Modelos de Predição/Otimização de Hiperparâmetros/Otimização_de_hiperparâmetros.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "aOEjYcJ6VAS6"
   },
   "source": [
    "# Otimização de hiperparâmetros\n",
    "![head](https://drive.google.com/uc?id=1E0NK-xpj28zAccdWf5-4bLRoa30jlmZ8)\n",
    "\n",
    "Fala galera, como vocês estão? Hoje, vamos trazer um assunto um pouco diferente do que foi falado até agora sobre modelagem, mas que é essencial para que você possa obter o melhor desempenho do seu modelo - a otimização de hiperparâmetros.<br>\n",
    "\n",
    "O que são hiperparâmetros e como otimizá-los? Por que se preocupar com isso? Em que etapa de um projeto de Ciência de Dados isso entra?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "2UzwQCNRkAYr"
   },
   "source": [
    "## Motivação\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "aF8zp1P83UC-"
   },
   "source": [
    "<figure>\n",
    "<center>\n",
    "<img src= 'https://drive.google.com/uc?id=1EmBb4OkO3WsQMtorvA9TqZLocXI8Y3Ll' />\n",
    "</center>\n",
    "</figure>\n",
    "\n",
    "\n",
    "Digamos que você esteja fazendo um projeto ponta a ponta de Ciência de Dados. Você passou muito tempo coletando e limpando os dados, depois analisou eles, tirou insights importantes e finalmente começou a modelar com Machine Learning.\n",
    "\n",
    "Olhando para alguma métrica, como acurácia ou R², você viu que seu modelo está razoável - dá para melhorar. Assim, você decide preparar ainda mais os dados para a modelagem, o que dá um up no modelo, mas ainda fica uma sensação de que dava para fazer melhor.\n",
    "\n",
    "E agora?\n",
    "\n",
    "Agora entra a otimização de hiperparâmetros.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "WZfrhrbukbUH"
   },
   "source": [
    "## Parâmetros vs. Hiperparâmetros \n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Parâmetros e hiperparâmetros são duas partes fundamentais de um modelo preditivo. Em um algoritmo machine learning, os **parâmetros são ajustados diretamente pelo processo de aprendizado** e influenciam diretamente na performance do algoritmo. Os coeficientes de uma regressão linear, os pesos de uma rede neural, as fronteiras das vizinhanças no kNN, todos esses são parâmetros que se ajustam ao treinar o modelo preditivo para um conjunto de dados.<br>\n",
    "\n",
    "Por sua vez, **hiperparâmetros são variáveis do algoritmo definidas antes do treinamento**. Eles representam características mais construtivas, como a métrica de performance de uma regressão (MSE, R²,…), o número de neurônios de uma rede neural (ou camadas), o número de vizinhos do kNN. Os hiperparâmetros são muito importantes para a performance do modelo preditivo e caso escolhidos de qualquer jeito, podem torná-lo inútil, ou muito longe do ótimo.   <br>\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "80EXohQwEyOA"
   },
   "outputs": [],
   "source": [
    "from sklearn.linear_model import Lasso\n",
    "from sklearn.datasets import load_boston\n",
    "\n",
    "lasso = Lasso(alpha = 1.0,    #-----------------------> alpha e tol: \n",
    "              tol = 0.0001)   #-----------------------> são hiperparâmetros\n",
    "\n",
    "lasso.fit(load_boston().data,load_boston().target)\n",
    "coefs = lasso.coef_  #--------------------------------> Os coeficientes são parâmetros"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "FQIEKbXkVAS8"
   },
   "source": [
    "<figure>\n",
    "<center>\n",
    "<img src= 'https://drive.google.com/uc?id=1C_HMnvOT-ZfWLlYBS0Rmgh1UOMYQEdZd' />\n",
    "</center>\n",
    "</figure>\n",
    "\n",
    "Do mesmo jeito que os parâmetros de um algoritmo são ajustados de forma a minimizar um erro ou maximizar uma performance, os hiperparâmetros também são. O problema é que testar hiperparâmetros exige que o algoritmo já esteja treinado no conjunto de dados. Assim, **para otimizar um hiperparâmetro é preciso treinar um modelo diversas vezes**, o que é muito demorado. Mas, isso não significa que não podemos ajustá-los — para isso existem diversos algoritmos que vamos falar a seguir."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "cF1x1yHFk3Bn"
   },
   "source": [
    "## Ajustando hiperparâmetros"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "PMDxNPcvk-Kl"
   },
   "source": [
    "Vamos ver os algoritmos mais populares de otimização de hiperparâmetros. Para compará-los, vamos à prática utilizando um modelo preditivo e um conjunto de dados.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "4BfhGa_TlZEZ"
   },
   "source": [
    "### O modelo preditivo"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "R3ZIWYpCjUVr"
   },
   "source": [
    " Vamos usar uma **[Árvore de Decisão](https://scikit-learn.org/stable/modules/generated/sklearn.tree.DecisionTreeClassifier.html)**, um modelo preditivo que formula comparações e desigualdades sobre os dados de entrada, para predizer uma variável de saída. Um ótimo aspecto desse modelo é que fácil de interpretá-lo, como na imagem a seguir:<br>\n",
    "\n",
    "\n",
    "<figure>\n",
    "<center>\n",
    "<img src= 'https://drive.google.com/uc?export=view&id=1mJbylXNgp2uUaVbGmzQ_pz5zs1KCE4Ge' />\n",
    "</center>\n",
    "</figure>\n",
    "Um hiperparâmetro muito importante para uma árvore de decisão é sua profundidade máxima, ou seja, o número de perguntas que podem ser feitas em sequência, pois ele ajuda muito a prevenir um problema comum deste modelo que é o sobreajuste, ou overfitting.\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "k1ruIIsYlicq"
   },
   "source": [
    "### O conjunto de dados\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "nC5UsOJzlgni"
   },
   "source": [
    "Em relação aos dados, vamos usar o [dataset de câncer de mama](https://scikit-learn.org/stable/datasets/index.html#breast-cancer-dataset), onde devemos classificar se um nódulo é maligno ou benigno com base em características dele, extraídas de uma imagem digitalizada de um tecido mamário (via PAAF - Punção Aspirativa de Agulha Fina)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "w8MAWoJ1VAS9"
   },
   "source": [
    "### Preparação"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "uvcgOb-zl85Z"
   },
   "source": [
    "Como neste caso é muito importante detectar os casos de câncer e evitar um alarme falso, precisamos prestar atenção tanto na taxa de positivos verdadeiros quando na de falsos positivos. Assim, a métrica de avaliação que vamos utilizar é o **AUC-ROC**, ou área sob a curva [ROC](https://en.wikipedia.org/wiki/Receiver_operating_characteristic), que representa a probabilidade do modelo distinguir corretamente a classe positiva da negativa. O melhor caso da AUC-ROC ocorre em 1 e o pior caso ocorre em 0.5 (por quê?). <br>\n",
    "\n",
    "$$TPR = \\frac{TP}{TP + FN} \\quad \\quad FPR = \\frac{FP}{FP + TN}$$\n",
    "\n",
    "<figure>\n",
    "<center>\n",
    "<img src= 'https://drive.google.com/uc?id=13UEKTB9S4DHGH3OUD5YmLZiJQWFZz_hB' />\n",
    "</center>\n",
    "</figure>\n",
    "\n",
    "Além disso, vamos repartir o conjunto de dados diversas vezes entre partes de  treino e partes de teste para garantir que o modelo performa bem em todo o conjunto - essa técnica é chamada de validação cruzada, ou **[cross-validation](https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.cross_val_score.html)**. Vamos usar essa estratégia para todos os algoritmos a seguir, também!\n",
    "\n",
    "<figure>\n",
    "<center>\n",
    "<img src= 'https://drive.google.com/uc?id=1zEQLIDI1-z2OxG6EzkihrDuxJ7OK1dHT' />\n",
    "</center>\n",
    "</figure>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Rva8gZ30VAS9"
   },
   "outputs": [],
   "source": [
    "# Importando o que precisamos\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.datasets import load_breast_cancer\n",
    "from sklearn.metrics import f1_score\n",
    "from sklearn.model_selection import cross_val_score"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "tdLsmmGsVATA"
   },
   "source": [
    "Vamos medir o quão bem a árvore de decisão performa sem realizar nenhum ajuste:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "aKJjqLKRVATB",
    "outputId": "31410c7e-3bff-4b24-d139-d5e010527e32"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Desempenho médio de AUC-ROC: 0.916\n"
     ]
    }
   ],
   "source": [
    "# Carregando o conjunto de dados\n",
    "entradas, saida = load_breast_cancer(return_X_y=True)\n",
    "\n",
    "# Vamos passar o modelo preditivo, os dados de entrada e saida, a funcao de desempenho e o numero de particoes para a funcao de \n",
    "# validacao cruzada\n",
    "desempenho_cross_val = cross_val_score(estimator = DecisionTreeClassifier(random_state=37),\n",
    "                                       X = entradas, \n",
    "                                       y = saida,\n",
    "                                       scoring = 'roc_auc',\n",
    "                                       cv = 3)\n",
    "\n",
    "print('Desempenho médio de AUC-ROC: {}'.format(round(desempenho_cross_val.mean(),3)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "K0nrLbXCVATE"
   },
   "source": [
    "Aparentemente já conseguimos um bom resultado mesmo sem otimizar o hiperparâmetro. Isso é razoavelmente comum, mas estamos diagnosticando casos de câncer, então cada pontinho vale! Então, vamos às técnicas de otimização de hiperparâmetros:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "WiMLpPNTVATF"
   },
   "source": [
    "### 1. Grid Search"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Z5pfWZtQBmP8"
   },
   "source": [
    "\n",
    "<figure>\n",
    "<center>\n",
    "<img src= 'https://drive.google.com/uc?id=18YG8avvFrHWU8AoXXSChil9JrPcCDCJe'width=\"600\" height=\"280\"/>\n",
    "</center>\n",
    "</figure>\n",
    "\n",
    "Possivelmente o caso mais ingênuo e mais simples. **O [Grid Search](https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html), ou busca em grade, é um algoritmo de busca que recebe uma conjunto de valores de um ou mais hiperparâmetros e testa todas as combinações dentro dessa vizinhança**. O algoritmo tabela qual foi o desempenho de cada configuração e ao final de todos os testes, fala qual é a melhor escolha. Atualmente, existem alternativas muito melhores que o Grid Search, mas vale a pena citá-lo por sua simplicidade.<br>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 84
    },
    "colab_type": "code",
    "id": "O39lxwPSVATF",
    "outputId": "d2a69883-2adf-4adc-ae1e-4cc6b81e247d"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Melhor comprimento máximo: 3\n",
      "Desempenho AUC-ROC do melhor modelo: 0.923\n",
      "Tempo para realizar a otimização:\n",
      "1 loop, best of 1: 1.15 s per loop\n"
     ]
    }
   ],
   "source": [
    "# Importando o Grid Search com cross-validation\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "\n",
    "# Vamos passar para o GridSearchCV o modelo preditivo, a vizinhança de hiperparâmetros,\n",
    "# a função de desemepnho e o número de partições do conjunto de dados:\n",
    "grid_search_cv = GridSearchCV(estimator = DecisionTreeClassifier(random_state = 37),\n",
    "                              param_grid = {'max_depth':range(1,50)}, # Testando comprimentos máximos de 1 a 50\n",
    "                              scoring='roc_auc', cv = 3)\n",
    "\n",
    "# Realizando a otimização por GridSearch para os dados de cancer de mama:\n",
    "grid_search_cv.fit(entradas,saida)\n",
    "\n",
    "#Vamos ver informações relevantes:  \n",
    "print('Melhor comprimento máximo: {}'.format(grid_search_cv.best_params_['max_depth']))\n",
    "print('Desempenho AUC-ROC do melhor modelo: {}'.format(round(grid_search_cv.best_score_,3)))\n",
    "print('Tempo para realizar a otimização:')\n",
    "%timeit -n 1 -r 1 grid_search_cv.fit(entradas,saida)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "tuCSapC5VATI"
   },
   "source": [
    "Podemos ver que o modelo selecionou uma árvore de decisão com máximo comprimento 3 e performou melhor que a árvore sem otimização! Também vimos o tempo de execução, por volta de 1.15 s, que vai ser útil para comparar com os outros algoritmos."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "A9vCYC0BVATJ"
   },
   "source": [
    "### 2. Random Search"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "92qn-wkSBopJ"
   },
   "source": [
    "Este algoritmo é muito parecido com o Grid Search, exceto por um motivo: ao invés de testar todas as combinações na vizinhança, **o [Random Search](https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.RandomizedSearchCV.html), ou busca aleatória, testa combinações aleatórias de hiperparâmetros**, conforme um número especificado de amostras a tirar. <br>\n",
    "\n",
    "Ele é uma alternativa para o Grid Search quando o conjunto de dados é muito grande, ou há um número muito grande de hiperparâmetros para otimizar. Para mostrar essas vantagens, vamos aproveitar para otimizar também outros hiperparâmetros da árvore de decisão."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 101
    },
    "colab_type": "code",
    "id": "IusaOXjWVATK",
    "outputId": "570c1180-09ed-4bdb-eb1a-c4bde3b5e50b"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Melhor comprimento máximo: 36\n",
      "Melhor minimo de amostras por folha: 15\n",
      "Desempenho AUC-ROC do melhor modelo: 0.97\n",
      "Tempo para realizar a otimização:\n",
      "1 loop, best of 1: 924 ms per loop\n"
     ]
    }
   ],
   "source": [
    "# Importando o Random Search com cross-validation\n",
    "from sklearn.model_selection import RandomizedSearchCV\n",
    "\n",
    "# Definindo o Random Search CV. Vamos fornecer o argumento n_iter, que fala quantas configurações de hparams testar:\n",
    "random_search_cv = RandomizedSearchCV(estimator = DecisionTreeClassifier(random_state = 37),\n",
    "                              param_distributions = {'max_depth':range(1,50),         # Testando comprimentos máximos de 1 a 50\n",
    "                                                     'min_samples_leaf':range(1,20)}, # O mínimo de amostras por folha, de 1 a 20\n",
    "                              scoring='roc_auc', \n",
    "                              cv = 3,\n",
    "                              n_iter = 50)\n",
    "\n",
    "# Realizando a otimização por GridSearch para os dados de cancer de mama:\n",
    "random_search_cv.fit(entradas,saida)\n",
    "\n",
    "#Vamos ver informações relevantes:\n",
    "print('Melhor comprimento máximo: {}'.format(random_search_cv.best_params_['max_depth']))\n",
    "print('Melhor minimo de amostras por folha: {}'.format(random_search_cv.best_params_['min_samples_leaf']))\n",
    "print('Desempenho AUC-ROC do melhor modelo: {}'.format(round(random_search_cv.best_score_,3)))\n",
    "print('Tempo para realizar a otimização:')\n",
    "%timeit -n 1 -r 1 random_search_cv.fit(entradas,saida)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "9O8CpXd7VATM"
   },
   "source": [
    "Muito bom! Conseguimos aumentar o desempenho do modelo e ainda diminuir o tempo de execução. Para conjuntos de dados maiores, essa melhoria aumenta mais ainda. Porém, ainda existem alternativas mais sofisticadas e mais rápidas que Random Search…"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "tyYrUapuVATN"
   },
   "source": [
    "### 3. Bayes Search\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "2O2dlWPhBfQq"
   },
   "source": [
    "<figure>\n",
    "<center>\n",
    "<img src= 'https://drive.google.com/uc?id=1jJ1L23Qitx3V6Ja2fR_oRqttDJS-DgVT'width=\"600\" height=\"300\"/>\n",
    "</center>\n",
    "</figure>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "7B2LN7jsa_fF"
   },
   "source": [
    "Subindo na liga dos algoritmos de otimização, nos deparamos com este método. **[Bayes Search](https://scikit-optimize.github.io/stable/modules/generated/skopt.BayesSearchCV.html#skopt.BayesSearchCV), ou busca bayesiana, tenta estimar qual é a combinação de hiperparâmetros que resultará na maior performance**, com base nas combinações testadas anteriormente.<br>\n",
    "\n",
    "A grande sacada é forma que esse algoritmo faz as estimativas: ele procura as regiões onde há menor confiança na distribuição levantada e, dentro dessas regiões, qual é a que pode ter um valor mais elevado para a performance. Isso fica mais claro com a gif acima!\n",
    "\n",
    "Assim, a cada iteração da busca bayesiana, a configuração de hiperparâmetros com a maior chance de melhoria é escolhida e a distribuição da performance do modelo é atualizada, permitindo a escolha de um novo ponto.\n",
    "\n",
    "O ganho de desempenho que esse algoritmo proporciona é brutal em relação aos anteriores e é uma das melhores escolhas para datasets gigantes e modelos com muitos hiperparâmetros. Vamos buscar ainda mais hiperparâmetros e ampliar o espaço de busca dos que já definimos:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 134
    },
    "colab_type": "code",
    "id": "j7c0XQ7pWBqj",
    "outputId": "ec751518-6704-47c3-b5c1-2e3446633a52"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Requirement already satisfied: scikit-optimize in /usr/local/lib/python3.6/dist-packages (0.7.4)\n",
      "Requirement already satisfied: numpy>=1.11.0 in /usr/local/lib/python3.6/dist-packages (from scikit-optimize) (1.18.2)\n",
      "Requirement already satisfied: joblib>=0.11 in /usr/local/lib/python3.6/dist-packages (from scikit-optimize) (0.14.1)\n",
      "Requirement already satisfied: pyaml>=16.9 in /usr/local/lib/python3.6/dist-packages (from scikit-optimize) (20.3.1)\n",
      "Requirement already satisfied: scipy>=0.18.0 in /usr/local/lib/python3.6/dist-packages (from scikit-optimize) (1.4.1)\n",
      "Requirement already satisfied: scikit-learn>=0.19.1 in /usr/local/lib/python3.6/dist-packages (from scikit-optimize) (0.22.2.post1)\n",
      "Requirement already satisfied: PyYAML in /usr/local/lib/python3.6/dist-packages (from pyaml>=16.9->scikit-optimize) (3.13)\n"
     ]
    }
   ],
   "source": [
    "#Instalando modulo necessario\n",
    "!pip install scikit-optimize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 101
    },
    "colab_type": "code",
    "id": "TlpW92y5VATO",
    "outputId": "0863413d-689f-46b9-cae7-3df21ed4cf3e"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Melhor comprimento maximo: 39\n",
      "Melhor minimo de amostras por folha: 16\n",
      "Desempenho AUC-ROC do melhor modelo: 0.97\n",
      "Tempo para realizar a otimização:\n",
      "1 loop, best of 1: 50.1 s per loop\n"
     ]
    }
   ],
   "source": [
    "# Filtrando avisos \n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')\n",
    "\n",
    "# Importando o Bayes Search com cross-validation:\n",
    "from skopt import BayesSearchCV\n",
    "\n",
    "# Importando espaços de busca\n",
    "from skopt.space import Integer\n",
    "\n",
    "# Definindo o Random Search CV. Vamos fornecer o argumento n_iter, que fala quantas configurações de hparams testar:\n",
    "bayes_search_cv = BayesSearchCV(estimator = DecisionTreeClassifier(random_state = 37),\n",
    "                              search_spaces = {'max_depth':Integer(1,50),        # Testando comprimentos máximos de 1 a 50\n",
    "                                              'min_samples_leaf':Integer(1,20)}, # O mínimo de amostras por folha, de 1 a 20\n",
    "                              scoring='roc_auc', \n",
    "                              cv = 3,\n",
    "                              n_iter = 50, random_state=37)\n",
    "\n",
    "# Realizando a otimização por GridSearch para os dados de cancer de mama:\n",
    "bayes_search_cv.fit(entradas,saida)\n",
    "\n",
    "#Vamos ver informações relevantes:\n",
    "print('Melhor comprimento maximo: {}'.format(bayes_search_cv.best_params_['max_depth']))\n",
    "print('Melhor minimo de amostras por folha: {}'.format(bayes_search_cv.best_params_['min_samples_leaf']))\n",
    "print('Desempenho AUC-ROC do melhor modelo: {}'.format(round(bayes_search_cv.best_score_,3)))\n",
    "print('Tempo para realizar a otimização:')\n",
    "%timeit -n 1 -r 1 bayes_search_cv.fit(entradas,saida) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "f3mVWGRAVATQ"
   },
   "source": [
    "Parece que chegamos no desempenho máximo: esse algoritmo de otimização gerou um modelo tão bom quanto o Random Search. No entanto, notamos uma demora maior de execução — o Bayes Search funciona melhor para treinamentos mais complexos e com espaços de busca muito grandes, então para nosso exemplo, Random Search bastaria. Mas fica o aprendizado para os projetos reais!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "z0bz8KmGVATS"
   },
   "source": [
    "## Conclusão"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "PipqjPcqBbIM"
   },
   "source": [
    "Nessa aula, vimos a diferença entre parâmetros e hiperparâmetros e comparamos os algoritmos de otimização de hiperparâmetros mais populares: Grid, Random e Bayes Search. Aplicamos eles a um caso prático de predição de câncer com árvores de decisão.\n",
    "\n",
    "Por fim, refletimos um pouco sobre o escopo de aplicação de cada um desses algoritmos. Se quiser ir além, existem [algoritmos ainda mais modernos para otimizar hiperparâmetros](https://ray.readthedocs.io/en/latest/tune-schedulers.html).\n",
    "\n",
    "E como sempre, para ficar por dentro de Inteligência Artificial basta seguir o Grupo Turing nas nossas redes: Medium, Facebook, Instagram e LinkedIn.\n",
    "\n",
    "Por hoje, é isso. Espero que tenha gostado!\n",
    "\n",
    "Para sugestões, dúvidas e feedbacks, segue meu contato!<br>\n",
    "\n",
    "Lucas Miura Semenzim<br>\n",
    "+11 99878 7085 <br>\n",
    "luc.ms22@gmail.com"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "LmkVz7h-BSmf"
   },
   "source": [
    "## Apêndices"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 503
    },
    "colab_type": "code",
    "id": "iDKjbMP1q5d-",
    "outputId": "b356e343-54ae-4699-f358-e071bf8d3d2c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Desempenho médio de AUC-ROC: 0.916\n",
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 5000 entries, 0 to 4999\n",
      "Data columns (total 3 columns):\n",
      "param_min_samples_leaf            5000 non-null float64\n",
      "param_min_weight_fraction_leaf    5000 non-null float64\n",
      "mean_test_score                   5000 non-null float64\n",
      "dtypes: float64(3)\n",
      "memory usage: 156.2 KB\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA5UAAAFPCAYAAADdkuqzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0\ndHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nOx9ebxmRXH2U3eWu80gwwwa9kGWyERZ\nlABKEhVjRMU1moBxR8nmkqhJxCQGSNDEJSbGaCTBfUEkG+FziVExXz5FQdkEFBCQZcbIDDMw+1rf\nH+dcaHre23Xqre4+/d7p5/d7f/e+7zmnT3Wf7uquqqf6EDOjoqKioqKioqKioqKiomIYjPUtQEVF\nRUVFRUVFRUVFRcXoohqVFRUVFRUVFRUVFRUVFUOjGpUVFRUVFRUVFRUVFRUVQ6MalRUVFRUVFRUV\nFRUVFRVDoxqVFRUVFRUVFRUVFRUVFUOjGpUVFRUVFRUVFRUVFRUVQ6MalRUVFRUVFUOCiMaJ6A+J\naKpvWSoqKioqKvpCNSorKioqKiqGx/sA/C8zb+pbkIqKioqKir5QjcqKiopiQURfJKJX9C1HLBDR\nx4joL/qWAwCI6GQiuoWINhDR8/uWpw8Q0R1E9MuWMpj5d5j547FkygEiegoR3d23HD6I6G1E9E8d\nzltIRNcT0b8Q0c8T0V/nkK+ioqKiYnZUo7KioiI52sX75taA+d/WuFokXcfMzxy1BfsI4TwAH2Dm\nRcz8b30LM4ogokVtn/5i37LkQjt2txHR+vbzfSJ6JxE9wlo2M7+DmV/T4dSjAPwrgEsA/AOAz1rv\nXVFRUVFhQzUqKyoqcuE5zLwIwOMBHA/gT3qWZ48EEc1v/z0EwA3GMvZ0/CqArQCeTkQ/M9tJsdur\ngPZ/FzMvBrAvgFcBOAnA/yOi6Rw3Z+ZrmfntzPwZZn4CM1+Z474VFRUVFbOjGpUVFRVZwcz3APgi\ngMcS0RIiuoyI7iWite3/B86cS0SXE9Fr2v8PJ6JvENH9RLSaiD7nnPe3RHQXET1ARN8lol90jk0S\n0cfb8m9qN1W52zm+PxH9cyvD7UT0BufYOUT0eSL6VBuVuZ6IjiSis4nop+09f2W2uhLRcUT0vfba\nzwGY8I6fRkTXENE6IvomER0dKIuJ6A1EdFtb/3cT0Zhz/NVt/dYS0ZeJ6BDv2t8lolsA3EJEPwLw\naAD/0Ubaxtt2uJSI7iOiW4notV47XNK2wwMAXtk+m79o5d5ARP9BREuJ6NPtc7iSiJZ3fEbnENHF\nRPSJtq1uIKLjneMHtVTHe4loDRF9oEu9B7Thy4jox20Zf+wdO4GIvtU+i1VE9AEiWjhbWS1egSZS\ndh2Al3rl3UFEf0RE1wHY2P5/iXfO3xLR+9v/te3/CCK6sJX1nvZZzJul3pPURBjXEtGNAH7eOz7r\nGAiBmbe0Bt1zASxFY2DOlBnqjz9HRF9p6/q/RPQ2p56fav+faOu7pn0mVxLRo9pjr2rLXt+Oh9/0\n6vPatg3va9t0/y71qaioqKgYHtWorKioyAoiOgjAswBcjUYHfRRN1OxgAJsBfGCWS/8cwH8CWALg\nQAB/5xy7EsCxAPYB8BkAnyeiGQPuzwAsR2NEPR3O4r81yv4DwLUADgDwNAC/R0TPcMp+DoBPtve9\nGsCXW7kPQEMh/fAs9VwI4N/aa/cB8Hk0ka2Z48cB+AiA30SzIP8wgEuJaHyW+gPAC9BEeR8P4HkA\nXt2W9TwAbwPwQjTRo/+L3SmBzwdwIoAVzHwYgDvRRo+ZeSuAiwDcDWB/AC8C8A4iOsW5/nlo6IZ7\nA/h0+9vpAF7WtsVhAL6F5nnuA+AmNG0/g9AzAhrD5KK2/EvR9oPWULoMwI/RPMcD2vO61hvtuSsA\nfKiVd380bX6gc8pOAL8PYBmAJ6LpC78zqKy2vEMAPKVti08DePmA084A8Oy2ThcBeBYRLXbq9Wtt\nWwD69v8YgB0ADgdwHIBfATAbdfTP0DyfwwA8A40xPFOPLmMgCGZeD+ArAH6xLXPW59LW/78AfKmt\n6+EAvjqg2FcAeASAg9A8q99Cox8A4KcATgOwFxpD9n1E9Pi2/FMAvBNN2+6Hpt9c1LUuFRUVFRVD\ngpnrp37qp36SfgDcAWADgHVoFnkfBDA54LxjAax1vl8O4DXt/58AcAGAAzvcby2AY9r/bwPwDOfY\nawDc3f5/IoA7vWvPBvDR9v9zAHzFOfacth7z2u+LATCAvQfI8EsAVgIg57dvAviL9v8PAfhz75of\nAnjyLHViAKc6338HwFfb/78I4Ezn2BiATQAOca49ZcAz+eX2/4PQGFWLnePvBPAxpx3+27v+cgB/\n7Hx/L4Avem11TcdndA6A/3KOrQCwuf3/iQDuBTB/QBnBenvnvh3ARc73aQDbZtpgwPm/B+BfA/L/\nyUz90BhjOwEc57Xvq71r/gfAy9v/nw7gR8O0P4BHoaHdTjq/nQHg67PIepvXd85CxzEwoKyPoe3D\n3u9/iXashJ5LK+fVs5R9DoBPtf+/Gs14OXq2Z+Bc928A3tj+fyEaeu7MsUUAtgNYLpVTP/VTP/VT\nP8N/aqSyoqIiF57PzHsz8yHc7Ji5mYimiOjDLSXxAQD/DWDvWWh8fwiAAHynpUe+euYAEb2lpcPd\nT0Tr0EQ4lrWH9wdwl1OO+/8hAPZv6XXr2mvfhmbRPoP/df7fDGA1M+90vgPNwtXH/gDuYWZ2fvux\nd+83e/c+qL1uNriy/9g59xAAf+uUcx+atjpglmsHyXofNxEnt3zper9t/O8PtovwjADgJ87/mwBM\nUJM7eBCAHzPzjgH371Jvt44P1oGZNwJY48h3JDX065+0ffEdnnw+Xo42YssNpfsbcCKALfw2+wwa\nowoAXoKHopTa9j8EwAIAq5y6fxjAI2eR1R8Dfj+UxkAXHICm/WfKnO25HATgRx3K+yQaVsBFRLSS\niN5FRAsAgIieSURXtPTWdWiYD+54f7B+zLwBzXMe1CcqKioqKiKhGpUVFRV94s0AfhbAicy8F5ro\nHtAsQB8GZv4JM7+WmfdHQxn9IDV5lr+IxuD8NQBLmHlvAPc7ZazCw2mOBzn/3wXg9tbYnfksZuZn\nRajbKgAHEJFbl4O9e5/v3XuKmUM7WbqyH4wmEjpT1m96ZU0y8zed813j1sdKAPvMUDOd8u/peH0Q\nHZ5RCHcBOJgGb07Tpd4zWAWn/YhoCg2tcgYfAvADAEe0ffFts8lHRE8CcASAs1sj9CdoIn4v8eT0\n2+zzAJ5CTd7wC/CQUalt/7vQRCqXOfXei5l/bpC8ft2xez80jQFqdnL+ZTQ015kyZ3sud6GhogfB\nzNuZ+VxmXgHgSWjori9v6eH/DOA9AB7V9qUv4KFntRKNUTsj2zSa5+y2ZUVFRUVFZFSjsqKiok8s\nRhPRWkdE++DhOXgPAxG9mB7axGctmkX2rraMHWgpkkT0djS5VjO4GM3ifwkRHQDgdc6x7wBYT80m\nKpNENI+IHktED9vIZEh8q5XrDUS0gIheCOAE5/g/AvgtIjqRGkwT0bM9w8LHH7T1OAjAGwHMbFb0\nD20dfw4AqNnE5cVdBWXmu9BQDd/ZbpByNIAzAXyqaxkCpGcUwnfQGEV/2bbRBBGd3B7T1PsSAKcR\n0S+0+a7n4eFz4GIADwDYQESPAfDbAZlegSaHcAUayvaxAB4LYBLAM2e7iJnvRUMb/igaQ+6m9ndV\n+zPzKjT5xe8lor2IaIyIDiOiJ89ya3cMHAjg9c6xoccANRs8PQEN/XRtWy8g/FwuA7AfEf1ee/1i\nIjpxQNlPJaLHtayFB9BQWHcBWAhgHE1f2kFEz0STTzqDzwJ4FREd2xqg7wDwbWa+Q6pPRUVFRcXw\nqEZlRUVFn/gbNAvx1QCuQLN5x2z4eQDfJqINaDZyeSMz34aGIvclADejob1twcOpfueh2QDldjQb\nhFyCJsqDlsZ6Ghqj4PZWjn9CQ800gZm3odmo5JVo6H+/DuBfnONXAXgtmg1p1gK4tT03hH8H8F0A\n1wD4P2jyx8DM/wrgr9BQBR8A8H0EjJtZcAaajXBWonkH4J8x838py5gN0jOaFe0zeg6aDV3uRPMs\nf7091rnezHwDgN9FEx1chabN73ZOeQsaSup6NAb/5/wygGZXUjQR179ro+czn9vRUDZ9CqyPz6CJ\n6n3G+13b/i9HY2Dd2NblEjQb0wzCuWja/XY0xugnZw4MOQb+kIjWo6GVfgJNn3xSSykOPpeW4vt0\nNM/0JwBuAfDUAff4mbZOD6DZ9OkbAD7ZXv8GNIbyWjTP7FKnPv8F4E/RRDNXodmc6PRAXSoqKioq\nIoAenu5TUVFRMbdBRL8N4HRmni2qUySIiNFQM2/tW5aKioqKioqKChc1UllRUTGnQUT7EdHJLUXw\nZ9Hkcf5r33JVVFRUVFRUVMwVVKOyoqJirmMhmp0x1wP4GhoK6Qd7laiioqKioqKioiOI6FQi+iER\n3UpEbx1w/BAi+ioRXUdElzt7UICIdhLRNe3nUv/aaDJW+mtFRUVFRUVFRUVFRUV5aDcsuxlNPvrd\nAK4EcAYz3+ic83kAlzHzx4noFACvYuaXtcc2MPOgV59FRY1UVlRUVFRUVFRUVFRUlIkTANzKzLe1\nmwBeBOB53jkr0LCxAODrA44nRzUqKyoqKioqKioqKioqysQBePiO6Xe3v7m4Fs2O80DzHuTFRDTz\nLuYJIrqKiK4gouenEnLQy6RHBsuWLePly5err/vuDbcDAMbGCLt2xaH/hspyj82fN4bpqQls3LQF\nO3bu6lTuwgXzsW37DlFW/1zNtaUh5rMpFWNjhInxBdiydTsAzIln1Wefc9uzzzYcpg1Gsb+Psn7R\nQNLtljZI9dy7yKW5d+y50p+nNHKU2OdKlWsQ+tI1o9JGKeUcZlyOytzQVU6pDcbGCLR13bodO3Ys\nSSFnDvzKk5fwmrXbVdd87/qNN6B51dYMLmDmC5S3fguADxDRKwH8N4B7AOxsjx3CzPcQ0aMBfI2I\nrmfmHynLFzHSRuXy5ctx1VVXdT6fjnpJe+GRmJ4ax/L998UdK+/Fxk1bH34eEaYmF2LT5m3oknMa\nKss/tmtsDONLFmH92g3ALtmoZCLMm1wI3rwNYA7K5p/rf9dA2wYxEWpPLfqsh4Sp6Qk85tD98YPb\nV2Ljpq1DP6s+4T8rS5+zYmp6Ao959H74wW2rsGHjFvmCRNC2Qcz+7iNl/+/zWeeC9GwsbaB97qFn\n6R+T5g3NvRc5eirGuHJlm55cqGqDUvtcqXL58J+73y9GRV+Mipzasv3noxmnKdtEKlsjp9QGTAS6\n+XO3RxO+B6xZux3f+o9jVNeML//mFmY+PnDKPQAOcr4f2P72IJh5JdpIJREtAvCrzLyuPXZP+/c2\nIrocwHEAohuVc57+Ske95MGPi02bt+GOlfdi0+Ztu10z1U50U5MLO93DL4uIMD01DiLa7diuXbvw\n0zUPYFcHgxIAmLlZrLeDLySbf67/XQNtG2jgts8ghJ6NFinrIUGq58ZNW3HTbfc8+IyGfVY54dfJ\nf1Z91mPjpq246UcrB05q0rPwj0vnh67120AqW9PfNXIBtv4v3auUPqttEw2kZ2NpA62eCz1L6Tn7\nx9U6lrrXT9NvtHL02efceklj3lK25pgWfntPT43jqEcfgOmp8YHfLXrQh6QXNSh5Xrf0E//5aMZH\nyjaRytbIKbUBM2Pnzg40vj0PVwI4gogOJaKFAE4H8LBdXIloGRHN2HVnA/hI+/sSIhqfOQfAyQBu\nRALMWaNykCHpItSxrUaNOwBjT4KubH0upiyQFJSlzfw22bxlO+5d+wA2b9FREWIgZT1zwm1Tv04x\n62DtzyFZtItuzQRtLTskt98mVoeXBjEXKSl1lWUhLJ2fcoxKZWt0mf+cJSNSU6+Qs2YQ/Hv31b6x\n4dYr9gLe4jDQYGB7+w4D53tMPejDH7c+Qv3Guj4Jla3V56Hj2jbRBAYszkkJku7ROlIr7GDmHQBe\nB+DLAG4CcDEz30BE5xHRc9vTngLgh0R0M4BHATi//f0oAFcR0bVoNvD5S3fX2JgY6VeKHH/88ezS\nX0NGZAhW2oBEKUmFlLS5lEjZPhb6SGzZSqbehhCiym3avM1Elwqd3ycN1EIDS1l2X7plkNwWpHy2\nizzas/ZeperRPqlwlv4vUW370otWXeV+BxC1Dhpqc0xIzy6mHvThj1sfKcelJf1JcxyI20+61iF2\n2VpdZJbtB5/9rkAFLRpPOHoRD0F/Hek6z2BORCqlqKQEyZs0NjaGRy7dC2Njg5srJQUwpreuFO9R\nSu+0hT4ChPuCtv1GyQvvIhTl0NCxu5TtImV0vAvlZljqeOyy3X4WW7do+nDM/pvy2W7avA13rlzT\necz3xWbQ6g+LLtM+u1BEXJJbirxLUdRckCKoEivArZc2yiwdD5VnfZYhWPSefy9ttEqKgKeMvKUs\nW9NPLEitU0O6R7p3StkqysZIG5VX33QHxlb8hrkcaQAsW7IIJx1zBJYtGfze0JSKQ5NDKUGim8wF\nSJOkNNHFzLUtBdrFrL/IjkkV1yzgS3GCaGGVO0Sfj1l2TqRcGE9OLMC+S/bC5MSCTvfy28C/PhWm\np8Zx1GH7d9a/1gW/BiFHkiR3aE+BQXLnWnBKhoRES49JHbfS2DXQ9rMQLFRQa/qHhvqprUeobElu\nv31HRadaytbqolF1qFfYMdJG5cT4giiKUxoAq9duwBXX3oLVazeY76WFNGGrodhwIRdyGg/ShBvT\ngCoF2klPs8jWTh6hsi15jSXBKndKx8ao9GFNPbV1srIZTOAyHCSSwbXbuA7IrWUv5FpwaiOoGlaA\n1jkZ02DthEj9zBKR0jIGNMe1elDT3p3Oddo35bMcVcdqxZ6LkTYqLdAMVu2OrTFhoRv6ddRuuJAL\nUp1iKlYiwuT4wqHKSrkY0tZRc7510R0TISdJr4v9iLDKHepnVqpmKR5kiXpooXrGpKnH1D3ubs9d\nkHJBqYncaOUuZZGtjaBqxobWcE6ZIuND+7xC8OX0U4EstF1LFFSrB2O2t9++lrKl/hwz6myBVReF\nrq+G89zCSBuVW7ZuH1pxanJGSoJmwtYsHID+Br5Up5gRK2bG5i3bh86FCcFiGMbMTfShnfRy0mpC\nVM/YeUsW9JWL6MOP9I6S7nKRMiptWaxazpWg7RcpI/Upjfa+0kF82Xw5YqZ/lOzwStn+oVQga86w\n5nhKyrrUT6zzkotOY7wHdkNsuraFJl0xWhhpo3LXLh5acboKa650amsyel8DX1LSMSdwyYtrqafF\nMCwpmpgTmnpoJ7qchkoulLLpiRVSVNpSr5ibSPQ5zlLe22J45GSW+DA7ICOlf5TkIMiJNes24rqb\n78SadRt3O6atoyUvL/m4NPSTmNT9mFFniR0ScnJb29tCk64YLYy0UWmBq7BGqVOHIqzayKSPUgd+\nTM9rSgM2p2GoaZNRjWb50E50MXf1LEVHxNz0JCXlWgtNvayRMs3Y0eoeaeGmQSlUZR85mSU+LPo7\nZ/qH1cEb894xMTE+H9MT45gYn7/bMWsOpQbWcRmCtZ+E5h2Jgu0jpg6Q2CEhJ7dVDgtNumK0sMca\nlS5id+qUFEp3sPs0jZS5XCUN/JyLW819LTTTlLkTJXvJQ7KpNxPx4FKkrH2mr/4vGSkpI07S+ZY2\ntWzmMkr9uWRZu0Lb56RcxlDZVsScw2L275xOQMu+CxJCawypjn3mB8ZMF5F0cmje6XMHfokd4n7v\nc71HRJg3b161TUYU9cElQEoK5W6D3aFplGT4pUTMxa3lvtERMXfCbZNSomyDEJItZuQ9NmWvrz4X\nsw9aI+spKaqprk2N0EJtVBkDVlpjaCFdstHdV/+2tomGoRGbsioX0E/ft6RZ+JB0sjjv9LQDv7Rn\nQSnrx6nJhRgfHy9PIVR0QjUqE8BC3dIoP5+mEZN6lRIabv8g9JVXJk3Wlvb3cydieqtjbiwgQVuW\nK5sUmbSUHZuyl6vPSd5lC6wbOKWkqI6K7vIRWqhp+0yfbRDdKTXLQto6xlMiV3oCELe9pyYX4uD9\nl0bJ6YsJS35gbKaJW542V99fB4QYNP6xPinYo4JNm7dh69at5XkMKzphjzUqUy6kLdQtS76P9aW8\nuYwLDbd/EEKL25SKVNpxzhJV8utkNVo0i4U+N7Rxn5cUmbTIGTuXNtdiTOtNzrmQ0Og5rVyhZ91n\ndCtn9GpU6ik929BC2jrGS02F0J6vcQJ2udfkeP87Vce8V+yx4JanzdWfmlyIg/db1olBI+nvlIGA\n2OycEGKWxczYuXNn/vf3RcTOnfOxbv0y1WeuYM4alVoKgwWaspIvRg0v5c3VJr5ckudPgiUaoIEk\nV8yoknXTA80EHrNPWvqctn1jwhq1KwU5DZFQH7SyCULPuk/6a87oVZ/1dHWy9OxiOkqtc1apqRBS\nPmFsBsLmreXpJgtijwW3vKE2pYlEYe0zvSGm46hkGntFXsxZo1Lq5H0tpFMuRq0v5c3VJr5cfgTQ\n0kZWYywEycsYM0dBujamErfIqd3Ew4dmc4CUG22URB233DunIRLqgxqq7CBovP85kTNSnLKeXRaJ\nMxTKPp09EkLOi5RjeKh5JpBPGEoD0NYjJsUypx601Nty7TBz1p0r1yRZJ+VMb9DcS9pgqE8dUFEW\n5qxRKXXymBN2KVELqxx9GRc5FWlMz3afu9nFfF2GBVa6Wsqxo5EtpcdYi5Q035goxZlWEkbFY99J\nTu7/tQcSQs6LlM9CO89o8gmtuqjP5xHzXiGKqnStZARZ6iWlwYSQ0hGthfpegejsnqLfK2TMWaNS\niipVxEXKhXDMZ+cvhC1bfGtyV2LDMrHFRMl5YhrZrBTsmMh1b2lcSd7/PWEhodU9o+Kxl5xSlk1V\nLLC2n9sn+3S8DUWpnOXakuoRgnWeDkXthoqWB4yglKkpIYyK08lHyg2GBs0z9ZUio4s5++C0O3pZ\nys51bclIuZiKyf0fOLkPmR8h5a70Sb8KIWb/LTlPLEQp8xGTgu1D2965jDUp0l5S9DYlNJuKSdeO\niqFdilPKh7X93OfRZx0tTm3/WpeK3AV96ffYrBX3uzb1RDKCUqamhDAqTicfKYM0g+aZ+kqR0cWc\nMSolIzLmYJYolJZFyqgi5WJK8+y07WvxwEnX9km/CsEil7VOfeWJ5cwJsTq0sjqeAnleKfN9JORs\nA82mYpprS4alXil3jbSWrdn4q5Q+1gmK9ztanLAWOXOyViR9XqpzJ+W+ADGvlZByQ6FNm+srRUYZ\nc8aolIzImEpG2tkwNOBiUlm0E3KfSfcWpMzP1JQteUf9431FEyVo5XJlsS7UYtZrbGwMj1y6F8bG\nGjUmTnSKnBBzrq2z4NG2WS5DRaI4avN93HpYn7OFlh6Sa9B3zaZiPrZs3YGNW7Ziy9YdA8suBdoI\nVKhNYvZPae7U6hPNxl8x8+slWOYCf5xKclqcsJa5ITZrRXweCoZRqeNSQk4HsKaN/GcXe7036q8U\n2ZMxZ4zKlEakj9DmAIO+u5iafPg7jiwYmJzu0NmkxWpOT1bMBWcIMelTPiQl7S+E+4omStDK5cqi\nXahpj2uw7z6L8ZQTj8K++ywGEB532oi0WU5nwaNts1wRwdg6UrOxRidk2rY/1A6Srlq69zSOPvJg\nLN17emDZpcAil98GMR2j0typ1ScWB2RJOjd0rSSnpQ0sc4MWkiPPLds/ZtXnpRqZMR3TWsMvlA4h\nXRvT+CWqOZWjjDnz4PqkO4Q8+gMHo2GxJEaNHJqMFDFJOYmG7lUyHdBChQMQbSFcUu6FRpahNlQY\nVq4t27Dqp/dj0xZ5QaTdDMocWVAseHI6xCzQRPyszznmxhAWGq+kq1av3YArrr0Fq9duUJedE9oF\nZkhfx8xVlKLhKfWJ9l6lIGaUyKprUqYVuWVbnAeD5Iy59kmZX5jTGdFc1M1ojOkIHfRsak7l6CK5\nUUlE84joaiK6rP1+KBF9m4huJaLPEdHC9vfx9vut7fHlqWULyKxSEqHzfe+PdbEUihoNpLMFIiYp\nJ9HQvXLmXmifpYYKZ/WehlCSYaGRJafcGzZuwfduvB0bNm4xl5VyMi8JmvEgMR1CbWZtg5htqKXx\nupB0FTNj4+Z+XgmggXaBGdLXOY0vqT1j0tZLfXYSdblPAyllWpFbtpUqm3LtkzK/0AKto3TT5m24\nc9XqgcclHRBzvty0ueZUjjJyRCrfCOAm5/tfAXgfMx8OYC2AM9vfzwSwtv39fe15vUCrJMTzHe+P\ndRLUGD0pd0DTwnIvi6LVPkvN84lpiJSMlJ7YUtDntv052yR0L8mItBgXKdkGUgTVUpaV7m25d0p0\nMpaHjPpLSFlPaUfjVMiZSuIj5jjskyZq6XPWed6CUPvHNtI15WnXJyEGgsUxpwVzzamcDUR0KhH9\nsA2+vXXA8UOI6KtEdB0RXU5EBzq/f4+IriGiG4jot1LJmNSobCv0bAD/1H4nAKcAuKQ95eMAnt/+\n/7z2O9rjT6OeCO9aI0bM5QpshpFSGfZp5FhyE31o6hGTxijJmtNjn3Nyl4wLC0r1EPf5+oGc/UjK\n99ZsdhYal9Y+ZFk8We6lvbbU/ixBo1Njy5W8noqdUmPBWqdQf7c4ALRyammipeQm9kldDrV/bL2n\nKS/mOraifxDRPAB/D+CZAFYAOIOIVninvQfAJ5j5aADnAXhn+/sqAE9k5mMBnAjgrUS0fwo5U0cq\n/wbAHwKY8TosBbCOmXe03+8GcED7/wEA7gKA9vj97fmzYmyMeldmgM3DVupAtk4WfRlj1uihxii1\nlq1Bn9GsmM8rl4dYiz4jlZY2kfqUJvImGZEWQy8l5d1Kz3Tvpc1XK7U/x0RsuVKONcmJmwopnZfW\necb9rjVQpfNLYZ6UyhLS0nal9tSMHW2buOfHdhaU4nwYcZwA4FZmvo2ZtwG4CE0wzsUKAF9r///6\nzHFm3sbMM0pxHAltv2QFE9FpAH7KzN+NXO5ZRHQVEV01j7cnUWba7eznYh6HlQYT0xjTIPbkbomS\nSmVr2tSyOYP12aU0eixlxcynmprUvWA8JFdOSIsQzaIvJsXM2oc049hKzXIXalK+WspnndNJpbk2\ntlwuK2CuLDatc1rKiHdoHxMLn/QAACAASURBVAar0ySlI2RU+kZITq1OldozF6MmNo28FOfDiOPB\nwFsLNyg3g2sBvLD9/wUAFhPRUgAgooOI6Lq2jL9i5pUphEwZqTwZwHOJ6A40FvUpAP4WwN5ENL89\n50AA97T/3wPgIABojz8CwBq/UGa+gJmPZ+bjd9KCdF5dxS6ec2XAhDya0uJKExFJiZSRSQmxqUUu\npMWuRS7pXhbEXJTHNKAGYkjaXJ/jv8vmC9F22zUYelpoqLVWhBZqljFrRUwqXMxrrWW7bTpy1NpE\nSBnx1lDetUg5z7uylWxgWtpQ63zLymZQzIfS8ymVhZEKO3fNw/3rF6s+AJbNBMzaz1lD3PotAJ5M\nRFcDeDIau2onADDzXS0t9nAAryCiR0Wq7sOQzKhk5rOZ+UBmXg7gdABfY+bfQBOSfVF72isA/Hv7\n/6Xtd7THv8aCptq1i6MkNPuwvhJgVBHyaEqLq1GdzK10WRdaWpeFBqa5ts/+qXVOhCDV2VJPiTZn\nyXlKGVGS+mvMRV/MqL0FsXWN+/wk51jOsaR1UoVgcZ5JDkQfW7buwMYtW7Fl646Bx902jU2FLXUu\nTjkepD6robyXBFc2LXssJyxtaKGoaqHpg1oaufR8SmXkFYbVMwGz9nOBd/zBwFsLNygHAGDmlcz8\nQmY+DsAft7+t888B8H0Avxi9BujnPZV/BOBNRHQrmpzJC9vfLwSwtP39TQB229lIA8vCI+dAz4mY\n3iRtHlNKGqQFMSdViZri90kLlcVC1/SRkkqrdU6EILWXZRxaqJ8xaaMxr5WgzcfUoM+NSyx9tCTn\nWMoNWizXSm20dO9pHH3kwVi697RYdmw6X6lzccx+pc3DC6HU9gIGyBbpHdBaWB17uRCT2TBUnYZ8\nPiVHoQvDlQCOoOa1jAvRBOsudU8gomVENGPXnQ3gI+3vBxLRZPv/EgC/AOCHKYTMYlQy8+XMfFr7\n/23MfAIzH87ML+Y2eZSZt7TfD2+P32a5Z8keuL4gKZWQIpEWJVKuRsxJNWZZKWlH0nFzH+1IT4lJ\nG5XO1U7AFkdGTljunfNazQQtRY1TRpW1srlIacSn3JjE6pCxIGYflMpavXYDrrj2FqxeuyGpXKME\nSz2lcbkntKGWPRYTJTmaQkjJbJBgeT6j0r59g5sNTF8H4MtoXtN4MTPfQETnEdFz29OeAuCHRHQz\ngEcBOL/9/SgA3yaiawF8A8B7mPn6FHJS394VC2hyKWP5M4a7lprt4zdt3iZO2ppzY8sROq6Va2xs\nDMuWLMLqtRuwa9furwGy3Ms/Pj01juX774s7Vt6LjZu2ivcOwS87ppwp0We/0Zwbs6xF0xN4zKH7\n4we3r8SGjVtM99Ii57ONCcvY8OGPuxCkMSsdt8iRcpyWqq817TeXUKrO9c/1x6GlrJh1lsalpQ20\nGFUdG0LJ7dlXH7Tqbw0GlvWDz36XmY83FdwjHnfUUv6XjzxTdc2RT/r0SNd5Bn3QX4tAyDuSMsqm\nkUM6rpVLoky65Wk9VZL31Er17ErdihmV82Ftk5jQRDGkc2OWBSBIg+lzLJWKZUsW4aRjjsCyJYvM\nZWmooVLUOGVU2cKakJCSjmaJku4JEaVByDkuLawLfxxayopZZ2lclrReGUVYdVHKNAFNe/tyWlhF\nKRkbktwVo405a1RKAz1Ez/IHjJbKpVEy0uAN3XuoRUqAMrlbYryzpbSWWmFZkEplW8610PK0irTU\nRWTKHIZNm7fhzpVrOlOAY9+7xPaW4NMFQ8/Hmt9joZXGdD70+V7QlAj1QckY6BOpdUIq2p0PywZm\n/jgMye3L6Z8bc82gHdMxqbY+StWx2n7jnq9xxA1CSrq9f9yyttQ4H/w+NzY2hkcu3QtjY2Od5K7Y\nczFnjUppoGu2kddG2SzeJY2cWg+PtKPXbuU5BqikRGIuSC3XdpmAh93gRqtIU3o4LdBOgho5c71H\nKzcsz0q6dteuXfjpmgcepL6GGANWD3HKBacGKftJymclQaOrSor6aJg7WuRsE02/8uXyx6GGEeOf\nm3LN4ENy6GqQkkGghdvvpD6ofa+iW09rjrZFp2rXTTEjlxq5/Si+JLdvhFbsOZizT1zrAXJhjbLF\n9OJYPFU+NBOCb4BaJ5O+DKiB93WMZf+4xWOsRV+LSm3/1MiZkzZTKu3LKleIMaB9dr4sKRecGqT0\ndFvkzDkmc3r7LRGonG1ijWDnalPL+iLG+S5izksp28+SPtKpDyreq+jWU1oDWB3qLqzroJiML43c\na9ZtxHU334k16zZ2ulfMlI6K0cKcNSpjKlptWZZ7S4vAlBN8KN/Kir4MKP++vrEcMxdGY6ACeWlh\nLrTPViOnVLbFSSIZa6VQa62GYIgxoH12pRqCKaP4KSljWuR0UoVgiUDlNDSsEexcbWpl5sR09sRE\nSjm0ush1MEh9UPteRbee0hogZptY9XFKxldIV02Mz8f0xDgmxueL5wLhHaBLov1XxMecNSolhChm\nMRF7c5dSPfwS+jKgfM+3FIW2tK/WQNVMEH2+/Dmlg0bT5/w2sHhetX1M3QYGQ9CFv1jSyh1z3OWk\nc1t0UUrKmBaWcRuzPXNGwjRy53QMlYSSqM+5oHXYuA6GlMZuzDWA9l4lIdQnfbk7rW02D7eZYsVo\nY481Kt1BUgptzpdrEHIq15jIlVej9Xz7coXk1EYeo7dnTy9/fpgIkR0w6jZy2qAUKqcPyWuuaUOr\nkZNy3Pn10OY1hZAyhSC7HEOO25h9tKSoqIuY+YB9IqazZ65GcrS6TDP2YjqhUvbBmGX3ORdL54ac\naSUb1hV27LFGpTu4Y1OeXGhzRPqcVHPmBoQgtVno3lrPt0RZdb9rI48xadD+y4VjJsJbIgtWqPN8\nI70Au08HSsxNOWJCq6sGLhxYHkddEFMXpaSMSbJY+mxJ+W0axKTLj4qBFdPZs6dEcmI60LXzfF+I\nOW9L/SSmzh2KEjyLM21UHUcV3bDHGJWa3JaYSl2bI1KK8vORc1MUqc00NA0iwvRk9+3AQ99jL/JC\nz9pfsPt9VEqET2Uo9ull9NtAY7iUlMNkWQD5csdcpAyVz+YsHNwIbcxcZQklLcJDi6/klGuDnDEh\njVMXUv/tk/avQUy9mPIVZinRZ//WOvL6arOc7ySOyfjS9u+QM62U/lqRBiNtVI6NUbRcRRc5J4ic\nxpoFJW2KEjquNb4kymrKeroLpoGKNkCbCyXCA+n6e8r8Ki00hktJ4ypmJNPv35YIoXYx6y8cQuyP\nkvKU+orSldQHczqHQgt6qf82P5Yf0YhpIOV8HUlMlCLHIGhzAlNBmrc1kPpczP0hYspWcj+psGOk\njcqFC+ZHy1V0kXOC6HOjAk1kp6TcF829JSUuleUeT6IMaXB0XKLN+e9U81EKBS3n5k+hOqekuMeG\n5tn5/dsSIdQuZkP9puQ8pZxROhdW3R6zz+bU3369XWea1H+19OFSx3VK504pOWqlrF0Gwe/vfbVZ\naAObkuD3wVLz5UvFzp1juH/9lOozVzDSRuW27Ts6d8yY+W0aSBNEn8aaJrJTilyDEHo+vvElRXI0\n1For3AVT7H4Q83lpKGhW2owGFsOlZG+pph5+/7ZECFPmJZW02O9rkxTrmCyFwiehE9W8daZJ/dfi\nMCipfSzOHWvOsMZ5bEHsNYJm7wQt+nKKx1zbxL6Xi4EORsV7QEOoOZVzGyNtVO7axUUaOS60u4/m\nhCayU4pcg6B5PlIkJ2e93WdfvKLtSEGzOCNyUmXnqrfUYmin3GAopRGfckOKkhCi8KUcO9qypTZ0\nnWmx5XbbqKR8TItzx9onNc7jUTHER4l54iLl2sZ6r9C1mzZvw52rVs+5+bIiPkbaqMyFmINTQk7l\nqKGv5YT23pYJW+MZlybgUZnYtNBQ0FLmcVjatyT6tgV99rEtW3dg45at2LJ1h/ralEZ8TEPQuljN\nGel0WTBWelrMqJDUhqGUgugRqIAzrJS5dqBck8OzPTRskVJyDQdBs3dCSXKHkHJto72XZkO4nBtO\nEhHmzZtXbZMRxUg/OM1GPT5yeXZLyvfJiZxUOUs0RnOtNAGX7PW1IGa0KwRpAp0rY0MD7UZeKfvc\n0r2ncfSRB2Pp3tPqa1Ma8TENVutiNXR+7Gez20LPQE+LSfvXtKF1Q6dQm0rOsJh5YjHhb15kXUNo\nrtfmc7roc76LqQNKdiTFlEOj23JGWKcmF2J8fHzPmeTnGEbaqNRs1ONDM6FoJ5++KAsloVSqnAVS\nsn/JXt9RgDSBpoyClgotbddST2lREnPnwpgoyWANnR+7D7r3cl/nYi3Lh7V9+4pSd5I7Up5YTMPC\nOs407e2331CvFZqlLCs05cXUASU7koaVYxA0/SRlhHVQZH3r1q2jvQDeg9HJqCSiZxPRHxLR22c+\nqQXrgtBGPZ0Gs2ZCUZybkrKQEhYF6L9vLCdluC9IUc9RqceowG/flDsqWiDJpZE7plEj3Vtynkk7\nDqdkJ/QVBdHq49D5Kam1Wjk7baYTCZqUgpgLY6l9Y+aJxWQQMNt2CA3pSYkaW1KqT1/zp1aH5nQk\nhWBlFMSEZdM8ZsbOnTsHTzIVxUM0KonoHwD8OoDXAyAALwZwSGK5OiG0UU+nzQI6ena1XuDQgMq5\nOLLQIbTXWik7LkY1382HxQjSYq5QbTXQLOSmJhfi4P2WZdkoRjLOLB54qWypz4n3VjjP1GUHIF1b\nSjTAcq/Y1Nq+2CAx29tq1Fg2XZqaXIiD918aRSdITitNe8emuLvlafWHBtprpY2U3PKsfU5t1Cs2\nLAudr31PuQWjum6qzve5hS6Ryicx88sBrGXmcwE8EcCRacWyQ+qoufLEfFg9VyElJC3yNJ5BrZyl\nUuP6RMxFt/TsSll054SaCproJeoD2z5gnJknUYXhp6HPaiM3mrKtUQlNNCBlf48ZebBGoftig8Sc\nwyw57hI61TES/VWijWqYErEp7ruV59S59x1yh9xVXIuc49bt39LrYUY1LcOCnCyJivzoYlRubv9u\nIqL9AWwHsF86kbojtFFPygRnS3nWBWVICfnH/IlMUmBum2nlTEmNs6BPg0qbDxeCNPmHqHUxPd8l\nRVs1OZjal6hr4Le9xGywTKJ+2VpjLXRv7fvxNGVrdI/2eMpcZg3VTQsrtbYvNohE45W+p1xIhwzW\nQfWIRX+V2kQyOrVya+iZbnkDdVMiZ5uEmLuK56TSaiKZkm6yUuJHEXuiIb0noYtReRkR7Q3g3QC+\nB+AOAJ9NKVRXWDbqcdFngrkWIeXoG5H+RJYyMdsSSUuJ2FQiDSzJ6oNPmv1ZhKh1VmqWxkC1IHbZ\nbpvkzB/JeS8pSqeRRWucWcqOiZS5zBqqW2zE1E3askLnS31Q+p6yL2gYGzFfkyC1Sczoo5aeGbrW\nN+xyGjUx2WIxdZUV7rOWdJOVEt8XLP2k0l3nNkSjkpn/nJnXMfM/o8mlfAwz/6l0HRFNENF3iOha\nIrqBiM5tf/8YEd1ORNe0n2Pb34mI3k9EtxLRdUT0eOkeoY16NIjdyVMmbmsiDZICizmBSPVyZZHu\nK3m6NYiZ66KFZsKQIpHayd+nModyhzQLnpQTgjYC4iNmxDXn4irmBG3pz1rjbC561a2RyZw61VKW\nxQkotZH0PeUCXzP3ap+t5nn48441+qhBaM4r1flrRc6N2CRYckFHRecump7A41csx6LpCfW1le46\nPIjoVCL6YWsnvXXA8UOI6KutDXU5ER3oHPsSEa0jostSyth199cnEdFL0GzY8zwienmHy7YCOIWZ\njwFwLIBTieik9tgfMPOx7eea9rdnAjii/ZwF4EPSDUIb9Whg7eSh6EDMXVElaCfvmLlIUr1cWbQb\nmVgmOskbndNrJt5ryEhkl/NDuUOaBU+fUTipzpp+ErMsCTEXclIk0rIo0UY5Y7Z3TFjuZY1Maowx\nCTF1k9b5YHGM5ozcS/d2YTVuNc9DGwW1GCKae6eMIvdp8FhehZISfvtaN0oqxeifmliI/fZdgqmJ\nfjf62pNARPMA/D0aW2kFgDOIaIV32nsAfIKZjwZwHoB3OsfeDeBlqeXssvvrJ9EI+gsAfr79HC9d\nxw1mdm1Z0H5C2vt5aBqDmfkKAHsTURG5mxJCAz3mrqgSrJNkzCiHfMFDCkXr+fYRUlIaY1cqy4pQ\nG1kikRKs77Cz9NmYjgrJG61pE+nclBRKy72sNK/Q9drxH7O9Y8JyL6ucoWi7tn1jzhWS80EzTmPO\nG1Zo5Na2Z0wquQbW9gzdO2UU2Sr3XKRUDpTLsDmUJQodE/fetx6Xf+dG3HvfevW1vW8ONbo4AcCt\nzHwbM28DcBEau8nFCgBfa///unucmb8KQP/AlOgSqTwewMnM/DvM/Pr284YuhRPRPCK6BsBPAXyF\nmb/dHjq/Dc++j4hmetYBAO5yLr+7/a14hBRa7F1RYyoOSy6SRQ7fyJGiAzFzSCT4nsSYlMoQpCid\nZYEjnWuROyWdypdb8kbHbJOYC/qUEScfFmeEVs6Y7W2BZWz411vlDI3jPhe+UhtZ2CExo1taXZTS\noM3qZHVg7Sehe8dmZbmwym1hTlkdBl2PaeHLZXXwaqLQKeFvzKhOJ+lpc6gRRxcb6VoAL2z/fwGA\nxUS0NINsD6KLUfl9AD8zTOHMvJOZjwVwIIATiOixAM4G8Bg0Ec99APyRpkwiOouIriKiq2jX1iJy\nokIKTdoVVYucOYAu/PbMmbslIXo+heNJ7Ivip925N4SUhl/MKJyEnIvylAsLH5r2t5ZlWXBaDYBU\nsI67lDo11Gdztp92nGrorzGjWzGj5dr+KjFmciEne0bb3iH6ptVgzRkBd8uW1jba9g+dH7uNuh7r\nAks/06STpNyJvRTs3EW4f+O46gNg2YO2TfM5a4hbvwXAk4noagBPBnAPgJ0RqyZiVqOSiP6DiC4F\nsAzAjUT0ZSK6dOajuQkzr0MTij2VmVe1FNetAD6KJqQLNJU/yLnswPY3v6wLmPl4Zj5+fHJRlpyo\nmLAqqL4mOmvuhVsPa2TBR8x8Ct+TaKH4WZS0dufe0L1SGn5aerEFOaNdOfVHX4a3NSpUio6NTVmN\nCbfPavOrJGien++kknRwaKyl7K8xo+Xa/hrTWI4JSe7stNGO9E2tXLGdzSG4Zft0TKtxqznf2kZd\nj3WBRZ9r0klKGVcFYvWMbdN+LvCOizYSM69k5hcy83EA/rj9bV1SqT2EIpXvAfBeAOcAeD6Ad7Tf\nZz5BENG+1LyKBEQ0CeDpAH5AbZ4kNaPo+WgioQBwKYCXU4OTANzPzKtC9wjt/pqLyqktO6aCyunp\ntuZeuPWIvRiNOfmEFldSe0uUMg20uweG7pXS8Mtp6KUs2+o0sUDDELCU5UPrkZcMgJDjKCViU1ZT\nIXZ+lUa/+E6qvqidUlkpaeeaha8VMfu/JFfOZ6mhb2rl6lVnOHRMq3Hrnp+SJRQblv6fM51kD8aV\nAI4gokOJaCGA09HYTQ+CiJYR0YxddzaAj2SWcXajkpm/Efp0KHs/AF8nouvQNMZXmPkyAJ8mousB\nXI8mCvoX7flfAHAbgFsB/COA35FuENr91bK4ignrYjXkNZMSni1KWhtN9He5DdUj9oI9l2ETk5ol\nQRt9Dd0rJZ2qJGqWpeyS8t0sfU4DySOvjWbtppsUUbicfTJ0PGV/9tvLml9lYVH02d9zQUvn1swj\nOY0FSa5SHGBWuSxMCGu09s6Vazo7XTXO5dgsIU09UzokK/KDmXcAeB2ALwO4CcDFzHwDEZ1HRM9t\nT3sKgB8S0c0AHgXg/Jnriej/Avg8gKcR0d1E9IwUcobor+uJ6IEBn/VE9IBUMDNfx8zHMfPRzPxY\nZj6v/f0UZn5c+9tLud0htmHE8u8y82Ht8aviVTOMmLRFqWztwBW9ZoGE55S5cj78XW59jKrCctsh\nJjXLh5XmHJr4UlI7+8pHS122ZYc6K63UvXcuaiZgp33tJqsiCpeyT2r6vzZ6a5krtEaP9vrQuTn1\ncc4IVAgx6dtaYyGns6IUaOWyOD4sz9J34IacfNp7xWYJae5dUhS0Ig6Y+QvMfGRrJ53f/vZ2Zr60\n/f8SZj6iPec13KQZzlz7i8y8LzNPMvOBzPzlFDKGIpWLmXmvAZ/FzLxXCmEsiDm55/QyWsqTEp6l\niS3UZtrFrL/LbcroS0pvnY9cuQCx83lCdGNLfqaPnEZPyrIH9skhd6iLEtGm/t/xZ3FsaKNwKfP0\nNP1fa1hb8yJDsOTS9WnISYvwvmSLGa2VNodLuaaYq7DoOsuzlMa8hdUSW39b2AkVFTnQZfdXENEx\nRPS69nN0aqGGQUylbfUy9jVpSnQejadWqwz9XW4XTU/g8SsOxaLpid3uZVV2mmdtXfRZJoU+jbMQ\n3Viqk6Z9c1GPU8NvI8sOddYos+befUZAQve2MjI095LK0vR//1innaUNeZEhWHLp+jRipEV4TNks\n/cKCqcmFOHj/pdFSIUqJ5mpRSvqD5dlK+sLCNovNdLCwE3L2sVHtzxV2iEYlEb0RwKcBPLL9fJqI\nXp9aMC1iLsqtXsZck6aWzpMzr2ZqYiH2e+QjMDWx+8LCOrmr5XYWfRYHgZXWGBQxYUTbSkPqCzkX\nxjHb31qW5vqSjAcLcuaoaSDlNlvzIkOQ+oEm4qqFNUcttAiPqV9ypnjs1iYKZ0JMR15JGNX0hxBi\nUsWl9sn53OfqvSrKQpdI5ZkATmx5u28HcBKA16YVqyxolVuuSVNrJObMq7n3vvW4/Ns34d771pvv\nZdlQwV/0WRwEUWiNBcLfZCmmJ9aCPtsv56RoacPo72hVQPN8pDr6jAJrjrELzbPU3lcTYc0ZtYhJ\nn9dCunfMeScm7VGC2yaxnQkhWUqO+qTUPzmp/y40aUISpD6mZcWllCXWfYB+56WKftHFqCQ8/OWZ\nO9vfikLKRWBfyg3Q7erZp5w+fDqsBTEXOFolHqKRau/to5TFgrTJUgg5NzbJiZwGrTXaoqHhxUT0\niKoT+bHkGFsM0py5zdK9+4Sl/+cc4zFpjxLchXLOubakfuEjpv4pJUdYywCzQMuKszjIJDpszH0v\nYr47vGK00MWo/CiAbxPROUR0DoArAFyYVKohUFJUKKYSKslQ7AvSi7s10CrxlO1fymLB32RJA22+\nlKbOfbZPrlxawL6AnxwffSPdj/zEjExqnmXK3GbJe1/SHJaS8pcSKY2PlAtlDSOpOETKKS4lRzhm\nmlBsdpPFQSYdd79b+1zxfbYiGUKvFDkUAJj5rwG8CsB97edVzPw3ecQbTbgDShu5mSuRnhC0dYj5\n4m4fFiVtfRalKF5LVFmKBPvUWk2dR3VzC23/tEZbNm8t3+mkpURa2sQyrrT3lfqgW55klMwVB+Jc\npa2nrFcpc4F2/RGTBhxyHudsnz51UUzquNZAjbnvxVzRZRV6hCKVlwAAEX2Vmb/HzO9vP1dnkk2F\nUqI+gO7lt1qqhYX+UAqs3ruUmzJp5LD2uVyKt086mk+t1dR5VDe3yLkAsuxS66NUfaFFqdTEUgyH\n1Cg1XcSKlPVyy/bHYUkbrMSmh7sIOY+rkbI7NA6tLsdrG1fEQMioHCOitwE4koje5H9yCdgV2slE\nw99PmRzte+cs9Ic+JyNNm2mTuEtRftb8TB+WnWU10PaDmLJYqLUSSs37Stk/LZtWDbrexahsTqQt\nq5TNLzQ5TXsKrHW2jocS4Y9Dq6GsaeOY9EutXCmdxxo5tMdDiM1uCpWdEpKu2hN1V8VghIzK09Fs\nyjMfwOIBn94xNkadk5B9aPj7kmII3Uua5HzvnIX+IE1GPhXRAr/O/u6NoTYbpSRuzbPVvhPTbaPp\nqXEc9egDOl8bc6EQkssqR8wNm2Ii5oScc0KV5JZ0Uym5W9r+HoKV3WHpCzFzD0uNvKeEtc6W6/tc\nCIfuHZv2r2mjmPRLrVwh53GfTlfLu659B7pWf4fQ52ZyVXdVzIaQdbEPgHcBeDUzn+t/MskXxMIF\n84deKIQGpOQxizmgrNGuUFkSFdGCgXV2EvZDbTZKNDD1s1VsWrBbO1D3iTrmQkGUK5IcsRFz3PUl\nhxaS3JJuCl2vdcyZoejvIVijK33pI2me2RO8/9ZXD+TcRCUmQveWNpPTyl3q6x00cvWeKzvkZkS+\nA12rv0PIGZWXdFXsCPa8efPskY8esWPnGNatX6j6zBXQbB2SiD4E4EQANwP4EoAvMfNPMsomYt70\nMuZDnjFwUBHRgwZM7EE3NjaGZUsWYfXaDdi1a9du97Lce3pqHMv33xd3rLw3+gu1fbk1kOoo1Tnl\n80gJjdwx21d7fsw+WOJ9u8jSF0qRY5AsKXWRpeyS2iwlSp0LciLUBn3WMWUflMrWzBXWMV1qP9Lo\nFyJSza2aNrKuXULHrfO6BaOqY6enxsE/uPiGjRs3PrZvWYbFYYf+DL/r3N9QXfOiV/z1d5n5+EQi\nZcOs3gBm/m1mfjyAcwAsAfAxIvoWEb2DiH6JiOblEnI27NrFWWgaPiTKaikRFB8WKqKGqjIIlufR\nJ38/JLcvh4XWq22f2N7srpD6QU7vfyn5U6XIMUgWi2wxveol5b6VRFcOYZQYHS76pFxrnm3KPig9\nd3+u0KRZaOW2RCpTjhWNftHOrRrKqvSspPbWRJ0lxFwnxaTa5sSmzduwdevW0VJ6FQ9CDDEz8w+Y\n+X3MfCqAUwD8D4AXA/h2auFKgZRQ7sMyUeZcbGmUijQx5cx58CeMvpSjhlqYGiE6Ssz2KZVKWBEf\nki4aFZp0Tlk0c4U0LktyVmhgoVzHvrcLqb371JMp+6TF2WlxHEnQ6Jeh5pUAZdWV1TpnlTLnadcj\nJelkF8yMnTt3lrUJQ0VniEYlER1GRDPunhMBHA7gT+dCmHY2SB4fIsL0ZBojRqOYrZOgRqlMTS7E\nwfstGzqHNQSpHgOVozNhWBLpNbJJC0btAinmIibkzY7pscy5kYO2ffpyLpTq8Y0NS7SxlIVXalmk\nSH7oXAmhfpbTYJKgestsIgAAIABJREFU0Yux5Qo9W6m9B87zQ8qm1ZOpnIB+2dZrQ7LF3HwLeHgb\naecV6f2Z7rO2zlmlOH+kcacNjlRUDIMuybD/DGAnER0O4AIABwH4TFKpEkCjqCWPj7ThjcZbqqUs\nhO6jpYmqvejOxhoxFZSWfjJwwhgykV6Ca7BqFoxdkNIQ13hi+/RYhuqhlStmPSz6Yq7CUk+rwyWl\nAyYmNHpQqzND+l5rMMWExdkQW67QvUuKHvpw5V40PYHHr1iORdMTQ5UVk2oupTfsNi4Vm2+ldHTE\nZFnkREpHRuz1iwbaYMmob9SzJ6PLg9vFzDsAvADA3zHzHwDYL61Y8aGZIHyqpz/4pHfvabyl/ncN\nzVSaBC25Av61/kvWYyoorYLvZGTGBKfx7KU0xDWe2D4nWFdOq6MiZj00+qLUBUps5Kxnrhzh2NDo\nQa3ODOn7Pmnplghfzj5liR7mxNTEQuy37xJMTQwXwU45VkKvyPDXCBJiRzY1KCW66CM74ysTtMGS\n8fHxshV9xazoYlRuJ6IzALwCwGXtb+W/YNCDZkBJOQjMjI2bh6MA+nL4Slq6d8hYCOXVaSEZ1jEV\nVJ/0ky6K+M5Vq7Fp87boE5GlPMn5oHk+fXosXTm1joqYHnkflvabq3TYnNTmmLosJyzPXptj6bZJ\nTlq6D+ldfKGoc0mL+5T6RIN771uPy79zI+69b32n8yVmVUx9FHpFhlZfNz/madtR0cmaTZW0zoQ+\nx5qWwVE36hlddDEqXwXgiQDOZ+bbiehQAJ9MK1Z8xMz/iUkD077HSLPpgWWx28mwngO7N0rP0rLJ\nQUqU/HxC8KldrpxWOmAI2j5jaT9t5CZmfy5l8STJoV0A9Zmrq4ElaqfNfc6ZqxiCNIdZos456xEz\np1JCqGxpl3b/WskBHDNyKd0rhEGsrNVrN2R5f2bOZ6vBIMMwtHeFC8mZUBK0DI66Uc/oosvurzcy\n8xuY+bPt99uZ+a/Si9YftHRBi4LSbvZiWUiMCqUvZj0kSPUs9cXRMZ9PTqMmRO2y0gFDyEmflBbV\nKelqpdBEJTm0/TenI8kCv14uxc9KT9NSyPrq7xYGTU4qp1SPlPe2bC7nyyU5GGPOFRYnqy/H0r2n\ncfSRB2Pp3tMA8r6+pGg92TF6q107pkQpRnpF/+iy++sRRHQJEd1IRLfNfHIIZ0HOJPCcG1iEIC1e\nfQMp1EZ9UiJjL0hDkKK5U5MLcfD+S3uffHyk7Dcxy/LbU0Pt0tIBQ+gzd0ta0MSUrRRvtSSHlkUR\nc4fnlI6igX2Sus0TMXOfS85VDJ2fM/oiRX6TOxSH3FxOSw3X6MmUeXm+HGvWbcR1N9+JNes2Akgb\nTbSk7+Q0djV5qSWxkSzsg2qQzi10ob9+FMCHAOwA8FQAnwDwKekiIpogou8Q0bVEdAMRndv+figR\nfZuIbiWizxHRwvb38fb7re3x5cNWCsjbyVPmMITg30davPpexlK9dZZJ0tr2AzcPSLSzbClIadT4\nzzZE7ZKcC9pnW8pkJS1oYi4OSvFWa+XopIsijcNQtCV2n3EXidI4i+lESYmci+yU+fJ9pj5YNpdL\nSQ3PmZc3MT4f0xPjmBifDyBvNNFC202JUsa4Fpo1hL/GKmUdOgogolOJ6IetnfTWAccPJqKvE9HV\nRHQdET2r/X0hEX2UiK5vbbKnpJKxi1E5ycxfBUDM/GNmPgfAsztctxXAKcx8DIBjAZxKRCcB+CsA\n72PmwwGsBXBme/6ZANa2v7+vPW9oWKhx2k6eMochNDFKG5tIHs2+ohqSMRwzn204AR+6b/KdZQtA\nSqNGPdk40SjrIsM9P+XEpV1kW3KdS4aljaV+EnMchu4Vu5+4z1oaZ6NCgy5lsS/ByngpJeqfEznr\nrKVN94WUcswVg0o9bp01VinPuXQQ0TwAfw/gmQBWADiDiFZ4p/0JgIuZ+TgApwP4YPv7awGAmR8H\n4OkA3ktESV7b0qXQre3NbyGi1xHRCwAMfkGjA24w886NBe2HAZwC4JL2948DeH77//Pa72iPP40M\nqywLNc7ayWMOkpDS0d4npddXA8kYtsiijRr7x336yah6DkuBuv2caJS0yNBQtVJOXFZqpmVhURKV\nKCY1Tns81r36XOCMCg16VBaBVlpoTNqoj5zGRClOK4lu7CMlI0mDlGuAmGNJu9bpa+6oa6yhcQKA\nW5n5NmbeBuAiNHaTCwawV/v/IwCsbP9fAeBrAMDMPwWwDsDxKYTsYlS+EcAUgDcAeAKAl6J5vYgI\nIppHRNcA+CmArwD4EYB13Lz3EgDuBnBA+/8BAO4CgPb4/QCWdquGDbHpJO711oEaUjpWz3dfGz9Y\nFWlIFm3U2GrgWlDKZF8KpGiU9tlqokRmGKiZlvFgZVloMCr0TB+acSbVIeWYjXnvlM+i1OfsI6ec\n2nGXM1e0r3neIod/vbUsSba+ENPRr13r5Jw7XIyK/igQD9pILVz7aQbnAHgpEd0N4AsAXt/+fi2A\n5xLRfGre4PEEAAelEDJoVLbh1l9n5g3MfDczv4qZf5WZr+hSODPvZOZjARyIxsp+jFVgIjqLiK4i\noquwY4u1uNnuYfLouMelvLCxsTE8culeGBsb/CgsVLmY9J6Um+NoEVPuPnd3DUW4Spn0ckLbL0qJ\nmFipmZaIiN8GKftzLlp/l+Ma5KSVpvT+D8z3znDfUYW2DSxtlpM1JMGy2VFKlpW27FD6QkzH9KiO\nFW2bxEyHGtU2KxzLZmyb9nPWEGWcAeBjzHwggGcB+GTLNP0IGiP0KgB/A+CbAHbGEtxF0Khk5p0A\nfsF6E2ZeB+DraN53uTcRzW8PHQjgnvb/e9Bazu3xRwBYM6CsC5j5eGY+HvMnrKINhD9YtYnF7nEp\nL2zZkkU46ZgjsGyJyCjudG8XMek9JXmXYsrd+3soZ4lwWTepyQXJKZISpfTJmPRtCVJkXerPORfO\nIUiU4ZKYEZqy/HrFpDY3P/LA49aog8UoLVU3adsgZLT78PVeKboIsO1RELMeW7buwMYtW7Fl646h\nyg6lL1ij+m55feY1xtTH2vWehaE3V3JBU2HnLuCBjfNVHwCrZ2yb9nOBV+yDNlIL136awZkALgYA\nZv4WgAkAy5h5BzP/PjMfy8zPA7A3gJtT1L3LSvBqIrqUiF5GRC+c+UgXEdG+RLR3+/8kmuTQm9AY\nly9qT3sFgH9v/78UD9FqXwTga9yThh64cFAkFrvHpbyw1Ws34Iprb8HqtRsGltVJtopZoZlcciMU\n4bJuUpMLWqfIqCDlQjnlhjbSccu9oy+cA5ThnMyI6LRSp14xqc1+LlLIeWmJCg36rrm2FAzV9h3f\nESjpvT4N7VJSOPz3UFqgrZMmNaLPNUBR+liBnLmgFQ/iSgBHUPMGjYVoNuK51DvnTgBPAwAiOgqN\nUXkvEU0R0XT7+9MB7GDmG1MI2cWonEATMTwFwHPaz2kdrtsPwNeJ6Do0jfEVZr4MwB8BeBMR3Yom\nZ/LC9vwLASxtf38TgN22y80Ff7BqE4tDx/1joVcsaMvWotTBHJ0y5kQOctbZci/JGdEX/Dr57xqz\nUsdLQcpIccwNbXy5JP1QSj/S5s+mREyjyK+XpR5SJCLkvLRSyS2USctY6VM/aN4RKDmDSzK0+3KQ\naR3mmrIlaPrkqBpnozKX+vDlLmmslAxu9pp5HYAvownQXczMNxDReUT03Pa0NwN4LRFdC+CzAF7Z\nBuceCeB7RHQTGhvsZank7GJU/lObS/ngBw8ZgrOCma9j5uOY+Whmfiwzn9f+fhszn8DMhzPzi5l5\na/v7lvb74e3x22xVi4ec9LacKHUw+3KZKZZO5GAgtTnSS9V9xNwoqRR6lS+z/64xy2YAOfPsJEiR\n4pgOgxCk+2jHsKUfxWx/IsL0ZBk6MmVUNNRmUnvGTF/Qym25Vpt36iLmONPQWQfVI3RvyRnsUz/7\n1G0p5/nQ2NE6zDVl+5Cca5o2yDnPWMZdn+s3y7opdn7sngRm/gIzH8nMhzHz+e1vb2fmS9v/b2Tm\nk5n5mJbq+p/t73cw888y81HM/MvM/ONUMnZZpf9dx98qWpRqrPkodTD7cvlUI43S9yMHA+sc6aXq\nPmJulFQKtMn+mo1kYhrhVkiR4lyySPfJudFUzDqXRJtO6bAJtVlMh1bshXBKinao7OjjrCOdVSun\nBJ/62aduSznPpBw7MXP8NJHLUdnPoPf1g2Ld5LZh73JXJMWsRiURPZGI3gxgXyJ6k/M5B8C8bBJm\nQGyl0deg0XpDS4l++fDl8mk0lgjfQGpzx507tf2kz/bVREg0FFVt9MT/HtpIJrYRHnNc90VHlozG\nnBtNxayzT5vWwGcu9ElpN+dsR3JoxTZKYlK0NWVrx1mo/TV01kFlWWi9/pzVp4OxKHZColQIqf00\nkUutY6Mvo1N6rinl0u547rahJYpcUT5CkcqFABYBmA9gsfN5AA9ttDMnoKXJSHAHjZX2olEMkodt\nVAevT6OJGZ2J6Q3VIvrOkB3L1lJU3UW8tU9pFpTa41I9QyjVYSAZjTkdWDHr7NOmNfCjnDEjC9px\npdkQxIf1VTQuYveDviJQWsdnqP2t+sKy+PXnrJxU5piIOUdpx44G2vaL6djIua7S6LaUqT0x27vP\n17pVxMesRiUzf4OZzwVwEjOf2/7/52hyLG/JJuGQUEdjDDSZEKy0F43Ckjxsc4V20NdrQGK3X8yd\nIX2EFLWWsuou4q19KufiSSNbqd5orQe+VKqWD0vfl6JAlkWelkZXik61jitNvykp/y/mRidzMV1B\nC22bhCDNFX22p8bxb4m8x4ZatyVK7dEi1Ia9v9atIiq65FS+k4j2omY72u8DuJGI/iCxXGb4gy/0\nrkktTUYD60SlUViShy0nXSLl+wv7moxiG0SW8qQ2CClqiaIaoh9r+1Sf0MhWkjfaRc7orHb8W/SF\nxRiWokAW/SB5zaWxooG2T8WkD0pRpJD+jjkWrEZMzPYvJZrYp2NI2yYhWaW5QtOesXWTezxlRK/P\n1KqYTIiY8NukRirnFrqs+Fcw8wMAng/giwAORcLtaGNhYEed5V2T2slCoyisE1XORbs7oViVuGYj\nDq0BmjJHJOWErslzlCC1QUxF7S7iU0bG+nw2EkYlSmGJzmqjtTGNiz6pcC5KzlGNSR+Uokgh/W3J\nNZTkKInJ0Jdus6bjWOYZbZ8MGWQxn6W2r0ttuFt5hoieJtXER8q0C61DINf86rdJjVSWAWr2zDlz\nwO9nEtHvdS2nyyp+AREtQGNUXsrM2wGUF5bw4HdU7bsmQxiV3EQLlWgoJe5MLpp3VOXcCTIm3Tjm\nvWPfN5eijpnPVuqzAcqJyFqpWi60dDQL9Tn2YjYVSs5RjUkflKJIIf1tyTWU5MgJS76mFmrdlmjX\n2iR6LgPFcqh+EmhDtzxrRC8kW6kMmEH3ziVLSTToiofhNwB8YsDvnwTw6q6FdDEqPwzgDgDTAP6b\niA5Bs1lP0bDQ9GIugEYpuuK20VCRLmdy0byjSjJAY3rUpHqlVHCWySfmvSRo2jdmPlufOU1Svygl\nV9FKmXShpaNJOtW9lzaq2afR7spqjcSHdJW1D8VsI6ksjf62jMtSnDWDEJPtodFt1nScUCQ5ZpQZ\nyEex1EbdpDZ0y7OmBYWuLykfU7p3ShqqZh6q6A3z26Dhw8DM2wB0VhCiUcnM72fmA5j5WdzgxwCe\nqpM1PfxOa1kcxFwASXz9lIvVmDQwSU7L5CItYGJ61KQIXkoFZ5l8tNTQUHlSWZr2jZnPlpIKLjkm\npH5RCjvBSpl0EZt65d6rz03CtPXSRHa0YyfUJlaU0ifn6qLQwvawjC1re4Yiyf4xK9W2lDzTlDTq\nlOMsZeqV9t6a/m7RsX6fK8VhW4ExInqU/+Og34KFSCcQ0SOI6K+J6Kr28140UcuioObQBxB9ARSg\nh+RcGKSMQKWcXFLSvlIiprKMSQ2Vzp2LkQfJMZEzSmrpF1bKpIvYuse9V58bOllo/9Ix7dgJtYkV\nuXTZXFn0aethad++qIWDINZDQbXtqy+knLNGhZoPpO1HoXpa8+l3K9vpc6U4xyrwbgD/h4ieTESL\n289TAFwG4D1dC+lCf/0IgPUAfq39PADgo3p5M6Ajh16CdQHkDkApgpdTYfUVgbKCiDA92f3dZbtd\nq3jvmVnORBuZSLQlzfORzrVEOS3na6OxGkiOCUuUNGZkzC8vJWUy5Zju07kg1cuiE7RjJ2YESrpX\nKlj1WClGqbYelvaN6QiNqQMkmqg2Ipjr2VrmLAmjQs0HHk5Rtba9Rg9amSdu2X6fK8lo35PBzJ8A\n8KcAzkOT8ngHgHMBvJ2ZP961nC5G5WHM/GfMfFv7ORfAo/Uip4WkHPvylKek9GlhyQXtU7H6G/nE\npDLHhFbxaoyHEKVp0PEQUk7A0vmhemqjsZp+EFrsd4FFbh9SvwhRhTRySYg9pkvxOEupD5aofsw2\nK8XYkmBd9JXSL3IuXq36xkXMPGpfn0jzig9LDn3K/i6lN4RQslHjt69LUbWOK818GpN5EnNslIpd\nuwjrN46pPiWAmb/IzE9m5qXt58nM/EVNGV1qspmIfmHmCxGdDGCzVtjU0CrHlChVSWkGbykLAWD3\njXx6pTIr7mXJQdPeKyZCE7T2vppFiJaC2qfDICSXBMnoCVGFNHLlRql6z+Jl10Y9NcdLenYhWBd9\npfQLadyViph51ACC+kT7nlaNbJb+rnU4WlhZJcFvX/f5WMeVZj7tM/ezIh+I6JlE9A0iWt1+vkFE\nz9KU0cWo/G0Af09EdxDRjwF8AMBvDSNwTkh0QQ2015aspLqilIUAsPtGPimpzH3lu1kNk5gITdDa\n+2oWIdqofs6XJlvklhCKhkm7GJY0Ti209JSLkJT5nZZc55KeXUqUOh9aIoAlL5pD/UrSJ9oN+mJS\nxy3XxqQb54S2fd3nY3WSWBwEUj1GxWFW8RCI6LUA/hwN5fXR7edcAOcQ0Vldy+my++s1zHwMgKMB\nPI6Zj2Pma4cTOx9iRi73xAFS6kIAiLsLqI+cz9qtR0l5Gykn6Jj1zPnS5JTPJ6ahPSrIuZFJSuqh\ndrEbS649FTGNOUsEsBSWxCCEjA2pz6Wkt8akTErHR2VsaVM6Qn3W2idj6slRMeorHobfB/ArzPw1\nZn6g/XwNwDPbY53QZffXvYnoDQDOAfAXRPR+Inr/sFL3hVBUY2xsDI9cuhfGxgY3R8yopxYle0RT\nIWWdrYtEDUbl2Ul5YyknaEsbxfSs9jmm52KennZx1OciRJNL5COm0Z8z52xUEdOYy8ksscC/l7Re\n8WGhglroraPS56xyhq7X6kHNpkFi2kRGpGSDVGQDMfN9/o/MvEZTSBet9AUAywFcD+C7zmekEIpq\n+BvB+PAHiPTuSQ36jJyVqvRT1lmbM2KBpR45F5h9TkaWNrI8K3/Diph9zjqmLc/e100x+5GmLGmD\noZSOC22dY+YSaeDLmfLZpexzOdGnruqLWeLfS1qv+HDbLGUqj/9stOukvvpg7A1vXEhtENOIz9kn\nUzpGK3rDA0R0jP9j+9v6roV0MSonmPlNzPxRZv74zEcjaQkIDVZ/I5hOyi3w7kkNJIWWMm+sVFpv\nyoVDTsqkpR4pn02fk5GPXIvEgWPa2bBCy0YIHbdGw83P3tFNMfuROhKheP9dTJS06VUIA+UMPDvN\nolvrOBqVKJM1j0yDUtvAX69IcNss9rwiUmsV66S+1iOxN7zZDYo20PTvUXUGVxSLNwO4lIjOIaLn\ntJ9zAfw7gDd1LaSLUflJInotEe1HRPvMfIaVui+EFs7+RjDSgJHePamBpBhSGkE5NzrRoK/8tdiw\n1MO666Sl7JzIZdD6Y1rasCLlBizavCYNfN0U81lrqFqbNm/DnSvXJOm/WjlLNQ58Of1n5+vnUHtL\nUc+UuXR9QtMmEvzzLWWn7HP+ekVzr9hzvqgDVq3urHv6mpesc1DoeutasS8WhYSS1hAVccDM/wPg\nRDR24SvbzxiAk9pjndDFqNwG4N0AvoWHqK9X6cQdLVgpkholn3KBKUFjsPaZgyahJFk0CMkt9QsL\nnU3r7Z8L7SvlfGg3Goi5AYuFShSThiQ9Zw1VS9ItEj3WAunZ+ohpmIQgPSv/u9+G6k06DBGSXLnl\n1vaNuXGJRgdY+lRsaO4V20kdotZq7zUqFMqY6zsJpRpvKeeZiv7AzD9h5rcz86+2nz9l5p9Q8yrJ\nTuhiVL4ZwOHMvJyZD20/jx5e7PIxNbkQB++/NAnHXouUilbjtfTrlHJRqIVGlpR5S1pY+ok02Wja\npKQFUky4cmudNzE3YNHIGftaTdlaOUJt0GkxlIkea3EQpKQPS9D0SV+XWyMkufq3tX3NfTBwvqXs\nPjfyiXVuF4SotaUaRFbknA9HxdDWYFTXE3MdRDSPiM4gorcQ0c+1v51GRN9E8yrJTuhiVN4KYNMQ\nAh5ERF8nohuJ6AYiemP7+zlEdA8RXdN+nuVcczYR3UpEPySiZ2jvGRWGnMmQMi3JS6PxJA40QJ1F\nYZ/1IiJMTizofu9EOWcStHlOIXSabGhwJFKSI6acMaHtYxq5+9woxtK+MRe3MZ9zFyN+WHqstn0t\nDgJrm4Si5Va5Xfi6vKTFaKjeKXWLtg005+d0OsWcW3OmlljvVdJayYWlz5Zap5woNe1qFEBEp7b2\n0a1E9NYBxw9u7a6riei6GfuKiH7DsbmuIaJdRHSsd/mFAF4DYCmAvyOiTwF4D4B3MfNxXWXsYlRu\nBHANEX2Y2teJULdXiuwA8GZmXgHgJAC/S0Qr2mPvY+Zj288XAKA9djqAnwNwKoAPEtG8rhWJiZRe\nXqsRE1MpaQa3v2jxc9Kkemm3QQ9hEI1s85btnSYvKW9Jg5Rbu2vht4n7fPz7Shv1+FHOUhaoMSNp\nVoTGYU45Yy5uc7aX5NCKGd2y6Exrm2ii5RakXKhZ55xQvXPqlpIW9BpZpP5eCmso9rPM5fCN7aQK\nIecaoFTk3CxxLqG1h/4ezXsjVwA4w7GpZvAnAC5ujcDTAXwQAJj50zM2F4CXAbidma/xrj0ewNOZ\n+WwAzwJwGoCTmfnfNHJ2WQ3/G4DzAXwTileKMPMqZv5e+/96ADcBOCBwyfMAXMTMW5n5djQR0hM6\nyBcdktKwDF6rZzamUtIMbskLKdVLuw16CAM3XenoBJDyljSwbO0+qB4WhAxF/74lURM16DNi6o95\ny0Y9eyL89pKMoJjRrT7pVrn6gq/HYi4wS6KrWepVUj00suTU1yUZJrnGjv8sJGaPBSnrVFL/DqHO\nj0PjBAC3MvNtzLwNwEVo7CYXDGCv9v9HAFg5oJwz2mt9bGPmXQDAzFsA3MbKd1QCHYxKdl4jwkO+\nUoSIlgM4DsC3259e14ZmP0JES9rfDgBwl3PZ3Qgbob3BMnit3ryUuzlaINVLuw16CKW0gWVrd+u9\nfWg2jpGelbQzqgV9Ro0s0OQLzZXoS8rFlGQExYxuacZZ7AWlRlbLvTQ7xWoh0eNzwlKvmNFcaxto\n2jSnvra8g1tqk5wRQQ0kZ69kdIag0Wux61EqSmE+jSC62EjnAHgpEd0N4AsAXj+gnF8H8NkBvz+m\ntcuuI6Lrne/XE9F1XYUUjUoiOoKILqEmN/K2mU/XGxDRIgD/DOD3mPkBAB8CcBiAYwGsAvDermW1\n5Z1FRFcR0VXYsUVzaTT0OXhjDshQWb4ytEw2wO7boIfuJSltX+6YRr5mwgjVaZh7W1BKWVL7jSr9\nR5MvlHPRnbI9U2445renVb9o7h2CtKBMCcu9NDvFamHVt5bx4F9rqVdM2l3sDYb6dFTvXuBweksy\nvix1TKlTJd2kZRi5so4SVb+idyybsW3az1lDlHEGgI8x84FoKKyfJKIH7TwiOhHAJmb+/oBrjwLw\nnPZzmvP9tPZvJ3Shv34UjSG4A8BTAXwCwKe6FE5EC9AYlJ9m5n8BAGb+X2be2YZZ/xEPUVzvAXCQ\nc/mB7W8PAzNfwMzHM/PxmD/h3ivbQq6vXK2c9xqoDJ3JJiW9yrpLYkxZLCiJSpQLUvuljBaE7p0z\n4pTTEJH6vsWrLkVTovfvRPpFg6Go4onubbm2pIiIJdKTMuJaSlkpyhsWlv0kJOPLUkeNITfMcRcS\ns0eqhyurts6l5MdasSeufVzs3AU8sIlUHwCrZ2yb9nOBV2wXG+lMABcDADN/C8AEgGXO8dMxOEoJ\nZv6x/wHwOOf/TuhiVE4y81cBUFv4OQCeLV1ETW+6EMBNzPzXzu/7Oae9AMCMxXwpgNOJaJyIDgVw\nBIDvdKtG3oWctEGLO6C0gyunURO6l7/437xlO1avW5+FXpVyl0StLC5KfpalQHp2oWiBtX1DRpBl\nYatFzgWi1Pctr5aRoimW/u1HJv3FbF/5QVqqeMx+ZNFjOaMUVvqxJZ8wZ0TPQkktleoZc8M8H5Lx\nZamj5IyU+kVK1oUPt95D1XnI/NjkTr/AvXyMSn7niOFKAEcQ0aFEtBCNgXipd86dAJ4GAER0FBqj\n8t72+xiAX8PgfMrZcJ5WyC6aZWsrzC1E9DoiegGALjuTnIxml6FT6OGvD3mXw9F9KoDfBwBmvgGN\nhX0jgC8B+F1m3tm1IjnzJaQNWtwBpfWwpYy6ae41NbkQB++37MFrl+49jaOPPBhL956OLmdOaKm1\nFiqLJZfLR6nUIe29pT4X8z1+IY9xykkvJ5Ohi4e+6+t2pP4qOX+0UdHJ8dnP7SuPTxvxyOms6Aux\no/wavWi51kdOp2CpESfN2mUQLBE/CyTqslZ3pRynlnpb8mNzUvelskd1fVgymHkHgNcB+DKajU8v\nZuYbiOg8Inpue9qbAbyWiK5FE5F8JT/UEX8JwF3M3Dl9EYB6gHQxKt8IYArAGwA8AcBLAbxCuoiZ\n/4eZiZmPZuf1Icz8MmZ+XPv7c5l5lXPN+cx8GDP/LDN/UVORnPkS0gYt7oDSUCWAtFE39b0cj5lf\n55hy9qkMpcm9o5KZAAAgAElEQVTFjaikjKCmVNLW9rTQ1aQ+55atnfgtHuNRnfS0Y4W5++t2pPaU\noniafsbM2Lx19mstZVugjXjkdFaEkHKRHLtOfUVks6dVFLgjq2btMgh9OTcluULzyiCU6qS19O+c\n1H1prs7JnNiT0NpQR7Z20vntb29n5kvb/29k5pOZ+ZjW5vpP59rLmfkk5S1/Uytjl91fr2TmDcx8\nNzO/ipl/lZmv0N4oNWIOICnqyczYuDnOu9+0cls8hZp7+R4z66Y0IUqwL1fKd64NtQicZeOCmJOL\ndtKMWbYEC11Ns0iJbUiE2mxUJj1prEjta33nrgauLNLY8PWLJtKQkualjXhodGxJhp9GlpIdMCnr\nETPiZHn22uhh6F7SPB57vWKph0Yu6V458zv7guT0S4k9gbEx10FELyWil7m/MfN3iOhlRPSSruXE\nJ9ZnxNgYJfGM+NTPQcdj8fetylIDzb1iK6QQJdi/V8yos+RBkzxu7qLcl9unOPWZwxCC9Vla6GqW\nRUrJi1kXOY0FbftqPfgahMrqEkEdth4pmQ3a9tQg5WJUO1ZSzlEWpKSo5qxHTAeZNnpY6gI/ZRRU\nulfM/M5RmZdSolTGRoUJrwfwrwN+/xc0tNpOGGmjcuGC+bN2WnNCeoC6ovVmx8SoKjQNJThmHbWL\n29AifqBcTj+xKNKYjoqSoaHJ9BlN1OiPkowFC+3LuqB3v8fWU65sgzYRCzEbtLmelgV4qH1zslIk\naGTJaZRYKaqlGFA+LONB65hLucBP6dSOqTNSRu1ilpWSdZESFsaGj5SbR1WosICZd+PFM/NGAJ0j\nPCP9FLdt3zFrp913n8V4yolHYd99FqvLlZKlLd7s2JsehFCSgnLlThkN8KFddISUoS+XlsKnkVN7\n3EeontbNSCyIWZalf0vXShtaWHJBNYjNZAj1I+uC3jXupAipxYD1mQxWZokL7QYrEj3ZRUpWirY9\nNRHsnA4tK0VV4zSZK4tZyYCKmU6S0qndV0S8z3WSlnVRChsq5hrOn2uJCPPmzRvtQTmamCSiaf9H\nIloMoLPyFx8cER1JRF8lou+3348moj9RiZoIu3bx7F6vLduw6qf3Y9OW+J5BC3JO0DnvlXKBb4GW\nhmShCFv6TWxDO1RPaeHsX+suSqzPKmcUWnOtXy9pQ4uUuaAWWHJzrQv6EG3dSlkVmQ4GZsluCJQl\nOVxSOsRCsO42GjPCakFKqr5fR8lxFBMWnWDVuTHTSUqmQmsQSsfJCW2OvP8qJg0sLJaYkObaqcmF\nGB8fH21K1mjiQgCXENEhMz8Q0XI0ryC5sGshXbwB/wjgbADbAYCZr0PzfpSisWHjFnzvxtuxYeOW\ngcf78k7FnKAtUbcu12uQkvoZE6NKH5YQipgMihKFXjPht5G7KLEuXnMuukP9W6KI+RtaaCJSo9LH\nNFTkQef70LSJxYCVGAOhayVIZfn9v8/IzW7Pw7DbqCbCap03cjoRXfh1XLNuI667+U6sWbcxuhzS\nvTWwzo851xwxy0oZqXcdpX3q66Ec07NsHCjBwmJJuVb059rNW7Zj+/btO8w3qlCBmd8D4N8B/DcR\nrSGi+wB8A8BlzPzuruV0MSqnmPk73m9FPHB3ox4f0uDsi4ZnXbi5x611iOk9tSjmmPQcidLUZ55e\nSoQiJoOObd66rXNUdLdnG2mrfCs0Y1wyoCQKqyYi1Wcf04xp7fi3RPm1i6eU+YTmspz+X8qzloxh\nCZp6xJx3cjp3/TpOjM/H9MQ4Jsbnq8uy0o01sBo9MfuoxPBwYTUaNfXW9knXUTpKawLLbt6d3mow\nSzvEpOVKc+3kxAIsWLBAPygrzGDmf2DmQwAsB3AIMx/CzB/SlNHFqFxNRIcBYAAgohcBWBW+JA9C\nG/VIKIWGpy1LsxlGTKUtlW1RzFOT/5+9d4+SLSvrBH9fRGZEPm7evPfWi4KioKBLEGwEpkCmtXup\ngI0MD7W7kVoLAXVAW7B94Mzg+AJ7sdruadslM+qAqxW0e0Ds9kGPyEPaxnZEsFRAgVZKuoSi3nXr\n3ps382ZGRsQ3f8SJjHP2OWfv/Z1v731O5M3fWrHujYxz9v7269v7e+4Bbn70NUH6L6VLU5cgsRJJ\nD6D5sQ2ZKj828u2WCkTm820lNpGWJaFTuv5TJkWRxMalhEZ4C013fjxSHoy1e6dkXZoIGWbRpvWw\nLaWJFi4Pjzy05w9Ju7W8LBW0PEAzF0w3aI0gaEKyHnz22oODg267+RxDENEPzj8AXgPgf6bZdSK3\nSMrxESpfB+BtAJ5MRF8C8P0A/qmY4gjIJ+pJqTk0EZJBucqSHCSkmqmQTEaMhi4dJlyxcCbaPKBK\n6nY9G9JKZIM0fjBm/0r6RHvoSGnJ0XgjSOjU8sCYbus23pVSEDEh7bOQniVaWkKhzb0zZJiFxlqu\n3f8kFj8XJN5NWtg8PKRCe3IPg0h1S9DmfilRCGjL1rzLzJhMJs0uRD+BBlvG5zSA2wD8LhF5hzw6\nhUpm/jwzPw/AdQCezMxfw8x3NSI5MPKJeqSLtc2DRqy6qtwIJJopSRC4ZkM2XVQ1Lh0mXJc7u9wa\nbQjN8CV1a2MZYyGkBtNEyJiblIeOkAfOkK7hgG4Op4wntCUXSSmIaCHxLDkukKxb6ToLeXh1wTZv\ntHu+ua41lnmJd1NouMIsumJhNZHv05SK5Zj7pQvOsBYLNKEPUrpO0A6Y+c0Vn+8D8PcA/K++5dQK\nlXlTaGYO/S4Ar8l97xSki7XNg4YNGrpchz4fwXt9OIgSjJ2H6aIaMr5KSlfMuA0XxAekjsQy5uEa\nu5guZl09pGs3yXy7QmZuBALM4UTxhJLEMZqyYqMtF1UTKQ/OMddtysNrSqWJrS5tf6aa/13lx1XI\n96kmo6oUMffL0LS0RdcJugVmPg/Ae+OwWSrnJtDbMHN3fUz2+W4Az1TQGAy2RD0uVwxNjFRMdxPN\n4u31eji3vXFkAZRqppj56I45LWx1SV1UY7pppGSsGndXbSKOpnRoEdPid1w1nBq3XRds5bnmRcp4\nWs3YmnQd13mSR0irvhbH5QCqCQ9xPSvZh7R8MOT8t7Wry+vM6ZqrCL+x9UmbYVkhIV0LbYYVdQGT\nCWNndyL6dBVE9HUAHvF9vlaonJs/AdwE4JnM/AZmfgOA/wHAzWpKA2BtuHqkXTK1TaETydisnBqX\nShMaOteHqzizvYn1YbW7q6vskG6otrpcLqomQh5SNFbR0DFo+e+mS3BMV6I2Y9Ik6OoGC6TrE00M\nn7Q8l4u1+a5kHqW8aD6lq3hXhLkuWfVdc9bG97oMiYtqSHfBLvFBTSyolGeG5LEmT8j3qfbco0lW\nFDMsKybadLk+QRwQ0V8Q0aeMz90A/iVmuXW84MPJbwCQ34lG2d+6h5y2yRWLJJn0LiunxqUyJB56\n5DI+/qnPFy6RPQ4LO+amGlIh4IJtHpkuwTHn0DLFpHUVXe2TkO6tpZ+EiTjySJ6VOZGreEphTpM4\npkuCiY3vdRmS5FHHxVprQpPsRcqbQuY7IKq/mzlkuILkN2B5wrJMuM7Xx3X+H3O8CMCLc58XAXgS\nMz+bmT/rW4iPUPkrAD5ORG8iojcB+BiAd4jJjYD9g8Mj7dLelRG+cO9DR5PYlaRGkgDDZDqagOeY\nMO/dOhEe3Eh5GLDNI9MlOOYhMGRs0bJoVqWQuq8dB7pcLtYa67k5v2POm70rI3zhnoeTjE1KYS5m\n4hgTGpc+ibVcGgrRJmzxzaETatnQpsXPhITfSHlTqnwHbe5hmrCsNuGK9Ze4y1ZZa/v9fvddF44Z\nmPlvjc8XAICIXkFEv+Nbjk/217cA+HbMfGofAfDtzPwvmhIeEvnsr67Ad5PphEyAoXELCwlNvKD5\n/rIKC1K6tX0WihapS3BK2PpEO9djzrN82dJ6QrqvhTwEtpmFT3NwNue3xgXb1Z+hkxvZkNICmPLA\nqXHpk7j8MTN2r3TDguqCrf83N4Z40i031obfhERMi58LNjdSF6RrJWS+AzNfRH4OavtHs1ZcBgsN\nUioTNGEAVdba4XB49Vg2OgYiGhDRNxPRrwO4F8BzAfzfvu97aQOY+c+Y+Wezz583pDUqXMJBSo1Q\nqs0/dEKK/PvLarWU0h3zUCiNp5AgZaC8xu3ORUfMzHv5/pfOi5BrOOQhsE1NdkhhTeNF4Yrr6qq2\nX7smU4YBaFz6JC5/be4z0vGwKtfWBrjxurPYWNN7vIT2RpA879UnS5iF3IybzM87Lb/QrJWYiGm9\nlRpSJH20d2WEg4ODbjHvqwBE9A1E9MsA/juAf4SZl+p5Zv52Zv5PvuUcGxNzTC1YyM0nJFLG/y0L\nukS3NJ5CgpSB8hq3Oy86PDPvSddh3rIm1ayGXMMhD4EhE/dorfoaSNthG0uNW64Wkj7ssgBl60OX\n8lKq3MyPX5cO3Rqh/8HzO/gvH/8MHjy/AyBsYsCUygjXHA2ZhdyFVJa2kMp4yW8+0PBvzdqSKns1\nYQAla+3a9ZhMJt1z2Tr+eD+AJwD4GmZ+RSZIisdhqYVKyZUimoyAXbXapYz/axOSzaVLdEsOV1Jo\nrQUh69K8K8m8J12Hectamy7qMRVeLmhcs2LSVRVH4+vS6vJCCUmXC5I+7LIAZaMttAIrP34pra8m\nXMoJCUKGL5hu5imTuaSMEXYhpmK0S+cEGyT8O6TnGhFhfdg9Q8oJouKZAD4K4PeI6ENE9J0A+tJC\nvIRKIrqBiF6Ufa6XVhILg9UV2UbX0G2jzcOAbWOMqQWTvhtTqxhac5tHm7GjIePfQrtCx6JTU7Z0\nHdpiAF2WS8m8aHMOudAV1yyXkk/jPhVTCJcKJjH5tQYa625KBZYW+StKpIfurrRjY32Amx99zRHd\nJh+L6VrbJeEgZDtj9llMSPh3SCGcmXHlII2yp8v759UEZv4EM7+RmZ8I4CcAPB3AKhH9LhG91rcc\np1BJRC8D8HEA/wTAywB8jIj+cUO6g2J0OK5lEiYj1rhttKlN7UrAv0lnm9rT4IkHcm4ey8rgumpN\n10ArKNtiAF2WS5f7T562Ls+hmK5ZJvLtdvGL2QuLertiIZHymjYt4HlIhQWNMCx1h02J/BUl2rEM\nCXGf5EICQsYyp5yfKd12XfxGM0fb3FttdMdUirjOy5qxtZ4d12/A49fbV2hc7WDmP2Lm7wVwE4Cf\nAfAc33d9LJU/AuBZzPwqZn4lgGcD+LFGlAZGPvurCZMRd0kDl4dGQy+FRntn0lkVXJ3KFTd4XbkN\nfFmFs65o2UNCOxaSPql81hLrWaJNMIe6JHTmoaXLloDF7F/z0NIV/qzlNW1ZRLSuzJL3U8ZzS5G/\noiSlJTKkctgMCQipVE25T6ScBy5+o6GtS3urje6Y3kleSsIcbOsh5dnxBDow85SZP8jM3+H7jo9Q\n2WPmB3LfH/Z5j4geS0S/T0SfIaJPE9H3ZX8/l/nrfi7792z2dyKitxLRnUT0KSJ6pm8jqpDy3igN\npBr6mAcPCSMoBVc7GJrNiiFFSObp2sCXBam07CkFIu1YSPrEfNYV65mnzXzWxXtM18+uCJmauHPA\nnoBFyi80CNmfKeNhQ/JnKdQKmIC0SGD2iS22UbJHVX23IaRy2LVWJFZmV9kxoaFTW5ernZrx0EDb\nB22dT0IK7a75fc3GSY6eOhDRC4jorzI56Y0Vv9+cyV1/nslRL8z+/ngiukJEn8g+3leESOEjVL6f\niD5ARK8molcD+B0A7/N4bwzgDcz8FMxMp68joqcAeCOADzPzrQA+nH0HgG8EcGv2eS2AXxC1xEDK\nu8o0kDIszRUMLo1nTC2YzYohRUif/JSH3a5A2l8hx04C7ViEnBe2381nvXhPzvVTs6aDH9QEcec2\nN8g211HMGOyYCHlQk0KjgAlNiwQxQyFsZUu9fmL2SVe9a1K6hqdU/mig9WJpS8CNKbQf4cT11Qoi\n6gP4OcxkpacAuD2TqfL4UQDvYeZnAHg5gJ/P/fY3zPz07PPdsei0CpU0m21vBfA2AE/LPm9n5v/N\nVTAz38vMf5b9fwfAZwE8BsBLAbwze+ydAL4p+/9LAfwKz/DHAM4Q0Y3yJs3QVYtTkAMOVwtTUoaU\n0g3BZsWQImY857IcQDWQ9lfIsQuJkC5nIeHqo8p4Fc9rVUyEbKM0jqarh1mz/yVCe5vrX3pQuxp4\nlQvB3Y1z69BWtiTRUWhIrUZdwbLQGRMuL5aYPFXDv5dFaD/meDaAO5n588w8AvBuzOSmPBjA6ez/\n2wDukVRARH0i+m8aIq1CJc9mxPuY+TeY+Qezz29KKyGixwN4BoCPAbiBme/NfroPwA3Z/x8D4Iu5\n1+7O/tYILrfRtjZkLdPIu9pphauUVrqQVgybMBwyLgwIe89f6PebQrq5t2WB0gqNbR1ipJYcybUq\nJkK20UW35DDrGruYc7+yHZ5Ce1cFZSCt5Sfm+LTpnpyH2X/mOrSVrV13mj4wBZM2D/C20AibJ8PV\nCpcXS8w9y8W/2zqP5Ou9ZmOKMxtX7/xwwEdGehOAVxDR3Zh5lH5v7rdbMrfYjxDR36+qgJknAP6K\niG5uSqSP++ufEdGzmlZARKcA/EcA38/Ml/K/ZUKraAYR0WuJ6A4iugPjfe/3Ylq3bJC6ybiQZ8wp\nLY0hGY62LJswrHElBGRJEbRzKP9+SobuUrh0BVKh0XaISakAkNZlO2x1yR1KcphNmczFRbdEaF8m\na0pMWrXxtc6yW3D3du3FIV2AXdC4aHcptMcWGtFlBY0NMfcKrZu0hjbXfE+lpCpZTM/ehMc/4cn4\n8rMrwettC9Mp4/LeoegD4Nq5bJN9vK/xyOF2AO9g5psAvBDArxJRD8C9AG7O3GJ/EMD/Q0Sna8o4\nC+DTRPRhInrv/ONLgI9Q+VUA/piI/oZmgZ9/QUSf8imciFYxEyj/PTP/Rvbn++durdm/8yRAXwLw\n2NzrN2V/K4CZ387MtzHzbVhZ8yEDQHsZp2K6yaS0NIZkONE3m4auhIAs+2PI5BjaQ5xms5HUnVIA\nlW7AqRQAgL0flmqt5OCaB5LDbMpkLq4+khzUlsmaEp3Whvc6O4sl2cXqeWjWg3QvDinAmjAVNJJ2\ndUnxkW9HW2es0IipLA6tjJAgShykJ0rKhyc8eSZMnrse19MId91z/mgtnN7olnI7IR6ayzbZ5+3G\n7z4y0ncCeA8AMPNHAawBuJaZD5j54ezvfwrgbwB8WQ0dPwbgRQB+EsBP5z5e8BEq/yGAJwD4egAv\nzip7seslmq2ifwvgs8z8b3I/vRfAq7L/vwrAb+f+/kqa4TkALubcZNWwCWAxD8ptMtaQ7XK1Q1KX\ntCxJ2aZVQtoHkue1G0TpfcUhTi18eNbd5UQ9GgWAa9wlFuyQaz45/7DMA4m7lNQFWIOQMdpdtdpr\nIW2XK75WA+bmF6trxlr6robXuayxG+sD3Pzoa5LERdqsRFrkFU1SJXeXvJ/ykI6FzQIuTczj+h5z\nnkh4srS/966McNd5wh7O4vr1Mzi7/yC+fHMf12xMcW59gutwgHPrkxPXVzv+BMCtRHQLEQ0wS8Rj\nWhC/AOC5AEBEX46ZUPkgEV1Hs0Q/IKInYJYQ9fNVlTDzRwD8NwBb2eez2d+84BQqmflvMZOOvz77\n/57PewC+GsC3Afh6WqSxfSGAnwLwfCL6HIDnZd+Bmf/v5wHcCeAXAXyPk/geBWEibWYliyn0aNol\n1bCFDPo2y9K4S0n7IGTdEmgPcdrD1hfuedjr3a7EYVTB5tarnb8SC3ZMb4SYcM1BjbtUzHmi7aO8\ntaVNl72YfaSNt5fAyx25Ia/T0BVSSeVXoWMcc79LaNPsadJ3pRZXCUKuNamnj0YhZsJmAXfNIfN8\nYbbDVnbKvbfksioRltdvwOPOXofrcIBnnhvjCdeMMewd4LHXMG6+FjizwSfXiHiAmccAXg/gA5gl\nPn0PM3+aiH6SiF6SPfYGAK8hok8CeBeAV2dhhv8AwKeI6BMA/gOA72bm81X1ENHLAHwcwD8B8DIA\nHyOif+xLp9OJmYh+AsBtAJ4E4JcBrAL4d5gJjbVg5j8EUDfbn1vxPAN4nYuePAarK+ivD9Ta1KqD\n8sb6AHtXRtEPc/PFedc9D3q1Q/K8ZlOU0hXTOiNxlzLplm56le1QuNP6Yr5ZtPH+XNu8e+XAWYZZ\nj3SepETMtaIdL1+k5EXSNkn6rMvzJD//2/QsCdlH5rxJuce52pFq7WihoXPuMVM3j1y/5+EaSxfM\n55tYa+vGUrJ3uOhSQ+DpE3Kt2c4YXnPIPF/k2mHro5Q81ayrlq71G46e/3u3nMWX7juPdZrlPjmz\nwTi9QdjaqLJJzdo8+719BXVXwczvg3GlIzP/eO7/n0GFbMbM/xGzUEQf/AiAZzHzAwBARNcB+D3M\nhFEnfCJjvxmzzK3z60HuIaItT+KiYjztYUTngPWZlqPqjps6DUidmf30BmFtOMB1153Dgw+ex/5B\ns1Tlw8EqDkaHzg179uwUT9o67e12MH8egHc9sekKCZPO9bUhHvPodTx25Ryu7NuZp0m3OZaSPliU\nt4+bv2wLzKeCtqsr0Ix1m/PEhS7T5gstL+oKujwWXaGt2V5QzU9c80Yyr5rxzPb7U4qu8mcpDzDb\noWmXayxjjrX8vHLgvU+HpFvDo3u9HrZOTfG4wWlMp6dE7UjZ92Zd9WMzyd7fx3DlIWxdf4hT63Mh\nkXBqow8iwtntDTxycS97d/bOpb3urLmrHL25QJnhYfh5pwLwEypHzMxEM/UJEW0KCYyG9dUeHnum\nj70rkyPh0SYsFjH7XqUVIRqjP72Ec6cZzIsuOrXhl21tMBjgzJkzuHDhAkYjXybT93xu8by0HvN5\nIsLq6ioOD21MW0pXNfJ1AbDWW0ln7xIee0MfzOue9awcMb/V1StYf9QKmOV9toC+H4bDIa677jo8\n+OCDODiQaRb9xkoLdxtNOsz+dj0vQbg2y8euvp1pD5zm/PV7vhndIdtYX1Z1G9rq3yLC8Do9/OjI\n87LDw8OKca+fN36/rx6VK+GZMXlCTDTfG+JCygPMdoRpl6ve8GvHRreUv9RDT3ev18PGxgQbK6uY\nTmXlLdrYN9oo4fUrXrxesu7q+77v+N1EuR3zdwHk3p0cCZWnNrrCh69avJ+IPoCZ+ywAfCsM66gN\nPkLle4jobQDOENFrAHwHZjGPrYMnY2z0DrCeCZJzgXIuQFYJjHUTtiwwMlZXmqU4Pjw8PNropdja\nlCyoKcajHQxXpxiuVr2XF+S49Pzq6gDb22dw8eIFHB7G3UTzdQFw1FvVrgmGq25lSXWb8u+aZZt9\nFA+rq32sD3vYXO9jsCJjnHHHyr8PVleHuPba6/DQQw/i8PDASVf5d7Ou+rr1bW4+tmbdYftfSpff\n3Ad0dMvbGG7sUvKiMtLxgLBY8LJTG2sOvleF+t+L43Ho2GfMd4s8wl52d4Q3937aJvx5QLkdXW6X\nDfV0p51Ddv4wo2ULF1cmDWhpPjZlXt983e3sTgrfXWdYyRm3fKatbvPpDT5xfe0AmPl/IaJ/hIUb\n7duZ+Td933dKTcz8r4no+QAuYRZX+ePM/KFG1AbGdMrYXl+4t+atkfPJORci80JjeZL3cMO167iy\ntwcgVMDwGDNrqHSRSA41DOAA6zUG1NXVVWyd3sbOpTkTMZ8fgQ8fwfrqIdZXYx+mFnUBcNRrb5cd\nh+hj17vs1dUBzp67Do+cnzPimAfMAxxeeaBhf5tjFY7O8jyxPcvYXJtitME4POQKuux0m3UVvx8a\nbXKNZbh2uejW0mKnK+ScM+mWrHEZP7D3r5S3SJ4Pu0Zl86RLAmiel4Xm5fn5PkWRH7sO2SaPMJFy\n35FhdXVOcxVdqcZeW095/7S3KxXk7aqn2zWHYu6PZtma+aw564yOlD5HpVkskIeHh6Xn56g2ZjRX\nStn7nwt8dl731sYEJ+gGhDGYBXiZ4pj5Q0T0sfnzRHSuLnNQSvR7XClIzmEKlPPJe3qjeChf31jH\no268Cffdezeu7F1uQEmXDhoLHB4eHh3Wq8ENDttNka/LJWhrLEwrWF/fwPhwJGjboo7ZBnKmoSDi\nolvT38V3y3Q27zP3PMk/O8oE8PmzrjYVfzfryn832+QeS3ubJe1y0d1sXlXTadKlm3N2usvf/Td7\nF+z9K53r/s/PFEHX4pHzD1Vq5KU4PBzjypU9HB6OPeoOOVYhEZaX2+a7qw/KPCIuraHgaleqsQ9d\nT1y6JV4uLjqKZdmft/O1mPtjueyQ81lCJ2N9NS+QupTWGgFWBrlCt4/Le2UD0AnSgYh2UKdNm+VS\nPe1Tjk/21+8C8GYA+5iZ8Sir+Ane1EZEXpA0XVurrJOnMxfZs9t9nNo6jfvuv4Qre3uZQLnXiAY7\n8wstcHZTgJWgrr+2NmdW4pWVAU5tncblnQsYj6XM+gA8Po+1wSHWBj79cwAeP4K1wQHWBtMG7y+g\no5uwsrKK8dh3XIt06uoGgAnWvDO7S571eX/+fWr0vX0s/NqspXUOkxb/8aqmM09X8zknnTf2PpPO\nQSBc/0r6k7F1qo/DfcZ43MSzpFjXysoKTm2t4fLqPsbjiYMOzVgtE2zt9OmDUPMiJVztyv8O4VqR\nrC3tHDPrspXXZM0vINt3pPzcvx9c78rorOqTUPzajrT7uB47u9WWSqlC9/QGY2djFZf3miiATxAC\nzBwkAauPk/4PAfgKZn48Mz+BmW9h5k4IlP3ewsU1b5Wcf7Y2+0cfYCFQbm1OMR4f4vLOBawNDgBM\nMwvl/IBCWF0dwNd11baA5gLU6moY7YukvNB1h0Jdf+3s9rCz28MjFycZU21mYZoxYz9mv7KygrX1\nDawcxc/K3s9jPB5j/8oexmO3xaNMxypObZ3ByorvWBXpnM/nZn1GWFnxn+/xyjb73j4WujZLUaRF\nMl5uOm3tNPuv+H1lZYDts9dlf3PDRou7TZp5Yn9XNv8Zk0lzpZpZV75P3HQ05w9lSPsz5jo1YWun\ntg/izfhRx2sAACAASURBVCMdXO1a/C7l1+Xnbe0w6ZC1WUKbfN8pIiQPLpflP89c70roTMsDioi7\np4VcO7OytjYZW5tTbG0C112zga1NZN8nWBvsY2tzkn23fRhnt1ewtTk7g8nyipwgFojoeiK6ef7x\nfc/H/fVvADQz4UVGvzcXKKvjJU8bmWDnljAT5t+baYvqNEShtVqS8lJq1aUaT7tGbTy2+dfrtKvF\nesIx8bmAOh6PxFpGPR0srnOO+SbaXDsaq2zXODdvsxay8Qo3NpX9KbpHtZ4WV5s0Y+l6t1x3/diP\nxyNcfOTBxmul/gAaYh3a5qxpIZX1Z8x1mhIx51EqSOeJ+bykHdI2S+pKue+426Hh5653bb8X12Va\nZaWJeHuadB7VWSGBWQjC1toZ7O/NvM3WN07h1JmbgAtfwP6+LIRsTtc1e5ews9stA8jVCCJ6CYCf\nBvBoAA8AeByAzwJ4qs/7PkLlDwP4oyym8iiAhZn/mZjaSPAVJE9vLixI/ZUBNjZPY2/3PCYl5jEB\nTcfYGB4CQ8al3WZZYGcoMok8Dc1xCAx9niP0VwiT8RizAw6hv7KatTeskGnvTy2KdIevy7c/XSjO\nm3J/u/o/FB0umHSYdIfEFCu9S9gY7gNDmati2HGOMffrxitkXebYlL9PD+8JOHa2Odh8LN1ro1i3\ne+y1a8X2fvOybXSXf5OuO8nzrjko+R2B146G36TsA9fz0nmSf95nPdQ9q6mraixT7Tsx95nm6K8M\nsbV9DjsXH8RkPD/i5vvEHJseBmsbGO37JHaMd+aSln1ptyfypKozwswwQh+XsTYYZWFDl3H5whew\nvy+3Py28Bg9PrJTdwD8H8BwAv8fMzyCirwPwCt+Xfdxf3wbgPwP4YwB/mvu0jl5vETd5eoNnKYkN\n0/rpzfHRJ4/J+ND7sJovQ/6Z4Ox2D6c3J4EESn/0V1axsXkO/cyNo78ywNb29eh7usoVQdl71VYR\nSX9KYbajXJedtnRgTHJuMf2VAbZO33DU32Y72kKZjiLdYetawXB4Cv3a63l6GKydQhUrCjmnUvZ9\n2LrMsXF9j7cW3GNpg7k27H0Uk5/Yoeu/yXiMg4PLmSLP9Zt03fk/7+pfye/6+Wz2qYbfuN5d1KXt\nA9nz0nkjWQ/+bXbV1e4epG2HDUq+Z/H2MPtssLaJ665/IgZr7ivbw6+dfNnmec7eB2e3gWvOreHs\nNmrPxb4ohw1NMwtlkxj3hTvx/Bx/glZxyMwPA+gRUY+Zfx/Abb4v+5wOVpn5BxuTFxlbm/0jYXIO\nv4UyZ3BlzBnB7FCjc0UIWZYUlQczkavcAu521PenFuV2FOtK18cNtI60eE53UA6n8ZTT0bxuV12D\ntQ2cu/bxOP/QXRiV3GZcc8qfrpRCSnsCUdy1ELJdbfaRDeX+k839ueA9GY9K/W/7LTRc/Sv9XTNW\nKffAYl26PpA8L29jcV5p1oOkbnc9MS1rdmjmyUyBez12Lj2Qszb6YTIeYefS/d7zZDI+xP7+Ja+x\n0vI5Z5/kznOuZ+toyZ+Xfb3yYrgIz+hYsbrdLgumU+Dy3tJekXKBiE4B+AMA/56IHgCw6/uyz+j9\nLhG9lohuJKJz809TakOilyXqmVkkFxZBOYoanrJGubkWLO7hyV9DOaNlzjzrYwvqyiv3Sb2FKTzs\nGs5UB1Sp1nFG14UKYVhuVXLXbZZlK1tGh07bah+70f5eJlDK3WZkfRJTS27C1b+SumTvxl0LIS3a\n9rLKY5vGG8Hsv2Zrvrr/0wrSrrGS/K4b95R7YLEubR/YnrfV60ZIbxE/QdGPD8q9mWzrUrZm1fOE\nmvIl2TyZjA9w4fyXPIXXeGvHPM+5+89NS5W3XV1ZMZIVnd4cO9xuT5AAL8Usj84PAHg/Znl1Xuz7\nso9EcDuyuEosXF/vEJMZAf0+HZnKNQffspti0c1Ld8AJ7VqYd+8p0i0VMk0U21ksy+yTwdomrr3h\n7+RcQEIe+nSuRLKy/OuyubZVweUuKJmzrg3DPAxIyna7IsraLcM0s1DKNxJ3n4TrAxnMtWO6oTen\nq7zmTYTkNyGFYRm0wl1zmAdIyYG9/L6t7LQIxwfliOlerxHOmrdZOx/DhnBIFTSu4vxpsJUtV4Rq\nBOsRdi4+EN0SPkO6kAMffnJpt49Luyu4tNvHIxenue9hPvFBOLvdQ/vhSyfI8F0AbmTmMTO/k5nf\nmrnDesE5Y5j5FhV5CXB6c4zJmBpruXZ2V9Bb7WNndyXLOsrY2b2E8Zgx66Li95WVAU4NzmDvwJ1F\nK3QcZd7FAUBBO6d13cpvdDMh5bqj4PUqF5CDK5dKh7561xV/WlK6H0volrqvhXW/8nAvzh0GJGW7\nhbN0bnt2mHPI3ich+0ACc05VrR0VXY018jKY7ZCsS/0aLo5tXMWGPx0m2gxvCMlTtXtHOrjcRv0T\nDGnGrk7p4V9WuhAOGb8xXUHt80Bjmde1WbYXaOASrCTnwROUcXabCvMgde6RE5SwBeCDRHQewK8B\n+HVmvt/3ZadQSUSvrPo7M/+KN4mR0O8tMlSd2TpALjktLuyYmvx65rgx3KvIoGhmRLNlbpNAt1mb\njHqmnas+rJqCoRsGYy5oLM1D3gEunL/b+6AcNu7DBlmsiv1QwgbdUrpcG13+9wDzYu+RHG3hNllZ\nu2NmGNYd3GTPhowjNctuTtdCIx8/Jiq4MKygM65iI+RYpxPGQvJUvZCUBhLhGECjGDM/mOtSp5QK\nqzTRCFxSYddWdjiln4mQSmwXTm9OjLJSZk8//tAYhE4QHsz8ZgBvJqKnAfhWAB8horuZ+Xk+71e6\nvxLRc4hoPfv6rNzn7wN4E4CXaAkPhdObY5zZKjOVM1ujwt912dWkz9sydmljhWyxLhVuGb2msaCu\n+EtZFsriBhIzQ6UZEyKLlzDHRxajYyJkzKTH+xtnG73vrjtc1kk77O6WKa1VMeNIO+Xqpqq7edxu\nyFjFMmRuuyHH2l5WWL7XzNPBN26sm4mUXG7R+d/dbdC45YbMaKvNrKxbW7aywrrphoNbie0KE/CH\nO3whplt7W/2tCX1wPV+/dsxz+wlaxQMA7gPwMIDrfV+q42ArAH6LiL6bmb83/wMRnQHw7qZUhkSv\n517Eiwl6iD6N0F/PaZv6q5hMIlhU+gOsbZzC/t55TCbmAinSYX+2Cv509/uEzbU94NRBVra0zbIs\nasW6zXYR+n3GZHKIfn/V2mZ5n+TfJWxsHKI/GYnfncGcJ8CiH2T9J2tHVb02FGnp9wmbm4foT6va\n7aLbVbek3dJ2LGD2V/X3Afb3LjccWwkk7Yg5L6QIR7eLTkk7ys82mSd1/MhcC+a8GWJz6xx2d+6v\nobP5nJWUFWfc/frE/XvVd655P97e6Yd8m2382vx/OIQfy3B8M2xZQL4PQ7ZbV5ZtfsKxH0ph9qeG\nX3Rpr/CvV0qH7fm22nQCPxDR9wB4GYDrAPw6gNcw82d836+0VDLzH2KWoOemip93AXQizrLfg0Cr\nwdkEzjTK/QE2tx6Ffl+vyTIxmRxmC6ZKi1akYzIZ4/DgMiaTOusLZTRmWsj+KtY2zqHf90juMhlh\nd+e+o4UreddFh7vuYh/k67b3j6v/7LRNJiPs1R4efVAcnzyk/SdrR329VTBpMcfaTrc5lva6dfPG\nH2Z/ub67IZuzTRF3XsSDi24Xb3LzrvyzZptl892G8loo9y8nE37q26Ufd//57Bpb83fX9+K78fZO\nOcz+zveR2V/h+EH4NWxbD3a6y+tQsraKZYffm+vhLqu+3W7eVb8fymH2Z3PetSx7hXbvtT3vN+79\n5b9XZHnxWADfz8xPZeY3SQRKwBJTycznAfxXIvpPWKyeHoCnAHhPU2pjYnszr5Uk9PoDTGsWf69P\nWFubabKmkxjazH2vp3r9IQZrQ4z2L2d0FOnu9dewtnEN9vfuy34foUcjrKyNsvfr21imI/+ujCHO\n6NzCaP9hQX/V1Q30+iygW0LbyKPspjjEygqwsrYLv2ylFJGWqrGs67Pis/Kx9J83zeZJHmYbbN9d\na9xGi+ZdE03WlX87JAhJd5k3mb+vYW1jC/t7u55j7b+mZXCthRF60wOcqm3nGoab53Cwe79HO7Rj\n1bwPws5J83fX9zwd5t4Zbv5qke8jgAp7p543mZCMpauP6n930e06Q9hQXbarXSHXcX1Z9nb78NxY\n/EYD7V6REs333qr3L+7Ok5KwVdjv91cxHA67oLG6KsHMP6x538eB/1/n/j8G8LfMfLem0lBY6U8N\nQXKBXn+Awdo1tYx4OjnA/t692aJoD9PJ6EgYAqrp5kK2Rz76+4zp1rexDG68mZp0yqGh258217hr\n0OuvYmV1C9OJnyKi1x9ibeNR2aEm9MYgGcvis/Kx9K+rXHa8A6d7jde3s/xukU53H5nP+yu0pO2Q\nQDa29nH1KYsTZaK1wzU/3fOXPK9RiMlfXAg5tuXfXd/zdBT3zjb7pExbfi8YFuanfg9rDlcf2X53\n0e1zhqhDm33igp225meZdrGsdBfRZM3XndXLGOHg4KB7E/IEXnCamJn5I7nP/9cVgdIFN7OcL+44\nwdW9/hB+bjZFOky6p5MDHOzeV7lw/Q6+vnS43jX7S1r24vnwG9mCtpibZJP+9j90u/pTM5YmYs79\nYtnzzafn5Sona6NmjdcdxBZ02vvI1q6ZMuHGrC0h2iFByLG1l2XjTWHnqxTSeTQXktyHnunkEOPD\nHUyDuKNJ+0jCg2P2v7lnhewTE837qDw/tXtYc1Qr2xZ1F3937b0m7GeIIuLt6+Gh4WVt8p88elgZ\nbMHvSvjlgW7PMsemK2N1ghBwznQi+hYi+hwRXSSiS0S0Q0SXPN77JSJ6gIj+Mve3NxHRl4joE9nn\nhbnffpiI7iSivyKif+hDfN+aqCfmwdkO+0HataBMum3taH7w1bWh6nc7Yyg+7xobDZMJPe55Wlz9\nvYb1Uzej118DMGO8h/vnvYRQeX+3BdmGIDlwSudUyLGWbpKu52UWvPYOuzrU9797vsY7WMjXiv88\nmnsr9GpjovzboV3TNuWFnl/4C6zuPmmOmHwvLk+17+s2JZaerubrUlp3V/eltGNbj5XBJjZPPxEr\ng80IdLSJ5nuvOTYmH+v1h1hfX18LSe1xARG9IJOR7iSiN1b8fjMR/T4R/TkRfWouYxHRs3Ny1yeJ\n6Jtj0eijPvlXAF7CzNvMfJqZt5j5tMd77wDwgoq//wwzPz37vA8AiOgpAF4O4KnZOz9PRH2/JnQP\nPm53iwUVj/lptElNXW7q2iGhpTsblYyWXn8Vq2unjw5XrsNWnplK+1uGeAd29yHF/8ApnVOadkkt\nk2XYrKA2C56b7i7N/6ZwzdeYfDDmWpG2Kx6dM9QpL7Rl2y3xxd/a9Q6pR8g9KWXdbfZnaOVaPJhC\n5BDDzUcdCSZhvbiaC6zj0S52L/0NxqNdj3quDlSNTTfCKLqNTCb6OQDfiFlum9sz2SmPHwXwHmZ+\nBmYy1c9nf/9LALcx89Mxk7HeRkRN7i9ywkeovJ+ZPystmJn/AMB5z8dfCuDdzHzAzP8dwJ0Anu18\ni2YZ3pqhHbeNsntrTNchjUUwpMuNT3kLxO0TGSSb5nh0GZcf+WuMR5e9310wU1l/SyA/sNfPi/L8\nDXlIkc0pmWXT5m7mg+Yu7W66i+jKQU33vt36Kp1HMjS3/LrXissluPl8l6KovJC6TLrKlgg96dzp\nJWgzBEY3n9vsz3DKtTLieiMU46LDeXHplChTjEc78EvsV4Xmwm+78KfFVMJOJwe4cuVKF7MstY1n\nA7iTmT/PzCPMrnZ8qfEMA5gb/bYB3AMAzLzHzPP00GuI6MLpI6neQUS/BuC3kLuwiJl/o2Gdryei\nVwK4A8AbmPkRAI8B8Me5Z+7O/mbFSr+HM9tj8FQe+Ey9IVaGZzA+2Gv0vg4LrdWMjhWMDw6T0xG+\nD8Jo49rsk2pI2nUZKHij2d69AqJdDDZiZ028AurtYehZj3temG1y9Y9mXjj6L9cuG93Vv/nTFXat\n+IxH0z4jUG8AntZnmaz73d1Ge9n6/o+jzZeNnWytVCOlVWJWl3bsbGW7f2tSdoh3fcoeK8rW0lbf\nf+2eP9pBXB56BdTbxar3upWs86pn06xxSZ9Rbw2r62dweOVe8DS0TCZbC3V0X9zZOBLSF0l+yknC\nJpNJUyn8OOMxAL6Y+343gK8ynnkTgA8S0fcC2ATwvPkPRPRVAH4JwOMAfFtOyAwKH0vlaQB7AL4B\nwIuzz4sa1vcLAJ4I4OkA7gXw09ICiOi1RHQHEd3x4MO7uHBxLhcTqOevpeHpCOODh7JFkhJFOt10\nyNolebe9PjAh7ZPjAs4YLkM3znC8n6/Hg6ql6f9iu2x0a9tkf186drLxkIB6A6wMrwX1FtbbPG3l\n33NUOfrI9q7r/TbnlKzueGMTE+U2muM+xOr6o7O/SWGf3655YStP/q4/tGXr+syO7vJY7T5Uj7Bt\nNtep+T1kO1LyhJhnoeZ9Il1LJt0XdzZwcWcDQLczDofCZArs7LHoA+DauWyTfV7boOrbAbyDmW8C\n8EIAv0pEPQBg5o8x81MBPAvADxNRlLhVp6WSmb89VGXMfP/8/0T0iwD+3+zrlzC7cHOOm7K/VZXx\ndgBvB4CnfNl1DHCm/Rji3DUb2UT2u14jnobQZg2YLc4FnXY6ys8LqHC+G7MPStQI+sT/3eMCzTiH\neL+IlPMiJGx0a9tU/37Yvi+Vbsx9+1owN/PZwfjGI+215pBSLbjkaYnZ/xos63xujvKcJBBVXyQv\nL6sI6ZwqlhdPuPJT2KbaVyRrpT3E3Ys1bZZayuzt0PHseJZ56fkwj9l8f9ipFGzSZvk6ZVy42Aew\nXvnbcbhWJQIeYubbLL/7yEnfiSyXDTN/NBMcrwXwwPwBZv4sEV0G8BWYeYwGRdI8x0R0Y+7rN2MW\nPAoA7wXwciIaEtEtAG4F8HFJ2dPJCOcf3sOFiytHWhHbJyY01gDt86HeDQ1Jn5jPxtRmdwVai3WX\nxvpqQ8y+N60l7rXg0qrX/y4tu711Gc+aslyot/iV5yRjFoYjF57c81vjCRHTCmQv2215P8DhlXs8\nD+ChrbntwO2tEM96a4O0Xlc7NDxbM5Yabw932avor26BetVJ8XT71ExIvLiz7nW+tp+xT64UaYg/\nAXArEd1CMw3hyzGTnfL4AoDnAgARfTlm8ZMPZu+sZH9/HIAnA7grBpFRsv8AABG9C8DXYmbSvRvA\nTwD4WiJ6OmZc/i4A3wUAzPxpInoPgM8AGAN4HTNPZDXKLiC/uLOZ+x3WZ+Ug9Pp7mE5WANQlsa3S\n4NggfT7Uu01h9r+kT8xnfd5Ngao2hZw3QN1Y9fpDz8uGq96X0hmjXU3QFTp8UbfOmrej11/D2nQd\n+3sbmE56kK8FQu/yBUwnfQAuZVoPK4MxxqNVz7Ilz4cbS/daCDlvUs5BWV3FfhjVzIscT/WeB3XI\n8+cQfdLGvpSH71py0+k3J7uwh/mibh8y+VEe2nlRv7/2+kMMeQ0Hu1X1ytvh/3sdnbaxtPVD1dmm\n6tkme4m27DQwYypnmXtvwMHu/ZhOTvLz1IGZx0T0egAfwGyAfymTnX4SwB3M/F4AbwDwi0T0A5gN\n+quZmYnoawC8kYgOMcsa9T3M/FAMOqNZKpn5dma+kZlXmfkmZv63zPxtzPx3mflpzPwSZr439/xb\nmPmJzPwkZv5dbf2S+5jk9+OZv5sX3Ia8tDemFide2WaKb1mftHfHqA0pr4MxETa1fswrLdq8bzDk\nfA6XdU/Tn9PJAfb37s0dVONlZ5TewWh/Pt69ca7s0CHnjXxvaFbPrC6TZ9pR5AnpeOZxuPJmhpT3\n3Ia8m7k9y06ZHy2gXSu2/VVy53N8aDLLmveVyta8JmutG2n6sGqtFDP3nqAOzPw+Zv6yTFZ6S/a3\nH88ESjDzZ5j5q5n5K7NrGz+Y/f1Xmfmp2d+eycy/FYtGp1BJRNtE9DO54NGfJqLtWASFguSqA+n9\neObvIS+4TSm4xD4whWIUXTnEmIfZ8uG2nWtqXNDfKZpvV7o7FqWCdMi6JZt93CtC0gkL0itcbM+H\nv8twMe9cwm/IeRPy7lSf+SnjmRKFQVfuAY2NtgQu3TqNKyykQXmtaJQkxe/uO59D9oluDknXh2TN\nh+W58RR/kj60KSpOsHzwsVT+EoBLAF6WfS4B+OWYRIVB8/uYXJpwc7GGvOA27v1t9ro0TKXM0MIx\niq4cYsyNrfy9m5u9OdelWnW7Vb+IsHeMyg5qoeeJfbNfbJpprRRaNL+bVnI/oftZqRVjcUB1z7GQ\n88ZcO3ZFkkRALf8e73BlVxRVfbehm54kgNn/yxO7JeFdOj6n6wO7oFieF6GUJNIzmQZSYVgD+ZrX\n8Nwiwiv+6su2/9ZdfnICOXyEyicy809kF25+npnfDOAJsQmLj7zm27QOutzAzEVgu+BWysTNsnWH\nMQmKTEVWT9wLsdP1QRl54cG0VJoKgDGm0wNMJzGu/wnZ5uaHbrc1y7V2JJC6oYd0Z7Nv9kWeobVS\nDLG2cWOSQ4z90G1C079SFzH3/J4fUMPOMTetebgUSRIBNeb8ddWt84jpihKkjKJ1q71wBTk0a8t/\nHYfog3pBsUhHmYfK+E1RmSk9k+ngq1CsgozHhgyVkiHmmc3GB7tiJDhBHPgIlVeyIE8AABF9NYAr\n8UhKA9NfP5Z1MPSBMayLX73GSF7P8YzpsW9sxTavDDawvvlYrAzCZxdu90CUb6fUmtUcUjf0sEjX\nTgBgSqOltR26TWhcPSV0+NSVP6C2eShx7xXtuKS6YR74m+95bfFfPyz6P6XXT5twr+PFGUTbB7KY\nSln8oPm7RJkZEjKFYtX7/jxWA73SJGbuD1vZJ5bJ4wyf7K//FMA7szhKAnAewKtjEpUCRSZlZo7V\n3aNzcXfBNPv9ASa9VezuDjCZhFhEPQz2DzAa9QBoBVVCf3cHkwlVlCWth9Dvr2IyOYSbWZjPut6t\neL6W7tDI1wV7vbtj7Ox+EaPROPtd0icSOtKmc2+GekFENE+sbe5Cn1S1Uz7u/d3zmEyAdO2YadFl\n/VtsV78/wNrGOezvn8ck826Q8wCCbE0Pav7vU5cvv/GBWbefhc/Ox6R80B/lsVK2oxX+27SuhTW8\nOOe6BM2ctPdR9RlEK+TU8D0HHRu9VezVnIXqf2+y5iWo63vLHua1bn14rJJukw4RP2kOk5/Mvw/6\n90Vx3z/B8sBpqWTmTzDzVwJ4GoC/y8zPYOZPxictNpppSy7uDp2fPCaTEXZ37jM28ubo91ewOjyF\nfj/EbTCc0VXF4GX19PurWNs4h76HS5r5bPldQr+/uKi7/Hs93XoU6y7W5ap3itHoMuZu0JI+kdGx\nvAjbJyERrux+f4CNrRuy8nzqso1tzDa75lTxd3PsJpND7O+dzw4tOh4Qc367+U1I1I+Xq80uulx8\n0gZzrHRIyYvC1RV33E3I1q2dNldZ9j4KfQaph5uOvZ37a+lw/R4L8nkhWbcx14pZtoyfaGDyk/n3\nRy6R9Tx8guMPp8RARGcAvBLA4wGsEM0YGzP/s6iUBYf7DqU4C4CDMsniYo6niZpMxjg8uIyJZ3yg\n5NBSx5DMw+nsb6PAByI7zLrtsPe/hm4ZHcsDSZ+YfeD6LoNpdQvb32Q5TErq6tI8KI9dkbdJeEbK\nNe3iN8W5ABVPtY2Xq83S39PNjXj7jB7+tMWdczp+YqNNP85hzyDN4aKjHTq18yIlL5MgLl3mWNWP\nXdW5envzxJp5XOETU/k+zATKvwDwp7lPx1EO+h7xo3B5/5SXhdFVdjzrgauuhTbKbRFpjn5/FcON\ns421d7Jni9/rD6/xDzRNhJ56y0Fzuru6Uenh3yemkOIWDvxRtrqZAlHzNV62DBTL0ihg3OhhMDgF\nP9YubaN97GTeDW1au+o1+lrtvn28ZJZh1+86XuUPvcUj5H7p8mKx1WX2b0jvBLsV3436sZcqeK9O\naMbSte50lmIZQq4V13xPeY4twn7+JvT7fZ8N7AQdhM/ArTHzDzLzLzPzO+ef6JQpMQ9Ov7y/hYu7\nQzxyiSxMPqSrSli46rJZRNztsv9uLzsm2nT1lAg9zd3/QtIhR3ubiQRuISWc0B7TRbKs/JEcbmUH\n4cFgA6fPPR4Dj2RR8jba615WRUj+0O5uQ8oDpgvNFTQSVFt229kv+/0BNrcedbSWyuvYX9Eakq6Y\nitCwIS9SLMte4RrL5u0w51xMxDYUpAsDMOHu/7lweXn/FIbDYRdTNJ/AAz5C5a8S0WuI6EYiOjf/\nRKdMiUu7A+zu5xeLLX5QtrjSum7VHwZcdMhjdPJlp4rDSI14msB2D9X+7WorpkyK1EI75/4f0goK\n+LnDVltbXJaZIkajPVw6fxdGoz0nndI2ug88y6kIkRza0x7EwkEnmOhitUJb2rgwv8pzzlcZ2q04\n0/r53ea+sizzXXsWcoGTKIlmiKXMd3n+xISk/yeTQxwcHBy3g+dVAx+hcgTg/wDwUSxcX++ISVQz\nLNxdL+4OnS5o+e8hXVWq6/L9rfy7zQ3VdVBwaZh1rlrLiZgazTb7TMq0ddaYMHS4kU5oLyeJkIyl\nfV25FDQ24VnuVldMFmWHnI/5H3jCCoGpkk646tFa7eyIJzjL1044ISekpc0n2ctirbnc/bqzxxXn\nXXfo1PFcnaeUDK4kQc3bUZ5zKddpSBdtcx22FVbk4+0x8dnAOovJlHFpT/Y5LvARKt8A4O8w8+OZ\n+Zbs84TYhEkxv6Pn8v6p7C927Wrxe9jFZTuYuK2HZWtA3UHOzSjNPpC4cYRmnJry4lm/TOgOr+25\nCsk2TVdsXPM+iBv/051DYB7l/rLH8JVRLzyX+zNdH5RdDf29F0ILgfb5rV13iz6V89SQrp0xrUKy\neWOnRVZWWotgPudAm+5+MkgUG2nRnN9oPKXCo7mSMOaaN2EKfnFdtE2kyTTup9g/ialcVvgM3J0A\n0kgWcwAAIABJREFU3L5ULWN+76Rv5ry4Vo/6sn3qzQuR9oOcnOHn3Tg0wq8U7SaKyEOq0eyKW6kL\n4YQNXR+0Gf/THDpBOjQvyW/A7fany9WwDjH7ROoS3LweN0K2s0sxqd1yDW2G8p7vUnjFTGTiH4sb\n0qrcJlztkLUzXRvl3goauLxc0q3DVOcXH8X+SUzl8sJHqNwF8AkiehsRvXX+iU2YHOa9ky4XkpgZ\nu+oPQK7EG2UhMqSwUHTjsAkPoS1OMvcH27ux0dy6G9dVKCRkyUbssXTNM5vK6YoHHd3xDs36ua/L\nYtv83rh09062mx1TkkgpZpKf0GtHY9kJ/XxTmHzMrqBJm6zLn+6wZceEmXlam+14gdBJxWyQeiuE\nDReRnFPDrqOYwnIRbsX+SUzl8sJHqPwtAG8B8EdYoitFUlqNwrq7duOy3PAWkuJVKLJsau26PaYI\n0g8/XyWXsPuUVr1xuTdFf6TsgzK66VqrpUvXp91MlV+OQe2OdTyl90eqskPXHZfW5rGgsROZxMym\n3l6m9gUkmaelkMfxpuN7srrCKWHDr6OQGYu1/b/cMZVXM5y78DJcH1KF8FqXZpfcS91d20RxU+Wo\n1sGwglq4y8tNLKw1/nfBNbmkOvQBxkaLtK6F9bzJ/G7vcnLNeJTR5Qvg61HtjZC+HWHHonjRtn4O\nhusT7V4Qo95qpGlzVV1lr5Vw87E4zw6Nss0L2ot1z5UTk8no6HAdKvt5SD5ZLrueP5fLNvfHcP0v\nyTwth2wsZN4Luj6QrD09H1zQGnNOadElV/4TpEWtpZKI3pP9+xdE9Cnzk47EFLBr0d1aF5uGx2Xq\n787VHUWNf8wU6VK3Ov/xCX+nVKo4srCWMjst0rqaz2+ZxjLs5eQhN7ZuJc+QIF2SCRviHjJ0c7DN\nmEw7Yl6jkq7NNm+G8m/aNb8QJlxll92o481RmeuttA/8+z+mi68s87QPNG6lMa8JkmTntVsmY1r5\nde3SZu7tqufPCWLD5v76fdm/LwLw4opPpxFy8bV5IEqJkIHz9j4N615i0p3yTqki2hzLVGno08XD\n6g88MZMXLSfaiz/s7rUIqcZWf4AM59Jui6cPDYn1VnvAzwsTrrLLdLUXG20XhsPVbdLRZb6m6QdJ\nu2K61rrCQ+ReQ5rrj/xpTauIO8FxQq1Qycz3Zv/+bdUnHYm+WNxTCcjSJ7uf1dxL2SWE0y7pGJpU\nMDEPwvUClC65iIs2MxFByLJ1SMXk5fWEu+9Rh3DXTsSvSwO3sqfu3tvl4WNShEsYoklKozlAAlXx\n9+HuWk57SMz3ofSQ7bI2FkM47GWntCq7rIn1wrAe+bpjJiwMi3SePzIPmbRny/rnQ5dtUz7Ynp0h\n5N7Rw+rqavsB8idoBOfpmIi+hYg+R0QXiegSEe0Q0aUUxEnQ6w8w4kfVaoRM2DdR2QLpjtZG68br\nD7fFo77/pS5Psrub4iU20SYiiDlPUlmg5Gn5/eHS6moQ2s0udCKk5pBfr7GI4dZezdGljMVpoLM2\n6udzPv5eV7apwInHP2Qul7J9uyx4t5O0Sbvmi+3orqAXFq4rXNrpB124U/g1X4/Qxg6b8sFet7vP\n/BXyg8EGtra2TjkfPEEn4WNy+VcAXsLM28x8mpm3mPl0bMKkmE5GwdwI9K5E7SCuG2+9m5EUUpen\nsstOvAOQrY+0iQjajNlJVU8qlyVt2dpDoO39lPxAGheWj+HWxpSlFJ67orjTWhslKI9PMf4+ZNkx\n+UdIl0uXtdFWtnsOxbyGYlkQU3ljV2K1qYzLQzuWXZkLKfvT1WaJQn402sPOzs7l0DSeIA18hMr7\nmfmz0oKJ6JeI6AEi+svc384R0Ycyy+eHiOhs9nfK7r+8M0sE9ExpfSHdCNzultq6bUh5v5J/3doD\nqI0OKd1xBSibK4w2EUFbMTvxrnNw/6651iPkAUcXy2LC/n5Md3nTwjTBZDzCZDLxrNvmPtXcXUrb\nDtfvxbratFqGTHJlh3t8wpUd19Mh5N17zV2ZXfNVdwiPefVPuvmu6wOZp1RIZYMUMmt52MRJqSDv\nT0miHhP2NssU8lMcHh62celw50FELyCiv8pkpTdW/H4zEf0+Ef15Jku9MPv7NdnfLxPR/xWTRh+h\n8g4i+jUiuj1zhf0WIvoWj/feAeAFxt/eCODDzHwrgA9n3wHgGwHcmn1eC+AXvKgPBrvQklYjr8le\nGm9jixuPkvIwayKdy3BY+LsKhW2DbMN1X8xtU2SEzuSbh1Tws/V3Ond5892ZBvhxOQ1wyrs5491J\nanOFjmlxcpeljavuCkIq6sL1d3g39Txcmdg1gnW8TNVyPtiOxVXuKdVefKeknd09A7jQPCGivs3m\nHLQp5LsR2tB1EFEfwM9hJi89BcDtRPQU47EfBfAeZn4GgJcD+Pns7/sAfgzAD8Wm02dnPA1gD8A3\nYJH59UWul5j5DwCcN/78UgDzey/fCeCbcn//FZ7hjwGcIaIbPWiLgjY1aABgz14azsphQmd90UCa\nvjrdJb1dcWUxIWH6cd1uiweeKrps97C62tFWJl+Ja5b0IBzS/W8yOcRo/9LR92U5AMkt3vnf7If/\nkH1QFuI3cea6WzEYbKrLltYdEzGFiTL8XQ/TKnSbC9Z6Om1KKkKvvwrfPb8ti6veUyolUl0VtjzI\n81VtmyVzcFn2rA7g2QDuZObPM/MIwLsxk53yYMxkNgDYBnAPADDzLjP/IWbCZVRUCpWZifXajJhv\nr/h8R8P6bphnlQVwH4Absv8/BsAXc8/dnf3NGxd3hw1JqkJKDZrp5mXPXhrSylFGc+uLBm6rhcva\n1RxxrZ6SPtS4A7oQdzPPC35locd+D6tdeDjEwd4jud9izsmiBUqiWDKFHG2iBztMN94D7Fz4IiaT\nAyedZSyTG+kCrsO/fE3XWx9dQnxYxEyeEy+EQ+ryLnE9jOtiXUTYWM/msFnpm9GicWuUoEtCY0jE\nDMvoDkLeUy5ZD1eL0B4APnLSmwC8gojuBvA+AN+bhrQF6iyV9wJ4FxF9RRbrWPpoK2bmWYopIYjo\ntUR0BxHd8cjFUEJ3e1kMy64tLhedNBtfShczH7ps1i4ddBenF58d4tSZm9DvDxu82+a9UDp3qaLg\nJ1XI2Nx207mhlxMJ+Lej3y9e06GLK3XBZbUPdy2Qu24NmgsemszTVSiOvb1/TSE+JKqs4/XXv5jo\njhu/y10zlUJM3uZwyh4NqhQZRR4ro8Xu1mgqVOJ5Qi0PQq6l5emTsMKdRBA/fsqIyZRwYU/2AXDt\nXLbJPq9tUPXtAN7BzDcBeCGAXyWipLEalZUx8ycxM6tuAPjTmk8T3D93a83+fSD7+5cAPDb33E3Z\n36poezsz38bMt53dXmtIRhFx4zhcdUndXdNsfHGTGsjoclm7YkIqiA/XzuSEC39Lg9QdMKaLn+zd\nLmeN9F/H8sy+xbKLSo94dwCGVPa4+7OsAAvlMaBxOw8tIOXH3t3GtMm2fJVpcd345Yds+56W5hC5\nvBaQsInp8v1g9ompTIvrCbUccK8l/0SO3e6TkGfL7ii1lhgPzWWb7PN243cfOek7AbwHAJj5owDW\nAFwbi+Aq1EqwzLzHzB9n5ndWfRrW914Ar8r+/yoAv537+yuzLLDPAXAx5ybbEM0PV3HdLYsMKaa7\naxnhknik3bDjMbuQdY9Gu7jw4F9jNNoFID0MyCx6IftfZ/12bbCa/te5oRfXjosOWWbffNkaF98Z\nwgmCMn4ht9LbhZyQAq1d0x021rZYVzyvCBkkyjR9PFvzg3GVy7ttT0uHlHkA4imi9W7Q9fcPjkZX\ncPniPRiNrmR1dcMFuE2UM2oXIfGg6XKfpFR2drkflgh/AuBWIrqFiAaYJeJ5r/HMFwA8FwCI6Msx\nEyofTElkrVBJRO+1fVwFE9G7AHwUwJOI6G4i+k4APwXg+UT0OQDPy74DM9/fzwO4E8AvAvgeSSOq\n4illgmF585EdLPw3lLKgEVN4015uXl/WsrgspLRCm4KJ7DAgPaS0lX3Xfl+pJMFNbOTXTmhFUXFd\nNnfxBcIKgm7XW9u8ssftuoSckO0ol71wqQwttEgUBGm9WCR9pM38Xb8+widgWR6XwDqU+Vw8RXT5\nzBDORbXf76O/MkC/389+T+kCHDLnQDgMBhs4ffbm2jsVZR403T0nhYzZXq4kTcsJZh4DeD2ADwD4\nLGZZXj9NRD9JRC/JHnsDgNcQ0ScBvAvAq7NQQxDRXQD+DYBXZzKZmTk2CGwmlP8Rs6DQdwH4GISr\nmZlvr/npuRXPMoDXScr3QVONc/XBbPXoAGliduC5Abs798MVayMte7EYfVAsa77xzeobqZiIWVZY\nuPqgOercRuO0o4j5YWDW9/a6TLpS0ilBeU7ZN1g/K12csS+uHUKvl8+gqK1Xsi7tkCfX8ecXknk1\nOxhfj72dBzI+ZrbR/F6kJbY2OlYm4GoFQTVirsu4d0W6Ub9fyua6q4+6ytsk0Lgqa+vS9J+Lf6eE\npB0p58xkcogDazIuF19sC7I9bd6nM7dobYx4V/rgeIOZ34eZES7/tx/P/f8zAL665t3HRyUugy2A\n81EA/ncAXwHgZwE8HzOf348w80dSEKeBJg7PbW0pa838Dzzp4q3Km1Hz2IyYm09MLW9Mt1EXJHXp\n4wclCHmPmctFVWKlc1nVdK600+lis01rwXbTFiu5jnReSQ7GsiRj2nv8FtbJrli/QyNmfLILKd3p\n2xae/WHjPyn3lXB1lfu+PSuzZn+UQ+Kaf4DLF+8OIGjJ6tWiCV8snluX34PgBO3DFlM5Yeb3M/Or\nADwHM9fU/0JEr09GnQo6c7vtmoSy8NbcHcvtrqYpK+TGF9d9QafljXnhuyYdu8Z9LV5/h0z2ohcW\n8i6qdmWORpgw16hrTXcV8jXsP6+aKOJ8lWn6/l3QHfoAH9NtV4KQoQ5ShBVoXYqkfF3tZV53QcJ/\n0ioEml9xoaUzLJ9M6d7dzhpPua9IlTXmnhiXVv+wixMsN6ypZoloSETfAuDfYeae+lYAv5mCsDZR\nFhJjaiVdlsuYAlNIaIRhXXbXmMwwX3a3rFvNIU/2smhn+JjJemEhrPW2S4mnTKRc4+FSu0uUaTF5\nphZF2pbDYm1Cuw7b8uAIq+CyI6aVP6RyWApJu7TjbH8/XfIiKcLOb/87pVOuK3nsbcxzrX8+j2VR\n6J7AD7ZEPb+CWaKdZwJ4MzM/i5n/OTNXXvXRFqqS9Pgh5mYVztoYN0NXyIB/DZ26y4VjMu582ctl\n3WouPNjaWT5o+adXd0OX8EYCXVIUE1orUco7BLtiaYjp6izFgjZ9/2joNu8M9IeeB6ZUSPpbnUPO\nV7nbrcR7JF5YiwuysdeOs+2+3qI7/HFRwppwjW1b5wL9+STe/ipTyJxgmWHbvV4B4FYA3wfgj4jo\nUvbZIaJLaciLB7vmxB7j545dlFxlYCLd1RF2piO7dygknWGZYcgrRVxj051NM6SQX2yn/fAUVwkS\nDu3O13i0yOpqb77GdHXWQDsWMrqL/W/eGSjDsmZXTOdBENZF1b520saNdmfsJWFDKRFWOeF/p3Ta\nNoc8O4YLa6mizUb3CZYbtpjKHjNvZZ/Tuc8WM59OSWQtaBbn5vlwYZG4mL4txk8SuyjX1rmu7tDE\nu/kzHRcz1Af816NLB3472tNOu1Acn7h3dRYPEinTlGsQc77GjL3VwmaVCylkug7dLldnzTzSWD11\nYyFxDzT7fzTaw6Xzd2E02mtUdzpI9zANdF4secj5SfO7O9tMuiSDdK3U/+4OG2ovQZPMbdcFf4+a\nkHNOjpju9FI6T2IqrxbI/Ww6BOZV7wN8WRNez/S1MX42piJ3nbAdRGTWRBudJtxuSTE3zXCuiG6m\nnjKbXTh3Yxfy41PO0hmyzWawv2tedCV+MCS0yoWYdNoOgSld7U3YXQs1d/PFtXo2vz/W7d1RvOdW\nT0sc6N3/mtMdNsyieV0Sa5UcKeMxXS6r5vO23yUJmqoQs93+6zJuvdozWxpo3dIlPLgrbT5BGCy1\nUDmdTryZtkwTHu+yculmZDuIuK2Jy5LYJGZ8a8ixdCFeXVKrkjk+eWtizDanjJFyYVlcKGPSKTkE\nxnS1174v6SOXFVQDzVjJXMLi0qKBy+IUlgdIlYThYK8rXvx36nG1uayG7P8u7Q0SukIi7pktJHTn\nP5NHnMRUXj1YaqFyptFtej9kW3c2ujRXroOexLWiS77qzV2JJJCOTUqXHBlt9s1cYglyXacRFq7N\nqKv9HRLSLKrxrBqyskPeWRc2IYiuHV1xzY+ZtTYmTG8Z194ZTiEZV9BobnXWl12PlC6TLpdVqZLW\nDum8aMfTpIyU3kxdOrPZILVKS2IqCf1+f8llk6sXx3jgQh62TISL80jpzhMSUro1rkQxoXGzk8N/\nvFybedpDdkhLcpv3ubWFeIdXjYu7nE5p3SHRlbEMKWh09wqGPNxuu/HQrot2W2VL3XTtiQU1dcV1\n47Ur1LviFqnrXxNd4WOhEc6i3e+vYjgchujsE7SAYytUhj1saeuqR7dcQvKMIaxQrnMl8oe0v2Ie\nWjRwX9URcnNKF+dxNbq6xJxTbbp2aizv8t+XH644o5CKupAInZQtbPhISotgN8qugi2xoLQk2T6j\nSWa0LK6g0v49/rzMREiL9mRyiIODg6YJTTqB8RR4eK8n+hwXHJ+WGOhO7IUL3bmUPc8YysldTEiZ\nRhoNnby/NIcWraXBtiEXLXptatXtfSrtg+Oqqa1Hm+7GkjmpXTs2uA+YLn6jQTcOea5YT3f/txNf\nGNq7IOR+qYnPjAvpumye1b2cWLB5ps2UHkjuUIhu7BXSxI32PpQq15bDmyEsL2JMJhPfbGUn6BiO\nkVAZ05LjQsy60rXDZAzcmYO/hBGHjeWSZC2LuSHHjEXUxHm4+6Cbm2BYdFeQ1liFQsLn0BGL37Qb\nl5eHyw1dmugrlaIuNO9p5zofLb8OCQ0trnlg48GhrYMaD6S0oRCafUimILD1idR6G3fOduP6kmrB\n+SSmclmx5APXwyJ5QDfcFmdId99bWNQnd0kL+31uJtpyL5RbGvzLLh8UunmFi6sPurUu46DLbeyO\nC5nLIyMev4nZBynHvq2xlPMezV3KUph8sj7+TcuvQ0JDi9sCWM+DwyeKCbd3yJAy/jucdVc6dvYs\nqrq105V9q0pwPompXF4stVDZ6/WPJmJ3Dk/Lk0zAzpRiasFlTFrKiMNCkrVMmsDJv49DxhKFhb0P\nwh4cQl64HA7dHRsgrAY5JDSeJaEvK5cgpQuqq49iXii+eF/aRp3VTY/6+DdpwrKUSdskc1STXTr0\nvq5RIIRMUBYvMaPUumt/XnYVjS2LqnbtpOVd/vclH4eYyqsZSy1USu6plEG3mYRcrMdFy16s155N\nrdzmeHeCxUTY/pXGEnVFsAmX6EHuXhwzTi+PbmYt1KLNON4y/F34YkLughovM3jxd5l3h6xumbJM\n4i4b2rVWGv9mQ3fjNe1I6SbdlgJB7hHTPGN/WA8kF/yVVvJ60nnQyfbuKgXLSUzlsmKphcr8PZVh\nrYPaFNIhmXi7cU710FmN7NnUuikkShFTIeAqe1kEG1lcqbw/24gLTqsBjoc2568JmQtfPMS02knr\nyv8u9e7Q1O0Wdv3dZcO79bdzJ2m7/LadBE7SusLSpfOICRlrHlKIl5UlvXrGFPSaK12l1tsueRKe\nIC6WXKhcIPSk7W4K6W4EV8utRvlDYThtcreQMllUyqzBtqtmYlr1dfdrthcX3A3LgBxtzl+J+9ry\nKJ1k67C5W6/cu8O/bBPycAR/17duQROekA4pLZNlxHPb1dER857y5SjLp+ymSld5bG7+e3es+icI\nj2MjVIbVUIZMIR0WXbFAyV1C8m5Oy3MoLCLdfY56xLn7U+uSGpPOtGW3g5hZgKVxSnHrliAlnVIv\nFtsctLushrSmpFVGStzBj8saba8dMXnCsgoAcndYDULGg6fbD3VKV41BIlVYygnawJILlYvsr2Gh\nCYQH2rPspITU9SLv5rQcdy+ZkLp86NCdPsm3a7ncWmzrMFY9cREzC7A8TkkC+5p3WbdstOjDFWQI\nddG8y2XVvbb85537IKeZw+msQtK62ysrHWKu0+7ElcrKDrl2Qr67PIrn0GNZLK8719WdIDSWWqhc\nWel1cnHmN/B2LTvd2XCL8T/FQ+CyaK5SClBuq1E7m6LWJTUsLXbYLKyx6omNuHMwZObeItx8sJ6H\numkh9HqzZDXz7/Y51XzOyS6ad5Xlclm1u/RJ553tIKeZw2mtQtK6JWUtxz5UhVCHdE1sbng+qEnO\nJYmD1F5PEi/pT1hIlFDSRID+XlztXld3gthYaqGSeZxoccoPDnx0MLPdMxSXzm5tuEUmb2r702mu\nNAdOacB+c7isRnKm35zO7mjCZbBZWGPVEx9djZeyo4mFu8gT7AkrptNDLATSmPfaajL96mJWNQd+\n10FOM4eXKXGSC8toQQl5SO/SPNAl5/K3nkv5QVyPjniQ0Onqb+l5ZFlj4rsGInoBEf0VEd1JRG+s\n+P1niOgT2eeviehC7rd/SUR/mX2+NRaNrQiVRHQXEf1F1vA7sr+dI6IPEdHnsn/POgtixvbmvm+t\nAV0WXAxrweRt9wyFhubQMYO9XaE2XFPbH1dzJRNiNBtEyux2bqZfrwSQx8PGs1ils7x1J0lEdxHP\nwie1cNt5guk6W3zWFWPWVmINrWJOl4wn5rVM0njObiSbc82j9pDyHla7oiNsXK8Mk8kEk/EIk8lE\n/K7Eei7nB809OvQCqMYzQhJ7K0sE6E7OdSJIakFEfQA/B+AbATwFwO1E9JT8M8z8A8z8dGZ+OoD/\nE8BvZO/+TwCeCeDpAL4KwA8R0ekYdLZpqfy6rPG3Zd/fCODDzHwrgA9n34NBtqHb433cQqaGYcV0\np7LDxvDCbrihXSjrIRVidEkPUgoubkGGa3+T0qm54LoI7RxNB40SKmSMX3eS5aSMsbTNC7fr7CqG\nG2e9DpR6yMrSKeZSrpVw7vVhQ0DCrQfXPGoLKS1fafMEyDAYrOPU9qMxGKx7nLk016rEy5Sso6sM\nnZt6yHh8+3lkWay3S4ZnA7iTmT/PzCMA7wbwUsvztwN4V/b/pwD4A2YeM/MugE8BeEEMIrvk/vpS\nAO/M/v9OAN8ke9292fhu6K7NRub+EE6wc0PHHO0Mb1kONEVIhZiYiVBiokp4Pth7pJF7j4nBYAOn\nzz0eg8GGmI4yunFwc0ETNzN7mgrfliGxg1vhkjLGsjmdQNm9vgtIawnT8dCQMZYhk3vFDDHoCtxW\nn5h1mWiPX49Ge7h0/i6MRnvOM9fyKAhintFc78bMEmzWZY+BX05PntbxGABfzH2/O/tbCUT0OAC3\nAPjP2Z8+CeAFRLRBRNcC+DoAj41BZFtCJQP4IBH9KRG9NvvbDcx8b/b/+wDc4FvY9uaBh8bNf0OX\nmvJ1m1N7FxeX0VVGrIE2k+9yoLzpLoRjbZvym7uUjrSIp4yQCJmme3d3XKpdCOki6bJYa2LN3a6z\n4e7BjekSLEUPg8Ep+GzbWmt52Ni6cJ4pMUMMymgrOZrU6hPSLT0lXHRPMRpdBjCFOxwkncAkQXj+\nrbtb3O7BYYPWJVuTdOmqwbVEdEfu81r3K7V4OYD/wMwTAGDmDwJ4H4A/wsx6+VEAcr9yD7QlVH4N\nMz8TM9/g1xHRP8j/yMyzIKgKENFr553+yMVFPGWZqWhc/qRMvXlMVVc1bG2i3Q1iWfs/XubOsLFb\n8WA/SEsPiBrvhCJ0Y7FM89H/4BA31jzOHa1Nfm/6bBXsHgPl+ayz1saMsZRAuo/b1rVMQVC2hjWP\nidUkdCrzj6JyYVkP6SG9stJ6GDVXELSNpjxBq6QqJl1yndWXH6MpcNcVEn0APMTMt+U+bzeK/RKK\n1sWbsr9V4eVYuL4CAJj5LVnI4fMx6+y/DtPaIloRKpn5S9m/DwD4Tcx8he8nohsBIPv3gZp33z7v\n9LPba0d/N5lKm24y6Q6cXYbG5S/kBnH8GJYfFpuZdi10JQbKrzZ3SvOq7274C+3HV1EkSRufzpU2\nJkK2I6bHQNnlNKS1tj1I16nteamCoGq86kNowt3V6eIfpnJhWdaSibYSZmnRXSHenUDSnye4lFTN\n7w2NeVY/5vgTALcS0S1ENMBMcHyv+RARPRnAWcyskfO/9Ynomuz/TwPwNAAfjEFkcqGSiDaJaGv+\nfwDfAOAvMeucV2WPvQrAb0vKPbV2OeKl7DJLZNgg8eUUisK6/DV3Q9K7Dvm7nIVFSFdOneW3OB7a\nO71Copy9sW7TDH+1T/26lc/f5VjjsqQeMbONpkTMdoTL8rk8CbBkkK5TG69yrUtXH9pCaELGkrva\nXFYuLGe+g656J7jQVSE+ZII4l5JKo5R1x1gT+v1+l/K9dALMPAbwegAfAPBZAO9h5k8T0U8S0Uty\nj74cwLszj885VgH8VyL6DIC3A3hFVl5wtDFwNwD4QyL6JICPA/gdZn4/gJ8C8Hwi+hyA52XfFWjz\nwmX/bGCu35f1QuaQLn8xk3q4ypIkqQkJbaKYYllay29zq6dOoPW3jJl0lp919UG8w5Jr/qYUvDXP\nSmPNlwcal0l/aMfd/rw2o3O4eRIW0tASG69yhbU0VyC497uQLpP5WMO0WB6rUso52h2loM29NeS5\nKWyW/Irs3cPhch14E4GZ38fMX8bMT2Tmt2R/+3Fmfm/umTcx8xuN9/aZ+SnZ5znM/IlYNCYXKrN0\nuF+ZfZ6a65iHmfm5zHwrMz+Pmc9Lyu31h8mEL82CahKjs4wXMmviTE3ETephPyhLktSEhCaGz1VW\nSLpc0Ai0MsuYHe7sgfGuAXFpZmNqvsPG/B3PtPEal0k3FnNBO+4h54lGwG1z3CWeJ2ldsF1ZxY/H\nWklrpetm1uy0SkF/lD11wl2zUm6zRlHtVl4eHBwst9/+VYxjZWLeXEvjjqBZUO6LuU0BqitbEgDY\nAAAeoklEQVQXMuugYbwpk3qU6dRohcNl5euqy40LMrqlm2DzBEJVyoRebxUhtM1uC4j8ztEFmsey\nhHy2yfNFdMcF2NYO7borKiu0WVBDKuqaZ89My4uK4QcuOotrz9Vf6VyXw1oyQyKcC3ZotKdIlXls\ndGdvllri/cc+ZJvdClzGZDJJb4Y/QRAcG6FyOjnA/t69OLW2E6H0cNd+uATS5U1SEzLOVPpuO4dw\nt3tmSNdlXSpx2eZc3y5tghsZnTEPMOUD/nR6KKjLdqVIvAuu5W7RMTP3hpuTurWi5XO2duhj3hcu\naVI6pe7fzSHbd7TCrf/zZviBi07b2gvbX9JY/W5aMkPuDaHfDeti6Q95eFNbbv+6M5e9nXFDG7p4\nj/AJwmCphcp+P6/MYEwnB4ihddQddmUCqc7t0Z4SPSbch914Kerth3CpW6NGCKoqLV9OTCG/3vIj\nd1mtb1dMrWybGl+pR4Auvq35YaDMH0K67aZDVRubuvmH5XNhE1HlXdKkYyVz/9bRrVOUhowNLcIM\nP9B4L6QMA5D2SVu8T743NOc38nmS8pqQBbpjebRDLvwWYRPam4Rp+eK4ZKY+QTWWWqisw/bmQcVf\nm1tfUm64GrdH90EhnlATUhjW1l0uPY5WzEdBcLD3SMM+kcFu+ZEJ9PZ2LWumQRc0fSRz5ZQdBtzK\nGfv8TpOERg7/zJouyPlzOEu8VMiR8CJJ2Xq6w7nGyZ531WOGH3Tl/kxprH5XIe+TpvtpXPf6rmal\njQe9R0y90O4+27isyLbxCOfGf4LuYemFyu2t6gQq25sjb6tdSutLWPdBWUr0uJlkwwnDbvhbQd1a\nsXgMzWTa9j4JL1g3T/DUDtNvNwFC89giKd0h14ZrfstcadtEOmHB5lUhV8RJroqQaujDXUMhRbGP\nXNcqNXeb7vYal5Sl88zp94c4deYm9PvDALTFg30Oh+2T0F5D9ehOfLcM8TxifLzDhhtng1gyu7UP\nnUCLpRcq69DrD3DtNVs4e9pHyLEfonWTPl7CFblbLqHXD5OMxI14ByLZeMSLZZELD7YrL7RXiISz\n/LTlRt1mAgSdO4+U7pBro7nFZHmsKWFhatmLYx86w21+LaWMEbZDwrtiXqu0LGvcVZa27H5/FcO1\nM0t9sG5TKRjSm6xbQk43M+ACdou1ZDyu1n3ouOLYCpXTyQij/YcxnYywvXmA7c19nFrbwfbmfo17\n7AIh401iQk6XxgIlu19ThpDumTLILBOyy7QlmkS9RdtEuEQFmtgtGdpLgBB2jYe09Gr7QJOEZlkg\njSf09yDQu5hpwgDiJUWRKCvaulapGeIl0LKVpS17NNrFhQf/GqPRrsfT7VnSyvN5YcVuUynohv+8\n0Ll2NqejCiGVnWHLdnldSPbA47IPnQA4JkJltQusPXHPTNBcfPKQWJikCKs9kgpjRetVSBcFd1kx\nYx7CHdptLsIuy7C0D2wWkpTxsCbKm2pb7mopES9b4PHts24gfDxhHjrer3F11giksrhdFzTXKtkR\nem3Yy4sZU6kt27+P2+Qn5nwuWrFTKgVl810yL0K6dmrf1Qnm3Yn7PdkDrx4cC6EyBIpC5n52NUl4\nZtimsCBzxdVlrU3pQhk6noL6K4XvvllUpZpBiUt2SqasCeB3I6R1PCTC3fknXTux6EoPSVKgcO2K\nGQepR3NXZ41AGvbAGDPB2wST8QiTySRQeW2ttXjXqphwtzGdN8lodAWXL96D0eiKx7saHmA/T2j3\n5nJtYVw79e/659RIibTC8QmWCcdGqKxL2KMq80jIHOHs6VnynyrLpgztCQsuWvKQx2va2xXzkKO5\nz6qqnTwZ59phu7xc5u5qswC6kJIp2w/pugN53AD+tuJPytbuYtr95pb1dvlDOFetuHFLLp66LIJ5\nc0tPTP6gX4f1/T8YrOPU9qMxGKwHojakx4Fmnca8YzTmnZfSNd9Hf2WAfr/v8WyRL2r2Ark3Wbo4\ndjvCeVmFiOMN51rbJUXeCdpE2guAImMuWF7cCZtQoNcfYLB2TRajORMomwuWhF5/gOnRAhuhRyOs\nrHVtwR1iZQVYWdtFM7cns12EXp9z7W6OXn+IwdrW0XjMvg8x2r98ND7N6RyhNz3Aqex7r08Yrh9i\nZToqlW3SUUZxrHV0AsC+4Nl83TDmnKwuezvN+eyCZL7L1oZ7PMLVZaejfs7I6WyPP8j700ar+ZuW\nv/jT0euvYW3jGuzv3XfUjou73c6yKcf80DbD/MA4O/g1Sda1gFRgNeu20ZI2XpPQ769m7XDfYynp\nw7rkT3XvahShLoRIWuM7bzQWQF0yF1bP6zrMPXUmk1GAOuxzrtzf9ufz0CqSwrjWzhCS3xwL8Bi4\ncn/bVLSCYyVUzuFntSRQbwCejrC9danityF4Oo/JJFDvgdz35qDeOvqDx2AyehA8rXIZqap7GKRu\nGZ1r6K2uYHr4CHhaJczI6JqVdx2mhxdqyhNRZ4yHZnxc7SBQ737wqarf7fWW2yyhUzoPir/n6wag\n7Pt6usOOqxbh1qlu3dnmTBWdtrraWf/VdFb97ktb1fwMxV/sz8547iomowdqeG7Tehe4uLOVe3+2\nr7jeLys/pQoaG0IL7ZK1XaGoq1E2XNwdZrGETeF/CHcffIsHZZn7cVEQcYdChBRcTDQXuJq7Z1ah\nODaL5C5NBMN4QqQJeR+Yc3Dx3a1ckCoyyn3aHP5j5/oeU0lyguXCsRQqXdje2skONduYHl4GT4Ez\nWw8tHqB1oH8KmDwMsF+sAGgN4H04DwO0AaxsAoOHAa4Qfs26S7Tk64J/vV4wy34YGNSULe6jHtDb\nBQYX4T7kCPpTDKNsZztstLjoJHsfWsl0zQPH84W6YdARsn/NNsYcu6q683UFrFs0v5X12uoSr7OE\nsNLmWmeOtSFpt9cafghYta3TjFZaqyzrws61VhK2t3ZmJeX2lQsX7W5lpvKTekOsDM/g/MPw8nyw\nCaG9/ipWVrcwnVRby+OCjTrN7wu4PX7sgvbl/S2BNTGkwOQq2/6ujBaX4OwvWLvfDSe8lQUkadma\ndknKhqoP7JZ5X1faGdzKiAE2tq7H3s4DmIjXdXMFjPt7aOtu/9iE5l1tuCqEyvmGX/jb5pfqDzW8\nD0zuWwhXLtAa0H9U9o7jAMRXgPHf1pdt1m1+p3Vg5WZg/IUZ3b71NmmHrUxxHw2B3pnZ885DoqA/\npSi10dEOGy1OOtlBv0UYcc2DUlXm7/m6jTiHoP1rtDHm2Jkw6wpZt2R+i+s1xt1Wl3SdpYSNNuc6\nc6wNSbudzzrqKtBaXdaZrYecgiUA8PQA08MHwdMDbG9Jx+wyqHcZGzfUez7sjh6P8cEeeHpwJITO\nvxdxBdTbw3AjrNt06NASF8qhJ4TL+6dyB2OJcBbT2iUt2/95l/VK40IZ022xOu7OX0i006YTOGdZ\n3m/A7s79AFjVB/YEW9p5UW6nLYmQDTJ3bnvSME1SMR86h8NhOf3+CZYCx1KorBIiC5ZIAKWDdgEu\nYcB8XHLwE5Ztez70gTNoO8yD8wEwvTD711m04FkpnIfbhAd+qzBi0uXqb8vveUUE78UVVFIKQSXB\nO+S8EcxvtRLKVlf5Nx/hxkZ3eHfazZp6DsHTU8bvVc9Kyw7xbB5VtG7mfpP0FxfceS/unM65w8LT\nNbbawnlxZxO9/h6mkxUAfQA9rAzGGI9Ws+9V8E2AE9L1Nlxd8/umH7lEAIbo9wdK61coxLQeFuHO\nbCqL55SUrYMZdyezstloywuFcosdACyyvE8mI2UfmHMwnrW37ELsD7+swXWutdLvzTGZjHF4eHji\nR7ukWHKhkrC9dRkupl0WKCGyLvgf4poeanJkHcWoPWiJNboMnm4CmGuN9fUWEb4d7nbVvyus2fMQ\nWN3G+ro3LWWHPMw2RX27qbeO3mAb09E14Gne0hB63uQRo2x7nKlu3ggpqazLt832cT+z9bDVnbaS\nn/mCNoCVa4DxeVS639tfTujaHBIKuitca333g4s7W6DeACvDazE+mI3Z/P9ly6Ib21u7AHYXpPWG\nWBmuYHxw2Ki8PPJWzwsX3Zk8NahKfJejxBA4Z66z20fLxJ24KlwSpnrB0G09dAk9EqHTfmAvx3NK\n4tvSCuUyK5udNlbxoHyW95R9IFM21GeMbwLXPCq67cqUJuGUKP3+ClZXV5dcNrl6sdQD11/pZYdK\ne6KH6kNA/mB3qsVkGEXk3adqnuhAQhQ3zHa421X/rgTUG6oEi3Ldi3mkLbuitmBjaaONp/uYjr6k\nPni2DbON5nfZvNFZ7DRz1DXuF3cfUxrL5oJkyDjThK7NIdGQ7tm+0VzxM/OYmbm08spsniz+H2Kf\nuWwpj3Bp9xqvhEEAwNNRJuyOsL0Vew+sd811C7dmfGbZ6qm78msBe/Zju3Db6xPGln4Pm53XtFTG\nTALUHBorW1VZezv3Ny4rJC0SSMc95Vjm55HU5TrkfP7al/w0Dg5+tjsT9wQiLLVQCR5XHv6PrBaD\nR2M6uifL+GceIhcHu5klJ/9smwgpNIZ0ddOWJWlX8z7QHfbLdRcFF23Z8TJ52oThtIqIdG3k6QjT\n8aUj10IRlU4Fgasd8fp03s7tzS9V1C0UEqvipI9iuoVlidx8fZJYRUqyZKLkFu1XlynI1ysnF/Nk\nkQXWxGr27AA8HVjrBeaxi67EPLZEPcOCNdDfepcyZrKqLkK/P8ZksgmfeXDkDrsf4yqDHgb7BxiN\negCGqLbG1Id+9fv1Qk/YGDRZ5tl0iJcESF+W5P10rsza5zUoziN7PKw7plLXZ5PJJPQ9UydIhCUX\nKqeYT9iqQ2Le1aL8e/4w0Dz4WY94h/CQljVpWeGtetbaoglQRUFGV3bcPrEJw+ks22nbOEBv5TSm\nfFBpucyjMqY6aHbjkJgLPfn/Z3RKrW62GGJaD2x5zGdRNeJ4S4+aSZYcz6vIMpKEleqSCJnVWW0v\nXpx5BFTF8wNlV1iXYmp7a6+UiMf1vYiiNdDvmq05rSkFyyrB2I9vXNwdRj10m8Ja2KQ2IWPQYt7h\n2Fw46NbdhSnbocksqxHEZdeAmLAlGHJn8i3Hzxbdv211L3577je/1bOtJ+gqlluozMG0Yly4uArq\nXQBPVzFrJoF6VzIt8Wq2IV975PJTfDYdinTo3Ha2ty4bwpXWsraAtKyQdbsQVpCpt2jL3y0iZZ+k\nrKuteiUu1vVxcHUujT1QvweebFmeCYX62FCAjrwotje/KLQWAgUhsvSTNsGQ5XcnWS0mdDIhEdRr\nstpub9YLpBd2rq1whfU52JrP275DrZycW1ndAqj/XZwu2AXjOjpnQu/CzTXWHPK/b9NErz/EiFMJ\nVPFiAjWCYXcspmnbkU9IBMgyy8roLApq0mtAyqifR9IkP7PSFmvElqRpTtdznvcjjvaeYBlwbIRK\n8/C/vXXJ8bxsg693a9IhH8uixaXda2oEVPt9aTJIy/J5XntIKSoMNNAI+f7vzmkMdzibo85Skg4p\nXW3Numx1yzwCTCuoDNK66mNDqbd25EVRFIyLgm6zmMv/v717D7arLO84/v2dk5xcgXAXoRIsKEaC\nARJgClSBQrEXLhU0KVXjWC8zhhYYp2KHiljLqKOACtOqtEBLB1AgFAsaBAQZRAjhknCrDZRqkA63\ncElicsLJ0z/Wu8k6++z73mufvXN+n5kzZ6+13vWuZ6/9nn32s993vavDs1yPGmYatW+dVH7serda\nakp5z2OdY+WeV/1JeKp92TB6tthXXq01xLXR96hqw2Unj+lNLB/u2r58DyKjehNrHauxYbblPTml\n4a7dvV1J471XQ2XlVaF8eV0DDA0VdTP4Tt6Xsrb2EsPxmpl3rM4+j/rnsPSe3dykSc3FOXZm2PIh\nqaOP3WwszagUy6YNa0c9j9I5OfaUb3f8+BOFpBOAb5JN931ZRHylbPtFwNFpcTqwW0TMknQ0cFGu\n6P7Awoi4sdMx9ndSqQGg0ht8I5r78Dv+H9Yb0ew34bUUNyx3zJFyNwwfOzS52xMntXMO6+1bqUcq\n/7zbUapbqe7xGf7ant4ZCt7NyaLGHmvre1Ns2cjI8Jq6cbR3m5FmVZqx9sWxw0w7deukZjV9/9it\ny3WTc02DwRkw8vqoOkrnv/Taz9qh+b+70bcfiZrDZcf2Jm4d7vrq643eRqS6gcEpTJ3+FjZu+D8g\nRiWRW0Y2s/aVetcaVlfpNgmN61xC1fykKdXLF3sz+HpxND7ssdnbetRPDAcYGprO8PAGoJcugyvu\nes565z8/CdDg4OQ67aD8/DUeZ71hz2OHb7feJuv9rYyMbHZPY8EkDQKXAscBa4Dlkm6KiMdLZSLi\nrFz5M4CD0vqfAvPS+p2A1cCtRcTZ50nlpBqzv9bduWdmfO2c7swm2mnlH6qLPXaRE7DU3re92Upr\nG8+ZUDulyNe9+XNdrx1UP2edPVZ7f9Nt3X6krqxH8JXXZrY9U2ojx2q8jRZ5q57addd77WuNeClP\nIstHsTR2rWO1hLLevSHHbg9lv0v3itzaazmZHWdMYXjjOraMbKozS2oljQ8jLbdu43YtD2Os17PT\n3DVntbcVOfSzvO5mhj1mz3EIOjSHxNDQdLbfaTavvfwMw8PrOlJnJxR5PWe9859P7uq1g/bOX+0E\ntJ026d7EnnQosDoingaQdA1wEvB4lfKLgPMqrD8V+FFEpycvyPR1UjlpytvYdc61o9Y9/+hJ6VHt\nDyH92ZPTPd29Jm/0B+f2jt3a617U8OaxsVUaptuJ4cn1667W217sNamNK7bNlSdnte95WU/5rMCt\nX4dbrFIPWmvJZZ0ZWlOP4KztSz2CBc7g2tTESc3OPNvMcbcm0q2oPeKl9kiHZkbLlL+f1bt2cez2\njQwMvsHQ9HVkPSjrc6XLbwlS/RYh1a2vX4SxifTMqetaTkgrJ7Nb/04bS45r/V2Xb8v9f1k/tYP3\n9RudTNRLHvLLg4NDbOnALKYlw8MbUkJUyOfTlhV7PWft81+rbLn2zl+9NpQd2wniNmNP4Ne55TXA\nYZUKStob2Ae4o8LmhcCFHY8u6eukcsWKFS9K+t9K2wYHBwemTJkytGnTpuFq0xMPDg4OeOribUuH\nXvddgCK7eHpOp/4WGjn/vaA8zlbiLp2zFp9zz7exRs5Rvt0U/do32kbrxdFsnM08xx5s/2+2s3rn\nr5uvZa8q6jNBJ95vOhlLh4/Z8+9l26LG29Al3Quqc/Ye7wDasvHlZTx5dbPXpEyV9EBu+bsR8d0W\nI1gIXBcRI/mVkvYA5gLLWqy3rr5OKiNi1/GOwbY9kh6IiPnjHYdtu9zGrBvczqxobmNmo0XECQVU\n+yzwO7nlvdK6ShYCn6mw/oPA0ogobErmgaIqNjMzMzMzs7YsB/aTtI+kIbLE8abyQpL2B3YE7q1Q\nxyLg6iKDdFJpZmZmZmbWgyLiDWAJ2dDVJ4DvR8Rjkr4k6cRc0YXANREx6kJbSbPJejrvKjJOlR3X\nbMKT9Mk2xrKb1eU2Zt3gdmZFcxszsxInlWZmZmZmZtYyD381MzMzMzOzljmptAlD0gmS/kvSaknn\nVNg+RdK1aft9aQw6ko6TtELSqvT7mG7Hbv2j1XaW2/42SeskfbZbMVt/aaeNSTpQ0r2SHkvvaVO7\nGbv1jzb+Z06WdGVqX09I+ny3Yzez7nNSaROCpEHgUuD9wBxgkaQ5ZcU+DqyNiH2Bi4CvpvUvAn8a\nEXOBjwL/1p2ord+02c5KLgR+VHSs1p/aaWOSJgFXAZ+OiHcD7wMKm17e+leb72WnAVPS/8xDgE+V\nf3lmZtseJ5U2URwKrI6IpyNiGLgGOKmszEnAlenxdcCxkhQRD0XEb9L6x4BpkqZ0JWrrNy23MwBJ\nJwP/Q9bOzCppp40dD6yMiEcAIuKl8htkmyXttLMAZqQvMaYBw8Br3QnbzMaLk0qbKPYEfp1bXpPW\nVSyTpm9+Fdi5rMwHgAcjYlNBcVp/a7mdSZoJfA44vwtxWv9q573sHUBIWibpQUl/04V4rT+1086u\nA9YDzwG/Ar4eES8XHbCZja9J4x2AWb+Q9G6y4T3Hj3cstk36InBRRKxLHZdmnTYJOBJYAGwAbpe0\nIiJuH9+wbBtzKDACvJXsRux3S7otIp4e37DMrEjuqbSJ4lmyG7+W7JXWVSyThu3sALyUlvcClgIf\niYinCo/W+lU77eww4GuSngHOBP5W0pKiA7a+004bWwP8LCJejIgNwC3AwYVHbP2onXb258CPI2Jz\nRDwP3APMLzxiMxtXTiptolgO7CdpH0lDwELgprIyN5FNxANwKnBHRISkWcDNwDkRcU/XIrZ+1HI7\ni4ijImJ2RMwGLgYuiIhLuhW49Y2W2xiwDJgraXpKAt4LPN6luK2/tNPOfgUcAyBpBnA48GRXojaz\nceOk0iaEdL3HErIPVU8A34+IxyR9SdKJqdg/k13btho4GyhNob4E2Bf4gqSH089uXX4K1gfabGdm\ndbXTxiJiLdnswsuBh8muD7+528/Bel+b72WXAjMlPUbW1i6PiJXdfQZm1m3KvlQyMzMzMzMza557\nKs3MzMzMzKxlTirNzMzMzMysZU4qzczMzMzMrGVOKs3MzMzMzKxlTirNzMzMzMysZU4qzcz6jKR5\nkv5ovOMwMzMzAyeVZmY9RVJIuiq3PEnSC5L+My3PAL4BPFCjjrdKuq6A2KZIui3dq/VDHajvTEm/\nkPQDSXPbqGexpEvajaeTJB0u6Xvp8dWSVko6q0b5KySdmh4/I2mXbsVqZmbWrknjHYCZmY2yHjhA\n0rSI+C1wHPBsbvsBwFkR8Xy1CiLiN8CpBcR2UKp/XvkGSYMRMdJMZRFxMXBxh2LrNe8HfizpLcCC\niNh3vAMyMzMrinsqzcx6zy3AH6fHi4Crc9veBXwS3uzd+pakn0t6OtfTNVvSo+nxYkk3SvpJ6gFb\nIulsSQ+lXsKdUrlPSFou6RFJ10uang9I0m7AVcCC1FP5u6m+r0p6EDitWh2Sdpe0NK1/WNJ8STMl\n3S7pQUmrJJ2UO9bZkh5NP2dWOkGSPibpl5LuB47Ird81HXt5+jmiwr6LJf2HpDsl/bek83Lb/kLS\n/SnO70gaTD9XpHhWlXoc0zDkX6ReyKWSdswd5ljgNuBWYM9U31F19sk7I3du9k/HO1TSvem1+7mk\nd1bZ18zMrKucVJqZ9Z5rgIWSpgIHAvfVKLsHcCTwJ8BXqpQ5APgzYAHwD8CGiDgIuBf4SCpzQ0Qs\niIj3AE8AH89XkHpG/xK4OyLmRcRTadNLEXFwRFxTo45vAXek9fOBXwIbgVMi4mDgaOAbyhwCfAw4\nDDgc+ISkg/KxSNoDOJ8smTwSmJPb/E3goohYAHwAuKzKOTk0bT+QLCGeL+ldwIeAI1Jv7AhwOjAP\n2DMiDoiIucDlqY5/BT4XEQcCq4DzUny7AJsj4lXgROCpdM7urrZPBS+mc/OPwGfTuieBo9Jr9wXg\ngir7mpmZdZWHv5qZ9ZiIWClpNlkv5S11it8YEVuAxyXtXqXMTyPideB1Sa8CP0zrV5ElVZANuf0y\nMAuYCSxrMNxrc4+r1XEM8OH03N4AXpM0GbhA0u8DW4A9gd3JksSlEbEeQNINwFHAQ7njHAbcGREv\npDLXAu9I2/4AmCOpVHZ7STMjYl1Z3D+JiJdyxzgSeAM4BFie9p8GPE92vt4u6dvAzcCtknYAZkXE\nXam+K4EfpMfHk/VQjlJnn3I3pN8ryL4QANgBuFLSfkAAk6vsa2Zm1lVOKs3MetNNwNeB9wE71yi3\nKfdYDZTZklvewtb/A1cAJ0fEI5IWp+M2Yn3ucTN1nA7sChwSEZslPQNMbfCYtQwAh0fExjrlosKy\ngCsj4vPlhSW9B/hD4NPAB4Gqk+6QXU95YcMRV1Z6jUbY+hr9PdkXBKekLx3ubPMYZmZmHeHhr2Zm\nvelfgPMjYlWXjrcd8FzqQTy9w3XcDnwK3pzNdnuyXrfnU0J5NLB3Kns3cLKk6cpmuj0lrcu7D3iv\npJ3TsU7LbbsVOKO0IGnMpELJcZJ2kjQNOBm4J8V5arp+lLR97zScdSAirgfOBQ5OQ1vXSjoq1fdh\n4C5lXZwHAg+XH7DaPlXiq2QHtk7atLiJ/czMzArlnkozsx4UEWvIrkXslr8jS9ZeSL+362Adfw18\nT9I5wEtk10z+O/BDSavIbo/yJEBEPCjpCuD+tO9lEZEf+kpEPCfpi2TXhL7C6ATur4BLJa0k+x/3\nM7LexXL3A9cDewFXRcQDAJLOJRveOgBsBj4D/Ba4PK0DKPVkfhT4pzQh0dPpeR0CPBQR5T2h1Nin\nUV8jG/56LtkwXDMzs56g6v/3zMzMOkfS7wHvjIjL6xYuNo7FwPyIWFJA3ecCq9PERWZmZhOCh7+a\nmVnhJC0im/l0m/4mMyK+7ITSzMwmGvdUmpmZmZmZWcvcU2lmZmZmZmYtc1JpZmZmZmZmLXNSaWZm\nZmZmZi1zUmlmZmZmZmYtc1JpZmZmZmZmLXNSaWZmZmZmZi37f7OMoXgihyHEAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 1224x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "'''Código fonte para a imagem principal da aula'''\n",
    "\n",
    "# Importando o que precisamos\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.datasets import load_breast_cancer\n",
    "from sklearn.metrics import f1_score\n",
    "from sklearn.model_selection import cross_val_score\n",
    "\n",
    "# Carregando o conjunto de dados\n",
    "entradas, saida = load_breast_cancer(return_X_y=True)\n",
    "\n",
    "# Vamos passar o modelo preditivo, os dados de entrada e saida, a funcao de desempenho e o numero de particoes para a funcao de \n",
    "# validacao cruzada\n",
    "desempenho_cross_val = cross_val_score(estimator = DecisionTreeClassifier(random_state=37),\n",
    "                                       X = entradas, \n",
    "                                       y = saida,\n",
    "                                       scoring = 'roc_auc',\n",
    "                                       cv = 3)\n",
    "\n",
    "print('Desempenho médio de AUC-ROC: {}'.format(round(desempenho_cross_val.mean(),3)))\n",
    "# Importando o Random Search com cross-validation\n",
    "from sklearn.model_selection import RandomizedSearchCV\n",
    "\n",
    "# Definindo o Random Search CV. Vamos fornecer o argumento n_iter, que fala quantas configurações de hparams testar:\n",
    "random_search_cv = RandomizedSearchCV(estimator = DecisionTreeClassifier(random_state = 37),\n",
    "                              param_distributions = {'min_weight_fraction_leaf':[ix/2001 for ix in range(1,200)], \n",
    "                                            'min_samples_leaf':range(1,450)} , \n",
    "                              scoring='f1', \n",
    "                              cv = 3,\n",
    "                              n_iter = 5000)\n",
    "\n",
    "# Realizando a otimização por GridSearch para os dados de cancer de mama:\n",
    "random_search_cv.fit(entradas,saida)\n",
    "\n",
    "# Transpondo os dados para um DataFrame\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "df = pd.DataFrame(random_search_cv.cv_results_).dropna()[['param_min_samples_leaf','param_min_weight_fraction_leaf','mean_test_score']].apply(lambda col: col.astype(np.float64))\n",
    "df.info()\n",
    "\n",
    "#Plotando o gráfico\n",
    "from matplotlib import pyplot as plt\n",
    "fig, ax = plt.subplots(figsize=(17,5))\n",
    "\n",
    "y = df['param_min_samples_leaf']\n",
    "x = df['param_min_weight_fraction_leaf']\n",
    "score = df['mean_test_score']\n",
    "\n",
    "cntr = ax.tricontourf(x, y, score, levels=20, cmap=\"cividis\")\n",
    "ax.scatter(x, y, c= 'w',alpha=0.1,s=1)\n",
    "\n",
    "fig.colorbar(cntr, ax=ax, label=\"valor AUC-ROC\")\n",
    "\n",
    "ax.set(\n",
    "    xlim=(min(x), max(x)),\n",
    "    ylim=(min(y), max(y)),\n",
    "    xlabel=\"Mínima fração de peso/folha\",\n",
    "    ylabel=\"Mínimo número de amostras/folha\",\n",
    ")\n",
    "ax.set_title('Paisagem de performance da Árvore de Decisão')\n",
    "\n",
    "plt.show()\n",
    "\n",
    "fig.savefig('/media/head.png')"
   ]
  }
 ],
 "metadata": {
  "colab": {
   "name": "Otimização de hiperparâmetros.ipynb",
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}