who-is-more-influential-AR.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "\n",
    "# المشروع الخامس"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "أهلاً بكم في المشروع الخامس من علم البيانات. في هذا المشروع، سنعمل معاً على التعرف على المؤثرين في الشبكات الاجتماعية باستخدام بيانات من موقع تويتر باستخدام العديد من السمات. ندعوك لقراءة تفاصيل المسابقة في كاقل:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- https://www.kaggle.com/c/predict-who-is-more-influential-in-a-social-network/data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "الهدف من هذا المشروع هو توقع المؤثرين في الشبكات الاجتماعية. البيانات الحالية مكونة من 5500 صف و 23 عمود. كل صف يحتوي على معلومات مستخدمين اثنين (11 متغير لكل منهم) بالإضافة إلى متغير الاستجابة والذي يحتوي على المستخدم الأكثر تأثيراً بين الاثنين (1 يعني أن A هو الأكثر تأثيراً بينما 0 يعني أن B هو الأكثر تأثيراً."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "\n",
    "# قراءة البيانات"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data = pd.read_csv('train.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['Choice', 'A_follower_count', 'A_following_count', 'A_listed_count',\n",
       "       'A_mentions_received', 'A_retweets_received', 'A_mentions_sent',\n",
       "       'A_retweets_sent', 'A_posts', 'A_network_feature_1',\n",
       "       'A_network_feature_2', 'A_network_feature_3', 'B_follower_count',\n",
       "       'B_following_count', 'B_listed_count', 'B_mentions_received',\n",
       "       'B_retweets_received', 'B_mentions_sent', 'B_retweets_sent', 'B_posts',\n",
       "       'B_network_feature_1', 'B_network_feature_2', 'B_network_feature_3'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_data.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "y = train_data['Choice']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data.drop(columns='Choice', inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "لنرى أولاً توزيع متغير الاستجابة:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Frequencies of each label in the dataset: \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "1    0.509455\n",
       "0    0.490545\n",
       "Name: Choice, dtype: float64"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print('Frequencies of each label in the dataset: ')\n",
    "y.value_counts(normalize=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "يوضح التوزيع أعلاه أن لدينا توازن بين الأصناف بحيث نستطيع المضي قدماً دون الاكتراث لهذا الأمر لاحقاً."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "لنلق الآن نظرة على أنواع المتغيرات لنتأكد أن جميع المتغيرات مقروءة بشكل سليم:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 5500 entries, 0 to 5499\n",
      "Data columns (total 22 columns):\n",
      " #   Column               Non-Null Count  Dtype  \n",
      "---  ------               --------------  -----  \n",
      " 0   A_follower_count     5500 non-null   int64  \n",
      " 1   A_following_count    5500 non-null   int64  \n",
      " 2   A_listed_count       5500 non-null   int64  \n",
      " 3   A_mentions_received  5500 non-null   float64\n",
      " 4   A_retweets_received  5500 non-null   float64\n",
      " 5   A_mentions_sent      5500 non-null   float64\n",
      " 6   A_retweets_sent      5500 non-null   float64\n",
      " 7   A_posts              5500 non-null   float64\n",
      " 8   A_network_feature_1  5500 non-null   int64  \n",
      " 9   A_network_feature_2  5500 non-null   float64\n",
      " 10  A_network_feature_3  5500 non-null   float64\n",
      " 11  B_follower_count     5500 non-null   int64  \n",
      " 12  B_following_count    5500 non-null   int64  \n",
      " 13  B_listed_count       5500 non-null   int64  \n",
      " 14  B_mentions_received  5500 non-null   float64\n",
      " 15  B_retweets_received  5500 non-null   float64\n",
      " 16  B_mentions_sent      5500 non-null   float64\n",
      " 17  B_retweets_sent      5500 non-null   float64\n",
      " 18  B_posts              5500 non-null   float64\n",
      " 19  B_network_feature_1  5500 non-null   int64  \n",
      " 20  B_network_feature_2  5500 non-null   float64\n",
      " 21  B_network_feature_3  5500 non-null   float64\n",
      "dtypes: float64(14), int64(8)\n",
      "memory usage: 945.4 KB\n"
     ]
    }
   ],
   "source": [
    "train_data.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "لنبدأ باستعراض ملخص للبيانات باستخدام describe:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>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",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>A_follower_count</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>649883.95</td>\n",
       "      <td>2028787.44</td>\n",
       "      <td>16.0</td>\n",
       "      <td>2663.75</td>\n",
       "      <td>45589.00</td>\n",
       "      <td>392738.00</td>\n",
       "      <td>36543194.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_following_count</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>12658.95</td>\n",
       "      <td>49008.67</td>\n",
       "      <td>0.0</td>\n",
       "      <td>322.00</td>\n",
       "      <td>778.00</td>\n",
       "      <td>2838.00</td>\n",
       "      <td>1165830.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_listed_count</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>5952.45</td>\n",
       "      <td>17339.14</td>\n",
       "      <td>0.0</td>\n",
       "      <td>85.00</td>\n",
       "      <td>932.00</td>\n",
       "      <td>6734.00</td>\n",
       "      <td>549144.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_mentions_received</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>2666.03</td>\n",
       "      <td>29165.43</td>\n",
       "      <td>0.1</td>\n",
       "      <td>3.45</td>\n",
       "      <td>48.77</td>\n",
       "      <td>349.82</td>\n",
       "      <td>1145218.99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_retweets_received</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>1032.37</td>\n",
       "      <td>10954.95</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.72</td>\n",
       "      <td>14.03</td>\n",
       "      <td>118.70</td>\n",
       "      <td>435825.87</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_mentions_sent</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>6.01</td>\n",
       "      <td>9.52</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.36</td>\n",
       "      <td>2.30</td>\n",
       "      <td>7.20</td>\n",
       "      <td>76.81</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_retweets_sent</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>1.11</td>\n",
       "      <td>1.91</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.10</td>\n",
       "      <td>0.34</td>\n",
       "      <td>1.32</td>\n",
       "      <td>16.29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_posts</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>9.09</td>\n",
       "      <td>18.31</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.63</td>\n",
       "      <td>3.56</td>\n",
       "      <td>10.69</td>\n",
       "      <td>193.07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_network_feature_1</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>5267.77</td>\n",
       "      <td>28946.78</td>\n",
       "      <td>0.0</td>\n",
       "      <td>12.00</td>\n",
       "      <td>195.00</td>\n",
       "      <td>1323.00</td>\n",
       "      <td>920838.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_network_feature_2</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>84.81</td>\n",
       "      <td>104.07</td>\n",
       "      <td>0.0</td>\n",
       "      <td>14.99</td>\n",
       "      <td>54.93</td>\n",
       "      <td>109.70</td>\n",
       "      <td>1121.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A_network_feature_3</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>3747.22</td>\n",
       "      <td>5757.34</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1180.81</td>\n",
       "      <td>2206.42</td>\n",
       "      <td>4390.38</td>\n",
       "      <td>144651.46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_follower_count</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>685486.86</td>\n",
       "      <td>2160612.35</td>\n",
       "      <td>20.0</td>\n",
       "      <td>2498.25</td>\n",
       "      <td>44027.00</td>\n",
       "      <td>370114.25</td>\n",
       "      <td>36543194.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_following_count</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>12738.26</td>\n",
       "      <td>50054.52</td>\n",
       "      <td>0.0</td>\n",
       "      <td>322.00</td>\n",
       "      <td>773.00</td>\n",
       "      <td>2838.00</td>\n",
       "      <td>664324.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_listed_count</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>5903.15</td>\n",
       "      <td>16298.46</td>\n",
       "      <td>0.0</td>\n",
       "      <td>75.00</td>\n",
       "      <td>890.00</td>\n",
       "      <td>6734.00</td>\n",
       "      <td>549144.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_mentions_received</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>2554.60</td>\n",
       "      <td>25088.73</td>\n",
       "      <td>0.1</td>\n",
       "      <td>3.26</td>\n",
       "      <td>48.77</td>\n",
       "      <td>374.37</td>\n",
       "      <td>1145218.99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_retweets_received</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>997.15</td>\n",
       "      <td>9342.01</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.71</td>\n",
       "      <td>14.03</td>\n",
       "      <td>107.08</td>\n",
       "      <td>435825.87</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_mentions_sent</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>6.10</td>\n",
       "      <td>9.73</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.36</td>\n",
       "      <td>2.25</td>\n",
       "      <td>6.87</td>\n",
       "      <td>76.81</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_retweets_sent</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>1.11</td>\n",
       "      <td>1.94</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.10</td>\n",
       "      <td>0.34</td>\n",
       "      <td>1.32</td>\n",
       "      <td>16.29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_posts</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>9.51</td>\n",
       "      <td>19.42</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.82</td>\n",
       "      <td>3.34</td>\n",
       "      <td>10.60</td>\n",
       "      <td>193.07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_network_feature_1</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>5254.93</td>\n",
       "      <td>26778.82</td>\n",
       "      <td>0.0</td>\n",
       "      <td>11.00</td>\n",
       "      <td>190.00</td>\n",
       "      <td>1323.00</td>\n",
       "      <td>920838.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_network_feature_2</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>85.02</td>\n",
       "      <td>106.69</td>\n",
       "      <td>0.0</td>\n",
       "      <td>15.18</td>\n",
       "      <td>54.93</td>\n",
       "      <td>112.19</td>\n",
       "      <td>1861.58</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B_network_feature_3</th>\n",
       "      <td>5500.0</td>\n",
       "      <td>3745.18</td>\n",
       "      <td>5518.40</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1206.50</td>\n",
       "      <td>2206.42</td>\n",
       "      <td>4349.91</td>\n",
       "      <td>75526.08</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                      count       mean         std   min      25%       50%  \\\n",
       "A_follower_count     5500.0  649883.95  2028787.44  16.0  2663.75  45589.00   \n",
       "A_following_count    5500.0   12658.95    49008.67   0.0   322.00    778.00   \n",
       "A_listed_count       5500.0    5952.45    17339.14   0.0    85.00    932.00   \n",
       "A_mentions_received  5500.0    2666.03    29165.43   0.1     3.45     48.77   \n",
       "A_retweets_received  5500.0    1032.37    10954.95   0.1     0.72     14.03   \n",
       "A_mentions_sent      5500.0       6.01        9.52   0.1     0.36      2.30   \n",
       "A_retweets_sent      5500.0       1.11        1.91   0.1     0.10      0.34   \n",
       "A_posts              5500.0       9.09       18.31   0.1     0.63      3.56   \n",
       "A_network_feature_1  5500.0    5267.77    28946.78   0.0    12.00    195.00   \n",
       "A_network_feature_2  5500.0      84.81      104.07   0.0    14.99     54.93   \n",
       "A_network_feature_3  5500.0    3747.22     5757.34   0.0  1180.81   2206.42   \n",
       "B_follower_count     5500.0  685486.86  2160612.35  20.0  2498.25  44027.00   \n",
       "B_following_count    5500.0   12738.26    50054.52   0.0   322.00    773.00   \n",
       "B_listed_count       5500.0    5903.15    16298.46   0.0    75.00    890.00   \n",
       "B_mentions_received  5500.0    2554.60    25088.73   0.1     3.26     48.77   \n",
       "B_retweets_received  5500.0     997.15     9342.01   0.1     0.71     14.03   \n",
       "B_mentions_sent      5500.0       6.10        9.73   0.1     0.36      2.25   \n",
       "B_retweets_sent      5500.0       1.11        1.94   0.1     0.10      0.34   \n",
       "B_posts              5500.0       9.51       19.42   0.1     0.82      3.34   \n",
       "B_network_feature_1  5500.0    5254.93    26778.82   0.0    11.00    190.00   \n",
       "B_network_feature_2  5500.0      85.02      106.69   0.0    15.18     54.93   \n",
       "B_network_feature_3  5500.0    3745.18     5518.40   0.0  1206.50   2206.42   \n",
       "\n",
       "                           75%          max  \n",
       "A_follower_count     392738.00  36543194.00  \n",
       "A_following_count      2838.00   1165830.00  \n",
       "A_listed_count         6734.00    549144.00  \n",
       "A_mentions_received     349.82   1145218.99  \n",
       "A_retweets_received     118.70    435825.87  \n",
       "A_mentions_sent           7.20        76.81  \n",
       "A_retweets_sent           1.32        16.29  \n",
       "A_posts                  10.69       193.07  \n",
       "A_network_feature_1    1323.00    920838.00  \n",
       "A_network_feature_2     109.70      1121.00  \n",
       "A_network_feature_3    4390.38    144651.46  \n",
       "B_follower_count     370114.25  36543194.00  \n",
       "B_following_count      2838.00    664324.00  \n",
       "B_listed_count         6734.00    549144.00  \n",
       "B_mentions_received     374.37   1145218.99  \n",
       "B_retweets_received     107.08    435825.87  \n",
       "B_mentions_sent           6.87        76.81  \n",
       "B_retweets_sent           1.32        16.29  \n",
       "B_posts                  10.60       193.07  \n",
       "B_network_feature_1    1323.00    920838.00  \n",
       "B_network_feature_2     112.19      1861.58  \n",
       "B_network_feature_3    4349.91     75526.08  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_data.describe().round(decimals=2).T # استخدمت ت لتسهيل القراءة (T)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "من خلال تفحص قيم الانحراف المعياري (std)، نجد أن أغلب المتغيرات لديها قيم عالية جداً بما يعكس توزيعاً عريضاً للبيانات، وهذا لا يساعدنا بشكل كبير. سنقوم بتجهيز البيانات وإنشاء بعض المتغيرات الجديدة."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "\n",
    "# تجهيز البيانات"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "الهدف من تجهيز البيانات هو تنظيف البيانات والتحقق منها بالإضافة إلى إضافة أي متغيرات جديدة قد تكون مهمة في عملية التصنيف. سنبدأ أولاً بالتأكد من عدم وجود أي بيانات مفقودة، وملئها بالطريقة المناسبة إن وجدت:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_data.isna().any().sum() # لا يوجد قيم مفقودة"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "كيف يمكن أن ننشيء أعمدة جديدة؟ سنقوم بإنشاء أعمدة جديدة تعبر عن الفرق في القيم بين المستخدم A والمستخدم B. فمثلا، بالإضافة إلى A_follower_count و B_follower_count، نستطيع إيجاد الفرق بين المتغيرين وإنشاء متغير ثالث يعبر عن الفرق. سأترك المهمة لك:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "\n",
    "#### ملاحضة : استخدمت طريقتين في انشاء الاعمدة الجديدة    \n",
    "    \n",
    "#### الاولى هي عن طريق استخدام فور لوب لتسريع العملية\n",
    "    \n",
    "#### الثانية هي الطريقة العادية"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# in this new function, we will perform the following operations and make a new\n",
    "# dataframe only for the new variables\n",
    "\n",
    "# A_follower_count - B_follower_count\n",
    "# A_following_count - B_following_count\n",
    "# A_listed_count - B_listed_count\n",
    "# A_mentions_received - B_mentions_received\n",
    "# A_retweets_received - B_retweets_received\n",
    "# A_mentions_sent - B_mentions_sent\n",
    "# A_retweets_sent - B_retweets_sent\n",
    "# A_posts - B_posts\n",
    "# A_network_feature_1 - B_network_feature_1\n",
    "# A_network_feature_2 - B_network_feature_2\n",
    "# A_network_feature_3 - B_network_feature_3\n",
    "\n",
    "def get_substracted_features(df):\n",
    "    \n",
    "    a_cols = [\n",
    "        'A_follower_count',\n",
    "        'A_following_count',\n",
    "        'A_listed_count',\n",
    "        'A_mentions_received',\n",
    "        'A_retweets_received',\n",
    "        'A_mentions_sent',\n",
    "        'A_retweets_sent',\n",
    "        'A_posts',\n",
    "        'A_network_feature_1',\n",
    "        'A_network_feature_2',\n",
    "        'A_network_feature_3',\n",
    "    ]\n",
    "    \n",
    "    b_cols = [\n",
    "        'B_follower_count',\n",
    "        'B_following_count',\n",
    "        'B_listed_count',\n",
    "        'B_mentions_received',\n",
    "        'B_retweets_received',\n",
    "        'B_mentions_sent',\n",
    "        'B_retweets_sent',\n",
    "        'B_posts',\n",
    "        'B_network_feature_1',\n",
    "        'B_network_feature_2',\n",
    "        'B_network_feature_3',\n",
    "    ]\n",
    "    \n",
    "    # الطريقة الاولى عن طريق الفور لوب\n",
    "    combined = pd.DataFrame()\n",
    "    \n",
    "    for i in range(len(a_cols)):\n",
    "        \n",
    "        combined[f'{a_cols[i][2:]}_subst'] = df[a_cols[i]] - df[b_cols[i]]\n",
    "    \n",
    "    return combined"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "إضافة إلى متغيرات الفرق، نستطيع كذلك استخدام متغيرات جديدة عبارة عن متغيرات النسبة. فمثلا، نستطيع حساب نسبة متابعي مستخدم ما إلى من يتابعهم، وذلك مثلاً بقسمة A_follower_count على A_following_count وهكذا. لنقم بكتابة الأمر في الخانة أدناه:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# in this new function, we will perform the following operations and make a new\n",
    "# dataframe only for the new variables\n",
    "\n",
    "# A_follower_count / A_following_count\n",
    "# B_follower_count / B_following_count\n",
    "# A_mentions_received / A_mentions_sent\n",
    "# A_retweets_received / A_retweets_sent\n",
    "# B_mentions_received / B_mentions_sent\n",
    "# B_retweets_received / B_retweets_sent\n",
    "# A_posts / A_listed_count\n",
    "# B_posts / B_listed_count\n",
    "\n",
    "def get_ratio_features(df):\n",
    "    \n",
    "    a = [\n",
    "        'A_follower_count',\n",
    "        'B_follower_count',\n",
    "        'A_mentions_received',\n",
    "        'A_retweets_received',\n",
    "        'B_mentions_received',\n",
    "        'B_retweets_received',\n",
    "        'A_posts',\n",
    "        'B_posts',\n",
    "    ]\n",
    "    \n",
    "    b = [\n",
    "        'A_following_count',\n",
    "        'B_following_count',\n",
    "        'A_mentions_sent',\n",
    "        'A_retweets_sent',\n",
    "        'B_mentions_sent',\n",
    "        'B_retweets_sent',\n",
    "        'A_listed_count',\n",
    "        'B_listed_count',\n",
    "    ]\n",
    "    \n",
    "    # الطريقة الثانية\n",
    "    combined = pd.DataFrame()\n",
    "    \n",
    "    combined['r_follower_count_A'] = df[a[0]] / df[b[0]]\n",
    "    combined['r_follower_count_B'] = df[a[1]] / df[b[1]]\n",
    "    combined['r_mentions_A'] = df[a[2]] / df[b[2]]\n",
    "    combined['r_RT_A'] = df[a[3]] / df[b[3]]\n",
    "    combined['r_mentions_B'] = df[a[4]] / df[b[4]]\n",
    "    combined['r_RT_B'] = df[a[5]] / df[b[5]]\n",
    "    combined['r_postlist_A'] = df[a[6]] / df[b[6]]\n",
    "    combined['r_postlist_B'] = df[a[7]] / df[b[7]]\n",
    "    \n",
    "    return combined"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "سنقوم الآن بدمج هذه العمليات في أمر واحد ليسهل تطبيقه مرة واحدة على بيانات الاختبار. قمنا كذلك بإضافة المزيد من اللمسات الإضافية والتي سنرى أثرها بعد قليل."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_process_features(df):\n",
    "    X_subst_features = get_substracted_features(np.log(df+1))     \n",
    "    X_ratio_features = get_ratio_features(df+1)    \n",
    "    X_features = pd.concat([np.log(df+1), X_subst_features, X_ratio_features], axis=1)\n",
    "\n",
    "    X_features['r_follower_count_A'] = np.log(X_features['r_follower_count_A'])\n",
    "    X_features['r_follower_count_B'] = np.log(X_features['r_follower_count_B'])\n",
    "    X_features['r_mentions_A'] = np.log(X_features['r_mentions_A'])\n",
    "    X_features['r_mentions_B'] = np.log(X_features['r_mentions_B'])\n",
    "    X_features['r_RT_A'] = np.log(X_features['r_RT_A'])\n",
    "    X_features['r_RT_B'] = np.log(X_features['r_RT_B'])\n",
    "    X_features['r_postlist_A'] = np.log(X_features['r_postlist_A'])\n",
    "    X_features['r_postlist_B'] = np.log(X_features['r_postlist_B'])\n",
    "    \n",
    "    return X_features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data = get_process_features(train_data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "لنلق الآن نظرة على الأعمدة بعد التجهيز:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['A_follower_count', 'A_following_count', 'A_listed_count',\n",
       "       'A_mentions_received', 'A_retweets_received', 'A_mentions_sent',\n",
       "       'A_retweets_sent', 'A_posts', 'A_network_feature_1',\n",
       "       'A_network_feature_2', 'A_network_feature_3', 'B_follower_count',\n",
       "       'B_following_count', 'B_listed_count', 'B_mentions_received',\n",
       "       'B_retweets_received', 'B_mentions_sent', 'B_retweets_sent', 'B_posts',\n",
       "       'B_network_feature_1', 'B_network_feature_2', 'B_network_feature_3',\n",
       "       'follower_count_subst', 'following_count_subst', 'listed_count_subst',\n",
       "       'mentions_received_subst', 'retweets_received_subst',\n",
       "       'mentions_sent_subst', 'retweets_sent_subst', 'posts_subst',\n",
       "       'network_feature_1_subst', 'network_feature_2_subst',\n",
       "       'network_feature_3_subst', 'r_follower_count_A', 'r_follower_count_B',\n",
       "       'r_mentions_A', 'r_RT_A', 'r_mentions_B', 'r_RT_B', 'r_postlist_A',\n",
       "       'r_postlist_B'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_data.columns"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "الخطوة الأخيرة في تجهيز البيانات هي في تسوية البيانات. في هذه المرة، لن نقوم بعمل ذلك يدوياً ولكن باستخدام أمر StandardScalar() من مكتبة scikit-learn."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import StandardScaler\n",
    "\n",
    "def preprocess_data(df, scaler=None):\n",
    "    '''\n",
    "    In this function, we will standardize data to have a mean of zero and a unit variance. The standardized data\n",
    "    and a scalar object (to be used for testing data) are returned.\n",
    "    '''\n",
    "    if not scaler:\n",
    "        scaler = StandardScaler()\n",
    "        scaler.fit(df.values)\n",
    "    df = pd.DataFrame(scaler.transform(df.values), columns=df.columns)\n",
    "    return df, scaler"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data, scaler = preprocess_data(train_data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "حتى نرى أثر الخطوات التي أجريناها على تجهيز البيانات، سنقم برسم المتغيرات ولكن بتمييز متغير الاستجابة. في الحالة المثالية، نريد أن نرى تمايزاً واضحا في أكبر عدد ممكن من المتغيرات بين الأصناف في متغير الاستجابة."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "features_names = ['following_count_subst', 'listed_count_subst', 'mentions_received_subst',\n",
    "       'retweets_received_subst', 'mentions_sent_subst', 'retweets_sent_subst', 'posts_subst',\n",
    "       'network_feature_1_subst', 'network_feature_2_subst', 'network_feature_3_subst',\n",
    "       'r_follower_count_A', 'r_follower_count_B', 'r_mentions_A',\n",
    "       'r_mentions_B', 'r_RT_A', 'r_RT_B', 'r_postlist_A', 'r_postlist_B']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_histograms(X_features, features_names, Y,  cols = 3, figsize=(8,20), **args):\n",
    "    fig = plt.figure(figsize=figsize)\n",
    "    total = len(features_names)\n",
    "    rows = total // cols \n",
    "    rows += total % cols\n",
    "    position = range(1,total + 1)\n",
    "    for k in range(total):\n",
    "        ax = fig.add_subplot(rows,cols,position[k])\n",
    "        plt.hist(X_features[features_names[k]].values[Y==0], label='Zero', **args)\n",
    "        plt.hist(X_features[features_names[k]].values[Y==1], label='One', **args)\n",
    "        plt.title(features_names[k])\n",
    "        plt.legend(loc='best')\n",
    "        plt.axis('off')\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x1440 with 18 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_histograms(train_data, features_names, y.values, cols = 3, figsize=(15,20), alpha=.5, bins=30)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "يوضح الشكل أعلاه أن لدينا توزيعين مختلفين للأصناف المختلفة وهذا خبر جيد جداً."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "\n",
    "# نمذجة البيانات"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "سنقوم الآن بنمذجة البيانات باستخدام عدة نماذج ونرى كيف يمكن أن نحسن من أداء النموذج. تحديدا، سنقوم بتدريب النموذج باستخدام البيانات  (training_data) ونرى أداء النموذج باستخدام عدد من المُدخلات. لنبدأ بالانحدار اللوجستي. في الخانة أدناه، قم باستيراد كلاً من مكتبة الانحدار اللوجستي، وأمر cross_val_score ثم استخدمهم في تدريب النموذج والتحقق من صحته."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.model_selection import cross_val_score\n",
    "\n",
    "model1 = LogisticRegression()\n",
    "scores = cross_val_score(model1, train_data, y, cv=3, scoring='accuracy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.7709077867434209"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scores.mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "الآن سنكرر نفس العملية ولكن باستخدام قيمة أخرى للمُدخل C وهو يعبر عن مقدار الضبط (regularization)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.7710894396780379"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model2 = LogisticRegression(C=3.0)\n",
    "scores = cross_val_score(model2, train_data, y, cv=3, scoring='accuracy')\n",
    "scores.mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "هل تلاحظ أي فرق في الدقة ؟"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "على الأغلب لاحظت أن هناك فرقاً طفيفاً في أداء النموذج باستخدام قيمة أخرى للمدخل. في الحقيقة كل قيمة ل C ستعطيك نتيجة مختلفة. فكيف نعرف القيمة الصحيحة التي يجب علينا استخدامها؟ سنقوم بالبحث عن هذه القيمة باستخدام  Grid Search."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "في الخانة التالية، تجد قيم مختلفة ل C وقيم مختلفة ل fit_intercept. ما عليك الآن هو قراءة الأمر GridSearchCV ومن ثم استخدام القيم من الخانة التالية في البحث عن أفضل نموذج ممكن لتصنيف البيانات. ابدأ أولا بقراءة توثيق هذا الأمر:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- http://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html#sklearn.model_selection.GridSearchCV"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "param_grid = {\n",
    " 'C':[0.0001, 0.001, 0.01 , 0.1, 0.3, 0.5, 0.7, 0.9, 1.0, 10., 100.],\n",
    " 'fit_intercept': [True, False]\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\muhammed\\anaconda3\\lib\\site-packages\\sklearn\\linear_model\\_logistic.py:763: ConvergenceWarning: lbfgs failed to converge (status=1):\n",
      "STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.\n",
      "\n",
      "Increase the number of iterations (max_iter) or scale the data as shown in:\n",
      "    https://scikit-learn.org/stable/modules/preprocessing.html\n",
      "Please also refer to the documentation for alternative solver options:\n",
      "    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression\n",
      "  n_iter_i = _check_optimize_result(\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(estimator=LogisticRegression(),\n",
       "             param_grid={'C': [0.0001, 0.001, 0.01, 0.1, 0.3, 0.5, 0.7, 0.9,\n",
       "                               1.0, 10.0, 100.0],\n",
       "                         'fit_intercept': [True, False]})"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.model_selection import GridSearchCV\n",
    "\n",
    "gsearch1 = GridSearchCV(model1, param_grid)\n",
    "gsearch1.fit(train_data,y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "بعد أن أنجزنا عملية البحث، سنقوم بطباعة أفضل \"تشكيلة\" من المدخلات والتي تملك أعلى درجة:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'C': 100.0, 'fit_intercept': False}"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gsearch1.best_params_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "وهنا سنقوم بطباعة أفضل درجة توصل لها البحث باستخدام التشكيلة أعلاه:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.7707272727272728"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gsearch1.best_score_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "عملية البحث عن نموذج عملية قد تأخذ وقتاً طويلاً جداً. فأعلاه، لدينا 22 نموذج مختلف (١١ ضرب ٢)، كل نموذچ يمر عبر التحقق المتقاطع ذو 5 أقسام بما معناه أننا نقوم بمطابقة النموذج 110 مرات! وهنا يأتي دور الإمكانيات المتقدمة في تسريع هذه العملية. بشكل عام، لا غنى عن أداء هذا البحث في أي نموذج معقد يتطلب العديد من المدخلات. لنكرر نفس الأمر أدناه ولكن مع نموذج الغابات العشوائية"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "GridSearchCV(estimator=RandomForestClassifier(),\n",
       "             param_grid={'max_features': ['auto', 'log2', 'sqrt'],\n",
       "                         'min_samples_leaf': [2, 3, 5],\n",
       "                         'min_samples_split': [2, 3, 5]})"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.ensemble import RandomForestClassifier\n",
    "\n",
    "param_grid = {\n",
    " 'min_samples_split':[2,3,5],\n",
    " 'min_samples_leaf':[2,3,5],\n",
    " 'max_features': ['auto', 'log2', 'sqrt']\n",
    "}\n",
    "\n",
    "model3 = RandomForestClassifier()\n",
    "gsearch2 = GridSearchCV(model3, param_grid)\n",
    "gsearch2.fit(train_data,y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "بعد أن أنجزنا عملية البحث، سنقوم بطباعة أفضل \"تشكيلة\" من المدخلات والتي تملك أعلى درجة:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'max_features': 'auto', 'min_samples_leaf': 5, 'min_samples_split': 5}"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gsearch2.best_params_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "وهنا سنقوم بطباعة أفضل درجة توصل لها البحث باستخدام التشكيلة أعلاه:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.7814545454545454"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gsearch2.best_score_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "\n",
    "# تسليم المشروع"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"direction:rtl; text-align:right\">\n",
    "قبل تسليم المشروع في موقع كاقل، سوف نستخدم النموذج الأفضل من النماذج أعلاه لمطابقته على جميع البيانات."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = LogisticRegression(**gsearch1.best_params_) # نلاحظ هنا طريقة استخدام المدخلات"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LogisticRegression(C=100.0, fit_intercept=False)"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(train_data, y) # training model on full data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_data = pd.read_csv('test.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "test_data = get_process_features(test_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_data = pd.DataFrame(scaler.transform(test_data.values), columns=test_data.columns)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_data['Choice'] = model.predict(test_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_data.index.name = 'Id'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_data.index = test_data.index+1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Id\n",
       "1    0\n",
       "2    1\n",
       "3    0\n",
       "4    0\n",
       "5    1\n",
       "Name: Choice, dtype: int64"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_data['Choice'].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_data['Choice'].to_csv('./social_media_analysis_submission.csv', header=True)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}