mappgraph/notebooks/train_GNN.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "train_GNN.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LKQ1mORvDvSx"
      },
      "source": [
        "T5_3"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0lcDoZ0o2Za3"
      },
      "source": [
        "## Import packages"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "P6QAxxEXvYeF",
        "outputId": "e37c68b3-3113-4bce-afc5-433c3bc9fdbc"
      },
      "source": [
        "!nvidia-smi"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Fri Sep  3 14:06:39 2021       \n",
            "+-----------------------------------------------------------------------------+\n",
            "| NVIDIA-SMI 470.63.01    Driver Version: 460.32.03    CUDA Version: 11.2     |\n",
            "|-------------------------------+----------------------+----------------------+\n",
            "| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n",
            "| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\n",
            "|                               |                      |               MIG M. |\n",
            "|===============================+======================+======================|\n",
            "|   0  Tesla T4            Off  | 00000000:00:04.0 Off |                    0 |\n",
            "| N/A   38C    P8     9W /  70W |      0MiB / 15109MiB |      0%      Default |\n",
            "|                               |                      |                  N/A |\n",
            "+-------------------------------+----------------------+----------------------+\n",
            "                                                                               \n",
            "+-----------------------------------------------------------------------------+\n",
            "| Processes:                                                                  |\n",
            "|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |\n",
            "|        ID   ID                                                   Usage      |\n",
            "|=============================================================================|\n",
            "|  No running processes found                                                 |\n",
            "+-----------------------------------------------------------------------------+\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sG9WNKzW8a8x",
        "outputId": "1f693f85-37e0-42f6-f604-fa10e1b5334e"
      },
      "source": [
        "# install StellarGraph if running on Google Colab\n",
        "import sys\n",
        "if 'google.colab' in sys.modules:\n",
        "  %pip install -q stellargraph[demos]==1.2.1"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[K     |████████████████████████████████| 435 kB 7.1 MB/s \n",
            "\u001b[K     |████████████████████████████████| 376 kB 52.4 MB/s \n",
            "\u001b[K     |████████████████████████████████| 45 kB 4.1 MB/s \n",
            "\u001b[?25h  Building wheel for mplleaflet (setup.py) ... \u001b[?25l\u001b[?25hdone\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "r1niN0z09sri"
      },
      "source": [
        "# verify that we're using the correct version of StellarGraph for this notebook\n",
        "import stellargraph as sg\n",
        "\n",
        "try:\n",
        "    sg.utils.validate_notebook_version(\"1.2.1\")\n",
        "except AttributeError:\n",
        "    raise ValueError(\n",
        "        f\"This notebook requires StellarGraph version 1.2.1, but a different version {sg.__version__} is installed.  Please see <https://github.com/stellargraph/stellargraph/issues/1172>.\"\n",
        "    ) from None"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AcQq7a-99v_n"
      },
      "source": [
        "import pandas as pd\n",
        "pd.options.mode.chained_assignment = None  # default='warn'\n",
        "import os\n",
        "import math\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "from stellargraph import StellarGraph\n",
        "import pandas as pd\n",
        "import numpy as np\n",
        "import stellargraph as sg\n",
        "from stellargraph.mapper import PaddedGraphGenerator\n",
        "from stellargraph.layer import DeepGraphCNN\n",
        "from stellargraph import StellarGraph\n",
        "from stellargraph import datasets\n",
        "from sklearn import model_selection\n",
        "from IPython.display import display, HTML\n",
        "from tensorflow.keras import Model\n",
        "from tensorflow.keras.optimizers import Adam\n",
        "from tensorflow.keras.layers import Dense, Conv1D, MaxPool1D, Dropout, Flatten,  BatchNormalization\n",
        "from tensorflow.keras.losses import binary_crossentropy, categorical_crossentropy\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from sklearn.metrics import classification_report\n",
        "import warnings\n",
        "warnings.filterwarnings('ignore')"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tQfYJXo0JUeN"
      },
      "source": [
        "root_path = 'https://raw.githubusercontent.com/soeai/MAppGraph/gh-pages/data'"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "V0Y8LA582yMR"
      },
      "source": [
        "## Config setting"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2YHZAXTQJJvK"
      },
      "source": [
        "N = 20\n",
        "t = 10\n",
        "k = 10  # the number of rows for the output tensor (k = 10, 20)\n",
        "T = 5\n",
        "overlap = 3 # note: overlap depends on T\n",
        "\n",
        "apps = ['diijam', 'baomoi', 'fptplay', 'iQIYI', 'bigo', 'myradio', 'spotify', 'nhaccuatui', 'soundcloud', 'sachnoiapp', \n",
        "        'phim247', 'popskid', 'truyenaudiosachnoiviet', 'vieon', 'voizfm', 'tunefm', 'wetv', 'zingmp3', 'truyenaudio', 'baohay24h',\n",
        "        'freefire', 'among_us', 'azar', 'comico', 'nimotv', 'mangatoon', 'medoctruyen', 'nhacvang', 'noveltoon', 'radiofm',\n",
        "        'vtvgo', 'tivi24h', 'tinder', 'tinmoi24h', 'tivi360', 'tiktok', 'linkedin', 'tiki', 'tinhte', 'lotus', 'tivi247',\n",
        "        'tivi_truyentranh_webtoon', 'tuoitre_online', 'vietnamworks', 'wallstreet_journal', 'cnn_news', 'bbc_news', 'twitter', \n",
        "        'weeboo', 'twitch', 'vnexpress', 'topcv', 'toc_chien', 'wesing', 'hago', 'google_meet', 'dubsmash', 'facebook','hahalolo', \n",
        "        'zalo', 'hello_yo', 'dan_tri', 'zoom', 'wikipedia', 'instagram', 'jobway', 'kaka', 'pinterest', 'quora', 'lazada', 'chess', \n",
        "        'cake', 'mobile_legend', 'co_tuong_online', 'ted', 'telegram', 'starmarker', 'skype', 'soha', 'tango', 'thanhnien', 'snapchat', \n",
        "        'tien_len', 'animal_restaurant', 'bida', 'cho_tot', 'messenger', 'netflix', 'nonolive', 'may', 'podcast_player', 'pubg', \n",
        "        'partying', 'kenh14', 'lienquan_mobile', 'likee_lite', 'reddit', 'sendo', 'shopee', 'the_guardian', 'ola_party']\n",
        "\n",
        "features = ['complete_max', 'complete_min', 'complete_mean', 'complete_mad', 'complete_std', 'complete_var', 'complete_skew',\n",
        "       'complete_kurt', 'complete_pkt_num', 'complete_10per', 'complete_20per', 'complete_30per', 'complete_40per', 'complete_50per', \n",
        "        'complete_60per', 'complete_70per', 'complete_80per', 'complete_90per', 'out_max', 'out_min', 'out_mean', 'out_mad', 'out_std',\n",
        "        'out_var', 'out_skew', 'out_kurt', 'out_pkt_num', 'out_10per', 'out_20per', 'out_30per', 'out_40per', 'out_50per', 'out_60per',\n",
        "        'out_70per', 'out_80per', 'out_90per', 'in_max', 'in_min', 'in_mean', 'in_mad', 'in_std', 'in_var', 'in_skew', 'in_kurt', \n",
        "        'in_pkt_num', 'in_10per', 'in_20per', 'in_30per', 'in_40per', 'in_50per', 'in_60per', 'in_70per', 'in_80per', 'in_90per', \n",
        "        'protocol', 'flows_num', 'flow_length_mean', 'flow_pkt_num_mean', 'flow_duration_mean', 'ip1', 'ip2', 'ip3', 'ip4'\n",
        "       ]"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JCENiV3-gyRf"
      },
      "source": [
        "train_graphs_folder = os.path.join(root_path, '%d_%d/train_graphs/N%d/t%d'%(T, overlap, N, t))\n",
        "test_graphs_folder = os.path.join(root_path, '%d_%d/test_graphs/N%d/t%d'%(T, overlap, N, t))"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JjW7cSVJ94-S"
      },
      "source": [
        "## Loading graphs for training and testing"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7glSmi_xsjgK"
      },
      "source": [
        "'''\n",
        "Load graphs for one app\n",
        "Input: app and folder that contains graphs of the app\n",
        "Output: List of graphs (StellarGraph objects) and List of labels\n",
        "'''\n",
        "def graphs_one_app(app, graphs_folder):\n",
        "  graphs = []\n",
        "\n",
        "  app_graph_folder = os.path.join(graphs_folder, app)\n",
        "  features_path = os.path.join(app_graph_folder, 'features.csv')\n",
        "  weights_path = os.path.join(app_graph_folder, 'weights.csv')\n",
        "\n",
        "  features_df = pd.read_csv(features_path, index_col=0)\n",
        "  weights_df = pd.read_csv(weights_path, index_col=0)\n",
        "\n",
        "  graph_num = features_df['graph_id'].iloc[-1]\n",
        "  # loop over all graphs of the app\n",
        "  for i in range(1, graph_num+1):\n",
        "    feature_df = features_df[features_df['graph_id'] == i]\n",
        "    feature_df = feature_df[['IP_port'] + features + ['graph_id']]\n",
        "    feature_df = feature_df.set_index('IP_port')\n",
        "    \n",
        "    weight_df = weights_df[weights_df['graph_id'] == i].reset_index(drop=True)\n",
        "\n",
        "    # drop graph_id column\n",
        "    feature_df = feature_df.drop(['graph_id'], axis=1)\n",
        "    weight_df = weight_df.drop(['graph_id'], axis=1)\n",
        "\n",
        "    if weight_df.shape[0] > 0:\n",
        "      graph = StellarGraph(feature_df, weight_df)\n",
        "      graphs.append(graph)\n",
        "\n",
        "  labels = [app]*graph_num\n",
        "    \n",
        "  return graphs, labels"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1kny4ma265qf"
      },
      "source": [
        "'''\n",
        "Load all graphs\n",
        "Input: folder that contains graphs\n",
        "Output: List of graphs (StellarGraph objects), List of graph_labels (dummy values) and List of labels (names of app)\n",
        "'''\n",
        "def generate_graphs(graphs_folder):\n",
        "  # build a list of graphs and labels: note that only apply for more than 2 classes\n",
        "  li = []\n",
        "  labels = []\n",
        "  idx = 0\n",
        "\n",
        "  for app in apps:\n",
        "    idx += 1\n",
        "    print('Loading {} ... {}/{}'.format(app, idx, len(apps)))\n",
        "    \n",
        "    one_app_graphs, one_app_labels = graphs_one_app(app, graphs_folder)\n",
        "    li.extend(one_app_graphs)\n",
        "    labels.extend(one_app_labels)\n",
        "    \n",
        "\n",
        "  graph_labels = pd.get_dummies(labels)\n",
        "  graphs = li\n",
        "\n",
        "  print('...............................................................')\n",
        "\n",
        "  return graphs, graph_labels, labels"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "p667q9A7r44A"
      },
      "source": [
        "Load graphs from csv files\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "mGabFHxmsAX9",
        "outputId": "980bacdd-a306-494e-88c8-1ca8c8d95695"
      },
      "source": [
        "train_graphs, train_graph_labels, _ = generate_graphs(train_graphs_folder)\n",
        "test_graphs, test_graph_labels, _ = generate_graphs(test_graphs_folder)\n",
        "train_size = len(train_graphs)\n",
        "\n",
        "graphs = train_graphs + test_graphs\n",
        "graph_labels = train_graph_labels.append(test_graph_labels, ignore_index=True)\n",
        "\n",
        "test_graph_labels = graph_labels[train_size:]"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Loading diijam ... 1/101\n",
            "Loading baomoi ... 2/101\n",
            "Loading fptplay ... 3/101\n",
            "Loading iQIYI ... 4/101\n",
            "Loading bigo ... 5/101\n",
            "Loading myradio ... 6/101\n",
            "Loading spotify ... 7/101\n",
            "Loading nhaccuatui ... 8/101\n",
            "Loading soundcloud ... 9/101\n",
            "Loading sachnoiapp ... 10/101\n",
            "Loading phim247 ... 11/101\n",
            "Loading popskid ... 12/101\n",
            "Loading truyenaudiosachnoiviet ... 13/101\n",
            "Loading vieon ... 14/101\n",
            "Loading voizfm ... 15/101\n",
            "Loading tunefm ... 16/101\n",
            "Loading wetv ... 17/101\n",
            "Loading zingmp3 ... 18/101\n",
            "Loading truyenaudio ... 19/101\n",
            "Loading baohay24h ... 20/101\n",
            "Loading freefire ... 21/101\n",
            "Loading among_us ... 22/101\n",
            "Loading azar ... 23/101\n",
            "Loading comico ... 24/101\n",
            "Loading nimotv ... 25/101\n",
            "Loading mangatoon ... 26/101\n",
            "Loading medoctruyen ... 27/101\n",
            "Loading nhacvang ... 28/101\n",
            "Loading noveltoon ... 29/101\n",
            "Loading radiofm ... 30/101\n",
            "Loading vtvgo ... 31/101\n",
            "Loading tivi24h ... 32/101\n",
            "Loading tinder ... 33/101\n",
            "Loading tinmoi24h ... 34/101\n",
            "Loading tivi360 ... 35/101\n",
            "Loading tiktok ... 36/101\n",
            "Loading linkedin ... 37/101\n",
            "Loading tiki ... 38/101\n",
            "Loading tinhte ... 39/101\n",
            "Loading lotus ... 40/101\n",
            "Loading tivi247 ... 41/101\n",
            "Loading tivi_truyentranh_webtoon ... 42/101\n",
            "Loading tuoitre_online ... 43/101\n",
            "Loading vietnamworks ... 44/101\n",
            "Loading wallstreet_journal ... 45/101\n",
            "Loading cnn_news ... 46/101\n",
            "Loading bbc_news ... 47/101\n",
            "Loading twitter ... 48/101\n",
            "Loading weeboo ... 49/101\n",
            "Loading twitch ... 50/101\n",
            "Loading vnexpress ... 51/101\n",
            "Loading topcv ... 52/101\n",
            "Loading toc_chien ... 53/101\n",
            "Loading wesing ... 54/101\n",
            "Loading hago ... 55/101\n",
            "Loading google_meet ... 56/101\n",
            "Loading dubsmash ... 57/101\n",
            "Loading facebook ... 58/101\n",
            "Loading hahalolo ... 59/101\n",
            "Loading zalo ... 60/101\n",
            "Loading hello_yo ... 61/101\n",
            "Loading dan_tri ... 62/101\n",
            "Loading zoom ... 63/101\n",
            "Loading wikipedia ... 64/101\n",
            "Loading instagram ... 65/101\n",
            "Loading jobway ... 66/101\n",
            "Loading kaka ... 67/101\n",
            "Loading pinterest ... 68/101\n",
            "Loading quora ... 69/101\n",
            "Loading lazada ... 70/101\n",
            "Loading chess ... 71/101\n",
            "Loading cake ... 72/101\n",
            "Loading mobile_legend ... 73/101\n",
            "Loading co_tuong_online ... 74/101\n",
            "Loading ted ... 75/101\n",
            "Loading telegram ... 76/101\n",
            "Loading starmarker ... 77/101\n",
            "Loading skype ... 78/101\n",
            "Loading soha ... 79/101\n",
            "Loading tango ... 80/101\n",
            "Loading thanhnien ... 81/101\n",
            "Loading snapchat ... 82/101\n",
            "Loading tien_len ... 83/101\n",
            "Loading animal_restaurant ... 84/101\n",
            "Loading bida ... 85/101\n",
            "Loading cho_tot ... 86/101\n",
            "Loading messenger ... 87/101\n",
            "Loading netflix ... 88/101\n",
            "Loading nonolive ... 89/101\n",
            "Loading may ... 90/101\n",
            "Loading podcast_player ... 91/101\n",
            "Loading pubg ... 92/101\n",
            "Loading partying ... 93/101\n",
            "Loading kenh14 ... 94/101\n",
            "Loading lienquan_mobile ... 95/101\n",
            "Loading likee_lite ... 96/101\n",
            "Loading reddit ... 97/101\n",
            "Loading sendo ... 98/101\n",
            "Loading shopee ... 99/101\n",
            "Loading the_guardian ... 100/101\n",
            "Loading ola_party ... 101/101\n",
            "...............................................................\n",
            "Loading diijam ... 1/101\n",
            "Loading baomoi ... 2/101\n",
            "Loading fptplay ... 3/101\n",
            "Loading iQIYI ... 4/101\n",
            "Loading bigo ... 5/101\n",
            "Loading myradio ... 6/101\n",
            "Loading spotify ... 7/101\n",
            "Loading nhaccuatui ... 8/101\n",
            "Loading soundcloud ... 9/101\n",
            "Loading sachnoiapp ... 10/101\n",
            "Loading phim247 ... 11/101\n",
            "Loading popskid ... 12/101\n",
            "Loading truyenaudiosachnoiviet ... 13/101\n",
            "Loading vieon ... 14/101\n",
            "Loading voizfm ... 15/101\n",
            "Loading tunefm ... 16/101\n",
            "Loading wetv ... 17/101\n",
            "Loading zingmp3 ... 18/101\n",
            "Loading truyenaudio ... 19/101\n",
            "Loading baohay24h ... 20/101\n",
            "Loading freefire ... 21/101\n",
            "Loading among_us ... 22/101\n",
            "Loading azar ... 23/101\n",
            "Loading comico ... 24/101\n",
            "Loading nimotv ... 25/101\n",
            "Loading mangatoon ... 26/101\n",
            "Loading medoctruyen ... 27/101\n",
            "Loading nhacvang ... 28/101\n",
            "Loading noveltoon ... 29/101\n",
            "Loading radiofm ... 30/101\n",
            "Loading vtvgo ... 31/101\n",
            "Loading tivi24h ... 32/101\n",
            "Loading tinder ... 33/101\n",
            "Loading tinmoi24h ... 34/101\n",
            "Loading tivi360 ... 35/101\n",
            "Loading tiktok ... 36/101\n",
            "Loading linkedin ... 37/101\n",
            "Loading tiki ... 38/101\n",
            "Loading tinhte ... 39/101\n",
            "Loading lotus ... 40/101\n",
            "Loading tivi247 ... 41/101\n",
            "Loading tivi_truyentranh_webtoon ... 42/101\n",
            "Loading tuoitre_online ... 43/101\n",
            "Loading vietnamworks ... 44/101\n",
            "Loading wallstreet_journal ... 45/101\n",
            "Loading cnn_news ... 46/101\n",
            "Loading bbc_news ... 47/101\n",
            "Loading twitter ... 48/101\n",
            "Loading weeboo ... 49/101\n",
            "Loading twitch ... 50/101\n",
            "Loading vnexpress ... 51/101\n",
            "Loading topcv ... 52/101\n",
            "Loading toc_chien ... 53/101\n",
            "Loading wesing ... 54/101\n",
            "Loading hago ... 55/101\n",
            "Loading google_meet ... 56/101\n",
            "Loading dubsmash ... 57/101\n",
            "Loading facebook ... 58/101\n",
            "Loading hahalolo ... 59/101\n",
            "Loading zalo ... 60/101\n",
            "Loading hello_yo ... 61/101\n",
            "Loading dan_tri ... 62/101\n",
            "Loading zoom ... 63/101\n",
            "Loading wikipedia ... 64/101\n",
            "Loading instagram ... 65/101\n",
            "Loading jobway ... 66/101\n",
            "Loading kaka ... 67/101\n",
            "Loading pinterest ... 68/101\n",
            "Loading quora ... 69/101\n",
            "Loading lazada ... 70/101\n",
            "Loading chess ... 71/101\n",
            "Loading cake ... 72/101\n",
            "Loading mobile_legend ... 73/101\n",
            "Loading co_tuong_online ... 74/101\n",
            "Loading ted ... 75/101\n",
            "Loading telegram ... 76/101\n",
            "Loading starmarker ... 77/101\n",
            "Loading skype ... 78/101\n",
            "Loading soha ... 79/101\n",
            "Loading tango ... 80/101\n",
            "Loading thanhnien ... 81/101\n",
            "Loading snapchat ... 82/101\n",
            "Loading tien_len ... 83/101\n",
            "Loading animal_restaurant ... 84/101\n",
            "Loading bida ... 85/101\n",
            "Loading cho_tot ... 86/101\n",
            "Loading messenger ... 87/101\n",
            "Loading netflix ... 88/101\n",
            "Loading nonolive ... 89/101\n",
            "Loading may ... 90/101\n",
            "Loading podcast_player ... 91/101\n",
            "Loading pubg ... 92/101\n",
            "Loading partying ... 93/101\n",
            "Loading kenh14 ... 94/101\n",
            "Loading lienquan_mobile ... 95/101\n",
            "Loading likee_lite ... 96/101\n",
            "Loading reddit ... 97/101\n",
            "Loading sendo ... 98/101\n",
            "Loading shopee ... 99/101\n",
            "Loading the_guardian ... 100/101\n",
            "Loading ola_party ... 101/101\n",
            "...............................................................\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iyDhfLE9MIpU"
      },
      "source": [
        "generator = PaddedGraphGenerator(graphs=graphs)"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fVXBLvOISbXz"
      },
      "source": [
        "## Build GNN"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CiKbAm3kMjay",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3e4a5acb-d8bd-46ad-dfdc-9bb54e509309"
      },
      "source": [
        "layer_sizes = [1024, 1024, 1024, 512]\n",
        "\n",
        "dgcnn_model = DeepGraphCNN(\n",
        "    layer_sizes=layer_sizes,\n",
        "    activations=[\"tanh\", \"tanh\", \"tanh\", \"tanh\"],\n",
        "    k=k,\n",
        "    bias=False,\n",
        "    generator=generator,\n",
        ")\n",
        "x_inp, x_out = dgcnn_model.in_out_tensors()\n",
        "\n",
        "#------------------------------------------------------------------------------\n",
        "x_out = Conv1D(filters=256, kernel_size=sum(layer_sizes), strides=sum(layer_sizes))(x_out)\n",
        "x_out = MaxPool1D(pool_size=2)(x_out)\n",
        "\n",
        "x_out = Conv1D(filters=512, kernel_size=5, strides=1)(x_out)\n",
        "\n",
        "x_out = Flatten()(x_out)\n",
        "\n",
        "x_out = Dense(units=1024, activation=\"relu\")(x_out)\n",
        "x_out = Dropout(rate=0.25)(x_out)\n",
        "\n",
        "predictions = Dense(units=len(apps), activation=\"softmax\")(x_out)\n",
        "\n",
        "#------------------------------------------------------------------------------\n",
        "\n",
        "model = Model(inputs=x_inp, outputs=predictions)\n",
        "\n",
        "# using exponentialDecay to decrease the learning rate after 10 epochs\n",
        "# lr =  initial_lr * decay_rate ^ (step / decay_steps)\n",
        "batch_size = 256\n",
        "decay_epoch = 20\n",
        "\n",
        "lr_schedule = keras.optimizers.schedules.ExponentialDecay(\n",
        "    initial_learning_rate=0.0001,\n",
        "    decay_steps=(train_size//batch_size)*decay_epoch,\n",
        "    decay_rate=0.9)\n",
        "optimizer = keras.optimizers.Adam(learning_rate=lr_schedule)\n",
        "\n",
        "model.compile(\n",
        "    optimizer=optimizer, loss=categorical_crossentropy, metrics=[\"acc\"],\n",
        ")"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.7/dist-packages/tensorflow/python/util/deprecation.py:617: calling map_fn_v2 (from tensorflow.python.ops.map_fn) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use fn_output_signature instead\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "u7DG1_5EWa6r",
        "outputId": "1e31547d-7e85-4c38-c8de-c04b3e8e284d"
      },
      "source": [
        "model.summary()"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"model\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_1 (InputLayer)            [(None, None, 63)]   0                                            \n",
            "__________________________________________________________________________________________________\n",
            "dropout (Dropout)               (None, None, 63)     0           input_1[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "input_3 (InputLayer)            [(None, None, None)] 0                                            \n",
            "__________________________________________________________________________________________________\n",
            "graph_convolution (GraphConvolu (None, None, 1024)   64512       dropout[0][0]                    \n",
            "                                                                 input_3[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "dropout_1 (Dropout)             (None, None, 1024)   0           graph_convolution[0][0]          \n",
            "__________________________________________________________________________________________________\n",
            "graph_convolution_1 (GraphConvo (None, None, 1024)   1048576     dropout_1[0][0]                  \n",
            "                                                                 input_3[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "dropout_2 (Dropout)             (None, None, 1024)   0           graph_convolution_1[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "graph_convolution_2 (GraphConvo (None, None, 1024)   1048576     dropout_2[0][0]                  \n",
            "                                                                 input_3[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "dropout_3 (Dropout)             (None, None, 1024)   0           graph_convolution_2[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "graph_convolution_3 (GraphConvo (None, None, 512)    524288      dropout_3[0][0]                  \n",
            "                                                                 input_3[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf.concat (TFOpLambda)          (None, None, 3584)   0           graph_convolution[0][0]          \n",
            "                                                                 graph_convolution_1[0][0]        \n",
            "                                                                 graph_convolution_2[0][0]        \n",
            "                                                                 graph_convolution_3[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "input_2 (InputLayer)            [(None, None)]       0                                            \n",
            "__________________________________________________________________________________________________\n",
            "sort_pooling (SortPooling)      (None, 35840, 1)     0           tf.concat[0][0]                  \n",
            "                                                                 input_2[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "conv1d (Conv1D)                 (None, 10, 256)      917760      sort_pooling[0][0]               \n",
            "__________________________________________________________________________________________________\n",
            "max_pooling1d (MaxPooling1D)    (None, 5, 256)       0           conv1d[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "conv1d_1 (Conv1D)               (None, 1, 512)       655872      max_pooling1d[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "flatten (Flatten)               (None, 512)          0           conv1d_1[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "dense (Dense)                   (None, 1024)         525312      flatten[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "dropout_4 (Dropout)             (None, 1024)         0           dense[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "dense_1 (Dense)                 (None, 101)          103525      dropout_4[0][0]                  \n",
            "==================================================================================================\n",
            "Total params: 4,888,421\n",
            "Trainable params: 4,888,421\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "L1Ij6mb6MnDR"
      },
      "source": [
        "gen = PaddedGraphGenerator(graphs=graphs)\n",
        "\n",
        "train_gen = gen.flow(\n",
        "    list(train_graph_labels.index - 1),\n",
        "    targets=train_graph_labels.values,\n",
        "    batch_size=256,\n",
        "    symmetric_normalization=False,\n",
        ")\n",
        "\n",
        "test_gen = gen.flow(\n",
        "    list(test_graph_labels.index - 1),\n",
        "    targets=test_graph_labels.values,\n",
        "    batch_size=256,\n",
        "    symmetric_normalization=False,\n",
        ")"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0epzLQdnBigR"
      },
      "source": [
        "## Training"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yWLELgWLMyTz",
        "outputId": "e9033bad-08e2-4694-d05d-46f0bdf74fce"
      },
      "source": [
        "class CustomSaver(keras.callbacks.Callback):\n",
        "    def on_epoch_end(self, epoch, logs={}):\n",
        "        if epoch % 10 == 0 & epoch > 0:  # or save after some epoch, each k-th epoch etc.\n",
        "            self.model.save(os.path.join(models_folder, \"model_{}.hd5\".format(epoch)))\n",
        "# create and use callback:\n",
        "saver = CustomSaver()\n",
        "\n",
        "epochs = 150\n",
        "history = model.fit(\n",
        "    train_gen, callbacks=[saver], epochs=epochs, verbose=1, validation_data=test_gen, shuffle=True,\n",
        ")"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/150\n",
            "259/259 [==============================] - 132s 442ms/step - loss: 4.5149 - acc: 0.0261 - val_loss: 3.7438 - val_acc: 0.1629\n",
            "Epoch 2/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 3.6978 - acc: 0.1234 - val_loss: 2.9574 - val_acc: 0.2766\n",
            "Epoch 3/150\n",
            "259/259 [==============================] - 113s 438ms/step - loss: 2.9874 - acc: 0.2575 - val_loss: 2.4867 - val_acc: 0.3633\n",
            "Epoch 4/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 2.6180 - acc: 0.3224 - val_loss: 2.1820 - val_acc: 0.4426\n",
            "Epoch 5/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 2.2470 - acc: 0.4091 - val_loss: 1.9512 - val_acc: 0.4888\n",
            "Epoch 6/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 2.0611 - acc: 0.4521 - val_loss: 1.8510 - val_acc: 0.5171\n",
            "Epoch 7/150\n",
            "259/259 [==============================] - 113s 438ms/step - loss: 1.9023 - acc: 0.4925 - val_loss: 1.6267 - val_acc: 0.5705\n",
            "Epoch 8/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 1.7901 - acc: 0.5175 - val_loss: 1.4667 - val_acc: 0.6149\n",
            "Epoch 9/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 1.6403 - acc: 0.5539 - val_loss: 1.4768 - val_acc: 0.6116\n",
            "Epoch 10/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 1.5008 - acc: 0.5922 - val_loss: 1.3206 - val_acc: 0.6516\n",
            "Epoch 11/150\n",
            "259/259 [==============================] - 113s 438ms/step - loss: 1.3794 - acc: 0.6253 - val_loss: 1.1910 - val_acc: 0.6902\n",
            "Epoch 12/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 1.3250 - acc: 0.6414 - val_loss: 1.2193 - val_acc: 0.6665\n",
            "Epoch 13/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 1.2522 - acc: 0.6569 - val_loss: 1.0841 - val_acc: 0.7165\n",
            "Epoch 14/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 1.2123 - acc: 0.6655 - val_loss: 1.0802 - val_acc: 0.7137\n",
            "Epoch 15/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 1.1721 - acc: 0.6762 - val_loss: 1.0417 - val_acc: 0.7227\n",
            "Epoch 16/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 1.0834 - acc: 0.7020 - val_loss: 0.9784 - val_acc: 0.7439\n",
            "Epoch 17/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 1.0593 - acc: 0.7067 - val_loss: 0.9527 - val_acc: 0.7470\n",
            "Epoch 18/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 1.0603 - acc: 0.7067 - val_loss: 0.9924 - val_acc: 0.7394\n",
            "Epoch 19/150\n",
            "259/259 [==============================] - 114s 438ms/step - loss: 0.9910 - acc: 0.7250 - val_loss: 0.8908 - val_acc: 0.7663\n",
            "Epoch 20/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.9679 - acc: 0.7316 - val_loss: 0.8402 - val_acc: 0.7797\n",
            "Epoch 21/150\n",
            "259/259 [==============================] - 113s 437ms/step - loss: 0.9427 - acc: 0.7369 - val_loss: 0.8176 - val_acc: 0.7883\n",
            "Epoch 22/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.9109 - acc: 0.7467 - val_loss: 0.8315 - val_acc: 0.7844\n",
            "Epoch 23/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.8805 - acc: 0.7522 - val_loss: 0.8387 - val_acc: 0.7819\n",
            "Epoch 24/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.8611 - acc: 0.7612 - val_loss: 0.8279 - val_acc: 0.7853\n",
            "Epoch 25/150\n",
            "259/259 [==============================] - 115s 443ms/step - loss: 0.8513 - acc: 0.7607 - val_loss: 0.7794 - val_acc: 0.7998\n",
            "Epoch 26/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.8373 - acc: 0.7644 - val_loss: 0.7713 - val_acc: 0.8010\n",
            "Epoch 27/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.8012 - acc: 0.7757 - val_loss: 0.7988 - val_acc: 0.7875\n",
            "Epoch 28/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.7954 - acc: 0.7758 - val_loss: 0.7131 - val_acc: 0.8137\n",
            "Epoch 29/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.7573 - acc: 0.7876 - val_loss: 0.6932 - val_acc: 0.8230\n",
            "Epoch 30/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.7277 - acc: 0.7957 - val_loss: 0.7121 - val_acc: 0.8163\n",
            "Epoch 31/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.7385 - acc: 0.7897 - val_loss: 0.7101 - val_acc: 0.8144\n",
            "Epoch 32/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.7205 - acc: 0.7956 - val_loss: 0.6805 - val_acc: 0.8277\n",
            "Epoch 33/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.7151 - acc: 0.7977 - val_loss: 0.6817 - val_acc: 0.8240\n",
            "Epoch 34/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.6793 - acc: 0.8077 - val_loss: 0.6429 - val_acc: 0.8337\n",
            "Epoch 35/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.6626 - acc: 0.8107 - val_loss: 0.6219 - val_acc: 0.8414\n",
            "Epoch 36/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.6319 - acc: 0.8201 - val_loss: 0.6596 - val_acc: 0.8299\n",
            "Epoch 37/150\n",
            "259/259 [==============================] - 115s 442ms/step - loss: 0.6336 - acc: 0.8186 - val_loss: 0.6559 - val_acc: 0.8316\n",
            "Epoch 38/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.6309 - acc: 0.8192 - val_loss: 0.6459 - val_acc: 0.8356\n",
            "Epoch 39/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.6436 - acc: 0.8184 - val_loss: 0.5974 - val_acc: 0.8512\n",
            "Epoch 40/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.6157 - acc: 0.8252 - val_loss: 0.6098 - val_acc: 0.8445\n",
            "Epoch 41/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.5761 - acc: 0.8374 - val_loss: 0.5725 - val_acc: 0.8566\n",
            "Epoch 42/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.5547 - acc: 0.8420 - val_loss: 0.5935 - val_acc: 0.8514\n",
            "Epoch 43/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.5632 - acc: 0.8386 - val_loss: 0.6132 - val_acc: 0.8452\n",
            "Epoch 44/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.5675 - acc: 0.8373 - val_loss: 0.6047 - val_acc: 0.8455\n",
            "Epoch 45/150\n",
            "259/259 [==============================] - 115s 443ms/step - loss: 0.5460 - acc: 0.8421 - val_loss: 0.5508 - val_acc: 0.8663\n",
            "Epoch 46/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.5135 - acc: 0.8546 - val_loss: 0.5536 - val_acc: 0.8617\n",
            "Epoch 47/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.5251 - acc: 0.8489 - val_loss: 0.5534 - val_acc: 0.8604\n",
            "Epoch 48/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.5216 - acc: 0.8513 - val_loss: 0.5551 - val_acc: 0.8594\n",
            "Epoch 49/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.5011 - acc: 0.8567 - val_loss: 0.5483 - val_acc: 0.8641\n",
            "Epoch 50/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.4853 - acc: 0.8619 - val_loss: 0.5568 - val_acc: 0.8584\n",
            "Epoch 51/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.4777 - acc: 0.8638 - val_loss: 0.5242 - val_acc: 0.8703\n",
            "Epoch 52/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.4966 - acc: 0.8549 - val_loss: 0.5461 - val_acc: 0.8604\n",
            "Epoch 53/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.4796 - acc: 0.8616 - val_loss: 0.5190 - val_acc: 0.8728\n",
            "Epoch 54/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.4838 - acc: 0.8607 - val_loss: 0.5438 - val_acc: 0.8671\n",
            "Epoch 55/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.4520 - acc: 0.8698 - val_loss: 0.4845 - val_acc: 0.8839\n",
            "Epoch 56/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.4349 - acc: 0.8755 - val_loss: 0.4833 - val_acc: 0.8844\n",
            "Epoch 57/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.4421 - acc: 0.8712 - val_loss: 0.5041 - val_acc: 0.8759\n",
            "Epoch 58/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.4503 - acc: 0.8691 - val_loss: 0.4997 - val_acc: 0.8810\n",
            "Epoch 59/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.4045 - acc: 0.8835 - val_loss: 0.5038 - val_acc: 0.8802\n",
            "Epoch 60/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.4190 - acc: 0.8782 - val_loss: 0.4988 - val_acc: 0.8769\n",
            "Epoch 61/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.4217 - acc: 0.8788 - val_loss: 0.4735 - val_acc: 0.8844\n",
            "Epoch 62/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.4109 - acc: 0.8803 - val_loss: 0.4957 - val_acc: 0.8785\n",
            "Epoch 63/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.3816 - acc: 0.8904 - val_loss: 0.4688 - val_acc: 0.8881\n",
            "Epoch 64/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.3810 - acc: 0.8900 - val_loss: 0.5143 - val_acc: 0.8752\n",
            "Epoch 65/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.4133 - acc: 0.8793 - val_loss: 0.4580 - val_acc: 0.8901\n",
            "Epoch 66/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.3617 - acc: 0.8953 - val_loss: 0.4728 - val_acc: 0.8901\n",
            "Epoch 67/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.3756 - acc: 0.8904 - val_loss: 0.4581 - val_acc: 0.8902\n",
            "Epoch 68/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.3568 - acc: 0.8959 - val_loss: 0.4744 - val_acc: 0.8850\n",
            "Epoch 69/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.3384 - acc: 0.9016 - val_loss: 0.4385 - val_acc: 0.8973\n",
            "Epoch 70/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.3496 - acc: 0.8983 - val_loss: 0.4717 - val_acc: 0.8900\n",
            "Epoch 71/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.3545 - acc: 0.8962 - val_loss: 0.4690 - val_acc: 0.8884\n",
            "Epoch 72/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.3384 - acc: 0.9011 - val_loss: 0.4655 - val_acc: 0.8908\n",
            "Epoch 73/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.3280 - acc: 0.9044 - val_loss: 0.4435 - val_acc: 0.8978\n",
            "Epoch 74/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.3353 - acc: 0.9006 - val_loss: 0.4348 - val_acc: 0.8981\n",
            "Epoch 75/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.3153 - acc: 0.9072 - val_loss: 0.4409 - val_acc: 0.9007\n",
            "Epoch 76/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.3076 - acc: 0.9114 - val_loss: 0.4289 - val_acc: 0.8992\n",
            "Epoch 77/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.3002 - acc: 0.9130 - val_loss: 0.4407 - val_acc: 0.8986\n",
            "Epoch 78/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.3077 - acc: 0.9081 - val_loss: 0.4221 - val_acc: 0.9056\n",
            "Epoch 79/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.2903 - acc: 0.9145 - val_loss: 0.4580 - val_acc: 0.8933\n",
            "Epoch 80/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.2974 - acc: 0.9112 - val_loss: 0.4274 - val_acc: 0.9019\n",
            "Epoch 81/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.2762 - acc: 0.9190 - val_loss: 0.4234 - val_acc: 0.9024\n",
            "Epoch 82/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.2748 - acc: 0.9185 - val_loss: 0.4174 - val_acc: 0.9057\n",
            "Epoch 83/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.2795 - acc: 0.9177 - val_loss: 0.4285 - val_acc: 0.9053\n",
            "Epoch 84/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.2865 - acc: 0.9157 - val_loss: 0.4359 - val_acc: 0.9022\n",
            "Epoch 85/150\n",
            "259/259 [==============================] - 113s 438ms/step - loss: 0.2613 - acc: 0.9235 - val_loss: 0.4127 - val_acc: 0.9076\n",
            "Epoch 86/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.2529 - acc: 0.9252 - val_loss: 0.4080 - val_acc: 0.9098\n",
            "Epoch 87/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.2483 - acc: 0.9267 - val_loss: 0.4139 - val_acc: 0.9105\n",
            "Epoch 88/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.2480 - acc: 0.9269 - val_loss: 0.4009 - val_acc: 0.9115\n",
            "Epoch 89/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.2493 - acc: 0.9248 - val_loss: 0.4210 - val_acc: 0.9094\n",
            "Epoch 90/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.2408 - acc: 0.9280 - val_loss: 0.4185 - val_acc: 0.9067\n",
            "Epoch 91/150\n",
            "259/259 [==============================] - 113s 438ms/step - loss: 0.2433 - acc: 0.9282 - val_loss: 0.4489 - val_acc: 0.8996\n",
            "Epoch 92/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.2411 - acc: 0.9282 - val_loss: 0.3980 - val_acc: 0.9139\n",
            "Epoch 93/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.2224 - acc: 0.9338 - val_loss: 0.4086 - val_acc: 0.9115\n",
            "Epoch 94/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.2212 - acc: 0.9333 - val_loss: 0.4187 - val_acc: 0.9096\n",
            "Epoch 95/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.2237 - acc: 0.9329 - val_loss: 0.4156 - val_acc: 0.9131\n",
            "Epoch 96/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.2156 - acc: 0.9347 - val_loss: 0.4227 - val_acc: 0.9134\n",
            "Epoch 97/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.2051 - acc: 0.9385 - val_loss: 0.4221 - val_acc: 0.9108\n",
            "Epoch 98/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.2107 - acc: 0.9347 - val_loss: 0.4193 - val_acc: 0.9130\n",
            "Epoch 99/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1959 - acc: 0.9414 - val_loss: 0.4139 - val_acc: 0.9142\n",
            "Epoch 100/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.1990 - acc: 0.9400 - val_loss: 0.4018 - val_acc: 0.9179\n",
            "Epoch 101/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1929 - acc: 0.9421 - val_loss: 0.4096 - val_acc: 0.9143\n",
            "Epoch 102/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.2010 - acc: 0.9379 - val_loss: 0.4040 - val_acc: 0.9149\n",
            "Epoch 103/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.1881 - acc: 0.9420 - val_loss: 0.4107 - val_acc: 0.9149\n",
            "Epoch 104/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1832 - acc: 0.9424 - val_loss: 0.4006 - val_acc: 0.9172\n",
            "Epoch 105/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1754 - acc: 0.9467 - val_loss: 0.3975 - val_acc: 0.9205\n",
            "Epoch 106/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1731 - acc: 0.9468 - val_loss: 0.3934 - val_acc: 0.9226\n",
            "Epoch 107/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.1742 - acc: 0.9469 - val_loss: 0.4024 - val_acc: 0.9196\n",
            "Epoch 108/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1740 - acc: 0.9472 - val_loss: 0.4078 - val_acc: 0.9169\n",
            "Epoch 109/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1688 - acc: 0.9492 - val_loss: 0.4158 - val_acc: 0.9174\n",
            "Epoch 110/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.1947 - acc: 0.9397 - val_loss: 0.4229 - val_acc: 0.9131\n",
            "Epoch 111/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1588 - acc: 0.9506 - val_loss: 0.3879 - val_acc: 0.9249\n",
            "Epoch 112/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.1610 - acc: 0.9505 - val_loss: 0.3929 - val_acc: 0.9226\n",
            "Epoch 113/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1545 - acc: 0.9524 - val_loss: 0.3929 - val_acc: 0.9204\n",
            "Epoch 114/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1574 - acc: 0.9500 - val_loss: 0.3913 - val_acc: 0.9251\n",
            "Epoch 115/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1401 - acc: 0.9572 - val_loss: 0.3863 - val_acc: 0.9277\n",
            "Epoch 116/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.1390 - acc: 0.9567 - val_loss: 0.3850 - val_acc: 0.9271\n",
            "Epoch 117/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.1384 - acc: 0.9564 - val_loss: 0.4038 - val_acc: 0.9213\n",
            "Epoch 118/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.1409 - acc: 0.9569 - val_loss: 0.3965 - val_acc: 0.9226\n",
            "Epoch 119/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.1424 - acc: 0.9548 - val_loss: 0.3969 - val_acc: 0.9259\n",
            "Epoch 120/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.1317 - acc: 0.9591 - val_loss: 0.3945 - val_acc: 0.9269\n",
            "Epoch 121/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.1376 - acc: 0.9567 - val_loss: 0.3925 - val_acc: 0.9277\n",
            "Epoch 122/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.1604 - acc: 0.9524 - val_loss: 0.4275 - val_acc: 0.9173\n",
            "Epoch 123/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.1348 - acc: 0.9583 - val_loss: 0.3895 - val_acc: 0.9291\n",
            "Epoch 124/150\n",
            "259/259 [==============================] - 114s 439ms/step - loss: 0.1152 - acc: 0.9649 - val_loss: 0.3889 - val_acc: 0.9307\n",
            "Epoch 125/150\n",
            "259/259 [==============================] - 113s 436ms/step - loss: 0.1133 - acc: 0.9641 - val_loss: 0.4020 - val_acc: 0.9287\n",
            "Epoch 126/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.1145 - acc: 0.9638 - val_loss: 0.4266 - val_acc: 0.9193\n",
            "Epoch 127/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1364 - acc: 0.9548 - val_loss: 0.4007 - val_acc: 0.9286\n",
            "Epoch 128/150\n",
            "259/259 [==============================] - 115s 442ms/step - loss: 0.1137 - acc: 0.9637 - val_loss: 0.4056 - val_acc: 0.9295\n",
            "Epoch 129/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1079 - acc: 0.9657 - val_loss: 0.3908 - val_acc: 0.9290\n",
            "Epoch 130/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1051 - acc: 0.9665 - val_loss: 0.4002 - val_acc: 0.9295\n",
            "Epoch 131/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1053 - acc: 0.9661 - val_loss: 0.4021 - val_acc: 0.9292\n",
            "Epoch 132/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.0985 - acc: 0.9690 - val_loss: 0.3996 - val_acc: 0.9307\n",
            "Epoch 133/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.1026 - acc: 0.9670 - val_loss: 0.4180 - val_acc: 0.9281\n",
            "Epoch 134/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.1001 - acc: 0.9679 - val_loss: 0.4162 - val_acc: 0.9269\n",
            "Epoch 135/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.1048 - acc: 0.9660 - val_loss: 0.4204 - val_acc: 0.9279\n",
            "Epoch 136/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.0973 - acc: 0.9683 - val_loss: 0.4067 - val_acc: 0.9291\n",
            "Epoch 137/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.0936 - acc: 0.9699 - val_loss: 0.4043 - val_acc: 0.9310\n",
            "Epoch 138/150\n",
            "259/259 [==============================] - 114s 438ms/step - loss: 0.0879 - acc: 0.9718 - val_loss: 0.4075 - val_acc: 0.9314\n",
            "Epoch 139/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.0941 - acc: 0.9693 - val_loss: 0.4310 - val_acc: 0.9264\n",
            "Epoch 140/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.0950 - acc: 0.9694 - val_loss: 0.4217 - val_acc: 0.9260\n",
            "Epoch 141/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.0856 - acc: 0.9717 - val_loss: 0.4163 - val_acc: 0.9309\n",
            "Epoch 142/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.0844 - acc: 0.9724 - val_loss: 0.4137 - val_acc: 0.9310\n",
            "Epoch 143/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.0799 - acc: 0.9745 - val_loss: 0.4146 - val_acc: 0.9336\n",
            "Epoch 144/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.0794 - acc: 0.9749 - val_loss: 0.4182 - val_acc: 0.9312\n",
            "Epoch 145/150\n",
            "259/259 [==============================] - 114s 442ms/step - loss: 0.0821 - acc: 0.9725 - val_loss: 0.4242 - val_acc: 0.9306\n",
            "Epoch 146/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.0779 - acc: 0.9744 - val_loss: 0.4129 - val_acc: 0.9346\n",
            "Epoch 147/150\n",
            "259/259 [==============================] - 115s 442ms/step - loss: 0.0762 - acc: 0.9751 - val_loss: 0.4215 - val_acc: 0.9297\n",
            "Epoch 148/150\n",
            "259/259 [==============================] - 114s 441ms/step - loss: 0.0737 - acc: 0.9753 - val_loss: 0.4132 - val_acc: 0.9347\n",
            "Epoch 149/150\n",
            "259/259 [==============================] - 114s 440ms/step - loss: 0.0683 - acc: 0.9782 - val_loss: 0.4158 - val_acc: 0.9325\n",
            "Epoch 150/150\n",
            "259/259 [==============================] - 115s 443ms/step - loss: 0.0679 - acc: 0.9781 - val_loss: 0.4278 - val_acc: 0.9346\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-6TF6ideNQrE",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 585
        },
        "outputId": "2585f972-2fe0-4793-bed5-b66fd2eab32b"
      },
      "source": [
        "sg.utils.plot_history(history)"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAfAAAAI4CAYAAACV/7uiAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOzdd3yV5cH/8c91Ts7J3glhE/aeIu5VqauOLpXWqvSxQn20tePXulprfWpbux55WmlddVuq1iq1Vq0DUVQgjCJDIAkjgeyE7OSs6/fHHUISwgiEc07g+3698krOfa5zn+vElm+ubay1iIiISN/iinQFREREpOcU4CIiIn2QAlxERKQPUoCLiIj0QQpwERGRPigm0hU4WllZWTY3NzfS1RARETkmVq1aVWmtze56vc8HeG5uLnl5eZGuhoiIyDFhjNnR3XV1oYuIiPRBYQtwY8yfjTHlxpj1B3jeGGP+zxiTb4xZZ4yZEa66iYiI9DXhbIE/AVx0kOcvBka3fc0D/hiGOomIiPRJYQtwa+1SoPogRa4AnrKOj4E0Y8yA8NRORESkb4mmMfBBQFGHx8Vt10RERKSLaArww2aMmWeMyTPG5FVUVES6OiIiImEXTQG+CxjS4fHgtmv7sdY+bK2daa2dmZ2939I4ERGR4140Bfhi4Lq22einArXW2pJIV0pERCQahW0jF2PMX4BzgSxjTDHwE8ADYK39E/AacAmQDzQBXw9X3URERPqasAW4tfYrh3jeAjeHqToiIiLt/MEQLf4gxhiSYvePxkAwRG2zn/qWAIGQJRiyFNc08WFBFcu3VVHb7MdlDC5juP9LU5g1POOY17nPb6UqIiInNmstlQ0+8ssbaPEHmTokjYxEb6cygWCI/IoGCsob2dPso7bZz86qJjaV1LG5rJ4Wf6i97LDMBKYMTiMp1s3WsgbyKxrY0+Tv9r29MS5OGprOmH7JhKwlZOn2D4BjQQEuIiJhZa2l0RfEbQyxMS4afQFWbKvmw4IqmnxBxg9IZmxOMhaoafTR7A8yLDOR0TlJuI1hw+46PtlVy5bSevIrGsgvb6C2uXPAjshOpF9yLKEQNPuDbC3vHNIA6Qkexg9I4auzhpGR6CE2xk1rIMj6XXWs2l5NSyDEqH5JXDJ5AP1T4kiN95AcF0OM24XbGDISvUwfmkacxx3G394+CnAREemR1TtreObjHRgMibFuErwxJHrdJMTGcO7YbEZmJ7WXrWxo5ZW1u9lUUsenpXWU7GlhT7OfYMjud9/YGBdxHjd/WbHzgO9tDNi2l2YmehnZL4nPTRnAqOwkRvVLIjbGxeqde1i1o4a6Zj8uF6QlePjqrGFMGZzK2P7JZCR6SYnzEOdxYYzp9d9PuCjARUSEUMiyYXcdaQkeBqfHdxts5XUt3P/6Zv62upjUeA9JsTE0+QI0+oL4Ak7r9rdvuvntlVO5ePIA1uys4ZvPrKKsrpXs5FjG9U9myuA00hM8pMR5CFlo8QeJcRlOyk1nxtB0YmNclNW1srmsHo/LkJ7oxRvjYltFI1vK6/EFQkwelMrkQan0S4nr9rOcMiLzmP6uooWxdv+/gvqSmTNnWh0nKiLi8AVCPPReAa9vKMXjdhHncVq1cTFuYj0ufIEQzf4gBhidk8z4AckUVzfzwqpidlY3AZCR6GXG0DSumDaICybm4A9aHllayCPvFxIIWm44azg3nzeq01ivPxiitLaFby9aw5qde7hs6kDeWF9KTmosf7zmJCYNSo3Qb6TvM8asstbO3O+6AlxEJPpZa9lUUk9LIIjLGOqa/Wwpq6egooGUOA9Th6SRHBfDz17dxOayemblZhDrcdHiD9Lid2ZYtwZCeGNcJHjdBIKW/IqG9pbz6SMz+cL0QfiCIf5TtIdl+VXs2tNMRqIXA1Q1+rhkcn9+cOE4hmclHrCeLf4gP3p5PS+uKuas0Vn835zppHeZUCY9owAXEYmg2mY//ynagz8Yal+GtPfLeRyiviXA5tJ6NpbUMSgtnl99eQppCV5CIcs9/9jAUx/t2O++6QkeGluD+IJOEA9IjeN/rpjE7Ak5h6xTIBiisLKRxNgYBqXFd3ouGLJ8kF/J8yuLaA2E+NZnRjF1SNphfda9f2yMyUkixh1N+4X1TQpwEZGjZK3tNDbsD4Yoq2thUFrnMeO6Fj8le1po8Qepamxl8drd/Gt9Ka2BUHe37SQrycuYnGTyttcwOD2ex+aezMNLC/nLip3MPT2Xc8dmYy3Ee92M7pdEZlIsrYEgn5bUs6O6ifPGZpMc5zkmn18iQwEuInIYAsEQ/ynew8C0eAakOq3SzaX1/PQfG1i9s4aBafEMTk+gurGVLaUN+IIhcjMT+NKMwYzITmLxf3bx7qcV7S1igJS4GK6YNoiLJ/UnKS4Gt8vgdhliXAa3y1mS5HYb4j3u9vXLK7dXM++pPBpaA/iDlv8+dyQ/uHBsn541LUdGAS4iJ7S6Fj/lda0keN3Ee9zEe93ExrgIhiw7qpvYWlbP0q2VvL6+lOpGHwAn56aTm5nIS2t2kRwXw2VTBlJR30rxnibSE7xMGJBC/9Q43txQxkeFVYDTgr586iBmDEsj3uMssTrStcLbKxv53vNr+cy4ftx83iiF97HSVA1xqeDq5r+RvwV2fACZoyF92L7rZRuhKh/6T4L04c76NoBQ0Pne3b2OkAJcRE5INY0+Hnm/kCc/3E6jL9jpOZcBY0z7muR4j5vzx/fjwon92V7ZyD8/KWFzWT1zTh7KDy8ce9DJWEXVTeza08zMYenH77hv0A8126G6EOpLIT0XciZB4gGWbYVC0LIHfA3gSXC+sOBrAn+TE3LuWLAh2LPDubfbA0NOgZSBB69LoNUJ0PJNULkV/I1O2CZkwIzrIXXQ/q/ZtQoK3m0LWQtVBVD0MezZCZ5EGDgN+k+G+HSITYbyjbDxH9BaC8YF4y6FsRfDur9C4ZJ9941NBU88tNY79bj2ZRh53pH8hrulABeRPscfDFFU3cSOqia2VzXS2OrsQ20wjMhOZOLAFDxuF+9vrWTplgrK61sIWmdNczBkCVlLUXUTTf4gl04ZyPnj+tEaCNLkC9LsD9LsCxKyluFZSYzJSWJMTvJ+LeXWQJDYmMjstLUfa/e19LrTWOUE0s6PoHobeBPBm+QE4uQv798qtNYJwYpPobESmiqd1mhjpVP2/J9AygCnbFUBPHkZ1HVzynP/yXDlk5A50nlc8A784ztQW+SE85FIHQpZo50gT8hwPlt9CTSUOd+bqjoUNk6AxsRCS1vYTvwijDgXPHHgb4bVT0HR8s7vkZTj/G4GzYC6Eti92vmDwNfgPO9NgvGXw4TLndfmPe78QZI8AGbdCLlnQ/kGKFkHIT/EpjjBP/nKfb+LXqAAF5E+YdWOGh58N5+CigaKa5q73bGrO4PS4hmelYjLZXAbcLucgyUyk2L5+hm5jMlJPsY1Pwz+FqdVV/qJEza5Z0Lq4MN77drn4LUfQHJ/GDh931fOJChaASsfgS1vABbcXsgY4QRXS60TOtnj4JzbnDCsLYayDbDldac13ZE3CRIyoaEckvrBda84reI/X+S0mj97L2SNcZ6r3gal6+CDB5ygvvppJwBfv8MpM+5zkJjl/CHhbwZfo/MHiCfRCVwbdFr1AGlDna5oXz3sXA7FK6BmB9TtdsI6Mdv57Hu/kvo7IdlvvNO9HdPWO1KzA5b/yQnsvUEMTm/BKTfB1DlOyGLAdYCeklDQaU3v/aNgL1+T83kHztj3fmGgABeRiNnT5OPR97fR4g+SFBdDUmzbV1wM/VPiGNvfafn+/u2t/OHdfLKTY5mZm8HwzERysxLJzUwgNyuR1HgPbmPwh0JsLWtgY0kdja0BzhyVxah+SeEZIw4GnNZYQiZkjz14izgUdAJ767+dr6LlTmh1lDECcs+C4WfDoJOc1/ganMDa2w288RV4Ya7zfFIO7F6zf0s4MRtmXAejL4AB05yWJzjd2JsWwzs/g6qt+8q7Y533HHMhDD7ZeX1C5r7XFa+CZ7/klPMmQmMFXP8Pp5u5q+pt8NzVULkFsDD2Evjiw21BGSG+RuePkECr8zvPHter49LhpAAXkV4VClnK61tJS/B06nb+pLiWldurOWVEBhMGpPBBfiU/eGEd5fUtxHvc+41D75Ua76G22c+XZgzmnssn9M5SqOYa2P6B08WbNmz/sA36oWKz84980Oe0TjNH72uZ7R1nrdvtdNsWr4RNr0JztfN85mgnAN0eJzACLc51a53x3N1r9rUC+0+BUec7reb+k53W3LalzteOZdBa16XyBkac47TSl9zvdPNe+3cnTAHqy6BkrdN9mzHc6eo9WKswGHDGbWNinVZ/yqBDtyLLN8FTn3fqdu3fYeipBy7bUuv0EGSMgLN/eODWrfSYAlxEjoq1zl7ZL64q5qOCKrZXNdIaCJGR6OWaU4Zy8aQBPL5sGy+uLm4/bCI7OZaK+lZG9UvigaunMWlQKsGQpdEXoKElQENrgOKaJjaV1FNQ3sAFE3O4aFLbmGtjJexa7XTxNlc7YdxU7fzsa3KCov8kGDDVaZl6Om9EQtAPT13hhCNAQpZTtt94pzu1eKXThdxS2/l1canO/ZqqnJnGoQ6nXHmTYexFMP4yp3X36auw7X3nDwNvEsTE7fsjIXmAc59BJzljsXvHkrsTDEDpf6B0vfM5PAlQth7WPAu1O53Av/5ViD+8jVR6VUOFE+C9OKYrPaMAF5HDsnd3MG+M04Ky1vL2pnJ+8+ZmPi2tx+t2ccaoTEb1S2JwegIf5FeyZNMuTjMbWGUm8rUzxjBn1lBWbq/mvS0VDM1I4NbzRzutdGudsdeStU7XrCfeaQ32n+K02JprYMUjsOZpZ2ZwR3FpzuzghAyIiXe6gxvKnOfcXicop34Fpl/r3OuNu+CjP8AF94E3AYrznLHnyi1OSzk+w5lRPPIzzuQjd4wzkaloufOHQ0JG26zkKZA6xBl3TRnotLY7CoWOXWszFHL+0Og3zvnDQk5ICnCRE1hJbTOrd+xhZm46Od2c4NTkC/DGhlLe+bSCpVsq8AdDzB6fw+wJOfx9dTHvby7hjMx6Ljl1GhedNJbUhA4hVpyH7+/fwlu1kcYRF5P4tWf3jTXmvw07P3a6l5urne7i7mYxJ2Q647DbP3C6nEee7yzDGTjdGbuMT+9+/LKhAnblwY4PneVBZZ/AkFNh3CXw77th1jy45NedXxMKOl3iyQOc0BaJcgpwkRPQ7j3N/HFJAa+u3MxU+ynvhaYwYWAap4/MZOLAVIZnJfL6hlKeW76T2mY/WUle5g4uZVzzanaWVUOghYnuIma4C/CE2sZ34zOcyVWxKeCKcUI5eYDTtZz3Z2cN7iW/gbfugY8fdF7jTXImNA2eCaMvhGGnO61xf5OzhKngHWfp06CT4MzvOl3GPWWtM1P7zbuclvyQU5xu5zDOFhY5FhTgIsexF/KKeOqjHVw1czBfnDGY2mY/C5fk8/zKYmJp4dXU3zKs6RN2p87gF55beKM0of0UKpeBCyf258aTUpi+6TeYdYsAsC4PIZeXUOYoPLmnOV3JTZXOjOP6EmhtcMZGc8+Ec++AuBR4+154/7eQMhjqimHWfLjgfzovxTnWGiqcLvhp10DyoQ/0EIl2CnCRPmzx2iLeWLMN600kLsbNFdMHcc4QDyz5BfXVpSzZXE6MC4LBEB43rA8O47nQbC4+aTR31d9H/La34NSbnElRQR/B6ddSET+c7YFsxngryahe60zo8jXCGd+Gs/6fM27cU9bCq9+FT16AyxY4m4eIyFFRgIuES2Ol00L1NTrdxjkTD1g0FLK4XF2WNtXuclqzsclYa3nyzRVMW3YTua5ybk38FetbstnT7OP93McZUPIOJaYfgZBlYHoigZClrtlHP18xIU8CrpyJziSoz/0OTr7BGfv91w9hy5sQbN33nonZTrf2uXc6E6aOVsCnrmuRXnKgANcMDpHetHO5s91kx3A87RaYfY8ze7m1ATa+DKEA7+8K8fsVDYyadibzzxvLsGTjdEEv/xM2Pp3q6Tfzt6pcLvn0NrLdDXjjEnjS+ysavvE6zz/+AAN3/5tnU+dxV9m5/OlrMxg6aQAxQBxA2QZcyxbA+r/BeXc54Q3OLOqrn3FmN9ftctYqpw52llX15iYoCm+RY04tcJHe4m+GP53ptD4v/BnWm0Rw0z+JWfWYMzM690xY+aizrWUH9Tae90OTmeYpYmCohJdds8kIlHG26xPn+ZhMEuf+DZf1O38cZI7GVnzKGu9JfHHPt/jySUP4zZVTu69ToDW8488i0uvUAhfpbTU7nD2oJ1zuhOQ7P4OqfOy1r7AkMIH/fX0L63edzw8H9eMbuxfgLlqOGfc56mbcxPV/202qrefBS9KJyX+Lsza/SaP18sCA37E7fSb9U+OxofVM3PMOmRf+EJM21HnPLzwEL1yPSR3ChBue5Xf5LVw0qf+B66jwFjluqQUuciit9fD3bzqHRpx7u9PVXFcCj13g7JKVOsTZQGTpr2mYfC1zy79C3o4aBqfHc/64fry+oZSYumLi3ZbYnFH4gyG2Vzbx4k2nMWVw285ahzplqqOCd52DH7QzlsgJQS1wOXG11DoHLWSOgpO/0f1hDHs1VTs7dg0/y9klzN+C/9k5eHZ+4GybWbMNLvw5PPMlZ2OSS/8X1jwDS39FS+JALlg/m0YauO8Lk7jypCF4Y1z8+NIJvL+1ko+3VbFxdx355Q387AuT9oU39Gz8uRfPGRaRvkstcDn+7V2b7ElwNg7JmeS0XpM6HE0Yl+YE9CcvQrCVQEIO/0z9CmmlyzjHruS7vpsYE1/LTcHnwJOIDfr45+QFPFc5ktS4GCb51/FSvsWVNYpHrptJblZipD+1iBwn1AKX488nLzrfJ3z+wFti1pfCRwth8pXO7mDr/kro03/iKt8EBUugtcNBFp5EqsZcyUs1I5m6+69c0fQAAMvG3M55E6/nV69/yvb6VO6JeY7fer/Jo8vTGD/AT3l9Kysah3DapEx++aUpJMXq/1YicuzpXxqJftY6221mjd13aMTKR+Gf33d+fuuncPotzhaeni77fC/5JYQCzlKq+DQWNp/PA1uHcuVJg7n5vFEMTAjRWLWLTwsKeWiTlzfXNJMcG8O1pz3BmCFFpJkmzpj4eQDOGpXFtxclMn7rmQzLTOTxuRM5b1y/MP4iRET2URe6RL8Vj8Br/8/ZJ/uS3zjj0C/eAGMughnXwrL/g6KPnclk5/8EJn3JCfrKrfDgKc649yW/4vm8In744jomDEhha3k9BsOwzATyKxqwFrKSvPzXmcP52qnDSDnAWdTBkOWjgipm5qZ3OgNbRORY0U5s0jeVbYSHz3XOfa4tds5gdrmdgyq+9rf2M6Bt4RLMmz+G0nXODO2YOGdHtKAPvr2Wd4st33gqj9NHZvLY9SdTXt/CQ+8VUlzTxJTBaUwbmsZpIzIVyiISdRTgEl2K8+CNO2HKVU4LuTv+ZnjkM04Q3/Shs7vXkvuhugC++DA2NoWV22t4+uMdvLG+lLE5CXwrazWzWj8iPtaLNzaeHQMv4fdFI/jHf3Yzpn8Si+adpjFqEelTNIlNokPAB0t/5cwKd8VA0XLA7Nvqc6/6MnjrJ1C+Ea75GyRlO9cv+jngnF99wyPL+aiwiuS4GD4/fSBbyxuY/8kYrB0DgNftwrcyRIK3hKtPHsJ3Zo9WeIvIcUP/mkn4NFTAoq9C8QrnqMfP3guv3Az//B4E/ZAx3Bm3LnzX2azEBp2zoUfP7nQbXyDEN59ZzfJtVdx96QTmzBpCgtf5n3J5fQurd9RQUttCSW0LA1Pj+OJJgw84pi0i0lepC12OjYAPPvgdZIyAMRc6Lepnv+yMYX/hjzDxC045fwss+goUvLPvtalDnWMop86B7LGdbhsMWW5dtIZX15Xwqy9N4aqTh4TxQ4mIhJ+60KV3BVqdddijPwtJXZZS7T0Teu0zzmO31/mKiYO5r8LgDv879MTBnL9A/luQmAWZoyExs9u3LK5p4u5XNvDOp+XccfE4hbeInNAU4NJzDRXw12uc8evYlLbjKr+xbzOVZQuc8D77h07Ab3gZaoucLvOM4fvdbkVxE39ankN6gpfhWdWkJjRQ3+KnviVAXIyb9EQPVQ0+Hl5aCMBPLpvA18/Y/z4iIicSBbj0TOl6+MscZ2b4Jb+Bza/B67fB8j/CoJlOa/zjti7yc+9w1mMPmXXA2724qpg7XlpHWoIXtzH8bXVL+3MxLkMgtG+I54IJOfzk8okMSos/ph9RRKQvUIDL4Qv4nHFsgP/6Fwyc7rS8Ny12DvQoWuGczjXkVPj8H/ftmtaFtZZNJfW8sKqIx5dt54xRmSz86kmkJnho8gVoaAmQEu8hNsZFMGTZ0+ynxR9kcHpCGD+siEh0C2uAG2MuAhYAbuBRa+0vuzw/FHgSSGsrc7u19rVw1lEOYuMrUF/iLOsaON25ZgxMuML5AvA1OoeGdHO6Vihk+f07+Tz98Q4qG1oB+NqpQ/nJZRPxuJ2wT/DGtM8oB4hxG7KSdKa1iEhXYQtwY4wbeBD4LFAMrDTGLLbWbuxQ7EfA89baPxpjJgCvAbnhqqN00Nrg7GKWkLHv2oqHIGMkjPzMgV/n7f4ULl8gxA9f/A8vr93N+eP6cdGk/pw1Opv+qXHdlhcRkYMLZwt8FpBvrS0EMMYsAq4AOga4BVLafk4FdoexfuJrghf/C3Z+6JyhbVxw+R9g+jWwaxUUr4SL7j9g13h3mn1BPi2t47dvbuGD/Ep+cOFY/vvckZienH8tIiL7CWeADwKKOjwuBk7pUuYe4E1jzLeARGA23TDGzAPmAQwdOrTXK3pCstY53WvLv+CkuZCeC/lvwz9udc7OznscvEkw7audXhYIhvj9O/ks2VLBTeeM4MKJ/Z1bfVLCn94rYFNJHSHrTEj79ZencOVMLf0SEekN0TaJ7SvAE9ba3xpjTgOeNsZMstaGOhay1j4MPAzORi4RqGffYi3YkHMIyIGsfhL+8xyccxucd6dz7aS58OhsWHQNtNY5x3XGpbS/pLimie8sWkvejhqyk2P55jOrmTYkDV8gxMaSOsbkJHHLeaOYMDCV6UPTyElRd7mISG8JZ4DvAjo2vwa3XevoBuAiAGvtR8aYOCALKA9LDY9H5Z/CC3OhbjeMvwwmfRFyz4SYtolhoRAUvA2v/cAZ2z7ntn2vjU+Hr/wVHv2MMx4+a177U2uL9nDdY8sJWVgwZxqfmzyAl1bvYsHbW/G4DQ9cPY3Lpg7E7VJXuYjIsRDOAF8JjDbGDMcJ7jnAV7uU2QmcDzxhjBkPxAEVYazj8eWTF2Hxt50jN8dc6Cz3WvuMsyvawOnOsZuFS6Cxwvn5i4/u30rPGgXX/t051jPbOSRkc2k9cx9fQWqCh2duOIVhmc7EtatOHqLd0UREwiRsAW6tDRhjbgHewFki9mdr7QZjzL1AnrV2MfB94BFjzHdxJrTNtX19s/ZI+WghvHGHc272lU9AykBn3/HCd2HHMmfNduF7kHsWjPscjL6gU/d4J4NOojFrKsEWP6W1LVz72HJiY1w8e8OpDM3U2mwRkUjQYSbHoy1vwHNXw/hL4cuPg/vIT+IKBEN8/4X/8MrafQsC0hM8PD//NEbnJPdGbUVE5CB0mMmJonwTvHgDDJgCX3joqMP7O39dy6vrSrj+tGEMyUjAGMN5Y7MZkZ3Ui5UWEZGeUoAfT3yNTsvbm+ic8HWATVUOR22Tnx+/sp5X15Vwx8XjmH/OyF6sqIiIHC0F+PFk3fOwZwdctxhSBx32yz4qqOL9rRXEe9x4Y1ws31bN+1sr8ActP7xorMJbRCQKKcCPF9bCykchZzIMP/swX2L543sF/PqNze23ABiUFs/XzxjOZVMGMnlw6rGqsYiIHAUF+PGiaDmUrYfLFnR7kEhXe5p83PXyev65roRLpwzgV1+egsftotkfJDk2RludiohEOQV4XxUMQMjvrPEGp/UdmwKTrzzoy6oaWnn0g2089eF2mv1B7rh4HPPOHtEe2HtPBRMRkeimAO9LrIX/LILN/4TCpRBshdn3wMQvwoaX4eQbDjpx7cP8SuY/s4qG1gCfmzyAWz4zinH9D7D2W0REopoCvC/5zyJ4+ZuQMggmXO6czf367fD+75zW+MnfOOBLX1pdzG1/W8fwrEQe/OoMreEWEenjFOB9RcAHS34OA6bCjUucIz2thTVPw+t3wqjZkDV6v5dVNrTypyUFPPrBNk4bkcmfrj2J1PgjXxsuIiLRQQHeV6x6AvbshEv/d9953MbAjOtg/OX7bdhSXtfCwiUFLFq5k9ZAiDknD+GnV0wkNuYgJ5KJiEifoQDvC3yNsPTXMOxMGHn+/s/Hp7X/2BoI8viy7fz+7a20BkJ8Yfog5p8zklH9tHOaiMjxRAHeF3z8R2gshznPHnSJWE2jjysf+oj88gZmj+/Hjz43gdysI9+NTUREopcCPNqtegKW/BLGXgJDZh206K/e+JRtlY08et1MZk/ICU/9REQkIhTg0aSuBF79DmSNgdwznVPF8h5zus0/v/CgL121o4a/rCjixrOGK7xFRE4ACvBo8umrsOV1yH8bPvw/59oZt8L5PwHXgSefBYIhfvTyevqnxHHr7DFhqqyIiESSAjyaFK2ApP5w61ooXgluLww9tduim0vrWbGtCgtsKqlnU0kdC6+ZQVKs/pOKiJwI9K99NCleAUNOdrZHPcCBJG9vKuPR97fxUWFVp+sXT+rPxZP6h6OWIiISBRTg0aKhHGq2w8wbDljk/a0V3PBkHoPS4rntonFcPm0gcTHOmvCMRK8OIBEROYEowKNF0Qrn+5BTDljksQ+2kZ0cyzv/7xxtyCIicoLT0VORUpkPxXn7HhevAJfH2Sq1GwUVDSzZXMHXThmm8BYREQV4xLz6HXj6C9BS6zwuWumEtyeu2+JPfrgdr9vFV08ZGsZKiohItFKAR0JLHez8CFrrYOVjEPTD7jUH3KilttnPi6uKuWzqQLKTY8NcWRERiUYK8EgoXAKhgHMs6McLnSVjgWYYfHK3xZ9fWUSTL8jXz8gNa6gfQHsAACAASURBVDVFRCR6KcAjIf/fEJsCVzwIjRXw2g+c691MYCusaOChpYXMys1g0qDUMFdURESilQI83KyFrW/BiHOdr8EnQ9l6pzWeOqhT0a1l9Vz98MdYa7n38xMjUFkREYlWCvBwK98E9bth1GznZLEzv+tc79J9vn5XLVc//DEGWDTvVMb1Twl/XUVEJGppHXi45f/b+T5qtvN9zMUw43qY+Pn2Is/nFfHjl9eTmejl2RtPZbiOBBURkS4U4OG29d/Qb+K+7nKXCy53Di7xBUL86OVPeD6vmDNGZbJgznSykjTrXERE9qcAD6fWetj5MZz2390+/avXP+X5vGK+9ZlRfGf2GNwubY0qIiLdU4CHU+F7EPLv6z7v4IOtlTz6wTauPXUY379gbAQqJyIifYkmsYVT/lvgTYIhnY8IrWn08f0X1jIyO5E7LxkfocqJiEhfohZ4uFjrBPiIcyHG2+Gy5c6/f0J1o4/Hrj+ZeK/2ORcRkUNTCzxcKjZDbdF+3ecvrCrmX+tL+f4FY7VRi4iIHDYFeLh0XT4G7Khq5KeLN3DqiAxuPGtEhComIiJ9kQI8XLb+G7LHQdoQAALBEN/561rcLsPvrpqmGeciItIjCvBwaG1wTh/r0Pp+aGkha3bu4b4vTGZgWnwEKyciIn2RAjwctr8PQR+M/iwAoZDl2Y93cPaYbC6bOjDClRMRkb5IAX6sNJRDxZa2w0v+DZ5EGHoaAGuKathd28IXpiu8RUTkyGgZ2bHy/HVOt3naMGipheFnQ4yzLeqr60rwxriYPT4nwpUUEZG+KqwtcGPMRcaYzcaYfGPM7Qcoc5UxZqMxZoMx5rlw1q/XtNZD0QoYeT70Gw+hIEy5EnC6z1/7pIRzx2STHOeJcEVFRKSvClsL3BjjBh4EPgsUAyuNMYuttRs7lBkN3AGcYa2tMcb0C1f9etXOj8EG4fRvwcjzOj2Vt6OGsrpWPjdlQIQqJyIix4NwtsBnAfnW2kJrrQ9YBFzRpcyNwIPW2hoAa215GOvXe7a/Dy4PDDllv6f+uW43sTEuzlf3uYiIHIVwBvggoKjD4+K2ax2NAcYYY5YZYz42xlzU3Y2MMfOMMXnGmLyKiopjVN2jsP0DGDwTvAmdLgdDltfWl/KZcf1IitX0AxEROXLRNgs9BhgNnAt8BXjEGJPWtZC19mFr7Uxr7czs7OwwV/EQWupg91rIPXO/p1Zsq6aiXt3nIiJy9MIZ4LuAIR0eD2671lExsNha67fWbgO24AR637F3/Dv3rP2eemtTGV63i/PG9s2hfRERiR7hDPCVwGhjzHBjjBeYAyzuUuZlnNY3xpgsnC71wjDW8ehtfx/cXhh88n5Pvbu5nFNGZJCo7nMRETlKYQtwa20AuAV4A9gEPG+t3WCMudcYc3lbsTeAKmPMRuBd4AfW2qpw1bFXbH8fBu0//r2jqpHCikY+M06tbxEROXphbQpaa18DXuty7e4OP1vge21ffU9LLZT8B87+wX5PvfOpM6FeAS4iIr0h2iax9W07PwYb6nYC2zufljMiO5FhmYkRqJiIiBxvFOC9aefH4IrZb/y7yRdgeWG1Jq+JiEivUYD3pl2rIGcieDofD7osvwpfMKTucxER6TUK8N4SCsHuNTDopP2eeufTcpJiYzg5NyMCFRMRkeORAry3VBdAa91+AW6tZcnmcs4clYU3Rr9uERHpHUqU3rJrlfN94IxOlzeW1FFS26LucxER6VUK8N6yazV4EiF7bKfL/95YhjFwngJcRER6kQK8t+xaBQOngcvd6fJbm8qYMTSd7OTYCFVMRESORwrw3hDwQeknMKhz9/nuPc2s31XHbB0dKiIiveywA9wYM9cYc1U3168yxlzXu9XqY8o3QLB1vwlsb28qA+CzExTgIiLSu3rSAr8NqO7meiVwe+9Up4/atdr53mUC25sbyxielcjIbO2+JiIivasnAZ4L5HdzvbDtuRPXrtWQkAVpQ9sv1bf4+biwis9OyMEYE8HKiYjI8agnAV4LDO/m+kigoXeq00ftWuWMf3cI6ve2VOAPWo1/i4jIMdGTAP8X8GtjzIC9F4wxA4H76XLC2AmltR4qN+83/v3mhjLSEzycNCw9QhUTEZHjWU8C/IdAIlBgjMkzxuThdKkntj13Ytr0atsJZGe1Xyqvb+H19aV8bsoA3C51n4uISO877PPArbUVxpjpwDXA3tlaC4G/WGubj0Xl+oTVT0LmKBh2evulZz7agT8U4r/O6G7EQURE5OgddoADWGtbgMfavqRiM+z8CD57b/v4d7MvyNMf7+D8cTmMyE6KcAVFROR41ZN14LcbY27o5voNxpgTswt99VPg8sDUr7ZfemlNMTVNfm48S61vERE5dnoyBj4P2NzN9U3A/N6pTh8SaIW1z8G4SyApG4BQyPLY+9uYPCiVWcN1dKiIiBw7PQnwgUBxN9d3A4N6pzp9yKevQnM1zLi+/dKSLeUUVjbyjbOGa+23iIgcUz0J8HJgcjfXpwBVvVOdPmT1087GLSPOa7/0+vpSUuJiuGTygIO8UERE5Oj1JMBfAv63bSY6AMaYGcBvgRd7u2JRzd8CO5bB+MvB5fwKrbUs3VLJmaOz8Lh1RoyIiBxbPZmFfhcwDVhljNm7J3oG8D5wZ29XLKrtWgVBH+Se2X4pv7yB0roWzhqdHcGKiYjIiaIn68AbgXONMZ8B9m47tspa+84xqVk02/EhYGDoqe2X3ttSAcDZYxTgIiJy7PVoHbgxJh3IAdyAFzjTGHMmgLX23t6vXpTasQxyJkL8vm1Sl26tZGR2IoPS4iNYMREROVEcdoAbY04GXgcMkAJUAP2AJqAEODECPOiHohUw/Zr2Sy3+IMsLq/jKrKEHeaGIiEjv6clsq18DfwOygGbgDGAYsAbnrPATQ8k68DfC0NPaL63cXk1rIMQ56j4XEZEw6UmATwP+11obAkKA11pbjBPePz8WlYtKO5Y53zvsfb50SwVet4tTRmjzFhERCY+eBHgQ8Lf9XA4Mafu5EqclfmLY8SFkjITk/u2X3t9ayczcdBK8PZpSICIicsR6EuDrcFrhAB8DdxpjLgR+QfdbrB5/QiHY+WGn1nd5XQufltZr+ZiIiIRVT5qM9wF7j9f6MfBP4F84k9m+3Mv1ik7lG6GlFoad0X5pWUElAGeNzopUrURE5ATUk3Xgb3X4eTsw0RiTAdRYa+0xqFv02fGh871DC/zD/CpS4z1MGJASoUqJiMiJ6KgGba211YcudRwp3wDxGc4e6Djbp35YUMVpIzJxuXR4iYiIhI827e6J6kLIHAltJ43trG5i155mzhiVGeGKiYjIiUYB3hPV25wZ6G2W5TuHsJ0+SuPfIiISXgrww+VvgdpiyBjRfunDgkpyUmIZkZUYwYqJiMiJSAF+uGq2A7Y9wEMhy0cFVZwxMgtjNP4tIiLhpQA/XNWFzve2AN9SXk9Vo4/TRmr8W0REwk8BfrjaA3w4sG/8+wyNf4uISASENcCNMRcZYzYbY/KNMbcfpNyXjDHWGDMznPU7qOpCiEuDBGe/848KKhmelchAHR8qIiIRELYAN8a4gQeBi4EJwFeMMRO6KZcM3AosD1fdDkt1YacJbOt31TF9aFoEKyQiIieycLbAZwH51tpCa60PWARc0U25/wHuB1rCWLdD6xDgrYEgZfUtDM1IiHClRETkRBXOAB8EFHV4XNx2rZ0xZgYwxFr7z4PdyBgzzxiTZ4zJq6io6P2adhXwQW1Re4DvqmnGWhiSrgAXEZHIiJpJbMYYF/A74PuHKmutfdhaO9NaOzM7OwyngO3ZCTbk7MIGFNU0AzBELXAREYmQcAb4LvadIQ4wuO3aXsnAJGCJMWY7cCqwOComsnVZQlZU3QTAkAxNYBMRkcgIZ4CvBEYbY4YbY7zAHGDx3iettbXW2ixrba61NhfnzPHLrbV5Yaxj97oGeE0TXreLnOS4CFZKREROZGELcGttALgFeAPYBDxvrd1gjLnXGHN5uOpxRKoLITYFEpxNW4qrmxmUHq8TyEREJGKO6jjRnrLWvga81uXa3Qcoe2446nRYqgudDVzatkwtqmlicLq6z0VEJHKiZhJbVKsu6LQGvKi6SRPYREQkohTghxL0O7PQ2wK8oTVATZNfS8hERCSiFOCHUlsEoYBmoIuISFRRgB9K9Tbne7pziMneAB+sFriIiESQAvxQ9uxwvqcPAzps4qJJbCIiEkEK8EOp2QEuDyQPAJwWeILXTUaiN8IVExGRE5kC/FD27IC0IeByA1Bc08yQ9ASM0RpwERGJHAX4odTsgLRh7Q+La5o0gU1ERCJOAX4oe3a0j39baymqbtIENhERiTgF+MG0NkBTVXsLvKbJT6MvqE1cREQk4hTgB7Nnp/N97wz0vWvANQNdREQiTAF+MHuXkKXlAs4e6KBzwEVEJPIU4AdT02UNeHXbGnAFuIiIRJgC/GD27ABPYvsxooUVDWQlxZIUG9ZD3ERERPajAD+YmrYZ6G1rvreU1TOuf3KEKyUiIqIAP7g9+9aAh0KWLWUNjMlRgIuISOQpwA/E2n0tcJwd2Jr9Qcb2T4pwxURERBTgB9ZcA7769hb45rJ6ALXARUQkKijAD6Rmu/O9rQW+ubQOgNEKcBERiQIK8ANpXwM+FIDNZQ0MTo/XDHQREYkKCvAD2bsGvK0LfUtpPWPV+hYRkSihAD+QPTsgPh3iUvAFQhRUNDBGS8hERCRKKMAPpMMxoturGgmErFrgIiISNRTgB9LhGNHNpZqBLiIi0UUB3h1roXYXpA4BnB3Y3C7DiOzECFdMRETEoQDvTmsdBJohuT/gtMBzMxOI87gjXDERERGHArw79WXO9yQnwLeU1TNWE9hERCSKKMC701DqfE/OodkXZEd1k8a/RUQkqijAu9OhBV5Q0YC1msAmIiLRRQHenfoS53tyDhX1rQAMSI2LYIVEREQ6U4B3p6EMPAkQm0JlgxPgmYmxEa6UiIjIPgrw7tSXQlIOGEN1ow+AjCRvhCslIiKyjwK8Ow1l7UvIqhp9xMa4SPRqCZmIiEQPBXh39rbAgaoGH5mJXowxEa6UiIjIPgrw7nRogVc3tqr7XEREoo4CvCtfo7MT294WeKNPE9hERCTqKMC7qt+7icsAYF8XuoiISDRRgHfV0LaJS7LTAq9u9JGhABcRkSijAO9qbws8qT9NvgDN/iCZSepCFxGR6BLWADfGXGSM2WyMyTfG3N7N898zxmw0xqwzxrxtjBkWzvoBHVrg/alqcNaAqwtdRESiTdgC3BjjBh4ELgYmAF8xxkzoUmwNMNNaOwV4EfhVuOrXrr4U3F6IT9+3iYsCXEREokw4W+CzgHxrbaG11gcsAq7oWMBa+661tqnt4cfA4DDWz9FhF7aqxrZtVLWMTEREokxMGN9rEFDU4XExcMpByt8A/OuY1qg7DZ03cQHtgy4ikREKhSguLqaxsTHSVZFjJDExkcGDB+Ny9bw9Hc4AP2zGmK8BM4FzDvD8PGAewNChQ3v3zevLIHMkgPZBF5GIqqysxBjD2LFjj+gfeIluoVCIXbt2UVlZSb9+/Xr8+nD+L2IXMKTD48Ft1zoxxswG7gIut9a2dncja+3D1tqZ1tqZ2dnZvVvLhlLtgy4iUWHPnj3k5OQovI9TLpeLnJwcamtrj+z1vVyfg1kJjDbGDDfGeIE5wOKOBYwx04GHcMK7PIx1cwRaobkGktoCXPugi0gEBYNBPB5PpKshx5DH4yEQCBzRa8MW4NbaAHAL8AawCXjeWrvBGHOvMebytmK/BpKAF4wxa40xiw9wu2OjfRe2vduoah90EYksNSCOb0fz3zesY+DW2teA17pcu7vDz7PDWZ/97F0DnrT3IBPtgy4iItFJAysddW2Bax90ERGJUgrwjtp3YWs7yKSxVZu4iIhE2BNPPEFMTFQumoooBXhH9aVg3JCQRZMvQIs/pH3QRUSOwOzZs5k7d26v3Ovqq69m1679Fi2d8PQnTUcNpZDUD1wuqhpaAO2DLiJyrPh8PrzeQ/8bGx8fT3x8fBhq1LeoBd7R2T+Aq54GnDXgoH3QRUR6au7cubz99ts8+eSTGGMwxvDEE09gjOHZZ5/lkksuITExkR//+MdYa7nxxhsZOXIk8fHxjBgxgjvvvJPW1n3bgHTtQt/7eNmyZcyYMYOEhAROOukkVq5cGYmPGzFqgXeUnut8AdXaB11EosxP/7GBjbvrwv6+Ewam8JPLJh52+QULFlBYWMiAAQNYsGABAHV1Tr1vu+027r//fh588EEArLX069eP5557jpycHNatW8f8+fPxeDz89Kc/PeB7hEIh7rjjDhYsWEB2djbf/e53ueqqq9i6desJM15+YnzKI6B90EVEjkxqaiper5f4+Hj693eW5ba0OMOS8+fP55prrulU/r777mv/OTc3l4KCAhYuXHjQALfW8sADDzBjxgwA7rnnHk499VQKCgoYO3Zsb3+kqKQAP4Aq7YMuIlGmJ63gaDVr1qz9rj3yyCM8+uijbN++ncbGRgKBAKFQ6KD3McYwderU9scDBw4EoKys7IQJcI2BH0C19kEXEel1iYmJnR6/8MIL3HzzzVx99dW89tprrFmzhrvvvhu/33/Q+7hcLtzuff8+793R7FDBfzxRC/wAKhtatQ+6iMgR8nq9BIPBQ5ZbunQp06dP53vf+177te3btx/Dmh0/1AI/gOpGn7rPRUSO0PDhw1m1ahUFBQVUVlYesEU9duxYPvnkE1555RUKCgpYsGABL730Uphr2zcpwA9A+6CLiBy573//+2RlZTF16lSys7NZtmxZt+Xmz5/Ptddey9e//nWmT5/O8uXLueeee8Jb2T7KWGsjXYejMnPmTJuXl9fr9z3jl+9wyvAMfnf1tF6/t4jI4di0aRPjx4+PdDXkGDvUf2djzCpr7cyu19UC70YoZCmvbyE7WS1wERGJTgrwblQ2tuIPWgamaes+ERGJTgrwbpTscTYcGJAaF+GaiIiIdE8B3o3de5oB1AIXEZGopQDvxu5apwWuABcRkWilAO9GyZ5m4jwu0hM8ka6KiIhItxTg3dhd28zA1HjtwiYiIlFLAd6N3XtaGJCmCWwiIhK9FODd2L3HaYGLiIhEKwV4F75AiIqGVgZoApuISMQ88cQTxMTsO29ryZIlGGMoLi4+6OuMMTzzzDNH/f5z585l9uzZR32fY0kB3kVZXQvWwiB1oYuIRI3TTz+dkpKS9nO/e8szzzzT7XynBQsW8MILL/Tqe/U2HSfaxd414APUhS4iEjW8Xi/9+/cP2/ulpqaG7b2OlFrgXZS0rwFXC1xE5Eg88sgjpKam0tLS0un6/fffz9ChQwkGg9x4442MHDmS+Ph4RowYwZ133klra+sB79ldF/q7777LlClTiIuLY8qUKbz77rv7ve6uu+5i/PjxJCQkMGTIEL75zW9SW1vbfs9rr70WcLrejTHMnTsX2L8L3VrLb37zG0aMGIHX62XkyJE88MADnd4rNzeXu+++m1tvvZWMjAxycnL47ne/SyAQ6Nkv8DCpBd7FLrXARSRa/et2KP0k/O/bfzJc/MvDLn7VVVfx7W9/m1deeYWrr766/fpTTz3F1772NYwx9OvXj+eee46cnBzWrVvH/Pnz8Xg8/PSnPz2s99i9ezeXXnopV111FYsWLWLXrl3ceuut+5WLj4/n4YcfZsiQIRQUFHDzzTfz7W9/myeffJLTTz+dP/zhD9xyyy2UlJS0l+/OwoUL+fGPf8yCBQs477zzePvtt/nOd75DcnIyN9xwQ3u53//+99x2220sX76cNWvWcM011zBp0qROZXqLAryLktpmUuM9JMbqVyMiciRSU1O54ooreOqpp9oDPC8vj40bN/LSSy/hcrm477772svn5uZSUFDAwoULDzvAFy5cSFZWFo888ggxMTFMmDCBn//851x22WWdyv3oRz/q9D6/+MUvmDNnDo8//jher7e9q/xQ3fO//OUv+da3vsW8efMAGD16NJs3b+a+++7rFM5nnXUWt99+e3uZxx9/nLfeeksBHg4le1q0haqIRKcetIIj7frrr+fyyy+nvLycfv368dRTTzFr1izGjh0LON3sjz76KNu3b6exsZFAIEAoFDrs+2/cuJFZs2Z1mql+5pln7lfupZde4oEHHiA/P5+6ujpCoRA+n4/S0tLDnhBXV1dHcXExZ599dqfr55xzDgsWLKCpqYmEhAQApk2b1qnMwIED2bZt22F/rp7QGHgXu/Y0M1CnkImIHJULLriArKwsnnvuOfx+P4sWLeL6668H4IUXXuDmm2/m6quv5rXXXmPNmjXcfffd+P3+Xq3D8uXLufLKKzn77LP5+9//zurVq/nTn/4EgM/n69X32svr9XZ6bIzp0R8mPaEWeBcltS3MzE2PdDVERPo0t9vNNddcw9NPP82IESOora1lzpw5ACxdupTp06fzve99r7389u3be3T/CRMm8PTTTxMMBnG73QAsW7asU5kPPviArKwsfvazn7Vfe/HFFzuV2Ru4He/TVUpKCoMHD2bp0qVceuml7dffe+89hg8f3t76Dje1wDtobA1Q2+xXF7qISC+47rrrWL16NT/5yU+49NJLycjIAGDs2LF88sknvPLKKxQUFLBgwQJeeumlHt37pptuoqKignnz5rFp0ybefvtt7rrrrk5lxo4dS0VFBY899hiFhYU89dRTLFy4sFOZ4cOHA7B48WIqKipoaGjo9v3uuOMOfv/73/PII4+wdetWHnroIf74xz9y55139qjevUkB3kFJbds54JqBLiJy1KZMmcK0adNYu3Yt1113Xfv1+fPnc+211/L1r3+d6dOns3z5cu65554e3XvQoEH84x//YMWKFUybNo1bb72V3/3ud53KXHrppdx1113ceeedTJ48mUWLFvHrX/+6U5mTTz6ZW2+9lfnz59OvXz9uueWWbt/vpptu4t577+XnP/85EyZM4P777+eXv/zlMZmcdriMtTZib94bZs6cafPy8nrlXku3VHDdn1fw13mncsqIzF65p4jIkdq0aRPjx4+PdDXkGDvUf2djzCpr7cyu19UC72DvLmzqQhcRkWinAO9gd20LxkB/zUIXEZEopwDvoGRPM/2SY/G49WsREZHopmVkHXzns2O45tRhka6GiIjIISnAOxiUFs8gjX+LSBSx1nZ73KUcH45mIrn6ikVEopTb7e713ckkuvj9/k7bwfZEWAPcGHORMWazMSbfGHN7N8/HGmP+2vb8cmNMbjjrJyISTdLS0igrKztmW3FKZIVCIcrKyo747PGwdaEbY9zAg8BngWJgpTFmsbV2Y4diNwA11tpRxpg5wP3A1fvfTUTk+JeVlUVxcTGbN2+OdFXkGElMTCQrK+uIXhvOMfBZQL61thDAGLMIuALoGOBXAPe0/fwi8AdjjLF9fbcZEZEj4HK5GDp0aKSrIVEqnF3og4CiDo+L2651W8ZaGwBqgf22RDPGzDPG5Blj8ioqKo5RdUVERKJXn5zEZq192Fo701o7Mzs7O9LVERERCbtwBvguYEiHx4PbrnVbxhgTA6QCVWGpnYiISB8SzgBfCYw2xgw3xniBOcDiLmUWA9e3/fxl4B2Nf4uIiOwvbJPYrLUBY8wtwBuAG/iztXaDMeZeIM9auxh4DHjaGJMPVOOEvIiIiHTR548TNcZUADt68ZZZQGUv3i9a6HP1Lfpcfcvx+LmOx88EffNzDbPW7jfhq88HeG8zxuR1d+5qX6fP1bfoc/Utx+PnOh4/Exxfn6tPzkIXERE50SnARURE+iAF+P4ejnQFjhF9rr5Fn6tvOR4/1/H4meA4+lwaAxcREemD1AIXERHpgxTgIiIifZACXEREpA9SgIuIiPRBCnAREZE+SAEuIiLSBynARURE+iAFuIiISB8UtuNEj5WsrCybm5sb6WqIiIgcE6tWrars7jSyPh/gubm55OXlRboaIiIix4Qxptsjs9WFLiIi0gcpwEVERPogBbiIiEgfpAAXERHpgxTgIiIifZACXEREpA9SgIuIiPRBCnAREZE+SAHegbWWFn8w0tUQERE5JAV4Bz94cR3n//a9SFdDRETkkBTgHSTHxVDX7I90NURERA5JAd5BaryH+tYAwZCNdFVEREQOSgHeQUqcB4D6FrXCRUQkuinAO0iJdwK8rjkQ4ZqIiIgcnAK8g9S2AK/VOLiIiEQ5BXgHKXHO8eh16kIXEZEopwDvYF8XugJcRESimwK8A3Whi4hIX6EA76C9Ba4udBERiXIK8A4SvW7cLqNZ6CIiEvUU4B0YY0iJi1EXuoiIRD0FeBcp8R51oYuISNRTgHeRGu9RC1xERKKeAryLlDiPlpGJiEjUU4B3kRIfQ12LJrGJiEh0U4B3oS50ERHpCxTgXagLXURE+gIFeBcp8R5aAyFa/MFIV0VEROSAFOBdaDc2ERHpCxTgXbSfSKbd2EREJIopwLtQC1xERPoCBXgXOpFMRET6AgV4FylxOhNcRESiX1QGuDHGbYxZY4x5NdzvnRLfNgauzVxERCSKRWWAA7cCmyLxxmqBi4hIXxB1AW6MGQx8Dng0Eu8f53ETG+NSgIuISFSLugAHHgB+CIQOVMAYM88Yk2eMyauoqOj1CuhIURERiXZRFeDGmEuBcmvtqoOVs9Y+bK2daa2dmf3/27vvMLmu+v7j7++07b1oV1qtqq1eLMu2jMA2BuOKbcAE0zEQh18oJgkphJCEhBACxIQOjg2EZoMxYOOGCzbukmVZlqxi1VXX9r47Ozsz5/fHGUmr9cpNuzszq8/reeaZmTvlnrtXms895557TlXVqJdD46GLiEimy6gAB1YCl5tZA3ALcL6Z/Wy8C1GcG9JALiIiktEyKsCdc591ztU556YDVwN/dM69b7zLoRq4iIhkO+Vw+QAAIABJREFUuowK8Eyhc+AiIpLpQukuwPE45x4GHk7HujWlqIiIZDrVwEdQkhemKxrHOZfuooiIiIxIAT6C4rwQiaSjN6Y5wUVEJDMpwEeg0dhERCTTKcBHoBnJREQk0ynAR3BkTnAFuIiIZCgF+AiONKFrRjIREclQGXsZWVp07IHeFkry5gJqQhcRkcylGvhQf/ovuOU9R+cEV4CLiEiGUoAPlVsK/R0UHWlCV4CLiEhmUoAPlVcK8X6CyRhFOSE1oYuISMZSgA+VW+rv+zsozgvT2acAFxGRzKQAHyqvzN9HOygrCNPeF0tveURERI5DAT7UkBp4WX6EDjWhi4hIhlKAD5WXCvBoB6X5ETrUhC4iIhlKAT7UkBp4aZ6a0EVEJHMpwIcaUgMvyw/T2T9IIqkpRUVEJPMowIfKLfH3/b4J3TkN5iIiIplJAT5UMAyRwiO90AF1ZBMRkYykAB8uNRpbaX4EQOfBRUQkIynAh8sr9b3QU1OKdijARUQkAynAh0vVwMsO18B71YQuIiKZRwE+XKoGXqYmdBERyWAK8OGOzEgWImBoMBcREclICvDhUjXwQMD8aGz9qoGLiEjmUYAPl1sKg30Qj6VGY1MNXEREMo8CfLhjxkMPqxe6iIhkJAX4cMNmJFMvdBERyUQK8OFeNCOZauAiIpJ5FODDHVMDD2soVRERyUgK8OGGnQPviyUYiCfSWyYREZFhFODDDZ0TPDWYi64FFxGRTKMAH+6YOcE1GpuIiGQmBfhwwTCEC46cAweNhy4iIplHAT6SwzOSpWrgnRqNTUREMowCfCS5Jalz4KkauM6Bi4hIhlGAjyS3FKKdOgcuIiIZSwE+klQTel4kSE4ooF7oIiKScRTgI0lNKQqkhlNVDVxERDKLAnwkqRo44Cc00WhsIiKSYRTgI8kthVgPJAY1I5mIiGQkBfhIjgzm4juyqRe6iIhkGgX4SIYNp6oauIiIZBoF+EiOGU41TEffIM659JZJRERkiIwLcDPLNbPVZvacmW00sy+MeyGOmVI0Qjzp6BmIj3sxREREjifjAhwYAM53zi0BlgIXmdmKcS3BkBp4SWo0Nl0LLiIimSTjAtx5Pamn4dRtfNuvj9TA2zUam4iIZKSMC3AAMwua2TqgCbjfObdq2OvXmtkaM1vT3Nw8+gUYUgMvL/A18FYN5iIiIhkkIwPcOZdwzi0F6oAzzWzhsNdvcM4td84tr6qqGv0ChHIglAf9HdSV5QOwr61v9NcjIiLyGmVkgB/mnOsAHgIuGveVF1RCbzPVRTnkhYM0tCrARUQkc2RcgJtZlZmVph7nARcAW8a9IMWToesAZsa0inwaWnrHvQgiIiLHE0p3AUZQC/yfmQXxBxi/cs7dOe6lKJ4CB9cBML2igG1N3eNeBBERkePJuAB3zq0HTkt3OSieDC/cDc4xrTKfP25pIpF0BAOW7pKJiIhkXhN6xiieAvEo9LczvaKAWCLJwc7+dJdKREQEUIAfX8kUf9+5j2kVvif6bnVkExGRDKEAP57iVIB3HWB6RQEADa3qyCYiIplBAX48xZP9fdd+aopziYQCqoGLiEjGUIAfT+EksCB07ScQMKaV61IyERHJHArw4wkEoagWug4AMK2iQDVwERHJGArwl1IyBbr2AzC9Ip/dbb0kk5oXXERE0k8B/lKKJ0OnD/BplQVEB5M0dQ+kuVAiIiIK8JdWPMU3oTvH9NSlZLt0HlxERDKAAvylFE+BeP+RwVwAdutSMhERyQAK8Jcy5FKyyaV5hIOmWclERCQjKMBfSkmdv+86QDBgTC3PVw1cREQyggL8pRyugXfuA/ysZKqBi4hIJlCAv5Qjg7kcvhbc18Cd06VkIiKSXgrwlzJsMJfpFQX0xRI09+hSMhERSS8F+Mspngxdvglds5KJiEimUIC/nJIpx9TAAY2JLiIiaacAfzlDBnOZUpZHMGCqgYuISNopwF9O8WQY7IP+dsLBAHVleZoXXERE0k4B/nKKp/h7zUomIiIZZMwD3MwKzexSMztlrNc1Jkqm+vuOPYCflayhRZeSiYhIeo16gJvZL8zsU6nHYWAV8Htgo5ldNtrrG3OVqeOO5s2Ar4F3D8Rp642lsVAiInKyG4sa+HnA46nHbwWKgFrgX4HPj8H6xlZusa+FN20BODIrmUZkExGRdBqLAC8HGlOPLwB+45xrBH4BzBuD9Y296nnQ5Gvg0ys1K5mIiKTfWAR4MzAj9fgC4KHU43wgOQbrG3vV86DlBUjEqSvLI2CqgYuISHqNRYDfCvzczB4AioH7U8uXAtvGYH1jr3o+JGLQtpOcUJDJpXmqgYuISFqNRYD/HfA/wPPABc65w1XVycD/jsH6xl7VXH/ftAnQrGQiIpJ+odH+QudcHLh+hOVfG+11jZuqOYD58+ALrmRaRT53bTiY7lKJiMhJbCwuI1tiZguGPL/EzG41s381s1E/YBgX4Twon3lMDbyjb5COPl1KJiIi6TEWTeg/ABYBmFkd8GugEPhz4ItjsL7xUT0Pmv2lZJqVTERE0m0sAnwO8Gzq8duBp51zFwMfAN41BusbH9XzoHUHDEaPXEqmMdFFRCRdxiLAI0A09fg84J7U461AzRisb3xUzwOXgNZt1JenBnNpUQ1cRETSYywC/AXgKjOrx18H/kBqeS3QPgbrGx/V8/1902Zyw0FqS3LZ2dKT3jKJiMhJaywC/AvAl4BdwGPOuTWp5W/haNN69imfBYHwkY5sp9WXsmpnmyY1ERGRtBj1AHfO3Q7UA6cDlw556UHgb0d7feMmFPETm6SGVD331CoOdUXZ1qRauIiIjL8xmU7UOdfonFsHRMwsN7XsSefcprFY37ipmnskwM85tQqAP73QnM4SiYjISWpMAtzMrjGz7UAP0GNm28zsQ2OxrnFVsxA6dkNfG7UleZw6qZA/bVWAi4jI+BuLgVyuA74L3AG8I3W7E/iumX1ytNc3rqae5e/3PQ34ZvTVu9roi8XTWCgRETkZjUUN/JPAdc65v3bO3Z66/RXwV8B1Y7C+8TN5GVgQ9q4C4NxTq4klkjy1szXNBRMRkZPNWAT4VHyHteEeTL2WvSL5ULsY9q4GYPn0MnLDAZ0HFxGRcTcWAb4PP4DLcOelXstudWfC/mcgESc3HOTsmRU6Dy4iIuNuLAL8e8A3zew/UxOZXGJmXwa+gT83nt2mngmDfdD4PODPgze09ml+cBERGVdjcR341/Bzgr8X33ntTuA9wGecc//9Up81s6lm9pCZbTKzjakOcZnlcEe2VDP6uXOqAXhwc1O6SiQiIiehsboO/DvOuXqgBChxztU75773Cj4aB/7GOTcfWAF83Mzmj0UZX7OSOiiafKQj24zKAuZMKuLe5w+luWAiInIyGZX5uc3svpd5/chj59xbjvc+59xB4GDqcbeZbQamAJkzAIyZb0ZP1cABLl5Uwzce3EZTV5Tq4tw0Fk5ERE4Wo1UD3/8qbq+ImU0HTgNWjVIZR8/Us6BzD3QdAOCSRbU4B3/YqFq4iIiMj1GpgTvnrhmN7znMzAqB24BPO+e6Rnj9WuBagPr6+tFc9Ssz9Dz4gis5dVIRs6sLuWvDQd5/9vTxL4+IiJx0xuQc+IkwszA+vH/unPvNSO9xzt3gnFvunFteVVU1vgUEqFkEoVzY89SRRZcsrGH1rjaauwfGvzwiInLSyagAN3+y/CZgs3Pu+nSX57hCEZj5Rlh/C0R9A8HFi2pJOrhvk5rRRURk7GVUgAMrgfcD55vZutTtknQXakTn/T30t8OT3wFgbk0RMyoLuGeDAlxERMZeRgW4c+4x55w55xY755ambnenu1wjmnwazLvcB3hfG2bGxQtreHJnKwc7+9NdOhERmeAyKsCzzhs/B7EeeOzrALz7TN+h7oeP7UpnqURE5CSgAD8R1XNhydWw+gboOsjU8nwuW1zLL1btobNvMN2lExGRCUwBfqLO+VuIR2Gj7zD/F+fMojeW4Gerdqe5YCIiMpEpwE9UxSworT9ySdn8ycWce2oVP3p8F9HBRJoLJyIiE5UCfDTUn+3HRncOgI+dO4uWnhi/fib7Z08VEZHMpAAfDfUroKcR2n3ntRUzy1lSV8JPnmxIa7FERGTiUoCPhqkr/P0eP2y7mfGO0+vY2tjDlkMvGglWRETkhCnAR0PVXMgtgT1PHll0yaJaggHj9nUH0lgwERGZqBTgoyEQ8BOcDBkbvbIwh5WzK/n9cwdwqXPjIiIio0UBPlrqV0DLC9DXdmTR5Usms6+9n7V7OtJYMBERmYgU4KPl8HnwvUenL79wwSQioQC/f07N6CIiMroU4KNlyjIIhI9pRi/KDXP+nGruXH+AeCKZxsKJiMhEowAfLeE8mLz0mAAHuHzpZFp6Yjy5szVNBRMRkYlIAT6a6lfAgbUweHQ2svPnVlOUE+K3a/ensWAiIjLRKMBH06zzIRGD7Q8eWZQbDnLZklruef4QPQPxNBZOREQmEgX4aJr+Bsgrg813HLP4qtPr6B9McPeGg2kqmIiITDQK8NEUDMOcS+GFeyA+cGTxsvoyZlQWaGx0EREZNQrw0Tb/Chjogp1/OrLIzLjq9DpW72pjT2tfGgsnIiIThQJ8tM08F3KKYfPtxyx+22lTMIPb1qoWLiIiJ04BPtpCOTDnYthyFyQGjyyeXJrHylmV3LZ2H8mkhlYVEZETowAfC/Muh/52aHgMkkno2AvJJO9cXse+9n4e3d6S7hKKiEiWC6W7ABPS7DdBuAB++zGI9fjbRV/mouXXUlmYw/890cC5p1alu5QiIpLFVAMfC+E8WHkdlE2Hpe+Bitmw/pfkhIK896x6/riliV0tvekupYiIZDEF+Fg57+/hI3+AS74Kyz4AB56Ftl2896x6wkHjJ082pLuEIiKSxRTg42H+Ff5+0+1UF+dy6aJabl2zTyOziYjIa6YAHw9l02HyMtj4WwA+tHIGPQNxbtPALiIi8hopwMfLgivh4Dpo28XSqaUsnVrKdx7azt42DewiIiKvngJ8vMy/0t9v+h0AX37HIgbiSd79v09xsLP/JT4oIiLyYgrw8VI2LdWM7gN8bk0xP/3ImXT2DfLe/11Fc/fAy3yBiIjIUQrw8bTgbb4ZvWU7AIvrSvnRNWdwsDPKJ36xloRGaBMRkVdIAT6eFv8ZBEKw9sdHFi2fXs6/X7mQVbva+P6fdqSvbCIiklUU4OOpqMaPk77uF8dMN/qOZVO4bHEt19+/lWf3tKexgCIiki0U4OPt9GugrxU2//7IIjPjP962iJriXK67ZZ3Oh4uIyMtSgI+3mW/014Wv+dExi0vywnzj6qU0dkW57FuP8nRDW3rKJyIiWUEBPt4CAVj2Qdj9GDRvha33wY8ugY2/Zfn0cn77lyvJCwe5+oanuPHRnTinjm0iIvJilu0BsXz5crdmzZp0F+PV6WmC6+dBbin0paYWrZ4P/+8JMKMrOsjf3bqeezce4qIFNXzlnYspzg2nt8wiIpIWZvaMc2758OWqgadDYTUsvto/vvircPFXoGkTHFgLQHFumO+9bxn/dOk87t/cyOXfeoxtjd1pLLCIiGQaBXi6XP4t+Mw2OOtaWHI1hPJg7U+PvGxmfPQNM7nl2hX0DMT5xC+eJZ5IprHAIiKSSRTg6RII+BtAbomfsez52yB27NjoZ0wv54tXLuKFxm5+9tTuNBRUREQykQI8Uyx7Pwx0wabbX/TShQsm8YZTKrn+/q209ugSMxERUYBnjmkroXwmPPvTF71kZvzLW+fTF0vwtfteSEPhREQk0yjAM4UZnPY+2P04fPdsf2nZvZ+FgR4AZlcX8aHXTeeWp/dy/f1baeuNpbnAIiKSTqF0F0CGWP5h6DoA3Yf8aG2rvg/b7oOrfgS1i/n0Baeyt72Pbz64jRse2cGb5k0iFDDiScdli2q5eFFturdARETGScZdB25mPwQuA5qccwtf7v1ZeR34K9XwGNz2Uehrgyu+A4vfCcD2pm5+8KedPLGjlVDQ6IslaO+NcevHzua0+rI0F1pEREbT8a4Dz8QAPwfoAX5y0gc4QG8L/OoDsHc1fPD3MO3sF72ls2+QS7/1KM7BXZ96PaX5kTQUVERExkLWDOTinHsE0EDghxVUwtU/h9J6+OX7oGPPi95SEo7z3XctoKk7ymdufU7Dr4qInAQyLsBfCTO71szWmNma5ubmdBdn7OWVwbtvgcQg3Pwe6G48+tqOP8LXF7D4sY/zj5fM44HNTbzz+0/ywKZGkkkFuYjIRJVxTegAZjYduFNN6MNsewBufhcEQnD6hyBSCI/+N+QWQ7QT98Hf87PGaXz/4R3s7+hndnUh154zkyuXTiESyspjNRGRk17WnAMHBfhLat0Bj10P624Gl4Al74YLvwTfWwll0+Cae4gnHXdtOMj3/7STzQe7mFScw8fOncX7VkwjHFSQi4hkEwX4RNPeAG07/fziZvD0jXDX38D7boPZbwbAOcej21r47sPbeWpnG3MmFfGFKxawYmZFessuIiKvWNZ0YjOzm4EngTlmts/MPpLuMmWksukw63wf3gCnfQBK6uGP/wGpgzIz45xTq7j5z1fwg/efTs9AnKtveIpP3fwsjV3R9JVdREROWMYFuHPu3c65Wudc2DlX55y7Kd1lygqhCJz7d35K0jU/POYlM+PCBTU88Nfn8qk3ncK9Gw9x/tce5nsP76CzfzBNBRYRkRORkU3or8ZJ24Q+kkQcfvFO3zP9sq/7kd1GsLu1l3/7/SYe3NJEJBTggvmTuGRhLQunFFNfno8drtWLiEjaHa8JXUOpTiTBEFx9M9z6QbjzryDWCys+fnTa0pRpZXnc9LbJbNsV41cNufx6fQt3rT8IQFFOiGXTynj97ErOObWKOTVF6dgSERF5GaqBT0TxGNz2Ydj8eyibAWf9hZ/pbOfDsOtRaNkKidS0pBYkWTWXpqkX8lD1B3n+QDdP7WxlR3MvAFedXse/vHU+Rbnh9G2PiMhJLKt6ob8aCvDjSCZg0+/gqe/DvtV+WSgX6ldAzSIf7Hll0LgR9q6Chkdh8bv8mOvBMAc6+vnZU7v5/p92MLU4xOcvmsHMusnUlOSSH1HDjYjIeFGAn8wOrINoJ0w9C8K5L37dOT8gzB//HU69CK78HuSXQyJOw0M3kfv4Vwkno1ww8FXaKOb8udV85arFVBbmjP+2iIicZBTg8vKevhHu+gzgoKAKgjnQtY9kzRJo3MjeqW/ll5P/gRsf20VJXpj/eddSVs6uTHepRUQmNAW4vDL71sDuJ/x58t5mWPpemPdWeOBf4fH/gWvuZXNkAZ+8+Vl2NPfwztPr+Mxb5lBd7Gv2iaTDgEBAPdlFREaDAlxOTKwXvn2mH3f9Lx6hL2F8/f6t/PiJBsLBABcvrGVXSw8bD3RRnBfm0kW1vHVJLfNqi3XOXETkBCjA5cRtvhN++V7f2W3ldTBpAXsbtrH6jh9Q3PocuXn5FBQWscHm8p8HTyMa9x8rygkxs7qQz186j+XTy9O7DSIiWUYBLifOObj3s/5ceXLQ92RvbwAclM/y99Eu6GshUbWAp+f8NZuT02nr6uH+7d1s7QzwF+fO4tNvPoWcUDDNGyMikh0U4DJ6elthw62w7T6oOwMW/xlUzPKvOecvX7vvn6Fzz5GPuECIeyZdy8d3vY5JxflctLCGCxfUcNaMcp0vFxF5CQpwGV+DUXj+Nn/uPBiG7Q/AljtpnfxGvhj+JHfviDEQT7JkciHfnfYwUzbdBKEcf216ca2/Vr1mCZz6FpKRYlp7Y1QV6bI1ETn5KMAlvZyD1f8L930OghEGF1zFE5GVFK3+OsvcJtbnn81AXhWRwS7KYwepHdhJyA1yMFzPBwb/kW3RYt5wSiV/d+FcFtWVpHtrRETGjQJcMkPjRnji27DxNxCP4iIF3D31M3x+10LMjJyQH7c9OhBl6eBzfCv8TWLhYn678DvcuK6PGQObed2sSt737g9Qkh859rvjA7DlLjjwrB86tmoOTD4Nwnlp2FARkdGhAJfM0t8O2+7359DLZxz/fQfWwc/eDoNRXDyKuQQADwZfT+17v8f8mfXQtBnW/hSeuxn628CCkHoflXPgQ3dBYdU4bJSIyOhTgEv2at4Kj3wVSuuh/mz2b36SSWuvp9GVY8W1TO7eAIEwzL0Uln2AgfrX09m4m4GdT1L36N9j5TPhQ3f64WFFRLKMAlwmlM7tT9H9q4/RH43x6+R5NM58G62umG2NPRzqih553znB57kp8lUGyuZQsOIaLNbtO9Yl437Cl2TC19aTCT98bNk0yK+E9l3Q/IIfuGbldZA75Lx7rA8i+WnYahE5GSnAZULa2dzDzav3cPeGQ5QXRDilupDplQVUFuZQXhDmmd3tHHj6Dq53XyPHBlOfMgiEcIEQBIJYIOiXRTuBIf8fIkUw2OuD/aL/BAvA6hth92Mw+81w/j9B1TxY9zN/Xr92CVz1QwjoGncRGT0KcDlpRQcT/Obx57nhoS20JXK4bNlMDnRGeXZPB539gxTlhCjJD7NyehHvn2csKB7AymdAUa3vEHfnp+Hgc/7LSqfBKW+B53/tz+PnlkK0AypP9ePHn/FRuORrYGN4bfsL9xztpCciE54CXE56TV1R/u3OTdy94SCzqgpZVl9GTUkunf2DNHcP8NALTfTFEsyqKuDihbW8ef4kFk8pIeASsPG3JCNFrAmfzuM72zl9UpCVzb8k2LYVln8Ypr8B7v88PPEtePMX4PWf9iuNdsL6X8GzP4XO/b42n18Og33Q1+ab8wsqobAaapfCiv8HxZP9Zxs3wa5H/GQyJVMgmYQH/tmvI68MrrkXquem7w8qIuNCAS6Skkg6giOM/tYzEOeu9Qe4be1+1jS0kXR+HPfplQXUl+ezbm8H+zv6j7y/KCfE4qkldEfjtPXGWFpXzFfsm+RvvR3yyiEY8bXzeNQPTDN5GfS1+pp7OA/yKyCcD30t0H0I9q/1ze+L3+WHqG141K8omAPLr/Hv2fQ7WPo+PzCOGXz4D75zX8s26NoPU8+ESIH/XNcB2LsaZpxzbAe+roPQug1yiv05/pwSfx8Mj+FfXUReKwW4yKvQ3hvj4a1NPLung10tvexp62NaRQFvP20K582pYu2edu7ZcIitjd2U5kcozA3x4OZGchjkf2c/yWllUcIkfJgueqe/Ht2MWDxJc88Ak0tyseHN7O0N8Nj/wLqfQ2ENnPERmP0mWH0DPPtz39nuLV+Esz8BTZvgRxdDpNCHfkdq2NpgDkw7Gwa6Yf8zflnhJLj8WzDrTfDUd+GhL0G8nxfJK/cHD2d8FCpnv/Y/3t6n/cHHGR85tvPfeBqMQjg3PesWGWUKcJExtretj3+7cxP3b2okJxTg3FOrOGN6OQc6+9nV0suull72tfeTSDrOn1vNt99z2shTrcb6/LCyQzvDte2E3hZfwz6ywtXwm2uhep7vVFdaDzsfhh0P+Rr+3EugZrGfy73xeSiZCp17Yc4lcOafw2C/n3wm2gkDXf56+s2/9xPVTFkOkxZA1VxIxKBzn289mLQApp7lWw/2r/EHCYU1/iCltB7+9GV47Ovgkv50wZv+xc8pHwiM/EdLJqFrH7Tu8GWrO/PETgv0tcFD/wFrfgRv+Gt44+fGtj+CZK6+ttS/w8rxXW+s17e+jWKLlgJcZJysaWjjzvUHuef5gzR2DZAfCTKjsoDplQXMrCzAOfjuw9tZOKWEmz54Bt3RQR7d1sL2ph56B+L0xuLUluSxdGopp9WXMq2iYMT1OOfYcqibOZOKXnpCmPgAPPyffjrY8/8J5l9x/FDrboS1/+cPApo3++Z+8DXp3JKjNf3Dckv8QQDO1+D72+C098GS98CDX4C9q3wrQeEkKKrx5/oLayAY8h0D9z8Lse5jv3PWm/wBRuWp/v0W9KcHug5ASZ3vwGfmD2ieuwUOrvOnIoIRP8nOQDdMXuoPLl73Kbjg3/z7nYOO3X5woKZN/qqCnGIIhPzy9gb/2VCuP4A6fB8p8OUomuzXXbvk2Nr9QI9/z+G/qXP+YCecd/R0xisVj/mytO6Ath1HD2xmX+BbRo53IDRW2hvg99dBTxO87QdQu3jk98VjsOFXsOoH/m91zt9C/YoTWO9u2HKnP5gN5/p91N/u/66hXH9aaNrrIKfoxZ9t3AhPfsf3PUkO+n971fOhdKrvmFpQ5fd7pMB/bzLu3xftOrqOrv3+oDUQhrrT/emvWA80b/F9WUrrfSfSgmq/fKDbHyTveQoObYAP3gHTX//at38YBbjIOEsmHe19McoLIi9qLn9gUyOfvPlZEklHLJEEoDg3RHFemLxwkH3t/fQP+tHkLl5Ywz9cPPeYIN/e1MM//W4DT+1s4x3L6vjKVYtHPK9/Qg4HUSjn6A9lXxvse9r/0E1Z7meh6z7oJ65peByWvd8PqHP48xt/61sKehqP3robfRP+pIUw5XTfP6Bilg/2zbf7MfN7Go9frsIaqDwF9jzpf3xLpvpWglivH9nvwi/5loO7PwNrboJTL/brO/jc0QMSjGMuGQzl+isM8kr9AU88mroN+IAeepARjPgOhy4Jrdt9P4dgxIdDOA869vrLDzFfzppF/rW8Un/VQl6ZfxxNtXo0b/ah0H3Qb7dLHl1XTonvv9C+C6a9Hq74tn/e0+RbTQIhf4ATCPlbMOxrnMMPHHpb/fDFW+/1+7J8ZqqzpPn1BUL+M+E8fzAUKYCD633rjQX8uAf97XDBv/v+GBb0f9M9q/zpkg23+tCrXgC9TdDbDPVn+3X1NPp/CzPO8S1FOUU+7Jq2+G0Y7PcBGi7w623cCHufGmHHm//bxXohMeDLXFDtyxbK9UEa7Uz1Mcn3B5Jl031n0KZNvnw9Tcfu95EEI1A8xR8sxnp9ICdTl6AGc/xkS537jy478m8oD+qW+xZK5UIVAAAShElEQVSqpe85OkPjKFCAi2SY9fs6+MWqPSyYXMw5p1YdE9DxRJJtTT3c+/whbnhkJ/FkkrfMr6EgJ0gsnuSuDQfJCwc5b041dzx3gLefNoWvvnMJSedYv6+TkrwQs6tHqJ1kimTy+LXJ+ADsfsJ32utt8iFdPMXX4Nt2QcNj/gd51vn+R7p63sjf45y/MuDpm6Bitq+V1y7199ULfAAMdPnvz6986dptrNeXp2mzD5e9T/uwrDzFH0BEO3znwME+XzsrmerD5NB6H1a9ralQH8YCPkxLp/mQL07V8itmQfmso50Pn/0Z/OEffXlfiUiR/2wox29ny1a/nRWn+L4U7buPDjf8UmaeB5d/2wfi7R+Hrfe8+D2BMExfCWd/0vfZGOyHZ34Ea3/i1184yR8M7XnKH2gdFi7wgRzO9d8x2Of/zoWTYNE7YOFV/u8x2O/LnlPsW24Go75lp+FRv09ivf77I4W+Rah8Bix598gjLybi/qA01uNvycTRg5+cIl+eoa0p4NfXtMl/d9l0f2orEfetE32tkFPoP1M8Zcw6girARbJUU1eU6+/fymPbW0gkHUnnWDm7ks9ePI+qohy+9eA2/vv+rcytKWJfez89A3EA5tcWc+Vpk7l8yRRqSo7t0DUQT9DSE6OtJ8b0ynyKcidwD3TnMuM8eHwA+jt82B++EqFyzivvbNe5zwd5pMCHXE5xahTB+NGRBeMDvvbb0+hbS5KDvnm7YpbvoFiz0H9XYtCHD+b/Nsm4b64eHHILRmDaymNPDWz8rW8NSCb8wceU04+98uGlDPTA7sf9uict8Act431KIEspwEUmsBsf3clta/dzWn0pK2dV0tQd5XfrDvDc3g7MYMWMCs6aWc7Wxm7W7engQOfR4WYjwQBnz6rgzfOqWTq1jDk1RURCATr7Btnb3kd9RT7FEzngRTKcAlzkJLSrpZffPbuf29ftp6G1j6nleSypK2XOpCKqi3MoyfPDzd63qZHdrX0AhINGfiREZ78/x5cbDnDxwlrevmwKi6aUUDp8GlcRGVMKcJGTmHOO3liCwpwRLltLvb6nrY8N+zvZsL+TnmicaRX5TC7N46mdrdy+7gDdUd80X5YfZnplATMqfM/66GCChtZeDnVGOX1aGZcunsziKSW09A6wt62P6qJcppZr8heR10oBLiKvWXQwwePbW9jZ3MvOll4aWnppaO3lYGeUUMCYWp5PRUGE5/Z1MJhwhIPGYML/tpjBOadU8e4z65lVVUA46M97tvYO0NIT42BHP3va+jnY2c8Z08t595n15EU0IYzIYQpwERl10cEEoYARSoVyZ98g9206xAuHuqkry6O+Ip/1+zq5ZfXeY6Z5HS4/EqS8IMK+9n4qCiJ88HXTWT6tjNmTCqkqzHnxqHUiJxEFuIikTTyRZNWuNtp6YwwmkjgH5YURqgpzqCnJpSJ1rfzqXW18+6HtPLK1+chn88JBKosiVBbmEAkGSDpH0vkx7Z1z5ISDzKoqZHZ1IQWRIL2xBNHBBJFggNxIkOqiHM45pUq1eslaCnARyRpNXVG2Nvawvambve39tPb45vbBRJKAGcGAYQbBgNETjbO9uYeOvsHjfl9BJMiFC2tYPq2cSChAKGAMxBP0DiRIOkd9eT6zqgupL88/0sQvkimOF+Aj92gREUmj6uJcqotzef0pr2wca+ccrb0xYvEkBZEQuZEAgwlHXyzO9qYebn/2AHdvOMhv1u5/ye+JhALMry1m4ZRiEkk41Omvq182rYxzT61ifm0x3dE43dE4RbkhJhXnEgkF6BmIs6u5l75YnMV1party7hQDVxETgoD8QTtvYMMJpIMJpLkhIMUpIK2obWPHU09bDnUxfp9nWw62EVOKEBNSS45oSDrU53zhjODkrzwMbX/cND8pXo1RVQX5VJdnEN9eT4zKguoKc596XHrh4gOJvjFqj38bNVultaV8i9vXUBJvq7HPxmpCV1E5DXqGYjzxPYW9rT1UZwXpignRHc0zv6Oflp6BphcmsesqkLCQWN1Qxurdraxu7WX9mHN+qGAUZATIj8SJBQ0kknfepB0kHCOoBmVRRGqi3LZsL+T5u4BFk0pYdPBLqoKc/jilQuZWp5PXyxOMGDUleVTlh/GzEimxtXPDY9u7b87OshgwlFeoOv/00UBLiIyzg7P/767xV9+t7+jn76BOL2xBMmkw8wIGATMCAQgnnC09AzQ2DVAVVEOf3neLM6aWcH6fR381S/XsaP5xeOp50eCBM3oicVxDpZOLeXSRbW8/pRK8sL+QCEUCBAMGKGAEUsk6Y8l6B/0t2gsQXFemDk1Rcec/9/Z3MOPn2jg18/sIxZP8o5ldfzlG2cdd3Y8GTsKcBGRLBYdTHD/pkYCZuRHggwmkuxr72dfez8OR1FOCAf8cUsTGw+8wklPhoiEAsytKSIWT3Kgo5+uaJxIMMBbl0wmPxLkl2v2kkg6Lpg3iXcur+PcU6uOXD441NbGbn7yZAMPbGridbMquGblDBbVlZz4H+AkpgAXETlJNLT0sn5/J4lkksGEI5F0xBNJ4klHTihIXiRAXjhIburW0jPgz/0f6CI3HGRyaS7TKgq4fMlkqopyAGjsinLTY7v49TP7aOuNUZIXprIwQm44SCjgB+6JDibY2dJLJBRg5awKVu9qozeWYH5tse8DUJJLRWGEwpwQBZEQwYDhcIQCAaaU5VFfnk9uOEh7b4z2vhjBgFEQCREOBWjvjdHcMwAOplXkM/UkumJAAS4iIicsFk/y8AtNPLi5iZ6BONHBBPGkIxwMEA4ai+tKedcZUykviNAVHeTWNfv445ZGDnZEOdDZT3Qw+fIreQWCAaOqMIfKoggVBTlUph5HggEOdEQ50NFPbyzOYMKPFzCtIp+5Nf5AIjccIBQIUJgborIwh4qCCDnhAAEznIPeWJyeaJyAGeWFEQoiwbQOJqQAFxGRtHLOMRBP0jMQp3cgTtKBAdF4gv3t/ext62MgnqS8IEJpfoRE0l8KGIsnKSvwg/mAY1dLHw0tvTR2RWlJjRHQ0jNAa0+MeDLJpOJcaktyKckLEwwEcM6xq6WXXa29vJbIiwR92PtWiwD5Ef+4OC9EdXEuk4pyyY8EsVR/hgvmTxrV8f91HbiIiKSVmR1ptvdhfNTcmuJX/D2nTysfcblz/nTBSOfmAfpjCfZ39BGLOwYTSbqj8SNj8scTSRKpdC/MCVGYEyKRdLT3xWjtjdE3kOr4l+oA2BeLs78jyrN7OmjtjR2znplVBeMygU/GBbiZXQR8AwgCNzrnvpzmIomISBYwM0LB4zd150WCzK4uGvX1xuJJBuIJkg5wjNtAPhkV4GYWBL4DXADsA542szucc5vSWzIREZGRRUIBIqHx71CXaV34zgS2O+d2OudiwC3AFWkuk4iISMbJtACfAuwd8nxfatkxzOxaM1tjZmuam5uHvywiIjLhZVqAvyLOuRucc8udc8urqqrSXRwREZFxl2kBvh+YOuR5XWqZiIiIDJFpAf40cIqZzTCzCHA1cEeayyQiIpJxMqoXunMubmafAP6Av4zsh865jWkuloiISMbJqAAHcM7dDdyd7nKIiIhkskxrQhcREZFXQAEuIiKShbJ+MhMzawZ2j+JXVgIto/h9mULblV20XdllIm7XRNwmyM7tmuace9E101kf4KPNzNaMNOtLttN2ZRdtV3aZiNs1EbcJJtZ2qQldREQkCynARUREspAC/MVuSHcBxoi2K7tou7LLRNyuibhNMIG2S+fARUREspBq4CIiIllIAS4iIpKFFOBDmNlFZvaCmW03s39Id3leCzObamYPmdkmM9toZtellpeb2f1mti11X5busr4WZhY0s2fN7M7U8xlmtiq1z36ZmgQnq5hZqZn92sy2mNlmMzt7IuwvM/ur1L/B583sZjPLzcb9ZWY/NLMmM3t+yLIR949530xt33ozW5a+kr+042zXV1P/Dteb2W/NrHTIa59NbdcLZnZhekr98kbariGv/Y2ZOTOrTD3Pmv01EgV4ipkFge8AFwPzgXeb2fz0luo1iQN/45ybD6wAPp7ajn8AHnTOnQI8mHqeja4DNg95/l/A151zs4F24CNpKdWJ+QZwr3NuLrAEv31Zvb/MbArwKWC5c24hfnKiq8nO/fVj4KJhy463fy4GTkndrgW+N05lfC1+zIu3635goXNuMbAV+CxA6jfkamBB6jPfTf1mZqIf8+LtwsymAm8B9gxZnE3760UU4EedCWx3zu10zsWAW4Ar0lymV805d9A5tzb1uBsfBlPw2/J/qbf9H3Blekr42plZHXApcGPquQHnA79OvSXrtsvMSoBzgJsAnHMx51wHE2B/4SdLyjOzEJAPHCQL95dz7hGgbdji4+2fK4CfOO8poNTMasenpK/OSNvlnLvPORdPPX0KqEs9vgK4xTk34JzbBWzH/2ZmnOPsL4CvA38HDO25nTX7ayQK8KOmAHuHPN+XWpa1zGw6cBqwCpjknDuYeukQMClNxToR/4P/D5hMPa8AOob84GTjPpsBNAM/Sp0auNHMCsjy/eWc2w98DV/bOQh0As+Q/fvrsOPtn4n0O/Jh4J7U46zeLjO7AtjvnHtu2EtZvV0K8AnKzAqB24BPO+e6hr7m/LWDWXX9oJldBjQ5555Jd1lGWQhYBnzPOXca0Muw5vIs3V9l+NrNDGAyUMAIzZoTQTbun5djZp/Dn477ebrLcqLMLB/4R+Cf012W0aYAP2o/MHXI87rUsqxjZmF8eP/cOfeb1OLGw01DqfumdJXvNVoJXG5mDfjTG+fjzx2XpppoITv32T5gn3NuVer5r/GBnu37683ALudcs3NuEPgNfh9m+/467Hj7J+t/R8zsQ8BlwHvd0YFCsnm7ZuEPJJ9L/X7UAWvNrIbs3i4F+BBPA6ekeslG8B027khzmV611Hnhm4DNzrnrh7x0B/DB1OMPArePd9lOhHPus865OufcdPy++aNz7r3AQ8BVqbdl43YdAvaa2ZzUojcBm8jy/YVvOl9hZvmpf5OHtyur99cQx9s/dwAfSPVuXgF0Dmlqz3hmdhH+NNXlzrm+IS/dAVxtZjlmNgPf6Wt1Osr4ajnnNjjnqp1z01O/H/uAZan/e1m9v3DO6Za6AZfge17uAD6X7vK8xm14Pb45bz2wLnW7BH+++EFgG/AAUJ7usp7ANp4H3Jl6PBP/Q7IduBXISXf5XsP2LAXWpPbZ74CyibC/gC8AW4DngZ8COdm4v4Cb8efxB/E//h853v4BDH81yw5gA74Xftq34VVs13b8OeHDvx3fH/L+z6W26wXg4nSX/9Vs17DXG4DKbNtfI900lKqIiEgWUhO6iIhIFlKAi4iIZCEFuIiISBZSgIuIiGQhBbiIiEgWUoCLyJgzs/NSs0DVvfy7ReSVUICLiIhkIQW4iIhIFlKAi5wEzOyTZrbFzKJmts3MPnd4THIzazCz/0jNhNZlZi1m9iUzCwz5fJGZ/cDMms1swMzWmNlbhq2j2sx+ZGaNqfW8YGYfHlaUeWb2iJn1mdkmM7t4HDZfZEIKvfxbRCSbmdm/AtcAn8YPjzkP+D6QC3w+9bZP4qdrPQM/z/P3gUb8hDEAP0y99j78OOcfA+40s8XOuS1mlgf8CegH3gvsBGYD5cOK8zXg7/FDV/4j8Eszm+acax/drRaZ+DSUqsgElppKsQV4u3Pu3iHLPwB80zlXmpqhaa9z7g1DXv8S8H7n3FQzm40f8/tS59zdQ96zFljnnPuwmX0EP6b0bOfcvhHKcR5+IpN3uNQMeWY2CT+X9kXOuT+M9raLTHSqgYtMbAuAPOA2Mxt6tB4Ecs2sKvX8yWGfexz4rJkVA/NTyx4Z9p5HgLNTj08HNo0U3sOsO/zAOddoZglg0ivaEhE5hgJcZGI7fB77nfiZ9oZrG8eyAMRGWKa+OCKvgf7jiExsG4EoMNM5t32EWyL1vhXDPvc6YL9zriv1HQDnDHvPOfipQgGeAebrOm+R8aMAF5nAnHM9wJeAL5nZx81sjpktMLOrzey/hrx1qZn9q5mdambvAa4D/jv1HTvwc3d/18wuNLO5ZvYNYCHw1dTnbwZ2A3eY2ZvNbIaZvcnM3jVe2ypyslETusgE55z7dzM7CHwCH8r9+Ob0Hw9527eAacAaYBD4Nkd7oAN8FB/WPwOKgQ3AZc65Lal19JnZucBXgFuAQqAB+PJYbZfIyU690EVOcqle6Dc6576Y7rKIyCunJnQREZEspAAXERHJQmpCFxERyUKqgYuIiGQhBbiIiEgWUoCLiIhkIQW4iIhIFlKAi4iIZKH/D10cpq/jblpjAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 504x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jYiAVVb2BnZ3"
      },
      "source": [
        "## Testing"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gAogToc6-7Pt",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "67ba6fac-ebca-47ae-a561-a3f4d1689204"
      },
      "source": [
        "# use the model to predict on testing data and get predicted labels\n",
        "pred_prob = model.predict(test_gen)\n",
        "pred_labels = np.argmax(pred_prob, axis=-1) \n",
        "pred_labels = [str(x+1) for x in list(pred_labels)]\n",
        "\n",
        "# get the true labels of the testing data\n",
        "test_labels = np.argmax(test_graph_labels.values, axis=-1)\n",
        "test_labels = [str(x+1) for x in list(test_labels)]\n",
        "\n",
        "# show the result as classification report\n",
        "print(classification_report(test_labels, pred_labels, target_names=apps, digits=4))"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
      "                          precision    recall  f1-score   support\n",
      "\n",
      "                  diijam     0.9627    0.9748    0.9687       159\n",
      "                  baomoi     0.9253    0.9200    0.9226       175\n",
      "                 fptplay     0.9615    0.9653    0.9634       259\n",
      "                   iQIYI     0.9153    0.9643    0.9391       168\n",
      "                    bigo     0.9454    0.9454    0.9454       183\n",
      "                 myradio     0.9638    0.9433    0.9534       141\n",
      "                 spotify     0.9328    0.9058    0.9191       138\n",
      "              nhaccuatui     0.9126    0.9176    0.9151       182\n",
      "              soundcloud     0.8967    0.8919    0.8943       185\n",
      "              sachnoiapp     0.8981    0.9151    0.9065       106\n",
      "                 phim247     0.9653    0.8910    0.9267       156\n",
      "                 popskid     0.9474    0.9524    0.9499       189\n",
      "  truyenaudiosachnoiviet     0.9649    0.9565    0.9607       115\n",
      "                   vieon     0.8915    0.8712    0.8812       132\n",
      "                  voizfm     0.9713    0.9902    0.9807       205\n",
      "                  tunefm     0.8837    0.9620    0.9212       158\n",
      "                    wetv     0.9707    0.9803    0.9755       203\n",
      "                 zingmp3     0.9560    0.9102    0.9325       167\n",
      "             truyenaudio     0.9205    0.9205    0.9205       151\n",
      "               baohay24h     0.9787    0.9840    0.9813       187\n",
      "                freefire     0.9389    0.9602    0.9494       176\n",
      "                among_us     0.9452    0.8214    0.8790       168\n",
      "                    azar     0.9355    0.8192    0.8735       177\n",
      "                  comico     0.7554    0.8528    0.8012       163\n",
      "                  nimotv     0.9670    0.9778    0.9724       180\n",
      "               mangatoon     0.9836    0.9783    0.9809       184\n",
      "             medoctruyen     0.9873    0.9748    0.9810       159\n",
      "                nhacvang     0.9944    0.9890    0.9917       181\n",
      "               noveltoon     0.9326    0.9022    0.9171       184\n",
      "                 radiofm     0.9777    0.9511    0.9642       184\n",
      "                 tivi247     0.9494    0.9375    0.9434       160\n",
      "                   vtvgo     0.8258    0.9343    0.8767       137\n",
      "                 tivi24h     0.8994    0.9728    0.9346       147\n",
      "                  tinder     0.9086    0.9408    0.9244       169\n",
      "               tinmoi24h     0.9714    0.9605    0.9659       177\n",
      "                 tivi360     0.9800    0.9735    0.9767       151\n",
      "                  tiktok     0.9661    0.9500    0.9580       120\n",
      "                linkedin     0.9933    0.9933    0.9933       149\n",
      "                    tiki     0.9645    0.9967    0.9803       300\n",
      "                  tinhte     0.7882    0.7701    0.7791        87\n",
      "                   lotus     0.9821    0.9940    0.9880       166\n",
      "tivi_truyentranh_webtoon     0.9888    0.9944    0.9915       177\n",
      "          tuoitre_online     0.9200    0.9471    0.9333       170\n",
      "            vietnamworks     0.9874    0.9632    0.9752       163\n",
      "      wallstreet_journal     0.9716    0.9500    0.9607       180\n",
      "                cnn_news     0.9268    0.9421    0.9344       121\n",
      "                bbc_news     0.9600    0.9351    0.9474       154\n",
      "                 twitter     0.9071    0.9338    0.9203       136\n",
      "                  weeboo     0.8333    0.8387    0.8360       155\n",
      "                  twitch     0.9779    0.9852    0.9815       135\n",
      "               vnexpress     0.9828    0.9884    0.9856       173\n",
      "                   topcv     0.9053    0.8947    0.9000       171\n",
      "               toc_chien     0.9468    0.9468    0.9468        94\n",
      "                  wesing     0.9492    0.8550    0.8996       131\n",
      "                    hago     0.8897    0.8768    0.8832       138\n",
      "             google_meet     0.9697    0.9697    0.9697       198\n",
      "                dubsmash     0.9045    0.9415    0.9226       171\n",
      "                facebook     0.9366    0.9172    0.9268       145\n",
      "                hahalolo     0.9829    0.9829    0.9829       175\n",
      "                    zalo     0.9830    0.9719    0.9774       178\n",
      "                hello_yo     0.8654    0.7542    0.8060       179\n",
      "                 dan_tri     0.9848    0.9606    0.9726       203\n",
      "                    zoom     0.9355    0.9575    0.9464       212\n",
      "               wikipedia     0.8750    0.8526    0.8636       156\n",
      "               instagram     0.9511    0.9943    0.9722       176\n",
      "                  jobway     0.9636    0.9464    0.9550       168\n",
      "                    kaka     0.7633    0.8716    0.8139       148\n",
      "               pinterest     0.6866    0.9030    0.7801       165\n",
      "                   quora     0.9766    0.9435    0.9598       177\n",
      "                  lazada     0.7707    0.8681    0.8165       182\n",
      "                   chess     0.9697    0.9648    0.9673       199\n",
      "                    cake     0.9767    0.9438    0.9600       178\n",
      "           mobile_legend     0.9880    0.9704    0.9791       169\n",
      "         co_tuong_online     0.9718    0.9829    0.9773       175\n",
      "                     ted     0.8561    0.6685    0.7508       178\n",
      "                telegram     0.9477    0.9106    0.9288       179\n",
      "              starmarker     0.9432    0.9222    0.9326        90\n",
      "                   skype     0.9653    0.9653    0.9653       144\n",
      "                    soha     0.9693    0.9634    0.9664       164\n",
      "                   tango     0.9839    0.9839    0.9839       248\n",
      "               thanhnien     0.9581    0.9091    0.9329       176\n",
      "                snapchat     0.9780    0.9674    0.9727       184\n",
      "                tien_len     0.7556    0.8160    0.7846       125\n",
      "       animal_restaurant     0.9531    0.9242    0.9385       132\n",
      "                    bida     0.8333    0.9322    0.8800        59\n",
      "                 cho_tot     0.7917    0.8444    0.8172        90\n",
      "               messenger     0.9448    0.9059    0.9249       170\n",
      "                 netflix     0.9822    0.9940    0.9881       167\n",
      "                nonolive     0.9118    0.9568    0.9337       162\n",
      "                     may     0.9912    0.9825    0.9868       114\n",
      "          podcast_player     0.9701    0.9586    0.9643       169\n",
      "                    pubg     0.9570    0.9570    0.9570       186\n",
      "                partying     0.8851    0.9506    0.9167       162\n",
      "                  kenh14     0.9697    0.9816    0.9756       163\n",
      "         lienquan_mobile     0.9655    0.9655    0.9655       145\n",
      "              likee_lite     0.9861    0.8765    0.9281       162\n",
      "                  reddit     0.9492    0.9600    0.9545       175\n",
      "                   sendo     0.9355    0.9508    0.9431       183\n",
      "                  shopee     0.9943    0.9943    0.9943       175\n",
      "            the_guardian     0.8831    0.8831    0.8831       154\n",
      "               ola_party     0.9103    0.8114    0.8580       175\n",
      "\n",
      "                accuracy                         0.9346     16561\n",
      "               macro avg     0.9325    0.9316    0.9313     16561\n",
      "            weighted avg     0.9364    0.9346    0.9347     16561\n",
      "\n"
     ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CLjP43YH-uOY"
      },
      "source": [
        "-------------------------------------------------------------"
      ]
    }
  ]
}