playground_10_recurrent_light.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "playground_10_recurrent_light.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true,
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/darja/NeuralUni/blob/master/playground_10_recurrent_light.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "metadata": {
        "id": "cJrkB3Zmv1rc",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Определение тональности отзывов на фильмы с сайта IMDB\n",
        "\n",
        "* Замените полносвязную сеть на рекуррентную. Попробуйте три архитектуры сети:\n",
        "    * SimpleRNN\n",
        "    * GRU\n",
        "    * LSTM\n",
        "*  Используйте DropOut, чтобы снизить переобучение:\n",
        "    * Попробуйте вставить слой SpatialDropout1D после слоя Embedding\n",
        "    * Используйте параметры dropout и recurrent_dropout для рекуррентных слоев\n",
        "* Меняйте архитектуру сети и гиперпараметры обучения, чтобы повысить качество распознавания тональности:\n",
        "    * Измените длину вектора в слое Embedding\n",
        "    * Изменяйте количество нейронов в рекуррентном слое\n",
        "    * Попробуйте использовать двунаправленные рекуррентные слои (Bidirectional)\n",
        "    * Попробуйте использовать несколько рекуррентных слоев в сети\n",
        "    * Изменяйте размер мини-выборки для обучения\n",
        "    * Попробуйте применить предварительно обученные векторы (GloVe, Word2Vec и т.п.)\n",
        "* После подбора лучших гиперпараметров, обучите сеть еще раз на полном объеме данных без разделения на обучающий и проверочный наборы.\n",
        "\n",
        "Составьте отчет, который включает:\n",
        "\n",
        "* Место в соревновании, которое вам удалось достичь\n",
        "* Описание архитектуры нейронной сети и гиперпараметров обучения\n",
        "* Ссылку на ноутбук с кодом обучения\n",
        "* График качества обучения на обучающем и проверочном наборах данных"
      ]
    },
    {
      "metadata": {
        "id": "BanK632ZwX8K",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Подготовка окружения"
      ]
    },
    {
      "metadata": {
        "id": "V3Yus6d7vDWR",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.layers import Dense, Flatten, Embedding, Dropout, SpatialDropout1D, SpatialDropout2D, LSTM, SimpleRNN, GRU\n",
        "from tensorflow.keras.layers import CuDNNGRU, CuDNNLSTM\n",
        "from tensorflow.keras.optimizers import SGD\n",
        "from tensorflow.keras import utils\n",
        "from keras.preprocessing.text import Tokenizer\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "from google.colab import files\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "465OE0IvDBb-",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "embedding_dim = 200\n",
        "batch_size = 32\n",
        "maxlen = 100\n",
        "num_words=20000"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "qEXhLdkxwdfw",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Загрузка данных с Kaggle"
      ]
    },
    {
      "metadata": {
        "id": "gftsTedzwnm-",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Подключение Kaggle"
      ]
    },
    {
      "metadata": {
        "id": "ywRUgD-ZwgZG",
        "colab_type": "code",
        "outputId": "36403056-f69d-4424-cbe5-d36791c95771",
        "colab": {
          "resources": {
            "http://localhost:8080/nbextensions/google.colab/files.js": {
              "data": "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",
              "ok": true,
              "headers": [
                [
                  "content-type",
                  "application/javascript"
                ]
              ],
              "status": 200,
              "status_text": ""
            }
          },
          "base_uri": "https://localhost:8080/",
          "height": 88
        }
      },
      "cell_type": "code",
      "source": [
        "file = files.upload()\n",
        "!mkdir ~/.kaggle\n",
        "!mv kaggle.json ~/.kaggle"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "     <input type=\"file\" id=\"files-8e9d73cd-dbcd-457d-a015-b8683d75b12a\" name=\"files[]\" multiple disabled />\n",
              "     <output id=\"result-8e9d73cd-dbcd-457d-a015-b8683d75b12a\">\n",
              "      Upload widget is only available when the cell has been executed in the\n",
              "      current browser session. Please rerun this cell to enable.\n",
              "      </output>\n",
              "      <script src=\"/nbextensions/google.colab/files.js\"></script> "
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Saving kaggle.json to kaggle.json\n",
            "mkdir: cannot create directory ‘/root/.kaggle’: File exists\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "OYyPQk_twqkV",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Загрузка обучающего набора"
      ]
    },
    {
      "metadata": {
        "id": "qKrMzQnWws4P",
        "colab_type": "code",
        "outputId": "35c5f258-2bbf-437e-858d-469f8ca65751",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 255
        }
      },
      "cell_type": "code",
      "source": [
        "!kaggle competitions download -c neural-university-imdb-spring2019\n",
        "!unzip imdb_train.npz.zip\n",
        "!unzip imdb_test.npy.zip"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Warning: Your Kaggle API key is readable by other users on this system! To fix this, you can run 'chmod 600 /root/.kaggle/kaggle.json'\n",
            "Downloading imdb_sample_submission.csv to /content\n",
            "  0% 0.00/77.1k [00:00<?, ?B/s]\n",
            "100% 77.1k/77.1k [00:00<00:00, 72.7MB/s]\n",
            "Downloading imdb_test.npy.zip to /content\n",
            "  0% 0.00/3.31M [00:00<?, ?B/s]\n",
            "100% 3.31M/3.31M [00:00<00:00, 54.2MB/s]\n",
            "Downloading imdb_train.npz.zip to /content\n",
            " 38% 5.00M/13.3M [00:00<00:00, 29.9MB/s]\n",
            "100% 13.3M/13.3M [00:00<00:00, 52.7MB/s]\n",
            "Archive:  imdb_train.npz.zip\n",
            "  inflating: imdb_train.npz          \n",
            "Archive:  imdb_test.npy.zip\n",
            "  inflating: imdb_test.npy           \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "UCwSyc5rxbY3",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Загрузка наборов данных"
      ]
    },
    {
      "metadata": {
        "id": "j55MK93axgq2",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# обучающий набор\n",
        "data = np.load('imdb_train.npz') \n",
        "x_train = data['x'] \n",
        "y_train = data['y'] \n",
        "\n",
        "# тестовый набор\n",
        "x_test = np.load('imdb_test.npy')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "bTYubvBL00j0",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# IMDB from Keras for adjusting architecture\n",
        "from keras.datasets import imdb\n",
        "\n",
        "(ix_train, iy_train), (ix_test, iy_test) = imdb.load_data(path=\"imdb.npz\")\n",
        "\n",
        "lx_test = np.append(ix_train, ix_test)\n",
        "ly_test = np.append(iy_train, iy_test)\n",
        "\n",
        "lx_test = pad_sequences(lx_test, maxlen=maxlen)\n",
        "# ix_test = pad_sequences(ix_test, maxlen=maxlen)\n",
        "\n",
        "# print(\"Test on %s, %s\" % (lx_test.shape, ly_test.shape))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "kYWrKmttxvy1",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Подготовка данных для нейронной сети"
      ]
    },
    {
      "metadata": {
        "id": "XxeoTZ2-xyhY",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "x_train = pad_sequences(x_train, maxlen=maxlen)\n",
        "x_test = pad_sequences(x_test, maxlen=maxlen)\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "_GYZEADvLY1E",
        "colab_type": "code",
        "outputId": "56febda3-cc04-46fd-d195-d6d93a94a2ec",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "cell_type": "code",
      "source": [
        "print(\"Train on %s, %s\" % (x_train.shape, y_train.shape))\n",
        "print(\"Test on %s, %s\" % (lx_test.shape, ly_test.shape))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on (40000, 100), (40000,)\n",
            "Test on (50000, 100), (50000,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "ZYqc-dj54OD1",
        "colab_type": "code",
        "outputId": "e0060887-0b49-41ca-b2b9-87574db7e0ef",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "cell_type": "code",
      "source": [
        "# tmp_x_test = []\n",
        "# tmp_y_test = []\n",
        "\n",
        "# for i in range(len(lx_test)):\n",
        "#     if (lx_test[i] not in x_train):\n",
        "#         tmp_x_test.append(lx_test[i])\n",
        "#         tmp_y_test.append(ly_test[i])\n",
        "        \n",
        "print(len(tmp_x_test))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "xqJckHaa-7db",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Загрузка векторных представлений"
      ]
    },
    {
      "metadata": {
        "id": "gv3sYxh8_CK8",
        "colab_type": "code",
        "outputId": "1751147f-1de7-4432-cb0a-eb1499b77c1b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 357
        }
      },
      "cell_type": "code",
      "source": [
        "!wget http://nlp.stanford.edu/data/glove.6B.zip\n",
        "!unzip glove.6B.zip"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2019-04-07 16:46:12--  http://nlp.stanford.edu/data/glove.6B.zip\n",
            "Resolving nlp.stanford.edu (nlp.stanford.edu)... 171.64.67.140\n",
            "Connecting to nlp.stanford.edu (nlp.stanford.edu)|171.64.67.140|:80... connected.\n",
            "HTTP request sent, awaiting response... 302 Found\n",
            "Location: https://nlp.stanford.edu/data/glove.6B.zip [following]\n",
            "--2019-04-07 16:46:12--  https://nlp.stanford.edu/data/glove.6B.zip\n",
            "Connecting to nlp.stanford.edu (nlp.stanford.edu)|171.64.67.140|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 862182613 (822M) [application/zip]\n",
            "Saving to: ‘glove.6B.zip’\n",
            "\n",
            "glove.6B.zip        100%[===================>] 822.24M  20.9MB/s    in 45s     \n",
            "\n",
            "2019-04-07 16:46:58 (18.5 MB/s) - ‘glove.6B.zip’ saved [862182613/862182613]\n",
            "\n",
            "Archive:  glove.6B.zip\n",
            "  inflating: glove.6B.50d.txt        \n",
            "  inflating: glove.6B.100d.txt       \n",
            "  inflating: glove.6B.200d.txt       \n",
            "  inflating: glove.6B.300d.txt       \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "0TqXJmXLDYvf",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "from keras.datasets import imdb\n",
        "word_index = imdb.get_word_index()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "BLH_JyAj_llZ",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "embeddings_index = dict()\n",
        "with open('glove.6B.%dd.txt' % embedding_dim, 'r') as f:\n",
        "    for line in f:\n",
        "        values = line.split()\n",
        "        word = values[0]\n",
        "        word_vector = np.asarray(values[1:], dtype='float32')\n",
        "        embeddings_index[word] = word_vector\n",
        "        \n",
        "embedding_matrix = np.zeros((num_words, embedding_dim))\n",
        "for word, i in word_index.items():\n",
        "    if i < num_words - 3:\n",
        "        embedding_vector = embeddings_index.get(word)\n",
        "        if embedding_vector is not None:\n",
        "            embedding_matrix[i + 3] = embedding_vector"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "MNCkcWiEyHGl",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Подготовка нейронной сети"
      ]
    },
    {
      "metadata": {
        "id": "0tIVZg4OeEfX",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def plot_training(acc, val_acc):\n",
        "    plt.plot(acc, label='Train')\n",
        "    plt.plot(val_acc, label='Val')\n",
        "    plt.legend()\n",
        "    plt.show()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "nOBG7FgByPlp",
        "colab_type": "code",
        "outputId": "91fe529d-3107-4236-b31a-0d3b4cc0e1c9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 643
        }
      },
      "cell_type": "code",
      "source": [
        "#del model\n",
        "\n",
        "acc = []\n",
        "val_acc = []\n",
        "\n",
        "# Создание нейронной сети\n",
        "model = Sequential()\n",
        "model.add(Embedding(num_words, embedding_dim, input_length=maxlen))\n",
        "model.add(SpatialDropout1D(0.5))\n",
        "model.add(CuDNNGRU(200))\n",
        "model.add(Dense(1, activation='sigmoid'))\n",
        "\n",
        "# Загрузка векторных представлений\n",
        "model.layers[0].set_weights([embedding_matrix])\n",
        "model.layers[0].trainable = False\n",
        "\n",
        "model.summary()\n",
        "\n",
        "# Сборка\n",
        "model.compile(optimizer='rmsprop', \n",
        "              loss='binary_crossentropy', \n",
        "              metrics=['accuracy'])\n",
        "\n",
        "# Обучение\n",
        "history = model.fit(x_train, \n",
        "                    y_train, \n",
        "                    epochs=2,\n",
        "                    batch_size=batch_size,\n",
        "                    validation_split=0.1)\n",
        "\n",
        "# График\n",
        "acc.extend(history.history['acc'])\n",
        "val_acc.extend(history.history['val_acc'])\n",
        "plot_training(acc, val_acc)\n",
        "\n",
        "# Оценка качества\n",
        "model.evaluate(lx_test, ly_test, verbose = 1)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_31 (Embedding)     (None, 100, 200)          4000000   \n",
            "_________________________________________________________________\n",
            "spatial_dropout1d_31 (Spatia (None, 100, 200)          0         \n",
            "_________________________________________________________________\n",
            "cu_dnngru_23 (CuDNNGRU)      (None, 200)               241200    \n",
            "_________________________________________________________________\n",
            "dense_31 (Dense)             (None, 1)                 201       \n",
            "=================================================================\n",
            "Total params: 4,241,401\n",
            "Trainable params: 241,401\n",
            "Non-trainable params: 4,000,000\n",
            "_________________________________________________________________\n",
            "Train on 36000 samples, validate on 4000 samples\n",
            "Epoch 1/2\n",
            "36000/36000 [==============================] - 19s 520us/sample - loss: 0.5961 - acc: 0.6719 - val_loss: 0.4192 - val_acc: 0.8085\n",
            "Epoch 2/2\n",
            "36000/36000 [==============================] - 15s 418us/sample - loss: 0.4617 - acc: 0.7800 - val_loss: 0.3723 - val_acc: 0.8322\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAD8CAYAAACMwORRAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBo\ndHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzt3X10VNd57/GvpJGEBBohQKC3AQzG\nD2BjbJBsMNjYBhHHsZMmN3Zvm7SLxFlpEtza6Wp63dysm+UmTZq6KW3S29ZpQ72ymsTJrWNf59Zt\nEAYcbBJHEhhjGx4bY4PeAAF6Re8z5/5xjsQg9DJIMxrNzPNZi8XMmXNGezPip6O9n7NPmuM4GGOM\nSS7p8W6AMcaY6LNwN8aYJGThbowxScjC3RhjkpCFuzHGJCELd2OMSUK+SHYSkR3AOsABHlHV6rDX\ntgOfBIJAjao+GvbaAuAY8FFV3RfFdhtjjBnDuGfuIrIJWKaq64GHgO+EveYHvgTcrqobgZUisi7s\n8CeAE9FtsjHGmPFEMiyzGXgOQFWPAgVeqAP0eX9miYgPyAUuAIjI3UAHcCTajTbGGDO2SIZlioDa\nsOfN3rZ2Ve0Rkcdxz867gadV9W0RyQK+CnwE+NtIGtLc3DHhS2ULCnJpaema6OEJyfqcGlKtz6nW\nX5h8nwsL89JG2h7RmPswQ2/kncF/GbgOaAf2iMhq3FD/Z1VtFZGI3rSgIBefL2MCzXEVFuZN+NhE\nZX1ODanW51TrL8Smz5GEeyPumfqgEqDJe7wCOKGq5wBEZD+wFvgAkCEiDwNLgVtE5AFVfXO0LzLJ\nn1w0N3dM+PhEZH1ODanW51TrL0y+z6P9YIhkzH0X8HEAEVkDNKrqYEveB1aISI73vBx4R1U3qOo6\nVV0H/AfwhbGC3RhjTHSNe+auqgdEpFZEDgAhYLuIbAPaVPVZEXkC2CsiA8ABVd0f2yYbY4wZT9p0\nWfJ3MhOq9qtcarA+J79U6y9EZVhmxAlVu0LVGGOSkIW7McYkIQt3Y4xJQhOpczfGGHMVQk6Ii/1d\ntPd10NbbTntfB+29HbT1tTP3dD53LriD9LTonmtbuBtjzAT1B/tp7+ukva+dNi+w2/vavRAffNxJ\ne18HISc04ntk+7JZP+9Wcnw5I74+URbuxhgTxnEcugd63MDu7XCDuq/dC+4ON8T7OmjvbadroHvM\n98pM9+HP8rMoL0B+dh7+rDz8Wf5Lj7PzWF62iK62YNT7YeFujEkJwVCQjv7OoeGQ9qEzbS+wB4dL\n+jroDw2M+V4zfbn4s/0E8krxZ/nxZ88iP8uPPyvPC273cY5vBmlpI1YqXnqvrFy6iH75p4W7MSah\n9Qb7Lo1jjzCmPfi4s/8iDqNfTpOelo4/K4/imUUjn2V7ge3PziMzffpH5/RvoTEm5YRPQA6FdNgQ\nSVtvBx3e495g35jvNSMjG39WHgtmFg6dXfu9wM7P8g89npmZG/VJzXiycDfGTJn+0MAIk44dQ+Pb\n3aEuzne1jjkBCZBGGrMyZzIvZ+5lgR0e1u4QiZ/sjKwp7OH0YeFujJmU8AnIwcAOHw65+gnIPBbl\nBbywzrt8TNsL7rzMWWSkT3yJ8FRg4W6MGVH4BOSlipGRy/7Gm4DM9eWETUBePiwSPqa9sLiQc+c6\np6iHyc3C3ZgUM9YEZHjZ39VMQA6vEgl/fDUTkONVlpjIWbgbkwRGnIAc4XF7Xwc9wd4x3ys7I4v8\nLD8LZhZemnQcYUw72SYgk42FuzHTWH+wn/PdLUOTjpdf+Rg+IRnZBOTcnDlXVInkZ/uHJiD9WXnM\n8GVPYQ9NrFi4GzPFLk1AXqoSGekqyI7eDi4OjH37SV+6j/wrJiDDzrK9xzYBmXoiCncR2QGsAxzg\nEVWtDnttO/BJIAjUqOqjIuIDvo97/1Qf8Ceq+nK0G2/MdDI0ARk+fj3iWiORT0BeMydATlru5XXZ\nYWPakVwBaVLTuOEuIpuAZaq6XkRWADuB9d5rfuBLwLWqOiAiu0RkHe6Nsy+q6kYRuR74V+CWmPXC\nmBjqDfaNOY49+LizL9IJyAVXXPl4+VWQs8jMyARS885EJjoiOXPfDDwHoKpHRaRARPyq2g70eX9m\niUgnkAtcAP4N+LF3fDMwN+otN2YShk9AXjkscim8I52AnJ9fOBTSI41p2wSkmUqRhHsRUBv2vNnb\n1q6qPSLyOHAC6AaeVtW3vf36vb8fBX403hcpKMjF55v4mGBhYd6Ej01U1ucr9Qf7aevpoKWnjZbu\nNlp72mntaaO1u50W7+/BbcFxJiD92bNYMGses3PyKZiRz+wcP7Nn+CnIyXf/nuH+PSNzRrS7eZlU\n+5xTrb8Qmz5PZEJ1aIDPG5b5MnAd0A7sEZHVqnrYe307sAa4f7w3bWkZe+JoLKn4q2sq9dlxHHqC\nPWTMDPH+6aahqx7DL6oZPMuOdAJyYV4Zfu+MOn+EtUYimoDsgY6efjqGzmOiL5U+Z0i9/kJUbpA9\n4vZIwr0R90x9UAnQ5D1eAZxQ1XMAIrIfWAscFpGHcEP9t1Q1dt/9JmEFQ0E6+y9ePhwybK2RwccR\nTUBm5VGaV3JZxcjwqyBzfDk2AWlSQiThvgt4HHhSRNYAjao6+GPmfWCFiOSoajdQDrwgIkuAzwGb\nVLUnBu0201hfsG+E9UWuvAryaicg5+fPITs0w1trJKzsLytvaALSGOMaN9xV9YCI1IrIASAEbBeR\nbUCbqj4rIk8Ae0VkADigqvtF5Bu4k6gviMjgW21V1bHX5jTTVsgJ0dXfPW7FSKQTkP6svCsmIPO8\nwB5tAjIVf2U3ZqLSHGf0M6ep1NzcMeGGpOJ/+mj1eSA0cMVwyKWV/Nwx7ba+djr6Ogk6o98KbPAK\nyOHLrV7+eBb+LP+Er4C0zzn5pVp/ISpj7iOOM9oVqklocAJyKLB724eWXQ2/yUGkE5D+rDwWDq3m\n57/yKki7AtKYacfCPYGEnBAd3p3U6/oHqGs+M+paI/2hseewc3w55HsTkP6sy9fKzg8b07YJSGMS\nk4X7NDA4ATnWhTSRTkDmZc6ieOb8Ee6y7rcJSGNSiIV7jDiOM3QF5BWBPezmBz3BsQuKsjKyyPcm\nIAfPqIsL5pExkD0U2PnZfrsC0hgzxML9Kg1OQF42HHLZrcQ6IpqABLwlWAsmNAGZihNPxpjIWbhz\naQJyKKS9O9O0DZ+A7OvgYv84E5BpGfiz/ZdNQF4xpp3ttwlIY0xMJXW4h09AurcS67ziysernoCc\nWTzqXdZtAtIYM10kfLg3dDZR21pLw/nmK8a0O/o6I5qALJo5P+wu64NrjVxadyQvK48sm4A0xiSQ\nhA/37x35Aee6z1+2bXACsjB/0bDFoS49tglIY0wyS/hwf+iGT9Druwg9mUNlfzN8sV2C1RhjpruE\nD/eFeWVWOWKMMcPYmIQxxiQhC3djjElCFu7GGJOELNyNMSYJRTShKiI7gHWAAzyiqtVhr20HPgkE\ngRpVfVREMoGngEXe9k+p6okot90YY8woxj1zF5FNwDJVXQ88BHwn7DU/8CXgdlXdCKwUkXXA7wKt\n3ra/AL4Zi8YbY4wZWSTDMpuB5wBU9ShQ4IU6QJ/3Z5aI+IBc4IJ3zLPePruBDdFstDHGmLFFEu5F\nQHPY82ZvG97Nrx8HTgAngVdV9e3wY1Q1BDgikhXFdhtjjBnDRC5iGloVyzuD/zJwHdAO7BGR1WMd\nM5qCglx8vomvklhYmDfhYxOV9Tk1pFqfU62/EJs+RxLujXhn6p4SoMl7vAI4oarnAERkP7A27JjD\n3uRqmqr2jfVFWlrGXkp3LKl4har1OTWkWp9Trb8QlRtkj7g9kmGZXcDHAURkDdCoqoMteR9YISI5\n3vNy4B3vmAe8bfcDeyfUamOMMRMy7pm7qh4QkVoROQCEgO0isg1oU9VnReQJYK+IDAAHVHW/iGQA\nlSLyMtALbItdF4wxxgyX5jijr3c+lZqbOybcEPtVLjVYn5NfqvUXojIsM+Kcpl2haowxScjC3Rhj\nkpCFuzHGJCELd2OMSUIW7sYYk4Qs3I0xJglZuBtjTBKycDfGmCRk4W6MMUnIwt0YY5KQhbsxxiQh\nC3djjElCFu7GGJOELNyNMSYJWbgbY0wSsnA3xpgkFNENskVkB7AOcIBHVLXa214K/DBs1yXAY8A+\nYCeQDWQAX1TV2ug12xhjzFjGPXMXkU3AMlVdDzwEfGfwNVVtUNU7VfVOYAtwCnge+GPgWVW9Czfs\n/yIGbTfGGDOKSIZlNgPPAajqUaBARPwj7LcNeEZVO4FzwFxve4H33BhjTJj+gSAt7T0xee9IhmWK\ngPAhlWZvW/uw/T4DbPUe7wB+IyK/D/iBjZNspzHGJI3Wzl72Hmxg32sNdPcG2fGHG5g5IzOqXyOi\nMfdhrrgZq4isB46p6mDgfwn4qar+hYjcB/w18LGx3rSgIBefL2MCzXEVFuZN+NhEZX1ODanW52Tu\n7/H6Vn6+/wS/PFTPQNBhVk4mD265joWlBaSnj3if6wmLJNwbcc/UB5UATcP2uQ/YHfZ8A/AV73EV\n8A/jfZGWlq4ImjIyu2N6arA+J79k7G8o5HDonXNU1dTxdl0rAEVzcqmsCHDb9UWUlc6eVJ9H+2EY\nSbjvAh4HnhSRNUCjqg5vSQXwdNjz48CtuMM5FcA7V9tgY4xJZF09A7z8eiO7a+s51+aOq19/zRwq\nywPcsGQO6WnRPVMfbtxwV9UDIlIrIgeAELBdRLYBbar6rLdbMXA27LBvAN8XkQe9538UxTYbY8y0\ndaalixdr6tl/pIneviCZvnQ23VTClrVllBbOmrJ2RDTmrqqPDdt0eNjrq4Y9bwLunVzTjDEmMTiO\nw7FTrVRV13H4+DkcYPasLO5bv4hNN5UyKye6k6WRmMiEqjHGGNxSxl+/dYaq6nrqmzsBuKY4j8qK\nAOUyH19G/BYBsHA3xpir1NbZy95DDew71EB7Vz/paWlULJ9PZUWApSV+0mI8nh4JC3djjInQydMd\n7K6p49WjZxgIOuRm+/jgrQu5e00Zc/NnxLt5l7FwN8aYMYRCDq8dP0dVdR3qlTIumJPL1vIybruh\nmOysiV+fE0sW7sYYM4Lu3gH2v97E7pq6uJQyTpaFuzHGhDnb0sXu2npefr2JHq+U8Y7VJVSWT20p\n42RZuBtjUp7jOOipVqpq6njtHbeUMX9WFveuW8Smm0rIy82KdxOvmoW7MSZl9Q+E+M3RM1RV13Hq\nrFvKuLgoj60VAcqXx7eUcbIs3I0xKaftYh/7DjWw92A97V39pKVB+fL5bC0PsLR0epQyTpaFuzEm\nZZw600FVTR2vvuWWMuZk+7jnloXcvbaUefk58W5eVFm4G2OSWijkcPi4uyrjsVNeKWNBDlvKA2xY\nVcSMrOSMweTslTEm5XX3DvDy603srq2judUtZVy5uIDK8gCrls6d9qWMk2XhboxJKmdbu91VGV9v\npKcviC8jnTtWF7OlPEBZApUyTpaFuzEm4TmOw9t1reyqvryU8YNeKaM/AUsZJ8vC3RiTsIZKGWvq\nOHXGLWVc5JUyViR4KeNkWbgbYxJOu1fKuOdQA+0X+9xSRimksiLAtaX5SVHKOFkRhbuI7ADWAQ7w\niKpWe9tLgR+G7boEeExVfyQifwJ8EugHvjB4jDHGTNSpMx3srqnn12+dHipl/MAtATavKWPe7OQq\nZZysccNdRDYBy1R1vYisAHYC6wFUtQG409vPB+wDnheR64H/DpQDNwIfASzcjTFXLRRyOPyuuyrj\nYCnj/IIcKssD3HZDETnZNgAxkkj+VTYDzwGo6lERKRARv6q2D9tvG/CMqnaKyH3AT1V1ADjo/THG\nmIh19w7w8pEmXqyp52xrNwArFhVQWRHgxhQoZZysSMK9CKgNe97sbRse7p8BtnqPFwNBEfkvIBP4\nY1U9zBgKCnLx+Sa+LnJhYd6Ej01U1ufUkGp9Dqan8x+vvMeuV0/S1TNApi+dylsW8uE7lrK42B/v\n5sVELD7jifw+c8WPSxFZDxwLO5tPAzKADwIbgH8BKsZ605aWrgk0xVVYmEdzc8eEj09E1ufUkCp9\ndhyHd+rbeOn1Jn79RhOOA/kzs/jo7dew6ebSoVLGZPy3mOxnPNoPhkjCvRH3TH1QCdA0bJ/7gN1h\nz8/ghr0DvCwiiyNuqTEmZfQPhKg+5t5g+uQZN+AWLfBKGVekdinjZEUS7ruAx4EnRWQN0Kiqw3/M\nVABPhz3/T+BzwI9FZDlQF43GGmOSQ/vFPva91sDegw20eaWMa6WQB7YIhbMyrZQxCsYNd1U9ICK1\nInIACAHbRWQb0Kaqz3q7FQNnw475tYh8UER+5W3aHuV2G2MSUN3ZTqpq6vj1m2cYCIbIyc5ga0WA\nzWvLKJydkzLDUFMhzXGceLcBgObmjgk3JBW/IazPqSEZ+hxyHF4/fp6qmjqOnmwBYP7sHLaUl7Fh\nVfFlpYzJ0N+rFYUx9xF/zbECUWNMTHT3DvDKkSZ219ZztiWslLHcK2VMt6GXWLJwN8ZE1bnWbnbX\nuqsydve6qzJuvLGYyvIAgfmpsypjvFm4G2MmbbCUsaqmjoNvN+M44J+ZxQduWcidN5Xin5l6qzLG\nm4W7MWbCBoIhqo+eZVdNHSdPu+PGCxfMorI8wC0rFpDps1LGeLFwN8ZctfauPl461MCewVJGYM11\nhVSWl3FdYLaVMk4DFu7GmIjVe6WMv/JKGWdkuaWMd68tY76tyjitWLgbY8YUchxef/c8VdWXlzJu\nLi9j47BSRjN92KdijBlRT98Arxw5ze6aOs54pYzLF86msiLA6qXzrJRxmrNwN8Zc5lxbN3tqG3jp\ncCPdvQP4MtLYuKqYLeVlLFyQWitUJjILd2MMjuNwvKGNquo6asNLGSvcVRnzrZQx4Vi4G5PCBoIh\nqo+dpaq6jvcHSxnnz6KywkoZE52FuzEpqKOrj32vNbLnYD1tnW4p483L5rG1ImCljEnCwt2YFFLf\n3Mlur5Sxf8AtZawsD7C53EoZk42FuzFJLuQ4vHHCLWV88323lLFw9gy2rA2w8UYrZUxW9qkak6R6\n+gY48MZpqmrqOXPBvY3l8oWzqSwPsPpaK2VMdhGFu4jsANYBDvCIqlZ720uBH4btugR4TFV/5L2+\nADgGfFRV90Wx3caYUZxv6+HFg/X88rVGurxSxg2riqgsD1gpYwoZN9xFZBOwTFXXi8gKYCewHkBV\nG4A7vf18wD7g+bDDnwBORLfJxpjhHMfh3YZ2dtXUcVCbCTkO/txMPrxhMXetKbNSxhQUyZn7ZuA5\nAFU9KiIFIuJX1fZh+20DnlHVTgARuRvoAI5Esb3GmDADwRA1x85SVVPHe01uKWNgvrsq460r55Pp\ny4hzC028RBLuRUBt2PNmb9vwcP8MsBVARLKArwIfAf528s00xoTr6OrjJa+UsTWslLGyPIAstFJG\nM7EJ1Su+a0RkPXAs7Gz+MeCfVbVVRCJ604KCXHyTOMsoLEy9sUTrc2oI7/PJ0+38fP8J9tbU0Tfg\n3mD6w7cv4b6NSyieNzOOrYyeVP+MoyWScG/EPVMfVAI0DdvnPmB32PMPABki8jCwFLhFRB5Q1TdH\n+yItLV2RtXgEdlPd1JCqfT5ztp03TlygqvrUUCnjvPwZbCkPsHFVMbkzfOCEkuLfJlU/40neIHvE\n7ZGE+y7gceBJEVkDNKrq8JZUAE8PPlHVDYOPReQp4Kmxgt0Yc6XeviAvHHiPZ/ce57RXyigBd1XG\nm6yU0Yxj3HBX1QMiUisiB4AQsF1EtgFtqvqst1sxcDZ2zTQmdZxv62HPwXpe8koZM9LTuO0Gt5Rx\nUVHqDVmYiUlzHCfebQCgubljwg2xX+VSQzL32XEc3m1sd1dl9EoZ83Iz+dCGJdwq88iflR3vJk6J\nZP6MRxOFYZkRf4WzK1SNiaOBYIgaPUtVdT3vNbn1CGWFs6isKGPdygWUFM9OubAz0WHhbkwcdHb3\n89Jr7g2mWzp6SQNuunYelRUBllspo4kCC3djplDDuYu8WFPHgTdO0zcQIjsrgy1ry9hcXsaCgtx4\nN88kEQt3Y2Is5Di8+d4FqqrreOO9C4BXyri2jI03lriljMZEmX1XGRMjvX1BDrzp3mC66bxbynhd\nwF2V8eZlVspoYsvC3Zgou9B+aVXGiz1uKeP664uorChjcZE/3s0zKcLC3ZgoebehjaqaOmqOuaWM\ns3Iyuf+2xdy1ppTZKVLKaKYPC3djJmEgGKJWm6mqqeNE42Ap40wqywOsu36Brcpo4sbC3ZgJGLWU\nsbyM5YsKrJTRxJ2FuzFXofHcRXbX1nPgSJNbypiZwea1ZWxZW8aCOVbKaKYPC3djxuF4pYy7aup4\n44RbyjjXP4Mt5WXcfmMxuTMy49xCY65k4W7MKHr7g/zqjdNUhZUyLivLd0sZr5tHRnp6nFtozOgs\n3I0Z5kJ7D3sONvDSaw1hpYwL2FIe4JpiK2U0icHC3RjPu41tVFVfXsp4322LuevmUgryrJTRJBYL\nd5PSgiGvlLG6jne9UsbSwVLGlQvIyrRSRpOYLNxNSurs7mf/4UZePFjPhfZeAFYvnUtlRYAVVspo\nkkBE4S4iO4B1gAM8oqrV3vZS4Idhuy7BvTn2T4Hv494/1Qf8iaq+HMV2GzMhTecvsrumnlfeaKKv\n3ytlXOOuylhkpYwmiYwb7iKyCVimqutFZAWwE1gPoKoNwJ3efj5gH/A88HvARVXdKCLXA/8K3BKL\nDhgzHsdxePP9C1RV13PkxHkA5vqz2bwxwB2rrZTRJKdIztw3A88BqOpRESkQEb+qtg/bbxvwjKp2\nisi/AT/2tjcDc6PVYGMi1dsf5FdvnmZ3TT2N5y4CcG1ZPlutlNGkgEjCvQioDXve7G0bHu6fAbYC\nqGo/0O9tfxT40eSaaUzkWjp62XOwnn2HLpUyrrt+AZVWymhSyEQmVK+YaRKR9cCx4WfzIrIdWAPc\nP96bFhTk4pvEIkuFhal3V3jr8+XePtXC8788wcuHGwiGHPJys3hwy3Xce9ti5ubnTGEroyvVPudU\n6y/Eps+RhHsj7pn6oBKgadg+9wG7wzeIyEO4of5b3pn8mFpauiJoysjsjumpYaQ+B0MhDr59jqrq\nOo43tAFQOm8mlRWXShlDfQMJ+2+Vap9zqvUXJt/n0X4wRBLuu4DHgSdFZA3QqKrDW1IBPD34RESW\nAJ8DNqlqz4RabMwYLvb088vDjbxYe6mU8UavlHGllTIaM364q+oBEakVkQNACNguItuANlV91tut\nGDgbdthncCdRXxCRwW1bVbUvai03KanpvLsq4ytH3FLGrMx07lpTypa1ZRTPnRnv5hkzbaQ5jhPv\nNgDQ3Nwx4YbYr3LJzXEc3nq/hZdeb6Lm6BkA5viz2by2jDtWlzAziUsZU+lzhtTrL0RlWGbEX1Pt\nClUzbfWFlTI2DJYyluZTWRFgjZUyGjMmC3cz7QyWMr70WiOd3f1uKePKBTxQKRTk2LesMZGw/ylm\n2nivqZ2q6jqqj50lGHKYOcPHh9Yv4u41ZRTkZafkr+zGTJSFu4mrYCjEobfPsaumjuP1biljybyZ\nVJaXse76IrJtVUZjJsTC3cRFV08/vzzcxIu1dZz3ShlXLZlLZUUZ1y+eY6WMxkyShbuZUqcvdLG7\npo5Xjpymtz/oljLeXMqWcitlNCaaLNxNzDmOw1snW6iqruP1d91VGQvysvnwhsXcvrqEWTnJW8po\nTLxYuJuY6esP8uu3zlBVXTdUyri01E9leYA11xXiy7BSRmNixcLdRF1LRy97DzWw71DDUCnjrSsX\nsKW8jKUl+fFunjEpwcLdRM37p91Sxt8cvbyU8a6bS5njnxHv5hmTUizczaQMljJW1dTxjlfKWDw3\nl8qKAOutlNGYuLFwNxNyqZSxnvPt7sKfVspozPRh4W6uypkLXeyuqeflI01uKaMvnTtvdldlLJln\npYzGTBcW7mZcjuNwNKyU0cEtZbx/w2LusFJGY6YlC3czqqFSxpo6Gpq9UsYSv7cqo5UyGjOdWbib\nK7R29rL3YAN7vVLG9LQ0blkxn8ryAEtLrZTRmEQQUbiLyA5gHeAAj6hqtbe9FPhh2K5LgMeA/wM8\nBSwCgsCnVPVE9JptYuHk6Q52Vdfxm6NnhkoZP7huIZvXlFkpozEJZtxwF5FNwDJVXS8iK4CdwHoA\nVW0A7vT28wH7gOeB3wVaVfUTIrIV+Cbw27HogJmcUMjh0DvNVFXX8XZYKeOW8gC3XV9EdpaVMhqT\niCI5c98MPAegqkdFpEBE/KraPmy/bcAzqtopIpuBH3jbd+P+QDDTSFfPAPtfd28wfa7NLWW84Zo5\nVFYEuP6aOaRbKaMxCS2ScC8CasOeN3vbhof7Z4CtYcc0A6hqSEQcEcka6wbZBQW5+HwTP0ssLMyb\n8LGJaiJ9bjzXyc/3n+DF6lN09wbJyszgnvWLuX/jNSws8segldFln3PyS7X+Qmz6PJEJ1StO6URk\nPXBshLP5UY8ZrqWlawJNcaXiHXqups+O43DsVCtV1XUcPn5uqJTx3nWL2HRT6VAp43T/N7TPOfml\nWn8hKjfIHnF7JOHeiHsmPqgEaBq2z324wy/DjzksIplA2lhn7SY2+gcGV2Wsp765E4Briv1srQiw\nVqyU0ZhkFkm47wIeB54UkTVAo6oO/zFTATw97JgHgF8A9wN7o9BWE6G2TndVxr2HGujoslJGY1LR\nuOGuqgdEpFZEDgAhYLuIbAPaVPVZb7di4GzYYT8BKkXkZaAXd7LVxNjJ0x1U1dTx6ltWymhMqoto\nzF1VHxu26fCw11cNex4EPjW5pplIuKWM7qqMb9e1AlA0J5fK8jJuu6HYShmNSVF2hWqCutjdz67f\nnGJ3WCnj9dfMobI8wA1LrJTRmFRn4Z5gzrZ0sbu2nleOnKa7d4BMXzqbbiphy9oySgtnxbt5xphp\nwsI9ATiOg55qpaqmjtfecUsZ5/hncO+6hdyxuoS83Kx4N9EYM81YuE9j/QNBXn3rLFU1ddSdHSxl\nzKOyIsAHNy6l5cLFOLfQGDNdWbhPQ20X+9h7sJ59hxpo90oZK5bPp7IiwNISP2lpaVajbowZk4X7\nNHLqTAdV1XW8evQMA0GH3GwrC+dvAAAK+0lEQVQf99zqljLOzbdSRmOmo+9+dweqR7lw4Tw9PT2U\nlJTi9+fzjW88MeZxL7zwc2bOnMXHP/7hmLTLwj3OQiGH146fo6q6DvVKGRcMlTIWMSPLPiJjprM/\n/MMvAm5YnzjxLg8//GhEx9177/2xbJaFe7x09w6w//UmdtfUXSplXFxAZUWAG5bMtVJGYxLYwYM1\nPP30v9HV1cXDD3+RQ4dq2bfvRUKhEOvXb+DTn/4s3//+k8yePZubb17Fzp1PkZaWzsmT73HnnZv5\n9Kc/O+k2WLhPsbOt3eyuqePl15vo6QuS6UvnjtUlbCkvo8xKGY2ZlJ/uOU71sbPj73gVKpbP58G7\nr73q49599zg//vHPyMrK4tChWv7hH/6F9PR0HnzwI/z2b//uZfu+9dab/OhHzxAKhXjggfst3BOF\n4zi8XdfKrupLpYz5s7K8VRmtlNGYZHTttcvIynL/b8+YMYOHH/4sGRkZtLa20t5++QK6IsuZMSO6\n82oW7jHUPxDiN0fPUFVdxymvlHFxkVvKWLF8vlW8GBNlD9597YTOsmMhM9NdSvv06SZ+8pMfsnPn\nD8nNzeX3fu/BK/bNyIj+MiEW7jHQdrGPfYca2HuwnvauftLSoFwKqawIcG1pPmk2nm5MymhtbaWg\noIDc3FxUj3H69Gn6+/tj/nUt3KPo1JlLqzIOBB1ysn3cc8tC7l5byrz8nHg3zxgTB8uWXUdOTi6f\n//ynWbXqJj7ykY/x7W9/ixtvXB3Tr5vmOE5Mv0Ckmps7JtyQeN69JRRyOHzcXZXx2CmvlLEghy3l\nATasil0po92xJjWkWp9Trb8QlTsxjTgUYGfuE9TdO8DLR5p4saaes63dAKxcXEBleYBVS62U0RgT\nXxGFu4jsANYBDvCIqlaHvRYAfgxkAQdV9XMiMgv4AVAAZAOPq+ovot34eGhu7ebF2nr2v95Id28Q\nX0Y6d6wuZsvaAGXzrZTRGDM9jBvuIrIJWKaq60VkBbATWB+2y7eBb6vqsyLyv0VkIfBhQFX1z0Sk\nBNgDLI9B+6fEYCljVU09h95pxnHcUsZ7bnVLGf1WymiMmWYiOXPfDDwHoKpHRaRARPyq2i4i6cDt\nwO94r28HEJFzwI3e8QXAuai3fAoMlTLW1HHqjFvKuKgoj61WymiMmeYiCfcioDbsebO3rR0oBDqA\nHd7Ns/er6p+p6tMisk1EjuOG+4ei3O6YavdKGfccaqD9Yh9pabBWCqksD7CszEoZjTHT30QmVNOG\nPS4F/g54H/gPEfkQbqCfUtV7RGQ18H2gfKw3LSjIxeebeCF/YWHehI8d9F5jG8//8gQvHaqnfyBE\n7gwfv7VpKfdtXMKCObmTfv9oi0afE431OfmlWn8hNn2OJNwbcc/UB5UATd7jc8BJVX0XQEReBK4H\nrgF+AaCqh0WkREQyvBtnj6ilpWsCzXdNppQoFHI4/K67KuNgKeP8ghwqywPcdkMROdk+CAanXXmW\nlYylhlTrcyL29w/+4FN88Yt/yvLlK4a2/dM//T35+bP5nd/55GX7HjxYw89+9lO+/vW/GtoWhVLI\nEbdHEu67gMeBJ72hl0ZV7QBQ1QEROSEiy1T1HWAtbuVMELgVeEZEFgGdYwV7PHT3DvDKkSZ2h5Uy\nrljkrsp4o5UyGmMiVFn5Afbsqbos3Pft28N3v/tPcWxVBOGuqgdEpFZEDgAhYLuIbAPaVPVZ4FHg\nKW9y9QjwcyAX2CkiL3lf43Ox6sDVGqmU8fYbi6kst1JGY8zV27x5K5///EN84Qt/BMCxY0cpLCzk\n/fff4ytf+R9kZmaSl5fHn//5X05puyIac1fVx4ZtOhz22nFg47DXO4ErV8eJE8dxeKe+jarqOg4O\nljLOzOIDtyzkzptLrZTRmCTxs+P/j0Nnj0T1PW+ev4qPXXvfqK8XFMyhpKSUt956g5Urb2DPnioq\nK++ho6ODr37165SUlPK1r/0vXn31V+TmTt3cXVJfoToQHFyVsZ6TZ9wxrYULZnmljAvI9FkpozFm\n8ior7+HFF6tYufIGXnnll/zjP+7k+PG3+da3vk4wGKSxsYG1ayss3Cer/WIf+15rYO/BBtoGSxmv\nc1dltFJGY5LXx669b8yz7FjZtOkufvCDnVRWfoBAYCF+v59vfvNrPPHE37J48TX8zd98a8rblFTh\nXn+2k101dfz6zTMMBEPkZGewtSLA5rVlFM62VRmNMbGRmzuTpUuX8YMf/CuVlfcAcPFiJwsWFNHR\n0cHBg7UsXbpsStuU8OEechx+89Zp/n332xw92QLA/Nk5bCkvY8OqYreU0RhjYqyy8h6+/vWv8tWv\nfg2Aj33sAT7/+YcIBBbyiU/8Pjt3fo/PfvYLU9aehF/y929+8hpvvHcB8EoZy71SxvTkHnpJxHrg\nybI+J79U6y/Ykr+jCsyfRcn8PG5bOZ+FC1LvyjZjjBlJwof7A3ddm5I/7Y0xZixWC2iMMUnIwt0Y\nY5KQhbsxxiQhC3djjElCFu7GGJOELNyNMSYJWbgbY0wSsnA3xpgkNG2WHzDGGBM9duZujDFJyMLd\nGGOSkIW7McYkIQt3Y4xJQhbuxhiThCzcjTEmCSXUeu4isgNYBzjAI6paHfbaFuAbQBB4QVW/Fp9W\nRtc4fb4L+CZunxX4jKqG4tLQKBqrz2H7fBNYr6p3TnHzYmKczzkA/BjIAg6q6ufi08roGqfP24FP\n4n5v16jqo/FpZXSJyA3A/wV2qOrfD3stqhmWMGfuIrIJWKaq64GHgO8M2+U7wH8DNgBbRWTlFDcx\n6iLo8/eAj6vqBiAPuGeKmxh1EfQZ77O9Y6rbFisR9PnbwLdV9RYgKCILp7qN0TZWn0XED3wJuF1V\nNwIrRWRdfFoaPSIyE/gu8OIou0Q1wxIm3IHNwHMAqnoUKPC+CRCRJcAFVa3zzlxf8PZPdKP22bNW\nVeu9x83A3CluXyyM12dww+5/TnXDYmis7+104Hbgee/17ap6Kl4NjaKxPuc+788sEfEBucCFuLQy\nunqBe4HG4S/EIsMSKdyLcANsULO3baTXzgLFU9SuWBqrz6hqO4CIFANbcb8hEt2YfRaRbcBLwPtT\n2qrYGqvPhUAHsENEXvaGo5LBqH1W1R7gceAEcBJ4VVXfnvIWRpmqDqhq9ygvRz3DEinchxvxjt8R\nvJbIruiXiMwHfg58QVXPT32TYm6ozyIyB/gU7pl7Mksb9rgU+DtgE3CziHwoLq2KrfDP2Q98GbgO\nuAa4VURWx6thcTLpDEukcG8k7AwOKAGaRnmtlBF+9UlAY/V58D/BfwJfUdVdU9y2WBmrz3fjnsnu\nB54F1niTcolurD6fA06q6ruqGsQdr71+itsXC2P1eQVwQlXPqWof7ue9dorbN9WinmGJFO67gI8D\niMgaoFFVOwBU9X3ALyKLvTG6+7z9E92offZ8G3fW/b/i0bgYGetz/ndVXamq64CP4laOfDF+TY2a\nsfo8AJwQkWXevmtxK6MS3Vjf2+8DK0Qkx3teDrwz5S2cQrHIsIRaFVJE/hK3SiIEbAduBtpU9VkR\nuQP4lrfrM6r613FqZlSN1mfgF0AL8Kuw3X+kqt+b8kZG2Vifc9g+i4GnkqgUcqzv7WuBp3BPxo4A\nn0+Sktex+vwHuENwA8ABVf3T+LU0OkRkLe4J2WKgH2jAnSh/LxYZllDhbowxJjKJNCxjjDEmQhbu\nxhiThCzcjTEmCVm4G2NMErJwN8aYJGThbowxScjC3RhjkpCFuzHGJKH/D9XL0fLWCzkkAAAAAElF\nTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "50000/50000 [==============================] - 15s 300us/sample - loss: 0.3750 - acc: 0.8335\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.37498897178649904, 0.83354]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 102
        }
      ]
    },
    {
      "metadata": {
        "id": "xR56qMYIehHx",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Обучение с размороженными весами"
      ]
    },
    {
      "metadata": {
        "id": "glrTrCWReglC",
        "colab_type": "code",
        "outputId": "993e2246-b664-4d20-924f-9a7b3f8effd0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 775
        }
      },
      "cell_type": "code",
      "source": [
        "model.layers[0].trainable = True\n",
        "\n",
        "model.compile(optimizer='rmsprop', \n",
        "              loss='binary_crossentropy', \n",
        "              metrics=['accuracy'])\n",
        "model.summary()\n",
        "\n",
        "# Обучение\n",
        "history = model.fit(x_train, \n",
        "                    y_train, \n",
        "                    epochs=6,\n",
        "                    batch_size=batch_size,\n",
        "                    validation_split=0.1)\n",
        "\n",
        "# График\n",
        "acc.extend(history.history['acc'])\n",
        "val_acc.extend(history.history['val_acc'])\n",
        "plot_training(acc, val_acc)\n",
        "\n",
        "# Оценка качества\n",
        "model.evaluate(lx_test, ly_test, verbose = 1)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_31 (Embedding)     (None, 100, 200)          4000000   \n",
            "_________________________________________________________________\n",
            "spatial_dropout1d_31 (Spatia (None, 100, 200)          0         \n",
            "_________________________________________________________________\n",
            "cu_dnngru_23 (CuDNNGRU)      (None, 200)               241200    \n",
            "_________________________________________________________________\n",
            "dense_31 (Dense)             (None, 1)                 201       \n",
            "=================================================================\n",
            "Total params: 4,241,401\n",
            "Trainable params: 4,241,401\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "Train on 36000 samples, validate on 4000 samples\n",
            "Epoch 1/6\n",
            "36000/36000 [==============================] - 13s 350us/sample - loss: 0.3807 - acc: 0.8272 - val_loss: 0.3027 - val_acc: 0.8650\n",
            "Epoch 2/6\n",
            "36000/36000 [==============================] - 11s 293us/sample - loss: 0.3189 - acc: 0.8602 - val_loss: 0.2806 - val_acc: 0.8755\n",
            "Epoch 3/6\n",
            "36000/36000 [==============================] - 14s 401us/sample - loss: 0.2778 - acc: 0.8811 - val_loss: 0.2947 - val_acc: 0.8765\n",
            "Epoch 4/6\n",
            "36000/36000 [==============================] - 19s 528us/sample - loss: 0.2524 - acc: 0.8945 - val_loss: 0.2681 - val_acc: 0.8810\n",
            "Epoch 5/6\n",
            "36000/36000 [==============================] - 19s 533us/sample - loss: 0.2294 - acc: 0.9049 - val_loss: 0.2640 - val_acc: 0.8910\n",
            "Epoch 6/6\n",
            "36000/36000 [==============================] - 19s 530us/sample - loss: 0.2082 - acc: 0.9148 - val_loss: 0.2571 - val_acc: 0.8900\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAD4CAYAAAAXUaZHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nCBGB/j+0JrrAWmV3BwAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "50000/50000 [==============================] - 27s 536us/sample - loss: 0.1867 - acc: 0.9265\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.186725788718462, 0.92648]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 103
        }
      ]
    },
    {
      "metadata": {
        "id": "wGYiEkiuhiPE",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Дообучение SGD"
      ]
    },
    {
      "metadata": {
        "id": "3BL4XyJyKMnK",
        "colab_type": "code",
        "outputId": "1814365e-132f-479f-935c-71e2850c7cd3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 707
        }
      },
      "cell_type": "code",
      "source": [
        "# model.layers[0].trainable = False\n",
        "\n",
        "opt = SGD(lr=0.0001)\n",
        "model.compile(loss=\"binary_crossentropy\", optimizer=opt, metrics=[\"accuracy\"])\n",
        "model.summary()\n",
        "\n",
        "history = model.fit(x_train, \n",
        "                    y_train, \n",
        "                    epochs=4,\n",
        "                    batch_size=16,\n",
        "                    validation_split=0.1)\n",
        "\n",
        "acc.extend(history.history['acc'])\n",
        "val_acc.extend(history.history['val_acc'])\n",
        "plot_training(acc, val_acc)\n",
        "\n",
        "model.evaluate(lx_test, ly_test, verbose = 1)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_31 (Embedding)     (None, 100, 200)          4000000   \n",
            "_________________________________________________________________\n",
            "spatial_dropout1d_31 (Spatia (None, 100, 200)          0         \n",
            "_________________________________________________________________\n",
            "cu_dnngru_23 (CuDNNGRU)      (None, 200)               241200    \n",
            "_________________________________________________________________\n",
            "dense_31 (Dense)             (None, 1)                 201       \n",
            "=================================================================\n",
            "Total params: 4,241,401\n",
            "Trainable params: 241,401\n",
            "Non-trainable params: 4,000,000\n",
            "_________________________________________________________________\n",
            "Train on 36000 samples, validate on 4000 samples\n",
            "Epoch 1/4\n",
            "36000/36000 [==============================] - 97s 3ms/sample - loss: 0.1858 - acc: 0.9271 - val_loss: 0.2571 - val_acc: 0.8917\n",
            "Epoch 2/4\n",
            "36000/36000 [==============================] - 58s 2ms/sample - loss: 0.1849 - acc: 0.9275 - val_loss: 0.2574 - val_acc: 0.8913\n",
            "Epoch 3/4\n",
            "36000/36000 [==============================] - 48s 1ms/sample - loss: 0.1853 - acc: 0.9264 - val_loss: 0.2578 - val_acc: 0.8907\n",
            "Epoch 4/4\n",
            "36000/36000 [==============================] - 48s 1ms/sample - loss: 0.1855 - acc: 0.9265 - val_loss: 0.2581 - val_acc: 0.8905\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAD4CAYAAAAXUaZHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nknAXQggLknAXQggLknAXQggL+v834OjCdNJoLwAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "50000/50000 [==============================] - 23s 465us/sample - loss: 0.1849 - acc: 0.9282\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.18488824370741844, 0.9282]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 105
        }
      ]
    },
    {
      "metadata": {
        "id": "ZJE0aUOMOmUO",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "plot_training(acc, val_acc)\n",
        "model.evaluate(lx_test, ly_test, verbose = 1)\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "v6sYUe16yJ9B",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Проверка на тестовом наборе"
      ]
    },
    {
      "metadata": {
        "id": "bESrUJht1H6u",
        "colab_type": "code",
        "outputId": "0ea347b1-89f7-4670-a8c7-9d28ed3d2e7b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "cell_type": "code",
      "source": [
        "model.evaluate(lx_test, ly_test, verbose = 1)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "50000/50000 [==============================] - 12s 241us/sample - loss: 0.1841 - acc: 0.9288\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.1840726106572151, 0.9288]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 56
        }
      ]
    }
  ]
}