support_vector_machines.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "support_vector_machines.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.8.0"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/VasiaKoum/Pattern-Recognition-ML/blob/main/support_vector_machines.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "a63bXjKFI-2f"
      },
      "source": [
        "ΕΠ08 Αναγνώριση Προτύπων – Μηχανική Μάθηση<br>\n",
        "# 3η Εργασία: Support Vector Machines\n",
        "__ΚΟΥΜΑΡΕΛΑ ΒΑΣΙΛΙΚΗ 1115201600074__<br><br>\n",
        "Σε αυτή την εργασία καλείστε να διερευνήσετε την επίδοση των support vector machines\n",
        "(SVMs) στην ταξινόμηση εικόνων, και πιο συγκεκριμένα στην ταξινόμηση εικόνων οι οποίες\n",
        "απεικονίζουν χειρόγραφα ψηφία."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "q2TnEO9IJhr3"
      },
      "source": [
        "## Σύνολο Δεδομένων (Dataset)\n",
        "*Καλείστε να διερευνήσετε την επίδοση των support vector machines στο πρόβλημα της\n",
        "αναγνώρισης χειρόγραφων ψηφίων. Για το σκοπό αυτό θα χρησιμοποιήσετε το σύνολο\n",
        "δεδομένων MNIST και υλοποιήσεις αλγορίθμων της βιβλιοθήκης scikit-learn. Το σύνολο\n",
        "δεδομένων MNIST αποτελείται από 70000 εικόνες χειρόγραφων ψηφίων και, τυπικά,\n",
        "χωρίζεται σε τρία υποσύνολα: training set (50000 εικόνες), validation set (10000 εικόνες),\n",
        "test set (10000 εικόνες). Κάθε εικόνα έχει διάσταση 28 x 28 pixels και απεικονίζει ένα\n",
        "χειρόγραφο ψηφίο*\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1X9QVR8kJ5Vw"
      },
      "source": [
        "from keras.datasets import mnist\n",
        "from sklearn.pipeline import Pipeline\n",
        "from sklearn.model_selection import GridSearchCV\n",
        "from sklearn.preprocessing import StandardScaler, Normalizer\n",
        "from sklearn.model_selection import KFold\n",
        "from sklearn.decomposition import PCA\n",
        "from sklearn.svm import SVC\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "# from google.colab import files\n",
        "import pandas as pd\n",
        "import joblib\n",
        "import time\n",
        "\n",
        "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
        "print(\"Train size set: \", len(x_train), \"- Test size set: \", len(x_test))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HDJoaVo2JncW"
      },
      "source": [
        "\n",
        "\n",
        "*Ζητείται να φορτώσετε τα δεδομένα του συνόλου MNIST και να μετατρέψετε κάθε\n",
        "εικόνα σε μορφή διανύσματος διάστασης 28 x 28 = 784. Στη συνέχεια\n",
        "κανονικοποιήστε (normalize) τα δεδομένα στο διάστημα [0,1].*\n",
        "<br>\n",
        "<br>\n",
        "Η κανονικοποίηση θα γίνει με τη χρήση του Normalizer από τη βιβλιοθήκη της sklearn. Μετατρέπουμε τα δεδομένα σε 2D (60000, 784) ώστε να τα χρησιμοποιήσουμε χωρίς σφάλμα στο training\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2QaApmc4J55E"
      },
      "source": [
        "def dim_3dto2d(dataset):\n",
        "    dataset_reshaped = dataset\n",
        "    if len(dataset.shape) == 3:\n",
        "        samples, x1, x2 = dataset.shape\n",
        "        dataset_reshaped = dataset.reshape((samples,x1*x2))\n",
        "    return dataset_reshaped\n",
        "\n",
        "x_train = dim_3dto2d(x_train)\n",
        "x_test = dim_3dto2d(x_test)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HWyloApGJqj7"
      },
      "source": [
        "## Ταξινομητής SVM\n",
        "*Στα SVMs υπάρχουν διάφορες επιλογές που μπορεί να επηρεάσουν την απόδοση\n",
        "τους σε προβλήματα ταξινόμησης. Παραδείγματα τέτοιων επιλογών αποτελούν ο\n",
        "τύπος του πυρήνα (kernel) και οι τιμές των διάφορων παραμέτρων. Ζητείται να\n",
        "εξετάσετε την επίδοση των SVMs για διαφορετικές τιμές παραμέτρων/πυρήνων ώστε\n",
        "να καθορίσετε το συνδυασμό παραμέτρων/πυρήνων που οδηγούν στη μεγαλύτερη\n",
        "ακρίβεια ταξινόμησης. Για αυτό το πείραμα να χρησιμοποιήσετε 60000 εικόνες για\n",
        "εκπαίδευση (training) και 10000 παραδείγματα για δοκιμές (test). Να αναφέρετε τις\n",
        "τιμές των παραμέτρων, δηλαδή τύπο πυρήνα, τιμές των C και gamma που οδηγούν\n",
        "στις καλύτερες επιδόσεις τόσο στο σύνολο εκπαίδευσης όσο και στο σύνολο δοκιμής\n",
        "(test set).*\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "X4qZ1NwsZNI8"
      },
      "source": [
        "def data_subset(dataset, per_train, per_test):\n",
        "    train_len = int(len(dataset[0])*per_train/100)\n",
        "    test_len = int(len(dataset[1])*per_test/100)\n",
        "    x_train = dataset[0][:train_len]\n",
        "    x_test = dataset[1][:test_len]\n",
        "    y_train = dataset[2][:train_len]\n",
        "    y_test = dataset[3][:test_len]\n",
        "    return x_train, x_test, y_train, y_test\n",
        "\n",
        "def SVM_classifier(dataset=[], per_train=100, per_test=100, preproc_method=\"Normalizer\", params=[]):\n",
        "    if preproc_method != \"Normalizer\" and preproc_method != \"Scaler\" or not (1 <= per_train <= 100) or not (1 <= per_test <= 100):\n",
        "        print(\"Run with: preproc_method = 'Normalizer' or 'Scaler', per_train in [1,100], per_test in [1,100]\")\n",
        "        return\n",
        "    start = time.time()\n",
        "\n",
        "    if len(dataset) != 4:\n",
        "        print(\"Run with dataset=[x_train, x_test, y_train, y_test]\")\n",
        "        return\n",
        "\n",
        "    x_train, x_test, y_train, y_test = data_subset(dataset, per_train, per_test)\n",
        "    print(\"Training set size: \", len(x_train), \" - Test set size: \", len(x_test))\n",
        "\n",
        "    transformer = Normalizer()\n",
        "    if preproc_method == \"Scaler\":\n",
        "        transformer = StandardScaler()\n",
        "    pipeline = Pipeline(steps=[('transformer', transformer), ('SVM', SVC())])\n",
        "    parameters = params\n",
        "    cv = KFold(n_splits=5, shuffle=True, random_state=42)\n",
        "    grid = GridSearchCV(pipeline, param_grid=parameters, cv=cv)\n",
        "    grid.fit(x_train, y_train)\n",
        "    print(\"Score = %3.2f\" %(grid.score(x_test, y_test)))\n",
        "    y_pred = grid.predict(x_test)\n",
        "\n",
        "    exe_time = (time.time() - start)/60\n",
        "    print(\"Execution time\", exe_time)\n",
        "    dataset_subset = [x_train, x_test, y_train, y_test, y_pred]\n",
        "    return grid, dataset_subset"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9OCCjqHilkBs"
      },
      "source": [
        "Επιλογή ποσοστού για το training & testing του SVM επί του συνόλου δεδομένων:\n",
        "$per$_ $train$, $per$_ $test$\n",
        "<br>\n",
        "π.χ. $per$_ $train=20$ τότε θα χρησιμοποιηθούν $60000*20/100=12000$"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qQMi9dAImOTz"
      },
      "source": [
        "set_per_train = 100\n",
        "set_per_test = 100"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8bvHi7Uey3SC"
      },
      "source": [
        "Θα χρησιμοποιηθούν οι παρακάτω συναρτήσεις (για kernels):<br>\n",
        "* Linear Function $K(x_i, x_j)=x_i*x_j$\n",
        "* Polynomial(poly) Function $K(x_i, x_j)=(1+x_i*x_j)^d$\n",
        "* Gaussian radial basis Function(rbf) Function $K(x_i, x_j)=exp(-\\gamma ||x_i-x_j||)^d$\n",
        "<br><br>\n",
        "Η παράμετρος gamma $(\\gamma)$ χρησιμοποιείται μόνο για την Gaussian radial basis Function."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "QqC_69qLdHPc",
        "outputId": "e0ea282d-4ce6-424e-85ec-451b26db8022"
      },
      "source": [
        "dataset_ = [x_train, x_test, y_train, y_test]\n",
        "\n",
        "# parameters = [ {\"SVM__kernel\": [\"linear\", \"poly\"], \"SVM__C\": [1, 10, 100, 1000]}, \n",
        "#            {\"SVM__kernel\": [\"rbf\"], \"SVM__C\": [1, 10, 100, 1000], \"SVM__gamma\": [10,1,0.1,0.01]} ]\n",
        "\n",
        "parameters = [ {\"SVM__kernel\": [\"linear\", \"poly\"], \"SVM__C\": [0.1, 1, 10, 100]}]\n",
        "\n",
        "grid, dataset_subset = SVM_classifier(dataset=dataset_, per_train=set_per_train, per_test=set_per_test, params=parameters)\n",
        "joblib.dump(grid, 'mnist_svm_classifier.pkl')\n",
        "# files.download('mnist_svm_classifier.pkl')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training set size:  60000  - Test set size:  10000\n",
            "Score = 0.99\n",
            "Execution time 170.43815218210221\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['mnist_svm_classifier.pkl']"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jtGRTui1vGBl"
      },
      "source": [
        "Θα γίνει εκπαίδευση του SVM classifier και το GridSearchCV θα αποθηκευτεί σε αρχείο pickle"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 97
        },
        "id": "BPxcyMAyupgT",
        "outputId": "c4b5ad91-73d7-45e1-d9cf-e052b5aadf8c"
      },
      "source": [
        "grid = joblib.load(\"mnist_svm_classifier.pkl\")\n",
        "print(\"Best score\", grid.best_score_,\"with best parameters\", grid.best_params_)\n",
        "df = pd.DataFrame(grid.cv_results_)\n",
        "if 'param_SVM__gamma' in df.columns:\n",
        "    df = df[['param_SVM__kernel', 'param_SVM__C', 'param_SVM__gamma', 'mean_test_score', 'mean_fit_time', 'mean_score_time']]\n",
        "else:\n",
        "    df = df[['param_SVM__kernel', 'param_SVM__C', 'mean_test_score', 'mean_fit_time', 'mean_score_time']]\n",
        "df.sort_values(by=['mean_test_score'], ascending=False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Best score 0.9851666666666666 with best parameters {'SVM__C': 10, 'SVM__kernel': 'poly'}\n"
          ],
          "name": "stdout"
        },
        {
          "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>param_SVM__kernel</th>\n",
              "      <th>param_SVM__C</th>\n",
              "      <th>mean_test_score</th>\n",
              "      <th>mean_fit_time</th>\n",
              "      <th>mean_score_time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>5</th>\n",
              "      <td>poly</td>\n",
              "      <td>10</td>\n",
              "      <td>0.985167</td>\n",
              "      <td>151.403805</td>\n",
              "      <td>57.376795</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7</th>\n",
              "      <td>poly</td>\n",
              "      <td>100</td>\n",
              "      <td>0.985167</td>\n",
              "      <td>153.981053</td>\n",
              "      <td>57.455523</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>poly</td>\n",
              "      <td>1</td>\n",
              "      <td>0.984133</td>\n",
              "      <td>153.064651</td>\n",
              "      <td>55.753010</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>poly</td>\n",
              "      <td>0.1</td>\n",
              "      <td>0.970667</td>\n",
              "      <td>263.845512</td>\n",
              "      <td>79.912391</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>linear</td>\n",
              "      <td>10</td>\n",
              "      <td>0.941633</td>\n",
              "      <td>119.636093</td>\n",
              "      <td>52.653587</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>linear</td>\n",
              "      <td>1</td>\n",
              "      <td>0.939917</td>\n",
              "      <td>151.433372</td>\n",
              "      <td>66.311614</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6</th>\n",
              "      <td>linear</td>\n",
              "      <td>100</td>\n",
              "      <td>0.933733</td>\n",
              "      <td>154.234085</td>\n",
              "      <td>46.046697</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>linear</td>\n",
              "      <td>0.1</td>\n",
              "      <td>0.927300</td>\n",
              "      <td>306.041010</td>\n",
              "      <td>110.010452</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "  param_SVM__kernel param_SVM__C  mean_test_score  mean_fit_time  \\\n",
              "5              poly           10         0.985167     151.403805   \n",
              "7              poly          100         0.985167     153.981053   \n",
              "3              poly            1         0.984133     153.064651   \n",
              "1              poly          0.1         0.970667     263.845512   \n",
              "4            linear           10         0.941633     119.636093   \n",
              "2            linear            1         0.939917     151.433372   \n",
              "6            linear          100         0.933733     154.234085   \n",
              "0            linear          0.1         0.927300     306.041010   \n",
              "\n",
              "   mean_score_time  \n",
              "5        57.376795  \n",
              "7        57.455523  \n",
              "3        55.753010  \n",
              "1        79.912391  \n",
              "4        52.653587  \n",
              "2        66.311614  \n",
              "6        46.046697  \n",
              "0       110.010452  "
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CxdThQc7JvOr"
      },
      "source": [
        "## Εφαρμογή PCA\n",
        "*Στη συνέχεια, να εφαρμόσετε PCA στα δεδομένα επιλέγοντας 3 διαφορετικές τιμές\n",
        "για τη διατηρούμενη διακύμανση και για κάθε τιμή διακύμανσης εκτελέστε ξανά τη\n",
        "μέθοδο SVM χρησιμοποιώντας τις παραμέτρους που οδήγησαν στην καλύτερη\n",
        "επίδοση στο παραπάνω ερώτημα. Για κάθε εκτέλεση, αναφέρετε τον αριθμό των\n",
        "συνιστωσών (components) που διατηρούνται καθώς και την ακρίβειας ταξινόμησης.\n",
        "Επίσης, καταγράψτε τους χρόνους εκτέλεσης κάθε πειράματος και εξαγάγετε\n",
        "συμπεράσματα σχετικά με μια πιθανή αντιστάθμιση (trade-off) μεταξύ ακρίβειας\n",
        "ταξινόμησης, μείωσης διαστάσεων και χρόνου εκτέλεσης του αλγορίθμου.*\n",
        "<br>\n",
        "<br>\n",
        "Εάν θέλουμε η **διατηρούμενη διακύμανση** να είναι σε κάποιο συγκεκριμένο ποσοστό, ορίζουμε ως number_comp = %ζητούμενη_διακύμανση, αλλιώς εάν επιθυμούμε συγκεκριμένο αριθμό συνιστωσών θα πρέπει να ισχύει: number_comp>0\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "B6pf0ceESAmL"
      },
      "source": [
        "def plot_variance(pca, number_comp):\n",
        "    cumExplainedVar=np.cumsum(pca.explained_variance_ratio_)\n",
        "    floor = np.max(cumExplainedVar)\n",
        "    max_components = np.argwhere((cumExplainedVar >= floor) & (cumExplainedVar < (floor+0.1*floor))).flatten()[0]\n",
        "    plt.figure(figsize=(16, 5))\n",
        "    plt.subplot(1, 2, 1)\n",
        "    plt.bar(range(number_comp),pca.explained_variance_ratio_[:number_comp])\n",
        "    plt.xlabel('Αριθμός Συνιστωσών')\n",
        "    plt.ylabel('Διατηρούμενη Διακύμανση');\n",
        "\n",
        "    plt.subplot(1, 2, 2)\n",
        "    for i in np.arange(0.8,0.9,0.001):\n",
        "        plt.axhline(y=i, color= \"whitesmoke\")\n",
        "    plt.plot(cumExplainedVar)\n",
        "    plt.plot(max_components,cumExplainedVar[max_components],'ro') \n",
        "    plt.xlabel('Αριθμός Συνιστωσών')\n",
        "    plt.ylabel('Συσσωρευτική Διατηρούμενη Διακύμανση');\n",
        "    plt.show()\n",
        "\n",
        "def PCA_nComponents(x_train, x_test, number_comp, grid, plot):\n",
        "    x_train_pca = PCA(n_components=number_comp, svd_solver='full').fit(x_train)\n",
        "    x_train_components = x_train_pca.transform(x_train)\n",
        "    x_test_pca = PCA(n_components=x_train_pca.n_components_, svd_solver='full').fit(x_test)\n",
        "    x_test_components = x_test_pca.transform(x_test)\n",
        "\n",
        "    if plot:\n",
        "        plot_variance(x_train_pca, x_train_pca.n_components_)\n",
        "\n",
        "    print (\"Components = \", x_train_pca.n_components_ ,\";\\nTotal explained variance = \", round(x_train_pca.explained_variance_ratio_.sum(),5)  )\n",
        "    dataset_ = [x_train_components, x_test_components, y_train, y_test]\n",
        "\n",
        "    fix_params = {}\n",
        "    for key, value in grid.best_params_.items():\n",
        "        d = {key:[value]}\n",
        "        fix_params.update(d)\n",
        "\n",
        "    grid_pca, dataset_subset = SVM_classifier(dataset=dataset_, per_train=set_per_train, per_test=set_per_test, params=fix_params)\n",
        "    joblib.dump(grid_pca, 'mnist_pca_svm_classifier.pkl')\n",
        "    # files.download('mnist_pca_svm_classifier.pkl')\n",
        "    # grid_pca = joblib.load(\"mnist_svm_classifier.pkl\")\n",
        "    print(\"Best score\", grid_pca.best_score_,\"with best parameters\", grid_pca.best_params_)\n",
        "    df = pd.DataFrame(grid_pca.cv_results_)\n",
        "    if 'param_SVM__gamma' in df.columns:\n",
        "        df = df[['param_SVM__kernel', 'param_SVM__C', 'param_SVM__gamma', 'mean_test_score', 'mean_fit_time', 'mean_score_time']]\n",
        "    else:\n",
        "        df = df[['param_SVM__kernel', 'param_SVM__C', 'mean_test_score', 'mean_fit_time', 'mean_score_time']]\n",
        "    df.sort_values(by=['mean_test_score'], ascending=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8UjzpWGSs3mz"
      },
      "source": [
        "Εκτέλεση της PCA_nComponents(δεδομένα εκπαίδευσης, test δεδομένα, αριθμός συνιστωσών, το GridSearchCV της παραπάνω εκπαίδευσης, True ή False για την εκτύπωση γραφήματος). "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "fOkL12ERuf-k",
        "outputId": "348b6aae-1e26-497a-ce05-f039b9f8e3fb"
      },
      "source": [
        "PCA_nComponents(x_train, x_test, 0.40, grid, True)\n",
        "PCA_nComponents(x_train, x_test, 0.65, grid, True)\n",
        "PCA_nComponents(x_train, x_test, 0.90, grid, True)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA7YAAAE9CAYAAAA/Ev6qAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8vihELAAAACXBIWXMAAAsTAAALEwEAmpwYAABFzElEQVR4nO3deZydZX3//9cnO9n3ELKQQIYlbAFComUTFQRRQEUF96qNtVJb/doWl4dSqi3Wn2vFhSpuVRFRIRUUEFyACiRIABMIhCSQBbKSkH2Z+fz+OHfCYUwyJ5M5c2bOvJ6Px3lwr+e8b8Rc+dz3dV9XZCaSJEmSJHVW3WodQJIkSZKkA2FhK0mSJEnq1CxsJUmSJEmdmoWtJEmSJKlTs7CVJEmSJHVqFraSJEmSpE6tR60DtKXhw4fnhAkTah1DklQnHnjggdWZOaLWOToz22ZJUlvaW9tc1cI2Is4Fvgx0B76VmVc1238G8CXgeOCSzLyhbN87gU8Uq5/OzO+19HsTJkxg9uzZbZRektTVRcRTtc7Q2dk2S5La0t7a5qp1RY6I7sDVwHnAZODSiJjc7LCngXcBP2p27lDgU8B0YBrwqYgYUq2skiRJkqTOq5rv2E4DFmTmwszcDlwHXFh+QGYuzsyHgaZm574KuD0z12bmc8DtwLlVzCpJkiRJ6qSqWdiOAZaUrS8ttlX7XEmSJElSF9LpR0WOiBkRMTsiZq9atarWcSRJkiRJ7ayahe0yYFzZ+thiW5uem5nXZObUzJw6YoQDV0qSJElSV1PNwnYW0BAREyOiF3AJMLPCc28FzomIIcWgUecU2yRJkiRJepGqFbaZuRO4jFJB+ihwfWbOjYgrI+ICgIg4JSKWAm8EvhkRc4tz1wL/Rqk4ngVcWWyTJElVEhHnRsT8iFgQEZfvYf+hEXFHRDwcEb+LiLG1yClJUnNVncc2M28Bbmm27ZNly7ModTPe07nXAtdWM58kSSopm6bvbEqDNs6KiJmZOa/ssP8P+H5mfi8iXg78B/D29k8rSdKLdfrBoyRJUptocZo+SvPS31ks/3YP+yVJqomqPrGVJEmdxp6m2pve7JiHgNcDXwZeBwyIiGGZuWZvX9rU1MTmzZvbOqskSS9iYbsHEy6/udYR9sviq86vdQRJUtfwEeCrEfEu4A+UZixobH5QRMwAZgCceOKJ7ZlPktRF2RVZkiRBBVPtZebyzHx9Zp4IfLzYtq75F5VPxVfFvJIk7WZhK0mSoIJp+iJieETs+rvDR3GQR0lSB2FhK0mSKpqmD3gZMD8iHgdGAZ+pSVhJkprxHVtJkgRUNE3fDcAN7Z1LkqSW+MRWkiRJktSpWdhKkiRJkjo1C1tJkiRJUqdmYStJkiRJ6tQsbCVJkiRJnZqFrSRJkiSpU7OwlSRJkiR1aha2kiRJkqROzcJWkiRJktSpWdhKkiRJkjo1C1tJkiRJUqdmYStJkiRJ6tQsbCVJkiRJnZqFrSRJkiSpU7OwlSRJkiR1aha2kiRJkqROrUetA0iSpLYVEacD44Huzfdl5vfbP5EkSdVlYStJUh2JiOuAccAcoKnZ7gQsbCVJdcfCVpKk+nJCZh5d6xCSJLUn37GVJKm+zIuI0bUOIUlSe/KJrSRJ9WUopeL2fmDbro2ZeUHtIkmSVF0WtpIk1Zcrah1AkqT2ZldkSZLqy5sz8/eUBo+anZm/L9ZbFBHnRsT8iFgQEZfvYf/4iPhtRDwYEQ9HxKvbOLskSa1iYStJUn3524j4E/Aw8GhEzIqIyS2dFBHdgauB84DJwKV7OO8TwPWZeSJwCfC1to0uSVLrWNhKklRfdgCXZeahmTkeuBz47wrOmwYsyMyFmbkduA64sNkxCQwslgcBy9sosyRJB8TCVpKk+vJkZv7frpXMvAPoX8F5Y4AlZetLi23lrgDeFhFLgVuAv9/TF0XEjIiYHRGz9ye4JEmt5eBRkiTVl2eL92N/TOkG9tuBJ9rouy8FvpuZn4+IlwI/iIhjM7Op/KDMvAa4BmDq1KnZt2/fNvp5SZL2zCe2kiTVl7cBhwE3AT8F+gHvruC8ZcC4svWxxbZy7wGuB8jMPwJ9gOEHmFeSpAPmE1tJkupIZi6PiI8AwzNz4X6cOgtoiIiJlAraS4C3NDvmaeAVwHcj4mhKhe2qNogtSdIBsbCVJKmORMSPgWOBHRFxEKWBpO5o6bzM3BkRlwG3At2BazNzbkRcSWnaoJnA/wP+OyI+RGkgqXdlZlbtYiRJqpCFrSRJ9eXzmTkbICKOB74WEVMz87MtnZiZt1AaFKp82yfLlucBp7ZxXkmSDpjv2EqSVEfKitrhwHGUuhV/uqahJEmqMgtbSZLqSET8W0TMAh4AzgJ+BoysbSpJkqrLrsiSJNWXdcBfZ+afax1EkqT24hNbSZLqy++aF7UR8c5ahZEkqT34xFaSpPry3oj4NnAv8CClEZKPAL5X01SSJFWRT2wlSaovKylNxbMFGAOcCfyipokkSaoyn9hKklRf3ggcs2t+2Yi4AvgN8I1ahpIkqZqq+sQ2Is6NiPkRsSAiLt/D/t4R8ZNi/30RMaHY3jMivhcRj0TEoxHx0WrmlCSpjqwCLoqIXTevp9YyjCRJ7aFqhW1EdAeuBs4DJgOXRsTkZoe9B3guMycBXwR2TR7/RqB3Zh4HnAy8b1fRK0mS9undwAxgZUSsBj4D/H1tI0mSVF3VfGI7DViQmQszcztwHXBhs2Mu5IXBLG4AXhERQendoH7F3eaDgO3A81XMKklSXcjMJzPzvMwcmpnDM/NsYECtc0mSVE3VfMd2DLCkbH0pMH1vx2TmzohYDwyjVOReCDwD9AU+lJlr9/QjETGD0p1pxo8f35b5JUnqdCLiO5RuEO/SHTgLsJGUJNWtjjp41DSgETgEGALcFRG/ycyFzQ/MzGuAawCmTp2azfdLktTF/LJsuSfwMuBHtYkiSVL7qGZhuwwYV7Y+tti2p2OWFt2OBwFrgLcAv87MHZTeEbqH0uAXf1HYSpKkF2Tmz5ptui4iZtUkjCRJ7aSa79jOAhoiYmJE9AIuAWY2O2Ym8M5i+WLgzmJ6gqeBlwNERD/gJcBjVcwqSVJdioijgCdqnUOSpGqq2hPb4p3Zy4BbKb3fc21mzo2IK4HZmTkT+Dbwg4hYAKylVPxCaTTl70TEXCCA72Tmw9XKKklSvYiIDZTesY3i0xfYsmt7Zg6sZT5Jkqqhqu/YZuYtwC3Ntn2ybHkrpal9mp+3cU/bJUnSvmWmIyBLkrqcjjp4lCRJaoWI6A+cT2kMi2eAm4obxpIk1a1qvmMrSZLaQURMLlu9FTia0hzwfwU8FBGvrEkwSZLaiU9sJUnq/CZExFsy8xPAmZm5c9eOiDiE0mtBU2oVTpKkarOwlSSpk8vMWyKisVje2Wx3I6UBpCRJqlsWtpIk1YHMvBVeNCryLiuBf65JKEmS2omFrSRJdcRRkSVJXZGDR0mSVEci4iXN1odExP9UeO65ETE/IhZExOV72P/FiJhTfB6PiHVtFFuSpANSUWEbEW+KiIUR8Xyzz4aIeL7aISVJUsW+FBFnAkTERcBs4P6WToqI7sDVwHnAZODSZqMtk5kfyswpmTkF+C/g520bXZKk1qm0K/J/AOdl5vxqhpEkSQfsNcCNxdPUg4CzM3NhBedNAxbsOjYirgMuBObt5fhLgU8deFxJkg5cpV2RV1rUSpLUKYwHPgkcC3wPGBwRJ1Vw3hhgSdn60mLbX4iIQ4GJwJ172T8jImZHxOxVq1btT3ZJklql0ie2syLiJ8CNwLZdGzPTLkiSJHUsn6c0KvIi4K+LbQm8vA1/4xLghsxs3NPOzLwGuAZg6tSpuadjJElqS5UWtgOAzcA5ZdsS362RJKlDycyzWnnqMmBc2frYYtueXAJ8oJW/I0lSm6u0sP16ZrY48IQkSaqtiPgv4Cbg95m5oxgU6meZeVELp84CGiJiIqWC9hLgLXv4/qOAIcAf2zS4JEkHoNJ3bP8YER+MiF9ExE8j4gMR4Ry4kiR1PLcAPwBWR8QfgRXA71o6KTN3ApcBtwKPAtdn5tyIuDIiLig79BLgusy0i7EkqcOotDgN4BjgK5SK4XcAJwAzqpRLkiS1zseALwG/BjYAb6T0fu2XWjoxM2+hVBiXb/tks/Ur2iamJEltp9LCdnFmvq9s/Y6IeKgagSRJ0gEZmJmfLVv/bETcWKswkiS1h0oL260RcWhmPgUQEZOAndWLJUmSWukPETGTF2YyOBVYX9NEkiRVWaWF7aeAeyPiMaA7cAjwzqqlkiRJrfVB4E3AS4v1/wOuq10cSZKqr6LCNjN/GhE3A0dQelI7PzN3VDWZqmLC5TfXOkLFFl91fq0jSFKnUwzq9JPiA0BEzKCYV1aSpHpUUWEbEedQGihqJKXREr8AWNhKktTBRMRJzTYNAv4RC1tJUh2rtCvyBcB/AlspjYj8x4h4c2bOr1oySZLUGp8vW05gDaVpfCRJqluVdkXe3SBGxEeBVZQmZh9apVySJKkVMvOs5tsi4pRaZJEkqb1U2hX5MOBc4DxgGjAb+Ncq5pIkSQcgIiYDlxafCVTeS0uSpE6n0kbuJkoTvX8F+ENmbqteJEmS1BoRMYEXitnuwPWUXifqPCMHSpLUCpV2RT6u+baIGJ+ZT7d9JEmStL8i4o/AMEqjIb81Mx8p25c1CyZJUjvoVslBEXFPRHw6IqZHRO+IOBn4UZWzSZKkyq0A+gOjgJERETXOI0lSu6mosAUuB94D/Ap4HLgP+FKVMkmSpP2UmRcBR1Ma3PFfgKUR8bWIOJPK23tJkjqlShu6q4GLMnNoZh4KnAr8v+rFkiRJ+ysz12fmdzLzHOBEYB7w78C42iaTJKm69ucO7iNly/cDG9o4iyRJaiOZuTIzv5qZpwKH1TqPJEnVVOmoyF8CHo2I3wPbganAd6uUSZIktaHMfKrWGSRJqqZKR0W+NiJuo1TQAlzpiMiSJEmSpI6g0lGRz6DUjWlt8Tk8Iu6rZjBJkiRJkipRaVfkfypb7glMAX7e5mkkSdIBiYiDgb8BxgPdm+/PzHe3eyhJkqqs0q7Iry1fj4hBwM1VSSRJkg7ETcCdwG1AU42zSJLULip9Ytvc8zgnniRJHVGfzPxorUNIktSeKipsI+K/gNy1CpwE7IyIrwBk5gerE0+SJO2nmyPi1Zl5S62DSJLUXip9Yju7hXVJktQx/D1weURspzRFXwCZmQNrG0uSpOqp9B3b7wFExHBgbWb6zo4kSR1QZg5o7bkRcS7wZUqDTn0rM6/awzFvAq6g1JProcx8S2t/T5KktrLXwjYiemTmzmL5v4HTgC3AIRHxK+DDmflc+8SUJEmViIijgPcCI4FfZ+aPKjyvO3A1cDawFJgVETMzc17ZMQ3AR4FTM/O5iBjZ5hcgSVIr7GsAqBkR8dJi+fuZeXRmngSMBe4D/qfq6SRJ0v76T+B+4Abg9RHx2woL0GnAgsxcmJnbgeuAC5sd8zfA1btubGfmyjbMLUlSq+2rsP0mcClAZt61a2Nm7szMbwCHVTmbJEnaT5l5QWZen5kzM/NiYCalG9ItGQMsKVtfWmwrdwRwRETcExH3Fl2X/0JEzIiI2RExe9WqVa25DEmS9steuyJnZiPwQfiLUZEBJuEAUpIkdTgRcRBwFnBu8dkJ3NhGX98DaABeRqkH1x8i4rjMXFd+UGZeA1wDMHXq1ESSpCprzajITcANmfmHKuSRJEkHZhFwF3Ab8LnMXNLC8bssA8aVrY8ttpVbCtyXmTuARRHxOKVCd9aBRZYk6cBUWtj+JDO3lm+IiFdm5m+qkEnabxMuv7nWESq2+Krzax1BUn0bU/S62l+zgIaImEipoL0EaD7i8Y2UXlP6TjFTwhHAwgPIKklSm9jXO7blfhkR/QAion8xSvK/t3RSRJwbEfMjYkFEXL6H/b0j4ifF/vsiYkLZvuMj4o8RMTciHomIPhVmlSSpK/tg+UpEDI6Ib7Z0UjETwmXArcCjwPWZOTciroyIC4rDbgXWRMQ84LfAP2XmmraNL0nS/qu0sP0K8KuIuBj4E/Ak8NJ9nVA2bcB5wGTg0oiY3Oyw9wDPZeYk4IvAZ4tze1AadflvM/MYSu/y7KgwqyRJXdklABHxJYDi/ddplZyYmbdk5hGZeXhmfqbY9snMnFksZ2Z+ODMnZ+ZxmXldVa5AkqT9VFFX5MycGRGbgJ8Dr87Meyo4bfe0AQARsWvagHllx1xIaZJ3KE1L8NWICOAc4OHMfKj4fe8GS5JUmX7FzeV3RMRHKI2NcVCNM0mSVFUVFbYR8QilUZF3Aj+LiJUAmXn8Pk7b07QB0/d2TGbujIj1wDBK7+xkRNwKjACuy8z/3Eu2GcAMgPHjx1dyOZIk1bM/AY8AvwF+CfQE7q5pIkmSqqzSwaNeU9UUf6kHcBpwCrAZuCMiHsjMO5of6JQCkiS9yLuAoyi9J3sUcDjwq1oGkiSp2ip9x3Z4Zj5V/qH03uu+VDJtwO5jivdqBwFrKD3d/UNmrs7MzcAtwEkVZpUkqSsbC2yk1L5uAuYCP65pIkmSqqzSwva9ETEnIr4REe+LiP/iL6cAaG73tAER0YvSYBYzmx0zE3hnsXwxcGdmJqVRF4+LiL5FwXsmL343V5Ik7dnNlLog30xpLtt5wPaaJpIkqcoq7Yq8ktI7tlsovRd7JvC1fZ1QvDO7a9qA7sC1u6YNAGYXIyx+G/hBRCwA1lKM5JiZz0XEFygVxwnckpmdZ6JSSZJqJDOPK1+PiEOB/65RHEmS2kWlhe0bgWOKp6lExBWUBqX4xr5OysxbKHUjLt/2ybLlrcV37+nc/6E05Y8kSWqlzHwqIgbVOockSdVUaWG7CrgoIv63mMB9ahUzSZKkVoqID5evUhqIcd2u7Zn5hZoEkySpiiotbN8NfBX4dkQ0AnOAD1YrlCRJarUBzdYf3ct2SZLqRkWFbWY+CZxXvi0iDq5KIkmS1GqZ+a+1ziBJUnurdFTkPbm9zVJIkqQ2EREfj4i5EbE0Ih6PiE8VsxNIklS3KipsI+KRiHi4+DwSEYuB31U1mSRJqkhEzIiII4rVp4ETMnMscBrQB7i2ZuEkSWoHlb5j+5qy5QTWZubGKuSRJEn770fAfwF/nZk/2LUxM1cCH40I54KXJNW1St+xfWrXckQcDnwoIt6cmcdWLZkkSapIZm6MiBnwF6MiA0wCnvrLsyRJqh+VdkU+JCI+FBGzKL1b2x8YUdVkkiSpYpm5o1gcUPbpB9wLvK5WuSRJag/7fGJb3P29FJgI/Ay4LDPvK/a9sfrxJEnS/sjMfy0Gi9r1zu38sqJXkqS61FJX5K8CDwJvy8y72yGPJEk6ABFxJvB9YDEQwLiIeGdm/qGmwSRJqqKWCtvRwBuBz0TEoZSe2l6XmbOqnkySJLXGF4BzMnM+QDFa8o+Bk2uaSpKkKtpnYZuZa4BvAN+IiLHAm4GvR8QwfMdWqroJl99c6wgVW3zV+bWOIKmk566iFiAzH4+InrUMJElStVU63Q+ZuRT4PPD54u7vJVVLJUmSWmt2RHwL+J9i/a3A7BrmkSSp6ioubMtl5uPAlW2cRZIkHbj3Ax8APlis3wV8rXZxJEmqvoqm+5EkSZ1DZm6jNPjjFcCngKuLbS2KiHMjYn5ELIiIy/ew/10RsSoi5hSf97ZpeEmSWqlVT2wlSVLHFBHnAtcACymNijwxIt6Xmb9q4bzuwNXA2cBSYFZEzMzMec0O/UlmXlaF6JIktVrFhW1EjAPGA92b73MKAUmSOowvAC/PzAUAEXE4cDOwz8IWmAYsyMyFxXnXARcCzQtbSZI6nIoK24j4AvA6So1bU7PdCVjYSpLUMfTYVdQWFgIbKjhvDLCkbH0pMH0Px70hIs4AHgc+lJlL9nCMJEntqtIntucDR2bm9mqGkSRJB2xmRNwCXE/p5vMbKXUrfj1AZv78AL77f4EfZ+a2iHgf8D3g5c0PiogZwAyA8ePHH8DPSZJUmUoHj1oM9KpiDkmS1AYy8yPAfwMnAi8DVgEHAa8FXrOPU5cB48rWxxbbyr97TdlAVN8CTt5Lhmsyc2pmTh0xwmnvJUnVV+kT243AnIi4A9g9smJmfnDvp0iSpFrIzF8Av9jP02YBDRExkVJBewnwlvIDImJ0Zj5TrF4APHqgWSVJaguVFrYzi48kSerAIuJg4G/Y+4CP797TeZm5MyIuA24tzrs2M+dGxJXA7MycCXwwIi4AdgJrgXdV5yokSdo/FRW2mfm9iDgWGAncn5kbqxtLkiS10k3AncBt/OWAj/uUmbcAtzTb9smy5Y8CH22DjJIktalKR0X+LbCZUjfkqRHxqcz8TlWTSZKk1uhTFKCSJHUZlXZFfntmLgWIiFHAVyNiWma+v3rRJNWrCZffXOsI+2XxVefXOoK0P26OiFcXT18lSeoSKhoVuayoPQl4LzAauLiKuSRJUuv8PfDLiNgSEc9HxIaIeL7WoSRJqqZKuyL/ADiL0rQ/twIfpjR6oiRJ6kAyc0CtM0iS1N4q7Yp8I3BZZq6vYhZJknSAIuLfM/Njtc4hSVJ7qqgrMvBweVEbEQMj4rwqZZIkSa13bvMNEdG5XmyXJGk/VfrE9vvASyPi5Zl5Z2Y+HxFXAb+qYjZJkrT/xkbEV8rWhwL9ahVGkqT2UGlh26v45zeBhiplkSRJB24L8ECxnMAqSvPaSpJUtyotbDMiPgH0iYgPUJrTdkv1YkmSpFZam5nfq3UISZLaU6Xv2L4LGAicAwwDzqc07Y8kSepY/qPWASRJam+VzmP758z858x8NDOvzMyLsUuyJEkdUWNEvOid2oj4l1qFkSSpPVQ6j+1vKb2ns0t3YBzwi2qEkiRJrTYJmBsRjwEPAscAG2obSZKk6qr0HduPlC33BF4GjGzzNJIk6UAdAcwDbqVU0B4PPFnTRJIkVVlFhW1mPtBs070RcX8V8kiSpAMzLTOPK1u/NiKcnk+SVNcq7Yo8tHwVOAV4PCKGAJGZa6sRTpIk7benI+JDwE3ANuBUSm23JEl1q9KuyA9Qese2ecP4p2L7YW0ZSpIktdo7gX8Dfl6szwLeU7s4kiRVX6VdkSdWO4gk1YMJl99c6wgVW3zV+bWOoCrIzNXA+8u3RcRJwLLaJJIkqfoq7Yp8KHAJpZGQnwF+lJmLqhlMkiQduIjoCfwPMLnWWSRJqpa9zmMbEX8VEbv2fwNYCdwGbAdujoi3tkM+SZK0HyJiQ0Q8v+sDPA18t8axJEmqqpae2F4NvD8zzyvfGBHfAP4P+GG1gkmSpP2XmQOabyue2kqSVLf2+sQ2M/8P+CaURkUu/wBHA93bKaMkSdpPUfKKiPg28Gyt80iSuqAf/hAmTIBu3Ur//GH1novutbAFyMw5xeIDZZ9ZwH8Af9PSl0fEuRExPyIWRMTle9jfOyJ+Uuy/LyImNNs/PiI2RsRHKrscSZK6toh4SUR8BVgKfAKYDWytbSpJUpfzwx/CjBnw1FOQWfrnjBlVK273WdjukpkTyz6HZ+YrMvOefZ0TEd0pdWU+j9KAFZdGRPOBK94DPJeZk4AvAp9ttv8LgJPKS5LUgoj494h4Evgc8AQwNTPPysyvU5rPtpLv2OcN6bLj3hARGRFT2ya9JKmerN+yg23/cjls3vziHZs3w8c/XpXfrHRU5EHAp4Azik2/B67MzPX7OG0asCAzFxbfcR1wITCv7JgLgSuK5RuAr0ZEZGZGxEXAImBTZZciSVKX9l7gKeDrwE2ZuV/tZ9kN6bMpPe2dFREzM3Nes+MGAP8A3NcmqSVJndZzm7bzxMqNPLFyA0+s2MiCYnnF89tYuGwvs8w9/XRVslRU2ALXAn8G3lSsvx34DvD6fZwzBlhStr4UmL63YzJzZ0SsB4ZFxFbgXyg1rnZDliSpZaMptZuXAl+IiHuA64BfVnh+JTekAf6NUg+rf2qL0JKkji0zWb1xO0+s3FAqXFds3L28euP23cf17dWdhpH9OW3SCBpG9Wfb98dw0DNL//ILx4+vSs5KC9vDM/MNZev/GhFzqpBnlyuAL2bmxojY54ERMQOYATC+Sv+SJEnq6DKzEfg18OuI6A28hlKR+3WgfwVf0eIN6Yg4CRiXmTdHhIWtJNWRzOTZ57cWhetGFux6CrtqI+s279h93IA+PWgY2Z9XHDWKhlH9mTSyPw2jBjB6YB+6dSur3T53Vemd2vLuyH37wmc+U5X8lRa2WyLitMy8GyAiTgW2tHDOMmBc2frYYtuejlkaET2AQcAaSg3pxRHxn8BgoCkitmbmV5v/SGZeA1wDMHXq1KzweiRJqluZuQ34GfCziBgIXHSg31nMbf8F4F0VHOtNZ0nqoJqakmXrtuzuNvxCIbuRjdt27j5uSN+eNIwcwKuPG03DyP40jBxAw6j+jBzQm5YePgLw1reW/vnxj5e6H48fXypqd21vY5UWtu8Hvle8axvAWuCdLZwzC2iIiImUCthLgLc0O2Zm8T1/BC4G7szMBE7fdUBEXAFs3FNRK0mS9i0znwe+X8GhLd2QHgAcC/yu+AvNwcDMiLggM2c3+01vOktSjTU2JUvWbt79DuyCsgJ2y47G3ceNGNCbhpH9ecNJY5g0akBRxPZnWP/eBx7irW+tWiHbXEWFbTHtzwnFXd9djWRL5+yMiMuAWynNeXttZs6NiCuB2Zk5E/g28IOIWECpWL6kdZchSZIO0D5vSBcDRg7ftR4RvwM+0ryolSS1rx2NTTy1ZvPursOlQnYjT67ayPadTbuPGz2oD5NG9ufSaeNpGFUqXieN7M/gvr1qmL7tVDoq8hBKoyKfXqzfTWlU5DX7Oi8zbwFuabbtk2XLW4E3tvAdV1SSUZIktV6FN6QlSTWybWcji1dv/osRiBet3sSOxhc6x4wdchANI/tzesPw0vuvI/tz+Mj+DOzTs4bpq6/SrsjXAXdT6i4M8FbgJ8ArqxFKkiS1XkScDoynVKC+SGbutVtySzekm21/2YGllCTtydYdjTy5auOLRiB+YuVGnlqzmcamUgEbAYcO7cukkQN4xdGjdr8De/jIfvTtVWmJV18qvepxmflvZeufjog3VyOQJEnaPxHxX8A/ZGZTMU3POGAO0NTs0KSy920lSVW2adtOnly18cWjEK/cyNNrN5PFA9ju3YIJw/pyxMgBnH/c6OIJ7AAOG9GPPj3/4t5ll1ZpYbs2Ii4Bri/WL6bUVUmSJNXeH4GfAm8ATsjMo2ucR5JUyEyWr9/KnKfX8dDSdTy+otSVeNm6FyaZ6dk9OGx4f44dM4jXnThm9wjEE4b1o1ePbjVM33lUWtheAlwNfA0YCHQDNkXE+4DMzIFVyidJklqQmT+KiDuK1XkRMTozn6lpKEnqojZt28nDS9fz4JLnmPP0OuYsWcfKDdsA6NWjGw0j+3PKhCFcOnIck4oC9tChfenR3QL2QFQ6KvJS4MIqZ5EkSa2UmSuKxaGUitv7gW1l+y+oSTBJqmONTcmClRuZs+Q55ixZx4NPl57IFq/CMnF4P06dNJwp4wZz4vjBHHXwQJ/AVknFbxZHxDj2PhDFH9oylCRJarUrah1AkurVqg3bigK2VMg+vHQ9G7ftBGDQQT05YdxgXnXMwUwZP5gpYwczpF99TKXTGVQ63c8XgNcB89jzQBQWtpIkdQz3FdPp7RYRzmIgSftp645G5i5fz4NFd+IHn163+73YHt2Co0cP5HUnjtn9NHbi8H5ERI1Td12VPrE9HzgyM7dXM4wkSTpgv4yICzNzU0T0B74InAD8psa5JKnDykwWr9nMnCXP7S5kH33m+d3zw44ZfBBTxg3mr0+dwJRxgzl2zCBHJe5gKi1sFwO9AAtbSZI6tq8Av4qIrwD/DlwL/G1tI0lSx7Ju83bmLHnhSexDS9exbvMOAPr16s7xYwfz3tMPKz2NHTeYkQP71DixWlJpYbsRmFOMuFg+EMUHq5JKkiS1SmbOjIhNwM+BV2fmPbXOJEm1tKOxicee2fCiUYoXrt4EQAQcMXIA5x5zMFPGDWbK+ME0jBxA9252Ke5sKi1sZxYfSZLUgUXEI5TGv9gJ/CwiVgJk5vE1DSZJ7SAzWbZuS+lp7NPreHDJOv68bD3bdpaGCRrevzcnjh/MG04ey4njB3P82MH0713xeLrqwCr9X3F+Zt5b1SSSJKktvKbWASSpvWzctpOHl67b/V7snCXrWFXMGdu7RzeOHTOIt7/k0NIoxeMGM2bwQQ7wVKcqLWy/BpxUviEiZmTmNW0fSZIktVZmPtV8W0RcgdMASerkGpuSJ1Zu2N2d+MGn1/H4yg1kMWfsYcP7cfqk4UwZP5gTxw3hqNED6NndOWO7ikoL274RcSKw6/bGUOAfAQtbSZI6kIh4fbNNg4HXY2ErqZNZuWHr7u7Ec55ex8NL17FpeyNQmjN2yrjBnHdc8W7suMEM7uucsV1ZpYXtGODzlArbBFYC769WKElSxzLh8ptrHaFii686v9YRau21ZcsJrAHeUqMsklSRrTsa+fOy9bufxM5Z8uI5YycfMpA3nDy2mDN2CBOG9bVLsV6k0sJ2QWa+vKpJJElSW/hUZj5d6xCStDeZyZK1W5i1eO3u92IffeZ5dja9MGfsieNLc8aeOH4wxxzinLFqWaWF7dyqppAkSW3lRv5yXIzXZOYvaxNHUleXmTy5aiP3LVrLfQvXcv+itTz7/FYA+vfuwfFjBzHjjMM4cfwQpowbzIgBvWucWJ1RpYXtO5tviIiTM/OBNs4jSZIOTI+IGMKLx8X4HGBhK6ldNDUljz27gfsXreG+RaVCds2m7QCMHNCb6YcNY9rEoZwyYYhzxqrNVFrY3h0RC4H/BR4EjgU+DJxarWCSJKlVjgQe4MXjYvxrTRNJqms7G5uYu/x57lu0hvuLQvb5rTuBUrfiM48cwfSJQ5k+cRiH+m6sqqTSwvbfKU35MxnYAEwD3lqtUJIkqdXmZeaJtQ4hqX5t29nII0vXl7oWL1rLA4vX7h6teOLwfrz6uNFMmziUaROHMnZI3xqnVVdRaWH7WWB6Zi4HiIgxwA3Az6oVTJIktcpztQ4gqb5s2d7Ig0ue2/1+7J+efo5tO5sAOHLUAF5/0limTRzK9IlDGTmwT43TqquqtLDdDjSVra8FtrR9HEmSdCCcxUDSgdq4bSezF6/d3a34oaXr2NGYdAuYfMhA3jr9UKYfNpRTJgxlaD/njlXHUGlh+zHggYh4klKRO7HYJkmSOpCI+DileWsHAZuBHwL/kZnbaxpMUoe1fvMO7l+8dvdgT39etp6mLM0fe9zYQbz7tIm8ZOIwTp4whIF9etY6rrRHFRW2mXlLREyiNCAFwOOZubl6sSRJUis9DZyQmTsjYiTwIeBa4G0tnRgR5wJfBroD38rMq5rt/1vgA0AjsBGYkZnz2ji/pCpbtWEbsxav5b6FpUJ2/ooNZEKvHt2YMm4wl501iWkTh3HSoYPp26vS52BSbVX8X2pmbgHm7FqPiJsz8/xqhJIkSa2TmT8oW14JfDQiWiw+I6I7cDVwNrAUmBURM5sVrj/KzG8Ux18AfAE4ty3zS2p7z6zfwv2L1nLvwtJT2SdXbQLgoJ7dOfnQIZxfDPZ0wrjB9OnZvcZppdapqLCNiP+iNGXALoOB/tUIJEmSWi8iPtxs0yTgqQpOnQYsyMyFxfdcB1wI7C5sM/P5suP78eK/G0jqADKTJWu3cG/Z1DtPry11tBzQuwenTBzKG6eOY9rEoRw3ZhA9u3ercWKpbVT6xHZ22XICa4A72j6OJEk6QAPKlpuAeynNPd+SMcCSsvWlwPTmB0XEB4rv6wXscaCqiJgBzAAYP358RaEltU5m8uSqTbvnkL1v4VqefX4rAEP69mTaxKG8868mMH3iUI4ePZDu3ZxDVvWp0ndsv1e+HhGnAZ+n9J6NJEnqIDLzX6v8/VcDV0fEW4BPAO/cwzHXANcATJ061ae6Uhtqakoee3YD9y9aUwz4tJbVG0tjw40Y0JvpxbQ70w8bxqQR/elmIasuouJ3bCPiREqjLL4JeBY4CgtbSZI6lIj4QWa+vWx9EPCbzDylhVOXAePK1scW2/bmOuDrrQ4qqSI7G5uYu/z50tPY4qns81t3AjBm8EGc0TCC6YcNZdrEYUwY1pcIC1l1TfssbCPiCOBSSgXtBuCnwMsyc1FELGqHfJIkaf8cBRAR92TmqZm5PiIqmZ9jFtAQERMpFbSXUGr/d4uIhsx8olg9H3gCSW1q285GHlm6nvsWreW+RWt5YPFaNm1vBGDi8H6cd+zoopAdytghfWucVuo4Wnpi+xilhu7izHyk2T67FkmS1PH0j4hxwIkRcTCwA+jd0knF9ECXAbdSmu7n2sycGxFXArMzcyZwWUS8svjO59hDN2RJ+2fbzkb+9NQ67l1Yehr7p6efY9vOJgCOGNWf1500hukThzFt4lBGDexT47RSx9VSYft6Sndsb4uI3wDXA7/OzB1VTyZJklrjm5QGffwYcBNwOHBFJSdm5i3ALc22fbJs+R/aLKXURWUm81ds4O4nVnPXE6u5f9FatuxoJAImjx7IW6cfyrSJpSeyQ/v1qnVcqdPYZ2GbmTcCN0ZEP0pD/s8AvhURtwADqx9PkiTtj8z8EvAlgIj4KtC32TQ9ktrZyue3cveC1dz9xGruXrCalRu2AXDYiH68aepYTmsYwbSJQxl0UCVvDUjak0pHRd4E/Aj4UUQMAd4IHFrNYJIkaf9FRG/g/cDpxaa7I+Lrmbm1hrGkLmXL9kbuW7SGu54oFbPzV2wAYGi/Xpw6aTinTxrOqQ3DGTP4oBonlepHxaMi75KZz1Eawv+ato8jSZIO0HeBzcBXi/W3AD+gdFNaUhU0NiVzl6/fXcg+8NRzbG9solePbpwyYQgXnXgUpzcMZ/LogU6/I1XJfhe2kiSpQ5uSmUeXrf82IubVLI1Up5as3by7e/E9T65m3ebSEDRHjx7Iu06dwGmThnPKhKEc1Kt7jZNKXYOFrSRJ9WVbRLwkM+8FiIjplAaTknQAnt+6gz8+uWb3e7KLVm8CYNTA3rziqFGc3jCcUycNZ8SAFgchl1QFFraSJNWXdwDfj4jBxfp4YH5EPAJkZh5fs2RSJ7KjsYmHlqzjridWc9cTq3ho6Xoam5K+vbozfeJQ3v6SQzm9YTiTRvYnwu7FUq1Z2EqSVEcy82FgSjHYozMYSBXKTBau3rR7Gp57F65h47addAs4buxg3n/m4ZzWMJyTxg+hV49utY4rqRkLW0mS6kxEHA4cW6zOzcwFtcwjdVRrN23nnuI92bueWMXy9aXBw8cNPYjXnnAIZzQM568OH86gvk7DI3V0FraSJNWRiHiQUvs+H2gETomIn2bmv9Q2mVR7W3c08sBTz5VGL16wirnLnycTBvbpwV8dPpy/O2s4pzcM59Bh/WodVdJ+srCVJKm+vCEzF+5aiYiewDzAwlZdTmby2LMbuPuJ1fzhiVXMWryWrTua6NEtOGn8ED70yiM4vWE4x40ZRI/udi+WOjMLW0mS6siuojYiTgNOAl4K3FHTUFI7WvH81mI+2VXcvWANqzduA2DSyP5ccsp4Tm8YzvTDhtG/t38NluqJ/4+WJKk+jQAOL/7pdD+qW5u37+S+hWt3dy9+fMVGAIb168Wpk4ZzWkOpe/HoQQfVOKmkarKwlSSpjkREZMkvgF8U254APl/bZFLbaGxKHlm2nrufWMVdT6zmT08/x47GpHePbkybOJQ3nDSW0xqGc/TBA+nWzWl4pK6iqoVtRJwLfBnoDnwrM69qtr838H3gZGAN8ObMXBwRZwNXAb2A7cA/Zead1cwqSVKdeDoi/pfSe7U7gTMoDSQldVpL1m7e/UT2ngVrWL9lBwCTRw/k3adO5LSG4ZwyYSh9enavcVJJtVK1wjYiugNXA2cDS4FZETEzM+eVHfYe4LnMnBQRlwCfBd4MrAZem5nLI+JY4FZgTLWySpJUR04EXgccAwRwJ/DDmiaS9tP6LTv445Ori2J2NU+t2QzAwQP7cM7kUZzWMJxTJw1neP/eNU4qqaOo5hPbacCCskEsrgMupHQHeZcLgSuK5RuArxZdqB4sO2YucFBE9M7MbVXMK0lSPXh3Zv5nrUNI+2PX6MW3z1vBb+ev5KEl62hK6NerOy85bBjv+qsJnN4wnMNH9CfC7sWS/lI1C9sxwJKy9aXA9L0dk5k7I2I9MIzSE9td3gD8aW9FbUTMAGYAjB8/vm2SS5LUeV0VES8BzgR2AHcB/5CZy2sbS3qxnY1NzFr8HLfPW8Htjz7LkrVbiIDjxw7msrMmcVrDCKaMG0yvHk7DI6llHXrwqIg4hlL35HP2dkxmXgNcAzB16tRsp2iSJHVUTcBPgTcC3YC3A99lH22p1F42bdvJHx5fxe3zVnDn/JWs27yDXj26cdqk4fzdyybxiqNHMnJAn1rHlNQJVbOwXQaMK1sfW2zb0zFLI6IHMIjSIFJExFhKozm+IzOfrGJOSZLqyYLM/HGx3AhcGxEfrGUgdW0rN2zlN/NWcvu8Z7nnyTVs39nE4L49efmRIznnmFGc3jCCfs4pK+kAVfNPkVlAQ0RMpFTAXgK8pdkxM4F3An8ELgbuzMyMiMHAzcDlmXlPFTNKklRv1kfEmyk9tQ3grcDK2kZSV5KZPLlqI7fNW8Ht81bw4NPrABg39CDeNv1Qzp48ilMmDKFHd7sYS2o7VStsi3dmL6M0onF34NrMnBsRVwKzM3Mm8G3gBxGxAFhLqfgFuAyYBHwyIj5ZbDsnM22YJUnat3cDXwS+Rmm6n98B76phHnUBjU3Jn54u3pedt4JFqzcBcPzYQfy/s4/g7GNGceSoAQ78JKlqqtrvIzNvAW5ptu2TZctbKb0D1Py8TwOfrmY2SZLqVL/M9H1aVd2W7Y3cvWA1t897ljseXcmaTdvp2T14yWHDePepE3jl5FGMHnRQrWNK6iJ8oUGSpPryx4j4EHAWpSe2vwWuycydLZ0YEecCX6bU0+pbmXlVs/0fBt5bfO8qSlMLPdXG+dWBrdm4jTseW8nt81Zw1xOr2LqjiQF9enDWkSM5e/IozjxyBAP79Kx1TEldkIWtJEn1JYBjgK9QGhX5HcAUiqnx9npSRHfgauBsSlP0zYqImZlZPv/8g8DUzNwcEe8H/hN4c5tfgTqURas3cfu8Z7l93goeeOo5mhIOGdSHN00dx9mTRzF94jCn5JFUcxa2kiTVl8WZ+b6y9Tsi4qEKzptGaUTlhQARcR1wIbC7sM3M35Ydfy/wtjbIqw6mqSl5aOm63e/LPrFyIwBHjx7IZS9v4JzJozjmkIG+LyupQ7GwlSSpvmyNiEN3dRGOiEmUug63ZAywpGx9KTB9H8e/B/jVnnZExAyKJ8Tjx4+vJLNqbOuORv745Bpum7eCOx5dwcoN2+jeLZg+cShvmT6eVx49inFD+9Y6piTtlYWtJEn15VPAvRHxGKV3ZQ+hNLVem4mItwFTgTP3tD8zrwGuAZg6dWq25W+r7azfvIM755eeyv5+/io2bW+kX6/unHnkCM6ePIqzjhzJ4L69ah1TkipiYStJUh3JzJ9GxM3AEZSe1M7PzB0VnLoMGFe2PrbY9iIR8Urg48CZmbmtDSKrHS1Zu3l3F+P7F6+lsSkZOaA3F544hrMnj+Klhw2jT8/utY4pSfvNwlaSpDoSEYdSmhd+HPAM8CNgUQWnzgIaImIipYL2EuAtzb77ROCbwLnOLd85ZCZzlz/PbUUx++gzzwPQMLI/f3vmYZw9+WCOHzOIbt18X1ZS52ZhK0lSffkGcD3wKHAkcHNEfCYzf7ivkzJzZ0RcBtxKqQvztZk5NyKuBGZn5kzgc0B/4KfFwEFPZ+YFVbwWtcL2nU3ct2gNt89bwW/mrWD5+q10C5h66FA+/uqjOXvyKCYM71frmJLUpixsJUmqI5l5Xvl6RHwD+D9gn4Vtce4twC3Ntn2ybPmVbRRTbWzD1h38bv4qbpu3gt/NX8mGrTvp07MbZzSM4ENnH8HLjxrJsP69ax1TkqrGwlaSpDoSEUObbZpE6Qms6swz67fwm3kruG3eCu5duIYdjcmwfr0479iDOXvywZw2aTgH9fJ/ekldg4WtJEn15QEggQCagMXA39QykNpGZjJ/xQZum1t6X/aRZesBOGx4P9596kTOnjyKE8cPobvvy0rqgixsJUmqI5k5sdYZ1HZ2NjYxa/FzpZGMH32WJWu3EAFTxg3mn889knMmH8ykkf1rHVOSas7CVpKkOhARn8jMT+9l35cy8x/bOZJaaWdjE/c8uYaZc5Zzx2MrWLd5B716dOO0ScP5u5dN4hVHj2TkgD61jilJHYqFrSRJ9eH1wKcBIuJU4KHM3FjsO6NmqVSRzOShpeu58cFl/PLhZ1i9cRsD+vTg7MmjOGfyKE5vGEG/3v61TZL2xj8hJUmqIxHxCeDtQLeI+NvMvIPS+7bqgBat3sSNDy5j5kPLWbR6E716dOMVR43kwiljOOuoEfTu4eBPklQJC1tJkupDv4i4AdgCnABMBr4TEX8E7LfagazasI1fPrycGx9cxkNL1xMBLz1sGO8/83BedezBDDqoZ60jSlKnY2ErSVJ9OAz4RmZ+sVj/U0ScAvwb0FC7WALYuG0nt819lhvnLOeeBatpbEomjx7Ix159FBecMIaDB3nvQZIOhIWtJEl1IDP/4jFfZm4H/qX4qJ3taGziD4+v4sY5y7l93rNs3dHE2CEH8bdnHsZFU8bQMGpArSNKUt2wsJUkSWojmckDTz3HjXOWcfPDz/Dc5h0M6duTi08ey0VTxnDyoUOI8JVnSWprFraSJEkH6IkVG7hxzjJumrOcpc9toU/Pbpw9+WAumnIIZxwxgp7du9U6oiTVNQtbSZKkVnhm/RZmzlnOjXOW8+gzz9Mt4LSGEXz47CM455iD6e/0PJLUbvwTV5IkqULrt+zg139+hhsfXM69i9aQCSeMG8ynXjuZ1xx/CCMG9K51REnqkixsJUmS9mHrjkZ+N38lNz64nDsfW8n2xiYmDu/HP7yigQunjGHi8H61jihJXZ6FrSRJUjNNTcm9i9Zw04PLueXPz7Bh606G9+/NW18ynoumjOH4sYMcBEqSOhALW0mSJEojGs975nlumrOcmXOW8+zzW+nXqzuvOvZgLpoyhr86fBg9HARKkjokC1tJktSlLVm7mZkPLefGB5fxxMqN9OgWvOzIEXz8/KN55dGjOKhX91pHlCS1wMJWkiR1OWs3befmR57hpgeXMfup5wA4ZcIQPn3RsZx/3GiG9OtV44SSpP1hYStJkrqELdsbuf3RFdz04DJ+//gqdjYlR4zqzz+96kguOOEQxg3tW+uIkqRWsrCVJEl1a2djE/c8uYabHlzGrXOfZdP2Rg4e2If3nDaRC6eM4ejRAxwESpLqgIWtJEmqK5nJQ0vXc+ODy/jlw8+weuM2BvTpwWtPOIQLp4xh+sShdOtmMStJ9cTCVpIk1YVFqzdx44PLmPnQchat3kSvHt14xVEjuXDKGM46agS9ezgIlCTVKwtbSZIEQEScC3wZ6A58KzOvarb/DOBLwPHAJZl5Q7uHbGbVhm3870PLuWnOMh5aup4IeMnEYbz/zMN51bEHM+ignrWOKElqBxa2kiSJiOgOXA2cDSwFZkXEzMycV3bY08C7gI+0f8IXbNy2k1v//Cw3zlnGPQtW05QwefRAPvbqo7jghDEcPKhPLeNJkmrAwlaSJAFMAxZk5kKAiLgOuBDYXdhm5uJiX1N7h9u+s4k/PL6KG+cs4zePrmDrjibGDjmI97/scC6aMoaGUQPaO5IkqQOxsJUkSQBjgCVl60uB6a35ooiYAcwAGD9+/AEH+597n+Lzt83nuc07GNK3JxefPJaLpozh5EOHOKKxJAmwsJUkSW0sM68BrgGYOnVqHuj3De/fi9MaRnDRlEM444gR9Oze7YAzSpLqi4WtJEkCWAaMK1sfW2yruXOPHc25x46udQxJUgfmLU9JkgQwC2iIiIkR0Qu4BJhZ40ySJFXEwlaSJJGZO4HLgFuBR4HrM3NuRFwZERcARMQpEbEUeCPwzYiYW7vEkiS9wK7IkiQJgMy8Bbil2bZPli3PotRFWZKkDsUntpIkSZKkTs3CVpIkSZLUqVnYSpIkSZI6NQtbSZIkSVKnVtXCNiLOjYj5EbEgIi7fw/7eEfGTYv99ETGhbN9Hi+3zI+JV1cwpSZIkSeq8qlbYRkR34GrgPGAycGlETG522HuA5zJzEvBF4LPFuZMpzZ93DHAu8LXi+yRJkiRJepFqPrGdBizIzIWZuR24Driw2TEXAt8rlm8AXhERUWy/LjO3ZeYiYEHxfZIkSZIkvUg157EdAywpW18KTN/bMZm5MyLWA8OK7fc2O3fMnn4kImYAM4rVjREx/8CjV81wYHVbf2l8tq2/cb95XfuhA1wXVOHa6vW6oENcm9e1H9r4ug5t02/rgh544IHVEfFUG3xVVf576QS64nV3xWuGrnndXfGaoWted1te8x7b5moWtu0iM68Brql1jkpExOzMnFrrHG3N6+p86vXavK7OpV6vSy+WmSPa4nu66n8vXfG6u+I1Q9e87q54zdA1r7s9rrmaXZGXAePK1scW2/Z4TET0AAYBayo8V5IkSZKkqha2s4CGiJgYEb0oDQY1s9kxM4F3FssXA3dmZhbbLylGTZ4INAD3VzGrJEmSJKmTqlpX5OKd2cuAW4HuwLWZOTcirgRmZ+ZM4NvADyJiAbCWUvFLcdz1wDxgJ/CBzGysVtZ21Cm6TLeC19X51Ou1eV2dS71el6qjq/730hWvuyteM3TN6+6K1wxd87qrfs1RekAqSZIkSVLnVM2uyJIkSZIkVZ2FrSRJkiSpU7OwbQcRcW5EzI+IBRFxea3ztJWIuDYiVkbEn2udpS1FxLiI+G1EzIuIuRHxD7XO1BYiok9E3B8RDxXX9a+1ztSWIqJ7RDwYEb+sdZa2FBGLI+KRiJgTEbNrnaetRMTgiLghIh6LiEcj4qW1zqSOq17b0X2p1zZ2X+q1/d2Xem+bW1Kvbffe1Gub3pL2avN9x7bKIqI78DhwNrCU0mjRl2bmvJoGawMRcQawEfh+Zh5b6zxtJSJGA6Mz808RMQB4ALios/9vFhEB9MvMjRHRE7gb+IfMvLfG0dpERHwYmAoMzMzX1DpPW4mIxcDUzKyridwj4nvAXZn5rWLk/L6Zua7GsdQB1XM7ui/12sbuS722v/tS721zS+q17d6bem3TW9Jebb5PbKtvGrAgMxdm5nbgOuDCGmdqE5n5B0qjWdeVzHwmM/9ULG8AHgXG1DbVgcuSjcVqz+JTF3e2ImIscD7wrVpnUcsiYhBwBqWR8cnM7Ra12oe6bUf3pV7b2H2p1/Z3X+q5bW6JbXfX0J5tvoVt9Y0BlpStL6XO/5CuJxExATgRuK/GUdpE0eVnDrASuD0z6+K6gC8B/ww01ThHNSRwW0Q8EBEzah2mjUwEVgHfKbqgfSsi+tU6lDos29EuqN7a332p47a5JV+iftvuvanHNr0l7dbmW9hKexER/YGfAf+Ymc/XOk9byMzGzJwCjAWmRUSn794WEa8BVmbmA7XOUiWnZeZJwHnAB4ruiZ1dD+Ak4OuZeSKwCegS701Kalk9tr/7Uo9tc0u6QNu9N/XYprek3dp8C9vqWwaMK1sfW2xTB1a85/Iz4IeZ+fNa52lrRReQ3wLn1jhKWzgVuKB4b+U64OUR8T+1jdR2MnNZ8c+VwC8odcvs7JYCS8ueStxAqdGT9sR2tAup9/Z3X+qsbW5JXbfde1OnbXpL2q3Nt7CtvllAQ0RMLF6WvgSYWeNM2odiIIdvA49m5hdqnaetRMSIiBhcLB9EaSCWx2oaqg1k5kczc2xmTqD0/687M/NtNY7VJiKiXzGACkW3nXOATj9CamY+CyyJiCOLTa8A6nZwGB0w29Euol7b332p17a5JfXcdu9NvbbpLWnPNr9HNb5UL8jMnRFxGXAr0B24NjPn1jhWm4iIHwMvA4ZHxFLgU5n57dqmahOnAm8HHineeQH4WGbeUrtIbWI08L1ihNFuwPWZ2SWG1+/ERgG/KP1djx7AjzLz17WN1Gb+HvhhUagsBP66xnnUQdVzO7ovddzG7ku9tr/7YtvcddRzm96Sdmnzne5HkiRJktSp2RVZkiRJktSpWdhKkiRJkjo1C1tJkiRJUqdmYStJkiRJ6tQsbKUOLiIujIgjap2jViLiNRFxVK1zSJK0i22zbbM6HgtbqUIRcVFEZGv+II+I/hHxy4h4rPhn/wrPex0wJTMf34/fOi0imiJiTtnn1fubuSOIiIuAl2Rm3c/pJ0naf7bN7c+2WR2V0/1IFYqInwCHUJpE/FP7ee4VwPLMvCYi3g+Myswr2j7l7gnub6KU80vV+A1JkjoC22ZJu/jEVqpAcRf3NOA9wCXN9p1V3O19PCKujpIJEfHnssNeD3y/WP4B8Ibi3O9GxMVl37Wx+GdExOci4s8R8UhEvLnsmH8ptj8UEZ9pnjVLd6veBrwnIs4qO293pogYEhH3RET3iPhIRHy62H5KRFy3l2xXFde5qrg7vjUiTi/uOi+IiPXF8veL478cER9o9u/qB8WdbiLi2Ii4PyLmR0Rj8Z2XR8SIiFhaTFa/67zDI2JOsbzHvJKkrsW22bZZKmdhK1XmQuDXRbejNRFxMkBE9KbUKL4aOBo4AnjdHs4fD9xbNAB3A+Na+L3XA1OAE4BXAp+LiNERcR7wWuCUzDwB+PyeTs7M54E3Aj+OiMF72P8c8CQwDbgVOLfYPgs4rvk5EXEqcB5wYnEtD1PqhnRXZk4B3gvclZlTMvMdxWlvBH7W7KdnArsa5C8CX8rMI4F3ADdl5lWU/v39MjMby/I+CfSMiIZK8kqSugTbZttmaTcLW6kylwK77j5eV6wDHAMsysyFxR/2PwZO38P5WTQsUyg1QC29A3Aa8OPMbMzMFcDvgVMoNaTfzcwtxZeu3cd3nA3cl5nr9rJ/DTAuMx8BRkXEyGL7dmB4s2OnATdm5pbit28CzmIvImII0Cczn42IbhHxhWLXY8CxxfLJwPXF8k+AM4rlY4B5xfe8PiLOKD+3wrySpPpn22zbLO1mYSu1ICKGAi8HvhURi4F/At4UEQEMZM8NYfNtyyNiVLF8CLCs/CfKlnu0UeaGIuf7yjY3NTvsSOCZYvl24FURcTzQF1jIi69hJy/+86JbsW2X5tcbZfsHA+8ulhvLjmnkhWsv/77yc18GnLSHc/eUV5LURdg2A7bN0otY2Eotuxj4QWYempkTMnMcsIgX7v5Oi4iJEdENeDOl7kxrKTWsu1wP/HWx/G5euBsKpbu9RMRLgd7F+yt3AW8u3rMZQemO6f2UGo13RsRBewtb5PgecHlmPlu2awUwNCIOKhqdw4D/K/bdSqkb1deBv8/MJkp3jQcV+38HXBQRfSOiH6UuSb8v++5VwJBdK8Xd6m0RcXSRfWdEHAOcCvy57DvfUiy/rbhmgEeAM4q/nBwDvDQiehX/nubuI68kqeuwbbZtll4sM/348bOPD/Bb4Nxm2z5I6Q/ulwF/AG4G5gPfALoVx3yV0h/2hwD9i2OeBH4J9CuO+W6xfRbwc+A2Su/mBPA5Sg3NI8Cby377n4tttwKH7SHv6ZTuoM4p+7yj2Pf3wBPAbODksnOGFud8rGzbccXvX1msf6BY/zPwwT387s+Bh8vWX1/81p+Acyh1V3oYOLrYP4bSXwYeBn4DjC229y37d/VPwDXF8v+3r7x+/Pjx46frfGybbZv9+Gn+cbof6QBExMuAj2Tma1p5/ncpDcZwQyvO/TvgkMz8RGt+W5KkemTbLHVNdkWWOq+/pnSHWZIkdQy2zVKN+MRWkiRJktSp+cRWkiRJktSpWdhKkiRJkjo1C1tJkiRJUqdmYStJkiRJ6tQsbCVJkiRJnZqFrSRJkiSpU/v/AeuS8iuIMU+7AAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 1152x360 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Components =  7 ;\n",
            "Total explained variance =  0.40812\n",
            "Training set size:  60000  - Test set size:  10000\n",
            "Score = 0.22\n",
            "Execution time 3.4092468778292337\n",
            "Best score 0.8646166666666666 with best parameters {'SVM__C': 10, 'SVM__kernel': 'poly'}\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x360 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Components =  21 ;\n",
            "Total explained variance =  0.65424\n",
            "Training set size:  60000  - Test set size:  10000\n",
            "Score = 0.13\n",
            "Execution time 2.68569176197052\n",
            "Best score 0.9766166666666667 with best parameters {'SVM__C': 10, 'SVM__kernel': 'poly'}\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x360 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Components =  87 ;\n",
            "Total explained variance =  0.90011\n",
            "Training set size:  60000  - Test set size:  10000\n",
            "Score = 0.13\n",
            "Execution time 9.064098803202311\n",
            "Best score 0.9868833333333333 with best parameters {'SVM__C': 10, 'SVM__kernel': 'poly'}\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 565
        },
        "id": "1kzClV7AGhNi",
        "outputId": "63122cd4-7fe5-4d72-86b5-dbd0917ac2f0"
      },
      "source": [
        "from google.colab import drive\n",
        "import sys\n",
        "drive.mount('/content/drive')\n",
        "pkl = \"/content/drive/MyDrive/ML/mnist_svm_classifier.pkl\"\n",
        "pkl_pca = \"/content/drive/MyDrive/ML/mnist_pca_svm_classifier.pkl\"\n",
        "\n",
        "grid = joblib.load(pkl)\n",
        "grid_pca = joblib.load(pkl_pca)\n",
        "print(\"--------> For Training set size:  12000  - Test set size:  2000 and linear, poly, rbf and C,gamma = 1, 10, 100\")\n",
        "print(\"Best score\", grid.best_score_,\"with best parameters\", grid.best_params_)\n",
        "df = pd.DataFrame(grid.cv_results_)\n",
        "if 'param_SVM__gamma' in df.columns:\n",
        "    df = df[['param_SVM__kernel', 'param_SVM__C', 'param_SVM__gamma', 'mean_test_score', 'mean_fit_time', 'mean_score_time']]\n",
        "else:\n",
        "    df = df[['param_SVM__kernel', 'param_SVM__C', 'mean_test_score', 'mean_fit_time', 'mean_score_time']]\n",
        "df.sort_values(by=['mean_test_score'], ascending=False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n",
            "--------> For Training set size:  12000  - Test set size:  2000 and linear, poly, rbf and C,gamma = 1, 10, 100\n",
            "Best score 0.9711666666666667 with best parameters {'SVM__C': 10, 'SVM__kernel': 'poly'}\n"
          ],
          "name": "stdout"
        },
        {
          "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>param_SVM__kernel</th>\n",
              "      <th>param_SVM__C</th>\n",
              "      <th>param_SVM__gamma</th>\n",
              "      <th>mean_test_score</th>\n",
              "      <th>mean_fit_time</th>\n",
              "      <th>mean_score_time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>5</th>\n",
              "      <td>poly</td>\n",
              "      <td>10</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.971167</td>\n",
              "      <td>10.827205</td>\n",
              "      <td>3.162293</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>poly</td>\n",
              "      <td>1</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.969500</td>\n",
              "      <td>10.859758</td>\n",
              "      <td>3.085548</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>13</th>\n",
              "      <td>rbf</td>\n",
              "      <td>10</td>\n",
              "      <td>1</td>\n",
              "      <td>0.964333</td>\n",
              "      <td>9.367451</td>\n",
              "      <td>3.342008</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>10</th>\n",
              "      <td>rbf</td>\n",
              "      <td>1</td>\n",
              "      <td>1</td>\n",
              "      <td>0.955833</td>\n",
              "      <td>10.070767</td>\n",
              "      <td>3.393352</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>poly</td>\n",
              "      <td>0.1</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.944667</td>\n",
              "      <td>16.570929</td>\n",
              "      <td>4.075422</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>12</th>\n",
              "      <td>rbf</td>\n",
              "      <td>10</td>\n",
              "      <td>0.1</td>\n",
              "      <td>0.939833</td>\n",
              "      <td>6.962410</td>\n",
              "      <td>2.870550</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>linear</td>\n",
              "      <td>1</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.932000</td>\n",
              "      <td>7.347908</td>\n",
              "      <td>2.919110</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>linear</td>\n",
              "      <td>10</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.928167</td>\n",
              "      <td>5.223567</td>\n",
              "      <td>2.228071</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7</th>\n",
              "      <td>rbf</td>\n",
              "      <td>0.1</td>\n",
              "      <td>1</td>\n",
              "      <td>0.919667</td>\n",
              "      <td>21.206935</td>\n",
              "      <td>5.310530</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>9</th>\n",
              "      <td>rbf</td>\n",
              "      <td>1</td>\n",
              "      <td>0.1</td>\n",
              "      <td>0.917000</td>\n",
              "      <td>14.780004</td>\n",
              "      <td>4.819524</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>linear</td>\n",
              "      <td>0.1</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.902833</td>\n",
              "      <td>17.521289</td>\n",
              "      <td>5.047727</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>14</th>\n",
              "      <td>rbf</td>\n",
              "      <td>10</td>\n",
              "      <td>10</td>\n",
              "      <td>0.859667</td>\n",
              "      <td>51.416728</td>\n",
              "      <td>7.035161</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>11</th>\n",
              "      <td>rbf</td>\n",
              "      <td>1</td>\n",
              "      <td>10</td>\n",
              "      <td>0.852167</td>\n",
              "      <td>51.208838</td>\n",
              "      <td>7.026770</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6</th>\n",
              "      <td>rbf</td>\n",
              "      <td>0.1</td>\n",
              "      <td>0.1</td>\n",
              "      <td>0.806500</td>\n",
              "      <td>38.435451</td>\n",
              "      <td>7.058010</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>8</th>\n",
              "      <td>rbf</td>\n",
              "      <td>0.1</td>\n",
              "      <td>10</td>\n",
              "      <td>0.179000</td>\n",
              "      <td>50.770108</td>\n",
              "      <td>7.521460</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "   param_SVM__kernel param_SVM__C  ... mean_fit_time  mean_score_time\n",
              "5               poly           10  ...     10.827205         3.162293\n",
              "3               poly            1  ...     10.859758         3.085548\n",
              "13               rbf           10  ...      9.367451         3.342008\n",
              "10               rbf            1  ...     10.070767         3.393352\n",
              "1               poly          0.1  ...     16.570929         4.075422\n",
              "12               rbf           10  ...      6.962410         2.870550\n",
              "2             linear            1  ...      7.347908         2.919110\n",
              "4             linear           10  ...      5.223567         2.228071\n",
              "7                rbf          0.1  ...     21.206935         5.310530\n",
              "9                rbf            1  ...     14.780004         4.819524\n",
              "0             linear          0.1  ...     17.521289         5.047727\n",
              "14               rbf           10  ...     51.416728         7.035161\n",
              "11               rbf            1  ...     51.208838         7.026770\n",
              "6                rbf          0.1  ...     38.435451         7.058010\n",
              "8                rbf          0.1  ...     50.770108         7.521460\n",
              "\n",
              "[15 rows x 6 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 97
        },
        "id": "bfof_cvyIMlu",
        "outputId": "47b76bb5-1dfe-401e-e0a1-38b57f05f877"
      },
      "source": [
        "print(\"Best score\", grid_pca.best_score_,\"with best parameters\", grid_pca.best_params_)\n",
        "df = pd.DataFrame(grid_pca.cv_results_)\n",
        "if 'param_SVM__gamma' in df.columns:\n",
        "    df = df[['param_SVM__kernel', 'param_SVM__C', 'param_SVM__gamma', 'mean_test_score', 'mean_fit_time', 'mean_score_time']]\n",
        "else:\n",
        "    df = df[['param_SVM__kernel', 'param_SVM__C', 'mean_test_score', 'mean_fit_time', 'mean_score_time']]\n",
        "df.sort_values(by=['mean_test_score'], ascending=False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Best score 0.9713333333333333 with best parameters {'SVM__C': 10, 'SVM__kernel': 'poly'}\n"
          ],
          "name": "stdout"
        },
        {
          "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>param_SVM__kernel</th>\n",
              "      <th>param_SVM__C</th>\n",
              "      <th>mean_test_score</th>\n",
              "      <th>mean_fit_time</th>\n",
              "      <th>mean_score_time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>poly</td>\n",
              "      <td>10</td>\n",
              "      <td>0.971333</td>\n",
              "      <td>2.097583</td>\n",
              "      <td>0.376625</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "  param_SVM__kernel param_SVM__C  ...  mean_fit_time  mean_score_time\n",
              "0              poly           10  ...       2.097583         0.376625\n",
              "\n",
              "[1 rows x 5 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ToSDEbqkqbKP"
      },
      "source": [
        "## Συμπεράσματα\n",
        "\n",
        "\n",
        "*   Για το πρώτο κομμάτι, ο SVM εκπαιδεύτηκε για το 100% του συνόλου δεδομένων με τις εξής παραμέτρους: kernel = linear,poly,rbf(20% του συνόλου δεδομένων) | C = 1, 10, 100 (0.1,1,10 για το 20% του συνόλου δεδομένων) | gamma(μόνο για rbf) = 0.1,1,10. Σύμφωνα με το GridSearchCV, οι παράμετροι με την καλύτερη απόδοση είναι: {'SVM__C': 10, 'SVM__kernel': 'poly'}. Παρατηρούμε πως καλύτερη Kernel Function είναι η poly, το οποίο σημαίνει πως τα δεδομένα ταξινομούνται καλύτερα με πολυωνυμική συνάρτηση. Σχετικά με την υπερπαράμετρο C, η οποία ελέγχει το σφάλμα, όταν επιλέγουμε μικρή τιμή, συνήθως έχουμε χαμηλό σφάλμα, κάτι το οποίο δεν σημαίνει και καλό decision boundary και αντίστοιχα για υψηλό C, έχουμε υψηλό σφάλμα. Για το συγκεκριμένο dataset η ιδανική τιμή για το C είναι το 10. Το $\\gamma$ ρυθμίζει την καμπυλότητα του ορίου απόφασης (decision boundary). Έτσι, όταν επιλέγουμε μικρό $\\gamma$ έχουμε λιγότερη καμπυλότητα στο decision boundary και όταν επιλέγουμε υψηλό $\\gamma$ έχουμε περισσότερη καμπυλότητα. Η τιμή 1 για το $\\gamma$ επιτυγχάνει την καλύτερη ακρίβεια, όταν χρησιμοποιούμε την rbf function. Ο χρόνος εκπαίδευσης κυμαίνεται από 151 έως 306 δευτερόλεπτα (μέσος όρος) ανάλογα την επιλογή του kernel και παρατηρούμε πως για μικρό C ο χρόνος είναι μεγαλύτερος.\n",
        "\n",
        "*   Με την εφαρμογή του PCA και τη μείωση διαστάσεων, παρατηρούμε πως η διατηρούμενη διακύμανση αυξάνεται με περισσότερο ρυθμό για <60 συνιστώσες, ενώ για παραπάνω συνιστώσες μειώνεται ο ρυθμός αύξησης. Για 87 συνιστώσες επιτυγχάνουμε καλύτερο Cross-validated score (of the best_estimator) 0.98 με τις παραμέτρους {'SVM__C': 10, 'SVM__kernel': 'poly'}"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Nsu1rUfxuD_k"
      },
      "source": [
        "## Πηγές\n",
        "* [Support Vector Machines](https://scikit-learn.org/stable/modules/svm.html)\n",
        "* [Multiclass Classification with Support Vector Machines (SVM), Dual Problem and Kernel Functions](https://towardsdatascience.com/multiclass-classification-with-support-vector-machines-svm-kernel-trick-kernel-functions-f9d5377d6f02)\n",
        "* [sklearn.preprocessing.Normalizer](https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.Normalizer.html)\n",
        "* [Explained variance in PCA](https://https://ro-che.info/articles/2017-12-11-pca-explained-variance)\n",
        "\n",
        "\n"
      ]
    }
  ]
}