Assessment.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a href=\"https://www.nvidia.com/en-us/deep-learning-ai/education/\"> <img src=\"images/DLI Header.png\" alt=\"Header\" style=\"width: 400px;\"/> </a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "L53DPe5wAy3x"
   },
   "source": [
    "# Workshop Assessment "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "K11_XvhuA1A0"
   },
   "source": [
    "Welcome to the assessment section of this course. In the previous labs you successfully applied machine learning and deep learning techniques for the task of anomaly detection on network packet data. Equipped with this background, you can apply these techniques to any type of data (images or audio) across different use cases. In this assessment, you will apply supervised and unsupervised techniques for intrusion detection on the NSL KDD dataset."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If you are successfully able to complete this assessment, you will be able to generate a certificate of competency for the course. Good luck!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Objectives"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "K11_XvhuA1A0"
   },
   "source": [
    "This assessment seeks to test the following concepts:\n",
    "\n",
    "1.   Building and training an Xgboost model.\n",
    "2.   Building and training an autoencoder neural network.\n",
    "3.   Detecting anomalies using different thresholding methods.\n",
    "\n",
    "The total duration of the assessment is 2 hrs, however, if you are unable to complete the assessment today, you are more than welcome to return to it at a later time to try and complete it then."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Tk6sSnV9OqRt"
   },
   "source": [
    "## Section 1: Preparation - Done for You"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Tk6sSnV9OqRt"
   },
   "source": [
    "### The Dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "0npSD6otPj_A"
   },
   "source": [
    "We will be using the NSL-KDD dataset published by the University of New Brunswick in this assessment. While the dataset is similar to the KDD dataset used throughout the workshop in terms of the features used, it varies in the following respects:\n",
    "\n",
    "1.   Removal of redundant and duplicate records in the dataset to prevent classifiers from overfitting a particular class.\n",
    "2.   The number of selected records from each difficulty level group is inversely proportional to the percentage of records in the original KDD data set making the task of unsupervised classification slightly more challenging."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "nI-iMkWrvhel"
   },
   "source": [
    "### Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "LiJdr60kqm7t"
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import os\n",
    "import random as python_random\n",
    "\n",
    "import xgboost as xgb\n",
    "import matplotlib.pyplot as plt\n",
    "import pickle\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "\n",
    "from tensorflow.keras import optimizers\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.layers import Input, Dense, Dropout\n",
    "from tensorflow.keras.utils import plot_model\n",
    "\n",
    "from collections import OrderedDict\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import roc_curve, auc, confusion_matrix\n",
    "from sklearn.metrics import roc_auc_score,confusion_matrix,classification_report,roc_curve\n",
    "\n",
    "# We will use our own accuracy score functions for the sake of grading this assessment\n",
    "from assessment import xgb_accuracy_score, autoencoder_accuracy_score\n",
    "\n",
    "from tensorflow.keras.models import load_model, model_from_json\n",
    "\n",
    "np.random.seed(42)\n",
    "python_random.seed(42)\n",
    "tf.random.set_seed(42)\n",
    "os.environ['PYTHONHASHSEED']=str(42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load the Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "sbxUus80Gh7k"
   },
   "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>duration</th>\n",
       "      <th>protocol_type</th>\n",
       "      <th>service</th>\n",
       "      <th>flag</th>\n",
       "      <th>src_bytes</th>\n",
       "      <th>dst_bytes</th>\n",
       "      <th>land</th>\n",
       "      <th>wrong_fragment</th>\n",
       "      <th>urgent</th>\n",
       "      <th>hot</th>\n",
       "      <th>...</th>\n",
       "      <th>dst_host_srv_count</th>\n",
       "      <th>dst_host_same_srv_rate</th>\n",
       "      <th>dst_host_diff_srv_rate</th>\n",
       "      <th>dst_host_same_src_port_rate</th>\n",
       "      <th>dst_host_srv_diff_host_rate</th>\n",
       "      <th>dst_host_serror_rate</th>\n",
       "      <th>dst_host_srv_serror_rate</th>\n",
       "      <th>dst_host_rerror_rate</th>\n",
       "      <th>dst_host_srv_rerror_rate</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>tcp</td>\n",
       "      <td>ftp_data</td>\n",
       "      <td>SF</td>\n",
       "      <td>491</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>25</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.00</td>\n",
       "      <td>normal</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>udp</td>\n",
       "      <td>other</td>\n",
       "      <td>SF</td>\n",
       "      <td>146</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>1</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.60</td>\n",
       "      <td>0.88</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>normal</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>tcp</td>\n",
       "      <td>private</td>\n",
       "      <td>S0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>26</td>\n",
       "      <td>0.10</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>neptune</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>tcp</td>\n",
       "      <td>http</td>\n",
       "      <td>SF</td>\n",
       "      <td>232</td>\n",
       "      <td>8153</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>255</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.01</td>\n",
       "      <td>normal</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>tcp</td>\n",
       "      <td>http</td>\n",
       "      <td>SF</td>\n",
       "      <td>199</td>\n",
       "      <td>420</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>255</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>normal</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 42 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   duration protocol_type   service flag  src_bytes  dst_bytes  land  \\\n",
       "0         0           tcp  ftp_data   SF        491          0     0   \n",
       "1         0           udp     other   SF        146          0     0   \n",
       "2         0           tcp   private   S0          0          0     0   \n",
       "3         0           tcp      http   SF        232       8153     0   \n",
       "4         0           tcp      http   SF        199        420     0   \n",
       "\n",
       "   wrong_fragment  urgent  hot  ...  dst_host_srv_count  \\\n",
       "0               0       0    0  ...                  25   \n",
       "1               0       0    0  ...                   1   \n",
       "2               0       0    0  ...                  26   \n",
       "3               0       0    0  ...                 255   \n",
       "4               0       0    0  ...                 255   \n",
       "\n",
       "   dst_host_same_srv_rate  dst_host_diff_srv_rate  \\\n",
       "0                    0.17                    0.03   \n",
       "1                    0.00                    0.60   \n",
       "2                    0.10                    0.05   \n",
       "3                    1.00                    0.00   \n",
       "4                    1.00                    0.00   \n",
       "\n",
       "   dst_host_same_src_port_rate  dst_host_srv_diff_host_rate  \\\n",
       "0                         0.17                         0.00   \n",
       "1                         0.88                         0.00   \n",
       "2                         0.00                         0.00   \n",
       "3                         0.03                         0.04   \n",
       "4                         0.00                         0.00   \n",
       "\n",
       "   dst_host_serror_rate  dst_host_srv_serror_rate  dst_host_rerror_rate  \\\n",
       "0                  0.00                      0.00                  0.05   \n",
       "1                  0.00                      0.00                  0.00   \n",
       "2                  1.00                      1.00                  0.00   \n",
       "3                  0.03                      0.01                  0.00   \n",
       "4                  0.00                      0.00                  0.00   \n",
       "\n",
       "   dst_host_srv_rerror_rate    label  \n",
       "0                      0.00   normal  \n",
       "1                      0.00   normal  \n",
       "2                      0.00  neptune  \n",
       "3                      0.01   normal  \n",
       "4                      0.00   normal  \n",
       "\n",
       "[5 rows x 42 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "col_names = [\"duration\",\"protocol_type\",\"service\",\"flag\",\"src_bytes\",\"dst_bytes\",\"land\",\"wrong_fragment\",\"urgent\",\"hot\",\"num_failed_logins\",\"logged_in\",\n",
    "             \"num_compromised\",\"root_shell\",\"su_attempted\",\"num_root\",\"num_file_creations\",\"num_shells\",\"num_access_files\",\"num_outbound_cmds\",\n",
    "             \"is_host_login\",\"is_guest_login\",\"count\",\"srv_count\",\"serror_rate\",\"srv_serror_rate\",\"rerror_rate\",\"srv_rerror_rate\",\"same_srv_rate\",\n",
    "             \"diff_srv_rate\",\"srv_diff_host_rate\",\"dst_host_count\",\"dst_host_srv_count\",\"dst_host_same_srv_rate\",\"dst_host_diff_srv_rate\",\n",
    "             \"dst_host_same_src_port_rate\",\"dst_host_srv_diff_host_rate\",\"dst_host_serror_rate\",\"dst_host_srv_serror_rate\",\"dst_host_rerror_rate\",\n",
    "             \"dst_host_srv_rerror_rate\",\"label\"]\n",
    "\n",
    "df = pd.read_csv(\"data/KDDTrain+_20Percent.txt\", header=None, names=col_names, index_col=False)\n",
    "\n",
    "text_l = ['protocol_type', 'service', 'flag', 'land', 'logged_in', 'is_host_login', 'is_guest_login']\n",
    "\n",
    "df.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "jeFncMHLLRsD"
   },
   "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",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>normal</th>\n",
       "      <td>13449</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>neptune</th>\n",
       "      <td>8282</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ipsweep</th>\n",
       "      <td>710</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>satan</th>\n",
       "      <td>691</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>portsweep</th>\n",
       "      <td>587</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>smurf</th>\n",
       "      <td>529</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>nmap</th>\n",
       "      <td>301</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>back</th>\n",
       "      <td>196</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teardrop</th>\n",
       "      <td>188</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>warezclient</th>\n",
       "      <td>181</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>pod</th>\n",
       "      <td>38</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>guess_passwd</th>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>warezmaster</th>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>buffer_overflow</th>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>imap</th>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>rootkit</th>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>phf</th>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>multihop</th>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ftp_write</th>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>spy</th>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>loadmodule</th>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>land</th>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                 label\n",
       "normal           13449\n",
       "neptune           8282\n",
       "ipsweep            710\n",
       "satan              691\n",
       "portsweep          587\n",
       "smurf              529\n",
       "nmap               301\n",
       "back               196\n",
       "teardrop           188\n",
       "warezclient        181\n",
       "pod                 38\n",
       "guess_passwd        10\n",
       "warezmaster          7\n",
       "buffer_overflow      6\n",
       "imap                 5\n",
       "rootkit              4\n",
       "phf                  2\n",
       "multihop             2\n",
       "ftp_write            1\n",
       "spy                  1\n",
       "loadmodule           1\n",
       "land                 1"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Describe the different classes of Labels\n",
    "\n",
    "pd.DataFrame(df['label'].value_counts())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "2F3BqL2Lu1PN"
   },
   "source": [
    "### Data Preprocessing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "lsxg_35OMOlN"
   },
   "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>land</th>\n",
       "      <th>logged_in</th>\n",
       "      <th>is_host_login</th>\n",
       "      <th>is_guest_login</th>\n",
       "      <th>protocol_type_icmp</th>\n",
       "      <th>protocol_type_tcp</th>\n",
       "      <th>protocol_type_udp</th>\n",
       "      <th>service_IRC</th>\n",
       "      <th>service_X11</th>\n",
       "      <th>service_Z39_50</th>\n",
       "      <th>...</th>\n",
       "      <th>flag_REJ</th>\n",
       "      <th>flag_RSTO</th>\n",
       "      <th>flag_RSTOS0</th>\n",
       "      <th>flag_RSTR</th>\n",
       "      <th>flag_S0</th>\n",
       "      <th>flag_S1</th>\n",
       "      <th>flag_S2</th>\n",
       "      <th>flag_S3</th>\n",
       "      <th>flag_SF</th>\n",
       "      <th>flag_SH</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 84 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   land  logged_in  is_host_login  is_guest_login  protocol_type_icmp  \\\n",
       "0     0          0              0               0                   0   \n",
       "1     0          0              0               0                   0   \n",
       "2     0          0              0               0                   0   \n",
       "3     0          1              0               0                   0   \n",
       "4     0          1              0               0                   0   \n",
       "\n",
       "   protocol_type_tcp  protocol_type_udp  service_IRC  service_X11  \\\n",
       "0                  1                  0            0            0   \n",
       "1                  0                  1            0            0   \n",
       "2                  1                  0            0            0   \n",
       "3                  1                  0            0            0   \n",
       "4                  1                  0            0            0   \n",
       "\n",
       "   service_Z39_50  ...  flag_REJ  flag_RSTO  flag_RSTOS0  flag_RSTR  flag_S0  \\\n",
       "0               0  ...         0          0            0          0        0   \n",
       "1               0  ...         0          0            0          0        0   \n",
       "2               0  ...         0          0            0          0        1   \n",
       "3               0  ...         0          0            0          0        0   \n",
       "4               0  ...         0          0            0          0        0   \n",
       "\n",
       "   flag_S1  flag_S2  flag_S3  flag_SF  flag_SH  \n",
       "0        0        0        0        1        0  \n",
       "1        0        0        0        1        0  \n",
       "2        0        0        0        0        0  \n",
       "3        0        0        0        1        0  \n",
       "4        0        0        0        1        0  \n",
       "\n",
       "[5 rows x 84 columns]"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Create one-hot encoded categorical columns in the dataset\n",
    "\n",
    "cat_vars = ['protocol_type', 'service', 'flag', 'land', 'logged_in','is_host_login', 'is_guest_login']\n",
    "\n",
    "# Find unique labels for each category\n",
    "cat_data = pd.get_dummies(df[cat_vars])\n",
    "\n",
    "# Check that the categorical variables were created correctly\n",
    "cat_data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "okwxsa65MVF_"
   },
   "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>dst_host_srv_rerror_rate</th>\n",
       "      <th>root_shell</th>\n",
       "      <th>duration</th>\n",
       "      <th>dst_host_diff_srv_rate</th>\n",
       "      <th>same_srv_rate</th>\n",
       "      <th>hot</th>\n",
       "      <th>src_bytes</th>\n",
       "      <th>dst_host_srv_count</th>\n",
       "      <th>srv_count</th>\n",
       "      <th>urgent</th>\n",
       "      <th>...</th>\n",
       "      <th>dst_host_srv_serror_rate</th>\n",
       "      <th>dst_host_same_src_port_rate</th>\n",
       "      <th>dst_host_serror_rate</th>\n",
       "      <th>dst_host_rerror_rate</th>\n",
       "      <th>rerror_rate</th>\n",
       "      <th>srv_rerror_rate</th>\n",
       "      <th>dst_bytes</th>\n",
       "      <th>num_root</th>\n",
       "      <th>num_shells</th>\n",
       "      <th>dst_host_same_srv_rate</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.00</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.03</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0</td>\n",
       "      <td>491</td>\n",
       "      <td>25</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.17</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.17</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.00</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.60</td>\n",
       "      <td>0.08</td>\n",
       "      <td>0</td>\n",
       "      <td>146</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.88</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.00</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>26</td>\n",
       "      <td>6</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.01</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0</td>\n",
       "      <td>232</td>\n",
       "      <td>255</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>8153</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.00</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0</td>\n",
       "      <td>199</td>\n",
       "      <td>255</td>\n",
       "      <td>32</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>420</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1.00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 34 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   dst_host_srv_rerror_rate  root_shell  duration  dst_host_diff_srv_rate  \\\n",
       "0                      0.00           0         0                    0.03   \n",
       "1                      0.00           0         0                    0.60   \n",
       "2                      0.00           0         0                    0.05   \n",
       "3                      0.01           0         0                    0.00   \n",
       "4                      0.00           0         0                    0.00   \n",
       "\n",
       "   same_srv_rate  hot  src_bytes  dst_host_srv_count  srv_count  urgent  ...  \\\n",
       "0           1.00    0        491                  25          2       0  ...   \n",
       "1           0.08    0        146                   1          1       0  ...   \n",
       "2           0.05    0          0                  26          6       0  ...   \n",
       "3           1.00    0        232                 255          5       0  ...   \n",
       "4           1.00    0        199                 255         32       0  ...   \n",
       "\n",
       "   dst_host_srv_serror_rate  dst_host_same_src_port_rate  \\\n",
       "0                      0.00                         0.17   \n",
       "1                      0.00                         0.88   \n",
       "2                      1.00                         0.00   \n",
       "3                      0.01                         0.03   \n",
       "4                      0.00                         0.00   \n",
       "\n",
       "   dst_host_serror_rate  dst_host_rerror_rate  rerror_rate  srv_rerror_rate  \\\n",
       "0                  0.00                  0.05          0.0              0.0   \n",
       "1                  0.00                  0.00          0.0              0.0   \n",
       "2                  1.00                  0.00          0.0              0.0   \n",
       "3                  0.03                  0.00          0.0              0.0   \n",
       "4                  0.00                  0.00          0.0              0.0   \n",
       "\n",
       "   dst_bytes  num_root  num_shells  dst_host_same_srv_rate  \n",
       "0          0         0           0                    0.17  \n",
       "1          0         0           0                    0.00  \n",
       "2          0         0           0                    0.10  \n",
       "3       8153         0           0                    1.00  \n",
       "4        420         0           0                    1.00  \n",
       "\n",
       "[5 rows x 34 columns]"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Separate the numerical columns\n",
    "\n",
    "numeric_vars = list(set(df.columns.values.tolist()) - set(cat_vars))\n",
    "numeric_vars.remove('label')\n",
    "numeric_data = df[numeric_vars].copy()\n",
    "\n",
    "# Check that the numeric data has been captured accurately\n",
    "\n",
    "numeric_data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "5cx8YpFAMYss"
   },
   "outputs": [],
   "source": [
    "numeric_cat_data = pd.concat([numeric_data, cat_data], axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "wkWZO8Gevx6I"
   },
   "source": [
    "## Assessment Task 1: Data Selection\n",
    "\n",
    "The first part of this assessment checks whether you understand the data you are working with. If successful, you should be able to load and split the data in order to begin learning from it.\n",
    "\n",
    "In the code block below, replace each #### FIX ME #### with solutions which:\n",
    "1.   Determine the number of classes in the dataset.\n",
    "2.   Set the variable test_size to the fraction of the dataset you would like to use for testing."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "g3Swkp2xMh_N"
   },
   "outputs": [],
   "source": [
    "# Capture the labels\n",
    "labels = df['label'].copy()\n",
    "\n",
    "# Convert labels to integers\n",
    "le = LabelEncoder()\n",
    "integer_labels = le.fit_transform(labels)\n",
    "num_labels = len(np.unique(integer_labels))\n",
    "\n",
    "# Split data into test and train\n",
    "x_train, x_test, y_train, y_test = train_test_split(numeric_cat_data,\n",
    "                                                    integer_labels,\n",
    "                                                    test_size= 0.2, \n",
    "                                                    random_state= 42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "RXkWHSGKMieW"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(20153, 118)\n",
      "(20153,)\n",
      "(5039, 118)\n",
      "(5039,)\n"
     ]
    }
   ],
   "source": [
    "print(x_train.shape)\n",
    "print(y_train.shape)\n",
    "print(x_test.shape)\n",
    "print(y_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "JO4wJzNc4Nin"
   },
   "outputs": [],
   "source": [
    "# Make sure to only fit the the scaler on the training data\n",
    "scaler = MinMaxScaler()\n",
    "x_train = scaler.fit_transform(x_train)\n",
    "x_test = scaler.transform(x_test)\n",
    "\n",
    "# Convert the data to FP32\n",
    "x_train = x_train.astype(np.float32)\n",
    "x_test = x_test.astype(np.float32)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "76kCWgjL5K13"
   },
   "source": [
    "## Assessment Task 2 : XGBoost - Set the XGBoost Parameters"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "76kCWgjL5K13"
   },
   "source": [
    "Treat the question as a **multi-class** supervised learning problem and train a **GPU-accelerated XGBoost model** on the given dataset. Refer to the [documentation](https://xgboost.readthedocs.io/en/latest/parameter.html) or your previous tasks to fix the parameter list. You may reference the notebooks from previous sections by opening the file explorer on the left-hand side of the JupyterLab screen.\n",
    "\n",
    "This task checks that you know how these parameters impact training."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "LJ56HUZDEoQC"
   },
   "outputs": [],
   "source": [
    " params = {\n",
    "    'num_rounds':        10,\n",
    "    'max_depth':         8,\n",
    "    'max_leaves':        2**8,\n",
    "    'alpha':             0.9,\n",
    "    'eta':               0.1,\n",
    "    'gamma':             0.1,\n",
    "    'learning_rate':     0.1,\n",
    "    'subsample':         1,\n",
    "    'reg_lambda':        1,\n",
    "    'scale_pos_weight':  2,\n",
    "    'tree_method':       'gpu_hist',\n",
    "    'n_gpus':            1,\n",
    "    'objective':         'multi:softprob',\n",
    "    'num_class':         num_labels,\n",
    "    'verbose':           True\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "2mOlFfUe9DGC"
   },
   "source": [
    "## Assessment Task 3: Model Training"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "2mOlFfUe9DGC"
   },
   "source": [
    "In this next task, you will prove that you can build and fit an accelerated XGBoost Model.\n",
    "1.   Initiate training by referring to the [XGBoost API](https://xgboost.readthedocs.io/en/latest/python/python_api.html) documentation.\n",
    "2.   Fit the model on test data to obtain the predictions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "h5kYwpdIErQU"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[14:59:34] WARNING: ../src/learner.cc:576: \n",
      "Parameters: { \"n_gpus\", \"num_rounds\", \"scale_pos_weight\", \"verbose\" } might not be used.\n",
      "\n",
      "  This could be a false alarm, with some parameters getting used by language bindings but\n",
      "  then being mistakenly passed down to XGBoost core, or some parameter actually being used\n",
      "  but getting flagged wrongly here. Please open an issue if you find any such cases.\n",
      "\n",
      "\n",
      "[14:59:34] WARNING: ../src/learner.cc:1115: Starting in XGBoost 1.3.0, the default evaluation metric used with the objective 'multi:softprob' was changed from 'merror' to 'mlogloss'. Explicitly set eval_metric if you'd like to restore the old behavior.\n",
      "[0]\ttest-mlogloss:2.06621\ttrain-mlogloss:2.06616\n",
      "[1]\ttest-mlogloss:1.69947\ttrain-mlogloss:1.69931\n",
      "[2]\ttest-mlogloss:1.44913\ttrain-mlogloss:1.44858\n",
      "[3]\ttest-mlogloss:1.25917\ttrain-mlogloss:1.25811\n",
      "[4]\ttest-mlogloss:1.10647\ttrain-mlogloss:1.10492\n",
      "[5]\ttest-mlogloss:0.97973\ttrain-mlogloss:0.97773\n",
      "[6]\ttest-mlogloss:0.87253\ttrain-mlogloss:0.87010\n",
      "[7]\ttest-mlogloss:0.78036\ttrain-mlogloss:0.77761\n",
      "[8]\ttest-mlogloss:0.70024\ttrain-mlogloss:0.69715\n",
      "[9]\ttest-mlogloss:0.62976\ttrain-mlogloss:0.62647\n"
     ]
    }
   ],
   "source": [
    "dtrain = xgb.DMatrix(x_train, label=y_train)\n",
    "dtest = xgb.DMatrix(x_test, label=y_test)\n",
    "evals = [(dtest, 'test',), (dtrain, 'train')]\n",
    "\n",
    "num_rounds = params['num_rounds']\n",
    "\n",
    "model = xgb.train(params, dtrain, num_rounds, evals=evals)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "nUUoK7fjJpZO"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0.03264967 0.03270135 0.03265994 ... 0.03264955 0.03267992 0.03264939]\n",
      " [0.0191389  0.01916919 0.01914492 ... 0.01913883 0.01915663 0.01913873]\n",
      " [0.02149575 0.02152978 0.02150252 ... 0.02149567 0.02151566 0.02149556]\n",
      " ...\n",
      " [0.02154769 0.0215818  0.02155447 ... 0.02154761 0.02156765 0.0215475 ]\n",
      " [0.02154769 0.0215818  0.02155447 ... 0.02154761 0.02156765 0.0215475 ]\n",
      " [0.02154769 0.0215818  0.02155447 ... 0.02154761 0.02156765 0.0215475 ]]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([17, 11,  9, ..., 11, 11, 11])"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds = model.predict(dtest)\n",
    "print(preds)\n",
    "\n",
    "true_labels = y_test\n",
    "true_labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "DtPSdZeLKCZ4"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([17, 11,  9, ..., 11, 11, 11])"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# If predictions > 0.5, pred_labels = 1 else pred_labels = 0\n",
    "\n",
    "pred_labels = np.argmax(preds, axis=1)\n",
    "pred_labels"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get the accuracy score for your model's predictions. In order to pass this part of the assessment, you need to attain an accuracy greater than 90%."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "qm8dMyMyKMAG"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "XGBoost Accuracy Score : 0.9948402460805715\n"
     ]
    }
   ],
   "source": [
    "# NOTE: We are using our own accuracy score function in order to help grade the assessment,\n",
    "# though it will behave here exactly like its scikit-learn couterpart `accuracy_score`.\n",
    "xgb_acc = xgb_accuracy_score(true_labels, pred_labels)\n",
    "print ('XGBoost Accuracy Score :', xgb_acc)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Assessment Task 4: Implement a Confusion Matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Show that you can determine the performance of your model by implementing a confusion matrix."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cm = confusion_matrix(true_labels, pred_labels)\n",
    "\n",
    "def plot_confusion_matrix(cm, title='Confusion matrix', cmap=plt.cm.Greens):\n",
    "    plt.figure(figsize=(10,10),)\n",
    "\n",
    "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    plt.title(title)\n",
    "    plt.colorbar()\n",
    "    \n",
    "    plt.tight_layout()\n",
    "\n",
    "    width, height = cm.shape\n",
    "\n",
    "    for x in range(width):\n",
    "        for y in range(height):\n",
    "            plt.annotate(str(cm[x][y]), xy=(y, x), \n",
    "                        horizontalalignment='center',\n",
    "                        verticalalignment='center')\n",
    "    plt.ylabel('True label')\n",
    "    plt.xlabel('Predicted label')\n",
    "\n",
    "plot_confusion_matrix(cm)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "ndrKZP0EmEBd"
   },
   "source": [
    "# Autoencoder Model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "MxndyyB5KdhM"
   },
   "source": [
    "As the second major part of this assessment, you get to train your own autoencoder neural network to understand inherant clusters in your data. Build an autoencoder treating this as a brinary classification problem. Feel free to open the file viewer on the left of the JupyterLab environment to view the notebooks from previous sections if you need a reference to guide your work."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "MxndyyB5KdhM"
   },
   "source": [
    "![alt text](https://drive.google.com/uc?id=1gexBTwBnK_LtTmxrZp_opHD1xaAd2oum)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "CNwbM0GESLFy"
   },
   "source": [
    "## Assessment Task 5: Set the Hyperparameters "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 149,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "EPuAaf6zmH40"
   },
   "outputs": [],
   "source": [
    "input_dim = x_train.shape[1]\n",
    "\n",
    "# Model hyperparameters\n",
    "batch_size = 512\n",
    "# Latent dimension: higher values add network capacity \n",
    "# while lower values increase efficiency of the encoding\n",
    "latent_dim = 4\n",
    "\n",
    "# Number of epochs: should be high enough for the network to learn from the data, \n",
    "# but not so high as to overfit the training data or diverge\n",
    "max_epochs = 20\n",
    "\n",
    "learning_rate = 0.00001"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "gI4-Yzz8UD1r"
   },
   "source": [
    "## Assessment Task 6: Build the Encoder Segment"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "gI4-Yzz8UD1r"
   },
   "source": [
    "1.   Fix the dimensions of the input (number of features in the dataset) in the input layer.\n",
    "2.   Define the hidden layers of the encoder. We recommended using at least 3-4 layers.\n",
    "3.   Consider adding dropout layers to the encoder to help avoid overfitting.\n",
    "4.   Experiment with different activation functions (relu, tanh, sigmoid etc.).\n",
    "\n",
    "Feel free to open the file viewer on the left of the JupyterLab environment to view the notebooks from previous sections if you need a reference to guide your work."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "metadata": {},
   "outputs": [],
   "source": [
    "# The encoder will consist of a number of dense layers that decrease in size \n",
    "# as we taper down towards the bottleneck of the network: the latent space.\n",
    "\n",
    "input_data = Input(shape=(input_dim,), name='encoder_input')\n",
    "\n",
    "# Hidden layers\n",
    "encoder = Dense(units= 96, activation= 'tanh', name='encoder_1')(input_data)\n",
    "encoder = Dropout(.1)(encoder)\n",
    "encoder = Dense(64,activation='tanh', name='encoder_2')(encoder)\n",
    "encoder = Dropout(.1)(encoder)\n",
    "encoder = Dense(48,activation='tanh', name='encoder_3')(encoder)\n",
    "encoder = Dropout(.1)(encoder)\n",
    "encoder = Dense(16,activation='tanh', name='encoder_4')(encoder)\n",
    "encoder = Dropout(.1)(encoder)\n",
    "\n",
    "# Bottleneck layer\n",
    "latent_encoding = Dense(latent_dim, activation='linear', name='latent_encoding')(encoder)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 151,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "539Kkh1KnC1r"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_18\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " encoder_input (InputLayer)  [(None, 118)]             0         \n",
      "                                                                 \n",
      " encoder_1 (Dense)           (None, 96)                11424     \n",
      "                                                                 \n",
      " dropout_72 (Dropout)        (None, 96)                0         \n",
      "                                                                 \n",
      " encoder_2 (Dense)           (None, 64)                6208      \n",
      "                                                                 \n",
      " dropout_73 (Dropout)        (None, 64)                0         \n",
      "                                                                 \n",
      " encoder_3 (Dense)           (None, 48)                3120      \n",
      "                                                                 \n",
      " dropout_74 (Dropout)        (None, 48)                0         \n",
      "                                                                 \n",
      " encoder_4 (Dense)           (None, 16)                784       \n",
      "                                                                 \n",
      " dropout_75 (Dropout)        (None, 16)                0         \n",
      "                                                                 \n",
      " latent_encoding (Dense)     (None, 4)                 68        \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 21,604\n",
      "Trainable params: 21,604\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# We instantiate the encoder model, look at a summary of it's layers, and visualize it.\n",
    "\n",
    "encoder_model = Model(input_data, latent_encoding)\n",
    "\n",
    "encoder_model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "XCTvID5mhH1g"
   },
   "source": [
    "## Assessment Task 7: Build Decoder Segment"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "XCTvID5mhH1g"
   },
   "source": [
    "1.    Fix the dimensions of the input to the decoder. \n",
    "2.    Grow the network from the latent layer to the output layer of size equal to the input layer.\n",
    "3.    Experiment with different activation functions (tanh, relu, sigmoid etc.)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "metadata": {},
   "outputs": [],
   "source": [
    "# The decoder network is a mirror image of the encoder network.\n",
    "decoder = Dense(16, activation='tanh', name='decoder_1')(latent_encoding)\n",
    "decoder = Dropout(.1)(decoder)\n",
    "decoder = Dense(48, activation='tanh', name='decoder_2')(decoder)\n",
    "decoder = Dropout(.1)(decoder)\n",
    "decoder = Dense(64, activation='tanh', name='decoder_3')(decoder)\n",
    "decoder = Dropout(.1)(decoder)\n",
    "decoder = Dense(96, activation='tanh', name='decoder_4')(decoder)\n",
    "decoder = Dropout(.1)(decoder)\n",
    "                \n",
    "# The output is the same dimension as the input data we are reconstructing.\n",
    "reconstructed_data = Dense(units = input_dim, activation='linear', name='reconstructed_data')(decoder)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "VKzj0ibOnRsw"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_19\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " encoder_input (InputLayer)  [(None, 118)]             0         \n",
      "                                                                 \n",
      " encoder_1 (Dense)           (None, 96)                11424     \n",
      "                                                                 \n",
      " dropout_72 (Dropout)        (None, 96)                0         \n",
      "                                                                 \n",
      " encoder_2 (Dense)           (None, 64)                6208      \n",
      "                                                                 \n",
      " dropout_73 (Dropout)        (None, 64)                0         \n",
      "                                                                 \n",
      " encoder_3 (Dense)           (None, 48)                3120      \n",
      "                                                                 \n",
      " dropout_74 (Dropout)        (None, 48)                0         \n",
      "                                                                 \n",
      " encoder_4 (Dense)           (None, 16)                784       \n",
      "                                                                 \n",
      " dropout_75 (Dropout)        (None, 16)                0         \n",
      "                                                                 \n",
      " latent_encoding (Dense)     (None, 4)                 68        \n",
      "                                                                 \n",
      " decoder_1 (Dense)           (None, 16)                80        \n",
      "                                                                 \n",
      " dropout_76 (Dropout)        (None, 16)                0         \n",
      "                                                                 \n",
      " decoder_2 (Dense)           (None, 48)                816       \n",
      "                                                                 \n",
      " dropout_77 (Dropout)        (None, 48)                0         \n",
      "                                                                 \n",
      " decoder_3 (Dense)           (None, 64)                3136      \n",
      "                                                                 \n",
      " dropout_78 (Dropout)        (None, 64)                0         \n",
      "                                                                 \n",
      " decoder_4 (Dense)           (None, 96)                6240      \n",
      "                                                                 \n",
      " dropout_79 (Dropout)        (None, 96)                0         \n",
      "                                                                 \n",
      " reconstructed_data (Dense)  (None, 118)               11446     \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 43,322\n",
      "Trainable params: 43,322\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# We instantiate the encoder model, look at a summary of its layers, and visualize it.\n",
    "autoencoder_model = Model(input_data, reconstructed_data)\n",
    "\n",
    "autoencoder_model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "PZFm_GT16QfY"
   },
   "source": [
    "## Assessment Task 8: Initiate Training of the Model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "PZFm_GT16QfY"
   },
   "source": [
    "1.   Fix the learning rate *Hint: Think in the order of 10e-4*.\n",
    "2.   Choose an appropriate error metric for the loss function (mse, rmse, mae etc.).\n",
    "3.   Think about whether you want to shuffle your dataset during training.\n",
    "4.   Initiate training of the autoencoder on the given dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "kYxZiu8JnT5O"
   },
   "outputs": [],
   "source": [
    "opt = optimizers.Adam(learning_rate=learning_rate)\n",
    "\n",
    "autoencoder_model.compile(optimizer=opt, loss='mse')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 155,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "1p83UCS4nWwy"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "40/40 [==============================] - 1s 6ms/step - loss: 0.0728 - val_loss: 0.0657\n",
      "Epoch 2/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0704 - val_loss: 0.0642\n",
      "Epoch 3/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0688 - val_loss: 0.0630\n",
      "Epoch 4/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0674 - val_loss: 0.0620\n",
      "Epoch 5/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0663 - val_loss: 0.0611\n",
      "Epoch 6/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0651 - val_loss: 0.0600\n",
      "Epoch 7/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0641 - val_loss: 0.0589\n",
      "Epoch 8/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0628 - val_loss: 0.0576\n",
      "Epoch 9/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0616 - val_loss: 0.0561\n",
      "Epoch 10/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0603 - val_loss: 0.0545\n",
      "Epoch 11/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0590 - val_loss: 0.0528\n",
      "Epoch 12/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0576 - val_loss: 0.0510\n",
      "Epoch 13/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0561 - val_loss: 0.0491\n",
      "Epoch 14/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0547 - val_loss: 0.0472\n",
      "Epoch 15/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0533 - val_loss: 0.0454\n",
      "Epoch 16/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0518 - val_loss: 0.0436\n",
      "Epoch 17/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0506 - val_loss: 0.0420\n",
      "Epoch 18/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0493 - val_loss: 0.0405\n",
      "Epoch 19/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0482 - val_loss: 0.0391\n",
      "Epoch 20/20\n",
      "40/40 [==============================] - 0s 3ms/step - loss: 0.0471 - val_loss: 0.0378\n"
     ]
    }
   ],
   "source": [
    "train_history = autoencoder_model.fit(x_train, x_train,\n",
    "        shuffle=True,\n",
    "        epochs=max_epochs,\n",
    "        batch_size=batch_size,\n",
    "        validation_data=(x_test, x_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 156,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "f7T5mPzInYui"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f0e703c9880>"
      ]
     },
     "execution_count": 156,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(train_history.history['loss'])\n",
    "plt.plot(train_history.history['val_loss'])\n",
    "plt.legend(['loss on train data', 'loss on validation data'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "l3Og5iic-zZe"
   },
   "source": [
    "## Assessment Task 9: Computing Reconstruction Errors"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "l3Og5iic-zZe"
   },
   "source": [
    "1.    Fit the trained model on the test dataset. \n",
    "2.    Compute the reconstruction scores using MSE as the error metric."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 157,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "o4PNXbdrnc_J"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "158/158 [==============================] - 0s 959us/step\n"
     ]
    }
   ],
   "source": [
    " # Reconstruct the data using our trained autoencoder model\n",
    "x_test_recon = autoencoder_model.predict(x_test)\n",
    "\n",
    "# The reconstruction score is the mean of the reconstruction errors (relatively high scores are anomalous)\n",
    "reconstruction_scores = np.mean((x_test - x_test_recon)**2, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 158,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "5qr6kiiPn0Xp"
   },
   "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 tr th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe thead tr:last-of-type th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th colspan=\"8\" halign=\"left\">recon_score</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th>count</th>\n",
       "      <th>mean</th>\n",
       "      <th>std</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>binary_labels</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2674.0</td>\n",
       "      <td>0.027385</td>\n",
       "      <td>0.009527</td>\n",
       "      <td>0.016644</td>\n",
       "      <td>0.020357</td>\n",
       "      <td>0.021943</td>\n",
       "      <td>0.033611</td>\n",
       "      <td>0.088658</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2365.0</td>\n",
       "      <td>0.049562</td>\n",
       "      <td>0.007239</td>\n",
       "      <td>0.017538</td>\n",
       "      <td>0.046280</td>\n",
       "      <td>0.050309</td>\n",
       "      <td>0.052710</td>\n",
       "      <td>0.077464</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              recon_score                                                    \\\n",
       "                    count      mean       std       min       25%       50%   \n",
       "binary_labels                                                                 \n",
       "0                  2674.0  0.027385  0.009527  0.016644  0.020357  0.021943   \n",
       "1                  2365.0  0.049562  0.007239  0.017538  0.046280  0.050309   \n",
       "\n",
       "                                   \n",
       "                    75%       max  \n",
       "binary_labels                      \n",
       "0              0.033611  0.088658  \n",
       "1              0.052710  0.077464  "
      ]
     },
     "execution_count": 158,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Store the reconstruction data in a Pandas dataframe\n",
    "anomaly_data = pd.DataFrame({'recon_score':reconstruction_scores})\n",
    "\n",
    "def convert_label_to_binary(labels):\n",
    "    my_labels = labels.copy()\n",
    "    my_labels[my_labels != 11] = 1 \n",
    "    my_labels[my_labels == 11] = 0\n",
    "    return my_labels\n",
    "  \n",
    "# Convert our labels to binary\n",
    "binary_labels = convert_label_to_binary(y_test)\n",
    "\n",
    "# Add the binary labels to our anomaly dataframe\n",
    "anomaly_data['binary_labels'] = binary_labels\n",
    "\n",
    "# Let's check if the reconstruction statistics are different for labeled anomalies\n",
    "anomaly_data.groupby(by='binary_labels').describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "-Yxh-RGrL4ER"
   },
   "source": [
    "## Assessment Task 10: Anomaly Detection"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "-Yxh-RGrL4ER"
   },
   "source": [
    "1.   Plot the area under the curve\n",
    "2.   Set the optimal threshold that separates normal packets from anomalous packets. \n",
    "3.   Threshold should be calculated as the difference between the true positive rate and false positive rate."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 159,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "syeIxcWoolif"
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAmkAAAJcCAYAAACixjPMAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAABUeklEQVR4nO3dedyVc/7H8ddHRUmyhaxZQkWJ7FtoFTXWbEOWMcxgGPvOWIYxGI19G/s+lihURIttQgiDbAlDQkJav78/zqnfLXV3V/e5r7O8nj3uh/ucc53rep9zut3vvt9riZQSkiRJKi6LZR1AkiRJv2ZJkyRJKkKWNEmSpCJkSZMkSSpCljRJkqQiZEmTJEkqQpY0qcJExFsR0THrHFmLiOsi4qw63uatEXFBXW6zUCLigIgYuJDP9e+gVAPhedKk7ETEx8BKwAzgB+BJ4OiU0g9Z5io3EdEHODyltG3GOW4FxqWUzsw4x7nAuimlA+tgW7dSBK9ZKkWOpEnZ2y2ltBSwMdAeOC3bOAsuIupX4raz5HsulT9LmlQkUkr/A54iV9YAiIgtI+L5iPguIl6vOkUUEctFxL8i4vOI+DYiHqny2K4RMSr/vOcjom2Vxz6OiE4RsUpETI6I5ao81j4ivo6IBvnbh0bEO/n1PxURa1ZZNkXEHyPifeD9ub2miOiZn9r6LiKejYhWc+Q4LSLezq//XxHRcAFewykR8QbwY0TUj4hTI+KDiJiUX+fu+WVbAdcBW0XEDxHxXf7+2VOPEdExIsZFxAkR8VVEfBERh1TZ3vIR8VhEfB8R/4mICyJi+Lw+y4jYtsrn9ml+JG+WZSOifz7nSxGxTpXnXZlf/vuIeCUitqvy2LkR8WBE3BkR3wN9ImLziHghv50vIuKqiFi8ynPaRMSgiPgmIr6MiNMjohtwOtA7/368nl+2aUTcnF/PZ/nXWC//WJ+IGBERV0TEBODc/H3D849H/rGv8tnfjIgNI+II4ADg5Py2Hqvy+XXKf18vn2vWZ/dKRKw+r/dWqigpJb/88iujL+BjoFP++9WAN4Er87dXBSYAu5D7B1Xn/O1m+cf7A/cBywINgB3y97cHvgK2AOoBB+e3s8RctvkM8LsqeS4Frst/3wsYA7QC6gNnAs9XWTYBg4DlgEZzeW3rAT/mczcATs6vb/EqOUYDq+fXMQK4YAFew6j8cxvl79sbWCX/XvXOb7t5/rE+wPA58t1aZXsdgenAX/JZdwF+ApbNP35v/mtJoDXw6Zzrq7LeNYFJwH75dS0PbFxlmxOAzfPv6V3AvVWee2B++frACcD/gIb5x84FpgG/yb/GRsCmwJb55VsA7wDH5ZdvAnyRX0/D/O0tqqzrzjlyPwxcDzQGVgReBn5f5f2bDhyT31ajqu8p0BV4BVgGCHJ/Z5rP+T7P4+/9SeT+3q+ff247YPmsfzb98qsYvjIP4JdflfyV/2X1Q/6XegKeBpbJP3YKcMccyz9FrrA0B2bOKhFzLHMtcP4c973L/5e4qr8gDweeyX8f+fKxff72E8BhVdaxGLnismb+dgJ2qua1nQXcP8fzPwM6VslxZJXHdwE+WIDXcOh83ttRQK/897MLRZXHZ5cHciVtMlC/yuNfkStA9ciVo/WrPHbBnOur8thpwMPzeOxW4KY5XvN/q3kN3wLt8t+fCwydz2s+bta2yZXE1+ax3LlUKWnk9oucQpWynX/+kCrv39g51jH7PQV2At7Lv1+Lzet9nuPv/ay/g+/O+pz88suvX3453Sll7zcppSbkisIGwAr5+9cE9s5PZX2Xn6bbllxBWx34JqX07VzWtyZwwhzPW53cKNOc/k1uGrA5sD254jesynqurLKOb8gVuVWrPP/Tal7XKsAns26klGbml5/X8z+pkrEmr+EX246Ig6pMj34HbMj/v5c1MSGlNL3K7Z+ApYBm5EaPqm6vute9OvBBNY//by7bACAiTozc9PLE/Gtoyi9fw5yveb2IeDwi/pefAr2oyvLzy1HVmuRG/b6o8v5dT25Eba7briql9AxwFXA18FVE3BARS9dw2wuSU6ooljSpSKSUniM36vD3/F2fkhtJW6bKV+OU0sX5x5aLiGXmsqpPgQvneN6SKaV75rLNb4GB5KYH9yc39ZaqrOf3c6ynUUrp+aqrqOYlfU7ulz+Q22+J3C/kz6osU3XfozXyz6npa5i97cjtK3cjcDS5qbJlyE2lRg1yzs94clN9q80j95w+Bdap5vG5yu9/djKwD7kR0mWAifz/a4Bfv45rgf8CLVNKS5Pb12zW8p8Ca89jc3Ou51NyI2krVHm/l04ptanmOb9cYUp9U0qbkpsOXo/cNOZ8n8dCvl9SJbCkScXlH0DniGgH3AnsFhFd8ztXN8zv4L5aSukLctOR10TEshHRICK2z6/jRuDIiNgiv0N344joERFN5rHNu4GDgL3y389yHXBaRLSB2TuW770Ar+V+oEdE7By5AxFOIFcEqpa8P0bEapE7eOEMcvvYLcxraEyuDIzPZz2E3EjaLF8Cq1Xdqb6mUkozgIfI7Sy/ZERsQO79mpe7gE4RsU/kDmhYPiI2rsGmmpArg+OB+hFxNjC/0agmwPfAD/lcR1V57HGgeUQcFxFLRESTiNgi/9iXQIuIWCz/Gr8gV9Yvi4ilI2KxiFgnInaoQW4iYrP8Z9WA3L6AP5MblZ21rXmVRYCbgPMjomX+s24bEcvXZLtSubOkSUUkpTQeuB04O6X0Kbmd908n94v7U3KjE7N+bn9Lbl+p/5Lbf+q4/DpGAr8jN/30Lbmd9ftUs9l+QEvgfyml16tkeRi4BLg3P5U2Gui+AK/lXXI7wv8T+BrYjdzpRqZWWexucuXgQ3JTXhcszGtIKb0NXAa8QK4UbETuQIRZngHeAv4XEV/X9DVUcTS5qcf/AXcA95ArnHPLMpbcvmYnkJsiHkVuZ/j5eYrcefLeIzf1+zPVT6sCnEhuBHQSuWI7q+SSUppE7qCN3fK53wd2zD/8QP6/EyLi1fz3BwGLA2+Te88fJDe1XhNL57f/bT77BHIHoQDcDLTOT6M+MpfnXk6u0A8kVzhvJndgglTxPJmtpExE7kS+h6eUBmedZUFFxCXAyimlg7POIql8OZImSfMRERvkp+EiIjYHDiN3ygpJKhjPGi1J89eE3BTnKuSmUy8DHs00kaSy53SnJElSEXK6U5IkqQiV3HTnCiuskFq0aJF1DEmSpPl65ZVXvk4pNVuY55ZcSWvRogUjR47MOoYkSdJ8RcQn819q7pzulCRJKkKWNEmSpCJkSZMkSSpCljRJkqQiZEmTJEkqQpY0SZKkImRJkyRJKkKWNEmSpCJkSZMkSSpCljRJkqQiZEmTJEkqQpY0SZKkImRJkyRJKkKWNEmSpCJkSZMkSSpCljRJkqQiZEmTJEkqQpY0SZKkImRJkyRJKkKWNEmSpCJkSZMkSSpCBStpEXFLRHwVEaPn8XhERN+IGBMRb0TEJoXKIkmSVGoKOZJ2K9Ctmse7Ay3zX0cA1xYwiyRJUkkpWElLKQ0FvqlmkV7A7SnnRWCZiGheqDySJEl16Ud+XKTnZ7lP2qrAp1Vuj8vf9ysRcUREjIyIkePHj6+TcJIkSQsjkbiQC1luwvqLtJ6SOHAgpXRDSqlDSqlDs2bNso4jSZI0V23Pe5Kmw3bnTM5kyXEbLNK66tdSpoXxGbB6ldur5e+TJEllrN15A5k4eVrWMWpdWmwGk/a+ikntB3EMx/CPdv+gHvUWen1ZlrR+wNERcS+wBTAxpfRFhnkkSSpK5VZqmjZqwMcX98g6Rq07ndP5K4M4l3M5m7MJYpHWV7CSFhH3AB2BFSJiHHAO0AAgpXQdMADYBRgD/AQcUqgskqTSUW6FpDaUa6kpN8dzPOuzPgdzcK2sr2AlLaW033weT8AfC7V9SZoXS0Bxs5ColHzN11zMxVzERTSjWa0VNMh2ulOSaqS2S5UlQFJtGMc4utCFj/iIfdiHzdm8VtdvSZMqQKmPHFmqJBWb93iPznTmO77jKZ6q9YIGljSpaNVmsbLkSFLtGcUoutKVRGIIQ9iEwlzZ0pIm1SKLlSRVhpVZmQd4gPVYr2DbsKRJC2B+JcxiJUnl67/8lw3YgI3ZmNd4jcUKfE0AS5pUhSVMkjQ3d3EXfejDjdxIH/oUvKCBJU0VoqbTkJYwSdKcruIqjuEYdmRH9mCPOtuuJU1lr915AwEsX5KkBZJInM/5nMM59KIX93IvDWlYZ9u3pKksVR05a9qoAa+f0yXjRJKkUvMar3Eu586e5qxfx7XJkqayMqucOW0pSVpUm7AJwxjGVmxVJ/ugzcmSprLhtKYkaVH9zM/8lt9yGIfRjW5swzaZZbGkqeRVHT1zWlOStLC+53t60YvneI5OdMo6jiVNpc3RM0lSbRjPeLrTndd5nbu4i/3YL+tIljSVHg8KkCTVpm/4hu3Yjk/4hEd5lF3YJetIgCVNJcSDAiRJhbAsy9Kd7uzBHmzHdlnHmc2SppLgtKYkqba9yqs0oQktackVXJF1nF+p++NJpQU0q6A5rSlJqi3P8Rwd6cgRHJF1lHlyJE1Fy6M2JUmF0I9+7MM+rMM63MmdWceZJ0uaio77nkmSCuV2budQDmVTNmUAA1ie5bOONE+WNGVmXhc9t5xJkgphBjO4nuvpSEce5mGa0CTrSNWypKlOzXn6DMuYJKnQEokpTKEhDelPfxrm/xQ7S5pq3bxGyMBiJkmqWzOZyZ/4E2/xFk/wBMuwTNaRasySphqrrnxVZRGTJBWDaUzjEA7hLu7iBE5gcRbPOtICsaRVqJoWrqosX5KkUjGZyezN3vSnPxdxEadyKkFkHWuBWNIqiPuDSZIqxaEcygAGcC3XciRHZh1noVjSytTcRsosZpKkSnEmZ7IHe7A3e2cdZaFZ0kqcp7GQJCnnEz7hbu7mVE6lTf5PKbOklSCnLSVJ+qW3eZsudOFHfuQADmAN1sg60iKzpBUxR8kkSZq///AfutOd+tTnOZ4ri4IGlrSi5GWRJEmqmSEMoSc9WYEVGMQg1mXdrCPVGktaEZhzxMxyJklSzUxkIuuwDv3pz6qsmnWcWmVJy1i78wYCWMokSVoAH/ERa7EWv+E37Mqu1C/DSrNY1gEq3cTJ03j9nC5Zx5AkqWRcwRWsx3oMYxhAWRY0sKRlqt15A2naqEHWMSRJKgmJxJmcyZ/5M7/hN2zO5llHKqjyrJ5FbM7TZziKJknS/M1kJkdzNNdyLYdxGNdzPfWol3WsgrKkFZgHBUiStOge4iGu5VpO4iQu4ZKSuw7nwrCkFYAnm5UkqXbtyZ4MZCCd6Zx1lDpjSatFnt9MkqTa8x3f0Yc+/JW/0opWFVXQwJJWazyVhiRJted//I9udONt3uYQDqEVrbKOVOcsabVgVkHzIABJkhbdR3xEZzrzBV/wOI/Thcr8/WpJqwUTJ09zBE2SpFrwAR+wPdvzEz8xmMFsxVZZR8qM50lbRJ7rTJKk2rMKq7A92zOUoRVd0MCRtEXmKJokSYtuOMPZkA1ZhmW4h3uyjlMUHElbBI6iSZK06B7iIXZmZ07ipKyjFBVL2kLyYAFJkhbdLdzC3uzNpmzK3/hb1nGKiiVtIVjQJEladJdxGYdxGJ3oxCAGsSzLZh2pqFjSFpAFTZKkRfc939OXvuzDPjzGYzSmcdaRio4HDiwAC5okSYtmBjMAWJqleYEXWImVyv5C6QvLkbQFMHHyNAuaJEkLaSpT2Z/9+QN/IJFYhVUsaNWwpNWQR3JKkrTwfuRHetKT+7mflrQkiKwjFT2nO2vAaU5Jkhbet3zLruzKi7zITdzEYRyWdaSSYEmrAU9YK0nSwkkkdmEXXuVVHuAB9mCPrCOVDEuaJEkqmCA4m7NpQAM60SnrOCXFkjYf7osmSdKCG81oXuM1fstv6U73rOOUJEvafDjVKUnSgnmRF9mFXViKpdiDPTwH2kLy6M5qOIomSdKCGcQgdmZnlmM5hjLUgrYILGnV8LxokiTV3IM8SA96sC7rMpzhtKBF1pFKmiVNkiTVijGMYXM25zmeY2VWzjpOybOkSZKkRfIZnwFwCqfwDM+wDMtkG6hMWNIkSdJCSSRO4RTa0IYP+ZAgWJzFs45VNjy6U5IkLbAZzOBIjuQmbuIojnL/swJwJE2SJC2QKUxhX/blJm7iTM7kaq5mMStFrXMkTZIkLZAruIIHeZDLuZzjOT7rOGXLkiZJkhbI8RxPO9p5JYECc2xSkiTN1+d8zt7szQQmsARLWNDqgCVNkiRVawxj2IZteJIneY/3so5TMZzulCRJ8/QGb9CFLkxnOs/wDJuxWdaRKoYjaZIkaa7+w3/YgR2oT32GMcyCVscsaZIkaa5WZ3W2YitGMIJWtMo6TsWxpEmSpF94jueYznRWZmUGMIA1WTPrSBXJkiZJkma7nuvZkR25jMuyjlLxLGmSJIlE4q/8lSM5kl3YhWM5NutIFc+SJklShUskTuZkTud09md/HuZhGtEo61gVz5ImSVKF+5APuZZrOZqjuYM7aECDrCMJz5MmSVLFmsEM6lGPdViH13mdtVmbILKOpTxH0iRJqkCTmERXutKXvgCswzoWtCJjSZMkqcJMYAKd6MSzPMsyLJN1HM2D052SJFWQz/iMLnThAz7gIR6iJz2zjqR5sKRJklQhfuRHtmM7vuZrnuRJOtIx60iqhiVNkqQK0ZjGnMIpdKADm7Jp1nE0H5Y0SZLK3HCGM53pdKQjv+f3WcdRDVnSJEkqYwMYwF7sRRva8DIvewRnCfHoTkmSytQ93EMvetGKVvSnvwWtxFjSJEkqQ9dwDQdwANuwDUMYwoqsmHUkLSBLmiRJZSaReIEX2JVdeYInWJqls46kheA+aZIklYmZzOQbvmEFVuAWbgHwOpwlzJE0SZLKwHSmcyiHshVb8T3f0yD/R6XLkiZJUon7mZ/Zi724jds4kANpQpOsI6kWON0pSVIJ+57v+Q2/YQhD6EtfjuGYrCOplljSJEkqYX/mzwxlKHdyJwdwQNZxVIssaZIklbC/8ld605vOdM46imqZ+6RJklRi3uM9DudwpjKVZjSzoJUpS5okSSXkVV5lW7alH/34hE+yjqMCsqRJklQihjKUHdmRRjRiOMNpScusI6mALGmSJJWAJ3iCrnRlFVZhOMNZj/WyjqQCK2hJi4huEfFuRIyJiFPn8vgaETEkIl6LiDciYpdC5pEkqVQ1pznbsi3DGMbqrJ51HNWBgpW0iKgHXA10B1oD+0VE6zkWOxO4P6XUHtgXuKZQeSRJKkUv8AIAG7MxgxjECqyQcSLVlUKOpG0OjEkpfZhSmgrcC/SaY5kEs6/62hT4vIB5JEkqGYnEX/gLW7M1D/FQ1nGUgUKeJ21V4NMqt8cBW8yxzLnAwIg4BmgMdJrbiiLiCOAIgDXWWKPWg0qSVExmMpPjOZ6+9KUPfehJz6wjKQNZHziwH3BrSmk1YBfgjoj4VaaU0g0ppQ4ppQ7NmjWr85CSJNWVaUyjD33oS1+O53hu5mbqe+75ilTIkvYZ/GLPxtXy91V1GHA/QErpBaAhONkuSapcIxjBndzJBVzAZVzGYpmPpygrhazm/wFaRsRa5MrZvsD+cywzFtgZuDUiWpEraeMLmEmSpKKUSARBRzoymtG0Zs5j7VRpClbPU0rTgaOBp4B3yB3F+VZE/CUiZk2unwD8LiJeB+4B+qSUUqEySZJUjMYznq3Zmqd4CsCCJqDAF1hPKQ0ABsxx39lVvn8b2KaQGSRJKmZjGUsXujCWscxkZtZxVETcE1GSpIz8l//Smc5MYhIDGci2bJt1JBURS5okSRkYy1i2YzsWYzGe4zna0S7rSCoyHjIiSVIGVmd1fs/vGc5wC5rmypImSVId6k9/xjCGILiAC2hJy6wjqUhZ0iRJqiO3czu96MXpnJ51FJUAS5okSXXgSq7kYA6mIx25mZuzjqMSYEmTJKmAEomzOZvjOI492IP+9KcJTbKOpRJgSZMkqYCmMIXBDOZQDuU+7mMJlsg6kkqEp+CQJKkApjGNKUxhKZZiIANpTGOCyDqWSogjaZIk1bLJTGZ3dqcXvZjBDJZiKQuaFpglTZKkWjSRiXSlKwMYwD7sQz3qZR1JJcrpTkmSasmXfEk3uvEWb3EP99Cb3llHUgmzpEmSVEv2ZV/e4z0e4zG60jXrOCpxlrR5aHfeQJo2apB1DElSCbmKq5jIRLZm66yjqAy4T9o8TJw8jdfP6ZJ1DElSkRvJSM7mbBKJNrSxoKnWWNIkSVpIz/AMO7Ijd3AH3/Jt1nFUZixpkiQthEd4hO50pwUtGMEIlmO5rCOpzFjSJElaQLdxG3uyJ5uwCc/xHKuwStaRVIYsaZIkLaClWZqudGUwgx1BU8FY0iRJqoFE4jVeA2B3dqc//WlM44xTqZxZ0iRJmo+ZzOSP/JHN2Gx2UfMyTyo0z5MmSVI1pjKVgzmYe7mXUziFjdk460iqEJY0SZLm4Sd+Yk/25Eme5BIu4WROzjqSKoglTZKkebibuxnIQG7kRg7n8KzjqMJY0iRJmkMiEQSHcRib5P9Idc0DByRJquIjPmJrtuYd3iEIC5oy40iaJEl5b/EWXejCZCbzPd9nHUcVzpE0SZKAl3iJ7dmeRGIoQ9mCLbKOpApnSZMkVbyRjGRndmZZlmUEI9iQDbOOJFnSJElqQxsO5ECGMYy1WCvrOBJgSZMkVbAHeZDv+I5GNOI6rqM5zbOOJM1mSZMkVaRLuZS92ZuLuTjrKNJcWdIkSRUlkTiN0ziZk+lNb/7CX7KOJM2Vp+CQJFWMGczgD/yBG7iB3/N7ruZq6lEv61jSXDmSJkmqGBOYwJM8yWmcxrVca0FTUXMkTZJU9n7iJ5ZgCVZkRUYximVZNutI0nw5kiZJKmvf8A2d6MSxHAtgQVPJsKTNRbvzBtK0UYOsY0iSFtEXfMEO7MArvMLO7Jx1HGmBON05FxMnT+Pji3tkHUOStAg+5EM605kv+ZIBDLCkqeRY0iRJZWca0+hCF77jO57hGTZn86wjSQvMkiZJKjsNaMA1XMOqrEob2mQdR1ooljRJUtkYyEDGMY5DOZQudMk6jrRIPHBAklQWHuABdmVXruZqpjM96zjSIrOkSZJK3o3cSG96szmb8zRPU9+JIpUBS9ocPP2GJJWWS7iEIziCbnRjIANZhmWyjiTVCkvaHCZOnsbr57gfgySVipnMZF/25REeYUmWzDqOVGscD5YklZwZzGAMY1if9TmVU0kkFnPcQWXGv9GSpJIyhSn0pjdbsiVf8iVBWNBUlhxJkySVjB/4gd3ZncEM5gquYCVWyjqSVDCWNElSSZjABHrQg5GM5FZu5WAOzjqSVFCWNElSSbiES3iN13iQB/kNv8k6jlRwljRJUkk4n/PZh33oQIeso0h1wj0tJUlF6w3eYGd2ZgITWIIlLGiqKJY0SVJRGsEItmd73uVdJjAh6zhSnbOkSZKKzpM8SWc6syIrMoIRrMd6WUeS6pwlTZJUVPrTn570ZH3WZzjDWZM1s44kZcKSJkkqKpuwCfuxH8/yLCuyYtZxpMxY0iRJmUskHuABpjOd5jTnNm6jKU2zjiVlypImScpUInEiJ7IP+3Art2YdRyoanidNkpSZ6UznCI7gX/yLYziGQzk060hS0XAkTZKUiZ/5mX3Yh3/xL87hHK7kSi+ULlXhSJokKRPv8z6DGcyVXMmxHJt1HKnoWNIkSXXqZ36mIQ3ZiI0YwxiP4JTmwXFlSVKdGcc4NmETruZqAAuaVA1H0iRJdeJ93qcznfmGb9iQDbOOIxU9S5okqeBGMYqudGUmM3mWZ9mETbKOJBU9pzslSQX1NV+zIzuyBEswjGEWNKmGHEmTJBXUCqzAFVzBTuzEGqyRdRypZFjSJEkFcQ/30JzmdKQjfeiTdRyp5DjdKUmqdVdzNQdwAFdwRdZRpJJlSZMk1ZpE4nzO52iOZjd24z7uyzqSVLKc7pQk1YqZzOQETuAf/IODOIibuZn6/pqRFpojaZKkWvM1X3Msx/Iv/mVBkxaRP0GSpEXyMz8zgQmsyqrcyq0sxmIEkXUsqeQ5kiZJWmjf8z3d6c7O7MwUplCPehY0qZY4kiZJWijjGU93uvM6r3Mbt7EES2QdSSorljRJ0gL7lE/pQhc+5mMe4RF60CPrSFLZsaRJkhbYn/gTn/M5T/EU27N91nGksmRJkyQtsOu5ns/5nHa0yzqKVLY8cECSVCPP8Rz7si9TmUozmlnQpAKzpEmS5usxHqMb3XiDN/iWb7OOI1UES5okqVp3cie7szsbsRFDGcpKrJR1JKkiWNIkSfN0EzfxW37LDuzA0zzNCqyQdSSpYljSJEnztCmbchAH0Z/+NKFJ1nGkimJJkyT9wkxm8jiPA9Ce9tzGbTSkYcappMpjSZMkzTaNaRzMwezGbgxhSNZxpIrmedIkSQBMZjK96c1jPMaFXEhHOmYdSapoljRJEhOZSE96MoxhXMM1HMVRWUeSKp4lTZLEcIbzEi9xN3ezL/tmHUcSlrRfaHfeQJo2apB1DEmqM1OZyuIsTg96MIYxrMZqWUeSlOeBA1VMnDyN18/pknUMSaoT/+W/tKIVT/EUgAVNKjKWNEmqQCMZyXZsx4/8yMqsnHUcSXNhSZOkCjOEIezIjizFUgxnuBdKl4pUjUtaRCxZyCCSpMIbzWi60501WIPhDGdd1s06kqR5mG9Ji4itI+Jt4L/52+0i4pqarDwiukXEuxExJiJOnccy+0TE2xHxVkTcvUDpJUkLpA1tOI/zGMpQVmXVrONIqkZNRtKuALoCEwBSSq8D28/vSRFRD7ga6A60BvaLiNZzLNMSOA3YJqXUBjhuQcJLkmrmOq5jDGMIglM4heVZPutIkuajRtOdKaVP57hrRg2etjkwJqX0YUppKnAv0GuOZX4HXJ1S+ja/na9qkkeSVDOJxFmcxVEcRV/6Zh1H0gKoSUn7NCK2BlJENIiIE4F3avC8VYGq5W5c/r6q1gPWi4gREfFiRHSb24oi4oiIGBkRI8ePH1+DTUuSZjKTozmaC7iAQzmUy7k860iSFkBNStqRwB/JFazPgI2BP9TS9usDLYGOwH7AjRGxzJwLpZRuSCl1SCl1aNasWS1tWpLK1zSmcSAHcg3XcCInchM3Ud/zl0slpSY/seunlA6oekdEbAOMmM/zPgNWr3J7tfx9VY0DXkopTQM+ioj3yJW2/9QglyRpHqYylY/5mIu5mFM4Jes4khZCTUbS/lnD++b0H6BlRKwVEYsD+wL95ljmEXKjaETECuSmPz+swbolSXMxkYlMYhKNacyzPGtBk0rYPEfSImIrYGugWUT8ucpDSwP15rfilNL0iDgaeCq//C0ppbci4i/AyJRSv/xjXfKn+JgBnJRSmrDwL0eSKteXfEk3urEyKzOAASzO4llHkrQIqpvuXBxYKr9Mkyr3fw/sVZOVp5QGAAPmuO/sKt8n4M/5L0nSQvqYj+lMZz7ncy7mYoLIOpKkRTTPkpZSeg54LiJuTSl9UoeZJEkL4G3epgtd+JEfGcQgtmbrrCNJqgU1OXDgp4i4FGgDNJx1Z0ppp4KlkiTVyExmsi/7MoMZPMdztKVt1pEk1ZKalLS7gPuAXcmdjuNgwJOVSVIRWIzFuId7aEhD1mGdrONIqkU1Obpz+ZTSzcC0lNJzKaVDAUfRJClDj/AIJ3MyiUQb2ljQpDJUk5I2Lf/fLyKiR0S0B5YrYCZJUjX+xb/Ykz0ZxjAmMznrOJIKpCbTnRdERFPgBHLnR1saL4QuSZm4nMs5gRPoQhce4iGWZMmsI0kqkPmWtJTS4/lvJwI7wuwrDkiS6tB5nMe5nMve7M0d3MESLJF1JEkFNM/pzoioFxH7RcSJEbFh/r5dI+J54Ko6SyhJAqAtbTmSI7mHeyxoUgWobiTtZnLX3nwZ6BsRnwMdgFNTSo/UQTZJqnhTmcoLvMAO7MDu+T+SKkN1Ja0D0DalNDMiGgL/A9bxsk2SVDd+4if2ZE8GM5h3eZe1WTvrSJLqUHUlbWpKaSZASunniPjQgiZJdeNbvmVXduVFXuR6rregSRWoupK2QUS8kf8+gHXyt4PcZTc9rbUkFcD/+B9d6co7vMN93MdeNbtcsqQyU11Ja1VnKSRJs93DPXzAB/SnP53pnHUcSRmp7gLrXlRdkurQDGZQj3ocx3H0opdTnFKFq8kVByRJBfYiL9Ka1rzDOwRhQZNkSZOkrA1iEJ3oxAxm0JCGWceRVCRqVNIiolFErF/oMFlqd95AmjZqkHUMSRXmQR6kBz1Yh3UYznDWYq2sI0kqEvMtaRGxGzAKeDJ/e+OI6FfgXHVu4uRpvH5Ol6xjSKogT/EUvenNZmzGszzLyqycdSRJRaQmI2nnApsD3wGklEaB/9STpEW1PdtzKqcykIEsy7JZx5FUZGpS0qallCbOcV8qRBhJKneJRF/68h3f0YhGXMiFNKZx1rEkFaGalLS3ImJ/oF5EtIyIfwLPFziXJJWdGczgSI7kT/yJf/GvrONIKnI1KWnHAG2AKcDdwETguAJmqnMeNCCp0KYylf3Yjxu4gTM4g+PK63+jkgqguisOzLJBSukM4IxCh8nKxMnT+PjiHlnHkFSmfuRH9mAPBjKQy7iMP/PnrCNJKgE1KWmXRcTKwIPAfSml0QXOJEll5Tu+433e5xZu4RAOyTqOpBIx35KWUtoxX9L2Aa6PiKXJlbULCp5OkkrY13zNsizLqqzK27ztiWolLZAancw2pfS/lFJf4Ehy50w7u5ChJKnUfciHbMEWnMRJABY0SQusJiezbRUR50bEm8CsIztXK3gySSpRb/Im27AN3/Ed+7Jv1nEklaia7JN2C3Af0DWl9HmB80hSSXue5+lBDxrTmGEMozWts44kqUTVZJ+0reoiiCSVuh/4gV70ohnNGMQg1mTNrCNJKmHzLGkRcX9KaZ/8NGfVKwwEkFJKbQueTpJKyFIsxf3cT2tasxIrZR1HUomrbiTtT/n/7loXQSSpVN3IjQTB4RzOjuyYdRxJZWKeBw6klL7If/uHlNInVb+AP9RNPEkqbpdwCUdwBP3oR/KyxpJqUU1OwdF5Lvd1r+0gklRKEomTOZlTOZX92Z9/82+CyDqWpDJS3T5pR5EbMVs7It6o8lATYEShg9UVr9spaUElEr/jd9zMzfyRP9KXvixWs9NOSlKNVbdP2t3AE8BfgVOr3D8ppfRNQVPVIa/bKWlBBcHarM1ZnMV5nOcImqSCqK6kpZTSxxHxxzkfiIjlyqmoSVJN/MAPfMAHtKMdp3N61nEklbn5jaTtCrxC7hQcVf+pmIC1C5hLkorKBCbQgx58wAd8yIc0oUnWkSSVuXmWtJTSrvn/rlV3cSSp+HzGZ3ShCx/wAfdxnwVNUp2oybU7t4mIxvnvD4yIyyNijcJHk6Tsvc/7bMu2jGUsT/AEveiVdSRJFaImhyNdC/wUEe2AE4APgDsKmkqSisRlXMYkJjGEIZ6oVlKdqklJm55SSkAv4KqU0tXgWL+k8jaTmQBcyZW8xEt0oEPGiSRVmpqUtEkRcRrwW6B/RCwGeGIxSWXrCZ5gC7ZgAhNYgiVYh3WyjiSpAtWkpPUGpgCHppT+B6wGXFrQVJKUkXu4h570ZEb+jyRlZb4lLV/M7gKaRsSuwM8ppdsLnkyS6ti1XMsBHMDWbM0QhrAiK2YdSVIFq8nRnfsALwN7A/sAL0XEXoUOJkl16SZu4g/8gR704EmepClNs44kqcJVdzLbWc4ANkspfQUQEc2AwcCDhQwmSXWpBz04iZO4kAtp4G63kopATfZJW2xWQcubUMPnFT0vri5VtulM52quZjrTaU5z/sbfLGiSikZNRtKejIingHvyt3sDAwoXqe54cXWpcv3Mz+zHfjzCI6zGap6kVlLRmW9JSymdFBF7ANvm77ohpfRwYWNJUuFMYhK96MUQhtCXvhY0SUVpniUtIloCfwfWAd4ETkwpfVZXwSSpEL7ma3ZhF17lVe7gDg7kwKwjSdJcVbdv2S3A48CewCvAP+skkSQV0Md8zEd8xMM8bEGTVNSqm+5sklK6Mf/9uxHxal0EkqRC+JZvWZZl6UAHPuIjlmKprCNJUrWqG0lrGBHtI2KTiNgEaDTHbUkqCa/xGhuwATdwA4AFTVJJqG4k7Qvg8iq3/1fldgJ2KlQoSaotwxjGruxKU5rSkY5Zx5GkGptnSUsp7ViXQSSptvWnP3uxFy1owUAGsjqrZx1JkmqsLE5KK0lz+piP2Z3daUMbhjLUgiap5NTkZLaSVHJa0II7uZNudGNpls46jiQtMEfSJJWNROIiLuJZngVgH/axoEkqWfMtaZFzYEScnb+9RkRsXvhoklRzM5nJcRzHGZzBgzyYdRxJWmQ1GUm7BtgK2C9/exJwdcESSdICmsY0+tCHvvTlOI6jL32zjiRJi6wm+6RtkVLaJCJeA0gpfRsRixc4lyTVyBSmsA/70I9+nM/5nMEZBJF1LElaZDUpadMioh65c6MREc2AmQVNVQfanTeQpo0aZB1D0iJqQAOWZmmu4ir+yB+zjiNJtaYmJa0v8DCwYkRcCOwFnFnQVHVg4uRpfHxxj6xjSFpI4xnPZCazBmtwO7c7eiap7My3pKWU7oqIV4CdgQB+k1J6p+DJJGkePuVTOtOZhjTkVV5lMQ9Ul1SG5lvSImIN4Cfgsar3pZTGFjKYJM3Nu7xLZzozkYk8zuMWNEllqybTnf3J7Y8WQENgLeBdoE0Bc0nSr7zKq3SjG0HwLM/SnvZZR5KkgqnJdOdGVW9HxCbAHwqWSJLmIpE4kRNZkiUZxCBa0jLrSJJUUAt8WaiU0qsRsUUhwkjS3CQSQXAf9zGFKazGallHkqSCq8k+aX+ucnMxYBPg84IlkqQq7uAOHuRBHuABmtEs6ziSVGdqssdtkypfS5DbR61XIUNJEkBf+nIQB/EDPzCVqVnHkaQ6Ve1IWv4ktk1SSifWUR5JIpE4L/9nd3bnbu6mIQ2zjiVJdWqeI2kRUT+lNAPYpg7zSBLncA7ncR6HcAj3c78FTVJFqm4k7WVy+5+Nioh+wAPAj7MeTCk9VOBskirU7uzOTGZyPud7JQFJFasmR3c2BCYAO/H/50tLgCVNUq2ZzGT+zb85kANpn/8jSZWsupK2Yv7IztH8fzmbJRU0laSKMpGJ9KQnwxjGRmxEO9plHUmSMlddSasHLAVznWuwpEmqFV/xFd3oxpu8yT3cY0GTpLzqStoXKaW/1FkSSRXnEz6hC134lE/pRz+60z3rSJJUNKorae6tK6mgRjKSr/maQQxiGw8kl6RfqO5ktjvXWQpJFWUSkwDYkz35gA8saJI0F/MsaSmlb+oyiKTKMIQhrMVaPMMzACzDMtkGkqQiVZPLQklSrXiER+hOd1ZmZTZgg6zjSFJRq8iS1u68gTRt1CDrGFJFuY3b2JM92ZiNGcpQVmGVrCNJUlGryJI2cfI0Xj+nS9YxpIoxjGH0oQ87sRODGcxyLJd1JEkqehVZ0iTVrW3Zluu4jsd5nKVYKus4klQSLGmSCmImMzmDM/iADwiC3/N7lmCJrGNJUsmwpEmqddOYxoEcyEVcxENe5leSFkpNLrAuSTX2Ez+xN3szgAFczMWcxElZR5KkkmRJk1RrJjKRXdmVEYzgBm7gd/wu60iSVLIsaZJqTT3qsRiLcS/3sg/7ZB1HkkqaJU3SIvuET1iO5WhCE57lWcJL/0rSIvPAAUmL5G3eZmu25lAOBbCgSVItsaRJWmgv8zLbsR0zmcnZnJ11HEkqK5Y0SQvlaZ5mJ3ZiGZZhBCPYiI2yjiRJZcWSJmmBTWUqv+N3rMVaDGc4a7N21pEkqex44ICkBbY4izOAAazIil6HU5IKpKAjaRHRLSLejYgxEXFqNcvtGREpIjoUMo+kRXMZl/Fn/kwisQEbWNAkqYAKVtIioh5wNdAdaA3sFxGt57JcE+BPwEuFyiJp0SQSZ3AGJ3Ii4xjHDGZkHUmSyl4hR9I2B8aklD5MKU0F7gV6zWW584FLgJ8LmEXSQprBDI7iKC7iIo7gCO7hHuq7p4QkFVwhS9qqwKdVbo/L3zdbRGwCrJ5S6l/diiLiiIgYGREjx48fX/tJJc3ToRzK9VzPaZzGdVxHPeplHUmSKkJm/xyOiMWAy4E+81s2pXQDcANAhw4dUmGTSaqqF73YiI04kROzjiJJFaWQJe0zYPUqt1fL3zdLE2BD4NmIAFgZ6BcRPVNKIwuYS9J8fMu3vMAL7MIu7MEeWceRpIpUyJL2H6BlRKxFrpztC+w/68GU0kRghVm3I+JZ4EQLmpStL/iCrnTlAz7gIz5iRVbMOpIkVaSC7ZOWUpoOHA08BbwD3J9Seisi/hIRPQu1XUkL70M+ZFu25UM+5FEetaBJUoYKuk9aSmkAMGCO++Z6gb+UUsdCZpFUvdGMpgtd+JmfeZqn2YItso4kSRXN4+glAdCPfgTBMIbRhjZZx5Gkiue1O6UKN5nJAJzGaYxilAVNkoqEJU2qYA/yIOuyLu/yLkHQjGZZR5Ik5VnSpAp1IzfSm960oIUHCEhSEbKkSRXoEi7hCI6gC10YyECWZdmsI0mS5mBJkyrM7dzOqZzKvuzLozxKYxpnHUmSNBce3SlVmL3Zm2/4hmM4xutwSlIRcyRNqgBTmMLpnM53fEcjGnEcx1nQJKnIWdKkMvcDP7Abu/FX/sqTPJl1HElSDTndKZWxb/iGHvTgZV7mFm5hX/bNOpIkqYYsaVKZ+oIv6EIX3uM9HuRBdmf3rCNJkhaAJU0qU9OZzkxm8gRPsBM7ZR1HkrSALGlSmfmIj1iDNVid1XmDNzxAQJJKlAcOSGXkeZ5nEzbhbM4GsKBJUgmzpEll4imeojOdaUYzjuCIrONIkhaRJU0qA/dzP7uxG+uxHsMYxpqsmXUkSdIisqRJJW484zmUQ9mSLXmWZ1mJlbKOJEmqBR44IJW4ZjRjMINpS1uWZMms40iSaokjaVIJSiRO5mRu4iYAtmRLC5oklRlLmlRipjOdwzmcS7mUN3kz6ziSpAKxpEklZApT6E1vbuEWzuZs/sE/so4kSSoQ90mTSsR0prMruzKYwVzBFRzHcVlHkiQVkCVNKhH1qc9O7MRv+S0HcVDWcSRJBWZJk4rcZ3zG53zOZmzGaZyWdRxJUh2xpElF7H3epzOdSSTe530WZ/GsI0mS6oglTSpSoxhFV7oyk5k8yZMWNEmqMB7dKRWh4QynIx1ZnMUZxjA2ZdOsI0mS6ljFlbR25w2kaaMGWceQqnU917MSKzGCEWzABlnHkSRloOKmOydOnsbHF/fIOoY0V1OZyuIszk3cxCQmsQIrZB1JkpSRihtJk4rVtVzLpmzKN3zDEixhQZOkCmdJkzKWSFzIhfyBP7AWa9GIRllHkiQVAUualKGZzOQETuBMzuS3/JZ/829LmiQJsKRJmTqf87mCKziWY7mVW2mAB7VIknIq7sABqZgcwREswzIcy7EEkXUcSVIRcSRNqmOTmMT5nM90ptOc5vyJP1nQJEm/4kiaVIe+5mu6053XeI2d2Ilt2CbrSJKkImVJk+rIp3xKF7rwMR/zCI9Y0CRJ1aqokubVBpSV93iPznTmO77jKZ5ie7bPOpIkqchVVEnzagPKyrd8Sz3q8SzP0p72WceRJJWAiippUl37lE9ZndXZgi14l3c9xYYkqcY8ulMqkP70Zz3W4w7uALCgSZIWiCVNKoC7uIvf8Bs2ZEO60z3rOJKkEmRJk2rZVVzFgRzIdmzHMzzjhdIlSQvFkibVotd5nWM4hl70YgADaEKTrCNJkkqUBw5Itagd7XiCJ+hEJ+r74yVJWgSOpEmLaBrTOIIjeI7nAOhGNwuaJGmRWdKkRTCZyezJntzIjbzMy1nHkSSVEf+5Ly2k7/menvRkKEO5mqv5A3/IOpIkqYxY0qSFMJGJ7MROvMEb3MVd7Md+WUeSJJUZpzulhdCEJrSnPY/yqAVNklQQjqRJC+Bd3qUhDVmTNbmJm7KOI0kqY46kSTX0Cq+wLdvyW35LImUdR5JU5ixpUg08y7PsyI40pjE3czNBZB1JklTmLGnSfPSjH93oxuqszghG0JKWWUeSJFUAS5pUjZnM5EIupC1tGcpQVmXVrCNJkiqEBw5I8zCd6dSnPo/zOA1p6HU4JUl1ypE0aQ6JxNmcTS96MZWpNKOZBU2SVOcsaVIVM5nJsRzL+ZzPyqzMYv6ISJIy4m8gKW8a0ziIg7iKqziBE7iJm7xQuiQpMxVT0tqdN5CmjRpkHUNF7AiO4C7u4iIu4lIu9TQbkqRMVcwwwcTJ0/j44h5Zx1ARO5Zj2Zqt+R2/yzqKJEmVM5Imzc1XfMU1XANAe9pb0CRJRaNiRtKkOX3CJ3SmM+MYxy7sQgtaZB1JkqTZHElTRXqbt9mGbRjPeAYxyIImSSo6jqSp4vyH/9Cd7tSnPs/xHG1pm3UkSZJ+xZE0VZwP+ICmNGU4wy1okqSiZUlTxfiSLwHYl315i7dYl3UzTiRJ0rxZ0lQRbuVW1mItRjACgIY0zDiRJEnVs6Sp7F3O5RzCIWzLtrSjXdZxJEmqEUuaylYicSZncgInsBd78RiPsRRLZR1LkqQasaSpbD3Mw1zIhRzO4dzLvSzBEllHkiSpxjwFh8rW7uzOAzzAnuzpdTglSSXHkTSVlZ/4iT704QM+IAj2Yi8LmiSpJFnSVDa+4zu60IXbuZ2XeCnrOJIkLRKnO1UWvuRLutKVt3mb+7iPvdk760iSJC0SS5pK3qd8yk7sxOd8zuM8The6ZB1JkqRF5nSnSt6yLEtLWjKYwRY0SVLZcCRNJes1XmNd1qUJTRjAgKzjSJJUqxxJU0kazGC2YzuO47iso0iSVBAVUdLanTeQpo0aZB1DteQhHqIHPVibtbmAC7KOI0lSQVRESZs4eRqvn+O+SuXgFm5hb/ZmUzblOZ6jOc2zjiRJUkFURElTeZjEJM7iLDrTmUEMYlmWzTqSJEkF44EDKnqJBEATmjCMYazGaizO4hmnkiSpsBxJU1GbwQyO4ihO4AQSibVZ24ImSaoIljQVralMZX/253qupyENs44jSVKdcrpTRelHfmRP9uQpnuJSLuVETsw6kiRJdcqSpqKTSOzGbjzHc9zETRzGYVlHkiSpzjndqaITBEdzNPdzvwVNklSxyn4kzRPZlo4P+ZBRjGKP/B9JkipZ2Ze0iZOn8fHFPbKOofl4kzfpSldmMpMudGEplso6kiRJmXK6U5l7gRfYnu0Jgmd4xoImSRKWNGVsIAPpRCdWYAVGMILWtM46kiRJRcGSpkyNYATrsi7DGEYLWmQdR5KkomFJUya+4RsAzuVcnud5VmbljBNJklRcClrSIqJbRLwbEWMi4tS5PP7niHg7It6IiKcjYs1C5lFxuIRL2IAN+IiPCILGNM46kiRJRadgJS0i6gFXA92B1sB+ETHnDkevAR1SSm2BB4G/FSqPspdInMIpnMqpdKITq7Jq1pEkSSpahRxJ2xwYk1L6MKU0FbgX6FV1gZTSkJTST/mbLwKrFTCPMjSDGRzBEfyNv3EUR3End3qhdEmSqlHIkrYq8GmV2+Py983LYcATc3sgIo6IiJERMXL8+PG1GFF15Qqu4CZu4kzO5GquZjF3h5QkqVpFcTLbiDgQ6ADsMLfHU0o3ADcAdOjQIdVhNNWSP/JHVmM19mXfrKNIklQSCjmc8RmwepXbq+Xv+4WI6AScAfRMKU0pYB7VsW/4hsM5nIlMpBGNLGiSJC2AQpa0/wAtI2KtiFgc2BfoV3WBiGgPXE+uoH1V2wG8bmd2Pudztmd77uAOXuXVrONIklRyCjbdmVKaHhFHA08B9YBbUkpvRcRfgJEppX7ApcBSwAMRATA2pdSztjJ43c5sfMAHdKITX/M1T/AEO7Jj1pEkSSo5Bd0nLaU0ABgwx31nV/m+UyG3r7o3mtF0pjPTmMYzPMNmbJZ1JEmSSpKH2KlWLc3SrM3aDGOYBU2SpEVgSVOteJVXmcEM1mANhjOcVrTKOpIkSSXNkqZFdh/3sSVbcimXAhBExokkSSp9ljQtkuu5nv3Yjy3ZkqM4Kus4kiSVDUuaFkoi8Vf+ypEcSQ968BRP0ZSmWceSJKlsWNK0UD7iI/7CXziAA3iIh2hEo6wjSZJUVorislAqHYlEEKzN2rzMy7ShjdfhlCSpAPztqhr7mZ/Zi724mZsB2IiNLGiSJBWIv2FVI5OYRA968BAP8SM/Zh1HkqSy53Sn5msCE+hOd17lVW7jNg7ioKwjSZJU9ixpqtZP/MT2bM8HfMBDPERPau3SqpIkqRqWNFVrSZbkEA6hAx3oSMes40iSVDEsaZqrUYziZ35mS7bkRE7MOo4kSRXHkqZfGcYwdmVXWtCC13jNIzglScqAv331C/3pTxe6sDIr8xiPWdAkScqIv4E1293czW/4Da1pzTCGsQZrZB1JkqSKVbYlrd15A2naqEHWMUpGIvEIj7AN2zCEIazIillHkiSpopXtPmkTJ0/j44t7ZB2j6CUSk5jE0izNHdzBTGZ6HU5JkopA2Y6kaf5mMpPjOZ6t2ZqJTGQJlrCgSZJUJCxpFWo60zmUQ7mSK9mZnWlCk6wjSZKkKixpFWjWhdJv4zbO4zz+wT88ilOSpCJTtvukad6O53ge5VH+yT85mqOzjiNJkubCklaBzuIsOtGJPdkz6yiSJGkenOOqEOMYxwmcwHSmswqrWNAkSSpylrQK8B7vsQ3bcBM38R7vZR1HkiTVgCWtzL3Kq2zLtkxmMs/yLK1pnXUkSZJUA5a0MjaMYezIjjSiEcMZTnvaZx1JkiTVkCWtjDWgAeuxHiMYwXqsl3UcSZK0ACxpZWg0owHYki15mZdZjdUyTiRJkhaUJa3M/JN/0pa2PMRDAASRcSJJkrQwLGllIpE4j/M4lmPpRS92YZesI0mSpEXgyWzLwKwLpfelL33ow43cSH0/WkmSSpojaWVgKEPpS1+O53hu5mYLmiRJZaAsf5u3O28gTRs1yDpGwSUSQdCRjrzAC2zBFu6DJklSmSjLkbSJk6fx+jldso5RUBOZyC7swlCGArkjOS1okiSVj7IsaeXuK75iR3ZkMIP5gi+yjiNJkgqgLKc7y9lYxtKZznzKpzzKox7FKUlSmbKklZDP+Ixt2IZJTGIgA9mWbbOOJEmSCsTpzhLSnObszu48x3MWNEmSypwjaSVgKENZM/+nL32zjiNJkuqAI2lFrh/96EIXjuO4rKNIkqQ6ZEkrYrdzO3uwB+1ox03clHUcSZJUhyxpRepKruRgDqYjHXmap1me5bOOJEmS6lDZlbRyuNrAFKZwG7exB3vQn/4sxVJZR5IkSXWs7A4cmDh5Gh9f3CPrGAtlJjOZylQa0pCneZomNPE6nJIkVaiyG0krVdOYxkEcxJ7syQxmsCzLWtAkSapglrQi8BM/sTu7cxd3sR3bsZgfiyRJFc+hmox9x3fsxm6MYATXcz1HcETWkSRJUhGwpGWsN715iZe4l3vZh32yjiNJkoqEJS1jf+WvjGc8XemadRRJklREymrnp1I5/cY7vMPf+TsAm7CJBU2SJP1KWY2klcLpN/7Df+hOdxrQgD70YQVWyDqSJEkqQmU1klbsnuZpdmInlmZphjPcgiZJkubJklZHHuZhdmEXWtCC4QxnHdbJOpIkSSpilrQ68hM/0YEOPMdzrMIqWceRJElFzpJWYGMYA8ABHMBQhrIcy2WcSJIklQJLWoEkEmdyJm1ow2u8BkA96mWcSpIklYqyOrqzWMxgBkdzNNdxHYdzOG1pm3UkSZJUYhxJq2VTmcoBHMB1XMcpnMIN3OAImiRJWmBlM5JWLCeyvZ3buY/7uIRLOJmTs44jSZJKVNmUtGI5ke1hHMa6rEtHOmYdRZIklTCnO2vB//gf3ejGB3xAEBY0SZK0yMqipGU51fkRH7Et2zKMYYxlbCYZJElS+Sn56c525w0E4PVzutT5tt/iLTrTmZ/5mad5mi3Zss4zSJKk8lSyJa3deQOZOHkaTRs1yKSgvcmbdKQjS7AEQxnKhmxY5xkkSVL5KtmSlvWBAmuxFl3owoVcyNqsnVkOSZJUnspin7S6NIhB/MAPLMVS3MM9FjRJklQQJVnSsjpQ4GZuphvdOI/z6nzbkiSpspRkSZs4eVqd74d2KZdyOIfThS6cy7l1um1JklR5SrKk1aVE4jRO42ROpje9eZRHaUzjrGNJkqQyZ0mbj6/4itu4jd/ze+7iLhZn8awjSZKkClCyR3cW2jSmUY96rMRKvMIrrMzKBJF1LEmSVCEcSZuLH/mRXdmVkzgJgOY0t6BJkqQ6ZUmbwzd8Q2c6M5jBtKFN1nEkSVKFKrnpzrc//56WBTr9xhd8QRe68B7v8QAPsAd7FGQ7kiRJ8xMppawzLJAlmrdMU754v9bXO53ptKUtYxnLozzKzuxc69uQJBXetGnTGDduHD///HPWUVRBGjZsyGqrrUaDBr8cSIqIV1JKHRZmnSU3klYo9anPRVxEc5qzBVtkHUeStJDGjRtHkyZNaNGiBRHuT6zCSykxYcIExo0bx1prrVVr6634fdJe4AXu4z4AfsNvLGiSVOJ+/vlnll9+eQua6kxEsPzyy9f66G1Fl7SneIpOdOI8zmMa07KOI0mqJRY01bVC/J2r2JJ2P/ezG7uxHusxhCE0oO6vBSpJkjQvFVnSbuAG9mVftmALhjCElVgp60iSpDJSr149Nt54YzbccEN22203vvvuu9mPvfXWW+y0006sv/76tGzZkvPPP5+qB/E98cQTdOjQgdatW9O+fXtOOOGEDF5B9V577TUOO+ywrGPM05QpU+jduzfrrrsuW2yxBR9//PFcl7vyyivZcMMNadOmDf/4xz9m33/uueey6qqrsvHGG7PxxhszYMAAAN5880369OlT+BeQV5ElbSxj6UY3nuIplmGZrONIkspMo0aNGDVqFKNHj2a55Zbj6quvBmDy5Mn07NmTU089lXfffZfXX3+d559/nmuuuQaA0aNHc/TRR3PnnXfy9ttvM3LkSNZdd91azTZ9+vRFXsdFF13EscceW6fbXBA333wzyy67LGPGjOH444/nlFNO+dUyo0eP5sYbb+Tll1/m9ddf5/HHH2fMmDGzHz/++OMZNWoUo0aNYpdddgFgo402Yty4cYwdO7ZOXkfFlLRE4lM+BeB8zqcf/ViSJTNOJUkqd1tttRWfffYZAHfffTfbbLMNXbp0AWDJJZfkqquu4uKLLwbgb3/7G2eccQYbbLABkBuRO+qoo361zh9++IFDDjmEjTbaiLZt2/Lvf/8bgKWWWmr2Mg8++ODsUZ8+ffpw5JFHssUWW3DyySfTokWLX4zutWzZki+//JLx48ez5557stlmm7HZZpsxYsSIX2170qRJvPHGG7Rr1w6Al19+ma222or27duz9dZb8+677wJw66230rNnT3baaSd23nlnfvzxRw499FA233xz2rdvz6OPPgrAxx9/zHbbbccmm2zCJptswvPPP7/Q7/Usjz76KAcffDAAe+21F08//TRznnLsnXfeYYsttmDJJZekfv367LDDDjz00EPzXfduu+3Gvffeu8gZa6IiTsExgxn8nt/zKI/yBm/QnObUr4yXLkkVr8Wp/Wt9nR9f3KNGy82YMYOnn3569tTgW2+9xaabbvqLZdZZZx1++OEHvv/+e0aPHl2j6c3zzz+fpk2b8uabbwLw7bffzvc548aN4/nnn6devXrMmDGDhx9+mEMOOYSXXnqJNddck5VWWon999+f448/nm233ZaxY8fStWtX3nnnnV+sZ+TIkWy44Yazb2+wwQYMGzaM+vXrM3jwYE4//fTZpfHVV1/ljTfeYLnlluP0009np5124pZbbuG7775j8803p1OnTqy44ooMGjSIhg0b8v7777PffvsxcuTIX+XfbrvtmDRp0q/u//vf/06nTp1+cd9nn33G6quvDkD9+vVp2rQpEyZMYIUVVpi9zIYbbsgZZ5zBhAkTaNSoEQMGDKBDh/8/ndlVV13F7bffTocOHbjssstYdtllAejQoQMXX3wxJ5988nzf80VV9k1lClM4gAP4N//mTM5kZVbOOpIkqQ7VtFDVpsmTJ7Pxxhvz2Wef0apVKzp37lyr6x88ePAvRnNmFYjq7L333tSrVw+A3r1785e//IVDDjmEe++9l969e89e79tvvz37Od9//z0//PDDL0bovvjiC5o1azb79sSJEzn44IN5//33iQimTfv/syV07tyZ5ZZbDoCBAwfSr18//v73vwO5U6WMHTuWVVZZhaOPPppRo0ZRr1493nvvvbnmHzZs2Hxf44Jo1aoVp5xyCl26dKFx48ZsvPHGs9+fo446irPOOouI4KyzzuKEE07glltuAWDFFVfk888/r9Us81LW050/8AO7sRv/5t9cwRWcz/leKF2SVHCz9kn75JNPSCnN3ietdevWvPLKK79Y9sMPP2SppZZi6aWXpk2bNr96fEFUPQ3EnOfsaty48ezvt9pqK8aMGcP48eN55JFH2GOP3GUQZ86cyYsvvjh7X6zPPvvsFwVt1muruu6zzjqLHXfckdGjR/PYY4/94rGq20wp8e9//3v2useOHUurVq244oorWGmllXj99dcZOXIkU6dOnetr22677WbvyF/1a/Dgwb9adtVVV+XTT3O7OE2fPp2JEyey/PLL/2q5ww47jFdeeYWhQ4ey7LLLst566wGw0korUa9ePRZbbDF+97vf8fLLL//ifW3UqNFcM9a2si5pF3ABz/AMt3Irx3Fc1nEkSRVmySWXpG/fvlx22WVMnz6dAw44gOHDh88uFpMnT+bYY4+dPXV20kkncdFFF80eTZo5cybXXXfdr9bbuXPn2cUP/n+6c6WVVuKdd95h5syZPPzww/PMFRHsvvvu/PnPf6ZVq1azC0yXLl345z//OXu5UaNG/eq5rVq1+sUO9hMnTmTVVVcFcvuhzUvXrl355z//OXvfsNdee23285s3b85iiy3GHXfcwYwZM+b6/GHDhs0ueFW/5pzqBOjZsye33XYbkNs3b6eddprrecy++uorAMaOHctDDz3E/vvvD+RGC2d5+OGHfzG9+9577/3idiGVdUk7m7MZxCAO5uCso0iSKlT79u1p27Yt99xzD40aNeLRRx/lggsuYP3112ejjTZis8024+ijjwagbdu2/OMf/2C//fajVatWbLjhhnz44Ye/WueZZ57Jt99+y4Ybbki7du0YMmQIABdffDG77rorW2+9Nc2bN682V+/evbnzzjtnT3UC9O3bl5EjR9K2bVtat24914K4wQYbMHHixNn7h5188smcdtpptG/fvtqjOM866yymTZtG27ZtadOmDWeddRYAf/jDH7jtttto164d//3vf38x+rawDjvsMCZMmMC6667L5ZdfPvvAjM8//3z2kZoAe+65J61bt2a33Xbj6quvZplllpn9mmYdlDFkyBCuuOKK2c8ZMmQIPXrUzRR62V1gfQxjOJmT+Rf/oilN6zCZJKkYvPPOO7Rq1SrrGGXtiiuuoEmTJhx++OFZR6lTU6ZMYYcddmD48OHUr//r3frn9ndvUS6wXlYjaa/zOtuyLUMZyljq5hwmkiRVmqOOOoolllgi6xh1buzYsVx88cVzLWiFUDZHd45gBD3oQROaMIQhtMJ/RUmSVAgNGzbkt7/9bdYx6lzLli1p2bJlnW2vLEbSnuEZOtOZlViJEYywoElShSu1XXlU+grxd64sStq6rEtnOjOMYazBGlnHkSRlqGHDhkyYMMGipjqTUmLChAk0bNiwVtdb0gcODGIQO7Mzi5VH15Qk1YJp06Yxbty4X50nTCqkhg0bstpqq9GgQYNf3L8oBw6U5D5picRf+StncAbXcR2/5/dZR5IkFYkGDRqw1lprZR1DWmQFHYKKiG4R8W5EjImIU+fy+BIRcV/+8ZciokVN1nsiJ3IGZ3AgB3Ioh9Z6bkmSpKwVrKRFRD3gaqA70BrYLyJaz7HYYcC3KaV1gSuAS+a33unLfsnlXM4xHMNt3EYDGszvKZIkSSWnkCNpmwNjUkofppSmAvcCveZYphdwW/77B4GdY27XbahiZqMfOJdzuZIr3RdNkiSVrULuk7Yq8GmV2+OALea1TEppekRMBJYHvq66UEQcARyRvznl3Dh39LmcW4jMKrwVmOPzVcnwsyttfn6ly8+utK2/sE8siQMHUko3ADcARMTIhT1KQtnz8ytdfnalzc+vdPnZlbaIGLmwzy3kfOFnwOpVbq+Wv2+uy0REfaApMKGAmSRJkkpCIUvaf4CWEbFWRCwO7Av0m2OZfsDB+e/3Ap5JpXbiNkmSpAIo2HRnfh+zo4GngHrALSmltyLiL8DIlFI/4GbgjogYA3xDrsjNzw2Fyqw64edXuvzsSpufX+nysyttC/35ldwVByRJkiqB57CQJEkqQpY0SZKkIlS0Ja1Ql5RS4dXgs/tzRLwdEW9ExNMRsWYWOTV38/v8qiy3Z0SkiPDUAEWkJp9fROyT/xl8KyLuruuMmrsa/L9zjYgYEhGv5f//uUsWOfVrEXFLRHwVEaPn8XhERN/8Z/tGRGxSk/UWZUkr1CWlVHg1/OxeAzqklNqSu9LE3+o2pealhp8fEdEE+BPwUt0mVHVq8vlFREvgNGCblFIb4Li6zqlfq+HP3pnA/Sml9uQOtLumblOqGrcC3ap5vDvQMv91BHBtTVZalCWNAl1SSnVivp9dSmlISumn/M0XyZ1DT8WhJj97AOeT+4fRz3UZTvNVk8/vd8DVKaVvAVJKX9VxRs1dTT67BCyd/74p8Hkd5lM1UkpDyZ2lYl56AbennBeBZSKi+fzWW6wlbW6XlFp1XsuklKYDsy4ppWzV5LOr6jDgiYIm0oKY7+eXH6ZfPaXUvy6DqUZq8vO3HrBeRIyIiBcjorp//avu1OSzOxc4MCLGAQOAY+ommmrBgv5uBErkslAqTxFxINAB2CHrLKqZiFgMuBzok3EULbz65KZcOpIbxR4aERullL7LMpRqZD/g1pTSZRGxFbnzjG6YUpqZdTAVRrGOpHlJqdJVk8+OiOgEnAH0TClNqaNsmr/5fX5NgA2BZyPiY2BLoJ8HDxSNmvz8jQP6pZSmpZQ+At4jV9qUrZp8docB9wOklF4AGpK7+LqKX41+N86pWEual5QqXfP97CKiPXA9uYLm/jDFpdrPL6U0MaW0QkqpRUqpBbl9CnumlBb6AsKqVTX5f+cj5EbRiIgVyE1/fliHGTV3NfnsxgI7A0REK3IlbXydptTC6gcclD/Kc0tgYkrpi/k9qSinOwt4SSkVWA0/u0uBpYAH8sd6jE0p9cwstGar4eenIlXDz+8poEtEvA3MAE5KKTkLkbEafnYnADdGxPHkDiLo4+BEcYiIe8j942eF/D6D5wANAFJK15Hbh3AXYAzwE3BIjdbr5ytJklR8inW6U5IkqaJZ0iRJkoqQJU2SJKkIWdIkSZKKkCVNkiSpCFnSJNW6iJgREaOqfLWoZtkfamF7t0bER/ltvZo/G/uCruOmWRe0jojT53js+UXNmF/PrPdldEQ8FhHLzGf5jSNil9rYtqTS4yk4JNW6iPghpbRUbS9bzTpuBR5PKT0YEV2Av6eU2i7C+hY50/zWGxG3Ae+llC6sZvk+QIeU0tG1nUVS8XMkTVLBRcRSEfF0fpTrzYjoNZdlmkfE0CojTdvl7+8SES/kn/tARMyvPA0F1s0/98/5dY2OiOPy9zWOiP4R8Xr+/t75+5+NiA4RcTHQKJ/jrvxjP+T/e29E9KiS+daI2Csi6kXEpRHxn4h4IyJ+X4O35QXyF1iOiM3zr/G1iHg+ItbPn3X+L0DvfJbe+ey3RMTL+WV/9T5KKh9FecUBSSWvUUSMyn//EbA3sHtK6fv8pYhejIh+c5wtfX/gqZTShRFRD1gyv+yZQKeU0o8RcQrwZ3LlZV52A96MiE3JndV7CyCAlyLiOWBt4POUUg+AiGha9ckppVMj4uiU0sZzWfd9wD5A/3yJ2hk4itw1FSemlDaLiCWAERExMH9tzF/Jv76dyV05BeC/wHb5s853Ai5KKe0ZEWdTZSQtIi4idwm8Q/NTpS9HxOCU0o/VvB+SSpQlTVIhTK5aciKiAXBRRGwPzCQ3grQS8L8qz/kPcEt+2UdSSqMiYgegNbnSA7A4uRGoubk0Is4kdy3Dw8iVoIdnFZiIeAjYDngSuCwiLiE3RTpsAV7XE8CV+SLWDRiaUpqcn2JtGxF75ZdrSu6i5XOWtFnldVXgHWBQleVvi4iW5C7302Ae2+8C9IyIE/O3GwJr5NclqcxY0iTVhQOAZsCmKaVpEfExuYIxW0ppaL7E9QBujYjLgW+BQSml/WqwjZNSSg/OuhERO89toZTSexGxCbnr6F0QEU+nlKobmav63J8j4lmgK9AbuHfW5oBjUkpPzWcVk1NKG0fEkuSu0fhHoC9wPjAkpbR7/iCLZ+fx/AD2TCm9W5O8kkqb+6RJqgtNga/yBW1HYM05F4iINYEvU0o3AjcBmwAvAttExKx9zBpHxHo13OYw4DcRsWRENAZ2B4ZFxCrATymlO4FL89uZ07T8iN7c3EduGnXWqBzkCtdRs54TEevltzlXKaWfgGOBEyKiPrn357P8w32qLDoJaFLl9lPAMZEfVoyI9vPahqTSZ0mTVBfuAjpExJvAQeT2wZpTR+D1iHiN3CjVlSml8eRKyz0R8Qa5qc4NarLBlNKrwK3Ay8BLwE0ppdeAjcjtyzUKOAe4YC5PvwF4Y9aBA3MYCOwADE4pTc3fdxPwNvBqRIwGrmc+MxX5LG8A+wF/A/6af+1VnzcEaD3rwAFyI24N8tneyt+WVKY8BYckSVIRciRNkiSpCFnSJEmSipAlTZIkqQhZ0iRJkoqQJU2SJKkIWdIkSZKKkCVNkiSpCP0fB2R/h4M//GwAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 720x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fpr, tpr, thresholds = roc_curve(binary_labels, reconstruction_scores)\n",
    "roc_auc = auc(fpr, tpr)\n",
    "\n",
    "plt.figure(figsize=(10,10))\n",
    "plt.plot(fpr, tpr, lw=1, label='ROC curve (area = %0.2f)' % roc_auc)\n",
    "plt.plot([0, 1], [0, 1], color='lime', linestyle='--')\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver operating characteristic')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 162,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.04293235\n"
     ]
    }
   ],
   "source": [
    "# We can pick the threshold based on the differeence between  the true positive rate (tpr) \n",
    "# and the false positive rate (fpr)\n",
    "optimal_threshold_idx = np.argmax(tpr - fpr)\n",
    "optimal_threshold = thresholds[optimal_threshold_idx]\n",
    "print(optimal_threshold)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 163,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "hqSbb5k1pa7c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.04293235\n",
      "Autoencoder Accuracy Score : 0.9239928557253423\n"
     ]
    }
   ],
   "source": [
    "# Use the optimal threshold value you just printed in the previous cell.\n",
    "thresh = optimal_threshold\n",
    "\n",
    "print(thresh)\n",
    "\n",
    "pred_labels = (reconstruction_scores > thresh).astype(int)\n",
    "\n",
    "results = confusion_matrix(binary_labels, pred_labels)\n",
    "# We are using our own accuracy score function in order to grade the assessment\n",
    "ae_acc = autoencoder_accuracy_score(binary_labels, pred_labels)\n",
    "print ('Autoencoder Accuracy Score :', ae_acc)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In order to pass the assessment, you need to an accuracy of at least 90%."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Confusion Matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This time, we'll create the confusion matrix for you."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 164,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "hqSbb5k1pa7c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion Matrix: \n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print ('Confusion Matrix: ')\n",
    "\n",
    "def plot_confusion_matrix(cm, target_names, title='Confusion matrix', cmap=plt.cm.Blues):\n",
    "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    plt.title(title)\n",
    "    plt.colorbar()\n",
    "    tick_marks = np.arange(len(target_names))\n",
    "    plt.xticks(tick_marks, target_names, rotation=45)\n",
    "    plt.yticks(tick_marks, target_names)\n",
    "    plt.tight_layout()\n",
    "\n",
    "    width, height = cm.shape\n",
    "\n",
    "    for x in range(width):\n",
    "        for y in range(height):\n",
    "            plt.annotate(str(cm[x][y]), xy=(y, x), \n",
    "                        horizontalalignment='center',\n",
    "                        verticalalignment='center')\n",
    "    plt.ylabel('True label')\n",
    "    plt.xlabel('Predicted label')\n",
    "\n",
    "\n",
    "plot_confusion_matrix(results, ['Normal','Anomaly'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "d8LUAaONQylG"
   },
   "source": [
    "## Assessment Task 11: Check Your Assessment Score"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Before proceeding, confirm your XGBoost model accuracy is greater than 95% and that your autoencoder accuracy is greater than 90%. If it isn't please continue work on the notebook until you've met these benchmarks."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "IR3BhxIQWyz9"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of the XGBoost Model:  0.9948402460805715\n",
      "Accuracy of the Autoencoder Model:  0.9239928557253423\n"
     ]
    }
   ],
   "source": [
    "print (\"Accuracy of the XGBoost Model: \", xgb_acc)\n",
    "print (\"Accuracy of the Autoencoder Model: \", ae_acc)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Run the following cell to grade your assessment."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 166,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Testing your XGBoost solution\n",
      "Required accuracy greater than 95%....\n",
      "Your XGBoost model is accurate enough!\n",
      "\n",
      "Testing your autoencoder solution\n",
      "Required accuracy greater than 90%....\n",
      "Your autoencoder model is accurate enough!\n",
      "\n",
      "You passed the assessment. Congratulations!!!!!\n",
      "\n",
      "See instructions below for how to get credit for your work.\n"
     ]
    }
   ],
   "source": [
    "from assessment import run_assessment\n",
    "run_assessment()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If the cell above tells you that you passed the assessment, read below for instructions on how to get credit for your work."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Get Credit for Your Work"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To get credit for your assessment and generate a certificate of competency for the course, return to the browser tab where you opened this JupyterLab environment and click the \"ASSESS TASK\" button, as shown below:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![get_credit](images/get_credit.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a href=\"https://www.nvidia.com/en-us/deep-learning-ai/education/\"> <img src=\"images/DLI Header.png\" alt=\"Header\" style=\"width: 400px;\"/> </a>"
   ]
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "name": "Final Assessment.ipynb",
   "provenance": [],
   "version": "0.3.2"
  },
  "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}