gradient_boosted_trees.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "gradient_boosted_trees.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "authorship_tag": "ABX9TyOBIEozSMXFG7SNdRT8/f+v",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/wolframalexa/FrequentistML/blob/master/gradient_boosted_trees.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Liwz4axEHPOY",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Assignment 4: xTreme Gradient Boosted Trees\n",
        "# Select a dataset for either classification or regression. It can be the same dataset \n",
        "# as the one you used in the previous assignments, or a new one. Use an out of the box package \n",
        "# for xTreme gradient boosting trees such as https://xgboost.readthedocs.io/en/latest/.\n",
        "# Use the same 80-10-10 split to tune your classifier/regression method and report your performance and output the feature importance. \n",
        "# Do the features reported make sense? If you are using the same dataset from assignment 1 or 2, do they agree with what you discovered using the Lasso penalty?"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QBjIb-CFITyK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#----------IMPORT PACKAGES\n",
        "\n",
        "import pandas as pd\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import random\n",
        "import xgboost as xgb\n",
        "from sklearn.metrics import mean_squared_error as mse"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bp3gMtF-IeYl",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#----------READ DATA\n",
        "\n",
        "data = 'https://raw.githubusercontent.com/wolframalexa/FrequentistML/master/datasets/life_expectancy_data.csv'\n",
        "dataframe = pd.read_csv(data, sep=',', header='infer')\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8aiwurWVLCGp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 304
        },
        "outputId": "fe90f2b4-1dc6-45e8-a85c-2fe4d7953d04"
      },
      "source": [
        "#----------CLEAN DATA\n",
        "\n",
        "# Add column to account for intercept \n",
        "ones_col = np.ones(len(dataframe))\n",
        "dataframe.insert(0, \"intercept\", ones_col, True) \n",
        "\n",
        "# Drop rows with missing data\n",
        "clean_data = dataframe.dropna(axis=0)\n",
        "\n",
        "# Rename columns\n",
        "clean_data.columns = ['intercept','country','year','status','life_exp','adult_mort','infant_mort','alcohol','percent_exp','hep_b','measles','bmi','under_five_mort','polio','tot_exp',\n",
        "                      'diphtheria','hiv_aids','gdp','population','thin_1-19','thin_5-9','income_comp_res','schooling'] \n",
        "\n",
        "# Remove rows with invalid data\n",
        "clean_data = clean_data[clean_data['infant_mort'] <= 1000] \n",
        "clean_data = clean_data[clean_data['measles'] <= 1000] \n",
        "clean_data = clean_data[clean_data['under_five_mort'] <= 1000]\n",
        "\n",
        "# Show data information                    \n",
        "clean_data.describe()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>intercept</th>\n",
              "      <th>year</th>\n",
              "      <th>life_exp</th>\n",
              "      <th>adult_mort</th>\n",
              "      <th>infant_mort</th>\n",
              "      <th>alcohol</th>\n",
              "      <th>percent_exp</th>\n",
              "      <th>hep_b</th>\n",
              "      <th>measles</th>\n",
              "      <th>bmi</th>\n",
              "      <th>under_five_mort</th>\n",
              "      <th>polio</th>\n",
              "      <th>tot_exp</th>\n",
              "      <th>diphtheria</th>\n",
              "      <th>hiv_aids</th>\n",
              "      <th>gdp</th>\n",
              "      <th>population</th>\n",
              "      <th>thin_1-19</th>\n",
              "      <th>thin_5-9</th>\n",
              "      <th>income_comp_res</th>\n",
              "      <th>schooling</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>count</th>\n",
              "      <td>1357.0</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.00000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1.357000e+03</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "      <td>1357.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>mean</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2007.941783</td>\n",
              "      <td>70.062122</td>\n",
              "      <td>161.971997</td>\n",
              "      <td>10.862196</td>\n",
              "      <td>4.720442</td>\n",
              "      <td>780.089679</td>\n",
              "      <td>81.265291</td>\n",
              "      <td>91.20339</td>\n",
              "      <td>40.379366</td>\n",
              "      <td>15.097273</td>\n",
              "      <td>85.208548</td>\n",
              "      <td>6.104576</td>\n",
              "      <td>85.543110</td>\n",
              "      <td>1.941489</td>\n",
              "      <td>6123.079003</td>\n",
              "      <td>7.627781e+06</td>\n",
              "      <td>4.215623</td>\n",
              "      <td>4.258364</td>\n",
              "      <td>0.641920</td>\n",
              "      <td>12.409506</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>std</th>\n",
              "      <td>0.0</td>\n",
              "      <td>4.041488</td>\n",
              "      <td>8.690290</td>\n",
              "      <td>123.624297</td>\n",
              "      <td>25.177559</td>\n",
              "      <td>4.054261</td>\n",
              "      <td>1883.611933</td>\n",
              "      <td>24.863380</td>\n",
              "      <td>193.88610</td>\n",
              "      <td>19.534123</td>\n",
              "      <td>36.859127</td>\n",
              "      <td>21.512464</td>\n",
              "      <td>2.279412</td>\n",
              "      <td>21.154592</td>\n",
              "      <td>6.180984</td>\n",
              "      <td>12197.905019</td>\n",
              "      <td>2.014408e+07</td>\n",
              "      <td>3.728639</td>\n",
              "      <td>3.802764</td>\n",
              "      <td>0.189328</td>\n",
              "      <td>2.745448</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>min</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2000.000000</td>\n",
              "      <td>44.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>0.010000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>2.000000</td>\n",
              "      <td>0.00000</td>\n",
              "      <td>2.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>3.000000</td>\n",
              "      <td>0.740000</td>\n",
              "      <td>2.000000</td>\n",
              "      <td>0.100000</td>\n",
              "      <td>5.668726</td>\n",
              "      <td>3.400000e+01</td>\n",
              "      <td>0.100000</td>\n",
              "      <td>0.100000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>4.200000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>25%</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2005.000000</td>\n",
              "      <td>65.200000</td>\n",
              "      <td>74.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>1.070000</td>\n",
              "      <td>40.806180</td>\n",
              "      <td>79.000000</td>\n",
              "      <td>0.00000</td>\n",
              "      <td>22.800000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>84.000000</td>\n",
              "      <td>4.600000</td>\n",
              "      <td>84.000000</td>\n",
              "      <td>0.100000</td>\n",
              "      <td>519.292285</td>\n",
              "      <td>1.557560e+05</td>\n",
              "      <td>1.500000</td>\n",
              "      <td>1.500000</td>\n",
              "      <td>0.542000</td>\n",
              "      <td>10.700000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>50%</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2008.000000</td>\n",
              "      <td>72.300000</td>\n",
              "      <td>144.000000</td>\n",
              "      <td>2.000000</td>\n",
              "      <td>4.040000</td>\n",
              "      <td>179.170133</td>\n",
              "      <td>92.000000</td>\n",
              "      <td>3.00000</td>\n",
              "      <td>46.400000</td>\n",
              "      <td>3.000000</td>\n",
              "      <td>94.000000</td>\n",
              "      <td>5.990000</td>\n",
              "      <td>94.000000</td>\n",
              "      <td>0.100000</td>\n",
              "      <td>1839.729450</td>\n",
              "      <td>1.143896e+06</td>\n",
              "      <td>2.700000</td>\n",
              "      <td>2.800000</td>\n",
              "      <td>0.690000</td>\n",
              "      <td>12.500000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>75%</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2011.000000</td>\n",
              "      <td>75.400000</td>\n",
              "      <td>219.000000</td>\n",
              "      <td>11.000000</td>\n",
              "      <td>7.440000</td>\n",
              "      <td>579.133164</td>\n",
              "      <td>96.000000</td>\n",
              "      <td>64.00000</td>\n",
              "      <td>56.700000</td>\n",
              "      <td>13.000000</td>\n",
              "      <td>97.000000</td>\n",
              "      <td>7.610000</td>\n",
              "      <td>97.000000</td>\n",
              "      <td>0.600000</td>\n",
              "      <td>5414.634326</td>\n",
              "      <td>5.737723e+06</td>\n",
              "      <td>6.500000</td>\n",
              "      <td>6.500000</td>\n",
              "      <td>0.762000</td>\n",
              "      <td>14.300000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>max</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2015.000000</td>\n",
              "      <td>89.000000</td>\n",
              "      <td>723.000000</td>\n",
              "      <td>549.000000</td>\n",
              "      <td>17.870000</td>\n",
              "      <td>18961.348600</td>\n",
              "      <td>99.000000</td>\n",
              "      <td>999.00000</td>\n",
              "      <td>77.100000</td>\n",
              "      <td>879.000000</td>\n",
              "      <td>99.000000</td>\n",
              "      <td>14.390000</td>\n",
              "      <td>99.000000</td>\n",
              "      <td>50.600000</td>\n",
              "      <td>119172.741800</td>\n",
              "      <td>1.986867e+08</td>\n",
              "      <td>19.700000</td>\n",
              "      <td>19.900000</td>\n",
              "      <td>0.936000</td>\n",
              "      <td>20.700000</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "       intercept         year  ...  income_comp_res    schooling\n",
              "count     1357.0  1357.000000  ...      1357.000000  1357.000000\n",
              "mean         1.0  2007.941783  ...         0.641920    12.409506\n",
              "std          0.0     4.041488  ...         0.189328     2.745448\n",
              "min          1.0  2000.000000  ...         0.000000     4.200000\n",
              "25%          1.0  2005.000000  ...         0.542000    10.700000\n",
              "50%          1.0  2008.000000  ...         0.690000    12.500000\n",
              "75%          1.0  2011.000000  ...         0.762000    14.300000\n",
              "max          1.0  2015.000000  ...         0.936000    20.700000\n",
              "\n",
              "[8 rows x 21 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "O-TxKELcLE2z",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#----------SEPARATE DATA\n",
        "\n",
        "# Choose seed\n",
        "np.random.seed(500)\n",
        "\n",
        "# Separate training and testing data\n",
        "train_data, validate_data, test_data = np.split(clean_data.sample(frac=1), [int(.8*len(clean_data)), int(.9*len(clean_data))])\n",
        "\n",
        "# Separate training inputs and outputs\n",
        "x_train = train_data.drop(['life_exp','country','status'], axis=1)  \n",
        "y_train = train_data['life_exp']\n",
        "\n",
        "# Separate validation inputs and outputs\n",
        "x_validate = validate_data.drop(['life_exp','country','status'], axis=1)  \n",
        "y_validate = validate_data['life_exp']\n",
        "\n",
        "# Separate testing inputs and outputs\n",
        "x_test = test_data.drop(['life_exp','country','status'], axis=1)\n",
        "y_test = test_data['life_exp']\n",
        "\n",
        "# Normalize testing validation and training inputs\n",
        "x_train = (x_train-x_train.min())/(x_train.max()-x_train.min()) \n",
        "x_train.intercept = np.ones(len(x_train))\n",
        "x_validate = (x_validate-x_validate.min())/(x_validate.max()-x_validate.min()) \n",
        "x_validate.intercept = np.ones(len(x_validate))\n",
        "x_test = (x_test-x_test.min())/(x_test.max()-x_test.min()) \n",
        "x_test.intercept = np.ones(len(x_test))\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IhP5A3u3aJ9Q",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 168
        },
        "outputId": "2969a922-b3df-4bb8-c411-03e39240710f"
      },
      "source": [
        "#--------TRAIN MODEL\n",
        "\n",
        "etas = np.linspace(0,1,10)\n",
        "depth_lim = 5\n",
        "estimators_lim = 5\n",
        "\n",
        "\n",
        "min_score = 100\n",
        "for depth in range(depth_lim): # nest these to get all possible models\n",
        "  for estimators in range(estimators_lim):\n",
        "    for rate in etas:\n",
        "      # make and fit model with these parameters\n",
        "      xgb_model = xgb.XGBRegressor(n_estimators = estimators, learning_rate=rate, objective=\"reg:squarederror\", max_depth=depth)\n",
        "      xgb_model.fit(x_train, y_train)\n",
        "      y_pred = xgb_model.predict(x_validate)\n",
        "\n",
        "      # score model and choose model with lowest error\n",
        "      score = mse(y_validate, y_pred)\n",
        "      if score < min_score:\n",
        "        min_score = score\n",
        "        best_model = xgb_model\n",
        "\n",
        "print(\"The best model is the following:\\n\",best_model)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "The best model is the following:\n",
            " XGBRegressor(base_score=0.5, booster='gbtree', colsample_bylevel=1,\n",
            "             colsample_bynode=1, colsample_bytree=1, gamma=0,\n",
            "             importance_type='gain', learning_rate=0.8888888888888888,\n",
            "             max_delta_step=0, max_depth=4, min_child_weight=1, missing=None,\n",
            "             n_estimators=4, n_jobs=1, nthread=None,\n",
            "             objective='reg:squarederror', random_state=0, reg_alpha=0,\n",
            "             reg_lambda=1, scale_pos_weight=1, seed=None, silent=None,\n",
            "             subsample=1, verbosity=1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TYIQcaT2dJZl",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 87
        },
        "outputId": "5650e123-b486-4652-e8c1-6c3072f9263a"
      },
      "source": [
        "#--------PREDICT AND REPORT FEATURE IMPORTANCE\n",
        "\n",
        "y_pred_test = best_model.predict(x_test)\n",
        "test_score = mse(y_test, y_pred_test)\n",
        "print(\"Score on test set:\",test_score)\n",
        "print(\"Feature importances:\\n\",list(zip(x_test.columns,xgb_model.feature_importances_)))\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Score on test set: 7.424513591207594\n",
            "Feature importances:\n",
            " [('intercept', 0.0), ('year', 0.0018047928), ('adult_mort', 0.025521992), ('infant_mort', 0.006940946), ('alcohol', 0.022453226), ('percent_exp', 0.010459741), ('hep_b', 0.0), ('measles', 0.00012279741), ('bmi', 0.0075482745), ('under_five_mort', 0.065146804), ('polio', 0.0), ('tot_exp', 0.0077739116), ('diphtheria', 0.0), ('hiv_aids', 0.605227), ('gdp', 0.0), ('population', 0.0), ('thin_1-19', 0.0), ('thin_5-9', 0.07071378), ('income_comp_res', 0.15453555), ('schooling', 0.021751178)]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wt2vWzm-mNqp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 431
        },
        "outputId": "64c9f41a-b294-4f22-8436-14655f74036f"
      },
      "source": [
        "#----------PLOT FEATURE IMPORTANCE\n",
        "\n",
        "fig = plt.figure()\n",
        "fig.patch.set_facecolor('xkcd:white') # necessary bc dark mode\n",
        "ax = fig.add_axes([0,0,1,1])\n",
        "features = x_test.columns\n",
        "importances = xgb_model.feature_importances_\n",
        "ax.bar(features, importances)\n",
        "plt.xticks(rotation=90)\n",
        "plt.title(\"Feature importance after running random forests\")\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAdUAAAGeCAYAAADCJptYAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOzdeVxUZf8//tcoiguriBsQiCQqgiiQoilgKOaClormUurtUpnV18olu3HJ9dZcyQrX1BTXBFPJLTF3yT1uDRURcMMFFZFtuH5/+JvzYQSEOXMh6P16Ph48HszMmfdcMxzmdZbruo5GCCFARERERqtQ1g0gIiJ6VTBUiYiIJGGoEhERScJQJSIikoShSkREJAlDlYiISBKGKkk1ffp0DB06tKyb8VL75ptvULNmTdSpU6esmyLFy7xODBo0CN98880Lea1X7e/+v0rDcarlh5OTE27duoWKFSsq9/3zzz+oV6+eUTWXLl2KwMBAGU18qUyaNAmXLl3CmjVryropJXbt2jW4uroiMTERtWrVwsqVK7F06VIcPHiwrJv2P2nQoEGwt7fH1KlTS/V1nv27vyhcv+Tjnmo5s23bNqSnpys/xgSqDLm5uWX6+mq9rO2+du0abGxspH2xluRzeFk/q2e9zO/DmL/7y/y+X0mCyg1HR0exe/fuAvenpaWJIUOGiDp16oh69eqJCRMmiNzcXCGEEJcuXRIBAQGiRo0awsbGRvTr10/cv39fCCHEgAEDhEajEVWqVBHVq1cXs2bNEn/88Yews7Mr8nUnTpwoevbsKfr37y/Mzc3FkiVLnvv6z5o4caLo37+/EEKIhIQEAUAsX75c2NvbCysrK/HDDz+I48ePC3d3d2FpaSlGjhypPHfFihWidevWYuTIkcLCwkK4urqKPXv2KI+npKSIbt26CWtra9GgQQMRHh6u97r5271o0SJRqVIlYWJiIqpXry48PDyEEEIsX75cNGrUSJiZmYn69euLH3/8Uamh+2zmzJkjbG1tRZ06dcTy5cuVxzMyMsTo0aPFa6+9JiwsLESbNm1ERkaGEEKII0eOCF9fX2FpaSk8PDzEH3/8UeTfecaMGcLZ2VmYmZmJxo0biy1btgghhNi9e7eoUqWK0Gg0onr16iIkJESYmpqKChUqiOrVqwtLS0shhBCZmZniiy++EA4ODqJWrVpixIgRSjt072HmzJmidu3aYsCAAQVeX/c5f/7556JGjRpiwoQJen+3/H+7nJwcIYQQfn5+4ptvvhGtW7cWZmZmokOHDiI1NVVv2ZUrVwoHBwdhY2Mjpk6d+tx1oqhlMzIyxPvvvy+srKxEo0aNxKxZswqsr/kBEGFhYcLFxUU4OTkJIYT49NNPhb29vTA3NxctWrQQBw4c0GtL7969xcCBA4WZmZlo0qSJOHHihPL4yZMnRfPmzYWZmZkICQkRffr0ERMmTFAeDw8PFw0aNBDW1taiW7duIiUlRa8t33//vXBxcRFmZmbim2++EZcuXRK+vr7C3Nxc9O7dW2RlZRV4D8/+3T/44AMhhBCRkZGiSZMmwtLSUvj5+Ym4uDjlOY6OjmLmzJnC3d1dVK5cWeTk5Dx3HVyxYoWoX7++MDMzE05OTmLNmjUiLi6u0PVr+/btonHjxsLMzEzUq1dPzJ49u8jPnwpiqJYjRYVqjx49xPDhw0V6erq4deuW8PHxUcIgPj5e7Nq1S2RmZorbt2+Ltm3bis8++6zImiUJVRMTE/Hrr78KrVYrMjIynvv6zyrsC3TEiBHiyZMn4vfffxempqaie/fu4tatWyI5OVnY2tqK/fv3CyGe/uNXrFhRzJ07V2RnZ4uIiAhhYWEh7t69K4QQom3btuKjjz4ST548EadOnRI1a9YUe/fuLbLdzwaFEEL89ttv4tKlSyIvL0/s379fVK1aVfz111/KZ1OxYkXx73//W2RnZ4vt27eLqlWrinv37gkhhPj444+Fn5+fSE5OFrm5ueLQoUMiMzNTJCcnixo1aojt27cLrVYrdu3aJWrUqCFu375d6Ge0YcMGkZKSIrRarYiIiBDVqlUT169fL/Tvs2LFCtGmTRu953/++eeiW7du4u7du+Lhw4eia9euYty4cXrvYcyYMSIzM1MJ2/x0n/PChQtFTk5OoZ9VYaHq7OwsLl68KDIyMoSfn58YO3as3rJDhw4VGRkZ4vTp06Jy5cpKCBS2ThS17NixY0W7du3EvXv3RFJSknB3dy82VAMDA8Xdu3eV97p69Wpx584dkZOTI+bMmSNq164tnjx5orTF1NRUbN++XeTm5opx48aJli1bCiGEyMrKEq+99pqy/m3cuFGYmJgoobp3715hY2Mj/vrrL5GZmSk++eQT0bZtW722BAcHiwcPHojz58+LypUri/bt24vLly+LtLQ00bhxY7Fy5cpC38ezf/eLFy+KatWqiV27dons7Gwxa9Ys0aBBAyWUHR0dRbNmzcS1a9dERkbGc9fB9PR0YW5uLi5cuCCEEOL69evi/PnzRa5fderUUTZE7t27p/x/UMkwVMsRR0dHZYvR0tJSdO/eXdy8eVNUrlxZ78tx7dq1wt/fv9Aav/76q/D09NSraWio5v+iMPT1C/sCTU5OVh6vUaOGiIiIUG6/++67Yt68eUKIp//gdevWFXl5ecrjPj4+YtWqVeLatWuiQoUK4uHDh8pj48aNU7bqn233s20pSvfu3cX8+fOVz6ZKlSpKkAghhK2trThy5IjQarWiSpUq4vTp0wVqzJw5s8AeYceOHYv8An1Ws2bNxNatW5U2PC9U8/LyRLVq1cSlS5eU+w4fPqzspf3xxx+iUqVKSogUZsWKFcLBwUHvvpKE6rfffqs8/v3334ugoCC9ZZOSkpTHfXx8xLp16wrULm7Z+vXri+joaOWxJUuWFBuqug2rolhZWSl/t4kTJ4q33npLeezvv/8WVapUEUIIERMTU2D98/X1VUJ1yJAh4quvvlIee/TokTAxMREJCQlKWw4ePKg83qJFCzFz5kzl9ujRo/U2ePN79u8+ZcoU0bt3b+W2VqsV9erVU/Y+HR0dxbJly5THn7cOpqenC0tLS7Fp06YCG1mFhaqDg4P48ccfxYMHDwptKz0fz6mWM1u3bkVaWhrS0tKwdetWJCYmIicnB3Xr1oWVlRWsrKwwYsQI3L59GwBw69Yt9O3bF3Z2drCwsMCAAQNw584do9rg4OCg/F7c65dE7dq1ld+rVq1a4HZ6erpy287ODhqNRrnt6OiI69ev4/r166hRowbMzc31HktJSSm03UXZuXMnWrVqhRo1asDKygo7duzQ+7xsbGxgYmKi3K5WrRrS09Nx584dZGZmokGDBgVqJiYmYuPGjcrnY2VlhYMHD+LGjRuFtmHVqlXw9PRUlj1//nyJ/2apqanIyMiAl5eX8vxOnTohNTVVWcbW1hZVqlR5bp2SfFbPyt8rVfe5GPJ4SZa9fv26XttK0s5nl5kzZw4aN24MS0tLWFlZ4cGDB3qf77OvnZmZidzcXFy/fr3Q9U/n+vXrerfNzMxgY2Ojtw4asq4/z7OvVaFCBTg4OBS5vj9vHaxevTrWr1+PH3/8EXXr1kWXLl1w4cKFIl978+bN2LFjBxwdHeHn54cjR46UqM30FEO1nHNwcICpqSnu3LmjhO3Dhw/x999/AwC+/vpraDQanDt3Dg8fPsSaNWsg8nXozv8FAQDVq1dHRkaGclur1ep9IT/7nOJeX7aUlBS99l+7dg316tVDvXr1cO/ePTx69EjvMTs7u0LbXdjtrKws9OzZE19++SVu3bqFtLQ0dO7cWe/1ilKzZk1UqVIFly9fLvCYg4MDBg4cqHw+aWlpePz4McaNG1dg2cTERAwbNgxhYWG4e/cu0tLS0LRp0yLb8Ox7qFmzJqpWrYq///5bea0HDx7ofVk/+5yS1H12vbh582axNUpD3bp1kZycrNxOSkoq9jn538uff/6J//znP9iwYQPu37+PtLQ0WFpaluhvXLdu3ULXP5169eohMTFRuf348WPcvXtXbx2U5dnXEkIgKSmpyPW9uHUwKCgIu3fvxo0bN9CoUSMMGzasQA0dHx8fREZG4vbt2+jRowdCQkKkv79XGUO1nKtbty46duyIL774Ag8fPkReXh4uX76MmJgYAMCjR49gZmYGS0tLpKSkYPbs2XrPr127Nq5cuaLcbtiwITIzM7F9+3bk5ORg6tSpyMrKUv36st2+fRsLFy5ETk4ONm7ciP/+97/o3LkzHBwc0Lp1a4wfPx6ZmZk4e/Ysli1bhgEDBhRZq3bt2rh69Sry8vIAANnZ2cjKyoKtrS1MTEywc+dO7Nq1q0TtqlChAoYMGYLRo0fj+vXr0Gq1OHLkCLKysjBgwABs27YNv//+O7RaLTIzM7F//369cNB5/PgxNBoNbG1tAQArVqzA+fPnn/sekpOTkZ2drbRj2LBh+H//7/8pRwtSUlLw+++/l+h9FMXT0xMHDhzAtWvX8ODBA8yYMcOoemqFhIRgxowZuH//PlJSUhAWFmbQ8x89egQTExPY2toiNzcXU6ZMwcOHD0v0XF9fX5iYmCjr35YtW3D8+HHl8ffeew8rVqzA6dOnkZWVha+//hotW7aEk5OTQW0siZCQEGzfvh179+5FTk4OvvvuO5iamqJ169aFLv+8dfDWrVuIjIzE48ePYWpqCjMzM1So8PSr/9n1Kzs7G7/88gsePHiASpUqwcLCQlmWSoaf1ktg1apVyM7ORpMmTWBtbY1evXophxYnTpyIkydPwtLSEl26dMG7776r99zx48dj6tSpsLKywpw5c2BpaYnFixdj6NChsLOzQ/Xq1WFvb6/69WVr2bIl4uPjUbNmTUyYMAGbNm2CjY0NAGDdunW4evUq6tWrh3feeQeTJ09+7vjb3r17A3h6SLdFixYwNzfHwoULERISAmtra6xduxbBwcElbtucOXPg7u4OHx8f1KhRA2PHjkVeXh4cHBwQGRmJ6dOnw9bWFg4ODpg9e7YS5vk1adIEX3zxBXx9fVG7dm2cO3cObdq0KfI127dvDzc3N9SpUwc1a9YEAMyaNQsuLi5o1aoVLCwsEBgYiIsXL5b4fRSmQ4cO6NOnDzw8PODl5YWuXbsaVU+t0NBQ2Nvbo379+ggMDESvXr1gampa4ucHBQWhU6dOaNiwIRwdHVGlSpUSH+quXLkytmzZgpUrV6JGjRpYv3693v9TYGAgvv32W/Ts2RN169bF5cuXERERYfB7LAlXV1esWbMGo0aNQs2aNbFt2zZs27YNlStXLnT5562DeXl5mDt3LurVq4caNWogJiYGP/zwA4DC16/Vq1fDyckJFhYW+PHHH/HLL7+Uynt8VXHyByo3OBCdnvXDDz8gIiKi1I6MEMnGPVUiKjdu3LiBQ4cOIS8vDxcvXsR3332Hd955p6ybRVRiJsUvQkT0YmRnZ2PEiBFISEiAlZUV+vbti48//rism0VUYjz8S0REJAkP/xIREUnCUCUiIpKkzM6p1qxZs1TGdxEREZWmq1evFjkLWpmFqpOTE2JjY8vq5YmIiFTx9vYu8jEe/iUiIpKEoUpERCQJQ5WIiEgShioREZEkDFUiIiJJGKpERESSMFSJiIgkYagSERFJwlAlIiKShKFKREQkSYlCNTo6Gq6urnBxccHMmTMLXWbDhg1o0qQJ3Nzc0K9fP6mNJCIiehkUO/evVqvFyJEjsXv3btjb28PHxwfBwcFo0qSJskx8fDxmzJiBQ4cOwdraGrdv3y7VRhMRFcZp3HbVz706s4vEltD/qmL3VI8fPw4XFxc4OzujcuXK6Nu3LyIjI/WWWbJkCUaOHAlra2sAQK1atUqntUREROVYsaGakpICBwcH5ba9vT1SUlL0lvnnn3/wzz//oE2bNmjVqhWio6Plt5SIiKick3Lpt9zcXMTHx2P//v1ITk5Gu3btcO7cOVhZWektFx4ejvDwcABAamqqjJcmIiIqN4rdU7Wzs0NSUpJyOzk5GXZ2dnrL2NvbIzg4GJUqVUL9+vXRsGFDxMfHF6g1fPhwxMbGIjY2Fra2thKaT0REVH4UG6o+Pj6Ij49HQkICsrOzERERgeDgYL1levTogf379wMA7ty5g3/++QfOzs6l0mAiIqLyqthQNTExQVhYGIKCgtC4cWOEhITAzc0NoaGhiIqKAgAEBQXBxsYGTZo0QUBAAGbPng0bG5tSbzwREVF5ohFCiLJ4YW9vb8TGxpbFSxPRK4pDauhFeF5+cUYlIiIiSRiqREREkjBUiYiIJGGoEhERScJQJSIikoShSkREJAlDlYiISBKGKhERkSQMVSIiIkkYqkRERJIwVImIiCRhqBIREUnCUCUiIpKEoUpERCQJQ5WIiEgShioREZEkDFUiIiJJGKpERESSMFSJiIgkYagSERFJwlAlIiKShKFKREQkCUOViIhIEoYqERGRJAxVIiIiSRiqREREkjBUiYiIJGGoEhERScJQJSIikoShSkREJAlDlYiISBKGKhERkSQMVSIiIkkYqkRERJIwVImIiCRhqBIREUnCUCUiIpKkRKEaHR0NV1dXuLi4YObMmQUeX7lyJWxtbeHp6QlPT08sXbpUekOJiIjKO5PiFtBqtRg5ciR2794Ne3t7+Pj4IDg4GE2aNNFbrk+fPggLCyu1hhIREZV3xe6pHj9+HC4uLnB2dkblypXRt29fREZGvoi2ERERvVSKDdWUlBQ4ODgot+3t7ZGSklJguc2bN8PDwwO9evVCUlJSobXCw8Ph7e0Nb29vpKamGtFsIiKi8kdKR6Vu3brh6tWrOHv2LDp06IAPPvig0OWGDx+O2NhYxMbGwtbWVsZLExERlRvFhqqdnZ3enmdycjLs7Oz0lrGxsYGpqSkAYOjQofjrr78kN5OIiKj8KzZUfXx8EB8fj4SEBGRnZyMiIgLBwcF6y9y4cUP5PSoqCo0bN5bfUiIionKu2N6/JiYmCAsLQ1BQELRaLYYMGQI3NzeEhobC29sbwcHBWLhwIaKiomBiYoIaNWpg5cqVL6DpRERE5YtGCCHK4oW9vb0RGxtbFi9NRK8op3HbVT/36swuEltCr7Ln5RdnVCIiIpKEoUpERCQJQ5WIiEgShioREZEkDFUiIiJJGKpERESSMFSJiIgkYagSERFJwlAlIiKShKFKREQkCUOViIhIEoYqERGRJAxVIiIiSRiqREREkjBUiYiIJGGoEhERScJQJSIikoShSkREJAlDlYiISBKGKhERkSQMVSIiIkkYqkRERJIwVImIiCRhqBIREUnCUCUiIpKEoUpERCQJQ5WIiEgShioREZEkDFUiIiJJGKpERESSMFSJiIgkYagSERFJwlAlIiKShKFKREQkCUOViIhIEoYqERGRJCUK1ejoaLi6usLFxQUzZ84scrnNmzdDo9EgNjZWWgOJiIheFsWGqlarxciRI7Fz507ExcVh3bp1iIuLK7Dco0ePsGDBArRs2bJUGkpERFTeFRuqx48fh4uLC5ydnVG5cmX07dsXkZGRBZb797//jbFjx6JKlSql0lAiIqLyrthQTUlJgYODg3Lb3t4eKSkpesucPHkSSUlJ6NKly3NrhYeHw9vbG97e3khNTVXZZCIiovLJ6I5KeXl5GD16NL777rtilx0+fDhiY2MRGxsLW1tbY1+aiIioXCk2VO3s7JCUlKTcTk5Ohp2dnXL70aNHOH/+PPz9/eHk5ISjR48iODiYnZWIiOh/TrGh6uPjg/j4eCQkJCA7OxsREREIDg5WHre0tMSdO3dw9epVXL16Fa1atUJUVBS8vb1LteFERETlTbGhamJigrCwMAQFBaFx48YICQmBm5sbQkNDERUV9SLaSERE9FIwKclCnTt3RufOnfXumzJlSqHL7t+/3+hGERERvYw4oxIREZEkDFUiIiJJGKpERESSMFSJiIgkYagSERFJwlAlIiKShKFKREQkCUOViIhIEoYqERGRJAxVIiIiSRiqREREkjBUiYiIJGGoEhERScJQJSIikoShSkREJAlDlYiISBKGKhERkSQMVSIiIkkYqkRERJIwVImIiCRhqBIREUnCUCUiIpKEoUpERCQJQ5WIiEgShioREZEkDFUiIiJJGKpERESSMFSJiIgkYagSERFJwlAlIiKShKFKREQkCUOViIhIEoYqERGRJAxVIiIiSRiqREREkjBUiYiIJClRqEZHR8PV1RUuLi6YOXNmgcd//PFHuLu7w9PTE2+++Sbi4uKkN5SIiKi8KzZUtVotRo4ciZ07dyIuLg7r1q0rEJr9+vXDuXPncPr0aYwZMwajR48utQYTERGVV8WG6vHjx+Hi4gJnZ2dUrlwZffv2RWRkpN4yFhYWyu+PHz+GRqOR31IiIqJyzqS4BVJSUuDg4KDctre3x7Fjxwos9/3332Pu3LnIzs7Gvn375LaSiIjoJSCto9LIkSNx+fJlzJo1C1OnTi10mfDwcHh7e8Pb2xupqamyXpqIiKhcKDZU7ezskJSUpNxOTk6GnZ1dkcv37dsXW7duLfSx4cOHIzY2FrGxsbC1tVXRXCIiovKr2FD18fFBfHw8EhISkJ2djYiICAQHB+stEx8fr/y+fft2vP766/JbSkREVM4Ve07VxMQEYWFhCAoKglarxZAhQ+Dm5obQ0FB4e3sjODgYYWFh2LNnDypVqgRra2v8/PPPL6LtRERE5YpGCCHK4oW9vb0RGxtbFi9NRK8op3HbVT/36swuEltCr7Ln5RdnVCIiIpKEoUpERCQJQ5WIiEgShioREZEkDFUiIiJJGKpERESSMFSJiIgkYagSERFJwlAlIiKShKFKREQkCUOViIhIEoYqERGRJAxVIiIiSRiqREREkjBUiYiIJGGoEhERScJQJSIikoShSkREJAlDlYiISBKGKhERkSQMVSIiIkkYqkRERJIwVImIiCRhqBIREUnCUCUiIpKEoUpERCQJQ5WIiEgShioREZEkDFUiIiJJGKpERESSMFSJiIgkYagSERFJwlAlIiKShKFKREQkCUOViIhIEoYqERGRJAxVIiIiSUoUqtHR0XB1dYWLiwtmzpxZ4PG5c+eiSZMm8PDwwFtvvYXExETpDSUiIirvig1VrVaLkSNHYufOnYiLi8O6desQFxent0zz5s0RGxuLs2fPolevXhgzZkypNZiIiKi8KjZUjx8/DhcXFzg7O6Ny5cro27cvIiMj9ZYJCAhAtWrVAACtWrVCcnJy6bSWiIioHCs2VFNSUuDg4KDctre3R0pKSpHLL1u2DG+//Xahj4WHh8Pb2xve3t5ITU1V0VwiIqLyy0RmsTVr1iA2NhYxMTGFPj58+HAMHz4cAODt7S3zpYmIiMpcsaFqZ2eHpKQk5XZycjLs7OwKLLdnzx5MmzYNMTExMDU1ldtKIiKil0Cxh399fHwQHx+PhIQEZGdnIyIiAsHBwXrLnDp1CiNGjEBUVBRq1apVao0lIiIqz4oNVRMTE4SFhSEoKAiNGzdGSEgI3NzcEBoaiqioKADAV199hfT0dPTu3Ruenp4FQpeIiOh/QYnOqXbu3BmdO3fWu2/KlCnK73v27JHbKiIiopcQZ1QiIiKShKFKREQkCUOViIhIEoYqERGRJAxVIiIiSRiqREREkjBUiYiIJGGoEhERScJQJSIikoShSkREJAlDlYiISBKGKhERkSQMVSIiIkkYqkRERJIwVImIiCQp0fVUiYiIDOE0brtRz786s4uklrxY3FMlIiKShKFKREQkCUOViIhIEoYqERGRJAxVIiIiSRiqREREkjBUiYiIJGGoEhERScJQJSIikoShSkREJAlDlYiISBKGKhERkSQMVSIiIkkYqkRERJIwVImIiCRhqBIREUnCUCUiIpKEoUpERCQJQ5WIiEgShioREZEkJQrV6OhouLq6wsXFBTNnzizw+IEDB9CiRQuYmJhg06ZN0htJRET0Mig2VLVaLUaOHImdO3ciLi4O69atQ1xcnN4yr732GlauXIl+/fqVWkOJiIjKO5PiFjh+/DhcXFzg7OwMAOjbty8iIyPRpEkTZRknJycAQIUKPJpMRET/u4pNwZSUFDg4OCi37e3tkZKSourFwsPD4e3tDW9vb6SmpqqqQUREVF690F3L4cOHIzY2FrGxsbC1tX2RL01ERFTqig1VOzs7JCUlKbeTk5NhZ2dXqo0iIiJ6GRUbqj4+PoiPj0dCQgKys7MRERGB4ODgF9E2IiKil0qxoWpiYoKwsDAEBQWhcePGCAkJgZubG0JDQxEVFQUAOHHiBOzt7bFx40aMGDECbm5upd5wIiKi8qbY3r8A0LlzZ3Tu3FnvvilTpii/+/j4IDk5WW7LiIiIXjIcA0NERCQJQ5WIiEgShioREZEkDFUiIiJJGKpERESSMFSJiIgkYagSERFJwlAlIiKSpESTPxAR0avNadx2o55/dWYXSS15uXFPlYiISBLuqRKVI9xbIHq5cU+ViIhIEoYqERGRJAxVIiIiSRiqREREkjBUiYiIJGGoEhERScJQJSIikoShSkREJAlDlYiISBKGKhERkSQMVSIiIkkYqkRERJIwVImIiCRhqBIREUnCUCUiIpKEoUpERCQJQ5WIiEgSk7JuwP8Cp3HbjXr+1ZldJLWEiIhKE0OVGPpERJIwVImIqNwzZuP/RW7485wqERGRJAxVIiIiSRiqREREkjBUiYiIJGFHJZKOvYmJ6H9ViUI1Ojoan332GbRaLYYOHYpx48bpPZ6VlYX3338ff/31F2xsbLB+/Xo4OTmVRnuJypX/tQ2I/7X3S2SoYkNVq9Vi5MiR2L17N+zt7eHj44Pg4GA0adJEWWbZsmWwtrbGpUuXEBERgbFjx2L9+vWl2vD/ZfxiIyIqn4o9p3r8+HG4uLjA2dkZlStXRt++fREZGam3TGRkJD744AMAQK9evbB3714IIUqnxUREROVUsXuqKSkpcHBwUG7b29vj2LFjRS5jYmICS0tL3L17FzVr1pTc3BfnZRloTERE5YdGFLNLuWnTJkRHR2Pp0qUAgNWrV+PYsWMICwtTlmnatCmio6Nhb28PAGjQoAGOHTtWIFTDw8MRHh4OALhw4fR27EIAACAASURBVAIaNWok9c08T2pqKmxtbctdLdYrX/XKc9tYr/zUYr1Xv97zXL16FXfu3Cn8QVGMw4cPi44dOyq3p0+fLqZPn663TMeOHcXhw4eFEELk5OQIGxsbkZeXV1zpF8rLy6tc1mK98lWvPLeN9cpPLdZ79eupVew5VR8fH8THxyMhIQHZ2dmIiIhAcHCw3jLBwcH4+eefATzds23fvj00Go3xmwNEREQvkWLPqZqYmCAsLAxBQUHQarUYMmQI3NzcEBoaCm9vbwQHB+Nf//oXBg4cCBcXF9SoUQMREREvou1ERETlSsVJkyZNKm6h119/HaNGjcJnn32Gdu3aAQACAgLg6uoK4GnwhoSE4NNPP8WwYcNgbW1dqo1Wy8vLq1zWYr3yVa88t431yk8t1nv166lRbEclIiIiKhnO/UtERCQJQ5WIiEgShmoJ5OXl4fDhw1JrZmVlleg+IiJ6ebzSoTp27NgS3VecChUqYOTIkTKapPD19S3RfYaQ9X5ly8zMxNy5c/Huu++iZ8+emDdvHjIzM42ue/LkSSxcuBCLFi3CyZMnJbS0fDp06FCJ7jPEmTNnEBYWhrCwMJw5c0Z1nZs3byIqKgrbtm3DzZs3jWrTi3D//n2cPXvWqBqy1zutVovr16/j2rVryo8a6enp2LRpE+bNm4eFCxciOjoaeXl5Rrfv/v37ePjwodF1gKfr7ePHjwEAa9aswejRo5GYmCildnnxSndUatGiRYGV3sPDQ9U/1ZdffglfX1+8++67Ro3BvXnzJlJSUjBgwACsXbtWmSP54cOH+PDDD3HhwgXVtWW8X3d390LfnxACGo1G1WcXEhICc3NzDBgwAACwdu1apKWlYePGjQbX0pkyZQo2btyId999FwCwdetW9O7dG998843BtRYsWIDBgwfD3NwcQ4cOxalTpzBz5kx07NjRoDohISHYsGFDgc/QmM8OKPzvWth9JbVgwQIsWbJE+ex+/fVXDB8+HKNGjTKoztKlSzFlyhS0b98eQgjExMQgNDQUQ4YMUdUune3bt+Pvv//W2/AKDQ1VXc/f3x9RUVHIzc2Fl5cXatWqhTZt2mDu3LkG15K53gHAokWLMHnyZNSuXRsVKjzdx1GzrmzYsAFz5syBh4cH/vjjD7Ru3Rp5eXk4d+4cfvnlF7i7uxtU7/r16xg3bhwiIyORnp4OOzs7AMCQIUMwYcIEVKpUyaB6Oh4eHjhz5gzOnj2LQYMGYejQodiwYQNiYmIMrtWtW7cC31WWlpbw9vbGiBEjUKVKFVVtNFrZzTtRehYvXiyaNm0qqlWrJtzd3ZUfJycn0b9/f1U1zczMhEajEZUqVRLm5ubCzMxMmJubG1xn5cqVwt/fX5iZmYmAgADh7+8v/P39RXBwsNi8ebOqtuneb9WqVY1+v1evXn3ujxqNGzcu0X2GaNiwoXjy5IlyOyMjQzRs2FBVLQ8PDyGEENHR0eKdd94R58+fF82bNze4zvXr14UQRX+Ghjp8+LCYM2eOsLe3F999953yM3HiRKXNari7u4v09HTldnp6unB3dze4TsOGDcWdO3eU23fu3FH9N9AZMWKEGDhwoLC3txeTJk0STZs2FUOGDDGqpqenpxBCiCVLlojQ0FAhhFD1foWQu94JIUSDBg30PkO13N3dxePHj4UQQqSmpiqz4J05c0b4+voaXC8gIED88ccfQgghNm/eLD7//HORnp4uJkyYIIYNG6a6nbr/q8mTJ4ulS5fq3WeoTz/9VLz33nsiKipKREVFif79+4uPPvpIfPzxx2LAgAGq22isV/Ii5f369cPbb7+N8ePHY+bMmcr95ubmqFGjhqqajx49ktK2Dz74AAMHDsS6devQv39/KTVlvl9HR0fl91u3buHEiRMAgDfeeAO1atVS1b4WLVrg6NGjaNWqFQDg2LFj8Pb2VlVLp169esjMzFS2RrOyspStaUOJ//9owY4dOzBw4EC4ubmpuspS3bp1AfzfZ/jw4UPk5uaqahMAZGdnIz09Hbm5uXrrn4WFBTZt2qS6rhACFStWVG5XrFhR1fu1sbGBubm5ctvc3Bw2Njaq2wUAhw8fxtmzZ+Hh4YGJEyfiiy++wNtvv21UzdzcXNy4cQMbNmzAtGnTjKolc70DAAcHB1haWhrVJuDp37Rq1aoAgOrVq+P27dsAnu4Zqjl0e/fuXfj7+wMA3n33XUybNg3Vq1fH1KlTjZqz3dzcHDNmzMDq1avx559/Ii8vDzk5OapqHT58WPl+Ap7uufr4+ODEiRNwc3NT3UZjvZKhamlpCUtLS6xbtw4nT57EwYMHodFo0KZNG9WhCjw9txAfH693WEo3GYYhKlSogHnz5kkLVUtLS5iZmeHUqVN6oWiMDRs24KuvvoK/vz+EEBg1ahRmz56NXr16GVzrr7/+QuvWrfHaa68BAK5duwZXV1flMKkhh7pGjRoFjUYDS0tLuLm5oUOHDtBoNNi9ezfeeOMNg9sGPB0w3rFjRyQkJGDGjBl49OiRcihOjZ9++gkTJ05ElSpVlMNTGo0GV65cMaiOn58f3nzzTZw9exYTJ05U3Z5nDR48GC1btsQ777wD4OkhzH/9618G13FxcUHLli3RvXt3aDQaREZGwsPDQzmsOnr0aINr6oKhWrVquH79OmxsbHDjxg2D6+QXGhqKoKAgvPnmm/Dx8cGVK1fw+uuvq6pV1Hr36aefAgAWLlxoUD1nZ2f4+/ujS5cuMDU1Ve439LPr3LkzOnXqhHbt2iE6Ohq9e/cGANy7d0/VBpOtrS3WrFmDgIAAbNmyBU5OTgCehrcx52nXr1+PtWvXYvny5ahTpw6uXbuGr776SlWt9PR0XLt2Te97JT09HQBQuXJl1W001it9TvXbb7/Fhg0bpJz/WLp0KRYsWIDk5GR4enri6NGj8PX1xb59+1S1bdy4cahZsyb69OmD6tWrK/cbE/rdu3fHokWLlJXMGM2aNcPu3buVvdPU1FQEBgaq6tRSXEcEQzYEdHNMF0V3XV9D5OXl4fTp03B2doaVlRXu3r2LlJQUeHh4GFwLeDoD2ZEjR6Rd+tDX1xdHjhyRUktHt7EJAG3btkXz5s0NrjF58uTnPq5mQ+Dbb7/FqFGjsHfvXowcORIajQZDhw7Ft99+a3Ct0iB7/SvqM1Tz2e3YsQNxcXFo1qwZOnToAADKnmD+wC6Ja9eu4csvv0RcXBw8PT0xe/Zs1K1bF3fv3sX+/fvRs2dPg9unk5iYiPj4eAQGBiIjIwNarVbviEdJ7dixAx9++CEaNGgAIQQSEhKwePFi+Pv7Y8mSJfj8889Vt9EYr3Sourq64syZM8qhmidPnsDT0xMXL140uJa7uztOnDiBVq1a4fTp07hw4QK+/vprbNmyRVXb6tevX+A+NXsz+bVr1w6nTp3CG2+8oRfUUVFRBtdyd3fHuXPnlNt5eXlo1qyZ3n0ltWfPHgQGBurd9/PPP6sKwJLq2bMnNm/eXKJlhRD45ZdfcOXKFYSGhuLatWu4efOm6j3fTp06YcuWLahWrZqq5z/ro48+QkpKCnr37q33d9VtLJbUw4cPYWFhgXv37hX6uKEbdPkPg+rcuXNH2sZEVlYWMjMzVR8e1R3VKIqhe5WlSbeHZWZmVsYtKV1LlixBeHg47t27h8uXLyM+Ph4ffvgh9u7dq6peVlaW0rnT1dW17Don5fNKHv7VkXn+o0qVKnp1GjVqpCqcdRISElQ/tygyt+Y7deqEoKAgvPfeewCeHrbp3LmzqlpTpkzB5s2bMWfOHKSnp2Po0KEwNTUt1VA1ZOPk448/RoUKFbBv3z6EhobC3NwcPXv21DtfY4gZM2agdevWaNmypd4egtov8czMTNjY2OgdFdFoNAaHar9+/fDbb7/By8ur0N7Jhm7QvfHGGwgPD1fOlW/evBnjx4/HP//8Y1AdANi3bx/at29f5Eaqoe8VgHLe/tChQ4iLi0OfPn0AABs3bkSTJk0MqlVUr3gdtT27z58/j4EDByobOjVr1sSqVauknhN8dgNZLWN6nOt8//33OH78OFq2bAng6VEd3flfNf766y9cvXoVubm5ylG0999/36g2GuuVDlWZ5z/s7e2RlpaGHj16oEOHDrC2tjbq/GVOTg5++OEHHDhwAMDTbv8jRoxQ3VUdeHoOTlbnotmzZ2Pz5s3KeMjhw4cr5+AMFRMTg++++w6enp4AnoasLqxLiyHDno4dO4aTJ08qh0Ctra2RnZ2t+rVHjBiB9u3bw93d3ahzszorVqwwugYA/PbbbwDkbdD98ssvGDJkCPz9/XH9+nXcvXtX9emQmJgYtG/fHtu2bSvwmJoNCOD/DsX+8MMPOHjwIExMnn7dffjhh2jbtq1BtXSf3ffffw8AGDhwIICnYy2NGWI3fPhwzJ07FwEBAQCA/fv3Y9iwYQZPNlPUxogQQtr4YRkHNU1NTfXOd+bm5qr+/AYOHIjLly/D09NT6Xin0WgYqqXpnXfe0QsCXW82NX799VcAwKRJkxAQEIAHDx6gU6dOqut99NFHyMnJwccffwwAWL16NT766CMsXbpUdU2ZnYuAp4dQjTl3onP//n0cP34cDRo0QHJyMhITE5W9o/KgUqVK0Gq1SntSU1ONCsOcnBxVYyCLkpycjFGjRikbOG3btsWCBQtgb29vUJ3i9jJatGhhUD13d3dMmDABAwcOhLm5OQ4cOGBwm3QmT56MvLw8vP322wgJCVFVoyi6yQt0h7fT09Nx//59g2roNqB3796NU6dOKffPmjULLVq00Ot1b4jHjx8rgQo8/Y7STY5giD59+qB///6F/k/JmGgFALp06WJ0DT8/P0yfPh1PnjzB7t27sXjxYnTr1k1VrdjYWMTFxZWb7xHFCx7C88JlZGSICxcuSKn1559/iuXLlwshhLh9+7a4cuWK6lqFjTM0Zuyh7vm3bt1Sbt++fVt1zc2bNwsXFxdhYWFh1LhcIYR4/fXXxbJly4QQT/8eo0aNUjV2zhC6sYklsWbNGtGtWzdhZ2cnvv76a9GwYUOxYcMG1a89fvx48dNPP4nr16+Lu3fvKj9qBQYGiuXLl4ucnByRk5MjVqxYIQIDAw2uoxsTXdhPQECAwfWGDBki/Pz8xJUrV0R0dLRwdXUVYWFhBtfJz8vLy6jnF2b58uXitddeEx988IF4//33hZOTk1i5cqWqWs2aNRMHDx5Ubh88eFA0a9ZMddt69OghpkyZIhISEkRCQoL49ttvRY8ePQyu06JFC3Hu3LlCH7O3t1fdPtny8vJEeHi46NWrl+jZs6cIDw8XeXl5qmr16tVLGRtenrzSoRoVFSUaNmwonJychBBCnDp1SnTr1k1VrUmTJomuXbuK119/XQghREpKimjdurXqtjVv3lxcunRJuX358mXVg6B1mjZtqndbq9UWuK+kGjRoIOLi4oxqj05iYmKB+2JiYoyum5WVJc6cOSPOnj0rsrKy9B77/fffDar13//+V4SFhYlFixYZ/b6dnJxE/fr1C/yoVdiXtjFf5LLMmzdP7wsxLS3N6Ikaxo4dK2bPni2uXbsmZYNE58aNG2Lr1q1i69at4saNG6rr/PXXX8LDw0M4OjoKR0dH0axZM3Hy5EnV9e7duydGjRolmjdvLpo3by4+/fRTce/ePYPrHDhwoND/MyGEOHHihMH1dBvBQgiRlJQk2rdvL6ysrISvr6+4ePGiwfWEECI3N1e4urqqem5h/P39hZWVlejYsaPo1q2b8lPWXunev15eXti3bx/8/f2VQzZNmzbF+fPnDa7l6emJU6dOoUWLFkottVMeAsDevXsxePBgODs7QwiBxMRErFixQu9QkKG++uornD17Vq9zkbu7O/7zn/8YXKtNmzZGzy+rk5GRge+++w7Xrl3DkiVLEB8fj4sXL6Jr166qa27fvr1Ad/qffvrJoIkCiuoFq6N2eNOTJ0+wePFiZXx027Zt8eGHHypjMA311ltvYfDgwcrfdd26dVixYoXqHpMyz+fnHx7x5MkT5ObmqhoeoSOzV/yFCxfQqFGjIg97G3q4G4ByWF/3takbM+3l5aX0GXgV5O+UFBISgsDAQAwdOhSRkZEICwtTve7JHPZX1NSGfn5+Rtc2xisdqq1atcLRo0fRvHlzo4PwjTfewPHjx5WV7fHjx/D19TVqcu6srCylB7Grq6vBY8kKk79zUdu2bQ3uXKTr8BATE4ObN2+iR48eeu1S02GkT58+8PLywqpVq3D+/HlkZGSgdevWOH36tMG1dBo1aoTffvsNLi4uAIDLly+jS5cuBs2dXL9+fWg0Gr0vSEB9b1idkJAQWFhYKJN7rF27Fg8ePMCGDRtU1UtMTMSoUaNw5MgRaDQatG7dGgsXLlT9xTR06FDk5OQoHXlWr16NihUrGnw+X/bwCNmGDx+O8PDwQjdUNRqNqk5V/fr1Q2xsLIKDgyGEwG+//QYPDw9cvXoVvXv3xpgxY0pU5/PPP8f8+fMLnb8WUDcM7lnG9NbN/1xPT0+9/9X836eGkjnsr7x6pTsqubm5Ye3atdBqtYiPj8fChQvRunVrVbVCQkIwYsQIpKWlYcmSJVi+fDmGDRumum1arRa///670h18z549ANTNQpNfz5490aFDB2V6vHv37hm0x5W/92W1atWwa9cu5bbaXpiXL1/G+vXrsW7dOqWusdty5ubmSqACT2emMXQPqTSGNQFPh0nExcUptwMCAgwewpGfo6Oj1C+dEydO6E3i0b59ezRr1szgOrKHRwBPj2rMnTsX165dQ3h4uFFHNcLDwwEAf/zxh1Ftyi85ORknT55UxpNOnjwZXbp0wYEDB+Dl5VXiUNX1Hv7yyy+lte1ZxvyPJScn49NPP4UQAqmpqcjJyVGOZKidVhCQM+zvzTffxMGDB2Fubl7o0DBZV9RR65UO1UWLFmHatGkwNTVFv379EBQUpPpqEpUrV0ZgYCAsLCxw8eJFTJkyRZm1RI1u3bqhSpUq0oZdAPrT41WoUEHVHpes4Rv5Va5cGU+ePFH+AS5fvmz0Xrm3tzc6d+6MkJAQaDQabNy4ET4+PsqetiHhf+jQIXh6eqJ69epYs2YNTp48ic8//1z1nqDsuY4TEhKwaNEiZQNMR23QVqxYEZcvX0aDBg0APB3Tm38u4JKSOTxCZ/DgwfDy8lKGlNjZ2aF3795GnSoA/m9DJ39PWDVDL27fvq237laqVAm3bt1C1apVDVqnvby8AACnT5/GZ599pvfYggULpBzCNKa37uzZs5Xfvb29kZ6eDmtra9y8eRPBwcGq6xb3vkoye5huJjBZ87FLVwbncV9KEyZMEA0aNBC9e/cWO3fuVN1jTUftVTKex8XFRaSmpkqplZSUJHr06CFsbW2Fra2tePfdd0VSUpKqWrt27RLt2rUTNWvWFP369ROOjo7KFTDUGjRoUJE/gwcPNqiWu7u7yMvLE6dPnxaenp4iLCxMtGvXzuA2NW3aVLi7u4tGjRoJjUYjHB0dhZOTk9BoNEZdlcfDw0MsWLBA7Nu3T+zfv1/5UWvPnj3CwcFB+Pn5CT8/P+Ho6Cj27dtncJ2vvvpKTJs2Tbi6uopdu3aJHj16iK+//lp1u4T4v96/+XtvG9srftKkScLf31/UqlVLDBo0SNSuXVv07NlTVa0pU6aI5s2bi0mTJolJkyYJLy8vMXnyZJGeni769etncL3COica0nO9rE2fPl1qvZK89/wd2Ar7KWuvdKgGBgaK+/fvK7fv3bunXBJJjby8PBEdHS369OkjGjRoIMaPH6/Xg9cQY8aMMbiHanGCgoKUyz8ZS9YwDp07d+6I3377TWzbtq1A8J8/f97Y5hpF1uWoSuOyeUII8cYbb6h+bmGePHkipk6dKgICAsQ777wjpk+frnc5s5LSarXShkfo+Pr6ioyMDOXzv3TpkvDx8TGqZtOmTYVWq1XC+ebNm0atyydOnBDz588X8+fPV9WzVggh1q5dK7p27SqsrKz0eq76+/uL9u3bq25bYdSOACgJY0csqKmn613v5ORU4MeYXvayvNKHf+/cuQMrKyvltrW1tVHnfDQaDerUqYM6derAxMQE9+/fR69evdChQweDe9i2atUK77zzDvLy8lCpUiUp5wNkTo+XmpqKwYMHK7cHDRqE+fPnq26bjY1NkYejBg4caHCHin/++QcfffQRbt26hfPnz+Ps2bOIiopSdXhfdzmqNWvW4MCBA6ovRyXrCkHP+uyzzzB58mR07NhR7++qpvcq8PSwp4WFBf79738DeNqRauDAgQZfNL5ChQoYNmxYkX0LDJl/WWfy5Mno1KkTkpKS0L9/fxw6dAgrV640qMazqlatigoVKsDExAQPHz5ErVq1kJSUpLqet7e30ZcubN26NerWrYs7d+7giy++UO43NzdXdSGHFzGjUlH1X7TS6gshyysdqhUqVNC7NFBiYqLqcz4LFizAqlWrULNmTQwdOhSzZ89GpUqVkJeXh9dff93gUB09ejSOHDlS7JyihpA5PZ6NjQ3WrFmjN4zD2GtlFkXNP+awYcMwe/ZsjBgxAsDTXt39+vVTFaq6y1EtW7bM6MtRlYZz585h9erV2Ldvn/J3Vdt7FZDfkaooanpPd+jQQTknLYTAggULjJ6g39vbG2lpaRg2bBi8vLxgZmYGX19fo2oay9HREY6OjtKuPvQiZlQqjOzZjAz9LoiKitIbGmbsuXcpynI3ubRFR0cLBwcHMWDAANG/f3/x2muviejoaFW1QkNDizyEp2aygLZt2wqtVquqLUWReS7m6tWrolu3bqJmzZrC1tZWdO/evcjB5cZScwjJ29tbCKH/nsvDhAiloUGDBgUmtzBG//79xZEjR5TbR48eFQMHDpRWX0ftocHk5GRx6NAhERMTo/zIkpCQIM6cOaN3X1mefjhy5Ijw9vYW1atXF5UqVRIVKlRQNXNZWc2opOY758aNGyIyMlJERUUVmIijqPdQmLFjx4r27duLZcuWiWXLlonAwEAxfvx4g9sj2yu7p5qXl4cHDx7g5MmTOHr0KABg/vz5qrd6n3ftyMaNGxtcT3dx4rffftuoixPn9/bbbyM8PBzdunXTq6lmEgPZwzhkq1mzJi5fvqxsKW/atAl169ZVVevo0aMYNWoU/vvf/yI7OxtarRZmZmZ48OCBzCar1rRpU6Slpam+OMKzZF40XraxY8di/fr1cHNz09srb9eunZT6uott56fm9IMsn3zyCSIiItC7d2/ExsZi1apVqq7yM3/+fFhYWBT6mG7e8tKguxh6SS1duhRTpkxB+/btlfnJQ0NDMWTIEABP1/WS2rFjB06fPq2sJx988AGaN2+O6dOnG9Qm2V7pyR+8vb0RGxtb1s0olMyLE+vInI3mgw8+wIIFC5Rz0vfv38cXX3yB5cuXq25fUXSTdBjiypUrGD58OA4fPgxra2vUr18fv/zyi6rzmt7e3oV+sc2YMcPgWqXB398fZ8+ehY+Pj97GktqNHpkXjX8eNZMEuLq64uzZs1ImQikpYyYzMJbuOyr/pDSl2Z4ZM2Zg/PjxJV4+NTUVS5YsKTCcS+33gKurKw4fPqycSrp79y5at26t6jKaHh4e2L9/v7LTcO/ePeV/pSy9snuqABAYGIg5c+agT58+erN3qJ1+TqbiwnPUqFFYtGiRQTWLO4G/e/fuEo+tPXv2bIFOXmr/0d96660Cs+zkv8/QQAWejl8cPHgwAgICcO/ePVhYWODnn39GaGioqja6uLhAq9WiYsWKGDx4MJo3b15uQvV5R0nUkBWaCxYsKHSMpe6+WbNmGVzT2dkZOTk5LzRUy/IqJ9WqVUN2djY8PT0xZswY1K1bF3l5eaX2ehs3bjQoVLt37462bdsiMDBQ1VjmZ9nY2OhN0mJubq66r8b48ePRvHlzBAQEQAiBAwcOqL5akEyv9J6qzD23F03GBYGNqdmsWTPs378f1tbWAJ5uBfr5+Rl0sePMzExkZGQgICAA+/fvVzohPHz4EJ06dTJoSsFnderUCVZWVmjRooXeP3v+npQl1a5dO+zZswdDhw5FnTp1ULduXaxcuVJv1qHyrCQD5ktDYeuT2r2sUaNGQaPRICUlBWfOnMFbb70l5QLvJVEa/2sllZiYiFq1aiEnJwfz5s3DgwcP8PHHH+vNFiaToX+fZ6coNNb777+Pc+fOoXv37tBoNIiMjISHh4fS49nQ0183btzQu350nTp1pLVVrVd6T7W8d71+0QzZfvriiy/g6+uL3r17QwiBTZs2YcKECQa93k8//YT58+fj+vXr8PLyUl7fwsICn3zyiUG1npWcnIzo6GijauisXr0aeXl5CAsLw7x585CUlGTwUJCyVJq9Owuzbt06rF27FgkJCXqz6zx69Ej1USDdEBUvL68CM/aU9p5k/lmhXjTdUYOqVasadeqnpAz9LLt27YodO3agc+fOUl6/QYMGykxewNM9YUD97EgnTpxQev9qNBrV12aV6ZXeU5U5j+iLVtZ7qgAQFxenDNto37696mEXixYtwqhRo1Q9tyjDhw/HqFGj4O7uLqXekydPlA47L5sXvaeVmJiIhIQEjB8/Xu9wm26MpYmJ+m314g4pq9GtWze899576N69u95poLJU3FC60jovaOieqrm5OR4/fgxTU1Np4+llGTduHE6cOKFcuGLdunXw8fFhR6XSVBpXR3lRSqOzQkm+fEvrcmiHDx8u0NlBzdyrui+j3NxcxMfHw9nZGaampso/u5ovo23btuHLL79EdnY2EhIScPr0aYSGhpbr3s/5leXhy1u3bukdfjO2h7LMQ8o6MTExWL9+PbZv3w4fHx/07dsXXbt2RZUqVYxqqzFeVGexZ02fPh1ff/11qdQuidjYWEybNg2JiYl63wVq/m89PDz0ev9qtVo0b96cHZVKU2lcHUWWjRs3FuiOnv8+NVvmCPWgtgAAFENJREFUWVlZBTp45L+vsOEEz/Ly8pJ+ObSBAwfi8uXL8PT0VM5/ajQaVaH622+/Gfyc4kyaNAnHjx+Hv78/gKfnkV6mUwdltU5v3LgRX375Jfz9/ZXhEbNnz0avXr0MrlUah5R1/Pz84OfnB61Wi3379mHJkiUYMmRIme5tlVZoFtdbt6SBWhrXogWA/v37Y/bs2dIuJJKWlqasH+VlCNwrHaqlcXUUWWbMmFEgVPPfN2jQIINr+vr6FvgnyH9fUVOZ5Zc/TO7du4f4+Hijz9nFxsYiLi5Oyrmx0vgyqlSpEiwtLfXuK8seoYV53sXAV69eXSZtmjp1Kk6cOKHsnaampiIwMFBVqMqetu9ZT548wbZt27B+/XqcPHlSuZZsWct/+bLs7Gzk5OSgevXqqgNfVm/duXPnIjw8vNCOf8bM5mVra2vUVW7yK6+9f1/pUJ00aVKBeURL49Jmhti5cyd27NiBlJQUfPrpp8r9Dx8+VH0u6ubNm0hJScGTJ09w6tQpvV62GRkZqmouXboUCxYsQHJyMjw9PXH06FG0bt1a1QWomzZtips3b6qenKG0ybzubml49mLgycnJehcDN2TAvEx5eXl6h3ttbGxUDweRPW1ffiEhITh+/Dg6deqETz75BH5+ftIut2is/B10hBCIjIxUNcRMJyMjQ9VQpmeVxrVogafDw4YOHVqgd7ea6zS/99578Pf3V04/zJo1q1z0/n2lz6kCTwcX6+YRbdWqldHziBrrzJkzOHXqFCZOnIgpU6Yo95ubmyMgIEAZwmKIn3/+GStXrkRsbKzeRN/m5uYYNGiQqhXW3d0dJ06cQKtWrXD69GlcuHABX3/9dYn2dp8VEBCA06dP44033pAyeYFsGRkZmDZtmnJB9qCgIPz73/8uN0c1PD09lYuB684turu7GzS8qTSMGTMGZ86cUeaHXr9+PTw8PIz6Ut+yZQvGjh2L27dvQzy9ipbRHWN+//13aeMsXwRjziF/8803aN26tbTeuoC8/hAAMGDAAFy4cKHAjFlqJ5NISUkpcH5W1uxbar3SoVrcpANlKTc316hekoXZvHkzevbsKaWWj48PTpw4AU9PTxw7dgympqZwc3PD33//bXCtmJiYQu+XcSFmGXSdJ/J/cZT1dH35tWzZEseOHVO+bHNzc9GiRYsyb9/YsWPRsmVL5aLRbdu2xdGjR40KVRcXF2zbtk3V1J/P2rdvH9q3b1/khqCajU3Z8rctLy8PsbGxiImJUb3HLru3blH9IdSOG3Z1dVU1e1JhiprSsqw31l/Jw7+6SQfu3LmD+/fv6x0OTUlJKdO2lWZX+q5du2Lt2rUFtirVzDJkb2+PtLQ09OjRAx06dIC1tbXq85l+fn565wQzMjKg1WpV1SoN/fv3x5w5c9C0adNyc1gwPz8/P0yfPh1PnjzB7t27sXjx4nIxHm/37t2YNWuWXjhNnDjRqFCtXbu2lEAFnm7MtW/fHtu2bQNQsNNdeQhVXdsAwMTEBE5OToiMjFRdT+14z6LI7A8BPD13HhcXJ+WqSFu3bsXFixfLzRElnVdyT3XBggXKpAN2dnZ6kw4MGzbM6IkHjFGaXek7deoES0tLeHl5GT3LUH4xMTF48OABOnXqpGqg/LPnBOPj4/XOCZa1N998U9nbKo/y8vKwbNky7Nq1C0IIBAUFYejQoWXWmeqHH37A4sWLceXKFb2B/I8ePUKbNm2wZs0ag2vq9thiYmJw8+ZN9OjRw+hzbjqZmZnYvHlzgSMRaqe0LI9Kq7du7969sXDhQmn9IRo3bozLly+jfv36Rg+Fe/vtt7Fx40aYmZlJaZssr2So6pTGpAPlWdOmTXH+/PmybkYB5fWcoM7evXuxbt06KZ0nSsOWLVvQpUuXcrNF/uDBA9y/f7/QyR/UDn8ZPHhwkY8Zc84NKHxKS41GY9QVoWS5cuUKPvvsMxw9ehQajQa+vr6YN28enJ2dDaozfPhwhIeHIyAgoMBjanrrduvWDRqNBo8ePZLaH6KonQpDdibKckrLknilQxWQe5JdJtld6QH5swzJUl7PCerI7jwh2+DBg7Fv3z60a9cOffr0QadOnaSfj3+VldeNTeDpFZpGjhypdPaKiIjAokWLcOzYsTJtV0xMDIQQGDt2LP7zn/8o9+vuM6Z9Z86cwZ9//gng6Xn4Zs2aGfT8n3/++bmPl/VwqVc6VGWfZC8t+bvSGzPOqkmTJrh06ZKUQysyjRkzBlZWVli1ahUWLVqExYsXo0mTJpg2bVqZtktHZueJ0pKTk4OdO3di/fr1OHjwIDp06IClS5eWdbOkK2zPbf78+YVeHKOkyuvGJgC9S77pNGvWzKiLOfx/7d1/TNT1HwfwJwIN+SG4MVvQWrnmdQiK8sNDE7ko9NQVtR1XUSajwnPZauIkbY1WyzY2CBfRL2eSUGcszXLaMs9I4Ii7Y2NEHO1K0BEmDnRw3Hkd9/2D3Scu8Fv3uffx+dznXo/NJYf77KVc9/p83u/36/Vi+SAxV4eruWL+r2pra/HRRx9xq0DHjx/nfj6+mpiYQFRUFPfZ7nK54HA4EB0dzSs2ViSdVOVyOdNN9kDztx0bi6WVQBDbnuA/lZSUYM+ePUwOTwSS0+nEmTNncPjwYbS0tGBkZETokJhj+eQWiJaWrO3duxeLFy/G448/jrCwMOh0OoyOjmLPnj0AfG8LyupBIhD75sB0Qm5vb+d6ME9MTCAnJ4fXz0KhUODs2bPcnur4+DgKCgrQ1tbGKzZWJJ1UWW+ys8T6KP1Mf/75p1cXpLvuusvva/pDrHeUHiwPTwSC5wn1/PnzyMvLQ1FREQoKCiS5BMzyyU2o/rq++H9P4HzagrJ6kAjEvjnwd/27p++y3W5HVlYWr/MVc42lYz2qjg/p/V85w8jICFJSUkTZdID1UXpg+u+1e/duDA0NYcmSJRgYGIBcLudVW8pSfn6+1x3l5OSkKO4oPViNkAuUhoYGaDQafPDBB6I5rBQoKpUKb7/9tteT2+bNm7lBD758oIshaf4b1j2mWXUvi4+PR3x8PNc3nZWSkhKsWbMGjz76KIDpspjS0lJe14qJiYHZbOZONhuNRixcuJBZrHxJ+klV7E0HWFu5ciXOnTuHBx98EF1dXdDr9Th69CgOHTokaFxivaMk4sP6yU3snE4n6uvruZmgeXl5KCsrQ2RkpE/XCdRp3UAwm81eDUNWrVrF6zpGoxEajQZJSUkApgeW63Q6ZGRkMIuVD0k/qYoxeXqOg9+KP4eoIiMjuf6rU1NTUCqVeOmll3hfj5V/3lGaTCZR3FGKnad+duZJcQCimmnJWjBNB2JBq9XC6XRi586dAKaHI2i1Wp8PoZWXl3Mnc0+cOMG97nlNLAwGA5YvX859Fty4cQMdHR1Ys2aNz9f6/fff0dXVhcHBQXz55Zfo6OgQxTkNSSZVMX8YeXrztra2ore3FxqNBsD0GC1/D8okJCRgfHwcubm5KC4uxpIlS0QxlLm2thZqtRpJSUlwu90YHh6GTqcTOizRa2hoAMC+S44YBUNLwUDo7Oz02i9+4IEHfC4xAf5+gHA6nbMeJiYnJ/0LkiGtVut1mjg2NnbWa//VG2+8AbVajbGxMej1epSXl0Or1QpejiTJpOpZWhDjh5Gnhqq+vh4XLlzgDpvs2LED69ev53VNz8zUr776ClFRUaipqUFjYyOuX78ueNcYl8uFH3/8EX19fVzZikwm83l5KxSp1WqYTCbR9KsOpJaWFq6loGee78z/SjWphoeHw2q1cidsf/vtN16N/2ee1p05Ks9zWlcsPD9PjwULFniV/vjC8+906tQpPPfcc9iyZQteffVVJnH6Q5JJNRiMjo7ixo0b3MGL8fFxjI6O8rqWZ2bqjh07uNmaQhdAe4SHh+Ozzz7Dyy+/LNiIsmA1NTWFt956C/39/aiurp71fTF0BGIlLi4O1dXVSE1N5ZIpIL65tqxVVVVBqVRyHZQuXrzIazzlk08+CZVKxfy0LmtLly7FwYMHodVqAQDvvfeez92jPJKTk1FWVobvvvsOe/fuhcPh4D16kCVKqgKpqKiYNWC3srKS17Vu3ryJpqYmtLW1zbl8JvRd/rp16/DCCy9Ao9F4LUfz7UcaKj7//HOcOHECf/31lyhXXVgaHx8HAFgsFnR2duKRRx6B2+3G119/jezsbIGjC5x169ahrKwM33//PRISErBx40bk5OT4fJ1AndZl7f3338eLL76IN998E2FhYcjPz+dmt/rq2LFjOHPmDMrLy5GQkIA//vgDVVVVjCP2naRP/4rd0NAQPv30U8jlcthsNiQlJfGaBXjhwgU0Njbi2LFjePjhh72+J4Z2e6z6kYaq06dPQ6VS3fL7R44cEc3KhL9yc3Nx6tQpxMXFAZhevtyyZQt3OlZqioqKsGjRIhQXFwMAmpqaMDY2hi+++ELgyIRx4MABvPLKK0KH4RdKqgL5+OOPUVtbi8uXLyM9PR0GgwE5OTl+JZpDhw7xrvkiwWuuVnLBSiaTobu7mysHcTgcWLFihejbSPKVkpKC3t7ef30tVEjhvUzLvwKpra1FZ2cnFAoF9Ho9+vr6sG/fPr+uWVpaKsoBAleuXMG+ffswNDSE06dPo7e3F+3t7XQDwIiU7ou3bduG7Oxsr+YA27dvFzaoAFq9ejUMBgMUCgUAoKOjg6sQCEVSeC9TUhVIVFQU16rL4XDgvvvu8/tu/FZ9P4VOqtu3b0dJSQnXQH/ZsmXQaDSUVBmR0mGe/fv3Q6VScVNMDh8+zLs5QDAwmUxYu3Yt10p0cHAQMpmM61ssllaZ80UK72VKqgK58847MTY2hsLCQjz00ENYvHix323VjEajKAcIjIyMoKioCAcOHAAw3ZaRT9kAmZsU7u5nWr16dcgcYhN7i8z5JoX3MiVVgRw/fhwAUFlZCaVSievXr2PTpk1+XZNV30/WYmJicO3aNS7ZGwwGxMfHCxxVcJiamkJzczOKiopu+WfEVIdIfBMM/Ynnk1qtFjoEv9FBJQlRKpWi7PtpNpuxa9cu/Pzzz1i+fDmuXr2K5uZmryJ1cmuZmZkwGo1Ch0GI3/r7+6HVanHlyhX09PSgu7sbJ0+eFEXTBlYoqUqIWAcI2O12vPvuu/j2228RFxeHnJwc7Nq1i9tTJv9fRUUFEhMTZ9X5iqmon5D/YsOGDaiqqkJZWRk3Ozo1NRU9PT0CR8YOJVUScFSL55+5JrdIcWILkb6srCx0dnZi1apVXFKV2sQq2lOVADEPEACAnp4er7o7pVLp9/CAUBJqk1uIdCUmJsJqtXKfU83NzaI7A+IvSqoSIOYBAgDV4vnLZrOhuroag4OD+PDDD/Hrr7/CYrFg69atQodGiE/q6urw/PPPo6+vD8nJybjnnntw9OhRocNiipZ/ScDJ5XJYLJZZtXgREREhWYvnK41Gg4yMDDQ0NKCnpwc2mw1r166V1JIZCS0TExOYmpri2lFKCT2pkoCjWjz/WK1W6HQ6rll6dHS0JOr5SOgZGxtDQ0PDrK5vBw8eFDAqtiipkoCjWjz/3HbbbZicnOT2oaxWq1fJFCHBYvPmzVAoFEhLS8OCBQuEDicgKKkSInKvv/46Nm3ahEuXLqG4uBitra345JNPhA6LEJ/Z7fY5ZwNLCe2pEhIErl27BoPBALfbDYVCgcTERKFDIsRnNTU1iI2NxdatW71WW6RUc01JlRCR+rcRWKHSH5dIR11dHfbv34+EhARuO0NqNdeUVAkRKc9wd7vdDqPRiJUrV8LtdqO7uxuZmZlob28XOEJCfLN06VL89NNPkl5pkeZOMSESoNfrodfrcccdd8BsNsNoNMJkMqGrqwvJyclCh0eIz+69915ER0cLHUZA0UElQkTOYrEgLS2N+zo1NRW//PKLgBERwk9MTAzS09OhVCq99lSppIYQMm9WrFiBZ599Fk899RQAoLGxkSb8kKBUWFiIwsJCocMIKNpTJUTk7HY76uvr0dLSAgDIzc2FVqulKT8kKN28eRP9/f0AAJlMhsjISIEjYouSKiGEkHlx/vx5PPPMM7j77rvhdrtx6dIlHDlyBLm5uUKHxgwlVUJErrW1FZWVlRgYGPBq7SalMgQSGjIyMtDU1ASZTAZgemj5E088AZPJJHBk7NCeKiEiV1paipqaGmRkZCA8PFzocAjhzel0cgkVAJYtWwan0ylgROxRUiVE5OLj46FSqYQOgxC/ZWZmzjp0J7UxkLT8S4jIVVRUwOVy4bHHHvMqQ6COSiTYOBwO1NXVcTOg169fj507d0pqQAQlVUJEztNZydPWze12IywsDOfOnRMyLEJ8NjExgaioKG4bw+VyweFwSKohBC3/EiJyeXl5s17zJFhCgkl+fj7Onj2L2NhYAMDk5CQKCgrQ1tYmcGTsUFIlROQ8H0DAdM3qN998A7lcLmBEhPBjt9u93s+xsbGw2WwCRsQeJVVCRG737t1eX5eXl2Pjxo0CRUMIfzExMTCbzdx5AJPJhIULFwocFVuUVAkJMjabDZcvXxY6DEJ89s4770CtViMpKQlutxvDw8PQ6XRCh8UUJVVCRC4tLY3bQ3W5XLh69Spee+01gaMixHdZWVno6+uDxWIBQG0KCSECGBgY4H4fERGB22+/HRERdD9MglNbWxsuXrzo1R1s27ZtAkbEFiVVQggh8+Lpp5+G1WpFeno6V1YTFhYmqdFvlFQJIYTMC7lcjt7eXkmXhC0QOgBCCCGhITU1FcPDw0KHEVC0MUMIIWRejIyMICUlBdnZ2V6tCU+ePClgVGzR8i8hhJB58cMPP8z5+oYNG+Y5ksChpEoIIYQwQnuqhBBCAur+++8HAMTFxWHRokXcL8/XUkJPqoQQQggj9KRKCCGEMEJJlRBCCGGEkiohhBDCCCVVQgghhBFKqoQQQggj/wNlddue7x1eaQAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qV5-9ONOvHNe",
        "colab_type": "text"
      },
      "source": [
        "The most important features are HIV/AIDS, income, adult mortality, alcohol use, the under-five mortality rate, the thinness rate for children 5-9, and schooling. It makes sense that these would have an impact on life expectancy.\n",
        "\n",
        "This agrees for the most part with the results for project #1 when using Lasso. The intercept and GDP are not important here even though they were included when using the Lasso method, while alcohol usage was included here but not using Lasso. This could be because of correlations between features that results in one being included in Lasso and the other being included when using gradient boosted trees.\n",
        "\n",
        "XGB returns a score on the test set of 7.4, compared to 12.3 using linear regression - the model is more accurate."
      ]
    }
  ]
}