assn3_nb.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Dhananjay42/cs6910-assn3/blob/main/assn3_nb.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "90nAoRsb0UqA"
      },
      "source": [
        "##CS6910: Assignment-3\n",
        "-- Submitted by Dhananjay Balakrishnan, ME19B012"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ixsJcIep0eUZ"
      },
      "source": [
        "##Setup and Loading the Dataset"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "cdw6pniAfp28",
        "outputId": "c5433b72-1659-45fc-8f9c-dad216e3916c"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/\n",
            "Collecting python-Levenshtein\n",
            "  Downloading python_Levenshtein-0.21.0-py3-none-any.whl (9.4 kB)\n",
            "Collecting Levenshtein==0.21.0 (from python-Levenshtein)\n",
            "  Downloading Levenshtein-0.21.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (174 kB)\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m174.1/174.1 kB\u001b[0m \u001b[31m8.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25hCollecting rapidfuzz<4.0.0,>=2.3.0 (from Levenshtein==0.21.0->python-Levenshtein)\n",
            "  Downloading rapidfuzz-3.0.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.0 MB)\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m3.0/3.0 MB\u001b[0m \u001b[31m75.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25hInstalling collected packages: rapidfuzz, Levenshtein, python-Levenshtein\n",
            "Successfully installed Levenshtein-0.21.0 python-Levenshtein-0.21.0 rapidfuzz-3.0.0\n"
          ]
        }
      ],
      "source": [
        "!pip install python-Levenshtein"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "AF-X9IyU3Kru"
      },
      "outputs": [],
      "source": [
        "import csv\n",
        "from __future__ import unicode_literals, print_function, division\n",
        "from io import open\n",
        "import unicodedata\n",
        "import string\n",
        "import re\n",
        "import random\n",
        "\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "from torch import optim\n",
        "import torch.nn.functional as F\n",
        "\n",
        "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
        "import Levenshtein\n",
        "import math\n",
        "import pickle\n",
        "import numpy as np\n",
        "import copy\n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns\n",
        "import random"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "UVnNtzrR0pUt",
        "outputId": "e9bad216-15e8-4334-f37c-01f6b81876d6"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /content/gdrive/\n"
          ]
        }
      ],
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/gdrive/')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "QUrlh1ep0bLk"
      },
      "outputs": [],
      "source": [
        "data_dir = '/content/gdrive/MyDrive/CS6910_A3/aksharantar_sampled/tam/'"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "i2Dr1QDU0qGD"
      },
      "outputs": [],
      "source": [
        "def obtain_data(dir):\n",
        "  x = []\n",
        "  y = []\n",
        "\n",
        "  with open(dir, 'r') as file:\n",
        "    reader = csv.DictReader(file, fieldnames=['x', 'y'])\n",
        "  \n",
        "    for row in reader:\n",
        "      x.append(row['x'])\n",
        "      y.append(row['y'])\n",
        "    \n",
        "  return x, y"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "faHyeHRY3lVl"
      },
      "outputs": [],
      "source": [
        "x_train, y_train = obtain_data(data_dir + 'tam_train.csv')\n",
        "x_test, y_test = obtain_data(data_dir + 'tam_test.csv')\n",
        "x_val, y_val = obtain_data(data_dir + 'tam_valid.csv')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "3ya5oC143xZ8",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "dba59296-cdf7-4e47-fd7b-842a7958887e"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "thottacharya தொட்டாச்சார்ய\n",
            "menmaithaan மென்மைதான்\n"
          ]
        }
      ],
      "source": [
        "for i in range(0, 2):\n",
        "  print(x_train[i], y_train[i])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "0s9Mght5EB7b"
      },
      "outputs": [],
      "source": [
        "start_token = 0\n",
        "end_token = 1"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "KItsHUzs9zPP"
      },
      "outputs": [],
      "source": [
        "#First we define a Language class, which will be used to easily load our dataset.\n",
        "class Language:\n",
        "  def __init__(self, name):\n",
        "    self.name = name\n",
        "    self.char2index = {}\n",
        "    self.index2char = {0: \"SOS\", 1: \"EOS\", 2: \"unknown\"} #mapping from index to character\n",
        "    self.n_chars = 3  # Count SOS, EOS, and unknown. \n",
        "    self.max_size = 2 #to find the maximum length of the dataset we're training our model on\n",
        "\n",
        "  def update_vocab(self, x):\n",
        "    for word in x:\n",
        "      if len(word) + 2 > self.max_size:\n",
        "        self.max_size = len(word) + 2\n",
        "\n",
        "      for letter in word:\n",
        "        if letter not in self.char2index.keys():\n",
        "          self.char2index[letter] = self.n_chars\n",
        "          self.index2char[self.n_chars] = letter\n",
        "          self.n_chars = self.n_chars + 1\n",
        "  \n",
        "  def get_index(self, character):\n",
        "    if character in self.char2index.keys():\n",
        "      return self.char2index[character]\n",
        "    else:\n",
        "      return 2\n",
        "  \n",
        "  def get_character(self, index):\n",
        "    if index == 2:\n",
        "      return '$'\n",
        "    else:\n",
        "      return self.index2char[index]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Dt5eia4S_tLQ"
      },
      "outputs": [],
      "source": [
        "english = Language('eng')\n",
        "tamil = Language('tam')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "OQBYGHJl_2fQ"
      },
      "outputs": [],
      "source": [
        "english.update_vocab(x_train)\n",
        "tamil.update_vocab(y_train)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "yCsWlmkwAlwC"
      },
      "outputs": [],
      "source": [
        "def encoded_word(language, word):\n",
        "  coded = [language.get_index(letter) for letter in word]\n",
        "  coded.append(end_token)\n",
        "  return coded\n",
        "\n",
        "def get_pairs(lang1, lang2, inputs, targets):\n",
        "  return [(torch.tensor(encoded_word(lang1, x), dtype=torch.long, device=device).view(-1, 1), torch.tensor(encoded_word(lang2, y), dtype=torch.long, device=device).view(-1, 1)) \n",
        "  for (x,y) in zip(inputs,targets)]\n",
        "\n",
        "def decoded_word(language, encoded_word):\n",
        "  if encoded_word[-1] == end_token:\n",
        "    encoded_word = encoded_word[:-1]\n",
        "\n",
        "  characters = [language.get_character(num) for num in encoded_word]\n",
        "  decoded = ''.join(characters)\n",
        "  return decoded"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "4VDupcI8m3aS"
      },
      "outputs": [],
      "source": [
        "train_data = get_pairs(english, tamil, x_train, y_train)\n",
        "test_data = get_pairs(english, tamil, x_test, y_test)\n",
        "val_data = get_pairs(english, tamil, x_val, y_val)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "63vUwSCMWDMn"
      },
      "source": [
        "#RNN without Attention"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "0zU4N_wh70Wc"
      },
      "outputs": [],
      "source": [
        "class Encoder(nn.Module):\n",
        "  def __init__(self, inp_vocab_size, embedding_size, n_layers, hl_size, dropout, cell_type, bidirectional):\n",
        "    super(Encoder, self).__init__()\n",
        "    self.vocab_size = inp_vocab_size\n",
        "    self.embedding_size = embedding_size\n",
        "    self.n_layers = n_layers\n",
        "    self.hl_size = hl_size\n",
        "    self.bidirectional = bidirectional\n",
        "    self.cell_type = cell_type\n",
        "    self.dropout = dropout\n",
        "\n",
        "    if cell_type == 'RNN':\n",
        "      self.cell = nn.RNN(self.embedding_size, self.hl_size, num_layers = self.n_layers, dropout = self.dropout, bidirectional = self.bidirectional).to(device)\n",
        "    elif cell_type == 'GRU':\n",
        "      self.cell = nn.GRU(self.embedding_size, self.hl_size, num_layers = self.n_layers, dropout = self.dropout, bidirectional = self.bidirectional).to(device)\n",
        "    elif cell_type == 'LSTM':\n",
        "      self.cell = nn.LSTM(self.embedding_size, self.hl_size, num_layers = self.n_layers, dropout = self.dropout, bidirectional = self.bidirectional).to(device)\n",
        "    else:\n",
        "      print('Wrong Cell Type.')\n",
        "      exit()\n",
        "    self.embedding_layer = nn.Embedding(self.vocab_size, self.embedding_size).to(device)\n",
        "  \n",
        "  def forward(self, input, hidden, c = 0):\n",
        "    embedded = self.embedding_layer(input).view(1, 1, -1)\n",
        "    if self.cell_type != 'LSTM':\n",
        "      output, hidden = self.cell(embedded, hidden)\n",
        "      \n",
        "      return output, hidden\n",
        "\n",
        "    else:\n",
        "      output, (hidden, c) = self.cell(embedded, (hidden, c))\n",
        "\n",
        "      return output, hidden, c\n",
        "  \n",
        "  def init_hidden(self):\n",
        "    if self.bidirectional:\n",
        "      return torch.zeros(self.n_layers*2, 1, self.hl_size, device = device)\n",
        "    else:\n",
        "      return torch.zeros(self.n_layers, 1, self.hl_size, device = device)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "FU3vbDoy1cm8"
      },
      "outputs": [],
      "source": [
        "class DecoderVanilla(nn.Module):\n",
        "  def __init__(self, out_vocab_size, embedding_size, n_layers, hl_size, dropout, cell_type, bidirectional):\n",
        "    super(DecoderVanilla, self).__init__()\n",
        "    self.vocab_size = out_vocab_size\n",
        "    self.embedding_size = embedding_size\n",
        "    self.n_layers = n_layers\n",
        "    self.hl_size = hl_size\n",
        "    self.softmax = nn.LogSoftmax(dim=1)\n",
        "    self.cell_type = cell_type\n",
        "    self.bidirectional = bidirectional\n",
        "    self.dropout = dropout\n",
        "\n",
        "    if self.bidirectional:\n",
        "      self.linear = nn.Linear(2*self.hl_size, self.vocab_size).to(device)\n",
        "    else:\n",
        "      self.linear = nn.Linear(self.hl_size, self.vocab_size).to(device)\n",
        "\n",
        "    if cell_type == 'RNN':\n",
        "      self.cell = nn.RNN(self.embedding_size, self.hl_size, num_layers = self.n_layers, dropout = self.dropout, bidirectional = self.bidirectional).to(device)\n",
        "    elif cell_type == 'GRU':\n",
        "      self.cell = nn.GRU(self.embedding_size, self.hl_size, num_layers = self.n_layers, dropout = self.dropout, bidirectional = self.bidirectional).to(device)\n",
        "    elif cell_type == 'LSTM':\n",
        "      self.cell = nn.LSTM(self.embedding_size, self.hl_size, num_layers = self.n_layers, dropout = self.dropout, bidirectional = self.bidirectional).to(device)\n",
        "    else:\n",
        "      print('Wrong Cell Type.')\n",
        "      exit()\n",
        "    \n",
        "    self.embedding_layer = nn.Embedding(self.vocab_size, self.embedding_size).to(device)\n",
        "  \n",
        "  def forward(self, input, hidden, c = 0):\n",
        "    embedded = self.embedding_layer(input).view(1, 1, -1)\n",
        "    output = F.relu(embedded)\n",
        "\n",
        "    if self.cell_type != 'LSTM':\n",
        "      output, hidden = self.cell(output, hidden)\n",
        "      output = self.linear(output[0])\n",
        "      output = self.softmax(output)\n",
        "      return output, hidden\n",
        "    else:\n",
        "      output, (hidden, c) = self.cell(output, (hidden, c))\n",
        "      output = self.linear(output[0])\n",
        "      output = self.softmax(output)\n",
        "      return output, hidden, c"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "class seq2seq_vanilla():\n",
        "  def __init__(self, inp_language, out_language, embedding_size, n_layers, hl_size, dropout = 0.2, cell_type = 'LSTM', lr = 0.01, teacher_forcing_ratio = 0.5,bidirectional_flag = False):\n",
        "    self.encoder = Encoder(inp_language.n_chars, embedding_size, n_layers, hl_size, dropout, cell_type, bidirectional = bidirectional_flag)\n",
        "    self.decoder = DecoderVanilla(out_language.n_chars, embedding_size, n_layers, hl_size, dropout, cell_type, bidirectional = bidirectional_flag)\n",
        "    self.lr = lr\n",
        "    self.teacher_forcing = teacher_forcing_ratio\n",
        "    self.max_length = out_language.max_size\n",
        "    self.cell_type = cell_type\n",
        "\n",
        "    self.encoder_optimizer = optim.SGD(self.encoder.parameters(), lr=self.lr)\n",
        "    self.decoder_optimizer = optim.SGD(self.decoder.parameters(), lr = self.lr)\n",
        "\n",
        "    self.loss_fn = nn.NLLLoss()\n",
        "\n",
        "  def train_step(self, input, target):\n",
        "    encoder_hidden = self.encoder.init_hidden()\n",
        "    encoder_c = self.encoder.init_hidden()\n",
        "\n",
        "    self.encoder_optimizer.zero_grad()\n",
        "    self.decoder_optimizer.zero_grad()\n",
        "\n",
        "    input_length = input.size(0)\n",
        "    target_length = target.size(0)\n",
        "\n",
        "    loss = 0\n",
        "\n",
        "    for i in range(0, input_length):\n",
        "      if self.cell_type != 'LSTM':\n",
        "        encoder_output, encoder_hidden = self.encoder.forward(input[i], encoder_hidden)\n",
        "      else:\n",
        "        encoder_output, encoder_hidden, encoder_c = self.encoder.forward(input[i], encoder_hidden, encoder_c)\n",
        "    \n",
        "    decoder_input = torch.tensor([[start_token]], device=device)\n",
        "\n",
        "    decoder_hidden = encoder_hidden\n",
        "    decoder_c = encoder_c\n",
        "\n",
        "    num = random.random()\n",
        "\n",
        "    if num < self.teacher_forcing:\n",
        "      #here, we use teacher forcing. \n",
        "      for j in range(0, target_length):\n",
        "        if self.cell_type != 'LSTM':\n",
        "          decoder_output, decoder_hidden = self.decoder.forward(decoder_input, decoder_hidden)\n",
        "        else:\n",
        "          decoder_output, decoder_hidden, decoder_c = self.decoder.forward(decoder_input, decoder_hidden, decoder_c)\n",
        "\n",
        "        loss = loss + self.loss_fn(decoder_output, target[j])\n",
        "        decoder_input = target[j]#.unsqueeze(0)\n",
        "\n",
        "    else:\n",
        "      #here, there is no teacher forcing. the predictions themselves are used. \n",
        "      for j in range(0, target_length):\n",
        "        if self.cell_type != 'LSTM':\n",
        "          decoder_output, decoder_hidden = self.decoder.forward(decoder_input, decoder_hidden)\n",
        "        else:\n",
        "          decoder_output, decoder_hidden, decoder_c = self.decoder.forward(decoder_input, decoder_hidden, decoder_c)\n",
        "\n",
        "        loss = loss + self.loss_fn(decoder_output, target[j])\n",
        "        value, index = decoder_output.topk(1)\n",
        "        decoder_input = index.squeeze().detach()\n",
        "        if decoder_input.item() == end_token:\n",
        "          break\n",
        "  \n",
        "    loss.backward()\n",
        "    self.encoder_optimizer.step()\n",
        "    self.decoder_optimizer.step()\n",
        "\n",
        "    return loss.item()/target_length\n",
        "  \n",
        "  def predict(self, input, target):\n",
        "    #here, we use the model to inference. \n",
        "    with torch.no_grad():\n",
        "      encoder_hidden = self.encoder.init_hidden()\n",
        "      encoder_c = self.encoder.init_hidden()\n",
        "\n",
        "      input_length = input.size(0)\n",
        "      for i in range(0, input_length):\n",
        "        if self.cell_type != 'LSTM':\n",
        "          encoder_output, encoder_hidden = self.encoder.forward(input[i], encoder_hidden)\n",
        "        else:\n",
        "          encoder_output, encoder_hidden, encoder_c = self.encoder.forward(input[i], encoder_hidden, encoder_c)\n",
        "\n",
        "      decoder_input = torch.tensor([[start_token]], device=device)\n",
        "      decoder_hidden = encoder_hidden\n",
        "      decoder_c = encoder_c\n",
        "\n",
        "      outputs = []\n",
        "      for i in range(0, self.max_length):\n",
        "        if self.cell_type != 'LSTM':\n",
        "          decoder_output, decoder_hidden = self.decoder.forward(decoder_input, decoder_hidden)\n",
        "        else:\n",
        "          decoder_output, decoder_hidden, decoder_c = self.decoder.forward(decoder_input, decoder_hidden, decoder_c)\n",
        "\n",
        "        value, index = decoder_output.data.topk(1)\n",
        "        decoder_input = index.squeeze().detach()\n",
        "        outputs.append(decoder_input.item())\n",
        "        if decoder_input.item() == end_token:\n",
        "          break\n",
        "\n",
        "      return outputs\n",
        "  \n",
        "  def predict_beam(self, input, beam_size):\n",
        "    #this is used to predict b outputs, where b is the beam size. \n",
        "    with torch.no_grad():\n",
        "      encoder_hidden = self.encoder.init_hidden()\n",
        "      encoder_c = self.encoder.init_hidden()\n",
        "\n",
        "      input_length = input.size(0)\n",
        "      for i in range(0, input_length):\n",
        "        if self.cell_type != 'LSTM':\n",
        "          encoder_output, encoder_hidden = self.encoder.forward(input[i], encoder_hidden)\n",
        "        else:\n",
        "          encoder_output, encoder_hidden, encoder_c = self.encoder.forward(input[i], encoder_hidden, encoder_c)\n",
        "\n",
        "      decoder_input = torch.tensor([[start_token]], device=device)\n",
        "      decoder_hidden = encoder_hidden\n",
        "      decoder_c = encoder_c\n",
        "\n",
        "      possible_outputs = [] #a list of lists containing the top \"beam_size\" set of the best possible outputs at an instance\n",
        "      next_inputs = [] #list containing the tensors that are to be fed as the next input\n",
        "      decoder_hiddens = [] #list containing the hidden values corresponding to the given outputs\n",
        "      decoder_cs = [] #list containing the cell states (in case we use an LSTM)\n",
        "\n",
        "      if self.cell_type != 'LSTM':\n",
        "        decoder_output, decoder_hidden = self.decoder.forward(decoder_input, decoder_hidden)\n",
        "      else:\n",
        "        decoder_output, decoder_hidden, decoder_c = self.decoder.forward(decoder_input, decoder_hidden, decoder_c)\n",
        "        \n",
        "      values, indices = decoder_output.data.topk(beam_size) #we unpack the top b values and their corresponding indices\n",
        "      values = torch.exp(values) #we would like to deal with a probability instead of a log-probability, so we do e^val\n",
        "\n",
        "      for j in range(0, beam_size):\n",
        "        #we do on iteration and populate the above lists. we will then use these lists to iteratively get our best b guesses. \n",
        "        value = values[0, j]\n",
        "        index = indices[0, j]\n",
        "        possible_outputs.append([index.item()])\n",
        "        next_inputs.append(index.squeeze().detach())\n",
        "        decoder_hiddens.append(decoder_hidden)\n",
        "        if self.cell_type == 'LSTM':\n",
        "          decoder_cs.append(decoder_c)\n",
        "                    \n",
        "      for k in range(1, self.max_length):\n",
        "        #now, we make predictions for the other timesteps\n",
        "        temp_probabilities = np.zeros([beam_size, beam_size])\n",
        "        temp_indices = []\n",
        "        temp_hiddens = []\n",
        "        temp_cs = []\n",
        "\n",
        "        for i in range(0, beam_size):\n",
        "          #we first iterate through all possible current inputs and their corresponding hidden states\n",
        "          curr_input = next_inputs[i]\n",
        "          hidden = decoder_hiddens[i]\n",
        "          if self.cell_type == 'LSTM':\n",
        "            c = decoder_cs[i]\n",
        "          \n",
        "          if self.cell_type != 'LSTM':\n",
        "            decoder_output, decoder_hidden = self.decoder.forward(curr_input, hidden)\n",
        "          else:\n",
        "            decoder_output, decoder_hidden, decoder_c = self.decoder.forward(curr_input, hidden, c)\n",
        "            temp_cs.append(decoder_c)\n",
        "            \n",
        "          #following this, we get the top b values for each of these inputs\n",
        "          temp_hiddens.append(decoder_hidden)\n",
        "          values, indices = decoder_output.data.topk(beam_size)\n",
        "          values = torch.exp(values)\n",
        "\n",
        "          temp_temp_indices = []\n",
        "          for j in range(0, beam_size):\n",
        "            #we now iterate through these b predictions correponding to a given input, and construct a probability table\n",
        "            #this probability table gives the conditional probability of the output, given the input sequence. \n",
        "            value = values[0, j]\n",
        "            index = indices[0, j]\n",
        "            temp_probabilities[i][j] = value.item()\n",
        "            temp_temp_indices.append(index.squeeze().detach())\n",
        "\n",
        "            if curr_input.item() == end_token:\n",
        "              temp_temp_indices.append(torch.tensor([[end_token]], device= device))\n",
        "          \n",
        "          temp_indices.append(temp_temp_indices)\n",
        "        \n",
        "        #we then choose the best b probabilities from these\n",
        "        mat = np.array(temp_probabilities)\n",
        "        idx = np.argpartition(mat, mat.size - beam_size, axis=None)[-beam_size:]\n",
        "        results = np.column_stack(np.unravel_index(idx, mat.shape)) \n",
        "\n",
        "        updated_outputs = []\n",
        "\n",
        "        decoder_hiddens = []\n",
        "        decoder_cs = []\n",
        "        next_inputs = []\n",
        "\n",
        "        for i, result in enumerate(results):\n",
        "          #we finally iterate through these best sequences and update our next_inputs, hiddens, and our possible_outputs. \n",
        "          x = result[0]\n",
        "          y = result[1]\n",
        "          next_inputs.append(temp_indices[x][y])\n",
        "          decoder_hiddens.append(temp_hiddens[x])\n",
        "          if self.cell_type == 'LSTM':\n",
        "            decoder_cs.append(temp_cs[x])\n",
        "\n",
        "          arr = copy.deepcopy(possible_outputs[x])\n",
        "          if arr[-1] == end_token:\n",
        "            pass\n",
        "          else:\n",
        "            arr.append(temp_indices[x][y].item())\n",
        "\n",
        "          updated_outputs.append(arr)\n",
        "        \n",
        "        possible_outputs = copy.deepcopy(updated_outputs)\n",
        "\n",
        "      #we finally return a set of predictions instead of one prediction.             \n",
        "      return possible_outputs\n",
        "\n",
        "  def evaluate_beam(self, data, beam_size, print_flag = True):\n",
        "    #same as the evaluate fn, except we check for the multiple outputs generated \n",
        "    #by the beam based predict method, and choose the best one to get a good performance. \n",
        "    correct = 0\n",
        "    character_wise = 0\n",
        "    count = 0\n",
        "    total_distance = 0\n",
        "\n",
        "    for pair in data:\n",
        "      input = pair[0]\n",
        "      target = pair[1]\n",
        "      predictions = self.predict_beam(input, beam_size)\n",
        "      target = target.tolist()\n",
        "      target = [t[0] for t in target]\n",
        "      tar_word = decoded_word(tamil,target)\n",
        "\n",
        "      mini = 2\n",
        "\n",
        "      for pred in predictions:\n",
        "        pred_word = decoded_word(tamil,pred)\n",
        "        if pred_word == tar_word:\n",
        "          correct = correct + 1\n",
        "        \n",
        "        dist = min((Levenshtein.distance(pred_word, tar_word)/max(len(tar_word),len(pred_word))), 1)\n",
        "        if dist < mini:\n",
        "          mini = dist\n",
        "      \n",
        "      if print_flag:\n",
        "        if count%500 == 0:\n",
        "          print([decoded_word(tamil, pred) for pred in predictions], tar_word)\n",
        "        count = count + 1\n",
        "      \n",
        "      total_distance = total_distance + mini\n",
        "    \n",
        "    avg_distance = total_distance/len(data)\n",
        "    char_acc = 1 - avg_distance\n",
        "    acc = correct/len(data)\n",
        "    return acc, char_acc\n",
        "\n",
        "\n",
        "  def evaluate(self, data, print_flag = True):\n",
        "    correct = 0\n",
        "    character_wise = 0\n",
        "    count = 0\n",
        "    total_distance = 0\n",
        "\n",
        "    for pair in data:\n",
        "      input = pair[0]\n",
        "      target = pair[1]\n",
        "      pred = self.predict(input, target)\n",
        "      target = target.tolist()\n",
        "      target = [t[0] for t in target]\n",
        "\n",
        "      if print_flag:\n",
        "        if count%500 == 0:\n",
        "          print(decoded_word(tamil,pred), decoded_word(tamil,target))\n",
        "        count = count + 1\n",
        "\n",
        "      pred_word = decoded_word(tamil,pred)\n",
        "      tar_word = decoded_word(tamil,target)\n",
        "      \n",
        "      if pred_word == tar_word:\n",
        "        correct = correct + 1\n",
        "      \n",
        "      total_distance = total_distance + min((Levenshtein.distance(pred_word, tar_word)/max(len(tar_word),len(pred_word))), 1)\n",
        "    \n",
        "    avg_distance = total_distance/len(data)\n",
        "    char_acc = 1 - avg_distance\n",
        "    acc = correct/len(data)\n",
        "\n",
        "    return acc, char_acc"
      ],
      "metadata": {
        "id": "w1nrOtq609CU"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "-UWaDWHPnXPN"
      },
      "outputs": [],
      "source": [
        "model = seq2seq_vanilla(inp_language = english, out_language = tamil, embedding_size = 64, n_layers = 2, hl_size = 128)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "uKHso_WvDWjg"
      },
      "outputs": [],
      "source": [
        "n_iters = 75000"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "asm1dpLKgJTi",
        "outputId": "282b649f-cddd-4648-9b4d-473dd1a1d71e"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "4096\n"
          ]
        }
      ],
      "source": [
        "print(len(val_data))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "-_nDKVgTNmnE"
      },
      "outputs": [],
      "source": [
        "training_pairs = [random.choice(train_data) for i in range(0, n_iters)]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "8yiauv0nN19Q"
      },
      "outputs": [],
      "source": [
        "# train_loss = 0\n",
        "# for i in range(0, n_iters):\n",
        "#   training_pair = training_pairs[i]\n",
        "#   x = training_pair[0]\n",
        "#   y = training_pair[1]\n",
        "#   loss = model.train_step(x, y)\n",
        "#   train_loss = train_loss + loss\n",
        "\n",
        "#   if (i+1)%5000 == 0:\n",
        "#     print('------------------------------------------------')\n",
        "#     print('train loss is:', train_loss/5000)\n",
        "#     test_acc, char_acc = model.evaluate(val_data)\n",
        "#     print(f'test accuracy is {test_acc} and character-wise accuracy is {char_acc}')\n",
        "#     train_loss = 0"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#RNN with Attention"
      ],
      "metadata": {
        "id": "NzzXCox-cqpB"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "class DecoderAttn(nn.Module):\n",
        "  def __init__(self, out_vocab_size, max_length, embedding_size, hl_size, dropout, cell_type, bidirectional):\n",
        "    super(DecoderAttn, self).__init__()\n",
        "    self.vocab_size = out_vocab_size\n",
        "    self.embedding_size = embedding_size\n",
        "    self.hl_size = hl_size\n",
        "    self.softmax = nn.LogSoftmax(dim=1)\n",
        "    self.cell_type = cell_type\n",
        "    self.bidirectional = bidirectional\n",
        "    self.dropout = dropout\n",
        "    self.max_length = max_length\n",
        "\n",
        "    self.dropout_layer = nn.Dropout(self.dropout).to(device)\n",
        "\n",
        "\n",
        "    if self.bidirectional:\n",
        "      self.linear = nn.Linear(2*self.hl_size, self.vocab_size).to(device)\n",
        "      self.attention = nn.Linear(2*self.hl_size + self.embedding_size, self.max_length).to(device)\n",
        "      self.attn_combine = nn.Linear(2*self.hl_size + self.embedding_size, self.embedding_size).to(device)\n",
        "\n",
        "    else:\n",
        "      self.linear = nn.Linear(self.hl_size, self.vocab_size).to(device)\n",
        "      self.attention = nn.Linear(self.hl_size + self.embedding_size, self.max_length).to(device)\n",
        "      self.attn_combine = nn.Linear(self.hl_size + self.embedding_size, self.embedding_size).to(device)\n",
        "\n",
        "\n",
        "    if cell_type == 'RNN':\n",
        "      self.cell = nn.RNN(self.embedding_size, self.hl_size, num_layers = 1, bidirectional = self.bidirectional).to(device)\n",
        "    elif cell_type == 'GRU':\n",
        "      self.cell = nn.GRU(self.embedding_size, self.hl_size, num_layers = 1, bidirectional = self.bidirectional).to(device)\n",
        "    elif cell_type == 'LSTM':\n",
        "      self.cell = nn.LSTM(self.embedding_size, self.hl_size, num_layers = 1, bidirectional = self.bidirectional).to(device)\n",
        "    else:\n",
        "      print('Wrong Cell Type.')\n",
        "      exit()\n",
        "    \n",
        "    self.embedding_layer = nn.Embedding(self.vocab_size, self.embedding_size).to(device)\n",
        "  \n",
        "  def forward(self, input, hidden, encoder_outputs, c = 0):\n",
        "    embedded = self.embedding_layer(input).view(1, 1, -1)\n",
        "    embedded = self.dropout_layer(embedded)\n",
        "\n",
        "    if self.bidirectional:\n",
        "      concatenated = torch.cat((embedded[0], hidden[0], hidden[1]), 1)\n",
        "    else:\n",
        "      concatenated = torch.cat((embedded[0], hidden[0]), 1)\n",
        "    #print(embedded.shape, hidden.shape, concatenated.shape)\n",
        "    #print(embedded.shape, hidden.shape, concatenated.shape)\n",
        "\n",
        "    attn_weights = F.softmax(self.attention(concatenated), dim=1)\n",
        "    attn_applied = torch.bmm(attn_weights.unsqueeze(0), encoder_outputs.unsqueeze(0))\n",
        "\n",
        "    output = torch.cat((embedded[0], attn_applied[0]), 1)\n",
        "    output = self.attn_combine(output).unsqueeze(0)\n",
        "\n",
        "    if self.cell_type != 'LSTM':\n",
        "      output, hidden = self.cell(output, hidden)\n",
        "      output = self.linear(output[0])\n",
        "      output = self.softmax(output)\n",
        "      return output, hidden, attn_weights\n",
        "    else:\n",
        "      output, (hidden, c) = self.cell(output, (hidden, c))\n",
        "      output = self.linear(output[0])\n",
        "      output = self.softmax(output)\n",
        "      return output, hidden, c, attn_weights"
      ],
      "metadata": {
        "id": "45CGDuE99oVG"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class seq2seq_attn():\n",
        "  def __init__(self, inp_language, out_language, embedding_size, hl_size, dropout = 0.2, cell_type = 'LSTM', lr = 0.005, teacher_forcing_ratio = 0.5,bidirectional_flag = True):\n",
        "    self.max_length = max(out_language.max_size,  inp_language.max_size)\n",
        "    self.encoder = Encoder(inp_language.n_chars, embedding_size, 1, hl_size, dropout, cell_type, bidirectional = bidirectional_flag)\n",
        "    self.decoder = DecoderAttn(out_language.n_chars, self.max_length, embedding_size, hl_size, dropout, cell_type, bidirectional = bidirectional_flag)\n",
        "    self.lr = lr\n",
        "    self.teacher_forcing = teacher_forcing_ratio\n",
        "    self.cell_type = cell_type\n",
        "    self.bidir = bidirectional_flag\n",
        "\n",
        "    self.encoder_optimizer = optim.SGD(self.encoder.parameters(), lr=self.lr)\n",
        "    self.decoder_optimizer = optim.SGD(self.decoder.parameters(), lr = self.lr)\n",
        "\n",
        "    self.loss_fn = nn.NLLLoss()\n",
        "    #decay^30000 = 0.2\n",
        "    self.decay_rate = 2**(math.log2(0.2)/30000)\n",
        "    #self.decay = lambda epoch: self.decay_rate\n",
        "    self.encoder_scheduler = torch.optim.lr_scheduler.MultiplicativeLR(self.encoder_optimizer, lr_lambda=self.decay)\n",
        "    self.decoder_scheduler = torch.optim.lr_scheduler.MultiplicativeLR(self.decoder_optimizer, lr_lambda=self.decay)\n",
        "  \n",
        "  def decay(self, epoch):\n",
        "    return self.decay_rate\n",
        "\n",
        "  def train_step(self, input, target):\n",
        "    encoder_hidden = self.encoder.init_hidden()\n",
        "    encoder_c = self.encoder.init_hidden()\n",
        "\n",
        "    self.encoder_optimizer.zero_grad()\n",
        "    self.decoder_optimizer.zero_grad()\n",
        "\n",
        "    input_length = input.size(0)\n",
        "    target_length = target.size(0)\n",
        "\n",
        "    loss = 0\n",
        "\n",
        "    if self.bidir == True:\n",
        "      encoder_outputs = torch.zeros(self.max_length, 2*self.encoder.hl_size, device=device)\n",
        "    else:\n",
        "      encoder_outputs = torch.zeros(self.max_length, self.encoder.hl_size, device=device)\n",
        "\n",
        "    for i in range(0, input_length):\n",
        "      if self.cell_type != 'LSTM':\n",
        "        encoder_output, encoder_hidden = self.encoder.forward(input[i], encoder_hidden)\n",
        "        encoder_outputs[i] = encoder_output[0, 0]\n",
        "      else:\n",
        "        encoder_output, encoder_hidden, encoder_c = self.encoder.forward(input[i], encoder_hidden, encoder_c)\n",
        "        encoder_outputs[i] = encoder_output[0, 0]\n",
        "    \n",
        "    decoder_input = torch.tensor([[start_token]], device=device)\n",
        "\n",
        "    decoder_hidden = encoder_hidden\n",
        "    decoder_c = encoder_c\n",
        "\n",
        "    num = random.random()\n",
        "\n",
        "    if num < self.teacher_forcing:\n",
        "      #here, we use teacher forcing. \n",
        "      for j in range(0, target_length):\n",
        "        if self.cell_type != 'LSTM':\n",
        "          decoder_output, decoder_hidden, decoder_attention = self.decoder.forward(decoder_input, decoder_hidden, encoder_outputs)\n",
        "        else:\n",
        "          decoder_output, decoder_hidden, decoder_c, decoder_attention = self.decoder.forward(decoder_input, decoder_hidden, encoder_outputs, decoder_c)\n",
        "\n",
        "        loss = loss + self.loss_fn(decoder_output, target[j])\n",
        "        decoder_input = target[j]#.unsqueeze(0)\n",
        "\n",
        "    else:\n",
        "      #here, there is no teacher forcing. the predictions themselves are used. \n",
        "      for j in range(0, target_length):\n",
        "        if self.cell_type != 'LSTM':\n",
        "          decoder_output, decoder_hidden, decoder_attention = self.decoder.forward(decoder_input, decoder_hidden, encoder_outputs)\n",
        "        else:\n",
        "          decoder_output, decoder_hidden, decoder_c, decoder_attention = self.decoder.forward(decoder_input, decoder_hidden, encoder_outputs, decoder_c)\n",
        "\n",
        "        loss = loss + self.loss_fn(decoder_output, target[j])\n",
        "        value, index = decoder_output.topk(1)\n",
        "        decoder_input = index.squeeze().detach()\n",
        "        if decoder_input.item() == end_token:\n",
        "          break\n",
        "      \n",
        "            \n",
        "    loss.backward()\n",
        "    self.encoder_optimizer.step()\n",
        "    self.decoder_optimizer.step()\n",
        "\n",
        "    self.encoder_scheduler.step()\n",
        "    self.decoder_scheduler.step()\n",
        "\n",
        "    return loss.item()/target_length\n",
        "  \n",
        "  def predict(self, input, target):\n",
        "    with torch.no_grad():\n",
        "      encoder_hidden = self.encoder.init_hidden()\n",
        "      encoder_c = self.encoder.init_hidden()\n",
        "\n",
        "      if self.bidir == True:\n",
        "        encoder_outputs = torch.zeros(self.max_length, 2*self.encoder.hl_size, device=device)\n",
        "      else:\n",
        "        encoder_outputs = torch.zeros(self.max_length, self.encoder.hl_size, device=device)\n",
        "\n",
        "      input_length = input.size(0)\n",
        "      for i in range(0, min(input_length, self.max_length)):\n",
        "        if self.cell_type != 'LSTM':\n",
        "          encoder_output, encoder_hidden = self.encoder.forward(input[i], encoder_hidden)\n",
        "          encoder_outputs[i] = encoder_output[0, 0]\n",
        "        else:\n",
        "          encoder_output, encoder_hidden, encoder_c = self.encoder.forward(input[i], encoder_hidden, encoder_c)\n",
        "          encoder_outputs[i] = encoder_output[0, 0]\n",
        "\n",
        "      decoder_input = torch.tensor([[start_token]], device=device)\n",
        "      decoder_hidden = encoder_hidden\n",
        "      decoder_c = encoder_c\n",
        "\n",
        "      outputs = []\n",
        "      for i in range(0, self.max_length):\n",
        "        if self.cell_type != 'LSTM':\n",
        "          decoder_output, decoder_hidden, decoder_attention = self.decoder.forward(decoder_input, decoder_hidden, encoder_outputs)\n",
        "        else:\n",
        "          decoder_output, decoder_hidden, decoder_c, decoder_attention = self.decoder.forward(decoder_input, decoder_hidden, encoder_outputs, decoder_c)\n",
        "\n",
        "        value, index = decoder_output.data.topk(1)\n",
        "        decoder_input = index.squeeze().detach()\n",
        "        outputs.append(decoder_input.item())\n",
        "        if decoder_input.item() == end_token:\n",
        "          break\n",
        "\n",
        "      return outputs\n",
        "\n",
        "\n",
        "  def evaluate(self, data):\n",
        "    correct = 0\n",
        "    character_wise = 0\n",
        "    count = 0\n",
        "    total_distance = 0\n",
        "\n",
        "    for pair in data:\n",
        "      input = pair[0]\n",
        "      target = pair[1]\n",
        "      pred = self.predict(input, target)\n",
        "      target = target.tolist()\n",
        "      target = [t[0] for t in target]\n",
        "\n",
        "      # if count%500 == 0:\n",
        "      #   print(decoded_word(tamil,pred), decoded_word(tamil,target))\n",
        "      # count = count + 1\n",
        "\n",
        "      pred_word = decoded_word(tamil,pred)\n",
        "      tar_word = decoded_word(tamil,target)\n",
        "      \n",
        "      if pred_word == tar_word:\n",
        "        correct = correct + 1\n",
        "      \n",
        "      total_distance = total_distance + min((Levenshtein.distance(pred_word, tar_word)/max(len(tar_word),len(pred_word))), 1)\n",
        "    \n",
        "    avg_distance = total_distance/len(data)\n",
        "    char_acc = 1 - avg_distance\n",
        "    acc = correct/len(data)\n",
        "\n",
        "\n",
        "    return acc, char_acc"
      ],
      "metadata": {
        "id": "YkdMjiL8B5Kx"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model = seq2seq_attn(inp_language = english, out_language = tamil, embedding_size = 128, hl_size = 128)"
      ],
      "metadata": {
        "id": "3lD7KsLN_FCF",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d19e229b-fa3c-4739-95f0-8801da22a761"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/rnn.py:71: UserWarning: dropout option adds dropout after all but last recurrent layer, so non-zero dropout expects num_layers greater than 1, but got dropout=0.2 and num_layers=1\n",
            "  warnings.warn(\"dropout option adds dropout after all but last \"\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# train_loss = 0\n",
        "# for i in range(0, n_iters):\n",
        "#   training_pair = training_pairs[i]\n",
        "#   x = training_pair[0]\n",
        "#   y = training_pair[1]\n",
        "#   loss = model.train_step(x, y)\n",
        "#   train_loss = train_loss + loss\n",
        "\n",
        "#   if (i+1)%5000 == 0:\n",
        "#     print('------------------------------------------------')\n",
        "#     print('train loss is:', train_loss/5000)\n",
        "#     test_acc, char_acc = model.evaluate(val_data)\n",
        "#     print(f'test accuracy is {test_acc} and character-wise accuracy is {char_acc}')\n",
        "#     train_loss = 0"
      ],
      "metadata": {
        "id": "FA4xyw70_FCH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "voDdPUQRcmr9"
      },
      "source": [
        "#WandB Setup\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nslhYuf3cr-k",
        "outputId": "751382ea-70f6-42f5-b0ee-3db543adb63e"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/\n",
            "Collecting wandb\n",
            "  Downloading wandb-0.15.3-py3-none-any.whl (2.0 MB)\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m2.0/2.0 MB\u001b[0m \u001b[31m30.2 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25hRequirement already satisfied: Click!=8.0.0,>=7.0 in /usr/local/lib/python3.10/dist-packages (from wandb) (8.1.3)\n",
            "Collecting GitPython!=3.1.29,>=1.0.0 (from wandb)\n",
            "  Downloading GitPython-3.1.31-py3-none-any.whl (184 kB)\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m184.3/184.3 kB\u001b[0m \u001b[31m21.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25hRequirement already satisfied: requests<3,>=2.0.0 in /usr/local/lib/python3.10/dist-packages (from wandb) (2.27.1)\n",
            "Requirement already satisfied: psutil>=5.0.0 in /usr/local/lib/python3.10/dist-packages (from wandb) (5.9.5)\n",
            "Collecting sentry-sdk>=1.0.0 (from wandb)\n",
            "  Downloading sentry_sdk-1.23.1-py2.py3-none-any.whl (205 kB)\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m205.1/205.1 kB\u001b[0m \u001b[31m24.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25hCollecting docker-pycreds>=0.4.0 (from wandb)\n",
            "  Downloading docker_pycreds-0.4.0-py2.py3-none-any.whl (9.0 kB)\n",
            "Requirement already satisfied: PyYAML in /usr/local/lib/python3.10/dist-packages (from wandb) (6.0)\n",
            "Collecting pathtools (from wandb)\n",
            "  Downloading pathtools-0.1.2.tar.gz (11 kB)\n",
            "  Preparing metadata (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "Collecting setproctitle (from wandb)\n",
            "  Downloading setproctitle-1.3.2-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (30 kB)\n",
            "Requirement already satisfied: setuptools in /usr/local/lib/python3.10/dist-packages (from wandb) (67.7.2)\n",
            "Requirement already satisfied: appdirs>=1.4.3 in /usr/local/lib/python3.10/dist-packages (from wandb) (1.4.4)\n",
            "Requirement already satisfied: protobuf!=4.21.0,<5,>=3.19.0 in /usr/local/lib/python3.10/dist-packages (from wandb) (3.20.3)\n",
            "Requirement already satisfied: six>=1.4.0 in /usr/local/lib/python3.10/dist-packages (from docker-pycreds>=0.4.0->wandb) (1.16.0)\n",
            "Collecting gitdb<5,>=4.0.1 (from GitPython!=3.1.29,>=1.0.0->wandb)\n",
            "  Downloading gitdb-4.0.10-py3-none-any.whl (62 kB)\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m62.7/62.7 kB\u001b[0m \u001b[31m8.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25hRequirement already satisfied: urllib3<1.27,>=1.21.1 in /usr/local/lib/python3.10/dist-packages (from requests<3,>=2.0.0->wandb) (1.26.15)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.10/dist-packages (from requests<3,>=2.0.0->wandb) (2022.12.7)\n",
            "Requirement already satisfied: charset-normalizer~=2.0.0 in /usr/local/lib/python3.10/dist-packages (from requests<3,>=2.0.0->wandb) (2.0.12)\n",
            "Requirement already satisfied: idna<4,>=2.5 in /usr/local/lib/python3.10/dist-packages (from requests<3,>=2.0.0->wandb) (3.4)\n",
            "Collecting smmap<6,>=3.0.1 (from gitdb<5,>=4.0.1->GitPython!=3.1.29,>=1.0.0->wandb)\n",
            "  Downloading smmap-5.0.0-py3-none-any.whl (24 kB)\n",
            "Building wheels for collected packages: pathtools\n",
            "  Building wheel for pathtools (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for pathtools: filename=pathtools-0.1.2-py3-none-any.whl size=8791 sha256=ac302d639c494f3acefec8c109b11452c4d8e965d2a85034350f997f5dcc347b\n",
            "  Stored in directory: /root/.cache/pip/wheels/e7/f3/22/152153d6eb222ee7a56ff8617d80ee5207207a8c00a7aab794\n",
            "Successfully built pathtools\n",
            "Installing collected packages: pathtools, smmap, setproctitle, sentry-sdk, docker-pycreds, gitdb, GitPython, wandb\n",
            "Successfully installed GitPython-3.1.31 docker-pycreds-0.4.0 gitdb-4.0.10 pathtools-0.1.2 sentry-sdk-1.23.1 setproctitle-1.3.2 smmap-5.0.0 wandb-0.15.3\n"
          ]
        }
      ],
      "source": [
        "!pip install wandb"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Uqzur-eWq6gY"
      },
      "outputs": [],
      "source": [
        "import wandb"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "7AEcrBGrq6C8",
        "outputId": "9c7bb5c2-9fb1-4d85-9817-949ac9bac212"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[34m\u001b[1mwandb\u001b[0m: Logging into wandb.ai. (Learn how to deploy a W&B server locally: https://wandb.me/wandb-server)\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: You can find your API key in your browser here: https://wandb.ai/authorize\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: Paste an API key from your profile and hit enter, or press ctrl+c to quit: \n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \u001b[32m\u001b[41mERROR\u001b[0m API key must be 40 characters long, yours was 41\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: Logging into wandb.ai. (Learn how to deploy a W&B server locally: https://wandb.me/wandb-server)\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: You can find your API key in your browser here: https://wandb.ai/authorize\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: Paste an API key from your profile and hit enter, or press ctrl+c to quit: \n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: Appending key for api.wandb.ai to your netrc file: /root/.netrc\n"
          ]
        }
      ],
      "source": [
        "!wandb login\n",
        "#63a3fc8a45a3ca461290e711c17782295dfe1975"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "8U4exWKvrA6t"
      },
      "outputs": [],
      "source": [
        "entity_name = \"clroymustang\"\n",
        "project_name = \"cs6910-assignment-3\""
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#Vanilla Hyperparameter Sweep"
      ],
      "metadata": {
        "id": "UMqWK_hLGz6s"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "6YMELBegw5Qu"
      },
      "outputs": [],
      "source": [
        "n_iters = 75000"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "2INJlUhew5Qv"
      },
      "outputs": [],
      "source": [
        "training_pairs = [random.choice(train_data) for i in range(0, n_iters)]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "6RvJeUKyrURQ"
      },
      "outputs": [],
      "source": [
        "def sweep():\n",
        "  defaults = {\n",
        "      'embedding_size': 32,\n",
        "      'n_layers': 1,\n",
        "      'hl_size': 32,\n",
        "      'bidir': False,\n",
        "      'dropout': 0,\n",
        "      'cell_type': 'GRU'\n",
        "  }\n",
        "  wandb.init(config = defaults)\n",
        "  config = wandb.config\n",
        "\n",
        "  vanilla_model = seq2seq_vanilla(english, tamil, embedding_size = config.embedding_size, n_layers = config.n_layers, hl_size = config.hl_size, dropout = config.dropout, \\\n",
        "                          cell_type = config.cell_type, lr = 0.01, teacher_forcing_ratio = 0.5, bidirectional_flag = config.bidir)\n",
        "  \n",
        "  train_loss = 0\n",
        "  for i in range(0, n_iters):\n",
        "    training_pair = training_pairs[i]\n",
        "    x = training_pair[0]\n",
        "    y = training_pair[1]\n",
        "    loss = vanilla_model.train_step(x, y)\n",
        "    train_loss = train_loss + loss\n",
        "\n",
        "    if (i+1)%5000 == 0:\n",
        "      test_acc, char_acc = vanilla_model.evaluate(val_data)\n",
        "      train_loss = train_loss/5000\n",
        "      log_dict = {\"train_loss\":train_loss, \"test_accuracy\":test_acc, \"test_similarity\":char_acc}\n",
        "      wandb.log(log_dict)\n",
        "      train_loss = 0\n",
        "  \n",
        "  sweep_name = str(config.cell_type) + \"_emb\" + str(config.embedding_size) + \"_hl\" + str(config.hl_size) + \"_nl\" + str(config.n_layers) + \"_d\" + \\\n",
        "  str(config.dropout) + \"_bidir\" + str(config.bidir)\n",
        "  wandb.run.name = sweep_name\n",
        "\n",
        "  wandb.run.save()\n",
        "  wandb.run.finish()\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "dtIQrWBarYGl"
      },
      "outputs": [],
      "source": [
        "hyperparameters = {\n",
        "      \"embedding_size\": {\n",
        "          'values': [32, 64, 128]\n",
        "      },\n",
        "      \"n_layers\": {\n",
        "          'values' : [1, 2, 3]\n",
        "      },\n",
        "      \"hl_size\": {\n",
        "          'values' : [32, 64, 128]\n",
        "      },\n",
        "      \"bidir\": {\n",
        "          'values': [True, False]\n",
        "      },\n",
        "      \"dropout\":{\n",
        "        'values': [0, 0.2, 0.3]\n",
        "      },\n",
        "      \"cell_type\": {\n",
        "          'values' : ['RNN', 'GRU', 'LSTM']\n",
        "      }\n",
        "  }"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "ewDLFM5wsuy_"
      },
      "outputs": [],
      "source": [
        "sweep_configuration = {\n",
        "    'method': 'bayes',\n",
        "    'name': 'vanilla_sweep',\n",
        "    'metric': {\n",
        "        'goal': 'maximize', \n",
        "        'name': 'test_similarity'\n",
        "        },\n",
        "    'parameters': hyperparameters\n",
        "}"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "6JISeTaQsyH4"
      },
      "outputs": [],
      "source": [
        "#sweep_id = wandb.sweep(sweep_configuration, entity=entity_name, project=project_name)\n",
        "sweep_id = 'td3auwvp'"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "SVQKj1pB_PRG"
      },
      "outputs": [],
      "source": [
        "#wandb.agent(sweep_id, sweep, project = project_name)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "tinkering"
      ],
      "metadata": {
        "id": "By9O3hoHDjzU"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "#Attention Hyperparameter Mini-Sweep\n",
        "\n"
      ],
      "metadata": {
        "id": "oedBo8q0G7zb"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "yS8-sOE1HS2x"
      },
      "outputs": [],
      "source": [
        "hyperparameters = {\n",
        "      \"embedding_size\": {\n",
        "          'values': [128]\n",
        "      },\n",
        "      \"n_layers\": {\n",
        "          'values' : [1]\n",
        "      },\n",
        "      \"hl_size\": {\n",
        "          'values' : [128]\n",
        "      },\n",
        "      \"bidir\": {\n",
        "          'values': [True, False]\n",
        "      },\n",
        "      \"dropout\":{\n",
        "        'values': [0.2]\n",
        "      },\n",
        "      \"cell_type\": {\n",
        "          'values' : ['GRU', 'LSTM']\n",
        "      },\n",
        "      \"attention\": {\n",
        "          'values' : [True, False]\n",
        "      }\n",
        "  }"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def sweep():\n",
        "  defaults = {\n",
        "      'embedding_size': 128,\n",
        "      'n_layers': 1,\n",
        "      'hl_size': 128,\n",
        "      'bidir': False,\n",
        "      'dropout': 0.2,\n",
        "      'cell_type': 'GRU',\n",
        "      'attention': False\n",
        "  }\n",
        "  wandb.init(config = defaults)\n",
        "  config = wandb.config\n",
        "\n",
        "  if config.attention:\n",
        "    model = seq2seq_attn(english, tamil, embedding_size = config.embedding_size, hl_size = config.hl_size, dropout = config.dropout, cell_type = config.cell_type, \\\n",
        "                         lr = 0.005, teacher_forcing_ratio = 0.5, bidirectional_flag = config.bidir)\n",
        "  else:\n",
        "    model = seq2seq_vanilla(english, tamil, embedding_size = config.embedding_size, n_layers = config.n_layers, hl_size = config.hl_size, dropout = config.dropout, \\\n",
        "                          cell_type = config.cell_type, lr = 0.01, teacher_forcing_ratio = 0.5, bidirectional_flag = config.bidir)\n",
        "  \n",
        "  train_loss = 0\n",
        "  for i in range(0, n_iters):\n",
        "    training_pair = training_pairs[i]\n",
        "    x = training_pair[0]\n",
        "    y = training_pair[1]\n",
        "    loss = model.train_step(x, y)\n",
        "    train_loss = train_loss + loss\n",
        "\n",
        "    if (i+1)%5000 == 0:\n",
        "      test_acc, char_acc = model.evaluate(val_data)\n",
        "      train_loss = train_loss/5000\n",
        "      log_dict = {\"train_loss\":train_loss, \"test_accuracy\":test_acc, \"test_similarity\":char_acc}\n",
        "      wandb.log(log_dict)\n",
        "      train_loss = 0\n",
        "  \n",
        "  sweep_name = str(config.cell_type) + \"_emb\" + str(config.embedding_size) + \"_hl\" + str(config.hl_size) + \"_d\" + \\\n",
        "  str(config.dropout) + \"_bidir\" + str(config.bidir) + \"_attn\" + str(config.attention)\n",
        "\n",
        "  wandb.run.name = sweep_name\n",
        "\n",
        "  wandb.run.save()\n",
        "  wandb.run.finish()"
      ],
      "metadata": {
        "id": "UtNrEIqaIj2P"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "aIOr7BkNHS2-"
      },
      "outputs": [],
      "source": [
        "sweep_configuration = {\n",
        "    'method': 'grid',\n",
        "    'name': 'attention_sweep',\n",
        "    'metric': {\n",
        "        'goal': 'maximize', \n",
        "        'name': 'test_similarity'\n",
        "        },\n",
        "    'parameters': hyperparameters\n",
        "}"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "T222FjZUHS2-"
      },
      "outputs": [],
      "source": [
        "#sweep_id = wandb.sweep(sweep_configuration, entity=entity_name, project=project_name)\n",
        "sweep_id = '6l900368'"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "2Vuz1HrUHS2-"
      },
      "outputs": [],
      "source": [
        "#wandb.agent(sweep_id, sweep, project = project_name)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#Best Vanilla Model"
      ],
      "metadata": {
        "id": "lpvi55EDGS4v"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "best_params = {\n",
        "    'cell_type':'LSTM', 'dropout':0, 'embedding_size':128, 'hl_size':128, 'n_layers':3, 'bidirectional_flag':False\n",
        "}"
      ],
      "metadata": {
        "id": "r9s_hJYd4c85"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "DOpat_BRatpu"
      },
      "outputs": [],
      "source": [
        "model = seq2seq_vanilla(inp_language = english, out_language = tamil, **best_params)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "vipYXEV9atqC"
      },
      "outputs": [],
      "source": [
        "n_iters = 75000"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "nu4UpeoqatqD"
      },
      "outputs": [],
      "source": [
        "training_pairs = [random.choice(train_data) for i in range(0, n_iters)]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "9yZGQ6xFatqD"
      },
      "outputs": [],
      "source": [
        "train_loss = 0\n",
        "for i in range(0, n_iters):\n",
        "  training_pair = training_pairs[i]\n",
        "  x = training_pair[0]\n",
        "  y = training_pair[1]\n",
        "  loss = model.train_step(x, y)\n",
        "  train_loss = train_loss + loss\n",
        "\n",
        "  if (i+1)%5000 == 0:\n",
        "    print('------------------------------------------------')\n",
        "    print('train loss is:', train_loss/5000)\n",
        "    test_acc, char_acc = model.evaluate(val_data)\n",
        "    print(f'test accuracy is {test_acc} and character-wise accuracy is {char_acc}')\n",
        "    train_loss = 0"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "with open('best_vanilla_model.pkl', 'wb') as file:\n",
        "  pickle.dump(model, file)\n",
        "  print('Object successfully saved.')"
      ],
      "metadata": {
        "id": "N_DGJ1J-oicd"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "pickled_model = pickle.load(open('best_vanilla_model.pkl', 'rb'))"
      ],
      "metadata": {
        "id": "MzgtFB3I8aEM"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "#Beam Search"
      ],
      "metadata": {
        "id": "zss0cVzw9eeT"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#we will now perform beam search for various values of B and compare the results\n",
        "# for b in range(2, 6):\n",
        "#   test_acc, char_acc = pickled_model.evaluate_beam(val_data, beam_size = b, print_flag = False)\n",
        "#   print(f'With beam size = {b}, we get a test accuracy of {test_acc} and a character-wise similarity as {char_acc}.')"
      ],
      "metadata": {
        "id": "df_QrEEN9Zfm"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "b = [1, 2, 3, 4, 5]\n",
        "test_acc = [0.377685546875, 0.388671875, 0.3896484375, 0.3916015625, 0.39599609375]\n",
        "test_sim = [0.8217747135037511, 0.8271914291907654, 0.8307840395134558, 0.8324873635588594, 0.8336391599211628]\n",
        "\n",
        "plt.plot(b, test_acc)\n",
        "plt.title('Validation Accuracy v/s Beam Size')\n",
        "plt.xlabel('Beam Size')\n",
        "plt.ylabel('Validation Accuracy')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 489
        },
        "id": "3GivbIFtNEaY",
        "outputId": "61fb5d71-1129-44bb-a7d7-0eb510a9681e"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Text(0, 0.5, 'Validation Accuracy')"
            ]
          },
          "metadata": {},
          "execution_count": 51
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.plot(b, test_sim)\n",
        "plt.title('Validation Similarity v/s Beam Size')\n",
        "plt.xlabel('Beam Size')\n",
        "plt.ylabel('Validation Similarity')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 489
        },
        "id": "8sRUqzcYOqNv",
        "outputId": "f41c6b74-857a-426e-904e-5134a67c7b7a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Text(0, 0.5, 'Validation Similarity')"
            ]
          },
          "metadata": {},
          "execution_count": 52
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#Best Attention Model"
      ],
      "metadata": {
        "id": "aUOXGR5iMdC5"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "best_params = {\n",
        "    'cell_type':'GRU', 'dropout':0.2, 'embedding_size':128, 'hl_size':128, 'bidirectional_flag':True\n",
        "}"
      ],
      "metadata": {
        "id": "MN0CuzmM43ef"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model = seq2seq_attn(inp_language = english, out_language = tamil, **best_params)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yga0HyWEMq1t",
        "outputId": "60d47649-6801-4d1c-ba26-d776cba6c43b"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/rnn.py:71: UserWarning: dropout option adds dropout after all but last recurrent layer, so non-zero dropout expects num_layers greater than 1, but got dropout=0.2 and num_layers=1\n",
            "  warnings.warn(\"dropout option adds dropout after all but last \"\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "train_loss = 0\n",
        "for i in range(0, n_iters):\n",
        "  training_pair = training_pairs[i]\n",
        "  x = training_pair[0]\n",
        "  y = training_pair[1]\n",
        "  loss = model.train_step(x, y)\n",
        "  train_loss = train_loss + loss\n",
        "\n",
        "  if (i+1)%5000 == 0:\n",
        "    print('------------------------------------------------')\n",
        "    print('train loss is:', train_loss/5000)\n",
        "    test_acc, char_acc = model.evaluate(val_data)\n",
        "    print(f'test accuracy is {test_acc} and character-wise accuracy is {char_acc}')\n",
        "    train_loss = 0"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Byhl-2rvM6fQ",
        "outputId": "b20a4b84-97ec-49ad-b629-d72bc7f53f0c"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "------------------------------------------------\n",
            "train loss is: 1.903665394523643\n",
            "test accuracy is 0.043701171875 and character-wise accuracy is 0.6178185361722799\n",
            "------------------------------------------------\n",
            "train loss is: 1.119872172133049\n",
            "test accuracy is 0.1669921875 and character-wise accuracy is 0.7318605761966062\n",
            "------------------------------------------------\n",
            "train loss is: 0.8010369807866555\n",
            "test accuracy is 0.299072265625 and character-wise accuracy is 0.7975959612835202\n",
            "------------------------------------------------\n",
            "train loss is: 0.5968623266388231\n",
            "test accuracy is 0.364013671875 and character-wise accuracy is 0.8230459448770827\n",
            "------------------------------------------------\n",
            "train loss is: 0.4952203252231501\n",
            "test accuracy is 0.3828125 and character-wise accuracy is 0.8301898847232598\n",
            "------------------------------------------------\n",
            "train loss is: 0.42420165220586803\n",
            "test accuracy is 0.414794921875 and character-wise accuracy is 0.8393633873834718\n",
            "------------------------------------------------\n",
            "train loss is: 0.3863839278969744\n",
            "test accuracy is 0.421142578125 and character-wise accuracy is 0.843775230974292\n",
            "------------------------------------------------\n",
            "train loss is: 0.3740708054971908\n",
            "test accuracy is 0.440185546875 and character-wise accuracy is 0.8477520920692774\n",
            "------------------------------------------------\n",
            "train loss is: 0.3472717947092251\n",
            "test accuracy is 0.45361328125 and character-wise accuracy is 0.8514962777667255\n",
            "------------------------------------------------\n",
            "train loss is: 0.33671376505050166\n",
            "test accuracy is 0.46435546875 and character-wise accuracy is 0.8526871368633189\n",
            "------------------------------------------------\n",
            "train loss is: 0.3413252448223062\n",
            "test accuracy is 0.45751953125 and character-wise accuracy is 0.8516383261656298\n",
            "------------------------------------------------\n",
            "train loss is: 0.3308446500447423\n",
            "test accuracy is 0.47265625 and character-wise accuracy is 0.853970973697839\n",
            "------------------------------------------------\n",
            "train loss is: 0.333306979548212\n",
            "test accuracy is 0.470703125 and character-wise accuracy is 0.854185014524833\n",
            "------------------------------------------------\n",
            "train loss is: 0.3196475235398716\n",
            "test accuracy is 0.46484375 and character-wise accuracy is 0.8548459217176734\n",
            "------------------------------------------------\n",
            "train loss is: 0.30728711889718124\n",
            "test accuracy is 0.47509765625 and character-wise accuracy is 0.8572123446253838\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "with open('best_attn_model.pkl', 'wb') as file:\n",
        "  pickle.dump(model, file)\n",
        "  print('Object successfully saved.')"
      ],
      "metadata": {
        "id": "yCl2P4KcM7dG",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "26be594a-ea5d-4de1-bc40-0935dc45b538"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Object successfully saved.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#Evaluation of Vanilla Model on Test-Dataset"
      ],
      "metadata": {
        "id": "WQLB_x7Gfym0"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "vanilla_model = pickle.load(open('/content/gdrive/MyDrive/CS6910_A3/best_models/best_vanilla_model.pkl', 'rb'))"
      ],
      "metadata": {
        "id": "cjLAlp_kOyJx"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "test_acc, test_sim = vanilla_model.evaluate(test_data, False)\n",
        "print(f'The model has test accuracy: {test_acc} and test similarity:{test_sim}.')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "G-Fm2ek3gjqh",
        "outputId": "8a98c1d1-46ea-4e09-9d58-e27b6f7b6dc2"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/rnn.py:812: UserWarning: RNN module weights are not part of single contiguous chunk of memory. This means they need to be compacted at every call, possibly greatly increasing memory usage. To compact weights again call flatten_parameters(). (Triggered internally at ../aten/src/ATen/native/cudnn/RNN.cpp:982.)\n",
            "  result = _VF.lstm(input, hx, self._flat_weights, self.bias, self.num_layers,\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "The model has test accuracy: 0.320556640625 and test similarity:0.7873471098288715.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "vanilla_values = []\n",
        "for test_pair in test_data:\n",
        "  input = test_pair[0]\n",
        "  target = test_pair[1]\n",
        "  pred = vanilla_model.predict(input, target)\n",
        "\n",
        "  target = target.tolist()\n",
        "  target = [t[0] for t in target]\n",
        "  input = input.tolist()\n",
        "  input = [i[0] for i in input]\n",
        "\n",
        "  tar_word = decoded_word(tamil, target)\n",
        "  inp_word = decoded_word(english, input)\n",
        "  pred_word = decoded_word(tamil, pred)\n",
        "\n",
        "  vanilla_values.append([inp_word, tar_word, pred_word])"
      ],
      "metadata": {
        "id": "gkbOt25WkcPP"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import csv\n",
        "header = ['input', 'target', 'prediction']\n",
        "with open('vanilla.csv', 'w') as file:\n",
        "    writer = csv.writer(file)\n",
        "    writer.writerow(header)\n",
        "    writer.writerows(vanilla_values)"
      ],
      "metadata": {
        "id": "myhUq7g9luS5"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "scores = []\n",
        "with_score = []\n",
        "for entry in vanilla_values:\n",
        "  input = entry[0]\n",
        "  tar_word = entry[1]\n",
        "  pred_word = entry[2]\n",
        "  score = 1 - min((Levenshtein.distance(pred_word, tar_word)/max(len(tar_word),len(pred_word))), 1)\n",
        "  entry.append(score)\n",
        "  with_score.append(entry)"
      ],
      "metadata": {
        "id": "UGVrLUPqyjk_"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from operator import itemgetter"
      ],
      "metadata": {
        "id": "GW4fVWpI0p-s"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "sort = sorted(with_score, key=itemgetter(3))"
      ],
      "metadata": {
        "id": "j_K6w9Dt1HZm"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "sort[0:20]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sjZBAtmJ1Z5I",
        "outputId": "4c074593-77bd-43f2-c55a-82b6eafe4de0"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[['kartu', 'சிம்', 'கற்டு', 0.0],\n",
              " ['vendum', 'குளிக்க', 'வெண்டும்', 0.0],\n",
              " ['varukinrana', 'உலா', 'வருகின்றன', 0.0],\n",
              " ['vitum', 'கலந்து', 'விடும்', 0.0],\n",
              " ['okay', 'ஓகே', 'கசோய்', 0.0],\n",
              " ['bee', 'பீ', 'ீய', 0.0],\n",
              " ['vazhi', 'நல்ல', 'வழி', 0.0],\n",
              " ['kei', 'சாவி', 'கெய்', 0.0],\n",
              " ['veentum', 'குளிக்க', 'வீண்டும்', 0.0],\n",
              " ['nava', 'நவ', 'வா', 0.0],\n",
              " ['shankar', 'பிரபு', 'நாங்கர்', 0.0],\n",
              " ['civinki', 'ஒட்டகச்', 'வினிங்கி', 0.0],\n",
              " ['pirakaasa', 'சைவப்', 'பிராசசச', 0.0],\n",
              " ['qey', 'சாவி', 'ஜெய்', 0.0],\n",
              " ['baee', 'பை', 'ீமே', 0.0],\n",
              " ['iowa', 'ஐயோவா', 'ஹேய', 0.0],\n",
              " ['payanikalai', 'சுற்றுலாப்', 'பயனிகளை', 0.0],\n",
              " ['high', 'ஹை', 'விகி', 0.0],\n",
              " ['ina', 'ஐநா', 'னன', 0.0],\n",
              " ['sankar', 'பிரபு', 'சங்கர்', 0.0]]"
            ]
          },
          "metadata": {},
          "execution_count": 109
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "sort[-20:]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "a1D28Ri61xxM",
        "outputId": "3a5dc5c9-910c-4800-c446-9813ae0d194a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[['chopra', 'சோப்ரா', 'சோப்ரா', 1.0],\n",
              " ['nanmai', 'நன்மை', 'நன்மை', 1.0],\n",
              " ['seidhen', 'செய்தேன்', 'செய்தேன்', 1.0],\n",
              " ['arivathu', 'அறிவது', 'அறிவது', 1.0],\n",
              " ['pali', 'பலி', 'பலி', 1.0],\n",
              " ['senti', 'சென்டி', 'சென்டி', 1.0],\n",
              " ['vaangkum', 'வாங்கும்', 'வாங்கும்', 1.0],\n",
              " ['thoalviyaith', 'தோல்வியைத்', 'தோல்வியைத்', 1.0],\n",
              " ['thimuka', 'திமுக', 'திமுக', 1.0],\n",
              " ['uyirinam', 'உயிரினம்', 'உயிரினம்', 1.0],\n",
              " ['ulagaip', 'உலகைப்', 'உலகைப்', 1.0],\n",
              " ['kattamaikka', 'கட்டமைக்க', 'கட்டமைக்க', 1.0],\n",
              " ['saerkkaiyaal', 'சேர்க்கையால்', 'சேர்க்கையால்', 1.0],\n",
              " ['puthumaigalai', 'புதுமைகளை', 'புதுமைகளை', 1.0],\n",
              " ['viradham', 'விரதம்', 'விரதம்', 1.0],\n",
              " ['thattaan', 'தட்டான்', 'தட்டான்', 1.0],\n",
              " ['scotch', 'ஸ்காட்ச்', 'ஸ்காட்ச்', 1.0],\n",
              " ['veechuthal', 'வீசுதல்', 'வீசுதல்', 1.0],\n",
              " ['thaathup', 'தாதுப்', 'தாதுப்', 1.0],\n",
              " ['sattama', 'சட்டமா', 'சட்டமா', 1.0]]"
            ]
          },
          "metadata": {},
          "execution_count": 110
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "confusion_matrix_vanilla = np.zeros([tamil.n_chars - 2,tamil.n_chars - 2])\n",
        "for entry in vanilla_values:\n",
        "  target = entry[1]\n",
        "  predicted = entry[2]\n",
        "  for i in range(0, min(len(target), len(predicted))):\n",
        "    t = target[i]\n",
        "    p = predicted[i]\n",
        "    a = tamil.get_index(t) - 2\n",
        "    b = tamil.get_index(p) - 2\n",
        "    confusion_matrix_vanilla[a, b] = confusion_matrix_vanilla[a, b] + 1"
      ],
      "metadata": {
        "id": "oscDbvHrG4gF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "labels = [u'{}'.format(tamil.get_character(i)) for i in range(2, tamil.n_chars)]"
      ],
      "metadata": {
        "id": "f4bTnMk-KiQh"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "for i in range(0, tamil.n_chars - 2):\n",
        "  for j in range(0, tamil.n_chars - 2):\n",
        "    if confusion_matrix_vanilla[i][j]>500:\n",
        "      confusion_matrix_vanilla[i][j] = 0"
      ],
      "metadata": {
        "id": "UXQdcFCvJ-AJ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# -*- coding: utf-8 -*-\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "from matplotlib.pyplot import figure\n",
        "from matplotlib.font_manager import FontProperties\n",
        "import warnings\n",
        "warnings.filterwarnings('ignore') # setting ignore as a parameter\n",
        "\n",
        "\n",
        "font_prop = FontProperties(fname='/content/Latha.ttf', size=6)\n",
        "ax = plt.subplot()\n",
        "for label in (ax.get_xticklabels() + ax.get_yticklabels()):\n",
        "    label.set_fontproperties(font_prop)\n",
        "\n",
        "sns.heatmap(confusion_matrix_vanilla[:30, :30], xticklabels = labels[:30], yticklabels = labels[:30])"
      ],
      "metadata": {
        "id": "4DCSNBWlNVtV"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "#Evaluation of Attention Model on Test-Dataset"
      ],
      "metadata": {
        "id": "R3-l1Dk5g6GJ"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "attn_model = pickle.load(open('/content/gdrive/MyDrive/CS6910_A3/best_models/best_attn_model.pkl', 'rb'))"
      ],
      "metadata": {
        "id": "i7G5IEnrhBX_"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "test_acc, test_sim = attn_model.evaluate(test_data)\n",
        "print(f'The model has test accuracy: {test_acc} and test similarity:{test_sim}.')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "brJh27VOg3E-",
        "outputId": "a430c71f-3ff5-4202-87ba-123eb03f5351"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "The model has test accuracy: 0.37548828125 and test similarity:0.8192080796912453.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "attn_values = []\n",
        "for test_pair in test_data:\n",
        "  input = test_pair[0]\n",
        "  target = test_pair[1]\n",
        "  pred = attn_model.predict(input, target)\n",
        "\n",
        "  target = target.tolist()\n",
        "  target = [t[0] for t in target]\n",
        "  input = input.tolist()\n",
        "  input = [i[0] for i in input]\n",
        "\n",
        "  tar_word = decoded_word(tamil, target)\n",
        "  inp_word = decoded_word(english, input)\n",
        "  pred_word = decoded_word(tamil, pred)\n",
        "\n",
        "  attn_values.append([inp_word, tar_word, pred_word])"
      ],
      "metadata": {
        "id": "8MVwdJgHhKbD"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import csv\n",
        "header = ['input', 'target', 'prediction']\n",
        "with open('attention.csv', 'w') as file:\n",
        "    writer = csv.writer(file)\n",
        "    writer.writerow(header)\n",
        "    writer.writerows(attn_values)"
      ],
      "metadata": {
        "id": "MLmIjgL1nE2d"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "scores = []\n",
        "with_score2 = []\n",
        "for entry in vanilla_values:\n",
        "  input = entry[0]\n",
        "  tar_word = entry[1]\n",
        "  pred_word = entry[2]\n",
        "  score = 1 - min((Levenshtein.distance(pred_word, tar_word)/max(len(tar_word),len(pred_word))), 1)\n",
        "  entry.append(score)\n",
        "  with_score2.append(entry)"
      ],
      "metadata": {
        "id": "TiB4hJK_uRuU"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "sort = sorted(with_score2, key=itemgetter(3))"
      ],
      "metadata": {
        "id": "lcxv9HG-EUkk"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "sort[:20]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "vDCx8a9VEYGI",
        "outputId": "ae71ac16-8dd3-4afb-8766-bc5d4f714a94"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[['kartu', 'சிம்', 'கற்டு', 0.0, 0.0],\n",
              " ['vendum', 'குளிக்க', 'வெண்டும்', 0.0, 0.0],\n",
              " ['varukinrana', 'உலா', 'வருகின்றன', 0.0, 0.0],\n",
              " ['vitum', 'கலந்து', 'விடும்', 0.0, 0.0],\n",
              " ['okay', 'ஓகே', 'கசோய்', 0.0, 0.0],\n",
              " ['bee', 'பீ', 'ீய', 0.0, 0.0],\n",
              " ['vazhi', 'நல்ல', 'வழி', 0.0, 0.0],\n",
              " ['kei', 'சாவி', 'கெய்', 0.0, 0.0],\n",
              " ['veentum', 'குளிக்க', 'வீண்டும்', 0.0, 0.0],\n",
              " ['nava', 'நவ', 'வா', 0.0, 0.0],\n",
              " ['shankar', 'பிரபு', 'நாங்கர்', 0.0, 0.0],\n",
              " ['civinki', 'ஒட்டகச்', 'வினிங்கி', 0.0, 0.0],\n",
              " ['pirakaasa', 'சைவப்', 'பிராசசச', 0.0, 0.0],\n",
              " ['qey', 'சாவி', 'ஜெய்', 0.0, 0.0],\n",
              " ['baee', 'பை', 'ீமே', 0.0, 0.0],\n",
              " ['iowa', 'ஐயோவா', 'ஹேய', 0.0, 0.0],\n",
              " ['payanikalai', 'சுற்றுலாப்', 'பயனிகளை', 0.0, 0.0],\n",
              " ['high', 'ஹை', 'விகி', 0.0, 0.0],\n",
              " ['ina', 'ஐநா', 'னன', 0.0, 0.0],\n",
              " ['sankar', 'பிரபு', 'சங்கர்', 0.0, 0.0]]"
            ]
          },
          "metadata": {},
          "execution_count": 113
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "sort[-20:]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qykfy0J_EZfL",
        "outputId": "8571680a-05c3-425f-8aeb-a0929b4bebe6"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[['chopra', 'சோப்ரா', 'சோப்ரா', 1.0, 1.0],\n",
              " ['nanmai', 'நன்மை', 'நன்மை', 1.0, 1.0],\n",
              " ['seidhen', 'செய்தேன்', 'செய்தேன்', 1.0, 1.0],\n",
              " ['arivathu', 'அறிவது', 'அறிவது', 1.0, 1.0],\n",
              " ['pali', 'பலி', 'பலி', 1.0, 1.0],\n",
              " ['senti', 'சென்டி', 'சென்டி', 1.0, 1.0],\n",
              " ['vaangkum', 'வாங்கும்', 'வாங்கும்', 1.0, 1.0],\n",
              " ['thoalviyaith', 'தோல்வியைத்', 'தோல்வியைத்', 1.0, 1.0],\n",
              " ['thimuka', 'திமுக', 'திமுக', 1.0, 1.0],\n",
              " ['uyirinam', 'உயிரினம்', 'உயிரினம்', 1.0, 1.0],\n",
              " ['ulagaip', 'உலகைப்', 'உலகைப்', 1.0, 1.0],\n",
              " ['kattamaikka', 'கட்டமைக்க', 'கட்டமைக்க', 1.0, 1.0],\n",
              " ['saerkkaiyaal', 'சேர்க்கையால்', 'சேர்க்கையால்', 1.0, 1.0],\n",
              " ['puthumaigalai', 'புதுமைகளை', 'புதுமைகளை', 1.0, 1.0],\n",
              " ['viradham', 'விரதம்', 'விரதம்', 1.0, 1.0],\n",
              " ['thattaan', 'தட்டான்', 'தட்டான்', 1.0, 1.0],\n",
              " ['scotch', 'ஸ்காட்ச்', 'ஸ்காட்ச்', 1.0, 1.0],\n",
              " ['veechuthal', 'வீசுதல்', 'வீசுதல்', 1.0, 1.0],\n",
              " ['thaathup', 'தாதுப்', 'தாதுப்', 1.0, 1.0],\n",
              " ['sattama', 'சட்டமா', 'சட்டமா', 1.0, 1.0]]"
            ]
          },
          "metadata": {},
          "execution_count": 114
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "confusion_matrix_attn = np.zeros([tamil.n_chars - 2,tamil.n_chars - 2])\n",
        "for entry in vanilla_values:\n",
        "  target = entry[1]\n",
        "  predicted = entry[2]\n",
        "  for i in range(0, min(len(target), len(predicted))):\n",
        "    t = target[i]\n",
        "    p = predicted[i]\n",
        "    a = tamil.get_index(t) - 2\n",
        "    b = tamil.get_index(p) - 2\n",
        "    confusion_matrix_attn[a, b] = confusion_matrix_attn[a, b] + 1"
      ],
      "metadata": {
        "id": "laa_em8TEb02"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "for i in range(0, tamil.n_chars - 2):\n",
        "  for j in range(0, tamil.n_chars - 2):\n",
        "    if confusion_matrix_attn[i][j]>500:\n",
        "      confusion_matrix_attn[i][j] = 0"
      ],
      "metadata": {
        "id": "PgYWIUK2WI1a"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# -*- coding: utf-8 -*-\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "from matplotlib.pyplot import figure\n",
        "from matplotlib.font_manager import FontProperties\n",
        "import warnings\n",
        "warnings.filterwarnings('ignore') # setting ignore as a parameter\n",
        "\n",
        "\n",
        "font_prop = FontProperties(fname='/content/Latha.ttf', size=6)\n",
        "ax = plt.subplot()\n",
        "for label in (ax.get_xticklabels() + ax.get_yticklabels()):\n",
        "    label.set_fontproperties(font_prop)\n",
        "\n",
        "sns.heatmap(confusion_matrix_attn[:30, :30], xticklabels = labels[:30], yticklabels = labels[:30])"
      ],
      "metadata": {
        "id": "vxV3AKYlVwSr"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#what predictions does attention get write that vanilla gets wrong?\n",
        "attention_better = []\n",
        "for i in range(0, len(attn_values)):\n",
        "  vanilla = vanilla_values[i]\n",
        "  attention = attn_values[i]\n",
        "  if attention[2] == attention[1] and vanilla[2] != vanilla[1]:\n",
        "    attention_better.append([attention[0], attention[1], vanilla[2], attention[2]])"
      ],
      "metadata": {
        "id": "SXttCZucV2QC"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "len(attention_better)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "3KNrsR60X-8I",
        "outputId": "8bd56569-ff16-4483-f275-a6864ad13bdd"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "504"
            ]
          },
          "metadata": {},
          "execution_count": 191
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "rand = random.sample(attention_better, 20)"
      ],
      "metadata": {
        "id": "s1goqt6sYGbS"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "for entry in rand:\n",
        "  print(entry)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YsjYp49bYeOf",
        "outputId": "bb9b6292-cbef-4c68-e8e5-ff139b204836"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "['vealaiyaiyum', 'வேலையையும்', 'வேளையையும்', 'வேலையையும்']\n",
            "['kaithal', 'கைதல்', 'கைத்ல', 'கைதல்']\n",
            "['oththuzhaikkinranar', 'ஒத்துழைக்கின்றனர்', 'ஒத்துழைக்கின்றர்', 'ஒத்துழைக்கின்றனர்']\n",
            "['maruththuvarai', 'மருத்துவரை', 'மறுத்துவரை', 'மருத்துவரை']\n",
            "['aayiraththai', 'ஆயிரத்தை', 'ஆயிர்ததை', 'ஆயிரத்தை']\n",
            "['kidaikkakkoodiya', 'கிடைக்கக்கூடிய', 'கிடைக்கட்கூடிய', 'கிடைக்கக்கூடிய']\n",
            "['piramukar', 'பிரமுகர்', 'பிரமுக்்', 'பிரமுகர்']\n",
            "['mutharkattamaaga', 'முதற்கட்டமாக', 'முதர்கட்டாக', 'முதற்கட்டமாக']\n",
            "['seiyappattiruppaar', 'செய்யப்பட்டிருப்பார்', 'செய்யப்பட்டுப்பார்', 'செய்யப்பட்டிருப்பார்']\n",
            "['mandabaththinullea', 'மண்டபத்தினுள்ளே', 'மண்டதத்தினில்லே', 'மண்டபத்தினுள்ளே']\n",
            "['karumpalakaiyil', 'கரும்பலகையில்', 'கர்ம்பகளைகளி்', 'கரும்பலகையில்']\n",
            "['magizhunthu', 'மகிழுந்து', 'கசிழுந்து', 'மகிழுந்து']\n",
            "['ontruthirandulla', 'ஒன்றுதிரண்டுள்ள', 'ஒன்றுதிர்்்்ல்', 'ஒன்றுதிரண்டுள்ள']\n",
            "['seimadhi', 'செய்மதி', 'செய்மாிி', 'செய்மதி']\n",
            "['melton', 'மெல்டன்', 'மெல்ட்ன்', 'மெல்டன்']\n",
            "['kodukkappattikkum', 'கொடுக்கப்பட்டிக்கும்', 'கொடுக்கப்படிட்குக்ம்', 'கொடுக்கப்பட்டிக்கும்']\n",
            "['purinthum', 'புரிந்தும்', 'புறிந்தும்', 'புரிந்தும்']\n",
            "['seithiyaalargalukum', 'செய்தியாளர்களுக்கும்', 'செய்தியாளர்களுகும்', 'செய்தியாளர்களுக்கும்']\n",
            "['purivadhu', 'புரிவது', 'புறிவது', 'புரிவது']\n",
            "['guha', 'குஹா', 'கு', 'குஹா']\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#Attention Heatmaps"
      ],
      "metadata": {
        "id": "nVWBcEL-d7kT"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "random_test_data = random.sample(test_data, 9)"
      ],
      "metadata": {
        "id": "u3DRfFGYgoxD"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "heatmaps = []\n",
        "out_words = []\n",
        "inputs = []\n",
        "\n",
        "for i in range(9):\n",
        "    sample = random_test_data[i]\n",
        "    input = sample[0]\n",
        "    input_w = input.tolist()\n",
        "    input_w = [a[0] for a in input_w]\n",
        "    word_in = decoded_word(english, input_w)\n",
        "\n",
        "    inputs.append(word_in)\n",
        "    target = sample[1]\n",
        "\n",
        "\n",
        "    with torch.no_grad():\n",
        "      encoder_hidden = attn_model.encoder.init_hidden()\n",
        "\n",
        "      if attn_model.bidir == True:\n",
        "        encoder_outputs = torch.zeros(attn_model.max_length, 2*attn_model.encoder.hl_size, device=device)\n",
        "      else:\n",
        "        encoder_outputs = torch.zeros(attn_model.max_length, attn_model.encoder.hl_size, device=device)\n",
        "\n",
        "      input_length = input.size(0)\n",
        "      for j in range(0, input_length):\n",
        "        encoder_output, encoder_hidden = attn_model.encoder.forward(input[j], encoder_hidden)\n",
        "        encoder_outputs[j] = encoder_output[0, 0]\n",
        "\n",
        "      decoder_input = torch.tensor([[start_token]], device=device)\n",
        "      decoder_hidden = encoder_hidden\n",
        "\n",
        "      outputs = []\n",
        "      decoder_attentions = torch.zeros(attn_model.max_length, attn_model.max_length)\n",
        "      for j in range(0, attn_model.max_length):\n",
        "        decoder_output, decoder_hidden, decoder_attention = attn_model.decoder.forward(decoder_input, decoder_hidden, encoder_outputs)\n",
        "        decoder_attentions[j] = decoder_attention.data\n",
        "\n",
        "        value, index = decoder_output.data.topk(1)\n",
        "        decoder_input = index.squeeze().detach()\n",
        "        outputs.append(decoder_input.item())\n",
        "        if decoder_input.item() == end_token:\n",
        "          break\n",
        "    \n",
        "    out_word = decoded_word(tamil, outputs)\n",
        "    \n",
        "    heatmaps.append(decoder_attentions[:j+1])\n",
        "    out_words.append(out_word)\n"
      ],
      "metadata": {
        "id": "gEBkioA3fwLU"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt\n",
        "from matplotlib.pyplot import figure\n",
        "from matplotlib.font_manager import FontProperties\n",
        "import warnings\n",
        "warnings.filterwarnings('ignore') # setting ignore as a parameter"
      ],
      "metadata": {
        "id": "eAvAW4VUqZNS"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "\n",
        "fig, axes = plt.subplots(3, 3, figsize=(25, 15))\n",
        "\n",
        "font_prop = FontProperties(fname='/content/Latha.ttf', size=7)\n",
        "#ax = plt.subplot()\n",
        "# for label in (axes.get_yticklabels()):\n",
        "#     label.set_fontproperties(font_prop)\n",
        "\n",
        "xlabels = [english.get_character(i) for i in range(0, english.n_chars)]\n",
        "\n",
        "for i, ax in enumerate(axes.flatten()):\n",
        "    ax.set_yticklabels(ax.get_yticklabels(), fontproperties=font_prop)\n",
        "    ylabels = [letter for letter in out_words[i]]\n",
        "    xlabels = [letter for letter in inputs[i]]\n",
        "    arr = np.array(heatmaps[i])\n",
        "    arr = arr[:,0:len(inputs[i])]\n",
        "    sns.heatmap(arr, ax=ax, cmap='YlGnBu', xticklabels = xlabels, yticklabels = ylabels)\n",
        "    ax.set_title(f'input - {inputs[i]}')\n",
        "\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 718
        },
        "id": "0v3sfayThRe6",
        "outputId": "5ce666a2-d11d-434c-852b-ded6417793b7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 2500x1500 with 18 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAB5QAAATFCAYAAABvmuoWAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAA9hAAAPYQGoP6dpAAEAAElEQVR4nOzde5yN5f7/8feaGdZgMjFmnHJOIqf2iIhINBVmU8mm7bQLEWHiq+lgZnSYnbCpMLFDB5WypaIIoQOVUmp3cAgpOeeQ00xm7t8f/axtWWvGWnO67zXX67kf67Gba92Hz73WWN7W577vy2VZliUAAAAAAAAAAAAAAM4TZncBAAAAAAAAAAAAAABnoqEMAAAAAAAAAAAAAPCLhjIAAAAAAAAAAAAAwC8aygAAAAAAAAAAAAAAv2goAwAAAAAAAAAAAAD8oqEMAAAAAAAAAAAAAPCLhjIAAAAAAAAAAAAAwC8aygAAAAAAAAAAAAAAv2goAwAAAAAAAAAAAAD8oqEM5GHevHlyuVzauXOn3aWEBLtfrwEDBigqKsqWfQMAAOTG7oyEgqtdu7YGDBhgy77JuAAAIL/IoUUjv9lw586dcrlcmjdvXqHX5DQdOnRQhw4dPD+vWbNGLpdLCxcutK8oAAVCQxkIId99951SU1MJgQAAAMg3MqXznDx5UqmpqVqzZo3dpQAAABQZcigAhC4aykAe+vbtq1OnTqlWrVp2lyLpz9CVlpbm2NDltNcLAADACZyWkZyeKU108uRJpaWl0VAGAACFihzqLLVq1dKpU6fUt29fu0sBgKDRUAbyEB4ersjISLlcLrtLscWJEyeCWt6u1+v06dPKyckp1n0CAAAEyvRMWdiCzagAAACmIoc6i8vlUmRkpMLDw4t1v5Zl6dSpU8W6TwAlDw1lIA/+5hmpXbu2unbtqo8++kgtW7ZUZGSk6tatqxdeeMHvuh988IGGDBmimJgYlS9fXv369dPhw4e9lnW5XEpNTfXZ/7nzccybN089e/aUJF133XVyuVxyuVyFdhVDamqqXC6XvvvuO/Xp00cVKlRQ27ZtJUlff/21BgwYoLp16yoyMlJVqlTRP/7xDx06dMjvMZ/7euXk5Cg1NVXVqlVT2bJldd111+m7777zmWvkt99+05gxY9SkSRNFRUWpfPnyuummm7Rp0yavfZydb+PVV1/VQw89pOrVq6ts2bI6duyYZ5ndu3ere/fuioqKUmxsrMaMGaPs7OxCeZ0AAACCZUqmXL16tVwul9544w2f515++WW5XC6tX7/eM/bDDz/otttuU8WKFRUZGakWLVrorbfe8lrv7PGvXbtWw4YNU1xcnC655BJJ/8uvP/zwg26//XaVL19eMTExGjlypE6fPu21nblz56pjx46Ki4uT2+1Wo0aNNHPmTJ86LcvSo48+qksuucSTXb/99luf5QLNrllZWRo/frzi4+MVHR2tcuXKqV27dlq9erVnmZ07dyo2NlaSlJaW5nlPzn8vA8m4kyZNUps2bRQTE6MyZcooPj6eeeoAADCYKTlU+l823LZtmwYMGKCLL75Y0dHRGjhwoE6ePOm1rF3Z0N8cygMGDFBUVJS2b9+uhIQElStXTtWqVdOECRNkWZbX+jk5OZo6daquuOIKRUZGqnLlyhoyZIjP+3H2PV6+fLlatGihMmXK6Nlnnw3q2N9880116dJF1apVk9vtVr169fTII4/4/Y511qxZqlevnsqUKaOWLVvqww8/9Fnm3GN47LHHdMkllygyMlLXX3+9tm3b5rXMhx9+qJ49e6pmzZpyu92qUaOGRo8eTVMcsFmE3QUAoWjbtm267bbbdOedd6p///6aM2eOBgwYoPj4eF1xxRVeyw4fPlwXX3yxUlNTtXnzZs2cOVM//fSTpzEaqGuvvVb33nuvnnrqKT3wwANq2LChJHn+v7D07NlT9evX1+OPP+4JLStWrND27ds1cOBAValSRd9++61mzZqlb7/9Vp988kmex5GcnKyJEyeqW7duSkhI0KZNm5SQkODzJd/27du1ePFi9ezZU3Xq1NG+ffv07LPPqn379vruu+9UrVo1r+UfeeQRlS5dWmPGjFFmZqZKly4tScrOzlZCQoJatWqlSZMmaeXKlZo8ebLq1aunoUOHFuprBQAAUBAlLVN26NBBNWrU0Pz589WjRw+v5+bPn6969eqpdevWkqRvv/1W11xzjapXr677779f5cqV02uvvabu3bvrP//5j8/6w4YNU2xsrMaPH+9zhfLtt9+u2rVrKz09XZ988omeeuopHT582OtL0ZkzZ+qKK65QYmKiIiIi9Pbbb2vYsGHKycnRPffc41lu/PjxevTRR3XzzTfr5ptv1saNG3XDDTcoKyvLa5+BZtdjx47p3//+t3r37q1Bgwbp999/13PPPaeEhAR99tlnat68uWJjYzVz5kwNHTpUPXr00C233CJJatq0qWd/gWbcadOmKTExUXfccYeysrL06quvqmfPnlqyZIm6dOkS9HsKAABKppKWQ891++23q06dOkpPT9fGjRv173//W3FxcXriiSc8y9iVDXOTnZ2tG2+8UVdffbUmTpyoZcuWKSUlRWfOnNGECRM8yw0ZMkTz5s3TwIEDde+992rHjh165pln9OWXX+rjjz9WqVKlPMtu3rxZvXv31pAhQzRo0CA1aNAgqGOfN2+eoqKilJSUpKioKL3//vsaP368jh07pieffNKz3HPPPachQ4aoTZs2GjVqlLZv367ExERVrFhRNWrU8DnWf/7znwoLC9OYMWN09OhRTZw4UXfccYc+/fRTzzKvv/66Tp48qaFDhyomJkafffaZnn76af3yyy96/fXX83wtARQhC0Cu5s6da0myduzY4RmrVauWJcn64IMPPGP79++33G63dd999/msGx8fb2VlZXnGJ06caEmy3nzzTc+YJCslJcVn/7Vq1bL69+/v+fn111+3JFmrV68ulOM7V0pKiiXJ6t27t89zJ0+e9Bl75ZVXfF6H81+vvXv3WhEREVb37t291k1NTbUkeR3b6dOnrezsbK/lduzYYbndbmvChAmesdWrV1uSrLp16/rU1b9/f0uS1/KWZVlXXnmlFR8fn/cLAAAAUERMypTJycmW2+22jhw54hnbv3+/FRER4VXb9ddfbzVp0sQ6ffq0ZywnJ8dq06aNVb9+fc/Y2eNv27atdebMGa99nc2viYmJXuPDhg2zJFmbNm3yjPnLswkJCVbdunW96ixdurTVpUsXKycnxzP+wAMP5Du7njlzxsrMzPRa7vDhw1blypWtf/zjH56xAwcO5Pr+BZNxzz/OrKwsq3HjxlbHjh19tgsAAEo+k3Lo2Wx4bsayLMvq0aOHFRMT4zVmVzbcsWOHJcmaO3euZ+xs1hsxYoRnLCcnx+rSpYtVunRp68CBA5ZlWdaHH35oSbLmz5/vtZ9ly5b5jJ99j5ctW+ZznIEce27LDRkyxCpbtqwnw2dlZVlxcXFW8+bNvTLvrFmzLElW+/btPWNnv9Nt2LCh17LTpk2zJFnffPNNnvtOT0+3XC6X9dNPP/k8B6B4cMtrIB8aNWqkdu3aeX6OjY1VgwYNtH37dp9lBw8e7HV22NChQxUREaF33nmnWGoN1t133+0zVqZMGc9/nz59WgcPHtTVV18tSdq4cWOu21q1apXOnDmjYcOGeY2PGDHCZ1m3262wsD8/krKzs3Xo0CFFRUWpQYMGfvfRv39/r7ryOoZ27dr5fW8AAADsVBIzZb9+/ZSZmel1m+UFCxbozJkz+vvf/y7pz1sCvv/++7r99tv1+++/6+DBgzp48KAOHTqkhIQEbd26Vbt37/ba7qBBg3Kda+7cKymk/2XNc1+bc3Pj0aNHdfDgQbVv317bt2/X0aNHJUkrV65UVlaWRowY4XW1zahRo3z2GWh2DQ8P99xJJycnR7/99pvOnDmjFi1a5Jmj/Qkk4557nIcPH9bRo0fVrl27oPcFAABKtpKYQ8/yl5kOHTrkNV2eXdkwL8OHD/f8t8vl0vDhw5WVlaWVK1dK+vOq3ejoaHXu3NmTnw8ePKj4+HhFRUV5TakiSXXq1FFCQoLPfgI59vOXO5vZ27Vrp5MnT+qHH36QJH3++efav3+/7r77bk/mlf68jXd0dLTf4xw4cKDXsmd/D8/93Tt33ydOnNDBgwfVpk0bWZalL7/80u92ARQ9bnkN5EPNmjV9xipUqOAzX4Uk1a9f3+vnqKgoVa1a1WvukqJw6tQprxAgSVWqVLngenXq1PEZ++2335SWlqZXX31V+/fv93ru/H2c66effpIkXXrppV7jFStWVIUKFbzGcnJyNG3aNM2YMUM7duzwmo8jJiYmoDolKTIy0jMH3Vm5vTcAAAB2KomZ8vLLL9dVV12l+fPn684775T05+2ur776ak8m3LZtmyzL0sMPP6yHH37Y73b279+v6tWre37OLftJvq9NvXr1FBYW5vXafPzxx0pJSdH69et95tA7evSooqOjPdn1/O3FxsYWKLs+//zzmjx5sn744Qf98ccfAR3T+QLNuEuWLNGjjz6qr776SpmZmZ7xYG5HCQAASr6SmEPPOv/Yzua4w4cPq3z58pLszYb+hIWFqW7dul5jl112mSR5XuetW7fq6NGjiouL87uN87+zzS1rBnLs0p9T1Dz00EN6//33vZrxZ5eTlOtrVKpUKZ/jOSuv9+esXbt2afz48Xrrrbd8fifz+i4aQNGioQzkQ25XR1j/f87hwnJu+AjWggULNHDgQK+xQOrzd9Xv7bffrnXr1mns2LFq3ry5oqKilJOToxtvvFE5OTn5rvFcjz/+uB5++GH94x//0COPPKKKFSsqLCxMo0aN8ruP3K5Ozu29AQAAcJqSmin79eunkSNH6pdfflFmZqY++eQTPfPMM57nz2a7MWPG+L1qQvI9ITG37OfP+c3TH3/8Uddff70uv/xyTZkyRTVq1FDp0qX1zjvv6F//+le+8myg2fWll17SgAED1L17d40dO1ZxcXEKDw9Xenq6fvzxx4D3F0jG/fDDD5WYmKhrr71WM2bMUNWqVVWqVCnNnTtXL7/8ctDHCAAASq6SmkOlCx+bndmwIHJychQXF6f58+f7ff78kw/95edAj/3IkSNq3769ypcvrwkTJqhevXqKjIzUxo0bNW7cuAId04Xen+zsbHXu3Fm//fabxo0bp8svv1zlypXT7t27NWDAgEJ7PQEEj4YyUMS2bt2q6667zvPz8ePHtWfPHt18882esQoVKujIkSNe62VlZWnPnj1eY8FcWZCQkKAVK1bkr+hzHD58WKtWrVJaWprGjx/vGd+6desF161Vq5akP69COfesuEOHDvmcXbZw4UJdd911eu6557zGjxw5okqVKhXkEAAAAEJeKGXKv/3tb0pKStIrr7yiU6dOqVSpUurVq5fn+bNXK5QqVUqdOnUKatv+bN261Strbtu2TTk5Oapdu7Yk6e2331ZmZqbeeustrysizr8t4NnsunXrVq8rKg4cOJDv7Lpw4ULVrVtXixYt8nrdU1JSvNYrjCuI//Of/ygyMlLLly+X2+32jM+dO7fA2wYAAOYKpRwaCDuzYW5ycnK0fft2z1XJkrRlyxZJ8mTaevXqaeXKlbrmmmuCOtnyXIEe+5o1a3To0CEtWrRI1157rWd8x44dXsud+xp17NjRM/7HH39ox44datasWdA1fvPNN9qyZYuef/559evXzzNeFL8LAILDHMpAEZs1a5bXre1mzpypM2fO6KabbvKM1atXTx988IHPeuefxVeuXDlJ8glo/lStWlWdOnXyeuTH2bPGzj8DcOrUqRdc9/rrr1dERIRmzpzpNX7uFSrn7uf8fbz++us+8+cBAACYKJQyZaVKlXTTTTfppZde0vz583XjjTd6fZEWFxenDh066Nlnn/X5klH680u6YEyfPt3r56efflqSPK+Nvzx79OhRn0Zrp06dVKpUKT399NNey/rLvYFmV3/7/vTTT7V+/Xqv5cqWLSspsPckN+Hh4XK5XF7v986dO7V48eJ8bxMAACCUcmgg7MyGeTn3+1LLsvTMM8+oVKlSuv766yX9eQfJ7OxsPfLIIz7rnjlzJqDXNNBj97dcVlaWZsyY4bVcixYtFBsbq4yMDGVlZXnG582bl+9c62/flmVp2rRp+doegMLDFcpAEcvKytL111+v22+/XZs3b9aMGTPUtm1bJSYmepa56667dPfdd+vWW29V586dtWnTJi1fvtznDLbmzZsrPDxcTzzxhI4ePSq3262OHTvmOndGYShfvryuvfZaTZw4UX/88YeqV6+u9957z+eMNH8qV66skSNHavLkyUpMTNSNN96oTZs26d1331WlSpW8zkrs2rWrJkyYoIEDB6pNmzb65ptvNH/+/Fzn2wAAADBJqGXKfv366bbbbpMkv196TZ8+XW3btlWTJk00aNAg1a1bV/v27dP69ev1yy+/aNOmTQHva8eOHZ6suX79er300kvq06eP54qIG264QaVLl1a3bt00ZMgQHT9+XLNnz1ZcXJxXQzs2NlZjxoxRenq6unbtqptvvllffvmlJ7ueK9Ds2rVrVy1atEg9evRQly5dtGPHDmVkZKhRo0Y6fvy4Z7kyZcqoUaNGWrBggS677DJVrFhRjRs3VuPGjQN+Hbp06aIpU6boxhtvVJ8+fbR//35Nnz5dl156qb7++uuAtwMAAHCuUMuhF2JnNsxNZGSkli1bpv79+6tVq1Z69913tXTpUj3wwAOeW1m3b99eQ4YMUXp6ur766ivdcMMNKlWqlLZu3arXX39d06ZN8+Tvgh57mzZtVKFCBfXv31/33nuvXC6XXnzxRZ+mealSpfToo49qyJAh6tixo3r16qUdO3Zo7ty5+f5O9/LLL1e9evU0ZswY7d69W+XLl9d//vMfv/N7AyheXKEMFLFnnnlGDRs21Pjx4zVv3jz17t1bb775plczddCgQRo3bpw++OAD3XfffdqxY4dWrFjhOWvvrCpVqigjI0P79+/XnXfeqd69e+u7774r8mN4+eWXlZCQoOnTpys5OVmlSpXSu+++G9C6TzzxhB5++GFt2LBBY8aM0bZt2/Tee+/JsixFRkZ6lnvggQd03333afny5Ro5cqQ2btyopUuXqkaNGkV1WAAAACEj1DJlt27dVKFCBUVHR3t92XhWo0aN9Pnnn6tLly6aN2+e7rnnHmVkZCgsLMxrmpVALFiwQG63W/fff7+WLl2q4cOHe91usEGDBlq4cKFcLpfGjBmjjIwMDR48WCNHjvTZ1qOPPqq0tDR9+eWXGjt2rH788Ue99957Pq9hoNl1wIABevzxx7Vp0ybde++9Wr58uV566SW1aNHCZ9///ve/Vb16dY0ePVq9e/fWwoULg3odOnbsqOeee0579+7VqFGj9Morr+iJJ55Qjx49gtoOAADAuUIth16IndkwN+Hh4Vq2bJn27t2rsWPHasOGDUpJSfE5MTMjI0OzZs3S/v379cADDyg5OVnvv/++/v73v+uaa64ptGOPiYnRkiVLVLVqVT300EOaNGmSOnfurIkTJ/psc/DgwZoxY4Z+/fVXjR07Vh9++KHeeuutfH+nW6pUKb399ttq3ry50tPTlZaWpvr16+uFF17I1/YAFB6XFchM9gCCNm/ePA0cOFAbNmzw+4VRSfTcc8/prrvu0s8//6xLLrkk1+WOHDmiChUq6NFHH9WDDz5YjBUCAACEllDNlGfOnFG1atXUrVs3n7nkCktqaqrS0tJ04MCBgOamAwAAQOBCNYeGmgEDBmjhwoVed68BACfiCmUAhWbPnj1yuVyqWLGiZ+zUqVM+y52da6RDhw7FVBkAAACK0+LFi3XgwAH169fP7lIAAAAAAEABMYcygALbt2+fFi5cqIyMDLVu3Vply5b1PLdgwQLNmzdPN998s6KiovTRRx/plVde0Q033BDQrVgAAAAQOj799FN9/fXXeuSRR3TllVeqffv2dpcEAAAAAAAKiIYygAL7/vvvNXbsWLVs2VKzZ8/2eq5p06aKiIjQxIkTdezYMVWuXFkjR47Uo48+alO1AAAAKCozZ87USy+9pObNm2vevHl2lwMAAAAAAAoBcygDAAAAAAAAAAAAAPxiDmUAAAAAAAAAAAAAgF80lAEAAAAAAAAAAAAAftFQBgAAAAAAAAAAAAD4FWHfrrfYt2uHKFMzxe4SbHdyV6rdJdjOJZfdJQBwlMuKdW9lavYu8DZO7XqlECoBnMj5eTVU8uSpXWl2lwAAKDTkVcA5nJNXnZRLyZ4AYDryalHgCmUAAAAAAAAAAAAAgF82XqEMAADs5nJxbhkAAACci7wKAAAAJzMlr9JQBgDAYC5uVgIAAAAHI68CAADAyUzJqzSUAQAwmCln0AEAACA0kVcBAADgZKbkVTOOEgAAAAAAAAAAAAAQNK5QBgDAYKacQQcAAIDQRF4FAACAk5mSV2koAwBgMJfLZXcJAAAAQK7IqwAAAHAyU/IqDWUAAIxmxhl0AAAACFXkVQAAADiZGXnVjKMEAAAAAAAAAAAAAASNK5QBADCYKXN8AAAAIDSRVwEAAOBkpuRVGsoAABjMlMADAACA0EReBQAAgJOZkldpKAMAYDAXs18AAADAwcirAAAAcDJT8qoZRwkAAAAAAAAAAAAACBpXKAMAYDBTbskCAACA0EReBQAAgJOZkldpKAMAYDBTAg/MsXv3bv34449+n7v22muLuRoAAFBQ5FUAAAA4mSl5lYYyAAAGMyXwwBwffPCBnnrqKZ/xTZs26eTJkzZUBAAACoK8CgAAACczJa/SUAYAAECJ0bt3b/Xu3dtnvH79+jZUAwAAAPj6/vvvNXnyZP3666+6/PLLNXToUPIqAABwNDPa5gAAwC9XIfwPCAUuV96/q5mZmTp27JjXIzMzq5iqAwAAuSGvoqSYMmWK578TExNVv359DR06VLVr11aPHj20fv36PNcnrwIA4Eym5FUaygAAGMzlCivwAygJ0tPTFR0d7fVIT3/W7rIAADAeeRUlRUZGhrZu3aotW7ZIkq699lo1aNBAvXv31ssvv6zU1NQ81yevAgDgTKbkVW55DQCAwUIlsACB+u677/T555/7jP/+++95rpecnKykpCSvMbd7V6HWBgAAgkdeRUlhWZZuuOEGz899+vSRJP3973/XI488oh07duS5PnkVAABnMiWv0lAGAABAibFjxw69++67np+PHz+u5cuXq0OHDnmu53a75Xa7zxstXfgFAgAAwEgulyvPpnFYWN5fRpNXAQCAnWgoAwBgMFPOoIM5unTpoi5duniNLVmyRI8//rhNFQEAgIIgr6KksCxLN998s894ly5ddM8998iyLB06dEgxMTE2VAcAAPLLlLxKQxkAAKOZEXhgjtdee00ZGRleY5mZmdq+fbtNFQEAgIIhr6JkyMjI0PHjx33G69SpI0m6++67tWzZMt1xxx3FXRoAACgQM/IqDWUAAAxmyhl0MEerVq0UHR3tNRYWFqZGjRrZVBEAACgI8ipKiuuvvz7P50ePHl1MlQAAgMJkSl6loQwAgMFMCTwwR61atVSrVi27ywAAAIWEvAoAAAAnMyWvmnGUAAAAAAAAAAAAAICgBdxQ7tixY1HWAQAAbOBSWIEfAAAAQFEhrwIAAMDJTMmrAd/y+ueff/bbVI6NjdWsWbN85qo7V2ZmpjIzM73G3O4sud2lgygVAAAUNlNuyQIAAIDQRF4FAACAk5mSV4OaQ3ncuHE+YwcOHFBiYqIWLVqkmJgYv+ulp6crLS3NaywlZbhSU0cEs3sAAFDIXC6X3SUAAAAAuSKvAgAAwMlMyasBN5RdLpcSEhL8Ple/fn0lJiZq8eLFio2N9Xk+OTlZSUlJXmNu964gSwUAAAAAAAAAAAAAFKeAG8qWZfm95XViYqJGjRqlefPmqWLFin7Xdbvdcrvd541yu2sAAOxmyi1ZAAAAEJrIqwAAAHAyU/JqwA3luXPn6sSJEz7jNWvWlPTnVcoAACC0uGRG4AEAAEBoIq8CAADAyUzJqwE3lNu2bVuUdQAAABuYcgYdAAAAQhN5FQAAAE5mSl414ygBAAAAAAAAAAAAAEEL+AplAABQ8phyBh0AAABCE3kVAAAATmZKXqWhDACAwUyZ4wMAAAChibwKAAAAJzMlr9JQBgDAZIacQQcAAIAQRV4FAACAkxmSV2koAwAAACHq1K40u0sISK1mC+0u4YK++byl3SUEpHypmnaXAAAA4MNJufSrQ1vsLsHjtqFH7S7BY9trV9ldAgAghNFQBgDAYKbM8QEAAIDQRF4FAACAk5mSV2koAwBgMJfLZXcJAAAAQK7IqwAAAHAyU/IqDWUAAAzmkhln0AEAACA0kVcBAADgZKbkVTOOEgAAAAAAAAAAAAAQNK5QBgDAYKbM8QEAAIDQRF4FAACAk5mSV2koAwBgMkPm+AAAAECIIq8CAADAyQzJqzSUAQAwmRkn0AEAACBUkVcBAADgZIbkVUMOEwAAAAAAAAAAAAAQLK5QBgDAZIbckgUAAAAhirwKAAAAJzMkr9JQBgDAZIYEHgAAAIQo8ioAAACczJC8SkMZAACTMfkFAAAAnIy8CgAAACczJK8acpgAAAAAAAAAAAAAgGBxhTIAAAazDLklCwAAAEITeRUAAABOZkpepaEMAIDJzMg7AAAACFXkVQAAADiZIXmVhjIAACYLMyTxAAAAIDSRVwEAAOBkhuRVGsoAAAAoMX744Qd9++23OnHihM9z/fr1s6EiAAAA4MI+/vhjXXPNNXaXAQAA4BcNZQAATGbIHB8o+U6dOqW+ffvq66+/Vnx8vCIivGOuy+WioQwAQCgir8IQAwcO1JYtW+wuAwAABMuQvEpDGQAAk5mRd2CAtLQ0VaxYUT/88IPCwsI84z/99JMWLVqkW2+9Nc/1MzMzlZmZ6TXmdmfJ7S5dJPUCAIAAkVdhCMuy8nyevAoAgEMZkldpKNvo1K40u0uwXb2+G+0uwXYvTb7I7hIcoXVcfbtLAMxkyBwfKPneffddrV692quZLEldu3bVX/7yF7344ovauDH33JGenq60NO9slpIyXKmpI4qkXgAAECDyKgzhusDVTeRVAAAcypC8SkMZAAAAIc/lcik7O9tn/NSpU2rUqJHeeeedPNdPTk5WUlKS15jbvatQawQAAADyi7wKAADsREMZAACTGTLHB0q+bt26afr06UpNTfUanz9/vvr376/OnTvnub7b7Zbb7T5vlNsHAgBgO/IqIIm8CgCAYxmSV2koAwBgMjPyDgzwwAMPqGfPnrrlllvUtWtXxcTE6JdfftGSJUtUp04dzZgxw+4SAQBAfpBXAQAA4GSG5FUaygAAmMyQOT5Q8pUpU0ZLlizRu+++q48++khffvmlLrnkEv3zn/9Us2bN7C4PAADkF3kVJczjjz+ulStXeo1ZlqXdu3fbVBEAACgQQ/IqDWUAAFDspk+frieffFJ79+5Vs2bN9PTTT6tly5a5Lj916lTNnDlTu3btUqVKlXTbbbcpPT1dkZGRxVg1QsFNN92km266ye4yAAAAAL+6deum+Ph4u8sAAAAICg1lAABMZsMJdAsWLFBSUpIyMjLUqlUrTZ06VQkJCdq8ebPi4uJ8ln/55Zd1//33a86cOWrTpo22bNmiAQMGyOVyacqUKcV/AAAAACg+ZlzwAYM0adJETZo0sbsMAABQWAzJq2F2FwAAAOxjuVwFfgRrypQpGjRokAYOHKhGjRopIyNDZcuW1Zw5c/wuv27dOl1zzTXq06ePateurRtuuEG9e/fWZ599VtDDBwAAgMPZkVelP++oU7t2bUVGRqpVq1YXzJ5Tp05VgwYNVKZMGdWoUUOjR4/W6dOn87VvAAAAhA5T8ioNZQAATBbmKvgjCFlZWfriiy/UqVOn/5UQFqZOnTpp/fr1ftdp06aNvvjiC08o2r59u9555x3dfPPN+T9uAAAAhIZizqvS/+6ok5KSoo0bN6pZs2ZKSEjQ/v37/S5/9o46KSkp+v777/Xcc89pwYIFeuCBBwp69AAAAHA6Q/IqDWUAAFAgmZmZOnbsmNcjMzPT77IHDx5Udna2Kleu7DVeuXJl7d271+86ffr00YQJE9S2bVuVKlVK9erVU4cOHfiCDgAAAEWCO+oAAADAyezIqzSUAQAwmavgj/T0dEVHR3s90tPTC63ENWvW6PHHH9eMGTO0ceNGLVq0SEuXLtUjjzxSaPsAAACAQxVCXg3mBEjuqAMAAICgGJJXaSgDAGAyl6vAj+TkZB09etTrkZyc7Hd3lSpVUnh4uPbt2+c1vm/fPlWpUsXvOg8//LD69u2ru+66S02aNFGPHj30+OOPKz09XTk5OYX+kgAAAMBBCiGvBnMCJHfUAQAAQFAMyas0lAEAMFkhzPHhdrtVvnx5r4fb7fa7u9KlSys+Pl6rVq3yjOXk5GjVqlVq3bq133VOnjypsDDvyBIeHi5JsiyrkF4IAAAAOFIh5NVgToDMD+6oAwAAYDBD8mpEoVUDAAAQgKSkJPXv318tWrRQy5YtNXXqVJ04cUIDBw6UJPXr10/Vq1f3nIXXrVs3TZkyRVdeeaVatWqlbdu26eGHH1a3bt08jWUAAAAgN263O9cTHs9X0DvqSFKTJk104sQJDR48WA8++KDPyZEAAADAuUIhr9JQBgDAZK7i32WvXr104MABjR8/Xnv37lXz5s21bNkyz21adu3a5RViHnroIblcLj300EPavXu3YmNj1a1bNz322GPFXzwAAACKVzHn1XPvqNO9e3dJ/7ujzvDhw/2uwx11AAAADGZIXqWhDACAyVw2dJQlDR8+PNeAs2bNGq+fIyIilJKSopSUlGKoDAAAAI5iQ17ljjoAAAAImCF5lYYyAAAms6mhDAAAAATEhrzKHXUAAAAQMEPyqsuy7d47W+zZLRylXt+Ndpdgu5cmX2R3CY7QOq6+3SUADnFZse7t0ltfKvA2tv3n74VQCeBE5NXCUqvZQrtLuKBvPm9pdwkBKV+qpt0lADAeeRVwDvKqP18dcs7rctvQo3aX4LHttavsLgEAigl5tShwhTIAACYLu/AiAAAAgG3IqwAAAHAyQ/IqDWUAAEzGLa8BFIOfNt1mdwkXVLZWmt0lBOTkT86fT95Stt0lBMQl5jUFQgJ5FUCQmscU75VpefnsxR/sLsGRLrtqhd0leGzZ0NnuEgCEOkPyKg1lAABMZkbeAQAAQKgirwIAAMDJDMmrhlyIDQAAAAAAAAAAAAAIFlcoAwBgMCvMkFPoAAAAEJLIqwAAAHAyU/IqDWUAAExmyBwfAAAACFHkVQAAADiZIXmVhjIAACYzI+8AAAAgVJFXAQAA4GSG5FXmUAYAAAAAAAAAAAAA+BVUQ/nWW28tqjoAAIAdwlwFfwAAAABFhbwKAAAAJzMkrwZ1y+uvv/5ac+bM8Rlv2LChWrdunet6mZmZyszM9Bpzu7PkdpcOZvcAAKCwGTLHBwAAAEIUeRUAAABOZkheDXoO5dWrV/uMhYeH59lQTk9PV1pamtdYSspwpaaOCHb3AACgMJmRdwAAABCqyKsAAABwMkPyalANZZfLpRdffDHonSQnJyspKclrzO3eFfR2AAAAAAAAAAAAAADFJ6iGsmVZ6tOnj894586dNXDgwFzXc7vdcrvd541yu2sAAGwXInN0AAAAwFDkVQAAADiZIXk1qIbyxIkTdezYMZ/xBg0aFFpBAACgGBkSeAAAABCiyKsAAABwMkPyalAN5R49ehRVHQAAwAaWGXkHAAAAIYq8CgAAACczJa+G2V0AAAAAAAAAAJjkl19+0W+//WZ3GQAAAAGhoQwAgMnCXAV/AAAAAEWFvIoS5plnnlGFChXUpEkT1alTR/Xr19ecOXPsLgsAAOSXIXmVhjIAACZzuQr+AAAAAIoKeRUlyLPPPqu5c+dqzZo1Onz4sI4ePao33nhDp0+ftrs0AACQX4bk1aDmUAYAACVMiJwBBwSiY8eOev/99+0uAwAAFCbyKkqQuXPn6rXXXlO9evU8Y40bN1bjxo1trAoAABSIIXmVhjIAACbjXiUoQX7++Wd17NjRZzw2NlazZs1SdHR0rutmZmYqMzPTa8ztzpLbXbrQ6wQAAEEgr6IEsSxLW7du1fDhw/X5558rKytLzZs3V1pamjp06JDnuuRVAAAcypC8SkMZAAAAJca4ceN8xg4cOKDExEQtWrRIMTExftdLT09XWlqa11hKynClpo4okjoBAABgnq5du6pLly665ZZbtHbtWsXExOj7779Xv379tGvXrjzXJa8CAAA70VAGAMBkITJHBxAIl8ulhIQEv8/Vr19fiYmJWrx4sWJjY32eT05OVlJSkteY2533l3oAAKAYkFdRguzYsUPt2rXTkiVLtHfvXtWoUUNbtmxRp06dLrgueRUAAIcyJK/SUAYAwGSGzPEBM1iW5feW14mJiRo1apTmzZunihUr+l3X7XbL7XafN8rtAwEAsB15FSXIl19+qQ8//FCHDx/WJ598oiNHjmjkyJFq1arVBdclrwIA4FCG5FUaygAAGMwy5Aw6mGHu3Lk6ceKEz3jNmjUl/XmVMgAACC3kVZQkZcuW1a+//qrLLrtMO3fu1KRJkxQREaHU1FR17tzZ7vIAAEA+mJJXaSgDAACgRGjbtq3dJQAAAAC5+utf/6rhw4dr8eLFGjlypB5++GFFRERoyJAh2r59u93lAQAA5IqGMgAAJguzuwAAAAAgD+RVlCBJSUn69ttvVadOHf3xxx9avny5Tpw4oSZNmthdGgAAyC9D8ioNZQAATGbIHB8AAAAIUeRVlCARERF6/vnn9c033+ijjz7Snj17VL9+ffXq1cvu0gAAQH4ZkldpKAMAYDJD5vgAAABAiCKvogRq0qQJVyUDAFBSGJJXDbkQGwAAAAAAAAAAAAAQLK5QBgDAZIbckgUAAAAhirwKAAAAJzMkr9JQBgDAZGbkHQAAAIQq8ioAAACczJC8SkMZAACDWYacQQcAAIDQRF4FAACAk5mSV5lDGQAAAAAAAAAAAADgF1coAwBgMkPOoAMAAECIIq8CAADAyQzJqzSUAQAwmcuMwAMAAIAQRV4FAACAkxmSV2koAwBgMia/AAAAgJORVwEAAOBkhuRVQw4TAAAAAAAAAAAAABAsrlAGAMBkhtySBQAAACGKvAoAAAAnMySv0lAGAMBkYWYEHgAAAIQo8ioAAACczJC8SkMZtvrxxb/YXYLt6v5lid0lOML2jfXtLsF2hzO32F2C7Sq4L7O7BPMYEngA4EKio2rbXUKJcerMQbtLCEjZiMp2lwAgEORVACGsovtyu0twpAGznfP9T7OX9tldgsf6vzlnhtKyEbF2lwCEDkPyqnM+oQAAAAAAAAAAAAAAjsIVygAAGMwyZI4PAAAAhCbyKgAAAJzMlLxKQxkAAJNxrxIAAAA4GXkVAAAATmZIXqWhDACAyQw5gw4AAAAhirwKAAAAJzMkrxrSNwcAAAAAAAAAAAAABIsrlAEAMFmYGWfQAQAAIESRVwEAAOBkhuRVGsoAAJjMkMADAACAEEVeBQAAgJMZkldpKAMAYDIz8g4AAABCFXkVAAAATmZIXmUOZQAAAAAAAAAAAACAX1yhDACAwSxDbskCAACA0EReBQAAgJOZkldpKAMAYDKXGYEHAAAAIYq8CgAAACczJK/SUAYAwGSGnEEHAACAEEVeBQAAgJMZkleZQxkAAAAAAAAAAAAA4BdXKAMAYDIzTqADAABAqCKvAgAAwMkMyas0lAEAMFgY9yoBAACAg5FXAQAA4GSm5FUaygAAGMxl0xl006dP15NPPqm9e/eqWbNmevrpp9WyZctclz9y5IgefPBBLVq0SL/99ptq1aqlqVOn6uabby7GqgEAAFDc7MqrQFFZvny5FixY4Pe5OXPmFHM1AACgoEzJqzSUAQBAsVqwYIGSkpKUkZGhVq1aaerUqUpISNDmzZsVFxfns3xWVpY6d+6suLg4LVy4UNWrV9dPP/2kiy++uPiLh6M0btxYw4YN07Bhwwq8rczMTGVmZnqNud1ZcrtLF3jbAAAAwFlVq1ZVq1atfMZTUlLyXI+8CgAA7GTIhdgAAMAfl6vgj2BNmTJFgwYN0sCBA9WoUSNlZGSobNmyuZ6NP2fOHP32229avHixrrnmGtWuXVvt27dXs2bNCnj0CHVpaWlKSEgolG2lp6crOjra65Ge/myhbBsAAOSfHXlV+vOOOrVr11ZkZKRatWqlzz77LM/ljxw5onvuuUdVq1aV2+3WZZddpnfeeSd/O0eJ1rRpUw0ZMsTncdFFF+W5HnkVAABnMiWvcoUyAAAGcxXzPVmysrL0xRdfKDk52TMWFhamTp06af369X7Xeeutt9S6dWvdc889evPNNxUbG6s+ffpo3LhxCg8PL67S4UC33nproW0rOTlZSUlJXmNu965C2z4AAMif4s6rEnfUgT0u9LtOXgUAwJlMyas0lAEAMFhh5B3/t15zy+12+yx78OBBZWdnq3Llyl7jlStX1g8//OB3+9u3b9f777+vO+64Q++88462bdumYcOG6Y8//rjgbeGAQPn/neX2gQAA2M2OOenOvaOOJGVkZGjp0qWaM2eO7r//fp/lz95RZ926dSpVqpQkqXbt2sVZMgxAXgUAwJlMyavc8hoAABSI/1uvpRfa9nNychQXF6dZs2YpPj5evXr10oMPPqiMjIxC2wcAAABKrszMTB07dszrcf4JkWedvaNOp06dPGPB3FGncuXKaty4sR5//HFlZ2cXyfEgtL322mvq2LGj1+O6667T7t277S4NAADYJBTyKlcoAwBgsMI4g87/rdd8r06WpEqVKik8PFz79u3zGt+3b5+qVKnid52qVauqVKlSXre3btiwofbu3ausrCyVLs1Z+QAAACVVYeTV9PR0paWleY2lpKQoNTXVZ1nuqIOi1qpVK0VHR9tdBgAAKCSm5FUaygAAGMxVCPcqye321v6ULl1a8fHxWrVqlbp37y7pzyuQV61apeHDh/td55prrtHLL7+snJwchYX9WfCWLVtUtWpVmskAAAAlXGHk1WBOgMyPc++oEx4ervj4eO3evVtPPvkkDWX4qFWrlmrVqmV3GQAAoJCYkldpKAMAYDA75vhISkpS//791aJFC7Vs2VJTp07ViRMnPHN+9OvXT9WrV/fcNnvo0KF65plnNHLkSI0YMUJbt27V448/rnvvvbf4iwcAAECxKoy8GswJkNxRBwAAAMEwJa8G1Tf/4IMPNHz4cD300EP66quvJEkff/yxOnTooC5duujLL78MZnMAAMBAvXr10qRJkzR+/Hg1b95cX331lZYtW+a5TcuuXbu0Z88ez/I1atTQ8uXLtWHDBjVt2lT33nuvRo4cqfvvv9+uQwAAAEAJde4ddc46e0ed1q1b+13nmmuu0bZt25STk+MZ4446AAAAKAp25dWAr1D+5ptvNGjQIN1333367bffdNttt2nSpEl65JFH9PTTT+vYsWMaNmxYrhM+AwAA5wmz4QplSRo+fHiut7hes2aNz1jr1q31ySefFHFVAAAAcBo78ip31AEAAECgTMmrATeU161bp0GDBmnw4MGSpISEBI0dO1YRERHKycnRpk2bcr2UOjMzU5mZmV5jbneW3G7O0gQAwE523PIaAAAACJQdebVXr146cOCAxo8fr71796p58+Y+d9QJC/vfTf/O3lFn9OjRatq0qapXr66RI0dq3LhxxV88AAAAipUpeTXghvJVV12lvn37qmbNmqpSpYratWunPXv2aN68ebrhhhvUtWtXvfjii37XTU9PV1pamtdYSspwpaaOCLhQAABQ+GgoAwAAwMnsyqvcUQcAAACBMCWvBtxQbtasmSIiIjR69GjVqlVL69atU05Ojq666iqtXbtWPXv21I4dO9SkSROfdZOTk5WUlOQ15nbvynfRAAAAAAAAAAAAAICiF3bhRf70xhtvqEmTJtq0aZPuvvtu/fLLL57nmjZtqueee06jRo3yu67b7Vb58uW9HtzuGgAA+7lcrgI/AAAAgKJCXgUAAICTmZJXA24onz59WpdeeqkGDx6szz77TLfccotiY2O1YcMGSVKtWrW0d+/eIisUAAAUPldYwR8AAABAUSGvAgAAwMlMyasB3/I6MTFRV199tX755Rf16dNHZcqU0WOPPaYuXbqoVq1a+uWXX/TYY48VZa0AAKCQhcgJcAAAADAUeRUAAABOZkpeDbihXL58ea1fv16vvPKKMjMz9dhjjykmJkbbtm3T119/rerVq6tOnTpFWSsAAAAAAAAAAAAAoBgF3FCWpOjoaN19991eY+XLl1fbtm0LtSgAAFA8TDmDDgAAAKGJvAoAAAAnMyWvBtVQBgAAJYspgQcAAAChibwKAAAAJzMlr9JQBgDAYGGGBB4AAACEJvIqAAAAnMyUvEpDGQAAAIDxdn79V7tLKDFqN15sdwkB2ffDYLtLuCCXDPlmAgAAGOWB5rXsLsHjnkbb7S7B4/LpZewuwWNhn612l+ARX8k5vy/hrtJ2lwDYhoYyAAAGM+WWLAAAAAhN5FUAAAA4mSl5lYYyAAAGMyXwAAAAIDSRVwEAAOBkpuRVGsoAABjMZcokHwAAAAhJ5FUAAAA4mSl5NczuAgAAAAAAAAAAAAAAzsQVygAAGMyUW7IAAAAgNJFXAQAA4GSm5FUaygAAGMyUwAMAAIDQRF4FAACAk5mSV2koAwBgMFMCDwAAAEITeRUAAABOZkpeZQ5lAAAAAAAAAAAAAIBfXKEMAIDBwgw5gw4AAAChibwKAAAAJzMlr9JQBgDAYKbckgUAAAChibwKAAAAJzMlr9JQBgDAYC4mvwAAAICDkVcBAADgZKbkVRrKAAAYzJQz6AAAABCayKsAAABwMlPyqiF9cwAAAAAAAAAAAABAsLhCGQAAg7lMOYUOAAAAIYm8CgAAACczJa/SUAYAwGCG5B0AAACEKPIqAAAAnMyUvEpDGQAAg5kSeAAAABCayKsoSX744Qd9++23OnHihM9z/fr1s6EiAABQUKbkVRrKAAAAKDGWL1+uBQsW+H1uzpw5xVwNAAAAIJ06dUp9+/bV119/rfj4eEVEeH8l63K5aCgDAABHo6EMAIDBTDmDDuaoWrWqWrVq5TOekpKS53qZmZnKzMz0GnO7s+R2ly7U+gAAQHDIqygJ0tLSVLFiRf3www8KCwvzjP/0009atGiRbr311gtug7wKAIAzmZJXaSgDNvt+Q1u7S4BDXHXNj3aXYLuvP6lgdwm2KxtxWbHuL8yQwANzNG3aVE2bNvUZnzRpUp7rpaenKy0tzWssJWW4UlNHFGp9AAAgOORVlATvvvuuVq9e7dVMlqSuXbvqL3/5i1588UVt3Lgxz22QVwEAcCZT8ioNZQAADGZK4AFcFzhdNDk5WUlJSV5jbveuoiwJAAAEgLyKksDlcik7O9tn/NSpU2rUqJHeeeedC26DvAoAgDOZklfDLrwIAAAAULK53W6VL1/e68HtAwEAAFAYunXrpunTp/uMz58/X3PnzlXnzp0vuA3yKgAAsBNXKAMAYLAwl2V3CUCheu2115SRkeE1ZlmWdu/ebVNFAACgIMirKAkeeOAB9ezZU7fccou6du2qmJgY/fLLL1qyZInq1KmjGTNm2F0iAADIJ1PyKg1lAAAMZsotWWCOVq1aKTo62u4yAABAISGvoiQoU6aMlixZonfffVcfffSRvvzyS11yySX65z//qWbNmtldHgAAKABT8ioNZQAADMbcFyhpatWqpVq1atldBgAAKCTkVZQkN910k2666Sa7ywAAAIXIlLxqynECAAAAAAAAAAAAAILEFcoAABjMlDk+AAAAEJrIqwAAAHAyU/IqDWUAAAxmyhwfAAAACE3kVQAAADiZKXmVhjIAAAZj7gsAAAA4GXkVAAAATmZKXjXlOAEAAAAAAAAAAAAAQeIKZQAADGbKLVkAAAAQmsirAAAAcDJT8ioNZQAADOZyWXaXAAAAAOSKvAoAAAAnMyWv0lAGAMBgppxBBwAAgNBEXgUAAICTmZJXmUMZAAAAAAAAAAAAAOAXVygDAGAwziwDAACAk5FXAQAA4GSm5FUaygAAGCzMkDk+AAAAEJrIqwAAAHAyU/IqDWUAAAxmyhwfAAAACE3kVQAAADiZKXnVlCuxAQAAAAAAAAAAAABB4gplAAAMxpllAPAnd/jFdpdQYlw5rYfdJQSk07sH7S7hglbcWMHuEi4ozMXXCiha5FUAQFGKLl3X7hI8Prpzs90leEz6JsruEjze+2Wv3SV4jG0abXcJHu5w59RiOlPyKv/yAwDAYKbckgUAAAChibwKAAAAJzMlr9JQBgDAYGEuy+4SAAAAgFyRVwEAAOBkpuRVU67EBgAAAAAAAAAAAAAEiSuUAQAwmCm3ZAEAAEBoIq8CAADAyUzJqzSUAQAwGLcqAQAAgJORVwEAAOBkpuRVGsoAABjMlDk+AAAAEJrIqwAAAHAyU/JqwI3zjh07FmUdAAAAAAAAAAAAAACHCfgK5Z9//tlvUzk2NlazZs1SdHR0rutmZmYqMzPTa8ztzpLbXTqIUgEAQGEzZY4PAAAAhCbyKgAAAJzMlLwa1C2vx40b5zN24MABJSYmatGiRYqJifG7Xnp6utLS0rzGUlKGKzV1RDC7BwAAhcyUwAMAAIDQRF4FAACAk5mSVwNuKLtcLiUkJPh9rn79+kpMTNTixYsVGxvr83xycrKSkpK8xtzuXUGWCgAAClvAc18UsunTp+vJJ5/U3r171axZMz399NNq2bLlBdd79dVX1bt3b/31r3/V4sWLi75QAAAA2MquvAoAAAAEwpS8GvBxWpaljh07+jymTp2qVq1aad68eapYsaLfdd1ut8qXL+/14HbXAACYacGCBUpKSlJKSoo2btyoZs2aKSEhQfv3789zvZ07d2rMmDFq165dMVUKAAAAU02fPl21a9dWZGSkWrVqpc8++yyg9V599VW5XC517969aAsEAACA0Yo7rwZ8hfLcuXN14sQJn/GaNWtK+vMqZQAAEFrCXFax73PKlCkaNGiQBg4cKEnKyMjQ0qVLNWfOHN1///1+18nOztYdd9yhtLQ0ffjhhzpy5EgxVgwAAAC72JFXz54AmZGRoVatWmnq1KlKSEjQ5s2bFRcXl+t6nAAJAABgHlPyasBXKLdt21YJCQk+j4YNGwa9UwAA4AxhroI/MjMzdezYMa9HZmam3/1lZWXpiy++UKdOnf5XQ1iYOnXqpPXr1+da54QJExQXF6c777yz0F8DAAAAOFdh5NVgnXsCZKNGjZSRkaGyZctqzpw5ua5z7gmQdevWLcARAwAAIJSYkldNubU3AADwI6wQHunp6YqOjvZ6pKen+93fwYMHlZ2drcqVK3uNV65cWXv37vW7zkcffaTnnntOs2fPLujhAgAAIMQURl7lBEgAAAAUFVPyKg1lAABQIMnJyTp69KjXIzk5uVC2/fvvv6tv376aPXu2KlWqVCjbhJk+/vhju0sAAAA24QRIAAAAOFko5NWA51AGAAAlT35uqXI+t9stt9sd0LKVKlVSeHi49u3b5zW+b98+ValSxWf5H3/8UTt37lS3bt08Yzk5OZKkiIgIbd68WfXq1StA9TDFwIEDtWXLFrvLAAAAQSqMvJqcnKykpCSvsUDz64VwAiQAAIDZTMmrNJQBADCYy2UV6/5Kly6t+Ph4rVq1St27d5f0Z4N41apVGj58uM/yl19+ub755huvsYceeki///67pk2bpho1ahRH2SgBLKt4f9cBAEDhKIy8ygmQcIqOHTvq/ffft7sMAABQiEzJqzSUAQAwWGGcQRespKQk9e/fXy1atFDLli01depUnThxQgMHDpQk9evXT9WrV1d6eroiIyPVuHFjr/UvvvhiSfIZB/LicuX9y56ZmekzN43bnSW3u3RRlgUAAC6guPMqJ0CiKP3888/q2LGjz3hsbKxmzZql6OjoXNclrwIA4Eym5FUaygAAoFj16tVLBw4c0Pjx47V37141b95cy5Yt88z7sWvXLoWFhdlcJUyTnp6utLQ0r7GUlOFKTR1hU0UAAMAunACJojRu3DifsQMHDigxMVGLFi1STEyM3/XIqwAA4Cw78ioNZQAADGZX23b48OF+z5iTpDVr1uS57rx58wq/IBjP/1w1u2yqBgAAnGVHXuUESBQVl8ulhIQEv8/Vr19fiYmJWrx4sWJjY32eJ68CAOBMpuRVGsoAABgsrJjnUAacyv9cNdw+EAAAu9mVVzkBEkXBsiy/t7xOTEzUqFGjNG/ePFWsWNHvuuRVAACcyZS8SkMZAACD2TGHMlCUHn/8ca1cudJrzLIs7d6926aKAABAQZBXUZLMnTtXJ06c8BmvWbOmpD+vUgYAAKHFlLxKQxkAAAAlRrdu3RQfH293GQAAAICPtm3b2l0CAABAvtBQBgDAYKacQQdzNGnSRE2aNLG7DAAAUEjIqwAAAHAyU/IqDWUAAAwWbncBAAAAQB7IqwAAAHAyU/IqDWUAAAwW5rLsLgEAAADIFXkVAAAATmZKXg2zuwAAAAAAAAAAAAAAgDNxhTIAAAYzZY4PAAAAhCbyKgAAAJzMlLxKQxkAAIOZEngAAAAQmsirAAAAcDJT8ioNZQAADBZuSOABAABAaCKvAgAAwMlMyavMoQwAAAAAAAAAAAAA8IsrlAEAMJgpt2QBAABAaCKvAgAAwMlMyas0lAEAMFiYy7K7BAAAACBX5FUAAAA4mSl5lYYyAAAGM+UMOgAAAIQm8ioAAACczJS8yhzKAAAAAAAAAAAAAAC/uEIZAACDhdtdAAAAAJAH8ioAAACczJS8SkMZAACDmXJLFgAAAIQm8ioAAACczJS8SkMZsJk7/GK7S4BDvLbsUrtLsN3/bciyuwTbPdO6ePcX5rKKd4cAgBJveUKc3SUEZO+pH+wu4YJave78v6cntTtmdwkBaV25st0lXFDpsIvsLsGRyKsAAFPUjGpgdwkej7f4xe4SPO5eV8ruEjwyfjhsdwkeA+r/ZncJXsqXrml3CR7F3d81Ja8yhzIAAAAAAAAAAAAAwC+uUAYAwGDhhtySBQAAAKGJvAoAAAAnMyWv0lAGAMBgpszxAQAAgNBEXgUAAICTmZJXaSgDAGAwUwIPAAAAQhN5FQAAAE5mSl6loQwAgMFMCTwAAAAITeRVAAAAOJkpeTXM7gIAAAAAAAAAAAAAAM7EFcoAABgs3GXZXQIAAACQK/IqAAAAnMyUvEpDGQAAg3GrEgAAADgZeRUAAABOZkpepaEMAIDBTJnjAwAAAKGJvAoAAAAnMyWvmtI4BwAAAAAAAAAAAAAEiSuUAQAwmCln0AEAACA0kVcBAADgZKbkVRrKAAAYLNxl2V0CAAAAkCvyKgAAAJzMlLxKQxkAAIOZcgYdAAAAQhN5FQAAAE5mSl5lDmUAAAAAAAAAAAAAgF9coQwAgMFMOYMOAAAAoYm8CgAAACczJa/SUAYAwGCmBB4AAACEJvIqAAAAnMyUvMotrwEAMFi4q+APwCluvfVWu0sAAACFjLyKkoS8CgBAyWNKXuUKZQAAAJQIX3/9tebMmeMz3rBhQ7Vu3TrPdTMzM5WZmek15nZnye0uXag1AgAAwFzkVQAAEKpoKAMAYLAwl2V3CUChWr16tc9YeHj4Bb+gS09PV1pamtdYSspwpaaOKNT6AABAcMirKGnIqwAAlCym5FUaygAAGIy5L1CSuFwuvfjii/laNzk5WUlJSV5jbveuwigLAAAUAHkVJQl5FQCAkseUvEpDGQAAg4WFyBwdQCAsy1KfPn18xjt37qyBAwfmua7b7Zbb7T5vlNsHAgBgN/IqShLyKgAAJY8peZWGMgAAAEqEiRMn6tixYz7jDRo0sKEaAAAAwBt5FQAAhCoaygAAGCzckDPoYIYePXrYXQIAAChk5FWUJORVAABKHlPyKg1lAAAMFuay7C4BAAAAyBV5FQAAAE5mSl6loQwAgMFMmeMDAAAAoYm8CgAAACczJa+G2V0AAAAAAAAAAAAAAMCZcm0oN27cWDNmzCiUnWRmZurYsWNej8zMrELZNgAAyL8wV8EfAAAAQFEhrwIAAMDJTMmruTaU09LSlJCQUCg7SU9PV3R0tNcjPf3ZQtk2AADIv7BCeAAAAABFhbwKAAAAJzMlr+Y6h/Ktt95aaDtJTk5WUlKS15jbvavQtg8AAPLHFSJnwAEAAMBM5FUAAAA4mSl5NdeGcmFyu91yu93njZYujl0DAAAAAAAAAAAAAPKpWBrKAADAmQw5gQ4AAAAhirwKAAAAJzMlr9JQBgDAYKbckgUAAAChibwKAAAAJzMlr9JQBgDAYGF2FwAAAADkgbwKAAAAJzMlr5pynAAAAAAAAAAAAACAIHGFMgAABnO5LLtLAAAAAHJFXgUAAICTmZJXaSgDAGAwQ6b4AAAAQIgirwIAAMDJTMmrNJQBADCYy5TEAwDAeaqUudzuEi7o055n7C7hgrb/ftjuEgLSaelpu0u4oPSWe+0uISDXVL6sWPdHXgUAoPhFlbrE7hI85rU7ZXcJHl8e+sXuEjwW7nBWe+9k9s92l+AxolHDYt2fKXmVOZQBAAAAAAAAAAAAAH456xQGAABQrAw5gQ4AAAAhirwKAAAAJzMlr3KFMgAABgtzFfyRH9OnT1ft2rUVGRmpVq1a6bPPPst12dmzZ6tdu3aqUKGCKlSooE6dOuW5PAAAAEoOu/IqAAAAEAhT8ioNZQAADOYqhEewFixYoKSkJKWkpGjjxo1q1qyZEhIStH//fr/Lr1mzRr1799bq1au1fv161ahRQzfccIN2796dj70DAAAglNiRVyVOgAQAAEBgTMmrNJQBAECxmjJligYNGqSBAweqUaNGysjIUNmyZTVnzhy/y8+fP1/Dhg1T8+bNdfnll+vf//63cnJytGrVqmKuHAAAACbgBEgAAAA4mR15lYYyAAAGc7kK/ghGVlaWvvjiC3Xq1MkzFhYWpk6dOmn9+vUBbePkyZP6448/VLFixeB2DgAAgJBT3HlV4gRIAAAABM6UvBoRfJkAAKCkKIwpOjIzM5WZmek15na75Xa7fZY9ePCgsrOzVblyZa/xypUr64cffghof+PGjVO1atW8mtIAAAAomYo7r549ATI5OdkzxgmQAAAAyI0peZUrlAEAMFhhzPGRnp6u6Ohor0d6enqR1PvPf/5Tr776qt544w1FRkYWyT4AAADgHMWdV/M6AXLv3r0B1cwJkAAAAOYwJa9yhTIAACiQ5ORkJSUleY35O3tOkipVqqTw8HDt27fPa3zfvn2qUqVKnvuZNGmS/vnPf2rlypVq2rRpwYoGAACAMYLJqwV19gTINWvWcAIkAAAAAhIKeZWGMgAABgsrhHuy5Hb7FX9Kly6t+Ph4rVq1St27d5ckz3wdw4cPz3W9iRMn6rHHHtPy5cvVokWLghcNAACAkFDceZUTIAEAABAMU/Iqt7wGAMBghXFLlmAlJSVp9uzZev755/X9999r6NChOnHihAYOHChJ6tevn9ccIE888YQefvhhzZkzR7Vr19bevXu1d+9eHT9+PJ9HDQAAgFBR3Hn13BMgzzp7AmTr1q1zXW/ixIl65JFHtGzZMk6ABAAAMIgpeZUrlAEAMJjLZRX7Pnv16qUDBw5o/Pjx2rt3r5o3b65ly5Z55v3YtWuXwsL+d87bzJkzlZWVpdtuu81rOykpKUpNTS3O0gEAAFDM7MirSUlJ6t+/v1q0aKGWLVtq6tSpPidAVq9e3TOv3RNPPKHx48fr5Zdf9pwAKUlRUVGKiooq9voBAABQfEzJqzSUAQBAsRs+fHiut7hes2aN1887d+4s+oIAAACA/48TIAEAAOBkduRVGsoAABisEKb4AAAAAIqMXXmVEyABAAAQCFPyKg1lAAAM5qKjDAAAAAcjrwIAAMDJTMmrNJQBADBY2IUXAULOe++9p59//llRUVFq0aKF6tWrZ3dJAAAgn8irKInIqwAAlBym5FVTjhMAAAAl1Ntvv+3573/84x+aMGGCPvjgAy1YsEBt27bV888/b2N1AAAAMB15FQAAhDquUAYAwGCm3JIFJdt9992nbt26SZJWrFihXbt2yfX/f7m3bdumfv36qX///nluIzMzU5mZmV5jbneW3O7SRVM0AAAICHkVJQF5FQCAksuUvEpDGQAc4i+V6ttdgu3iymy2uwTjGJJ3UMJZlqWwsP/deKdUqVKqW7eupk2bpptuukkHDx684DbS09OVlpbmNZaSMlypqSMKvV4AABA48ipKAvIqAAAllyl5lYYyAAAGM+UMOpRsLpdLp0+fVk5Ojho3bqyvv/5an376qYYNG6abbrpJlmVdcBvJyclKSkryGnO7dxVVyQAAIEDkVZQE5FUAAEouU/IqDWUAAACENMuy1KFDB8/PZcuWVfPmzXXq1ClJUnh4+AW34Xa75Xa7zxvl9oEAAAAoOPIqAAAIdTSUAQAwmCEn0KGEmz9/vk6dOiXLsjR27FgNGDBA69at07hx4yRJ1atX1+zZs9WrVy+VL1/e5moBAEAwyKsoCcirAACUXKbkVRrKAAAYLMyUxIMSrWXLlp7/fvvtt/X2229r2LBhnvE5c+Zo+PDhys7O1t13321XmQAAIB/IqygJyKsAAJRcpuRVGsoAABjMkLwDg1SpUkWDBg3yGqtVq5befvttmyoCAAAFQV5FSUNeBQCgZDElr4bZXQAAAAAAAAAAAAAAwJm4QhkAAIO5XJbdJQAAAAC5Iq8CAADAyUzJqzSUAQAwmCm3ZAEAAEBoIq8CAADAyUzJqzSUAQAwmMuUxAMAAICQRF4FAACAk5mSV5lDGQAAAAAAAAAAAADgF1coAwBgMENOoAMAAECIIq8CAADAyUzJqzSUAQAwGLcqAQAAgJORVwEAAOBkpuRVGsoAABjMlDk+AAAAEJrIqwAAAHAyU/KqKY1zAAAAAAAAAAAAAECQuEIZAACjGXIKHQAAAEIUeRUAAABOZkZepaEMAIDBXIYEHgAAAIQm8ioAAACczJS8SkMZAACDuVzMfgEAAADnIq8CAADAyUzJq2YcJQAAAAAAAAAAAAAgaFyhDACA0cy4JQsAAKEozOX8f7JfWr6B3SUEZHXXLLtLuKA9J3+zuwSHIq8CAGCyiLAydpfgcVVsfbtL8Ggec9LuErz8dPxnu0uwkRl51fn/OgUAAEXGlDk+AAAAEJrIqwAAAHAyU/IqDWUAAIxmRuABAABAqCKvAgAAwMnMyKs0lAEAMJjLFWZ3CQAAAECuyKsAAABwMlPyqhlHCQAAAAAAAAAAAAAIGlcoAwBgNDNuyQIAAIBQRV4FAACAk5mRV2koAwBgMJchgQcAAAChibwKAAAAJzMlr9JQBgDAYKYEHgAAAIQm8ioAAACczJS8yhzKAAAAAAAAAAAAAAC/uEIZAACjcW4ZAAAAnIy8CgAAACczI6/SUAYAwGAulxm3ZAEAAEBoIq8CAADAyUzJqzSUAQAwmhmBBwAAAKGKvAoAAAAnMyOvmnEdNgAAAAAAAAAAAAAgaFyhDACAwVyGnEEHAACA0EReBQAAgJOZkldpKAMAYDRuVgIAAAAnI68CAADAyczIqzSUAQAwmCln0AEAACA0kVcBAADgZKbkVTPa5gAAAAAAAAAAAACAoHGFMgAABnO5zDiDDgAAAKGJvAoAAAAnMyWv0lAGAMBoZgQeAAAAhCryKgAAAJzMjLxKQxkAAIO5mP0CAAAADkZeBQAAgJOZklfNOEoAAAAAAAAAAAAAQNC4QhkAAKOZcUsWAAAAhCryKgAAAJzMjLxKQxkAAIO5XGYEHgAAAIQm8ioAAACczJS8SkMZAACjmRF4AAAAEKrIqwAAAHAyM/IqcygDAAAAAAAAAAAAAPziCmUAAAzm4twyAAAAOBh5FQAAAE5mSl6loQwAgNHMuCULAAAAQhV5FQAAAE5mRl6loQwAgMFchgQeAAAAhCbyKgAAAJzMlLxqxnXYAAAAAAAAAAAAAICgcYUyAAAGc7nMOIMOAAAAoYm8CgAAACczJa/SUAYAwGjcrAQAAABORl4FAACAk5mRV2koAwBgMFPm+AAAAEBoIq8CAADAyUzJq2a0zQEAAAAAAAAAAAAAQeMKZQAAjGbGGXQAAAAIVeRVAAAAOJkZeZUrlAEAMJjL5SrwIz+mT5+u2rVrKzIyUq1atdJnn32W5/Kvv/66Lr/8ckVGRqpJkyZ655138rVfAAAAhBbyKgAAAJzMlLxKQxkAAKOFFcIjOAsWLFBSUpJSUlK0ceNGNWvWTAkJCdq/f7/f5detW6fevXvrzjvv1Jdffqnu3bure/fu+u9//xv0vgEAABBqyKsAAABwMjPyqsuyLCvoSgvFFnt2CwBwrF9ObLa7BNtdUq5bse7PUsFfc5caBLV8q1atdNVVV+mZZ56RJOXk5KhGjRoaMWKE7r//fp/le/XqpRMnTmjJkiWesauvvlrNmzdXRkZGwYoH8kReBYCSItvKsruEC9pzcofdJQSEvEpehZOQVwHAVH/knLS7BC8/Hf/Z7hI8Li1PXj1fYeRVrlAGAMBgrkL4X2Zmpo4dO+b1yMzM9Lu/rKwsffHFF+rUqZNnLCwsTJ06ddL69ev9rrN+/Xqv5SUpISEh1+UBAABQcpBXAQAA4GTG5FXLUKdPn7ZSUlKs06dP212KbXgNeA0si9fAsngNLIvXwLJ4DQoiJSXFkuT1SElJ8bvs7t27LUnWunXrvMbHjh1rtWzZ0u86pUqVsl5++WWvsenTp1txcXGFUj9QXELhc4YaC08o1BkKNVpWaNQZCjVaVmjUGQo1WlZo1BkKNRYX8irgn5M+J6jF+bVYlrPqoRZqoRZqKUlCIa/aeMtrex07dkzR0dE6evSoypcvb3c5tuA14DWQeA0kXgOJ10DiNSiIzMxMnzPm3G633G63z7K//vqrqlevrnXr1ql169ae8f/7v//T2rVr9emnn/qsU7p0aT3//PPq3bu3Z2zGjBlKS0vTvn37CvFIgKIVCp8z1Fh4QqHOUKhRCo06Q6FGKTTqDIUapdCoMxRqLC7kVcA/J31OUIvza3FaPdRCLdRCLSVJKOTViEAPBgAAwJ/cwo0/lSpVUnh4uE9Q2bdvn6pUqeJ3nSpVqgS1PAAAAHAu8ioAAACcLBTyKnMoAwCAYlO6dGnFx8dr1apVnrGcnBytWrXK64y6c7Vu3dpreUlasWJFrssDAAAA+UVeBQAAgJPZlVe5QhkAABSrpKQk9e/fXy1atFDLli01depUnThxQgMHDpQk9evXT9WrV1d6erokaeTIkWrfvr0mT56sLl266NVXX9Xnn3+uWbNm2XkYAAAAKKHIqwAAAHAyO/KqsQ1lt9utlJSUgC8hL4l4DXgNJF4DiddA4jWQeA2KU69evXTgwAGNHz9ee/fuVfPmzbVs2TJVrlxZkrRr1y6Fhf3vJipt2rTRyy+/rIceekgPPPCA6tevr8WLF6tx48Z2HQKQL6HwOUONhScU6gyFGqXQqDMUapRCo85QqFEKjTpDoUanIq/CFE76nKAW59ciOaseaqEWaqEWk9mRV12WZVmFfiQAAAAAAAAAAAAAgJDHHMoAAAAAAAAAAAAAAL9oKAMAAAAAAAAAAAAA/KKhDAAAAAAAAAAAAADwi4YyjDdgwAB1797d7jIA2KRDhw4aNWqU3WUAAIAA8Pc27MTvHwCYic///+G1AAqXXX+m+LOM/IiwuwDAbtOmTZNlWXaXAQAAAADG69Chg5o3b66pU6faXQoAADjPokWLVKpUKbvLAADYgIaypKysLJUuXdruMmCT6Ohou0sAHIfPRQBAqODvLAAAAKB4VKxY0e4SEAT+rQSgMBl5y+sOHTpo+PDhGjVqlCpVqqSEhAS7SypWy5YtU9u2bXXxxRcrJiZGXbt21Y8//mh3WbYx9ZbXv//+u+644w6VK1dOVatW1b/+9S8jb3XBn4c/mf65mJOTo//7v/9TxYoVVaVKFaWmptpdEoAS4oUXXlBMTIwyMzO9xrt3766+ffvaVJV/OTk5Sk9PV506dVSmTBk1a9ZMCxcutLssH07/O+tsfcOHD1d0dLQqVaqkhx9+2HF3xAmFDHTixAn169dPUVFRqlq1qiZPnmx3SX6FwmsZCjVKf/7bbO3atZo2bZpcLpdcLpd27txpd1leQiU3hsp7DsAeHTp00L333uuYzzMnfWY5KX846XU5y+7vDp2WtZ32byin/FvJSb+7TqrFSZ8vkrNyrZPeJ+TOyIayJD3//PMqXbq0Pv74Y2VkZNhdTrE6ceKEkpKS9Pnnn2vVqlUKCwtTjx49lJOTY3dpKEZJSUn6+OOP9dZbb2nFihX68MMPtXHjRrvLKnb8efgfkz8Xn3/+eZUrV06ffvqpJk6cqAkTJmjFihV2lwWgBOjZs6eys7P11ltvecb279+vpUuX6h//+IeNlflKT0/XCy+8oIyMDH377bcaPXq0/v73v2vt2rV2l+bD6X9nPf/884qIiNBnn32madOmacqUKfr3v/9td1leQiEDjR07VmvXrtWbb76p9957T2vWrHFkXg2F1zIUapT+nI6odevWGjRokPbs2aM9e/aoRo0adpflJVRyY6i85wDs46TPMyd9ZjkpfzjpdXESJ2VtJ/4bygn/VnLS766TanHS54vE3wPIB8tA7du3t6688kq7y3CMAwcOWJKsb775xu5SbNG/f3/rr3/9q91lFKtjx45ZpUqVsl5//XXP2JEjR6yyZctaI0eOtK8wBzD1z4PJn4vt27e32rZt6zV21VVXWePGjbOpIgAlzdChQ62bbrrJ8/PkyZOtunXrWjk5OTZW5e306dNW2bJlrXXr1nmN33nnnVbv3r1tqso/p/+d1b59e6thw4Ze7++4ceOshg0b2ljVhTktA/3+++9W6dKlrddee80zdujQIatMmTKOz6tOey39cXKN7du3d+x7HMq50cnvOYDi5/TPM7s+s5yeP5zwWW7339NOytpO/DeUU/+t5ITfXbtrcdrnC38PID+MvUI5Pj7e7hJss3XrVvXu3Vt169ZV+fLlVbt2bUnSrl277C0MxWb79u36448/1LJlS89YdHS0GjRoYGNV9uDPw/+Y/LnYtGlTr5+rVq2q/fv321QNgJJm0KBBeu+997R7925J0rx58zRgwAC5XC6bK/ufbdu26eTJk+rcubOioqI8jxdeeMGRt5ly+t9ZV199tdf727p1a23dulXZ2dk2VuXN6Rnoxx9/VFZWllq1auUZq1ixoiPzqtNfSyk0agwVoZIbec8BXIiTPs+c8pnltPzhlNfFaZyStZ36bygn/FvJSb+7TqnFaZ8vEn8PIHgRdhdgl3Llytldgm26deumWrVqafbs2apWrZpycnLUuHFjZWVl2V0aUOz48/A/Jn8ulipVyutnl8vFLVUAFJorr7xSzZo10wsvvKAbbrhB3377rZYuXWp3WV6OHz8uSVq6dKmqV6/u9Zzb7bajpDyZ/HdWYSEDFZ5QeC1DocZQESq5kfccwIU46fOMzyz/eF2czan/hnLCv5Wc9LvrpFqchr8HECxjG8qmOnTokDZv3qzZs2erXbt2kqSPPvrI5qpQ3OrWratSpUppw4YNqlmzpiTp6NGj2rJli6699lqbqys+/HkAABSXu+66S1OnTtXu3bvVqVMnx80J2qhRI7ndbu3atUvt27e3u5yQ9+mnn3r9/Mknn6h+/foKDw+3qSJvoZCB6tWrp1KlSunTTz/15NXDhw9ry5YtjvodDYXXMhRqPFfp0qUddTV/KAq19xyA2Zz0meWk/OGk18VpnJK1+TeUf0763XVSLU76fHEaJ71PyBsNZcNUqFBBMTExmjVrlqpWrapdu3bp/vvvt7ssFLOLLrpI/fv319ixY1WxYkXFxcUpJSVFYWFhjrr9ZlHjzwMAoLj06dNHY8aM0ezZs/XCCy/YXY6Piy66SGPGjNHo0aOVk5Ojtm3b6ujRo/r4449Vvnx59e/f3+4SQ8quXbuUlJSkIUOGaOPGjXr66ac1efJku8vyCIUMFBUVpTvvvFNjx45VTEyM4uLi9OCDDyoszFmzNoXCaxkKNZ6rdu3a+vTTT7Vz505FRUWpYsWKjnvfnS7U3nMAZnPSZ5aT8oeTXhencUrW5t9Q/jnpd9dJtTjp88VpnPQ+IW/8thomLCxMr776qr744gs1btxYo0eP1pNPPml3WbDBlClT1Lp1a3Xt2lWdOnXSNddco4YNGyoyMtLu0ooNfx4AAMUlOjpat956q6KiotS9e3e7y/HrkUce0cMPP6z09HQ1bNhQN954o5YuXao6derYXVrI6devn06dOqWWLVvqnnvu0ciRIzV48GC7y/IIlQz05JNPql27durWrZs6deqktm3bOmJOuHOFwmsZCjWea8yYMQoPD1ejRo0UGxvLvGn5EGrvOQCzOe0zyyn5w2mvi5M4KWvzbyhfTvrddVItknM+X5zGae8TcueyLMuyuwjATr1791Z4eLheeuklu0ux1YkTJ1S9enVNnjxZd955p93lAABQ4lx//fW64oor9NRTT9ldCopQhw4d1Lx5c02dOtXuUgAAAIAShawNAPbhCmUY68yZM/ruu++0fv16XXHFFXaXU+y+/PJLvfLKK/rxxx+1ceNG3XHHHZKkv/71rzZXBgBAyXL48GG98cYbWrNmje655x67ywEAAAAAAACCwhzKMNZ///tftWnTRtddd53uvvtuu8uxxaRJk7R582aVLl1a8fHx+vDDD1WpUiW7ywIAoES58sordfjwYT3xxBNq0KCB3eUAAAAAAAAAQeGW1wAAAAAAAAAAAAAAv7jlNQAAAAAAAAAAAADALxrKAAAAAAAAAAAAAAC/aCgDAAAAAAAAAAAAAPyioQwAAAAAAAAAAAAA8IuGMgAAAAAAAAAAAADALxrKAAAAAAAAAAAAAAC/aCgDAAAAAAAAAAAAAPyioQwAAAAAAAAAAAAA8IuGMgAAAAAAAAAAAADALxrKAAAAAAAAAAAAAAC/aCgDAAAAAAAAAAAAAPyioQwAAAAAAAAAAAAA8IuGMgAAAAAAAAAAAADALxrKAAAAAAAAAAAAAAC/aCgDAAAAAAAAAAAAAPyioQwAAAAAAAAAAAAA8IuGMgAAAAAAAAAAAADALxrKAAAAAAAAAAAAAAC/aCgDAAAAAAAAAAAAAPyioQwAAAAAAAAAAAAA8IuGMgAAAAAAAAAAAADALxrKMMq8efPkcrm0c+dOu0vB/7dz5065XC5NmjTpgsumpqbK5XIVaT1r1qyRy+XSmjVrinQ/AACgZCN3+iqOLOdEZ38XPv/880LZ3oABA1S7du18rVu7dm0NGDCgUOoAAAAlB9nVWzB5a8CAAYqKiirw/gq6DTvVrl1bXbt2tbsMAEWMhjJgo++++06pqalGhLV33nlHqampdpcBAABgJKfmzscff1yLFy+2uwwAAAA4iNOy68mTJ5WamlqgC1AKYxt2ctp7AqD40VCGUfr27atTp06pVq1adpci6c+/iNPS0oz4i/idd95RWlpagbbx0EMP6dSpU4VUEQAAQNEhd/ryl+VoKBevzZs3a/bs2XaXAQAAHIbs6m327NnavHmz5+eTJ08qLS2twA3lgm7DTna/JwDsF2F3AUBxCg8PV3h4uN1llAgnTpxQuXLlinWfERERiojgYwsAADgfudMXWc5+brfb7hIAAIADkV29lSpVyu4SAMBxuEIZRvE3H8jZOR4++ugjtWzZUpGRkapbt65eeOEFv+t+8MEHGjJkiGJiYlS+fHn169dPhw8f9lrW5XL5vb3zuXOWzZs3Tz179pQkXXfddXK5XIU6d+/vv/+uUaNGqXbt2nK73YqLi1Pnzp21ceNGzzIffvihevbsqZo1a8rtdqtGjRoaPXq0z5UjZ+fx+PHHH3XzzTfroosu0h133BHwNgYMGKDp06d7Xpuzj/PNmjVL9erVk9vt1lVXXaUNGzZ4Pe9v3r1Tp07p3nvvVaVKlXTRRRcpMTFRu3fv9nkPfvrpJw0bNkwNGjRQmTJlFBMTo549e3JWHQAAKBKm5M6FCxfK5XJp7dq1Ps89++yzcrlc+u9//yvJN8u5XC6dOHFCzz//vKemszUHkt22b98ul8ulf/3rXz77XrdunVwul1555ZWAtyf977X/+OOPlZSUpNjYWJUrV049evTQgQMHvJbNyclRamqqqlWrprJly+q6667Td999F9A8xYcPH1bLli11ySWXeK5+efPNN9WlSxdVq1ZNbrdb9erV0yOPPKLs7Ow8tyVJkyZNUps2bRQTE6MyZcooPj5eCxcu9FmOOZQBAIA/pmRXKbDvTM+dQ3nnzp2KjY2VJKWlpXnqOf84du/ere7duysqKkqxsbEaM2aMJ8cVxjbOCjT3zZ07Vx07dlRcXJzcbrcaNWqkmTNn+iwXyPsc6Htyod+V3377TWPGjFGTJk0UFRWl8uXL66abbtKmTZt86gLgPJweDkjatm2bbrvtNt15553q37+/5syZowEDBig+Pl5XXHGF17LDhw/XxRdfrNTUVG3evFkzZ87UTz/9pDVr1vhtkubm2muv1b333qunnnpKDzzwgBo2bChJnv8vqLvvvlsLFy7U8OHD1ahRIx06dEgfffSRvv/+e/3lL3+RJL3++us6efKkhg4dqpiYGH322Wd6+umn9csvv+j111/32t6ZM2eUkJCgtm3batKkSSpbtmzA2xgyZIh+/fVXrVixQi+++KLfel9++WX9/vvvGjJkiFwulyZOnKhbbrlF27dvz/OswAEDBui1115T3759dfXVV2vt2rXq0qWLz3IbNmzQunXr9Le//U2XXHKJdu7cqZkzZ6pDhw767rvvPMcDAABQlEpa7uzSpYuioqL02muvqX379l7PLViwQFdccYUaN27sd90XX3xRd911l1q2bKnBgwdLkurVqycpsOxWt25dXXPNNZo/f75Gjx7tte358+froosu0l//+teAt3euESNGqEKFCkpJSdHOnTs1depUDR8+XAsWLPAsk5ycrIkTJ6pbt25KSEjQpk2blJCQoNOnT+f5mh08eFCdO3fWb7/9prVr13qOed68eYqKilJSUpKioqL0/vvva/z48Tp27JiefPLJPLc5bdo0JSYm6o477lBWVpZeffVV9ezZU0uWLPGbjQEAAAJR0rKrFNh3pueKjY3VzJkzNXToUPXo0UO33HKLJKlp06aeZbKzs5WQkKBWrVpp0qRJWrlypSZPnqx69epp6NChhbKNswLNfTNnztQVV1yhxMRERURE6O2339awYcOUk5Oje+65x+sYL/Q+B/KeBPK7sn37di1evFg9e/ZUnTp1tG/fPj377LNq3769vvvuO1WrVi3f7yuAYmABBpk7d64lydqxY4dnrFatWpYk64MPPvCM7d+/33K73dZ9993ns258fLyVlZXlGZ84caIlyXrzzTc9Y5KslJQUn/3XqlXL6t+/v+fn119/3ZJkrV69ulCO71zR0dHWPffck+cyJ0+e9BlLT0+3XC6X9dNPP3nG+vfvb0my7r///nxv45577rH8feTs2LHDkmTFxMRYv/32m2f8zTfftCRZb7/9tmcsJSXFaxtffPGFJckaNWqU1zYHDBjg8x74q3P9+vWWJOuFF17wjK1evbrI3hMAAGAOk3Jn7969rbi4OOvMmTOesT179lhhYWHWhAkTPGPnZznLsqxy5cp51XlWoNnt2WeftSRZ33//vWcsKyvLqlSpktd2A93e2de+U6dOVk5Ojmd89OjRVnh4uHXkyBHLsixr7969VkREhNW9e3evbaamplqSvPZ9dpsbNmyw9uzZY11xxRVW3bp1rZ07d17wmIcMGWKVLVvWOn36tGesf//+Vq1atfJcNysry2rcuLHVsWNHr/Hzfy8AAAAsy6zsGsh3pufnrQMHDuRa+9nvTc/NvZZlWVdeeaUVHx9fqNuwrMBzn79smZCQYNWtW9drLND3Oa/3JNBtnD592srOzvZad8eOHZbb7fY5dgDOwy2vAUmNGjVSu3btPD/HxsaqQYMG2r59u8+ygwcP9rpidujQoYqIiNA777xTLLUG6uKLL9ann36qX3/9NddlypQp4/nvEydO6ODBg2rTpo0sy9KXX37ps/y5Z8Pldxu56dWrlypUqOD5+ez74e89OGvZsmWSpGHDhnmNjxgxIs86//jjDx06dEiXXnqpLr74Yq9b2gAAABSlkpg7e/Xqpf3793vd8m7hwoXKyclRr1698rXNQLPb7bffrsjISM2fP98ztnz5ch08eFB///vfg97eWYMHD/a6kqZdu3bKzs7WTz/9JElatWqVzpw5E1AOPeuXX35R+/bt9ccff+iDDz5QrVq1cj3m33//XQcPHlS7du108uRJ/fDDD7lu9/x1Dx8+rKNHj6pdu3bkXAAAUCAlMbsG8p1pftx9991eP7dr1y7P7zXzu41Ac9+5yx09elQHDx5U+/bttX37dh09etRr2WDe59wEsg23262wsD9bUtnZ2Tp06JCioqLUoEEDcisQAmgoA5Jq1qzpM1ahQgWfeT4kqX79+l4/R0VFqWrVqkU+F++pU6e0d+9er0deJk6cqP/+97+qUaOGWrZsqdTUVJ8QsGvXLg0YMEAVK1b0zM1x9laF5weLiIgIXXLJJT77CWYbeTn/PTjbXPb3Hpz1008/KSwsTHXq1PEav/TSS32WPXXqlMaPH68aNWrI7XarUqVKio2N1ZEjR4KqEwAAoCBKYu688cYbFR0d7XU76AULFqh58+a67LLL8l1DINnt4osvVrdu3fTyyy97xubPn6/q1aurY8eOQW/vrAtl07ON5fNzZ8WKFb1OkjxX3759tX//fq1du1bVq1f3ef7bb79Vjx49FB0drfLlyys2NtbTFL9QXl2yZImuvvpqRUZGqmLFip7bKpJzAQBAQZTE7BrId6bBioyM9MyRfFZur1NBtxFo7vv444/VqVMnlStXThdffLFiY2P1wAMPSPLNlsG8z7kJZBs5OTn617/+pfr163tl8q+//prcCoQAGsqApPDwcL/jlmUV6n6ys7Pzve6CBQtUtWpVr0debr/9dm3fvl1PP/20qlWrpieffFJXXHGF3n33XU8tnTt31tKlSzVu3DgtXrxYK1as0Lx58yT9+Rf8uc49g+zc4wlmG3kp6vdgxIgReuyxx3T77bfrtdde03vvvacVK1YoJiYmqDoBAAAKoiTmTrfbre7du+uNN97QmTNntHv3bn388cf5vjpZCi679evXT9u3b9e6dev0+++/66233lLv3r29smuwWbAo3qdbbrlFR44c0bRp03yeO3LkiNq3b69NmzZpwoQJevvtt7VixQo98cQTkvLO1R9++KESExMVGRmpGTNm6J133tGKFSvUp0+fQv+9AgAAZimJ2fVC35nmR26vU2FvI9Dc9+OPP+r666/XwYMHNWXKFC1dulQrVqzQ6NGjJflmy8J4nwPZxuOPP66kpCRde+21eumll7R8+XKtWLFCV1xxBd/PAiEgwu4CgFCzdetWXXfddZ6fjx8/rj179ujmm2/2jFWoUEFHjhzxWi8rK0t79uzxGjv3NnoXkpCQoBUrVgRVa9WqVTVs2DANGzZM+/fv11/+8hc99thjuummm/TNN99oy5Ytev7559WvXz/POsHsI5htBHOsgapVq5ZycnK0Y8cOr7Mgt23b5rPswoUL1b9/f02ePNkzdvr0aZ/3CQAAwClCKXf26tVLzz//vFatWqXvv/9elmUF1FDOra5gstuNN96o2NhYzZ8/X61atdLJkyfVt2/ffG8vEGdvV71t2zavu+UcOnQo1ys5RowYoUsvvVTjx49XdHS07r//fs9za9as0aFDh7Ro0SJde+21nvEdO3ZcsJb//Oc/ioyM1PLly+V2uz3jc+fODfq4AAAA8iuUsmte35n6UxjfaxbGNgLNfW+//bYyMzP11ltveV05vHr16nzvuzDqX7hwoa677jo999xzXuNHjhxRpUqVCrx9AEWLhjIQpFmzZmngwIGeOUFmzpypM2fOeAWOevXq6YMPPvBZ7/yz7cqVKydJAX2RFcgZdmdlZ2fr+PHjio6O9ozFxcWpWrVqyszMlPS/s8bOPUvMsiy/V0zkJphtnHusF198ccD7yEtCQoIefPBBzZgxQ//61788408//bTfWs8/q+7pp58u0BmQAAAARSkUcudZnTp1UsWKFbVgwQJ9//33atmypc+0JP6UK1fOb03BZLeIiAj17t1bL7/8sr7//ns1adJETZs2zff2AnH99dcrIiJCM2fOVOfOnT3jzzzzTJ7rPfzwwzp27JiSk5MVHR2toUOHeuqTvHN1VlaWZsyYccFawsPD5XK5vI5l586dWrx4cTCHBAAAUCChkF0D+c7Un7JlywZcT1FuI9Dc5y9bHj16tEAnHAbznuTGXyZ//fXXtXv3br9TGAJwFhrKQJCysrJ0/fXX6/bbb9fmzZs1Y8YMtW3bVomJiZ5l7rrrLt1999269dZb1blzZ23atEnLly/3OdOqefPmCg8P1xNPPKGjR4/K7XarY8eOiouLK1CNv//+uy655BLddtttatasmaKiorRy5Upt2LDBc1XG5Zdfrnr16mnMmDHavXu3ypcvr//85z9BzY0RzDbi4+MlSffee68SEhIUHh6uv/3tbwU6zvj4eN16662aOnWqDh06pKuvvlpr167Vli1bJHmfOde1a1e9+OKLio6OVqNGjbR+/XqtXLlSMTExBaoBAACgqIRC7jyrVKlSuuWWW/Tqq6/qxIkTmjRpUkDrxcfHa+XKlZoyZYqqVaumOnXqqFWrVkFnt379+umpp57S6tWrPbeJPldhZ8HKlStr5MiRmjx5shITE3XjjTdq06ZNevfdd1WpUqU8r+B48skndfToUd1zzz266KKL9Pe//11t2rRRhQoV1L9/f917771yuVx68cUXA7rNYJcuXTRlyhTdeOON6tOnj/bv36/p06fr0ksv1ddff52v4wMAAAhWKGTXQL4z9adMmTJq1KiRFixYoMsuu0wVK1ZU48aN1bhx44D3XRjbCDT33XDDDSpdurS6deumIUOG6Pjx45o9e7bi4uJ8rgYPVGG8J127dtWECRM0cOBAtWnTRt98843mz5+vunXr5qsmAMWLOZSBID3zzDNq2LChxo8fr3nz5ql379568803vb40GjRokMaNG6cPPvhA9913n3bs2KEVK1Z4zuQ6q0qVKsrIyND+/ft15513qnfv3vruu+8KXGPZsmU1bNgwffXVV0pJSdHo0aM9QS4pKUnSn1/6vf3222revLnS09OVlpam+vXr64UXXgh4P8Fs45ZbbtGIESO0bNky9e3bV7179y7wcUrSCy+8oHvuucczj3NWVpYWLFggSYqMjPQsN23aNPXr10/z58/Xfffdpz179mjlypWKiooqlDoAAAAKWyjkznP16tVLx48fl/Tn3HSBmDJliuLj4/XQQw+pd+/emjlzpqTgs1t8fLyuuOIKhYWF6Y477vB5viiy4BNPPKGHH35YGzZs0JgxY7Rt2za99957sizLK4f6k5GRob/97W8aOHCg3nzzTcXExGjJkiWqWrWqHnroIU2aNEmdO3fWxIkTL1hHx44d9dxzz2nv3r0aNWqUXnnlFT3xxBPq0aNHvo8NAAAgWKGQXQP5zjQ3//73v1W9enWNHj1avXv31sKFC4Pef0G3EWjua9CggRYuXCiXy6UxY8YoIyNDgwcP1siRI4Ou+azCeE8eeOAB3XfffVq+fLlGjhypjRs3aunSpapRo0a+6wJQfFxWMDOrAwabN2+eBg4cqA0bNqhFixZ2l4M8fPXVV7ryyiv10ksv+f1CEQAAwMnInflz5ZVXqmLFilq1apVtNRw5ckQVKlTQo48+qgcffNC2OgAAAIoL2RUAzMAVygBC2qlTp3zGpk6dqrCwMF177bU2VAQAAIDi9vnnn+urr75Sv379im2fueVQSerQoUOx1QEAAAAAQFFjDmUAIW3ixIn64osvdN111ykiIkLvvvuu3n33XQ0ePJjbpQAAAJRw//3vf/XFF19o8uTJqlq1qnr16lVs+16wYIHmzZunm2++WVFRUfroo4/0yiuv6IYbbtA111xTbHUAAAAAAFDUaCgDCGlt2rTRihUr9Mgjj+j48eOqWbOmUlNTucUgAACAARYuXKgJEyaoQYMGeuWVVy44d3Fhatq0qSIiIjRx4kQdO3ZMlStX1siRI/Xoo48WWw0AAAAAABQH5lAGAAAAAAAAAAAAAPjFHMoAAAAAAAAAAAAAAL9oKAMAAAAAAAAAAAAA/KKhDAAAAAAAAAAAAADwK8K+XW+xb9cFUKZmit0lBO3UrjS7SwAABOyyYt1bmZq9C7yNU7teKYRKACdyTl51UgYlWwKA6cirgHM4J6+ey0nZ9Vwnd6XaXYJfLrnsLgEAShjyalGwsaEMAADs5nJxsxIAAAA4F3kVAAAATmZKXjXjKAEAAAAAAAAAAAAAQeMKZQAADObi3DIAAAA4GHkVAAAATmZKXqWhDACAwUy5JQsAAABCE3kVAAAATmZKXqWhDACAwUwJPAAAAAhN5FUAAAA4mSl51YyjBAAAAAAAAAAAAAAEjSuUAQAwmMvlsrsEAAAAIFfkVQAAADiZKXmVhjIAAEbjZiUAAABwMvIqAAAAnMyMvEpDGQAAg5kyxwcAAABCE3kVAAAATmZKXjXjKAEAAAAAAAAAAAAAQeMKZQAADGbKGXQAAAAITeRVAAAAOJkpeZWGMgAABnNxsxKUMEeOHNH+/fv9PnfZZZcVczUAAKCgyKsAAABwMlPyKg1lAAAMZsoZdDDHwoUL9dhjj/mM//rrr8rMzLShIgAAUBDkVQAAADiZKXmVhjIAAABKjLvuukt33XWXz3j9+vVtqAYAAAAAAAAIfWa0zQEAgF8uV1iBH4AT/frrr1q6dKnee+89HT9+XC6Xy+6SAABAPpBXUZJ88sknnv/OyMiwsRIAAFBYTMmrXKEMAIDBQiWwAMG4//779eKLL6p58+bKzMzUV199pVOnTuW5TmZmps8tsd3uLLndpYuyVAAAcAHkVZQk/fr105YtWyRJU6ZM0d133x3wuuRVAACcyZS8asZRAgAAv1yF8D/ASZ566ilt2bJF27Zt09KlS7Vy5UqtXLlSWVlZea6Xnp6u6Ohor0d6+rPFVDUAAMgNeRUliWVZqlatmqpWraodO3bokksu0bXXXqv169dfcF3yKgAAzmRKXuUKZQAAAJQYCxYs0CuvvKIyZcpIko4dO6YaNWpo586dea6XnJyspKQkrzG3e1dRlQkAAAADuVwuffLJJ7IsSx06dNCqVav06aefavDgwfrmm2/yXJe8CgAA7BTUFcrM8wEAQMliyhwfMMfJkydVrlw5SdKsWbNUo0YN1atX74LZ1e12q3z58l4Pbh8IAID9yKsoSSzL0oABAzRw4EBFRESobt266tatm44dO3bBdcmrAAA4kyl5Nagq+/Xr5/nvKVOmBLxeZmamjh075vXIzMz7toMAAKDomRJ4YI6bb75Zjz32mCTpueee0+bNm7Vv3z7NnTvX5soAAEB+kFdRksydO1fjxo3TuHHjdNFFF2nEiBFq06aNhg8fbndpAAAgn0zJq0FVmd95PpjjAwAAZzIl8MAcDz/8sLZv365WrVrp2LFj+tvf/qarr75at99+u92lAQCAfCCvoiRp27atEhISlJCQoDfeeEM1a9bUv/71L40dO9bu0gAAQD6ZkleDmkM5v/N8MMcHAABOFRqBBQhUZGSkFi9erLVr1+qLL75Q6dKl1b59ezVp0sTu0gAAQL6QV1Ey1apVi0YyAAAlghl5NaiG8tl5PiR55vmIi4vT/fffn+d6brdbbrf7vFHm+AAAAEDRaN++vdq3b293GQAAAAAAAEDIC6qhPHfuXJ04cULSn1cdjxgxQmvXrv1/7N17nFVlvT/w754BNiCBIjcjEsF7mBgqoaKkKJmSlkcRTURLLcLUMVO8AVpOapqa5u0EmnnXjoV3RclrmpfSyiBCxYPiLQVDGBT27w9/znFkuGwYZq09z/vda71intlrre/ebGY+7u9az2OdDwCoUJUypQoAAGmSVwEAyLNU8mpZDeWddtqp/s+bb7553HTTTfHzn/88dttttyYvDABY+1IJPAAAVCZ5FQCAPEslr5bVUP4k63wAQOUrJLLGBwAAlUleBQAgz1LJq2k8SwAAAAAAAADKttp3KAMAlS+VKVkAAKhM8ioAAHmWSl7VUAaAhBUKhaxLAACA5ZJXAQDIs1TyqoYyACQslSvoAACoTPIqAAB5lkpeTeNZAgAAAAAAAFA2dygDQMIKri0DACDH5FUAAPIslbyqoQwACUtlShYAACqTvAoAQJ6lklc1lAEgYakEHqh0C2dPzLqEeht84eqsS6j3r798NesS6rVv1T3rEgBaJHkVKk+esusntd8wn3X956VTsi6hUVUFrQOAVZFKXk3jWQIAAAAAAABQNpcZAUDCUlnjAwCAyiSvAgCQZ6nkVQ1lAEhZIlOyAABQoeRVAADyLJG8qqEMAAlLZY0PAAAqk7wKAECepZJX03iWAAAAAAAAAJTNHcoAkLBCoZB1CQAAsFzyKgAAeZZKXtVQBoCEFUxWAgBAjsmrAADkWSp5VUO5TAtnT8y6hLL974LpWZewWr6y12tZl1C2f04bknUJAGVJZY0PAAAqk7wKAECepZJX03iWAAAAAAAAAJTNHcoAkLJE1vgAAKBCyasAAORZInlVQxkAUmauEgAA8kxeBQAgzxLJqxrKAJCyRK6go+VZvHhxvP7661EqlRqMd+zYMdZdd91sigIAmp68SoWSVwEgEYnkVQ1lAAAqxvvvvx9HHXVU3HTTTdGhQ4do06ZN/fcKhUIcfvjh8eMf/zjDCgEASJm8CgC0RBrKAJCyRK6go+U488wzo0OHDvHmm29Gx44dsy4HAFjb5FUqjLwKAIlJJK9qKANAyhJZ44OW4/e//3089dRT0a5du6xLAQCag7xKhZFXASAxieRVDWUASFgpkSvoaDkWLVq0Vj6cq6uri7q6ugZjxeLiKBbbLGcPAKA5yKtUGnkVANKSSl5NpG8OAEBLsM0228TZZ58dw4YNa9Lj1tbWRqdOnRpstbWXN+k5AABo+eRVAKAl0lAGgJQVmmCDZnThhRfGs88+GwcccECTHnfcuHExb968Btu4cUc16TkAgNUgr1Jh5FUASEwiedWU1wCQsqoKSSzw//Xs2TNuuOGGJj9usViMYrH4qVHTBwJA5uRVKoy8CgCJSSSvaigDQMoSWeMDAIAKJa8CAJBnieRVU14DAAAAAAAA0Ch3KANAytK4gA4AgEolrwIAkGeJ5FUNZQBIWSJrfAAAUKHkVQAA8iyRvKqhDAApS2SNDwAAKpS8CgBAniWSV62hDAAAAAAAAECj3KEMAClL4wI6AAAqlbwKAECeJZJXNZQBIGWJrPEBAECFklcBAMizRPLqKk95veuuu67NOgCALBSaYAMAgLVFXgUAIM8yyquXXHJJ9O7dO9q2bRsDBw6MJ598coWPv+CCC2KzzTaLdu3aRa9eveK4446LRYsWrfL5Vrmh/Morr8Suu+66zDZixIiYN2/eCvetq6uL+fPnN9jq6havcpEAQMvS3IEHAADKIa8CAJBXN954Y9TU1MT48ePjmWeeia233jqGDRsWb7zxRqOPv+666+Kkk06K8ePHxwsvvBC/+tWv4sYbb4yTTz55lc9Z1pTXJ5544jJjb775Znz961+P3/72t7H++us3ul9tbW1MnDixwdj48WNjwoSjyzk9ANDESoXmv2Xj48Bz2WWXxcCBA+OCCy6IYcOGxfTp06Nbt27LPP7jwDNp0qTYYYcdYsaMGTF69OgoFApx/vnnN3v9AAA0H3kVAIA8yyKvnn/++XHEEUfEYYcdFhERl112Wdxxxx0xadKkOOmkk5Z5/GOPPRY77rhjHHTQQRER0bt37xg5cmQ88cQTq3zOVW4oFwqFGDZsWKPf22STTeLrX/963HbbbdG1a9dlvj9u3LioqalpMFYszl7lIgGAtSSDNT6yCDwAAFQoeRUAgDxrgrxaV1cXdXV1DcaKxWIUi8VlHrt48eJ4+umnY9y4cf9XQlVVDB06NB5//PFGj7/DDjvEb37zm3jyySdj++23j1mzZsWdd94ZhxxyyCrXuMpTXpdKpUanvL7gggti4MCBcdVVV0Xnzp0b3bdYLEbHjh0bbMVim1UuEgBYS5p5jY+PA8/QoUPrx1Yl8Dz99NP10wx+HHi+9rWvlXdyAAAqTxPk1caXYqtb9lwhrwIAUKYmyKu1tbXRqVOnBlttbW2jp3vrrbdiyZIl0b179wbj3bt3j7lz5za6z0EHHRRnnHFG7LTTTtG6devo27dvDBkyZO1MeT158uRYsGDBMuOf//znI+Kju5QBgPSUcwXdigLPP/7xj0aPf9BBB8Vbb70VO+20U5RKpfjwww/ju9/9blmBBwCAdDW+FNv4mDBhwjKPlVcBAGhujc/0vOxnq6tr2rRpcdZZZ8Uvf/nLGDhwYMycOTOOOeaYOPPMM+O0005bpWOsckN5p512Wu1CAYCcaoI1Psr5gG51NEXgAZrOa387NOsS6nXs89OsS6g3f9ayU6BmZUlpUdYl1KsutM26BKDSNUFerYQP6IC17/2Xx2ddQqM69jkn6xIaNX/Wj7IuoVEfLH0/6xIa1bqqfdYlAFlpgry6vJtzGtOlS5eorq6O119/vcH466+/Hj169Gh0n9NOOy0OOeSQ+M53vhMREVtttVUsWLAgjjzyyDjllFOiqmrlE1qvckMZAGiBmmCNj3I+oMsq8AAAUKGaIK9Wwgd0AABUqCbIq+Vo06ZNDBgwIKZOnRr77rtvREQsXbo0pk6dGmPHjm10n/fff3+ZTFpdXR0RHy15vCokWgBIWROs8VEsFqNjx44NtuV9YPfJwPOxjwPPoEGDGt2nKQIPAAAVqgnyajnkVQAAytLMeTUioqamJq688sq4+uqr44UXXojvfe97sWDBgjjssMMiImLUqFExbty4+scPHz48Lr300rjhhhvixRdfjPvuuy9OO+20GD58eH1uXRl3KAMAzaqmpiYOPfTQ2HbbbWP77bePCy64YJnA07Nnz6itrY2IjwLP+eefH9tss039FILlBh4AAFhV8ioAAHk2YsSIePPNN+P000+PuXPnRv/+/ePuu++O7t27R0TE7NmzG1zweOqpp0ahUIhTTz015syZE127do3hw4fHT37yk1U+p4YyAKSsCdb4KFcWgQcAgAolrwIAkGcZ5NWIiLFjxy53iutp06Y1+LpVq1Yxfvz4GD9+/Gqfr1DKbO6dGdmcNkH/u2B61iWslq/s9VrWJZTtn9OGZF0CUPE2bdazbbzfb9b4GDNv/VYTVAJ5JK/mXcc+P826hHrzZ52UdQn1lpQWZV1CvepC26xLAJqcvAr5Ia+2BB37nJN1CY2aP+tHWZfQqA+Wvp91CY1qXdU+6xKAevLq2uAOZQBIWdXKHwIAAJmRVwEAyLNE8moiTxMAAAAAAACAcrlDGQBSltEaHwAAsErkVQAA8iyRvKqhDAApSyPvAABQqeRVAADyLJG8qqEMAAkrVSWSeAAAqEjyKgAAeZZKXrWGMgAAAAAAAACNcocyAKQskTU+AACoUPIqAAB5lkhedYcyAKSs0AQbVIBHH3006xIAgNUhrwIAkGeJ5FV3KANAyhJZ4wMOO+ywmDFjRtZlAADlkldpge6999545ZVXokOHDrHttttG3759sy4JAFhdieRVdygDANDilUqlrEsAACBRU6ZMqf/z4YcfHmeccUY89NBDceONN8ZOO+0UV199dYbVAQCsnDuUASBliazxAYWVvNfr6uqirq6uwVixuDiKxTZrsywAYGXkVVqA448/PoYPHx4REffdd1/Mnj27Pp/OnDkzRo0aFYceeugKjyGvAkBOJZJXNZQT8Ll1Nsu6hNXyxD1ZV5COTbatzBf7n08Ny7oEqHxp5B1Yqdra2pg4cWKDsfHjx8aECUdnVBEAEBHyKi1CqVSKqqr/myiydevW0adPn7jwwgtjzz33jLfeemulx5BXASCnEsmrGsoAkLJE1viAlRk3blzU1NQ0GCsWZ2dUDQBQT16lBSgUCrFo0aJYunRp9OvXL5577rl44oknYsyYMbHnnnuu0vIs8ioA5FQieVVDGQCA5BWLxSgWi58aNX0gAABrrlQqxZAhQ+q/bt++ffTv3z8WLlwYERHV1dUrPYa8CgBkSUMZAFKWyBV0pOOss86K+++/v8FYqVSKOXPmZFQRALBG5FVagGuvvTYWLlwYpVIpTjjhhBg9enQ89thjceKJJ0ZERM+ePePKK6+MESNGRMeOHTOuFgAoSyJ5VUMZABJWSiPvkJDhw4fHgAEDsi4DAGgi8iotwfbbb1//5ylTpsSUKVNizJgx9eOTJk2KsWPHxpIlS+K73/1uVmUCAKshlbyqoQwAKUvkCjrSsdVWW8VWW22VdRkAQFORV2lhevToEUcccUSDsQ033DCmTJmSUUUAwBpJJK9qKANAygppBB4AACqUvAoAQJ4lklersi4AAAAAAAAAgHxyhzIApCyRKVkAAKhQ8ioAAHmWSF7VUAaAlJmrBACAPJNXAQDIs0TyqoYyAKQskTU+AACoUPIqAAB5lkheTaRvDgAAAAAAAEC53KEMAClLZI0PAAAqlLwKAECeJZJXNZQBIGGlRKZkAQCgMsmrAADkWSp5VUMZAFJm8QsAAPJMXgUAIM8SyauJPE0AAAAAAAAAyuUOZQBIWSJrfAAAUKHkVQAA8iyRvKqhDAApS2SND6BlemvmmKxLyKVum/131iXUe3vG2KxLACqdvAq0cO/+67isS6go3XOUdT/pzRlHZl1Co6oLbbMuAVq+RPJq2Q3le++9N1555ZXo0KFDbLvtttG3b9+1URcA0BwSuYIOAIAKJa8CAJBnieTVla6hPGXKlPo/H3744XHGGWfEQw89FDfeeGPstNNOcfXVV6/VAgEAAAAAAADIxkrvUD7++ONj+PDhERFx3333xezZs6Pw/2/fnjlzZowaNSoOPfTQFR6jrq4u6urqGowVi4ujWGyzunUDAE0hjQvoAACoVPIqAAB5lkheXekdyqVSKaqqqqKqqirmzJkTrVu3jk033TTuuuuu2HjjjeOtt95a6Ulqa2ujU6dODbba2sub5AkAAKuvVFVY4w0AANYWeRUAgDxLJa+u9A7lQqEQixYtiqVLl0a/fv3iueeeiyeeeCLGjBkTe+65Z5RKpZWeZNy4cVFTU9NgrFicvfpVAwBNo0ICCwAAiZJXAQDIs0Ty6kobyqVSKYYMGVL/dfv27aN///6xcOHCiIiorq5e6UmKxWIUi8VPjZruGgAAAAAAACDPVtpQvvbaa2PhwoVRKpXihBNOiNGjR8djjz0WJ554YkRE9OzZM6688soYMWJEdOzYca0XDAA0oUIaV9ABAFCh5FUAAPIskby60oby9ttvX//nKVOmxJQpU2LMmDH145MmTYqxY8fGkiVL4rvf/e7aqxQAaHpVWRcAAAArIK8CAJBnieTVlTaUP6lHjx5xxBFHNBjbcMMNY8qUKU1aFADQTBK5gg4AgAolrwIAkGeJ5NVE+uYAAAAAAAAAlKusO5QBgBamKo0r6AAAqFDyKgAAeZZIXtVQBoCUJRJ4AACoUPIqAAB5lkhe1VAGgISVElnjg3TMmTMn/vWvfzX6vZ133rmZqwEA1pS8CgBAnqWSVzWUAQBoMR566KG46KKLlhn/y1/+Eu+//34GFQEAAABAZdNQBoCUVWVdADStkSNHxsiRI5cZ32STTTKoBgBYY/IqAAB5lkhe1VAGgJQlMiULFFbyXq+rq4u6uroGY8Xi4igW26zNsgCAlZFXqWD9+vWLMWPGxJgxY9b4WPIqAORUInk1kb45ANCoqsKab9AC1NbWRqdOnRpstbWXZ10WACCvUsEmTpwYw4YNa5JjyasAkFOJ5FV3KAMA0GL8/e9/j6eeemqZ8ffee2+F+40bNy5qamoajBWLs5u0NgAA0rLffvs12bHkVQAgSxrKAJCyCrkCDlbViy++GHfddVf91//5z3/innvuiSFDhqxwv2KxGMVi8VOjpg8EgMzJqxAR8ioA5FYieVVDGQBSlkbeISF77bVX7LXXXg3Gbr/99jjrrLMyqggAWCPyKgAAeZZIXtVQBoCElRK5go503HTTTXHZZZc1GKurq4tZs2ZlVBEAsCbkVQAA8iyVvKqhDABAizFw4MDo1KlTg7GqqqrYcsstM6oIAAAAACqbhjIApKyQxhV0pGPDDTeMDTfcMOsyAICmIq8CAJBnieRVDWUASFkiU7IAAFCh5FUAAPIskbyqoUxudS5ulnUJyZjx1B5Zl7BaJj47O+sSyvbbvxWzLmG1PDmybdYllK1Y3WnlDyIijbwDAEClklcBAMizRPJqVdYFAAAAAAAAAJBP7lAGgIRVubQMAIAck1cBAMizVPJqIk8TAGhMobDmGwAArC3yKgAAeZZVXr3kkkuid+/e0bZt2xg4cGA8+eSTK3z8u+++G9///vdjgw02iGKxGJtuumnceeedq3w+DWUASFgqgQcAgMokrwIAkGdZ5NUbb7wxampqYvz48fHMM8/E1ltvHcOGDYs33nij0ccvXrw4dt9993jppZfilltuienTp8eVV14ZPXv2XOVzmvIaAGhWHweeyy67LAYOHBgXXHBBDBs2LKZPnx7dunVb5vEfB55u3brFLbfcEj179oyXX3451l133eYvHgCAFk9eBQAgz84///w44ogj4rDDDouIiMsuuyzuuOOOmDRpUpx00knLPH7SpEnx73//Ox577LFo3bp1RET07t27rHNqKANAwgoZzAGYReABAKAyyasAAORZU+TVurq6qKurazBWLBajWCwu89jFixfH008/HePGjasfq6qqiqFDh8bjjz/e6PF///vfx6BBg+L73/9+/O53v4uuXbvGQQcdFCeeeGJUV1evUo2mvAaAhDX3lCwfB56hQ4fWj5UTeLp37x79+vWLs846K5YsWbImTx0AgArQFHm1rq4u5s+f32D79Ad2H5NXAQAoR1Pk1dra2ujUqVODrba2ttHzvfXWW7FkyZLo3r17g/Hu3bvH3LlzG91n1qxZccstt8SSJUvizjvvjNNOOy3OO++8+PGPf7zKz1NDGQAS1twf0GUVeAAAqEypfEAHAEBlaoq8Om7cuJg3b16D7ZN3IK+ppUuXRrdu3eKKK66IAQMGxIgRI+KUU06Jyy67bJWPYcprAGCN1NbWxsSJExuMjR8/PiZMmNAkx/9k4Kmuro4BAwbEnDlz4txzz43x48c3yTmAytSmqmPWJeTSk0/tnnUJ9QZc/3rWJdS74Cvzsy6h3k49Ns66hHqFaP7phKG5jRs3LmpqahqMNTZ94OqSV4E1UVVonXUJFeXHv/t61iU0at/752VdQqMuGPhy1iU0qk/HTbMuoVGyMVlZ3vTWjenSpUtUV1fH6683/O/t119/PXr06NHoPhtssEG0bt26wfTWW2yxRcydOzcWL14cbdq0Wel5NZQBIGGFJpirpJwP6LIKPAAAVKamyKuV8AEdAACVqSnyajnatGkTAwYMiKlTp8a+++4bER9d4Dh16tQYO3Zso/vsuOOOcd1118XSpUujquqjgmfMmBEbbLDBKmdVU14DQMKaYkqWYrEYHTt2bLAt7wO7Twaej30ceAYNGtToPjvuuGPMnDkzli5dWj9WbuABAKAyNUVeLYe8CgBAOZo7r0ZE1NTUxJVXXhlXX311vPDCC/G9730vFixYEIcddlhERIwaNarBlNnf+9734t///nccc8wxMWPGjLjjjjvirLPOiu9///urfE4NZQBIWFVhzbdyZRF4AACoTPIqAAB5lkVeHTFiRPzsZz+L008/Pfr37x9//vOf4+67747u3btHRMTs2bPjtddeq398r1694p577ok//elP8cUvfjF+8IMfxDHHHBMnnXTSKp/TlNcAQLMaMWJEvPnmm3H66afH3Llzo3///ssEno+nXon4v8Bz3HHHxRe/+MXo2bNnHHPMMXHiiSdm9RQAAGjB5FUAAPJu7Nixy53ietq0acuMDRo0KP74xz+u9vk0lAEgYaszpUpTaO7AAwBAZZJXAQDIs6zyanPTUAaAhKUSeAAAqEzyKgAAeZZKXtVQBoCEFVJJPAAAVCR5FQCAPEslrzbaUF68eHG8/vrrUSqVGox37Ngx1l133eaoCwAAAAAAAICMNWgov//++3HUUUfFTTfdFB06dIg2bdrUf69QKMThhx8eP/7xj5u9SABg7ShUZV0BAAAsn7wKAECepZJXGzSUzzzzzOjQoUO8+eab0bFjx6xqAgCaSSIzsgAAUKHkVQAA8iyVvNqgofz73/8+nnrqqWjXrl1W9QAAzSiVwAMAQGWSVwEAyLNU8mqDhvKiRYvWSjO5rq4u6urqGowVi4ujWGyznD0AAAAAAAAAyFqDhvI222wTZ599djzwwANxzz33NNlJamtrY+LEiQ3Gxo8fGxMmHN1k5wAAypfKFXSk49Zbb4158+Y1+r3DDz+8masBANaUvAoAQJ6lklcbNJQvvPDCOP744+OAAw5o0pOMGzcuampqGowVi7Ob9BwAQPmqEgk8pOOPf/xjzJ07d5nx3//+9xrKAFCB5FVaonvvvTdeeeWV6NChQ2y77bbRt2/frEsCAFZTKnm1QUO5Z8+eccMNNzT5SYrFYhSLxU+Nmu4aALKWyhV0pOPcc89t8PX06dPjM5/5TDzxxBMZVQQArAl5lZZgypQpMXz48Ij4aNacGTNmRN++feO9996LY489Nn7605/GoYcemnGVAMDqSCWvtlr5QwAAoPKcc845cf7550exWIw333xzhY+tq6uLurq6BmPF4uIoFl0ECQDAmjn++OPrG8r33XdfzJ49Owr//9PnmTNnxqhRo1baUJZXAYAsVWVdAACQnUJhzTfIk08us3LppZfGjBkz4q9//Wv06NFjhfvV1tZGp06dGmy1tZev7XIBgJWQV2kJSqVSVFVVRVVVVcyZMydat24dm266adx1112x8cYbx1tvvbXSY8irAJBPqeRVDWUASFihqrDGG+TJ7bffXv/nQqEQH374YXzmM5+JWbNmrXC/cePGxbx58xps48YdtbbLBQBWQl6lJSgUCrFo0aJ4//33o0+fPjF//vy4/PLL6y+GLJVKKz2GvAoA+ZRKXjXlNQAkrFKugINV9fHdHx/r0qVLRHz0Id6SJUuWu1+xWIxisfipUdMHAkDW5FVaglKpFEOGDKn/un379tG/f/9YuHBhRERUV1ev9BjyKgDkUyp5VUMZAIAW4+O7P1blLg8AAGgO1157bSxcuDBKpVKccMIJMXr06HjsscfixBNPjIiInj17xpVXXhkjRoyIjh07ZlwtAMCyNJQBIGGpXEFHOkqlUnz1q19t9HsPPPBAM1cDAKwpeZWWYPvtt6//85QpU2LKlCkxZsyY+vFJkybF2LFjY8mSJfHd7343qzIBgNWQSl7VUAaAhKUSeEjH5MmTY8GCBVmXAQA0EXmVlqZHjx5xxBFHNBjbcMMNY8qUKRlVBACsiVTyqoYyACSsKpHAQzp22mmnrEsAAJqQvAoAQJ6lklc1lAEgYalcQQcAQGWSVwEAyLNU8mpV1gUAAAAAAAAAkE/uUAaAhBVcWgYAQI7JqwAA5FkqeVVDGQASlsqULAAAVCZ5FQCAPEslr2ooA0DCCqkkHgAAKpK8CgBAnqWSVxO5ERsAAAAAAACAcrlDGQASlsgFdAAAVCh5FQCAPEslr2ooA0DCUgk8AABUJnkVAIA8SyWvaigDQMJSCTwAAFQmeRUAgDxLJa9aQxkAAAAAAACARrlDGYhCVOYlNCdv3SXrEsr264v/mXUJq2X28HZZl1C23p+pzF9xrZv5Uq+qyvznD8AK9O24WdYl1Ht8xPysS6i365T8/NLbYYM5WZdQ7yfbrpd1CQ20rlon6xLIGXkVgE8as2XvrEto1Lc3ezfrEhq1yYUdsi6hUbceMjPrEhq1VefOWZfQqDZVnbIuoVFVhcr8/LWppZJX/W0DQMJSCTwAAFQmeRUAgDxLJa9qKANAwqoKpaxLAACA5ZJXAQDIs1TyqjWUAQAAAAAAAGiUO5QBIGGpTMkCAEBlklcBAMizVPKqhjIAJMxUJQAA5Jm8CgBAnqWSVzWUASBhqazxAQBAZZJXAQDIs1TyaiqNcwAAAAAAAADK5A5lAEhYKmt8AABQmeRVAADyLJW8qqEMAAkzVQkAAHkmrwIAkGep5FUNZQBIWCpX0AEAUJnkVQAA8iyVvJpK4xwAAAAAAACAMrlDGQASViiUsi4BAACWS14FACDPUsmrGsoAkLBUpmQBAKAyyasAAORZKnlVQxkAEmbtCwAA8kxeBQAgz1LJq6k8TwAAAACAzOy3335ZlwAAsFrcoQwACatKZI0PWqZ+/frFmDFjYsyYMWt8rLq6uqirq2swViwujmKxzRofGwBYffIqLclzzz0XkyZNWmZ8iy22iEGDBq1wX3kVAPIplbzqDmUASFhVYc03yMrEiRNj2LBhTXKs2tra6NSpU4OttvbyJjk2ALD65FVamgcffHCZbcaMGSvdT14FgHxKJa+u9h3KS5cujbvvvjs6deoUO+64Y1PWBAA0E1eWUcmacsrAcePGRU1NTYOxYnF2kx0fAFg98iotSaFQiGuuuWa19pVXASCfUsmrZTWU77333thjjz0iIuKoo46K559/PhYtWhS77757nHvuuWulQAAAWNuKxWIUi8VPjZo+EACAplMqleKggw5aZnz33XePww47bIX7yqsAQJbKaiiPHTu2fgqWe++9N1566aVYsmRJfPnLX17hftb4AIB8qpQpVQAASJO8SktyzjnnxPz585cZ32yzzTKoBgBoCqnk1bIayqVSKT772c9GqVSKt956K3r16hV9+vSJX/ziFyvcr7a2NiZOnNhgbPz4sTFhwtHlVwwANJmqQinrEgAAYLnkVVqSb3zjG1mXAAA0sVTyalkN5UKhEH/84x+jVCrFkCFDYurUqfHEE0/EkUceGc8///xy97PGBwDkUypX0AEAUJnkVQAA8iyVvFr2HcqjR4/+aMdWraJPnz7RrVu3OOmkk1a4nzU+AAAAAAAAACpPWQ3lyZMnx4IFCyLio7uOjz766PjDH/4QY8eOXSvFAQBrV1XWBQAAwArIqwAA5FkqebWshvJOO+1U/+fNN988brrppvj5z38eu+22W5MXBgCsfams8QEAQGWSVwEAyLNU8mpZDeVP2nDDDeOEE05oyloAgGaWyhofAABUJnkVAIA8SyWvpnInNgAAAAAAAABl0lAGgIRVFdZ8AwCAtUVeBQAgz7LKq5dcckn07t072rZtGwMHDownn3xylfa74YYbolAoxL777lvW+TSUASBhVU2wrY7mDjwAAFQmeRUAgDzLIq/eeOONUVNTE+PHj49nnnkmtt566xg2bFi88cYbK9zvpZdeih/+8IcxePDgss+poQwACasqlNZ4K1cWgQcAgMokrwIAkGdZ5NXzzz8/jjjiiDjssMNiyy23jMsuuyzat28fkyZNWu4+S5YsiYMPPjgmTpwYffr0Kf95lr0HAMAayCLwAADAqpJXAQDIq8WLF8fTTz8dQ4cOrR+rqqqKoUOHxuOPP77c/c4444zo1q1bfPvb316t82ooA0DCmnuNj6wCDwAAlakp8mpdXV3Mnz+/wVZXV9fo+eRVAADK0dx59a233oolS5ZE9+7dG4x379495s6d2+g+jzzySPzqV7+KK6+8cvWf52rvCQBUvKZY46MSAg8AAJWpKfJqbW1tdOrUqcFWW1vb6PnkVQAAytHcebVc7733XhxyyCFx5ZVXRpcuXVb7OK2apBoAoCKVe4dxY2pra2PixIkNxsaPHx8TJkxY42M3VeABIBttqjpmXUK9qXuXvy7V2rLv1Kwr+D9nPPt21iU0cNLW87MuoV77Vt2yLqFeIaqzLiEzTZFXx40bFzU1NQ3GisXimh845FUAPlKsXjfrEho189h8Zohf/O3DrEto1Hl//SDrEhr1i0Ezsy6hUZ2L+Vzmo7oJ8mM5mjuvdunSJaqrq+P1119vMP76669Hjx49lnn8v/71r3jppZdi+PDh9WNLly6NiIhWrVrF9OnTo2/fviutUUMZAFgjlRB4AABIV7FYXOUGsrwKAEBzKyevtmnTJgYMGBBTp06NfffdNyI+yp9Tp06NsWPHLvP4zTffPJ5//vkGY6eeemq89957ceGFF0avXr1W6bwaygCQsEJhze/WqoTAAwBAZWqKvFoOeRUAgHI0d16NiKipqYlDDz00tt1229h+++3jggsuiAULFsRhhx0WERGjRo2Knj17Rm1tbbRt2zb69evXYP911103ImKZ8RXRUAaAhDXFlCzlyiLwAABQmeRVAADyLIu8OmLEiHjzzTfj9NNPj7lz50b//v3j7rvvju7du0dExOzZs6OqqqpJz6mhDAAJa9pYsWqyCDwAAFQmeRUAgDzLKhWOHTu20Rl0IiKmTZu2wn2vuuqqss+noQwANLvmDjwAAFAOeRUAAP6PhjIAJKwqgzU+AABgVcmrAADkWSp5VUMZABKWxRofAACwquRVAADyLJW8qqEMVKzWVe2zLqFs06/cNOsSVss/5r2SdQllG3rn+1mXsFr+sHfzni+VwAMAQGWSVwEAyLNU8mpWa0UDAAAAAAAAkHPuUAaAhFVnXQAAAKyAvAoAQJ6lklc1lAEgYVWFUtYlAADAcsmrAADkWSp51ZTXAJCwqsKabwAAsLbIq7Qkf/zjH7MuAQBoYqnkVXcoAwAAAACsZaNGjYo77rgjSqVl72TadNNNM6gIAGDVaCgDQMIq5Qo4WB1Lly6Nu+++Ozp16hQ77rhj1uUAAKtBXqWl2WOPPZYZKxQKMWvWrAyqAQDWVCp5VUMZABJWnUjgIR333ntv/Yd0Rx11VDz//POxaNGi2H333ePcc8/NuDoAoFzyKi1Jhw4d4oEHHoh11103Pvzww3jnnXeia9euWZcFAKyBVPKqhjIAJCyVK+hIx9ixY2PGjBkR8VFz+aWXXoolS5bEl7/85RXuV1dXF3V1dQ3GisXFUSy2WWu1AgArJ6/SknzhC1+IX/ziF/HYY4/FtGnTol27dtGuXbs477zz4sADD1zhvvIqAORTKnm1KusCAACgqZRKpfjsZz8bG2ywQbz66qvRq1ev2HXXXeMXv/jFCverra2NTp06Ndhqay9vpqoBAEjBvvvuG+PHj4/27dvH22+/Hf/+97/j8ccfjxNOOGGl+8qrAECW3KEMAAmrKpSyLgGaVKFQiD/+8Y9RKpViyJAhMXXq1HjiiSfiyCOPjOeff365+40bNy5qamoajBWLs9d2uQDASsirtCT33HNPHHfccTF16tTYcssto1u3bjFr1qwYM2bMSveVVwEgn1LJqxrKAJCwVKZkIR2lUilGjx4dERGtWrWKPn36RLdu3eKkk05a4X7FYjGKxeKnRk0fCABZk1dpSZ566ql49NFH47zzzos5c+bE7Nmzo1evXvG5z31upfvKqwCQT6nkVQ1lAEhYddYFQBObPHlyLFiwICI+uovj6KOPjj/84Q8xduzYjCsDAFaHvEpL0r59+3jllVdi0003jdmzZ8fxxx8fS5cujYkTJ8awYcOyLg8AWA2p5FUNZQAAWoyddtqp/s+bb7553HTTTfHzn/88dttttwyrAgCAiAMOOCCOO+64+O1vfxvHHHNMnHHGGVEoFOK73/1uvPjii1mXBwCwXBrKAJCwVKZkIU0bbrhhnHDCCVmXAQCsAXmVluToo4+OWbNmxec+97lYtGhRXH/99bF48eL44he/mHVpAMBqSiWvaigDQMKqCqWsSwAAgOWSV2lJCoVCXHDBBTF27Nh4+OGH49VXX41NNtkkvvGNb2RdGgCwmlLJqxrKAJCw6kSuoAMAoDLJq7REG2+8cWy88cZZlwEANIFU8mpZDeV58+ZFp06d1lYtAEAzS2VKFgAAKpO8CgBAnqWSV8tqKG+33XZxxx13RKm07O3bm266aZMVBQAAAAAAAED2yp7yeo899lhmrFAoxKxZs5qkIACg+aRyBR0AAJVJXgUAIM9SyatlNZQ/+9nPxjXXXBO9evUq6yR1dXVRV1fXYKxYXBzFYpuyjgMANK1UAg8AAJVJXgUAIM9SyatlNZQPOuigGDlyZPzwhz+Mzp07N/jezjvvvNz9amtrY+LEiQ3Gxo8fGxMmHF3O6QGAJlZdWHYZCwAAyAt5FQCAPEslr5bVUN55551jzJgxccwxx0SPHj2iqqoqIj6a8vqxxx5b7n7jxo2LmpqaBmPF4uzVKBcAAAAAAACA5lJWQ3nixIlx6aWXxhFHHBERETNnzowPPvggtthiixXuVywWo1gsfmrUdNcAkLWqrAsAAIAVkFcBAMizVPJqWc/z73//e4wcOTIiIs4666zYaaed4qtf/WoceeSRa6U4AGDtqiqs+QYAAGuLvAoAQJ6lklfLukN54403jquuuirGjh0bt956a7zwwgtRLBZj4403Xlv1AQBrUaUEFgAA0iSvAgCQZ6nk1bIayueee27ss88+MXny5HjnnXdi9913j7lz58YxxxyztuoDAACANVas7pR1CfV+P/T9rEuod9HfFmZdQgPj/rQ06xLq1fSbmXUJ9T63zuezLqFeq1Tm9AOAFqBN1WeyLqFRNVt1yLqERs19f0bWJTTqJ3/O5+tVrH496xIadfZ2/bIuoUUqq6Hcp0+feP755+Ppp5+OWbNmRdu2bWPQoEHRpUuXtVUfALAWVRdKWZcAAADLJa8CAJBnqeTVshrKHxswYEAMGDCgqWsBAJpZKlOyAABQmeRVAADyLJW8uloNZQCgZUgl8AAAUJnkVQAA8iyVvGrlGwAAAAAAAAAa5Q5lAEhYKlfQAQBQmeRVAADyLJW8qqEMAAmrTiTwAABQmeRVAADyLJW8qqEMAAmrKpSyLgEAAJZLXgUAIM9SyavWUAYAAAAAAACgUe5QBoCEubIMAIA8k1cBAMizVPKqhjIAJKwqkTU+AACoTPIqAAB5lkpe1VAGgIRVJxJ4AACoTPIqAAB5lkpeTeVObAAAEjBv3rysSwAAAACAFsUdygCQsKpCKesSoEltt912cccdd0SptOx7e9NNN82gIgBgTcirAADkWSp5VUMZABKWyhofpGWPPfZYZqxQKMSsWbMyqAYAWBPyKgAAeZZKXtVQBoCEpRJ4SMdnP/vZuOaaa6JXr15ZlwIANAF5FQCAPEslr2ooAwDQYhx00EExcuTI+OEPfxidO3du8L2dd955ufvV1dVFXV1dg7FicXEUi23WSp0AALR8/fr1izFjxsSYMWPW+FjyKgCQJQ1lgGbUqqpd1iWsln7rVd66o7ftPiPrEipCVdYFQBPbeeedY8yYMXHMMcdEjx49oqrqo3d5oVCIxx57bLn71dbWxsSJExuMjR8/NiZMOHqt1gsArJi8SiWbOHFi9O/fv0mOJa8CQD6lklc1lAEgYYVEpmQhHRMnToxLL700jjjiiIiImDlzZnzwwQexxRZbrHC/cePGRU1NTYOxYnH2WqsTAFg18iqVbL/99muyY8mrAJBPqeRVDWUASFgieYeE/P3vf48rr7wyIiLOOuusuOiii6JYLMawYcPiiiuuWO5+xWIxisXip0ZNHwgAWZNX4SPyKgDkUyp5NZU7sQEASMDGG28cV111VURE3HrrrfHCCy/ECy+8ELfffnu2hQEAAABAhXKHMgAkLJUpWUjHueeeG/vss09Mnjw53nnnndh9991j7ty5ccwxx2RdGgCwGuRVAADyLJW8qqEMAAkzVQktTZ8+feL555+Pp59+OmbNmhVt27aNQYMGRZcuXbIuDQBYDfIqAAB5lkpe1VAGgIQVCqWsS4C1YsCAATFgwICsywAA1pC8CgBAnqWSV1NpnAMAAAAAAABQJncoA0DCElniAwCACiWvAgCQZ6nkVQ1lAEhYIZXEAwBARZJXAQDIs1TyqoYyACQskbwDAECFklcBAMizVPKqNZQBAAAAAAAAaJSGMgAkrKqw5tvquOSSS6J3797Rtm3bGDhwYDz55JPLfeyVV14ZgwcPjvXWWy/WW2+9GDp06AofDwBAyyGvAgCQZ6nkVQ1lAEhYoQm2ct14441RU1MT48ePj2eeeSa23nrrGDZsWLzxxhuNPn7atGkxcuTIePDBB+Pxxx+PXr16xR577BFz5sxZjbMDAFBJ5FUAAPIslbxaKJVKpdWotQnMyOa0ACThnbrK/D2zXnHvZj3f39+9fY2PseW65dU8cODA2G677eLiiy+OiIilS5dGr1694uijj46TTjpppfsvWbIk1ltvvbj44otj1KhRq1UzrJrK/DkC5N8HS9/PuoR6F/3tzaxLaODl/7TKuoR6Nf3+k3UJ9T63zuezLqFeq6qtm/V88iqsiLwKVKZSZNSWWom57+fz5+rZz62TdQmNKlbn8+/x7O12a9bzpZJX3aEMADSbxYsXx9NPPx1Dhw6tH6uqqoqhQ4fG448/vkrHeP/99+ODDz6Izp07r60yAQBoQerq6mL+/PkNtrq6ukYfK68CANDcKiGvaigDQMKaYkqWcgLPW2+9FUuWLInu3bs3GO/evXvMnTt3lWo+8cQT47Of/WyD0AQAQMvUFHm1trY2OnXq1GCrra1t9HzyKgAA5Uglr+ZnLikAoNmtzhodn1ZbWxsTJ05sMDZ+/PiYMGFCExy9oZ/+9Kdxww03xLRp06Jt27ZNfnwAaA6tq9pnXUK947faMOsSGli8dH7WJdSreSLrCv7PF9dbtQ+GmsORmzfvlNdNkVfHjRsXNTU1DcaKxWITHHlZ8ioArFyhSX7DN70e7TfOuoRGnb3de1mX0Kg/zH0r6xJyIZW8qqEMAAmraoLEU07g6dKlS1RXV8frr7/eYPz111+PHj16rPA8P/vZz+KnP/1p3H///fHFL35xzYoGAKAiNEVeLRaLq/yBnLwKAEA5UsmrprwGANZIsViMjh07NtiWF4DatGkTAwYMiKlTp9aPLV26NKZOnRqDBg1a7jnOOeecOPPMM+Puu++ObbfdtsmfAwAARMirAADkW1Z5dbl3KPfr1y/GjBkTY8aMKfugn1ZXV7fMWorF4uIoFtus8bEBgNWXxQRDNTU1ceihh8a2224b22+/fVxwwQWxYMGCOOywwyIiYtSoUdGzZ8/6dULOPvvsOP300+O6666L3r17168F0qFDh+jQoUMGzwAAgOYirwIAkGep5NXlNpQnTpwY/fv3X8On9JHG11YcGxMmHN0kxwcAVk+hUGr2c44YMSLefPPNOP3002Pu3LnRv3//uPvuu6N79+4RETF79uyoqvq/SVQuvfTSWLx4cfzXf/1Xg+OsrXWaAQDID3kVAIA8SyWvFkql0lp/po3foTzbHcoArDXv1M3IuoTVsl5x72Y937/mT1njY/TtOLwJKoE8qsyfIwCVbPHS+VmXUK/mifeyLqHeF9dbnHUJ9Y7cfFiznk9ehRWRVwGaUimWZF1CoxYvyU8u/aQ/zH0r6xIatUfPrzXr+VLJq8u9Q7kpNb6YtGYyAAAAAAAAQJ41S0MZAMinQhaLfAAAwCqSVwEAyLNU8qqGMgAkrGrlDwEAgMzIqwAA5FkqeVVDGQASlsoVdAAAVCZ5FQCAPEslr6bSOAcAAAAAAACgTO5QBoCEJXIBHQAAFUpeBQAgz1LJqxrKAJCwVKZkIU3/+7//G+3bt4/OnTtnXQoAsJrkVQAA8iyVvGrKawBIWKEJNsibiy++ONZbb73YaqutYqONNopNNtkkJk2alHVZAMBqkFcBAMizVPKqO5QBAGgxLr/88pg8eXJMmzYttt5664iI+Otf/xoPPfRQxpUBAMBH7r333njllVeiQ4cOse2220bfvn2zLgkAYIU0lAEgYVWVcgkcrKLJkyfHTTfd1OBDuX79+kW/fv0yrAoAWF3yKi3BlClTYvjw4RERcfjhh8eMGTOib9++8d5778Wxxx4bP/3pT+PQQw/NuEoAYHWkklc1lAEgYYnkHRJSKpXin//8Z4wdOzaeeuqpWLx4cfTv3z8mTpwYQ4YMWe5+dXV1UVdX12CsWFwcxWKbtVwxALAi8iotwfHHH1/fUL7vvvti9uzZUfj/Cy7OnDkzRo0atdKGsrwKAPmUSl61hjIAJKxQKK3xBnmy9957x1577RUdOnSIP/zhDzFjxoyYOHFijBo1aoX71dbWRqdOnRpstbWXN1PVAMDyyKu0BKVSKaqqqqKqqirmzJkTrVu3jk033TTuuuuu2HjjjeOtt95a6THkVQDIp1TyqjuUAQBoMV588cUYPHhw3H777TF37tzo1atXzJgxI4YOHbrC/caNGxc1NTUNxorF2WuzVAAAElEoFGLRokWxdOnS6NevXzz33HPxxBNPxJgxY2LPPfeMUmnlHyTLqwBAljSUASBhqUzJQjqeffbZePjhh+Odd96JP/7xj/Huu+/GMcccEwMHDlzhfsViMYrF4qdGTR8IAFmTV2kJSqVSg+VX2rdvH/3794+FCxdGRER1dfVKjyGvAkA+pZJXNZQBIGGFVBIPyWjfvn28+uqrsemmm8ZLL70UP/vZz6JVq1YxYcKE2H333bMuDwAok7xKS3DttdfGwoULo1QqxQknnBCjR4+Oxx57LE488cSIiOjZs2dceeWVMWLEiOjYsWPG1QIA5Uglr2ooA0DCEsk7JGSfffaJsWPHxm233RbHHHNMnHbaadGqVas46qijYtasWVmXBwCUSV6lJdh+++3r/zxlypSYMmVKjBkzpn580qRJMXbs2FiyZEl897vfzapMAGA1pJJXNZQBAGgxampq4m9/+1tstNFG8cEHH8Q999wTCxYsiK222irr0gAAIHr06BFHHHFEg7ENN9wwpkyZklFFAAArp6EMAAmryroAaGKtWrWKq6++Op5//vl45JFH4rXXXotNNtkkRowYkXVpAMBqkFcBAMizVPKqhjIAJCyVNT5Iz1ZbbeWuZABoAeRVAADyLJW8qqEMQIvUvlX3rEuoEIkkHgAAKpS8CgBAnqWRVzWUASBhhUQCDwAAlUleBQAgz1LJq6lM7Q0AAAAAAABAmdyhDAAJKxRcWwYAQH7JqwAA5FkqeVVDGQCSlsaULAAAVCp5FQCAPEsjr2ooA0DCUlnjAwCAyiSvAgCQZ6nk1TTuwwYAAAAAAACgbO5QBoCkpXEFHQAAlUpeBQAgz9LIqxrKAJCwQsFkJQAA5Je8CgBAnqWSVzWUASBpaVxBBwBUhjZVHbMuod4vBn0m6xLqvffB7KxLyJC8CgA0j0JUZ11Co4rV62ZdQqN27rE06xJyIo28mkbbHAAAAAAAAICyuUMZABJWSOQKOgAAKpO8CgBAnqWSVzWUASBhqQQeAAAqk7wKAECepZJXNZQBIGlWvwAAIM/kVQAA8iyNvJrGswQAAAAAAACgbO5QBoCEFQppTMkCAEBlklcBAMizVPKqhjIAJC2NwAMAQKWSVwEAyLM08qqGMgAkrJBI4AEAoDLJqwAA5FkqedUaygAAAAAAAAA0yh3KAJA015YBAJBn8ioAAHmWRl7VUAaAhKUyJQsAAJVJXgUAIM9SyasaygCQsEIhjcADAEBlklcBAMizVPJqGvdhAwAAAAAAAFA2dygDQNLSuIIOAIBKJa8CAJBnaeRVDWUASFjBZCUAAOSYvAoAQJ6lklc1lAEgaWlcQQcAQKWSVwEAyLM08moabXMAAAAAAAAAyuYOZQBIWKGQxhV0AABUJnkVAIA8SyWvaigDQNLSCDwAAFQqeRUAgDxLI69qKANAwgpWvwAAIMfkVQAA8iyVvJrGswQAAAAAAACgbO5QBoCkpTElCwAAlUpeBQAgz9LIqxrKAJCwQiKBBwCAyiSvAgCQZ6nkVQ1lAEhYoZBG4AEAoDLJqwAA5FkqedUaygAAAAAAAAA0yh3KAJA015YBAJBn8ioAAHmWRl7VUAaAhKWyxgcAAJVJXgUAIM9SyasaygCQtDQCDwAAlUpeBQAgz9LIq2nchw0A5Moll1wSvXv3jrZt28bAgQPjySefXOHjb7755th8882jbdu2sdVWW8Wdd97ZTJUCAJAieRUAgDxr7ryqoQwACSsUCmu8levGG2+MmpqaGD9+fDzzzDOx9dZbx7Bhw+KNN95o9PGPPfZYjBw5Mr797W/Hs88+G/vuu2/su+++8de//nVNnz4AADknrwIAkGep5NVCqVQqlV1pk5iRzWkBSELdknlZl7BaitXbNev5SjF9jY9RiM3KevzAgQNju+22i4svvjgiIpYuXRq9evWKo48+Ok466aRlHj9ixIhYsGBB3H777fVjX/7yl6N///5x2WWXrVnxsELyKkDKSpHRxyWNeO+D2VmXUK9j692b9XzyKqyIvApAdhYt+XfWJTSqbfWXm/V8qeRVdygDQMIKTfC/cixevDiefvrpGDp0aP1YVVVVDB06NB5//PFG93n88ccbPD4iYtiwYct9PAAALYe8CgBAnqWSV1uVVSUAwKfU1dVFXV1dg7FisRjFYnGZx7711luxZMmS6N69e4Px7t27xz/+8Y9Gjz937txGHz937tw1rBwAgBTIqwAA5FlF5NVSC7No0aLS+PHjS4sWLcq6lFVWiTWXSupuTpVYc6lUmXVXYs2lkrqbUyXWvLaNHz++FBENtvHjxzf62Dlz5pQiovTYY481GD/hhBNK22+/faP7tG7dunTdddc1GLvkkktK3bp1a5L6YW3J08+LPNVSKuWrHrWoRS1qUUvLJ6/CiuX1Z4S6yqOu8qirPOoqj7rKk9e6mlMl5NUM11BeO+bPnx+dOnWKefPmRceOHbMuZ5VUYs0R6m5OlVhzRGXWXYk1R6i7OVVizWtbOVfQLV68ONq3bx+33HJL7LvvvvXjhx56aLz77rvxu9/9bpl9Pv/5z0dNTU0ce+yx9WPjx4+P2267Lf7yl7802fOAppannxd5qiVv9ahFLWpRi1paPnkVViyvPyPUVR51lUdd5VFXedRVnrzW1ZwqIa9aQxkAWCPFYjE6duzYYGss7EREtGnTJgYMGBBTp06tH1u6dGlMnTo1Bg0a1Og+gwYNavD4iIj77rtvuY8HAIBPklcBAMizSsir1lAGAJpVTU1NHHroobHtttvG9ttvHxdccEEsWLAgDjvssIiIGDVqVPTs2TNqa2sjIuKYY46JXXbZJc4777zYa6+94oYbboinnnoqrrjiiiyfBgAALZS8CgBAnmWRVzWUAYBmNWLEiHjzzTfj9NNPj7lz50b//v3j7rvvju7du0dExOzZs6Oq6v8mUdlhhx3iuuuui1NPPTVOPvnk2GSTTeK2226Lfv36ZfUUAABoweRVAADyLIu82uIaysViMcaPH7/cW8HzqBJrjlB3c6rEmiMqs+5KrDlC3c2pEmvOo7Fjx8bYsWMb/d60adOWGdt///1j//33X8tVQdPK08+LPNUSka961KIWtahFLTRGXiUlef0Zoa7yqKs86iqPusqjrvLkta68a+68WiiVSqXV3hsAAAAAAACAFqtq5Q8BAAAAAAAAIEUaygAAAAAAAAA0SkMZaHJDhgyJY489NusyVlul18/a5f0BVCo/v6Bl8G95xbw+y+e1ASqNn1urxutUHq8XqfGep6m0yroAYOWGDBkS/fv3jwsuuCDrUpLw29/+Nlq3bp11GQDQpPLy+02uAchGXn4PAKyqPP7ckmXL4/UiRd73tFQayjSJxYsXR5s2bbIuA5pE586dsy4BAJqc328AafN7AKg0fm4BQH60qCmvb7nllthqq62iXbt2sf7668fQoUNjwYIFWZe1QnfffXfstNNOse6668b6668fe++9d/zrX//KuqyVGjJkSIwdOzaOPfbY6NKlSwwbNizrklaqUl/r0aNHxx/+8Ie48MILo1AoRKFQiJdeeinrslZq6dKl8aMf/Sg6d+4cPXr0iAkTJmRd0iqrpGlA3nvvvTj44INjnXXWiQ022CB+/vOfV0T9lfrvEWBVXHHFFfHZz342li5d2mB8n332icMPPzyjqvLx+y1PuSYvv4t+/etfx/rrrx91dXUNxvfdd9845JBDmr2evLwuER+9Z3/wgx/kIlPmOXPdcccd0alTp7j22mszOX+e3jN5ksfXJS/vWSBf8vx5at5+buUpy35aHj+H83qVJ4/ZIa915em/Ez4tr+/7PL7nI/L5/mL5WkxD+bXXXouRI0fG4YcfHi+88EJMmzYtvvnNb0apVMq6tBVasGBB1NTUxFNPPRVTp06Nqqqq+MY3vrHMB5B5dPXVV0ebNm3i0UcfjcsuuyzrclaqUl/rCy+8MAYNGhRHHHFEvPbaa/Haa69Fr169si5rpa6++upYZ5114oknnohzzjknzjjjjLjvvvuyLqvFqampiUcffTR+//vfx3333RcPP/xwPPPMM1mXtVKV+u8RYFXsv//+8fbbb8eDDz5YP/bvf/877r777jj44IMzrCx7eco1efldtP/++8eSJUvi97//ff3YG2+8EXfccUcmFyDk5XX5WF4yZV4z13XXXRcjR46Ma6+9NrOfL3l7z+SF1wWoBJX6eWpW8pRlPy0vmemTvF7lyWt2yGtdefw7jMjv+z6vr1de3180rsVMef3aa6/Fhx9+GN/85jdjww03jIiIrbbaKuOqVm6//fZr8PWkSZOia9eu8fe//z369euXUVWrZpNNNolzzjkn6zJWWaW+1p06dYo2bdpE+/bto0ePHlmXs8q++MUvxvjx4yPio/fKxRdfHFOnTo3dd98948pajvfeey+uvvrquO6662K33XaLiIjJkyfHZz/72YwrW7lK/fcIsCrWW2+92HPPPRv8fL7llluiS5cu8ZWvfCXj6rKVp1yTl99F7dq1i4MOOigmT54c+++/f0RE/OY3v4nPf/7zMWTIkGar42N5eV0+lodMmdfMdckll8Qpp5wSU6ZMiV122SWzOvL2nskLrwtQCSr189Ss5CnLfloeMtOneb3Kk9fskNe68vh3GJHf931eX6+8vr9oXIu5Q3nrrbeO3XbbLbbaaqvYf//948orr4x33nkn67JW6p///GeMHDky+vTpEx07dozevXtHRMTs2bOzLWwVDBgwIOsSylLJr3Ul+uIXv9jg6w022CDeeOONjKppmWbNmhUffPBBbL/99vVjnTp1is022yzDqlaNf49AS3fwwQfHrbfeWj+N8bXXXhsHHnhgVFW1mPhd8fL0u+iII46Ie++9N+bMmRMREVdddVWMHj06CoVCs9eSp9clIh+ZMo+Z65Zbbonjjjsu7rvvvkybyRH5e8/khdcFqASV+nkqy8pDZqokeXy98pod8lpXHv8O8yyvr1de3180rsV8olVdXR333Xdf3HXXXbHlllvGL37xi9hss83ixRdfzLq0FRo+fHj8+9//jiuvvDKeeOKJeOKJJyIiYvHixRlXtnLrrLNO1iWUpZJf60rUunXrBl8XCgVTVVDPv0egpRs+fHiUSqW444474pVXXomHH344+emu8yZPv4u22Wab2HrrrePXv/51PP300/G3v/0tRo8e3ex1ROTrdYmQKZdnm222ia5du8akSZMyn5Y0b++ZvPC6AJWgUj9PZVkyU3ny+HrlNTvkta48/h3mWV5fr7y+v2hci2koR3z0j2DHHXeMiRMnxrPPPhtt2rSJ//mf/8m6rOV6++23Y/r06XHqqafGbrvtFltssYWrANeSSn+t27RpE0uWLMm6DHKmT58+0bp16/jTn/5UPzZv3ryYMWNGhlWtXKX/ewRYFW3bto1vfvObce2118b1118fm222WXzpS1/KuqxcyEOuyePvou985ztx1VVXxeTJk2Po0KGZrLOVx9clD/KYufr27RsPPvhg/O53v4ujjz46szq8ZxrndQEqSaV9npq1PGTZSuL1WjV5zQ55rSvvvO9XjfdX5Wkxayg/8cQTMXXq1Nhjjz2iW7du8cQTT8Sbb74ZW2yxRdalLdd6660X66+/flxxxRWxwQYbxOzZs+Okk07KuqwWqdJf6969e8cTTzwRL730UnTo0CE6d+5sykziM5/5TBx66KFxwgknROfOnaNbt24xfvz4qKqqymSKzFVV6f8eAVbVwQcfHHvvvXf87W9/i29961tZl5Mbecg1efxddNBBB8UPf/jDuPLKK+PXv/51JjXk8XXJg7xmrk033TQefPDBGDJkSLRq1SouuOCCZq/Be6ZxXhegUlTi56lZy0OWrSRer1WT1+yQ17ryzvt+1Xh/VZ4W8y7u2LFjPPTQQ/G1r30tNt100zj11FPjvPPOiz333DPr0parqqoqbrjhhnj66aejX79+cdxxx8W5556bdVktUqW/1j/84Q+juro6ttxyy+jatas1BKh3/vnnx6BBg2LvvfeOoUOHxo477hhbbLFFtG3bNuvSlqvS/z0CrKpdd901OnfuHNOnT4+DDjoo63JyIw+5Jo+/izp16hT77bdfdOjQIfbdd99Masjj65IXec1cm222WTzwwANx/fXXx/HHH9/s5/eeaZzXBagUlfh5atbykGUriddr1eQ1O+S1rrzzvl813l+Vp1DKesElAJrMggULomfPnnHeeefFt7/97azLAQDKsNtuu8UXvvCFuOiii7IuhZWQuQAAAEhJi5nyGiBFzz77bPzjH/+I7bffPubNmxdnnHFGRETss88+GVcGAKyqd955J6ZNmxbTpk2LX/7yl1mXQyNkLgAAAFKmoQxQ4X72s5/F9OnTo02bNjFgwIB4+OGHo0uXLlmXBQCsom222SbeeeedOPvss2OzzTbLuhyWQ+YCAAAgVaa8BgAAAAAAAKBRVVkXAAAAAAAAAEA+aSgDAAAAAAAA0CgNZQAAAAAAAAAapaEMAAAAAAAAQKM0lAEAAAAAAABolIYyAAAAAAAAAI3SUAYAAAAAAACgURrKAAAAAAAAADRKQxkAAAAAAACARmkoAwAAAAAAANAoDWUAAAAAAAAAGqWhDAAAAAAAAECjNJQBAAAAAAAAaJSGMgAAAAAAAACN0lAGAAAAAAAAoFEaygAAAAAAAAA0SkMZAAAAAAAAgEZpKAMAAAAAAADQKA1lAAAAAAAAABqloQwAAAAAAABAozSUAQAAAAAAAGiUhjKZueqqq6JQKMRLL72UdSm5MW3atCgUCjFt2rSsS6mv5ZZbbsm6lCYzYcKEKBQKDcYKhUKMHTt2jY7bFMcAAGgqcvaaGTJkSAwZMiTrMgAAWhQZFaCyaSjD//f3v/89JkyYINSsJXfeeWdMmDAh6zJW22OPPRYTJkyId999N+tSAAAqipwNAEDeVHJGrfTPWYHKpKFMZg455JBYuHBhbLjhhlmXEhEfhYiJEydWZIioBHfeeWdMnDgx6zJW22OPPRYTJ07UUAYAck/OBgAgb2TUplPpn7MClalV1gWQrurq6qiurs66DAAAaFHkbAAA8kZGzbdSqRSLFi2Kdu3aLfO9RYsWRZs2baKqyv2JkDI/AchMY+tm9O7dO/bee+945JFHYvvtt4+2bdtGnz594te//nWj+z700ENx1FFHxfrrrx8dO3aMUaNGxTvvvNPgsYVCodEpQHr37h2jR4+uP97+++8fERFf+cpXolAoNOlaxi+//HKMGTMmNttss2jXrl2sv/76sf/++6/SFXAPP/xw7L///vH5z38+isVi9OrVK4477rhYuHBh/WN+9rOfRaFQiJdffnmZ/ceNGxdt2rSpf10++bw/aVXWiqurq4u99947OnXqFI899tgq1zd69Oi45JJLIiLqX9tPrmX8s5/9LHbYYYdYf/31o127djFgwIBG127+eK3i2267Lfr16xfFYjG+8IUvxN13373MYx955JHYbrvtom3bttG3b9+4/PLLV/jcVnTMCRMmxAknnBARERtttFF9/Z/++1tZXWvyPgAAWFUp5ewJEyZEoVCImTNnxujRo2PdddeNTp06xWGHHRbvv/9+g8dOnjw5dt111+jWrVsUi8XYcsst49JLL13h8UulUnTp0iVqamrqx5YuXRrrrrtuVFdXN5i95uyzz45WrVrFf/7zn4iIeO6552L06NHRp0+faNu2bfTo0SMOP/zwePvtt5vkuQMAVJJUMupzzz0XhUIhfv/739ePPf3001EoFOJLX/pSg8fuueeeMXDgwAZjd911VwwePDjWWWed+MxnPhN77bVX/O1vf6v//so+Z126dGlccMEF8YUvfCHatm0b3bt3j6OOOmqZ1+nj1/6ee+6JbbfdNtq1axeXX355TJs2LQqFQtxwww1x6qmnRs+ePaN9+/Yxf/78NX5tgMrmDmVyZ+bMmfFf//Vf8e1vfzsOPfTQmDRpUowePToGDBgQX/jCFxo8duzYsbHuuuvGhAkTYvr06XHppZfGyy+/XP+Lb1XtvPPO8YMf/CAuuuiiOPnkk2OLLbaIiKj//zX1pz/9KR577LE48MAD43Of+1y89NJLcemll8aQIUPi73//e7Rv3365+958883x/vvvx/e+971Yf/3148knn4xf/OIX8b//+79x8803R0TEAQccED/60Y/ipptuqm98fuymm26KPfbYI9Zbb701eg4LFy6MffbZJ5566qm4//77Y7vttlvl+o466qh49dVX47777otrrrlmmWNfeOGF8fWvfz0OPvjgWLx4cdxwww2x//77x+233x577bVXg8c+8sgj8dvf/jbGjBkTn/nMZ+Kiiy6K/fbbL2bPnh3rr79+REQ8//zzsccee0TXrl1jwoQJ8eGHH8b48eOje/fujT63lR3zm9/8ZsyYMSOuv/76+PnPfx5dunSJiIiuXbuWVdeavA8AANZUS8zZHzvggANio402itra2njmmWfiv//7v6Nbt25x9tln1z/m0ksvjS984Qvx9a9/PVq1ahVTpkyJMWPGxNKlS+P73/9+o8ctFAqx4447xkMPPVQ/9txzz8W8efOiqqoqHn300fq8+vDDD8c222wTHTp0iIiI++67L2bNmhWHHXZY9OjRI/72t7/FFVdcEX/729/ij3/8Y1mvIwBAS9XSMmq/fv1i3XXXjYceeii+/vWvR8RHObGqqir+8pe/xPz586Njx46xdOnSeOyxx+LII4+s3/eaa66JQw89NIYNGxZnn312vP/++3HppZfGTjvtFM8++2z07t17pZ+zHnXUUXHVVVfFYYcdFj/4wQ/ixRdfjIsvvjieffbZePTRR6N169b1j50+fXqMHDkyjjrqqDjiiCNis802q//emWeeGW3atIkf/vCHUVdXF23atFnj1waocCXIyOTJk0sRUXrxxRfrxzbccMNSRJQeeuih+rE33nijVCwWS8cff/wy+w4YMKC0ePHi+vFzzjmnFBGl3/3ud/VjEVEaP378MuffcMMNS4ceemj91zfffHMpIkoPPvhgkzy/T3r//feXGXv88cdLEVH69a9/XT/24IMPLlNDY/vW1taWCoVC6eWXX64fGzRoUGnAgAENHvfkk08uc45PP++P7bLLLqVddtllmVpuvvnm0nvvvVfaZZddSl26dCk9++yzK31ujdX3/e9/v7S8HzmfPsbixYtL/fr1K+26664NxiOi1KZNm9LMmTPrx/7yl7+UIqL0i1/8on5s3333LbVt27bB+f/+97+Xqqurl6lhVY957rnnLvN+LfcYq/o+AABYEynl7PHjx5cionT44Yc3GP/GN75RWn/99RuMNZbFhg0bVurTp0+DsU/n4nPPPbdUXV1dmj9/fqlUKpUuuuii0oYbbljafvvtSyeeeGKpVCqVlixZUlp33XVLxx133ArPd/311y/z9wAAkIKUMupee+1V2n777eu//uY3v1n65je/Waquri7dddddpVKpVHrmmWca1P7ee++V1l133dIRRxzR4Fhz584tderUqcH48j5nffjhh0sRUbr22msbjN99993LjH/82t99990NHvvxZ8J9+vRpNM8C6TLlNbmz5ZZbxuDBg+u/7tq1a2y22WYxa9asZR575JFHNriq6nvf+160atUq7rzzzmapdVV9cu2JDz74IN5+++3YeOONY911141nnnlmlfddsGBBvPXWW7HDDjtEqVSKZ599tv57I0aMiKeffjr+9a9/1Y/deOONUSwWY5999lnt2ufNmxd77LFH/OMf/4hp06ZF//79V6u+FfnkMd55552YN29eDB48uNHXZujQodG3b9/6r7/4xS9Gx44d698fS5YsiXvuuSf23Xff+PznP1//uC222CKGDRvW6PlXdsxVsSrHWJP3AQDAmmqJOftj3/3udxt8PXjw4Hj77bcbTM33ySw2b968eOutt2KXXXaJWbNmxbx585Z77MGDB8eSJUsaLPkyePDgGDx4cDz88MMREfHXv/413n333Qav7yfPt2jRonjrrbfiy1/+ckSE7AcA8P+1xIz68eeaCxYsiIiPZjb82te+Fv3796/Pjw8//HAUCoXYaaedIuKj2W3efffdGDlyZLz11lv1W3V1dQwcODAefPDBlZ735ptvjk6dOsXuu+/e4BgDBgyIDh06LHOMjTbaaLmflx566KGNrqcMpEtDmdz5ZBPwY+utt94y6zxERGyyySYNvu7QoUNssMEGa31N2oULF8bcuXMbbCt7/Omnnx69evWKYrEYXbp0ia5du8a77767wg+vIiJmz54do0ePjs6dO0eHDh2ia9euscsuu0RENNh3//33j6qqqrjxxhsj4qP13m6++ebYc889o2PHjqv9XI899tj405/+FPfff/8y08yUU9+K3H777fHlL3852rZtG507d46uXbvGpZde2uj+K3t/vPnmm7Fw4cJl3hsR0WDalnKOuSpW5Rhr8j4AAFhTLTFnf+zTz+3j5V4++dweffTRGDp0aKyzzjqx7rrrRteuXePkk0+OiBXn1i996UvRvn37Bh/+DR48OHbeeed46qmnYtGiRfXf+/gDwYiIf//733HMMcdE9+7do127dtG1a9fYaKONVno+AICUtMSMOnjw4Pjwww/j8ccfj+nTp8cbb7xRnx8/mSm33HLL6Ny5c0RE/POf/4yIiF133TW6du3aYLv33nvjjTfeWGmd//znP2PevHnRrVu3ZY7xn//8Z5ljfJxNG7Oi7wFpsoYyuVNdXd3oeKlUatLzLFmyZLX3vfHGG+Owww5rMLai+o4++uiYPHlyHHvssTFo0KDo1KlTFAqFOPDAA2Pp0qUrrHH33XePf//733HiiSfG5ptvHuuss07MmTMnRo8e3WDfz372szF48OC46aab4uSTT44//vGPMXv27AbrxkXEctcTWbJkSaOv/T777BM33HBD/PSnP41f//rXUVVV1WCfVa1veR5++OH4+te/HjvvvHP88pe/jA022CBat24dkydPjuuuu26Zx6+N90dTHHNVjrG67wMAgKbQEnP2x1b23P71r3/FbrvtFptvvnmcf/750atXr2jTpk3ceeed8fOf/3yFWax169YxcODAeOihh2LmzJkxd+7cGDx4cHTv3j0++OCDeOKJJ+Lhhx+OzTffPLp27Vq/3wEHHBCPPfZYnHDCCdG/f//o0KFDLF26NL761a/KfgAA/19LzKjbbrtttG3bNh566KH4/Oc/H926dYtNN900Bg8eHL/85S+jrq4uHn744fjGN75Rv8/H+fCaa66JHj16LHPMVq1W3spZunRpdOvWLa699tpGv//JrBoRK7wD2d3JwKdpKFPR/vnPf8ZXvvKV+q//85//xGuvvRZf+9rX6sfWW2+9ePfddxvst3jx4njttdcajC2v0dqYYcOGxX333bfKj7/lllvi0EMPjfPOO69+bNGiRcvU9WnPP/98zJgxI66++uoYNWpU/fjyzj1ixIgYM2ZMTJ8+PW688cZo3759DB8+vMFjGns9IiJefvnl6NOnzzLj++67b+yxxx4xevTo+MxnPhOXXnrpatW3vNf31ltvjbZt28Y999wTxWKxfnzy5MmNPn5lunbtGu3atau/qu+Tpk+fvlrHjCjv/bE8q/s+AABobpWSs1fVlClToq6uLn7/+983uAtmVaYOjPjoLpOzzz477r///ujSpUtsvvnmUSgU4gtf+EI8/PDD8fDDD8fee+9d//h33nknpk6dGhMnTozTTz+9fryxjAoAwKqplIzapk2b2H777ePhhx+Oz3/+8/VTeg8ePDjq6uri2muvjddffz123nnn+n0+XkqvW7duMXTo0BUef3m19+3bN+6///7YcccdNYSBJmfKayraFVdcER988EH915deeml8+OGHseeee9aP9e3bNx566KFl9vv0VWnrrLNORMQqNfc22GCDGDp0aINtRaqrq5e5au0Xv/jFSq+M+/gKvU/uWyqV4sILL2z08fvtt19UV1fH9ddfHzfffHPsvffe9c/rY3379o0//vGPsXjx4vqx22+/PV555ZXl1jFq1Ki46KKL4rLLLosTTzxxtepb3utbXV0dhUKhwWvx0ksvxW233bbcelakuro6hg0bFrfddlvMnj27fvyFF16Ie+65Z7WOGVHe+2NFta3O+wAAoLlVSs5eVY3l1nnz5q3yRYwff/h3wQUXxE477VT/Id7gwYPjmmuuiVdffbXB2n+NnS8i4oILLliTpwEAkLRKyqiDBw+OJ554Ih588MH6nNilS5fYYost6meU/GR+HDZsWHTs2DHOOuusBs/xY2+++eZKaz/ggANiyZIlceaZZy6z/4cffuimFmCNuEOZirZ48eLYbbfd4oADDojp06fHL3/5y9hpp53i61//ev1jvvOd78R3v/vd2G+//WL33XePv/zlL3HPPfdEly5dGhyrf//+UV1dHWeffXbMmzcvisVi7LrrrtGtW7c1rnPvvfeOa665Jjp16hRbbrllPP7443H//ffH+uuvv8L9Nt988+jbt2/88Ic/jDlz5kTHjh3j1ltvXe7avt26dYuvfOUrcf7558d7770XI0aMWOYx3/nOd+KWW26Jr371q3HAAQfEv/71r/jNb35TfxXc8owdOzbmz58fp5xySnTq1ClOPvnksuobMGBARET84Ac/iGHDhkV1dXUceOCBsddee8X5558fX/3qV+Oggw6KN954Iy655JLYeOON47nnnlthTcszceLEuPvuu2Pw4MExZsyY+PDDD+MXv/hFfOELX1jtY35c/ymnnBIHHnhgtG7dOoYPH75Mw35FVvd9AADQ3ColZ6+qPfbYI9q0aRPDhw+Po446Kv7zn//ElVdeGd26dVvmbpXGDBo0KFq1ahXTp0+PI488sn585513rp/B55MfCHbs2DF23nnnOOecc+KDDz6Inj17xr333hsvvvhi0z85AIBEVFJGHTx4cPzkJz+JV155pUFO3HnnnePyyy+P3r17x+c+97n68Y4dO8all14ahxxySHzpS1+KAw88MLp27RqzZ8+OO+64I3bccce4+OKLI2L5n7PusssucdRRR0VtbW38+c9/jj322CNat24d//znP+Pmm2+OCy+8MP7rv/6rSZ4fkB53KFPRLr744thiiy3i9NNPj6uuuipGjhwZv/vd7xpM+3HEEUfEiSeeGA899FAcf/zx8eKLL8Z99923TCOwR48ecdlll8Ubb7wR3/72t2PkyJHx97//vUnqvPDCC2PUqFFx7bXXxvHHHx+vvfZa3H///dGhQ4cV7te6deuYMmVK9O/fP2pra2PixImxySabxK9//evl7jNixIh477334jOf+UyD6V4+NmzYsDjvvPNixowZceyxx8bjjz8et99+e4MAszwnn3xy/OhHP4pTTjklLrnkkrLq++Y3vxlHH3103H333XHIIYfEyJEjIyJi1113jV/96lcxd+7cOPbYY+P666+Ps88+u8EaIuX64he/GPfcc0907do1Tj/99Jg0aVJMnDhxjY653XbbxZlnnhl/+ctfYvTo0TFy5MgGVwauitV9HwAANLdKydmrarPNNotbbrklCoVC/PCHP4zLLrssjjzyyDjmmGNWaf911lknttlmm4iI2GmnnerHP/5wsFevXrHhhhs22Oe6666LYcOGxSWXXBLjxo2L1q1bx1133dVEzwgAID2VlFF32GGHqK6ujs985jOx9dZb149/cvrrTzvooINi6tSp0bNnzzj33HPjmGOOiRtuuCH69+/fYA3n5X3OGhFx2WWXxRVXXBFvvPFGnHzyyTFu3Lh44IEH4lvf+lbsuOOOTfb8gPQUSk29uj00g6uuuioOO+yw+NOf/hTbbrtt1uUAAECLIGcDAJA3MipA9tyhDAAAAAAAAECjNJQBAAAAAAAAaJSGMgAAAAAAAACNsoYyAAAAAAAAAI1yhzIAAAAAAAAAjdJQBgAAAAAAAKBRGsoAAAAAAAAANEpDGQAAAAAAAIBGtcru1DMyOWu7z4/P5LwREQtnT8zs3ABUik2b9WztPj9yjY+xcPb1TVAJ5FE2eTVFWWb01PhvEmDNyauQH/JqOWTO8siNQOWSV9eGDBvKAEDWCgWTlQAAkF/yKgAAeZZKXtVQBoCEFax+AQBAjsmrAADkWSp5NY1nCQAAAAAAAEDZ3KEMAAlLZUoWAAAqk7wKAECepZJXNZQBIGGpBB4AACqTvAoAQJ6lklc1lAEgYYVCIesSAABgueRVAADyLJW8mkbbHAAAAAAAAICyuUMZAJLm2jIAAPJMXgUAIM/SyKtlNZRnzJgR06ZNa/R73bp1i3333bcJSgIAmksqa3wAAFCZ5FUAAPIslbxaVkP5zTffjMcff7zB2P/8z//EN77xjejTp4+GMgBUmFQCDwAAlUleBQAgz1LJq2U1lHfcccfYcccdIyJi+vTp8frrr8ejjz4akydPXivFAQAAAAAAAJCdstdQfuqpp+Kggw6K1q1bR5cuXWLWrFkxZsyY+OUvf7k26gMA1qJCImt8AABQmeRVAADyLJW8WnZD+aCDDoodd9wxdt999xgxYkRsuummsXDhwrj44otj7Nixje5TV1cXdXV1DcaKxcVRLLZZvaoBgCaRypQsAABUJnkVAIA8SyWvlvUs//3vf8fMmTNjvfXWi/vuuy9qa2ujuro6zj///LjuuuuWu19tbW106tSpwVZbe/kaFw8ArJlCoWqNNwAAWFvkVQAA8iyVvFrWHcodO3aMzp07x6BBg+Kvf/1rvP766xERsd5668V777233P3GjRsXNTU1DcaKxdmrUS4A0JQqJbAAAJAmeRUAgDxLJa+W1VBu1apVTJkyJc4555zo2LFj/PjHP4577rknIiLmz5+/3P2KxWIUi8VPjZruGgAAAABIw7vvvhtvvPFGo9/bdNNNm7kaAIBVV/YayoMGDYr/+Z//qf+6UCjEggULYp111mnSwgCAta8QhaxLAACA5ZJXaUluueWW+MlPfrLM+Kuvvhp1dXUZVAQArKlU8mrZDeXGHHLIIfFf//VfTXEoAKAZpTIlCwAAlUlepSX5zne+E9/5zneWGd9kk00yqAYAaAqp5NU1bihPnjw5/vSnP8X3v//9pqgHAGhGqQQeAAAqk7xKS/Tqq6/Gs88+G61bt44ddtghCoU07mwCgJYolby6xg3lHXbYIXbYYYemqAUAAAAAoMU66aST4pprron+/ftHXV1d/PnPf46FCxeudL+6urplpsUuFhdHsdhmbZUKAFAvjbY5ANCoQqFqjTcAAFhb5FVakosuuihmzJgRM2fOjDvuuCPuv//+uP/++2Px4sUr3be2tjY6derUYKutvbwZqgYAViSVvNokaygDAJWqMgILAACpkldpOW688ca4/vrro127dhERMX/+/OjVq1e89NJLK9133LhxUVNT02CsWJy9NsoEAMqSRl5N41kCAI1K5Qo6AAAqk7xKS/L+++/HOuusExERV1xxRfTq1Sv69u0bl1122Ur3LRaL0bFjxwab6a4BIHup5NXKqBIAAAAAoIJ97Wtfi5/85CcREfGrX/0qpk+fHq+//npMnjw548oAAFbMlNcAkLBKuQIOAIA0yau0JKeddloceOCBMXDgwJg/f34ceOCBMW/evDjggAOyLg0AWE2p5FUNZQBIWMFkJbRAL7zwQpx33nnx6quvxuabbx7f+973YpNNNsm6LABgNcirtCRt27aN2267Lf7whz/E008/HW3atIlddtklttpqq6xLAwBWUyp5VUMZABKWyhV0tHznn39+1NTURETE17/+9fjOd74T++yzT7z44ovxjW98I6688soYNGjQcvevq6uLurq6BmPF4mLr0gFAxuRVWqJddtkldtlll6zLAACaQCp5NbmG8sLZEzM7d/sNszn3gpdPz+S8ERGFKGR2bgAgHZdddlkMHz48SqVSRETsvPPOsf7668eXv/zlGDJkSJxwwglxzz33LHf/2tramDixYVYbP35sTJhw9FqtGwAAAADyLrmGMgDwfwoFF/7QMpRKpdhjjz3qvz7ooIMiIuJb3/pWnHnmmfHiiy+ucP9x48bV3+H8sWJxdtMXCgCURV4FACDPUsmrGsoAkLBUpmSh5SsUCitsGldVrfi9XiwWo1gsfmrUdNcAkDV5FQCAPEslr2ooA0DCCpFG4KHlK5VK8bWvfW2Z8b322iu+//3vR6lUirfffjvWX3/9DKoDAFaXvAoAQJ6lklc1lAEAqHiXXXZZ/Oc//1lmfKONNoqIiO9+97tx9913x8EHH9zcpQEAAABARdNQBoCEpTIlCy3fbrvttsLvH3fccc1UCQDQlORVAADyLJW8qqEMAAlLJfAAAFCZ5FUAAPIslbyqoQwACUtljQ8AACqTvAoAQJ6lklfTeJYAAAAAAAAAlM0dygCQskSmZAEAoELJqwAA5FkiebWsZ7l06dKora2NzTbbLNZZZ53o27dvnHLKKbFw4cK1VR8AsBYVClVrvAEAwNoirwIAkGep5NWyqjzxxBPjkUceiUmTJsWrr74aDz30UBSLxfjWt761tuoDANaiQqGwxhsAAKwt8ioAAHmWSl4ta8rrdu3axc033xzt27ePiIhOnTrF6aefHr17914btQEAAAAAAACQobIaymeccUaDr99555247bbbolevXivcr66uLurq6hqMFYuLo1hsU87pAYAmVihvshIAAGhW8ioAAHmWSl4tq6G8ePHiOOWUU+KWW26JOXPmRIcOHWLw4MHxm9/8ZoX71dbWxsSJExuMjR8/NiZMOLr8igGAJlMpa3QAAJAmeRVoKgtnT1z5g6i3zoZnZl1CRfnPyydnXUJFKUR11iVAk0klr5bVUP7Rj34UrVq1iieffDI6d+4cjzzySLz//vsrvUN53LhxUVNT02CsWJxdfrUAQNOqkDU6AABIlLwKAECeJZJXy2ooP/LII/HAAw9Ex44d48QTT4x777031llnnfjv//7vuPXWW5e7X7FYjGKx+KlR010DAAAAAAAA5FlZDeXevXvHfffdF/vtt1888MAD9Q3lTTfddG3VBwCsTWnMyAIAQKWSVwEAyLNE8mpZDeUzzjgjhg4dGrfddlu0bds2dt1113j77bdjzJgxa6s+AGBtSmRKFgAAKpS8CgBAniWSV8tqKG+55Zbxj3/8I6ZMmRKvvPJKrL/++rHbbrtFnz591lZ9AMDalEjgAQCgQsmrAADkWSJ5tayGckREx44d4+CDD14btQAAAAAAAACQI2U3lAGAFiSRNT4AAKhQ8ioAAHmWSF7VUAaAhJUSmZIFAIDKJK8CAJBnqeRVDWUASFkaeQcAgEolrwIAkGeJ5NVEbsQGAAAAAAAAoFzuUAaAlFUlcgkdAACVSV4FACDPEsmrGsoAkLJE1vgAAKBCyasAAORZInlVQxkAUpZG3gEAoFLJqwAA5FkieVVDuRm9//L4TM670Ta3ZXLeiIgXn903k/POW/xiJuft1GajTM4LAFDJFs6emHUJyVjw4WtZl5CMTfrfn3UJSXn1r4dkXQIAAECLpaEMAClLZI0PAAAqlLwKAECeJZJXNZQBIGWJrPEBAECFklcBAMizRPKqhjIApCyNvAMAQKWSVwEAyLNE8qqGMgAAAADAWvaPf/wj/va3v8WCBQuW+d6oUaMyqAgAYNVoKANAyhJZ4wMAgAolr9ICLFy4MA455JB47rnnYsCAAdGqVcOPZAuFgoYyAFSqRPKqhjIApCyNvAMAQKWSV2kBJk6cGJ07d45//OMfUVVVVT/+8ssvx29/+9vYb7/9VnqMurq6qKurazBWLC6OYrFNk9cLAJQhkbxatfKHAAAtValQWONtdVxyySXRu3fvaNu2bQwcODCefPLJFT7+ggsuiM022yzatWsXvXr1iuOOOy4WLVq0WucGAKByyKu0BHfddVf89Kc/bdBMjojYe++9489//nPsu+++Kz1GbW1tdOrUqcFWW3v5WqoYAFhVqeRVDWUAoFndeOONUVNTE+PHj49nnnkmtt566xg2bFi88cYbjT7+uuuui5NOOinGjx8fL7zwQvzqV7+KG2+8MU4++eRmrhwAgBTIqzS1QqEQS5YsWWZ84cKFseWWW8Yrr7yy0mOMGzcu5s2b12AbN+6otVEuAJBzWeRVDWUASFlVYc23Mp1//vlxxBFHxGGHHRZbbrllXHbZZdG+ffuYNGlSo49/7LHHYscdd4yDDjooevfuHXvssUeMHDlypVfdAQDQAsirtADDhw+PSy65ZJnxa6+9NiZPnhy77777So9RLBajY8eODTbTXQNADiSSV8tqKL/77rsxY8aMRjcAoAIV1nyrq6uL+fPnN9g+vbbXxxYvXhxPP/10DB06tH6sqqoqhg4dGo8//nij++ywww7x9NNP1wecWbNmxZ133hlf+9rX1vz5AwCQb/IqLcDJJ58cTz31VHzzm9+MSZMmxe9+97u45JJLYsKECbHRRhvFL3/5y6xLBABWVyJ5tayG8i233BLDhg1bZttqq63KOQwAkBeFwhpvja/lVdvo6d56661YsmRJdO/evcF49+7dY+7cuY3uc9BBB8UZZ5wRO+20U7Ru3Tr69u0bQ4YMMYUgAEAK5FVagHbt2sXtt98eRxxxRPzrX/+K+++/P/7zn//ET3/607jrrrti3XXXzbpEAGB1JZJXW636KxLxne98J77zne8sM77JJpuUcxgAoAUZN25c1NTUNBgrFotNdvxp06bFWWedFb/85S9j4MCBMXPmzDjmmGPizDPPjNNOO63JzgMAQMskr5IXe+65Z+y5555ZlwEA5Ewl5NWyGsofe/XVV+PZZ5+N1q1bxw477BCFworn966rq1vm1uxicbF1PgAga6uxRsenFYvFVQ44Xbp0ierq6nj99dcbjL/++uvRo0ePRvc57bTT4pBDDqm/qG2rrbaKBQsWxJFHHhmnnHJKVFWVNeEKAACVRF4FACDPEsmrZSfak046Kbbbbrv45S9/Geecc0707t075syZs8J9Gr9V+/JyTw0ANLUmWOOjHG3atIkBAwbE1KlT68eWLl0aU6dOjUGDBjW6z/vvv79MqKmuro6IiFKpVF4BAABUFnkVAIA8SySvlnWH8kUXXRQzZsyImTNnRrt27SIi4s9//nNst912K9yv8Vu1Z5dzagBgbVjJLCNrQ01NTRx66KGx7bbbxvbbbx8XXHBBLFiwIA477LCIiBg1alT07Nmzfp2Q4cOHx/nnnx/bbLNN/ZQsp512WgwfPrw++AAA0ELJqwAA5FkiebWshvKNN94Y119/fX0zef78+dGrV6946aWXVrhf47dqm+4aAFI0YsSIePPNN+P000+PuXPnRv/+/ePuu++O7t27R0TE7NmzG1wxd+qpp0ahUIhTTz015syZE127do3hw4fHT37yk6yeAgAALZi8CgBAnmWRVwulMube2WabbeL++++P9ddfP6644oo44YQTolAoxNFHHx1nnnlmGU81ImJGmY9ndW20zW2ZnfvFZ/fN5LzzFr+YyXk7tdkok/MCLcmmzXq2jff7zRofY+at32qCSiCP5FVangUfvpZ1CcnYpP/9WZeQlFf/ekjWJSREXoX8kFdZe9bZsNzP+9P2n5dPzrqEilIIM3iwNsmra0NZayh/7Wtfq+9W/+pXv4rp06fH66+/HpMnT14rxQEAa1lVE2wAALC2yKsAAORZInm1rCmvTzvttDjwwANj4MCBMX/+/DjwwANj3rx5ccABB6yt+gCAtSmDNT4AAGCVyasAAORZInm1rIZy27Zt47bbbos//OEP8fTTT0ebNm1il112ia222mpt1QcAAAAAAABARspqKH9sl112iV122aWpawEAmlsaF9ABAFCp5FUAAPIskby6Wg1lAKBlKFUlkngAAKhI8ioAAHmWSl7VUAaAlCWyxgcAABVKXgUAIM8SyatVWRcAAAAAAAAAQD5pKANAygpNsEGOLF26NGpra2OzzTaLddZZJ/r27RunnHJKLFy4MOvSAIDVIa8CAJBnieRVDWUASFlVYc03yJETTzwxHnnkkZg0aVK8+uqr8dBDD0WxWIxvfetbWZcGAKwOeRUAgDxLJK9aQxkAUpbIGh+ko127dnHzzTdH+/btIyKiU6dOcfrpp0fv3r1XuF9dXV3U1dU1GCsWF0ex2GZtlQoArAp5FQCAPEskr2ooJ2DWs/tkXUKzG/CttzI57++v+CCT80ZEbN6pTybnrSr4MQIVLY28Q0LOOOOMBl+/8847cdttt0WvXr1WuF9tbW1MnDixwdj48WNjwoSjm7xGAKAM8ipAJha8fFrWJVSUdTc+P+sSKsq7M2uyLqGifLD0/axLqCitm3tu5kTyqk4QAAAtxuLFi+OUU06JW265JebMmRMdOnSIwYMHx29+85sV7jdu3LioqWn4H7TF4uy1WSoAAAAAVAQNZQBIWYWs0QGr6kc/+lG0atUqnnzyyejcuXM88sgj8f7776/0DuVisRjFYvFTo6a7BoDMyasAAORZInlVQxkAUpZI4CEdjzzySDzwwAPRsWPHOPHEE+Pee++NddZZJ/77v/87br311qzLAwDKJa8CAJBnieRVDWUASFgpjbxDQnr37h333Xdf7LfffvHAAw/UN5Q33XTTrEsDAFaDvAoAQJ6lklc1lAEAaDHOOOOMGDp0aNx2223Rtm3b2HXXXePtt9+OMWPGZF0aAAAAAFQkDWUASFkiU7KQji233DL+8Y9/xJQpU+KVV16J9ddfP3bbbbfo06dP1qUBAKtDXgUAIM8SyasaygCQskIagYe0dOzYMQ4++OCsywAAmoK8CgBAniWSVzWUASBliVxBBwBAhZJXAQDIs0TyatXKHvDPf/4znn/++eaoBQAAAAAAAIAcWWFDefz48fGVr3wl9tprrzj11FObqyYAoLlUNcEGAABri7wKAECeJZJXV1jmTTfdFJdeemlcfPHFcf/99zdXTQBAcykU1nwDAIC1RV4FACDPEsmrK1xD+cMPP4yzzjor2rRpE5dccklz1QQANJdE1vgAAKBCyasAAORZInl1hQ3lQqEQjz/+eHPVAgAAAAAAAECOrLCh3FTq6uqirq6uwVixuDiKxTbNcXoAYDlKFTKlCgAAaZJXAQDIs1TyarMs9VxbWxudOnVqsNXWXt4cpwYAVqSqCTYAAFhb5FUAAPIskby6wjuUS6VSk5xk3LhxUVNT02CsWJzdJMcGANZAImt8AABQoeRVAADyLJG8usKG8o033tgkJykWi1EsFj81arprAAAAAAAAgDxb4Y3UX/rSl5qrDgAgC4XCmm8AALC2yKsk4tFHH826BABgdSSSV1d4hzIA0MIlMiULAAAVSl4lEYcddljMmDEj6zIAgHIlklc1lAEgZWnkHQAAKpW8SiJKpVLWJQAAqyORvKqhDAAAAACQocJKprusq6uLurq6BmPF4uIoFtuszbIAACJiJWsoAwAtW6mqsMYbAACsLfIqfKS2tjY6derUYKutvTzrsgAgeankVXcoA0DKKiSwAACQKHkVIiJi3LhxUVNT02CsWJydUTUAQL1E8qqGMgCkbCXTqgEAQKbkVYiIiGKxGMVi8VOjprsGgMwlklc1lAEAAAAAmsFZZ50V999/f4OxUqkUc+bMyagiAICV01AGgJRVZV0AAACsgLxKCzN8+PAYMGBA1mUAAE0lkbyqoQwAKUtkShYAACqUvEoLs9VWW8VWW22VdRkAQFNJJK9qKANAyqrSCDwAAFQoeRUAgDxLJK9qKCegEGm8mT/pues+n8l5txr1v5mcNyLi4f/+Vybn7dGubybnjYioKvgRBgBQjnVabZB1Ccl4Z/7MrEsAAACAJqEbAwApS+QKOgAAKpS8CgBAniWSVzWUASBhpUTW+AAAoDLJqwAA5FkqeVVDGQBSVpV1AQAAsALyKgAAeZZIXk3kaQIAAAAAAABQLncoA0DKEpmSBQCACiWvAgCQZ4nkVQ1lAEhZVRqBBwCACiWvAgCQZ4nkVQ1lAEhZIoEHAIAKJa8CAJBnieRVaygDAAAAAAAA0Ch3KANAytK4gA4AgEolrwIAkGeJ5FUNZQBIWOn/tXfHYXKV5d3479klOREhCYGQQAhEIkEwwtYAMQrNljeYWgkFQQGxgShYxVBgAWFrYbNYXQpFsQJFchlEG4SrSLERG8EIKQLCK0ilvr6kgHT5gQkBkcUAs7o7vz94TV2ySZjN7Jwz+3w+Xue64MmcuZ9zmMx+3fuc5ySyJAsAAI1JXgUAoMhSyatDbij39/fHypUrY9y4cfGe97ynlnMCAOqllEbgAQCgQcmrAAAUWSJ5taqG8u233x7vfe97IyLiL//yL+ORRx6JV199NY444oi47LLLhmWCAAAAAAAAAOSjqoby4sWLY82aNRHxWnP5ySefjL6+vnjXu961xf3K5XKUy+UBY1nWG1k2usrpAgA1lciSLAAANCh5FQCAIkskr1bVUK5UKrH77rtHpVKJ5557LqZOnRp77713fPnLX97ifl1dXdHZ2TlgrKNjcSxZckb1MwYAaieNvAMAQKOSVwFoAHfctyDvKTCCve0vfp73FBrK48tb6lswkbxaVUO5VCrFj370o6hUKtHa2hqrVq2K+++/Pz7+8Y/HI488stn92tvbo62tbcBYlnUPbcYAQM00NeU9AwAA2Dx5FQCAIkslr1Z9h/Ipp5zy2o7bbRd777137LrrrnHBBRdscb8syyLLsteNWu4aAAAAAAAAoMiqaihfd911sWHDhoh47a7jM844I1avXh2LFy8elskBAMOrlMiSLAAANCZ5FQCAIkslr1Z1I/ahhx4a8+fPj/nz58e//Mu/xJ577hlf/OIX47zzzhuu+QEAw6hU2vZtKK666qqYNm1ajBkzJmbPnh0PPPDAFl//61//Oj71qU/FbrvtFlmWxYwZM+K73/3u0IoDANAw5FUAAIoslbxa1R3Kf2ivvfbSSAaABlfK4RK6m266Kdra2uKaa66J2bNnxxVXXBHz58+PRx99NHbddddNXt/b2xtHHHFE7LrrrnHzzTfHlClT4r//+79j/PjxdZ87AAD1Ja8CAFBkqeTVITeUAQCG4gtf+EKcdtppsWjRooiIuOaaa+K2226LZcuWxQUXXLDJ65ctWxa/+tWv4t57741Ro0ZFRMS0adPqOWUAABIirwIAUGR55NWqlrwGAEaWWizJUi6Xo6enZ8BWLpcHrdfb2xsPPvhgzJs3b+NYU1NTzJs3L+67775B9/nXf/3XmDNnTnzqU5+KSZMmxcyZM+Pzn/989PX1Dcs5AQCgOORVAACKLJW8qqEMAAmrReDp6uqKcePGDdi6uroGrffcc89FX19fTJo0acD4pEmTYu3atYPu88QTT8TNN98cfX198d3vfjcuvPDCuPzyy+Nv//Zva34+AAAoFnkVAIAiSyWvWvIaABJWqsGlZe3t7dHW1jZgLMuybX/j/6e/vz923XXXuPbaa6O5uTlmzZoVTz/9dFx22WXR0dFRszoAABSPvAoAQJGlklc1lAGAbZJl2RsOOLvssks0NzfHunXrBoyvW7cuJk+ePOg+u+22W4waNSqam5s3ju23336xdu3a6O3tjdGjRw998gAAjHjyKgAARdYIedWS1wCQsFosyVKN0aNHx6xZs2LVqlUbx/r7+2PVqlUxZ86cQfd5z3veE4899lj09/dvHFuzZk3stttufjkHADDCyasAABRZKnlVQxkAEtZU2vatWm1tbbF06dK4/vrr4+c//3l88pOfjA0bNsSiRYsiImLhwoXR3t6+8fWf/OQn41e/+lWceeaZsWbNmrjtttvi85//fHzqU5+q1WkAAKCg5FUAAIoslbxqyWsASFi1V8DVwvHHHx/r16+Piy66KNauXRstLS2xcuXKmDRpUkREdHd3R1PT/1zzNnXq1Pje974XZ599dhxwwAExZcqUOPPMM+P888+v/+QplJkzZ8bpp58ep59++ja/V7lcjnK5PGAsy3ojy9xVBAB5klcBACiyVPJqqVKpVGp+JG/ImnzKwjAq9/06t9on3Fne+ouGwYxxvbnUjYi46I/yuSbmzdvtlktdUjGjrtX2X/bv2/we/+ejf1yDmUD1vvWtb0VLS0tMnz59m99ryZIl0dnZOWCso2NxLFlyxja/N5CmN+3ZkfcUkvJKd+fWX0SNyKtQHH6/CkXxv9f/V95TaCgHT9wn7yk0lOknPZj3FBrK48tPrGu9VPKqO5QBIGF5XEEHtXLsscfW7L3a29ujra1twFiWddfs/QGAoZFXAQAoslTyqoYyACSslEriga3IsiyyLHvdqOWuASBv8ioAAEWWSl7VUAaAhJWatv4aAADIi7wKAECRpZJXEzlMAAAAAAAAAKrlDmUASFgiK7IAANCg5FUAAIoslbzqDmUASFiptO0bAAAMF3mVRjZz5sy4+uqra/Je5XI5enp6Bmzlcm9N3hsAGLpU8qqGMgAkLJXAAwBAY5JXaWSdnZ0xf/78mrxXV1dXjBs3bsDW1fWVmrw3ADB0qeRVS14DAAAAANTYscceW7P3am9vj7a2tgFjWdZds/cHANgSDWUASFhTg1wBBwBAmuRVeE2WZZFl2etGR+cyFwDgf6SSVzWUASBhjbKkCgAAaZJXAQAoslTyak2eoXzPPffU4m0AgDpL5RkfAAA0JnkVAIAiSyWv1qShvGjRolq8DQAAAAAAAAAFUpMlryuVyhb/vFwuR7lcHjCWZb2RZZ7zAQB5KqXykA8AABqSvAoAQJGlkldrcodyaSv3Y3d1dcW4ceMGbF1dX6lFaQBgG6SyJAsAAI1JXgUAoMhSyas1uUN5a9rb26OtrW3AWJZ116M0ALAFjRJYAABIk7wKAECRpZJX69JQzrIssix73ajlrgEgb6kEHgAAGpO8CgBAkaWSV2uy5DUAAAAAAAAAI09Vdyh//vOfj+9///sDxiqVSjz99NM1nRQAUB9NiVxBBwBAY5JXAQAoslTyalUN5QULFsSsWbOGay4AQJ2lsiQLAACNSV4FAKDIUsmrVTWU3/GOd8Q73vGO4ZoLAFBnJQ+/AACgwORVAACKLJW8mshhAgAAAAAAAFCtqu5QBgBGllSWZAEAoDHJqwAAFFkqeVVDGQASVkol8QAA0JDkVQAAiiyVvKqhDAAJSyTvAADQoORVAACKLJW86hnKAAAAAAAAAAzKHcoAkLBUrqADAKAxyasAABRZKnlVQxkAEpZK4AEAoDHJqwAAFFkqeVVDGWooax6fW+1v/a/f5VL37rVP5lI3ImLuzaNzqbviqEdzqbvrm96SS92IiOZSPuea4deUSOABgHr76p0n5z0FGBHkVQAawUET9857Coxgz957e95TaDAn1rVaKnnVM5QBAAAAAAAAGJQ7lAEgYalcQQcAQGOSVwEAKLJU8qqGMgAkrKlUyXsKAACwWfIqAABFlkpe1VAGgISlcgUdAACNSV4FAKDIUsmrnqEMAAAAAAAAwKDcoQwACXNlGQAARSavAgBQZKnkVQ1lAEhYKs/4AACgMcmrAAAUWSp5VUMZABKWyjM+AABoTPIqAABFlkperepO7B/96EfDNQ8AAAAAAAAACqaqO5QXLlwYt912W1Qqm96+PWPGjJpNCgCoj1Se8QEAQGOSVwEAKLJU8mrVS16/973v3WSsVCrFE088UZMJAQD1k8qSLAAANCZ5FQCAIkslr1bVUN5hhx3iBz/4QYwfPz5+97vfxQsvvBATJ04crrkBAMOsVNp01REAACgKeRUAgCJLJa9WdSf229/+9vjyl78c73vf+2LHHXeMfffdN6ZMmRI33njjcM0PAAAAAAAAgJxUdYfy0UcfHR/84AfjmGOOieeffz6233776O7ujve85z1xwgknbHa/crkc5XJ5wFiW9UaWjR7arAGAmkhlSRYAABqTvAoAQJGlklerukP5e9/7Xpx99tnx+OOPx/777x+HHHJIvPOd74xFixZtcb+urq4YN27cgK2r6yvbNHEAYNs11WADAIDhIq8CAFBkqeTVqu5Q/vGPfxz33HNPXH755fH0009Hd3d3TJ06NfbYY48t7tfe3h5tbW0DxrKsu/rZAgA11ZTIMz4AAGhM8ioAAEWWSl6tqvG9/fbbx1NPPRUREd3d3XHOOefEcccdF9/73ve2uF+WZTF27NgBm+WuAQAAAAAAAIqtqjuUP/ShD8XZZ58dt9xyS5x55plx8cUXR6lUik984hPxi1/8YrjmCAAMk1Se8QEAQGOSVwEAKLJU8mpVDeUzzjgjnnjiidhjjz3i1VdfjW9+85vR29sbBxxwwHDNDwAYRo3yjA4AANIkrwIAUGSp5NWqGsqlUimuuOKKWLx4cdx9993xzDPPxD777BPHHHPMcM0PABhGqVxBBwBAY5JXGUkOP/zw+MEPfpD3NACAGkolr1bVUP69t771rfHWt7611nMBAAAAABiRnnrqqTj88MM3GZ84cWJce+21MW7cuM3uWy6Xo1wuDxjLst7IstE1nycAwOsNqaEMAIwMTaVK3lMAAIDNklcZac4///xNxtavXx9HHXVU3HLLLbHzzjsPul9XV1d0dnYOGOvoWBxLlpwxLPMEAN6YVPKqhjIAJCyVJVkAAGhM8iojSalUivnz5w/6Z/vss08cddRRceutt8bEiRM3+fP29vZoa2sbMJZl3cMyTwDgjUslr2ooA0DCmvKeANTYr3/963j22WcH/bMZM2bUeTYAwLaSVxlJKpXKoEteH3XUUXHWWWfF1772tZgwYcKg+2ZZFlmWvW7UctcAkLdU8qqGMgAAI8bNN98cn/vc5zYZf+aZZzZ55hwAANTTddddFxs2bNhkfM8994yI1+5SBgAonA3PRAAAJHlJREFUIg1lAEhYKs/4IB2nnnpqnHrqqZuM++UcADQmeZWR5NBDD817CgBAjaWSVzWUASBhqTzjg/Q888wz8ZOf/CRGjRoV7373u6NU8mEHgEYkrwIAUGSp5FUNZQBIWCqBh7RccMEF8Y1vfCNaWlqiXC7Hww8/HK+88soW9ymXy5ssiZ1lvZFlnksHAHmSVwEAKLJU8qqGMowQTaV8/jrP3e2tudSNiPj349bnUveK/xyTS93/++LzudSNiPjynC03YobLjqP2zKVuRH5/p4Bt8w//8A+xZs2aeOyxx+JNb3pTREQ8/PDDcfDBB29xv66urujs7Bww1tGxOJYsOWPY5goAAAAAjaAp7wkAAPlpqsE2FFdddVVMmzYtxowZE7Nnz44HHnjgDe134403RqlUiqOPPnqIlRnpbrrpprjiiis2NpN7enpi6tSp8eSTT25xv/b29njxxRcHbO3tf1mHGQMAWyKvAgBQZKnkVQ1lAEhYU6myzVu1brrppmhra4uOjo546KGH4sADD4z58+fHs88+u8X9nnzyyTj33HPjsMMOG+rhkoCXX3453vzmN0dExLXXXhtTp06N6dOnxzXXXLPF/bIsi7Fjxw7YLHcNAPmTVwEAKLJU8qqGMgAkrKm07Vu1vvCFL8Rpp50WixYtiv333z+uueaa2H777WPZsmWb3aevry9OOumk6OzsjL333nsbjpiR7s/+7M/ic5/7XEREfPWrX41HH3001q1bF9ddd13OMwMAhkJeBQCgyFLJqxrKAMA2KZfL0dPTM2Arl8uDvra3tzcefPDBmDdv3saxpqammDdvXtx3332brXHxxRfHrrvuGh/72MdqPn9GlgsvvDCeeOKJmD17dvT09MQJJ5wQ73rXu+JDH/pQ3lMDAHIirwIAUGSNkFc1lAEgYbV4xkdXV1eMGzduwNbV1TVoveeeey76+vpi0qRJA8YnTZoUa9euHXSfH/7wh/HVr341li5duq2HSwLGjBkTt956a1x66aVx2mmnxXHHHRdf//rX4wtf+ELeUwMAhkBeBQCgyFLJq9sNeU8AoOENZUmV12tvb4+2trYBY1mWbfsbR8RLL70Uf/EXfxFLly6NXXbZpSbvSRrmzp0bc+fOzXsaAMA2klcBACiyVPKqhjIAJKxUqmzze2RZ9oYDzi677BLNzc2xbt26AePr1q2LyZMnb/L6xx9/PJ588slYsGDBxrH+/v6IiNhuu+3i0UcfjenTp2/D7AEAKDJ5FQCAIkslr1ryGgCom9GjR8esWbNi1apVG8f6+/tj1apVMWfOnE1e/7a3vS0eeeSRePjhhzduRx11VPzJn/xJPPzwwzF16tR6Th8AgBFOXgUAoMjyyqvuUAaAhNViSZZqtbW1xcknnxwHHXRQHHLIIXHFFVfEhg0bYtGiRRERsXDhwpgyZUp0dXXFmDFjYubMmQP2Hz9+fETEJuMAAIw88ioAAEWWSl7VUAaAhOWxVMnxxx8f69evj4suuijWrl0bLS0tsXLlypg0aVJERHR3d0dTk0VUAACQVwEAKLZU8mqpUqls++LeQ7Imn7LAiPHy79bnUveK/3w5l7r/98XRudSNiPjynFdyqbvjqD1zqRsR0VTK65qrGXWtduGD39/m9/jsrHk1mAkUkbwKDN0Njz+R9xSS8uHpe+c9hYTIq1Ac8ioURSX68p5CQylFc95TaCg7vqUr7yk0lJd+cV1d66WSV11OCQAAAAAAAMCgNttQnjlzZlx99dU1KVIul6Onp2fAVi731uS9AYChaypt+wYAAMNFXgUAoMhSyaubXc+zs7MzWlpaalKkq6srOjs7B4x1dCyOJUvOqMn7AwBD0yiBBQCANMmrADQCSzgznPa/7GN5T4EtSCWvbrahfOyxx9asSHt7e7S1tQ0Yy7Lumr0/ADA0/u8OAABFJq8CAFBkqeTVzTaUaynLssiy7HWjo+tRGgAAAAAAAIAhqktDGQAopqZSJe8pAADAZsmrAAAUWSp5VUMZABKWyjM+AABoTPIqAABFlkpe1VAGgISlEngAAGhM8ioAAEWWSl5tynsCAAAAAAAAABSTO5QBIGHNiVxBBwBAY5JXAQAoslTyqoYyACQslSVZAABoTPIqAABFlkpe1VAGgIQ1lSp5TwEAADZLXgUAoMhSyaueoQwAAAAAAADAoNyhDAAJS2VJFgAAGpO8CgBAkaWSVzWUASBhzXlPAAAAtkBeBQCgyFLJqxrKAJCwVK6gAwCgMcmrAAAUWSp5VUMZaFjbbzcxl7qfPuCVXOpe9fN1udSNiLjwwSyXun/19sdzqRsR8ZYd35JL3eY6B5CmUqW+BQEgEce9JZ+sCiONvAoAQJGlkleb8p4AAAAAAAAAAMXkDmUASFi974gGAIBqyKsAABRZKnlVQxkAEpbKMz4AAGhM8ioAAEWWSl7VUAaAhKUSeAAAaEzyKgAARZZKXvUMZQAAAACAYXb44YfnPQUAgCFxhzIAJCyVK+gAAGhM8iojyVNPPTVoU3nixIlx7bXXxrhx4za7b7lcjnK5PGAsy3ojy0bXfJ4AwBuXSl7VUAaAhDWXKnlPAQAANkteZaQ5//zzNxlbv359HHXUUXHLLbfEzjvvPOh+XV1d0dnZOWCso2NxLFlyxrDMEwB4Y1LJqxrKAJAwz74AAKDI5FVGklKpFPPnzx/0z/bZZ5846qij4tZbb42JEydu8uft7e3R1tY2YCzLuodlngDAG5dKXtVQBgAAAAAYZpVKZdAlr4866qg466yz4mtf+1pMmDBh0H2zLIssy143arlrAKA+NJQBIGGpPOMDAIDGJK8yklx33XWxYcOGTcb33HPPiHjtLmUAoLGkklerbijffvvt8dRTT8UOO+wQBx10UEyfPn045gUA1EEqgQcAgMYkrzKSHHrooXlPAQCosVTy6lYbyitWrIgFCxZERMRHP/rRWLNmTUyfPj1eeumlOOuss+KSSy6Jk08+edgnCgDUXnOpkvcUAABgs+RVAACKLJW8utWG8jnnnLOxoXzHHXdEd3d3lEqvtdsfe+yxWLhw4VYbyuVyOcrl8oCxLOuNLPOcDwAAAAAAAICiatraCyqVSjQ1NUVTU1M8/fTTMWrUqJgxY0b827/9W7z1rW+N5557bqtFurq6Yty4cQO2rq6v1OQAAIChaypt+wYAAMNFXgUAoMhSyatbvUO5VCrFq6++Gv39/TFz5sz46U9/Gvfff3+cfvrp8b73vS8qla3fyt3e3h5tbW0DxrKse+izBgBqolECCwAAaZJXAQAoslTy6lYbypVKJVpbWzf++/bbbx8tLS3xyiuvREREc3PzVotkWRZZlr1u1HLXAJC3VAIPAACNSV4FAKDIUsmrW20oL1++PF555ZWoVCpx3nnnxSmnnBL33ntvnH/++RERMWXKlFi6dGkcf/zxMXbs2GGfMAAAAAAAAAD1sdWG8iGHHLLxn1esWBErVqyI008/feP4smXLYvHixdHX1xef+MQnhm+mAEDNNSdyBR0AAI1JXgUAoMhSyatbbSj/ocmTJ8dpp502YGyvvfaKFStW1HRSAEB9NJUqeU8BAAA2S14FAKDIUsmrVTWUAYCRpSnvCQAAwBbIqwAAFFkqeTWV4wQAAAAAAACgSu5QBoCENSXyjA8AABqTvAoAQJGlklc1lAEgYc2JBB4AABqTvAoAQJGlklc1lAEgYU2lSt5TAACAzZJXAQAoslTyqmcoAwAAAAAAADAoDWUASFhTads3KIpjjz027ykAADUmrwIAUGSp5FVLXgNAwholsMAb8dOf/jSWLVu2yfh+++0Xc+bM2eK+5XI5yuXygLEs640sG13TOQIA1ZFXAQAoslTyqoYyQJW2a3pTLnX/6u1Tc6kbEfFi7y9yqXv5I/mc64iIt4///3Kpe8L0mXWtZ6kSRpo777xzk7Hm5uatNpS7urqis7NzwFhHx+JYsuSMms4PAKiOvAoApO6+YyfkPQW2IJW8qqEMAMCIUCqV4hvf+MaQ9m1vb4+2trYBY1nWXYtpAQAAAEBD01AGgISVElmShTRUKpX48Ic/vMn4EUccEYsWLdrivlmWRZZlrxu13DUA5E1eBQCgyFLJqxrKAJCwRPIOibj00kujp6dnk/F99903h9kAALUgrwIAUGSp5FUNZQBIWCpX0JGGY445Ju8pAAA1Jq8CAFBkqeTVVJ4VDQAAAAAAAECV3KEMAAlzZRkAAEUmrwIAUGSp5FUNZQBIWKlUyXsKAACwWfIqAABFlkpe1VAGgIQl8ogPAAAalLwKAECRpZJXU7kTGwAAAAAAAIAquUMZABJWSuUSOgAAGpK8CgBAkaWSVzWUASBhieQdAAAalLwKAECRpZJXq24o33777fHUU0/FDjvsEAcddFBMnz59OOYFANRBUyqJBwCAhiSvAgBQZKnk1a0+Q3nFihUb//mjH/1oXHzxxfHv//7vcdNNN8Whhx4a119//bBOEAAYea666qqYNm1ajBkzJmbPnh0PPPDAZl+7dOnSOOyww2KnnXaKnXbaKebNm7fF1wMAwLaSVwEAKLJ659WtNpTPOeecjf98xx13xN133x3XX3993HLLLXH33XfHV77yla0WKZfL0dPTM2Arl3urmigAUHulGmzVuummm6KtrS06OjrioYceigMPPDDmz58fzz777KCvv+uuu+LEE0+MO++8M+67776YOnVqvPe9742nn356CNUBAGgk8ioAAEWWSl4tVSqVypZesM8++8Tjjz++8d+bmppi7733ji996Uvxvve9L2bMmBFr1qzZYpElS5ZEZ2fngLGOjsWxZMkZb3iiAKmrRF9utV/s/UUudS9/ZHQudSMi3j7+d7nUPWH6n9a13v/59Xe2+T32H39kVa+fPXt2HHzwwXHllVdGRER/f39MnTo1zjjjjLjgggu2un9fX1/stNNOceWVV8bChQuHNGd4Y7accQG2pLf/pbynkJTRTTvmPYWEzKhrNXkVtkReBUhBfyWf31M2qqbS/nWtl0pe3eodyqVSKV599dV4+eWXY++9946enp74yle+Em1tbRERsZV+dEREtLe3x4svvjhga2//yzc0QQBg+NTiCrrBVyIpD1qvt7c3HnzwwZg3b97Gsaamppg3b17cd999b2jOL7/8cvz2t7+NCRMmDOWQAQBoIPIqAABFlkpe3WpDuVKpRGtraxx++OEREbH99ttHS0tLvPLKKxER0dzcvNUiWZbF2LFjB2xZlt9dZwBA7XR1dcW4ceMGbF1dXYO+9rnnnou+vr6YNGnSgPFJkybF2rVr31C9888/P3bfffcBoQkAADZHXgUAoMgaIa9ut7UXLF++PF555ZWoVCpx3nnnxSmnnBL33ntvnH/++RERMWXKlFi6dGkcf/zxMXbs2DdcGADI31Ce0fF67e3tG1cu+b0sy2rwzpu65JJL4sYbb4y77rorxowZMyw1AAAoDnkVAIAiSyWvbrWhfMghh2z85xUrVsSKFSvi9NNP3zi+bNmyWLx4cfT19cUnPvGJIUwdAMhLUw0ST5Zlbzjg7LLLLtHc3Bzr1q0bML5u3bqYPHnyFvf9+7//+7jkkkvi+9//fhxwwAFDni8AAI1DXgUAoMhSyatbXfL6D02ePDlOO+20AU3mvfbaK1asWKGZDAANqBbP+KjG6NGjY9asWbFq1aqNY/39/bFq1aqYM2fOZve79NJL47Of/WysXLkyDjrooCqrAgDQqORVAACKLJW8WlVDGQBgW7W1tcXSpUvj+uuvj5///OfxyU9+MjZs2BCLFi2KiIiFCxdGe3v7xtf/3d/9XVx44YWxbNmymDZtWqxduzbWrl0bv/nNb/I6BAAARjB5lVqZOXNmXH311TV5r3K5HD09PQO2crm3Ju8NADSWPPLqVpe8BgBGrlKpUveaxx9/fKxfvz4uuuiiWLt2bbS0tMTKlStj0qRJERHR3d0dTU3/c83bP/7jP0Zvb28cd9xxA96no6MjlixZUs+pAwBQZ/IqjayzszNaWlpq8l5dXV3R2dk5YKyjY3EsWXJGTd4fABiaVPJqqVKp1P9IIyJiTT5lARpUJfpyq/1i7y9yqXv5I6NzqRsR8fbxv8ul7gnT/7Su9R7vWbHN7zF97IIazASKSF4Fhq63/6W8p5CU0U075j2FhMyoazV5FV5TLpejXC4PGMuy7siy/P5/MwD10V/J5/eUjaqptH9d66WSV92hDAAJK1X7kA4AAKgjeRVek2VZZFn2ulHNZADIWyp51TOUAQAAAAAAABiUO5QBIGGuLAMAoMjkVQAAiiyVvKqhDAAJS2VJFgAAGpO8CgBAkaWSVzWUASBhieQdAAAalLwKAECRpZJXNZQBGkQpmnOrPX70W3Op+9lZuZSNiIiXf7c+v+IAQMMb3bRj3lMAAACAmtBQBoCEpbIkCwAAjUleBQCgyFLJqxrKAJCwRPIOAAANSl4FAKDIUsmrGsoAkLCmVBIPAAANSV4FAKDIUsmrTXlPAAAAAAAAAIBicocyACQskQvoAABoUPIqAABFlkpe1VAGgISVSpW8pwAAAJslrwIAUGSp5FUNZQBIWCpX0AEA0JjkVQAAiiyVvOoZygAAAAAAAAAMyh3KAJCwUiqX0AEA0JDkVQAAiiyVvKqhDAAJSyTvAADQoORVAACKLJW8qqEMAAnz7AsAAIpMXgUAoMhSyasaygCQsFSWZAEAoDHJqwAAFFkqeXWzjfOZM2fG1VdfXZMi5XI5enp6Bmzlcm9N3hsAAAAAAACA4bHZO5Q7OzujpaWlJkW6urqis7NzwFhHx+JYsuSMmrw/ADBUiVxCBwBAg5JXAYC0NZUsNlxsaeTVUqVSqQx3kXK5HOVyecBYlnVHlo0e7tIAMCQv/259LnW33+49da33Qvk72/weO2VH1mAmUERr8p4AABTQjLpWk1dhS+RVANiUvDoc6nJZQ5ZlkWXZ60Y1kwEgb6XSZp9+AQAAuZNXAQAoslTyahpHCQAAAAAAAEDVLLwOAElL4xkfAAA0KnkVAIAiSyOvaigDQMJKiQQeAAAak7wKAECRpZJXNZQBIGlpBB4AABqVvAoAQJGlkVc9QxkAAAAAAACAQblDGQASViq5tgwAgOKSVwEAKLJU8qqGMgAkLY0lWQAAaFTyKgAARZZGXtVQBoCElRIJPAAANCZ5FQCAIkslr6ZxHzYAAAAAAAAAVXOHMgAkLJUr6AAAaEzyKgAARZZKXtVQBoCkWawEAIAik1cBACiyNPKqhjIADGJM8/i8p1AXpVIaV9ABANCY5FUAAIoslbyaRtscAAAAAAAAgKq5QxkAkpbGFXQAADQqeRUAgCJLI69qKANAwkqJBB4AABqTvAoAQJGlklc1lAEgaZ5+AQBAkcmrAAAUWRp5NY2jBAAAAAAAAKBq7lAGgISlsiQLAACNSV4FAKDIUsmrGsoAkLBSKY3AAwBAY5JXAQAoslTyqoYyACQtjcADAECjklcBACiyNPKqZygDAAAAAAAAMCh3KANAwkquLQMAoMDkVQAAiiyVvKqhDABJS2NJFgAAGpW8CgBAkaWRVzWUASBhpVIagQcAgMYkrwIAUGSp5NU07sMGAAAAAAAAoGruUAaApKVxBR0AAI1KXgUAoMjSyKsaygCQsJLFSgAAKDB5FQCAIkslr2ooA0DS0riCDgCARiWvAgBQZGnk1TTa5gAAAAAAAABUzR3KAJCwUiJX0AEA0JjkVQAAiiyVvKqhDAAJK5XSCDwAADQmeRUAgCJLJa9qKANA0jz9AgCAIpNXAQAosjTyahpHCQAAAAAAAEDV3KEMAAlL5RkfAAA0JnkVAIAiSyWvaigDQNLSCDwAADQqeRUAgCJLI69a8hoAElYqlbZ5G4qrrroqpk2bFmPGjInZs2fHAw88sMXX//M//3O87W1vizFjxsQ73vGO+O53vzukugAANBZ5FQCAIkslr2ooAwB1ddNNN0VbW1t0dHTEQw89FAceeGDMnz8/nn322UFff++998aJJ54YH/vYx+InP/lJHH300XH00UfHf/7nf9Z55gAApEBeBQCgyPLIq6VKpVKp1QFUZ00+ZQHgDeiv/DaXuk2lt9e1XiUe3eb3KMW+Vb1+9uzZcfDBB8eVV14ZERH9/f0xderUOOOMM+KCCy7Y5PXHH398bNiwIb7zne9sHHvXu94VLS0tcc0112zb5GGL5FUA2NSMulaTV2FL5FUA2JS8+nq1yKvuUAaAhJVq8L9yuRw9PT0DtnK5PGi93t7eePDBB2PevHkbx5qammLevHlx3333DbrPfffdN+D1ERHz58/f7OsBABg55FUAAIosmbxaaTCvvvpqpaOjo/Lqq68mUzu1unnWdswjv26etR3zyK+bd+28dHR0VCJiwNbR0THoa59++ulKRFTuvffeAePnnXde5ZBDDhl0n1GjRlVuuOGGAWNXXXVVZdddd63J/GGkSPH7J0/Od/041/XjXNePc11f8irUj++36jhf1XG+quN8Vcf5qo7zVVuNkFdzXPJ6aHp6emLcuHHx4osvxtixY5OonVrdPGs75pFfN8/ajnnk1827dl7K5fImV8xlWRZZlm3y2meeeSamTJkS9957b8yZM2fj+Kc//elYvXp13H///ZvsM3r06Lj++uvjxBNP3Dh29dVXR2dnZ6xbt66GRwKNLcXvnzw53/XjXNePc10/znV9yatQP77fquN8Vcf5qo7zVR3nqzrOV201Ql7d7o0eDADAYDYXbgazyy67RHNz8yZBZd26dTF58uRB95k8eXJVrwcAgD8krwIAUGSNkFc9QxkAqJvRo0fHrFmzYtWqVRvH+vv7Y9WqVQOuqPtDc+bMGfD6iIg77rhjs68HAIChklcBACiyvPKqO5QBgLpqa2uLk08+OQ466KA45JBD4oorrogNGzbEokWLIiJi4cKFMWXKlOjq6oqIiDPPPDPmzp0bl19+ebz//e+PG2+8MX784x/Htddem+dhAAAwQsmrAAAUWR55teEaylmWRUdHxxu+9Xsk1E6tbp61HfPIr5tnbcc88uvmXbtRHH/88bF+/fq46KKLYu3atdHS0hIrV66MSZMmRUREd3d3NDX9zyIq7373u+OGG26Iv/mbv4m//uu/jn322SduvfXWmDlzZl6HAIXk+6e+nO/6ca7rx7muH+e62ORVGDrfb9VxvqrjfFXH+aqO81Ud5ytfeeTVUqVSqdT8SAAAAAAAAABoeJ6hDAAAAAAAAMCgNJQBAAAAAAAAGJSGMgAAAAAAAACD0lCuQmtra5x11ll5T4Nh4L8t9VaPz1yKn+sUjxkARjo/3wEAACBfGsqQOL+gYzj5fAEAAACwrfyOCSBfGsoAAAAAAAAQEb29vXlPoaE4X2loqIbyypUr49BDD43x48fHzjvvHEceeWQ8/vjjeU9rWOVxzNdee23svvvu0d/fP2D8z//8z+OjH/3osNaeNm1aXHHFFQPGWlpaYsmSJcNa9/Vuu+22GDduXCxfvnzYa+X5uT7llFNi9erV8aUvfSlKpVKUSqV48sknh71uXsfc2toaf/VXfxWf/vSnY8KECTF58uS6fLY2bNgQCxcujB122CF22223uPzyy4e95u/19/fX/Xh/L6/PV17HnOLPKKAY8vr5lirf94wk3/nOd2L8+PHR19cXEREPP/xwlEqluOCCCza+5tRTT42PfOQjeU1xRPI9AowkfpYMTX9/f3R1dcVb3vKWeNOb3hQHHnhg3HzzzXlPq7Dy+h1To/L5ql5ra2ssXrw4zjrrrNhll11i/vz5eU+p0JyvNDVUQ3nDhg3R1tYWP/7xj2PVqlXR1NQUxxxzzCaNz5Ekj2P+4Ac/GM8//3zceeedG8d+9atfxcqVK+Okk04atrpFccMNN8SJJ54Yy5cvr8vx5vm5/tKXvhRz5syJ0047LX75y1/GL3/5y5g6deqw183zmK+//vp485vfHPfff39ceumlcfHFF8cdd9wxrDXPO++8WL16dXz729+O22+/Pe6666546KGHhrXm7+VxvL+X1+crr2NO8WcUUBx5ft+nxvc9I8lhhx0WL730UvzkJz+JiIjVq1fHLrvsEnfdddfG16xevTpaW1vzmeAI5XsEGEn8LBmarq6u+PrXvx7XXHNN/OxnP4uzzz47PvKRj8Tq1avznloh5fU7pkbl8zU0119/fYwePTruueeeuOaaa/KeTuE5X+kpVSqVSt6TGKrnnnsuJk6cGI888kjMnDlz2Ou1trZGS0vLJnfQ1lO9jvnoo4+OnXfeOb761a9GxGt3LXd2dsZTTz0VTU3Ddx3CtGnT4qyzzhrwPIyWlpY4+uijh/VOm9//t91nn33iM5/5THz729+OuXPnDlu9LfG5Hr5jbm1tjb6+vrj77rs3jh1yyCFx+OGHxyWXXDIsNX/zm9/EzjvvHP/0T/8UH/zgByPitQs09thjj/j4xz8+rOc9j+MdbA71/HwV4Zh/r95/l4F0Fem7L0W+74dfEfLqSDZr1qw48cQT49xzz41jjjkmDj744Ojs7Iznn38+Xnzxxdhjjz1izZo1sc8+++Q91RHL9wjQ6PwsqU65XI4JEybE97///ZgzZ87G8VNPPTVefvnluOGGG3KcXXHJhG+Mz9fQtLa2Rk9PT91uAmp0zleaGuoO5f/6r/+KE088Mfbee+8YO3ZsTJs2LSIiuru7853YMMrrmE866aT41re+FeVyOSIili9fHieccMKwNpPzdvPNN8fZZ58dd9xxR12byT7X9T3mAw44YMC/77bbbvHss88OW73HH388ent7Y/bs2RvHJkyYEPvuu++w1fxD9T7eIsjrmFP8uwwUR4rf93nxfc9IM3fu3LjrrruiUqnE3XffHR/4wAdiv/32ix/+8IexevXq2H333TUAasz3CDDS+FlSncceeyxefvnlOOKII2KHHXbYuH3961/3CAS2mc/X0M2aNSvvKTQU5ys92+U9gWosWLAg9tprr1i6dOnGZ/zOnDlzRD/wO69jXrBgQVQqlbjtttvi4IMPjrvvvju++MUvDmvNiIimpqZ4/U3zv/3tb4e9bkTEH/3RH8VDDz0Uy5Yti4MOOihKpVJd6vpc1/eYR40aNeDfS6XSiF5aLrXjjcjvmFP8uwwUR4rf93nxfc9I09raGsuWLYv/+I//iFGjRsXb3va2aG1tjbvuuiteeOGF3FZuGsl8jwAjjZ8l1fnNb34TERG33XZbTJkyZcCfZVmWx5QYQXy+hu7Nb35z3lNoKM5Xehqmofz888/Ho48+GkuXLo3DDjssIiJ++MMf5jyr4ZXnMY8ZMyY+8IEPxPLly+Oxxx6LfffdN975zncOe92JEyfGL3/5y43/3tPTE7/4xS+GvW5ExPTp0+Pyyy+P1tbWaG5ujiuvvHLYaxbhcz169Ojo6+urW70iHHM9TZ8+PUaNGhX3339/7LnnnhER8cILL8SaNWuS+D9U9f585SW1zzVAqnzfMxL9/tmXX/ziFzfm09bW1rjkkkvihRdeiHPOOSfnGY4svkeAkcjPkursv//+kWVZdHd3J/G7oVpJ5XdM28rnCxguDdNQ3mmnnWLnnXeOa6+9Nnbbbbfo7u6OCy64IO9pDau8j/mkk06KI488Mn72s5/FRz7ykbrUPPzww+NrX/taLFiwIMaPHx8XXXRRNDc316V2RMSMGTPizjvvjNbW1thuu+2G/Zkcef83jnjtudX3339/PPnkk7HDDjvEhAkThnVp8yIccz3tsMMO8bGPfSzOO++82HnnnWPXXXeNz3zmMyN6+fg/VO/PV15S+1wDpMr3PSPRTjvtFAcccEAsX75840W1f/zHfxwf+tCH4re//a1fRNaY7xFgJPKzpDo77rhjnHvuuXH22WdHf39/HHroofHiiy/GPffcE2PHjo2TTz457ykWUiq/Y9pWPl/AcGmYb9ympqa48cYb48EHH4yZM2fG2WefHZdddlne0xpWeR/z4YcfHhMmTIhHH300PvzhD9elZnt7e8ydOzeOPPLIeP/73x9HH310TJ8+vS61f2/fffeNH/zgB/HNb35z2K+gzPu/cUTEueeeG83NzbH//vvHxIkTh/25XUU45nq77LLL4rDDDosFCxbEvHnz4tBDD03mGRP1/nzlJcXPNUCKfN8zUs2dOzf6+vqitbU1IiImTJgQ+++/f0yePDn23XfffCc3wvgeAUYqP0uq89nPfjYuvPDC6Orqiv322y/+9E//NG677bZ4y1vekvfUCiuV3zHVgs8XMBxKldc/sBYAAAAAAAAAooHuUAYAAAAAAACgvjSUAQAAAAAAABiUhjIAAAAAAAAAg9JQBgAAAAAAAGBQGsoAAAAAAAAADEpDGQAAAAAAAIBBaSgDAAAAAAAAMCgNZQAAAAAAAAAGpaEMAAAAAAAAwKA0lAEAAAAAAAAYlIYyAAAAAAAAAIPSUAYAAAAAAABgUP8/WiZmpKr5hq4AAAAASUVORK5CYII=\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(english.n_chars)"
      ],
      "metadata": {
        "id": "S5HkWmcmhWGN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "cc703d6a-bbed-4a95-c2a1-928a35dc1c6b"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "29\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "ssmZnrIOizHD"
      },
      "execution_count": null,
      "outputs": []
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "provenance": [],
      "collapsed_sections": [
        "63vUwSCMWDMn",
        "lpvi55EDGS4v",
        "zss0cVzw9eeT",
        "aUOXGR5iMdC5"
      ],
      "include_colab_link": true
    },
    "gpuClass": "standard",
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}