "    model.add(layers.Activation(activations.relu))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(layers.MaxPooling2D((2, 2)))\n",
    "    model.add(layers.Conv2D(16, (3, 3), input_shape=(width, height, depth), padding='same'))\n",
    "    model.add(layers.Activation(activations.relu))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(layers.Conv2D(16, (3, 3), input_shape=(width, height, depth), padding='same'))\n",
    "    model.add(layers.Activation(activations.relu))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(layers.MaxPooling2D((2, 2)))\n",
    "    model.add(layers.Conv2D(32, (3, 3), input_shape=(width, height, depth), padding='same'))\n",
    "    model.add(layers.Activation(activations.relu))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(layers.Conv2D(32, (3, 3), input_shape=(width, height, depth), padding='same'))\n",
    "    model.add(layers.Activation(activations.relu))\n",
    "    model.add(layers.MaxPooling2D((2, 2)))\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(128))\n",
    "    model.add(layers.Activation(activations.relu))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(Dropout(0.2))\n",
    "    model.add(Dense(128, input_shape = (128,)))\n",
    "    model.add(layers.Activation(activations.relu))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(Dropout(0.2))\n",
    "    model.add(Dense(classes, input_shape = (128,)))\n",
    "    model.add(layers.Activation(activations.softmax))\n",
    "    return model\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training des CNN\n",
    "Rufen Sie die im vorigen Abschnitt implementierte Funktion `generateCNN()` auf und weisen Sie die von der Funktion zurückgegebene Architektur der Variablen `model` zu. Durch Aufruf der Funktion `model.summary()` erhalten Sie eine Übersicht des erzeugten Netzes.\n",
    "\n",
    "Für das Training soll der `Adam`-Algorithmus aus dem Modul `tensorflow.keras.optimizers` benutzt werden. `Adam` implementiert ein *Stochastic Gradient Descent*-Lernverfahren, welches die Lernraten für die Gewichte individuell und dynamisch anpasst.\n",
    "\n",
    "In den folgenden zwei Codezellen, werden die Trainingsparameter konfiguriert:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " conv2d (Conv2D)             (None, 32, 32, 8)         608       \n",
      "                                                                 \n",
      " activation (Activation)     (None, 32, 32, 8)         0         \n",
      "                                                                 \n",
      " batch_normalization (BatchN  (None, 32, 32, 8)        32        \n",
      " ormalization)                                                   \n",
      "                                                                 \n",
      " max_pooling2d (MaxPooling2D  (None, 16, 16, 8)        0         \n",
      " )                                                               \n",
      "                                                                 \n",
      " conv2d_1 (Conv2D)           (None, 16, 16, 16)        1168      \n",
      "                                                                 \n",
      " activation_1 (Activation)   (None, 16, 16, 16)        0         \n",
      "                                                                 \n",
      " batch_normalization_1 (Batc  (None, 16, 16, 16)       64        \n",
      " hNormalization)                                                 \n",
      "                                                                 \n",
      " conv2d_2 (Conv2D)           (None, 16, 16, 16)        2320      \n",
      "                                                                 \n",
      " activation_2 (Activation)   (None, 16, 16, 16)        0         \n",
      "                                                                 \n",
      " batch_normalization_2 (Batc  (None, 16, 16, 16)       64        \n",
      " hNormalization)                                                 \n",
      "                                                                 \n",
      " max_pooling2d_1 (MaxPooling  (None, 8, 8, 16)         0         \n",
      " 2D)                                                             \n",
      "                                                                 \n",
      " conv2d_3 (Conv2D)           (None, 8, 8, 32)          4640      \n",
      "                                                                 \n",
      " activation_3 (Activation)   (None, 8, 8, 32)          0         \n",
      "                                                                 \n",
      " batch_normalization_3 (Batc  (None, 8, 8, 32)         128       \n",
      " hNormalization)                                                 \n",
      "                                                                 \n",
      " conv2d_4 (Conv2D)           (None, 8, 8, 32)          9248      \n",
      "                                                                 \n",
      " activation_4 (Activation)   (None, 8, 8, 32)          0         \n",
      "                                                                 \n",
      " max_pooling2d_2 (MaxPooling  (None, 4, 4, 32)         0         \n",
      " 2D)                                                             \n",
      "                                                                 \n",
      " flatten (Flatten)           (None, 512)               0         \n",
      "                                                                 \n",
      " dense (Dense)               (None, 128)               65664     \n",
      "                                                                 \n",
      " activation_5 (Activation)   (None, 128)               0         \n",
      "                                                                 \n",
      " batch_normalization_4 (Batc  (None, 128)              512       \n",
      " hNormalization)                                                 \n",
      "                                                                 \n",
      " dropout (Dropout)           (None, 128)               0         \n",
      "                                                                 \n",
      " dense_1 (Dense)             (None, 128)               16512     \n",
      "                                                                 \n",
      " activation_6 (Activation)   (None, 128)               0         \n",
      "                                                                 \n",
      " batch_normalization_5 (Batc  (None, 128)              512       \n",
      " hNormalization)                                                 \n",
      "                                                                 \n",
      " dropout_1 (Dropout)         (None, 128)               0         \n",
      "                                                                 \n",
      " dense_2 (Dense)             (None, 43)                5547      \n",
      "                                                                 \n",
      " activation_7 (Activation)   (None, 43)                0         \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 107,019\n",
      "Trainable params: 106,363\n",
      "Non-trainable params: 656\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "NUM_EPOCHS = 15 # Number of training epochs \n",
    "INIT_LR = 1e-3 # Initial Learning Rate for ADAM training\n",
    "BS = 64 # Size of minibatches\n",
    "opt = Adam(learning_rate=INIT_LR, decay=INIT_LR / (NUM_EPOCHS * 0.5)) \n",
    "\n",
    "model = generateCNN(32, 32, 3, 43)\n",
    "model.summary()\n",
    "model.compile(loss=\"categorical_crossentropy\", optimizer=opt, metrics=[\"accuracy\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Für das Training sollen nicht nur die vorhandenen Trainingsbilder eingesetzt werden, sondern zusätzlich Bilder die Augmentierungen der Trainingsbilder sind. Augmentierte Bilder können mit dem `ImageDataGenerator` des Moduls `tensorflow.keras.preprocessing.image` erzeugt werden. Der Code für die Erzeugung des in diesem Projekt eingesetzten Objekts ist unten gegeben. \n",
    "\n",
    "**Aufgabe:** Erklären Sie was in dieser Codezelle definiert wird."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span style=\"color:#900C3F\">Dokumentation unserer Ergebnisse</span> \n",
    "\n",
    "1. Der Parameter `rotation_range` meint das Maß - in Grad (hier: 10 Grad) - in welchem die Bilder rotieren können, denn nicht alle Bilder können gleich ausgerichtet fotografiert werden - ein wenig Abweichung muss also mit einberechnet werden. \n",
    "2. `zoom_range` meint etwas Ähnliches wie `rotation_range`, nur geht es hier um eine Abweichung bzgl. des Zooms - sprich, ein paar Bilder wurden vielleicht etwas näher aufgenommen als andere.\n",
    "3. Ebenfalls ähnlich ist `width_shift_range`. Hier bezieht es sich auf die Abweichung des Verkehrszeichens auf der X-Achse."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# construct the image generator for data augmentation\n",
    "aug = ImageDataGenerator(\n",
    "    rotation_range=10,\n",
    "    zoom_range=0.15,\n",
    "    width_shift_range=0.1,\n",
    "    height_shift_range=0.1,\n",
    "    shear_range=0.15,\n",
    "    horizontal_flip=False,\n",
    "    vertical_flip=False,\n",
    "    fill_mode=\"nearest\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Das Training wird mit folgender Codezelle ausgeführt.\n",
    "\n",
    "**Aufgabe:** Erklären Sie die Argumente der Funktion `fit_generator()`. \n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span style=\"color:#900C3F\">Dokumentation unserer Ergebnisse</span> \n",
    "\n",
    "1. ```aug.flow()``` transformiert die Eingabebilder so, dass unter dem Strich mehr Eingabedaten zur Verfügung stehen.\n",
    "2. ```validation_data``` definiert die Testdatensätze, welche für die Modellevaluation verwendet werden.\n",
    "3. ```steps_per_epoch``` spezifiziert wie viele \"Schritte\" eine Epoche durchläuft. Sprich wie oft eine Epoche 'X' Batches bestehend aus einer Stichprobe von 'Y' Elementen ausführt.\n",
    "4. ```epochs``` definiert dann die Anzahl an Epochen insgesamt.\n",
    "5. ```class_weight``` bietet das Gewicht zu jedem Klassenlabel. Die Gewichte können auch dazu verwendet werden, bestimmten unterpräsentierten Klassen beispielsweise 'mehr Aufmerksamkeit' während der Trainings zu schenken. In unserem Fall repräsentiert das jeweilige Gewicht den tatsächlichen Anteil an den zur Verfügung stehenden Daten.\n",
    "6. ```verbose``` gibt an, in welcher Art der Fortschritt der Trainings angezeigt werden soll. Im vorliegenden Fall in der Art des ```progress bar```, also einem Balken."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Für die Ausführung der folgenden Zelle muss das Dictionary `classWeight` (siehe Erklärung oben) angelegt sein. Dieses enthält für jede Klasse den Klassenindex als key und den relativen Anteil dieser Klasse in den Trainingsbildern als Value. Wenn z.B. 30% aller Trainingsdaten zur Klasse 0 gehören, dann wäre der Value zum Key 0 der Wert 0.3."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{38: 0.05, 2: 0.06, 11: 0.03, 5: 0.05, 23: 0.01, 9: 0.04, 10: 0.05, 25: 0.04, 1: 0.06, 12: 0.05, 30: 0.01, 35: 0.03, 13: 0.06, 41: 0.01, 6: 0.01, 18: 0.03, 31: 0.02, 40: 0.01, 3: 0.04, 4: 0.05, 8: 0.04, 21: 0.01, 15: 0.02, 7: 0.04, 17: 0.03, 37: 0.01, 14: 0.02, 24: 0.01, 26: 0.02, 28: 0.01, 16: 0.01, 33: 0.02, 36: 0.01, 22: 0.01, 19: 0.01, 29: 0.01, 0: 0.01, 39: 0.01, 20: 0.01, 34: 0.01, 42: 0.01, 32: 0.01, 27: 0.01}\n"
     ]
    }
   ],
   "source": [
    "trainX = np.array(trainX)\n",
    "trainY = np.array(trainY)\n",
    "testX = np.array(testX)\n",
    "testY = np.array(testY)\n",
    "\n",
    "classWeight = dict()\n",
    "\n",
    "for label in trainY:\n",
    "    weight_keys = [key for key in classWeight.keys()]\n",
    "    if not(label in weight_keys):\n",
    "        classWeight[label] = 1\n",
    "    else: \n",
    "        classWeight[label] = classWeight[label] + 1\n",
    "    \n",
    "for item in classWeight:\n",
    "    classWeight[item] = round(classWeight[item] / len(trainY), 2)\n",
    "\n",
    "print(classWeight)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[INFO] training network...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "<ipython-input-18-51669aec0121>:3: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.\n",
      "  H = model.fit_generator(\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "612/612 [==============================] - 38s 61ms/step - loss: 0.0581 - accuracy: 0.4640 - val_loss: 1.1784 - val_accuracy: 0.6384\n",
      "Epoch 2/15\n",
      "612/612 [==============================] - 32s 52ms/step - loss: 0.0210 - accuracy: 0.7317 - val_loss: 0.6914 - val_accuracy: 0.7882\n",
      "Epoch 3/15\n",
      "612/612 [==============================] - 32s 52ms/step - loss: 0.0131 - accuracy: 0.8176 - val_loss: 0.4712 - val_accuracy: 0.8492\n",
      "Epoch 4/15\n",
      "612/612 [==============================] - 34s 55ms/step - loss: 0.0096 - accuracy: 0.8620 - val_loss: 0.3716 - val_accuracy: 0.8869\n",
      "Epoch 5/15\n",
      "612/612 [==============================] - 38s 61ms/step - loss: 0.0077 - accuracy: 0.8906 - val_loss: 0.3961 - val_accuracy: 0.8793\n",
      "Epoch 6/15\n",
      "612/612 [==============================] - 32s 52ms/step - loss: 0.0062 - accuracy: 0.9096 - val_loss: 0.3350 - val_accuracy: 0.8973\n",
      "Epoch 7/15\n",
      "612/612 [==============================] - 32s 52ms/step - loss: 0.0055 - accuracy: 0.9219 - val_loss: 0.3078 - val_accuracy: 0.9039\n",
      "Epoch 8/15\n",
      "612/612 [==============================] - 34s 55ms/step - loss: 0.0047 - accuracy: 0.9308 - val_loss: 0.2531 - val_accuracy: 0.9234\n",
      "Epoch 9/15\n",
      "612/612 [==============================] - 34s 56ms/step - loss: 0.0044 - accuracy: 0.9366 - val_loss: 0.3101 - val_accuracy: 0.9059\n",
      "Epoch 10/15\n",
      "612/612 [==============================] - 34s 55ms/step - loss: 0.0039 - accuracy: 0.9433 - val_loss: 0.2687 - val_accuracy: 0.9186\n",
      "Epoch 11/15\n",
      "612/612 [==============================] - 36s 58ms/step - loss: 0.0036 - accuracy: 0.9461 - val_loss: 0.1826 - val_accuracy: 0.9423\n",
      "Epoch 12/15\n",
      "612/612 [==============================] - 36s 58ms/step - loss: 0.0035 - accuracy: 0.9508 - val_loss: 0.1914 - val_accuracy: 0.9405\n",
      "Epoch 13/15\n",
      "612/612 [==============================] - 35s 58ms/step - loss: 0.0031 - accuracy: 0.9551 - val_loss: 0.2040 - val_accuracy: 0.9365\n",
      "Epoch 14/15\n",
      "612/612 [==============================] - 36s 58ms/step - loss: 0.0029 - accuracy: 0.9570 - val_loss: 0.2041 - val_accuracy: 0.9363\n",
      "Epoch 15/15\n",
      "612/612 [==============================] - 34s 55ms/step - loss: 0.0028 - accuracy: 0.9594 - val_loss: 0.2223 - val_accuracy: 0.9353\n"
     ]
    }
   ],
   "source": [
    "# compile the model and train the network\n",
    "print(\"[INFO] training network...\")\n",
    "H = model.fit_generator(\n",
    "    aug.flow(trainX, one_hot_econded_trainY, batch_size=BS),\n",
    "    validation_data=(testX, one_hot_econded_testY),\n",
    "    steps_per_epoch=trainX.shape[0] // BS,\n",
    "    epochs=NUM_EPOCHS,\n",
    "    class_weight=classWeight,\n",
    "    verbose=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Visualisieren Sie die Entwicklung der *Accuracy* über dem Fortschritt der Trainingsepochen. Plotten Sie dabei die entsprechenden Kurven der Accuracy auf den Trainings- und auf den Testdaten in einen Graphen."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span style=\"color:#900C3F\">Dokumentation unserer Ergebnisse</span> \n",
    "\n",
    "Im folgenden Graph erkennt man gut, dass die Testdaten die Trainingsdaten gut annähern, wobei ein gewisser Grad an Verbesserung durchaus möglich ist (siehe peaks an Abweichungen). Dennoch sollte man darauf achten, overfitting zu vermeiden. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 700x500 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(H.history['accuracy'])\n",
    "plt.plot(H.history['val_accuracy'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluation des gelernten Modells\n",
    "\n",
    "Wenden Sie das gelernte CNN an, um für alle Bilder des Testdatensatzes die Art des Verkehrsschildes zu bestimmen. Evaluieren Sie die Qualität des CNN indem Sie einen `classification_report()` aus dem Modul `sklearn.metrics` erzeugen. \n",
    "\n",
    "1. Welche Metriken werden im Report angezeigt? Beschreiben Sie diese kurz?\n",
    "2. Diskutieren Sie die Klassifikationsgenauigkeit des CNN anhand des Reports.\n",
    "\n",
    "Zeigen Sie 5 Bilder an, die nicht korrekt klassifiziert wurden. Läßt sich die Fehlklassifikation erklären?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span style=\"color:#900C3F\">Dokumentation unserer Ergebnisse</span> \n",
    "\n",
    "1. precision / recall / f1-score / support\n",
    "\n",
    "`Precision` - die Precision beschreibt wie viele der als Klasse X vorhergesagten auch tatsächlich Klasse X sind <br>\n",
    "`Recall` - der Recall zeigt wie viele derer, die wirklich zu Klasse X gehören, auch als Klasse X erkannt wurden <br>\n",
    "`F1-Score` - der F1-Score berechnet das harmonische Mittel aus Precision und Recall. <br>\n",
    "`Support` - der Support gibt an wie häufig die Klasse X in y_true vorkommt <br>\n",
    "`Accuracy` - totale Anzahl der korrekten Vorhersagen geteilt durch die totale Anzahl der Vorhersagen\n",
    "`Macro Average` - Durchschnitt des ungewichteten Mittels pro Klasse \n",
    "`weighted Average` - Durchschnitt des gewichteten Mittels pro Klasse \n",
    "\n",
    "\n",
    "Der Klassifikationsreport anhand der Testdaten zeigt ein sehr gutes Ergebnis. Die `Precision` fällt nie unter 0.81 und hat einen macro-Durschschnittswert von 0.94. Das zeigt, dass die `y_pred`, also die Vorhersage in ~0.94 der Fälle der tatsächlichen Klasse entspricht. Ebenso ist der `Recall` sehr hoch. Dieser fällt nur einmal auf 0.7 und hat einen macro-Durschschnittswert von 0.92. Es werden also ~0.92 aller Elemente einer Klasse auch tatsächlich dieser Klasse zugeordnet. Somit ergibt sich auch eine hohe `Accuracy` von 0.95. \n",
    "\n",
    "Auffällig ist jedoch, dass sich `Macro-average` und `Mean-average` leicht unterscheiden. Bei dem `Macro-average` wird danach gewichtet wie hoch der Anteil der Elemente einer Klasse an allen Elementen ist. Bei einem sehr unbalanciertem Datensatz z.B. einem Datensatz in dem protokolliert wird, wie viele von 100 Patienten auf die Krankheit Y positiv getestet wurden, kann es zu großen Differenzen der Klassen kommen und zu Gewichten von a=0.99 (negativ getestet) und b=0.01 (positiv getestet). In diesem Fall wäre a eine Mehrheitsklasse und b die Minderheitsklasse. Die Gewichte sind weit auseinander und unbalanciert. Wenn also wie bei der Berechnung des Macro-average die Gewichtung miteinbezogen wird, dann muss man darauf achten, dass die Klassen nicht zu unbalanciert sind. \n",
    "\n",
    "Bei der Berechnung des `Mean-average` werden diese Gewichte nicht beachtet. Somit würden a und b in diesem Fall gleich behandelt werden. \n",
    "In dem Fall von den Daten zu der Verkehrszeichenerkennung sind die Klassen ebenfalls nicht gleich verteilt. Das erkennt man an den unterschiedlichen Werten in der `Support`-Spalte. Allerdings sind die Differenzen nicht so signifikant wie bei dem Beispiel mit den Patienten. Es ergibt sich dennoch ein unterschiedlicher Wert für den `Weighted-Average`. Bei einem balancierten Datensatz würden die Werte von `Macro-Average` und `Mean-Average` übereinstimmen. Bei einem Klassenungleichgewicht wird die `Accuracy` ein ungenaues Ergebnis liefern, weil die Minderheitsklasse eventuell nicht erkannt wird. Darauf sollte man also achten bei dem Lesen eines solchen Klassifikationsreports. \n",
    "\n",
    "Die fehlerhafte Klassifikation der 5 ausgewählten Bilder lässt sich unter Anderem mithilfe der Kontrastverstärkung erklären. Vor allem bei Bild Nummer 1 ist es selbst mit den Augen nur schwer zu erkennen, welches Schild gemeint ist. \n",
    "Außerdem spielen auch schwierige Lichtverhältnisse eine nicht zu vernachlässigende Rolle. Beispielsweise erkennt man im letzten Bild einen regelrechten gelben Fleck - wohl aufgrund erhöhter Sonneneinstrahlung.\n",
    "Ein Beispiel von besonders wenig Lichteinstrahlung erkennt man im dritten Bild. \n",
    "\n",
    "Vor allem die falschen Klassifikationen aufgrund unterschiedlichster Lichtverhältnisse sollten im Bestfall 'vernichtet' werden, da es im Straßenverkehr immer wieder zu plötzlichen Verdunkelungen durch Wolken, Regen oder Tunnel kommen kann."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.94      0.97      0.95        60\n",
      "           1       0.98      0.98      0.98       720\n",
      "           2       0.98      0.95      0.97       750\n",
      "           3       0.97      0.95      0.96       450\n",
      "           4       0.98      0.97      0.97       660\n",
      "           5       0.87      0.97      0.92       630\n",
      "           6       0.95      0.95      0.95       150\n",
      "           7       0.97      0.86      0.91       450\n",
      "           8       0.86      0.96      0.91       450\n",
      "           9       0.90      0.99      0.94       480\n",
      "          10       1.00      0.95      0.97       660\n",
      "          11       0.88      0.87      0.88       420\n",
      "          12       1.00      0.98      0.99       690\n",
      "          13       0.99      1.00      0.99       720\n",
      "          14       0.98      1.00      0.99       270\n",
      "          15       0.98      0.94      0.96       210\n",
      "          16       0.92      0.97      0.94       150\n",
      "          17       1.00      0.94      0.97       360\n",
      "          18       0.99      0.71      0.83       390\n",
      "          19       0.66      0.88      0.76        60\n",
      "          20       0.76      0.80      0.78        90\n",
      "          21       0.90      0.49      0.63        90\n",
      "          22       0.97      0.86      0.91       120\n",
      "          23       0.73      0.85      0.78       150\n",
      "          24       0.92      0.90      0.91        90\n",
      "          25       0.88      0.99      0.93       480\n",
      "          26       0.97      0.88      0.92       180\n",
      "          27       0.48      0.77      0.59        60\n",
      "          28       0.90      0.92      0.91       150\n",
      "          29       0.61      0.98      0.75        90\n",
      "          30       0.66      0.55      0.60       150\n",
      "          31       0.99      0.93      0.96       270\n",
      "          32       1.00      0.97      0.98        60\n",
      "          33       0.97      0.99      0.98       210\n",
      "          34       0.91      1.00      0.95       120\n",
      "          35       0.98      0.97      0.97       390\n",
      "          36       0.92      0.99      0.95       120\n",
      "          37       0.81      0.97      0.88        60\n",
      "          38       1.00      0.98      0.99       690\n",
      "          39       0.96      0.90      0.93        90\n",
      "          40       0.73      0.91      0.81        90\n",
      "          41       0.88      0.73      0.80        60\n",
      "          42       1.00      0.68      0.81        90\n",
      "\n",
      "    accuracy                           0.94     12630\n",
      "   macro avg       0.90      0.90      0.89     12630\n",
      "weighted avg       0.94      0.94      0.94     12630\n",
      "\n"
     ]
    }
   ],
   "source": [
    "y_pred = model.predict(testX)\n",
    "y_true = testY\n",
    "y_preds = y_pred.argmax(axis = 1)\n",
    "\n",
    "report = classification_report(y_true, y_preds, labels=list(range(43)))\n",
    "print(report)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7ff1e528ad60>"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 700x500 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "inputs = []\n",
    "\n",
    "for input, prediction, label in zip(testX, y_pred, y_true):\n",
    "    if(prediction is not label):\n",
    "        #print(input, 'has been classified as ', prediction, 'and should be ', label)\n",
    "        inputs.append(input)\n",
    "    \n",
    "io.imshow(inputs[0])\n",
    "#print(inputs[:5])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7ff1f6e55580>"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 700x500 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "io.imshow(inputs[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7ff1e5dd55e0>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 700x500 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "io.imshow(inputs[2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7ff1e7721d30>"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 700x500 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "io.imshow(inputs[3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7ff1e79700a0>"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAe0AAAHqCAYAAAApsRZDAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAA9hAAAPYQGoP6dpAAAuZElEQVR4nO3deXxV9Z3/8fcne8hCwhIIAgICVhSrApaRKmiL2upUW61aq7Xt1M7Ysb8uTrUynQ79ddSujp2qnY621bZjdbTW1loUV9SiuFBFBAyyLyFsIQlkT77zRxLnNoTlfExIvuT1fDzuA3PvfXsOJ+feN+fec+/HQggCAAB9X1pvrwAAADg4lDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABAJShsAgEhQ2gAARCKjt1egMzMzSSMk1fT2ugAAcAgVSNoc9vOtZ32utNVW2Bt7eyUAAOgFIyVt2teNfbG0ayTpsg+/V1mZ6YmCDz27zLXAT39iuit3ysgcV+7Vdb53JaYP9P26duVVuHLHvf9SV25r+iBX7pK//Zwrt3tPkyu36rc3u3L3/el+V+66O/7syv3o7y9w5VbteCtx5pU397iWpTTfPr2lyveC2q7dDa6c5Pva5oKCXFdu/IQSV277lmpX7gjneq7YVOnKfeny8125IYXbXbmlG7a5chVlO105tTi/5jtkJ440Nrfo18+/Lh3gVeYeK20z+4Kkr0kqlfSmpC+HEJ472HxWZnri0k4zS3T/DtnZvs2Ql5PpyuVk+Z7g8pzr2Zjjy+Xn+Z4A9mQMcOXM+fvz5gqdf7/crEP7b93cbN9+lp3w8SNJGenO01zSki9LktKdZe99rHulOdczI8O5XZy/h0zv8px/v9ycLFdugPe50/kcmLRL3mHe0nYu7yD0yIloZnaxpFsk3SDpREnPSZpnZqN7YnkAAPQHPXX2+Fcl/SyEcGcIYXkI4cuSNki6qoeWBwDAYa/bS9vMsiRNkTS/003zJZ3S3csDAKC/6Ik354ZISpfU+eynCknDO9/ZzLIlpb5rX9AD6wQAQPR68stVOr+Db11cJ0nXS6pKufBxLwAAutATpb1dUov2Pqou0d5H35J0k6SBKZeRPbBOAABEr9tLO4TQKOlVSbM73TRb0sIu7t8QQqjuuIhvQgMAoEs99YHTmyX9ysxekfSCpM9LGi3pP3toeQAAHPZ6pLRDCPeZ2WBJ31Tbl6sslfThEMK6nlgeAAD9QY99tVMI4XZJt/fU/x8AgP6mL373eJucoVJWsq+6Gzdiq2tRixck/45mSbpjlW95pUOLXLmFJXmu3BnjB7pyd/7x3125unrf91dfOKvUlbvkonNcubIBVa5c6fC9Prl4UPIHJP8+Ykn6/sPPunInjz0yceaLE3xfWrit3rctl5T6vkr2ybIdrlxxvu/rMyeMKXTltju/I31oke8xW77V9x3i40qLXbk128tduZdW+HKjsnxfXztpku/85kkT/8aVW122KnGmrqFRembxAe/HPG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABAJShsAgEhQ2gAARILSBgAgEn12yteu6o3KzExPlKms9U2X+to501y548dvd+XOOnqcK3fHsweeANOVxet9k4aWLd/kyl1wxlRXbtxQ3wSmq75xnytXXDDAlZs02jfla9IxvilmH7zo+67cB044OXHmxPGtrmWVP/+6K7fhvm+6cqXDfBPvJh1Z4splDPA9ho6fOMyVW1lW58qVDvZNTSstTj4RTpKKWppduW1VTa5cufmWl+FsutUrlrhyxcOyEmey6w9um3CkDQBAJChtAAAiQWkDABAJShsAgEhQ2gAARILSBgAgEpQ2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESiz075mlpqysmyRJlJ4ya5ljVq+lhXbuxHZ7hyI1e85cqF53yTcV4u2+rKDS0Z5Mqt21Xryi1attaVG1romw524RknuHIF05JPz5Kkc95+3pXLOMo33W3ySeckzhQPznEta/OmZa7c2FlHu3IjdvomtKVv3ujKpeUEV66iZosrN+qoo1y5xroqV+7RBX925dZt8y1vxrFjXDk11rti5ZWNrtzEI30V+cffJX88tLQe3IQ9jrQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFgIvuk1PcXMCiVV3X31bA3ITja96aQc30SW8+9d6Mrt3OObunXd5dNcuQ9MHO7KLV6x3pX75m9ed+W2VdW5chOdk40+e/XnXbnBOUWuXPoAX27auGJXbszIMa5c09adiTPl5eWuZbVasol8HYYdNdKVSysc6MqtW7/NlVu88PeuXM26N125TcvLXLmK1ixX7sUVvu2ydO1mV66w0DelrbW52Zdz9lyrsx7HDi9KnGlpbdWyNVslaWAIoXpf9+NIGwCASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABAJShsAgEhQ2gAARILSBgAgEpQ2AACRoLQBAIiEbyzWITBzTIkKc5NNrNnQ6lvWiKICV64gt9GVW/rmRlfumOJBrtz6nb7JP+edPs6VS59xoit3SsPfuHJZJb7pZ+e+xzd1a8nDj7pyT935oCu3trzWlStwTN4aku17Sqhs9j34dsuX+8i0ia7cikzfvnnRP1/iyqXlXevK3fuLb7lyzQ//0ZUbW+gbZ7VnhO85qXhIqSt32fQiV27jWztcuYff9k0/GzdySOJMU3NLx5Sv/eJIGwCASFDaAABEottL28zmmlnodNnS3csBAKC/6an3tN+U9MGUn1t6aDkAAPQbPVXazSEEjq4BAOhGPfWe9gQz22xma8zsXjPznYYMAADe0RNH2oskfUpSmaRhkr4haaGZHRtC2Ou8ezPLlpSdcpXv81cAABzmuv1IO4QwL4Tw2xDCGyGEJySd037TFfuIXC+pKuXi+xAzAACHuR7/yFcIYY+kNyRN2MddbpI0MOUysqfXCQCAGPX4N6K1v/x9jKTnuro9hNAgqSHl/j29SgAARKknPqf9AzObaWZjzex9kh6QVCjp7u5eFgAA/UlPHGmPlPQbSUMkbZP0oqTpIYR1PbAsAAD6jW4v7RCC75v1AQDAflkIvukuPcXMCiVVfWzGRGVmpCfKXnnWca5lPvBimSunVt+7C5XVe1y5XXt8U8WU54vN+OanXLkTa8e7cidl+ib/5C552JV77E8PuHL5Db4NOqPI+YtwnufRHJJP0Cqra3Ytq2BArivXtNv3WJiYn33gO3XhscpqV670CN/ylmWe5spd+dPvuHIP/M9PXbnVzz3iyunIfZ1fvH/l63y/99zcA0/B6sr6tVWu3GMvrnTlivOTPx5aWlu1uny7JA0MIexzR2VgCAAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABCJnpin3S2OHVGgnMxkq7d+83bXsoqLc1y5F5ZtdOXWbapx5Rqbm1y5y390tSt3evaJrty4bZWu3JKf/4srNyOr1pX7SPYgV+6NFt+EorezfPvZa2GIK1eb2ZI4M3bCCNey3t7lm6L0WtUqV27K7t2u3IRs31PeMbXJJg52KG5c6Mrd+LcfduWuu//XrtwzpUe4cmWP+KaKrVyx1pVbWOWbStkSfI+9cUMHu3L5+ckn7DW3tGp1+YHvx5E2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEIk+O+Xr3NNPU35udqJMaYNvutQjm9a4cidMO9aVm3vro67caWdf7srNqDnOlctcuMyVK3/iXlfudN8gJd3vnO527DG+39+y4rGu3IVf+YIrN75wlCuXk5fryh1KF8k3ua767VdduYoXF7ly/3H3ba7cZ4sGunJnNe105b517jmu3HW/fsyVy/mYK6Zta29w5RZvqnDligp8j4UCM1dOjv26SQc3GYwjbQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASfXbKV15GkfIzchJlHln0kmtZ35u/2JWbNm6jK3fs5NGu3OlTfdO6Zo71Le+lb97syp0cWly527bucuVGnDXLlZv0pbmu3Pi84a5ceqbv4VaQ65zW5Z5Q5BGcMd/fLXfy6a5c0cSTXbmrZk1x5W67+kuu3LiGelfu80N8U8V+cMUlrtz/+8kvXblnTzjVlTu55s+u3OMrNrlyg/OTTZrscPzQ4sSZxuYWSQeeYsaRNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABAJC8E5naeHmFmhpKqPTZmkzIz0RNldLbt9C830TRq6YMIwVy5r9gWu3BXnzXDl7rjss67cJfW+SUMLaqpcueLLP+HKzfj8N1y5ppYBrlxzi2+KWXNzsyuXn1/gyskx5Ms7F6yystIXdD7/1Dn3zdwc39SmnFxfrn77clfuJ1df5splrNvjyk0syHPlth01wZWb+ZU5rty937nOlStbssKV2y5nN0wclThT19SsLz3ykiQNDCFU7+t+HGkDABAJShsAgEgkLm0zO83MHjazzWYWzOz8Trebmc1tv73OzJ4xs2O7bY0BAOinPEfaeZJel3T1Pm6/VtJX22+fJmmLpMfNzPnGHAAAkKSMpIEQwjxJ8yTJ7K9PW7G2K74s6YYQwoPt110hqULSpZJ++u5WFwCA/qu739MeK2m4pPkdV4QQGiQtkHRKVwEzyzazwo6LJI7IAQDoQneX9vD2Pys6XV+Rcltn10uqSrls7OZ1AgDgsNBTZ493/vCldXFdh5skDUy5jOyhdQIAIGqJ39M+gC3tfw6XVJ5yfYn2PvqW9M7L5w0dP3d+nxwAALTp7iPtNWor7tkdV5hZlqSZkhZ287IAAOhXEh9pm1m+pPEpV401sxMk7QwhrDezWyTNMbOVklZKmiOpVtI97351AQDovzwvj0+V9HTKzze3/3m3pE9L+p6kXEm3SyqWtEjSmSGEGv9qAgAAz+e0n9F+5gqEtgkkc9svAACgm3T3iWjd5smyVYlPSivIy3Et66RjfBN8Rkw9wpVrck5t+uMP7nTlZtT4XuRIy/Gt5/OW78p96+PXuHJ19cmmwXWYP/8RV271qtWu3PAjk0/+kaRTZ/imuxUWJP/9mXPOV11dnSv36ouLXbnnnv2DK1dRvsuVKxzqm+j33X//gSv3hf96wJW75pxzXLmLinzPnT989TVXbsqqna7cqBNHuHKDdm458J268Fau7znwsV3Jl9fU3HpQ92NgCAAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABCJPjvla8bkUcrMSDa9qb6qxbWsnGZXTKFppit3/NijXLnf3fJdV+6k9CxX7s7twZW77s5fuHKN2bmu3Dev/Yor99Jrr7lyGa2+h83AgXmu3PMLn3Tl/vWbNyXOFOT4fgfla1e5cj/72S2u3NBhvqlbpcPf48qVrXnGlfvcZ/7BlbvzZ7e6clNPPdGVu/f5Ja7czCLfFKyy+c+5cp/5wS2u3D2vf9mVm1jq65Tm119InGkUU74AADisUNoAAESC0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASlDYAAJGgtAEAiESfnfL17YtmKD832XSqsqVrXMtasn23K5c2sdCV27bxDVdueEO1K7c9d6gr91KTK6bPjh7lypWVLXLlVm7yTZiadea5rtzlF13kym1au9yVu/b6f3XlHj3h6cSZj3/8LNey5v3mf1y5sSVjXbl/+rZvmxQVFbtyf/j1RFfuljt+5Mq9udL3XPbZb/sm7H199nRX7poRvufAf1v6mCun1mtdsS2jfdPrHnn4cVduw9bkndLSenBTFTnSBgAgEpQ2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACLRZ6d8Pb98g3Kzk61e5a4a17ImnjjZlWvKHeDKbVj6piv3njzf8hbt8E0Hu+lf/s2Vy8jOdOVKS49x5S677CpX7tRT3u/KDSoucuUKC1tcuaF5vglFf3np1cSZT11xqmtZg4dWunLDSo5z5UpKhrlyGRm+fXPG2ae5cnc/4Ju6tWfHDleuddwYV650VIErt6Gh1pUbmXFwE6062/yWb1Ke5bS6chNLfNtl6NijEmeampq1+rEXDng/jrQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASPTZKV/bK7YpJzM9UWbxpirXslbWlrly588scuVa1m9w5cYMyHLlHtnim8STM/49rlxudrZveSNGunIXX/hxV253jW/62ZIlr7lyryx43JVbXbHdlbv6Qx90pJI95joMKxrqyr22ao8rFzJ907ok33QpmS+3eu1aV25rxRZXrjX4Jsn97eyxrlzZA2+4cpnm288aa337y9DBvuekyrwGV27JhnWJM80tBzeJjCNtAAAiQWkDABCJxKVtZqeZ2cNmttnMgpmd3+n2u9qvT7282G1rDABAP+U50s6T9Lqkq/dzn0cllaZcPuxYDgAASJH4RLQQwjxJ8yTJzPZ1t4YQgu9MCgAA0KWeek97lpltNbMyM7vDzEr2dUczyzazwo6LpIIeWicAAKLWE6U9T9InJZ0h6RpJ0yQ9ZWb7Ouf+eklVKZeNPbBOAABEr9s/px1CuC/lx6Vm9oqkdZLOkfRgF5GbJN2c8nOBKG4AAPbS41+uEkIoN7N1kibs4/YGSe98gn0/75MDANCv9fjntM1ssKRRksp7elkAABzOEh9pm1m+pPEpV401sxMk7Wy/zJX0W7WV9BhJN0raLul3725VAQDo3zwvj0+V9HTKzx3vR98t6SpJkyV9SlKR2or7aUkXhxBq/KsJAAA8n9N+RtL+3ng+y702KYbk5yg3K9nqZQzyfan/0CG+QRUF+UWuXFGr70v9s4Lv3YwZpcWuXF6+dyiDz6E+m6Gxsc6Vu/eeX7py8x6e58oVD/EN4zjhBM/AF9/vfFSJ7zH00MKlrly9K+X92/nl5PiG/KSZ77GeNyDXlRt9zBBX7tXGZleuocX3aPdul5Y03/Iycgb6cvVbk4cYGAIAwOGF0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJDyjOQ+Jx9ZUKjMjPVHmrPeOcy1rQ2uy5XSorNntyoX6JlduR4Nvos6ySl/uhFpfLs+VOvTS0327/4z3n+bKleTmuHJPLFjoyt3z6/sTZ/7p61e7lpWefoQr15q22JVrlm+i36E27kjfdklL8x1PVVVVu3KV63y5nDTfc2edc2JhWrrzuXp7rTPX4MoV52YnzjQx5QsAgMMLpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASPTZKV8VO/YoPT3Zvyl+//ybrmUVjZ/syk0pL3fldublu3I7GltcuWU7fdPI1r65zJUbPMm3PXWIJzcVFQ125S688BJXrum8C1y5M/8wz5X70bMLEmd2yTfZrWiMb5+eOcs3mW+Ae18xZ86ntGSoK+ed8tXa6tsuN9zxiit3SXamK7dkj3OCYGGhK7eu3PccmFuY5crlVCefKpYemPIFAMBhhdIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABAJShsAgEhQ2gAARILSBgAgEpQ2AACRoLQBAIgEpQ0AQCT67JSvLdvrlZaWbCJPxuCBrmWteWO5K3f8R5NPcpGk4kG+iUgbm+pcuZmDcly5P9z3G1fuuPM+5sq1tO505eY/9pwrd/x73+fKlZaWuHJpzglTazZsduWqqrYlzjQ0vOFa1ojJW1y5cbsObrJRZ88//rQr9773TXPlnv3TU66cahtcsYEDfc9laQc5Kaqzqh31rtzQMXmuXHmDb8pXRoFvytfI1iZX7rm1vv36ybLkz2UhHNyENo60AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABAJShsAgEj02Slfw0uKlJ6e7N8UBWm+v862mj2uXKhId+VmzJjhyj3+xwdduQ/l+7ZLWYVv6lZanW+Cjwb4ppGVvfaqK/fw/X9w5U7/4NmuXMXbK125xxfMc+VO+9ApiTOD06pdywqtNa5ckW/gne65+79cucVPznflli9b68pl5Q9z5U468XhXruoN32S+idm+57Ka5oObTNXZ8Uf7pq3d+asfu3JNVRWunLKzXbGTJiT/vTe3tGrhsvID3o8jbQAAIkFpAwAQiUSlbWbXm9nLZlZjZlvN7CEzO7rTfczM5prZZjOrM7NnzOzY7l1tAAD6n6RH2jMl3SZpuqTZantPfL6Z5aXc51pJX5V0taRpkrZIetzMCt796gIA0H8lOkMphPBXZ+CY2WckbZU0RdKzZmaSvizphhDCg+33uUJShaRLJf20G9YZAIB+6d2+pz2w/c+O04zHShou6Z1TNEMIDZIWSEp+KisAAHiH+yNf7UfVN0t6PoSwtP3q4e1/dj6/vkLSkfv4/2RLSj2vnpfRAQDowrs50r5V0vGSPtHFbZ0/vGddXNfheklVKZeN72KdAAA4bLlK28x+LOkjkk4PIaSW7Jb2P4d3ipRo76PvDjep7WX2jstIzzoBAHC4S/qRLzOzWyV9TNIZIYQ1ne6yRm3FPTslk6W2s84XdvX/DCE0hBCqOy6SfF+tBADAYS7pe9q3qe0s8PMk1ZhZxxF1VQihLoQQzOwWSXPMbKWklZLmSKqVdE83rTMAAP1S0tK+qv3PZzpd/xlJd7X/9/ck5Uq6XVKxpEWSzgwhcAQNAMC7kPRz2nYQ9wmS5rZfAABAN+mzU76y09KUkZZs8syiVRtcy5r53tGuXHr5867coiLfBJ/hx01x5fLXLnflzm71TT+7/eqrDnynLvzjnXe4cpMmnejKrVixzpX7xc9vc+WqahpcuQ86p4p9/KKLEmdCq28qVXZWiStXemSZKzd16mBXbmuFb59+32m+qVSnf/gjrpw5Jw/+8NofunJfGTnIlbuhbLsrd+OdX3Llvn/XDa7c0rVbDnynLqzY6HuBuHxHbeJM2/HugTEwBACASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABAJShsAgEhQ2gAARILSBgAgEpQ2AACRoLQBAIhEn53ytWV7ldLTkv2bYuigga5llQ7Jd+XWrfBN+dp+xARX7qtXftyVu/VK30Sdrw31TW5atGqRK1e3/lVX7tyLfNtl0lTf5Kby8k2uXEZmtis3YcJEV66gIPl+nZnpfEoIQ12x3NwiV+7ciz/oymVkZDpzySYOdthdVe7Krb1/jis3Lcv3+1tRXe/KnTVrpitXmeWbYta0aYcrt6PS9/c7eYxv+tmFnzghcaa2oVlX3LbwgPfjSBsAgEhQ2gAARILSBgAgEpQ2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIhIUQensd/oqZFUqquuiDJyozM9lknYljh7mWWTBggCv3ypI1rlzxUWe5clNnHOPKnVy53JVb9Zt7XLnpeb7t+a2tvsk/c37+X67ciPee7srJfBOfZL6YO+cK9q3ng32p3VPnyqWn+45Tams2u3KLbv9HV65s3jJX7pOlQ1y5H272PfZufOgxV+4bN37WlXvwkddcucrqRlfuxnPf68pNspbEmT2NzfrAPa9I0sAQQvW+7seRNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABCJjN5egX2ZccpRys3JTJQZN7zEtawnX33blcsYXOzK2a5Vrtziv/immJ33mU+4ck8/8bQrd2T1dlduzrA8V+6rF13myn3xk1e4cqVnXuDKDT1mrCvXrGSPgw6FAwsTZ1pbW13L8uZ2bN3qytVWV7pym55Y5MpVvfEzV27da+Wu3CdLh7pyf6j0jYS79MpbXbk53/hnV27bSt9zbnNj8ulZkjRjymhXbndGsyu3oGJ34kx908H93TjSBgAgEpQ2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACLRZ6d8KT1HSs9KFFnw0lrXoo6fMMaVW7/5VVeuvLLGldu4psyV+9nRvl/zVT+93ZW76ZOXu3IzGhtcuV9MHunK/fe8B125n9//K1duxpQTXbmjTv6oK1dtexJn0tPTXcuq3VPryq1+5XVXbsOql1y5gfWNrtznRvsm+p0+cogr9531vulSV17rm0b2yDLfRL/1G5e5crvk288K8nNduZpq33PLSSW+yYqlrcn3s5rGFv3rQdyPI20AACJBaQMAEIlEpW1m15vZy2ZWY2ZbzewhMzu6033uMrPQ6fJi9642AAD9T9Ij7ZmSbpM0XdJstb0nPt/M8jrd71FJpSmXD7/L9QQAoN9LdIZSCOHs1J/N7DOStkqaIunZlJsaQghb3v3qAQCADu/2Pe2B7X/u7HT9rPaXz8vM7A4zK3mXywEAoN9zf+TLzEzSzZKeDyEsTblpnqT7Ja2TNFbStyU9ZWZTQgh7nXdvZtmSslOuKvCuEwAAh7N38zntWyUdL+n9qVeGEO5L+XGpmb2itgI/R1JXH4y9Xjqoj6cBANCvuV4eN7MfS/qIpNNDCBv3d98QQrnaSnvCPu5yk9peZu+4+L4pAwCAw1yiI+32l8R/LOmjkmaFENYcRGawpFGSyru6vf0l84aU+ydZJQAA+o2kR9q3SbpM0qWSasxsePslV5LMLN/MfmBmf2NmY8xslqSHJW2X9LtuXG8AAPqdpO9pX9X+5zOdrv+MpLsktUiaLOlTkorUdnT9tKSLQwi+L9wGAACSkn9Oe7+vXYcQ6iSd9a7WCAAAdKnPTvn60V2PKz0t2av3Jx13nGtZM2f4JvE8+/LbrlzdnnpXrjn4Jv/84k7fyflpGV925a755X+7cj+55jpX7u7Xlx74Tl04v9h3/sQ1o337S/P2Lk/rOKBHf/EdV668Lvlko/qWVtey6ptbXLlj8rIPfKcuXDm085cwHpyMYt/yXqxNPjFNkhbsznflPv3Fm125Xz57vytXX/uEK7dys+87tBy7piQpN9s3HcwUXLnSsb5OGZdzwNO99lJd3yRpyQHvx8AQAAAiQWkDABAJShsAgEhQ2gAARILSBgAgEpQ2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBIWgm/6SU8xs0JJVbd9/4vKzU02kacgyzf5pzSv0pW774EXXLmXl29y5QYPKnTlSoeXuHLVu33bxfJHuXInjTvWlbv2K5e7cj/59hxXbsPyDa7cRPNNihpivslGRQmn5ElSerpv8N+Oet/kuiU1Ta5c/hDfPr2y2fd8991//wdX7qm/+CbzvfDYG67cxqqFrlzZKt9z0orN1a5cUcEAV66h2be/fO3Mk1y50wb4JgEuz0m+n9U1NuvzP31ekgaGEPa5YTnSBgAgEpQ2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIBKUNAEAkKG0AACLhG+lzCFxw8RUqLMxPlMnNzXUtq3Ljy67cKWW+6UuhxLfZL511nCv3nqmTXbkFT85z5dZs9k0MSsv0TWnb3DzQlWuceKord9b5R7tyVRUrXbldm7a7chWbkv8ejnBOkltZ6Zv2NHTUeFdu6vTTXDl74WlX7v//5jVXriHN99yyYUONK6cBvulZp072PUccOcj3WH+2zJer3FPnyv3PX1a5cqPOO96Ve9IxCbCxqeWg7seRNgAAkaC0AQCIBKUNAEAkKG0AACJBaQMAEAlKGwCASFDaAABEgtIGACASlDYAAJGgtAEAiASlDQBAJChtAAAiQWkDABCJPjvla9NDv1dVbk6iTE2m76/z5HOPunJ3zPdNjvnY5+a4coOzfROYfvTgQ65cTm6RK9dYk+nK1Q7KcuXuefRJV27HytWu3OOvr3DlXlj7lit31uyzXbni0cmnwq2xJteytqze5cq9+fIjrtxDv3vCldte+YYrt2rrwU1g6uz9kye5coVZvsfCs0vfduUqBiWbqNjh1Aklrtxr68yVU/Btl6lHD3Pl6gf6nnPL1u1InGluaT2o+3GkDQBAJChtAAAiQWkDABAJShsAgEhQ2gAARILSBgAgEpQ2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESiz075+uMTjykn4dSu7z/1umtZU8YOd+UGFfsm4wzIqXLlapp8/8a67fu3unK7d9e6cjPfN8WVW1++xZVrbmp25WS+SUP19b7t0txQ7cpltja4cicNGpQ4s6y8xrWsc8652JWr3rHTlVu8fLEvt9H39xs7Yqgrl2O+ffOi03zTwdZt2+rKzTzlGFdu1EDfY+jSHN8kwMXLkk/PkqTcQt9z/CklvorMufikxJna+ia9tKT8gPfjSBsAgEhQ2gAARCJRaZvZVWa2xMyq2y8vmNmHUm43M5trZpvNrM7MnjGzY7t/tQEA6H+SHmlvlPR1SVPbL09J+n1KMV8r6auSrpY0TdIWSY+bWUH3rC4AAP1XotIOITwcQvhTCKGs/fLPknZLmm5mJunLkm4IITwYQlgq6QpJAyRd2t0rDgBAf+N+T9vM0s3sEkl5kl6QNFbScEnzO+4TQmiQtEDSKfv5/2SbWWHHRRJH5QAAdCFxaZvZZDPbLalB0n9K+mgIYZnaCluSKjpFKlJu68r1kqpSLhuTrhMAAP2B50j7LUknSJou6SeS7jaz1A8Vhk73ty6uS3WTpIEpl5GOdQIA4LCX+JPjIYRGSW+3//iKmU2T9CVJ322/brik1E+Il2jvo+/U/1+D2o7aJUnm/MILAAAOd93xOW2TlC1pjdrOFp/9zg1mWZJmSlrYDcsBAKBfS3SkbWY3SponaYPaThi7RNIsSWeHEIKZ3SJpjpmtlLRS0hxJtZLu6cZ1BgCgX0r68vgwSb+SVKq2k8aWqK2wH2+//XuSciXdLqlY0iJJZ4YQfF/4CwAA3pGotEMIf3eA24Okue0XAADQjfrslK8xJ43XgJysRJkbJx3hWtaTL6925dav2+f5dfs1eNwQV270+ye4clPeM8yV27zNN80qx+pcuQ9MH+/KrXx7lSv358XrXLmxw4tcue31rpheXr7SlZsw5ajEmRdXb3ct69wjT3TlPlDom4JVvt63TXKyfZPWPnfmdFduV6XvMXTT75515f7jsx868J26UFDkO71pQ4ZvStsjr/iec/+8yfec+8nbL3HliscMduUurEw+xax6d50099ED3o+BIQAARILSBgAgEpQ2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIRJ8dGFJX35Q805A8I0lNzS2uXEtLqytXV+sbqFFTs8eVaz7Efz/v9mxs9A2PaG72rWfbULrkWlp9y2t1Li+0+nKNjt9Dq/PvVlfn2zez6n1TVJqc+6b3d17f6HtuaWhy7tPOv9/u+kZXzup8x297Mg7tc26r87FQu6fBlauu8e2f6buT/96r9xzcssy7E/cUMztC0sbeXg8AAHrByBDCpn3d2BdL2ySNkFTTxc0Faiv0kfu4vb9iu3SN7dI1tsve2CZdY7t0rae2S4GkzWE/xdznXh5vX9ku/5XR1ueSpJoQgm8w7mGI7dI1tkvX2C57Y5t0je3StR7cLgf8f3EiGgAAkaC0AQCIRGyl3SDpW+1/4v+wXbrGduka22VvbJOusV261mvbpc+diAYAALoW25E2AAD9FqUNAEAkKG0AACJBaQMAEImoStvMvmBma8ys3sxeNbNTe3udepOZzTWz0OmypbfX61Azs9PM7GEz29y+Dc7vdLu1b6vNZlZnZs+Y2bG9tLqHxEFsk7u62Hde7KXVPSTM7Hoze9nMasxsq5k9ZGZHd7pPf9xXDma79Mf95SozW2Jm1e2XF8zsQym398q+Ek1pm9nFkm6RdIOkEyU9J2memY3uzfXqA96UVJpymdy7q9Mr8iS9Lunqfdx+raSvtt8+TdIWSY+bWcGhWb1ecaBtIkmP6q/3nQ8fgvXqTTMl3SZpuqTZavtGyPlmlpdyn/64rxzMdpH63/6yUdLXJU1tvzwl6fcpxdw7+0oIIYqLpEWSftLpuuWSburtdevFbTJX0mu9vR596SIpSDo/5WeTVC7pupTrsiXtkvT3vb2+vbFN2q+7S9JDvb1uvbxdhrZvm9PYV/a9Xdhf/mrb7JT0d725r0RxpG1mWZKmSJrf6ab5kk459GvUp0xof3lmjZnda2bjenuF+pixkoYrZd8JITRIWiD2nVntL4eWmdkdZlbS2yt0iA1s/3Nn+5/sK206b5cO/XZ/MbN0M7tEba9gvaBe3FeiKG1JQySlS6rodH2F2jZcf7VI0qcknSXpSrVti4VmNrhX16pv6dg/2Hf+2jxJn5R0hqRr1Pby3lNmlt2ra3WItE8TvFnS8yGEpe1X9/t9ZR/bReqn+4uZTTaz3Wr75rP/lPTREMIy9eK+0uemfB1A569vsy6u6zdCCPNSfnzDzF6QtErSFWp74OH/sO+kCCHcl/LjUjN7RdI6SedIerB31uqQulXS8ZLe38Vt/Xlf6XK79OP95S1JJ0gqknSBpLvNbGbK7Yd8X4nlSHu7pBbt/S+YEu39L51+K4SwR9Ibkib09rr0IR1n07Pv7EcIoVxtT8KH/b5jZj+W9BFJp4cQNqbc1K/3lf1sl730l/0lhNAYQng7hPBKCOF6tZ3c+SX14r4SRWmHEBolvaq2MxtTzZa08NCvUd/U/lLVMWo7QQJt1qjtAfbOvtN+jsRMse+8o/0tlVE6jPed9o/o3CrpY5LOCCGs6XSXfrmvHMR26Spz2O8v+2BqO+Gs1/aVmF4ev1nSr9pflnlB0ucljVbb+wz9kpn9QNLDktar7V9435BUKOnu3lyvQ83M8iWNT7lqrJmdIGlnCGG9md0iaY6ZrZS0UtIcSbWS7jnU63qo7G+btF/mSvqt2p50x0i6UW2vaP3uUK7nIXabpEslnSepxsw6jpKqQgh1IYTQH/cVHWC7tO9Lc9XP9hczu1Ft7+VvkFQg6RJJsySd3av7Sm+fQp/wdPsvSFqrtpMCXlXKRxL640XSvZI2S2qUtEltD6pJvb1evbAdZqntfaTOl7vabze1PemUS6pX2xmex/X2evfWNpGUK+kxSVvb95117deP6u317uFt0tX2CJI+nXKf/riv7He79OP95WcpfbNV0hOSZvf2vsJoTgAAIhHFe9oAAIDSBgAgGpQ2AACRoLQBAIgEpQ0AQCQobQAAIkFpAwAQCUobAIBIUNoAAESC0gYAIBKUNgAAkaC0AQCIxP8CsR0wOhpwJYEAAAAASUVORK5CYII=",
      "text/plain": [
       "<Figure size 700x500 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "io.imshow(inputs[4])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## <span style=\"color:#900C3F\">Fazit</span> \n",
    "\n",
    "Auch der letzte Versuch aus der Reihe war wieder sehr interessant. Wir haben mehr darüber gelernt wie CNN funktionieren. Vor allem konnte man das sehen, beim Betrachten der Werte u.a. der Accuracy, die von Epoche zu Epoche stetig stiegen. Genau wie bei der Gesichtserkennung funktioniert die Klassifizierung erstaunlich gut, wenn man auch die Punkte der unbalance\n",
    ", wie im vorherigen Dokuabschnitt beschrieben, beachten muss. \n",
    "\n",
    "\n",
    "Abschließend können wir sagen, dass das ganze Modul sich wirklich gelohnt hat. Auch wenn es stimmt, was am Anfang gesagt wurde, dass es sehr zeitintensiv sein wird, hat man umso mehr gelernt. Geholfen haben dabei natürlich auch die sehr spannenden Themen z.B. der Versuch der Empfehlungssysteme/ Musikempfehlungen, die einem jeden Tag begegnen und bei denen man sich jetzt besser vorstellen kann wie sie funktionieren. \n",
    "\n",
    "Wir danken Ihnen an dieser Stelle für die Möglichkeit, diese Versuche durchzuführen. :)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  },
  "nav_menu": {},
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}