Classificação_de_Sequências.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "Classificação de Sequências.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "Adaptado por Bruno Menezes (https://github.com/brunoleomenezes) e será utilizado apenas para fins acadêmicos."
      ],
      "metadata": {
        "id": "qXAZolrdI-Zp"
      }
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gmai_ryJVhd5"
      },
      "source": [
        "# Classificando nomes com uma *Character-Level RNN*\n",
        "\n",
        "Esse notebook foi criado com base no tutorial do PyTorch: <br> \n",
        "https://github.com/spro/practical-pytorch/blob/master/char-rnn-classification/char-rnn-classification.ipynb"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "goO_pZBdVzrf"
      },
      "source": [
        "### Problema: Dado um nome próprio de entrada, classificar esse nome de acordo com a nacionalidade a que ele pertence.\n",
        "\n",
        "Entrada: **Hinton**\n",
        "\n",
        "(-0.47) Scottish\n",
        "\n",
        "(-1.52) English\n",
        "\n",
        "(-3.57) Irish\n",
        "\n",
        "\n",
        "-\n",
        "\n",
        "Entrada: **Schmidhuber**\n",
        "\n",
        "(-0.19) German\n",
        "\n",
        "(-2.48) Czech\n",
        "\n",
        "(-2.68) Dutch"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lliVTIikWKQf"
      },
      "source": [
        "### Import de bibliotecas"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Y2lbiDXXUNGM"
      },
      "source": [
        "# Para os dados\n",
        "import unicodedata\n",
        "import string\n",
        "import sys, random, os\n",
        "\n",
        "# Para o processamento e análise\n",
        "import torch\n",
        "from torch import nn\n",
        "import numpy as np\n",
        "import seaborn as sns\n",
        "sns.set_style('darkgrid')\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iIvPHo3Or62y"
      },
      "source": [
        "args = {\n",
        "    'lr': 5e-5,\n",
        "    'regularizacao': 1e-7,\n",
        "    'num_epocas': 40,\n",
        "}\n",
        "args['device'] = torch.device('cuda' if torch.cuda.is_available() else 'cpu')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mVssEJvAiaHp"
      },
      "source": [
        "## Dados de entrada"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XjtvooS3WsIB"
      },
      "source": [
        "### Importando dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Smrb3GKqWsVu",
        "outputId": "ae629a8f-4194-4dad-a5dd-efe989365933",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 598
        }
      },
      "source": [
        "# comente as duas linhas seguintes caso rode mais de uma vez\n",
        "!wget https://download.pytorch.org/tutorial/data.zip #\n",
        "!unzip data.zip #\n",
        "############################################################"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-10-17 22:08:22--  https://download.pytorch.org/tutorial/data.zip\n",
            "Resolving download.pytorch.org (download.pytorch.org)... 99.86.38.96, 99.86.38.37, 99.86.38.106, ...\n",
            "Connecting to download.pytorch.org (download.pytorch.org)|99.86.38.96|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 2882130 (2.7M) [application/zip]\n",
            "Saving to: ‘data.zip’\n",
            "\n",
            "data.zip            100%[===================>]   2.75M  --.-KB/s    in 0.05s   \n",
            "\n",
            "2020-10-17 22:08:22 (60.4 MB/s) - ‘data.zip’ saved [2882130/2882130]\n",
            "\n",
            "Archive:  data.zip\n",
            "   creating: data/\n",
            "  inflating: data/eng-fra.txt        \n",
            "   creating: data/names/\n",
            "  inflating: data/names/Arabic.txt   \n",
            "  inflating: data/names/Chinese.txt  \n",
            "  inflating: data/names/Czech.txt    \n",
            "  inflating: data/names/Dutch.txt    \n",
            "  inflating: data/names/English.txt  \n",
            "  inflating: data/names/French.txt   \n",
            "  inflating: data/names/German.txt   \n",
            "  inflating: data/names/Greek.txt    \n",
            "  inflating: data/names/Irish.txt    \n",
            "  inflating: data/names/Italian.txt  \n",
            "  inflating: data/names/Japanese.txt  \n",
            "  inflating: data/names/Korean.txt   \n",
            "  inflating: data/names/Polish.txt   \n",
            "  inflating: data/names/Portuguese.txt  \n",
            "  inflating: data/names/Russian.txt  \n",
            "  inflating: data/names/Scottish.txt  \n",
            "  inflating: data/names/Spanish.txt  \n",
            "  inflating: data/names/Vietnamese.txt  \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xBgek-v83EA6",
        "outputId": "00558d41-1840-4427-bc0e-31a60bf6efa5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 142
        }
      },
      "source": [
        "# Transforma um arquivo de nomes em listas e/ou arrays (nomes, rotulos)\n",
        "def readLines(filename):\n",
        "    lines     = open(filename).read().strip().split('\\n')\n",
        "    nomes     = [unicodedata.normalize('NFKD', line).encode('ascii', 'ignore') for line in lines]\n",
        "    categoria = filename.split('/')[-1].split('.')[0]\n",
        "    rotulos   = np.repeat( categoria, len(nomes) )\n",
        "\n",
        "    return nomes, rotulos \n",
        "\n",
        "\n",
        "root_path = 'data/names/'\n",
        "arquivos = sorted(os.listdir(root_path))\n",
        "categorias = [a[:-4] for a in arquivos]\n",
        "\n",
        "dados, rotulos = [], []\n",
        "samples_perclass = []\n",
        "\n",
        "for file_name in arquivos:\n",
        "  retorno = readLines(os.path.join(root_path,file_name))\n",
        "  dados.append(retorno[0])\n",
        "  rotulos.append(retorno[1])\n",
        "\n",
        "  samples_perclass.append( (file_name, len(retorno[0])) )\n",
        "\n",
        "\n",
        "print(samples_perclass, )\n",
        "print('Minimo amostras', min(samples_perclass, key= lambda k: k[1]), '\\n' )\n",
        "\n",
        "print(dados[categorias.index('Portuguese')][0:10])\n",
        "print(rotulos[categorias.index('Portuguese')][0:10])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[('Arabic.txt', 2000), ('Chinese.txt', 268), ('Czech.txt', 519), ('Dutch.txt', 297), ('English.txt', 3668), ('French.txt', 277), ('German.txt', 724), ('Greek.txt', 203), ('Irish.txt', 232), ('Italian.txt', 709), ('Japanese.txt', 991), ('Korean.txt', 94), ('Polish.txt', 139), ('Portuguese.txt', 74), ('Russian.txt', 9408), ('Scottish.txt', 100), ('Spanish.txt', 298), ('Vietnamese.txt', 73)]\n",
            "Minimo amostras ('Vietnamese.txt', 73) \n",
            "\n",
            "[b'Abreu', b'Albuquerque', b'Almeida', b'Alves', b'Araujo', b'Araullo', b'Barros', b'Basurto', b'Belo', b'Cabral']\n",
            "['Portuguese' 'Portuguese' 'Portuguese' 'Portuguese' 'Portuguese'\n",
            " 'Portuguese' 'Portuguese' 'Portuguese' 'Portuguese' 'Portuguese']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1HlMtAzxfneB"
      },
      "source": [
        "### Convertendo os dados para tensor"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WMoQMcoRkSZd"
      },
      "source": [
        "**Convertendo os rótulos para tensor**\n",
        "\n",
        "Representação One-Hot de 18 categorias de idiomas que queremos prever."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7HVlWioYjlUA",
        "outputId": "2cc57f26-6f12-4938-e663-27721cb207e8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "def label2tensor(rotulos):\n",
        "  rotulos_tns = torch.zeros( len(rotulos), 1, dtype=torch.int64 )\n",
        "  for k, rotulo in enumerate(rotulos):\n",
        "    idx = categorias.index(rotulo)\n",
        "    rotulos_tns[k][0] = idx\n",
        "  return rotulos_tns\n",
        "\n",
        "rotulos_arabe = rotulos[0]\n",
        "rotulos_tns = label2tensor(rotulos_arabe)\n",
        "print(type(rotulos_tns), rotulos_arabe[0], rotulos_tns[0])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "<class 'torch.Tensor'> Arabic tensor([0])\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9JCS8K8XkWNQ"
      },
      "source": [
        "**Convertendo os nomes para tensor**\n",
        "\n",
        "Aqui também usaremos a representação One-Hot, porém teremos que trabalhar com uma lista de tensores, pois os nomes tem comprimentos diferentes. Mais à frente no curso aprenderemos a lidar com isso da forma certa!"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ix07lvp9fmZi",
        "outputId": "bcd484cb-0d24-4f1e-a3b2-b349ecb179b9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 105
        }
      },
      "source": [
        "caracteres_validos = string.ascii_letters\n",
        "print(caracteres_validos)\n",
        "tam_dicionario = len(caracteres_validos)\n",
        "\n",
        "def nome2tensor(nome):\n",
        "  tns = torch.zeros( len(nome), tam_dicionario )\n",
        "  \n",
        "  for k, letra in enumerate(nome.decode('utf-8')):\n",
        "    idx = caracteres_validos.find(letra)\n",
        "    tns[k, idx] = 1\n",
        "\n",
        "  return tns\n",
        "\n",
        "dados_arabe = dados[0]\n",
        "dados_tns = [nome2tensor(dado) for dado in dados_arabe]\n",
        "print(dados_arabe[0].decode('utf-8')[0],'\\n', dados_tns[0][0])\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ\n",
            "K \n",
            " tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "        1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.])\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sU4QqWqgBqxc"
      },
      "source": [
        "**Amostrando batch balanceado**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GYfkmF8q5mT4",
        "outputId": "834b5ce1-d371-4abb-ded5-4eda4d810260",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "num_amostras = min(samples_perclass, key= lambda k: k[1])[1]\n",
        "\n",
        "def sample_batch(size=num_amostras):\n",
        "  dados_batch, rotulos_batch = [], []\n",
        "  for cat in categorias:\n",
        "    \n",
        "    amostras_cat = dados[categorias.index(cat)]\n",
        "    idx = np.random.choice(range(len(amostras_cat)), size=size)\n",
        "    \n",
        "    dados_batch.extend([ r for k, r in enumerate(dados[categorias.index(cat)]) if k in idx])\n",
        "    rotulos_batch.extend([ r for k, r in enumerate(rotulos[categorias.index(cat)]) if k in idx])\n",
        "\n",
        "  dados_tns = [nome2tensor(dado) for dado in dados_batch]\n",
        "  return dados_tns, label2tensor(rotulos_batch)\n",
        "\n",
        "dados_batch, rotulos_batch = sample_batch()\n",
        "print(len(dados_batch), dados_batch[0].size(), rotulos_batch.size())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1131 torch.Size([5, 52]) torch.Size([1131, 1])\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "joBV8pWElOO5"
      },
      "source": [
        "### Por fim, vamos separar dados de treino e teste."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AisQtZfilR9v",
        "outputId": "0440d116-a5a4-427f-9b08-a3013d8795f7",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "\n",
        "Xtrain, Xtest, Ytrain, Ytest = split(dados_tns, rotulos_tns, test_size=0.2)\n",
        "print(len(Xtrain), len(Xtest))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "909 228\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QnZL-mUXmMtA"
      },
      "source": [
        "## Modelo Recorrente\n",
        "\n",
        "* Implemente um modelo para classificação de nomes próprios (série de caracteres) usando apenas camadas *RNNCell*, *Linear* e ativação *LogSoftmax*\n",
        "* Cada entrada (caracter) possui dimensão (52): alfabeto maiúsculo e minúsculo\n",
        "* *Hidden size* possui dimensão (256): hiperparâmetro \n",
        "* Saída possui dimensão (18): vetor de probabilidade de classes\n",
        "* Batch size = 1 **pra não termos que lidar com as sequências de tamanho variável.**\n",
        "\n",
        "### Links úteis\n",
        "\n",
        "RNNCell: https://pytorch.org/docs/stable/generated/torch.nn.RNNCell.html#torch.nn.RNNCell\n",
        "\n",
        "Linear: https://pytorch.org/docs/stable/generated/torch.nn.Linear.html#torch.nn.Linear\n",
        "\n",
        "Non-linear activations: https://pytorch.org/docs/stable/generated/torch.nn.LogSoftmax.html#torch.nn.LogSoftmax"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PY9NvNN8kT-z",
        "outputId": "7b2e829c-6f56-4e53-8615-86460495451e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 105
        }
      },
      "source": [
        "class RNN(nn.Module):\n",
        "    def __init__(self, tam_entrada, tam_feature, tam_saida):\n",
        "        super(RNN, self).__init__()\n",
        "        \n",
        "        self.tam_entrada = tam_entrada\n",
        "        self.tam_feature = tam_feature\n",
        "        self.tam_saida   = tam_saida\n",
        "        \n",
        "        self.rnn    = nn.RNNCell(self.tam_entrada, self.tam_feature)\n",
        "        self.linear = nn.Linear(self.tam_feature, self.tam_saida)\n",
        "        self.softmax = nn.LogSoftmax()\n",
        "    \n",
        "    def forward(self, nome):\n",
        "      \n",
        "        # Inicialize o estado interno da RNN\n",
        "        hidden = torch.zeros(1, self.tam_feature).to(args['device'])\n",
        "        \n",
        "        for letra in nome:\n",
        "          letra = letra.unsqueeze(0)\n",
        "          hidden = self.rnn(letra, hidden)\n",
        "        \n",
        "        saida = self.linear(hidden)\n",
        "        saida = self.softmax(saida) \n",
        "        return saida\n",
        "\n",
        "tam_feature = 256\n",
        "model = RNN(tam_dicionario, tam_feature, len(categorias))\n",
        "model.to(args['device'])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "RNN(\n",
              "  (rnn): RNNCell(52, 256)\n",
              "  (linear): Linear(in_features=256, out_features=18, bias=True)\n",
              "  (softmax): LogSoftmax(dim=None)\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 42
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SVSrUbUns5WH"
      },
      "source": [
        "## Loss e Otimizador"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0UwVsprXr5BV"
      },
      "source": [
        "criterion = nn.NLLLoss().to(args['device']) \n",
        "optimizer = torch.optim.Adam(model.parameters(), lr=args['lr'], weight_decay=args['regularizacao'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mNex6BA1s4_W"
      },
      "source": [
        "## Treinamento\n",
        "\n",
        "A otimização é um processo que tem uma raíz muito bem definida de passo a passo, sempre fazemos:\n",
        "* Carregar os dados e colocar no dispositivo de hardware adequado\n",
        "* Forward do dado na rede\n",
        "* Cálculo da função de custo (no nosso caso uma função composta)\n",
        "* Passos de Otimização\n",
        "  * Zerar os gradientes do otimizador (`optimizer.zero_grad()`)\n",
        "  * Calcular os gradientes com base na loss (`loss.backward()`)\n",
        "  * Passo de otimização (`optimizer.step()`)\n",
        "\n",
        "Apesar de cada solução ter pequenas variações em um ou mais passos do fluxo, o esqueleto é sempre o mesmo. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N3AvGhzMscZT"
      },
      "source": [
        "def forward(X, Y, etapa):\n",
        "  if etapa == 'Treino': model.train()\n",
        "  else: model.eval()\n",
        "\n",
        "  acuracia = 0.\n",
        "  loss_epoca = []\n",
        "  for k, (dado, rotulo) in enumerate(zip(X, Y)):\n",
        "      \n",
        "    dado = dado.to(args['device'])\n",
        "    rotulo = rotulo.to(args['device'])\n",
        "\n",
        "    saida = model(dado)\n",
        "    loss = criterion(saida, rotulo)\n",
        "    loss_epoca.append(loss.detach().cpu().numpy())\n",
        "\n",
        "    _, pred = torch.max(saida, axis=-1)\n",
        "    acuracia += 1 if pred[0].item() == rotulo[0].item() else 0\n",
        "\n",
        "    if etapa == 'Treino':\n",
        "      # Otimização\n",
        "      optimizer.zero_grad()\n",
        "      loss.backward()\n",
        "      optimizer.step()\n",
        "\n",
        "  loss_epoca = np.asarray(loss_epoca).ravel()\n",
        "  acuracia   = acuracia/float(len(loss_epoca))\n",
        "  print('\\n','*'*15 + etapa + '*'*15 )\n",
        "  print('Epoca: {:}, Loss: {:.4f} +/- {:.4f}, Acurácia: {:.4f}'.format(epoca, loss_epoca.mean(), \n",
        "                                                                        loss_epoca.std(), \n",
        "                                                                        acuracia\n",
        "                                                                       )) \n",
        "  return loss_epoca.mean(), acuracia"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RaxYaVprzIS3",
        "outputId": "130a7851-70ad-483d-9f0e-673166b7bf45",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "loss_treino, loss_test = [], []\n",
        "acc_treino, acc_test = [], []\n",
        "\n",
        "dados_test, rotulos_test = sample_batch(size=5)\n",
        "for epoca in range(200):\n",
        "\n",
        "  dados_tns, rotulos_tns = sample_batch()\n",
        "\n",
        "  loss, acuracia = forward(dados_tns, rotulos_tns, 'Treino')\n",
        "  loss_treino.append(loss)\n",
        "  acc_treino.append(acuracia)  \n",
        "\n",
        "  loss, acuracia = forward(dados_test, rotulos_test, 'Teste')\n",
        "  loss_test.append(loss)\n",
        "  acc_test.append(acuracia)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:23: UserWarning: Implicit dimension choice for log_softmax has been deprecated. Change the call to include dim=X as an argument.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            " ***************Treino***************\n",
            "Epoca: 0, Loss: 2.6588 +/- 0.7154, Acurácia: 0.1538\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 0, Loss: 3.0990 +/- 0.6658, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 1, Loss: 2.5624 +/- 0.6972, Acurácia: 0.1411\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 1, Loss: 3.5574 +/- 1.1812, Acurácia: 0.0337\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 2, Loss: 2.5786 +/- 0.7952, Acurácia: 0.1326\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 2, Loss: 3.2838 +/- 0.9332, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 3, Loss: 2.5490 +/- 0.7683, Acurácia: 0.1566\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 3, Loss: 3.2704 +/- 0.9530, Acurácia: 0.0787\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 4, Loss: 2.5506 +/- 0.6784, Acurácia: 0.1467\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 4, Loss: 3.1761 +/- 0.8476, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 5, Loss: 2.4938 +/- 0.6715, Acurácia: 0.1538\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 5, Loss: 3.2236 +/- 0.9805, Acurácia: 0.0899\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 6, Loss: 2.4875 +/- 0.6751, Acurácia: 0.1527\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 6, Loss: 3.4099 +/- 1.1842, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 7, Loss: 2.4685 +/- 0.7033, Acurácia: 0.1715\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 7, Loss: 3.3505 +/- 1.1318, Acurácia: 0.0787\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 8, Loss: 2.4286 +/- 0.7183, Acurácia: 0.1771\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 8, Loss: 3.3212 +/- 1.1812, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 9, Loss: 2.3480 +/- 0.7648, Acurácia: 0.2135\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 9, Loss: 3.2992 +/- 1.1396, Acurácia: 0.0787\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 10, Loss: 2.2333 +/- 0.8165, Acurácia: 0.2628\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 10, Loss: 3.4025 +/- 1.2242, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 11, Loss: 2.1914 +/- 0.8388, Acurácia: 0.2652\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 11, Loss: 3.5825 +/- 1.4361, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 12, Loss: 2.0904 +/- 0.8072, Acurácia: 0.2811\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 12, Loss: 3.6550 +/- 1.6437, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 13, Loss: 1.9521 +/- 0.8541, Acurácia: 0.3392\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 13, Loss: 3.8424 +/- 1.8282, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 14, Loss: 1.9139 +/- 0.8581, Acurácia: 0.3396\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 14, Loss: 3.7393 +/- 1.8009, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 15, Loss: 1.8230 +/- 0.8520, Acurácia: 0.3586\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 15, Loss: 3.9523 +/- 2.0300, Acurácia: 0.0787\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 16, Loss: 1.8270 +/- 0.9337, Acurácia: 0.3588\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 16, Loss: 4.0092 +/- 2.1207, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 17, Loss: 1.7671 +/- 0.9247, Acurácia: 0.3815\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 17, Loss: 4.0532 +/- 2.1060, Acurácia: 0.0449\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 18, Loss: 1.7521 +/- 0.9118, Acurácia: 0.3851\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 18, Loss: 4.0227 +/- 2.1165, Acurácia: 0.0449\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 19, Loss: 1.6407 +/- 0.9134, Acurácia: 0.4249\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 19, Loss: 4.0732 +/- 2.2075, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 20, Loss: 1.7504 +/- 1.0137, Acurácia: 0.3979\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 20, Loss: 4.2476 +/- 2.2990, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 21, Loss: 1.6860 +/- 0.9903, Acurácia: 0.4174\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 21, Loss: 4.1449 +/- 2.1607, Acurácia: 0.0787\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 22, Loss: 1.6625 +/- 0.9977, Acurácia: 0.4262\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 22, Loss: 4.2496 +/- 2.2903, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 23, Loss: 1.7028 +/- 1.0531, Acurácia: 0.4241\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 23, Loss: 4.0308 +/- 2.1379, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 24, Loss: 1.6980 +/- 0.9930, Acurácia: 0.4222\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 24, Loss: 4.0906 +/- 2.2608, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 25, Loss: 1.7454 +/- 1.0326, Acurácia: 0.4046\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 25, Loss: 4.1551 +/- 2.3122, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 26, Loss: 1.6940 +/- 1.0290, Acurácia: 0.4287\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 26, Loss: 4.0935 +/- 2.3307, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 27, Loss: 1.7091 +/- 1.0584, Acurácia: 0.4149\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 27, Loss: 3.8285 +/- 2.2271, Acurácia: 0.0787\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 28, Loss: 1.6978 +/- 1.0123, Acurácia: 0.4126\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 28, Loss: 3.5983 +/- 2.1740, Acurácia: 0.0899\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 29, Loss: 1.6810 +/- 0.9963, Acurácia: 0.4236\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 29, Loss: 3.9901 +/- 2.3002, Acurácia: 0.0562\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 30, Loss: 1.7034 +/- 1.1028, Acurácia: 0.4242\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 30, Loss: 3.6140 +/- 2.1514, Acurácia: 0.1124\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 31, Loss: 1.7308 +/- 1.0862, Acurácia: 0.4292\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 31, Loss: 3.5416 +/- 1.7906, Acurácia: 0.0787\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 32, Loss: 1.6915 +/- 1.0885, Acurácia: 0.4197\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 32, Loss: 3.4912 +/- 2.0786, Acurácia: 0.1348\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 33, Loss: 1.7112 +/- 1.1079, Acurácia: 0.4461\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 33, Loss: 3.6068 +/- 1.8657, Acurácia: 0.1011\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 34, Loss: 1.7577 +/- 1.1316, Acurácia: 0.4440\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 34, Loss: 3.3826 +/- 1.8742, Acurácia: 0.0899\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 35, Loss: 1.7402 +/- 1.1589, Acurácia: 0.4508\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 35, Loss: 3.4952 +/- 1.8042, Acurácia: 0.0674\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 36, Loss: 1.7651 +/- 1.1779, Acurácia: 0.4439\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 36, Loss: 3.3793 +/- 1.8133, Acurácia: 0.1236\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 37, Loss: 1.7464 +/- 1.1789, Acurácia: 0.4462\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 37, Loss: 3.3131 +/- 1.7617, Acurácia: 0.1236\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 38, Loss: 1.8853 +/- 1.3399, Acurácia: 0.4436\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 38, Loss: 3.4640 +/- 1.7014, Acurácia: 0.1124\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 39, Loss: 1.8198 +/- 1.2021, Acurácia: 0.4149\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 39, Loss: 3.5022 +/- 1.7294, Acurácia: 0.0899\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 40, Loss: 1.9053 +/- 1.2794, Acurácia: 0.4355\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 40, Loss: 3.4485 +/- 1.7043, Acurácia: 0.1011\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 41, Loss: 1.9622 +/- 1.2464, Acurácia: 0.3859\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 41, Loss: 3.2727 +/- 1.5387, Acurácia: 0.0899\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 42, Loss: 1.8840 +/- 1.2130, Acurácia: 0.4194\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 42, Loss: 3.1905 +/- 1.5143, Acurácia: 0.1348\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 43, Loss: 1.8523 +/- 1.2085, Acurácia: 0.4208\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 43, Loss: 3.2952 +/- 1.5671, Acurácia: 0.1011\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 44, Loss: 1.9978 +/- 1.2678, Acurácia: 0.3945\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 44, Loss: 3.3431 +/- 1.5862, Acurácia: 0.0899\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 45, Loss: 1.9703 +/- 1.1854, Acurácia: 0.3810\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 45, Loss: 3.4104 +/- 1.5760, Acurácia: 0.0899\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 46, Loss: 2.0420 +/- 1.2120, Acurácia: 0.3615\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 46, Loss: 3.1751 +/- 1.4771, Acurácia: 0.1236\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 47, Loss: 1.9939 +/- 1.1858, Acurácia: 0.3707\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 47, Loss: 3.2053 +/- 1.4508, Acurácia: 0.1461\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 48, Loss: 2.0734 +/- 1.1963, Acurácia: 0.3573\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 48, Loss: 3.1959 +/- 1.5324, Acurácia: 0.1348\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 49, Loss: 2.1303 +/- 1.1434, Acurácia: 0.3413\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 49, Loss: 3.1759 +/- 1.4427, Acurácia: 0.1348\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 50, Loss: 2.0885 +/- 1.0899, Acurácia: 0.3313\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 50, Loss: 3.1209 +/- 1.3495, Acurácia: 0.1461\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 51, Loss: 2.0828 +/- 1.0618, Acurácia: 0.3510\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 51, Loss: 3.0394 +/- 1.3172, Acurácia: 0.1573\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 52, Loss: 2.0064 +/- 1.0259, Acurácia: 0.3687\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 52, Loss: 3.1028 +/- 1.3102, Acurácia: 0.1124\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 53, Loss: 2.0202 +/- 1.0121, Acurácia: 0.3632\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 53, Loss: 2.9630 +/- 1.2231, Acurácia: 0.1124\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 54, Loss: 1.9958 +/- 1.0279, Acurácia: 0.3719\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 54, Loss: 3.0088 +/- 1.2935, Acurácia: 0.1124\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 55, Loss: 1.9724 +/- 1.0669, Acurácia: 0.3912\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 55, Loss: 3.0537 +/- 1.2694, Acurácia: 0.1348\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 56, Loss: 1.9467 +/- 0.9893, Acurácia: 0.3710\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 56, Loss: 3.0325 +/- 1.1648, Acurácia: 0.1011\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 57, Loss: 1.9432 +/- 1.0354, Acurácia: 0.3970\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 57, Loss: 2.9779 +/- 1.1809, Acurácia: 0.1348\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 58, Loss: 1.9415 +/- 1.0148, Acurácia: 0.3862\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 58, Loss: 3.0052 +/- 1.2106, Acurácia: 0.1011\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 59, Loss: 1.9214 +/- 1.0637, Acurácia: 0.3969\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 59, Loss: 2.9283 +/- 1.2911, Acurácia: 0.1348\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 60, Loss: 1.9570 +/- 1.0555, Acurácia: 0.3804\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 60, Loss: 2.9669 +/- 1.3497, Acurácia: 0.1348\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 61, Loss: 1.9316 +/- 1.0502, Acurácia: 0.3828\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 61, Loss: 2.9648 +/- 1.2804, Acurácia: 0.1124\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 62, Loss: 1.9575 +/- 1.0537, Acurácia: 0.3727\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 62, Loss: 2.8554 +/- 1.2321, Acurácia: 0.1461\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 63, Loss: 1.9200 +/- 1.0554, Acurácia: 0.4025\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 63, Loss: 2.8697 +/- 1.1775, Acurácia: 0.1124\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 64, Loss: 1.9255 +/- 1.0454, Acurácia: 0.3788\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 64, Loss: 2.8201 +/- 1.1343, Acurácia: 0.1461\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 65, Loss: 1.9046 +/- 1.0098, Acurácia: 0.3848\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 65, Loss: 2.6751 +/- 1.0808, Acurácia: 0.1685\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 66, Loss: 1.9256 +/- 1.0647, Acurácia: 0.3926\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 66, Loss: 2.6702 +/- 1.1300, Acurácia: 0.1685\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 67, Loss: 1.9632 +/- 1.0424, Acurácia: 0.3648\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 67, Loss: 2.6274 +/- 1.1144, Acurácia: 0.1461\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 68, Loss: 1.9714 +/- 1.0549, Acurácia: 0.3813\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 68, Loss: 2.7636 +/- 1.2426, Acurácia: 0.1348\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 69, Loss: 1.9493 +/- 1.0298, Acurácia: 0.3757\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 69, Loss: 2.6294 +/- 1.1105, Acurácia: 0.1910\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 70, Loss: 2.0512 +/- 1.0642, Acurácia: 0.3626\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 70, Loss: 2.5820 +/- 1.1373, Acurácia: 0.1798\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 71, Loss: 2.0546 +/- 1.0752, Acurácia: 0.3717\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 71, Loss: 2.6035 +/- 1.1301, Acurácia: 0.2135\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 72, Loss: 2.0016 +/- 1.0107, Acurácia: 0.3722\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 72, Loss: 2.5786 +/- 1.1261, Acurácia: 0.2247\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 73, Loss: 2.0664 +/- 1.0174, Acurácia: 0.3531\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 73, Loss: 2.5058 +/- 1.0989, Acurácia: 0.2247\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 74, Loss: 2.0574 +/- 1.0423, Acurácia: 0.3610\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 74, Loss: 2.4262 +/- 1.0592, Acurácia: 0.2360\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 75, Loss: 2.0598 +/- 1.0114, Acurácia: 0.3393\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 75, Loss: 2.5032 +/- 1.1168, Acurácia: 0.2472\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 76, Loss: 2.0694 +/- 0.9981, Acurácia: 0.3814\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 76, Loss: 2.5771 +/- 1.2300, Acurácia: 0.2360\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 77, Loss: 2.0811 +/- 1.0063, Acurácia: 0.3545\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 77, Loss: 2.6892 +/- 1.2624, Acurácia: 0.2247\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 78, Loss: 2.0663 +/- 1.0000, Acurácia: 0.3451\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 78, Loss: 2.7077 +/- 1.3155, Acurácia: 0.2022\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 79, Loss: 2.0370 +/- 0.9762, Acurácia: 0.3660\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 79, Loss: 2.6230 +/- 1.2644, Acurácia: 0.2022\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 80, Loss: 2.0647 +/- 1.0399, Acurácia: 0.3703\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 80, Loss: 2.5242 +/- 1.1480, Acurácia: 0.2135\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 81, Loss: 1.9725 +/- 1.0260, Acurácia: 0.4048\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 81, Loss: 2.4903 +/- 1.0872, Acurácia: 0.2247\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 82, Loss: 1.9510 +/- 1.0080, Acurácia: 0.3934\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 82, Loss: 2.4945 +/- 1.1819, Acurácia: 0.2022\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 83, Loss: 1.9619 +/- 1.0761, Acurácia: 0.4021\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 83, Loss: 2.4142 +/- 1.0845, Acurácia: 0.2135\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 84, Loss: 1.9557 +/- 1.0941, Acurácia: 0.4057\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 84, Loss: 2.3156 +/- 1.0658, Acurácia: 0.2472\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 85, Loss: 1.9210 +/- 1.0533, Acurácia: 0.4067\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 85, Loss: 2.4911 +/- 1.1703, Acurácia: 0.2135\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 86, Loss: 1.9170 +/- 1.0816, Acurácia: 0.4066\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 86, Loss: 2.4945 +/- 1.1671, Acurácia: 0.2135\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 87, Loss: 1.9326 +/- 1.0271, Acurácia: 0.3940\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 87, Loss: 2.4370 +/- 1.1938, Acurácia: 0.2247\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 88, Loss: 1.9568 +/- 1.0580, Acurácia: 0.3920\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 88, Loss: 2.5179 +/- 1.2867, Acurácia: 0.2135\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 89, Loss: 1.9581 +/- 1.0558, Acurácia: 0.4023\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 89, Loss: 2.5562 +/- 1.3072, Acurácia: 0.2022\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 90, Loss: 1.9616 +/- 1.0700, Acurácia: 0.4096\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 90, Loss: 2.6409 +/- 1.3552, Acurácia: 0.1910\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 91, Loss: 1.9765 +/- 1.0840, Acurácia: 0.4062\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 91, Loss: 2.4099 +/- 1.1926, Acurácia: 0.2360\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 92, Loss: 1.9915 +/- 1.0719, Acurácia: 0.4011\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 92, Loss: 2.2929 +/- 1.1337, Acurácia: 0.2584\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 93, Loss: 2.0667 +/- 1.0265, Acurácia: 0.3566\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 93, Loss: 2.1165 +/- 0.9872, Acurácia: 0.2809\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 94, Loss: 1.9752 +/- 1.0220, Acurácia: 0.3882\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 94, Loss: 2.0768 +/- 1.0164, Acurácia: 0.3483\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 95, Loss: 1.9221 +/- 1.0426, Acurácia: 0.4078\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 95, Loss: 2.1757 +/- 1.1135, Acurácia: 0.3034\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 96, Loss: 1.9532 +/- 1.0487, Acurácia: 0.4107\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 96, Loss: 2.1872 +/- 1.0805, Acurácia: 0.3258\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 97, Loss: 1.9656 +/- 1.0949, Acurácia: 0.3929\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 97, Loss: 2.1208 +/- 1.0806, Acurácia: 0.3146\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 98, Loss: 1.9196 +/- 1.0713, Acurácia: 0.4156\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 98, Loss: 2.1932 +/- 1.1551, Acurácia: 0.2921\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 99, Loss: 1.9197 +/- 1.0788, Acurácia: 0.4236\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 99, Loss: 2.2992 +/- 1.2273, Acurácia: 0.2809\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 100, Loss: 1.8551 +/- 1.0603, Acurácia: 0.4504\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 100, Loss: 2.3380 +/- 1.2741, Acurácia: 0.2472\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 101, Loss: 1.8672 +/- 1.0716, Acurácia: 0.4279\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 101, Loss: 2.5722 +/- 1.4782, Acurácia: 0.2472\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 102, Loss: 1.8852 +/- 1.0913, Acurácia: 0.4384\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 102, Loss: 2.4064 +/- 1.3447, Acurácia: 0.2472\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 103, Loss: 1.8898 +/- 1.1053, Acurácia: 0.4374\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 103, Loss: 2.3193 +/- 1.2765, Acurácia: 0.2697\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 104, Loss: 1.8639 +/- 1.0739, Acurácia: 0.4269\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 104, Loss: 2.2107 +/- 1.1744, Acurácia: 0.2584\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 105, Loss: 1.8559 +/- 1.0193, Acurácia: 0.4061\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 105, Loss: 2.0212 +/- 1.1010, Acurácia: 0.3820\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 106, Loss: 1.8512 +/- 1.0705, Acurácia: 0.4303\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 106, Loss: 2.0697 +/- 1.1337, Acurácia: 0.3596\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 107, Loss: 1.8422 +/- 1.0792, Acurácia: 0.4332\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 107, Loss: 1.9947 +/- 1.0893, Acurácia: 0.3708\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 108, Loss: 1.8212 +/- 1.1069, Acurácia: 0.4475\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 108, Loss: 2.0039 +/- 1.1302, Acurácia: 0.3596\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 109, Loss: 1.8574 +/- 1.1137, Acurácia: 0.4301\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 109, Loss: 2.0319 +/- 1.1781, Acurácia: 0.3596\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 110, Loss: 1.7817 +/- 1.0836, Acurácia: 0.4405\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 110, Loss: 2.0589 +/- 1.2166, Acurácia: 0.3483\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 111, Loss: 1.7934 +/- 1.1120, Acurácia: 0.4609\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 111, Loss: 1.9791 +/- 1.1216, Acurácia: 0.3708\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 112, Loss: 1.7854 +/- 1.0983, Acurácia: 0.4564\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 112, Loss: 2.1134 +/- 1.2480, Acurácia: 0.3146\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 113, Loss: 1.7939 +/- 1.0971, Acurácia: 0.4535\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 113, Loss: 2.1542 +/- 1.2867, Acurácia: 0.3258\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 114, Loss: 1.8137 +/- 1.1018, Acurácia: 0.4417\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 114, Loss: 2.2528 +/- 1.3501, Acurácia: 0.3371\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 115, Loss: 1.8084 +/- 1.1122, Acurácia: 0.4485\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 115, Loss: 2.1707 +/- 1.2982, Acurácia: 0.3258\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 116, Loss: 1.7579 +/- 1.0843, Acurácia: 0.4641\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 116, Loss: 2.1165 +/- 1.2571, Acurácia: 0.3146\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 117, Loss: 1.8193 +/- 1.1012, Acurácia: 0.4517\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 117, Loss: 2.0205 +/- 1.1859, Acurácia: 0.3483\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 118, Loss: 1.7904 +/- 1.1105, Acurácia: 0.4580\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 118, Loss: 1.9687 +/- 1.1722, Acurácia: 0.3933\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 119, Loss: 1.7024 +/- 1.0550, Acurácia: 0.4853\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 119, Loss: 1.9089 +/- 1.1918, Acurácia: 0.3708\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 120, Loss: 1.8248 +/- 1.1079, Acurácia: 0.4442\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 120, Loss: 1.8843 +/- 1.1406, Acurácia: 0.4157\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 121, Loss: 1.7811 +/- 1.0609, Acurácia: 0.4465\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 121, Loss: 1.9228 +/- 1.1748, Acurácia: 0.4045\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 122, Loss: 1.7265 +/- 1.1024, Acurácia: 0.4969\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 122, Loss: 1.9131 +/- 1.1726, Acurácia: 0.4157\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 123, Loss: 1.7902 +/- 1.1305, Acurácia: 0.4551\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 123, Loss: 1.9308 +/- 1.2326, Acurácia: 0.3820\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 124, Loss: 1.7829 +/- 1.0931, Acurácia: 0.4429\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 124, Loss: 1.9109 +/- 1.2196, Acurácia: 0.4157\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 125, Loss: 1.7377 +/- 1.0869, Acurácia: 0.4609\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 125, Loss: 1.9719 +/- 1.2213, Acurácia: 0.3820\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 126, Loss: 1.7732 +/- 1.1093, Acurácia: 0.4584\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 126, Loss: 1.9443 +/- 1.1915, Acurácia: 0.4045\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 127, Loss: 1.7916 +/- 1.1084, Acurácia: 0.4508\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 127, Loss: 1.9156 +/- 1.1481, Acurácia: 0.3483\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 128, Loss: 1.7653 +/- 1.1238, Acurácia: 0.4711\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 128, Loss: 1.8004 +/- 1.0940, Acurácia: 0.3933\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 129, Loss: 1.7076 +/- 1.0911, Acurácia: 0.4817\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 129, Loss: 1.7925 +/- 1.1091, Acurácia: 0.3820\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 130, Loss: 1.7481 +/- 1.0878, Acurácia: 0.4605\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 130, Loss: 1.7277 +/- 1.1020, Acurácia: 0.4607\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 131, Loss: 1.7486 +/- 1.1326, Acurácia: 0.4844\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 131, Loss: 1.7518 +/- 1.0949, Acurácia: 0.4607\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 132, Loss: 1.6965 +/- 1.0855, Acurácia: 0.4825\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 132, Loss: 1.7705 +/- 1.1292, Acurácia: 0.4382\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 133, Loss: 1.7244 +/- 1.0894, Acurácia: 0.4630\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 133, Loss: 1.7243 +/- 1.1194, Acurácia: 0.4494\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 134, Loss: 1.7321 +/- 1.0849, Acurácia: 0.4630\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 134, Loss: 1.6993 +/- 1.0763, Acurácia: 0.4382\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 135, Loss: 1.6713 +/- 1.1131, Acurácia: 0.4951\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 135, Loss: 1.7006 +/- 1.1005, Acurácia: 0.4270\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 136, Loss: 1.6931 +/- 1.0902, Acurácia: 0.4794\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 136, Loss: 1.7180 +/- 1.1140, Acurácia: 0.4157\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 137, Loss: 1.6624 +/- 1.0856, Acurácia: 0.4690\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 137, Loss: 1.7047 +/- 1.1038, Acurácia: 0.4382\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 138, Loss: 1.7000 +/- 1.1330, Acurácia: 0.4628\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 138, Loss: 1.6739 +/- 1.0928, Acurácia: 0.4382\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 139, Loss: 1.6405 +/- 1.1044, Acurácia: 0.4913\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 139, Loss: 1.7301 +/- 1.1069, Acurácia: 0.4382\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 140, Loss: 1.6870 +/- 1.1108, Acurácia: 0.4892\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 140, Loss: 1.7134 +/- 1.1331, Acurácia: 0.4270\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 141, Loss: 1.6252 +/- 1.0628, Acurácia: 0.4991\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 141, Loss: 1.7246 +/- 1.1223, Acurácia: 0.4382\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 142, Loss: 1.6128 +/- 1.0946, Acurácia: 0.5035\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 142, Loss: 1.7286 +/- 1.1137, Acurácia: 0.4382\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 143, Loss: 1.6511 +/- 1.0945, Acurácia: 0.4943\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 143, Loss: 1.7751 +/- 1.1075, Acurácia: 0.3933\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 144, Loss: 1.6389 +/- 1.1111, Acurácia: 0.4916\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 144, Loss: 1.7521 +/- 1.1529, Acurácia: 0.4157\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 145, Loss: 1.6233 +/- 1.1374, Acurácia: 0.5158\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 145, Loss: 1.6832 +/- 1.0788, Acurácia: 0.4719\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 146, Loss: 1.5537 +/- 1.0332, Acurácia: 0.5062\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 146, Loss: 1.7535 +/- 1.1934, Acurácia: 0.4719\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 147, Loss: 1.5703 +/- 1.0391, Acurácia: 0.4978\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 147, Loss: 1.7437 +/- 1.1728, Acurácia: 0.4719\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 148, Loss: 1.5946 +/- 1.1053, Acurácia: 0.5230\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 148, Loss: 1.7447 +/- 1.1920, Acurácia: 0.4494\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 149, Loss: 1.5977 +/- 1.0663, Acurácia: 0.5190\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 149, Loss: 1.6880 +/- 1.1334, Acurácia: 0.4607\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 150, Loss: 1.5973 +/- 1.0797, Acurácia: 0.5101\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 150, Loss: 1.6814 +/- 1.1304, Acurácia: 0.4944\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 151, Loss: 1.5174 +/- 1.0482, Acurácia: 0.5372\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 151, Loss: 1.6406 +/- 1.1423, Acurácia: 0.4944\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 152, Loss: 1.5920 +/- 1.1267, Acurácia: 0.5300\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 152, Loss: 1.6453 +/- 1.0793, Acurácia: 0.5056\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 153, Loss: 1.5819 +/- 1.1389, Acurácia: 0.5185\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 153, Loss: 1.6043 +/- 1.0633, Acurácia: 0.4719\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 154, Loss: 1.5404 +/- 1.1056, Acurácia: 0.5241\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 154, Loss: 1.5752 +/- 1.0294, Acurácia: 0.4944\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 155, Loss: 1.5246 +/- 1.1207, Acurácia: 0.5389\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 155, Loss: 1.5275 +/- 1.0379, Acurácia: 0.5281\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 156, Loss: 1.5323 +/- 1.1324, Acurácia: 0.5404\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 156, Loss: 1.5279 +/- 1.0847, Acurácia: 0.5618\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 157, Loss: 1.4946 +/- 1.1062, Acurácia: 0.5432\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 157, Loss: 1.4938 +/- 1.0371, Acurácia: 0.5169\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 158, Loss: 1.5639 +/- 1.1371, Acurácia: 0.5114\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 158, Loss: 1.5085 +/- 1.1095, Acurácia: 0.4944\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 159, Loss: 1.5532 +/- 1.1362, Acurácia: 0.5182\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 159, Loss: 1.4816 +/- 1.0661, Acurácia: 0.5281\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 160, Loss: 1.5185 +/- 1.1515, Acurácia: 0.5378\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 160, Loss: 1.5953 +/- 1.1644, Acurácia: 0.5056\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 161, Loss: 1.5311 +/- 1.1194, Acurácia: 0.5374\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 161, Loss: 1.5283 +/- 1.1646, Acurácia: 0.5393\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 162, Loss: 1.4708 +/- 1.0810, Acurácia: 0.5370\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 162, Loss: 1.5533 +/- 1.1874, Acurácia: 0.5056\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 163, Loss: 1.4980 +/- 1.1489, Acurácia: 0.5526\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 163, Loss: 1.5478 +/- 1.1350, Acurácia: 0.4831\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 164, Loss: 1.4930 +/- 1.1256, Acurácia: 0.5450\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 164, Loss: 1.5041 +/- 1.0970, Acurácia: 0.4944\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 165, Loss: 1.5171 +/- 1.1430, Acurácia: 0.5271\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 165, Loss: 1.4694 +/- 1.0359, Acurácia: 0.5169\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 166, Loss: 1.4632 +/- 1.1057, Acurácia: 0.5578\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 166, Loss: 1.4557 +/- 1.0383, Acurácia: 0.4831\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 167, Loss: 1.4517 +/- 1.1124, Acurácia: 0.5603\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 167, Loss: 1.4311 +/- 1.0770, Acurácia: 0.5281\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 168, Loss: 1.4540 +/- 1.1501, Acurácia: 0.5590\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 168, Loss: 1.3917 +/- 1.0308, Acurácia: 0.5393\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 169, Loss: 1.4483 +/- 1.1698, Acurácia: 0.5673\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 169, Loss: 1.4158 +/- 1.0772, Acurácia: 0.5281\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 170, Loss: 1.4422 +/- 1.0903, Acurácia: 0.5636\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 170, Loss: 1.4544 +/- 1.1003, Acurácia: 0.5056\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 171, Loss: 1.4708 +/- 1.1413, Acurácia: 0.5433\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 171, Loss: 1.4068 +/- 1.0801, Acurácia: 0.5618\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 172, Loss: 1.4072 +/- 1.1169, Acurácia: 0.5747\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 172, Loss: 1.4610 +/- 1.0714, Acurácia: 0.5618\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 173, Loss: 1.4226 +/- 1.0768, Acurácia: 0.5685\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 173, Loss: 1.4069 +/- 1.0753, Acurácia: 0.5169\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 174, Loss: 1.3988 +/- 1.1268, Acurácia: 0.5911\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 174, Loss: 1.4438 +/- 1.0960, Acurácia: 0.5281\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 175, Loss: 1.4023 +/- 1.1098, Acurácia: 0.5712\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 175, Loss: 1.4355 +/- 1.1008, Acurácia: 0.5393\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 176, Loss: 1.4256 +/- 1.1246, Acurácia: 0.5582\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 176, Loss: 1.4462 +/- 1.1341, Acurácia: 0.5056\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 177, Loss: 1.4080 +/- 1.0995, Acurácia: 0.5698\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 177, Loss: 1.3921 +/- 1.1114, Acurácia: 0.5393\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 178, Loss: 1.4500 +/- 1.1249, Acurácia: 0.5539\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 178, Loss: 1.4203 +/- 1.0988, Acurácia: 0.5281\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 179, Loss: 1.3989 +/- 1.0581, Acurácia: 0.5684\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 179, Loss: 1.3916 +/- 1.0835, Acurácia: 0.5618\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 180, Loss: 1.4137 +/- 1.0911, Acurácia: 0.5504\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 180, Loss: 1.3878 +/- 1.0849, Acurácia: 0.5506\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 181, Loss: 1.4051 +/- 1.1126, Acurácia: 0.5774\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 181, Loss: 1.4010 +/- 1.0850, Acurácia: 0.5281\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 182, Loss: 1.4160 +/- 1.1487, Acurácia: 0.5707\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 182, Loss: 1.4311 +/- 1.1468, Acurácia: 0.5169\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 183, Loss: 1.3576 +/- 1.1062, Acurácia: 0.5825\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 183, Loss: 1.3778 +/- 1.1031, Acurácia: 0.5506\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 184, Loss: 1.3730 +/- 1.0985, Acurácia: 0.5708\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 184, Loss: 1.3772 +/- 1.0898, Acurácia: 0.5730\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 185, Loss: 1.4127 +/- 1.1238, Acurácia: 0.5689\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 185, Loss: 1.3313 +/- 1.0822, Acurácia: 0.5730\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 186, Loss: 1.3661 +/- 1.1112, Acurácia: 0.5665\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 186, Loss: 1.3705 +/- 1.1043, Acurácia: 0.5281\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 187, Loss: 1.3820 +/- 1.1079, Acurácia: 0.5702\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 187, Loss: 1.3392 +/- 1.0598, Acurácia: 0.5843\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 188, Loss: 1.3201 +/- 1.0887, Acurácia: 0.6028\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 188, Loss: 1.3103 +/- 1.0379, Acurácia: 0.5506\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 189, Loss: 1.3515 +/- 1.0983, Acurácia: 0.5727\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 189, Loss: 1.2992 +/- 1.0228, Acurácia: 0.5955\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 190, Loss: 1.3895 +/- 1.1191, Acurácia: 0.5792\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 190, Loss: 1.3031 +/- 1.0797, Acurácia: 0.5955\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 191, Loss: 1.3397 +/- 1.1128, Acurácia: 0.5935\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 191, Loss: 1.3073 +/- 1.0545, Acurácia: 0.5618\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 192, Loss: 1.2826 +/- 1.0593, Acurácia: 0.6110\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 192, Loss: 1.2760 +/- 1.0541, Acurácia: 0.6180\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 193, Loss: 1.3077 +/- 1.1290, Acurácia: 0.5941\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 193, Loss: 1.2918 +/- 1.0574, Acurácia: 0.5955\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 194, Loss: 1.2494 +/- 1.0829, Acurácia: 0.6210\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 194, Loss: 1.2546 +/- 1.0909, Acurácia: 0.5730\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 195, Loss: 1.3357 +/- 1.1607, Acurácia: 0.6004\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 195, Loss: 1.2907 +/- 1.0884, Acurácia: 0.5955\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 196, Loss: 1.3001 +/- 1.0569, Acurácia: 0.5942\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 196, Loss: 1.2676 +/- 1.0932, Acurácia: 0.6067\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 197, Loss: 1.2746 +/- 1.0578, Acurácia: 0.5973\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 197, Loss: 1.2598 +/- 1.0760, Acurácia: 0.6067\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 198, Loss: 1.3084 +/- 1.1038, Acurácia: 0.5935\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 198, Loss: 1.2667 +/- 1.1193, Acurácia: 0.6180\n",
            "\n",
            " ***************Treino***************\n",
            "Epoca: 199, Loss: 1.2387 +/- 1.0492, Acurácia: 0.6158\n",
            "\n",
            " ***************Teste***************\n",
            "Epoca: 199, Loss: 1.2656 +/- 1.1382, Acurácia: 0.6180\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "D-jhpXs_z7vb"
      },
      "source": [
        "## Análise de Convergência"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gM4BTf7lz7Q6",
        "outputId": "79b46b63-c135-4721-acab-d84c4a2e30a6",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 258
        }
      },
      "source": [
        "fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(12,3))\n",
        "\n",
        "ax1.plot(loss_treino[1:], label='Train')\n",
        "ax1.plot(loss_test[1:], label='Test')\n",
        "ax1.set_title('Model Convergence - Loss')\n",
        "ax1.set_xlabel('epochs')\n",
        "ax1.set_ylabel('Loss')\n",
        "ax1.legend()\n",
        "\n",
        "ax2.plot(acc_treino, label='Train')\n",
        "ax2.plot(acc_test, label='Test')\n",
        "ax2.set_title('Model Convergence - Accuracy')\n",
        "ax2.set_xlabel('epochs')\n",
        "ax2.set_ylabel('Accuracy')\n",
        "ax2.legend()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7fcc15fd0a90>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 45
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAtAAAADgCAYAAAAqslEYAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOzdd3gc1bnA4d/MbFNZadUsS7It2ZZ9XDEuYIwxHS6dEDqhkxCSUEISUrhJICSEQBICSUjovQUIBC49dGNwN8b1uMhWtYolraTVSlvn/jErWbYlS7JWxfJ5n8ePd2fOzHw7Go2+PXOKZpomiqIoiqIoiqL0jD7YASiKoiiKoijKgUQl0IqiKIqiKIrSCyqBVhRFURRFUZReUAm0oiiKoiiKovSCSqAVRVEURVEUpRdUAq0oiqIoiqIovaASaKVXhBAFQghTCGHrQdkrhRCfD0RciqIow5m69yrK0NLtL6Jy4BJCbAdygVwp5c4Oy1cBhwJjpZTbByU4Kw4HcCvwLaw4a4CPgDsGM67hSAhhAhOklFsGOxZFGe7UvVfpSAihAVuBVinllMGOR4kPVQM9/G0DLm57I4SYDiQOXji7eQU4C7gESAVmACuAEwYzqI56UtujKIrSCXXv7YNhdu89GhgBjBNCHDaQBx5m53FIUSd2+HsGuBz4W+z9FcDTwO/aCgghUmPrTwX8wCPA76WUUSGEAdwNXAk0An/uuPPYtvcCpwFR4AngNillZF9BCSFOBE4CJkopS2OLG4AHOpTJBR4EjgLqgLullI/E1t0OTAFagXOAEuAKKeVyIcTPgMOklOd12Nf9gCalvHFfMQshrgS+AyyNnbd/CiH+AjwJHANI4D3gWCnlUbF9T4qdv9lYNTm/klK+FFv3JNAMFGDdRNcDl0gpt8bWTwXui20bAu6XUv5eCKEDP43F4gE+BK6TUtbt67z2Vjc/+0LgMawasxDwoZTywlhtyr1YtVcuoBi4WEq5Np6xKcoBTt17UffemCuA14GE2OtlHc5PV3EYwM+Aa7CS703ANwAD68uZXUoZju3jE+BZKeWjXZzHJ7CurRmAGTuPP5BSemPbjwbuBxZgVay+APwIqASOkVKuiZUbAWwH8qWUNX04H8OCqoEe/hYDKUKIybFfyIuAZ/co8zesWohxWDeqy4GrYuu+A5wBzATmAOftse2TQBgojJU5Gfh2D+I6EVja4QbemReBMqxHjOcBvxdCHN9h/VmxMh7gDeDvHbY7TQjhBoh97guA53sY81ygCMgG7sT6w9IMjMS6+V3RVlAIkQT8N7bvEVjn9x9CiI6P6S4CfgOkAVti+yQW3wfAu7HPWIh1swa4AetmeUxsXT0d/sDF0b5+9r8F3o/FPYpdicDJWH+QJsa2vQCo7YfYFOVApu696t6LECIR6xw+F/t3UawJTXdx/AjrCcZpQApwNdaXrJ7Y8zxqwF2xY0wGRgO3x2IwgDexKkIKgDzgRSllEOvneWmH/V6MVZFy0CfPoGqgDxZtNSGfAhuA8rYVHW7sh0opm4AmIcSfgcuwah8vAO5ru9kKIe4Cjo29zsb65fZIKVuA5liNwbXAQ93ElAHs6Gpl7BvxfOB0KWUr8JUQ4tHY5/goVuxzKeXbsfLPAD8EkFIWCyFWYtWOPA0cD/illIt7GHOFlPJvsf2awLnANCmlH1gvhHiq7Rxg/YHbLqV8IvZ+lRDi38D5WDdugNeklEtj+3sOqwambdtKKWVbzVIrsCT2+jrgeillWWy724ESIcRlbbUOfdWDn30IyMdqx1kGtHVKCgFuYBLWH+IN8YhHUYYhde9V995vAgGsyggbYAdOB17rJo5vAz+VUsrY+9WxeNw9OGb7ecT6wrIl9g+gRghxL3Bb7P3hWIn1LR0+X9u9/ingZSHEz6WUJta1eU+PPvVBQCXQB4dngM+AsVg3tY4ysX6hizssK8b6FgrWL1bpHuva5Me23SGEaFum71G+K7VYNZhdyQXqYn9YOh57Tof3lR1e+wGXEMIWuwk8j/Vt+Wmsdn5tNSA9ibnj6yys35Ou1ucDc4UQ3g7LbFjnvKs4k2OvR2N1LOlMPvCaECLaYVkEq0ahvGNBIcQ7WI/eAL4rpXyui33uqbuf/U+xaqGXCiHqgT9LKR+XUn4khPg7Vq1MvhDiVeAnUsrGHh5XUQ4W6t6r7r1XAC/Fzk04luRfgZVA7yuOfa3rzm7XQezLS1sTDTfWea/vcJzizr4cSCmXCCH8wLFCiB1YNeRv7GdMw45KoA8CsVqBbVjf/q/ZY/VOdtU0ro8tG8OuG8UOrF8wOqxrU4r1zTpzP76ZfwDcJIQY1fZNfw8VQLoQwt3hRt4xru68DPxZCDEKqzZkXi9iNju8rsH6Bj8Kqw0a7H4+SoFPpZQn9TCujkqxaqC6Wne1lHJRdzuRUp66H8eGbn72UspKrMfICCGOAj4QQnwmpdwipfwr8NdYm7iXgFuAX+1nHIoyLKl778F9742dg+OBw4UQ58YWJ2J94cjsQRzjgT37ljR32E9bpcXIPcqYe7z/fWzZdCllnRDiG+xqdlMKjOnwBWhPT2E146gEXok9lVBQCfTB5BogTUrZLDr0yo113ngJuFMIcTmQjtX26k+xIi8BNwoh3sT6xf15h213CCHex7pZ/grwYdW0jJJSfrqvYKSUHwgh/ov1Tf86rMdTCVgd04JSyseFEF8AdwkhfoJVY3JNbH23pJQ1sY4VTwDb2poZ9Dbm2Pl5FbhdCPFtrD8kl2N1nAGr7dgfhBCXYbUXA6vTna8HTRveBO4VQvwQ+CfgAKZIKZdgdeC5UwhxReyPcBZwpJTy9Z58/i44hBCuDu9DWD/fTn/2QojzgS9jf2TrsW7AUWH1IteBlVjXRCtWhyBFUfam7r37EfMwufdehpX8H7fH8i+waumf3EccjwK/FUKsx2p+MR0oj53fcuBSIcRDWLXZ47uJw43VUbRBCJGHVeHRZinWl7U/CCFuw6ptn93hC8SzWNdIU+zzKDGqE+FBQkq5VUq5vIvVN2DdoIuw2j49DzweW/cIVo/d1VgJ06t7bHs51i/9eqwk6xUgp4dhnQe8DfwL65d7LdZjwg9i6y/G6tRQgfW46zYp5Qd776ZLz2N1mHl+j+W9jfl6rI4+lViPB1/AqkkhVkNzMlYtQkWszN2As7vgYtueBJwZ224zu26092M9KntfCNGE1SFpbnf77MY6oKXDv6vY98/+MGCJEMIXi+UmKWURVoeWR7DOXTHWI+E/9jE2RRmW1L23TzEf6PfeK4B/SCkrO/7DStKv6CaOe7G+RL2PVdP8GNYXHbCeDN6Cde+dipWQ78tvgFlYP+u36HAtSWvUljOxmmeUYHUevbDD+lKs688EFvb+FAxfmmnuWdOvKMq+CCHuBkZKKa/otrCiKIoSF+reOziEEI9jdUz85WDHMpSoJhyK0g1hjTXqANZg1cpeQ8+Gi1IURVH2k7r3Dj4hRAHWSCIzBzmUIUcl0IrSPTfWo8NcoAprQoO+tEVWFEVRuqfuvYNICPFb4GbgLinltsGOZ6hRTTgURVEURVEUpRdUJ0JFURRFURRF6QWVQCuKoiiKoihKLxxwbaCj0agZifS+2YlhaOzPdv1BxdI5FUvnVCydO1BjsduNnVizrB0U1D07vlQsQzcOULF0ZajEsj9xdHXPPuAS6EjExOv193o7jydxv7brDyqWzqlYOqdi6dyBGktWlru4+1LDh7pnx5eKZejGASqWrgyVWPYnjq7u2aoJh6IoiqIoiqL0gkqgFUVRFEVRFKUXVAKtKIqiKIqiKL1wwLWBVhTlwBWJhKmvryEcDvZ5X1VVGkNlHPvOYrHZHKSlZWEY6ja7p55cB0P959tT6jpQlOFJ/UbHkxnFtf55Wiecgxb2k7DuOfxzbgRNVfQrCkB9fQ0uVyJJSSPRNK1P+zIMnUgkGqfI+mbPWEzTpLm5kfr6GjIzcwYxsqGpJ9fBUP759pS6DhRl4CwqqiMcjXJMYeaAHE9ldnFkq/oK9yc/J2HDCySsfZqkpX/CqN042GEpypARDgdJSkrpc/I81GmaRlJSSlxq2ocjdR0oihIPju0f4n7v+2Ca3P9ZEfd8uGW3p0V6cxWel04j9en5mCufjOuxVQIdR7a6TQA4Sj7BUboQAKOxZDBDUpQhZ7gnTW0Ols+5vw6W83OwfE5FGQz27R/g2vIGwZrNbK/1U+0LUlzf0r4+YdU/se1cx6LWfF7ebo/rsVUTjjgy6jcDYC//EqJha5lKoBVlSGho8HLTTd8HoK6uFl3X8XjSAHjkkaew27u+uW7cuJ53332LH/7wlgGJdbAIIU4B7gcM4FEp5R86KXMBcDtgAqullJcMaJBxoK4FRRkeKipKKARqN36CySQAlhbXU5DmQvdXk7DuWZrGf4Or1p7L92cUxPXYKoGOI6N+M6amo0UCu5apBFpRhoTUVA9PPvk8AI899hAJCYlccsll7evD4TA2W+e3xEmTpjBp0pQBiXOwCCEM4AHgJKAMWCaEeENKub5DmQnAL4D5Usp6IcSIwYm2b9S1oCjDQ6hhBwDeTQuBSXgS7KzaXs13112IrWE7pqbzefblsLaZWaNS43pslUDHka1uM8H8E3CUfAq6jYg7D72xFAC9sQTP6xfTeOJ94DlmkCNVFAXgzjtvx+FwsGmT5JBDZnDCCSdz//1/JhgM4HS6uPXWXzNmTAErVy7nxRef5Z577uOxxx6iqqqSiopyqqqquOCCi7noogOuErYzhwNbpJRFAEKIF4GzgfUdynwHeEBKWQ8gpawe8Cj7SbyuhfPPv2iwP4qiDDv+YISLnlrOLccXsmB8BgBl3hayInWgQWHrarKTHcwbm06WfAabvh3/jGsJ5czh0+JUnLYWpox0xzUmlUDHS8iP0VRK6+QLMW0JYHOhBRowvNsASFz1EEZjMUnL/gKTVQKtKG+tq+KNtZX7vb2mwZ4ji501bSSnT83u1X5qaqp58MHHMQyD5mYfDzzwCDabjWXLlvDQQw9w551/3GubkpJi/vrXB/H7/Vxyybmcd975aJqx359liMgDSju8LwPm7lFmIoAQYhFWM4/bpZTv7munhqHh8STutqyqSsMwrC44b66t5PU1O/oW+R7Onp7DGdNG9qisrmvouoamadTUVPPII0+2XwsPPvgYNpuNpUuX8PDD/+Cuu/7UHrdh6Oi6RklJMQ888DB+fzMXXvhNzjvvfGy23ZuAaNre52B/GYYet3311VCJZajEASqWrvQ1lqKSenY0BlhTvpMzRzWA080rpRGuxkuT5iaXOo4f2cJx0zKZtul1ipMOYUnu9Wyva2FFWRWzx6SRlZEc13OiEug4sXm3AhBOn4D/sB8CkPT5HThKP0NrqcW18V9EXWk4Sj8jVPk1uAoHM1xFUWKOO+5EDMNKfn0+H7/73e2UlZWgaRrhcLjTbebNm4/D4cDhcJCWlkZdXR0ZGVkDF/TgsQETgGOBUcBnQojpUkpvVxtEIiZer3+3ZaZptg8LF42ae30Rgs6/IPVUNGr2eNi5aNSMxWBy3HEnAhqRSJSGhkbuu+/XsWtBJxwOEYlE2/cbiUSJRk3mzZuPYdhwu1NJS0ujpmYnI0bs/iXONPc+B/vL40mM2776aqjEMlTiABVLV/oay+rtdQAcK+/AvuZTAIrdf8CphSnPOwl32asssG1krq+BVK2Oy+u/w2evrW3f/oQJGXi9/v2KIyur85prlUDHiREbgSOSNqF9WSRlNFq4haSlf0YLt+I9+1+kvvEtjGUPwYK9a7UU5WBy+tTsXtcWdxSvcYJdLlf760cffZBZs+Zw111/YseOCm644budbmO3O9pf67pOJBLpcxxDQDkwusP7UbFlHZUBS6SUIWCbEGITVkK9bH8P2tV1MBjjQHd1LVRXV/L973+n022G6bWgKH327PIyXvt6By9dOafP+yqqtZLe7MB2IimjMJrKKKhfBDbwTDqe5h0fcoSxEUf5NqKOFKbOOp2TPclkJjt4ZlkZp0yOf3cNlUDHia36a0zdRiS1oH1ZNGUMAK51zxLIP4HwyNmEc+Zgq94wSFEqirIvPp+PrCyrJvntt/9vkKMZcMuACUKIsViJ80XAno27/wNcDDwhhMjEatJRNKBRDpCO18Jbb70xyNEoyoHn4807KalvYWO1j/npSfu/o5CfyI6vgQxyqMGbfSaeplc4Ul8HgO4eiS1/Hq6aZZi6jVDOYVw9b1z75rNHe/r4STrX7+NACyEMIcQqIcSbnaxzCiH+JYTYIoRYIoQo6O944sle/DFpL5yAY9t/SVj/PIGxp4CxqzYikmJV5mhmlJaZ11nLknPQmuLb3k9RlPj41rcu58EHH+Cqqy456GoSpZRh4HrgPWAD8JKUcp0Q4g4hxFmxYu8BtUKI9cDHwC1SytrBibh/HczXgqL0VWsowvrKJgAWx5pf9GSb77z4FR9v3rnb8oS1T/P7+puZk9pMquanTMuhmBym6VYfs2jiCEK5R2A0lmDzFhHKPSK+H6YLA1EDfRPWzTilk3XXAPVSykIhxEXA3cCFAxBTn2mBRtwf/wSjuYrUt6/C1B00z/vFbmUibiuBDo2Y0f4DjSaNhOZqiITAiO+g3oqi9Mw113TeNGPatEN48cVX299fe601VvCsWXOYNWtOp9s+88xLQ2ra6b6QUr4NvL3Hsl93eG0CP4r9GxZ6ci0Yhs63v/09oPtrQVEUWLujiXDUxG5oLN5ev8+y72+sJsFuUNkU4KvyRur82zh6fAaGbk1CFK7bTjJhrsraBGXwRomdWZFcxhllAEQSswnmzmvfXyhvYBLofq2BFkKMAk4HHu2iyNnAU7HXrwAnCCEOiGmbkhb/Ad1fQ+NJfyeSnId/zg1EU/N3L2RPwDfvf/Ed83urRwwQTc5Bw0T3D5vRnxRFURRFUZBVPi54cjkvfVWBrsHZ00aypqKRptZQp+VN0+TeT4r46RvrefTLYtxOG8neDdS+ewcJK/8BkQCBBmu0ppmhFQAsb3TTkmINxBC1J4MjiUjGJKLOVExbIuGs6QPyWfu7Bvo+4KdAV4PvtQ+bJKUMCyEagAxgZxflOx0SqSfiOpxL/XZs658nOutqEg6/hOhhF+HQdBydlT3+xzg7vNVGWEl2qlaP6ZmAtvENtJIviJ6814RfA2I4DXMTTyqWzvU1lo7Dl8UrnqGis1jiOXyZoijKUBQ1TfRYJeGS4nq21frZVutn0ohkThRZvLJ6B59t3km+28GLK8u58ehx+EMRvP4QiQ6D2uYgugZ1/hB3nzmZcR/+jqnbvoZtUJtYiOatACC7dgkAjvR8jj7cAx8/TzQp1jlQNwhM/Ib1dF8fmO59/XYUIcQZQLWUcoUQ4th47bezIZF6Ip7DuSR/fA82zcA7/TqivdynoaWTDjRXbieYPA33mjdwFr1D3eG/7nbb/jCchrmJJxVL5/oaS8fhy/pqKDWb6CqWzoYv62pIJEVRlAPNI18W8+LKcm4+dhynT8lmy85mUlw2NGDe2DQOzUslJ8XJC8tKcdt1Pti0k1mjPby3oZplJV5uPckauez2UwVNrWGOHZdK5keb+dJxJPOCX/Dwu5/zPVs1aGCEmzENJ3+97Dhs9RsBq/1zG9/Rdw7oZ+/PNH0+cJYQ4jTABaQIIZ6VUl7aoUzbsEllQggbkAoM6Q4pmr8Gl3yF1qmXWu2Ze6ltG6PZeiShBbxo4RYIt4AtIa6xKoqiKIqi7MlWuQJ71SpaZny719sGw1HWVzYRiER57Mtikpw2fvPuJsbv/IiLS1/j/IRUxl90P05XIoaucd6MXP62cBtt7XNfXlWOrWIJF7GF51ZcjKHBcYWZuOwGtsoVGJEWpp5wGdEPlnFGXoCcmgaIjQkfcY/CZuhEPOMwNZ1o0v4PhdpX/ZZASyl/AfwCIFYD/ZM9kmeAN4ArgC+B84CPYp1Uhixb3Sa0aJjAuFP3a3vT6cG0JaD7rJE49Nb69v+jySqBVhRFURSlfyV8/TjOrW/TcsjVoPW8KdznRbX89r1N1PmtNs1ZyQ5euHw2Fz21gpxNTzE+tB49bFLfsJ5w4mEAnDV9JA9/WQzAgnEZfLCphn/Y3+U0YylnVk5hXOZ0XHZrMit7xWIAgnnzMJNzme0sxTBDRF3p6K11RN2jrEBsLlqmX0Uo9/B4nZJeG/BxoIUQdwDLpZRvAI8BzwghtgB1WOOODmlGo3URRFLyuynZBU0Ddw56Ww10q3fX/8m5cYlRURRFURSlK1ZlYAi9uYpock6X5bbV+klyGIxwW725nl5aitOmc+fpk9jRGGDOGA+pCXbm5ntwbK1hvZnPNG07RlMZ4RwrgfYk2PnFqZMItoYYl5HIB5tqmGbfAVG4wfYab2Uf2X48e/liwmkTMBMzrdrmqlUABEfNx7Xl/4i0JdBA84Lf9Mep6bEBSaCllJ8An8RedxwSqRU4fyBiiBejoQRTt+3zguuO6c7B2LMGuqUONdKoovSfhgYvN91kDUtXV1eLrut4PGkAPPLIU9jt+x5WcuXK5djtdqZPn9HvsSr9S10LykEtGsbwWvMf6U1lXeYz4ajJ917+mgS7zrOXzSLRbrB5ZzP/M2kEJ0/afWa/w8d4yN5az8LodKbp2zEay3Zb/63Dx+D1+omaJqcJD6NKdtBqS+VkVlDvrgImQjSCfccyAhPPASCSMgpH+SIAQrEEOtohgR5saibCXtIbS6xvQLqx/ztJyUUvWQJmFC3QYO23dd/jJCqK0jepqR6efPJ5AB577CESEhK55JLLerz9qlUrSEhIVEnTMKCuBeVgZjSWoEUC1usONcV7WlZST21zEIA/friF6+YX4AtEmJC196yCR+TYcGkhSskm4spAbyrbqwyArmn87kgXenGY5tk34FryO47TVwEL0JtK0UM+wiOs36uOyXIo70ia59xE68Sz+/LR40ol0D3klK8SzpqG0VhMdH+bb8SY7hx0XyVaoAEt1jJeC6gEWlEG2saNG/j73/+C3+/H4/Fw6623k5mZycsvv8jrr/8bwzAoKBjLddfdwOuvv4qu67z//jvcfPMtzJgxc7DDV+JIXQvKwcKo29z++tkPFzPDcxITRyTvVe7dDdUkOw3OmjaS51eUI7KtEYQKM/dOoLOwcphoYjbRlFEYvs4TaOv4mwDQ8+cT3lRIas0yGgFbLK5w+kSA3ZprRJJG4p97Sy8/af9SCXRPRIK4P7yZ4LhTMBpLCIw/pG/7c+eiRYMY9VvbF+ktPZvqUlGGC+fGV3BteHG/t9c0DdPcvc9x6+SLCEw6r4d7MLnvvj9y111/Ji0tjQ8/fJ+HH36AW2+9jWeffZKXX34Dh8NBU1MTbrebs8/+Zq9rKpXudXUddPbz7aneXQfQ3bXw6qtvYhg2dS0oB5TKxlaWF9Vy1LiM9mVR0+TthQu5GPCZLjzBHSwprmfiiGRM0+TxJSVML3mKiYG1fFJ/EyeJLC6cmcfzK8p5ZlkpAOMzkyAawfPaNwnmHYn/iJ+1Tw53ymHTiFZuwqi1hpnTG4rx/Od8zHMfx97QiPuDmwmOORoTjbCnkFDuETg3/QeiEYx6K7GOpMUmSYkl0FF7MtiH3nj6KoHuAaOxBM2MYC/5BD3UTCRlTJ/2Z6bkAWDbua59maaacCjKgAoGgxQVbeXmm38AQDQaISMjE4Dx4ydwxx2/ZMGCY1mw4NhBjFIZCN1dC7ff/r8cddQx6lpQDigPflrEc0tLuOuMyZwosgBYWdqAo2ELXmcW9pRsxtbXsajWGqu+pL6FBxcV85JzIePYSCT0Xc6aPpLcVBcTspLYXNNMboqTZKcN56b/YK9cga1mLa3Tr0RvrgJg7JhxRJpH4Sj+EEyTxBV/xfBVECn+HLuvCcNXTsL6F6w8yp5AKPcIEtY9i23nOmx1m4kkjcR0pgC7aqDbJ0sZYlQC3QNtNcV6qBmASGofE+jY9raaNe3LVBto5WATmHReL2sJdxePiVTGjh3HQw89sdfyP/7xPlavXsWiRZ/x9NOP89RT+19TruxbV9fBQE+Us69rYc2ar1i48FN1LSgHlDXlVh+r3763iYkjkhmTlsCb66v4nl6Bc+RkdGcyYxrWsLWmGUyTlVusUcZmuioxAiafXjoSI9tKZheMz2BzTTOFWclgRklc/lci7tHovgoSVj3YPh5zNCmbiDsPLdyKrforXPLfAGg7JbaW5vbYwmnWBCqh3LmANXydUb+ZSKz5hrWvkdZYz4lDM4EeOvPgDmGG10qgzdgw4H1tA03qaABsNWut/TlS0FtVEw5FGUh2ux2vt561a78GIBwOU1S0lWg0SnV1FbNmzeF737sRn89HS0sLiYlJtLQMjRkhlfjq7lqYPfswdS0oB5RQJMqGykb+Z1IWJiaPfFmMPxjh401VTNArMDMmEnGPJitazba6Zuyb3+CKZWdwbGo19lifrMTGovb9HTPeagZSmJmIvWIxtvpNNM+9hcDEb5Cw7hmMxlJMmwvT4SbqtnKc5M9+CeiEMyaj7dyEUbeZUOY0TN1GOHMKANHkHMKpBTi2vY+tbnN7Yg2AYSeSkk+0j0/9+4uqge4Bw7uVaEIWEXcu9urVfW7CgSuVqMONrU4CEPGMVU04FGWAaZrO7353N/fd9yd8Ph+RSIQLLriYMWPyueOOX9Hc7MM0Tc477yLcbjfz5y/gV7/6GQsXfqo6jg0z3V8LzZhmVF0LypD27PIy1u1o5MfHF7LTFyAUMTmmMJOMJAf/WlmOaZqkh2twGq00pU2ASIBEM0By2It/+xI8BPhuwkdgDdDR3iYZYHJ2Mj8+bjzHFmZglMaGlss9AtPhxiX/jWPbe0QTs0HTiLitZqr26tW0TL0U05ZAwrpnMaIhWg69jqYT7iWaMrp9361TLyX5i98BEEnvkEADDWc+g+lw9+dp228qge4Bm7eIsGccgcIzQNPb2+fsN00j6s7DVrsRE41IagH2qq/iE6yiKN265prvtr9+4IFH9lr/z38+tteyMWPy1eP7Yagn18KezUnUtaAMBW2dbDXNejr+6uoKSr2trCxr4MxpIwEr8Z2Rm8LLX1Xw3sYabpvQCqXWSBdtTUfztOAwBrAAACAASURBVJ00lq0jF5jT9AEAUYd7t9E6NE3jW6O9RBLTMRrLrPkwkrIJOZIw0TCaKwnFhsNr6/xn6jb8s36Ao2whWrgFgHD6BCKx2uc2LVMvI3HlP9Bb6winTdxtXTS1II5nLL5UE44eMLxFRNLG0XrIVXjP+7+47DMSe8RhOlOIJmSgqSYciqIoiqL00LX/Ws0fPtgCgNcfotTbylnTsmkNRXlqaSmpCXbyUl2McDv54THjuG5+PueNagSskS7aZlQu1CpI9W8DwB7xE7UnE8o9Alv9rgRab64k7eXTSFj3XGzylVzQDUxn6q7mGLG2yqYzhYh7FK2TLyaaMnq3pLhjG+d2jiT8s36AaXN1vn6IUjXQ3dBavegttUQ84+O632jsEYfp9GC60tCDTRAJgbHvGbAURVH6ixDiFOB+wAAelVL+YY/1VwJ/BMpji/4upXx0QINUFAWvP8RX5Y18Vd7ImdOy8baEADh9ajbjM5P4yydFTMtNaa+dvmBmbPSvD7cQTcjCdKURcaQQdaZyur6anGgdQWc6jkCdlVynT8BR8glEw6DbsJd/iWZGsO1ch+Erb2+mAVZTDvvOdUQ6dParv/B9zNjQc23D0rUNXdeZlkOvpXXyBZguT9zPVX9RNdDdaOtAGO8Euq0GOupKI+pKB0ALeON6DEVRlJ4SQhjAA8CpwBTgYiHElE6K/ktKeWjsn0qeFWUQfL3Dqkm2Gxp/+mgrX1c0YmgwOdvNBTPzOGFiJmfPyAUzinPjy2jBJgBs9ZsJt7Uz1g1COYdxnLkUgPDUiwGrljicNhEtGsJosEbmsJcvBqxJUPSmsvaOgrBrJI22kTjAqoVGt+poTZcHM3mk1e7ZntD5B9I0TFdaPE7NgFEJdDfslSuBXTPjxEvbtzfT5Wm/aNRkKsrBYH8nxzjQHICf83Bgi5SySEoZBF4E+m3e3APw/OyXg+VzKv3rq7IGzn9iGW+ts8ZbXl3eiE3XuOX4QtZVNvHCinIKs5JJsBvYdI0/nDmFc2bm4dzyJikf3oxz02tgmrGh4nZ11AvlHoFuhgFoFecRThcERy8gkjEJAFuNNTKNfccS633dJnRf5e410HlHEknJJ5zddWdas/BkgvnHx/ekDDLVhGNfTBPXhhcJjTiUaGofh67bQ1sPVKsGOpZAB+qJxPUoijK02GwOmpsbSUra9WhxODJNk+bmRmw2x2CH0ht5QGmH92XA3E7KnSuEOBrYBNwspSztpMw+qetAGQ6qmgK4bDqpCf3b9PKLbXX8+D/rCEdN/vzxVo4cm8aaigYmZSfzjekj+XjzTr7cXs+0nD1GqzCjJC6/HwBbnURvrkQPNu02VFwo9wirqOEk4hlH/cUfWiuiEaION/aKJQRHHYWtfguRlHyMRqtGuuM026bLQ91li/b5GSKn34fPO7yGflQJ9D7YqlZiq5M0HXt33PfdPsOO00M0yeota9Rtti7m2BSZLYd8m8CEM+N+bEUZLGlpWdTX1+Dz9b25Ul+meo63zmKx2RykpWUNUkT95v+AF6SUASHEd4GngH1WKxmGhsez+zS8ycljKC8vp6amrMuf4VD/+fZ0O6fTSUHBGGy2+CRZhqHvdT4Hy1CJZbDiuPTZlUzJSeGecw/pUyz/+GQrNkPj2gXj9loXDEe595MixqQnctc50/jWY0v54ydFrK/yccnho0lLS+IP5x7C5U8s47QZue3H1hf/Hb14IVqdxDQcOJu2YQ+WAJAwejquthhTDsd0JIOnAE/6Hgn46CNwVS7B3rDKej/nKvjodgAScwpJ6MXnHI7Xikqg98G14UVMWyKBCfF/imm60gmOOopQ7lwiaYWE0wWuDf+iddpl2Hauw165gnD6JJVAK8OKYdjIzMyJy748nkS8Q6RGYyjF0gflwOgO70exq7MgAFLK2g5vHwXu6W6nkYjZ6blJTd337GJD6Zz2NRafLwSEhkQs8TRUYhmMOEzTZNvOZnTY7dj7E8tzS4rxBSKcNSkLl90AoM4f5Cf/WY/LrlNc5+e+c6YxLsXJtUfm88/Pt2MCIsM6ViLwylVzoC2WSIjMj+8AVyqBgpMwnanYSz8jULQEG+B1jcPsEGPiId/GdKXRskfcCSMOI3nrfzEX/oVI4gi8Y84kg9sBaNAyifbicx7I10pWVufjUB8UCbStejVakwbuQ7ov3IF9xwqCo46yvp3Fm6bRcPaucURbp1xM8ue3Y+xcj70i1li/qSz+x1UURencMmCCEGIsVuJ8EXBJxwJCiBwp5Y7Y27OADQMboqIMnnWVTeSnJZDstNHQEiYYMaloaO3VPnyBMMnOXalXU2uYal8QgEXb6jhhovXUanmJlzU7rI6BR45N48ixVlPPq+aO4bjCTL7YXsfRsdkB92Q0FqNFw4RP/B2No88kYdWDuOQrOLa9Rzhd7NVZzz/3J53up32a7Zo1+Ob/mmhStjXsbksd0eT4VIQcyPotgRZCuIDPAGfsOK9IKW/bo8yVDMCQSIkr/4HRVAznv9vzjaJhjIZtBMeeFO9wOtUqziXpi9+TsP459Cbr75PeWDIgx1YURZFShoUQ1wPvYQ1j97iUcp0Q4g5guZTyDeBGIcRZQBioA64ctIAVZQA1toa45oWv+M68MVxzRD5VPmu6vobWME2tYdyuztMp0zT5YNNO5hWksWZHIzf+ey1iRDK3njSBKSPdFNU2t5d9d0N1ewK9scqH3dB477p5uOz6bn0FCjISKcjouhmCUWfNIGhmCgAisTbP9urVtEy/osefOZx1CKYtEdOeQMvUS61laYUYRgkYql1/f9ZAB4DjpZQ+IYQd+FwI8Y6UcvEe5f4lpby+H+Owhomr+LJX2xiNJWjREOG0zscsjDfTlUar+CaudS9gxi5Mo6kczChoarAURVH6n5TybeDtPZb9usPrXwC/GOi4FGWwbazyEYmaFNdZM+pVNwXa11U0tCJcnT+p/nhLLbe+uYHrF4yltjmI06ZT2djKo18Wc+8509i600qgjy3MYNG2Orz+EJ5EOxurfRRmJnWZmO9L+wQoGRPAz65h64BQzhE935Fhxzf/l9YU3bExnVtmfh+tZWevYxqO+i0zk1KaUkpf7K099m9QeoREk0eitdRBJNB94Rij3prdJzJACTTEHqPoBnrIRyh7Flo0iO6rxCn/DZHgbmVd618g6bNfwRDpZKMoiqIow9WGKiudKfNaTTaqfbvyifLGzptxRKImDy7aDsCyknpWVzQydaSbk0QWy0u9BMNRtu70k2g3+N5RBUSiJk8uLcU0TTZV+xAj9q/5qFG3yRqowJEEWFNrmzYXAMHczgbW6VrrtMsJjvuf9vfBghMITL5wv+Iabvq1DXRsYP4VQCHwgJRySSfFejUkUmc9urujZVr9YjxGI3h6Nhyd3mIN1ZKcPx1c8e052mUvUM84ovNvRl94D9rsy+DtlXi2/gtj0Z8JuwzMGVZzRK3kS4xPfoZmRnFMOQWzcP+bmQyVnrGgYumKiqVzKhZFUQbKxiprIpLyhl010BpWrWC5t6XTbT6QNWyr9TMmLYGvyhsJR00uP2wU03NSeGX1DlaVN1BU28z4zETGZSRx2pRsXv6qnGMLM2hoDXebQKe8daU1WyCAZtB4ysMEC07AVreZcNqEXTWkmm5NjBJqxkzad+ddpef6NYGWUkaAQ4UQHuA1IcQ0KeXaDkV6PSRSVz2698Wup+MBfDu2E6Znw0q5d2yAxGy8rTZojW/P0X32Ap16HfqYM9EiAdIBVj8PQHjLZzTlfwOiEdJfu5ZIyhg0M4r54R14M+btdzOPodIzFlQsXVGxdO5AjaWrHt2Kogxd62M10HX+EC2hCNW+IFnJDgLhKOVddCT8YFMN2W4nNywYyy1vrAdgRm4qs0anYjc0vthWx9adfo4utDoDXntkPu9vrObnb1p9cydl7zuBtpcvJpw1neCoo0hY8xSOorcJjjkWw7uV4OgFuzUx8B39W/XEOs4GpHGtlNILfAycssfyWill23OQR4HZ/XH89nGWfZU93sao3zygzTfaaTrRlNHts/wYPqtDob3Cqrw3GrZj+Mrxz76B5sN/hH3nOuxlXwx8nIqiKIpygPO2hDod47vM28JpDy1mVVkDDS0hq51zrEa43NtKdVOAbLeT3FRXewLdEopQ12w1twxHTZaXejkiP43Zoz24tBAF2g4O9bSSYDeYmZfKG2srqW8JMT7TamqRk+Lif0+eSG1zEEODwtjyToX86CEfgXGn4D/iZ4Ty5uEoX4zeVIoWCbR3HGwTHjmbcM6ceJwyJabfEmghRFas5hkhRAJwErBxjzIdx0HptyGR2uZn1/1Ve63TWurQ/Hs0iDdNjPqtg5NAt7ElEEm0HrVEEzIxGovRfRXYdq4DsL51jjnOKlq7vtNdGDvXq2+ciqIoitKJ0voWTn1wMT99Yz2+QHi3dW+uq6LGF+SJJSVsjNU+nzAxE7CacVT7AoxwO8lLdbUPZXf7O5L593zM7e9KVpR68QUizC1Iw+2y8XDyo3zi/DH5Lx6B7t3GJXNGMSErmbn5HhaMS28/7mlTsvnZCYWcd2hu+5jQndGbrXwmmmjlN6HcuRiNxTi3vAlAOH1inM6S0pX+bMKRAzwVawetAy9JKd8cjCGRTKcH0+ZC9+3Ya537wx9iNGyn/uKPQbcuVt1fhR5sJJw2vj/C6bGoexSGvxr/zO+R/MVvsZcvxlYnMXW7ldwbDqKudIy2HrcdOLa9T+rbV+P9xkuE8o4chOgVRVEUZehatK2OcNTks621XPjkcn6wYCynTcnGNE3e21iNoWt8ub2eioZWEu0GJ4ks/vH5dsq8rVQ1BZhXkI7d0PlkSy2+QJhF2+ooyEjinfVVfL61Fg04bIwHgNluL4HAKJzNZTjKPmf+tMuYPza907jOOzS329iNWIVgNNamuW1K7qSlfyacNpFw9qFxOEPKvvRbAi2l/BqY2cnygR8SSdMgeWT7N7aObDvXYTRX4dz6FoEJZwFWuyKAcPZe4Q+oSMpobDVraZl6KYnL78de8SWGb4f1zTI21F04bQK2ur0T6ISvHwfAqC9SCbSiKMoAe+WrCtIT7Rw/cdhN5z6k7GwO8t+tdZw4Lm23sZJ7YklxPaM9Lm47RXDvJ0Xc9o4kK9lBgt2gzNvKjUeP5aEviinztvDnb0wjL9VFksNAVvtoCUUZ4XYyJi2Bp5eZ/OGDzQTCUX51+mQ+21jFY4tLmJydjCfBmsLdFW4klDubSEUIe8ViAhPORm+uIpI+oZsoO6c3VwMQjT2pDmdOJepwoweb8M+5UQ1/OwAOipkIAUx3Dnrz7m2gtaAPI5ZUJ674G4HCM0HTcJR+RtTpIZzVu5kL480/6wcEx54CjiSC+cfh3PIW6DaCBSe2l4mkFeLc+qbVVCN28zDqNuMo+9x6rWYzVBRFGRBl3hbW7WiiORTh7g+3YOgad+s6df4gYkQyU0aqDqTx9vKqch5fUsqEK+fsc3IRgJL6Fv7yyVZuP0WQ6DBYWdrAqVNGMCMvlYcvnMFZjyzh6aVlJDgMHIbGOYfkkJfqwmkzmB9rZjHKk8DyUi8AI5IdzB+bzmiPi/c21pDsNDisII2JaS5K61uYW7Brxj8t4MV0pRHKPQJ7+WLcH9+CY/sH1F32RXsz097Q/bEEum1b3SA0egFG3RYrl1H63UGTQOPOQS9ftdsiw1sEQGDcqTiL3iFp0R00z/819tJPCY5e0N6kY7BEMqcQyZwCQMshV+Pa/DpgfdNsL5M+AX19A1rLTsxEq6YjYfXDmLoD05GMrhJoRVGUAfHPz7fzvqwBYGZeCrX+ED95fV37+gsOzeWWEwaxb80wtL7Sap+8sszbbQL9/IoyPi+q48NNNRRkJOIPRZibbyW5TpvORbPy+Mfn2wG4fsFYkp22vZ4gjPa4kNU+Ulw2DslNwdA1Lpk9irs/3ML8sVaTDpuucecZk3dtFI2gBRqJOj1E0wWuza9jbH0LgIRVD9F81K/pLb25yvo77/S0L2s84X40MzzoucvB4qBJoE33SIzmyt1rar1bAWg+/CdEknNIXP0IRlMpRnMV/tHHDGa4ewlnzyKUdQj2mq8JZ+1KoMOxnra2+s2EErOw1azBtf5FWg65GlvtegxfeVe7VBRFUeJobWUTh+alcNgYD+cfmktTIMKb6yo5pjCTp5aW8vraSm4+bjw2vXdNDZTOmabJhtj4zCvLGpiQlczKsgamjnQzZ4xnt7KtoQjvbbRqbT/eUsu4+hYMDeaM3lXuvBm5vLiynMPGeLj8sFGdHvPaIws4alwGx07IIMlhpVBnTM1mYVEt587ovO2yFmxEw8R0edrbKpu2RGv4uXVP45/1fcxEq4OiY8ubGA3brQ0NJy1TL0ULNePa+JL1OSadj5mYhe6vJpqY1Z7PAGBPGJzZ6g5SB00CjTsHLdyCFmzEdKYCYNRvxdR0Ip4Cmo/6DRgOElc9CEBwzNGDGe3eNA3/nJtIXnQH4cxp7Yvbhqox6jYTyp1H8me/wkzIwH/4j0heeBv28kWDFbGiKMpBo94fpKKhlfNm5HDZYdbkXWmJ8P2jxgLWVM0fb97J9jo//5U1jM9I5ORJalKLvihvaKWhNYzDprOitIHV5Y1UxqbYvvP0Sbud348278QXiDA9J4XlJV5WlTVw8qQRu02V7XbZ+M+3D8dl07tsTz02I5Gxe9R0u+wG939zepdxaq1Wk4+oy0MkrZBwxhQChWcQGH86juf/S+Lqh2medytaSx0p730PrUMarIX9GA3bcW182XofCeI/7IfozdXtHQiVwXHQtDI3U60bWvInP28fjcPwFhF1jwbDCZpG85G/pOHUx/DNu5Vocve9YAdacNz/UHfZIkzHrsHVo8k5RO1J2Oo3Y3iLsFcuxz/7ekxnKhF3ntVxMhIaxKgVRVEOfNEuhgQ1TZM6f7B9oo2u2jm3TYqxstTLU0tKeGdDdf8EehBpm177nENz2dkcpLIpwD1nTWFCVhIPLtpOOBIF4Kmlpdz9wRbGpCVw0zFjCUdNwlGTa4/ce2biBLvR686I3dEDVgJtOtNA06i/6H38c24kkjaewISzSPj6SbTWeuw7lqBh4v3GS9Rct5VAwYkkrHoIp3wV/yFXE0nMRm+0Jmu2aqBVAj2YDp4EeuJp+A/9Ls5t75O06LeA1YQj7Bm3W7nguP+hZdb3ByPE/aNphDOnYa9YjL30UwACBdbU3lH3KDQzit689/B9iqIoCgTC0d0m0lhR6mV5iXe3hHnJtjqO/dsittftPttkMBzl1jc3cNqDi3nlqwo0up49Lj8tEadN51+rKoiYUNUU6LSc0nPrK5uwGxqXzrUS4TmjUzm2MIPr5hdQ6m3lzXVVFNU28/eF2zh0VAr3nTON6bkp5KclcNHMPEZ5EgYkzo410Hvyz74RLewnYfWj2CuWYBpOQiPngOHEP+cm9GAj6DZaZn2fqDuvfWAAvbmqfZI4ZXAcPE04dBvN83+FFmjAufUtiASweYtoyZs32JH1WaDwdNwLf03C2meIpOQTTbVuJhG31YbLaCojmjJmMENUFEUZcvzBCGc8vISfHD+e06Zks3VnMz94ZQ2RqMmIZAenTM7mO/PG8MmmGlpCUV77egc3HzueptYwt72zkXWVTdT5QyTaDT4vqmNsemJ7u9g9GbrGhKwk1u6w2uxWd0igw1ETXQM9zjWfw5lpmqwub2RiVjIiO5nvH1XA8RMy0TSNBePSGZ+ZyDsbqtsbQ/z4uEJGp1kJ80tXzWEgz/SuGui9E+hIhiAw/jQSvn6caGIWoZGzdw1Tmz2TlmlXEEnOIZo0kkjKaOxVX0G4FT3gVU04BtlBUwPdJlhwEnqwiYSvn0ALtxDxHPg9ogPjz8DUdGz1mwmO3tV2OxqbDlxvLEVvLFGzEiqKonRQWt9CUyDMkuJ6oqbJ7/+7mWSHwW2nTGTiiGSeXlbKcyvKWFlSD8Bb66oIhKM8t6KMhUV1zM1P456zpvDT2MgaU3L2PUxd21TQAA2tYZqDYf73zQ2c8PcvuPHfa/rvgw5D726sZs2ORk4UWWiaxlVzx5CfbrVN1jSNIwvSWbOjkaXF9XgS7Iz2uNq31TUt7s009mVfNdAAzbNvQg82YfMWtXcybOM75k5aZl9vbe/OQ/dVdJiFUCXQg+ngS6BHL8A0nCR9cSdRVxqBwtMHO6Q+M5NGtE+WEhy9oH15JNaOO3nRHWQ8cyRpzx+Dc9N/BiVGRVGUoabU2wLAhkofX2yr4+uKRm48ehxnTB3JX86ZxvQcNx/Inawpb2BydjINrWEeW1zMiyvLOWFiJnecNonjJmRy6pQRXH7YKM49JGefx2tLoCdkJQGwrNjL+7IGZ6wTXDAc7d8PPEzU+YPc8+EWZuSmcPGsvE7LHJbvIRQx+WjzTg7JTYlbwuzc/DpJn98BQNIXd+KUr0A0jPu/N2Cr/rrTbXbVQKd2uj6SNZVAwckAhPKO6LQMWE+VtWgIW431ZUsl0IProEugsScSHDUfDZPmI36O6UrrfpsDQMu0y4gkjSQ06qhdC20uIokj0AMNtEy5BNOeTMp/ryf50/8dvEAVRRmyhBCnCCGkEGKLEOLn+yh3rhDCFELMGcj44q0slkC3jYyRYNc5ZfKupOSYwky27GwmFDG5cu4YZo5K5YklpTQHI3z7iF0d0HRN44ajxzE9N2WfxzuiII3J2clcFEv6FhdbNdvnzMghHDXZvLM53h/xgGGaJttr/bu1R+/IFwjz8lcVVDS08uW2enyBCD85fjxGF0MCHpqXit3QiJowo5ufS2845askrn6YhNWPkrjqn7jWv4BRvxXXptesSc06obV6iTrcoHfdarZ5/i9pmX6F1f65C9FYs0xn0TsAhPdzFkMlPg6eNtAdtMy8jkhKPq1TLh7sUOImOP506sbvXZveOvVbmPYkWmZeB9EwyQtvI2HtU7SKcwmPnDUIkSqKMhQJIQzgAeAkoAxYJoR4Q0q5fo9ybuAmYMnARxlfZd5WAEzgvQ3VzB+XgcO2q17p2MIM/r5wG2AlYccVZrC0xEtzMEJhrBa5N3JSXDx96SxK6q3E/cvt9WjAqZNH8PjiEtbtaGLqMJmtsC0Rbqv5rfMHWby9npL6FjTgRJHF+Mxd5/DvC7fz9LJS5o9N5+cnFjIyZVeTi03VPn7wyhq8LSFOn5qNTdNIcdmYOKLzDptgjaYxPSeFlWUNHBLHBLqtE1/y57cDYKvfglG/2VpXt7nTbfSAt9P2zx1FPOPwHX3nvsu0JdDb3iOSnGuNIqYMmoOvBhoI5R1J89G/PSjmivcf/mMreQbQbdYQfQkZJC390+AGpihKvxFCnCmE6O0N7nBgi5SySEoZBF4Ezu6k3G+Bu4HWPoY56MoaWhgVaxsbMWFewe5PJPPTExmbnkh+eiIZSQ40TWNufhrHT8js03FHJFudxCoaWsnzuMhPSyA90c76ysY+7benvipraK997y8/eGUNf/zImqzs9TU7+OZjy7jtHcnji0t4dHEJd76/K9lcuLWWp5eVMnNUKivLvFzzwlcU1e6qjX99TSUtoQhzRqfy+dZalpd6OTQvtdtOl8dOyCTVZetyZJReM030pjKiDishjyTnobfU4qj4EgBb3Sao2UDac0ejN1e2b6a1erts/9wbbQm0Fm612kqrTqeDavhnkMruHEn4Z12Po/QzbJUrwDTRAgNz01YUZcBcCGwWQtwjhJjUw23ygNIO78tiy9oJIWYBo6WUb8UnzMFV5m1lek4K2W4nYDWx2NPtpwruObfrSTL2h8tukBqbwGNseiKapjFlpLt9Wuq+ikS77jBumiY/fn0d939aFJdjdSYcNVlV1sCrqyt4aVUFv3t/M5Oyk3n60pl8cfMCfnzceNbsaGTtjkZaQxH+8MFmJmQl8bdzp/PYxYcSMeFHr+2aAv3L7XUcNsbDuTNyaWgNU97QysxRnbcn7ujCmbn837VzcdnjM7W1FvCih5ppmfk9fEfdju+o2wBwFL0LgN5Ygr7239i8RdhqdsWvBxq6bP/cK/ZEoq50YN9tpZWBcVA24TjYtU65mKQld+OSr6I3fE36p3dTd+UyTMfweHSoKAc7KeWlQogU4GLgSSGECTwBvCClbNqffcZqtO8FruzNdoah4fEkdl9wr+30/dqupwKhCNW+ABNGpmC3G2yq8jGtIGOvckd6EjEMnUgkvh38cjwJNFQ2MTkvFY8nkdkF6Sz6eAuGy47bZe9yu+7Oy9YaH9/45xc89K3ZlHtbuO/Dzbx744L2GffKvS00tob5ekcTqakJfepc11UsRTU+wrEk/o8fbWF8VhJPXz23vXnMpfPH8tAXxbz41Q6m5aZQ7Qty7wWHkp2ZTHZmMlfPL+CP72/CluCg1heg1NvKVfPHcsqhefz6nY2EIiZHT8puP3Z/XyuYJoRboWUnAM5RUzAnnQmNZfAuGM1VmLodLRpC//p5AJLwkpiaAOEWbKEGzLS8+MToGQOVdbjEsbi62V+/n5deGCqxxDMOlUAfhExHMoH8E3FufRNtux0t5MNWs6Z9JA9FUQ58UspGIcQrQALwQ+Ac4BYhxF+llP/P3nmHx1Feffue2V4krbqsLtnyY8sV94Ix2GAwzQECoSeEl0BCIIGQkPYCIeRLedNIQkhCCDUECAQwxGCaCwbbuHePi2R1S7K6dqUts/P9sStZstW77LmvS5dmZ55n5syuNHvmzDm/88cOppQAbZMqU8PrWogAJgNrhRAAScBKIcSViqJs7cwOVdWorfV0trlTXC57n+b1lPwqD5oGsVYDXzw/m0AXdg6GLXH2kJOc7DBTW+tBxNrQNHh3Z0mXLb47s+WV7SXMzYjm88Jamv1BfvHuAcobvFR7/Ly9vYhLcxMB2Hok5ARWu33szq8iPdrGY+8fQpYkfrRsfK/OoTNbdh+rBmBSijOpHwAAIABJREFUUgT7jjfw/SXj8DQ203bk1VOTeH5LMav3l7MgKxoRbW3dl8sUcrQPFtWwvagOgOmJTgJNPmanu9hZXE+y3Xhy/CD/rZiP/peIj76De+GPMQENhgQCtR7Qook1OZD9bnxpi7AUfIzUGErd8J4oRlv/J+xbH0dTfXiT5tI4ADZGRGRgajhOrTwGutnfYL8vvWGk2NIXO+LjOw4uDpoDLYSwAusBS/g4rymK8vApYyzA88BMoAr4kqIoxwbLJp2TeMevwNqmYthYsRt/ygLkxjIcm35J09SvEkiYOowW6ujo9BUhxJXAbcA4QtfYOYqiVAgh7MB+oCMHeguQI4TIIuQ4Xw/c2LJRUZQ6oDX5VwixFnigK+d5JNMiYZfmsmEzGaDzoO+g0JI2kh0XiobNSHWRHGnhjd1lXTrQHVFU08Sv1xzlikmJrR0UW9pcO8wGPlQqWx3oQ5Unc4t3ldazq6SelXvLcVoM/PCinAGRe8sPd2z8/dWTOeH2MS7u9ILLr5+bxaSkCD47VsOts9sXwyVHhfLSy+qa2XSsmlSXtbUJyneXjKO8wYuxE/WNwcBUsQfZ34hVeR04mYuMJKFGj0Ou2IUvaxnmwnVImgqA7K5AChQgN4duJrorIuwp7gU/RvI16vnPI4DBzIH2AksURZkGTAcuEUKcmrRzO1CjKMo44HeEClN0hgBf+gUEzZFo8RNRnSkh/UpNw7nuh1iV13C9/gUsh1cOt5k6Ojp94xrgd4qiTFEU5f8URakAUBTFQ+i6exqKogSAbwKrgQPAq4qi7BNCPBp2yM8YKhu9vLw9FFxPG6J2zqeSGWPHZpLJDDf/MMgSX5g6hq1FdRyr6l2E7LP8kJO2uaCGvWUNLMiKJj3axqLsGFZMSWJTQQ3vH6xg87EaDle6SXNZibIaeXvvcX695ggOs4FGr0pxbTP+XqSq+NVga6pGW45Ve0hwmnHZTB06zwBGWWLJ+Hh+vGw86dHtP4MxYQWOsvpmDpQ3Mi3lZP5wqsvGzLQeOqOBJuTa/B6eTefIDaHSAFPZFjSjvZ0zrMaEovaBuEmoLV2A7QnInvLWedB5E5XeEnSOQdXl60YEgxaBVhRFA1oqIkzhn1P/01YAj4SXXwP+JISQwnN1BhOjlfrL/oEjfgzBj36GqWIX5vz3sBz7APfMezEXf4Jz/Y/xZSzBUHOYQNyk1vaiOjo6I55HgLKWF0IIG5CoKMoxRVE+6mySoiirgFWnrHuok7HnD4ilw8B9b+zjWLWH7y0dh8s+xKHnMNdMG8MFOXGh6HeYKycn8bfPCnhr73G+tTi7x/v6LJwyUdHoA+DS3ER+fnkuRllCqWjkpW0l/Oi/B7EYZRxmA+ekRuELBPkkr5qkCAvfWzqO+9/cx77jDXzrP3vIiLHzyyty20n6dcTXXtyGy2Lk4UtEu/X5VZ7WG4O+EGM3YTHKHChv7DSC3RNse1/EsekXVN22A83Sdyk7Q8PJTCY1Mq1d9NefeA7mo6tQo3MIJJ6DbI1ENUYguyuQm04QiJ2AoebIyai1zhnDoKpwCCEMQoidQAXwgaIop+qGtlZ9h6MfdcDpVRw6g4I/eR7ET8QfPxVDfQHOdT8iEJuLZ879NC56FLm5mqg3ryP6tSuwb/ndcJuro6PTc/4NtA0lquF1Zz1Vbh9KRSNfm5/BtdOTh80Ok0FuTeNoIdZhZmZaFJ/mVfd4P81+lW1FdSwee/Krc1JSBHazAbNRZvKYCO5ZlMUPL8ohqGlUe/yMj3eyeFwsCU4zf7xmCvMyozEZJF7aVkxRbTMb8qr531UHuz32npJQq+y2aJpGQXUTWbF9d6AlSSIpwtIaWR8b17d9yZ4KJNWLqWxLn20BkBuK0Qg5zWpE+86Hzbk3Un3zBjSzk4bFP0e96Q2CjgRkdxlyYym+jAupvnUzvrGX9ssGnZHHoBYRKoqiAtOFEC7gDSHEZEVR9vZnnyO1ors3jDRbrNlzYBPInkrU6/6JKyYSYhYSHH8ppkOr0Ex27Idex7z0QaR9oRwwLXspRA7sl89Ie190W05Ht6VjRpItYYxhLWcAFEXxCSH0R0jAzpJQUdqMtAGQFRsEFmTF8Lu1eZTWNXOwvIH5WTHtotSnsqWwFm8gyNXTxpBf7aGwpomJSSd1jyVJ4tY5oRzjY9UeXtpWQk68g0VjY7lyclJrzvO4OAcHyhuxmWSun5HCM5uL2FdWT361h3f3V2CQJX5xRS52c8iWhuYAdU1+AE40eolzhm4GKhp9ePxqvyLQAGOirBQcq2m1rS9IgVAqjKl0E6pzDFKgiUDSzJMDNA1z3ip86UvA1Ekqj+pDdpfjT12IuXhDazfAVmQjmj0+tGyyg9VO0J6AoTH0AEiNSCXoSOyT/TojmyFR4VAUpVYIsQa4BGjrQLdUfRcLIYxAFKFiwk4ZqRXdvWGk2VJnG0+swUJT7o24HRNbK3ulhT/DnL4MJCORH3wTnr0cY8VOALzZy6lf/tSA2zKS3hfdltPRbemY3tjSWUX3AFMZVsdYCSCEWAGcGIoDj3R2FNdhNcpM6KKL3XAyPzOG35HHT1crbC2q4/Z56dy1MLPdmBNuHw+8uY8bZqTwzOeFJDjNzExzcfmkRLYX1xHZiQzeHfMziLWbmZsR0rtuWzA4IdHJgfJGFmXHcuvsNF7ZXsrj6/LYXdZArN1ERaOPTQU1rU1kSutO9tE5UN7IIqeFao+Ph98NRa5z+9lRcUxkyCGPtBqJc/Tt3k/yhx3o4g1YjryN5Gug+tZNrZKtxvIdRL13J/VLf4d3wrUd7kNuLEVCwzv2cgz1RfjbOuCd0NZhPjVirXPmMGgpHEKI+HDkuSX/7iLg1GdCK4Evh5e/CHys5z8PPZrVRfWNa3Gf+0j79fZ4vOKLeLMvJmiOxFSxE885d9E8/mpMJZ9BUB0eg3V0dLrjLuCHQohCIUQR8CBw5zDbNCJoae1sNIzMPmKZMTaSIixsDcu3vbXnOIFTCvs+UirZd7yBH686yNETHr63NAeLUea2uen8sYumL06LkVvnpHWY2zwxMeRULhXxOC1GLpuUyI6SepxmA8/dPAOnxcBnbVJLSupOdjLcfzwkLf7bNUfZW9bAT5aLAXCgQ4WEY+McfVYGaXWgK/dgaChG9tZh3fNc63ZTuINgS7S4Iwz1odbdanQ21bd8ildc0+1xg/aTKirBSL3d9pnKYEagxwDPCSEMhBz1VxVFeUcI8SiwNRwZeRp4QQhxBKgmJJukMwx0+U9utOGZ8Q1Mlbtxz30Qy9F3sB76D8YTewkkTBs6I3V0dHqEoihHgXlCCGf49cC0uBvFHK9v5rP8ao5UuvnagozhNqdTJEliYXYM/9lVxv/MT+epjYWsO1rF0vHxrWPWHg3pN6dEWUmIsLB4XP9Lh5ZNiMcbCHJeOJf6uunJvLWnjHsXZxPnMDMvI5pP86v50yf57Cmt59zsUEe8xAgLB8obCWoam47VcJGIb5XM6w/JLQ50f3Kp/W40JCQ0/PFT0Gyx2Hc8iaEun+aJN2AqDZVlyZ7yTvdhaAg70L0oAgw6TjrQqlOPQJ+p9MiBFkI4gCZFUYJCiPHABOBdRVH8nc1RFGU3cE4H6x9qs9wMdPzcRGdE0TTzm7TEG3wpCwEwFX+qO9A6OiMUIcRlwCTAGm58gqIojw6rUcPIrz8+yrqjVUiE8oxHMncuyGDZhHimJUfx9t5y/rOrrNWBrvX42FFUyy2z07h7UdaAHdNhNnL9jJPOXmasnY/uXtDaBntBVgwfHjrBc5+HpNlsJgNRNhOz0118mlfNoYpG6poDzE4fGLm2pHAKx9g+5j8D4G8ikDANzWjBM+c7aOYIIlZ/HevhlZiOb0d2h5qeyO6KTnchNxShSTJBx5geH7YlAh20xXWeW60z6unpM6z1hC7CKcD7wC3As4NllM7IRnMkEIgej+Xof3Gu/xGG6kPDbZKOjk4bhBB/Ab4E3ANIhAIVIzfsOsj4AkE2F9SwYnISH39zQb/TCwabaLuZGakuDLLEpZMS2VpUywl3qCZ03aETqBqcPwBR5+6wtilebLnpaMlH3nismrRoG1OTI6lp8vOXTwsABsyBnpQUwe3z0rlQxHc/uBOkgIegPY66q17Hn7KAQPwUam7eQMOS32CsOYzsa0BDQvZ07kAbGkoIOpLA0HO5Q9UeisDr+c9nNj11oKWwAP/VwJ8VRbmWUGRD5yzFl7oQU8UubHueI+LjB0DTU9d1dEYQCxRFuZVQo6qfAPOB3vVpPoPYXlxLcyDIBTlxOC1DUjs/YCwT8QS1UN4zwBqlgliHmYlDfBMQ6zDz88sn8tT103DZTAQ1SI22cWluIslRVj7NryYr1k6809L9znqA0SBz18JMkg6/gHX/y33ah+R3oxlPTwHxjr2MgGssAP6U+d1EoItPV97oDrODoMmBGqHnP5/J9NiBFkLMB24C/hte17mujs4ZT9OMr9Ow6Kc0nvsIpvLtWA6/Odwm6ejonKRFIsEjhEgG/ITqUs5KPs2vwWKUmTlCpeu6Ymycg3FxDt5XKlGDGhuOVjE/Mxp5GFo5Xyjiw50AQ+9jWrQdi1Hm/vNDTV/mDFD0uS22ff/Euu+FPs2V/B60jlIoZAON5/8cz4xvEoifguwu7zQIZKjNb+0w2Bs8s++nedJNvZ6nM3roqQP9beAHwBvh1q7ZwJrBM0tnpBN0JtM89Taapn4Vf9xk7NufGG6TdHR0TvJ2WAXp/4DtwDHgpWG1aBiobfLz9KYCPlQqmZkW1S4lYTSxbEI8u0vrefdAOXVNfuZnRg+rPS2ttFPDLbjPGxvLw5eM55bZAx9xlZuqWgv5ekvIge44h9qfsgD3/O8TdCQhBX1I3trT5zfXYvCUE4ju/cObpnPuxJ+2qNfzdEYPPXKgFUVZpyjKlYqi/FIIIQMnFEW5d5Bt0xkNSDK+rItCedD+kaHJq6NzNhO+Rn+kKEqtoiivE8p9ntBZS+4zmfcPVvCXTwvwBoJcOTlpuM3pM1dOTsJuMvCLD48gSTAnY3gd6EXZMYyJtDAz/aSe9OWTkk7rrNhvtCBScw1yUxX4m7offwpSwAMdpHC0pUUxQ3ZXgOrF9foKrHtfBMBQcwQANeaszX7S6YIeOdBCiJeEEJFhNY69wH4hxHcH1zSd0UIgfgqSFsRYdaDdenP+B0i+hmGySkfn7ERRlCDwRJvXXkVR6obRpGGjstGHUZb48O757WTgRhuxDjO3zE7FGwgyJSUKl63nBW2DQVKklZV3zEUMch625K1D0kL9BnodhVZ9SMFApxHoFloUM2RPBdYD/8Z0fBuOzb8CvwdjTahAPhA9rvfG65zx9DSFI1dRlHrgC8C7QBYhJQ4dHQLxkwEwVu5pXSfX5hO16rZ2ovU6OjpDxkdCiGuEEEOfKDuCOOH2EWM3DUu+8EBz06xUsmLsXDn17Elll5tONiaWG4pB9fa4YF3yuwE6zoFuQ0vXQENDCfbtT6A6k5Gbq7HtexFD9RE0o7X3RYQ6ZwU9daBNQggTIQd6ZVj/WZdd0AEg6BhD0BaLseKkA20q39Hut46OzpByJ/BvwCuEqBdCNAgh6ofbqKGmyu0jboBUIYYbm8nAq7fN4svzM4fblCGjrQNtrM0j9rm52Lf9oUdzpXDKh2bqJoUjHIG27XoKQ0MRjYt/ji/1XOzbn8RYuZuAaxzIozN3Xmdw6amez18JFaHsAtYLITKAs+5irNMJkkQgfjKmNhFoY8Wu8O+dvdqVXJuPfddTeKbfSbAPlc86OjqgKMrIFjoeIqrcvoHPy9UZMqQ2DrT56CrkphPYtj9J0+Qvo1m7Vvw4GYHuOoVDMzvRjHaM1Qr++Cn4Mpagmey43rwWc1MlzeOv6v+J6JyR9MiBVhTlD0Db274CIcQFg2OSzmgkEDcFW/FfINAMRiumsONscJcjN5YRdHb/2FHKW0PMqzciqV6QZBrPe2ywzdbROSMRQpzX0XpFUdYPtS3DSZXHP+Kbpuh0jtxUDYBmtGEuC7fd9jdi2/0PPHPu73KuFPCE53bfClx1JGKsy8cz61sgSfhT5uMbMxdz2WbUPihw6Jwd9LSVdxTwMNByUV4HPAqclYUpOqfjT5iKPRgg9tlZeGZ/G2PlXvxJMzEd34axYie+HjjQ8vZnCFpcqDE5mI+ugnN/oj8609HpG22LvK3AHGAbsGR4zBl61KBGjcdHbLhzns7oQ24ORaADcZMwHd9KIEagRqZj2/s8ntn3QRe57VJYFaq7FA4A1ZWNZnLgy1rWus4z+z7MK6/HnzCln2ehc6bS0xSOfxBS37gu/PoW4BlCnQl1dPBlXkjD4v+H5fBbODc8AkBT7k0YK3ZjKt+JL3t51zvQgkiFn+HLWIovfQmR738dU9nn+FPm994Y1Ye5eAO+jLPGV9DRaYeiKFe0fS2ESAN+3908IcQlwOOEGmX9XVGUX5yy/S7gbkAFGoGvKYqyf6DsHkhqmvwENXQHehQjNVURNDlRXdmYjm/FnzKPQNwULMc+wFBzuEt5ud440A0XhR+wSyfLwvxp51J1yyaCejtunU7oaRHhWEVRHlYUJS/88xMgezAN0xllGMw0T76V+uV/J2gLyUX5k+cQiMvFVLq5++lVB5GaqvGlLMCbuRTNaMVy9J0+mWLb+zxR79yKoUrp03wdnTOQYmBiVwOEEAZC8nfLgVzgBiFE7inDXlIUZYqiKNOBXwG/HQxjB4Iqtw+AON2BHrXITVVotljUsAqGL3k+vpR5AJhKN3U5t9WB7kEKh2aJQrOc3qUyGJnaZZRb5+ympxHoJiHEuYqibAAQQiwEeq9qrnPGo1mjqV/2JyyHVxKMzMA77gqcnz2G+diHyI1lqDE5+JPnnTbPXPIZAP7k+WCy48tYgjlvNSx6rNcXMMuRtwEw1Begxor+n5SOzihDCPFHTiolycB0Qh0Ju2IOcERRlLzwPl4GVgCtEeawnGkLDkawGlOLA61HoEcvclM1QVsM/qQZBK3R+FMWoFmjUR1JmEo20Tz51s4nt+RAd1NEqKPTV3rqQN8FPB/OhQaoAb48OCYNDk0+lW1FtcxIjULS7ygHFX/qQvypCwFomvpVrPv/ReSq25E0FU020nDB/+GdcG27OaaSjWiujNAdP+BLX4Ll6CoMVQdQ404NgnWO3FCC6fi28HLf2r/q6JwBbG2zHAD+pSjKp93MSQGK2rwuBuaeOkgIcTdwP2CmBznVBoOEy9V9FPD0eXKf5rXg0UIFaFljIvu1n4GwZSA5m2wx+mrQIlNwTL0UdcpyosLf3VLmQiwFn2KIsoEkdWiHbPQDEBkXC/ahe7/Ops+nN4wUWwbSjp6qcOwCpgkhIsOv64UQ3wZ2D4gVg8z24loefvcQx+ubuXFmCt9enK070UOFwUzj+T8n4oN78cy4G0vee0R8/AD+1HNPKnMEmjCVfIY2cUXrNF/G+QCYC9fQ1AsH2nL0vwBokgFDfR8caE3D9epypDm3Q9a13Y/X0RmZvAY0K4qiQig9QwhhVxTF098dK4ryBPCEEOJG4Md0E0xRVY3a2t4f1uWy92leC0WVjQCYAmq/9jMQtgwkZ5MtMe4T+GIm0XjKMaxxs4nY9zoNBftQXdkd2mGrr8UJ1HoA39C9X2fT59MbRootfbEjPr5jJZ+e5kADIce5zSO8LjVkhBBpQog1Qoj9Qoh9QohvdTDmfCFEnRBiZ/jnod7Y01N2FNeRGm1j+cQEXtpWwtInNnLj89vwBoKDcTidU/CnLKD6K1tpnnobDRf8EklTsR78d+t2S/77yL4GgpOuaV0XdCThj5uEueDjHh9H8tZj2/Ms/vgpqK4sDI29d6Alby2mE3uR83p+XB2dEchHQNsWbDbgw27mlABpbV6nhtd1xsuEmmuNSKrcPpwWA1aTruQzKtE05KZqNFvMaZt86YvRJBnr/pc6nS75PWiSDAbrYFqpcxbTKwf6FLoL4QaA7yiKkgvMA+7uoCAF4BNFUaaHfx7thz2dcvu8DP71P3P5yXLBtxdnc252DIcr3by15/hgHE6nC4JRmfhSFmI98DJooRsY68FXUZ3JaJmL2o31p1+AqWxrOzH9TtE0nGu/j9xQQuOiRwlGpCD3IQJtCKd9SJUHez1XR2cEYVUUpbHlRXi5u+eWW4AcIUSWEMIMXA+sbDtACJHT5uVlwOEBsnfAqXL7ibXr+c+jFcnfiBT0EbTFnrYtGJmGd9yV2PY8jxTWij5tfsATKiDUnzbrDBL9caC7LB5RFKVMUZTt4eUG4AChHLthQ5IkbpqVyk+WC6anRPLc54U0+9XhNOmspDn3Bgz1hZiKP0VuLMVUuJ7mCde2kxACaB7/BZAMRK7+BnJDKZK7ov2OfG4MNUcg0IyxfDvWIyvxzLmfwJjZqM7UVme4N7TmTVfnhZrC6OiMTtxCiBktL4QQM+mm8FtRlADwTWA1oev1q4qi7BNCPCqEuDI87JvhJ4o7CT2FHLG1MCfcXr2AcBTTEjjpyIEG8My6FwJN2HY/3fF8v0cvINQZVLrMgRZCNNCxoyzR/vFglwghMoFzgI70zOYLIXYBpcADiqLs62pfA1WQct9Fgi8/u4UL/7yRCyck8KNLJxI/RC1fR0oyPQyTLTOuRlv/QyLz30Krz0FCwzzn1tNtcc1Avex3mN++m9jn56BZIgncuxfMTqSiTRheuAJJUwlOvTHcjtWKedHdmC125IRs5P3VuOwamHt+EZUDISdd0lRcainETR7os+81Z/3fSyfotnTJt4F/CyFKCV2vk4AvdTdJUZRVwKpT1j3UZvm0VLyRSENzgH3HG7h2evJwm6LTR4wnQq6AGpXV4XY1ZjyBMbNaFZxORfK70Uw9dlN0dHpNlw60oij97oEqhHACrwPfPkUCCUKyShmKojQKIS4F3gRyTt1HWwaqICU31sYfr5nMhrxq3thdxoYjJ3j6hulkxAz+l+BISaaH4bPFmb0cy8G3wRaHb8xc6qREXGrwdFvSV2BebsFUvgP79ido2vEmXnE1tkPrcWoq3uxLMO95BcxOfBlLqW8yQpMHiymBSKChuGux/VNxVOS3PuduOrYTr2X45c71v5eOGa22dFaQMpAoirJFCDEBECdXKf5BP/AIYc3hE/hVjWUifrhN0ekjptLNaEYrgYSpnY4JxAgsR95G1U6P80n+JuiBBrSOTl/pTwpHtwghTISc538qivKfU7eHixIbw8urAJMQIm4wbWrLvMwYHlgyjhdvmQnA/646iF/VCwuHAu/4q5H9bgz1BadJ2p2KL/sS3PMeRHWmYDn0BgDGmsOojiQaz/sZyEZkb10o5SOMGhGqheptGoehoZhAVCaabMJYrTdi0RmdhKXmHIqi7FUUZS/gFEJ8Y7jtGipWH6wg1WUlN2nwb1Z0BgdzySb8SbPA0Hkajhqdg+ytgzbpfYbaPKL/tRRj9UG0Xjx91NHpLYPmQAshJOBp4ICiKB12qxJCJIXHIYSYE7anBxVjA0tWrJ0fLRvPgfJGXtyqawcPBf7keaiORDSjDe+4y7ufIMl4x6/AXLQeqakKQ80R1Ogcgo5EmqbehmpPwJd+QevwlvarvdWClhtKCEZlQuw4DLoDrTN6uUNRlNqWF4qi1AB3DKM9Q4I3EOTfO0vZWlTLMhGvy5WOUqTmWgxVB/AnnyZD3o5A+OmidOJQ6zpz3mqM1QqG+sIedSHU0ekrPW2k0hcWArcAe8IFJwA/BNIBFEX5C/BF4OtCiAChApfrFUUZls5WF+TEMSvdxcq9x/nKnDT9wjvYyAYaz3sMyVuPZnb2aEpzzhewb/8zlqPvYqw+TFPu9QC45/8Q9+z7wXhSrijoSESTzRjqCnpllqGhGG/CNIzOGIyF3bcg19EZoRiEEFLL9TTcpvuMr6h7ZnMhT28qJDcpgi/q+c+jFlPZ50hoHXatbYsaE8r4lE4o4JoZnnvyuq2ZdAdaZ/AYNAc63Pa7Sy9UUZQ/AX8aLBt6y/KJCfx09SH2H29g0pjI4TbnjMeXvbxX49XYiagRqVj3vYgU8KBGh9PlJRlOvVBKMoH4SZjKd/T8AH4PcnMNakQqWlQshv3/wVB9qFc51Do6I4T3gFeEEH8Nv74TeHcY7RkSjlS6yY618+yN0/UgyCjGVLoZzWDBn3hOl+OC9kSC5kg4ocA4IKhiKv0cb9bFmIvWE7REdTlfR6c/DGoO9GhjSU4cZoPEewcrh9sUnY6QJHwZSzCd2AucjD50hj95LsbynRDoUr2rlZZ86WBEKkFxGRoSlqOrupmlozMieRD4GLgr/LOHXignjVbKG7wkRVp053mUYyrdhD9xerunih0iSagxOaEINGCsOoDsq8c79jJqr3oNz+wu+73p6PQL3YFug9NiZGF2LB8olWgdVPXqDD++jCWty4Ho7hzoeUhBX4+j0C0OtBqRChFJBMbM0h1onVGJoihBQrKhx4A5wBJC2s5nNBWNXhKHSI70bMdQcwR68T0peet71BRL8jVgrNzTbfpGC4HoHKTKA5iKP8WivA6Erv2BhGkEI/Q0Hp3BQ3egT+Hc7Biq3D7yqkaGPJZOe3wpC9BkM0FrNFonAvst+JNmoSFhKu1BLrMWxLbr72gGC2r0WAC8Yy/DWLUfQ23eQJiuozPoCCHGCyEeFkIcBP4IFAIoinJBOGXujMUXCFLt8ZPg1B3owcZUsIaYl85H2vXPHs9xrv0+Ue/c2u04Y9lWJC3Ycwc6fjKSpwrXW1/CvuspAq5s3XHWGRIGs4hwVDIjNZQztb24jrFxugTOiMNkx5d1EajdS9pqVhdq7MQeOdC2nU9hLlpPw+JfoFmjAfBmXYxzwyOYC9bQ5Bp+PWgdnR5wEPgEuFxRlCMAQoj7htekoaGi0QugR6AHG03DsSUkrGX49DeQdgUYTN1OM1XsQm4shWAA5M5dD3PpJjTZiD9pZo9nBeIkAAAgAElEQVTMac69CVvmDBrr3QCoUZk9mqej01/0CPQppERZSXCa2V5U2/1gnWGhftkT1C9/qkdj/clzMB3fBloX+t6ahm330/hSF9E86abW1cHINNTIDEzFn/bXZB2doeJqoAxYI4R4SgixlG6Kuc8Uyht0B3ooMBV/gql8B96si5FqC7AcfrP7SYEm5PpCpKC/W2UkU+kmAvFTTy8M7wyDCS1tHv6U+fhT5hN0junZPB2dfqI70KcgSRIz01xsL67T86BHKrIRZEOPhgZiJyAFmpAbSjodY6jLx9BYinfsZXBK8ZEv9VxMpRtDUZPemtlY2qOcPx2dgUJRlDcVRbkemACsIdTSO0EI8aQQYtnwWje4tDjQCboDPXhoGo4tv0d1JFG/7Am0xCnYt/4BgmqX0ww1eUho4eVDnQ/Ughgr9oQaqOjojHB0B7oDZqRGUe3xc6y6Z+oNOiOXFqk7Y83hTseYitYD4Es797Rt/tRzkX0NGCt29/rYUStvJnL1Xb2ep6PTXxRFcSuK8pKiKFcAqcAOQsocZywVegR60DGVbsRU9jmeGd8AoxX13Acw1uVjObKyy3ltr7/G6s6vxbKnAinoQ3VlDpTJOjqDhu5Ad8DsDBcS8OqOzqOWOqODlk5VhpojnY4xF32CGpke6kB4Cr7UBaExvUzjkGvzMdYcwlyyMZT3p6MzTCiKUqMoyt8URVk63LYMJuUNXiKtRmymnj2d0uk5svs4jk8exrnux6j2BJpzbwBAE5cRiBGhKPQpaXKWwysxHt8GgKHmMJpkQLUnYKgORaAN1Yew7nm2/XHqw1KizpRBPiMdnf6jO9AdkBJl4/oZKby2q4zPC2qG2xydfqBZownaYlsv2ic3BEMSTMEAppLP8KWeHn0G0GyxBGInYCrb1KvjmgvXtC5bjrzTa7t1dHR6R0WjT48+DxKOTb/CtucZ5OYqPHO/B8awpLgk0zTtDow1hzGcOKmSKNcXEfHhvTg2/hwAY/Uh1KhM1LjcVvm7iLUPErH+xxhLP2+d105KVEdnhKM70J3wjXMzSXNZeWLDseE2RaefBKLHYWyJQPvcONc8SNzfBPatj2Mq2Yjsa8CXdl7n82MmYKg52qtjmgvWEHBl44+fiuXwW/0xX0dHpweUN+ga0IOBXF+IRXmdpim3UfXVXTTnXt9ue8u101y6sXWdffsTSMFASIM/0Iyh5jBqTA6B6PEYaw5jKt6AqWwLGhKOrY+fPJbuQOuMInQHuhOsJgOXTUpk//EGqj2+4TZHpx+o0TkYag6H1Db2Podt/z8J2mKx7fo7tj3PErS48GVe2Pl8V3aoCLGHHQ0JNGEu+Qxf+gV4c1ZgqtjVpxxqHR2dnlPR4NU1oAcB+/YnQTLQdE7H9RzBiGTUiLRWuVC5oRTrgVcIRGUhqV5MZZ9jqM0nEJ0T6hqoeon84B5UeyKeOd/BXLQO4/HtQCgCHbRGg1mXkNUZ+egOdBfMz4wBYPMZlMbR7FfPuhsCNXocsrcOqekE5oKPCcROpGHpb5G9tVjyV9MsrumyZawaPRYJDUPdsR4dz779z0iqF1/2JTTn3kDQ4sKx+VcDdDY6OoOHEOISIYQihDgihPh+B9vvF0LsF0LsFkJ8JITIGA47T6XRG6CmyU9yVDetn3V6jal4A77MJV3Kw/lT5oUcaE3DvuPPgEbDxU+iIeFc/79Imoo/ZSG+9MX40hcTiB5L46Kf4Jl2B0FrNPZwFNrQUIwakTZEZ6aj0z90B7oLJiQ6cdlMbMw/cxzo7761n2uf2UpluOnA2UBLIaHp+DZMx7fiy1iCP3k+gRgB0FoQ0xlquIlKTzoSGst3YN/6B5rFNfhT5qNZIvHMuBtz4dqedUTU0RkmhBAG4AlgOZAL3CCEyD1l2A5glqIoU4HXgBFxZ1hYE3o6lBFtG2ZLzjACzRjqC1qvlZ3hS56H3FyNqXgD1v3/oll8kUD8ZNTYCRhrj+JPPAd/6kKCzmTqrvgndVe9jm/c5WB20DTta1gKPsJYsRu5oYRghF5AqDM60B3oLpAlibkZLjYdqyF4BmhC7yyuY1NBDfXNAX72/mGa/V1rd54pBOKnEDQ5iFj7IFIwgC/jApAkGhc9inv2faixE7qeH9XiQOd3eyzbrr+jWaJoXPTT1nVNU75C0BKF9cAr/TsRHZ3BZQ5wRFGUPEVRfMDLwIq2AxRFWaMoiif8chMhibxhp6AmZFJ6jO5ADySG2jwkLYgaDkJ0Rkvb7aj/fgWCATwzvwmEHGsAz6xvn6ax30LT1ND10b71cQwNRXr+s86oQW/l3Q3zM2NYfbCSI5Vuxic4h9ucfvHUxgJi7CZumJHCExuOsfiPn/Kdi8Zz3ZSkIbfF41OxmWQkSULTNHaW1LN6zVGKq9zcODOV+ZnRrdukTi68PUWzRlO94FHi1n2HoDkSf2KoRaw/dSH+1IXd78DsQHUkYqw9iqE2D9UxBkwdf1Gbyne0Rp5PrrThSz8fc8HHIfUPSb9v1RmRpABFbV4XA3O7GH878G53OzUYJFyuHnaVazdP7vG8ck8AWYJJGbFYjAP//9UbWwabobRFKgl1DbSlT8XWwTFbbYmagLrkEagtQEueQWRG+MHFontQE8din3Y59k6v43a0OXdh+eSXAFgSszH18vzO1s+nO3RbBteOQXOghRBpwPNAIqABf1MU5fFTxkjA48ClgAf4iqIo2wfLpr4weUwEAAfKG0a1A51f5eHzwlruWZTFLbNTEYlOntlcxN8+yecLExMwD8KXTgsBNcj7SiUldc0syo4h0mriume34rKZyIq1U1jtobTei8NswGE28K3/7OWmmakkR1l49vMi/n799H7lNh6r9nDjh2NYNfZGkhMSwGDq9T5UVzamko1EH3qLpkk34T7vZITZVLIRy+G38My6F0N9IU2Tbjltvi9jKdbDb2Gs2EUg8Zw+n4uOzkhACHEzMAtY3N1YVdWorfV0N+w0XC57j+cdKqtnTKSVpsZmBqP9VW9sGWyGzBYtiL14LwZJptYwBjo4ZjtbJv7PyQ0t66QEEF+Buq4/FWn8LcRsegLZ30ijIQFfL8/vrPx8eoBuy8DYER8f0eH6wYxAB4DvKIqyXQgRAWwTQnygKMr+NmOWAznhn7nAk3Qd8Rhy0qJtOC0G9h9vZMWU4bam9zzynkKi0wyALMGlkxKRJIn5mTFIwD2v7+WTvCqWjo8fNBs+PHSCh99VAFh9oIL5WTEEghoiwckJt4/xCU7uXJjJF2al4W5o5rdrjvLPbcWt8z86VMkts/teWLJyz3H8KvxKuo3/N3din/ahusZiLgnJNFkP/Qf3gh+B0YqpcC1Rq25HUr2g+gEIJE4/bb4v4wI0ScZ87EPdgdYZqZQAbf/RUsPr2iGEuBD4EbBYUZQRUUxRUO0hQ0/fGDBsu57GtuvvqNFjUSMzuiyyHgg0q4umqV/Fse0PqFEjoi5VR6dbBs2BVhSlDCgLLzcIIQ4QekTY1oFeATyvKIoGbBJCuIQQY8JzRwSyJDEhMYID5Q3DbUqvOVbl4b/7ypGAKJuJOenRxDnMrdtnp0eTGGnhH5sKeWV7CZdNSmTFlM4rrfvKZ/nVuGwmHlw6jh+8c4CCmhIunhDPY5e1d2btZiM+g8x3l47Dbjbi9gXYXVrPuiNVfXagA0GNVQcqWu3wBYJ9ira3FBJ6My/CcuwDLPkfoBmtRL53J4GYHGRPJdaDr6JJMv74qafN16zRBJJmYcl7D8+cBzrNB9TRGUa2ADlCiCxCjvP1wI1tBwghzgH+ClyiKErF0Jt4OkFNo7CmiZlpruE25YzBeGIvhoYiDA1FeLMuHpJjembdSyBhKmo3BYs6OiOFIUnGFEJkAucAp8oQdJRzN+JKcHMTIzhc6cYXCHY/eATwaV41T3ySz2u7SjHIElE2E7VNfpbnJrQbZ5AlrpqewqFKN4dPuHns/cP8cX3+gBZMBjWNzQU1zM1wsXR8HFOTQ7nBN8zo/GOWJYl7zsvi+xfmcMG4OHaX1vdZem/9kRNUuX2smJKE26eyrbi2T/vx5lyJe+53qb/4SVTnGJzrfkDku7cTiJtI3YpX8OasQEILFdt0omHaPOE6jNUK5rxVfbJBR2cwURQlAHwTWA0cAF5VFGWfEOJRIcSV4WH/BziBfwshdgohVg6Tua1UNHhpDgRJ1xU4BgzZffLeSI3OGZqDGq34si/Rgws6o4ZBLyIUQjiB14FvK4pS39/9DUVByqnMHhvL81uKON6sMjW1/3nQg51M/+qufXyWVwXAJZMSWT4piSfWHuXKmWk4Le0/8nuX5nBeThzTUl08tuoAz28potLj51fXTMFqMvTZhoIqNzuL6oh2mKj2+Fmam0R0tINffXEqG/OqWDjx9MLFjt6XS6en8LeNBWwuqeeG2ek9Pn5RjYc/fHSEt3aXkuqy8ciKyXygVLKxsI7l00NV3r5AEG8gSIT19H+D02xxZUPKDzADLLwfacdzBHMuhvn3EGWJhJk3wK6nkFNndf7ZzrsVbe/fidz8SwLTv9B1LnZTLdKxtWgTrhwxxRcwcgpBQLdlMFAUZRWw6pR1D7VZ7rzj0DBxrDqUz6incAwcsvs4qnMMhsYy/PGTh9scHZ0RyaA60EIIEyHn+Z+KovyngyE9yrlry1AUpJxKRkQo7WHzkUrSnb0vQBtIW05F0zTuf3Mf6dE27jt/LJqmsbe0jlSXlcpGH1dNTmRGahQLbplBoMlHbVP7SK7LZSfHZcXT2Mx9izKJtxn5w/p8iqs9PHjhOBKcZqJsJmRJoqSuiac2FvKV2WkU1Hj45Gg1DywZe5qj/dTGAv72Wah6uyWWMCXBQW2th1iTzOUivsPz7+h9SbYZGBfn4NF3DrDuYAU7iuuId1q4amoS10xLbh13vL6Zx9flU1LXRLXHz4lGL0aDzC2zUrlldhqBJh+Lx8byxo4Sbp2ZwuoDFbywtRg1qPHKV2YSYzd3a0sr424I/QA0AU0esOZgn30fvqxlBLr4bM1zfkDUf79M8+bnaM69sdNxzo9/gO3AKzRNuhlW/J7auuZOxw4lI6UQBEavLZ0VpOj0nooGL7/86AhOi4Hx8aO3yHukIXsq8I69nKZp/4Pqyhpuc3R0RiSDqcIhAU8DBxRF+W0nw1YC3xRCvEyoeLBuJOU/t5AUYSHVZeWfW4tZPjHhtCjucLK5oIYNedXYTDJ3LcykxuOnvjnA3edmcsXkJEyGnmfpSJLELbPTSImy8tC7Cje/EBJEMRkkrpmWzLaiWg5Xull35ARur4pG6GnbN87NxGyUcZiNbDxWzVOfFXCRiGfF5CR++dFh4hzmdrnXvUGSJP5y3VR+uvoQm47VMD8zhtL6Zn7x4REiLEaWTUhga2EtP151kGa/yvSUKLLjHIyJsHDV1DEkRJxs7XvH/AzeVyq54+WdFNc2Mystip0l9fxhfT6PXNLPvDtJwjPnO90O82UswR8/Fdv2P9M84UsgG0DTQJKQfA2Y81YTSJyOVfkPamQ6tn0vEjTJsOAxXf5OR+cUfvr+IWo8fv70xSlE2fof3DgbMeetRm6qxJ84AzUuF1QfcnMNQUcCavTY4TZPR2fEMpie4ELgFmCPEGJneN0PgXQARVH+QuhR4aXAEUIydrcNoj19RpIkHrlEcOcru/jFh4dPK34bLjRN42+fFWIzyTT5g6w5fKI1GiwSI3rlPLdlyfh4xic4OVjeSJXbx4GKRl7eXoIswQ8uyuHl7SXMSrORHGXlpW0lvLnnOJFWI9dMG8Mr20vJjrPz0MXjsZoM/Pu22QSC/cupjrKZ+PUXJrVqQvvVIF9/dTcPvavw188KKKxpIjnKyp++OIVxcR3nH0NIUeULU5J4fVcZl09K5KGLx/Pkp8d4ZnMRnxfUcPGEBL61OLtftnaLJOGZeTdR792J5eg7+BNnEP3aFQRtsUjeOgzu42iyCbQgtStexrr/ZRzb/oAjaMK96CeDa5uOziii0RtgS2EtN89KZfKYyO4n6JyGoeYIUe/eDkAgehw1N65F9lQCELQndDVVR+esZzBVODZw8gl+Z2M04O7BsmEgmZYSxc2z03ju8yLuPS+7XWRzqKly+3DZTOwtq2dPWT3fWzqOF7cU8e7+CnKTnBgkunQke0Kqy0aq62RO4WW5CTT5g5w3NparpiQhSRKBoIbTYsRqlPlAqeSZzUWckxLJY5dNbHXkDbKEQR6YopCWhiomg8z/rcjlhS3FFNY0ccWkRK6fkdKjnO17zstianIkyyYkIEkSX52bjgRsKazjX9tLuH5GCokRFpr9Kp8X1FDt8XPe2Fjs5r7ng5+KL3s5gRiBc+33CTqTQfUStLiQLFG4534P+/Yn8KfMJxiZjmfud7H5KrDtexH3gh+CYfj+7nR0RhKbjtWgBjUWZccMtymjFlNYmrNp4pewHXgFyV2B7C4HIOhIHE7TdHRGPCMnF2EUcGluAs99XsT6o1V8cXpy9xMGgWa/yjX/2MKXzknG7VMxGyQuzU2gyu3jmc2FFNU2kR3nGPBuXLPTo1uXWxxZoyxxx/yQZuf1M1LYXlTHzHQXxgFymLsi2m7m3j5Eix1mI5fmnvxisJoMfP3cLK6sa+Lqp7fw752lTE+J5Ddr8yiuCYn/T0x08rurJhPbxzSU05Bk6i5/gaj/3oqx6iD1y57Em3NF62bvxOtCaR0AkkRQXI5xz8uYynfiTx5RMuk6OkNCQA2yubCW+ZnRyOHrz4a8KqKsxrMv+ly+D0vBTrzjr+r3rkylm1AdiTRPuhnbgVcwl25GCxc36w60jk7X6A50L8iKsZMebWPdkeFzoA9VunH7VF7bVYbZILMgKwaH2chNM1NZe+QER094uHzS0F/4TAaZuZnR3Q8coaRE2Vg8Lo7nPi/iOSA7zsGvV+TiDQR5dPUhLvvbZqaOiWBcvJOZaVEsyIrBapT50yf5JEZYuO6c3qkvBiOSqbn6LYzVBwkkzTx9QBspJy19PhoSptJNugOtc1aQV+Xmk20l3DIjGVmS+OBQJQ+tUnjo4vEkRlj47/5yPs2rZl5m9IA94RoVaBrGt79BxIlDeHNW9K8uQtMwlWzCnzKPQPwUNKMdU+kmArGhehA9hUNHp2t0B7oXSJLE4rGxvLS9hA15VWTHOvrVYrov7D8eauhS3xwAaO0gGGE18vjVU3hw5X7OHxc7pDadKXxtfgZ+NciSnDium5dJU2NI+SI7zsG7+yvYUVzLqv3l/HtnKdE2E4vGxrBybzkOs4ErJyf1WvYvYLRz3D6J0wX9TsEWjRo7AVPJJpj1rb6dnI7OKOJYlYc/rT3K+Fgb8zNj2FUSUkB9YsMxvAEVXyCIT9W4SAxeB9WRiPnYh0jlewCQPZX9ihIb6vIxeMrxJM8H2Yh/zGxMpZsIWl1okkzQFjdQZuvonJHoZf295PycONSgxn1v7OOe1/eg9rNArrccKG8gxm5iQoITs0Fi0diT+X+JERaevekcFo/TL3x9YVy8g99dNZkrJie1S4EZF+fgnvOy+MeN5/Dh3Qv487VTiHGYWLm3nHFxDtw+lXVHqtrty68GeWZzIb/48DAfHapE0zRe2FLEXa/uotYTavm9ck8ZX/j75+wu7V4e3Zc8D9PxLaD2raGMjs5oYtHYWOKdFl7ZXgrArpJ6kiIsVLl9SEi8etss1t+78My+1gWacX78AIaao6HXmoZ96+/RwlFnuaG417uUmmtwfvwAkrceU2mor5k/ZV7od/I8jNUKxqqDIedZHri6Dx2dMxE9At1LpiZH8vjVkzlU0cgTG47x8eETQxoFOXC8kdykCL5xbiZl9V4cZv0jHEqMssTs9GievfEcPj58gsXjYrnhuW38a3sJO0vqiLGbibAaeWvPcY6ccOMwG3h9Vxk/WS74+8ZCPH6Vu1/bzV+/NI3PC2tRNXjs/UO8ePMMAF7bVUp6tI3Z6dFYjDKaprF633HMgQksCzTj+s/VuOc9iD9t0TC/Ezo6g4fJIHPD7DT+sOYI+483cPSEmzsWZBDnMDMuzkFK1JnfNMVUuhnbgZeRm2uov/RpTIVrMVXsQp19J4Ytf8XQUNxx+lcXmAs+wnbgZXyZSzFW7iVojkR1haTqvFnLcGz+Jeb89wnETRqMU9LROaPQI9B9YEFWDLfMTiMj2sYzmwuHpMX38fpmCmuaOFbtITcxgpx4J+eN1VM1hgurycCluYk4zEYuy01k3/EG3t5XzlMbC/jNmqP41SC/XjGJD78xn/RoG4++p4Sc53MzOVTp5t0DFewqqSfNZSW/ysM/txXz5p7j/G5tHve9sY/739iLLxDk+28f4Jsv7+SuHSmsir8DuekEER8/AMFQCo+pcC3mvHf7dA5yXQGR792JsXLPQL41OjoDwvWz07AYZb7/9n40YFpyJFdNHcOU5LOjaLAlQmzJX42hch+Orb9HdSYTXPQ9oG8RaGP1kdbfhppDqDE5rfUWaqzAm3UxEhpBh57/rKPTHXr4so8YZInb56fz0CqFa5/Zwv+7IheR4OTl7SVcMSmxQ1H/vCo3WTE9b/f7wpYi1h6p4iIRz5835ONTNTQgN0nvZDaS+PKcNMYnOJmbEY1PDeL2BdpFyO49L5sH3trH/MxovjwnjXf2lfPy9hJOuH08uHQcn+VX8/yWIiKtJiYlRbAkJ44/fpLPV/+1E6Wike9cmENZjYdvbLuAf8yawJK932Hn+/8gtXYzE6tWo0kGaq57N9QEoadoGhFrv4+5+BPMBR9Rf/Ff8aUtwvnJw6gRKQSSZqAZLL2OcOnoDBTxERa+vjCT36/LQ5Zg0pgRfN3TNCLfu4PmCdfhy1rW83mql6h3vox7zgMExsxqt8lcuolA9HhkdxmuN65B9jfScN7PsNqiCVpcGOrbONBBlchVt9E8+cv446cQ9d4d1F/0Jwx1x4hY813QNBou+CWGmsMAGGoOYaw+jDezfWd2z6xvYclfrRcQ6uj0AN2B7gfLJybispn43/8e5IUtRVw1ZQyPr8ujyu07rSHHB0olP3znAPcsyuLeZR13vVt9oIIIq5EFWTHkVbl5YsMxAHaX1jM+3kGUzcS+sgYm6Q70iMJqMnBBTigX044B1yk3T+eNjeGBC8YyNzMaSZK4ICeOZz8vAmBaSiRTkyO56YXtNHpV7lmUxZLxcaw/WsWu0npunZ3GXYvHUlPjZkthLY8eySBXTuKio4+iahJvR97AZYH30VZ/l9+O+T1zs2JZmHW6Lq75yDtgsODLWIKp7HPMheswF3+Ce/b9WI68g+OznyLN/g62fS+0m1e3/Gl82RcP0juno9M1189I4aNDlciSNKLT1SR/I5a899DMEb1yoA21xzAXbyCQMLW9Ax1owli+k6ZpX8WfNBPzsY/QLFE0516PFVAj09pFoA1VB7EUfEzQkYjkd2M6vg1z8QYM1QqypxLNYMZ6+C0M1YcAMJVtQW46gRqd086eQMJUGhb/gkDi9H69Hzo6ZwMj94o0SpifGcPC7Bg+y68hJazI8eaeMu6Yn4HdbKC8wYs3EOT3a0OFIP/YXMjcnDi25VWxKDuWtGgbmqbxzOYinvz0GBajzJ+vncof1uXhMBt44eYZ7CypY/G4WOwmA95AsNdqDzrDiyRJfGnGSZm7JeNDDrTTYiA71oFBlrgsN4G9ZQ2cnxOHLEk8dtkE1h+t4pppya37+J956Tz49gF+Y7iSn9pe5q30H/Pg/lSKE5P5eu1vmF31vzx56DYWn1dHxM4/E4idiGfWt1BlC5Hv342kqQQtUcjeOgB8KfND2yPTiPzoPhyfPoIakUrtilcwNBTjXPsg9i2/DTkE0lkkFaYzYjDIEn++diqqNrTF2r1FdlcAtDqo7fB7Qrru5tObWxnCTrCh+nC79abyHUhBH/7k+fgyl+LLXt5uezAi5WRxIWAuDTVEMVYfJmhPbN2nsfowgRhBMCIFU9EnyJ4KNMnQelw1pr0DDdA8+eaenraOzlmN7kAPALPSXKzaX8HKveXE2E1Ue/y8vfc4czOiueXF7TSHc6T/9+Lx/Oz9Q9z8jy0A/G5tHheJeCxGmXf2lXPh+Hi2F9dy+792Ikvw8CWC5ChrO6k83Xke/UxIcJLqspIVY2/VsP3xxYJgUGttQpMUaT1NW/r8nDjOSYnEHXUd9Rf/iHkBiC3Ywi/LZzA+4w6uLn+Kq9UNsAaKzdkkNqwhIv9DdqjZxMhGamfej690N47JV+JPnhv6opUkvOMuJ7jhEQzuctxzv0cwKoNgVAaeWd8m8qNvY9v+BN4J1+qNFXSGhdFwzZM9oe59xurDIWc5fMNpPrqKqPe+BkDDokdpnvrV9vPCjqzxFMfbVPo5GhL+U9I6WlAjUjEXrms9lql0EwCGmsMYw/nLxppDGGoO4U9ZQCB+KpZwrYQvdRHm4k8ACESP///t3Xl8VNXZwPHfvbMlk20SSFhNCCEcWWQR2V1QCyoWQeuCSlWo9bUudaNatdXq+7b6trWv+Lq2isUNd8W3YsGF4lI2WVxYjkAEIUAgkH2ZzHLfP+4kJJCg0cxMSJ7v58OHmTt37n3mzL3PnJx77jk/+LML0VlJBboNnJDtA6C0JsCVY7L5dEcpD31YQGayB4/T5NqTcklPdHHGgCxqAyEqQxYTctNZtHEvz68uxB8M87Mx2Vw1LofVO0qZu2IHvxjfhyGd5GaZzsYwDB67YEiTofKcpgHfMiGEaRg8ftHQhpnYvG6478cD2FFaw3GDTqZkxwReWLyEdRWpvFd7PMe4ynnOvIdRxnoWJp7LMztPY/m24Tw7bjjHJjXqBuRMpHbgJSR+PpfaARc1LPb3n0bgi6dJXn4/3tX/S+kFC8F3XNsWhhAdQH0LtBGsxqzcRTjF/uPXWbweC4Ng1hCSPn2I2gEXg+vg/RH1LcFm+TcQrAGn/ZrjgCaUloPlSWt2f+GU3hjBGozaEqyEdDVnYYYAABsJSURBVFy7VmA5PJj+Mnu8eMC59zPM2hJq0vOp6zmm4b3+vLNx7/wIy+klnBKfCcGE6AhkFI420CP1YCvxyBwff5o6iME9Uiksq+XOSf2ZfnwvzhhgtwpcOLwXN/+oP327JPGLE3N5deYJ/P2SYVw9vg+mYQ+R9tgFQ6Ty3MF1T00g3dv6qcHNQ7pSDO+dxjmDu2MYBsHsU8iddAOV2RN59IJhhLzd+Q/jN3zsm8ad+89g+bYSAJ771P7R3lvh5/73NvNxwX4qRs3mwCVLmrYym05Kf/IWJef/H5guUt67AeOrd3AUb/j+H1yIDsSoq8Coq8CsKmpY1rgbh1lVRNibSeX4uzFriknc8HyT99e3QBtYOEoKGpY7D3x1WP/kxkIpvQFwb/8A99a3MWtL8OedbW+z9gCWMwGz1j7fQ+n5hLocS9iThmWY+CN9tIPp/X7YTIZCdHLSAt1GRuf4WLRxH4O7p+J2mjxy/nHsLK2lT5cjj7rRPTWB7qmxnc1QdFwjs9MZmW1Pqf7sjOOpDQ4lwTmFwJMryE9N4IRjfLy8tpDLRh7Dnz7YwrrCcl77bDf5mUncNKEvIw/9u810EOw2nIoJ95O26Gp45VJ87hRKpr/X0MomOg6l1JnAHMABPKm1vv+Q108GHgSGANO11q/GPsr2I2XRNeBwE0rrg2U4MKwQzpLNBHJOBcCs3kvYm0Ww5yjqeo4h8fOnqRl6ZcP7HRU7CXm74aguwlmymVDmIAgHcZQWUNfn9Bb3Wz92c+r7NwJgYVBz3OUkfPU6AHXZpzZ02Qhl5IPpINB7PI7SAqykLEKp2QQzB0elTIToLKQC3UauPTGXC4f3wh25LO90mN9aeRYimlISnKRETvF5lx5PWoKT2mCYBV/s4dJn1wBw1xn9cToMnvhkO9e/9iULrhxFtxRPwzaCYYtQ2CKcdzbW1JdITgDz9VmkvH8TZVNflBasDkQp5QAeASYCO4FVSqm3tNaNLzl8A1wBzI59hO2Ps3QrBP1YzkTCKb0wAlWHtEDvbRhTua7vWSR/fDdmRWHDH59mRSGBY07C/OrNg0PMlW3HCAeO2D85lJFPyU8WYPjtWUytxC4EM49ruEnYnzcZT8E7WA4PodRsACom/DdGyA9A6bRXsdzJbV8gQnQiUoFuI2mJrmbHfhaiPchOt/tWpgEvXTGCVz/bTZLbwZTB3QF7hs1zn1zFgi92c9W4PvxrczF/+ddW9pT7sYDuKR5enTWWhdtLCaddzYzCB0j+16+pnHB/i5Voz+YFJGyYT9mU58CUVHMUGAVs0VoXACilXgSmAg0VaK31tshr0Z89qr2zLMyqIoyQn3DJFsLeLCzTibPk4IgaZvXehpbe+n7Irl3LCGYMAFcijuq91KblEkrr03AjoaPE/r+5ETIaa26M9lB6PuaeT6nLOR3LdBHy9W0496yEdOrHMpG+z0L8cPKrJkQn0z01getOym2yrFdaImNz03nziz0APLnsG1RWMmcP7EaFP8hLa3exZHMxDyzZSmnNCEqc07h+wwuEk7pRPeqWhu0EwxbLtx1gaM800ja9gnvnx7i/XkRdpH+maNd6ATsaPd8JjI5TLO2eUVfe0KLrLF5PXd5ZhJJ7kvjlsxhVe7ESu2DWFDfcV1DfD9lTsIjkpb/B8tg38oZSexPMHIx7x4cQqLZH8gCCvn6tjinQcxSE/FieVALdRxDKaH7OASHEDxe1CrRSai7wY2Cv1vqwzlZKqQnAAuDryKLXtdb3RiseIcSRnTekJ7MXrOdvy77hzAFZ3DkxnwSXg2DY4l29j78s2UppTYAHpg3imRU/I/dAMZNXP4y//7l2Sxfw2rpd/HnJVpKcYda6VwLg/ewp3IX/xrnvSwI9RlI16paG0QZEx+dwGPh8re/O5nCY3+t90dBsLPu+aXhoYOFM74U58ir4fC7pG58iPPpaDCtMQpfeuOvfmz0Wz2a7bzKBSgC8PfpBr2Mx5y0gY+uLGJUFWKm98WVlfvdY6p1xD1hhfA4XXLYAp2HiM6M3DGB7+Y7aSxwgsbSkvcTSlnFEswX678DDwDNHWOcjrfWPoxiDEOI7OrFvBleNy2Foz1RG5aQ3LHeaBpOOzeLFNYV0T01gfG4Gfbt4uf6ZGZxirKb21StYYwwiZfy1vPZZLfmZSYx1f417fzWF3oH02r0S1+6VBLKG4l37OBgmVWPvIBS2GsbBFu1CIXBMo+e9I8t+kFDIorS0utXv8/m8zb7PLP0a1951+Puf+0ND+16xOEq24Nz3JeHErvgarVPj7EKN2YOU/Gl4Vs+lImssPqDS9FEXeW9i5kiSN/8Tf58f4SzegKNyF2V0JZzcm7TeJ+L89xws00Wwy7GUtVBmLZVLU4Ef/qG/g+8WS+eJAySWlrSXWL5PHJmZzc/+HLU7gLTWHwIHorV9IUTbcpgGPx+b06TyXO+syDCM5w7vicM06O1L5L6LTuGp5Gswa0uYWPtPst+bSV7pRzyY+iKze60H4LKSmXwcGsRjvlspveBtKvpfSOLaJ3jklTc496mVVPqDMf2M4ohWAflKqVyllBuYDrwV55gOk7T6IVLfvR6jJg4/L5ZFyvs3kfLeDQ1jOIfd9o9r2GufI9XDr8YI1pD45bNNlgPU5U4kmJZL1ZjbqBp7B4GsoYST7fsQqkb/CsJBTH85/pyWR+AQQrQP8e4DPVYp9RmwC5ittV4f53iEEM0Y2D2FB88dzKmDu1NXXQfAsd1SOPbymwhbN7Jl41LyP7icJ90P2G2WheD39ePmE89kyc6xPLX8Gz5fuIlPNk5kseddZhXdy5v+/2L+6kJ+Pi4nvh9OAKC1DiqlrgMWYQ9jN1drvV4pdS/wqdb6LaXUSOANIB2YopS6R2s9KJZxunatsP/fvZK6vmfGcte4dn6Eq2htJA57wpJAz7F4ti1uGG0j1GUA4cSuuLcvAWiYWhsg5OtLyYyPGtbz95/W8Fqw+wj2Xyk/gUIcLeJZgV4D5GitK5VSk4E3gSPfdkwH7k8XJxJL8ySWw519vBeHwyTkPjxtZIybTIHxBI6KQvr0G4zjtctxDjibSUN7MTI/k/lrCnln416mDOnH28Z/M3PztbyWNodb11zOBaOz6dMlqcn2dpRU88dFmltGJdOnqxdSDx9zur2UC7SvWH4IrfVCYOEhy+5q9HgVdteOmDKrivCufICaIbNwlNt9j127lmPWlmAZDvwDLsS74s/UZU8g2ML01+6vF+MoLaBm+NVH2MefqRr3WyxPMxNZWRZJqx7EciZgBGtx7fwYy+kl2G04nm2LD7Y0GwaBnqPxbH0bgHBS832ZhRBHt7hVoLXW5Y0eL1RKPaqU6qq1Lj7S+9q6P108SCzNk1iad7TEkjHAbg0sAbhiDZguKK3GAG47vR8l1QEuGdELw8inMmcOOUtu5TXjNh55+N/83X0ptSHo19XLLxxv0mvfUv5dcyv3br8LI8lDycVLcG//AEwnYU8q7m+W4j7xF6wstHhxTSE3n5pHsid+7QGt+Y5a6k8nWubZvIDEDS/g2rMagJC3G56v38X88lnCCekEjjmJpE8fxFmymfLmKtDBGlKW3IbhL6Vm8OVNptOu5131PyRumE+g51j86rzDXnftWoZr90oqx95J0rI/4KjaQzCtD/68yTgObCKUntewbl3PMXi2vk3Y4wOH57BtCSGOfnH7xVFKdQeKtNaWUmoUdn/s/fGKRwjRhlxNW2MnD+zW5Lk//xzqsk/B/Nc9XLvlZYYlB3m9x2ySCj9kUuWTAPwj5Q9kBXZDGdS8/nMy977XZBv7y/dw947z8JRu4e9J7oND84VDeD++h7A3i9oTroveZxQx4yq0u0s4D2jCnjRqB04n6dM5ADiq9pCw6RV7vV0rwLLgkCnvEzbMx6zZZ69TtIZA7/FNXjcrdpGw8aXINpY3W4H2rppDyNuNmiEzSVz/HI7y7YS93Qil51Ex6ZEm6wZ62WM+1w9hJ4ToeKI5jN18YALQVSm1E7gbcAForR8Hzgd+oZQKAjXY08JaLWxOCNHBWJ40QpMeoDqlC+PXPsbwvGwSAvMJZCiCvn5kF7zN6nB/wGLE3vfY4lLsGHgdH2/8mrzqdVyoX+b31jpGezYxZ+35XFc0i8LiUv7gmcfJ1YsIYRLIO4tQeh7PrNzB6p2lPDB1EE6HzJ54VLHCuHavIJA1DNfedQR6jCTQaxx8OodAt+G4itaSuO6vAJg1xThKthDKyCd56Z32VQvsCU0CWUNx7vsC167lh1WgvWsfBSwCXQfj2rUC95Z/kLzsPizDoPLUP2JUJOMq/ITK8XeDM4FgRn+7Ap2UdWi0AIQyFGGPTyrQQnRgUatAa60v/pbXH8Ye5k4I0VkZBlVjbsO1ZzXetY8R9PWlfPJccCZgBqv5LHEGe8pryKh6liuKZ7BzRRd8id255/SzCX20jNHhTVRljuCGfa9yZtEqupvlpFWX8KZ7ChP9iwl98Ds2qet5+KMaLAzmrynkuB6p+Lwu+mTYreQLNxSxqaiSmyb0xTBkWL32xrF/E6a/jMrjrqA2UEUwczDBrKFUnXADtYMuJf2lMzFrDxD05eEs3Ypr13KMkJ/EL+dR13M04ZRjwDCpHnolKR/c0nDzXz2zqoiEDfOpVecT8uWRvOz3JH90FzgTMeuqSFp2H2aSj3BiF2oGzQAiswRue7fJCBtNGCaVp9xHODEj2sUjhIiTeI/CIYTo7Ewn5Wc8SsL6F6gZMgsrwR5dt3zKs0xpWGkaD5fVUBMI0zstgQSXA0e3pymrrKEu5zQqls0he/8aTKeH0iGzGJ41hnl/m801e14hc88SHkw6gyUZFzJ6+ZX8b/BcClOP55WZI9lYVMG9i74iFLYY1juN0/K7xqsUxKEs+4LkwdEuxhBOPXj/YvXoX0WWj8ZT8A61Ay4k8bO5uAr/jXvHUsLuVMonP93khsBAzzEkfvkMhPwNfZMT1z4O4SDVI67DrC0BwFG9l7Kz52FW7iZl6a8BqBx7R0Pf6WB6f4AWW6AB/PlTWnxNCHH0kwq0ECLuwkndqR518xHX6ZXW9MYvK39SwwQVjvE3U9noNS9QOeIGLl7Wn1nJK5lat4gzazUecxsjPAVML7ud1z/vzbOrdpCV7CbB5eChpQWcmJuB23mwi0cobBG2LFzS7SOm3AX/xLnqj3D+O7h3LSeU0rtJ5bmxQM8xeAreIdBzLM7iDSRsXgBA1Qk3HjaaRqDnaLyf/Q1X4XIC2adgVBeTuP5Z/P3PJZzWh3ByL8KuJEK+POpyToNwHd7VczBDtdQOvqxhO6GM+gp0jyiVgBCivZMKtBCiQ7psVA7j+v6U/r7LCM0/DU/FNipO+k8SP/sbz1h/5IIP3JQ4+vDERUMprw1w/Wtf8vrnu5l+/MEh8+YsLeD9r/bx9CXDyUqR0RRiySj+ClfROly7VlCXPaHF9WoGXkLYm0mw23CqRt9KsOsgcLipGXjJYevWZZ9CyJuFd82jlGWfgnfdExD0Uz3iensFh4vyyXMJJ/ewb0R0eCifPJcUr4nlTm7YTjDzOMonPYY/d2Jbf2whxFFCKtBCiA7JYRqoLLvSUzZ5Ls79G/AfewF1fU4n4aVpvGj9nqL8S+i96wscZdt4PnU7D674CT1ST2Xplv3cfGoeb28oorw2yG3/t4HHLxzK57vKeHt9EbdP7I/HKa3S0RLoORqAhPXPYdbsJ9BzTMsruxLx558DQDgth5rjr2l5XWciNcOvJvmTe/FsfJnEL+bhzz+nyRB0h95gGMw8DsvnhcbDFBqGdNEQopOTCrQQosMLZQ4ilGlPmBdOzcZ//it4P/wtastfMbZYhBMyGBWu46HQBj5Z+AazjU08VHE/Pwv+g8m+rVyxexbXvGKwtbiKqroQo3LSDxuaT7QdKyEdK2sgnkh3jLojVaBbqWbQDLxrHiX1g5uxMKge8cs227YQovOQCrQQotMJpedRMfUFKquLweHC8qRh7ttA2ivTmGosI4CT2UWz6eE8gOU3eT/lbp7fN45Ep0G/xJ2EPk7E0f0+ezQGERXhY8bh2LuBkLcb4bQ+bbdhl5eS897AWbKFcFIWoS6q7bYthOg05BqkEKLTsrxdsTxpAIQzB1Iz4z3KL/uEL074Ez2MA2z2DKbkosWYPYYy0/Uu0833yfYGSPAXs3PfESdNFT+QlTMOiExK0sbDC4Z9udTlTiSYNbRNtyuE6DykBVoIISLCqdkA5I6aygs1SfQ/dii+LlmUTXkegrVgmFTUGcx8+lN+ZeUhYzBEj5V9IpbTS13O6fEORQghDiMVaCGEOIRhGEyccMgIC84EAHyJ8I+rRstNhNGW1JXimWsPmxZeCCHaA6lACyFEK0nlOUbcSfGOQAghmiW/AkIIIYQQQrSCVKCFEEIIIYRoBalACyGEEEII0QpSgRZCCCGEEKIVpAIthBBCCCFEKxiWZcU7htbaB2yPdxBCCPE95QCZ8Q4ihiRnCyGOZs3m7KOxAi2EEEIIIUTcSBcOIYQQQgghWkEq0EIIIYQQQrSCVKCFEEIIIYRoBalACyGEEEII0QpSgRZCCCGEEKIVnPEOIBaUUmcCcwAH8KTW+v4Y7fcY4BmgG2ABf9Vaz1FK/Q74OfbwTgB3aK0XxiCebUAFEAKCWusTlFIZwEtAH2AbcKHWuiTKcajIPuv1Be4CfMSgXJRSc4EfA3u11oMjy5otB6WUgX3sTAaqgSu01muiHMufgClAHbAVmKm1LlVK9QE2Ajry9uVa66ujHMvvaOE7UUrdDvwM+3j6pdZ6UZRjeQlQkVV8QKnWelgMyqWl8zgux0xnIDm7IZ5tSM6WnP3d4/gdkq9jlq87fAu0UsoBPAKcBQwELlZKDYzR7oPALVrrgcAY4NpG+/4frfWwyL+oJ+JGTo3s84TI818D72ut84H3I8+jStuGaa2HASOwD9o3Ii/Holz+Dpx5yLKWyuEsID/y7yrgsRjE8i4wWGs9BPgKuL3Ra1sblU+bJZ0jxALNfCeR43g6MCjynkcj51rUYtFaX9TouHkNeL3Ry9Esl5bO43gdMx2a5OzDSM6WnP1d4wDJ1zHL1x2+Ag2MArZorQu01nXAi8DUWOxYa727/i8ZrXUF9l9dvWKx71aYCsyLPJ4HTIvx/k/HPpliNtGC1vpD4MAhi1sqh6nAM1prS2u9HPAppXpEMxat9WKtdTDydDnQu63219pYjmAq8KLW2q+1/hrYgn2uRT2WSIvBhcD8ttrft8TS0nkcl2OmE5CcfWSSs22dOmdLvm4xlpjl685Qge4F7Gj0fCdxSIiRyxbDgRWRRdcppT5XSs1VSqXHKAwLWKyUWq2UuiqyrJvWenfk8R7syx6xNJ2mJ1Y8ygVaLod4Hz+zgHcaPc9VSq1VSi1VSp0Uoxia+07iWS4nAUVa682NlsWkXA45j9vrMXO0axflJzm7RZKzjyzeOVvydUS083VnqEDHnVIqGfsSxo1a63LsSwR5wDBgN/BAjEI5UWt9PPYli2uVUic3flFrbWEn7JhQSrmBc4BXIoviVS5NxLocWqKUuhP7ctTzkUW7gWyt9XDgZuAFpVRqlMNoF9/JIS6m6Q94TMqlmfO4QXs5ZkTbkJzdPMnZR9YOcna7+D4O0WHzdWeoQBcCxzR63juyLCaUUi7sL/F5rfXrAFrrIq11SGsdBv5GG15KORKtdWHk/73Y/ddGAUX1lysi/++NRSwRZwFrtNZFkbjiUi4RLZVDXI4fpdQV2DdlXBo52Ylcftsfebwa+2aV/tGM4wjfSbzKxQmcR6MbmmJRLs2dx7SzY6YDkZwdITn7iNrV+dcecrbk64b9xiRfd4YK9CogXymVG/nreTrwVix2HOn78xSwUWv9l0bLG/evORf4MgaxJCmlUuofA5Mi+30LuDyy2uXAgmjH0kiTv0zjUS6NtFQObwGXKaUMpdQYoKzRZaCoiIxAcCtwjta6utHyzPobP5RSfbFveiiIciwtfSdvAdOVUh6lVG4klpXRjCXiR8AmrfXORjFGtVxaOo9pR8dMByM5G8nZ30G7Of/aS86WfB3bfN3hh7HTWgeVUtcBi7CHRJqrtV4fo92PB34KfKGUWhdZdgf2XeXDsC8hbAP+IwaxdAPeUEqB/b2/oLX+p1JqFfCyUupnwHbszv5RF/lBmEjTz/7HWJSLUmo+MAHoqpTaCdwN3E/z5bAQe3ibLdh3ns+MQSy3Ax7g3cj3VT/Mz8nAvUqpABAGrtZaf9ebSL5vLBOa+0601uuVUi8DG7AvWV6rtQ5FMxat9VMc3v8SolwutHwex+WY6egkZzeQnH1w35Kzv1sckq9jmK8Ny4p7tyEhhBBCCCGOGp2hC4cQQgghhBBtRirQQgghhBBCtIJUoIUQQgghhGgFqUALIYQQQgjRClKBFkIIIYQQohWkAi3Ed6CUmqCU+ke84xBCCPHtJGeLaJMKtBBCCCGEEK0g40CLDkUpNQP4JeAGVgDXAGXY05pOAvYA07XW+yIDzj8OeLGnE52ltS5RSvWLLM8EQsAF2FN9/g4oBgYDq4EZWmtLKXU/cA72APWLtdazY/RxhRDiqCY5WxytpAVadBhKqQHARcB4rfUw7ER6KZAEfKq1HgQsxZ6xCeAZ4Dat9RDgi0bLnwce0VoPBcYB9dN6DgduBAYCfYHxSqku2FOmDops57+i+ymFEKJjkJwtjmZSgRYdyenACGBVZArP07GTZhh4KbLOc8CJSqk0wKe1XhpZPg84WSmVAvTSWr8BoLWu1VpXR9ZZqbXeqbUOA+uAPtgtJbXAU0qp87CnAhVCCPHtJGeLo5Yz3gEI0YYMYJ7W+vbGC5VSvz1kve/bb8nf6HEIcGqtg0qpUdiJ/3zgOuC077l9IYToTCRni6OWtECLjuR94HylVBaAUipDKZWDfZyfH1nnEuBjrXUZUKKUOimy/KfAUq11BbBTKTUtsg2PUsrb0g6VUslAmtZ6IXATMDQaH0wIITogydniqCUt0KLD0FpvUEr9BlislDKBAHAtUAWMiry2F7vPHcDlwOORZFsAzIws/ynwhFLq3sg2LjjCblOABUqpBOzWlJvb+GMJIUSHJDlbHM1kFA7R4SmlKrXWyfGOQwghxLeTnC2OBtKFQwghhBBCiFaQFmghhBBCCCFaQVqghRBCCCGEaAWpQAshhBBCCNEKUoEWQgghhBCiFaQCLYQQQgghRCtIBVoIIYQQQohWkAq0EEIIIYQQrfD/Fg+oyNMiwQcAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 864x216 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9Dkx4uHTz-8l"
      },
      "source": [
        "## Usando o modelo"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Rul7kSlY0Aui",
        "outputId": "d681f413-7d1d-4b18-941c-7ad505afcb88",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 371
        }
      },
      "source": [
        "def predict(nome):\n",
        "  model.eval()\n",
        "\n",
        "  tns = torch.zeros( len(nome), tam_dicionario )\n",
        "  for k, letra in enumerate(nome):\n",
        "    idx = caracteres_validos.find(letra)\n",
        "    tns[k, idx] = 1\n",
        "  tns = tns.to(args['device'])\n",
        "\n",
        "  saida = model(tns)\n",
        "  topv, topi = saida.data.topk(3, 1, True)\n",
        "\n",
        "  print(nome)\n",
        "  for value, index in zip(topv[0], topi[0]):\n",
        "    print('(%.2f) %s' % (value, categorias[index]))\n",
        "  print('\\n')\n",
        "  \n",
        "predict('Mota')\n",
        "predict('Kamikaze')\n",
        "predict('Suarez')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mota\n",
            "(-1.63) Japanese\n",
            "(-1.81) Spanish\n",
            "(-1.87) Portuguese\n",
            "\n",
            "\n",
            "Kamikaze\n",
            "(-0.73) Japanese\n",
            "(-1.71) Russian\n",
            "(-2.61) Greek\n",
            "\n",
            "\n",
            "Suarez\n",
            "(-0.20) Spanish\n",
            "(-2.95) Portuguese\n",
            "(-3.49) Polish\n",
            "\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:23: UserWarning: Implicit dimension choice for log_softmax has been deprecated. Change the call to include dim=X as an argument.\n"
          ],
          "name": "stderr"
        }
      ]
    }
  ]
}