Presentation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "### Author: Pedram Daee -* pedram.daee@aalto.fi *\n",
    "# Ad Install Probability Prediction\n",
    "\n",
    "Goal: Predict install probability for ad install event given a set of features. Each row of data is the description of the impression and whether it concluded with an install event or not.\n",
    "\n",
    "This notebook goes through a typical data analysis workflow. There are two main sections:\n",
    "\n",
    "- The data: description of the data and feature engineering methods used. \n",
    "- The model: description of the preprocessing and modelling decisions."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## The data\n",
    "\n",
    "<a id='TL;DR'></a>\n",
    "<div class=\" alert alert-warning\">\n",
    "    <b>tl;dr.</b> \n",
    "\n",
    "  <li>The ultimate task is to recommend ads but the problem statement is not acknowledging this. Rather than modelling $p(y|x)$, where $y$ is the install event and $x$ contains the features, it would be more suitable to model $p(y|x,a)$ from the data. The conceptual difference is that $x$ is assumed to be given and we do not have any control in its creation, while $a$ is our action. Currently, $a$ is provided as a feature in $x$. This is my main criticism to the task, as I think the prediction and recommendation should be modelled jointly (and not having a model for recommending campaigns and then using the recommended campaings as part of the feature space of a model for predicting installs). </li>\n",
    "  <li>To create a representation for the action, we could have exploited the user interactions with the advertised ads by creating user ID vs campaign ID matrix and use a function (e.g., latent factorization) of it for action representation, but the user ID is missing from the data. However, user ID (history of campaigns recommended) is missing from the data but potentially important. I guess this is available as the data contains features such as installCount.</li>\n",
    "    <li> The install probability then can be modelled as a function of both $p(y|x)$ (prior install probability; what is done in this assignment) and $p(y|a,hist,\\theta)$ (a personalized model of the user, given the history of interaction of the user and some unknown parameters $\\theta$). The uncertainty in the second component can then help the decision making process (e.g., recommend an ad to gain more information about the user or to exploit the current best found ad).</li>\n",
    "  <li>I have done many quick sub-optimal hacks for feature engineering. These hacks are not the best ways to handle the data. My aim was to try to not add many features for categorical data coding at this stage, as I was more interested to just investigate a baseline solution and the feature contributions.</li>\n",
    "  <li>Quick summary of engineerings: I am removing the two time features and creating a new one expressing the difference between those two in minutes (also took care of nulls and added suggestions for other ways to exploit the time features), I am removing some of the categorical features, and replacing others by values such as frequency, I am replacing the software version by the first integer in the string, I am replacing countries by a feature representing prior knowledge for per country install ratio (a bit cheating),... </li>\n",
    "</div>\n",
    " "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following blocks imports common libraries, loads the training data, and prints some primitive statistics. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 2206990 entries, 0 to 2206989\n",
      "Data columns (total 17 columns):\n",
      "id                 object\n",
      "timestamp          object\n",
      "campaignId         object\n",
      "platform           object\n",
      "softwareVersion    object\n",
      "sourceGameId       int64\n",
      "country            object\n",
      "startCount         int64\n",
      "viewCount          int64\n",
      "clickCount         int64\n",
      "installCount       int64\n",
      "lastStart          object\n",
      "startCount1d       int64\n",
      "startCount7d       int64\n",
      "connectionType     object\n",
      "deviceType         object\n",
      "install            int64\n",
      "dtypes: int64(8), object(9)\n",
      "memory usage: 286.2+ MB\n",
      "None\n",
      "Number of installs: 26497,\n",
      "Percentage of installs: 0.012006,\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "tr_filename = \"training_data.csv\"\n",
    "# load the whole training data (no need to chunk it!)\n",
    "tr_data = pd.read_csv(tr_filename, sep=';')\n",
    "# print basic statistics about the data\n",
    "print(tr_data.info())\n",
    "Y = tr_data['install']\n",
    "print(\"Number of installs: {},\".format(Y.sum()))\n",
    "print(\"Percentage of installs: {:f},\".format((Y.sum()/Y.shape[0])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The training data contains $n=2206990$ instances with $d=16$ features and the corresponding responses $y \\in \\{0,1\\}$  indicates whether install was observed (install=1) or was not observed (install=0) after impression. The data set is very unbalanced as the occurrence of an install event is only $\\%1.2$.\n",
    "\n",
    "The following line prints the first five rows of the dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "with pd.option_context('display.max_rows', 5, 'display.max_columns', None, 'display.width', 100):\n",
    "    print(tr_data.head())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's look at the null values in the data. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of nulls per feature:\n",
      "id                      0\n",
      "timestamp               0\n",
      "campaignId              0\n",
      "platform                0\n",
      "softwareVersion         0\n",
      "sourceGameId            0\n",
      "country               101\n",
      "startCount              0\n",
      "viewCount               0\n",
      "clickCount              0\n",
      "installCount            0\n",
      "lastStart          163180\n",
      "startCount1d            0\n",
      "startCount7d            0\n",
      "connectionType          0\n",
      "deviceType              0\n",
      "install                 0\n",
      "dtype: int64\n",
      "number of installs in country nulls: 1\n",
      "number of installs in lastStart nulls: 2821\n"
     ]
    }
   ],
   "source": [
    "## check for duplicated data rows --> no need. IDs are all unique\n",
    "#print(\"There are {} duplicated rows \".format(tr_data.duplicated().sum()))\n",
    "\n",
    "print(\"Number of nulls per feature:\")\n",
    "print(tr_data.isnull().sum())\n",
    "\n",
    "print(\"number of installs in country nulls: {}\".format(Y[tr_data['country'].isnull()].sum()))\n",
    "print(\"number of installs in lastStart nulls: {}\".format(Y[tr_data['lastStart'].isnull()].sum()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "lastStart feature has a huge number of nulls where 2821 of those rows have install event (around 10 percent of the total rows with install event), therefore it is better to handle them properly, rather than just dropping the rows. I will take care of these bellow where we look deeper at individual features.\n",
    "\n",
    "Let's look at the features one by one"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "id\n",
      "count                      2206990\n",
      "unique                     2206990\n",
      "top       5c362dd39722fe1afb4a52a0\n",
      "freq                             1\n",
      "Name: id, dtype: object \n",
      "\n",
      "timestamp\n",
      "count                           2206990\n",
      "unique                          2204923\n",
      "top       2019-01-10T12:14:36.205+00:00\n",
      "freq                                  3\n",
      "Name: timestamp, dtype: object \n",
      "\n",
      "campaignId\n",
      "count                      2206990\n",
      "unique                        5704\n",
      "top       5c3bfb0b36c2c6cc18710e7b\n",
      "freq                         41740\n",
      "Name: campaignId, dtype: object \n",
      "\n",
      "platform\n",
      "count     2206990\n",
      "unique          2\n",
      "top       android\n",
      "freq      1591868\n",
      "Name: platform, dtype: object \n",
      "\n",
      "softwareVersion\n",
      "count     2206990\n",
      "unique        174\n",
      "top         8.0.0\n",
      "freq       360593\n",
      "Name: softwareVersion, dtype: object \n",
      "\n",
      "sourceGameId\n",
      "count    2.206990e+06\n",
      "mean     3.085592e+06\n",
      "std      1.376251e+07\n",
      "min      1.105500e+04\n",
      "25%      1.202538e+06\n",
      "50%      1.631784e+06\n",
      "75%      2.639966e+06\n",
      "max      1.316274e+08\n",
      "Name: sourceGameId, dtype: float64 \n",
      "\n",
      "country\n",
      "count     2206889\n",
      "unique        220\n",
      "top            US\n",
      "freq       359949\n",
      "Name: country, dtype: object \n",
      "\n",
      "startCount\n",
      "count    2.206990e+06\n",
      "mean     4.831307e+01\n",
      "std      9.039421e+01\n",
      "min      1.000000e+00\n",
      "25%      6.000000e+00\n",
      "50%      1.900000e+01\n",
      "75%      5.300000e+01\n",
      "max      3.997000e+03\n",
      "Name: startCount, dtype: float64 \n",
      "\n",
      "viewCount\n",
      "count    2.206990e+06\n",
      "mean     3.595887e+01\n",
      "std      7.899328e+01\n",
      "min      0.000000e+00\n",
      "25%      2.000000e+00\n",
      "50%      1.000000e+01\n",
      "75%      3.400000e+01\n",
      "max      2.886000e+03\n",
      "Name: viewCount, dtype: float64 \n",
      "\n",
      "clickCount\n",
      "count    2.206990e+06\n",
      "mean     2.099431e+00\n",
      "std      6.797011e+00\n",
      "min      0.000000e+00\n",
      "25%      0.000000e+00\n",
      "50%      0.000000e+00\n",
      "75%      2.000000e+00\n",
      "max      1.031000e+03\n",
      "Name: clickCount, dtype: float64 \n",
      "\n",
      "installCount\n",
      "count    2.206990e+06\n",
      "mean     5.703424e-01\n",
      "std      1.543641e+00\n",
      "min      0.000000e+00\n",
      "25%      0.000000e+00\n",
      "50%      0.000000e+00\n",
      "75%      1.000000e+00\n",
      "max      7.300000e+01\n",
      "Name: installCount, dtype: float64 \n",
      "\n",
      "lastStart\n",
      "count                           2043810\n",
      "unique                          2042213\n",
      "top       2019-01-16T10:21:58.859+00:00\n",
      "freq                                  2\n",
      "Name: lastStart, dtype: object \n",
      "\n",
      "startCount1d\n",
      "count    2.206990e+06\n",
      "mean     6.787689e+00\n",
      "std      1.136146e+01\n",
      "min      1.000000e+00\n",
      "25%      2.000000e+00\n",
      "50%      3.000000e+00\n",
      "75%      8.000000e+00\n",
      "max      1.466000e+03\n",
      "Name: startCount1d, dtype: float64 \n",
      "\n",
      "startCount7d\n",
      "count    2.206990e+06\n",
      "mean     1.978492e+01\n",
      "std      3.468965e+01\n",
      "min      1.000000e+00\n",
      "25%      3.000000e+00\n",
      "50%      9.000000e+00\n",
      "75%      2.300000e+01\n",
      "max      3.280000e+03\n",
      "Name: startCount7d, dtype: float64 \n",
      "\n",
      "connectionType\n",
      "count     2206990\n",
      "unique          3\n",
      "top          wifi\n",
      "freq      1497765\n",
      "Name: connectionType, dtype: object \n",
      "\n",
      "deviceType\n",
      "count       2206990\n",
      "unique        23637\n",
      "top       iPhone8,1\n",
      "freq          58565\n",
      "Name: deviceType, dtype: object \n",
      "\n",
      "install\n",
      "count    2.206990e+06\n",
      "mean     1.200594e-02\n",
      "std      1.089119e-01\n",
      "min      0.000000e+00\n",
      "25%      0.000000e+00\n",
      "50%      0.000000e+00\n",
      "75%      0.000000e+00\n",
      "max      1.000000e+00\n",
      "Name: install, dtype: float64 \n",
      "\n"
     ]
    }
   ],
   "source": [
    "for name in tr_data.columns:\n",
    "    print(name)\n",
    "    print(tr_data[name].describe(), \"\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *id*: \n",
    "* Indicates the impression ID. \n",
    "* All ID's are unique. \n",
    "* This feature will be discarded."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [],
   "source": [
    "# remove \"id\" column and save the data in a new variable\n",
    "tr_data_m = tr_data.drop(columns=\"id\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *timestamp* and *lastStart*:   \n",
    "* **timestamp**: time of the event in UTC.\n",
    "* **LastStart**: the time of users last start event (any campaign) (I assume in UTC).\n",
    "* These are the two time features in the data. Potentially can be useful as:\n",
    "* 1) their interaction might have an influence on the response variable. \n",
    "* 2) user's behavior may depend on the time of the day (e.g., people might be more likely to install an app in the evening or over the weekend than working hours(?)). To test this, we also need to convert the UTC time zone to a local time zone using data in the \"country\" feature and also add new features (e.g., \"weekend\" or \"holiday\") to the data.  \n",
    "* For now I only consider the first point due to time constrains. \n",
    "* In the following code box, first I replace the nan values in 'lastStart' with the corresponding values in 'timestamp' and then I define a new feature ('epoch_diff') as the difference in minutes between the 'timestamp' and 'lastStart', and drop the old features. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# replace nan values in 'lastStart' with values in 'timestamp'\n",
    "nan_idx = tr_data['lastStart'].isnull()\n",
    "tr_data_m['lastStart'].loc[nan_idx] = tr_data_m['timestamp'][nan_idx]\n",
    "\n",
    "# convert the string objects to datetime\n",
    "tr_data_m['timestamp'] = pd.to_datetime(tr_data_m['timestamp'],infer_datetime_format=True)\n",
    "tr_data_m['lastStart'] = pd.to_datetime(tr_data_m['lastStart'],infer_datetime_format=True)\n",
    "\n",
    "# convert times to epoch in minutes\n",
    "epoch_timestamp = pd.to_datetime(tr_data_m['timestamp']).astype(np.int64) // (10**9 * 60)\n",
    "epoch_lastStart = pd.to_datetime(tr_data_m['lastStart']).astype(np.int64) // (10**9 * 60)\n",
    "\n",
    "# create a new feature of difference in timestamp and lastStart in minutes\n",
    "tr_data_m['epoch_diff'] = epoch_timestamp - epoch_lastStart\n",
    "\n",
    "#print(tr_data_m['timestamp'][:10])\n",
    "#print(tr_data_m['lastStart'][:10])\n",
    "#print(tr_data_m['epoch_diff'][:10])\n",
    "\n",
    "# remove \"timestamp\" column\n",
    "tr_data_m = tr_data_m.drop(columns=\"timestamp\")\n",
    "tr_data_m = tr_data_m.drop(columns=\"lastStart\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *campaignId*: \n",
    "* The id of the advertising campaign. \n",
    "* I guess this is all we have about the recommended ads. \n",
    "* There are 5704 unique campaigns.\n",
    "* The dataset is biased based on this feature, since the campainId is selected by an intelligent algorithm.\n",
    "* The following code box looks at the distribution of campaigns over number of appearances and number of installs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# compute campain frequencies\n",
    "campainId_counts = tr_data_m['campaignId'].value_counts()\n",
    "#find the number of installs for each campainId\n",
    "install_1_frame = tr_data_m[tr_data_m['install'] == 1]\n",
    "install_per_campain = np.zeros(campainId_counts.shape[0])\n",
    "for i in range(campainId_counts.shape[0]):  \n",
    "    install_per_campain[i] = sum(install_1_frame['campaignId'] == campainId_counts.index[i])\n",
    "\n",
    "# plot campain frequencies against number of installs\n",
    "plt.figure()\n",
    "plt.plot(campainId_counts,install_per_campain,'.', label='campain')\n",
    "plt.plot([0,campainId_counts[0]],[0,campainId_counts[0]*(Y.sum()/Y.shape[0])], label='average install')\n",
    "plt.legend()\n",
    "plt.title(\"campain frequency against number of installs\")\n",
    "plt.xlabel(\"campain frequency\")\n",
    "plt.ylabel(\"number of installs\")\n",
    "\n",
    "# plot install ratio for frequent campains\n",
    "#install_ratio = install_per_campain/campainId_counts\n",
    "#plt.figure()\n",
    "#plt.plot(range(1000),100*install_ratio[:1000],'.')\n",
    "#plt.title(\"Install ratio of popular campains (more than {} appearance)\".format(campainId_counts[100]))\n",
    "#plt.xlabel(\"frequent campains (sorted)\")\n",
    "#plt.ylabel(\"install ratio\")\n",
    "\n",
    "plt.show()\n",
    "\n",
    "#print(campainId_counts)\n",
    "#print(install_per_campain)\n",
    "#print(install_ratio)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* The figure shows three things: 1) the correlation between number of installs and number of recommendation of the campaigns 2) huge number of campaigns have only been recommended very few times 3) A bit concerning point that there are a few extremely frequent recommended campaigns that have lower than average install ratio, while some better than average campaigns are not recommended that frequently. This may (or may not) indicate a sub-optimal performance in the recommendation part (e.g., see the blue point at 30000).\n",
    "* We need to find a good representation for campaigns (see next box)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a id='What I would have done..'></a>\n",
    "<div class=\" alert alert-warning\">\n",
    "    <b>What I would have done...</b> \n",
    "\n",
    "<br />\n",
    "It would be crucially important for the core recommendation task to have a feature representation for campaigns. It seems that the data does not have this information. I can think of two ways to do this: (i) try to generate features from the campaign itself (e.g., from the textual description of the campaign or the video).(ii) generate features for campaign based on the interaction data (e.g., if two users installed a couple of similar campaigns, then the information about one user can be generalized to the other one for the recommendation). It seems that approach (i) has not been successful. However, it should be possible to exploit the interaction data in (ii). For this, I would need to have a unique user ID to be able to create the sparse matrix of campaignId vs. user_id, where the elements indicate install or no-install event. I could then use matrix factorization and try to estimate this big matrix with two simpler matrix (e.g., AV), and learn these simpler (low dimensional) matrices by minimizing the reconstruction error. A simple solution then would be to use rows of A as a feature representation for campaigns.   \n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* As for now, I just represent each campaign with its frequency. Please note that this is not a good (or stable) approach since as the designer of the recommender system, we are also controlling the frequency of campaign observations. Anyway, this would be a hack for now (but not that optimal as we are losing a lot of information). Next code block creates this new feature."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# first conver the campainId_counts to a dict\n",
    "campainId_dict = campainId_counts.to_dict()\n",
    "\n",
    "campain_freq = np.zeros(tr_data_m.shape[0])\n",
    "for i in range(tr_data_m.shape[0]):\n",
    "    campain_freq[i] = campainId_dict[tr_data_m['campaignId'][i]]\n",
    "\n",
    "# add the new feature to the data frame\n",
    "tr_data_m.insert(1, 'campain_freq', campain_freq)\n",
    "\n",
    "# remove 'campaignId' column\n",
    "tr_data_m = tr_data_m.drop(columns=\"campaignId\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *platform*: \n",
    "* User's platform (two options: android and ios)\n",
    "* The following box convert this to binary feature."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "android    1591868\n",
      "ios         615122\n",
      "Name: platform, dtype: int64\n"
     ]
    }
   ],
   "source": [
    "# convert platform to binary values (1:ios, 0:android)\n",
    "print(tr_data_m['platform'].value_counts())\n",
    "tr_data_m['platform'] = tr_data_m['platform'] == 'ios' \n",
    "tr_data_m['platform'] = tr_data_m['platform'].astype(np.int64)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *softwareVersion*: \n",
    "* Software version of the device.\n",
    "* There are 174 different software versions.  \n",
    "* To simplify this feature, we extract the first number in the 'softwareVersion' string and consider it as the representative of the software version. The following code block builds this new feature 'softwareVersion_num' and drops 'softwareVersion' column."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {},
   "outputs": [],
   "source": [
    "# the following simple function finds the first number in an string\n",
    "def find_first_num(input_str):\n",
    "    for i, c in enumerate(input_str):\n",
    "        if c.isdigit():\n",
    "            start = i\n",
    "            while i < len(input_str) and input_str[i].isdigit():\n",
    "                i += 1\n",
    "            #print('Integer %d found at position %d' % (int(input_str[start:i]), start))\n",
    "            return int(input_str[start:i])\n",
    "    return 0\n",
    "\n",
    "# save the first number in the version string in an int array\n",
    "soft_ver_num = np.zeros(tr_data_m.shape[0])\n",
    "for i in range(tr_data_m.shape[0]):\n",
    "    soft_ver_num[i] = find_first_num(tr_data_m['softwareVersion'][i])\n",
    "\n",
    "# add the new feature to the data frame\n",
    "tr_data_m.insert(1, 'softwareVersion_num', soft_ver_num)\n",
    "\n",
    "# remove 'softwareVersion' column\n",
    "tr_data_m = tr_data_m.drop(columns=\"softwareVersion\")\n",
    "\n",
    "# uncomment the following two lines to see how the values have changed\n",
    "#print(tr_data_m['softwareVersion_num'].value_counts())\n",
    "#print(tr_data['softwareVersion'].value_counts())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *sourceGameId*: \n",
    "* The game id of the publishing game.\n",
    "* Similar analysis to 'campaignId' will be applied here."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# compute sourceGameId frequencies\n",
    "gameId_counts = tr_data_m['sourceGameId'].value_counts()\n",
    "#find the number of installs for each gameId\n",
    "install_1_frame = tr_data_m[tr_data_m['install'] == 1]\n",
    "install_per_gameId = np.zeros(gameId_counts.shape[0])\n",
    "for i in range(gameId_counts.shape[0]):  \n",
    "    install_per_gameId[i] = sum(install_1_frame['sourceGameId'] == gameId_counts.index[i])\n",
    "\n",
    "# plot gameId frequencies against number of installs\n",
    "plt.figure()\n",
    "plt.plot(gameId_counts,install_per_gameId,'.', label='gameID')\n",
    "plt.plot([0,max(gameId_counts)],[0,max(gameId_counts)*(Y.sum()/Y.shape[0])], label='average install')\n",
    "plt.legend()\n",
    "plt.title(\"gameId frequency against number of installs\")\n",
    "plt.xlabel(\"gameId frequency\")\n",
    "plt.ylabel(\"number of installs\")\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* Clear linear correlation (probably not much information in this coding). Let's use the frequency (similar to 'campaignId') as the new feature and remove the old feature (again we are losing information but this is just a quick hack)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {},
   "outputs": [],
   "source": [
    "# first convert the gameId_counts to a dict\n",
    "gameId_dict = gameId_counts.to_dict()\n",
    "\n",
    "gameId_freq = np.zeros(tr_data_m.shape[0])\n",
    "for i in range(tr_data_m.shape[0]):\n",
    "    gameId_freq[i] = gameId_dict[tr_data_m['sourceGameId'][i]]\n",
    "\n",
    "# add the new feature to the data frame\n",
    "tr_data_m.insert(1, 'gameId_freq', gameId_freq)\n",
    "\n",
    "# remove 'sourceGameId' column\n",
    "tr_data_m = tr_data_m.drop(columns=\"sourceGameId\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *country*: \n",
    "* The country of the user (total 220 countries).\n",
    "* This feature can also help us to convert the time zones to local time (see explanations above).\n",
    "* There are 101 nulls in this column (only one of them has install event). We drop the corresponding rows."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "101\n",
      "0\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# drop all the nulls (they are only from from 'country' feature) the dataset.\n",
    "print(tr_data_m['country'].isnull().sum())\n",
    "tr_data_m.dropna(axis=0, how='any',inplace = True)\n",
    "print(tr_data_m['country'].isnull().sum())\n",
    "tr_data_m.reset_index(inplace=True)\n",
    "\n",
    "# compute country frequencies\n",
    "country_counts = tr_data_m['country'].value_counts()\n",
    "#find the number of installs for each country\n",
    "install_1_frame = tr_data_m[tr_data_m['install'] == 1]\n",
    "install_per_country = np.zeros(country_counts.shape[0])\n",
    "for i in range(country_counts.shape[0]):  \n",
    "    install_per_country[i] = sum(install_1_frame['country'] == country_counts.index[i])\n",
    "\n",
    "# visualize the install ratio for the most popular countries\n",
    "install_per_country_ratio = install_per_country/country_counts\n",
    "plt.plot(install_per_country_ratio[:30]*100,'.')\n",
    "plt.plot([0,30],[100*Y.sum()/Y.shape[0],100*Y.sum()/Y.shape[0]], label='average install')\n",
    "plt.xticks(rotation=90)\n",
    "plt.title(\"Install ratio of frequent countries\")\n",
    "plt.xlabel(\"frequent countries\")\n",
    "plt.ylabel(\"install ratio\")\n",
    "plt.show()\n",
    "\n",
    "#print(country_counts)\n",
    "#print(install_per_country)\n",
    "#print(install_per_country_ratio)    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* I will use the ratio of installs of each country as a new feature value. This is to some extent justifiable as it encapsulates the prior knowledge we have about each country (e.g., people in the country \"TR\" are less likely to install a recommendation). At the same time it is a bit dangerous since we use the training data label information here. The following code box creates this new feature and removes the old one (note: It would have make sense to also have a feature for frequency to also cover sufficient statistics for the accuracy of the ratio)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [],
   "source": [
    "# first convert the install_per_country_ratio to a dict\n",
    "country_dict = install_per_country_ratio.to_dict()\n",
    "country_freq = np.zeros(tr_data_m.shape[0])\n",
    "for i in range(tr_data_m.shape[0]):\n",
    "    country_freq[i] = country_dict[tr_data_m['country'][i]]\n",
    "\n",
    "# add the new feature to the data frame\n",
    "tr_data_m.insert(1, 'country_ratio', country_freq)\n",
    "\n",
    "# remove 'country' column\n",
    "tr_data_m = tr_data_m.drop(columns=\"country\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### *startCount*, *viewCount*, *clickCount*, *installCount*, *startCount1d*, and *startCount7d* : \n",
    "* **startCount**: how many times the user has ever started a campaign ever\n",
    "* **viewCount**: how many times the user has ever viewed a campaign ever\n",
    "* **clickCount**: how many times the user has ever clicked a campaign ever\n",
    "* **installCount**: how many times the user has installed games from this network ever\n",
    "* **startCount1d**: how many times user has started (any) campaigns within the last 24h\n",
    "* **startCount7d**: how many times user has started (any) campaigns within the last 7 days\n",
    "* These seems to be in a a good format. Let's keep them that way.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### connectionType : \n",
    "* Internet connection type (three types: wifi, cellular, none)\n",
    "* The following code block creates two new features ('wifi' and 'cellular') and removes 'connectionType'."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "wifi        1497694\n",
      "cellular     691313\n",
      "none          17882\n",
      "Name: connectionType, dtype: int64\n"
     ]
    }
   ],
   "source": [
    "print(tr_data_m['connectionType'].value_counts())\n",
    "\n",
    "tr_data_m['wifi'] = tr_data_m['connectionType'] == 'wifi' \n",
    "tr_data_m['wifi'] = tr_data_m['wifi'].astype(np.int64)\n",
    "\n",
    "tr_data_m['cellular'] = tr_data_m['connectionType'] == 'cellular' \n",
    "tr_data_m['cellular'] = tr_data_m['cellular'].astype(np.int64)\n",
    "\n",
    "# remove 'connectionType' column\n",
    "tr_data_m = tr_data_m.drop(columns=\"connectionType\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  deviceType:              \n",
    "* Device model (total of 22476 models).\n",
    "* Would be interesting to see if there is a different install behavior between people with more high-tech devices or people with tablets, or different brands (iphone vs. Samsung). The following code block looks at the install ratio of most popular devices. \n",
    "* Due to time constrains, I will remove this column for now (may get back to this later)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# compute device frequencies\n",
    "device_counts = tr_data_m['deviceType'].value_counts()\n",
    "#find the number of installs for each device\n",
    "install_1_frame = tr_data_m[tr_data_m['install'] == 1]\n",
    "install_per_device = np.zeros(device_counts.shape[0])\n",
    "for i in range(device_counts.shape[0]):  \n",
    "    install_per_device[i] = sum(install_1_frame['deviceType'] == device_counts.index[i])\n",
    "\n",
    "# visualize the install ratio for the most popular device types\n",
    "install_per_device_ratio = install_per_device/device_counts\n",
    "plt.plot(install_per_device_ratio[:30]*100,'.')\n",
    "plt.plot([0,30],[100*Y.sum()/Y.shape[0],100*Y.sum()/Y.shape[0]], label='average install')\n",
    "plt.title(\"Install ratio of frequent devices\")\n",
    "plt.xticks(rotation=90)\n",
    "plt.xlabel(\"frequent devices\")\n",
    "plt.ylabel(\"install ratio\")\n",
    "plt.show()\n",
    "\n",
    "# remove 'deviceType' column for now (take care of this later)\n",
    "tr_data_m = tr_data_m.drop(columns=\"deviceType\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now let's look at the data after all the feature modifications."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "   index  country_ratio  campain_freq  gameId_freq  softwareVersion_num  platform  startCount  \\\n",
      "0      0       0.014333          58.0        528.0                 11.0         1          25   \n",
      "1      1       0.014333          58.0        204.0                 12.0         1          10   \n",
      "2      2       0.014333          58.0        528.0                 12.0         1          27   \n",
      "3      3       0.014333          58.0        111.0                 12.0         1          15   \n",
      "4      4       0.014333          58.0        528.0                 12.0         1          20   \n",
      "\n",
      "   viewCount  clickCount  installCount  startCount1d  startCount7d  install  epoch_diff  wifi  \\\n",
      "0         24           0             2             1             8        0        1591     0   \n",
      "1          9           2             0             1             4        0        2663     0   \n",
      "2         26           0             0             1             2        0        2941     0   \n",
      "3         14           2             0             3            14        0        1152     1   \n",
      "4         18           0             0            13            18        0           0     0   \n",
      "\n",
      "   cellular  \n",
      "0         1  \n",
      "1         1  \n",
      "2         1  \n",
      "3         0  \n",
      "4         1  \n"
     ]
    }
   ],
   "source": [
    "#tr_data_m = tr_data_m.drop(columns=\"index\")\n",
    "with pd.option_context('display.max_rows', 5, 'display.max_columns', None, 'display.width', 100):\n",
    "    print(tr_data_m.head())\n",
    "#tr_data_m"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And finally, save the data as a csv file. I also need to save the dictionaries and other variables necessary for feature transformation as I will need to do the same transformations on the test data (however, I am not saving them at the moment)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save the modified dataframe for future use\n",
    "tr_data_m.to_csv(r'tr_data_m.csv', index = None)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## The Method\n",
    "\n",
    "<a id='TL;DR'></a>\n",
    "<div class=\" alert alert-warning\">\n",
    "    <b>tl;dr.</b> \n",
    "\n",
    "  <li> The data is extremely unbalanced, methods with high flexibility may overfit and simpler methods will classify most points to no install class. </li>\n",
    "  <li> I use under-sampling of the no install class as a quick solution to the problem. There are better (more expensive) ways to handle this but given my time this is fine. </li>\n",
    "  <li> I only consider two baseline like methods in the assignment: decision tree and logistic regression, and one advanced model, random forest. I could investigate more powerful methods, or spend more time on hyperparameter tuning, but the time is limited and I also believe that spending more time on feature engineering and extracting more features (as explained in the previous section) would probably be more fruitful.</li>\n",
    "  <li> After balancing, I use 5-fold cross validation for evaluation of each method and then just divide the training data to 2 to investigate the outputs.</li>\n",
    "  <li> I use ROC (true positive rate (TPR) against the false positive rate (FPR); kind of suitable for unbalanced data, but we are balancing the data anyway!) and confusion matrix as the performance measure criteria.</li>\n",
    "  <li> Discussions about the results and modelling decisions are provided.</li>\n",
    "  \n",
    "</div>\n",
    " "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following code box reads the modified training data from the CSV file and removes the \"index\" and label (\"install\") columns from the data. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load the modified training data\n",
    "tr_filename = \"tr_data_m.csv\"\n",
    "tr_data_m = pd.read_csv(tr_filename)\n",
    "tr_data_m = tr_data_m.drop(columns=\"index\")\n",
    "Y = tr_data_m['install']\n",
    "tr_data_m = tr_data_m.drop(columns=\"install\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 2206889 entries, 0 to 2206888\n",
      "Data columns (total 14 columns):\n",
      "country_ratio          float64\n",
      "campain_freq           float64\n",
      "gameId_freq            float64\n",
      "softwareVersion_num    float64\n",
      "platform               int64\n",
      "startCount             int64\n",
      "viewCount              int64\n",
      "clickCount             int64\n",
      "installCount           int64\n",
      "startCount1d           int64\n",
      "startCount7d           int64\n",
      "epoch_diff             int64\n",
      "wifi                   int64\n",
      "cellular               int64\n",
      "dtypes: float64(4), int64(10)\n",
      "memory usage: 235.7 MB\n",
      "None\n",
      "number of data in class 1: 26496\n",
      "number of data in class 0: 2180393\n"
     ]
    }
   ],
   "source": [
    "# Print some basic statistics\n",
    "print(tr_data_m.info())\n",
    "print(\"number of data in class 1: {}\".format(sum(Y==1)))\n",
    "print(\"number of data in class 0: {}\".format(sum(Y==0)))\n",
    "#tr_data_m"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  Data standardization and balancing:              \n",
    "\n",
    "To standardize or not to standardize? I decided to do so as the scales are quite different in different columns. However, it would make sense to also run the methods without standardization or different normalization. The following code standardizes the data and saves them in data frame X. \n",
    "\n",
    "**Note**: The normalization should be done inside the CV folds on the training data only. However, here I do it once for both training and validation data, which is wrong but not drastically wrong given the data size that we have."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# the following code normalizes the data (zero mean, unit std)\n",
    "from sklearn import preprocessing\n",
    "# Get column names first\n",
    "names = tr_data_m.columns\n",
    "# Create the Scaler object\n",
    "scaler = preprocessing.StandardScaler()\n",
    "# Fit your data on the scaler object\n",
    "scaler.fit(tr_data_m)\n",
    "X = scaler.transform(tr_data_m)\n",
    "X = pd.DataFrame(X, columns=names)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following box performs under-sampling to take care of the unbalanceness of the data. This can be quite harmful as we are losing information by throwing the data from class 0 out. However, for now I prefer to go with this quick solution. The new data are stored in X_resampled, Y_resampled.\n",
    "\n",
    "* For now I am just using random undersampling. A better approach would be to cluster the data of class 0 and then use the prototypes as the representations of the class. This would be a bit expensive computationally (as k-means is expensive if there are many data), but doable (codes for this are commented bellow). See https://imbalanced-learn.readthedocs.io/en/stable/under_sampling.html for more information."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[(0, 26496), (1, 26496)]\n"
     ]
    }
   ],
   "source": [
    "from collections import Counter\n",
    "from sklearn.datasets import make_classification\n",
    "from imblearn.under_sampling import ClusterCentroids\n",
    "\n",
    "# ClusterCentroids makes use of K-means to reduce the number of samples. \n",
    "#Therefore, each class will be synthesized with the centroids of the K-means method instead of the original samples\n",
    "#print(sorted(Counter(Y).items()))\n",
    "#cc = ClusterCentroids(random_state=0)\n",
    "#X_resampled, Y_resampled = cc.fit_resample(X, Y)\n",
    "#print(sorted(Counter(Y_resampled).items()))\n",
    "\n",
    "# random undersampling\n",
    "from imblearn.under_sampling import RandomUnderSampler\n",
    "rus = RandomUnderSampler(random_state=0)\n",
    "X_resampled, Y_resampled = rus.fit_resample(X, Y)\n",
    "X_resampled = pd.DataFrame(X_resampled, columns=names)\n",
    "print(\"number of data for each class after resampling:\")\n",
    "print(sorted(Counter(Y_resampled).items()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  Regularized logistic regression on balanced data:              \n",
    "* Let's start with regularized (l2) logistic regression as a simple baseline on the balanced dataset and look at the cross validation performance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.61574012 0.65189213 0.60842385 0.63913734 0.66559331 0.62946899\n",
      " 0.65623765 0.63272208 0.65068522 0.63493906]\n"
     ]
    }
   ],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.model_selection import cross_val_score\n",
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "logreg = LogisticRegression(solver='lbfgs')\n",
    "\n",
    "roc_auc_scores = cross_val_score(logreg, X_resampled, Y_resampled, scoring = 'roc_auc', cv=5)\n",
    "print(roc_auc_scores)    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "It seems that the cross validation ROC scores are very close. The following plots the ROC Curve after dividing the data in two equal-size random sets."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import roc_auc_score\n",
    "from sklearn.metrics import roc_curve\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "X_train, X_val, Y_train, Y_val = train_test_split(X_resampled, Y_resampled, test_size=0.5)\n",
    "\n",
    "logreg.fit(X_train, Y_train)\n",
    "y_pred_lr = logreg.predict(X_val)\n",
    "\n",
    "logit_roc_auc = roc_auc_score(Y_val, y_pred_lr)\n",
    "fpr, tpr, thresholds = roc_curve(Y_val, logreg.predict_proba(X_val)[:,1])\n",
    "plt.figure()\n",
    "plt.plot(fpr, tpr, label='Logistic Regression (area = %0.2f)' % logit_roc_auc)\n",
    "plt.plot([0, 1], [0, 1],'r--')\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 for logistic regression')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.savefig('Log_ROC')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following function is used to plot the confusion matrix."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {},
   "outputs": [],
   "source": [
    "# This function is used to plot the confusion matrix and normalized confusion matrix\n",
    "# copied from: https://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html\n",
    "# reference: API design for machine learning software: experiences from the scikit-learn project, Buitinck et al., 2013.\n",
    "def plot_confusion_matrix(y_true, y_pred, classes,\n",
    "                          normalize=False,\n",
    "                          title=None,\n",
    "                          cmap=plt.cm.Blues):\n",
    "    \"\"\"\n",
    "    This function prints and plots the confusion matrix.\n",
    "    Normalization can be applied by setting `normalize=True`.\n",
    "    \"\"\"\n",
    "    if not title:\n",
    "        if normalize:\n",
    "            title = 'Normalized confusion matrix'\n",
    "        else:\n",
    "            title = 'Confusion matrix, without normalization'\n",
    "\n",
    "    # Compute confusion matrix\n",
    "    cm = confusion_matrix(y_true, y_pred)\n",
    "    if normalize:\n",
    "        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "        print(\"Normalized confusion matrix\")\n",
    "    else:\n",
    "        print('Confusion matrix, without normalization')\n",
    "\n",
    "    print(cm)\n",
    "\n",
    "    fig, ax = plt.subplots()\n",
    "    im = ax.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    ax.figure.colorbar(im, ax=ax)\n",
    "    # We want to show all ticks...\n",
    "    ax.set(xticks=np.arange(cm.shape[1]),\n",
    "           yticks=np.arange(cm.shape[0]),\n",
    "           # ... and label them with the respective list entries\n",
    "           xticklabels=classes, yticklabels=classes,\n",
    "           title=title,\n",
    "           ylabel='True label',\n",
    "           xlabel='Predicted label')\n",
    "\n",
    "    # Rotate the tick labels and set their alignment.\n",
    "    plt.setp(ax.get_xticklabels(), rotation=45, ha=\"right\",\n",
    "             rotation_mode=\"anchor\")\n",
    "\n",
    "    # Loop over data dimensions and create text annotations.\n",
    "    fmt = '.2f' if normalize else 'd'\n",
    "    thresh = cm.max() / 2.\n",
    "    for i in range(cm.shape[0]):\n",
    "        for j in range(cm.shape[1]):\n",
    "            ax.text(j, i, format(cm[i, j], fmt),\n",
    "                    ha=\"center\", va=\"center\",\n",
    "                    color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "    fig.tight_layout()\n",
    "    return ax"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's look at the confusion matrix for the validation data. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion matrix, without normalization\n",
      "[[7547 5699]\n",
      " [4589 8661]]\n",
      "Normalized confusion matrix\n",
      "[[0.56975691 0.43024309]\n",
      " [0.34633962 0.65366038]]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot non-normalized confusion matrix\n",
    "plot_confusion_matrix(Y_val, y_pred_lr, classes=['0','1'],\n",
    "                      title='Confusion matrix, without normalization')\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "plot_confusion_matrix(Y_val, y_pred_lr, classes=['0','1'], normalize=True,\n",
    "                      title='Normalized confusion matrix')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now let's train the model on all training data and look at the contribution of each coefficient."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY0AAAF5CAYAAABjpHbeAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzt3XeYXGXZx/HvL40eCCF0ktBEiogQIBQpAiovVQUUAUEFxIYNFV99pdgQxY4iUkRAmjUgSBOQFiChhSoYiZQAEkIXkpD7/eN5hszuzu6e7JyzM5v9fa5rr50558x9np2dmXvOUxURmJmZFTGk1QUwM7OBw0nDzMwKc9IwM7PCnDTMzKwwJw0zMyvMScPMzApz0rAOJK0n6Q5JL0o6UtISki6W9LykiyQdIOmKAnH+V9Jp/VHmvur8txZ8TEhap4Rzj8+xhpUQ615JO/ThcYX+l4uCgfB6HCjkcRoDk6QPAp8H3gy8CNwJfCsibmgy7unACxHxuXz/IODTwNYRMa+5UvepPOOBfwHDyz5/57+1wf5rgXMi4rS6bQGsGxEPN3nu8VT0d7XD+WzR5SuNAUjS54EfAd8GVgLGAj8H9ioh/Djg3k73/7GIftB0/lutDyQNbed4VrKI8M8A+gGWBV4C9u3hmMVISeWJ/PMjYLG6/buTrkyeA24CNs7b/wa8Dryaz3EeMAeYm+9/FDgEuKEu1obAlcCzwFPA/+btx5K+pdeOm5jP9RxwF7BD3b5rgW8AN5Kumq4AVsj7/g1EPv9LwFbAOsB1wPPAM8AFPTwXe5ISw3P5POt387e+qdPjvtVp/8/y9gCOAB4CZgMnk6/Y8/6PAPfnfZcD47op1/gca1i+vyowKT+PDwOH1R27BHBWjnk/8CXgsbr9jwA759tbAFOAF/L/4wc9PI+F/pcNyv5r4BfApcDLwM6k19z383meAk4Blqh7zJeAmaTX46G5LOv0JR6wAnBJ/p8+C1wPDMn7vgw8TnodPQjs1M3rseHrou75PAq4m/QauwBYvNXv/Xb5aXkB/LOQ/zB4NzCv9mHTzTHHA5OBFYExpA/rb+R9mwJPA1sCQ4GD85tksbz/WuDQulid32xvfNAAy+QPgi8Ai+f7W3Z+HLAaMAv4H9LV7S75/pi6c/4TeBPpA/Ja4IS8bzx1H65523nAV3OsxYFtu3ke3pQ/hHYBhucProeBEY3+1gaP77I/l+USYDnSFd5/gHfnfXvn+OsDw4CvATd1E7vD30VKgj/Pf88mOW7tA++EvH8UsHr+MOsuadwMHJRvLw1M7OF5LPS/bFD2X5M+TLep+x/8iJT0ls+PvRj4Tt1r9klSUloSOJuuSWNh4n2HlESG55+3AwLWAx4FVq37m9du8Hrs7XXxCHArKZEvT0rUR7T6vd8uP66eGnhGA89Ez9VFBwDHR8TTEfEf4DjgoLzvMOCXEXFLRLweEWcBr5GuBBbW7sCTEXFSRLwaES9GxC0NjjsQuDQiLo2I+RFxJenb8P/UHXNmRPwjIv4LXEj64OzOXFLV0qr5vN2147wf+EtEXBkRc0nfXJcAtl64P7OLEyLiuYj4N3BNXVk/Rvpguz//f74NbCJpXE/BJK0BbAt8Of89dwKnseB/th/w7YiYHRGPAT/pIdxcYB1JK0TESxExueDfVPR/WfPniLgxIuaTXj+HAZ+LiGcj4kXS3/6BuvKfGRH3RsQrpNdjM/HmAquQruLmRsT1kT7tXyddoWwgaXhEPBIR/2xwriKvi59ExBMR8SwpYfX0ehxUnDQGnlnACr30ulkVmFF3f0beBunD9guSnqv9AGvU7V8Ya5CuEHozDti30zm3Jb3xa56su/0K6Vtyd75E+mZ5a+459JFujuvwPOQPpEdJVz7N6K6s44Af1/2Nz+Zy9na+VYHah2PNjLrHrZrLXVN/u7OPkr5JPyDpNkm793LumqL/y0ZlGEO6gpha97f/NW+HYuVfmHjfI10ZXCFpuqSjASJ1Tvgs6ariaUnnS2r0ui7yuliY1+Og4qQx8NxMqmffu4djniB9gNWMzdsgvTm+FRHL1f0sGRHn9aEsjwJrFzzu7E7nXCoiTijw2C7d+yLiyYg4LCJWJX27/3k33WA7PA+SRPpwfLzAeRueuxePAh/r9HcuERE39fK4J4DlJS1Tt21sXTlnkqqlatbotsARD0XE/qSqye8Cv5O0VIG/pej/8o1T1d1+BvgvsGHd371sRNQ+aIuUv3C8fBX0hYhYC9gD+LyknfK+30bEtqT/e5Ceg86afV0Mak4aA0xEPA98HThZ0t6SlpQ0XNKukk7Mh50HfE3SGEkr5OPPyft+BRwhaUslS0nardMHVlGXACtL+qykxSQtI2nLBsedA+wh6V2ShkpaXNIOklZvcGxn/wHmA2vVNkjat+6xs0kfDq83eOyFwG6SdpI0nFRf/xqpjaeIp+rPW8ApwFckbZjLuaykfXt7UEQ8msv0nfzcbEy6Yji37u/4iqRRklYDPtVdLEkHShqTvz0/lze/ToPnsZOi/8tG5Z9Pel39UNKKuRyrSXpXXfk/LGl9SUuSXo99jidpd0nr5A/7F/Lf93oed/MOSYuRvlj9l2peF4Oak8YAFBE/II3R+Brpw+BR0gfJn/Ih3yS1GdwNTANuz9uIiCmk+uKfkT5wHyY1iPalHC+SGhP3IF3OPwTs2OC4R0ndgf+3rrxfpMDrL9eBfwu4MVdVTAQ2B26R9BKpsfQzEfGvBo99kNSe8lPSt9c9gD0iYk7BP/HHwD6SZkvqqR2hdr4/kr7Zni/pBeAeYNeC59qf1HD7BPBH4Jjc9gOpY8NjpHEWVwG/I33INfJu4N783PwY+EBuo2j0PNaXvdD/sgdfJr2WJue//SpSwzQRcRmpHeaafMzN+THd/Q09xgPWzfdfyrF+HhHXktozTiD9r58kXW39b+fAJbwuBjUP7jMbYCR9nJQMtm91WfpC0vqkhLpYLx06rA35SsOszUlaRdI2koZIWo9UnfLHVpdrYUh6j6QRkkaRrsYudsIYmJw0zNrfCOCXpAFrfwP+TBrTMZB8jFQ1+U9SO8PHW1sc66uWVk9Jejep3nUocFp3vWkk7QNcBGye6+TNzKwFWnalkeeXOZnUULgBsL+kDRoctwxwJNDTQCMzM+sHraye2gJ4OCKm514L59N4wr1vACeSutCZmVkLNT2XfxNWo+Mo0MdI8yG9QdLbgDUi4hJJRxUJusIKK8T48eNLK6SZ2WAwderUZyJiTG/HtTJpqMG2NxpYJA0BfkiBMQSSDgcOBxg7dixTprjZw8xsYUia0ftRra2eeoyO0wmszoKpLiDNbLkRcK2kR0gT6k2SNKFzoIg4NSImRMSEMWN6TZRmZtZHrUwatwHrSlpT0gjSDJaTajsj4vmIWCEixkfEeNJU33u695SZWeu0LGnkgT2fIi1Ucz9wYUTcK+l4SXu2qlxmZta9VrZpEBGXklbrqt/WcDKziNihP8pkZmbd84hwMzMrzEnDzMwKc9IwG4SmzpjNydc8zNQZs1tdFBtgWtqmYWb9b+qM2Rxw2mTmzJvPiGFDOPfQiWw2blSri2UDhK80zAaZydNnMWfefOYHzJ03n8nTZ7W6SDaAOGmYDTIT1xrNiGFDGCoYPmwIE9ca3eoiDRqLQrWgq6fMBpnNxo3i3EMnMnn6LCauNdpVU/1kUakWdNIwG4Q2GzdqQH5gDWSNqgUH4v/A1VNmZv2g6mrB/qr68pWGmVk/qLJasD+rvpw0zMz6SVXVgv1Z9eXqKTOzAa4/e8T5SsOsj6bOmO0eSNYW+rNHnJOGWR8sKt0nbdHRXz3iXD1l1gceVW2DlZOGWR94VLUNVq6eMusDj6q2wcpJw6yPPKraBiNXT5mZWWFOGmZtalGYEdUWPa6eMmtD7tJr7cpXGmZtyF16rV21NGlIerekByU9LOnoBvs/L+k+SXdLulrSuCrL4+oAaxfu0mvtqmXVU5KGAicDuwCPAbdJmhQR99UddgcwISJekfRx4ETg/VWUx9UB3fN0Gf3PXXqtXbWyTWML4OGImA4g6XxgL+CNpBER19QdPxk4sKrCLCoLpJTNybR13KXX2lErq6dWAx6tu/9Y3tadjwKXVVUYVwc05rp1M6vXyisNNdgWDQ+UDgQmANt3s/9w4HCAsWPH9qkwrg5orJZM586b72RqZi1NGo8Ba9TdXx14ovNBknYGvgpsHxGvNQoUEacCpwJMmDChYeIpwtUBXTmZmlm9ViaN24B1Ja0JPA58APhg/QGS3gb8Enh3RDzd/0U0cDI1swVa1qYREfOATwGXA/cDF0bEvZKOl7RnPux7wNLARZLulDSpRcU1MzNaPCI8Ii4FLu207et1t3fu90KZmVm3PCLczMwKc9IwM7PCnDTMzKwwJw0zMyvMScPMzApz0jAzs8KcNMzMrDAnDTMzK8xJw8zMCnPSMDOzwpw0zMysMCcNMzMrzEnDzMwKc9IwM7PCnDTMzKwwJw0zMyvMScPMzApz0jAzs8KcNMzMrDAnDTMzK8xJw8zMCnPSMDOzwlqaNCS9W9KDkh6WdHSD/YtJuiDvv0XS+P4vpZkNFlNnzObkax5m6ozZrS5K2xrWqhNLGgqcDOwCPAbcJmlSRNxXd9hHgdkRsY6kDwDfBd7f/6Vt3tQZs5k8fRYT1xrNZuNGtbo4ZtbJ1BmzOeC0ycyZN58Rw4Zw7qET/V5toJVXGlsAD0fE9IiYA5wP7NXpmL2As/Lt3wE7SVI/lrEUtRfjSVc8yAGnTfa3GFukDdRv65Onz2LOvPnMD5g7bz6Tp89qdZHaUsuuNIDVgEfr7j8GbNndMRExT9LzwGjgmcpKdeZuXbdtuDdscRjMeQXO3bfr/k0+CG87AF6eBRd+qMvuZ5bcjTnzxrFSzOKH+jkr/2FJWG6JBQds/SlYb1d45iG4+LNd4293FKy9I8y8G/76lS67H9jwc1z98nh2WuoR3nzvD7s+/t3fgVU2hn9eA3//ftf9e/wIVlgXHrwMbvpZ1/3v/SUsuzrc83u47Yyu+/f7DSw1Gu44F+78bdf9B1wEI5aEW38F9/6p6/4P/yX9vvEn8I/LO+4bvjgc+Pt0+7oTYfp1HfcvOQref066fdWx8OhtHfePXBXe96t0+7Kj4clpHfePXhv2/Em6PelImPXPjvtXfgvsekK6/fvD4IUnOu5fY3PY+dh0+4ID4ZVOH5RrbQ/bfyndPud9MPfVjvvf9C7Y5sh0u4LXHpt/BDZ6Hzz/GPzhY133N/naY6evw9gt4d+3wNXH8+Jrc3l95gtsFvD6tXDfXt9ng023HRCvvQ89eAmbj3iB+QFDBOs9vCLsOCntHyivvdrfU6FWJo1GVwzRh2OQdDhwOMDYsWObL1nJ1l1xaUYMG8LQeenFOHKJ8p72F1+by3EX38Mt817jhmEPcepqc1lmseGlxTdbGC/8dx7z8zt0fsC0x59ng01bW6ailllsOOuvMpIX/juPkUsMY5kl/D5qRBFdPoP758TSVsCxEfGufP8rABHxnbpjLs/H3CxpGPAkMCZ6KPSECRNiypQp1Ra+D6pq0zj5moc56YoHmR8wVPD5d67HJ3dcp7T4ZgujVhU7d958hrtdYECRNDUiJvR2XCuvNG4D1pW0JvA48AHgg52OmQQcDNwM7AP8raeE0c42GzeqkjfPxLVGM2LYkDfepBPXGl36OcyK2mzcKM49dKI7fSzCWpY0chvFp4DLgaHAGRFxr6TjgSkRMQk4HThb0sPAs6TEYnX8JrV2U9UXJGsPLaueqkq7Vk+ZmbWzotVThbrcStpW0ofz7TG5SsnMzAaZXpOGpGOALwO1/nbDgXOqLJSZmbWnIlca7wH2BF4GiIgngGWqLJSZmbWnIkljTu6xFACSlqq2SGZm1q6KJI0LJf0SWE7SYcBVwK+qLZaZmbWjXrvcRsT3Je0CvACsB3w9Iq6svGRmZtZ2Co3TyEnCicLMbJDrNWlIepEF8z2NIPWeejkiRlZZMDMzaz9Fqqc69JSStDdpWnMzMxtkFno9jYj4E/COCspiZmZtrkj11Hvr7g4BJtBgenIzM1v0FWkI36Pu9jzgEbqusGdmZoNAkTaND/dHQczMrP11mzQk/ZQeqqEi4shKSmRmZm2rpysNzy9uZmYddJs0IuKs/iyImZm1vyK9p8aQpkbfAFi8tj0i3O3WmlbV2ulmVo0ivafOBS4AdgOOIK3Z/Z8qC2WDw9QZszngtMnMmTefEcOGcO6hE504zNpckcF9oyPidGBuRFwXER8BJlZcLhsEJk+fxZx585kfMHfefCZPn9XqIplZL4pcaczNv2dK2g14Ali9uiLZYDFxrdGMGDaEufPmM3zYECauNbrVRbIBwFWarVUkaXxT0rLAF4CfAiOBz1VaKhsUNhs3inMPnegPACvMVZqtVyRp3BIRzwPPAztWXB4bZDYbN8pveiusUZWmXz/9q0ibxk2SrpD0UUn+75hZy9SqNIcKV2m2SJFpRNaVtAXwAeCrku4Dzo+Ic/p6UknLk3pkjSfNZbVfRMzudMwmwC9I1WGvA9+KiAv6ek4zG/hcpdl6iig+Ya2kFYAfAAdExNA+n1Q6EXg2Ik6QdDQwKiK+3OmYNwEREQ9JWhWYCqwfEc/1FHvChAkxZYoHs5uZLQxJUyNiQm/H9Vo9JWmkpIMlXQbcBMyk+UWY9gJqI87PAvbufEBE/CMiHsq3nwCeBsY0eV4zM2tCkYbwu4A/AcdHxM0lnXeliJgJEBEzJa3Y08G5emwE8M+Szm9mZn1QJGmsFQtTh5VJugpYucGury5knFWAs4GDI2J+N8ccDhwOMHbs2IUsqZmZFVWkIbxPq/RFxM7d7ZP0lKRV8lXGKqSqp0bHjQT+AnwtIib3cK5TgVMhtWn0pbxmZta7hV4jvCSTSHNYkX//ufMBkkYAfwR+ExEX9WPZzMysG0Uawrcpsm0hnQDsIukhYJd8H0kTJJ2Wj9kP2A44RNKd+WeTJs9rZmZN6LXLraTbI2LT3ra1C3e5NTNbeEW73Pa03OtWwNbAGEmfr9s1EujzGA0zMxu4emoIHwEsnY9Zpm77C8A+VRbKzMzaU0/LvV4HXCfp1xExox/LZGZmbarIOI3FJJ1KmifqjeO93KuZ2eBTJGlcBJwCnEaaONDMzAapIkljXkT8ovKSmJlZ2ysyuO9iSZ+QtIqk5Ws/lZfMzMzaTpErjdrI7S/WbQtgrfKLY2Zm7azI3FNr9kdBzMys/RWZRmRJSV/LPaiQtK6k3asvmpmZtZsibRpnAnNIo8MBHgO+WVmJzMysbRVJGmtHxInAXICI+C+gSktlZmZtqUjSmCNpCVLjN5LWBl6rtFRmZtaWivSeOgb4K7CGpHOBbYBDqiyUmZm1pyK9p66UdDswkVQt9ZmIeKbykpmZWdvptnpK0pvz702BccBM4AlgbN5mZmaDTE9XGp8HDgdOarAvAE9YaGY2yPQ0Nfrh+feO/VccMzNrZ0UG931S0nJ190dJ+kS1xTIzs3ZUpMvtYRHxXO1ORMwGDquuSGZm1q6KJI0hkt4YzCdpKGkpWDMzG2SKjNO4HLhQ0imkBvAjSOM2zMxskClypfFl4G/Ax4FPAlcDX2rmpHlNjislPZR/j+rh2JGSHpf0s2bOaWZmzes1aUTE/Ij4RUTsExHvi4hfRkSzy74eDVwdEeuSktDRPRz7DeC6Js9nZmYl6Glw34X59zRJd3f+afK8ewFn5dtnAXt3U4bNgJWAK5o8n5mZlaCnNo3P5t9VrJ2xUkTMBIiImZJW7HyApCGkgYUHATtVUAYzM1tIPSWNS4BNgW9GxEELG1jSVcDKDXZ9tWCITwCXRsSjdZ23ujvX4aTR64wdO3ZhimlmZguhp6QxQtLBwNaS3tt5Z0T8oafAEbFzd/skPSVplXyVsQrwdIPDtgLengcSLp3L81JEdGn/iIhTgVMBJkyYED2Vy8zM+q6npHEEcACwHLBHp30B9Jg0ejEJOBg4If/+c+cDIuKA2m1JhwATGiUMMzPrPz0ljVUi4uOS7sjf5Mt0Amnsx0eBfwP7AkiaABwREYeWfD4zMyuBIhrX5ki6PSI2rf3u53L12YQJE2LKlCmtLoaZ2YAiaWpETOjtuJ6uNGZJugZYU9KkzjsjYs9mCmhmZgNPT0ljN1LvqbNpvKaGmZkNMj2tpzEHmCxp64j4j6SlIuLlfiybmZm1mSJzT60j6T7gfgBJb5X082qLZWZm7ahI0vgR8C5gFkBE3AVsV2WhzMysPRVJGkTEo502NTthoZmZDUBF1tN4VNLWQEgaARxJrqoyM7PBpciVxhGkdTRWAx4HNsn3zcxskOn1SiMiniFNJ2JmZoNcr1caklaX9EdJT+eJBn8vafX+KJyZmbWXItVTZ5ImGFyVVEV1cd5mZmaDTJGkMSYizoyIefnn18CYistlZmZtqEjSeEbSgZKG5p8DyWM2zMxscCmSND4C7Ac8CcwE9snbzMxskCnSe+rfgGe0NTOzQr2nzpK0XN39UZLOqLZYZmbWjopUT20cEc/V7kTEbOBt1RXJzMzaVZGkMUTSqNodSctTbPoRMzNbxBT58D8JuEnS74AgNYp/q9JSmZlZWyrSEP4bSVOAdwAC3hsR91VeMjMzazuFqplyknCiMDMb5Aqtp2FmZgYtShqSlpd0paSH8u9R3Rw3VtIVku6XdJ+k8f1bUjMzq9eqK42jgasjYl3g6ny/kd8A34uI9YEtgKf7qXxmZtZAq5LGXsBZ+fZZwN6dD5C0ATAsIq4EiIiXIuKV/iuimZl11qqksVJEzATIv1dscMybgOck/UHSHZK+J2lov5bSzMw6qGyQnqSrgJUb7PpqwRDDgLeTRp//G7gAOAQ4vcG5DgcOBxg7dmwfSmtmZkVUljQiYufu9uUVAFeJiJmSVqFxW8VjwB0RMT0/5k/ARBokjYg4FTgVYMKECVFG+c3MrKtWVU9NAg7Otw8G/tzgmNuAUZJqCz69A48VMTNrqVYljROAXSQ9BOyS7yNpgqTTACLideAo4GpJ00ij0X/VovKamRktmngwImYBOzXYPgU4tO7+lcDG/Vg0MzPrgUeEm5lZYU4aZmZWmJOGmZkV5qRhZmaFOWmYmVlhThpmZlaYk4aZmRXmpGFmZoU5aZiZWWFOGmZmVpiThpmZFeakYWZmhTlpmJlZYU4aZmZWmJOGmZkV5qRhZmaFOWmYmVlhThpmZlaYk4aZmRXmpGFmZoU5aZiZWWFOGmZmVlhLkoak5SVdKemh/HtUN8edKOleSfdL+okk9XdZzcxsgVZdaRwNXB0R6wJX5/sdSNoa2AbYGNgI2BzYvj8LaWZmHbUqaewFnJVvnwXs3eCYABYHRgCLAcOBp/qldGZm1lCrksZKETETIP9esfMBEXEzcA0wM/9cHhH392spzcysg2FVBZZ0FbByg11fLfj4dYD1gdXzpislbRcRf29w7OHA4QBjx47tW4HNzKxXlSWNiNi5u32SnpK0SkTMlLQK8HSDw94DTI6Il/JjLgMmAl2SRkScCpwKMGHChCij/GZm1lWrqqcmAQfn2wcDf25wzL+B7SUNkzSc1Aju6ikzsxZqVdI4AdhF0kPALvk+kiZIOi0f8zvgn8A04C7groi4uBWFNTOzpLLqqZ5ExCxgpwbbpwCH5tuvAx/r56KZmVkPPCLczMwKc9IwM7PCnDTMzKwwJw0zMyvMScPMzApz0jAzs8KcNMzMrDAnDTMzK8xJw8zMCnPSMDOzwpw0zMysMCcNMzMrzEnDzMwKc9KwRdrUGbM5+ZqHmTpjdquLYrZIaMnU6Gb9YeqM2Rxw2mTmzJvPiGFDOPfQiWw2blSri2U2oPlKwxZZk6fPYs68+cwPmDtvPpOnz2p1kcwGPCcNW2RNXGs0I4YNYahg+LAhTFxrdKuLZDbguXrKFlmbjRvFuYdOZPL0WUxca7SrpsxK4KRhi7TNxo1ysjArkaunzMysMCcNMzMrzEnDzMwKc9IwM7PCnDTMzKwwJw0zMytMEdHqMpRK0n+AGU2EWAF4pqTiLCqxq47vsrcm/kCNXXX8wVr2cRExpreDFrmk0SxJUyJigmP3X3yXvTXxB2rsquO77D1z9ZSZmRXmpGFmZoU5aXR1qmP3e3yXvTXxB2rsquO77D1wm4aZmRXmKw0zMyvMScPMzApz0rBFjqTPFNnWx9jbFNnWrqp8bgYiSVfn399tdVkGCrdpZJL2BLbLd6+LiItLirtpT/sj4vYmYk8DGv0DlULHxn2NneMPBXYDxlO39kpE/KCZuDn2msCnG8Tes4TYt0fEpp223RERb6sodpdtTcQ/OyIO6m1bE/FLf24k/ZTGr0MAIuLIdoyd498HfBw4Bfgg6b1TH7+Z9+eXIuLE7v6GZsuezzEUuDwidm42VlFehAmQ9B1gC+DcvOlISVtHxFdKCP9zYFPgbtIL8i3ArcBc0gvpHU3Eviz/Pjv/PgB4BTiriZj1LgZeBaYB80uKWfMn4PR8jlJiS9qf9MZfU9Kkul3LAE0tEC5pK2BrYIykz9ftGgkMbSZ2Jxt2Ou9QYLNmg1b53ABT8u9tgA2AC/L9fYGpbRwb4OvA0cDqwEl0TBrNvj/vy7+n9HhUEyLidUmvSFo2Ip6v6jz1nDSS3YBNImI+gKSzgDuAMpLGI8BhETEtx94IOCoiDikh9jYRUV81crSkGyPi+BJiA6ze7NVKD16NiJ+UHPMmYCZpKoWT6ra/SErazRgBLE16zyxTt/0FYJ8mYyPpK8D/AktIeqG2GZhDOd0oK3tuIuIsAEmHADtGxNx8/xTginaNnc2MiF0lfb3E903N+4FLgOUi4sclx673KjBN0pXAy7WNZVzJNOLqKUDS3cAOEfFsvr88cG0ZH5iS7oyITXrb1tfYwKci4oZ8f2vg52XEzvG+C1wdEWW8OTvH/iCwLumN/1ptezPVAf1B0riIaGZus97if6ekK9x+J+lBYKu699EoYHJErNeusSVNjYjNyqxirIt9H7ArMAnYga5VX8+WdJ6DG22vJdyy+Uoj+Q5wh6RrSP/Y7SjnKgPgfkmnAeeQLncPBO4vKfZHgTMkLZtjPw8trghwAAAgAElEQVR8pKTYAJOBP0oaQqpOq7WVjCwh9luAg0iX/7XqqWarAwCQ9F7gu8CKpDKXWe7FJJ1K17aYpsud43xF0mrAuE7x/15G/IqfmxNY8D4C2B44toS4VcaeK+lMYDVJXa58m/y2fgrwV2AtUlVa56qvtZqIvSBQRcmhO77SyCStAmxO+sfeEhFPlhR3cVJDW62R/e/ALyLi1TLi53OMJP0vS63TlDQd2BuYFiW/UCQ9AGwcEXPKjJtjPwzsERFlJef62HeRPgymAq/XtkdEGfXrSDoB+ACpPrwWP8roIJDjV/bc5PgrA1vmu6W9j6qKLWkFYGdSIv165/1lfCBL+kVEfLzZOD3EX5f0xXcDYPHa9ogoJSl1Od9gThqS3hwRD3TXw6msqhJJSwBjI+LBMuLVxV0J+Dawaq6X3YB0CX96SfEvB3attfWUSdIFwKcj4ukKYt/Yqa2nzNhTI6Lphuke4j9ISqav9Xpw3+JX9tx0c743R8QDTTx+44hotj2qyHneGhF3lRxzZES8kKu7uyixeuoG4Bjgh8AewIdJn+3HlBG/y/kGedI4NSIOr7vkrRdlVDnkrrzfA0ZExJqSNgGOL6lr6WXAmcBXI+KtkoYBd0TEW5qNneP/mnQJfRkd2x3K6HJ7LbAxcFun2GU8Lz8GVib10KqP/YcSYh8LPA38sVPssj4ALgP2jYiXyojXIH5lz0035/t3RIxt4vGvA/8CzgPOi4j7ennIwsavrFuspEsiYndJ/8qxO1RPlXUlUNcuM6323pd0fUS8vYz4nQ3qNo2IODzf3LVzdVGuVirDMaTuvNfmc94paXxJsVeIiAtzzxsiYl5+k5XlX/lnRP4pUyXfgrKRpK7H76zbFkAZH4y1RscvdopdVlXAK8CdSoPO6j/Uy+oJU/pz06gtoLYLWK6vcbO7SW1f+wOTJL1MSiDnR8QjTcaGBe2LpXeLjYjd880bSNXS1zdz1dWDV3O740OSPgU8TmqzqsSgvtKo6WbAUym9KSTdEhFb1g+gknR3ST2zrgXeB1wZEZtKmgh8NyK2bzZ2lVoxIGmg6O+eMGWQ9CLwBeqSXJ2TImKFJmJ3eB9K2oLU5rMv8GhEbN3X2P1F0juAbYG3k75c3EFKIKV0w5W0OSn5LQd8A1gWODEiJpcRv7NBfaWRG9ZWI/WNfxsLLh9HAkuWdJp7cvfSobnB6khSn/kyfJ7UnW9tSTcCYyhhzEBNrrZrdMneVLVdVDwgKfeGaVTupnuWSfpQo+0R8ZtmY+c4lSaHip6b24B7IqLL6zpX5zWjczfVW4FbJX2BBZ1L+h5cupieR5w3XV0aEX+TdB2po82OwBHARkApSSMibss3XyK1Z1RqUCcN4F3AIaTRoPX19C+SBlqV4dPAV0nfwn4LXA58s9mg+XJ0cVLXw/VIb64HIw9+KslRdbcXJ13VzCspdpUDki6pu7048B7giRLiQnrj18feCbgdKCVp1NV/d1BiT5gqnpt9SP/PLiJizSZjf6+buAFc12RsgO/n3+8ltfWck+/vTxqY27Rc1bgUcDNwPbB5GR1A+iPhNTyvq6dA0vsi4vcVxB0KnBARX+z14L7Fvzkitqoidg/nvK6M6q/+rIbJCfaqssZSdIq9LHB2iV1iR9fdXZxUDbN8RHTpDlrS+Up7biR9pnOVS6Nt7RY7x/p7RGzX27Y+xv4haSqY14AbSe0bN0fEf5uM2+P7MCLKSKpdz+ukkUjajTTvT30/56anFZD0tyo+rHLs40gNhX8oexxFjl/fVXAI6YX/kyhhhG+OPwJ4U75b9lVS/XnWA/4SEetUEHs4cHdErF927Lpz3BAR21YUu7Tnppu2wSoniiwldo51P7BbREzP99cELi3z/yppaVL10VHAyhGxWFmx+9Ngr54CqM1jsySpvvE00uX2rSWFv0NpgriL6FgNU0ZPns+TLnvnSXqVckf3QhrAVusqOI/Uk+qjZQSWtANpYsVHcvw1JB0cJYx8zg2ztXIH8CTw5Wbj5tj1VQJDgfWBC8uInePXfzAOASbQca6rZuOX/tyo2okiq5xosd7ngGuVBrRCGvF/ePeHF5d7NL2d9KVrBnAGqZqqFP1QpdmBk0aydURsnHs1HSfpJMrpngmwPOnFXX+10WwXx20i4kZgTOeuwmUqoT66JycB74w84FHSm0hdKZseOBcRpX3INvD9utvzgBkR8ViJ8esnE5xHSqr7lRW8ouemyokiq4z9hoj4a+6o8ua86YH6AZaSdomIK/sYfglSm+nUiCirTbDehLrbb1RpVnAewNVTAEi6NSK2kDSZ1CA2i9QbZN0mYn43Ir4sad+IuKi0wtJhME/pk6w1ONfWdJ1nqelG30bdjsvqipxj1a+Pcm1EXNLT8QsZeyUWNIjfWkajZn+q8rlZVPXHe61MVVZp+kojuVjScqSeGreTrgR+1WTM/5H0NdLEh6UmDRZMsrZ6o4FVZQ0Ek3Q2sDZwJ3XzIFFOT6Epkk6n41ogZc7ftDkL1kf5TL46a3oSSkn7kV4n15KqeH4q6YsR8btmY+f4y5IGPr6xIBhpBoFSuiZX/NxUNhlilbGLFqGfzrPQqq7S7HK+wX6lkXuPTKz1MZe0GLB4s29SSd8j1YkuRRqB+8Yumnyxqx8mWcvnuR/YoKJG9sWAT5IGPYnUo+TnUcKcS0pT3devjzKUNL1KGQMq7wJ2qV1dSBpD6n301mZj53i/B+5hwUJaBwFvjYj3lhS/yuemyokiK51oscD52/ZKQx2nQapVaX4/Sp7rrmbQX2lExPzchrFVvv8ajUe2LmzcLwJflPTniNir2XidYj8DnC/p/uhhkjVJX4mI7zRxqntIfddnNhGjA0lXR8ROpG/PX6bj+JgyLQfU5oNatsS4QzpVR80ifbsry9oR8b66+8cprZtSpqqem6cq/FCvMvaAFhE79uf5Bn3SyK6Q9D4q6LraW8JoZqxFTwkj25c0ZXJfrQDcJ+lWyptUcJXcv3xPSefTdcRvGTMLV7k+yl+VZv89L99/PwuW3S3DfyVtGwsW1toGaKo/fydVPjdTlGYvrmIyxCpjF/FIP52nMHVcdriLKGFi0YbnHezVU/BGN8SlSJd2VXRd7encpfU1Lzt2d4OHmhk0JGkfUrfdbek6SVyUNaZFFa2PkmO/l7pqtYj4Y4mx30pqM6pdAcwGDinwBWFhzlHV2jFnNtgcUc70LZXFrjtHJZ0+qiKpx0k/I+K4Ss7rpNE7SRtGxL0Vxa6srrTqethmrpIk/V9EfKPk8rwLWKZzo7SkA4Cnm+gyiaR1gJVyV+f67dsBj0fEP/sau5vzjQSIiBd6O7ZgvMqem0VBd50+yupUsihx9VQxZwNt2QjWi6p7fDQzffwOpBk531DX3tFXx5EWoensatL6F818MP6IxvORvZL3NTpvYbmq4fnIC2jVkoWkTwNDI+JHzcSn2ucGeONqoKqJIiuLnU2gok4fVWnUc7JeVQnPSaOYKj98q4xddlffzhb6Daa0TslSwAqSRkGHmYVXbbI8S0bEfzpvjIgnJS3VZOzx0WAFuYiYonLWR/kIjb+YnEqaRbbZpFHlc1NT5USRVcaGCjp99INSuqgvLCeNYqr89nHQwj5A3awyVlP7hhER326iXFX5GPBZUoKYyoKk8QJwcpOxF5c0rPOoW6X5oZZoNnYP+5qNDakqpMt66RHxmqQyvlhU+dwAEJ0m/ZR0HnBVO8fWgmlhlqH8Th+V6ty1XtJSEfFyd8eXxUmjYr0NSoqIe/oQttaAvA1pMfkL8v196d9vHwv9YRZpVtIfS/p0RPy05PL8AfiVpE/V3jz5W/RPaH5amNskHRYRHQZ9Svoo5Q1KXCkinuq8rYzYVPvcdGddoM9LvfZT7O/3fkh7k7QVcDqwNDA2d6b4WER8opLzDaAqvJaRNDkiJvbxsVUOeLqGNH/T3Hx/OHBFf/XblrRRH5PeG48nJb36mYX73FtFaY30bwKHkiaGg/TBcjrwf9HELLr5w/uPwBwWJIkJpGVw39NsDySlxZ2OJK2AV+t2vBlwInByswM2Gzw3AtaghOem7hyNJkP8SuerhHaLneOvCcyMPJebpCVIHR8eKSN+lSTdQppkdVIsWB30nojYqJLzOWm8MQr3DOCy2kjZEmPfGBHblBmzLvaDwFYR8Wy+PwqYHOVNXV7ltBDHkBrDNwAuBXYFboiIplcelLQkqScMwMMR8V9Ji0U5o813JK26BnBvRPyt2Zh1sXcFjs7xA7iXtB5LaeNA8odhbRr0h6PJNR0WFZKmkCYunZPvjwBujIjNe35k66nxktJ3RUmzFHTm6qnkF6R57n8i6SLg11HeAvBVDko6gQUDtSCt4ndsCXFrTqS6qRv2Ad5KmsLiw/mb/Gklxb6hQVfjmymnB9yOpGmtbyq7/jgiLlOaPLPDlN+S1oyIfzUTO38B6GzdWnNJWYPkVO1EkVVOtDisvk0pIubkxDEQPJrHmEQu85GkNcMr4aQBRMRVwFVKk8XtD1wp6VHSpIXnNHnpPpLULfOd9aekhHrkiDhT0mXAlnnT0WUOYqPaqRv+G2kKl3l5TMLTQFPz/6t/1nx/hPQa+UmuMrmeNMDvzyXFnyRp17outxuQ1utotqqhpy7BpbweVe1kiJXFzv4jac+ImJTPtxfwTEmxq3YEab3x1YDHgCuAStozwNVTb1BaZvNAUm+mJ0gvzm2Bt0TEDi0sWhfqOKtlF1HOVBxI+jGpG2LpV0mSfk4a9/ABUj3+S8CdEfHhJmIeTFrzfQKpm2p9z6yzyvo2nc+1Mmmdi6OAUVHSOhVKK0h+CdiNtPb7b4ADIqLs+adKp2onQ6wsdo63Nuk9v1re9ChwUNmDNqsg6SzgsxExO98fBZxU4hiWjudz0gBJfyAtvnI2qWpqZt2+KRExodsHdx/zSxFxYnfdY5sZeKOOs1o2CF3aVByVT92QzzMeGNloHEQfYg0B9o+Ic3s9uG/xTyO1wzxFusq4Abi9c1fWJs+xNylxLAO8NyIeKiFm5fMU5Q/2Hera2JYnVSOVlTQqid3pPEuTPhdfLDNuldRguqBG28oy6Kun8ofMndHN1NN9SRhZrVqn8/xKTeuv3lHNfOvvTk9XSZI2bfYqKVd5fYwF1RhlG01a5vU50kyxz5SRMBp8uRgJTAc+LamM0b1VrmZYU+VkiFXGRp3WMZFU6jomFRsiaVTdlcbyVPjZ7isNaGoOpVbKvYQ+D4yNiMOVlqtcr6wGQkmrAz8ljQcJ0rfqz0QTy5t2ukqqf/HVemY1fZUk6f9IM8NeQMd12Z/t9kELf471gXeR1pYeGhGrNxnv4J72N9vltr+o2okiq4xd6TomVcrdtb8C/I70ntoP+FZEnN3jA/t6PicNkHQcab3h0qdGV1qk58t0HY9QxofjBaQxAx+KiI1yd8qbI2KTZmPn+FcCv2XB6noHkurXdykh9hKkxrptSS/064FfRAlrnktq1NMoIqKphvYce3fg7aRvpKNIvbKuj4gzmo2d4y8FvBoRr+f7Q4HFIuKVnh9ZOP7ipFmGN6Tj67HPVY6qdqLIfploUdKdnd83jba1q9xh4h2khHp1RNxX2ckiYtD/kBapn08auPVCvv9CSbGvIL1J7yd1iT0D+G5Jsafk33fUbburxOflziLb+hj7QlIX2x3zz6nAha1+LRQo98mkNTRWrSj+ZGDpuvtLk7r3lhX/ItJEkf8EDs6vzx+XUOYxDbavTPoS05axO8W7Gdi27v42ZcZflH4GfZsGQJTU86UboyPidEmfibQOxXW5vrQMc/I39oA3eoA0PYCtzjOSDmTBgkP7k1aqK8N60XHw0TVKS6mWQiWPNq+L8UlJ43LsJ/LzPyzKazhdPCJeqjvfS7kasizrRMS+kvaKiLMk/Ra4vMmYVU6G2B8TLQJ8HDgrt22I1F7VY5XhYFXmMpUDlqSri2zro9oYj5mSdsvjB5qq/65zDPBXYA1J55Kmuf5SSbEhzby6H2nKhpmkAXll9Zy6Q9IbU7NI2hK4sYfjC1Mabf7T/LMjaZBiKRPPSTqMVHf8y7xpdVKX5LK8XN9ZQNJmlLtyX+31+FxOrMuSFh5qxuJK05R0oJImiqww9hsi4s78JWZjUjf7t0UJvfkWSa2+1GnlD+lb6PLAXaT66eXzz3jg/pLOsTvpjbkRcA2pDWLPEv+G0aQ+/bsDK7T6OV2Ict9PqhJ8JP/MJ02bMQ24u8nY00hfiO7K91cCLi6p3HeS5puqrxKcVuLzsjmp6uj6/PMwsFmJ8Q/Nr/XtSL2zniZNbtdMzBOAM4Gl6rYtRZrXqqmq2CpjdzrPaNLkjbfn9+iPSbUEpcRflH4Ge/VUldN0AxALejI9T/rWW7bVSF1AhwHb5e6ZTQ1iq3KMSZ13lxCjO6WPNq/zWqQpJoA3JgIsrfNERNwm6c2kgX0CHogSJhOsc3Wkrpl/Jz8nSpP1NeNrpMkQZ0jqMlFkG8eudz7pOXlfvn8AqffdziWeY5Hg3lOAqpmmuxZ7LdK3lq1I36ZvBj4XEdNLiH0G6XL63hwbShh8J2mPiLi4u26g0ebdP6sYbV4X+0TSGI0PAZ8m9QC7LyK+2mTcd0TE39R4jqimvwjUnafLEsCSpkbEZiXErnKiyMpi5/hdnoO+Duxd1DlpZKpoUXlJk0lXLbXG5A8An46ILbt/VOHY90XEBs3GKXiuIaRePaWsWd1fyhxtnuMNIfWGeyfpSuBy4LRo8o0k6biIOKaqUfj56mVDUvvOF+t2jQS+GBEbNhM/n6NRQiplnfoqY+dY3ycNxL0wb9oH2DAijikj/qLESQNQhYvKK09b3Glbn9fn6BTndNIcM5X0yc49a44gPSdTSW0zP4iI71VxvrKowVrjjbYNJkoT8O1N6hAwqW7Xi8D5EXFTE7FrE0WeA3wQOkwUeUpEvLkdY3c6z4ukSS1rV+xDWTAwNKKE5QAWFYO9TaOmykXlr5F0NKnONEh9/P+Sh/oTzY1SPgu4WdKTpK62tVHVZc3Hs0FEvJAHUl1KGqQ4FWjLpJEHri1JBeuPS7owIvaTNI3G7TxNPeeqeG6oSLPw/lnSVhFxczOxGngXaaLI1YGT6Ng2+L9tHLvesqR2jDUj4nhJY4FVIuKWEs+xSHDSSKpcVP79+ffHOm3/COnDp5kG2jNI0x1MY8E3pDINz10b9wZ+FhFzJbXzpWmjjg1B+jb9syZjfyb/PhO4lTQLapl6GitU5nP+Hkn3krrx/pW0pslnI+KcvgaMNN7jbCqYKLLK2J2cTHoPvQM4nvSa+T2pN5vVcdJIVqCiReUjotmeKT35d+T5/ytyCqk77F3A3/OgtrZt04gF649/HfhRvkr6P9LiS019u44FMx8vQxqj8Szp6vF30Wld7z7GPw6oTXP9mYh4Lt8fRfqGXZZ3RsSXJL2HtPbCvqSu4H1OGlDtRJFVxq6zZURsKumOfM7ZGjiLMPUrJ43k2KoC57mDdqNrI3vTU1EDD+R2h4spf72LIaRFmFar2/Zvquk2XLZ9chXDtsAupA/dX7Bgsao+yx/ux0namHQVeZ2kxyKirK6ZG9cSRj7f7DwgtCzD8+//Ac6LiGdr3YdLcKWko6hmosgqYwPMze/V2uwKY6jm6n3Ac9IAIk3vUZWLgVeppgppCVKyqGJVwPmSPsWC3iTkNp/S1o2oUK0zw26kxtI/Szq25HM8TRopP4u0hnpZqp7m+mJJD5Cqpz6RPxybniQyq/Xw+mTdtmarYPsjNqSBfX8EVpT0LVLvqa+VFHuR4t5TvNFzovZEjCB9G3u5jB4Tku4usWG6X6kfphivgqRLgMdJA7Nq03DcGh3nuupr7I+TrjDGkKYTuaDM3mvqh2muc5XXCxHxeh7/MDLKXSZ4QMrdkndiwUyxla2zPZA5aTSgtHLaFhHRdO8MSd8lvQCvaL5kXWKXPs11p/iVTTFepfxB+G7S9B4PKa3D8JYy/gdKa1WfHxUuv6qKp7muakxSjl3JRJFVx7binDS6UeJYiveQGhmHkCaLq3WLLeMq5iLgAVL/9eNJXQbvj4jP9PhAG7QqHpN0DLAD6YP9UmBX4IaI2KedY9vCcdIAOk3dMIQ0bmP7KGE1P0nTSV1Wp5U9DkR5HeBaFVjuHnt5lLdGeKUrA1r/k3Q/FY1JymNY3kqazPGtklYijZbfo51j28JxQ3hS/8KbR+pmuldJsR8C7qlo4GDnaa6fpPlpruudSRrvsHW+/xhpER8njYGryjFJVU4UWWVsWwhOGkCUMJFdD2YC10q6jI7dYsvocntqbtT8GmlqiKUpd+bPtSPi/ZL2B4g0SVxp/TOtJSobkwRMkbQc8CvSl42XSAMhy1BlbFsIrp4CJK1OWrBnG1KPlRtIA6weKyF2wwnPaoO5moz9BRb0+qp9mD8HTC2joVbSTaTeJDfmgU9rk/r2b9FsbGsNSds32l52t3OVPFFkf8W23jlpAJKuBH4L1Lo1HggcEBG7tK5UvcsD+yaQxoJAGpdwG/Bm4KKIOLHJ+LuQrmI2IK0lvQ1wSERc20xcWzRVOVGkJ6FsH04agKQ7I2KT3rb1MfYY0hKsnbvFNt1YLely4H2R15SWtDSpf/97SFcbTU+bLmk0MJF0JTM5Ip5pNqb1P0k3RMS2ncYkQQm9+eomiryG1MOpfqLIyyJi/XaMbX3jNo3kGUkHsmDNi/1JI33LcC5pcNzupGnGDwb+U1LsscCcuvtzgXG57aHPi9NIuo9U7vMj4p/AX5orprVaRGybf/c0MWJfVTlRZJWxrQ+GtLoAbeIjpJG3T5IarvcBymocHx0RpwNzI+K6PPCu6fEf2W+ByZKOyW0nNwLnSVoKaGZA2P6kRvUrJN0i6bOSmppa3BZdEfHjPDHnt4BN8u0zSWuQNztRZGWxrW+cNJJvAAdHxJiIWJGURI4tKXatW+xMSbvlyedWLyNwRHwDOIzU+P08cEREHB8RL0fEAU3EvSsivhIRa5OmBB9HSk5/k3RYGWW3RdI+kWYWrk0U+WvSRJHtHtsWgts0WDBIrrdtfYy9O3A9sAaph9ZI4NiIuLjHB7YZSTsAPyQNDFusxcWxNlQ32PQ7pMGsvy3xfVRZbFs4vtJIhuTxDkDpM4vuS0rO90TEjqRvSe8pKXalJG0u6QeSZgDHAaeSlt40a+RxSb8kVfVeKmkxyvuMqTK2LQRfaVDtzKJVXsVURdK3STO5ziYtNHR+GWNWbNFW8USRlcW2heOkkVU1s6iku4AdouP6CNdFxFvKiF+F3Kh+XkT8o9VlMbP24qRRsf5YH6Eq+dvdF0gTFh7mCQvNzEmjH1S9PkJVJF1A6hv/oYjYSNISwM1lDHo0s4HJScO6JWlKREyob4ORdFeUsAKemQ1M7n1gPZmTry4CIE9Y2OeR5mY28HkaEevJMcBfgTUknUuesLClJTKzlnL1lDWU181YHXgFT1hoZpmThnVL0tSI2KzV5TCz9uE2DevJZEmbt7oQZtY+fKVh3cpTpL8JmAG8zIK1FzZuacHMrGWcNKxbksY12h4RM/q7LGbWHtx7yrpVSw6SVqRu1UEzG7zcpmHdkrSnpIeAfwHXAY8Al7W0UGbWUk4a1pNvkLrb/iOvmLYTaXVAMxuknDSsJ3MjYhZpvZEhEXEN4HmnzAYxt2lYT56TtDTwd+BcSU8D81pcJjNrIfeesm5JWgr4L+mK9ABgWeDcfPVhZoOQk4Z1S9JHgOsj4qFWl8XM2oOrp6wn44EDJY0HpgDXk5LInS0sk5m1kK80rFd5evTDgKOA1SJiaIuLZGYt4qRh3ZL0NdJ06EsDdwA3kK40Zra0YGbWMk4a1i1Jt5N6S/2FNLhvckS82tpSmVkrOWlYjyQtA2ybf/YDnoqIbVtbKjNrFTeEW7ckbQS8HdgemAA8SmoMN7NBylca1i1JtWqpG4DbImJui4tkZi3maUSsJ1dGxIkRcVMtYUj6TKsLZWat46RhPflQg22H9HchzKx9uE3DupC0P/BBYC1Jk+p2LQN4ChGzQcxJwxq5HZgJrACcVLf9ReDulpTIzNqCk4Y1cl5EbCrpnxFxXasLY2btw0nDGhkh6WBgK0nv7bwzIv7QgjKZWRtw0rBGjiBNhb4csEenfQE4aZgNUh6nYd2S9NGIOL3V5TCz9uGkYd2SNIJ01bFd3nQdcIoH+ZkNXk4a1i1JpwHDgbPypoOA1yPi0NaVysxayUnDuiXproh4a2/bzGzw8Ihw68nrktau3ZG0FvB6C8tjZi3m3lPWk6OAayRNz/fHAx9uXXHMrNWcNKwno4GNSMliL2Br4PlWFsjMWsvVU9aT/4uIF4CRwC7AKcAvWlskM2slJw3rSa39YjdSV9s/AyNaWB4zazEnDevJ45J+SVrm9VJJi+HXjNmg5i631i1JSwLvBqZFxEOSVgHeEhFXtLhoZtYiThpmZlaYqxrMzKwwJw0zMyvMScMGDUlHSrpf0rl9eOx4SR+solxmA4mThg0mnwD+JyIO6MNjx5PWTV8okob24VxmbctJwwYFSacAawGTJH1O0lKSzpB0m6Q7JO2Vjxsv6XpJt+efrXOIE4C3S7ozP/4QST+ri3+JpB3y7ZckHS/pFtLqh5tJuk7SVEmX515oncv3a0k/kXSTpOmS9snbl5Z0dS7LtE7lfEDSaZLukXSupJ0l3SjpIUlb5OMa/p1mfRYR/vHPoPgBHgFWyLe/DRyYby8H/ANYClgSWDxvXxeYkm/vAFxSF+sQ4Gd19y8Bdsi3A9gv3x4O3ASMyfffD5zRoGy/Bi4ifZHbAHg4bx8GjMy3VwAeBkS68pkHvCU/ZipwRt63F/Cnnv7OVv8v/DNwfzz3lA1W7wT2lHRUvr84MBZ4AviZpE1II+Lf1IfYrwO/z7fXI83fdaUkgKHAzG4e96eImA/cJ2mlvE3AtyVtB8wHVgNq+/4VEdMAJLy57P4AAAE5SURBVN0LXB0RIWkaKan09Hfe34e/y8xJwwYtAe+LiAc7bJSOBZ4C3kr6Bv9qN4+fR8fq3cXrbr8aEbUpWATcGxFbFSjTa53KB2mt9jHAZhExV9IjdeeqP35+3f35LHhvN/w7zfrKbRo2WF0OfFr567+kt+XtywIz8zf+g0hXBgAvAsvUPf4RYBNJQyStAWzRzXkeBMZI2iqfZ7ikDReinMsCT+eEsSMwbiEeC93/nWZ94qRhg9U3SO0Nd0u6J98H+DlwsKTJpKqpl/P2u4F5ku6S9DngRuBfwDTg+8DtjU4SEXOAfYDvSroLuJM0xXxR5wITJE0hXXU8sBCPhe7/TrM+8TQiZmZWmK80zMysMCcNMzMrzEnDzMwKc9IwM7PCnDTMzKwwJw0zMyvMScPMzApz0jAzs8L+H1r7Dm3mpuDQAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "logreg.fit(X_resampled, Y_resampled)\n",
    "# visualize the coefficients of the logistic regression\n",
    "plt.plot(X_resampled.columns,logreg.coef_[0][:], '.')\n",
    "plt.plot(X_resampled.columns,0 * logreg.coef_[0][:], '--')\n",
    "plt.title(\"Coefficients of the logistic regression\")\n",
    "plt.xticks(rotation=90)\n",
    "plt.xlabel(\"feature name\")\n",
    "plt.ylabel(\"coefficient value\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* Unsurprisingly, the coefficient corresponding to 'installCount' has the highest positive value. \n",
    "* Other contributions are also sensible, given how we simplified the features in the previous section.\n",
    "* If there was time, I would go back to the previous section and try to perform a more proper representation for campainId and gameId as it seems that those two features are not contributing much. An alternative direction forward would be to use non-linear models or feature transformation at the modelling stage."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  Decision tree on balanced data:     \n",
    "* Let's run decision tree. First we need to find a good depth for the tree so that we would not overfit."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best result for depth: 6\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.tree import DecisionTreeClassifier\n",
    "\n",
    "max_depth = 30\n",
    "# try different depths and save the CV results a\n",
    "roc_scores = np.zeros([max_depth, 5]) \n",
    "for depth in range(1,max_depth+1):\n",
    "    dt = DecisionTreeClassifier(max_depth=depth)\n",
    "    roc_scores[depth-1,:] = cross_val_score(dt, X_resampled, Y_resampled, scoring = 'roc_auc', cv=5)\n",
    "\n",
    "#print(roc_scores)   \n",
    "mean_roc = np.mean(roc_scores,1)\n",
    "best_depth = np.argmax(mean_roc)+1\n",
    "print(\"best result for depth: {}\".format(best_depth))\n",
    "plt.plot(range(1,len(roc_scores)+1),mean_roc,'.')\n",
    "plt.title(\"Average 5-fold CV error (ROC)\")\n",
    "plt.xlabel(\"decision tree max depth\")\n",
    "plt.ylabel(\"average ROC score\")\n",
    "plt.show()   "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Based on the above results, it seems that depth = 6 or 7 are reasonable choices. In the next block we use the training and validation data from the previous section to visualize the ROC curve and confusion matrix for decision tree."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion matrix, without normalization\n",
      "[[8776 4470]\n",
      " [5820 7430]]\n",
      "Normalized confusion matrix\n",
      "[[0.66253963 0.33746037]\n",
      " [0.43924528 0.56075472]]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "dt = DecisionTreeClassifier(max_depth=7)\n",
    "dt.fit(X_train, Y_train)\n",
    "y_pred_dt = dt.predict(X_val)\n",
    "\n",
    "dt_roc_auc = roc_auc_score(Y_val, y_pred_dt)\n",
    "fpr, tpr, thresholds = roc_curve(Y_val, dt.predict_proba(X_val)[:,1])\n",
    "plt.figure()\n",
    "plt.plot(fpr, tpr, label='Decision Tree (area = %0.2f)' % dt_roc_auc)\n",
    "plt.plot([0, 1], [0, 1],'r--')\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 for decision tree')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.savefig('Log_ROC')\n",
    "\n",
    "# Plot non-normalized confusion matrix\n",
    "plot_confusion_matrix(Y_val, y_pred_dt, classes=['0','1'],\n",
    "                      title='Confusion matrix, without normalization')\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "plot_confusion_matrix(Y_val, y_pred_dt, classes=['0','1'], normalize=True,\n",
    "                      title='Normalized confusion matrix')\n",
    "\n",
    "# visualize feature importances of the model\n",
    "plt.figure()\n",
    "plt.plot(X_resampled.columns,dt.feature_importances_, '.')\n",
    "plt.plot(X_resampled.columns,0 * dt.feature_importances_, '--')\n",
    "plt.title(\"Feature importances of decision tree\")\n",
    "plt.xticks(rotation=90)\n",
    "plt.xlabel(\"feature name\")\n",
    "plt.ylabel(\"Importance\")\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  Random forest on balanced data:     \n",
    "\n",
    "Here we also train a random forest. First let's try to roughly (not enough time to do this very accurately) tune some of the hyperparameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "max_n_estimators = [50, 100, 200, 400]\n",
    "max_depths = [7, 20, 100, 300]\n",
    "# try different parameters\n",
    "roc_scores = np.zeros([len(max_n_estimators)*len(max_depths), 2]) \n",
    "it = 0\n",
    "for num_estimate in max_n_estimators:\n",
    "    #print(num_estimate)\n",
    "    for depth in max_depths:\n",
    "        rf = RandomForestClassifier(n_estimators=num_estimate, max_depth=depth)\n",
    "        roc_scores[it,:] = cross_val_score(rf, X_resampled, Y_resampled, scoring = 'roc_auc', cv=2)\n",
    "        it = it + 1\n",
    "\n",
    "#print(roc_scores)   \n",
    "mean_roc = np.mean(roc_scores,1)\n",
    "#print(\"best result for depth: {}\".format(best_depth))\n",
    "plt.plot(mean_roc,'.')\n",
    "plt.title(\"Average 2-fold CV error (ROC)\")\n",
    "plt.xlabel(\"hyperparameter configuration (n_estimates and depth)\")\n",
    "plt.ylabel(\"average ROC score\")\n",
    "plt.show()   "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Based on the figure above (see the code to find mapping from x axis to the two parameters), it seems that the best depth and number of estimateors for the random forest is 7 and 200. Let's use these parameters for the validation. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 136,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion matrix, without normalization\n",
      "[[7675 5571]\n",
      " [4171 9079]]\n",
      "Normalized confusion matrix\n",
      "[[0.5794202  0.4205798 ]\n",
      " [0.31479245 0.68520755]]\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYoAAAEWCAYAAAB42tAoAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzs3Xd4FFX3wPHvSYDQQg019N6kBhBFxIICUsUCNhReEXvB96coKmBXFEV5RUVEsYAFBAQFQYp0EnoPJZAChJaEhPTc3x93g0sIyQay2ZTzeZ48ye5OOTM7mTNz79x7xRiDUkopdSleng5AKaVU/qaJQimlVJY0USillMqSJgqllFJZ0kShlFIqS5oolFJKZUkThZuIyL0istjTcXiaiNQRkVgR8c7DddYTESMixfJqne4kIjtFpPtlzHdZx6CIlBKR+SISLSI/53R+TxCREBG5OQ/WU+D2TW4oEonCcRDFO05Yx0RkuoiUdec6jTHfG2Nucec68qOM/7DGmCPGmLLGmFRPxuUpjoTV6EqWYYxpaYxZns16LkqOV3AM3gFUAyobY+68jPkLM4/sGxF5UERW5dX6MioSicKhrzGmLNAWaAeM9nA8l8WTV8mF5Qo9J4ro/q4L7DPGpOR0RldiLuDHkVv3Tb5ljCn0P0AIcLPT6/eABU6vfYAJwBHgODAFKOX0eX9gCxADHAB6Ot4vD3wFHAXCgTcAb8dnDwKrHH9PASZkiGku8Jzj75rAr8AJ4BDwlNN0Y4FfgO8c6/9PJttXHvjWMf9hYAzg5RTHauATIBrYA9yUYd6stmE1MBE47fisIfA3cAo4CXwPVHBMPwNIA+KBWOD/gHqAAYo5plkOvO5Y7llgMeDnFM8Djm04BbyS8bvLsN2lgA8c00cDqxzvpa9zqOM7PQm87DRfJ2AtEOXY7k+BEk6fG+BxIBg45HjvYyDU8R0EAdc5Te8NvOQ4Ns46Pq8NrHQsK86xP+52TN8HezxFAWuA1hmO1ReAbUAiUMx5HzhiD3TEcRz40PH+Ece6Yh0/XXA6Bh3TtAT+cnyXx4GXMtmn44AkINmxnOHYC8oxjv0ciT3WyjumT9/Xwx0xrMxkmd2BMMd2HcMeJxWB37HH7BnH37Wc5llO1sfJ/fx7nLycYR/5AB8BEY6fjwCfDLH8n2NbjgIDgN7APse+uWi/5Oa+Aa52fO9RwFagu9M6HgQOOrb5EHAv0BxIAFId643K83OoJ07ceb6RFx5EtYDtwMdOn38EzAMqAb7AfOBtp3/MaKCH46DwB5o5PvsN+BwoA1QFNgCPOH3h6YmiG/YkI47XFbEn05qOZQYBrwIlgAaOA+VWx7RjHQfmAMe0pTLZvm+xicfXcXDuA4Y7xZECPAsUB+52bE8lF7chBXgSe8IqBTRy7AsfoAr2ZPhRZvs6wz+Lc6I4ADRxLG858I7jsxaOf4Sujn0xwbHtl0oUkx3z+2NP1tc44kpf55eOdbTBnnSbO+brgP1nLeaYdjfwjNNyDfaEWil9fwP3AZUd84zCnvBKOj77L/aYagqIY32VnZbVyGnZ7bEnlM6OmIc69pmP0/7bgk00pTLuU2yCu9/xd1ng6sz2cybHoC/2pDgKKOl43fkS+3Us8J3T62HAfuyxWRaYDczIsN5vscdQZsdnd+xx9K7j+ynl2JeDgNKOWH4GfnOaZznZHyfdHMv70LH89H00HliHPZ6rYE/Kr2eI5VXs/8PD2GT1gyOOltiTcgN37BvssXoKm5i8sP9LpxxxlsFeADR1zF8DaJnxu/TIOdRTK87TjbT/aLHYLG2Apfx7FSzYK76GTtN34d8ryc+BiZkssxr25ON85zEEWJbJP6lgryi6OV4/DPzt+LszcCTDskcDXzsdmBddpTlN6+2Io4XTe48Ay53iiMCRpBzvbcBekbmyDUcutW7HNAOAzRn2dXaJYozT548Bfzr+fhX40emz0tgruIsSheOfLB5ok8ln6euslWGbB19iG54B5ji9NsCN2Wz3mfR1A3uB/peYLmOi+AzHScvpvb3A9U77b1gmx2/6SXAl9srWL8M0F+znTI7BIc7fUzbbNpYLT4ZLgcecXjfFJvBiTuvN9MTqmL6743ssmcU0bYEzTq+zO05mOn1Wxvk4wSaY3k6f3wqEOMUSz793zb6O+Ds7TR8EDHDHvsHeVc3IsMxF2AuGMti7jEFkSLh4OFEUpTqKAcYYX+yB0gzwc7xfBXtCChKRKBGJAv50vA/2yu5AJsuri70iOeo03+fYq5gLGPtNz8T+swLcgy2ySV9OzfRlOJbzEvYkni40i+3yw159H3Z67zD2yiVduCMG589rurgNF6xbRKqKyEwRCReRGGyRmB85c8zp73PYKzEcMZ1fnzHmHPZqKzN+2CvjzL6bLNcjIk1E5HfHgw0xwFtcvA0Zt3uUiOx2PO0ShS2yS5/nUsdIZuoCozJ837Wx257pujMYjr3K3iMiG0Wkj4vrzUmMGdXk4uOrGK4fowAnjDEJ6S9EpLSIfC4ihx3fwUqgQoan41w9TuK48DjJLF7n/XvK/PtwRbzj93Gnz+Od1pWdnO6busCdGb7/rkANx3bcDYzE/k8uEJFmLsbhVkUpUQBgjFkBTMcWa4Atv47H3uJVcPyUN7biG+yX3DCTRYVir8b9nOYrZ4xpeYlV/wjcISJ1sXcRvzot55DTMioYY3yNMb2dw85ik05ir2DqOr1XB1vfkM5fRCTD5xEubkPGdb/teK+1MaYctkhGspg+J45iiwYB+ygitogiMyexRQSZfTfZ+QxbV9PYsQ0vceE2gNN2iMh12CvBu4CKxpgK2OK79HkudYxkJhR4M8P3XdoY82Nm687IGBNsjBmCTebvAr+ISJms5rmMGDOK4OLjK4ULT67ZrT/j56OwV9+dHd9BN8f7Gb+HzBzFJj47g0hpLjxOMos3woXlXo6c7ptQ7B2F8/dfxhjzDoAxZpExpge22GkPtvg04zLyXJFLFA4fAT1EpK0xJg37ZUwUkaoAIuIvIrc6pv0KeEhEbhIRL8dnzYwxR7EVbB+ISDnHZw1F5PrMVmiM2YwtC50KLDLGRDk+2gDEiMgLjme0vUWklYh0dGVDHFdGPwFvioivIxE9h73ST1cVeEpEiovIndjKsYU53QYHXxwVaiLijy2fd3YcW157OX4B+orINSJSAlvEkumJw/G9TQM+FJGajv3WRUR8XFiPL7YsONZxxfaoC9OnYL+/YiLyKlDO6fOpwOsi0lis1iKSfuLKuD++BEaKSGfHtGVE5DYR8XUhbkTkPhGp4tj+9GMo1RFbGpfe978D1UXkGRHxcRwrnV1ZJ/Yi51kRqe94rPwtYJa5jCd/nPhiL9CiRKQS8FoO5v0F6CMiXR3HyXguPJf9CIwRkSoi4octqvouk+Xkhpzum++wx/itjmO2pIh0F5FaIlJNRPo5En8i9v8s/c7nOFDLsb15rkgmCmPMCWwF0yuOt17AVkitc9wGL8Fe7WCM2QA8hH3yJxpYwb9XEA9gi312Ycusf8FeCVzKj8DN2Iqz9FhSgb7YMtpD2CvlqdiiDVc9ia1nOYh98ucH7Ek03XqgsWPZbwJ3GGPSb9Vzug3jsBWy0cACbOWds7ex/6RRIvJ8DrYBY8xOx7bMxF41nsVW/CZeYpbnsZXIG7FPq7yLa8f089jiv7PYE/esbKZfBPyBfUjgMPZOxrk44UNssl6MTUBfYSsuwZZpf+PYH3cZYwKxdVSfYvf3fmz5s6t6AjtFJBb7JNZgY0yCo5juTWC1Y11XO89kjDmLrTjtiy3SCQZucHGd07BPKq3EHqMJ2O/pSnyE3UcnsRXPf7o6o+M4eRx7nB/F7scwp0newD4Ztg17fGxyvOcOOdo3xphQ7FOUL2GTeyj2YsvL8TMKe5dyGrgeWzcD9knDncAxETnpjg3JSvpTOKqQEpEHsY/UdvV0LDnluEKLwhYRHfJ0PEoVVUXyjkLlXyLS11HRWQZbj7Qd+9SPUspDNFGo/KY//zaUaowtWtHbXqU8SIuelFJKZUnvKJRSSmWpwHVS5efnZ+rVq+fpMJRSqkAJCgo6aYypkv2UFytwiaJevXoEBgZ6OgyllCpQRORw9lNlTouelFJKZUkThVJKqSxpolBKKZUlTRRKKaWypIlCKaVUljRRKKWUypLbEoWITBORSBHZcYnPRUQmich+EdkmIu3dFYtSSqnL5847iunYLpEvpRe2L5/GwAjsYDJKKaVyUVqaISE5NfsJs+C2BnfGmJUiUi+LSfoD3zo6fFsnIhVEpIZjMB2llFKXkJpmiDybQPiZeI7FJJCQnEbIyTh2REQTdS6ZfcfPci4pFYzh1n1ruTV47RWtz5Mts/25cPCXMMd7FyUKERmBveugTp06eRKcUkp5SlqaYXPoGU7HJXP4VBypaYb1h04TcjKO8Kh4ElPSLprHS6Bp9XL4lS3BNQ0rU+3UMe757n1abv6Hkw2ubOhtTyaKzIa4zLQrW2PMF8AXAAEBAdrdrVKqwDPGsO94LBHR8Ww6fIb4pFROn0ti/cHThEfFXzR9xdLFKe7tRbPqvtTzK0N9vzLUqliaWhVLUaN8SSqX9aGsT7H0hUNAAOzbCx98gN9TT0Hx4pcdqycTRRhOA6QDtXDfAOhKKeUx0eeS2XU0hpOxiSzdfZyYhBTWHjhFfIa6g2rlfGhUtSzXNfajVsVSBNSrRFVfHyqX8aFcqWKIZDqE/L/WrIGrrgJfX5g6Ffz8oHbtrOdxgScTxTzgCRGZCXQGorV+QilV0BljCDsTz9oDp/hlUxi7I2I4m5hy0XTXNfajc/1KXFWrAk2r+VLV1wcvr2wSwaWcOgUvvmiTw2uvwdix0K7dlW2IE7clChH5EegO+IlIGPAaUBzAGDMFWAj0xg4ufw54yF2xKKWUOxhj2BIaxZLdx1m57yTHYhI4cTbx/OcVSxenrl9prm3kR7vaFalWzodaFUvjV7ZE9ncHrgUA334Lzz8PZ87Af/9rf3KZO596GpLN5wZ43F3rV0opdwg9fY7Aw6dZujuSdQdPczLWJoZyJYtRopg3PVpUo3mNcvRqVZ1m1X1zJyFcygsvwPvvwzXXwJQpttjJDQrceBRKKZVX4hJT2BIaxexN4Ww6coaoc0mcOZcMQIXSxenSoDId6lbk9va1qFSmRN4EFR8PcXG2/mH4cGjc2P72cl+zOE0USinFv4+kbjocxb7jZ/lzx7GL6haaVfdl5PUNuapWeTrVq0Qx7zzuBenPP+Hxx6FtW/j1V2ja1P64mSYKpVSRlJCcytLdkaw5cJK9x84SePjMBZ/3aFGNWhVL0bSaLzc2r0pV35IeihSIiIBnnoGff7aJ4Ykn8nT1miiUUkWCMYbgyFg2HT7DnM3hbAw5TZqBEt5etPQvx31X16FJNV/a16lIy5rl3Fu3kBNLl8LAgZCUBK+/biurfXzyNARNFEqpQuvQyTiW7DrO/shYlu2NJNLxRFLN8iX5z3UN6Fy/Et2aVKF4XhchuSI52TaSa9MGeveGN96ARo08EoomCqVUoZCcmsZPgaEEhpxh99EYwqPiOZvwbx1DjxbV6NbYj/Z1K9K8ernLb7PgbjEx8MorsH49rF5tK61nzvRoSJoolFIFVnxSKnM2h/P3nkj+CT5xvg+kLg0q075uRepVLk2LGuW5uoEHKp5zyhj45Rd4+mk4dgweewwSE6F0aU9HpolCKVWwHDl1jvnbIvgn+ASBIWdISbPdv/VsWZ2bmldlYDv//J8UMjpxAoYOhT/+sC2q586Fjh09HdV5miiUUvlaWpoh5FQc01YfYvneE4SdsR3mNalWljsDanNLi2pc19iv4CUHZ+XKwcmT8NFH9vHXYvnr1Jy/olFKKf59QumH9Uf4JSiMWEd7hrI+xRjYzp/Hb2hEo6plPRzlFVq5Et5807aHKFsW1q1za6O5K6GJQimVL6SkprF413H+2nWclftOcCouCbCd593SohrXNvKjQZUCnhzA3jn8978wfTrUqwchIdCqVb5NEqCJQinlYaGnz/H16hDmbY3gZGwixb2FgLqVuK11DW5qXpUa5Ut5OsTcYQx8/bVNEjExMHo0jBmTLyqrs6OJQimVp9LSDGsOnGLGuhA2H4ki8mwiInB9kyoMbOfPrS2rU7K4t6fDdI/vvoMWLWwHfi1bejoal2miUEq5XXR8Msv3RrIq+CRzNoeTkmbwEmjlX55Hrm9Ij+bVqFM5/19Z59i5c/DWWzByJNSqZesjypfP18VMmdFEoZRyi8iYBJbsjmT+1giCjpwhKSUNX59idG5QiRubVeOugFr4lrz84TnzvYUL7RNMISHg7w+PPgoVK3o6qsuiiUIplWuMMSzdHcnk5fvZfCQKgLqVSzOkY216tqpBp/qV8M6vLaJzS1iY7cDv11+heXNYsQK6dfN0VFdEE4VS6oqkpRk2hJzmj+1HWbonkrAz8ZQvVZyeLavzwDV16dKgcv7pYC8vvPkmLFhgi5xGjYISeTROhRuJHWiu4AgICDCBgYGeDkOpIi0tzbD24Cne/mM3h07EEZeUCkAVXx8e796QezrXpUSxglUOf0U2bIBSpewIc6dOQXQ0NGjg6aguICJBxpiAy5lX7yiUUi4xxrA1LJqF248yb0sEx2ISKOYl+JX1YUyfFlzb0K9wVkhnJToaXnoJPvsM+vSBefOgcmX7U4hoolBKXZIxhpXBJ5m9KYxVwSfPN4JrX6cCz93ShN5X1aCsTxE8jRgDs2bBs89CZCQ8+aQdK6KQKoLfsFIqK6lphi2hUSzedYzftx4lPMr2rdSwShnu71KXQe1rUbtSEbtzyOi77+CBByAgAH7/HTp08HREbqWJQikFwPqDp/h+/RHWHjzFCccAP21qV2B41/oMbOdPxTIFv1L2iiQmwsGD9kmmu+6ClBSbLLwLaeNAJ5oolCriws6c49egcCYu2QdAgypleK5HE3q1qk6F0kU8OaRbtsy2gzh3DoKD7VCkDz3k6ajyjCYKpYqozUfOnO9jCaBrIz8+vKsNVcuV9HBk+UhkJDz/PMyYYZ9i+uKLPB+vOj/QRKFUERIdn8ysjUeYvSmcPcfOUsxLGNKpDkOvqUuz6uU8HV7+sn8/dOoEsbHw8sv2p1Qh6aAwhzRRKFUEBB8/y7TVIfy44QgADfzKMLZvCwa2r0X5UoW4G43LERNjBxJq2BCGD4dhw2y9RBGmiUKpQiotzbB8XyQzN4SyZPdx0gzc3LwagzvW5uYW1TwdXv4TFwfjx8OXX8K2bbYTv/ff93RU+YImCqUKmcSUVGZtDGXK8gNERCfgJTD0mnr857oG+FcomkUn2Zo/H554Ao4csXcRBWCMiLykiUKpQiA+KZWgw2f4Y8dRZm8KJz45lUZVy/JKnxYM6VSb0iX0Xz1TKSn2Udc5c+z4EP/8A127ejqqfEePHqUKsITkVD5bfoCvVh06P650QN2KDL2mHn1a1yhanfHlhDEgAsWKQY0a8M47tpV1IejAzx00UShVACWlpDFxyT6+XRNCXFIq7epU4MFr6tGlYWWq+urjrVlat86OE/Hll9C+PUye7OmI8j1NFEoVICmpaczeHM7HS4IJj4qnU71KPH5jI65vUsXToeV/Z87YDvw+/xxq1rSvlUvcmihEpCfwMeANTDXGvJPh8zrAN0AFxzQvGmMWujMmpQqiM3FJTFl5gF+DwjgZm0TdyqWZcl97eraq4enQCoZZs+Cpp+DkSTuo0Lhx4Ovr6agKDLclChHxBiYDPYAwYKOIzDPG7HKabAzwkzHmMxFpASwE6rkrJqUKEmMMy/edYNqqQ/wTfBKAsj7FeGNAK4Z0qlP4R4rLTXv2QL168Oef0K6dp6MpcNx5R9EJ2G+MOQggIjOB/oBzojBAenPQ8kCEG+NRqkBISzOsCD7Be3/uZffRGCqWtqPFDWhXU+8gXJWQAO++a+sg+va1RU5jxhSJDvzcwZ2Jwh8IdXodBnTOMM1YYLGIPAmUAW7ObEEiMgIYAVCnTp1cD1Sp/ODIqXPMWBfC7E3hnIpLwq+sD6/3b8ldHWvjU0xPcC5bsgQee8x23jdqlE0UxbX1+ZVwZ6LI7L4447irQ4DpxpgPRKQLMENEWhlj0i6YyZgvgC/ADoXqlmiV8pCwM+d4+489LNh2FLDda7zYqxl929SkZHFNEC47fhyeew5++AEaNYLFi6FHD09HVSi4M1GEAbWdXtfi4qKl4UBPAGPMWhEpCfgBkW6MS6l8YVdEDN+sCWH+tghSUg1DOtVm5PUNqVu5jKdDK5j++gt++QVefRVGj4aS+phwbnFnotgINBaR+kA4MBi4J8M0R4CbgOki0hwoCZxwY0xKeVxEVDxjftvBsr2ReItwXWM/XurdnMbV9CmcHNu61RYx3XEH3HsvXHst1K/v6agKHbclCmNMiog8ASzCPvo6zRizU0TGA4HGmHnAKOBLEXkWWyz1oDFGi5ZUoZOYksqa/aeYtzWCBduOkpKWRp/WNRnXryWVivrIcZcjNhZeew0+/tg+zTRggG1lrUnCLdzajsLRJmJhhvdedfp7F3CtO2NQypPWHTzFrI2hLNp5jHNJqfj6FKNf25o81r0hDaqU9XR4BdNvv8GTT0JYGIwYAW+/bZOEchvdu0rlMmMMszaGMnXVIfZHxlKimBe3XVWDfm1q0qVhZa2gvhLbt8PAgXDVVbYR3TXXeDqiIkEThVK5aOH2o7w2bycnzibSvEY5/ntrU+67uq4ODnQlkpNtr6433mgTxIIF9mkmfeQ1z2iiUCoXRJ5N4LlZW1m137agfqRbA/6vZzNtPX2l1qyBkSNh507Yu9c+9tq7t6ejKnI0USh1BdLSDPO3RfDS7O3EJaXSr01N3hzYCt+SerV7RU6fhhdftD281q4Ns2fbJKE8QhOFUpdpS2gUz/20hYMn4vCvUIqpQ9vQpWFlT4dV8CUkQNu2EBFhW1aPHQtlteLfkzRRKJVD+yNjeeePPSzZfZxyJYvx3h2tGdDWnxLFvDwdWsEWFmbHqS5ZEl5/3SaLNm08HZVCE4VSLouMSeCzFQf4ccMRElPSePi6+jzavZG2g7hS8fH2Edd337Utq/v2haFDPR2VcuJSohCREkAdY8x+N8ejVL4THZ/MJ0uDmbrqEAAd61Vkwp1ttKuN3LB4se3A78ABuO8+6NTJ0xGpTGSbKETkNuBDoARQX0TaAq8ZYwa6OzilPCk5NY0/dxxj3PxdnIxNpE/rGtx/dV061a+kY1HnhiefhE8/hcaNbY+vN93k6YjUJbhyRzEe2z34MgBjzBYR0ccPVKGV3mBu4pJ9HI9JpEb5knz9UEduaFrV06EVfKmp9re3N1x9Nfj5wQsvaAd++ZwriSLZGBOV4QpK+2NShY4xhr/3RDJu/i6OnD5Hs+q+vNqnJT1aVNOK6tywaZNtE3H//fZu4t57PR2RcpEriWK3iNwFeDl6gn0aWOfesJTKW6v3n+S1eTvZHxlLWZ9ivDmwFUM61sFLG8xdubNnbdffkyZBlSpQQ0fpK2hcSRRPAK8CacBsbG+wo90ZlFJ55WxCMh8s3sf0NSGUK1mMMbc15+6OtbXBXG5ZvBiGDbNtIkaOhLfeggoVPB2VyiFXEsWtxpgXgBfS3xCR27FJQ6kCKSkljVkbj/Dx0mBOxibRv21NxvdvpX0y5bYSJaBqVfj1V+iccSRkVVC4kijGcHFSeDmT95TK9xKSU/ly5UE+XLIPY6Cqrw/THgzgxmbVPB1a4ZCcDB9+CDEx8Oab0L07BAaCl9bxFGSXTBQicit2mFJ/EfnQ6aNy2GIopQqU3zaHM3nZfoIjY7nKvzz3dK7DoPa1tKI6t6xa9W8HfnfeCWlpNkFokijwsrqjiAR2AAnATqf3zwIvujMopXJLaprh920RfPjXPg6fOkedSqWZcl8Heraq7unQCo9Tp+wjrl99BXXqwPz50KePp6NSueiSicIYsxnYLCLfG2MS8jAmpXJFeFQ8z8zczMaQM/iVLcHztzRh5PUNKeatV7i56tQpmDkT/u//7NNNZbTFemHjSh2Fv4i8CbQAzreKMcY0cVtUSl2h6asPMXb+LgBG9WjCw90a6MhyuWn3bvjpJztudZMmcOQIVKrk6aiUm7iSKKYDbwATgF7AQ2gdhcqndkXE8Mnfwfyx4xhNq/nyWt8WXNPIz9NhFR7nztlK6vfft11/Dx9ue3zVJFGouZIoShtjFonIBGPMAWCMiPzj7sCUyolzSSmMm7eLWYGheHsJ/+lan//r2UwrqnPTn3/aDvwOHbK9u77/vm1Apwo9VxJFotj+Ow6IyEggHNBOb1S+kJyaxk+BoUxYtJcz55Lxr1CKn0Z2wb9CKU+HVrjExtquNypXhmXL7GOvqshwJVE8C5QFngLeBMoDw9wZlFKuOB2XxB2freHgyTiqlfPhu+Gd6dpYi5lyTWoq/PgjDBlii5mWLIFmzcDHx9ORqTyWbaIwxqx3/HkWuB9ARGq5MyilshN0+AyPfhdE1Llk7ZfJHYKC4JFH7O9SpWDQIB1trgjLMlGISEfAH1hljDkpIi2xXXncCGiyUHnuXFIK4+fbughjYN4T19K6lvYdlGuio+GVV2DyZNv1xsyZcPvtno5KeVhWLbPfBgYBW7EV2HOwPce+C4zMm/CU+teSXcd5bd5OwqPiub2dP4/d0IhGVct6OqzCZdAg+PtvePxxeOMNKF/e0xGpfCCrO4r+QBtjTLyIVAIiHK/35k1oSlnGGD5aEszHS4MpV7IYn93bnl5XaVfVuebgQfv0kq+vffTVyws6dvR0VCofyerZwQRjTDyAMeY0sEeThMprOyOieWj6Rj5eGsx1jf1Y/eKNmiRyS1KS7fa7ZUt79wC2h1dNEiqDrO4oGohIeg+xAtRzeo0xRgsuldskp6bx3p97+PKfQ5Tw9uKJGxrxzM2NtfuN3LJype3Ab/duuOMOeOopT0ek8rGsEsWgDK8/dWcgSqXbcOg0r/y2g73Hz3JNw8p8dHdbqpbTMZVzzcSJ8NxzUK8eLFgAvXt7OiKVz2XVKeDSvAxEqbQ0w9RVB3lr4R4qlC7OR3e3ZUA7f0+HVTikpUFcnK2HuO0//V9AAAAgAElEQVQ2OHECxoyB0qU9HZkqAFxpcKeU2x04Ectzs7awNSyaZtV9+fKBAGpX0pNYrti50xYzpY8016SJrZtQykVuLfAVkZ4isldE9otIpmNYiMhdIrJLRHaKyA/ujEflP6GnzzHi20Bu/nAF28Kjeeqmxix46jpNErnh3DkYPRratrV1EX36gDGejkoVQC7fUYiIjzEmMQfTewOTgR5AGLBRROYZY3Y5TdMYGA1ca4w5IyLah1QR8vu2CJ76cTNeIgzuWJvHujfSBJFbNm+2DeVCQuChh+C998BPuzdRlyfbRCEinYCvsH081RGRNsB/jDFPZjNrJ2C/MeagYzkzsW0zdjlN8zAw2RhzBsAYE5nzTVAFTUJyKlNWHOCjJcFUKlOCaQ92pG1tbV2dK4wBETvSXJ068M030K2bp6NSBZwrdxSTgD7AbwDGmK0icoML8/kDoU6vw4DOGaZpAiAiqwFvYKwx5k8Xlq0KqFXBJxk3fyfBkbG0q1OBD+9qS30/HRHtiqWkwKefwrx58NdftpfXFSs8HZUqJFxJFF7GmMO2p/HzUl2YL7Me2jIWkBYDGgPdsX1H/SMirYwxURcsSGQEMAKgTp06Lqxa5TdbQ6P4fOUBFm4/RrVyPjpudW7asMFWVm/eDL16QUwMVKzo6ahUIeJKogh1FD8ZR73Dk8A+F+YLA2o7va6F7QYk4zTrjDHJwCER2YtNHBudJzLGfAF8ARAQEKC1cQVIWprhp8BQRs/ZjjFwU7OqTLizDRXLlPB0aAVfbCy88AJ89hnUqAE//2z7ahLtRVflLlcSxaPY4qc6wHFgieO97GwEGotIfexgR4OBezJM8xswBJguIn7YoqiDroWu8rs9x2IYM2cHgYfP0MCvDJ/e054WNct5OqzCo3hxWL4cnnwSXn8dyum+Ve7hSqJIMcYMzumCjTEpIvIEsAhb/zDNGLNTRMYDgcaYeY7PbhGRXdjirP8aY07ldF0qf0lJTeNdR/cbAM/f0oRHuzfCW8eLuHL798P48bYbcF9fO15ESW21rtxLTDbPVYvIAWAvMAuYbYw5mxeBXUpAQIAJDAz0ZAgqCyv2nWDEt4EkpqRxQ9MqvNKnBQ2qaFfgVywx0T7i+uabUKKE7Xrjuus8HZUqQEQkyBgTcDnzZtvgzhjTEHgD6ABsF5HfRCTHdxiqcDPG8OHivQydtoHi3l7cd3Udvn6okyaJ3LBsmR1d7tVXYcAA2LNHk4TKUy41uDPGrAHWiMhY4CPge2CmG+NSBUhcYgqjZ29n3tYIerSoxqTB7ShVwtvTYRUOxti7iORk+PNPuPVWT0ekiiBXGtyVxTaUGww0B+YC17g5LlVAHDwRy31T1xMRncCj3RsyqkcT7Qr8SqWlwVdfQc+eULs2zJgBFSrYsauV8gBX/qN3AFcD7xljGhljRhlj1rs5LpXPGWP4aWMofT5ZRXR8MjOGd+KFns00SVypbduga1cYMQKmTrXv1aihSUJ5lCtFTw2MMWluj0QVGDsjohk9ezvbwqJpXLUsH9zVhta1tAuOKxIbC+PG2bEiKlaE6dPhgQc8HZVSQBaJQkQ+MMaMAn4VkYsejdIR7oqm2ZvCeO6nrQCM79+S+6+ui2gDrys3dix88AH85z/wzju2Cw6l8oms7ihmOX7ryHYKYwyzNoby4uzt+PoUY87j19Coqq+nwyrYQkPtYELNmsGLL9onmrp29XRUSl3kkgXKxpgNjj+bG2OWOv9gK7VVEZGaZhg7bycvzt5Os+q+LH3+ek0SVyIlBT78EJo3h0cese/5+WmSUPmWKzWPwzJ5b3huB6Lyp5CTcQz5Yh3frD3MoPa1mP9kV6r6akvgy7ZuHQQEwKhR0L277QZcqXwuqzqKu7GPxNYXkdlOH/kCUZnPpQqT7WHRDPlyHfHJqYzr15IHumh9xBVZsAD69oWaNWH2bFvUpPtTFQBZ1VFsAE5he32d7PT+WWCzO4NSnpWWZpiweC9TVhyguLcX85/oqp35XS5jICIC/P3h5pttP01PP237aVKqgLhkojDGHAIOYXuLVUVEQnIqD3y1gQ0hp7mhaRXeuv0qapTXZ/gvy7598Nhj9veuXVC2LIwZ4+molMqxrIqeVhhjrheRM1w44JAAxhhTye3RqTy1KyKGkd8FceT0Of57a1Me695Qi5ouR0KCfcT17bdtQ7n030oVUFkVPaUPd6ojshdyaWnG0S34QUoV92bSkHb0a1PT02EVTMeO2TGqg4NhyBD7dFN1HclPFWxZFT2lt8auDUQYY5JEpCvQGvgOiMmD+JSbnY5L4pEZgWwMOUOvVtV5tW8LLWq6HMnJdiChatVsopg8GXr08HRUSuUKVx6P/Q07DGpD4FtsG4of3BqVyhMJyamMnBHExpAzPNKtAZPvaa9JIqfS0mDKFGjYEMLC7FNMU6dqklCFiiuJIs0xpvXtwEfGmCcBf/eGpdztVGwiQ6fZSuuxfVswundzvHQEupzZuhWuuQYefRQaN7Z3FUoVQq4kihQRuRO4H/jd8V5x94Wk3O1YdAJ9P1nF+kOnGdWjCQ9eW9/TIRUsxsDzz0OHDnDwoO0GfMkSqK/7URVOrvQeOwx4DNvN+EERqQ/86N6wlLus3n+Se6eup4S3F9Mf6kj3plU9HVLBIwJnzsDw4fbppooVPR2RUm6V7ZjZACJSDGjkeLnfGJPi1qiyoGNmX76/9xzn4W+DSE0zfP+fzlzbSB9oc9nhw7ah3KuvQvv2tm7CS8feUAWHW8fMFpHrgP3AV8A0YJ+IXHs5K1Oe88f2owybHkjNCiVZ8d/umiRclZwM770HLVrAX3/B3r32fU0SqghxpehpItDbGLMLQESaAzOAy8pMKu/9ueMYj36/iQZVyjDz4aupWk479XPJmjW2d9cdO6B/f5g0CerU8XRUSuU5VxJFifQkAWCM2S0iJdwYk8olxhg+/Gsfn/y9n8plSjBrRBeq+Pp4OqyCY8kSiI6G336ziUKpIirbOgoRmQ4kYu8iAO4FShtjhro3tMxpHYVrtoVFMXbeTjYdieL6JlWYcGcbTRLZMcY+wVSlCvTqBYmJtuipbFlPR6bUFbuSOgpX7ihGAk8B/4ft52kl8MnlrEzljUMn47j/qw1Exyfz1I2NeLZHE+2zKTt79tj2EMuXw5132kTh42N/lCriskwUInIV0BCYY4x5L29CUlfiWHQCfSb9Q3Kq4bfHr6Vt7QqeDil/i4+Ht96Cd9+FMmXg88/tuNVKqfMu+eiGiLyE7b7jXuAvEclspDuVjwQdPs2AyauJS0pl8r3tNUm4Yv58eOMNuPtue1cxYoQ+0aRUBlndUdwLtDbGxIlIFWAh9vFYlc8YY5iy4iDvL9qDf8VSzBpxNZ0bVPZ0WPnXsWOwZQv07GmLmerVg06dPB2VUvlWVoki0RgTB2CMOSEiepmVDx2PSeDlOdtZsjuSaxpW5vP7O+BbUntYyVRqqi1aGj0aSpSAI0fsOBGaJJTKUlaJooHTWNkCNHQeO9sYc7tbI1PZWrY3kid/2Ex8ciqPdGvACz2bacd+l7JpE4wcCRs32iFJ//c/HUxIKRdllSgGZXj9qTsDUTnzv+X7ee/PvTSpVpaPB7ejeQ0d0/qSDh2ydw1+fvDDDzB4sO2vSSnlkqwGLlqal4Eo16SkpjF52QEmLtlHuzoV+HZYJy1qyowxsH07tG5te3X9+mvo2xcqaAW/Ujml9Q4FzCtzdzBxyT6ua+ynSeJSDh2CPn2gXTvYts2+d//9miSUukxuTRQi0lNE9orIfhF5MYvp7hARIyLaf9QlJKem8drcHfy4IZQHutRlxvDOmiQySkqy3X63bAkrVsCECbYzP6XUFXGlZTYAIuJjjEnMwfTewGSgBxAGbBSRec79Rjmm88W2/F7v6rKLmsizCQyfHsj28Gh6tarOy7c193RI+U9qqh1tLigIbr8dPvoIatf2dFRKFQqudDPeSUS2A8GO121ExJUuPDphx644aIxJAmYCmfWs9jrwHpDgethFx5r9J+n6zjK2h0fzap8W/O/e9vgU8/Z0WPlHTIz97e0Nw4bZBnS//qpJQqlc5ErR0ySgD3AKwBizFbjBhfn8gVCn12FkGGtbRNoBtY0xv5MFERkhIoEiEnjixAkXVl047D12lsd+2IQIfP1gR4Z1ra99NqUzBqZPhwYNYO5c+95jj9m6CaVUrnIlUXgZYw5neC/VhfkyO6Od76rW0YBvIjAquwUZY74wxgQYYwKqVKniwqoLvk1HznD3F2tJSTX8OOJqbmimQ5aet2sXdO8ODz0EzZpBw4aejkipQs2VRBEqIp0AIyLeIvIMsM+F+cIA5/v/WkCE02tfoBWwXERCgKuBeVqhDTM3HOH2/60h6lwyPzzcmfZ1dEzm8957D9q0sYMJTZ0KK1dCq1aejkqpQs2VyuxHscVPdYDjwBLHe9nZCDQWkfpAODAYuCf9Q2NMNHB+PE4RWQ48b4wp0oNNfL/+MC/P2UHTar5MHRpA7UqlPR1S/mCMbSRXvTrcey+8/74dN0Ip5XbZJgpjTCT2JJ8jxpgUEXkCWAR4A9OMMTtFZDwQaIyZl+NoCzFjDG8t3M2X/xyifZ0KfDu8M2V9XH4orfCKiICnn4brroOnnoIHHrA/Sqk8k+2ZSES+xKluIZ0xZkR28xpjFmJ7nXV+79VLTNs9u+UVZhMW7+XLfw5xY7OqfDKkHWWKepJITbX9Mb38sh1l7pprPB2RUkWWK2ejJU5/lwQGcuHTTOoKvblgF1/+c4jrm1Rh6gMB2rHfli128KCgILjlFpswtMJaKY9xpehplvNrEZkB/OW2iIqYd/7Yw5f/HOLaRpX5373tNUkAREfbIqdZs+x4EfpIsFIedTnlG/WBurkdSFE0f2sEU1YcoHP9SnzzUCeKeRfRrreMgZ9/huBgW9R0/fVw8CCULOnpyJRSuNYy+4yInHb8RGHvJl5yf2iF2+crDvDkj5vxr1CKjwe3K7pJ4sAB6N3bDkU6d66tjwBNEkrlI1neUYhtBtwG+3grQJox5qKKbZUzq/ef5O0/9tClQWWmPdiRUiWKYJcciYm207433oDixeHjj23L6mJFvBJfqXwoy/9KY4wRkTnGmA55FVBh9/36w7y9cA9VfH34/IEORTNJAISGwuuv2zEiPvoI/P2zn0cp5RGulHdsEJH2bo+kCHh/0R5enrMDgB8fvppyRa2b8BMn4FPHQImNGtmuOH7+WZOEUvncJe8oRKSYMSYF6Ao8LCIHgDhsH07GGKPJIwd+2xzO5GUH6NWqOhPvbkvJ4kXoTiItzY4w93//B2fPQo8e0LSp7dBPKZXvZVX0tAFoDwzIo1gKJWMMn/y9n4+W7KNJtbJFL0ns2AGPPgqrVtnW1VOm2CShlCowskoUAmCMOZBHsRQ6p+OSuHPKGg6ciOP6JlX4qKgliaQk22AuKQmmTYMHH9Q2EUoVQFkliioi8tylPjTGfOiGeAqNiKh47pyylvCoeAZ3rM1bA68qOo3p/v7btoUoUQJ++sl2Be7nl/18Sql8KavKbG+gLLY78Mx+1CWcik2kzyerCI+KZ+oDAbwzqHXRSBJhYTBoENx0E3z7rX2va1dNEkoVcFndURw1xozPs0gKiZiEZHpP+ofTcUl8dm97bm5RzdMhuV9Kin2a6ZVXbGd+b79tuwJXShUK2dZRKNfFJ6Vy9+frOB6TyGt9W9DrqhqeDilv3H8/zJwJvXrB5MlQv76nI1JK5aKsEsVNeRZFIZCWZnhx9jZ2H41hXL+WDL2mnqdDcq+oKNuKumxZePxxW+Q0aJBWVitVCF2yjsIYczovAynoxv++i7lbIniuR5PCnSSMsXcPzZvboiaw9RB33KFJQqlCqoj2RJe7/txxjOlrQhjcsTZP3NDI0+G4z/79cOutMGQI1KoF993n6YiUUnlAE8UVOhWbyGvzdlCrYinG9W9ZeJ9u+uEHaNUK1q+3Fdfr1kEH7QJMqaJAu+q8AueSUrjny/Ucj0nk10evwadYIWxMl5xse3cNCLDFS++9BzVrejoqpVQe0juKy3QuKYUHv97I3uNneaVPCzrUrejpkHJXZKR9munuu+3rJk3gu+80SShVBGmiuAyn45Lo/+lqNhw6zehezRjetRA9DpqWBl98YftjmjULWra0bSOUUkWWFj3lUGJKKk/9uJngyFhG92rGI9c39HRIuefgQVtBvXYtdO8On31mu99QShVpmihywBjD0z9uYdX+k7zcuzkPdytk3WSXL2/bR3zzjS120sddlVJo0VOO/LHjGH/uPMb9V9ctPEli3jy4/XZbvFS5su0W/IEHNEkopc7TROEiYwyfr7A9rr/Uu7mHo8kFR47AgAHQvz/s2wdHj9r3vfSQUEpdSM8KLkhITuX/ftnG1rBonr+lScEe5zolBSZMsC2rFy+Gd9+FzZttAzqllMqE1lG4YNz8XfwcFMbAdv482r2At7xOTYWpU+HGG+GTT6BePU9HpJTK5/SOIhu/bQ7nxw1H6NemJh/e1Qbvgtjy+swZeOEFO161jw+sXm3rJjRJKKVcoIkiC6fjknhm1haq+vrwxsBWSEGr4DUGvv/ePuL6wQewbJl9v3JlraxWSrlME8Ul2LEl1gLw/p1tKFeyuIcjyqF9+6BHD9suol49CAyEfv08HZVSqgDSOopLmLxsP8GRsYy5rTnXN6ni6XBy7plnbHL43/9gxAjwLsAV8Eopj9JEkYlVwSf5dNl+ujWpUrC65/jrL1vMVLu2bVXt4wPVq3s6KqVUAefWoicR6Skie0Vkv4i8mMnnz4nILhHZJiJLRaSuO+NxRWRMAi/8uo2KpYszaXDbglEvcewY3HMP3HKLfdwVoG5dTRJKqVzhtkQhIt7AZKAX0AIYIiItMky2GQgwxrQGfgHec1c8rkhKSaPfp6sJj4rnnUGtqVC6hCfDyV5aGkyZYu8ifv0VXnvNtpFQSqlc5M47ik7AfmPMQWNMEjAT6O88gTFmmTHmnOPlOsCjrb7eXLCLYzEJvDXwKm5tWQCuxt9+Gx591A4gtG0bjB0LJUt6OiqlVCHjzjoKfyDU6XUY0DmL6YcDf2T2gYiMAEYA1KlTJ7fiu8D2sGi+WXuYfm1qck9n96wjV5w9CydPQv36MHKk/T1kiD7uqpRyG3feUWR25jKZTihyHxAAvJ/Z58aYL4wxAcaYgCpVcv8JpBNnE3lkRiDlShbj1b4ZS8fyCWNgzhxo0cIOJmSMbQ9xzz2aJJRSbuXORBEG1HZ6XQuIyDiRiNwMvAz0M8YkujGeTJ1NSObeqes4fjaRLx4IwK+sT16HkL3Dh20biNtvh0qVYNIkTQ5KqTzjzqKnjUBjEakPhAODgXucJxCRdsDnQE9jTKQbY8mUMYaHvw1k3/FYPh7clqsbVM7rELK3di3cfLP9e8IEePppKKZPNSul8o7b7iiMMSnAE8AiYDfwkzFmp4iMF5H0JsLvA2WBn0Vki4jMc1c8mZm3NYJ1B09zT+c69G/rn5erzl5MjP3dvj0MGwa7d8OoUZoklFJ5TozJtNog3woICDCBgYFXvJy4xBRavraIymVKsOqFG/NP1+GnTsGLL9ouwHfuhLJlPR2RUqoQEJEgY0zA5cxbZPt6+m7dYQDeHdQ6fyQJY+Dbb22biK+/thXWWg+hlMoHimQ5xv7Is0xYvJeGVcpwU/Oqng4HoqPtaHPLl0OXLrYRXevWno5KKaWAIpgojkbHM3TaRrxE+HhwO8920WGMvWsoVw78/OCLL2D4cB2OVCmVrxSpM5IxhuHTAzkRm8jn93eglX95zwWzaJGtqA4Ls8ni55/h4Yc1SSil8p0idVZ6+4897Doaw6t9WtC9qYeKnI4ehcGDoWdPOHcOIvP8qWCllMqRIpMoFu88xhcrD9KjRTXu9VQXHZMn28rq336DceNs/0zt23smFqWUclGRqKM4Fp3AqJ+30rBKGT6624NdhwcFQefONmE0buyZGJRSKocK/R3FmbgkekxcwdmEFD64qy1lfPIwN8bE2JHmgoLs6//9z9ZNaJJQShUghTpRxCamcP+09cQmpvDeoNa0rV0hb1ZsDPzyCzRvbvtlWrHCvl+ypLaNUEoVOIU6UTw7aws7wmOYfE977upYO/sZcsOhQ9CnD9x5J1Stavtqeu65vFm3Ukq5QaFNFGv2n+SvXcfp2siP3lfVyLsVf/89rFwJEyfCxo22TkIppQqwQlmZnZZmuGfqesqXKs4XD3Rw/wr/+QcSE20vr//9Lzz4INTy6GB9SimVawrlHcUvQWEA3NGhFqVLuDEXnjxpe3bt1g3Gj7fv+fhoklBKFSqF7o4iMiaBMXN3UKdSaV7o2cw9KzEGpk+3dw/R0fDCC/DKK+5Zl3Kr5ORkwsLCSEhI8HQoSuWKkiVLUqtWLYoXL55ryyxUicIYw1MzN5OUksaU+zpQopibbpgWLrR3Etdeazvwa9XKPetRbhcWFoavry/16tXzbL9fSuUCYwynTp0iLCyM+vXr59pyC1XR06yNoaw7eJrhXevToma53F34uXOwerX9u3dvmDvXVlprkijQEhISqFy5siYJVSiICJUrV871O+RCkygSklN5Y8FuapYvyahbmuTuwv/4wyaEXr0gKsq2hejXTzvwKyQ0SajCxB3Hc6E5001YtJfYxBTeGNgq9yqww8Nte4jevW0l9fz5UCGPGu0ppVQ+USgSxYmzicxYd5ibm1flxmbVcmehkZHQogX8/ju88QZs3QrXX587y1bKibe3N23btqVVq1b07duXqKioXFluSEgIrdxQNDp27Fj8/f1p27Ytbdu25cUXX8z1daTbsmULCxcuvOTnmzdv5j//+Y/b1p8b3n77bRo1akTTpk1ZtGhRptMYY3j55Zdp0qQJzZs3Z9KkSQDMnTuX1q1b07ZtWwICAli1ahUAJ06coGfPnnm2DYWiMvvDv/aRmJLGYzc0uvKFhYeDv79tVf3663DbbdCw4ZUvV6lLKFWqFFu2bAFg6NChTJ48mZdfftnDUWXt2Wef5fnnn8/xfKmpqXh7uz708JYtWwgMDKR3796Zfv7WW28xZswYl5eXkpJCsWJ5d9rbtWsXM2fOZOfOnURERHDzzTezb9++i/bB9OnTCQ0NZc+ePXh5eRHpGH7gpptuol+/fogI27Zt46677mLPnj1UqVKFGjVqsHr1aq699lq3b0eBTxTL9kTy44Yj3NmhFu3rVLz8BUVHw5gx8PnnsG6d7f77qadyL1CV742bv5NdETG5uswWNcvxWt+WLk/fpUsXtm3bBkBsbCz9+/fnzJkzJCcn88Ybb9C/f39CQkLo1asXXbt2Zc2aNfj7+zN37lxKlSpFUFAQw4YNo3Tp0nTt2vX8chMSEnj00UcJDAykWLFifPjhh9xwww1Mnz6d3377jdTUVHbs2MGoUaNISkpixowZ+Pj4sHDhQipVquRS7EuXLuX5558nJSWFjh078tlnn+Hj40O9evUYNmwYixcv5oknnqBjx448/vjjnDhxgtKlS/Pll1/SrFkzfv75Z8aNG4e3tzfly5dnyZIlvPrqq8THx7Nq1SpGjx7N3XfffX59Z8+eZdu2bbRp0waADRs28MwzzxAfH0+pUqX4+uuvadq0KdOnT2fBggUkJCQQFxfH33//zfvvv89PP/1EYmIiAwcOZNy4cQAMGDCA0NBQEhISePrppxkxYoTL311m5s6dy+DBg/Hx8aF+/fo0atSIDRs20KVLlwum++yzz/jhhx/wctR7Vq1qx8spW7bs+Wni4uIuqH8YMGAA33//fZ4kigJf9PTWwt1UKlOCsf1c/2e8gDHw00+2A7/Jk2HkSL2DUB6RmprK0qVL6devH2Cfh58zZw6bNm1i2bJljBo1CmMMAMHBwTz++OPs3LmTChUq8OuvvwLw0EMPMWnSJNauXXvBsidPngzA9u3b+fHHHxk6dOj5J2N27NjBDz/8wIYNG3j55ZcpXbo0mzdvpkuXLnz77beZxjpx4sTzRU+LFi0iISGBBx98kFmzZrF9+3ZSUlL47LPPzk9fsmRJVq1axeDBgxkxYgSffPIJQUFBTJgwgcceewyA8ePHs2jRIrZu3cq8efMoUaIE48eP5+6772bLli0XJAmAwMDAC4rWmjVrxsqVK9m8eTPjx4/npZdeOv/Z2rVr+eabb/j7779ZvHgxwcHBbNiwgS1bthAUFMTKlSsBmDZtGkFBQQQGBjJp0iROnTp10bY/++yz57fd+eedd965aNrw8HBq1/63n7latWoRHh5+0XQHDhxg1qxZBAQE0KtXL4KDg89/NmfOHJo1a8Ztt93GtGnTzr8fEBDAP//8k+n3k9sK9B3Fsj2RBEfG8sj1DS6v+3Bj4Pbb7UBC7dvDvHkQEJD7gaoCISdX/rkpPj6etm3bEhISQocOHejRowdgy61feuklVq5ciZeXF+Hh4Rw/fhyA+vXr07ZtWwA6dOhASEgI0dHRREVFcb2jLu3+++/njz/+AGDVqlU8+eSTgD2h1q1bl3379gFwww034Ovri6+vL+XLl6dv374AXHXVVefvbjLKWPS0detW6tevT5Mm9onD9CK0Z555BuD8ST42NpY1a9Zw5513np83MTERgGuvvZYHH3yQu+66i9tvvz3b/Xb06FGqVKly/nV0dDRDhw4lODgYESE5Ofn8Zz169Dh/Z7R48WIWL15Mu3btzscUHBxMt27dmDRpEnPmzAEgNDSU4OBgKleufMF6J06cmG1s6dITu7PMnkpKTEykZMmSBAYGMnv2bIYNG3Y+CQwcOJCBAweycuVKXnnlFZYsWQLYu46IiAiXY7kSBTpRvLFgF1V8fXj4ugY5mzE5GYoXt4+5du0KN94Ijz0GOSg7VSq3pNdRREdH06dPHyZPnsxTTz3F999/z4kTJwgKCqJ48eLUq1fv/A8RwRQAAA1jSURBVF2Aj4/P+fm9vb2Jj4/HGHPJRyMzO2Glc16Wl5fX+ddeXl6kpKS4tA1ZLR+gTJkyAKSlpVGhQoXzdTLOpkyZwvr161mwYAFt27bNdBpnpUqVuqC9wCuvvMINN9zAnDlzCAkJoXv37hetPz3W0aNH88gjj1ywvOXLl7NkyRLWrl1L6dKl6d69e6btEZ599lmWLVt20fuDBw++qGK/Vq1ahIaGnn8dFhZGzZo1L5q3Vq1aDBo0CLCJ4aGHHrpomm7dunHgwAFOnjyJn58fCQkJlCpV6qLp3KHAFj0t2XWcAyfiGHl9Q/zK+mQ/Q7rly6F1a9tgDmDUKHjySU0SyuPKly/PpEmTmDBhAsnJyURHR1O1alWKFy/OsmXLOHz4cJbzV6hQgfLly59/Mub7778//1m3bt3Ov963bx9HjhyhadOmuRZ7s2bNCAkJYf/+/QDMmDHj/J2Ns3LlylG/fn1+/vlnwJ60t27dCtjil86dOzN+/Hj8/PwIDQ3F19eXs2fPZrrO5s2bn18f2DsKf39/wFYOX8qtt97KtGnTiI2NBWzxUGRkJNHR0VSsWJHSpUuzZ88e1q1bl+n8EydOZMuWLRf9ZPb0V79+/Zg5cyaJiYkcOnSI4OBgOnXqdNF0AwYM4O+//wZgxYoV5+/M9u/ffz4Jb9q0iaSkpPN3OPv27XPLU22Z+f/27j62qvqO4/j7A8IqjolO51SczPmAfaBVO1dkqWM6UjEqMwgqMFl0Bux8YmpgzLgno27zYU4YMhR0Qe0UN4jTMLMxdQaUbiJg1Y2hcU2MVscapuhs/e6P36/0Utt7T2vvU/t9JTe5D+ee8+03t/d7z++c8/0VZaEwM65ZtZlRI4Yln/+6pQUuuAAmTgydXkeOzG6QzvXBcccdR2VlJQ888AAzZsygsbGR6upqVq5cydixmXuXLV++nPr6esaPH7/Hr81LLrmE9vZ2KioqmD59OitWrNhjT+LjKikpYfny5ZxzzjlUVFQwZMgQ5syZ0+2yK1eu5K677qKyspKysjJWxx9tV199NRUVFZSXl1NbW0tlZSUTJ06kqamJqqoqGhoa9ljP2LFjaW1t3V1IrrnmGhYsWMCECRNob2/vMdZJkyZx/vnnM378eCoqKpg6dSo7d+6krq6OtrY2xo0bx7XXXktNTc3HzktZWRnTpk2jtLSUuro6Fi1atPuMp8mTJ+8eOpo/fz6rVq2ioqKCBQsWsGzZMgBWrVpFeXk5VVVV1NfX09DQsHuvcd26dZx++ukfO8YklGmXsdBUV1fbpbc/xHVrXuC7k8dycW2CA8/33w/19fDf/4ZGfgsXwogR2Q/WFbwXX3yRY489Nt9huD669dZbGTlyZMFfS5ENtbW1rF69mv32++jZnt19riX91cz6dBC2KPcoHm96g0NH7c1FX054bKKtLbTg2LQJrr/ei4RzA8TcuXP7dc+oWLS0tDBv3rxui0Q2FF2h+NCMv2x7i5O+8GmGDOmhp8k778D8+bB4cXg8c2aYt7q0NHeBOueyrqSkhFmzZuU7jJw78MADmTJlSs62V3SFYse74ZS3uvLPdr/AI49AWRncdBPE0/+Qws25bhTb8Ktz6WTj81x0haI1Foraow/c84Xm5nBNxBlnwD77hBbgt92WhwhdMSkpKeHtt9/2YuEGhI75KEpKSvp1vUV3HcX7be2ccPh+DBvapcZt3w5r18INN8C8eTB8eH4CdEVl9OjRNDc309LSku9QnOsXHTPc9aeiKxRtHxp1ZXHY6dlnYf16uPzyMG/1a69Bl6sonUtn2LBh/ToTmHMDUVaHniTVSXpZ0jZJH7kaRdInJDXE15+RNCbJek8+aK9wJXVNDdxySzh4DV4knHMuC7JWKCQNBRYBpwGlwHmSup52dCGww8yOBG4Fbsq03lG7dnLUySeGLq+XXQZbtoRjEs4557Iim3sUJwLbzGy7mf0PeAA4q8syZwH3xPsPAacowzx+o1vfRIcdBhs3hoPVn+rnubGdc87tIZvHKA4F/pXyuBn4Uk/LmFmbpFbg08BbqQtJuhjoaAz/vhobt3LCCVkJusgcQJdcDWKei06ei06ei059bu6VzULR3Z5B13MQkyyDmS0FlgJIauzrZegDjeeik+eik+eik+eik6TGvr43m0NPzcBhKY9HA12bp+9eRtJewL7Av7MYk3POuV7KZqHYCBwl6fOShgPnAmu6LLMGuCDenwr8yfzKJ+ecKyhZG3qKxxy+DawFhgJ3m9kLkn4INJrZGuAu4NeSthH2JM5NsOql2Yq5CHkuOnkuOnkuOnkuOvU5F0XXZtw551xuFV2vJ+ecc7nlhcI551xaBVsostX+oxglyMU8SU2SNkv6o6TD8xFnLmTKRcpyUyWZpAF7amSSXEiaFj8bL0i6L9cx5kqC/5HPSVon6bn4fzI5H3Fmm6S7Jb0paWsPr0vS7TFPmyUdn2jFZlZwN8LB738CRwDDgeeB0i7LXAIsiffPBRryHXceczERGBHvzx3MuYjLjQSeBDYA1fmOO4+fi6OA54D94uPP5DvuPOZiKTA33i8FXs133FnKRS1wPLC1h9cnA48RrmGrAZ5Jst5C3aPISvuPIpUxF2a2zszejQ83EK5ZGYiSfC4AfgT8BHgvl8HlWJJcfAtYZGY7AMzszRzHmCtJcmFAR7+fffnoNV0Dgpk9Sfpr0c4C7rVgAzBK0sGZ1luohaK79h+H9rSMmbUBHe0/BpokuUh1IeEXw0CUMReSjgMOM7NHchlYHiT5XBwNHC3paUkbJNXlLLrcSpKL7wMzJTUDjwKX5ia0gtPb7xOgcOej6Lf2HwNA4r9T0kygGjg5qxHlT9pcSBpC6EI8O1cB5VGSz8VehOGnrxD2Mp+SVG5m/8lybLmWJBfnASvM7GZJ4wnXb5Wb2YfZD6+g9Ol7s1D3KLz9R6ckuUDSqcBC4Ewzez9HseVaplyMBMqBP0t6lTAGu2aAHtBO+j+y2sw+MLNXgJcJhWOgSZKLC4HfAJjZeqCE0DBwsEn0fdJVoRYKb//RKWMu4nDLnYQiMVDHoSFDLsys1cwOMLMxZjaGcLzmTDPrczO0Apbkf+R3hBMdkHQAYShqe06jzI0kuXgNOAVA0rGEQjEY579dA3wjnv1UA7Sa2euZ3lSQQ0+WvfYfRSdhLn4KfBJ4MB7Pf83Mzsxb0FmSMBeDQsJcrAUmSWoC2oGrzezt/EWdHQlz8R3gV5KuJAy1zB6IPywl3U8YajwgHo+5DhgGYGZLCMdnJgPbgHeBbyZa7wDMlXPOuX5UqENPzjnnCoQXCuecc2l5oXDOOZeWFwrnnHNpeaFwzjmXlhcKV3AktUvalHIbk2bZMT11yuzlNv8cu48+H1teHNOHdcyR9I14f7akQ1JeWyaptJ/j3CipKsF7rpA04uNu2w1eXihcIdplZlUpt1dztN0ZZlZJaDb5096+2cyWmNm98eFs4JCU1y4ys6Z+ibIzzsUki/MKwAuF6zMvFK4oxD2HpyT9Ld5O6maZMknPxr2QzZKOis/PTHn+TklDM2zuSeDI+N5T4hwGW2Kv/0/E529U5xwgP4vPfV/SVZKmEnpurYzb3DvuCVRLmivpJykxz5b0iz7GuZ6Uhm6SfimpUWHuiR/E5y4jFKx1ktbF5yZJWh/z+KCkT2bYjhvkvFC4QrR3yrDTb+NzbwJfM7PjgenA7d28bw7wczOrInxRN8d2DdOBCfH5dmBGhu2fAWyRVAKsAKabWQWhk8FcSfsDXwfKzGwc8OPUN5vZQ0Aj4Zd/lZntSnn5IeDslMfTgYY+xllHaNPRYaGZVQPjgJMljTOz2wm9fCaa2cTYyuN7wKkxl43AvAzbcYNcQbbwcIPervhlmWoYcEcck28n9C3qaj2wUNJo4GEz+4ekU4ATgI2xvcnehKLTnZWSdgGvEtpQHwO8YmZ/j6/fA9QDdxDmulgm6fdA4pbmZtYiaXvss/OPuI2n43p7E+c+hHYVqTOUTZN0MeH/+mDCBD2bu7y3Jj7/dNzOcELenOuRFwpXLK4E3gAqCXvCH5mUyMzuk/QMcDqwVtJFhLbK95jZggTbmJHaQFBSt/ObxN5CJxKazJ0LfBv4ai/+lgZgGvAS8FszM4Vv7cRxEmZxuxFYBJwt6fPAVcAXzWyHpBWExnddCXjczM7rRbxukPOhJ1cs9gVej/MHzCL8mt6DpCOA7XG4ZQ1hCOaPwFRJn4nL7K/kc4q/BIyRdGR8PAt4Io7p72tmjxIOFHd35tFOQtvz7jwMTCHMkdAQn+tVnGb2AWEIqSYOW30KeAdolXQQcFoPsWwAJnT8TZJGSOpu78y53bxQuGKxGLhA0gbCsNM73SwzHdgqaRMwljDlYxPhC/UPkjYDjxOGZTIys/cI3TUflLQF+BBYQvjSfSSu7wnC3k5XK4AlHQezu6x3B9AEHG5mz8bneh1nPPZxM3CVmT1PmB/7BeBuwnBWh6XAY5LWmVkL4Yys++N2NhBy5VyPvHusc865tHyPwjnnXFpeKJxzzqXlhcI551xaXiicc86l5YXCOedcWl4onHPOpeWFwjnnXFr/B/h0s2x3A23GAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZIAAAF5CAYAAAC1Aay7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzs3XmcHFW5//HPNwlhX0KIigQSoogGVDQDREEEFQVZVVAQFdxww+W6gldlUa/gT6/KFRdkEZEdt6AiqCwKEskEwg4SIpEICISwryHP749zmtT09Mz0TFVNzyTf9+vVr+mu5anTPd39dJ06iyICMzOzoRrT6QKYmdno5kRiZmalOJGYmVkpTiRmZlaKE4mZmZXiRGJmZqU4kVhHSfqRpC93uhwrIklfk3SfpLs7dPzbJb1hGI6zuqTzJD0o6Zy6j2e9OZGMUvlD+rikRwq355eMuaOkRVWVsR0R8eGI+OpwHrMvkn4q6WudLkcVJG0MfAaYHhHP63R5arYP8FxgYkTsO1wHlXSQpMuG63gjmRPJ6LZHRKxVuN3ZycJIGtfJ45chaWyny1CxKcDiiLinnY1H8/+O9Fz/ERFLB7vjKH/eI0dE+DYKb8DtwBv6WDcT+BvwAHANsGNh3XuBm4CHgQXAh/LyNYHHgWXAI/n2fOCnwNcK++8ILGoqxxeAa4EngXF5v18A9wL/BD7Rz/N4Nn4jNvB54B7gLmBv4M3AP4D7gS8W9j0COBc4Kz+fq4CXF9a/BLgkvw43AHs2HfeHwO+BR4GDgaeBp/JzPy9vdyhwW45/I/CWQoyDgMuAbwFL8nPdtbB+feBk4M68/teFdbsD83LZ/ga8rLDuC8C/8zFvAV7fx2u3LvCz/DovBL5E+nH4hqb/5U9b7Nt4rb8A3A2cCkwAfpvjLcn3Jxf2uQT4KnB5LtuFwAaF9e/O5VgM/DeF9yiwKvDd/Frcme+vOpT/e9PzODL/z57Oz/X9+TX4Ui7LPfk1WjdvPxWIvN2/gL+08Zk5iPRZeTj/jw8gvbeeAJ7Jx32g098JHf0+6nQBfBviP66PRAJslD/Ib84fqJ3z40l5/W7ACwABrwUeA16Z1+1IIUnkZT9l4EQyD9gYWD0fcy7wFWA8MC1/CN/Ux/N4Nn6OvTTvuwrwQdKX2unA2sAW+cM7LW9/RP4C2Sdv/9n8QV8l3+YDX8zleF3+Iti8cNwHge1ymVdrfq55u31JiXEM8A5S0tkwrzsoH/+DwFjgI6QvSeX1vyMluQm5PK/Ny19J+oLbNu93YH4dVwU2B+4Anp+3nQq8oI/X7mfAb/JrM5X0pfv+vv6XTfs2Xutj8nFXByYCbwPWyDHPoWfyu4SUVF+Ut78EODqvm076Qt0hx/vfHL+RSI4CZgPPASaRvrS/OpT/e4vncgTw88Lj9+X//TRgLeCXwKmF1zPya7dmfh59fmbyNg+x/H2zIbBF4f9/Wae/C0bCreMF8G2I/7j0xfMI6RfUA40PPOkX5qlN214AHNhHnF8Dn8z3e3350F4ieV/h8bbAv5piHAac3Mfxn42fYz8OjM2P184f+m0L288F9s73jwBmF9aNIf2afU2+3Q2MKaw/AziicNyf9fdc+yjvPGCvfP8gYH5h3Rq5vM/LXzjLgAktYvyQ/CVaWHYLKbG/kJRk3gCs0k85xpLOAKcXln0IuKSv/2XT/juSfsmv1s82WwFLCo8vAb5UePxR4A/5/leAMwvr1szxG4nkNuDNhfVvAm4fyv+9RTmPoGci+TPw0cLjzUkJfxzLE8m0wvo+PzP5eTxASrCrN21zEE4kRISvkYxye0fEevm2d142BdhX0gONG7A96YsNSbtKmi3p/rzuzcAGJctxR+H+FOD5Tcf/IuliaDsWR8Qz+f7j+e9/CusfJ/3K7HXsiFhGqiJ5fr7dkZc1LCT9+mxV7pYkvUfSvMJz2ZKer9ezLaIi4rF8dy3SGdr9EbGkRdgpwGeaXqONSWch84FPkb4c75F0Zh+NKDYgnWkt7Of5DeTeiHii8FzXkPRjSQslPQT8BViv6fpRsQXYYyz/Xzyfnv+LR0m/6imsby5r8XkN9v/en1bHGkfP92Dze7blZyY/j3cAHwbukvQ7SS9usxwrDSeSFc8dpF9X6xVua0bE0ZJWJV27+Bbw3IhYj3SNQHnfVkNBP0r6pd3QqgVQcb87gH82HX/tiHhz6WfW2saNO5LGAJNZXg+/cV7WsAnp2kOrcvd6LGkK8BPgEFKLoPWA61n+evXnDmB9Sev1se7rTa/RGhFxBkBEnB4R25O+4IJU/dTsPtKv7Cn9PL+BND//z5B+vW8bEeuQqqmgved7Fz3/F2uQqsoa7mxR1roah7Q61lJ6Jqbm92zLzwxARFwQETuTfozdTHpPNMdYqTmRrHh+Duwh6U2SxkpaLTfrnUz6Bbsqqf55qaRdgTcW9v0PMFHSuoVl84A3S1pf0vNIv5b7cyXwkKQv5Pb9YyVtKWnryp5hTzMkvTW3vvkUqbpnNvB3UhL8vKRVJO0I7AGc2U+s/5Dq1RvWJH1Z3Asg6b2kM5IBRcRdwPnADyRNyGVofDH/BPiwpG2VrClpN0lrS9pc0uty0n+C9Ev8mRbxnwHOBr6e95sCfJr0/x+qtfPxHpC0PnD4IPY9F9hd0vaSxpOuiRS/X84AviRpkqQNSFVhZcranzOA/5K0qaS1gP8Bzoq+W3X1+ZmR9FxJe0pak/TeeoTl/4//AJPz812pOZGsYCLiDmAvUnXSvaRfW58jXSt4GPgE6QtoCfBOYFZh35tJH8IF+RT/+aTWPNeQroVcSLp43N/xnyF9YW9FuvB9H3ACqYVRHX5DqnpYQmo19NaIeDoingL2BHbNZfgB8J78HPtyIjA9P/dfR8SNwLeBK0hfGi8ltVhq17tJZw03k657fAogIrpJF5S/n8s9n1TfDinRH53LfDfp4vQX+4j/cVKyXEBqPXY6cNIgytfsu6SLz/eRkvEf2t0xIm4APpbLcBfpeRX7JH0N6Ca17ruO1MKurj47J5Het38hvQefIL1WLfX3mcm3z5DOcu4nXcf6aN71IlJrwLsl3VfHExktGq1LzEYdSUcAL4yId3W6LGYrM5+RmJlZKU4kZmZWiqu2zMysFJ+RmJlZKU4kZmZWykox8uUGG2wQU6dO7XQxzMxGlblz594XEZMG2m6lSCRTp06lu7u708UwMxtVJC0ceCtXbZmZWUlOJGZmVooTiZmZleJEYmZmpTiRmJlZKU4kZmZWihOJmTF34RKOu3g+cxe2mtDRrH8rRT8SM+vb3IVLOOCE2Ty1dBnjx43htA/MZMaUCZ0ulo0iPiMxW8nNXrCYp5YuY1nA00uXMXvB4oF3MitwIjFbyc2cNpHx48YwVrDKuDHMnDZx4J3MCly1ZbaSmzFlAqd9YCazFyxm5rSJrtayQav1jETSLpJukTRf0qEt1u8g6SpJSyXtU1i+k6R5hdsTkvbO634q6Z+FdVvV+RzMVgYzpkzgYzu90EnEhqS2MxJJY4HjgJ2BRcAcSbMi4sbCZv8CDgI+W9w3Ii4Gtspx1gfmAxcWNvlcRJxbV9nNzKx9dVZtbQPMj4gFAJLOBPYCnk0kEXF7Xresnzj7AOdHxGP1FdXMzIaqzqqtjYA7Co8X5WWDtR9wRtOyr0u6VtJ3JK061AKamVl5dSYStVg2qAniJW0IvBS4oLD4MODFwNbA+sAX+tj3YEndkrrvvffewRzWzMwGoc5EsgjYuPB4MnDnIGO8HfhVRDzdWBARd0XyJHAyqQqtl4g4PiK6IqJr0qQBJ/gyM7MhqjORzAE2k7SppPGkKqpZg4yxP03VWvksBUkC9gaur6CstpLwUCBm1avtYntELJV0CKlaaixwUkTcIOkooDsiZknaGvgVMAHYQ9KREbEFgKSppDOaS5tCnyZpEqnqbB7w4bqeg61YPBSIWT1q7ZAYEb8Hft+07CuF+3NIVV6t9r2dFhfnI+J11ZbSVhathgJxIjErz0Ok2ErDQ4GY1cNDpNhKw0OBmNXDicRWKjOmTHACMauYq7bMzKwUJxIzMyvFicTMzEpxIjEzs1KcSMzMrBQnEjMzK8WJxMzMSnEiMTOzUpxIzMysFCcSMzMrxYnEzMxKcSIxM7NSnEjMzKwUJxIzMyvFicTMzEqpNZFI2kXSLZLmSzq0xfodJF0laamkfZrWPSNpXr7NKizfVNLfJd0q6SxJ4+t8DmZm1r/aEomkscBxwK7AdGB/SdObNvsXcBBweosQj0fEVvm2Z2H5McB3ImIzYAnw/soLb2ZmbavzjGQbYH5ELIiIp4Azgb2KG0TE7RFxLbCsnYCSBLwOODcvOgXYu7oim5nZYNWZSDYC7ig8XpSXtWs1Sd2SZktqJIuJwAMRsXSIMc3MrGJ1ztmuFstiEPtvEhF3SpoGXCTpOuChdmNKOhg4GGCTTTYZxGHNzGww6jwjWQRsXHg8Gbiz3Z0j4s78dwFwCfAK4D5gPUmNBNhnzIg4PiK6IqJr0qRJgy+9mZm1pc5EMgfYLLeyGg/sB8waYB8AJE2QtGq+vwGwHXBjRARwMdBo4XUg8JvKS25mZm2rLZHk6xiHABcANwFnR8QNko6StCeApK0lLQL2BX4s6Ya8+0uAbknXkBLH0RFxY173BeDTkuaTrpmcWNdzMDOzgSn9yF+xdXV1RXd3d6eLYWY2qkiaGxFdA23nnu1mZlaKE4mZmZXiRGJmZqU4kZiZWSlOJGZmVooTiZmZleJEYmZmpTiRmJlZKU4kZmZWihOJmZmV4kRiZmalOJGYmVkpTiRmZlaKE4mZmZXiRGJmZqU4kZiZWSlOJGZmVooTiZmZleJEYmZmpdSaSCTtIukWSfMlHdpi/Q6SrpK0VNI+heVbSbpC0g2SrpX0jsK6n0r6p6R5+bZVnc/BzMz6N66uwJLGAscBOwOLgDmSZkXEjYXN/gUcBHy2affHgPdExK2Sng/MlXRBRDyQ138uIs6tq+xmZta+2hIJsA0wPyIWAEg6E9gLeDaRRMTted2y4o4R8Y/C/Tsl3QNMAh7AzMxGlDqrtjYC7ig8XpSXDYqkbYDxwG2FxV/PVV7fkbRquWKamVkZdSYStVgWgwogbQicCrw3IhpnLYcBLwa2BtYHvtDHvgdL6pbUfe+99w7msMNi7sIlHHfxfOYuXNLpopiZlVJn1dYiYOPC48nAne3uLGkd4HfAlyJidmN5RNyV7z4p6WR6X19pbHc8cDxAV1fXoBJY3eYuXMIBJ8zmqaXLGD9uDKd9YCYzpkzodLHMzIakzjOSOcBmkjaVNB7YD5jVzo55+18BP4uIc5rWbZj/CtgbuL7SUg+D2QsW89TSZSwLeHrpMmYvWNzpIpmZDVltiSQilgKHABcANwFnR8QNko6StCeApK0lLQL2BX4s6Ya8+9uBHYCDWjTzPU3SdcB1wAbA1+p6DnWZOW0i48eNYaxglXFjmDltYqeLZGY2ZIoYUbU+tejq6oru7u5OF6OHuQuXMHvBYmZOm+hqLTMbkSTNjYiugbar8xqJ9WPGlAlOICsg/0CwlZETiVlF3IjCVlYea8usIm5EYSsrJxKzirgRha2sXLVlVpEZUyZw2gdm+hqJrXTaTiSSpgCbRcSfJK0OjIuIh+srmtno40YUtjJqq2pL0geBc4Ef50WTgV/XVSgzMxs92r1G8jFgO+AhgIi4FXhOXYUyM7PRo91E8mREPNV4IGkcgxyA0czMVkztJpJLJX0RWF3SzsA5wHn1FcvMzEaLdhPJocC9pPGtPgT8HvhSXYUyM7PRo91WW6sDJ0XET+DZaXRXJ02Ja2bWMR6WpvPaTSR/Bt4APJIfrw5cCLy6jkKZmbXDw9KMDO1Wba0WEY0kQr6/Rj1FMjNrj4elGRnaTSSPSnpl44GkGcDj9RTJzKw9HpZmZGi3autTwDmSGlPlbgi8o54imZm1x8PSjAxtJZKImCPpxcDmgICbI+LpWktmZtYGD0vTeYMZtHFrYGre5xWSiIif1VIqMzMbNdoda+tU4FvA9qSEsjUw4PSLknaRdIuk+ZIObbF+B0lXSVoqaZ+mdQdKujXfDiwsnyHpuhzzWElq5zmYmVk92j0j6QKmxyAmeM99TY4DdgYWAXMkzYqIGwub/Qs4CPhs077rA4fn4wYwN++7BPghcDAwm9Qxchfg/HbLZWZm1Wq31db1wPMGGXsbYH5ELMjjdJ0J7FXcICJuj4hrgWVN+74J+GNE3J+Txx+BXSRtCKwTEVfkpPYzYO9BlsvMzCrU7hnJBsCNkq4EnmwsjIg9+9lnI+COwuNFwLZtHq/Vvhvl26IWy83MrEPaTSRHDCF2q2sX7VaN9bVv2zElHUyqAmOTTTZp87BmZjZY7Tb/vXQIsRcBGxceTwbu7GPbVvvu2LTvJXn55HZiRsTxwPEAXV1dHvLezKwm7bbamilpjqRHJD0l6RlJDw2w2xxgM0mbShoP7AfMarNcFwBvlDRB0gTgjcAFEXEX8HAuj4D3AL9pM6aZmdWg3Yvt3wf2B24lDdj4gbysTxGxFDiElBRuAs6OiBskHSVpTwBJW0taBOwL/FjSDXnf+4GvkpLRHOCovAzgI8AJwHzgNtxiy8yso9ROi15J3RHRJenaiHhZXva3iBgVo/92dXVFd3d3p4thZjaqSJobEQP2GWz3YvtjuXpqnqRvAncBa5YpoJmZrRjardp6d972EOBR0kX0t9ZVKDMzGz3aTSR7R8QTEfFQRBwZEZ8Gdq+zYGZmNjq0m0gObLHsoArLMSLNXbiE4y6ez9yFSzpdFDOzEavfaySS9gfeCUyTVGy6uzawQk9F5ik8zczaM9DF9r+RLqxvAHy7sPxh4Nq6CjUStJrC04nEzKy3fhNJRCzM/TweHWLv9lGrMYXn00uXeQpPM7N+DNj8NyKekfSYpHUj4sHhKNRI4Ck8zcza024/kieA6yT9kdT8F4CI+EQtpRohPIWnmdnA2k0kv8s3M+uQuQuX+AzZRqR2R/89Jfdsf1FedEtEPF1fscysyK0IbSRrd/TfHUkDNh4H/AD4h6QdaiyXmRW0akVoNlK0W7X1beCNEXELgKQXAWcAM+oqmJkt51aENpK1m0hWaSQRgIj4h6RVaiqTmTVxK0IbydpNJN2STgROzY8PAObWUyQza8WtCG2kajeRfAT4GPAJ0rzpfyFdKzEzs5Vcu622npT0feDPwDJSq62nai2ZmZmNCm0lEkm7AT8iTW0rYFNJH4oIT3NrZraSa3cY+W8DO0XEjhHxWmAn4DsD7SRpF0m3SJov6dAW61eVdFZe/3dJU/PyAyTNK9yWSdoqr7skx2yse067T9bMzKrXbiK5JyLmFx4vAO7pbwdJY0n9TnYFpgP7S5retNn7gSUR8UJSYjoGICJOi4itImIr0uyMt0fEvMJ+BzTWR0S/5TAzs3q1m0hukPR7SQdJOhA4D5gj6a2S+ppydxtgfkQsyNdTzgT2atpmL+CUfP9c4PWS1LTN/qQ+K2ZmNgK1m0hWA/4DvBbYEbgXWB/Yg76n3N0IuKPweFFe1nKbiFgKPAg097R6B70Tycm5WuvLLRKPmZkNo3Zbbb13CLFbfcHHYLaRtC3wWERcX1h/QET8W9LawC9IVV8/63Vw6WDgYIBNNtlkkEU3M7N2tdtqa1Pg48DU4j4RsWc/uy0CNi48ngzc2cc2iySNA9YF7i+s34+ms5GI+Hf++7Ck00lVaL0SSUQcDxwP0NXV1ZzAzMysIu12SPw1cCLp2siyNveZA2yWk9C/SUnhnU3bzAIOBK4A9gEuiogAkDQG2Bd4dnDInGzWi4j78hAtuwN/arM8ZmZWg7YntoqIYwcTOCKWSjoEuAAYC5wUETdIOgrojohZpOR0qqT5pDOR/QohdgAWRcSCwrJVgQtyEhlLSiI/GUy5zMysWsonAP1vJL0T2Ay4EHiysTwirqqvaNXp6uqK7u7uThfDzGxUkTQ3IroG2q7dM5KXki5qv47lVVuRH5uZ2Uqs3UTyFmCax9cyM7Nm7fYjuQZYr86CmJnZ6NTuGclzgZslzaHnNZL+mv+amdlKoN1EcnitpTAzs1Gr3Z7tl9ZdEDMzG536TSSSHqb3sCaQhjaJiFinllKZmdmo0W8iiYi1h6sgZmY2OrXbasvMzKwlJxIzMyvFicTMzEpxIjEzs1KcSMzMrBQnEjMzK8WJxMzMSnEiMTOzUpxIzMysFCcSMzMrpdZEImkXSbdImi/p0BbrV5V0Vl7/d0lT8/Kpkh6XNC/fflTYZ4ak6/I+x0pSnc/BzMz6V1sikTQWOA7YFZgO7C9petNm7weWRMQLge8AxxTW3RYRW+XbhwvLfwgcTJpDfjNgl7qeg5mZDazOM5JtgPkRsSBP0XsmsFfTNnsBp+T75wKv7+8MQ9KGwDoRcUVEBPAzYO/qi25mZu2qM5FsBNxReLwoL2u5TUQsBR4EJuZ1m0q6WtKlkl5T2H7RADHNzGwYtTtD4lC0OrNontukr23uAjaJiMWSZgC/lrRFmzFTYOlgUhUYm2yySduFNjOzwanzjGQRsHHh8WTgzr62kTQOWBe4PyKejIjFABExF7gNeFHefvIAMcn7HR8RXRHRNWnSpAqejpmZtVJnIpkDbCZpU0njgf2AWU3bzAIOzPf3AS6KiJA0KV+sR9I00kX1BRFxF/CwpJn5Wsp7gN/U+BzMzGwAtVVtRcRSSYcAFwBjgZMi4gZJRwHdETELOBE4VdJ84H5SsgHYAThK0lLgGeDDEXF/XvcR4KfA6sD5+WZmZh2i1PhpxdbV1RXd3d2dLoaZ2agiaW5EdA20nXu2m5lZKU4kZmZWihOJmZmV4kRiZmalOJGYmVkpTiRmZlaKE4mZmZXiRGJmZqU4kZiZWSlOJGZmVooTiZmZleJEYmZmpTiRmJlZKU4kZmZWihOJmZmV4kRiZmalOJGYmfVj7sIlHHfxfOYuXNLpooxYtU21a501d+ESZi9YzMxpE5kxZUKni2M2Ks1duIQDTpjNU0uXMX7cGE77wEx/nlpwIlkB1f3md5KywRqt75nZCxbz1NJlLAt4eukyZi9YPKrKP1xqTSSSdgG+B4wFToiIo5vWrwr8DJgBLAbeERG3S9oZOBoYDzwFfC4iLsr7XAJsCDyew7wxIu6p83mMNnW++f0LzQZrNL9nZk6byPhxY3h66TJWGTeGmdMmdrpII1JtiUTSWOA4YGdgETBH0qyIuLGw2fuBJRHxQkn7AccA7wDuA/aIiDslbQlcAGxU2O+AiOiuq+yjXZ1vfv9Cs8Eaze+ZGVMmcNoHZo7Ks6nhVOcZyTbA/IhYACDpTGAvoJhI9gKOyPfPBb4vSRFxdWGbG4DVJK0aEU/WWN4VRp1vfv9Cs8Ea7e+ZGVMmOIEMoM5EshFwR+HxImDbvraJiKWSHgQmks5IGt4GXN2URE6W9AzwC+BrERHNB5d0MHAwwCabbFLyqYw+db35/QvNBsvvmRVfnYlELZY1f+H3u42kLUjVXW8srD8gIv4taW1SInk36TpLzyARxwPHA3R1dfVKNDZ0/oVmg+X3zIqtzn4ki4CNC48nA3f2tY2kccC6wP358WTgV8B7IuK2xg4R8e/892HgdFIVmpmZdUidiWQOsJmkTSWNB/YDZjVtMws4MN/fB7goIkLSesDvgMMi4vLGxpLGSdog318F2B24vsbnYGZmA6gtkUTEUuAQUourm4CzI+IGSUdJ2jNvdiIwUdJ84NPAoXn5IcALgS9LmpdvzwFWBS6QdC0wD/g38JO6noOZmQ1MLa5Tr3C6urqiu9uthc3MBkPS3IjoGmg7j7VlZmalOJGYmVkpTiRmZlaKE4mZmZXiRGJmZqU4kZiZWSlOJGZmVooTiZmZleJEYmZmpTiRmJlZKU4kZmZWihOJmZmV4kRiZmalOJGYmVkpTiRmZlaKE4mZmZXiRGIjztyFSzju4vnMXbik00Uxq1Wd7/Xh/ByNq/0IZoMwd+ESDjhhNk8tXcb4cWM47QMzmTFlQqeLZVa5Ot/rw/05qvWMRNIukm6RNF/SoS3WryrprLz+75KmFtYdlpffIulN7ca00W32gsU8tXQZywKeXrqM2QsWd7pIZrWo870+3J+j2hKJpLHAccCuwHRgf0nTmzZ7P7AkIl4IfAc4Ju87HdgP2ALYBfiBpLFtxrRRbOa0iYwfN4axglXGjWHmtImdLpJZLep8rw/356jOqq1tgPkRsQBA0pnAXsCNhW32Ao7I988Fvi9JefmZEfEk8E9J83M82ohpo9iMKRM47QMzmb1gMTOnTXS1lq2w6nyvD/fnqM5EshFwR+HxImDbvraJiKWSHgQm5uWzm/bdKN8fKGa1Tt6t97It9oZtPghPPQan7dt7/VbvhFccAI8uhrPf03v91u+DLd8GDy6CX36o9/pXHwKb7wr33Qrnfar3+h0+Cy/YCe66Fv5wWO/1r/8KbLIt/Ovv8Oejeq/f5Ruw4cvgtovhL9/qvX6P78IGm8Et58Pfvt97/Vt/DOtOhut/AXNO6r3+7T+DNSfC1afBvNN7rz/gHBi/Blz5E7jh171Wz3jv79Ib//Jj4aILeq5cZTV41y/S/Uu/CQsu7bl+jQnwjp+n+386Au6Y03P9Os+Ht/0k3T//ULj7up7rJ74A9jw23Z/1CVh8W8/1z3sp7Hp0uv+LD8JDd/Zcv/HW8IYj0v2z3gWPNV3onPZaeO3n0/2fvw2efqLn+he9Cbb7RLrv917v9TW/93jv79Lfy4+Ff9T/3psBzAC4ncrfezMeW8KMxvOpWZ3XSNRiWbS5zWCX9z64dLCkbknd9957b78FNTOzoVNEy+/h8oGlVwFHRMSb8uPDACLiG4VtLsjbXCFpHHA3MAk4tLhtY7u8W78xW+nq6oru7u7qnpyZ2UpA0tyI6BpouzrPSOYAm0naVNJ40sXzWU3bzAIOzPf3AS6KlNlmAfvlVl2bApsBV7YZ08zMhlFt10jyNY9DgAuAscBJEXGDpKOA7oiYBZwInJovpt9PSgzk7c4mXURfCnwsIp4BaBWzrudgZmYDq61qayRx1ZaZ2eCNhKotMzNbCTiRmJlZKU4kZmZWihOJmZmV4kQrnfCMAAAgAElEQVRiZmalrBSttiTdCywc4u4bAPdVWJzhil13fJe9M/FHa+y647vs9cSeEhGTBtpopUgkZUjqbqf520iLXXd8l70z8Udr7Lrju+zDH7vIVVtmZlaKE4mZmZXiRDKw40dp7Lrju+ydiT9aY9cd32Uf/tjP8jUSMzMrxWckZmZWihOJmZmV4kRiKwVJn2xn2RBjb9fOspGoztdltJL05/z3mE6XZbTwNZI+SNoT2CE/vDQizqso7iv7Wx8RV5WMfx2tpx9WCh8vKxF7LLAbMJXCXDYR8b9DjdkUf1Pg4y3i71lB7Ksi4pVNy66OiFfUFLvXsiHGPjUi3j3QshLxK39dJP0ffUyBDRARnxhq7GGKfyPwEeBHwDtpmuJ7qJ9RSZ+PiG/2Vf6y5c7HGAtcEBFvKBtrMGqb2Go0k/QNYBvgtLzoE5JeHRGHVRD+B8ArgWtJb9CXkmZ/fJr05npdyfjn57+n5r8HAI8Bp5SMC3Ae8ARwHbCsgnjNfk2a7Oy8quJL2p/0ZbCppOJsmmsDi0vGfhXwamCSpE8XVq1DmnitCls0HXMsMKNs0DpfF6Ax+c92wHTgrPx4X2BuydjDEf8rpOm+JwPfpmciKfMZvTH/rW1ypIh4RtJjktaNiAfrOk4zJ5LWdgO2iohlAJJOAa4GqkgktwMfjIjrcuwtgc9GxEEVxAbYLiKK1SqHSro8Io6qIPbkMmc0bXgiIo6tOObfgLtIQ0V8u7D8YVIyL2M8sBbpc7R2YflDpKmjh0zSYcAXgdUlPdRYDDxFNU06a3tdIuIUAEkHATtFxNP58Y+AC8vEHo74wF0Rsaukr1T0uWl4B/BbYL2I+F6FcZs9AVwn6Y/Ao42FVZzx9MVVWy1IuhbYMSLuz4/XBy6p4ktU0ryI2GqgZWXiA4dExGX58auBH1QRP9cZ/zkiqviwtor/TmAz0pfBk43lZav76iZpSkQMdSy3gWJ/o6Iz4WEn6RbgVYXP0QRgdkRsPpLj51kBZ1RVPVmIeyOwKzAL2JHeVWb3V3ScA1stbyTgOviMpLVvAFdLupj0z96Bas5GAG6SdALwc9Jp8ruAmyqKDfB+4CRJ6+b4DwLvqyj2bOBXksaQquIa113WqSj+S4F3k6oOGlVbVVT3IemtwDHAc0jlrrLsq0o6nt7XdkqXOyIOk7QRMKUp9l/KxobaX5ejWf45AngtcEQFceuO/7Skk4GNJPU6Qy7xy/5HwB+AaaQquOYqs2lDjNtDnQmjLz4j6YOkDYGtSf/sv0fE3RXFXY10Ia9xIf8vwA8j4okq4heOsw7p/1tZPamkBcDewHVRwxtH0s3AyyLiqRpizwf2iIgqk3Yj9jWkL4m5wDON5RFRur5e0tHAfqT69UbsqKIBQo5f2+uS4z8P2DY/rOxzVGd8SRsAbyAl2K80ry/7RS3phxHxkTIxBoi/GenH8HRgtcbyiKgkUbU8phPJcpJeHBE399WyqqoqFkmrA5tExC1VxGuK/Vzgf4Dn53re6aTT/xMriH0BsGvj2lHVJJ0FfDwi7qkh9uVN146qjD03IkpfAO8j9i2k5PrkgBsPLX5tr0sfx3txRNxcMsbLIqLs9a12jvPyiLimwnjrRMRDuaq8lwqrti4DDge+A+wBvJf0XX94FfFbHtOJZDlJx0fEwYVT5aKooqoiNyv+f8D4iNhU0lbAURX+wjwfOBn474h4uaRxwNUR8dIKYv+UdPp9Pj2vYVTV/PcS4GXAnKb4VTT//R7wPFLLsGLsX1YQ+wjgHuBXTbFLfzHk/+e+EfFI2Vh9xK/tdenjeP+KiE1KxngG+CdwBnBGRNw4wC6DjV9LM11Jv42I3SX9M8ftUbVV1RlD4RrPdY3PvaS/RsRrqojfiq+RFETEwfnurs1VTblKqgqHk5oWX5KPOU/S1IpiA2wQEWfnVj9ExNL8wavCP/NtfL5VrbZfTKQmuY8BbywsC6CKL8zGxc3PNcWu4ovhMWCeUie54hd9VS1wKn9dWl1XaKwC1htq3IJrSdfS9gdmSXqUlFTOjIjbK4jfqOartJluROye715GqtL+a9mzsz48ka9j3irpEODfpGtgtfEZSQt9dNKqqoPZ3yNi22KnL0nXVtWsNv+qfxvwx4h4paSZwDER8doq4telUx2pRrpOtMApS9LDwGcoJL6Cb0fEBiXj9/gsStqGdB1pX+COiHh1mfh1k/Q6YHvgNaQfG1eTkkolTYIlbU1KhusBXwXWBb4ZEbOriN+Kz0gK8oW7jUht91/B8lPPdYA1KjrM9bmZ69h8UewTpDb9Vfk0qXnhCyRdDkyiZJ+Ghlzl1+pUv4rWSbV2pMqtcFqVvXSLNknvabU8In5WNnbdCaOm12UOcH1E9Hpf52rAspqbzV4JXCnpMyxvxDL04NJ59N9zvlRVa0RcJOlSUmOenYAPA1sClSSSiJiT7z5Cuj5SOyeSnt4EHETq0Vqs93+Y1DmsCh8H/pv0a+104ALga1UEzqezq5GaQW5O+sDdErnDVgU+W7i/GunMZ2lFsaHejlS/LdxfDXgLcGcFcSF9IRRjvx64CiidSAr16T1U2AKnjtdlH9L/speI2LRkbEjXGFvFDuDSCuJ/K/99K+n60c/z4/1JHYpLydWUawJXAH8Ftq6igUndCbDfY7tqqzdJb4uIX9QQdyxwdER8bsCNh36MKyLiVXXFb3G8S6uqNhvOapycdP9UxdlUi9jrAqdW1EhgYuHhaqTqm/Ujolez1CpU+bpI+mRzdU2rZSM4/l8iYoeBlg0h7ndIw9w8CVxOul5yRUQ8XjJuv5/DiKgiybY+thNJa5J2I41zVGyHXXq4BEkX1fHlVYh/JOli5C+r7uvR1GxxDOnDcGxU1FM5H2M88KL8sMqzqebjbA78LiJeWEPsVYBrI+IlVcfO8S+LiO1ril3Z69LHtcZKBsocpvg3AbtFxIL8eFPg91X9XyWtRap6+izwvIhYtYq4neCqrRaUxuxZg1R/eQLpVP3KisJfrTRI3jn0rL6pqrnlp0mnzUslPUG1PZXnsrzZ4lJSC673VxAXAEk7kgaXvD0fY2NJB0YFvbjzBeBG2QO4G/hC2bg5drFKYSzwEuDsimIXvyjHAF30HNerbPzKXxfVOyBk7fEL/gu4RKkjLqSRCw7ue/P25JZUryH9EFsInESq4qrEMFSH9uJE0tqrI+JluTXVkZK+TTXNRAHWJ73Zi2clpZuhStouIi4HJjU3Xa5KRfXb/fk28MbIHTUlvYjUrLN0Z7+IqOzLt4VvFe4vBRZGxKKKYhcHVFxKSrJvryh2Xa9LnQNlDkd8ACLiD7lBzIvzopuLHUMl7RwRfxxC6NVJ12DnRkSV1xgbugr3n60OreE4z3LVVguSroyIbSTNJl1wW0xqhbJZiZjHRMQXJO0bEedUVtjl8WsZaK7FcV5N7zGlSl9UzrF7NYOuuGl0cY6ZSyLit/1tP8jYz2X5Rfcrq7h4OlzqfF1WZHV/1qpUZ3Uo+IykL+dJWo/UOuQq0hnDT0rGfLOkL5EGf6w8kbB8oLnJrTqEVdHySdKpwAuAeRTGfaKC1klZt6QT6TmXShXzSzTGrNqa5XPMfDKfxZUejFPS20nvlUtIVUT/J+lzEXFuBbHXJXXUfHaSNdJICJU0ka75dalzQMja47dThGE6zqDUXR3a8pg+I+kpt1qZ2WgDL2lVYLWyH1xJ/49Uv7omqSfxs6uo4M2vmgeay8e4CZhe9UX8QvxVgY+ROmuJ1JrlB1HBOFNKUwMU55gZSxo6poqpAa4Bdm6chUiaRGr59PIKYv8CuJ7lE5O9G3h5RLy1bOwcv87Xpe4BIWuN38bxR+QZiXoO8dSoDv1W1DC2X4PPSJpExLJ8TeRV+fGTtO6hO9i4nwM+J+k3EbFX2Xgt4t8HnCnppuhnoDlJh0XEN4Z4mOtJ7ervGuL+LUn6c0S8nvRL+wv07MNTpfWAxvhX61YYd0xTVdZi0i/BKrwgIt5WeHyk0pwzVarrdflPzV/ydccflSJip+E+phNJaxdKehs1NKEdKImU7QfSXxLJ9iUNMT0UGwA3SrqSagdV3DC3gd9T0pn07rlcxajLdc4x8welkZHPyI/fwfIpj8t6XNL2sXyisu2AUv0NmtT5unQrjehc14CQdccfyO3DdJy2qOd0z71ERYOrtjy2q7Z6y00i1ySdFlbdhHagY1fWDr7q+H11eCrb0UnSPqRmxNvTe6C8qKrfjWqaYybHfiuFKrmI+FVFcV9OugbVOFNYAhzUxg+GwRyjrrl3Tm6xOKKCYWmGI34+Rm2NS6omqd9BTyPiyNqO7UQyeJK2iIgbaopdd6ur2uKXPZuS9OWI+GrFZXoTsHbzhW9JBwD3DLH5ZiPGC4Hn5mbXxeU7AP+OiNuGGrvFsdYBiIiHBtq2zXi1vS4rir4al1TRcGVF46qtoTkVGHEX2dpUZ0uTskPt70garfRZhesnQ3UkaXKfZn8mzR9S5gvzu7Qeg+2xvK7VcduSqykejDwhWSOBSPo4MDYivjvU2Fmdrwvw7BlDLQNlDkd8Umun2hqXVK1Va82iOhOgE8nQ1PllXHeTwjqaHjcM6QOnNNfLmsAGkiZAj1GXn1+yTGtExL3NCyPibklrlow9NVrM1BcR3So/x8z7aP1j5XjS6LplE0mdr0tDnQNlDkf8WhqX1KiSpvJD4UQyNHX+Qnn3UHZSH7O5NTR+jUTE/wyxXHX6EPApUtKYy/JE8hBwXMnYq0ka19yDWGk8rNXLxu5nXdnYES3mro+IJyVV8WOjztcFgGga+FTSGcCfqohdZ3wtH/JmbeppXFKL5ib+ktaMiEf72r5KTiTDbKBOVBFx/RBDNy5SbwdMB87Kj/dl+H6pDOkLLtJord+T9PGI+L+Ky/RL4CeSDml8qPIv7mMpP+zNHEkfjIgenVUlvZ8KXnNJz42I/zQvKxs3q/N16ctmQKlpdocp/rcG3mTkkvQq4ERgLWCT3GDjQxHx0dqOOUqq/0YUSbMjYuYQ9627k9bFpPGqns6PVwEuHI625ZK2LJEIn41BSoTFUZeH3EpGac76rwEfIA2QB+nL5kTgy1FidOH8pf4r4CmWJ44u0jTEbynT+klpsqxPkGYabDR/ngF8EziubAfTFq+LgI2p4HUpHKPVgJCHNZ9JjOD4mwJ3RR67TtLqpMYVt1cRvy6S/k4aaHZWLJ+F9fqI2LK2YzqR9JZ7E58EnN/o8Vth7MsjYrsqYzbFvwV4VUTcnx9PAGZHBUO9D8OQF4eTLrhPB34P7ApcFhGlZ3iUtAapBQ7A/Ih4XNKqUU2v+Z1IM9wB3BARF5WNmePuChyaYwdwA2k+m6r6qDS+HBtDxs+PknNirEgkdZMGcH0qPx4PXB4RW/e/Z2ep9XTe10QFIy30xVVbrf2QNE/AsZLOAX4aETdXFLvuTlRHs7yDGaTZEo+oKPY3qXdIin2Al5OG6Hhv/sV/QkWxL2vR7PkKqml9txNpGPC/VVknHRHnKw0g2mNodEmbRsQ/y8TOPwqabda4/FLV+1E1DwhZc/xxxetUEfFUTiYj3R25/0vk8n6CNId7bZxIWoiIPwF/Uhowb3/gj5LuIA3c+POSp/3rkJqHvrF4SCqql46IkyWdD2ybFx1aYce7uoekeDzSEDVLc7+Je4BScyhIeh6wEbC6pFfQs0XYGqVKu9ztpPfJsbm65a+kTom/qSD2LEm7Fpr/TifNdVK2mqK/psmVvB9V44CQwxEfuFfSnhExKx9vL+C+imLX6cOk+d83AhYBFwK1XR8BV231SWmK03eRWlHdSXqzbg+8NCJ27GDRWlLPET97iQqGGZH0PVJzyFrOpiT9gNQvYz/StYFHgHkR8d4SMQ8EDiJdu5hDzxZhp1R4JthIWm8nzXg3ISqY60Npps7PA7sBm5N6uR8QEVWPt1U51Tgg5DDFfwHpc79RXnQH8O4qO5rWQdIpwKciYkl+PAH4doX9a3of04mkN0m/JE1mcyqpWuuuwrruiOjqc+e+Y34+Ir7ZVzPdsp2F1HPEzxbhK5mDu/YhKQrHmgqs06qfxhBijQH2j4jTBtx4aPFPIF3X+Q/pbOQy4KrmprUl4u9NSiZrA2+NiFsriFn7uEz5i37HwvW69UnVT1UmktriF46zFum78uEq49ZFLYZBarWsSq7aapK/dOZFH8N0DyWJZI0qoeaxpCoxHK2yypwZ9Ke/sylJryx7NpWryz7E8iqQqk0kTbH7AGkU3fvKJpEWPzjWARYAH5dURS/lWuenyOocELL2+GqaC0ZSpXPB1GiMpAmFM5L1qfm73mckLajkmFGdlFsnfRrYJCIOVpoqdPMqLkJKmgz8H6mvSpB+eX8ySk4r23Q2VXxDNlqFVXE29WXSqLlnAc9eEG/8mq2CpJcAbyLN9T02IiaXiHVgf+vLNv8dLqpxoMy646vmuWDqkpuOHwacS/o8vR34ekSc2u+OZY7pRNKbpCNJcz9XPoy80qRHX6B3X4mqRrg9i9Sn4T0RsWVu3nlFRGxVQew/AqezfAbDd5Hq63cuGzvHX510UXB70gfgr8APo4I56CW1auUUEVHqYn6OvTvwGtIv1wmk1mB/jYiTKoi9JvBERDyTH48FVo2Ix/rfs+34q5FGXt6Cnu/HIVdXquYBIeuOX4g3r/lz02rZSJQbZbyOlGD/HBE31nrAiPCt6QY8DCwjdTR7KD9+qKLYF5I+uDeRmuaeBBxTYdm789+rC8uuqSj2vHaWlYh/Nqm57075djxwdqffD22U+zjSHCTPryH2bGCtwuO1SM2Mq4p/DmmgzNuAA/P783sVlHlSi+XPI/2oqeI1qS1+Id4VwPaFx9tVGX9FuvkaSQtRQWubfkyMiBMlfTLSPB6X5rrXqjyVf9kHPNvypHSnu+w+Se9i+QRO+5NmA6zK5tGz09TFStPYVkIV95ovxPiYpCk59p359R8X1VycXS0iHikc65FcfVmVF0bEvpL2iohTJJ0OXFAyZt0DQg7HgJMAHwFOyddKRLr+1W+V48qqqulAVyiS/tzOsiFq9EG5S9JuuW/DkOvSWzgc+AOwsaTTSMOCf76i2O8j1bfeTRoRdZ+8rCpXS3p26BlJ2wKX97N925R6zf9fvu1E6lxZyeB7kj5Iqo/+cV40mdREugqPFhsjSJpBtTMkNt6PD+REuy5pIqcyVlMagqUHVTcgZN3xAYiIefmHzctIzf5fERW0IlwhdfqUaCTdSL9U1weuIdV1r59vU4GbKjrG7qQP65bAxaTrGXtW/Dwmkvod7A5s0OnXdRDlvolUpXh7vi0jDQtyHXBtydjXkX44XZMfPxc4r6JyzyONr1WsTryuothbk6qd/ppv84EZFb7mH8jv9R1IrcLuIQ3wVybm0cDJwJqFZWuSxvEqXY1bd/xCzImkQSyvyp/T75FqFCqJvyLdXLXVU53DmQMQy1tPPUj6ZVyHjUjNUccBO+TmokPueFd3H5iCXSqK00rlveYLnow0fAbw7ICIlTTSiIg5kl5M6owo4OaoYEDFgj9Haib6F/LroTRYYRlfIg0IuVBSr4EyS8YejvgNZ5Jel7flxweQWv29ocJjrBDcaqsF1TOceSP2NNIvm1eRfnFfAfxXRCyoKP5JpFPxG3J8KNlpUNIeEXFeX01SYxQ0Ra2j13wh9jdJfUjeA3yc1PLsxoj47xIxXxcRF6n1mFilfhg0HafX1MuS5kbEjApi1zZQ5jDF7/U6DLVD8orOiaQPSoOeTaXQkScquDAraTbp7KZxwXo/4OMRsW3few0q/o0RMb2KWAMcZwypNVElc4gPpyp7zed4Y0gt8d5IOmu4ADghSny4JB0ZEYfXNZpAPsvZgnSt6HOFVesAn4uILcrEz8dolaR6LRvB8b9F6kB8dl60D7BFRBxeRfwViRNJC5JOJf3SmQc8kxdHFVU4ykM8Ny0b8vwmLeKfSBpXp/J247lFz4dJr8lc0rWe/42I/1f1saqmFnO/t1q2slAagHBvUoODWYVVDwNnRsTfSsRuDJT5c+Cd0GOgzB9FxIuHGns44heO8zBpYM/Gmf1Ylndmjaho+oQVga+RtNYFTC/zi7IfF0s6lFT/GqT+B7/LwxgQ5XtanwJcIeluUrPfRu/wKsYfmh4RD+WOX78ndaycC4zYRJI73K1BDfPBSzo7It4u6TpaXzsa8muumsfCijQy8W8kvSoirigTq4U3kQbKnAx8m57XGr84CuI3rEu6LrJpRBwlaRNgw4j4e4XHWCE4kbR2Palz010DbTgE78h/P9S0/H2kL6OyF4BPIg3lcB3Lf0lVZZXcxHJv4PsR8bSkkX5K26oBRZB+eX+/ZOxP5r8nA1eSRoetSn99map8zd8i6QZSk+I/kOaD+VRE/HyoASP1RzmVmgbKrDt+wXGkz9DrgKNI75lfkFrSWYETSWsbADdKupKew6WX7ncQEWVbxAzkX5HnT6jBj0jNcq8B/pI74Y3oaySxfD74rwDfzWdUXyZNaFXql3gsHxV6bVIfkvtJZ5rnRtNc60OIfSTQGBL8kxHxQH48gfQrvCpvjIjPS3oLae6KfUnN0oecSKD+gTLrjp9tGxGvlHR1PuYSjY6JrYadE0lrR9QVOI+VtBu9L+SXHrY7uzlfyziPCucMyReU/xMRGxWW/Yv6mjBXbZ9cPbE9sDPpy/iHLJ8AbMjyl/6Rkl5GOuO8VNKiiKiimejLGkkkH2tJ7sRalVXy3zcDZ0TE/Y1mzBX4o6TPUt9AmXXHfzp/XhujREyi+rP8FYITSQuRhi6py3nAE9RT9QSpZ++TVDwDY/4FeAjLW7CQryFVMufGMGg0mtiNdEH2N5KOqPgY95B6/S8mzWtfhbqHBD9P0s2kqq2P5i/L0oNkZo2WZR8rLKui+na44h8L/Ap4jqSvk1ptfami2CsUt9pqIbfWaLww40m/2h6topWGpGsruvA97DQMQ7HXRdJvgX+TOpM1hhm5MnqO7TXU2B8hnYlMIg2VclZVreY0DEOC5+qyhyLimdw3Y52oeLj30So3k349y0fRrXXu89HKiaQNSjPUbRMRpVuESDqG9Ia8sHzJWsavfFjwQuzahmKvW/6C3IU0dMmtSvNYvLSK/4PS3OFnRk3T36rmIcHr6jOVY9cyUOZwxbf2OJG0qaq+Hvmi5s9J4z49zfLmuZW0SZd0DnAzqX39UaTmizdFxCf73dFWSjX3mToc2JH0Rf97YFfgsojYp2zs4Yhv7XMiaaFpWIoxpH4lr40KZk2UtIDUfPa6OvqpKM/N3KhCy811L4hqZhmsbfZF6wxJN1FTn6ncv+blpMEsXy7puaQe/3uMhvjWPl9sb634RlxKavK6V0WxbwWur6mzI/QeFvxuyg8L3nAyqS/Gq/PjRaSJkZxIRq86+0zVOVDmcMS3NjmRtBAVDOTXj7uASySdT8/muVU1/z0+Xzz9Emnoi7WobkTUF0TEOyTtDxBpkLzK2opaR9TWZwrolrQe8BPSD5BHSB03q1J3fGuTq7ZakDSZNAHSdqSWMpeROoUtqiB2ywHfGh3QKoj/GZa3OGt8yT8AzC17MVjS30gtWC7PHbVeQOp7sE2ZuNY5kl7bannVTeBV8UCZwx3f+udE0oKkPwKnA40mlu8CDoiInTtXqvbkzohdpP4qkPpNzAFeDJwTEd8sEXtn0pnOdNLc3tsBB0XEJWXKbCumugfK9ECcI4cTSQuS5kXEVgMtG2LsSaSpb5ub55a+GJ7jXwC8LfI835LWIvVBeAvprKTUEPOSJgIzSWc7syPivpJFtg6QdFlEbN/UZwoqaEVYGCjzYlKrquJAmedHxEuGGns44tvg+RpJa/dJehfL5wzZn9RbuQqnkTr07U4akv1A4N6KYkOaKe6pwuOngSn5esaQJvyRdCOp3GdGxG3A78oX0zopIrbPf/sbHHKo6hwoczji2yCN6XQBRqj3kXoQ3026OL4PUNUF+IkRcSLwdERcmjsKVjIXSXY6MFvS4fl6zOXAGZLWBIbakW1/0kX7CyX9XdKnJJUagt1WXBHxvTw46deBrfL9k0lzwpcesr7u+DZ4TiStfRU4MCImRcRzSInliIpiN5rn3iVptzwA3+SKYhMRXwU+SLrA/iDw4Yg4KiIejYgDhhjzmog4LCJeQBo6fQopWV0k6YNVld1WOPtEGm25MVDmT0kDZY6W+NYmXyNpodGpb6BlQ4y9O/BXYGNSy7B1gCMi4rx+dxxhJO0IfIfUmW3VDhfHRqBC59hvkDrgnl7V52g44lv7fEbS2pjcFwOofMTVfUkJ/PqI2In0S+otFcWulaStJf2vpIXAkcDxpClPzVr5t6Qfk6qJfy9pVar9zqk7vrXJZyQt1Dniap1nO3WR9D+k0W2XkCZuOrOKPjW2YqtzoMzhiG/tcyLpQ10jrkq6Btgxes4vcWlEvLSK+HXIF+3PiIh/dLosZjbyOJEMs+GYX6Iu+RfgZ0iDNn7QgzaaGTiRdETd80vURdJZpHb774mILSWtDlxRRUdNMxu9nEisbZK6I6KreE1H0jVRwSyDZjZ6uYWDDcZT+SwkAPKgjUPqLW9mKw4PkWKDcTjwB2BjSaeRB23saInMrONctWVtyfOOTAYew4M2mlmBE4m1TdLciJjR6XKY2cjiayQ2GLMlbd3pQpjZyOIzEmtbHk7+RcBC4FGWz13xso4WzMw6yonE2iZpSqvlEbFwuMtiZiOHW21Z2xoJQ9JzKMzuaGYrN18jsbZJ2lPSrcA/gUuB24HzO1ooM+s4JxIbjK+Smv7+I89K93rSDIxmthJzIrHBeDoiFpPmaxkTERcDHmfLbCXnayQ2GA9IWgv4C3CapHuApR0uk5l1mFttWdskrQk8TjqTPQBYFzgtn6WY2UrKicTaJul9wF8j4tZOl8XMRg5XbdlgTAXeJWkq0A38lZRY5nWwTGbWYT4jsUHLQ8l/ELm8VEYAAATdSURBVPgssFFEjO1wkcysg5xIrG2SvkQaOn4t4GrgMtIZyV0dLZiZdZQTibVN0lWkVlq/I3VInB0RT3S2VGbWaU4kNiiS1ga2z7e3A/+JiO07Wyoz6yRfbLe2SdoSeA3wWqALuIN0wd3MVmI+I7G2SWpUaV0GzImIpztcJDMbATxEig3GHyPimxHxt0YSkfTJThfKzDrLicQG4z0tlh003IUws5HF10hsQJL2B94JTJM0q7BqbcDDo5it5JxIrB1XAXcBGwDfLix/GLi2IyUysxHDicTacUZEvFLSbRFxaacLY2YjixOJtWO8pAOBV0l6a/PKiPhlB8pkZiOEE4m148OkYePXA/ZoWheAE4nZSsz9SKxtkt4fESd2uhxmNrI4kVjbJI0nnZ3skBddCvzIHRPNVm5OJNY2SScAqwCn5EXvBp6JiA90rlRm1mlOJNY2SddExMsHWmZmKxf3bLfBeEbSCxoPJE0DnulgecxsBHCrLRuMzwIXS1qQH08F3tu54pjZSOBEYoMxEdiSlED2Al4NPNjJAplZ57lqywbjyxHxELAOsDPwI+CHnS2SmXWaE4kNRuN6yG6kZr+/AcZ3sDxmNgI4kdhg/FvSj0lT7P5e0qr4PWS20nPzX2ubpDWAXYDrIuJWSRsCL42ICztcNDPrICcSMzMrxdUSZmZWihOJmZmV4kRiKy1Jn5B0k6TThrDvVEnvrKNcZqONE4mtzD4KvDkiDhjCvlNJ89gPiqSxQziW2YjmRGIrJUk/AqYBsyT9l6Q1JZ0kaY6kqyXtlbebKumvkq7Kt1fnEEcDr5E0L+9/kKTvF+L/VtKO+f4jko6S9HfSLJMzJF0qaa6kC3Lrt+by/VTSsZL+JmmBpH3y8rUk/TmX5bqmct4s6QRJ10s6TdIbJF0u6VZJ2+TtWj5Ps1IiwjffVsobcDuwQb7/P8C78v31/n87d8waRRCGcfz/qIFD0TQeFkGtVFDEiCKkCbGx1ELRQkS/gIXgBxAEQbAMlsHGSoQUaUQsFYsoJmfUSCCpDNqKEPW812LmcCO3IXdTiPD8YGFm9mZnhuN4d3a5F/gI7AC2A43cfgCYzeUJYKZyrWvAZKU+A0zkcgAXc3kIeAE0c/0SMNVjbg+AR6SbvcPAUm7fBuzK5d3AEiDSDqkNHM19XgFT+dw5YHqjdf7r78LH/30415ZZcgY4K+lmrjeAfcAnYFLSKOmf/QcHuPYv4HEuHyLlK3sqCWArsFrTbzoiOsA7SXtym4A7ksaBDjACdM8tR0QLQNIC8CwiQlKLFGg2Wuf7AdZlBjhpo1mXgPMRsbiuUboFfAaOke7012r6t1n/qLhRKa9FRDe9jICFiBjbxJy+/zU/gMtAEzgRET8lrVTGqn6+U6l3+PNb77lOsxJ+R2KWPAGuK28TJB3P7cPAat4ZXCHtIAC+Ajsr/VeAUUlbJO0FTtWMswg0JY3lcYYkHeljnsPAlxxETgP7++gL9es0G5gDiVlym/T+Yl7S21wHuA9clfSS9FjrW26fB9qS5iTdAJ4Dy0ALuAe87jVIRPwALgB3Jc0Bb0jp+DfrIXBS0ixpd/Khj75Qv06zgTlFipmZFfGOxMzMijiQmJlZEQcSMzMr4kBiZmZFHEjMzKyIA4mZmRVxIDEzsyIOJGZmVuQ3NTdAXCxfZCQAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.ensemble import RandomForestClassifier\n",
    "\n",
    "rf = RandomForestClassifier(n_estimators=200, max_depth=7)\n",
    "rf.fit(X_train, Y_train)\n",
    "y_pred_rf = rf.predict(X_val)\n",
    "\n",
    "rf_roc_auc = roc_auc_score(Y_val, y_pred_rf)\n",
    "fpr, tpr, thresholds = roc_curve(Y_val, rf.predict_proba(X_val)[:,1])\n",
    "plt.figure()\n",
    "plt.plot(fpr, tpr, label='Random Forest (area = %0.2f)' % rf_roc_auc)\n",
    "plt.plot([0, 1], [0, 1],'r--')\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 for random forest')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.savefig('Log_ROC')\n",
    "\n",
    "# Plot non-normalized confusion matrix\n",
    "plot_confusion_matrix(Y_val, y_pred_rf, classes=['0','1'],\n",
    "                      title='Confusion matrix, without normalization')\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "plot_confusion_matrix(Y_val, y_pred_rf, classes=['0','1'], normalize=True,\n",
    "                      title='Normalized confusion matrix')\n",
    "\n",
    "# visualize feature importances of the model\n",
    "plt.figure()\n",
    "plt.plot(X_resampled.columns,rf.feature_importances_, '.')\n",
    "plt.plot(X_resampled.columns,0 * rf.feature_importances_, '--')\n",
    "plt.title(\"Feature importances of random forest\")\n",
    "plt.xticks(rotation=90)\n",
    "plt.xlabel(\"feature name\")\n",
    "plt.ylabel(\"Importance\")\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test data prediction\n",
    "Given the above validation results, we will use the random forest for the test data install probability prediction. \n",
    "* **Note:** This was not at all a clear win for random forest, as it seems that the baseline methods can also achieve almost the same average score. If I had time I would look deeper at each method and try to do a better validation before selecting the final method. Furthermore, as a general rule of thumb, the simpler methods should be more favorable in this situation. Anyway, I decided to go with random forest because I also consider the resulting feature importance to be more sensible. \n",
    "* Please note that we could have used more advanced methods (such as supervised deep learning methods) to learn a better feature representation and mapping for this task, However, for this assignment I am fine with the current report given my personal time.\n",
    "\n",
    "Let's load the test data and also look at the number of nulls)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 3740134 entries, 0 to 3740133\n",
      "Data columns (total 16 columns):\n",
      "id                 object\n",
      "timestamp          object\n",
      "campaignId         object\n",
      "platform           object\n",
      "softwareVersion    object\n",
      "sourceGameId       int64\n",
      "country            object\n",
      "startCount         int64\n",
      "viewCount          int64\n",
      "clickCount         int64\n",
      "installCount       int64\n",
      "lastStart          object\n",
      "startCount1d       int64\n",
      "startCount7d       int64\n",
      "connectionType     object\n",
      "deviceType         object\n",
      "dtypes: int64(7), object(9)\n",
      "memory usage: 456.6+ MB\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "import csv\n",
    "\n",
    "test_filename = \"test_data.csv\"\n",
    "# load the whole test data (no need to chunk it!)\n",
    "test_data = pd.read_csv(test_filename, sep=';',engine='python', quoting=csv.QUOTE_NONE)\n",
    "# print basic statistics about the data\n",
    "print(test_data.info())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of nulls per feature:\n",
      "id                      0\n",
      "timestamp               0\n",
      "campaignId              0\n",
      "platform                0\n",
      "softwareVersion         0\n",
      "sourceGameId            0\n",
      "country               281\n",
      "startCount              0\n",
      "viewCount               0\n",
      "clickCount              0\n",
      "installCount            0\n",
      "lastStart          275262\n",
      "startCount1d            0\n",
      "startCount7d            0\n",
      "connectionType          0\n",
      "deviceType              2\n",
      "dtype: int64\n"
     ]
    }
   ],
   "source": [
    "print(\"Number of nulls per feature:\")\n",
    "print(test_data.isnull().sum())\n",
    "#with pd.option_context('display.max_rows', 5, 'display.max_columns', None, 'display.width', 100):\n",
    "#    print(test_data.head())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We need to perform the same operations done on training data on the test data. However, we need to use statistics and dictionaries that were learned from the training data. I would have ideally implemented those as functions and here would just call them again (but did not do that here due to time constraints). The following is the preprocessing steps:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# replace nan values in 'lastStart' with values in 'timestamp'\n",
    "nan_idx = test_data['lastStart'].isnull()\n",
    "test_data['lastStart'].loc[nan_idx] = test_data['timestamp'][nan_idx]\n",
    "\n",
    "# convert the string objects to datetime\n",
    "test_data['timestamp'] = pd.to_datetime(test_data['timestamp'],infer_datetime_format=True)\n",
    "test_data['lastStart'] = pd.to_datetime(test_data['lastStart'],infer_datetime_format=True)\n",
    "\n",
    "# convert times to epoch in minutes\n",
    "epoch_timestamp = pd.to_datetime(test_data['timestamp']).astype(np.int64) // (10**9 * 60)\n",
    "epoch_lastStart = pd.to_datetime(test_data['lastStart']).astype(np.int64) // (10**9 * 60)\n",
    "\n",
    "# create a new feature of difference in timestamp and lastStart in minutes\n",
    "test_data['epoch_diff'] = epoch_timestamp - epoch_lastStart\n",
    "\n",
    "# convert the campains to frequency, based on values learned from training data\n",
    "campain_freq = np.zeros(test_data.shape[0])\n",
    "for i in range(test_data.shape[0]):\n",
    "    if test_data['campaignId'][i] in campainId_dict.keys():\n",
    "        campain_freq[i] = campainId_dict[test_data['campaignId'][i]]\n",
    "    else:\n",
    "        campain_freq[i] = 1    \n",
    "\n",
    "# add the new feature to the data frame\n",
    "test_data.insert(1, 'campain_freq', campain_freq)\n",
    "\n",
    "# convert platform to binary values (1:ios, 0:android)\n",
    "test_data['platform'] = test_data['platform'] == 'ios' \n",
    "test_data['platform'] = test_data['platform'].astype(np.int64)\n",
    "\n",
    "# save the first number in the version string in an int array\n",
    "soft_ver_num = np.zeros(test_data.shape[0])\n",
    "for i in range(test_data.shape[0]): \n",
    "    soft_ver_num[i] = find_first_num(test_data['softwareVersion'][i])\n",
    "\n",
    "# add the new feature to the data frame\n",
    "test_data.insert(1, 'softwareVersion_num', soft_ver_num)\n",
    "\n",
    "# convert the gameId to frequency, based on values learned from training data\n",
    "gameId_freq = np.zeros(test_data.shape[0])\n",
    "for i in range(test_data.shape[0]):\n",
    "    if test_data['sourceGameId'][i] in gameId_dict.keys():\n",
    "        gameId_freq[i] = gameId_dict[test_data['sourceGameId'][i]]\n",
    "    else:\n",
    "        gameId_freq[i] = 1 \n",
    "        \n",
    "# add the new feature to the data frame\n",
    "test_data.insert(1, 'gameId_freq', gameId_freq)\n",
    "\n",
    "# convert the country to frequency, based on values learned from training data\n",
    "country_freq = np.zeros(test_data.shape[0])\n",
    "for i in range(test_data.shape[0]):\n",
    "    if test_data['country'][i] in country_dict.keys():\n",
    "        country_freq[i] = country_dict[test_data['country'][i]]\n",
    "    else:\n",
    "        country_freq[i] = 0.01 # a default number a bit smaller than the prior install probability\n",
    "\n",
    "# add the new feature to the data frame\n",
    "test_data.insert(1, 'country_ratio', country_freq)\n",
    "\n",
    "# create connectionType features\n",
    "test_data['wifi'] = test_data['connectionType'] == 'wifi' \n",
    "test_data['wifi'] = test_data['wifi'].astype(np.int64)\n",
    "test_data['cellular'] = test_data['connectionType'] == 'cellular' \n",
    "test_data['cellular'] = test_data['cellular'].astype(np.int64)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Drop columns that were modified."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# remove \"id\" column\n",
    "test_data = test_data.drop(columns=\"id\")\n",
    "# remove \"timestamp\" column\n",
    "test_data = test_data.drop(columns=\"timestamp\")\n",
    "test_data = test_data.drop(columns=\"lastStart\")\n",
    "# remove 'campaignId' column\n",
    "test_data = test_data.drop(columns=\"campaignId\")\n",
    "# remove 'softwareVersion' column\n",
    "test_data = test_data.drop(columns=\"softwareVersion\")\n",
    "# remove 'sourceGameId' column\n",
    "test_data = test_data.drop(columns=\"sourceGameId\")\n",
    "# remove 'country' column\n",
    "test_data = test_data.drop(columns=\"country\")\n",
    "# remove 'connectionType' column\n",
    "test_data = test_data.drop(columns=\"connectionType\")\n",
    "# remove 'deviceType' column for now (take care of this later)\n",
    "test_data = test_data.drop(columns=\"deviceType\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Save the modified test data for later use."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save the modified testdata dataframe for future use\n",
    "test_data.to_csv(r'test_data_m.csv', index = None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Load the modified test data\n",
    "test_filename = \"test_data_m.csv\"\n",
    "test_data = pd.read_csv(test_filename)\n",
    "# let's look at the test data\n",
    "with pd.option_context('display.max_rows', 5, 'display.max_columns', None, 'display.width', 100):\n",
    "    print(test_data.head())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now standardize the test data based on the statistics of the training data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 135,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Get column names first\n",
    "names = test_data.columns\n",
    "# Fit your data on the scaler object (trained from trainign data)\n",
    "X_test = scaler.transform(test_data)\n",
    "X_test = pd.DataFrame(X_test, columns=names)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's evaluate the test data using the best method from validation phase, i.e., the random forest (not strongly the best, but fine)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# use the best (rough) parameters found\n",
    "rf = RandomForestClassifier(n_estimators=200, max_depth=7)\n",
    "# train using all the training data\n",
    "rf.fit(X_resampled, Y_resampled)\n",
    "# predict the test data\n",
    "y_pred_test = rf.predict(X_test)\n",
    "Y_test_predicted_prob = rf.predict_proba(X_test)[:,1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Save the predicted probabilities in a csv file."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 138,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_temp = pd.DataFrame(data={\"install\": Y_test_predicted_prob})\n",
    "df_temp.to_csv(\"./Y_pred_test.csv\", sep=',',index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We are done!"
   ]
  }
 ],
 "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": "3.6.5"
  },
  "latex_metadata": {
   "affiliation": "Aalto University",
   "author": "Pedram Daee",
   "title": "Unity Assignment"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}