demo10_a.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "lec10a.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8HOm2nh3tRAk",
        "colab_type": "text"
      },
      "source": [
        "Language prediction from character-wise surnames.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EOeIrijwtK6E",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 275
        },
        "outputId": "b5a45afa-c977-4954-d056-344bc6b207c2"
      },
      "source": [
        "%matplotlib inline\n",
        "\n",
        "!wget https://download.pytorch.org/tutorial/data.zip\n",
        "!unzip data.zip\n",
        "\n",
        "from __future__ import unicode_literals, print_function, division\n",
        "from io import open\n",
        "import glob\n",
        "import os\n",
        "\n",
        "def findFiles(path): return glob.glob(path)\n",
        "\n",
        "print(findFiles('data/names/*.txt'))\n",
        "\n",
        "import unicodedata\n",
        "import string\n",
        "\n",
        "all_letters = string.ascii_letters + \" .,;'\"\n",
        "n_letters = len(all_letters)\n",
        "\n",
        "# Turn a Unicode string to plain ASCII, thanks to http://stackoverflow.com/a/518232/2809427\n",
        "def unicodeToAscii(s):\n",
        "    return ''.join(\n",
        "        c for c in unicodedata.normalize('NFD', s)\n",
        "        if unicodedata.category(c) != 'Mn'\n",
        "        and c in all_letters\n",
        "    )\n"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-04-05 19:54:50--  https://download.pytorch.org/tutorial/data.zip\n",
            "Resolving download.pytorch.org (download.pytorch.org)... 13.226.45.116, 13.226.45.35, 13.226.45.86, ...\n",
            "Connecting to download.pytorch.org (download.pytorch.org)|13.226.45.116|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 2882130 (2.7M) [application/zip]\n",
            "Saving to: ‘data.zip.1’\n",
            "\n",
            "\rdata.zip.1            0%[                    ]       0  --.-KB/s               \rdata.zip.1          100%[===================>]   2.75M  15.6MB/s    in 0.2s    \n",
            "\n",
            "2020-04-05 19:54:50 (15.6 MB/s) - ‘data.zip.1’ saved [2882130/2882130]\n",
            "\n",
            "Archive:  data.zip\n",
            "replace data/eng-fra.txt? [y]es, [n]o, [A]ll, [N]one, [r]ename: N\n",
            "['data/names/French.txt', 'data/names/Chinese.txt', 'data/names/Czech.txt', 'data/names/Japanese.txt', 'data/names/German.txt', 'data/names/Portuguese.txt', 'data/names/Irish.txt', 'data/names/English.txt', 'data/names/Korean.txt', 'data/names/Russian.txt', 'data/names/Italian.txt', 'data/names/Spanish.txt', 'data/names/Polish.txt', 'data/names/Dutch.txt', 'data/names/Vietnamese.txt', 'data/names/Scottish.txt', 'data/names/Arabic.txt', 'data/names/Greek.txt']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CuO7F_LAuKAc",
        "colab_type": "text"
      },
      "source": [
        "Build a list of names per language."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4yTtu9O4uO-G",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Build the category_lines dictionary, a list of names per language\n",
        "category_lines = {}\n",
        "all_categories = []\n",
        "\n",
        "# Read a file and split into lines\n",
        "def readLines(filename):\n",
        "    lines = open(filename, encoding='utf-8').read().strip().split('\\n')\n",
        "    return [unicodeToAscii(line) for line in lines]\n",
        "\n",
        "for filename in findFiles('data/names/*.txt'):\n",
        "    category = os.path.splitext(os.path.basename(filename))[0]\n",
        "    all_categories.append(category)\n",
        "    lines = readLines(filename)\n",
        "    category_lines[category] = lines\n",
        "\n",
        "n_categories = len(all_categories)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_glHLOisumww",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "3b891887-2396-4772-caeb-012e01f4f9aa"
      },
      "source": [
        "print(category_lines['Chinese'][:5])"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['Ang', 'AuYong', 'Bai', 'Ban', 'Bao']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Y7eSBLonu5uH",
        "colab_type": "text"
      },
      "source": [
        "Convert names to Pytorch tensors. We will do it character-by-character, so each letter is one-hot encoded."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vJKvtCyOvF7U",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "b3d4e9f6-7ce6-4486-92bc-1b8c21e701b0"
      },
      "source": [
        "import torch\n",
        "\n",
        "def letterToIndex(letter):\n",
        "    return all_letters.find(letter)\n",
        "\n",
        "def letterToTensor(letter):\n",
        "    tensor = torch.zeros(1, n_letters)\n",
        "    tensor[0][letterToIndex(letter)] = 1\n",
        "    return tensor\n",
        "\n",
        "# Turn a line into an array of one-hot letter vectors\n",
        "def lineToTensor(line):\n",
        "    tensor = torch.zeros(len(line), 1, n_letters)\n",
        "    for li, letter in enumerate(line):\n",
        "        tensor[li][0][letterToIndex(letter)] = 1\n",
        "    return tensor\n",
        "\n",
        "print(letterToTensor('B'))\n",
        "\n",
        "print(lineToTensor('Bai').size())"
      ],
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "tensor([[0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "         0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "         0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "         0., 0., 0.]])\n",
            "torch.Size([3, 1, 57])\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NeXvc9POv_Ya",
        "colab_type": "text"
      },
      "source": [
        "Quick helper function to get names/languages."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "X-2r09njv9Ep",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 187
        },
        "outputId": "d2ec9a03-520c-4bbb-d175-e787bac16f19"
      },
      "source": [
        "import random\n",
        "\n",
        "def randomChoice(l):\n",
        "    return l[random.randint(0, len(l) - 1)]\n",
        "\n",
        "def randomTrainingExample():\n",
        "    category = randomChoice(all_categories)\n",
        "    line = randomChoice(category_lines[category])\n",
        "    category_tensor = torch.tensor([all_categories.index(category)], dtype=torch.long)\n",
        "    line_tensor = lineToTensor(line)\n",
        "    return category, line, category_tensor, line_tensor\n",
        "\n",
        "for i in range(10):\n",
        "    category, line, category_tensor, line_tensor = randomTrainingExample()\n",
        "    print('category =', category, '/ line =', line)"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "category = Italian / line = Tedesco\n",
            "category = Greek / line = Bilias\n",
            "category = Arabic / line = Zogby\n",
            "category = Japanese / line = Shintaro\n",
            "category = English / line = Lenihan\n",
            "category = French / line = Brisbois\n",
            "category = Polish / line = Gorski\n",
            "category = Italian / line = Ambrogi\n",
            "category = Czech / line = Pesek\n",
            "category = Korean / line = Byon\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zrTwovIywWVS",
        "colab_type": "text"
      },
      "source": [
        "Another helper function to convert output logits to language."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KgFTiKJMwZ0I",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def categoryFromOutput(output):\n",
        "    top_n, top_i = output.topk(1)\n",
        "    category_i = top_i[0].item()\n",
        "    return all_categories[category_i], category_i"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "10K096_AwcXb",
        "colab_type": "text"
      },
      "source": [
        "OK, now let's define our network and train the model. We will use a recurrent neural network (RNN) with a hidden state vector of size 128. It will pass each character sequentially (remember each character is one-hot encoded) and then finally perform a softmax to get the 18 classes."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jmxbzHTtwoxR",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import torch.nn as nn\n",
        "\n",
        "class RNN(nn.Module):\n",
        "    def __init__(self, input_size, hidden_size, output_size):\n",
        "        super(RNN, self).__init__()\n",
        "\n",
        "        self.hidden_size = hidden_size\n",
        "\n",
        "        self.i2h = nn.Linear(input_size + hidden_size, hidden_size)\n",
        "        self.i2o = nn.Linear(input_size + hidden_size, output_size)\n",
        "        self.softmax = nn.LogSoftmax(dim=1)\n",
        "\n",
        "    def forward(self, input, hidden):\n",
        "        combined = torch.cat((input, hidden), 1)\n",
        "        hidden = self.i2h(combined)\n",
        "        output = self.i2o(combined)\n",
        "        output = self.softmax(output)\n",
        "        return output, hidden\n",
        "\n",
        "    def initHidden(self):\n",
        "        return torch.zeros(1, self.hidden_size)\n",
        "\n",
        "n_hidden = 128\n",
        "rnn = RNN(n_letters, n_hidden, n_categories)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kkhIVj9Sx-uz",
        "colab_type": "text"
      },
      "source": [
        "OK, great. Let's train this network. Each loop of training will read each letter in, keep the hidden state for the next letter, and compare the final output with the target label using Cross Entropy Loss. This will be used to backpropagate through the weights."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4xdvVfHoyBO8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "lr = 0.005\n",
        "criterion = nn.NLLLoss()\n",
        "\n",
        "def train(category_tensor,line_tensor):\n",
        "  hidden=rnn.initHidden()\n",
        "  rnn.zero_grad()\n",
        "  for i in range(line_tensor.size()[0]):\n",
        "    output, hidden = rnn(line_tensor[i],hidden)\n",
        "  \n",
        "  loss = criterion(output,category_tensor)\n",
        "  loss.backward()\n",
        "\n",
        "  for p in rnn.parameters():\n",
        "    p.data.add_(-lr,p.grad.data)\n",
        "  \n",
        "  return output, loss.item()\n",
        "  \n",
        "def test(line_tensor):\n",
        "  hidden = rnn.initHidden()\n",
        "  for i in range(line_tensor.size()[0]):\n",
        "    output, hidden = rnn(line_tensor[i], hidden)\n",
        "\n",
        "  return output"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ebQOxHu23-nH",
        "colab_type": "text"
      },
      "source": [
        "OK! Now we can run a script to monitor the training process. Observe how the loss decreases over time. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ja1vaJJS4E4d",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 357
        },
        "outputId": "a7596952-45f7-4135-dc92-b24a8b7b7a22"
      },
      "source": [
        "n_iters = 100000\n",
        "print_every = 5000\n",
        "plot_every = 1000\n",
        "\n",
        "# Keep track of losses for plotting\n",
        "current_loss = 0\n",
        "all_losses = []\n",
        "\n",
        "\n",
        "for iter in range(1, n_iters + 1):\n",
        "    category, line, category_tensor, line_tensor = randomTrainingExample()\n",
        "    output, loss = train(category_tensor, line_tensor)\n",
        "    current_loss += loss\n",
        "\n",
        "    # Print iter number, loss, name and guess\n",
        "    if iter % print_every == 0:\n",
        "        guess, guess_i = categoryFromOutput(output)\n",
        "        correct = 'Y' if guess == category else 'N (%s)' % category\n",
        "        print('%d %d%% %.4f %s / %s %s' % (iter, iter / n_iters * 100, loss, line, guess, correct))\n",
        "\n",
        "    # Add current loss avg to list of losses\n",
        "    if iter % plot_every == 0:\n",
        "        all_losses.append(current_loss / plot_every)\n",
        "        current_loss = 0"
      ],
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "5000 5% 2.8812 Albuquerque / German N (Portuguese)\n",
            "10000 10% 2.3976 Odriscoll / Italian N (English)\n",
            "15000 15% 1.8143 Salib / Arabic Y\n",
            "20000 20% 3.0821 Ui / Korean N (Japanese)\n",
            "25000 25% 3.8161 Raske / Polish N (Dutch)\n",
            "30000 30% 1.1736 Yeo / Chinese N (Korean)\n",
            "35000 35% 2.7148 Knopf / Arabic N (German)\n",
            "40000 40% 2.6530 Kubin / Russian N (Czech)\n",
            "45000 45% 2.1725 Millar / Scottish N (English)\n",
            "50000 50% 0.4774 Soares / Portuguese Y\n",
            "55000 55% 1.9418 Blanco / Italian N (Spanish)\n",
            "60000 60% 0.6971 O'Neal / Irish Y\n",
            "65000 65% 1.6441 Arreola / Portuguese N (Spanish)\n",
            "70000 70% 2.7261 Seger / German N (Czech)\n",
            "75000 75% 1.1403 Phi / Vietnamese Y\n",
            "80000 80% 4.0694 Salazar / Polish N (Spanish)\n",
            "85000 85% 0.3784 Zhu / Chinese Y\n",
            "90000 90% 0.4804 Hew / Chinese Y\n",
            "95000 95% 0.2778 Yang / Korean Y\n",
            "100000 100% 0.8105 Remishevsky / Russian Y\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "csFMr2i14_yO",
        "colab_type": "text"
      },
      "source": [
        "Great! Seems like things are improving. We can now compute a confusion matrix to see how well we have done."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sRNGSY694_YB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "confusion = torch.zeros(n_categories, n_categories)\n",
        "n_confusion = 10000\n",
        "\n",
        "for i in range(n_confusion):\n",
        "    category, line, category_tensor, line_tensor = randomTrainingExample()\n",
        "    output = test(line_tensor)\n",
        "    guess, guess_i = categoryFromOutput(output)\n",
        "    category_i = all_categories.index(category)\n",
        "    confusion[category_i][guess_i] += 1\n",
        "\n",
        "# Normalize by dividing every row by its sum\n",
        "for i in range(n_categories):\n",
        "    confusion[i] = confusion[i] / confusion[i].sum()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3yqDdcGB5wj_",
        "colab_type": "text"
      },
      "source": [
        "Let's visualize the confusion matrix!"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KwLA93uT5psj",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 320
        },
        "outputId": "9c52d6fd-06fd-4281-9ca7-ca62278e07f5"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import matplotlib.ticker as ticker\n",
        "\n",
        "fig = plt.figure()\n",
        "ax = fig.add_subplot(111)\n",
        "cax = ax.matshow(confusion.numpy())\n",
        "fig.colorbar(cax)\n",
        "\n",
        "# Set up axes\n",
        "ax.set_xticklabels([''] + all_categories, rotation=90)\n",
        "ax.set_yticklabels([''] + all_categories)\n",
        "\n",
        "# Force label at every tick\n",
        "ax.xaxis.set_major_locator(ticker.MultipleLocator(1))\n",
        "ax.yaxis.set_major_locator(ticker.MultipleLocator(1))\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": 39,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAVwAAAEvCAYAAAAJoHlDAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0\ndHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nOydd5wdZfX/35/dJCQQCFVKaEGCGHoI\nSCdUUZpdmoiF8lWK+kPFhhEL9kIRKSKoCAgoRYKA9BZIJSFAKAEkgECoCYRks3t+fzzPzc7ezNyZ\n2dzde3c579drXjvlzDPPnb33zDPnOUVmhuM4jtPztDS6A47jOO8WXOE6juP0Eq5wHcdxeglXuI7j\nOL2EK1zHcZxewhWu4zhOL+EK13Ecp5dwhes4jtNLuMJ1nH6GAkdIOjVury9p+0b3ywF5pJnj9C8k\nnQN0AHua2fslrQLcZGbbNbhr73oGNLoDjuPUnQ+Y2WhJUwHM7DVJgxrdKcdNCk4BJC0v6XuSzo/b\nIyUd0Oh+OZm0SWoFDEDSGoQRr9NgXOE6RfgTsBDYMW4/B/yocd1xcjgD+CfwHkk/Bu4GftLYLjng\nNlynAJImmdkYSVPNbJu470Ez26rRfXPSkbQpsBcg4BYze6TBXXLwEa5TjEWShtD5ivpewojXaULi\n/+cpMzsbeAjYR9LKDe6WgytcpxjfB/4NrCfpEuAW4BuN7ZJTg6uAdkkbA+cC6wF/a2yXHHCTglMQ\nSasBOxBeUSeY2dwGd8nJQNKU6KXwDWCBmZ2ZNAc5jcNHuE4uknYG3jGz64GVgW9L2qDB3XKyaZN0\nKHAk8K+4b2AD++NEXOE6RTgHeFvSVsDXgCeBPze2S04NPkfwKPmxmT0laQTwlwb3ycFNCk4BEq+o\npwLPmdkfK/sa3TcnnTjJub6ZzWp0X5xOfITrFGGepG8BnwGul9SCv6I2LZIOBKYRJjqRtLWkaxvb\nKwdc4fZpejEC7NMEN7DPm9n/gHWBX/TAdZz6MA7YHngdwMymARs1skNOwBVu36ZXIsCikr0KWC7u\nmkuIZHKakzYze6Nqn4f2NgGucPs27zWznwNtAGb2NsFtq65IOhq4kuDTCTAcuLre13HqxkxJhwGt\n8a3nTODeRnfKcYXb1+mtCLAvAzsDbwKY2ePAe3rgOk59OAHYjPBduJTwf/tKQ3vkAJ6esa9THQG2\nM3BUD1xnoZktksLgWdIAopJ3mo/4pvOduDhNhCvcPoyZ3SxpCp0RYCf1UATYHZK+DQyRtA/wJeC6\nHriOUwckjQG+DWxI4jduZls2qk9OwP1wewFJuwAjzexPMTfpUDN7qg7t7gxMM7O3JB0BjAZ+Z2bP\n5JzXCqxJ1x/jf2vItwBfAPYlKPYbgQvMvzxNiaRZwNeBGSQmy2p9L8p+J5zu4Qq3h5H0fWAM8D4z\n20TSOsAVZrZzHdqeDmwFbEnwWPgj8Ckz273GOScQTBEv0vljNB/9dJ/4ED2apUeUn29Qf+42s11K\nyPt3opdwk0LP81FgG2AKgJk9L2nFOrW92MxM0sHA2TEC7As555xEUP6vFL2IpKdIsdmaWZ/w7ZS0\nCWHEtwFdFeKedbrENcBdwH+A9jq1uSx8X9IFhKxuSyZRzewfGfKlvxNO93CF2/Msikqx4kmwQh3b\nTkaA7VowAuxZoNpHM48xifXBwCeBVUu20UiuAP4AnE/PKMTlzeybPdBud/kcsCnhu7BkxApkKdzu\nfCecbuAmhR5G0snASGAf4HTg88DfzOzMOrS9FnAYMNHM7pK0PjDWzDITy0j6I/A+4Hq6jn5+XfLa\nk81s2+71fKm2lgM+ztKv5KfVqf269TWj/R8B95rZ+J66RhkkzTKz9xWQ+1pc3Yw6fCecfHyE28OY\n2S/jzP6bhC/1qWZ2c53a/p+kqwgKHYpFgP03LoPikoukZJKaFsKIt57fnWsII6zJ9Iwf8XWSvkS4\nN0mF8uqyNCppHmHkKELKyoWEIBSF5m2lZWl/GbhX0igzezhHrmLaKv2dcLqHj3B7mGhCeMfM2iW9\nj6B0bzCztjq0fTRwDLCqmb1X0kjgD2a217K2XXWd2xKbi4GngV/WKxOVpIfMbPMS8qUmqaINuhrr\nKzboskh6BHgv8BThAVN5AOROgkWz1FAze7Nne/nuxBVuN5C0E0v/2FNf4yVNBnYFViFUT51EsOse\nXod+TCMkKbk/UdxxhpltUeOcNQjlcTYj2GMr/a/XBFJpJJ0HnGlmMwrK30uYpJpMwiZrZlf1TA9z\n+5PmnvfbRrlVZSWHz3ILk/Q34DjCvZwIrERwL/QERXXGTQolkfQXwuhhGp0/diM7IbfM7O3oPXCO\nmf08Ksp60J0IsEuAy4EDCD+yzwIv1zohYetL8gYwOWai6haSZhD6OwD4nKTZFBuRlZ6kkrQ5MIqu\nD5l6JVE/B9gqJmj/f8AFhITfme55PUlFsUp6D4nPW4NRZvampMOBG4BTCA+zVIUb39oWmFlH3G4B\nBscIN6cGrnDLM4bwBS36aiBJOwKHE4IHAFrr1JfuRICtFt3HTjKzO2IbE3POGROXStsHANOB4yRd\nERPodIfuppL8l6QPF52kir7QYwkKdzzwIcLbRq3JxeEs7UZ2Z4Z40j3vrILueT2GpIOAXwHrAC8R\nPscjhLeaNAZKGgh8hND/topXTQa3AHsD8+P28sBNwE516H7/xsx8KbEQXIzWLiG/O3At8M24vRFw\nRs45rYQfy/qVJUOuhWDLvIKQzetoopmoRtsT4t8bgf0JPsJP5pxzJ8GuV9keCtwBDAEersM9fS+w\nXFwfC5wIrFxDfh7B3WkBYTJyHvBmDfkZ8V49GLfXBG6uIf8zgp16POEhcx1wbQ35O4BvAY8Ba8Vr\nzWjgd/RBYDVgatzeA/hjDfkTCak9xxPeLjYA7qohP63IvsSxESn7tmvU/Wnk0vAO9JWl8qMDbgNe\niwrr2spS4PzlC17nBIK3wcyoKGYA0+v4OQ4AhgGbx88yGTgo55xHgYGJ7eWAR+P61Dr0aRphJLlx\nVFq/AMbX8TM/EP9OJtgnVel/hvysygOgYPtrEWq97Rq31weOzDmn0EM1yn4MeJxgxinygJkU/z4I\ntFTWS96zATWO3QOMTmxvC9xXQ34KMDyxvXsjH0iNXNykUJxfduekaE74I2FUuH608x1rZl/KOKVw\n1E+crBlH56tvxfaZOftuZpUqrm8QRj5FuAS4X9I1cftA4G/RlpfnelSEDjNbLOljhMmzMyVNrXWC\npFUI7nBJm2zWK/8kSSsTAh8mE16F76vR/GxC0EAhFzULCdp/ndj+L7XNFamhtIQQ7TR+DhxoZo8U\n6Q/wuqShhDeTSyS9BLyV0o8jzOyvGTZ6SHymKr4CXCHpecJ3bi1CVZAsjgWuVij9M5rgj/7hYh+l\nf+FeCiVRqID6gpm9E7eHAGua2dMZ8vcDnyCMgiueBJluUNEFax8zW1ygL48CX2Xp2fpMZR37fwJL\ne1kclHOtMYT0jwD3mNmkvP4VJd6j3xLSCR5oodJsrXv0RcKDaV3C6HgHwggr19NC0obASmY2vYbM\nVYQcFdWhsSdWyd1tZrsk/HGXHKKGH66kJ4APFHmoRvl7rETujYorYuzH4YQ3mkuqryfpWDM7N9q4\nqzGrEXgSbb6V4IpZluPmGAce58Z+7W9mNSdq+yuucEsiaRKwk5ktituDCApouwz5+83sA5KmJhTu\ng2a2VZVc6aifStsl+/8gYcRdnUnqjhrnrJ+23+rk9iRpFMFj4j4zuzQ+FD5lZj/LkJ8BbEewR28t\naVPgJ2b2sSq5Tc3s0arAjWT/p2S0/9kM+YuLf6psij5U44gfwiv4WoQqG0VyI5Ttz85mdk+BfXua\n2a2JfnWhuj+SrqPrg2gU8ALBJJf7kO+PuEmhPAMqyhbAgltWreicZ6PfrsVRwUmEGeNquhP1c5uk\nXxBi5JM/xFRFEnnHzM7Iabea6+n84QwBRhDsnFmz3qWwEBF1YmL7KcLEVRbvmNk7kpC0XFSqaaGs\nXyMEhvwq7bJA6oi4O4pVBdIbJh6qs4HbJeU9VA9MrL9NSI+Z7H+qwo0K8WeEqhwiP/LtTMKrft6+\n3YFbq/pVqz/dMsP1Z3yEWxJJNxPsjNfG7YOBEy0jukvS6sDvCG40IrjPnFT0dTKnL7el7LZar9YK\nta5Gxn4UVdLVbYwGvmRmXyzR3bR2/m5mn0r443bBMvxwJf2TkKDlKwSl+RphUq8udkGFiL3TWdpv\nN9U2nmWTre5/xqv7EszsB8vQ7eR1nqCAzTe+5u9EuI+/SRxaCfho9VvYMvZpA0JO6P9IWh5oNbN5\n9Wq/r+AKtyQKdcMuIcwwi5Bp6Ugze6JO7d8MfNLMXo/bqwCXmdkH69T+6YTsYk/SVTmUijRTTkRb\nwTbWNrMXykZGVbWxO8FG+e/km0eVzCfj8XmSvksYuf3QzFIn5iTdTVCgvyGM5j5HmO0/NUO+lE22\n6tzcUFpJFxMe0snvxK8sO5S5kM033ruxBHPOHxKH5gHXWahdl3beaoT7swvhQXk3cFrW51cvhaD3\nBdykUBIzexLYIc4CY2bza8mrfHLqNSo/rCj3mkLEUFb7+7N0mG6tLFufBDbKUk4Z10jOYrcQFNbz\nRc/PwsxeiH9zFWtKn6qraAwn5A5I43tmdkU8Z2+C29kfgCz79xAzu0WSYt/GKYRopypcSqY3VEoo\nraRaobRbpnwntqlxiUmSLifH5mudgS8LrCp4JT6kUhUucBnBA+LjcftwQvTi3hnyXyaGoMfrPl7r\nO92fcYVbElWlElQMq62h5Momp26XtH7F/hdHf6mvIZL+QIjy2YMQTvoJ4IGc9h8CViZEIBUlmTB9\nMcGmm5m3oOhDJmV2f8khas/yL6miQah0MRD4K51eFNVU7vv+wHlmdr1CSsUsFsaR5+OSjicEBQyt\nIV/UJluhVCgt0CJpFTN7DUDSqtT+7a5ECZsvcAjB9SzJtwgBNWmsbWY/TGz/SFIttzAvQhpxhVue\nsqkEy8b9fwe4W9IdBMWzK+F1LI2dzGxLSdPN7AeSfkX4AddiZeBRhXDepHKoNWP8sJl1+fHFEVDW\nD7LQQ8bMulv5omwVjecknUvISfyz+NBsqSF/EuFBdiLwQ4KdONVzIVI2vWHZUNpfAfdJqtzvTwI/\nqSF/QZrXQbWQpA8R/GGHS0pOpK5EeLBmcZOkQ4C/x+1PEAKBsrhDXoQUcBtuaVQ+lWDp5NRxom2H\nuDnBMirxJlzOJhCikV4BZprZxjXaTk2okuMWNsXMRuftSxybZmZbZ7WXIp9WPWKeZfh2SnrAzLav\n9EHB7/S+GpNsywP7EaKbHpe0NrCFmd1UtI/1RNKJwDcJkWD7EyLN/mpmu9Y4ZxSdXhW3Wo1ct0X/\nXwpBOFsDp9HVXDIPuK0yok7IJ/P/rkDnHEALML/GG4kXIY24wi2JyqcSnEf4chZKTq3w3nU4wc56\nmoIP7FpmtpSpQNL3CO47ewFnE34M59eY3GklKORNC/a9MgL6FMFGV2Elwmvx9hnnlXrISHoaWI/g\nbSDCKPx/hFn/o81scpV8qSoacaJzjpktlDSWENH156RdNMr91sy+oqX9R4HstwDVIeWlpAGW4Zcr\n6S9m9pkC+7rldRBH2wI2ibtyAxnKohAgtL7VKYdyX8VNCuXZBThKIal1birBbrw2/54wctiTMPKY\nR7CXLhVYkbCjXSXpX4QUeZmTNxaSoM9K2ohzeJ6Qv/cgggmlwjxChFsX1P0KCDcDV5rZjbGdfQl2\n8j8R7seSya34QLqcULOraBWNq4AxkjYGziOYPP7G0uGlf4l/y/qPFkp5qe6H0nbxd44PzrSSQYMI\ntuYBdLW7v0l47c9iJ0Io8tOE/9V6kj5r2aHSFU+JQqHVCtnLfhH7N0LS1gSvBg98cHL5UBEhdTPK\nieBeNFoxl0CckU61C0oaTLCHLXHPkXSOxbDjDFYBZkp6gER8fdqX38welPQQ8EErEAywDDbZHczs\n6EQ7N0n6pZkdG+2tyWuYpPEWXNKKlioqlKuhMpKuZV7JoGjKy0oB0UL3SaFAaMX2+SZBGQIsIjw4\nqvtfufZFZvaMpOWtWI7aXwP7VkafClWOLyVdqaOM0GoyAkkILmTbA7fHfk5TiCZ81+EKtyTxi1zt\nkpQ2g92tKCegLY5gKlV+1yARglvFnwmjzcqr9GGEUdona3yE79U4tnRHw6h4PUmDirqSqXwFhBck\nfZPgbgQhEcqL8T6kffYpkrYzs7w8vhXaJB0KHElnlNRS1Y2VEYBRIesthjCKh/A59ie8GSxllzaz\nc+PfQgEOZnY6cLqk083sW0XOiawj6QaKJ0wamHzVN7PHopkhi5PoDK3eQzG0uoZ8m5m9ISm5711p\ny3SFWxIVdEkys2Pi36IZuSqcQSh2+B5JPya8Cn43Q3ZzMxuV2L5NUs3sXWZ2h1KifnL69BRwj6Rr\n6ToqznoFLlsB4TDCKOjquH1P3NdKsB9X8wHgcEnPxP7kVYj4HOFV/8cWEuOMoNN8kKS7CdF/JGkY\n4bOeSbCZpplcaoZUW1VynAQ3SNotRT7rlf+3wAcJqUMrbypLnZ9gkqQLCN9jCHMItZITFQ2trjBT\nIcKxVSHo4UTg3hry/RZXuOUp65KEStRAM7NLFJzs9yIoko9YdojmFEk7mNmEeJ0PUPuH0iXqh5D4\nezghCKBW1M+TcWmh2OtwqQoI0QvjhIzDaRF8paLurGCuButGAEY8r2jKy8k1jtXi64n1wYTX88lk\nvyVhZs9WjShr+YD/HyE4oXKP7iLYzrOYo5Du8mrgZkmvAbXu3QkEd8eFBNv5jUAtP+h+i3splETl\nXZJSa6BljWYk/ZFgZ5yW2DfOzMalyD5CGGlXXtXXJySVWUzGiE/dKDyZOLdodN0dwL8JI8vdCEEW\nD2ZdI9oMT2bph1KtnBBbEXyUIVQneLCG7FOkex1k5UbYgTBSfT9hoqcVeKuGZ8lGhHwZOxJMIPcB\nXzWz2Vl9iucVup8p561HMNF8POP4lQS77FmEt4GTgDFmdkiG/JLK0nG7lZCAPdf+q5zQ6tjWf7rx\nptc/sSbIgt6XFoJiOJcQXXQ04cd1Qg35R6B22Zsq+TkE/8wjE/umZMhuUGvJOOf++LdSfmUAORUl\nCNUhphJGMc8QRleb1ZAvVQEhft7/IzwItq0sNeRPIkTMnRaXGTn/g9USy3CC29RpNeQnEapPTCUo\n288Bp9eQn0DITzEgLkdU7nOB+/nfvPuZcr6oUdoIWJ3gOfEi4WH3V8LEXq3+V5dQujdDtpUa1TIy\nzrkFGFbmnP66NLwDfWmJX/T1CP6fvyC4D+2Tc07ZGmhTCCOG6wi+tQPIKWNDSMOXW6olyv6cMPP9\naPwc/yTYNmudcy+wR2J7bNYPspv3dXJJ+enAContFfIeGmWuSWeJmumJfZn/g7RrU6OkTdn7SRht\nnxGXswjJYv5ax/tftkbZNXnfsxT5/xLyMFc+R826fv11cRtuCcyKuySp03l+ReDh6IZVJJRWFnxp\nD5Q0jlB3bFjGNcpWZ4UQt/8FwqjwGOB6M7ug1mchKLclqSDN7Pb4Glrdn25VQACuk/QlgvJP3qNX\nM+RFV5tkO50uU0sLd3XNayFMetb67r8dXfGmSfo5IWl2rVDgGySdQvCyMIKXxXjFCLqUz1HofiZI\n2uUXA5daVegugKRvmNnPJZ1Jugkla1LuLUmjLboqKlT3WFCjP2muhWZmB2fI/4POPA6VfmX+v/oz\nrnDLU9Ql6VpCQuq7qvbvSvgBd0HBKX+teB4AZjZOUjthBJTGDwk+kP8xs20k7UF4nV2KOIG1rpmd\nDZwfJ8/WALaV9LqZXVnjs8xWiGqrzOwfQTCpdMHMdol/y/rjVvIUJCeHjFDhOI0/EWqs/TNuf4Qw\nesoi6Zq3mODgX8t17jMEBXs8wdtgPULodBYVT4pKzouKMjmE9M9R6H5WMLOLo3sgVrs0zXGS7iVn\n4jSFZI0ygLWpXaMs6VpYyfexlH246jtHVNBrEO5Jmfwi/YdGD7H72kJ4FW8nzNpPJ6OqLvAvQrx+\n9f4tCLlGl0k+HitcnZXgarVeYnsawVNhfeCWnM+8CuE1cArB3vhbYJUM2dI2vhL3Ptn/0YRZ9RMJ\nXiMHlGinFTi8xvGTCu7bjhB2Xdn+LOGBeQYh9+sy3U+CMhtHqOL8KiH0+WVCZF1au18hzCk8TTAd\nbZNzH5b0n+DeeDyhosNZtfof5bchmNWeJryFLWVDX5bvXH9dGt6BvrIQbVYUnKACJtZoa6kS0WXl\n4/7/ECY4ziJEBv2O7MmOiVXbZyXWJ2ScMzj+iM8iVF4dmNXHqvMK2fiAbyTWP1l17Ccp8o8CG6bs\n/zzwZMr+lQhpBs8i2KsVlcpTwDU1+rXUJCUpNtyoMFeN67sRAh4+TnjzuHJZ7ydh4vFmYERi30YE\nt6qv1jhvA8IIcmq8Z6cS/K6Xtf+bEPylHyXYkU8AnqnRj9Lfuf6+NLwDfWVJ/giBqwrIP17j2BPL\nKh/3L0949W0ljK5OIGNkktVGPLaUsor7LyfMcB9L8Ln8bcF7dSchAu4WwojvWkLV4lr3dErWscS+\nDwOPJZUHwSY9g/DqWi1/DXBR7P/fCaGldwBbZ/T7UMJk5WvJfhNGcEuNyEi8TRAmOMclttMmokrd\nz6gwV0/ZvwY5E6kJ2W1iO+116H9HvH8bJ/bNLvu9rfWd6++L23CLkzTyZ9kWk0ySdLSZnd+lkRCH\nnuYAX1g+ZVIq2b9TJT0JfMfMbkkcvz+j/WPJTlo+yqLvbPQPzktuXqFo+LAy1tO2MbPxCglxbpD0\nEeCLBFey3awqlWBko0T/LyDYzte37FwT90aZ1elq951HMB9V06rOLF970TVvcdpvq+z9HGgpqTnN\n7OVaobcKCb4/RLCr7kV40IyrQ/8/Ftu8TdK/CZOEtSa/uvOd69e4wi2OZaxn8RXgnwpZ/SsKcwzB\nkf6jyyJvNSaloqP55gQ/zGTe3q8CV8cQy0rinG2B5QiTTmksSdFnIflL1mW7YMWTv9S6p6n32ELp\nm88RlMi9wJ41FGiy/+2S5tSQxUKk2TOEAIYiXEpIFjOXMKt/FyyZAE3L2lb2ftbKXZEWZLAPYZT+\nYYJCuww4xszeqpbtTv/N7GrCd2gF4GDCd/Y9ks4B/mlL5xfuzneuX+ORZgWJ3gKVuP0hhBImkJ/f\ndg86Fd9MM7s15zql5Gu0c6zFZClV+/ek022sZvuJzwxdP3fqZ84YeVNDvtY9HWxmA6vkk+kflyMo\nsIpLWK32e6T/8ZwdCLP6N1UUW4ycG2pVGeG60Z+kfHV/0u7PrYTQ2asyRvxLN1Si/xnnr0Lw+Pi0\nZVeuLvyd6++4wnUcx+klajlzO47jOHXEFe4yICmruGNd5HvjGi7v8r19je70qb/gCnfZKPvF6c4X\nraev4fIu39vXcIXrOI7j9Cw+aVaAQcOG2OC1lnZCaHtjAQOHDVn6hMfSC562sZCBLJd6LIuy56TJ\nqzW7oMMie4dBGpx5fGn5BQxSymfO+B5ltp/hEbWo4x0GtaTId2S0z0IGpdwfy/Dca7OFDFTK/cz4\nGdTj/ndHXgPTPTYXdSxgUEvK/Qds8dJFfzM/L6AB5a6R1n6ta6glfTyX9T9+s/2VuWa2RupJBfng\nHivYK6/WyrUemDx94Y1mtt+yXKs7uB9uAQavtRI7nHtoYXnb+3/lLmBZJcuy5Ms9JFuHrVKufSjd\nJ2tL/zFmUeshkNr+okLl1DrlM5RDpnx7/o90mSj5Pxuw+pqlL9E+95VS8q2rr9aj7bcsv3wp+Rvf\nuLBbFTeSvPJqOw/cuH6uXOvaj6++rNfqDq5wHcfpNxjQkVlztfE0jQ1XUrukaYllwx66zlhJ/8qX\ndBynr2EYbdaeuzSKZhrhLjCzrdMOKMRAyqzsu7fjOO82fITbDSRtKGmWpD8T6letJ+nrkiZKmi7p\nBwm5RySdL2mmpJukMKsjaWNJ/5H0oKQpkt4bmx8q6UpJj0q6REWTBDiO09QYRrvlL42imRTukIQ5\noZLJfyTwezPbjFCddiQhO9TWhEoFuyXkzo5yrxNyekJI4HK2mW0F7ERnpYVtCIk3RhEyf+1c3RlJ\nx0iaJGlS2xu1qo04jtNMdGC5S6NoWpNCtOE+Y2YT4q594zI1bg8lKNr/Ak9ZZ1nxycCGklYEhpvZ\nPwEqWaLiYPYBM5sTt6cRynPfneyMmZ0HnAew0vvWdN85x+kDGNDeQIWaRzMp3DSSmZJEKFXdJQNW\nVMwLE7vaCVmYalEt3+z3wXGcgjRyBJtHM5kU8rgR+LykoQCShkt6T5awmc0D5sRE1UhaTlI5x0DH\ncfoUBrSZ5S6Nos+M7MzsJknvB+6LZoH5hGqntXw8PgOcK+k0Qu7UWpVaHcfp4xjmJoUimNnQqu2n\n6VqxADP7HaFQYjWbJ2R+mVh/HNizSnY2oVpAReb47vbZcZwmw6C9efVt8yjcpuaxNmzP5wqLf2jm\n66Wav2Gzlcv2qBTtrxVK/r9stJQL1aWjnPO5ttuilLxNnFFKvtlY/OJLpc9pff/Ictd45PFS8ho0\nqJR8+5tvlpKvByHSrHlxhes4Tj9CtNesa9lY+tKkmeM4Tk3CpJlylyJI2i8GXz0h6ZSU4+tLuk3S\n1BiM9eG8Nntd4UpaS9Jlkp6UNFnS+BhkkJrfQNIFkkb1dj8dx+l7BD9c5S55xOrXZxPKzY8CDk3R\nQ98F/m5m2xDKx/8+r91eNSnEENp/Aheb2SFx31bAQVnnmNkXe6l7juP0AzoKjmBz2B54wsxmA0i6\njFAa/uGEjAGVRNnDgOfzGu3tEe4eQJuZ/aGyw8weBO4iI7+BpNsljYnr8yX9OOZGmCBpzbh/DUlX\nxTwLEyXtHPfvnggXnhqjz0jLyeA4Tt+nxAh39Uroflyqy/4MB55NbM+J+5KMA46QNAcYD5yQ17/e\nVribE0Jv08jNbwCsAEyIuRHuBI6O+38H/MbMtiPkUbgg7j8Z+HIMGd4VWCBpX7JzMiyhSy6FLoFp\njuM0K4ZopyV3Aeaa2ZjEcl43LncocJGZrQt8GPiLpJo6tZm8FHLzGwCLgIqtdzKwT1zfGxiVSPq1\nUoxIuwf4taRLgH+Y2ZyocJmsntIAACAASURBVNNyMtyZvFCXXApatYk9+xzHSVInk8JzwHqJ7XXj\nviRfAPYDMLP7JA0GVgcyffp6W+HOBD6RcaxIfoM26yzClpRpAXaoJKhJ8FNJ1xOePvdI+iAZORkc\nx+n7GGKRlfQJT2ciMFLSCIKiPQQ4rErmv8BewEUxCnYw8HKtRnvbpHArsFzSXiJpS8Lr/rJwEwn7\niaSt49/3mtkMM/sZ4QZuSsmcDI7j9B1C4ENL7pLbjtli4HiCvniE4I0wU9JpkiqT/P8POFrSg8Cl\nwFGJAWEqvTrCNTOT9FHgt5K+CbwDPA1cvYxNnwicLWk64TPdCRwHfEXSHoTgk5nADWa2MCMnQ/nQ\nHsdxmo56BT6Y2XjCZFhy36mJ9YdJn2vKpNdtuGb2PPCplEPnJ2SOT6yPTawPTaxfCVwZ1+cCn065\nVuqsYY2cDI7j9GHMRLs1bzxXM02aNS0aOJABa1V7hGRzw+blnrBzj9mhlPya1zxZSp5FbeXk6UaZ\n8QXlqmK0DF0pXyhB++SH84UStK5Urv2yZdg1rFz7rFxOvmP55cq1D3TMmFVKvrXkZ+h4q+T/uGSZ\n9C7Zr5eBjiYO7XWF6zhOvyFMmjWvWmvenjmO45SkMmnWrLjCdRynX9FeHz/cHqFpHwUZSW42qUO7\n4ySdXI8+Oo7TXJSINGsITTnCrZHkZk3gsUb2zXGc5qajib0UmrVnWUlu9kkko3lO0p8AJB0h6YG4\n/9yYWq2Sz3JKTHZzS6L9UTEpzmxJJ/bqJ3Mcp8cIyWt8hFuW1CQ30en4VEkrEzKMnRWDGD4N7Gxm\nbZJ+Dxwu6QaCb+9uZvaUpFUTTW1KUOorArMknWNmXXynYjTcMQCDW1es/yd0HKfuGKKtPqG9PUKz\nKtxMornhr8CvzWyypOOBbYGJMXJsCCFqbAfgTjN7CsDMXk00c72ZLQQWSnqJYKqYk7xOMnnNsEFr\nevIax+kDmOGBD92gVpKbccAcM/tT3BbB1vutpJCkA2u0XyRRjuM4fQ41deBDsz4KUpPcSPoeIRVj\n0u56C/CJSgIaSatK2gCYAOwWs/1QZVJwHKcfYoQRbt7SKJpyZFcjyc3yhKzrD0TzwbVmdqqk7wI3\nxeS/bYSk4xOiwv5H3P8SnflzHcfppzRyUiyPplS4UDPJTZrs5cDlKftvAG6o2jeuanvz3Pbb2lj8\n/P+KdKXSaHFZYPXzJ5SS/8YTD5aSP33kNqXkAbCO8ueUoP3NN0vJa+Cgcu3Pm1dKfsD665aSX/xs\nbvmqrrxYLhmdBvT8T7P9jXL/g9bVyr0ktr/6ein5emCoXgnIe4SmVbiO4zhlCWXSm1etNW/PHMdx\nSlOsDHqjaF5jh+M4TkmMEGmWtxQhBk7NkvSEpFNSjv8mEYj1mKRcG0qPjnAlzU8mDXccx+lp6jHC\njdGqZxMm2ucQ/PyvjVUeADCzrybkTyBUHq+Jj3Adx+k3mKleI9ztgSfMbLaZLQIuAw6uIX8ooa5Z\nTXpc4UoaKumWmNNghqSD4/4NJT0q6RJJj0i6UtLy8dipkiZKekjSeTG6jJj/4Gcxb8JjknaN+1sl\n/SKeM13SsXH/2pLujEP+hxLy+0q6L/bpikpBScdx+jZh0qw1dwFWlzQpsRxT1dRw4NnE9py4bymi\n3/8IQvxATXpjhPsO8FEzG03IX/CrigIF3gf83szeD7wJfCnuP8vMtosuW0OAAxLtDTCz7YGvAN+P\n+74AvGFm2wHbESppjiCUNb7RzLYGtgKmSVod+C6wd+zTJOBr1Z2WdEzln9HWJTDNcZzmRUUDH+aa\n2ZjEct4yXPQQ4Eoza88T7A0vBQE/kbQboXrucELuAoBnzeyeuP5XQgTZL4E9JH2DEOiwKiHU97oo\n94/4dzKwYVzfF9hSUiUceBgwklAa/UJJA4GrzWyapN2BUcA9Ue8PAu6r7nQyl8JKWtVzKThOHyBM\nmtXFS+E5YL3E9rpxXxqHAF8u0mhvKNzDgTWAbWM2r6eBwfFYtSIzSYOB3wNjzOxZSeMS8tCZByGZ\nA0HACWZ2Y/XFo6LfH7hI0q+B14CbzezQZf5kjuM0HXWKNJsIjIxvys8RlOph1UKSNgVWIWXQlkZv\nmBSGAS9FZbsHsEHi2PqSdozrhwF306lc50bbalYSmyQ3Av8XR7JI2kTSCtG28qKZnQ9cAIwm5FjY\nWdLGUXYF1aGShOM4jacSaZa35LZjthg4nqBbHgH+bmYzJZ0m6aCE6CHAZWbFwkt7bIQraQBhNHoJ\ncJ2kGQR76aMJsVnAlyVdCDwMnGNmb0s6H3gI+B/hSZPHBQTzwpRoH34Z+AgwFvi6pDZgPnCkmb0s\n6SjgUkmVWtTfxStJOE6/oF5FJM1sPDC+at+pVdvjyrTZkyaFzYAnzWwusGP1QUkbAovN7IjqY2b2\nXYISrN4/NrE+l2jDNbMO4NtxSXJxXKrbuZUwuVYIDRrEgHWLx9q3z3mhsCxAy7ByCc5P33irUvLP\nn/yBUvIAw89YKv97TcrG/ressnIp+Y65r5SSh4GlpNuff7Fk++Uoe3/e3n906Wss/+9yOTZaVx5W\nSt7eXlBKXq0lE4HnTjnlYwZtHc3r7dojClfScYQJsK/0RPuO4zhpBJPCu0zhxlpkf8iReZpQSsdx\nHKduNHMuBU9e4zhOv6GObmE9Qq+PvSWtKelvChVzJ8eIr4/2dj8cx+mP1C20t0fo1StHD4KrCcUd\nNzKzbQluFYVmpKLng+M4TiYdsa5ZraVR9Laq3xNYFG28AJjZM2Z2Zo18CGMl3SXpWuDhuH2HpGvi\nKPmnkg6P+RVmSHpvPO9ASfdLmirpP5LWjPvHSbow5mWYLenEtI46jtP3CF4KrblLo+hthbsZMCXj\nWFY+BAgBCyeZWSVAYSvgOOD9wGeATWJ+hQuAE6LM3cAOZrYNIdPPNxLX2hT4ICEj0PcrARNJkrkU\nFnW83b1P6zhOr1KvwIeeoqGv6JLOBnYBFgHPkJ4PYRHwgJk9lTh1opm9ENt4Ergp7p9BSJADwUxx\nuaS1CfkSkudfb2YLgYWSXiLkdpiT7Fsyl8Kw5dbyXAqO00fwMumdzCSMVgEwsy8DexFyLVTyIWwd\nlxFmVlGkb1W1k0zf1ZHY7qDzIXImIevYFsCxpOdjgK45GRzH6cNUvBSadYTb2wr3VmCwpP9L7Fs+\n/k3Nh7AM1xpGZ3afzy5DO47j9CGa2UuhV0d2ZmaSPgL8JqZffJkwev0mcAXp+RC6yzjgCkmvERT9\niNrijuP0dczE4ndbpFktou31kIzDafkQbo9L5fzq7bFpx8zsGuCalOuPq9r2aDfH6Uc0c+CD2y6L\n0NGOzZtfWNzaFpVr/o15peQ1oFxiluG/eaCUPMCqd5RLqPPKzq+Vku94q9os31gGDF+nlHzHaiuV\nk5/+aL5QgqF3zColD9C+sFxlkrLyWm65fKGk/KBy31PK/WxSafZIM1e4juP0K1zhOo7j9AIVP9xm\npZR1WVJ7ogLuFYpVdgueu7WkD5fvouM4TnHqFdoraT9JsyQ9IemUDJlPSXpY0kxJf8trs+x03oLo\nI7s5weJyXJGTYg6ErQFXuI7j9BhmsLijJXfJQ1IrcDbwIULR2UMljaqSGQl8C9jZzDajQP7vZfGf\nuAvYWNKqkq6O+Q8mSNoydmacpL9Iugf4C3Aa8Ok4Qv50PH5yovMPxSoQSPpefLLcLenSilzMfzAm\nrq8eC1JSIw/D2pLuTIzKd437941ZyqbEkfrQZbgPjuM0EXUKfNgeeMLMZpvZIkJ6gIOrZI4Gzjaz\n1wDM7KW8RrulcOOI9UOEUNofAFPNbEuCS9efE6KjgL1jhdxTgcvjCPnyGm1vB3yckC/hQ8CYAl3K\nysNwGHCjmW0d25smaXVC+Z69zWw0oc7a11L6kcil8E6BLjiO02hK5FJYvfL7jssxVU0NB55NbM+J\n+5JsAmwi6Z442Nwvr39lJ82GSJoW1+8C/gjcT1CQmNmtklaTVPGZudbMyhVCgp2Ba8zsHeAdSdcV\nOGdf0vMwTAQujNFrV5vZNEm7Ex4E94T4CgaRUuK4Sy6FgWt4LgXH6SNYsRHsXDMrMpirxQCCnhlL\nyN1yp6QtzOz1WieUYUEcLS4hKq0sajlbLqbrCHtwlmDGOUn5Sh6GG6tPkLQbsD9wkaRfA68BN8dR\nt+M4/Yw6Ja95Dlgvsb0unakCKswB7jezNuApSY/ROdBLpR4xcHcBh0PIXUt4cryZIjcPSHrTP01M\nZCNpNJ2ht/cAB0oaHG2rB1Sds21c/0Rif2oeBkkbAC+a2fmE1I2jgQnAzpI2jrIrSNoEx3H6PGZ1\ns+FOBEZKGiFpECE69toqmasJo1uiqXITYHatRuvhhzuO8No+HXib7EQxtwGnRJPE6cBVwJGSZhLM\nEo8BmNlEhWTj04EXCXbiN2IbvwT+Hu0t1yfavoD0PAxjga9LagPmA0ea2cuSjgIulVQJnflu5fqO\n4/RlRHsdyqSb2WJJxxMGc63AhWY2U9JpwCQzuzYe21fSw4Ssg183s1dq9s6s+cyTkoaa2fzo53sn\ncIyZZSUu73GGDVzDdlzl44Xl2+fWvOdLoYGDynapHNZR+pSeDu1tNpottLd15WGl5AHaX38jX2gZ\nKB3aO6DceO6m+RdPXla76tBN1rbNzzwqV+7+/X66zNfqDs0aaXZe9HkbDFzcSGULgFT6y1MGa2/v\nsbbDBcor3Ff3LFflYsgda5aSb/t0KfHSDzFbvLiU/OIXXiwlz3PPl5NvKVnWpRvft7IP7pYR6+UL\nJbA5L5SSb0S+DM+l0A3M7LBG98FxnD6IBTtus9KUCtdxHKe7eImdHkBSZr5ESfd291zHcfouFifN\n8pZG0a9GuJIGmNliM9up0X1xHKcxNLNJoc+OcCtIGivpruhK9nDcNz/+Tc2lEI/9WNKDMSSv3IyP\n4zhNi5lyl0bR5xVuZDRwkplVBzAslUsh7l8BmGBmWxHczo6ubrBrLoWy0cmO4zQCs+ZWuP3FpPCA\nmT2Vsn+pXApx/yLgX3F9MrBP9YldcikMek8Tv6Q4jpOkmd3C+ssIN9Xhz8zuBHYjxEBfJOnIeKjN\nOiM+2uk/Dx7Heddjlr80in6taGIuhTlmdn4M4x1N1/SRjuP0IwzR0UAvhDz6tcIlJZdCY7vjOE5P\n08z2vz6rcM1saPx7O3B7xrGLgYuzzo3rVwJX9mBXHcfpLaxwPtyG0GcVbm9iAwfQvu4axU94aW6p\n9luHrlBKvuPtcnkOWtdcu5Q8QPuLudVCutB2eLmv0rwdqpPn12b5GzJzOqfS0z+5sj9qtZbLpdC2\nabk8BwCtEx8pJd8xrHANWAD0QrnP0LrKKqXkebWceCZNPMR1hes4Tr/CR7iO4zi9gAEdHc2rcBs6\nnSepPUaBVZbU2u8F26pEl60jKdMmK2lDSQ919zqO4zQxBpjylwbRaP+JBbGKb2X56bI2aGbPm9kn\n8iUdx+mP1MsPV9J+kmZJeiJtMCjpKEkvJwaMX8xrs9EKNxVJT0v6gaQpkmZI2jTuX0PSzZJmSrpA\n0jOxllDy3CUjWEmbSXog3ozpkkZGsVZJ58d2bpI0pJc/ouM4PYUVWHKQ1AqcDXyIUOX70FgUoZrL\nEwPGC/LabbTCHVJlUkjWAZhrZqOBc4CT477vA7ea2WYEV671c9o/DvhdzKUwhlBlE0JlzbNjO68T\ny7wnSeZSaFtczivAcZxGkZ9HoeCk2vbAE2Y228wWAZcBBy9r7xo9abZU2fUE/4h/JwMfi+u7AB8F\nMLN/S8orpHUf8B1J6wL/MLPHY1n3pxJ5FSYTClB2IZlLYaUV1mliRxPHcbpQ7Ne6uqRJie3z4m++\nwnDg2cT2HOADKe18XNJuhCK0XzWzZ1NkltBohVuLhfFvt3MdmNnfJN0P7A+Ml3QsoYzxwoRYO+Am\nBcfpDxhYMS+FuXUoInkdcKmZLYy65WJgz1onNNqkUJZ7gE8BSNoXqOlZLWkjYLaZnQFcA2zZ4z10\nHKfBqMCSy3NAMvpk3bhvCWb2iplVBm8XANvmNdpohVttw83zUvgBoQ78Q8Angf8B82rIfwp4SNI0\nYHM8cY3j9H/qMGlGSO06UtIISYOAQ4BrkwKSkiGcBwG5oX4NNSmYWWqsoJltmFifREhCA/AG8EEz\nWyxpR2C7yhMmkT/haYJyJbqZVSvxVyvHo8wv6/BRHMdpFuow4xJ1zPHAjUArcKGZzZR0GjDJzK4F\nTpR0ELCYoFeOymu3mW24aawP/F1SCyGJ+FKVGnqEt9/BphSPU9c2m5Zqvn3yzLI9KsXi557v0fa7\nc43l//diKfnPzEzLL5/NnzfNc2CpoqeTpFpHKfGWu6flC1XROmKDUvKLJ84oJd8nZo4rgQ/1aMps\nPDC+at+pifVvAd8q02afUrhm9jiwTaP74ThO89LMRST7lMJ1HMfJxXMp1KaSByGuf1jSY7Fag+M4\nTilk+UujaKoRrqS9gDMIE2PPFJAXILOSBjLHcfonxb0QGkJTjHABYrTG+cABZvZk3Pc1SQ/F5Stx\n34YxocSfgYeA9SR9XdLEmC/hB4k2r5Y0OeZMOCaxf76kH0t6UNIESWv27qd1HKdnKJAp7F2cLazC\ncsDVwEfM7FEASdsCnyOE0+0AHC2pMmE2Evh9zIXwvri9PbA1sG1U3gCfN7NtCXkUTpS0Wty/AjDB\nzLYC7iTF26FLLoUugWmO4zQ19fHD7RGaReG2AfcCX0js2wX4p5m9ZWbzCbkVdo3HnjGzCXF937hM\nBaYAmxIUMAQl+yAwgRA1Utm/CPhXXM/MpWBmY8xszECWW/ZP6DhO79BRYGkQzWLD7SBEhd0i6dtm\n9pMc+bcS6wJON7NzkwKSxgJ7Azua2duSbgcGx8NtZkucR7qdq8FxnCajjn64PUGzjHAxs7cJSWYO\nl/QF4C7gI5KWl7QCIUvYXSmn3gh8XtJQAEnDJb0HGAa8FpXtpgSzhOM4/Rz3UiiImb0qaT+CXfUk\n4CLggXj4AjObKmnDqnNukvR+4L6YenE+cATwb+A4SY8AswhmBcdx+jtN7KXQFAq3kgchrj8LjEgc\n/nWV7NMkciHEfb8DfpfS9IcKXO9KQjJzx3GcHqUpFG6zo5YWWgYXnzjrmPJwufYHDirbpVJY26Ly\nJ6mcHWzAOmvnCyVof/GlUvJ/ft96+UIJNp9crv8P71xuYrRjUVspeQ0pl3K5peT9B1j8VK7retdr\nDB6cL5Rk4MBS4rao5PfunXLiWTTSZJCHK1zHcfoPRlOH9rrCdRynf9HEI9yGeClIao8Jxx+SdJ2k\nlevY9gUZ1TUdx3kX0MxeCo1yC1sQywpvTkjc++V6NWxmXzSzckZUx3H6Dx5pVpP7CBUykXS7pDFx\nfXVJT8f1zSQ9EEfF0yWNlLSCpOtjPoSHKiXWq9o4J4bnzqzKsfC0pB9ImiJpRvTTdRynP1AnhStp\nv5i35QlJp9SQ+7gkq+idWjRU4UpqBfaiqlZQCscBv4sl1ccQShbvBzxvZlvFkfK/U877TqzMuSWw\nu6RkEcm5ZjYaOAc4OaVvS3IpLLI6TZ86jtOjFDEnFDEpRN10NsG1dBRwaJqpUtKKhJiB+4v0r1EK\nd0gs7Pg/YE3g5hz5+4BvS/omsIGZLQBmAPtI+pmkXc3sjZTzPiVpCiHPwmaEG1fhH/Fvbi6FQSrp\nPuM4TuPoUP6Sz/bAE2Y228wWAZcBB6fI/RD4GQWd2hpqwwU2IORCqNhwFyf6tETLmdnfCFUxFwDj\nJe1pZo8BowmK90eSltQaApA0gjBy3cvMtgSuT7YJS1KAeS4Fx+lHFBzhrl55g43LMVXNDAeeTWzP\nifs6ryONBtYzs+uL9q3RVXvflnQicLWk3wNPE2q7PwB8oiInaSNgtpmdIWl9YEtJjwKvmtlfJb0O\nfLGq+ZUISW7eiPluPwTc3tOfyXGcBlPMRjs3mhu7RSxk+2sKVOpN0vCRXcyPMB04FPgloSrvMYQR\naYVPAZ+R1EYwQ/wE2A74haQOQnrH/6tq90FJU4FHCU+qe3r8wziO01jq5/b1HCGla4V1474KKxJS\nDNwec7isBVwr6SAzm5TVaEMUbjKXQdw+MLGZnNj6bjz+U+CnVc3cGJfqtscm1o/KuP6GifVJwNg0\nOcdx+iD1UbgTgZHRNPkccAhw2JJLhDmj1SvbMf3rybWULTTBCLdP0CK0fIlY+LffLtW8WnvWlG6L\ny4c6ls2NsPj5F0pfoxQlcws8tG25LNPrTCiXJ+DFL5Srcdo+c1Yp+bKftzvnWHvJTNyLF5QS16Ce\nzRGSed06JBg3s8WSjicM6lqBC81spqTTgElmludZlYorXMdxnBTMbDwwvmrfqRmyY4u06QrXcZz+\nhedSKI6k+fHvhpIOKyC/oaSH4voYSWf0dB8dx2lS6hT40FM0ncJNsCEJI3URzGySmZ3YM91xHKdP\n4LkUusVPgV1j/oSvxpHsXTH/wRRJO1WfIGmspH/F9e0l3SdpqqR7Jb0v7j9K0j8k/VvS45J+3suf\ny3GcnqSJFW4z23BPIbhZHAAgaXlgHzN7R9JI4FJCXoUsHgV2jbONexN8dz8ej20NbEOINpsl6cxY\n2mcJ0Rf4GIDBLV282BzHaVJEfbwUeopmVrjVDATOkrQ1IRx3kxz5YcDFUTlbPL/CLZXcC5IeJoQY\nd1G4ZnYecB7AsIFrNLEZ3nGcJTTYRptHX1K4XwVeBLYimELykkX8ELjNzD4aK/3enji2MLHuuRQc\npz/RxAq3mW248wjhcxWGAS+YWQfwGYIzci2G0RmKd1Tde+c4TnPSxDbcZla404H2mGD8q8Dvgc9K\nehDYlJCYphY/B06P+RR8BOs47xKa2S2s6RRRJc+CmbUBe1YdTuZZ+GaUe5qQRAIzu51oOjCz++hq\n563kZbgIuChxvQPq1nnHcRpPE5sUmk7hNiUSGlg81l4Dyt3WjkVt5brTmmdNqZLvRkx7xxol63q+\n8GIp8bKfwdrbS8lDuanq2T96fyn5Fz5Trv8jvlUuz0HriivmC1XRPm9euRNaSuZeWLi4XPMrDysl\nn/vOWgRzLwXHcZzew0e4juM4vUMzu4X1+qSZpO/EKrrTYxTZB+rc/r05x+fX83qO4zQZTeyl0Ksj\nXEk7AgcAo81soaTVgbomzTSzpUJ+Hcd5l9BghZpHb49w1ybUEloIYGZzzex5SU9L+rmkGZIekLQx\ngKQDJd0f8yH8J9YmQ9I4SRdKul3S7FgXjXiskm1sbUl3xlH0Q5J2Tcj8OLqbTai06ThO30c0t1tY\nbyvcm4D1JD0m6feSdk8ce8PMtgDOAn4b990N7GBm2xDKFH8jIb8p8EFCOePvS6p2IzgMuDFWB94K\nmBb3rwBMMLOtgDuBo9M6KumYSkXPRR3lMt07jtM46qVwJe0naZakJySdknL8uDhInCbpbkmj8trs\nVYVrZvMJVXmPAV4GLpd0VDx8aeLvjnF9XeBGSTOArwObJZq73swWmtlc4CWgeqQ6EficpHHAFmZW\n8ZlZBPwrrk8mpIFM6+t5ZjbGzMYMailRXsdxnMZSBxuupFbgbEK171HAoSkK9W9mtkUc1P2cUMW3\nJr0+aWZm7WZ2u5l9HziezgxeydtQWT8TOCuOfI8FBidkauZDMLM7gd0I4b0XSToyHmozM8s6z3Gc\nPk59Js22B54ws9lmtojwhn1wl8uYvZnYXKFIy72qcCW9L2bvqrA18Exc/3Ti731xPZkP4bMlr7UB\n8KKZnQ9cAIzuVqcdx+k7FK/4sHrFZBiXY6paGk7XDIJz4r4uSPqypCcJI9zc4ge9PbobCpwpaWVg\nMfAEwbxwALCKpOmEkeuhUX4ccIWk14BbgRElrjUW+LqkNmA+cGRtccdx+gXFRrBzzaxWPu1ilzI7\nGzg7lgP7LjkDw15VuGY2GUir1ADwCzP7ZpX8NcA1Ke2Mq9rePLFeycVwMXBxyrlDE+tXAleW/BiO\n4zQxdQrtfQ5YL7G9Lp1v22lcBpyT16jbLwtgi9tpn/tq8RNUzlIzYMN1Ssl3vPhyKXlbuDBfqPoa\n02eVkm9ZYfly7b/1din5srkXsHJ5AgZfP7GU/Ijx5f7Huz5YztPlnu2XKyUPgBUb2i0RbyuXG4GW\ncv+D9ldK/GbqSJ3cviYCIyWNICjaQ6iqsShppJk9Hjf3Bx4nh6ZQuGa2YaP74DhOP6BOgQ+xNNfx\nwI2E3NsXmtlMSacBk8zsWuD4WL6rDXiNAvNMTaFwHcdx6kadAhvMbDwwvmrfqYn1k8q22bQJyCW1\nJ6LErohFJLNkj5J0Vlw/LuECliY7TtLJPdFnx3Eai0eadZ8FZrZ1nBBbBBxX5CQz+4OZ/blnu+Y4\nTrOiDstdGkUzK9wkdwEbS1pV0tUx09gESVtWCyZHsJJOlPRwlL8sITYqLQ+D4zh9nCJBD++WbGHd\nQdIAQnjdv4EfAFPN7COS9gT+TAieyOIUYETMTJYsYbApsAehSOUsSefEkj7J6x5D8BFmMOVm4B3H\naRyeD7d7DJE0DZgE/Bf4I7AL8BcAM7sVWE3SSjXamA5cIukIQqBFhbw8DF1yKQzU4OrDjuM0Kz7C\n7RYLYlKIJcQAiTLsT8incCDwHUlbxP018zA4jtN38RFu/bgLOBxA0lhCeN6baYKSWoD1zOw2QoXf\nYYTQYsdx+jM+wq0b44ALY86Ft6ntaNwK/FXSMIK3yBlm9no3RsmO4/QVvGpv90jmPEjsexX4SMr+\ni4CL4vq4xKFdUmTHVW1vXi3jOE7fpOKH26w0rcJtJjSgldbVViks3/7a66Xa73j5lXLyC94pJd8y\npPykn5W8hr1TLl9D69AVSsm3z3+rlDxWbpjTsly53AUdi9ryhRLctVW5JPbL3b5yvlAVi/aaW+6E\nkvkpWlYsaZErmduBkt2v23V7EVe4juP0K3yE6ziO0xt41d7yJPIozIzVdf9f9DrIO+/bBWQukvSJ\n+vTUcZxmQx35S6NozNlYPgAAD2ZJREFUSoVLZx6FzYB9CJFm3y9wXq7CdRynf+MKdxkws5cIIbbH\nK7AkMxiApH9JGivpp8ToNEmXxGNHxjwKD0r6S6LZ3STdG3Mp+GjXcfoLRpg0y1saRJ+w4ZrZ7Fi2\n+D01ZE6RdHwlOk3SZoQaQzuZ2VxJqybE1ya4jG0KXEtKmZ0uuRRaPV7CcfoKzTxp1vQj3GVgT+CK\nmC+h4sNb4Woz6zCzh0nJoxDll+RSGNRSzqXHcZwG0sSRZn1C4UraiJDz4CVCEppkv7uTWSbpNOqh\nZ47TT6hnAnJJ+0maJekJSaekHP9aIv3rLZI2yGuz6RWupDWAPwBnmZkBTwNbS2qRtB6wfUK8TdLA\nuH4r8ElJq8V2kiYFx3H6I5affLxIAvJowjybMGE/CjhU0qgqsanAGDPbkmCW/Hleu81qw62kZhxI\nGNH+Bfh1PHYP8BTwMPAIMCVx3nnAdElTzOxwST8G7pDUTrg5R/VS/x3HaRT1MRlsDzxhZrMBYgGD\ngwl6J1wmJMaqMAE4Iq/RplS4ZpYZcxhHuYdnHPsmITNYZfti4OIqmaOqtn1GzHH6EQVNBqtLmpTY\nPs/MzktsDweeTWzPAT5Qo70vADfkXbQpFa7jOE63MKBYzbK5ZjamHpeMBQ7GALvnybrCLYC1LWbx\n/14sLN+x6zal2m+5a2rZLpWi462SiV+6gXW0l5Jvb1tUSr51jTXKtf/yy6XkO94pl6ynp1m4+/9K\nn3Pq7Cn5QglO22h0Kfn2heUSFDWM+pgUngPWS2yvG/d1QdLewHeA3c0s9wY1/aSZ4zhOGerkpTAR\nGClphKRBwCEEn/3O60jbAOcCB8UArVxqKlxJt0n6YNW+r0h6Ks1NokpurKSdinTCcRynXtTDS8HM\nFgPHAzcSJuf/bmYzJZ0m6aAo9gtCFZkrYoTrtRnNLSHPpHApQbPfmNh3CPBZM7sz59yxwHzg3rxO\nOI7j1IU6BjaY2XhgfNW+UxPre5dtM8+kcCWwfxxSI2lDYB3gvZV8BpLWkHSVpIlx2TnKHQd8NWr+\nXWOWrjOqcxhIGhqdhqdImiHp4Mq1JD0az3tM0iWS9pZ0j6THJW0f5VaQdKGkByRNTZy/Wdw3LTom\nj4z7j0jsPzf62zmO0w8IgQ+WuzSKmgo3hsM+QHD+hTC6/TtdnyG/A35jZtsBHwcuMLOnCcEKv4lZ\nv+6KspUcBgcAP4373gE+amajgT2AX6mz8NjGwK8IOQ82BQ6L559MZ2aw7wC3mtn28fxfSFqBoPB/\nF3MrjAHmSHo/8Glg57i/nQwXM8dx+igdBZYGUcRLoWJWuCb+/QKwReL43sCoRHHGlSRl+bZebWYd\nwMOSKjkMBPxE0m6EWzGczvwGT5nZDABJM4FbzMwkzQA2jDL7AgdJOjluDwbWB+4jlEZfF/iHmT0u\naS9gW2Bi7O8QQrjwUnRJXsPyNW6P4zjNRCNHsHkUUbjXAL+RNBpY3swmS0oq3BZgBzPr4lej9Oq4\naTkMDgfWALY1szZJT9OZHyEp35HY7kj0XcDHzWxW1bUekXQ/sD8wXtKxUfZiM/tWrQ8MIXkNIXKN\nlbRq8/4HHcfppK9XfDCz+cBtwIWE0W41NwEnVDYkbR1X5wErFujDMOClqGz3AHITQFRxI3BCxQwR\nXTUqCW9mm9kZhIfGlsAtwCckvSfKrFok4YTjOH2F+uRS6CmK+uFeCmxFusI9ERgTJ6YeJthOAa4D\nPlqZNKvR9iXx/BnAkcCjBftU4YeEnAvTo9nhh3H/p4CHYk6GzYE/x3SM3wVukjQduJlgV3Ycp7/Q\n1xOQm9nVJNIYmtlFwEVxfS5hIqr6nMcIo8oKd1UdH5o4f8eMS2+ekD8qsf505ZiZLQCOTbn+T+mc\nmEvuvxy4PON6juP0ZayxJXTy8NBex3H6F3180uxdjwYNZMA66+ULRmzG7FLtl8tCUJ4Ba6UWtahJ\nmdwRAKRPkmYyYM3MakmptM99pZR8aUr2n/wi0l0pmWtiwPB1yrUPnLZROfl1J5RLlDdnpwWl5NVS\n8p62lRPPpHn1rStcx3H6F+poXpuCK1zHcfoPRkMDG/Lo9Wxhkr4jaWb0apgmqVZS37Tzt5b04cR2\nlyQ5ko6TdGSN88clgiQcx+lHiPyw3kYGRvTqCFfSjoSw3tFmtlDS6sCgks1UQnUrSSXGkkiSY2Z/\nqE9vHcfpkzTxpFlvj3DXJmRaXwjBJczMnpe0XUxq82BMLLOipMGS/hQT2kyVtEdMonMa8Ok4Ov4m\nSyfJWTKClXSiOqtqXpboxyhJt8ckOif28j1wHKcn6et+uHXkJuBUSY8B/yH4w94X/37azCZKWglY\nAJxEKGG2haRN47mbAKcSKmUeDyBpCDDfzH4Zt/dKXO8UYEQcTa+c2L8pIdHNisAsSeeYWZc50i65\nFFqLBMw5jtNw3IbbSQwT3pagyF4mKNpjgRfMbGKUeTMm/90F+Gvc9yjwDEHhlmE6cIlCzaHFif3X\nm9nCGHTxEp3JcpJ9Pc/MxpjZmEGtQ0pe1nGcRqGOjtylUfT6pJmZtZvZ7Wb2fUJG9Y/14OX2J9SW\nH03IEFYZ0SeT4rTj3hqO008oYE4oaFKQtJ+kWZKeUEqFG0m7xTzeiyv5vfPoVYUr6X2VROCRrQnl\nK9aWtF2UWTEqxruIuWolbUJIuTiLpZPipCbJ+f/t3W+IXNUZx/Hvb0ebrSls0ZVSVPzf2qS1+C+o\nWJpWCUqLsWhp1jdNLUSkRt8ISl+oRPAPvhCSpq1LaxVFVFIqS1uMmmj/0ZrdKEpjCC4JalQKSUys\nG8m6mccX50xyd/bOzpzunbt3t88HhsyceebcO8lycvfcc55HUg9wSqwdfzshSY6XRHduPjMKGXBj\nYYL1hFzgi4ABSYuawt4BVgJPdnp6ZV/ZfQFYF+dTJ4BRwvTC72L75wnzt1cAvwR+FZPaTAAr41zs\nS8AdMSnNfYQkORtipYfVmWPVgCck9RHyQKw1s/0t0kY65+aLYmYMlgCjZrYTIN50Xw682QiIOV2Q\nOs/eUOqAa2ZbgbzCknuAi3Paf5LTxz7goqbmVklyLsv5/N1Nr7/eHOOcm7s6XGfbL2kk83ow5sBu\nOAl4N/N6N5C0ZyCPz112YuIw9X37Ow6vjx1M6z/xqrt2zllJ8RPb30qKB9CCBUnxduhQ+6CM5FwN\niXoWLkyKt/G0jfx2ODEDRuK/cer5ANS+2JcU//7lE+2Dsv1vSszJsfzjtPgDaeEtdTbg7jGzCws6\nYsd8wHXOzR9mcLiQOYX3gGzGqpNj24yUvkrBOee6qphVCsPA2ZJOjxuuVgBDMz21Sg64kq6RZHHD\nQ+pnc3+PkbRGUnIdeefcHFPAgBv3AtxMKOG1HXjGzLbFceRqgLhDdjfwQ+DhWHFmWlWdUhgA/h7/\nvCv7hqRj4l9GEjO7s6Bzc85VlQEF1Swzsz9zNGdLo+3OzPNhwlRDxyp3hRtLrF9GKMe+IrYtlfQ3\nSUPEZRmSnpW0NWYeW9XUx0OxfZOkE2Pbo43FyXm5G8r8js65bjGwevvHLKncgEtY6/ZcrIm2V9IF\nsf184FYza2zvvcHMLiBkDrtF0gmxfSEwYmaLgb8w9Qr5c4Qtxbea2TcJa36npLKXtErSiKSR8ckV\n4J1zVWWEm2btHrOkigPuANDI7PVUfA2wxcx2ZeJukfQ68C/C3cTGDrY6R4tEPsHUtbhfJT93wyST\ncimod6bfyTlXFs8W1hlJxwPfBb4hyQi7xQz4EzCWiVtKuDK9xMwOSnoZaDUqVjc5pnOueJ4Pt2PX\nAY+b2almdpqZnQLsAr7VFNcHfBgH23OYvEutJ/YDcD3h5lvWDvJzNzjn5rziktd0Q9UG3AHgD01t\nv+fotELDc8AxkrYD9xOmFRrGgCWS/k24Wl6T/aCZjQM/IuRueB14gdZXx865ucSAer39Y5ZU6srO\nzL6T07YWWNvUdoiQxSevj9yMYGa2MvN8mPzcDc65ua7CUwqVGnArLeV/xXriPvueWlL44R070/r/\nP1QfG2sflNHT291fcuzTtJ8JO5iYj4P075zs8rTVOht3b02Kr305KbyFwrb2doUPuM65+cPAZnGd\nbTs+4Drn5peCdpp1Q9Vumh0h6UuSnoyVdbdK+qekHxTQ71JJfyziHJ1zFeSrFNIolGV4FvirmZ0R\nd5StoGnfsi/ncs5NYlbpVQqVHHAJy7nGzezXjQYze9vM1klaKWlI0mZgk6SFkh6JORFei6V2kFST\n9KCkYUlvSLqx+SAxp8Jrks4s76s557qqwle4Vb1CXAy8Os375wPnmtk+SfcCm83shlgrbYukFwkF\nKA+Y2UWSFgD/kPR8owNJlwLrgOVm9k7zAWJCnFUAvUqrHuCcmy2WXo2jRFUdcCeRtJ6QE2GcUEnz\nhVjbDGAZcLWk2+LrXkKF32XAuZnyxX2EfAvjwNeAQWCZmb2fd8xY32gQoK/WX91ZeOfcUQWmZ+yG\nqg6424BrGy/M7GeS+oFG0bfsgkMB15rZjmwHcR54tZltbGpfCnxAGJjPA3IHXOfcHFXhZWFVncPd\nDPRKuinTdlyL2I3A6jjAIum8TPtNko6N7V+RjswN7Ae+B9wXB2Dn3DxggNWt7aMTkq6UtEPSqKQ7\nct5fIOnp+P4rkk5r12clB1wzM+Aa4NuSdknaAjwG3J4Tfg9wLPBGLHFxT2z/DSFZ+asxr8LDZK7o\nzew/wPeB9ZJmXP7YOVcBVkwCckk1wvTlVcAiYEDSoqawnxKSaJ0FPAQ80K7fqk4pYGYfECs+5Hg0\nE/cJMGUFgoXtJj+Pj6yX44N4s2zxjE/WOVcZBd00WwKMmtlOAElPEYojvJmJWQ7cHZ9vAH4hSfGC\nMVdlB9wq+ai+d8/zHz/2ds5b/cCehK7y46f/+SjmGKnxrbfNz875dDt+Ss2PWT6f3FKopZxP68+0\n/jnNjZ8mN0Krczq1g3Ob1n/5cOOLtqG/g9BeSSOZ14PxRnnDScC7mde7gebfhI/EmNmEpAPACUzz\n9+0DbgfM7MS8dkkjZnZhp/2kxpdxDI/3+LlwTp0ysyu70W9RKjmH65xzs+w9QumuhpNjW25M3PXa\nB+ydrlMfcJ1zbqph4GxJp8fCsyuAoaaYIeDH8fl1hA1Y0y6B8CmFmRlsHzKj+DKO4fEeX/Yx/pdz\nKlWck72ZsLy0BjxiZtskrSFUBR8Cfgs8LmkU2Efrm/xHqM2A7JxzriA+peCccyXxAdc550riA65z\nzpXEB1znnCuJD7jOOVcSH3Cdc64kPuA651xJPgOkwI9Q0LipbgAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    }
  ]
}