8_Recurrent_Neural_Network.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "8_Recurrent_Neural_Network.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jacLnynuqUXZ"
      },
      "source": [
        "# 순환 신경망 (Recurrent Neural Network, RNN)\n",
        "\n",
        "- **순서가 있는 데이터**를 입력으로 받음\n",
        "\n",
        "- 변화하는 입력에 대한 출력을 얻음\n",
        "\n",
        "- 시계열(날씨, 주가 등), 자연어와 같이 **시간의 흐름에 따라 변화하고, 그 변화가 의미를 갖는 데이터** "
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3P0gpo4gqWUz"
      },
      "source": [
        "## Feed Forward Network vs Recurrent Network\n",
        "\n",
        "- Feed Forward Net (앞먹임 구조)\n",
        "  - 일반적인 구조의 신경망\n",
        "\n",
        "  - 입력 → 은닉 → 출력층 으로 이어지는 단방향 구조\n",
        "\n",
        "  - 이전 스텝의 출력의 영향을 받지 않음\n",
        "\n",
        "- Recurrent Net (되먹임 구조)\n",
        "  - 이전 층(Layer), 또는 스텝의 출력이 다시 입력으로 연결되는 신경망 구조\n",
        "\n",
        "  - 각 스텝마다 이전 상태를 기억 시스템(Memory System)  \n",
        "\n",
        "  - 현재 상태가 이전 상태에 종속\n",
        "\n",
        "  <br>\n",
        "\n",
        "  <img src=\"https://www.researchgate.net/profile/Engin_Pekel/publication/315111480/figure/fig1/AS:472548166115333@1489675670530/Feed-forward-and-recurrent-ANN-architecture.png\">\n",
        "\n",
        "  <sub>출처: https://www.researchgate.net/figure/Feed-forward-and-recurrent-ANN-architecture_fig1_315111480</sub>\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2h5HFH0BqYho"
      },
      "source": [
        "## 순환 신경망 구조\n",
        "\n",
        "<img src=\"https://colah.github.io/posts/2015-08-Understanding-LSTMs/img/RNN-unrolled.png\" width=\"700\">\n",
        "\n",
        "<br>\n",
        "\n",
        "- 입력 $x_t$에서 $t$는 시각을 뜻함\n",
        "\n",
        "- $X_0$에 대한 출력 $Y_0$이 다음 레이어에 전달\n",
        "\n",
        "- 각각의 입력에 대해 출력은 해당 레이어대로 출력값을 반환"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kowOGfSLqbSn"
      },
      "source": [
        "## 순환 신경망의 다양한 구조\n",
        "\n",
        "<img src=\"https://static.packt-cdn.com/products/9781789346640/graphics/2d4a64ef-9cf9-4b4a-9049-cb9de7a07f89.png\">\n",
        "  \n",
        "  <sub>출처: https://subscription.packtpub.com/book/big_data_and_business_intelligence/9781789346640/11/ch11lvl1sec80/introduction</sub>\n",
        "\n",
        "- one to one\n",
        "  - RNN\n",
        "\n",
        "- one to many\n",
        "  - Image Captioning \n",
        "\n",
        "  - 이미지에 대한 설명 생성\n",
        "\n",
        "- many to one\n",
        "  - Sentiment Classification\n",
        "\n",
        "  - 문장의 긍정/부정을 판단하는 감정 분석\n",
        "\n",
        "- many to many\n",
        "  - Machine Translation\n",
        "\n",
        "  - 하나의 언어를 다른 언어로 번역하는 기계 번역\n",
        "\n",
        "- many to many\n",
        "  - Video Classification(Frame Level)\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uiIA4Ue7-G5B"
      },
      "source": [
        "## 두 가지 정보(현재 입력, 이전 시각의 출력)을 처리하는 수식\n",
        "$\\qquad h_t = tanh ( \\ h_{t-1} W_h \\ + \\ x_t W_x + b) $\n",
        "\n",
        "- $W_x$ : 입력 $x$를 출력 $h$로 변환하기 위한 가중치\n",
        "\n",
        "- $W_h$ : 다음 시각의 출력으로 변환하기 위한 가중치\n",
        "\n",
        "- $h$는 '상태'를 기억\n",
        "\n",
        "- $h_t \\ $를 은닉 상태(hidden state) 또는 은닉 상태 벡터(hidden state vector)라고도 불림\n",
        "\n",
        "  <sub>출처: https://colah.github.io/posts/2015-08-Understanding-LSTMs/</sub>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lp3_sVIjHk0F"
      },
      "source": [
        "## 순환 신경망 레이어 (RNN Layer)\n",
        "\n",
        "- 입력: `(timesteps, input_features)`\n",
        "\n",
        "- 출력: `(timesteps, output_features)`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "n08yr0aAIbFD"
      },
      "source": [
        "import numpy as np"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZKIInEEZIcBj"
      },
      "source": [
        "timesteps = 100\n",
        "input_features = 32\n",
        "output_features = 64\n",
        "\n",
        "inputs = np.random.random((timesteps, input_features))\n",
        "\n",
        "state_t = np.zeros((output_features, ))\n",
        "\n",
        "W = np.random.random((output_features, input_features))\n",
        "U = np.random.random((output_features, output_features))\n",
        "b = np.random.random((output_features,))\n",
        "\n",
        "successive_output = []\n",
        "\n",
        "for input_t in inputs:\n",
        "  output_t = np.tanh(np.dot(W, input_t) + np.dot(U, state_t)+b)\n",
        "  successive_output.append(output_t)\n",
        "  state_t = output_t\n",
        "\n",
        "final_output_sequence = np.stack(successive_output, axis=0)"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bEZ8dQEEKxlQ"
      },
      "source": [
        "## 케라스의 순환층\n",
        "- `SimpleRNN` layer\n",
        "\n",
        "- 입력: `(batch_size, timesteps, input_features)`\n",
        "\n",
        "- 출력\n",
        "  - `return_sequences`로 결정할 수 있음\n",
        "  \n",
        "  - 3D 텐서\n",
        "    - 타임스텝의 출력을 모은 전체 시퀀스를 반환\n",
        "\n",
        "    - `(batch_size, timesteps, output_features)`\n",
        "\n",
        "  - 2D 텐서\n",
        "    - 입력 시퀀스에 대한 마지막 출력만 반환\n",
        "\n",
        "    - `(batch_size, output_features)`\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lZ1YCi1iKMC8"
      },
      "source": [
        "from tensorflow.keras.layers import SimpleRNN, Embedding\n",
        "from tensorflow.keras.models import Sequential"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GQ9LbTgoKxIl",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3cdc53bc-f274-4dc6-abaa-efe8706d4f50"
      },
      "source": [
        "model = Sequential()\n",
        "model.add(Embedding(1000, 32))\n",
        "model.add(SimpleRNN(32))\n",
        "model.summary()"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding (Embedding)        (None, None, 32)          32000     \n",
            "_________________________________________________________________\n",
            "simple_rnn (SimpleRNN)       (None, 32)                2080      \n",
            "=================================================================\n",
            "Total params: 34,080\n",
            "Trainable params: 34,080\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KDxE5Jv0Kc7O",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "53ef84fd-133a-4627-97bb-6925af448905"
      },
      "source": [
        "model = Sequential()\n",
        "model.add(Embedding(1000, 32))\n",
        "model.add(SimpleRNN(32 ,return_sequences=True))\n",
        "model.summary()"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential_2\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_2 (Embedding)      (None, None, 32)          32000     \n",
            "_________________________________________________________________\n",
            "simple_rnn_1 (SimpleRNN)     (None, None, 32)          2080      \n",
            "=================================================================\n",
            "Total params: 34,080\n",
            "Trainable params: 34,080\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E-MDIz1zcIyP"
      },
      "source": [
        "- 네트워크의 표현력을 증가시키기 위해 여러 개의 순환층을 차례대로 쌓는 것이 유용할 때가 있음\n",
        "\n",
        "  - 이런 설정에서는 중간층들이 전체 출력 시퀀스를 반환하도록 설정"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "E1tUaIe2L7GZ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3c262b02-719d-48cf-f053-999c6789e50b"
      },
      "source": [
        "model = Sequential()\n",
        "model.add(Embedding(1000, 32))\n",
        "model.add(SimpleRNN(32 ,return_sequences=True))\n",
        "model.add(SimpleRNN(32 ,return_sequences=True))\n",
        "model.add(SimpleRNN(32 ,return_sequences=True))\n",
        "model.add(SimpleRNN(32 ,return_sequences=True))\n",
        "model.add(SimpleRNN(32))\n",
        "model.summary()"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential_3\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_3 (Embedding)      (None, None, 32)          32000     \n",
            "_________________________________________________________________\n",
            "simple_rnn_2 (SimpleRNN)     (None, None, 32)          2080      \n",
            "_________________________________________________________________\n",
            "simple_rnn_3 (SimpleRNN)     (None, None, 32)          2080      \n",
            "_________________________________________________________________\n",
            "simple_rnn_4 (SimpleRNN)     (None, None, 32)          2080      \n",
            "_________________________________________________________________\n",
            "simple_rnn_5 (SimpleRNN)     (None, None, 32)          2080      \n",
            "_________________________________________________________________\n",
            "simple_rnn_6 (SimpleRNN)     (None, 32)                2080      \n",
            "=================================================================\n",
            "Total params: 42,400\n",
            "Trainable params: 42,400\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "j0CNAzN_deXE"
      },
      "source": [
        "## IMDB 데이터 적용"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W1SbNYRdmVTY"
      },
      "source": [
        "### 데이터 로드"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-27Tkihbcei2"
      },
      "source": [
        "from tensorflow.keras.datasets import imdb\n",
        "from tensorflow.keras.preprocessing import sequence"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KL-_piPIdms6",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "487f1d85-fe35-4cdf-8780-e7464a92cba7"
      },
      "source": [
        "num_words = 10000\n",
        "max_len = 500\n",
        "batch_size = 32\n",
        "\n",
        "(input_train, y_train), (input_test, y_test) = imdb.load_data(num_words = num_words)\n",
        "print(len(input_train))\n",
        "print(len(input_test))\n",
        "\n",
        "input_train = sequence.pad_sequences(input_train, maxlen=max_len)\n",
        "input_test = sequence.pad_sequences(input_test, maxlen=max_len)\n",
        "print(input_train.shape)\n",
        "print(input_test.shape)"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "25000\n",
            "25000\n",
            "(25000, 500)\n",
            "(25000, 500)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uf2EVnzceS4R"
      },
      "source": [
        "### 모델 구성"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "d7dfeDC5eMdj"
      },
      "source": [
        "from tensorflow.keras.layers import Dense"
      ],
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KnAUijOReWGu",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1063afa0-6ac1-4241-ba66-3ede88b5b2be"
      },
      "source": [
        "model = Sequential()\n",
        "\n",
        "model.add(Embedding(num_words, 32))\n",
        "model.add(SimpleRNN(32))\n",
        "model.add(Dense(1, activation='sigmoid')) #imdb는 마지막 결고값이 '긍정' or'부정'\n",
        "\n",
        "model.compile(optimizer='rmsprop',\n",
        "              loss='binary_crossentropy',\n",
        "              metrics=['acc'])\n",
        "model.summary()"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential_4\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_4 (Embedding)      (None, None, 32)          320000    \n",
            "_________________________________________________________________\n",
            "simple_rnn_7 (SimpleRNN)     (None, 32)                2080      \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 1)                 33        \n",
            "=================================================================\n",
            "Total params: 322,113\n",
            "Trainable params: 322,113\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pU_9-Hu7mcMa"
      },
      "source": [
        "### 모델 학습"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XdPtzPwNeonA",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "2f597715-bb41-4d26-9080-6189cb4031c3"
      },
      "source": [
        "historyt = model.fit(input_train, y_train,\n",
        "                     epochs=10,\n",
        "                     batch_size=128, \n",
        "                     validation_split=0.2)"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/10\n",
            "157/157 [==============================] - 98s 598ms/step - loss: 0.6005 - acc: 0.6705 - val_loss: 0.4171 - val_acc: 0.8342\n",
            "Epoch 2/10\n",
            "157/157 [==============================] - 94s 596ms/step - loss: 0.3734 - acc: 0.8448 - val_loss: 0.4738 - val_acc: 0.8140\n",
            "Epoch 3/10\n",
            "157/157 [==============================] - 93s 594ms/step - loss: 0.2813 - acc: 0.8899 - val_loss: 0.4536 - val_acc: 0.8044\n",
            "Epoch 4/10\n",
            "157/157 [==============================] - 94s 600ms/step - loss: 0.2127 - acc: 0.9196 - val_loss: 0.3459 - val_acc: 0.8720\n",
            "Epoch 5/10\n",
            "157/157 [==============================] - 95s 606ms/step - loss: 0.1660 - acc: 0.9413 - val_loss: 0.3716 - val_acc: 0.8598\n",
            "Epoch 6/10\n",
            "157/157 [==============================] - 95s 604ms/step - loss: 0.1088 - acc: 0.9638 - val_loss: 0.4446 - val_acc: 0.8362\n",
            "Epoch 7/10\n",
            "157/157 [==============================] - 94s 602ms/step - loss: 0.0726 - acc: 0.9764 - val_loss: 0.4913 - val_acc: 0.8224\n",
            "Epoch 8/10\n",
            "157/157 [==============================] - 94s 601ms/step - loss: 0.0557 - acc: 0.9819 - val_loss: 0.5480 - val_acc: 0.8396\n",
            "Epoch 9/10\n",
            "157/157 [==============================] - 94s 598ms/step - loss: 0.0387 - acc: 0.9884 - val_loss: 0.5685 - val_acc: 0.8326\n",
            "Epoch 10/10\n",
            "157/157 [==============================] - 94s 600ms/step - loss: 0.0269 - acc: 0.9918 - val_loss: 0.6121 - val_acc: 0.8384\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "afIyKdLsmj6N"
      },
      "source": [
        "### 시각화"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sendhEujeu8S"
      },
      "source": [
        " import matplotlib.pyplot as plt\n",
        " plt.style.use('seaborn-white')"
      ],
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3rjMFpiBe4Fa",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 507
        },
        "outputId": "7971d0b0-af3c-4a28-ce7e-5e73e306638e"
      },
      "source": [
        "loss = historyt.history['loss']\n",
        "acc = historyt.history['acc']\n",
        "val_loss = historyt.history['val_loss']\n",
        "val_acc = historyt.history['val_acc']\n",
        "\n",
        "epochs = range(1, len(loss)+1)\n",
        "\n",
        "plt.plot(epochs, loss, 'b--', label='training loss')\n",
        "plt.plot(epochs, val_loss, 'r--' ,label='validation loss')\n",
        "plt.grid()\n",
        "plt.legend()\n",
        "\n",
        "plt.figure()\n",
        "plt.plot(epochs, acc, 'b--', label='training acc')\n",
        "plt.plot(epochs, val_acc, 'r--', label='validation acc')\n",
        "plt.grid()\n",
        "plt.legend()\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAW4AAAD1CAYAAABwdB+7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3deViV1fbA8e9hUBLQRCEjtBQ1FDPnfg455XRLy7SExKGuY05ZWRqpeDVwyNK00lIrzYlSShsUG+zaTUzTHEAyMUNNZXAgUBCH8/tjeUASRZTDe4b1eR4ezvu+53BWO1zss9+91zaZzWYzSiml7IaL0QEopZQqHk3cSillZzRxK6WUndHErZRSdkYTt1JK2Rk3a79BTk4O8fHx+Pr64urqau23U0oph3Dx4kXS0tKoV68eHh4eBa5ZPXHHx8cTFhZm7bdRSimHtGzZMpo0aVLgnNUTt6+vb96bV6lSxdpvZzVJSUnUrFnT6DBsgrZFQdoeBWl75LuVtjh+/DhhYWF5OfRKVk/cluGRKlWqEBAQYO23s5rMzEy7jr8kaVsUpO1RkLZHvpJoi8KGmPXmpFJK2RlN3EopZWc0cSullJ3RxK2UUnZGE7dSStkZTdxKKWVnNHErpVRJunQJvvgCHnqIsvv2WeUtbmged1RUFLt27cJkMhEeHk79+vXzrh07dowXXniB8+fPU7duXSZPnmyVQJVSyqZlZsKHH8KcOXDgAAQE4JaSYpW3KrLHvXXrVpKTk4mOjiYyMpLIyMgC16dNm8a///1vVq1ahaurK0ePHrVKoEopZbMuXIC6deG558DPD6Kj4Y8/ONO6tVXersjEHRcXR4cOHQAIDAwkIyODrKwsAC5dusT27dtp3749ABEREfj7+1slUKWUshlmM/zwA4waJY/d3GDqVNiyBTZvhl69wN3dam9fZOJOT0+nYsWKecc+Pj6kpaUBcPLkSTw9PZk6dSpPPfUUb7zxRokGt3cvzJ9foj9SKaVuXk6ODIc0aADt2sHy5ZCcLNf69IEHHiiVMIpdq+TKvYXNZjMpKSn069ePu+66i8GDB/PDDz/Qtm3bq16XlJREZmZmsd7rzTd9+fDDSgQGHiAg4HxxQy1ROTk5JCYmGhqDrdC2KEjboyBHbY+y+/ZRbcAA3E6eJKdWLU5OmcLfjzyCOTsbrvHfeyttkXKd8fEiE7efnx/p6el5x6mpqXnVqipWrIi/vz/VqlUDoHnz5uzfv7/QxF2zZs1iF1uJiICPPoL162tSwp35YktMTKROnTrGBmEjtC0K0vYoyKHaY/t2OH4cHnkEAgPl+9NP49GuHf4mE0UNDN9KW3h7e1/zWpFDJS1btiQ2NhaAhIQE/Pz88PLyAsDNzY2qVavy559/5l2vXr36TQVZmKpV4cknYeFCuWGrlFJWd+ECrFoFrVpBkyYwdqyMY5cpA0uWQPv2YDIZGmKRPe5GjRoRHBxMaGgoJpOJiIgIYmJi8Pb2pmPHjoSHhzNu3DjMZjO1a9fOu1FZUp5/HlauhA8+kBu2SillNTExknQOHYIaNWDWLHjmGcMT9T/d0Bj3mDFjChwHBQXlPb777rtZsWJFyUZ1hWbNoFMnOHPGam+hlHJmv/0GFSvCHXfAbbdJwp4zB7p2BRvdbtEuVk6uXw/h4UZHoZRyGGYzxMbCv/4FdepIzxrkeONGeOwxm03aYCeJ22SSdv71V6MjUUrZvYULZbFMly6wcydMngwvvGB0VMViF4kbYPFiaNQIduwwOhKllN1JTc1//O234OkJH38sc7AnTJDVjnbEbhL344+DlxfMnm10JEopu2A2w08/ySpGf3+Ij5fzixbBtm2yYKZMGWNjvEl2k7grVIB//1tmmBw7ZnQ0SimblZsrvelmzWRK3zffyFBIpUpy3dPT5maJFJfdJG6QsgAXLsA77xgdiVLKJpw8CRs2QFSULEW3GDgQsrLg3XfhyBGYMQPuvNO4OEtYsZe8GykwUG72xsTAlCl2/0dTKVUc58/nF24aMQLWrYM//si/Hhoqc67LlJGhkHr1wMWu+qY3zK4SN0hv28dHk7ZSDi07W2Z8/PKLJOFffpENCn77Ta7n5kLDhjBoEDRtCo0bw+2357/+ij0DHJHdJW5L1diLF+WPqSZwpexcbi7s2SN1QQYOlH/Yo0bJtD2AKlUkOTdtKjccTSZ4/31jYzaY3SVugH374OGHZfiqc2ejo1FKFduOHfmzO3btkuQN0KYN3Huv9KQfeUSStb+/9tD+wS4Td/Xq8klq1ixN3ErZrEuXICmp4HDHjBnQvLnUAvn4YxniGDVKEnSTJvKPG2RGiLomu0zcZcrA8OEwfrxstlC3rtERKaW4dAnOnZPHe/dCixaQkSHHt90mY9I5OXLctSucPu2wNw+tzW5bbcgQ8PDQBTlK2YRNm6TXPH68HFevDk89JcMhu3bB33/LYph27eS6m5sm7Vtgty1XuTL06yeftq7Y50EpVZoOHICePWVsOjVVkjdID3vePFk1V7++JGpVYuw2cQOMGQMrVkhFRqVUKVuyRMYp16+XQk379slcamV1dv1nsFYt+VJKlZILF2TYw8dHNsbt00dWw/kXtYmXKkl23eMGudcxcSKsXm10JEo5uNhY2d184EA5vvdeGcPWpF3q7D5xly0r28NFRsrcfKVUCUtMlIUTXbrIPNw+fYyOyOnZfeI2mWD0aNlkYdMmo6NRysGsWgX33SczQl5/Xab59ehhdFROz+4TN0DfvlKx0bL7kFLqFuTmygIZkOl7I0fKQpoxY+QjrjKcQyTu226DoUNh7Vr5/VJK3QSzGT7/HIKDoXt3WVBj6RH5+hodnbqCQyRukJWUPXtK8SmlVDHt3AkPPSRbTbm7S31rXSBjs+x6OuCV7rwTPv3U6CiUskPffCNFf3x8pG7y4MG6YMbGOdyf1H37ZC9QpdR1ZGfD7t3yuE0bmYudlATDhmnStgMO939oyBA4eFBW4urvn1L/YDbLxq3jxslimj/+kBuOr75qdGSqGG6oxx0VFUVISAihoaHstvyVvqx9+/b07t2bvn370rdvX1JSUqwS6I0aPVpuiH/2maFhKGV7tmyRin29e8tNx2XLdJaInSqyT7p161aSk5OJjo7mwIEDhIeHEx0dXeA5CxYswNPT02pBFke3blCjhtwIf/JJo6NRykb88ovUwa5SBT74QCq0uboaHZW6SUX2uOPi4ujQoQMAgYGBZGRkkJWVZfXAbparKzz3HMTFwc8/Gx2NUgbKyoLvv5fHjRvDggWwf79sqKtJ264VmbjT09OpeEX5PR8fH9LS0go8JyIigqeeeoqZM2ditoF15888Ix2LPXuMjkQpA1y8KL3qWrXkI+ipU7LEeOBA8PIyOjpVAop9++6fiXnUqFE8+OCDVKhQgeHDhxMbG0uXLl2uel1SUhKZmZk3H2kxrV9vokwZM4mJJfPzcnJySCypH2bntC0KsqX2KLd1K3dMm4bHb79x9v77SZk1i5zjx+H48VKLwZbaw2i30hbXu19YZOL28/Mj/YqdClJTU/G9YhVV9+7d8x63bt2a33//vdDEXbNmTQICAm446JJy9GjJFC9LTEykTp06t/6DHIC2RUE20x5//CEfN6tWhRUrKBcSQnUDNtm1mfawAbfSFt7e3te8VuRQScuWLYmNjQUgISEBPz8/vC5/3MrMzGTAgAHkXt6hedu2bdSyoQLZr74K9erBmTNGR6KUlezalb9/X40aUvfht99kQwPdGd1hFdnjbtSoEcHBwYSGhmIymYiIiCAmJgZvb286duxI69atCQkJoWzZstStW7fQ3rZRHnlEVu4uXizrCpRyCGazlMKcNk12nylfXmaJ+PjIJrzK4d3QGPeYMWMKHAcFBeU97t+/P/379y/ZqEpI8+bQrJl0SIYO1dILygEkJMhNxi1bpPBTZCQ8+6zu3+dkHDqVmUzw/PMyA+rrr42ORqmbdO4cJCfLY19fyMyEd9+Vc+HhmrSdkEMnbpCKgQEBsHCh0ZEoVUyZmfDGGzJ2HRoqQyR+fjLP9dlnpZ6xckoOX83D3R2++KIENhXOzZX5sbpwQVlbairMmSOV+k6fls0Mxo3Lv643HZ2ewydukP1Nb8revdLDqVyZCl9+Ca1aQcOG0KRJ/lfNmjp4rkrWp5/KXfXHH4exY+VGjVJXcJqMs2kT1K8PN1QDKz1ddmaoX19u/gDnAgNljuyFCzBvHoSFyS7XJ07IazZuhOhoKUtoA6tHlR3ZtUsKPy1aJMfPPCMb9K5erUlbFcppErdlCfy8edd5Um6ujCnWrAnvvSfjiJfLXebcf798fN28WcYed+2S6mqWxUjz5sk4ZM2aUnmtY0eYNMnq/13KTpnN8MMPsnN6gwYynpeRIdfKlZNOgVLX4DSJu3ZtmeL67ruQk3ONJ40ZI18tWkiR+blzoXLlq5/n5ia98d69888tXQo7dsD770tZwpMnC+7o8Nhj8o90/HjZ1+/IEe2ZO7MhQ2Ts+tdfZVjk8GF44QWjo1J2winGuC2ef1621Vu+HP7978snf/1VCu/UqiVPePhhSbDFVaaMjH83bAiDBsm5S5fyr999d/6iCcvGmL17S68dZPuoevVkDzbleM6dkz/u3bvLJ7InnoBGjaB/f50doorNqRJ3u3bSUZ49G57pcgzT+Ffho48gJARWrIDq1eWrpFx503LOHPmenS3DLL/8IvMUQWYOdOokj/398298PvaYBKzsV2amDLvNmiWFc3Jy5P6J5f+3UjfBqRK3yQQz/pPNXSvfgNrTZEz7hRdk+KK03HYb/N//yZeFpyf8+CNs3y4J/ZdfZMzTz08S94ED8NJL8pphw7Q0pz0wm2HChPwpfe3bw4cfyr0PpW6RUyVugM57ZkL0RJlqNWOG3Ew0mru7TDVs1Sr/3N9/58/XPX4c4uNlP7bYWPjqK/DwMCZWdX1paXLD2mSSu+EPPSRT+po2NToy5UCc4+bkli2yJQ7Ac8+R+eUPvNYoht8v2UDSvpby5cFS1rFlS/j9d6mW9f338NRTMi1R2Y6dO+X/y113ySckkOl8q1Zp0lYlzrET96FDcgOweXOYOFHOlS/P2SZtmDIlvxqm3ejXT4L+/HMZm1fG+/lnqg4eLDelv/pKbnCXLy/X3JzuA60qJY6ZuLOyZHzx3ntleGH8+ALbvt9xh6yfWbxYZu3ZleeekwSRNy1GGSYtDdq1wyMxEaZOlY7C9On5c/uVshLHTNwrV8Jrr0GPHrBvH0yZctUNvdGj4exZmXZtdx5+WGasJCcXsaJIWcXp0/Ld1xdiYjiwfr3UErn9dmPjUk7DcRL3f/+b36t++mnYtk3mSFerVujT69eXG/1vvw3nz5demCVqzhyZZfLOO0ZH4jzWrIHAQBm/BujShUuensbGpJyO/SfupCTpWbdtKyvQzGYZW2zSpMiXvviiDE1ayo3YnWnT4NFHYcSI/IU8yjqys6Wdu3eHe+7R+fXKUPabuE+fluXpdevChg0yNLJpU7FKXj78sEyXrlLFinFak7u7FLZq21ZW4H35pdEROaa9e+GBB+STzQsvyAwlG9pbVTkf+03cP/8Mb74JffvKFjevvnrTS4f/+EN+hF3y8JCP7w0ayI0xrX9S8rZvl7n0X38tRcjKlDE6IuXk7Gu+0vr1MjQyYgR07iw3Hm+x55ObK52ptm2lDLJdKl9e2qZMGS2yX1JOn5b7JB07SuegWze9+ahshn30uPfuhX/9S77mz8+/m1gCH1fLlIEBAyAmBv7885Z/nHEqV5YEfvaszPfet8/oiOzX5s3yCaZnz/wZJJq0lQ2x7cR94oT0ruvXl3HFmTPlY6u7e4m+zYgR0lGdO7dEf6wxjh+X3nfHjlIqVN24ixdl44zWrWW65YYNmrCVTbLtxH38uOzyO2SIDEK/+CKULVvibxMQICW0Fy6UYm52rUYNqWeSkSHJOy3N6Ijsw/nzUrFv/Hj5Zfj114KFwJSyIbaduIODZTXaO+9YfTXa889LxU1LSRO71rChzDBJTpba4n//bXREts/dXRL1Bx9IwfYKFYyOSKlrsu3EDVLatBQ0ayblkh2mTPKDD0qBo5QUHTK5lnPn5C/25s1yHBkp+z3qDV5l424ocUdFRRESEkJoaCi7d+8u9DlvvPEGffv2LdHgSlulSvL9zBlj4ygxjzwiQ0zBwXJ85Y48zm7fPulhz54tez8qZUeKTNxbt24lOTmZ6OhoIiMjiby86/mVkpKS2LZtm1UCLG19+8rCHIdhmdseHi6LdJw9eZvNsqFBo0bySeSLL6RtlLIjRSbuuLg4OnToAEBgYCAZGRlkZWUVeM60adN4/vnnrRNhKWvYUBZg7thhdCQlzNNT9jwcPdq5F+l8/rlUVmzWTLaQ69rV6IiUKrYiF+Ckp6cTbPmoDfj4+JCWlobX5Wp7MTExNGvWjLvuuuu6PycpKYlMO5iy0aqVC+XK1WTSpCymTz+adz4nJ4fExEQDI7tFjz+O3/79VJo7l7SLF0kfMeKmf5Q9toXp7FnM5cpBrVpUmDqVjK5d5aZtCdy4tcf2sCZtj3y30hYpKSnXvFbslZPmK3prp0+fJiYmhg8//PC6bwJQs2ZNAiyb49q4QYPgnXcqMH9+Bfz95VxiYiJ16tQxNrBb9eGH4OqK77vv4lu7ttT2vgl21RaXLsHrr8Nbb8lenv7+UK8e/iX4FnbVHqVA2yPfrbSFt2UHrEIUOVTi5+dHenp63nFqaiq+l6fmbdmyhZMnTxIWFsaIESNISEggKirqpoK0JSNHylqM+fONjqSEmUyy43hoaKnN1jHUsWNSGmHcONnPs1w5oyNSqkQU2eNu2bIlc+fOJTQ0lISEBPz8/PKGSbp06UKXLl0AOHLkCK+88grhDnCjJzBQ6ja1bWt0JFbg5ibzlC1T3k6eBB8fY2OyhnXr5GZsVpbsljFwoE7zUw6jyMTdqFEjgoODCQ0NxWQyERERQUxMDN7e3nTs2LE0YjREt25GR2BFlgS2YQM88YTcsGvf3tiYStqHH0q93pUrpfSvUg7khsa4x4wZU+A4KCjoqucEBATw8ccfl0xUNmLtWtmX0m6rBhalcWPZIeixx2T3eHvfjXz/fqkxEhgo9Qvc3W+61K9Stsz2V04aKCNDqgZu2GB0JFZSqZL8x1WuLJUX9+41OqKbt2SJzOV89lk5Ll9ek7ZyWJq4ryMkBO68UxbXOSx/f/j2W+mdduoEqalGR1Q8f/8tq6b695dPEIsWGR2RUlanifs6ypSB4cOl2N7+/Q6860lgoPxHDhhg9WJeJSopSVZALl8O//mPDPdUrWp0VEpZnSbuIgwZAl5e8NJLd3HhgtHRWFH9+pL8TCbZy82ygYCtOXIkv7bIXXdB7dpyPHEiuLoaGZlSpca+ti4zQOXKMrNs27Z03NzsYwHRLcnJgXbtpOe6YYPxc5/T0mDjRulNf/+93ID08ZHzt90m+0Aq5WS0x30DWrWCLl1kuf6qVbJPscPy8JANcePiZKpgbm7pvv/p0zKJ3vK+06fLzYYVKyAoSDaI/v57nZOtnJr2uIvh/HmYMEGKyn32mWww45CeeEKWjQ4eLPtXLltmvWGIM2ekqpelV71jhxTB+vFH+Ys5dCj06iVj2W7666oUaI+7WNzdJbcEBkqp608+MToiKxo0SHq70dEwa1bJ/dycHEnSls2Mt2+XOrqzZ0sFw4gI+O9/8+eU16wplfw0aSuVR/81FNOdd0pe6dZNSn6cOJE/ddjhvPyyzDLp1evmf8alS7BlS/4Y9ebNsvPMmDFS/OmBB+Cbb6BFC+PH05WyE5q4b8Ltt8vsuV694M8/jY7Gyp55Rr5nZspd2qKS+MWLsHOn1EDp2FGGPbp2hVOnoEEDmV/Zvr1srQay+fPleu9KqRujifsmlSsnJT4sQ7+HDslu8S6OOvg0cyZMniwJvEWLgtd++01moHz/vXwcOX0a7r1Xzru6wldfQa1aMkVHKXXLHDXNlAo3N5nckJ4uw7D9+skNTIf06quyY/zgwdz+ySdSxMlSm33qVKntvWcP9OwpNzM3bsx/bfPmmrSVKkHa4y4BlSpJ3goPlxGBTz91wOHaMmVkLmTnztw5aZKca9lSFsBMmCCLd+65x8gIlXIa2uMuASYTvPKK7FGwfr2U/Dh1yuiorMDTE9at48js2TIrpFYtOV+zpiZtpUqRJu4SNHiwTBHctk1mtTkkb28yO3WSnrYuglHKEDpUUsJ69pTSGfXrGx2JUspRaY/bCpo3l1GFzEyZCffrr0ZHpJRyJJq4rSg1FXbvlr0rN20yOhqllKPQxG1FgYHw009SfbRzZ9kKTSmlbpUmbiurWlXqJdWvDz16OPD+lUqpUqOJuxRUqgTffQe9e9v/frxKKeNp4i4lXl6yn+0990jdpejo/IWHSilVHJq4DRATI5UFBw3CsbdDU0pZhSZuA/TsKavEFy2SYns5OUZHpJSyJ5q4DWAySaG9t96SnXQefhj+/tvoqJRS9uKGVk5GRUWxa9cuTCYT4eHh1L9iWeAnn3zCqlWrcHFxISgoiIiICEy6FPqGjBolNy6HDoWEBFm4o5RSRSmyx71161aSk5OJjo4mMjKSyMjIvGvZ2dl89dVXLFu2jJUrV/LHH3/wqy4TLJawMDh4MD9pnzljbDxKKdtXZOKOi4ujw+UdSgIDA8nIyCArKwuA2267jcWLF+Pu7k52djZZWVn4+vpaN2IHZClVvXy5bGSekGBsPEop21bkUEl6ejrBwcF5xz4+PqSlpeHl5ZV37v3332fJkiX069ePqlWrFvpzkpKSyMzMLIGQjZGTk0NiYqJV38PTsyznzlWlRQsX3nvvEPffb5t3LUujLeyJtkdB2h75bqUtUlJSrnmt2NUBzYVMPh48eDD9+vVj0KBBNG7cmMaNG1/1nJo1axIQEFDct7MZiYmJ1KlTx6rvUaeOrLDs2BEGDKhOTIwslbc1pdEW9kTboyBtj3y30hbe3t7XvFbkUImfnx/p6el5x6mpqXnDIadPn2bbtm0AeHh40Lp1a3bs2HFTQSpRvbrUN6ldW3aS37/f6IiUUramyMTdsmVLYmNjAUhISMDPzy9vmOTChQuMGzeOM5fvqO3Zs4fq1atbMVzncMcdUtP7/ffzN5lRSimLIodKGjVqRHBwMKGhoZhMJiIiIoiJicHb25uOHTsyfPhw+vXrh5ubG/feey8PPfRQacTt8CpUgKeflsfbtsE338j2aDrTUil1Q2PcY8aMKXAcFBSU97hHjx706NGjZKNSBSxfDrNnw7Fj8t3V1eiIlFJG0q3L7MAbb0iyfuMN2LkTFi+GGjWMjkopZRRd8m4HXFzg9dclYe/eLTNP4uONjkopZRRN3HbCZIJ+/WDPHhg5EurWlfNaXVAp56OJ285UqwZTp0ov/OhRWWm5cqXRUSmlSpMmbjt27hz4+sJTT0l97xMnjI5IKVUaNHHbserVZT/LyEjZnOG++2DdOqOjUkpZmyZuO+fmBuHhsHWrlIhdutToiJRS1qbTAR1Egwbwyy8yfAKQmAhpadC6tbFxKaVKnva4HUjZslC+vDyeOBHatoUxY3RrNKUcjSZuB/XhhzBkiCzaadwYtPaXUo5DE7eD8vKCefPkZuXp0/DAA/Dtt0ZHpZQqCZq4HVyXLrJo54UXoFUrOaeLdpSyb5q4nYCPD0yfDh4espt8/fqyw/ylS0ZHppS6GZq4ncy5c1KgavRo6NABkpONjkgpVVyauJ2Mry988QUsXCh1vu+7T25kFrIjnVLKRmnidkImEwwYIJUGGzaEJUs0cStlT3QBjhOrXh02boSMjPyiVT//DI8/bnRkSqnr0R63k3NxgYoV5fHMmdCjB/TtK1MIlVK2SRO3yjN9OkREwIoVMvb9zTdGR6SUKowmbpXH3R0mTYK4OFnA06mT7LqjlLItmrjVVZo2lSXyEyZAt25y7vx5Y2NSSuXTxK0KddttMHmyLN45f15WXYaH51cfVEoZRxO3KlJurox5T50KzZrBvn1ljQ5JKaemiVsVydNTFuysXQspKfDEE9V15olSBrqhxB0VFUVISAihoaHs3r27wLUtW7bQq1cvQkNDeeWVV7ikBTAcVrduEB8PYWEn2bsXvL3l/NGjxsallLMpMnFv3bqV5ORkoqOjiYyMJDIyssD1iRMnMmfOHFauXMmZM2f48ccfrRasMl7lyjBuXCrbtoGrK5w5I8MorVrJUnr9u62U9RWZuOPi4ujQoQMAgYGBZGRkkJWVlXc9JiaGKlWqAODj48OpU6esFKqyJS6Xf3NcXWUK4ZEj8OijUnlwyRKdhaKUNRWZuNPT06loWVqHJOe0tLS8Yy8vLwBSU1P56aefaNOmjRXCVLbKwwNGjoT9++Hjj6UOSv/+sv+lUso6il2rxFxINaITJ04wdOhQIiIiCiT5KyUlJZGZmVn8CG1ETk4OiYmJRodhE67VFo0bQ3Q0bN9+G7ffnk1iIsye7Yubm5mwsFNUrHjRgGitT383CtL2yHcrbZGSknLNa0Umbj8/P9LT0/OOU1NT8fX1zTvOyspi0KBBjB49mlaWLVYKUbNmTQICAm40ZpuTmJhInTp1jA7DJhTVFnXrynezWQpYRUfDRx/5MnAgvPgiVKtWSoGWEv3dKEjbI9+ttIW35e5/IYocKmnZsiWxsbEAJCQk4Ofnlzc8AjBt2jT69+9P69atbyo45bhMJli5EhIS4Mkn4d13ITBQ6n8rpW5ekT3uRo0aERwcTGhoKCaTiYiICGJiYvD29qZVq1Z8/vnnJCcns2rVKgC6du1KSEiI1QNX9qNuXfjoI1mJOWsWtGwp5+PjpUduOVZK3ZgbGuMeM2ZMgeOgoKC8x/Hx8SUbkXJY1apJ4raYPh2WLpXEPW4cPPxw/mwVpdS16T8TZZj582HOHDh8WBb31K8Plz+4KaWuQxO3Moynp0wlTErKn0poWZh76ZIs7lFKXU0TtzKcuzv06SNJ+9VX5dzatXD33fCf/8CJE8bGp5St0cStbIbJBGUvFx68+24Z+540ScbGn3sOkpMNDU8pm6GJW9mkhg1hzZqCUwk7ddLd6JUCTdzKxkFCqIgAABPBSURBVFmmEh44AIsWSa88O1s2NI6LMzo6pYyhiVvZhWrVpAIhwN69sG4dtGgBISHw55+GhqZUqdPErexO48Yy3j1pkpSSDQqSm5q5uUZHplTp0MSt7JKnJ0REwO+/yxj4f/8rs1OUcgaauJVdCwiQOeDffSfj3ykp0K4d6H4eypFp4lYOwTKN8I8/ZEFP69bQq5eOfyvHpIlbOZTmzWHfPhn//vJLGf8eP16nESrHoolbOZxy5QqOf//5pwyjgCZw5Rg0cSuHZRn/XrxYjnfvhv/7Px3/VvZPE7dyeK6u8j0tDY4e1fFvZf80cSun8dBD8NtvBce/p0wxOiqlik8Tt3IqV87/fuKJ/EU7ZrOUklXKHmjiVk4pIEB235k8WY6/+gqaNtXxb2UfNHErp2aZbQKQmqrj38o+aOJWCuja9er532+8YXRUShVOE7dSl/1z/nfFinL+4kUd/1a2RRO3Uv9gmf/973/L8bvv6vi3si2auJUqwl136fi3si2auJUqQo8eV49/z59vdFTKmWniVuoG/HP8u3ZtOZ+drePfqvTdUOKOiooiJCSE0NBQdu/eXeDauXPnGDt2LD169LBKgErZEsv4d/v2cvzqq9C1aw1eew0OHjQ2NuU8ikzcW7duJTk5mejoaCIjI4mMjCxwfcaMGdSpU8dqASply5o3h8qVLzBhAtSoAQ8+CEuWGB2VcnRFJu64uDg6dOgAQGBgIBkZGWRlZeVdf/755/OuK+VsnnwSliw5xJ9/QmQkpKfD99/LNbMZYmPh3DlDQ1QOqMjEnZ6eTkXLhFbAx8eHtLS0vGMvLy/rRKaUHbn7bggPlx3o331Xzu3aBV26wJ13wtCh8NNPWg9clQy34r7AfJO/eUlJSWRmZt7Ua21BTk4OiYmJRodhE7QtCrpWe7i6wnvvebJ2bQUWL/bmvfdcCAjIZe7cI9x7r+N2w/X3I9+ttEVKSso1rxWZuP38/EhPT887Tk1NxdfXt9hB1KxZk4CAgGK/zppiY2Pp3LnzDT33xRdfZPTo0VStWrXQ688++yzz5s276VgeeOABfv7555t+fWlKTEzU+xpXuF573HcfDB4MmZnw2Wfw6adl6NixBp6esGIFnDgBISFwE/+kbJb+fuS7lbbw9va+5rUih0patmxJbGwsAAkJCfj5+TnE8MiRI0f46quvbvj5AwcOvGbSBm4paSvH5+0N/frBF19IaVmAtWth5Ejw94du3SA6WqYXKlWUInvcjRo1Ijg4mNDQUEwmExEREcTExODt7U3Hjh0ZNWoUx48f5+DBg/Tt25devXrRrVu30oj9lkyePJndu3fz9ttvYzabOXz4MEeOHOGjjz7ilVdeISUlhbNnzzJy5EjatWvHq6++yrRp04iNjSUzM5ODBw9y6NAhwsPDadOmTV6PuW/fvrRo0YItW7Zw6tQp5s+fj6+vLy+99BJHjx6lYcOGrFu3jk2bNhUa1759+5g8eTIuLi54enoybdo0XF1dGT16NLm5ueTm5jJx4kSqVat21bng4OBSbkV1K1askHHxpUth2TJZ3PPoo7BmjVw3mwtWL1TK4obGuMeMGVPgOCgoKO/xnDlzSiSQtm2vPterFwwbBmfPwsMPX3396aflKz1diuJf6Ycfrv9+AwYMYNmyZYwYMYK5c+dy/vx5li9fzokTJ2jVqhWPP/44hw8f5rnnnqNdu3YFXnv8+HEWLFjApk2bWLlyJW3atClw3cvLi8WLFzNz5kw2bNhAtWrVOHfuHJ988gkbN25ksWUTxEJERkby8ssvc//997No0SKWLFlCUFAQd9xxB1FRURw+fJiDBw/y119/XXVO2Z/77oPp0yEqCv77XyhbVs4fPQotW8owSt++oH+T1ZV05eRl9evXB6B8+fLs2bOH0NBQxo4dy+nTp696bqNGjQCoUqVKoTdcmzRpknc9KyuLAwcO5L2mTZs2uLld++/lgQMHuP/++wEZ9967dy8NGjRg586dTJw4keTkZFq3bl3oOWW/XF1lUU/LlnKckQF168LMmVCvHjRsCG++CadOGRunsg3FnlViLdfrIZcrd/3rlSsX3cMuiru7OwBffvklGRkZLF++nNOnT/PEP7vycN3EC+Bq2Z0WmYVjNpvzzpmK8dn3/PnzuLi44Ofnx5o1a/j5559ZsWIFO3fuZMSIEYWeU46hTh3ZlSc1FVaulOGUl16SHnjFirJK09cXHOB2k7oJTtvjdnFx4cKFC1edP3XqFAEBAbi4uPDNN9+Qa9mU8BZUq1aN+Ph4AP73v/9x8eLFaz63Vq1a/PrrrwBs27aNevXqsXnzZjZv3kyrVq2YMGEC8fHxhZ5TjsfPD0aNgq1bpSrhXXfJ+WefhSpVZBglNhYK+VVWDsxmetylLTAwkL179xIVFVVg2k2nTp149tln2blzJz179qRKlSq8/fbbt/Re7dq1Y/Xq1Tz11FM0a9aM22+//ZrPHT9+PP/5z38wmUxUqFCBqVOncvr0aV566SUWLlyIyWRi1KhRVKlS5apzyrFdOalpwgRZWv/JJ9Ibr1IFXn4Znn9erh88CNWqyRCMckBmKzt8+LC5du3a5sOHD1v7raxq7969N/3aU6dOmdevX282m83m48ePmzt37lxSYRniVtrCERnZHjk5ZnNMjNn8xBNm85tvyrmMDLMZzGYPD7O5QQOzOSzMbI6KMpt37SqdmPT3I9+ttMX1cqfT9rhLk6enJ+vWrWPRokVcunSJV155xeiQlIMoWxYef1y+LFxc4IMPICFBvjZtkumGFSpA/fqwfz+EhspMlXr15HtwsPTQXZx28NS+aOIuBe7u7syePdvoMJST8PKCZ54peC4jI39OeHY2VKoE330nJWot1qyReeQJCbBuXX5iDwjQ+eS2RhO3Uk6gQoX8x/Xrw4YN8vjUKSmMlZAADzwg5/73P5nBYlG+vExNjI6WXvmxY7J5hL+/JnSjaOJWyolVrChzxy3zxwGGDJEFbZahFstX5cpyffZsmDEDbr9dErpluGXYMChipqwqIdrMSqmrVKokmyMXtq6rd2+Z4WJJ6KtWydfIkXL9tdfu4MiRq8fQLYlf3TpN3EqpYrn/fvmyMJvh9On8YZPKlS/w559yQ/Tvv+XcffeBZdfDuXPBwyM/oV85jKNujN5DvgHt27cnOzub999/P29xjMWZM2dob9mA8Bos1RVjYmL45ptvbjqOuXPnsnTp0pt+vVLWYDLJkIvF0KEn+OknSeaHD8uNzqio/Otvvimlblu2lOGWqlVh4sT86zt3whWbbKlCaI+7GAYPHlzs11jKx3bu3Fk3VFZOxWSSGSn/LMN/4AAcOgTx8TLUEh+fP4ySnQ2NGkkv/p578odaHntM9vdUwmkT9+OPP84777yDv78/f/31FyNHjmTJkiW8+OKLnD17lpycHCZMmJBXfApg3LhxdO7cmaZNmzJy5EjOnTtH48aN866vXbuWpUuX4uLiQq1atZgyZcpV5WMrVqxInz59mDFjBjt27ODixYuEhYXRvXv3QkvC+vv7Fxp/Ya///PPPWbp0Ke7u7gQFBREREVHoOaWM5OIiSfmee6Br14LXTCZYvTo/oSckyJL+KlUkcf/5J3TqlJ/QLd9r14YyZQz4jzGI7STuUq7r2qFDBzZu3EhYWBjfffcdnTp1Ii0tjSeffJIOHToQFxfHggULmDt37lWvXbNmDbVq1SI8PJyvv/46b0OG7OxsFi5cSPny5QkLC2Pfvn1XlY8FqUGyf/9+Vq5cydmzZ3n00UfzNlz+Z0nYp59++qr3v9brFy1axPvvv8+dd97J6tWrycnJKfSch4fHddtGKaN4eFy9oOj8efkCyM2V8fKEBJl3fumSnP/kE9m4OTERPv1UZrvcc4/sBVq5suNNW7SdxF3KOnXqxLRp0/IS96RJk6hcuTLvvvsuixYtIjc3l3LlyhX62gMHDtC0aVMAmjVrlne+QoUKDBs2LO85hZWEBYiPj897fbly5ahZsybJyclAwZKwxX19165dGT58OI8++ihdu3bFw8Oj0HNK2RN3d/kC6VmvXi2Pc3Jg3z5J4g8+KOe2b4dJkwpuyuzhIUW67rtP5qh//73MR7d8BQTIc+yJ7STuUq7rWqtWLVJTUzl27BiZmZlUr16dt99+mzvuuIPXX3+dPXv2MGPGjEJfazabcbm8NvjS5T/5ubm5TJ48mTVr1uDr68uQIUOu+d7/LO1qKd8KV5eELc7rhwwZQrdu3YiNjaV///4sXbq00HMVr7yTpJSd8vC4eoZLnz7Qo4ck9EOH8r8s4+ybN0Nho4WpqVIm99NPYcuWgom9WjXb67XbTuI2QNu2bZk1a1berJBTp05x7733AvDtt99y3vL57B+qV69OfHw8nTt3ztvg98yZM7i6uuLr68uxY8eIj4/n/PnzlC1b9qrysfXq1WPevHkMHjyYM2fOcOjQIe6+++4bjvtar581axYjRozgmWeeISkpiaNHj/LRRx9ddU4Tt3Jk5crJxhMNG1597eWX4bnn4K+/CiZ2y83RX3+FefMK7v1Zpowcm0wwa5b08K9M6vfcAzVqlMp/Wh6nTtwdO3YkNDSUtWvXAvDYY48xduxY1q9fT1hYGF9++SWrLZ/LrtC9e3eGDx9O//79825OVqxYkZYtW9KzZ0+CgoIYOHAgU6dO5eOPP76qfGyTJk2oV68eYWFhXLhwgRdffPGawzKFudbrPT09CQkJwdvbm6pVq1KnTh1++umnq84p5czKlpVEW1iyjYqCyEg4eVISenKylAWwFN86eBC+/lqW/VvUqCEzZUD+KBw9mp/Ug4PdscY/OZP5Wp/HS8iRI0d46KGH+O677wj457wgO5KYmKhJ7zJti4K0PQpyhvY4dw6OHJHkfu4cdOki5wcNknH05GTppS9YcIiBA6vd1HtcL3c6dY9bKaVuRtmyEBgoX1dasEC+m81w4gQcOXLWKu+vKyeVUqqEmUwybl62rHUGNDRxK6WUndHErZRSdkYTt1JK2ZkbStxRUVGEhIQQGhrKbkttxss2b97ME088QUhICO+8845VglRKKZWvyMS9detWkpOTiY6OJjIyksjIyALXX3vtNebOncuKFSv46aefSEpKslqwSimlbiBxx8XF5RVACgwMJCMjg6zLxXIPHz5MhQoVuPPOO3FxcaFNmzbExcVZN2KllHJyRc7jTk9PJzg4OO/Yx8eHtLQ0vLy8SEtLw8fHp8C1w4cPF3j9xYsXATh+/HhJxWyIlJSUvJWPzk7boiBtj4K0PfLdSltYcqYlh16p2AtwirvQMi0tDYCwsLDivpVSSjm9tLS0q2oZFZm4/fz8SE9PzztOTU3F19e30GspKSn4+fkVeH29evVYtmwZvr6+BSrfKaWUuraLFy+SlpZGvXr1rrpWZOJu2bIlc+fOJTQ0lISEBPz8/PDy8gIgICCArKwsjhw5QpUqVdi4cSMzZ84s8HoPD4+8GtNKKaVu3LWqht5QkamZM2fyyy+/YDKZiIiIYO/evXh7e9OxY0e2bduWl6w7derEgAEDSjZypZRSBVi9OqC9mzFjBtu3b+fChQsMGTKETp06GR2S4XJycujatSvDhg1z+g2Q165dy8KFC3Fzc2PUqFG0LWwLPidw5swZxo4dS0ZGBufPn2f48OE8aNmWxon8/vvvDBs2jKeffpo+ffpw7NgxXn75ZS5evIivry+vv/46ZUpgc0xdOXkdW7ZsYf/+/URHR7Nw4UKioqKMDskmzJs3jwoVKhgdhuFOnTrFO++8w/Lly5k/fz7fffed0SEZ5rPPPqN69ep8/PHHvPXWW1et93AGZ8+eZcqUKTS/Yjv6OXPm0Lt3b5YvX87dd9/NqlWrSuS9NHFfR9OmTXnrrbcAKF++PNnZ2YVOzXEmBw4cICkpyWl7lleKi4ujefPmeHl54efnx5QpU4wOyTAVK1bM2yP177//dspdlsqUKcOCBQsKTND4+eefeeihhwBo165dia1z0cR9Ha6urnk706xatYrWrVs7/cyY6dOnM27cOKPDsAlHjhwhJyeHoUOH0rt3b6defPbII49w9OhROnbsSJ8+fRg7dqzRIZU6Nze3qzbjzs7OzhsaqVSpUt706Ft+rxL5KQ7u22+/ZdWqVXzwwQdGh2Kozz//nAYNGlC1alWjQ7EZp0+f5u233+bo0aP069ePjRs3XrWZszNYs2YN/v7+LFq0iN9++43w8HBiYmKMDsumlOTtRE3cRfjxxx+ZP38+CxcudPrVYD/88AOHDx/mhx9+4Pjx45QpU4YqVarQokULo0MzRKVKlWjYsCFubm5Uq1YNT09PTp48SaVKlYwOrdTt2LGDVq1aARAUFERqaioXL150+k+o5cqVIycnBw8Pj0LXudwsHSq5jszMTGbMmMF7773H7bffbnQ4hps9ezarV6/mk08+4cknn2TYsGFOm7QBWrVqxZYtW7h06RKnTp3i7NmzTjm2CzLfeNeuXQD89ddfeHp6On3SBmjRogWxsbEAbNiwocRm2miP+zq+/vprTp06xejRo/POTZ8+HX9/fwOjUrbijjvuoHPnzvTq1QuA8ePH4+LinH2hkJAQwsPD6dOnDxcuXGDSpElGh1Tq4uPjmT59On/99Rdubm7ExsYyc+ZMxo0bR3R0NP7+/nTv3r1E3kvncSullJ1xzu6BUkrZMU3cSillZzRxK6WUndHErZRSdkYTt1JK2RlN3EopZWc0cSullJ3RxK2UUnbm/wE57RSBzs7RIQAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAD1CAYAAABA+A6aAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3deViU5frA8e+A4AK4oICJG0KFYRpmlmK5hFrWKW1RzK3USlPb1KNS/axMUtvLUlOzIklawO2UeCw9WaJ2jqZimEEuiMqiQiCigPP743aAERVF4J0Z7s91cQ3zvgxzzwPcPHO/z2Iym81mlFJK2T0nowNQSilVOTShK6WUg9CErpRSDkITulJKOQhN6Eop5SBqGfGk+fn5JCQk4OXlhbOzsxEhKKWU3SkqKiIjI4N27dpRp06dMucNSegJCQkMGTLEiKdWSim7t3TpUjp16lTm+GUl9L179/LUU0/x6KOPMnToUKtzmzZt4u2338bZ2Zk77riDcePGARAREcGOHTswmUyEh4fTvn374sd4eXkVB9W0adMKvyijJSUlERAQYHQYNkPbw5q2RwltC2sVbY+jR48yZMiQ4hx6vnITel5eHjNmzKBLly4XPP/aa6+xePFifHx8GDp0KH379uX48eMcOHCA6OhokpOTCQ8PJzo6uvgxljJL06ZNad68+RW/KFuRk5Nj1/FXNm0Pa9oeJbQtrF1te1ysVF3uRVFXV1cWLlyIt7d3mXMpKSk0aNCAa665BicnJ7p37058fDzx8fGEhoYC4O/vT3Z2Nrm5uRUOXimlVPnKTei1atW6YPEdICMjA09Pz+L7np6eZGRkkJmZSaNGjcocV0opVXWq5aLoxZaLSUpKIicnpzpCqBL5+fkkJiYaHYbN0Pawpu1RQtvCWkXbIy0t7ZLnryqhe3t7k5mZafVk3t7euLi4WB1PT0+/YBE/ICDArutqiYmJtG3b1ugwbIa2hzVtjxLaFtYq2h4eHh6XPH9VE4uaN29Obm4uhw4dorCwkPXr1xMSEkJISAhxcXEA7N69G29vb9zd3a/mqZRSSpWj3B56QkICs2fPJjU1lVq1ahEXF0evXr1o3rw5vXv35uWXX2bixIkA9OvXDz8/P/z8/AgKCiIsLAyTycT06dOr/IUopVRNV25Cb9euHZGRkRc9f8stt1gNSbSYNGnS1UVmkLi4OPr27XtZX7to0SKeffZZWrRoccHzY8eOZd68eZUZnlLKRhw+DEePQnY2ZGXJrbMzDBsm52fOhP/+1/q8vz+sXVt1MRkyU9RWHTp0iH/961+XndBHjx590WQOaDJXysZYxmeYTJKQDx4sSbhZWZCbC+cKDnz0EfzrX9YJ2cUF/vpLzo8fD7Gx1t+/RYuShL53LyQlQcOG0KwZ3HADXH991b4+TeilvPrqq+zcuZO5c+diNptJSUnh0KFDfPrpp0ybNo20tDTy8vKYMGECPXv25IUXXmDWrFnExcWRk5PDvn37OHjwIOHh4XTv3p1bb72VLVu2MGzYMLp27crmzZs5ceIE8+fPx8vLi8mTJ3P48GGCg4P5/vvv+emnn6zi+eSTT4iLi+Ps2bN0796d8ePH8/fffzNp0iRyc3Px8PDg7bffpqioqMwxNzc3g1pRKeP8/Te4uUlPedMmWLVKkvaBA3KblgaZmeDhAW+/DW+9VfZ7PP20JO6MDPn6Bg0kETdoAE2alHzdxIkwYoQcb9iw5Nbis8+q/vWez6YTeo8eZY8NHAhPPQV5edCvX9nzjz4qH5mZ8NBD1uc2bLj0840aNYqlS5cyfvx4PvjgAwoKCoiKiuLYsWN069aNAQMGkJKSwjPPPEPPnj2tHnv06FEWLlzITz/9xLJly+jevbvVeXd3dz777DPefPNN1q5dS8uWLTl9+jRfffUV69ev57OL/PSjoqJwcnLizjvv5NFHH2Xx4sV069aN4cOH8+mnnxIfH8+uXbvKHLNM7FLKURQVwZEj0Lgx1K0r5YwlS6wTdnY2/PEHXHcd/PqrJOwWLaBlS+jZE665RnrnAI89BnfeWTYh1zqXFadPl4+LCQmp+td8pWw6oRvNsv5M/fr12bVrF9HR0Tg5OZGVlVXmazt27AjIcgYXGltvWUinadOmZGVlkZycXPyY7t27U6tW2R9FnTp1GDp0KLVq1eLEiRNkZWXx+++/88wzzwDw6KOPAhAdHV3mmFL2JjdXkrKPjyTt3bth1iw5dvAgHDoEhYWwbp0k4oMH4csvoVUr8POD7t3l8wYN5PuNHQsTJoDTRcbyBQXJhyOx6YR+qR51vXqXPt+kSfk98vK4uLgAsHr1arKzs4mKiiIrK4uHzu/6wwUTcmml114wm82YzebiYyZLl6GU1NRUPv30U2JjY3Fzc+Pee+8t/j5nz54t873PP6aULTl7VnrXBw5A06bQurUk5AkTShL28ePytUuWyLvs/HzYuFF61yEhctuqlfS+AQYMgAceuPhzurpW9auyPTad0Kubk5MThYWFZY6fOHGC5s2b4+TkxL///W/OnDlz1c/VsmXL4rH6P//8M0VFRWWe09PTEzc3N3bv3k1qaioFBQW0a9eOzZs30759e5YtW0bt2rUveGzAgAFXHaNS5TGb5YLh4cOSsA8fhjZtoFs3qWfffTekpkJqaiCWP61XX4WXXoLateUCY8uW0KWL3FqSN8DNN8P+/Rd/7gv0g2o8Teil+Pv78/vvvxMREWE1I6tPnz6MHTuW3377jQcffJCmTZsyd+7cq3qunj178u233zJ48GA6d+5Mw9JXU4C2bdvi5uZGWFgYN998M2FhYbzyyit88MEH/POf/2TYsGG4ubnx5ptvYjabyxxTqjLs3SulDkuyPnwYAgPhySclmTdqJHXr0p54QhK6u7vUum+/HWrXPkbHjk1o2RIsK2n7+MCuXdX/mhyZyXyxhVaq0KFDh7jzzjv54YcfauzU/6ysLLZs2ULfvn1JS0tjxIgRrFmzppIjrF46vduarbbH6dPSOwZYvVouIlqS9ZEjEBAAixbJeT8/616ymxsMHgwLF8r9GTMkcV9zjQzNs3zUq2f9nLbaFkapaHuUlzu1h24QNzc3vv/+exYvXszZs2eZNm2a0SEpB/TeezLaw5KwDx+WskZCgpyfM0fq1HXrliTjxo1LHj9/viT/Zs0kaZ+/lMhLL1Xfa1Hl04RuEBcXF959912jw1AOoLBQShe//CJjr/fsgf/9T2rMe/bIsWbNpNRx113SA7dYtkx63fXrX7gmfZlz7JSN0ISulJ3JzpYkXKuW9KAnT5YhfyCJOyQEcnIkSZc3WblZs6qPV1UfTehK2TCzWUaCbNpU0gNPSICff4auXeUC5YgRksS7dpVyio7+qLk0oStlQ06fhm3bwMtLSiO//CKjREB63F26yAxoS8+6R48Lz6hWNZMmdKUMVFgoC0BZet///a8k9alT4fXXZSz2/PnS+77hBlmjRKmLuaoNLmqyXr16cerUKT7++GO2b99ude7kyZP06tXrko+3TCqKiYnh3//+d5XFqWzH2bMynX3hQvj0Uznm5CSzIt97T85PmAAxMfDcc3K+bl0Z833jjZrMVfm0h36VnnjiiSt+TOlleh+41Nxl5RAWLIAVKyA+XmZVgpRRHn1UEvovv8jsyovsxa7UZdOEXsqAAQP48MMPadasGampqUyYMIHPP/+ciRMnkpeXR35+Pi+99FLxol0AU6dOpW/fvtxyyy1MmDCB06dPc/PNNxefX7lyJV988QVOTk5ce+21zJgxo8wyvY0aNWLo0KHMmTOHbdu2UVRUxJAhQ+jfv/8Fl95tVmpowtGjR5k8eTIAhYWFzJ49m5YtW7J8+XIiIyNxcnLiscceo1+/fhc8pipXTg4sXSrLtr7xhhz75RdZw+Thh0suXpYeOnjDDcbEqhyPbZdcLFd8Sn989JGcy8u78HnLe9nMzLLnyhEaGsr69esB+OGHH+jTpw8ZGRk8/PDDREZG8vzzz7PQMkXuPCtWrODaa68lKirKagbYqVOnWLRoEcuWLeOvv/7ijz/+YNSoUXTu3Jnx48cXf92vv/7Kn3/+ybJly/jss8+YO3cuuefGolmW3r3jjjtYe952J+np6YwbN47IyEgefPBBoqKiyM3N5aOPPmLp0qUsXryYVatWXfCYqjw7d8rqfs2ayW1qKpw8KX9eS5ZIqeXjj2VEyrXX6kgUVTVsO6FXsz59+vDjjz8CktD79u1LkyZNiIuLY/Dgwbz55psXXDoXIDk5meDgYAA6d+5cfLxBgwY89dRTDB06lOTk5Is+PiEhgVtuuQWAevXqERAQwIEDBwDrpXctSd7Cy8uLyMhIhgwZwmeffUZWVhZ//fUXbdq0oU6dOtSvX5958+Zd8JiqHHFx0KGD9CUefFBKK9u3g7u7rICptW9VXWy75FLN6+dee+21pKenc+TIEXJycvDz82Pu3Ln4+PjwxhtvsGvXLubMmXPBx5rNZpzOLbxsWcr2zJkzvPrqq6xYsQIvLy+efPLJiz73+UvoFhQUFH+/85feLe3999+nW7duDB48mDVr1rBhwwacnJzKLKd7oWOqYv78U+rirVrJRcyePeH992HIEPD0NDo6VZNpD/08PXr04J133ikepXLixAlatmwJwLp16ygoKLjg4/z8/Eg4t0DGli1bABnt4uzsjJeXF0eOHCEhIaE4UZ+/TG+7du2sHnfw4EFatWpVbryW+MxmMz/88AMFBQW0adOGffv2cfLkSU6fPs1jjz12wWMGrMtmtwoKZPRJ796yHvd778l+kSDrbk+YoMlcGc+2e+gG6N27N2FhYaxcuRKA+++/nylTprBmzRqGDBnC6tWr+fbbb8s8rn///owbN44RI0YUXxRt1KgRISEhPPjggwQGBjJ69Ghef/11IiMjyyzT26lTJ9q1a8eQIUMoLCxk4sSJ1Dt/yboLGDRoEDNmzMDX15dhw4bx0ksvsW3bNp5++mkee+wxQHYxqlevXpljF9pYQ13YqFEQGSnbmc2YIfevucboqJSydlnL50ZERLBjxw5MJhPh4eFWozzWrVvHvHnzcHV15Z577mHo0KFs2bKFZ555hmuvvRaA6667jpdKLcumy+c6Jkdpj7NnYe1amdDzzjuyhOyWLbJhcL9+JXtOlsdR2qMyaFtYM2z53K1bt3LgwAGio6NJTk4mPDyc6OhoQGrFM2bMIDY2loYNG/L4448Xb07cuXNn3n///SsOWCmjZGTIiJQFC2T9FC8vWSvczw9uvdXo6JQqX7k19NI7yPv7+5OdnV080uLEiRPUr18fT09PnJycuO2229i0aVPVRqxUFTh5Uib3TJkiZZUvv4SUFFluVil7UW4PPTMzk6BSW2N7enqSkZGBu7s7np6enDx5kv379+Pr68uWLVvo3Lkzvr6+JCUlMWbMGLKzsxk/fjwhlo0CS0lKSiInJ6dyX1E1ys/PJzEx0egwbIY9tUdurhMrVzZgz57avPrqUQCmTWvAjTeeIiBA9oz966+rew57ao+qpm1hraLtkZaWdsnzV3xRtHTJ3WQyMWvWLMLDw/Hw8Ciu6bRu3Zrx48dz9913k5KSwvDhw1m7di2u523DHRAQoDV0B2IP7bF9u6wRHhUlvfKbb4YWLRrh7g6VHbo9tEd10bawVtH28Dh/y6jzlFty8fb2JjMzs/h+eno6Xl5exfc7d+5MVFQUCxYswMPDA19fX3x8fOjXrx8mk4mWLVvSpEmTcv+zKFXVvvwSOnaEL76AQYNg61ZZ3dDd3ejIlKoc5Sb0kJCQ4pUBd+/ejbe3N+6l/gJGjx7NsWPHyMvLY/369XTp0oWVK1eyePFiADIyMjh27Bg+Pj5V9BKUurA//pBVC6Oi5H6/fjJ+PDUVFi+GcxNzlXIY5ZZcOnbsSFBQEGFhYZhMJqZPn05MTAweHh707t2bgQMHMnLkSEwmE0888QSenp706tWLSZMmFU90efnll8uUW5SqCgUFsHy5lFXWrwcXF7nQCdCgATz9tLHxKVWVLquGPmnSJKv7gYGBxZ/36dOHPn36WJ13d3dn/vz5lRCeUldmwADZMKJVK4iIgJEjQd8cqppCp/4ru3X6NHz1Fdx9Nxw/LseefRZWr4bkZJg2TZO5qll06r+yO3v3luz6k5kpvfGkJOjcGc5NmVCqRtKEruxKSgpcf71Mv7/vPnjiCUniukStUprQlY2z7MF58qTctmgBn38uqx42bWp0dErZFq2hK5uTlweffSbbtbVrJ5tU5eeDZU7bsGGazJW6EE3oymZYEvabb8oGypmZ8nlqqixdq6v9KnVpmtCVoXJzZZLPbbfBuSXoGT0a/vMf2LMHJk6UVQ+VUuXTGroyxLZtsmlyVBTk5MjO95YeeLNm8qGUujKa0FW1KSyU0Slnz8LDD8Phw7KmyhNPQJcuWlJR6mppQldVymyWRbA+/hh++EHWV6ldWyYEtWkDjRoZHaFSjkMTuqoSWVmyquHChbBzJ7i5QViYlFdq15Zla5VSlUsviqpKYzbDqVPy+e7dMGGCLI41f76UVxYtgiZNjI1RKUemPXR11QoK4JNP4I032tC3L3z4IXTtCjt2QKn9xJVSVUx76KrCzGb4+msZoTJmDNStexbLToMmkyZzpaqb9tBVhU2fDjNmQFAQrFoFbdrs54YbdJsxpYyiCV1dkd9+gzp1IDBQ1hpv00am4js7g+4BrJSxtOSiLsu+fTB0KAQHw//9nxxr3Vqm6OtKh0rZBk3o6pIyM2XTiOuvh2+/halTZUy5Usr2aMlFXdLcufDBB1Jeefll8PU1OiKl1MVoQldWCgpksSw/P+jbF55/Xqbnt9VrnUrZPC25KKBkCGJQEIwdK1PzAerX12SulL3QhK745Re49VYYOBBcXWUI4qJFRkellLpSl1VyiYiIYMeOHZhMJsLDw2lfasbIunXrmDdvHq6urtxzzz0MHTq03Mco27JnDxw9CkuWlAxBVErZn3IT+tatWzlw4ADR0dEkJycTHh5OdHQ0AGfPnmXGjBnExsbSsGFDHn/8cUJDQzl48OBFH6OMt38/vPSS9MrHj4cRI2DIEBlfrpSyX+WWXOLj4wkNDQXA39+f7OxscnNzAThx4gT169fH09MTJycnbrvtNjZt2nTJxyjjZGbCc8/JEMRvvpGNl0HWKNdkrpT9K7eHnpmZSVBQUPF9T09PMjIycHd3x9PTk5MnT7J//358fX3ZsmULnTt3vuRjSktKSiInJ6cSX071ys/PJ9FOpkeuWNGA117z4dQpJx54IItx4zLx8Sms1Nmd9tQe1UHbo4S2hbWKtkdaWtolz1/xsEWzZSdfwGQyMWvWLMLDw/Hw8KB58+blPqa0gICAiz7GHiQmJtLWhoeAFBTAmTOyFvnhw9C7N0REQNu2jYDK31nC1tujuml7lNC2sFbR9vDw8Ljk+XJLLt7e3mRmZhbfT09Px6vUrr2dO3cmKiqKBQsW4OHhga+vb7mPUVXLbJaSSlBQyTT9O++E2FgdgqiUIys3oYeEhBAXFwfA7t278fb2tiqdjB49mmPHjpGXl8f69evp0qVLuY9RVWf9erjtNtmz09VVErlSqmYot+TSsWNHgoKCCAsLw2QyMX36dGJiYvDw8KB3794MHDiQkSNHYjKZeOKJJ/D09MTT07PMY1TVi4iAF16A5s11CKJSNdFl1dAnTZpkdT8wMLD48z59+tCnT59yH6OqRmoqFBZCq1bwwAPSKx83DurWNToypVR107Vc7Ngvv0gS79kTli2TNcpL/a9VStUwOvXfTn3yiSTy+vVl5yCllNKEbmcKC2Vy0KhR0L07bN2qI1eUUkITup3Jzobly+GZZ+D776FR5Q8nV0rZKa2h24nkZGjZEho3hm3bNJErpcrSHrod+P576NgRXnxR7msyV0pdiCZ0G2Y2w1tvwb33Qps2MhxRKaUuRhO6jcrPh0cfhUmTZGjizz9LyUUppS5GE7qN2rcPYmLglVcgOloW2FJKqUvRi6I25sABmfXZti38+Sc0bWp0REope6E9dBtime25ZInct6tkfuAAdXbuNDoKpWo0Teg24OxZWVRr8GDo1AnuucfoiK5QSgp07kyz8PCSYx9/DF9/DYcOGReXUjWMllwMlpMj+3muWgWjR8OHH8oCW3YjLw/694eTJ8mYOpXmIMNzXnlFdtUAWf7xtttg0CB46CEjo1XKoWlCN9jPP8s48/fflw2bTSajI7oCZrP8F9q+HVauJMffX46bTHJVd8cOiI8v+QgIkIR+6hSEhkLnztCli3y0aGHsa1HKAWhCN0haGvj4wN13Q1KSXAi1OzEx8OWXshD7vfditUGpqyvccot8PP20HCsslNv0dHBygvnz4d135ZivL3z0Edx3H5w+Lf8sdOdqpa6I1tAN8NFH4Ocny9+CnSZzgAEDZEzl1KmX9/W1zvUfWrWCjRtlYZqtW+XtyR13SFIH+O47aNBAyjTPPQdffSV1eqXUJWlCr0YFBTB2rMz47NULbrzR6IgqKClJxlc6OcHAgRWvE1l68RMmQFQU3HyzHL/2Wnj2WTk/f77U3lu2lHGcALt2waZNMvtKKVVMSy7VJDNTysf/+Q9MmQIzZ9rp9nDZ2VJecXaWxOpUBX2Cdu1g9mz5vKBAavFbt0oNHuCdd2Rsp4sLBAdLDb5rV9lI1a4uQihVuTShV5MvvoDNm+V2yBCjo6mgoiIJPjkZ1q2rmmR+PhcXGcvZqVPJsddfl1q75WLrxx/LleWBA+X8119D+/Zw/fVVH59SNkQTehXLyoKGDeW64N1323mOeekl+Ne/5CJA9+7GxeHjI0Ml+/eX+wUFJUMkz5yRutaxY3DXXdLwfftWzz8fpQymv+VVxGyG116D664rKTfbdTJftUp6xk88AWPGGB2NNReXkivLrq6weze8+qqUavr1k+m3P/xgbIxKVYPLSugREREMGjSIsLAwdp43vXvp0qUMGjSIwYMHM3PmTABiYmLo3r07w4YNY9iwYcybN6/yI7dheXkQFiYd2j59wNvb6IgqQc+e8H//Bx98YPt1ah8fafz9++Via+PG4OUl55KSYO9eQ8NTqqqUW3LZunUrBw4cIDo6muTkZMLDw4mOjgYgNzeXxYsXs3btWmrVqsXIkSP57bffAOjXrx9Tpkyp2uhtUEqKVAK2b5frepMn237+u6Rjx2Q8uLu7zP60J66usp7C4MElx155RS5k3H23lGP69NFyjHIY5f4mx8fHExoaCoC/vz/Z2dnk5uYC4OLigouLC3l5eRQWFnLq1CkaNGhQtRHbuNdek9F1q1bBP/9p58n8zBkZa967tyw44wjeeEOS+vbtktTbtoVPPzU6KqUqRbkJPTMzk0al9jzz9PQkIyMDgNq1azNu3DhCQ0Pp2bMnHTp0wM/PD5Ce/ahRoxgxYgS///57FYVvO06elNu335YRdna3wNaFPPusTACaMMFxerFNm0rp6MABWLpU9vOz/H6ePSsjeJSyU1c8ysVsNhd/npuby4IFC1izZg3u7u6MGDGCPXv20KFDBzw9PenRowfbt29nypQprFq1qsz3SkpKIicn5+pegYHy8/PZtSuRt97yZvNmN774Yj9ubtI+pWfB26OG0dFcM28emaNHk3HTTZf1gvLz80m0pxceHCzj2QsKIDERt40baTFmDLl33MGJoUM52bXrVb3Fsrv2qELaFtYq2h5paWmXPF9uQvf29iYzM7P4fnp6Ol7nLjAlJyfTokULPD09AejUqRMJCQk89NBD+J9bqCk4OJjjx49TVFSE83kzaQICAmjevPmVvSIbsmXLH0ybdj1xcdKJvemmwOLZ7Xbt559l5tPdd9Nk/nyaXOYMqMTERNq2bVvFwVUhT084fBiP+fPxePxxGR0zYQKMGgW1a1/xt7P79qhE2hbWKtoeHh4elzxf7vvokJAQ4uLiANi9ezfe3t64u7sD4OvrS3JyMvnnpmAnJCTQunVrFi5cyOrVqwHYu3cvnp6eZZK5vUtNhcGDW/PjjzKv5f33cYxkDrKmyv33ywgRB/u5XZKPD7z8spRjIiPlQvDrr5eUm85dO1LKVpWbgjp27EhQUBBhYWGYTCamT59OTEwMHh4e9O7dm1GjRjF8+HCcnZ0JDg6mU6dONG/enMmTJ7Ns2TIKCwuLhzM6kg8+gJQUV378EW6/3ehoKsnp0zIyxM8PvvnG6GiMU7s2DB0qs2LT02Wc+5kz0mMPDpbRMaGhdn7FWzmiy+pTTpo0yep+YGBg8edhYWGEhYVZnW/atCmRkZGVEJ7tGj8efH1Tuf12+y0ZWTGb4dFH5cLgsmWarEDawMdHPj9zRkov8+fLUMe2baUcM2yY9OSVsgEOMnSh+jVvDqGh9ntBt4w5cySRBwdrMr8Qyzj8gwfh88+hXj146ilZSwbkH6Iqy2yG3bvx+P57aasTJ4yOyKE5StW3Wn37rWwd17mz0ZFUku++g2nTZJnaGjgZ7IrUri298qFD4b//LVk07J//lBmoTz8tayPX9H+KubnyD3D5ckhKovh97JtvwsSJ8o9xzBho3VpKfJbbwEDHecdjNss/sNRUWWvIyws6dqzSp9SEXgGzZ8vf6623Gh1JJfjjD5lJedNN8Mknmogul8kka7lbeHnBZ5/BypVwww0wYQLO7doZF191y8+HH3+Ev/+WdS/q1pV3fEFBMHEif3l50aZOHUnYIMswp6XJEqSle+3R0bJq5n//K+vxlE72rVvL421hJ6uCAjhyRJL1oUNy26xZyYqfN94ocxpOnSp5zMiRsHhxlYalCf0KHToEv/4qgx8cQmam1I+WL5cygqqYf/5TeudffQXvvQdjx+IVFgbdusmyw7GxsiuTQyzsc05Wlry7W75cli/OzZUEHhYmo6OSk4t3PD+dmCjXHSxuvBH+9z/5PDtb1t3Zv7/kbW9Wltxfv956dFF8vOxk9d13khz9/KyT/nXXVc5wsz//lH1xU1NLPry85J8MyD/tpCTrx/TrV5LQe/SQay2+viUflj13q5B9JvQlS6SOOXy47BpRztjMyrR8udz27+8gZdOQkKrbqKKmqVNHfieHDYNt2zh+5AiNQFZ9fPhh+UDfueIAABZHSURBVJobbpA/9p49ZaRMw4YGBlwBR47IbFuTCZ5/Xv4WfXzgkUfkj6JXr5KvPZfMy9WgAXToIB8WoaGwc6f8kR0/Lsl1/35pP5CE//vvkthL71x16JAkz88/l3Olk72fnyRVJyeZAb19u3XCrltXHgPw+OOyG41F48byD9kiPFz+UZdO2Ofm4wAyDM4A9pnQXV3lBzBypOzn9sAD8kfUp0+VlwxiY+VdX2Cgnc8GfeEFqQe/9JIm88pmMsHNN3PG8gty441SWli/HjZskNLMRx/BmjWyVvsff0BCgqwx36SJoaFfUGKi9GSWL5d1LXbulNf0/POS+G69tep+h0wmSaaNG1tvcvLII/Jx9qwMLd23Tz6uuUbOZ2ZK2SYmRsojID33U6ck1gULZOkHV1cplfj6SsK3eP112dTc11fOn1/meeyxqnm9V8tsgJSUFPN1111nTklJqfg3OXvWbI6PN5vHjjWbGzUymzt1Kjl36NDVB3kBBQXyNNOmyf3ff/+9Sp6nykVFmc1gNj/5ZKV+W7ttjypy0fY4c0Z+d0+elPsvvyw/DzCbb7zRbJ4wwWz+9luz+fTp6gv2QnbtMpuvv74kts6dzeaICLP58OEr/laG/W4UFprNBw+azf/5j9n81Vclxw8fNpvT083moiJDwqpoe5SXO+2zhw7yn/u22+TjnXekxw5yUSYgQOp1w4fLBT/LWOKrVKuW1M+Liirl2xlj2zYZT3377TK9VVU/Fxf5vbUID5d3lxs2SC9+8WL5yMqS85Y63x13WL+tr0ynT8tzL18uF8gtI1DatIFnnpEt/3x9q+a5q5KzM7RoIR+lWXryDsYx3mvXri2/eCBvp+bMkR/kc8/JL+G990od8yoVFsqt3c6GT0uTKf1NmshM0Mutcaqq5eIiG11PmwZr18qoj61b5TjIsKoBA+TnFhwsv9dr11bOc3/7rVzE9PKS5YSXLi3pHLm7S0157Fj7TOY1kGMk9NLc3WUG36+/ylZkkybBb7+V/HHs3CmLT13hFc28PCmlLVxYBTFXl02b5B3MihWONdrC0bi6ymgRi//8Ry7ivfqq1JLnz7f+RXz1VfmZXs6knSNHJIlbLFwoPfOwMNkvNiMDZsyovNeiqpX9llwuxw03wKxZEBFRctHmzTdl4aU2bUpGJFh695ewdq38rl/Gl9quAQNkhEWp9e2VHXB1leGP3brBiy9KecSSvDMz5QJefr6UIW+6SX7Gw4ZJbx7koqvloubmzfJ1R45IKTIyUso4dvu2U5XmeD30Cyl9Bf6jj2SUgZ+fzGTz95feSTliYyUPlh65ZDcWL5ar/aDJ3BHUri1DB0HKMCdOSC9++nQZAvjRR/LuFOR3PTAQpk6VmuFrr8kwVcs7NC8vTeYOxLF76Bfi7i498+HDZQPQpUvBzU3OFRZKvbB/fxlOdm6CQkGBbCl3770llRu78dNPcoGrb1/poetMUMdTp470NO64Q5J6fn5JSbF3bxkTff/9ZS8MKodT8xJ6aS1aSM/F4s8/pSu+aJH0YIYMgeHD+SmzAydOmBgwwLhQK+TAAZl45e8v/7g0mdcMpcdMN2smS4OqGqFmlFwuV9u2sojO8uVSr5w7F4KDaXvsZ158Efr2saOpoXl58k7j9Gm5YFbDN+9Wqiao2T30C3F1lben998Px45BbCzNHurKjEHAlKky/HH4cEmWtiw6WmJdvRquv97oaJRS1UB76JfSuDHJPUfz/VpnzpxBLkQlJkoppmlTmk2ZIivK2aLHHpNJRP36GR2JUqqaaEIvx+LF8I9/yPrnPPecrBexfj089BBuGzfK2F2Qi1Bt28qiQs89J0vR/vqrlD6q07//LWPtQYawKaVqDC25lGP5chk80LjxuQNOTjLOt0cP/pw0ibaWgen5+TLbb9cuWfjHsg7ylCkyFv7kSbm98Ub5uPbayt9VOjERHnxQxh9v2KAXQZWqYTShX8Iff0iOHDv2Il9gMpWMKKhbV3rlIIu97Nsnyd2yBnJSkkxwOntW7ru6So9+zhxZxyM3V9bu8PWtWCLOypK6f926MllEk7lSNY4m9EuIjZXbK77+6ewsC4QFBJQc69BBeumJiZLoExLk1rKW+7p1Mk68YUNo10568e3ayTraXl6Xfr6iIlmEbP9+2TWmZcsrDFgp5Qg0oV/Cxo2yBHOlzceoU0fKIZYp2aXddBN8+GFJoo+Kkp1cQkMloX/xhYwltyT6G2+UHn6dOvDxx7K29oIFMtxSKVUjXVZCj4iIYMeOHZhMJsLDw2nfvn3xuaVLl7Jy5UqcnJxo164dL7zwAgUFBUydOpXDhw/j7OzM66+/Tgs7nKW2apUsUFgtWreWXeQtzGZZ9c6yzKdlD8Mff0SG3CA1+KwsWQ63UaPLWsJAKeW4yk3oW7du5cCBA0RHR5OcnEx4eDjR0dEA5ObmsnjxYtauXUutWrUYOXIkv/32G/v27aN+/fq89dZb/Pzzz7z11lu8++67Vf5iKpuTk4HLJptMstenxWOPyUdhocxo3bVLZoJali3QZK5UjVfusMX4+HhCQ0MB8Pf3Jzs7m9xzm7a6uLjg4uJCXl4ehYWFnDp1igYNGhAfH0/v3r0B6Nq1K9u2bavCl1A1Royw0Y2ga9WSUsvAgTB5stHRKKVsSLkJPTMzk0alVujz9PQkIyMDgNq1azNu3DhCQ0Pp2bMnHTp0wM/Pj8zMTDzP7azi5OSEyWTijKVMYAdOnJAStmXDGKWUsgdXfFHUXGpjiNzcXBYsWMCaNWtwd3dnxIgR7Nmz55KPKS0pKYmcnJwrDaHKrVxZn8JCX4KD95GYmH/Rr8vPzyfRrneKrlzaHta0PUpoW1iraHuklXNRr9yE7u3tTWZmZvH99PR0vM4No0tOTqZFixbFvfFOnTqRkJCAt7c3GRkZBAYGUlBQgNlsxvUC250FBATQvHSd2Ea8+KLUzgcO9LvkZuaJiYm0bdu2+gKzcdoe1rQ9SmhbWKtoe3hYhjlfRLkll5CQEOLi4gDYvXs33t7euLu7A+Dr60tycjL5+dKLTUhIoHXr1oSEhLBmzRoA1q9fz6233nrFgRvl1CkZAdi/P5dM5kopZWvK7aF37NiRoKAgwsLCMJlMTJ8+nZiYGDw8POjduzejRo1i+PDhODs7ExwcTKdOnSgqKmLTpk0MHjwYV1dXZs2aVR2vpVL8/bcMGNFBI0ope3NZNfRJkyZZ3Q8MDCz+PCwsjLDzsp9l7Lk98vGRBbmUUsreaFGhlMJC+O23kt27lFLKnmhCL2XjRpmVv3q10ZEopdSV04ReyvLlsjRKr15GR6KUUldOE/o5ZrMk9D59SmbTK6WUPdGEfs62bXDwoO1vFaqUUhejCf2cFStk3Pk//mF0JEopVTGa0M+ZMkVWpm3SxOhIlFKqYjShn+PmBt27Gx2FUkpVnCZ0ZCOgmTNLtvtUSil7pAkd2fktJkbXblFK2bcan8KOHIH4eNmfWSml7FmNT+grVsitDldUStm7Gp/Qly+HgAAICjI6EqWUujo1OqGbzVC3riyVazIZHY1SSl2dK96CzpGYTBAba3QUSilVOWp0D/3vv42OQCmlKk+NTej5+dC8OdjRZkpKKXVJNTahr1sHOTmy/rlSSjmCGpvQY2Ohfn3o2dPoSJRSqnLUyIReVAQrV8I994Crq9HRKKVU5aiRCf2XXyAzU2eHKqUcS41M6G3byvotd91ldCRKKVV5LmscekREBDt27MBkMhEeHk779u0BSEtLY9KkScVfl5KSwsSJEykoKOC9996jZcuWAHTt2pWxY8dWQfgV4+UFTz1ldBRKKVW5yk3oW7du5cCBA0RHR5OcnEx4eDjR0dEA+Pj4EBkZCUBhYSHDhg2jV69exMXF0a9fP6ZMmVK10VfA3r3w008waBB4eBgdjVJKVZ5ySy7x8fGEhoYC4O/vT3Z2Nrm5uWW+LjY2lr59++Jm4zssL10KTz4p49CVUsqRlJvQMzMzadSoUfF9T09PMjIyynzd119/zUMPPVR8f+vWrYwaNYoRI0bw+++/V1K4Vy82Frp1k7KLUko5kitey8VsNpc5tn37dtq0aYO7uzsAHTp0wNPTkx49erB9+3amTJnCqlWryjwuKSmJnJycCoRdMQcPurBrVwBTpqSRmHj8qr9ffn4+iYmJlRCZY9D2sKbtUULbwlpF2yMtLe2S58tN6N7e3mRmZhbfT09Px+u87u2GDRvo0qVL8X1/f3/8/f0BCA4O5vjx4xQVFeHs7Gz1uICAAJo3b17+q6gk330nt2PG+NC6tc9Vf7/ExETatm171d/HUWh7WNP2KKFtYa2i7eFRzoW/cksuISEhxMXFAbB79268vb2Le+IWu3btIjAwsPj+woULWb16NQB79+7F09OzTDI3QmIi3HQTtG5tdCRKKVX5yu2hd+zYkaCgIMLCwjCZTEyfPp2YmBg8PDzo3bs3ABkZGTRu3Lj4Mf/4xz+YPHkyy5Yto7CwkJkzZ1bdK7gCixbByZNGR6GUUlXjsmropceaA1a9caBMfbxp06bFwxlthdks65/b+CAcpZSqsBozU/Thh+Hpp42OQimlqk6NSOh//w2rVulCXEopx1YjEvp338GZM7oYl1LKsdWIhB4bC97ecNttRkeilFJVx+ETen6+9NDvvx9sYOSkUkpVmSueKWpvTp+G55+HcyMslVLKYTl8Qm/QAF55xegolFKq6jl0yaWoSMoturKiUqomcOiEHh8v+4auWGF0JEopVfUcOqHHxsrY87vvNjoSpZSqeg6b0M1mSeh33gn16xsdjVJKVT2HTeg7d8K+fTqZSClVczhsQl+zRhbjuu8+oyNRSqnq4bAJffJk2LULfK5+HwullLILDpvQnZwgKMjoKJRSqvo4ZEJfsgTGjoWCAqMjUUqp6uOQM0WXLIHsbHBxMToSpZSqPg7XQ09Ph59/1tEtSqmax+ES+sqVMga9f3+jI1FKqerlcAl9+XJo3Ro6dDA6EqWUql4OV0Nv00Y2sjCZjI5EKaWql8Ml9PffNzoCpZQyxmUl9IiICHbs2IHJZCI8PJz27dsDkJaWxqRJk4q/LiUlhYkTJ3LXXXcxdepUDh8+jLOzM6+//jotWrSomldQyv790KqV9s6VUjVTuTX0rVu3cuDAAaKjo5k5cyYzZ84sPufj40NkZCSRkZEsWbKEa665hl69erF69Wrq16/Pl19+yZgxY3jrrbeq9EWA7EzUvj2U+v+ilFI1SrkJPT4+ntDQUAD8/f3Jzs4mNze3zNfFxsbSt29f3NzciI+Pp/e5Pd+6du3Ktm3bKjnsstavh5wc6NWryp9KKaVsUrkll8zMTIJKzaH39PQkIyMDd3d3q6/7+uuv+eSTT4of4+npCYCTkxMmk4kzZ87g6upq9ZikpCRycnKu+kUAfPJJU+rVq4+v758kJpor5XuWJz8/n8TExGp5Lnug7WFN26OEtoW1irZHWlraJc9f8UVRs7lssty+fTtt2rQpk+Qv9RiAgIAAmjdvfqUhlFFUBD/9JLsT3XRT4FV/v8uVmJhI27Ztq+35bJ22hzVtjxLaFtYq2h4eHh6XPF9uycXb25vMzMzi++np6Xh5eVl9zYYNG+jSpYvVYzIyMgAoKCjAbDaX6Z1Xpi1bIC1NZ4cqpWq2chN6SEgIcXFxAOzevRtvb+8yPfFdu3YRGBho9Zg1a9YAsH79em699dbKjLmM4GDZN/See6r0aZRSyqaVW3Lp2LEjQUFBhIWFYTKZmD59OjExMXh4eBRf+MzIyKBx48bFj+nXrx+bNm1i8ODBuLq6MmvWrKp7BUDdurqRhVJKXVYNfdJ5YwFL98YBVq1aZXXfMva8OvzxB0RHw5gx4O1dLU+plFI2ye7XcvnqK3j5ZbkwqpRSNZndJ/TYWFm75ZprjI5EKaWMZdcJff9+2L5dR7copRTYeUJfsUJuNaErpZSdJ/SMDLjlFggIMDoSpZQynl0n9Ndeg82bjY5CKaVsg90m9IICuXWy21eglFKVy27T4UMPwYMPGh2FUkrZDrtM6Lm5EBcH1bBnhlJK2Q27TOhr1siGFjq6RSmlSthlQl++HBo3hpAQoyNRSinbYXcJ/cwZWL1aFuOq5XBbXCulVMXZXUosKoLZs6FDB6MjUUop22J3Cb1uXXjySaOjUEop22N3JRellFIXpgldKaUchCZ0pZRyEJrQlVLKQWhCV0opB6EJXSmlHIQmdKWUchCGjEMvOrej89GjR414+kqTlpaGh4eH0WHYDG0Pa9oeJbQtrFW0PSw505JDz2dIQs/IyABgyJAhRjy9UkrZtYyMDFq1alXmuMlsNpurO5j8/HwSEhLw8vLC2dm5up9eKaXsUlFRERkZGbRr1446deqUOW9IQldKKVX59KKoUko5CE3oFTRnzhwGDRrEgw8+yNq1a40Ox3D5+fmEhoYSExNjdCiGW7lyJffddx8PPPAAGzZsMDocQ508eZLx48czbNgwwsLC2Lhxo9EhGWLv3r2EhobyxRdfAHDkyBGGDRvGI488wjPPPMOZM2cq5Xk0oVfA5s2b+fPPP4mOjmbRokVEREQYHZLh5s2bR4MGDYwOw3AnTpzgww8/JCoqivnz5/PDDz8YHZKhYmNj8fPzIzIykvfee4+ZM2caHVK1y8vLY8aMGXTp0qX42Pvvv88jjzxCVFQUrVq14ptvvqmU59KEXgG33HIL7733HgD169fn1KlTFx1GVBMkJyeTlJREjx49jA7FcPHx8XTp0gV3d3e8vb2ZMWOG0SEZqlGjRmRlZQHw999/06hRI4Mjqn6urq4sXLgQb2/v4mNbtmzhzjvvBKBnz57Ex8dXynNpQq8AZ2dn6tWrB8A333zDHXfcUaNH68yePZupU6caHYZNOHToEPn5+YwZM4ZHHnmk0v5Q7dU999zD4cOH6d27N0OHDmXKlClGh1TtatWqVWZEyqlTp3B1dQWgcePGxUO5r/q5KuW71FDr1q3jm2++4ZNPPjE6FMMsX76cm266iRYtWhgdis3Iyspi7ty5HD58mOHDh7N+/XpMJpPRYRlixYoVNGvWjMWLF7Nnzx7Cw8P1Ost5KnOgoSb0Ctq4cSPz589n0aJFNXoG3IYNG0hJSWHDhg0cPXoUV1dXmjZtSteuXY0OzRCNGzcmODiYWrVq0bJlS9zc3Dh+/DiNGzc2OjRDbNu2jW7dugEQGBhIeno6RUVFNfodLUC9evXIz8+nTp06pKWlWZVjroaWXCogJyeHOXPmsGDBAho2bGh0OIZ69913+fbbb/nqq694+OGHeeqpp2psMgfo1q0bmzdv5uzZs5w4cYK8vLwaWTe2aNWqFTt27AAgNTUVNze3Gp/MAbp27UpcXBwAa9eu5fbbb6+U76s99Ar47rvvOHHiBM8++2zxsdmzZ9OsWTMDo1K2wMfHh759+zJw4EAAXnzxRZycam6/adCgQYSHhzN06FAKCwt5+eWXjQ6p2iUkJDB79mxSU1OpVasWcXFxvPnmm0ydOpXo6GiaNWtG//79K+W5dKaoUko5iJrbdVBKKQejCV0ppRyEJnSllHIQmtCVUspBaEJXSikHoQldKaUchCZ0pZRyEJrQlVLKQfw/QB8NQwvBHQEAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YLO526Sgf-U_",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "9114ffae-9e5d-4f95-c0bb-18cbe01954a9"
      },
      "source": [
        "model.evaluate(input_test, y_test)"
      ],
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "782/782 [==============================] - 36s 46ms/step - loss: 0.6217 - acc: 0.8339\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.6216864585876465, 0.8338800072669983]"
            ]
          },
          "metadata": {},
          "execution_count": 24
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VTmZ_XiZguM5"
      },
      "source": [
        "- 전체 시퀀스가 아니라 순서대로 500개의 단어만 입력했기 때문에 성능이 낮게 나옴\n",
        "\n",
        "- SimpleRNN은 긴 시퀀스를 처리하는데 적합하지 않음\n",
        "\n",
        "- SimpleRNN은 실전에 사용하기엔 너무 단순\n",
        "\n",
        "- SimpleRNN은 이론적으로 시간 $t$ 에서 이전의 모든 타임스텝의 정보를 유지할 수 있지만, 실제로는 긴 시간에 걸친 의존성은 학습할 수 없음\n",
        "\n",
        "- 그래디언트 소실 문제(vanishing gradient problem)\n",
        "  - 이를 방지하기 위해 LSTM, GRU 같은 레이어 등장\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oZXJIdDai7sV"
      },
      "source": [
        "# LSTM(Long Short-Term Memory)\n",
        "- 장단기 메모리 알고리즘\n",
        "\n",
        "- 나중을 위해 정보를 저장함으로써 오래된 시그널이 점차 소실되는 것을 막아줌\n",
        "\n",
        "  <img src=\"https://colah.github.io/posts/2015-08-Understanding-LSTMs/img/LSTM3-chain.png\">\n",
        "\n",
        "  <sub>출처: https://colah.github.io/posts/2015-08-Understanding-LSTMs/</sub>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "j3afhQOUlqG9"
      },
      "source": [
        "## IMDB 데이터"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tZOCK6EQl1QI"
      },
      "source": [
        "### 데이터 로드"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "00ZS9v3YmT_H"
      },
      "source": [
        "from tensorflow.keras.datasets import imdb\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences"
      ],
      "execution_count": 27,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UW-VG35_gwmt",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "99ca4bc7-11a1-41f9-965f-1970308d2b35"
      },
      "source": [
        "num_words = 10000\n",
        "max_len = 500\n",
        " \n",
        "(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words = num_words)\n",
        "print(len(x_train))\n",
        "print(len(x_test))\n",
        "\n",
        "pad_x_train = sequence.pad_sequences(x_train, maxlen=max_len)\n",
        "pad_x_test = sequence.pad_sequences(x_test, maxlen=max_len)\n",
        "print(pad_x_train.shape)\n",
        "print(pad_x_test.shape)"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "25000\n",
            "25000\n",
            "(25000, 500)\n",
            "(25000, 500)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uo8uhFyBl2k-"
      },
      "source": [
        "### 모델 구성"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tFkAN2Dzl3GC"
      },
      "source": [
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.layers import Dense, LSTM, GRU, Embedding"
      ],
      "execution_count": 29,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TPr9Ec-El50W",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8c3981a9-ac99-45b1-f512-74f813d69bb6"
      },
      "source": [
        "model = Sequential()\n",
        "\n",
        "model.add(Embedding(num_words, 32))\n",
        "model.add(LSTM(32))\n",
        "model.add(Dense(1, activation='sigmoid'))\n",
        "\n",
        "model.compile(optimizer='rmsprop',\n",
        "              loss='binary_crossentropy',\n",
        "              metrics=['acc'])\n",
        "model.summary()"
      ],
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential_6\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_6 (Embedding)      (None, None, 32)          320000    \n",
            "_________________________________________________________________\n",
            "lstm_1 (LSTM)                (None, 32)                8320      \n",
            "_________________________________________________________________\n",
            "dense_2 (Dense)              (None, 1)                 33        \n",
            "=================================================================\n",
            "Total params: 328,353\n",
            "Trainable params: 328,353\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vAPnX6ACmenz"
      },
      "source": [
        "### 모델 학습"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FGWhqX5GmNeR",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4570541c-1008-456c-99c4-3deac4b3b70f"
      },
      "source": [
        "history = model.fit(pad_x_train, y_train,\n",
        "                    epochs=10,\n",
        "                    batch_size=128,\n",
        "                    validation_split=0.2)"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/10\n",
            "157/157 [==============================] - 18s 85ms/step - loss: 0.5047 - acc: 0.7588 - val_loss: 0.3263 - val_acc: 0.8708\n",
            "Epoch 2/10\n",
            "157/157 [==============================] - 12s 78ms/step - loss: 0.2896 - acc: 0.8885 - val_loss: 0.3171 - val_acc: 0.8696\n",
            "Epoch 3/10\n",
            "157/157 [==============================] - 12s 78ms/step - loss: 0.2340 - acc: 0.9119 - val_loss: 0.3194 - val_acc: 0.8658\n",
            "Epoch 4/10\n",
            "157/157 [==============================] - 12s 78ms/step - loss: 0.1962 - acc: 0.9276 - val_loss: 0.2842 - val_acc: 0.8884\n",
            "Epoch 5/10\n",
            "157/157 [==============================] - 12s 78ms/step - loss: 0.1756 - acc: 0.9384 - val_loss: 0.2853 - val_acc: 0.8874\n",
            "Epoch 6/10\n",
            "157/157 [==============================] - 12s 77ms/step - loss: 0.1567 - acc: 0.9445 - val_loss: 0.4031 - val_acc: 0.8750\n",
            "Epoch 7/10\n",
            "157/157 [==============================] - 12s 77ms/step - loss: 0.1417 - acc: 0.9488 - val_loss: 0.3404 - val_acc: 0.8730\n",
            "Epoch 8/10\n",
            "157/157 [==============================] - 12s 77ms/step - loss: 0.1240 - acc: 0.9566 - val_loss: 0.3639 - val_acc: 0.8854\n",
            "Epoch 9/10\n",
            "157/157 [==============================] - 12s 76ms/step - loss: 0.1174 - acc: 0.9596 - val_loss: 0.3386 - val_acc: 0.8590\n",
            "Epoch 10/10\n",
            "157/157 [==============================] - 12s 76ms/step - loss: 0.1111 - acc: 0.9602 - val_loss: 0.3440 - val_acc: 0.8764\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gFiE5SpPmh_x"
      },
      "source": [
        "### 시각화"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jwQ6ZYHTyXHS",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 507
        },
        "outputId": "9d510081-47ec-48a4-a453-1464f155e474"
      },
      "source": [
        "loss = history.history['loss']\n",
        "acc = history.history['acc']\n",
        "val_loss = history.history['val_loss']\n",
        "val_acc = history.history['val_acc']\n",
        "\n",
        "epochs = range(1, len(loss)+1)\n",
        "\n",
        "plt.plot(epochs, loss, 'b--', label='training loss')\n",
        "plt.plot(epochs, val_loss, 'r--' ,label='validation loss')\n",
        "plt.grid()\n",
        "plt.legend()\n",
        "\n",
        "plt.figure()\n",
        "plt.plot(epochs, acc, 'b--', label='training acc')\n",
        "plt.plot(epochs, val_acc, 'r--', label='validation acc')\n",
        "plt.grid()\n",
        "plt.legend()\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NiZ6uiUPyXHN"
      },
      "source": [
        "### 모델 평가"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DZjW6iOV3k-J",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8845f72f-fca5-4074-f4e7-c11b534e0460"
      },
      "source": [
        "model.evaluate(pad_x_test, y_test)"
      ],
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "782/782 [==============================] - 23s 29ms/step - loss: 0.3829 - acc: 0.8652\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.38287097215652466, 0.8651599884033203]"
            ]
          },
          "metadata": {},
          "execution_count": 37
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Dc0wuvArGHgD"
      },
      "source": [
        "# GRU (Gated Recurrent Unit)\n",
        "- LSTM을 더 단순하게 만든 구조\n",
        "\n",
        "- 기억 셀은 없고, 시간방향으로 전파하는 것은 은닉 상태만 있음\n",
        "\n",
        "- reset gate\n",
        "  - 과거의 은닉 상태를 얼마나 무시할지 결정\n",
        "\n",
        "  - $r$ 값이 결정\n",
        "\n",
        "- update gate\n",
        "  -  은닉 상태를 갱신하는 게이트  \n",
        "\n",
        "  - LSTM의 forget, input gate 역할을 동시에 함\n",
        "  \n",
        "  <img src=\"https://miro.medium.com/max/1400/1*jhi5uOm9PvZfmxvfaCektw.png\" width=\"500\">\n",
        "\n",
        "<sub>출처: https://towardsdatascience.com/illustrated-guide-to-lstms-and-gru-s-a-step-by-step-explanation-44e9eb85bf21</sub>\n",
        "\n",
        "  ### $\\qquad z = \\sigma (x_t W^{(z)}_x + h_{t-1} W^{(z)}_h + b^{(z)} \\\\ \n",
        "  \\qquad r = \\sigma (x_t W^{(r)}_x + h_{t-1} W^{(r)}_h + b^{(r)}) \\\\\n",
        "  \\qquad \\tilde{i} = tanh (x_t W^{(i)}_x + (r \\odot h_{t-1}) W^{(i)}_h + b ) \\\\\n",
        "  \\qquad h_t = (1 - z) \\odot h_{t-1} + z \\odot \\tilde{h}$\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TvnFLeqSsOSh"
      },
      "source": [
        "## Reuters 데이터\n",
        "\n",
        "- IMDB와 유사한 데이터셋(텍스트 데이터)\n",
        "\n",
        "- 46개의 상호 배타적인 토픽으로 이루어진 데이터셋 \n",
        "  - 다중 분류 문제\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ONPaMDFEsftq"
      },
      "source": [
        "### 데이터셋 로드"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sFyvFVizsUVD"
      },
      "source": [
        "from tensorflow.keras.datasets import reuters"
      ],
      "execution_count": 38,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "C234q3AVsUSp",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e7a9b5ec-f6c2-402d-8636-1c5331f81f5f"
      },
      "source": [
        "num_words = 1000\n",
        "(x_train,y_train), (x_test, y_test) = reuters.load_data(num_words=num_words)\n",
        "\n",
        "print(x_train.shape)\n",
        "print(y_train.shape)\n",
        "print(x_test.shape)\n",
        "print(y_test.shape)"
      ],
      "execution_count": 39,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/reuters.npz\n",
            "2113536/2110848 [==============================] - 0s 0us/step\n",
            "2121728/2110848 [==============================] - 0s 0us/step\n",
            "(8982,)\n",
            "(8982,)\n",
            "(2246,)\n",
            "(2246,)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VioYGD8Ms4zH"
      },
      "source": [
        "### 데이터 전처리 및 확인"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2Sb0OxNlsUQT"
      },
      "source": [
        "from tensorflow.keras.preprocessing.sequence import pad_sequences"
      ],
      "execution_count": 40,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rNUkve6-sUN6"
      },
      "source": [
        "max_len = 300"
      ],
      "execution_count": 41,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jBxQZtOJsULl",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e1d04a7d-78b4-4bc8-ef3e-7fda45ec5b5d"
      },
      "source": [
        "pad_x_train = pad_sequences(x_train, maxlen=max_len)\n",
        "pad_x_test = pad_sequences(x_test, maxlen=max_len)\n",
        "\n",
        "print(len(pad_x_train[0]))"
      ],
      "execution_count": 42,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "300\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "a6hwQtn_sUJN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "dadc7b85-fcf9-4ca4-d51c-6ca402ff9026"
      },
      "source": [
        "pad_x_train[5]"
      ],
      "execution_count": 43,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "         0,   0,   1,   4,  37,  38, 309, 213, 349,   2,  48, 193, 229,\n",
              "       463,  28, 156, 635,  11,  82,  14, 156, 635,  11,  82,  54, 139,\n",
              "        16, 349, 105, 462, 311,  28, 296, 147,  11,  82,  14, 296, 147,\n",
              "        11,  54, 139, 342,  48, 193,   2, 361, 122,  23,   2,  28, 318,\n",
              "       942,  11,  82,  14, 318, 942,  11,  82,  54, 139, 122,   7, 105,\n",
              "       462,  23, 349,  28, 296, 767,  11,  82,  14, 296, 767,  11,  54,\n",
              "       139, 342, 229, 162,   7,  48, 193,  55, 408,  28, 258, 557,  11,\n",
              "        82,  14, 196, 557,  11,  82,  54, 139, 162,   7, 105, 462,  55,\n",
              "       349,  28, 191, 968,  11,  82,  14, 191, 785,  11,  54, 139,  17,\n",
              "        12], dtype=int32)"
            ]
          },
          "metadata": {},
          "execution_count": 43
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_P_JxBFntQOP"
      },
      "source": [
        "### 모델 구성\n",
        "- LSTM 레이어도 SimpleRNN과 같이 `return_sequences` 인자 사용가능"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1vG3ZGSksUG5"
      },
      "source": [
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.layers import GRU, Dense, Embedding"
      ],
      "execution_count": 44,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K_uMdxmysUEg",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "9eaa6930-0488-43a7-a98a-4289233e6664"
      },
      "source": [
        "model = Sequential()\n",
        "model.add(Embedding(input_dim=num_words, output_dim=256))\n",
        "model.add(GRU(256, return_sequences=True))\n",
        "model.add(GRU(128))\n",
        "model.add(Dense(46, activation='softmax'))\n",
        "\n",
        "model.compile(optimizer='adam',\n",
        "              loss='sparse_categorical_crossentropy',\n",
        "              metrics=['acc'])\n",
        "\n",
        "model.summary()"
      ],
      "execution_count": 45,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential_7\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_7 (Embedding)      (None, None, 256)         256000    \n",
            "_________________________________________________________________\n",
            "gru (GRU)                    (None, None, 256)         394752    \n",
            "_________________________________________________________________\n",
            "gru_1 (GRU)                  (None, 128)               148224    \n",
            "_________________________________________________________________\n",
            "dense_3 (Dense)              (None, 46)                5934      \n",
            "=================================================================\n",
            "Total params: 804,910\n",
            "Trainable params: 804,910\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qjgupW2puHXA"
      },
      "source": [
        "### 모델 학습"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "T22YsRQ8sUBg",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "2ef69b53-e2dd-4a52-a583-6e7441e2158f"
      },
      "source": [
        "history = model.fit(pad_x_train, y_train,\n",
        "                    batch_size=32,\n",
        "                    epochs=20,\n",
        "                    validation_split=0.2)"
      ],
      "execution_count": 46,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/20\n",
            "225/225 [==============================] - 33s 134ms/step - loss: 2.0513 - acc: 0.4615 - val_loss: 1.7393 - val_acc: 0.5292\n",
            "Epoch 2/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 1.6984 - acc: 0.5699 - val_loss: 1.6136 - val_acc: 0.6004\n",
            "Epoch 3/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 1.4259 - acc: 0.6422 - val_loss: 1.3835 - val_acc: 0.6500\n",
            "Epoch 4/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 1.2216 - acc: 0.6900 - val_loss: 1.2540 - val_acc: 0.6967\n",
            "Epoch 5/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 1.0455 - acc: 0.7360 - val_loss: 1.1174 - val_acc: 0.7362\n",
            "Epoch 6/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.9052 - acc: 0.7705 - val_loss: 1.0385 - val_acc: 0.7496\n",
            "Epoch 7/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.7990 - acc: 0.7993 - val_loss: 0.9957 - val_acc: 0.7568\n",
            "Epoch 8/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.6932 - acc: 0.8216 - val_loss: 0.9844 - val_acc: 0.7618\n",
            "Epoch 9/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.6149 - acc: 0.8490 - val_loss: 1.0088 - val_acc: 0.7596\n",
            "Epoch 10/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.5380 - acc: 0.8651 - val_loss: 0.9961 - val_acc: 0.7735\n",
            "Epoch 11/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.4753 - acc: 0.8803 - val_loss: 0.9835 - val_acc: 0.7652\n",
            "Epoch 12/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.4172 - acc: 0.8958 - val_loss: 1.0275 - val_acc: 0.7657\n",
            "Epoch 13/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.3615 - acc: 0.9094 - val_loss: 1.1055 - val_acc: 0.7557\n",
            "Epoch 14/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.3229 - acc: 0.9179 - val_loss: 1.1174 - val_acc: 0.7579\n",
            "Epoch 15/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.2844 - acc: 0.9286 - val_loss: 1.1545 - val_acc: 0.7596\n",
            "Epoch 16/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.2708 - acc: 0.9280 - val_loss: 1.1488 - val_acc: 0.7585\n",
            "Epoch 17/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.2146 - acc: 0.9429 - val_loss: 1.2093 - val_acc: 0.7629\n",
            "Epoch 18/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.1915 - acc: 0.9479 - val_loss: 1.2288 - val_acc: 0.7596\n",
            "Epoch 19/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.1671 - acc: 0.9534 - val_loss: 1.2827 - val_acc: 0.7579\n",
            "Epoch 20/20\n",
            "225/225 [==============================] - 29s 130ms/step - loss: 0.1563 - acc: 0.9552 - val_loss: 1.3324 - val_acc: 0.7557\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JGZhIiYeuSCD"
      },
      "source": [
        "### 시각화"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "INJu0-itsT_T",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 507
        },
        "outputId": "ebbb82f6-18bb-46d8-cd2c-191c741ae6f0"
      },
      "source": [
        "loss = history.history['loss']\n",
        "acc = history.history['acc']\n",
        "val_loss = history.history['val_loss']\n",
        "val_acc = history.history['val_acc']\n",
        "\n",
        "epochs = range(1, len(loss)+1)\n",
        "\n",
        "plt.plot(epochs, loss, 'b--', label='training loss')\n",
        "plt.plot(epochs, val_loss, 'r--' ,label='validation loss')\n",
        "plt.grid()\n",
        "plt.legend()\n",
        "\n",
        "plt.figure()\n",
        "plt.plot(epochs, acc, 'b--', label='training acc')\n",
        "plt.plot(epochs, val_acc, 'r--', label='validation acc')\n",
        "plt.grid()\n",
        "plt.legend()\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": 47,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dLZLATDBuTtj"
      },
      "source": [
        "### 모델 평가\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SVxVV-kmsT6D",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3a5c8cdb-e3e8-4228-fda0-920fe02115d0"
      },
      "source": [
        "model.evaluate(pad_x_test, y_test)"
      ],
      "execution_count": 48,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "71/71 [==============================] - 4s 55ms/step - loss: 1.3397 - acc: 0.7556\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[1.3396633863449097, 0.7555654644966125]"
            ]
          },
          "metadata": {},
          "execution_count": 48
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dYf8DhjTnzsm"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}