aspect_based_sent_analysis (1).ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "aspect_based_sent_analysis.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    },
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "670c65607459432fa40636b4b5d06c74": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_4187f6e244214bbebabbb67f8eb0a5d9",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_37ba1d85ee46420b867d1e40d3384bd6",
              "IPY_MODEL_717fc1379bcf42daa69ab83b01821aa4"
            ]
          }
        },
        "4187f6e244214bbebabbb67f8eb0a5d9": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "37ba1d85ee46420b867d1e40d3384bd6": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_db2e965d65b04f0c8ccd99ea1ee0367d",
            "_dom_classes": [],
            "description": "100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 500002,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 500002,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_8027c0cd01ec407695b2c4c109231d79"
          }
        },
        "717fc1379bcf42daa69ab83b01821aa4": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_1e0cc0bb41b8469e878fc53eecf2fc72",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 500002/500002 [01:26&lt;00:00, 5798.78it/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_aa09a8822b5e4719971ec01bbd6b4b53"
          }
        },
        "db2e965d65b04f0c8ccd99ea1ee0367d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "8027c0cd01ec407695b2c4c109231d79": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "1e0cc0bb41b8469e878fc53eecf2fc72": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "aa09a8822b5e4719971ec01bbd6b4b53": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        }
      }
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1TenNKsQT_we",
        "outputId": "9068b49d-0266-407a-db2d-34224a04ab74"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mounted at /content/drive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QkFXlDJGwELN"
      },
      "source": [
        "!pip install razdel\n",
        "!pip install navec\n",
        "!pip install slovnet\n",
        "!pip install ipymarkup\n",
        "!pip install pymorphy2\n",
        "!wget https://storage.yandexcloud.net/natasha-navec/packs/navec_hudlit_v1_12B_500K_300d_100q.tar\n",
        "!wget https://storage.yandexcloud.net/natasha-slovnet/packs/slovnet_syntax_news_v1.tar\n",
        "!wget https://storage.yandexcloud.net/natasha-navec/packs/navec_news_v1_1B_250K_300d_100q.tar\n",
        "\n",
        "!wget https://github.com/dkulagin/kartaslov/raw/master/dataset/kartaslovsent/kartaslovsent.csv\n",
        "\n",
        "!pip3 install colorama"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AQl_rTAFt88z"
      },
      "source": [
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "from torch.utils.data import Dataset, DataLoader\n",
        "\n",
        "from razdel import tokenize, sentenize\n",
        "from navec import Navec\n",
        "from slovnet.model.emb import NavecEmbedding\n",
        "from slovnet import Syntax\n",
        "\n",
        "import os\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "from tqdm.notebook import tqdm\n",
        "\n",
        "from sklearn.metrics import accuracy_score\n",
        "\n",
        "import pymorphy2\n",
        "\n",
        "from scipy.spatial.distance import cosine\n",
        "from sklearn.neighbors import  NearestNeighbors\n",
        "from sklearn.neighbors import KNeighborsClassifier as KNN\n",
        "\n",
        "from collections import defaultdict\n",
        "import re\n",
        "\n",
        "from json import dumps, load\n",
        "\n",
        "from colorama import Back\n",
        "\n",
        "\n",
        "np.random.seed(42)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        },
        "id": "i-nfeUXm0h1i",
        "outputId": "355e2718-8131-42ae-9fe5-2fef55c1b153"
      },
      "source": [
        "data = pd.read_csv(PATH)\n",
        "def count_words(row):\n",
        "    return len(row.split())\n",
        "\n",
        "r = data['text'].apply(count_words)\n",
        "plt.hist(r, bins=350)\n",
        "plt.title('Распределение числа слов в отзывах.')\n",
        "plt.xlabel('Число слов')\n",
        "plt.xlim((0, 700))  # лимит на 700 словах, но максимум слов порядка 1000\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6QqlVOLOzItf"
      },
      "source": [
        "path = '/content/navec_hudlit_v1_12B_500K_300d_100q.tar'\n",
        "navec = Navec.load(path)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9X4b-Vq5w6XE"
      },
      "source": [
        "class TextDataset(Dataset):    \n",
        "    def __init__(self):\n",
        "        self.targets = {\"neg\": 0, \"neu\": 1, \"pos\": 2}\n",
        "        self.max_len = 700\n",
        "        self.data = pd.read_csv(PATH).iloc[:70000]\n",
        "        #self.data = self.data.filter(lambda x: len(x['text'].split()) > 50)\n",
        "        \n",
        "    def __len__(self):\n",
        "        return len(self.data)\n",
        "    \n",
        "    def __getitem__(self, idx):\n",
        "        sample, target = self.data.iloc[idx]\n",
        "        sample = [_.text for _ in tokenize(sample)]\n",
        "        if len(sample) > self.max_len:\n",
        "          sample = sample[:self.max_len]\n",
        "        sample = sample + [\"<PAD>\"]*(self.max_len-len(sample))\n",
        "        sample = [navec.vocab.get(x, navec.vocab.unk_id) for x in sample]\n",
        "        target = self.targets[target]\n",
        "        return torch.LongTensor(sample), torch.LongTensor([target])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GzLNStpPz6lU"
      },
      "source": [
        "batch_size = 200\n",
        "dataset = TextDataset()\n",
        "train_size = int(0.8 * len(dataset))\n",
        "test_size = len(dataset) - train_size\n",
        "train_set, test_set = torch.utils.data.random_split(dataset, [train_size, test_size])\n",
        "train_iterator = DataLoader(train_set, batch_size=batch_size, shuffle=True)\n",
        "test_iterator = DataLoader(test_set, batch_size=batch_size)\n",
        "device = torch.device(\"cuda\") if torch.cuda.is_available() else torch.device(\"cpu\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "huLIP8yg8Bbg"
      },
      "source": [
        "class RNN(nn.Module):\n",
        "    def __init__(self, input_dim, embedding_dim, hidden_dim, output_dim, num_layers, dropout, bidirectional=False):\n",
        "        super().__init__()\n",
        "        self.num_layers = num_layers\n",
        "        self.hidden_dim = hidden_dim\n",
        "        self.embedding = NavecEmbedding(navec)\n",
        "        self.bi = bidirectional\n",
        "        self.rnn = nn.GRU(embedding_dim, hidden_dim, dropout = dropout, num_layers=self.num_layers, bidirectional=self.bi)\n",
        "        self.fc = nn.Linear(hidden_dim*self.num_layers, output_dim)\n",
        "        self.dropout = nn.Dropout(dropout)\n",
        "        self.softmax = nn.Softmax(1)\n",
        "\n",
        "    def init_hidden(self, batch_size):\n",
        "        mult = 2 if self.bi else 1\n",
        "        return torch.zeros(mult * self.num_layers, batch_size, self.hidden_dim).to(device)\n",
        "        \n",
        "    def forward(self, x):\n",
        "        #text,shape = [sent len, batch size]\n",
        "        \n",
        "        embedded = self.embedding(x)  # Делаем эмбеддинг\n",
        "        #embedded.shape = [sent len, batch size, emb dim]\n",
        "        #print(embedded.shape)\n",
        "        output, self.hidden =  self.rnn.forward(embedded, self.hidden)   # Прогоняем через Rnn\n",
        "        #hidden.shape = [1, batch size, hid dim]\n",
        "        #print(self.hidden)\n",
        "        x = self.fc(torch.cat((self.hidden[-2,:,:], self.hidden[-1,:,:]), dim = 1))\n",
        "        x = self.dropout(x)\n",
        "        return x.squeeze(0)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "MaXHWVGK9Tdf",
        "outputId": "7af83e55-f844-499d-fdd2-bb24253873dc"
      },
      "source": [
        "import torch.optim as optim\n",
        "\n",
        "INP_DIM = 700\n",
        "EMB_DIM = 300\n",
        "HID_DIM = 512\n",
        "NUM_LAYERS = 2\n",
        "DROPOUT = 0.75\n",
        "OUT_DIM = 3\n",
        "N_EPOCH = 15\n",
        "BIDIRECTIONAL = True\n",
        "\n",
        "\n",
        "model = RNN(INP_DIM, EMB_DIM, HID_DIM, OUT_DIM, NUM_LAYERS, DROPOUT, BIDIRECTIONAL).to(device)\n",
        "criterion = nn.CrossEntropyLoss().to(device)\n",
        "optimizer = optim.Adam(model.parameters(), lr=0.001)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/slovnet/model/emb.py:46: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at  /pytorch/torch/csrc/utils/tensor_numpy.cpp:180.)\n",
            "  torch.from_numpy(navec.pq.indexes),\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ho5dBjITI2MC",
        "outputId": "0629285c-84ea-46da-9dc8-41bab1c6ee92"
      },
      "source": [
        "losses_train = []\n",
        "losses_test = []\n",
        "\n",
        "for i in range(N_EPOCH):\n",
        "  model.train()\n",
        "  avg_loss = 0\n",
        "  for j, (x, y) in tqdm(enumerate(train_iterator, 1), total=len(train_iterator), position=0, leave=True):\n",
        "    y = y.squeeze(1).to(device)\n",
        "    x = x.permute(1, 0).to(device)\n",
        "    optimizer.zero_grad()\n",
        "    model.hidden = model.init_hidden(x.size(1)) \n",
        "    output = model.forward(x)\n",
        "    loss_train = criterion(output, y)\n",
        "    loss_train.backward()\n",
        "    avg_loss += loss_train.item()\n",
        "    torch.nn.utils.clip_grad_norm_(model.parameters(), 0.2)\n",
        "    optimizer.step()\n",
        "  losses_train.append(avg_loss / j)\n",
        "  avg_loss = 0\n",
        "  model.eval()\n",
        "  test_acc = []\n",
        "  for j, (x, y) in tqdm(enumerate(test_iterator, 1), total=len(test_iterator), position=0, leave=True):\n",
        "    y = y.squeeze(1).to(device)\n",
        "    x = x.permute(1, 0).to(device)\n",
        "    model.hidden = model.init_hidden(x.size(1)) \n",
        "    output = model.forward(x)\n",
        "    test_acc.append(accuracy_score(torch.argmax(output, axis=1).cpu(), y.cpu()))\n",
        "    loss_test = criterion(output, y)\n",
        "    avg_loss += loss_test.item()\n",
        "  losses_test.append(avg_loss / j)\n",
        "\n",
        "  print(\"\\nepoch\", i)\n",
        "  print(\"train_loss\", losses_train[-1])\n",
        "  print(\"test_loss\", losses_test[-1], \", test accuracy\", np.array(test_acc).mean())\n",
        "  print()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 280/280 [12:50<00:00,  2.75s/it]\n",
            "100%|██████████| 70/70 [01:32<00:00,  1.33s/it]\n",
            "  0%|          | 0/280 [00:00<?, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "epoch 0\n",
            "train_loss 0.8597637319139072\n",
            "test_loss 0.6090377994946071 , test accuracy 0.8067857142857144\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 280/280 [12:50<00:00,  2.75s/it]\n",
            "100%|██████████| 70/70 [01:33<00:00,  1.33s/it]\n",
            "  0%|          | 0/280 [00:00<?, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "epoch 1\n",
            "train_loss 0.8477341773254531\n",
            "test_loss 0.5552148529461451 , test accuracy 0.8022142857142859\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 280/280 [12:49<00:00,  2.75s/it]\n",
            "100%|██████████| 70/70 [01:32<00:00,  1.32s/it]\n",
            "  0%|          | 0/280 [00:00<?, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "epoch 2\n",
            "train_loss 0.8414186296718461\n",
            "test_loss 0.5551061621734075 , test accuracy 0.8027142857142857\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 280/280 [12:48<00:00,  2.74s/it]\n",
            "100%|██████████| 70/70 [01:32<00:00,  1.32s/it]\n",
            "  0%|          | 0/280 [00:00<?, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "epoch 3\n",
            "train_loss 0.8330604527677808\n",
            "test_loss 0.5697154326098306 , test accuracy 0.7899285714285713\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 280/280 [12:48<00:00,  2.74s/it]\n",
            " 31%|███▏      | 22/70 [00:29<01:03,  1.32s/it]"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "id": "O7vB1EScJ4Er",
        "outputId": "4c19e53b-c413-4420-c45f-887c82e77a76"
      },
      "source": [
        "plt.plot(losses_train, label=\"train\")\n",
        "plt.plot(losses_test, label =\"test\")\n",
        "plt.legend()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7f5985ff3590>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 23
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_vKDBdYBNGMO",
        "outputId": "01dcbe81-cc11-44a4-c7e1-1c47fe5ab305"
      },
      "source": [
        "results = []\n",
        "model.eval()\n",
        "for x, y in train_iterator:\n",
        "    y = y.squeeze(1).to(device)\n",
        "    x = x.permute(1, 0).to(device)\n",
        "    model.hidden = model.init_hidden(x.size(1)) \n",
        "    output = model.forward(x)\n",
        "    results.append(accuracy_score(torch.argmax(output, axis = 1).cpu(), y.cpu()))\n",
        "\n",
        "print(np.array(results).mean())\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.831375\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6RoZTj4KXmlw",
        "outputId": "6643b784-10f8-4173-a91f-44be52e509ff"
      },
      "source": [
        "results = []\n",
        "model.eval()\n",
        "for x, y in test_iterator:\n",
        "    y = y.squeeze(1).to(device)\n",
        "    x = x.permute(1, 0).to(device)\n",
        "    model.hidden = model.init_hidden(x.size(1)) \n",
        "    output = model.forward(x)\n",
        "    results.append(accuracy_score(torch.argmax(output, axis = 1).cpu(), y.cpu()))\n",
        "\n",
        "print(np.array(results).mean())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.803642857142857\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "m6B7Eg5J4S8b"
      },
      "source": [
        "bi = \"\" if BIDIRECTIONAL else \"non \"\n",
        "#torch.save(model.state_dict(), f\"{PATH_SAVE}navec_{EMB_DIM}_h_{HID_DIM}_GRU({bi}bi)_drop_{DROPOUT}_{N_EPOCH}_epochs.pth\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "7sam_K_2H22s",
        "outputId": "e9f95b06-71d9-422d-dba6-730a8568a6a4"
      },
      "source": [
        "bi = \"\" if BIDIRECTIONAL else \"non \"\n",
        "model.load_state_dict(torch.load(f\"{PATH_SAVE}navec_{EMB_DIM}_h_{HID_DIM}_GRU({bi}bi)_drop_{DROPOUT}_{N_EPOCH}_epochs.pth\"))\n",
        "model.eval()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "RNN(\n",
              "  (embedding): NavecEmbedding(id='hudlit_v1_12B_500K_300d_100q', indexes=[...], codes=[...])\n",
              "  (rnn): GRU(300, 512, num_layers=2, dropout=0.75, bidirectional=True)\n",
              "  (fc): Linear(in_features=1024, out_features=3, bias=True)\n",
              "  (dropout): Dropout(p=0.75, inplace=False)\n",
              "  (softmax): Softmax(dim=1)\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 11
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "C0Wq5xYsVC4t"
      },
      "source": [
        "## Анализ атрибутивной тональности и визуализация"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2fGPLnQCDTQq"
      },
      "source": [
        "## kNN только по эталонным словам (словам из словаря тональности)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "R3iKOfBhqKmu"
      },
      "source": [
        "Возьмем проекцию на 2 измерения с помощью PCA"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nOcY9vGy8oT1"
      },
      "source": [
        "emb = torch.Tensor([navec[vec] for vec in navec.vocab.words])\n",
        "words = navec.vocab.words"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V1ImCNTMDmSF"
      },
      "source": [
        "n_dim = 2\n",
        "u, s, vh = torch.pca_lowrank(emb)\n",
        "s[n_dim::] = 0\n",
        "emb_2d = torch.matmul(emb, vh[:, :n_dim])\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 195
        },
        "id": "n25up2a7FhcJ",
        "outputId": "4f568b13-21b3-4052-8b99-ea6e2485f8fa"
      },
      "source": [
        "tonal_dict = pd.read_csv(\"/content/kartaslovsent.csv\", sep=\";\")\n",
        "tonal_dict.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>term</th>\n",
              "      <th>tag</th>\n",
              "      <th>value</th>\n",
              "      <th>pstv</th>\n",
              "      <th>ngtv</th>\n",
              "      <th>neut</th>\n",
              "      <th>dunno</th>\n",
              "      <th>pstvNgtvDisagreementRatio</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>абажур</td>\n",
              "      <td>NEUT</td>\n",
              "      <td>0.08</td>\n",
              "      <td>0.185</td>\n",
              "      <td>0.037</td>\n",
              "      <td>0.580</td>\n",
              "      <td>0.198</td>\n",
              "      <td>0.00</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>аббатство</td>\n",
              "      <td>NEUT</td>\n",
              "      <td>0.10</td>\n",
              "      <td>0.192</td>\n",
              "      <td>0.038</td>\n",
              "      <td>0.578</td>\n",
              "      <td>0.192</td>\n",
              "      <td>0.00</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>аббревиатура</td>\n",
              "      <td>NEUT</td>\n",
              "      <td>0.08</td>\n",
              "      <td>0.196</td>\n",
              "      <td>0.000</td>\n",
              "      <td>0.630</td>\n",
              "      <td>0.174</td>\n",
              "      <td>0.00</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>абзац</td>\n",
              "      <td>NEUT</td>\n",
              "      <td>0.00</td>\n",
              "      <td>0.137</td>\n",
              "      <td>0.000</td>\n",
              "      <td>0.706</td>\n",
              "      <td>0.157</td>\n",
              "      <td>0.00</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>абиссинец</td>\n",
              "      <td>NEUT</td>\n",
              "      <td>0.28</td>\n",
              "      <td>0.151</td>\n",
              "      <td>0.113</td>\n",
              "      <td>0.245</td>\n",
              "      <td>0.491</td>\n",
              "      <td>0.19</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "           term   tag  value  ...   neut  dunno  pstvNgtvDisagreementRatio\n",
              "0        абажур  NEUT   0.08  ...  0.580  0.198                       0.00\n",
              "1     аббатство  NEUT   0.10  ...  0.578  0.192                       0.00\n",
              "2  аббревиатура  NEUT   0.08  ...  0.630  0.174                       0.00\n",
              "3         абзац  NEUT   0.00  ...  0.706  0.157                       0.00\n",
              "4     абиссинец  NEUT   0.28  ...  0.245  0.491                       0.19\n",
              "\n",
              "[5 rows x 8 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "id": "bOxdVJKfGnsy",
        "outputId": "5235adc6-06b2-435c-e72a-489183de137d"
      },
      "source": [
        "plt.bar(tonal_dict.tag.unique(), tonal_dict.tag.value_counts())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<BarContainer object of 3 artists>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 14
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 66,
          "referenced_widgets": [
            "670c65607459432fa40636b4b5d06c74",
            "4187f6e244214bbebabbb67f8eb0a5d9",
            "37ba1d85ee46420b867d1e40d3384bd6",
            "717fc1379bcf42daa69ab83b01821aa4",
            "db2e965d65b04f0c8ccd99ea1ee0367d",
            "8027c0cd01ec407695b2c4c109231d79",
            "1e0cc0bb41b8469e878fc53eecf2fc72",
            "aa09a8822b5e4719971ec01bbd6b4b53"
          ]
        },
        "id": "2IgjE5CJZ-Sp",
        "outputId": "5565af4c-7480-499c-ec75-ff3d47bdc685"
      },
      "source": [
        "morph = pymorphy2.MorphAnalyzer()\n",
        "\n",
        "cnt = 0\n",
        "adjs = []\n",
        "for w in tqdm(words):\n",
        "  m = morph.parse(w)\n",
        "  if m[0].tag.POS in [\"ADJF\", \"ADJS\"]:\n",
        "    adjs.append(cnt)\n",
        "  cnt+=1"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "670c65607459432fa40636b4b5d06c74",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, max=500002.0), HTML(value='')))"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "OzKxpXQgaAIZ",
        "outputId": "65c77248-2f3c-4ee0-bbc6-a567e1f47f4f"
      },
      "source": [
        "print(len(adjs))\n",
        "words = np.array(words)\n",
        "words_adjs = words[adjs]\n",
        "emb2d_adjs = emb_2d[adjs, :]\n",
        "emb_adjs = emb[adjs, :]\n",
        "emb2d_words_adjs = {word: emb for word,emb in zip(words_adjs, emb2d_adjs)}\n",
        "#words_emb2d_adjs = {emb: word for word,emb in zip(words_adjs, emb2d_adjs)}\n",
        "emb_words_adjs = {word: emb for word,emb in zip(words_adjs, emb_adjs)}\n",
        "#words_emb_adjs = {emb: word for word,emb in zip(words_adjs, emb_adjs)}\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "98840\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lQgH1p77DZw6",
        "outputId": "97d4b45e-1ea2-47a2-90db-f2aaeefdc0ed"
      },
      "source": [
        "standard_words = [word for word in tonal_dict.term if word in emb_words_adjs]\n",
        "print(f\"Total standard words: {len(standard_words)}\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Total standard words: 6440\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-orybMdBJ4Tz"
      },
      "source": [
        "import random\n",
        "random.shuffle(standard_words)\n",
        "# разделим на train и test\n",
        "standard_words_train = standard_words[:int(0.8*len(standard_words))]\n",
        "standard_words_test = standard_words[int(0.8*len(standard_words)):]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kJfftvGGGDpT"
      },
      "source": [
        "classes = {\"PSTV\": 0, \"NEUT\": 1, \"NGTV\": 2}\n",
        "dict_pairs_train = dict(tonal_dict[[\"term\", \"tag\"]][tonal_dict.term.isin(standard_words_train)].itertuples(index=False))\n",
        "X_train = np.array([emb_words_adjs[word].numpy() for word in dict_pairs_train])\n",
        "y_train = np.array([classes[dict_pairs_train[word]] for word in dict_pairs_train])\n",
        "\n",
        "dict_pairs_test = dict(tonal_dict[[\"term\", \"tag\"]][tonal_dict.term.isin(standard_words_test)].itertuples(index=False))\n",
        "X_test = np.array([emb_words_adjs[word].numpy() for word in dict_pairs_test])\n",
        "y_test = np.array([classes[dict_pairs_test[word]] for word in dict_pairs_test])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 450
        },
        "id": "bEZ7qu2-4m6h",
        "outputId": "f992e74e-ba0a-420d-d014-60cfda59197f"
      },
      "source": [
        "from sklearn.metrics import f1_score\n",
        "\n",
        "scores = []\n",
        "_n_neighbors = [5, 10, 20, 25, 30, 35, 40, 50, 55, 60]\n",
        "for n_neigh in _n_neighbors:\n",
        "  print(n_neigh)\n",
        "  estimator = KNN(n_neighbors=n_neigh, metric='euclidean', n_jobs=-1)\n",
        "  estimator.fit(X_train, y_train)\n",
        "  predicted = estimator.predict(X_test)\n",
        "  scores.append(f1_score(predicted, y_test, average=\"micro\"))\n",
        "\n",
        "plt.plot(_n_neighbors, scores)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "5\n",
            "10\n",
            "20\n",
            "25\n",
            "30\n",
            "35\n",
            "40\n",
            "50\n",
            "55\n",
            "60\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[<matplotlib.lines.Line2D at 0x7f0007b653d0>]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 37
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pszs03-LG3DI",
        "outputId": "1c71b62a-80f8-43eb-f89a-09e1a7f0d8ca"
      },
      "source": [
        "estimator = KNN(n_neighbors=25, metric='euclidean', n_jobs=-1)\n",
        "estimator.fit(X_train, y_train)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='euclidean',\n",
              "                     metric_params=None, n_jobs=-1, n_neighbors=25, p=2,\n",
              "                     weights='uniform')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 38
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "czm9wbVcjncX"
      },
      "source": [
        "from sklearn.externals import joblib\n",
        "joblib.dump(estimator, 'KNN.pkl')\n",
        "estimator  = joblib.load('KNN.pkl' , mmap_mode ='r')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "LDmllInPbCpO",
        "outputId": "d08c0028-5c8f-4617-ed94-87833c0664ce"
      },
      "source": [
        "# from sklearn.naive_bayes import GaussianNB\n",
        "# estimator = GaussianNB()\n",
        "# estimator.fit(X_train, y_train)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "GaussianNB(priors=None, var_smoothing=1e-09)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 28
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "534F-yL6JLkh",
        "outputId": "464e608f-1ee7-451d-ec09-9cf54af89bc2"
      },
      "source": [
        "from sklearn.metrics import classification_report\n",
        "\n",
        "predicted = estimator.predict(X_test)\n",
        "print(classification_report(predicted, y_test, target_names=[\"PSTV\", \"NEUT\", \"NGTV\"]))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "        PSTV       0.51      0.83      0.64       139\n",
            "        NEUT       0.90      0.81      0.85       889\n",
            "        NGTV       0.68      0.70      0.69       260\n",
            "\n",
            "    accuracy                           0.79      1288\n",
            "   macro avg       0.70      0.78      0.72      1288\n",
            "weighted avg       0.81      0.79      0.79      1288\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 863
        },
        "id": "IWk3NkI1YA1K",
        "outputId": "be53a4a9-8ac1-4d2a-f5d2-b81ab6316980"
      },
      "source": [
        "coords_2d = np.array([emb2d_words_adjs[word].numpy() for word in dict_pairs_test])\n",
        "\n",
        "plt.figure(figsize=(15, 15))\n",
        "plt.scatter(coords_2d[predicted == 0][:, 0], coords_2d[predicted == 0][:, 1], label=\"PSTV\")\n",
        "# plt.scatter(coords_2d[predicted == 0][:, 0], coords_2d[predicted == 0][:, 1], label=\"NEUT\")\n",
        "plt.scatter(coords_2d[predicted == 2][:, 0], coords_2d[predicted == 2][:, 1], label=\"NGTV\")\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA3IAAANOCAYAAABUbz43AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOzdcXBcd33v/c/PipJsHqg3iUO5WpvE0KBLYgmLCrjF9Lk8cYtIuQmqe2Pi207LFG5uH+jNkIeKWjMd1/UMjwVixs+Tp72Xmxs60KG3QQNCGEKqlngKJW15UCIjE6gKU+ZOvC4PiVu5A9nUinyeP1YraaVzVnt2zzm/3++c92smo/hoLR2fPbt7vuf7/X2/JggCAQAAAAD8scP2DgAAAAAA4iGQAwAAAADPEMgBAAAAgGcI5AAAAADAMwRyAAAAAOCZq2zvQJRdu3YFt9xyi+3dAAAAAAArnnzyyeeCILgp7HvOBnK33HKL5ubmbO8GAAAAAFhhjPmfUd+jtBIAAAAAPEMgBwAAAACeIZADAAAAAM84u0YOAAAAQLEsLy/r/PnzeuGFF2zvSqauvfZa7d69W729vW3/HQI5AAAAAE44f/68XvrSl+qWW26RMcb27mQiCAJdvHhR58+f1969e9v+e5RWAgAAAHDCCy+8oBtvvLEwQZwkGWN04403xs5CEsgBAAAAcEaRgriGTv7NBHIAAAAA4BkCOQAAAABY1dPTo/3792vfvn2655579Pzzz0uSPvShD+n222/X4OCg9u/fr69//ev6xV/8Re3fv18/9VM/pZ07d2r//v3av3+/fu/3fk/j4+NNP/fs2bN6zWtek9h+0uwEAAAAAFaVSiWdPXtWkvTLv/zL+tjHPqaf+Zmf0Re/+EU99dRTuuaaa/Tcc8/p8uXL+tznPidJ+ou/+At99KMf1Re/+EVJ0t/93d/pbW97m06ePLn2cx955BEdOXIksf0kkAMAAADgpZn5qiZnF3Vhqaa+ckljI/0aHaok9vN/9md/VgsLC7rlllu0a9cuXXPNNZKkXbt2tfx7r371q3X99dfr61//ut74xjdKkqampjQ7O5vYvlFaCQAAAMA7M/NVjU+fU3WppkBSdamm8elzmpmvJvLzX3zxRT322GMaGBjQW9/6Vj3zzDN69atfrfe+9736yle+su3fP3LkiB555BFJ0t/8zd/ohhtu0K233prIvkkEcgAAAAA8NDm7qNryStO22vKKJmcXu/q5tVpN+/fv1/DwsF7xilfo3e9+t17ykpfoySef1EMPPaSbbrpJ73znO/WJT3yi5c955zvfqc985jO6cuVK4mWVEqWVAAAAADx0YakWa3u7Nq6R26inp0dvectb9Ja3vEUDAwP65Cc/qXe9612RP2fPnj3au3evvvKVr+izn/2s/vqv/7qr/dqMjBwAAAAA7/SVS7G2d2NxcVHf/e531/589uxZ3Xzzzdv+vSNHjuiBBx7QK1/5Su3evTvRfSKQAwAAAOCdsZF+lXp7mraVens0NtKf+O/60Y9+pF/7tV/TbbfdpsHBQX3729/W8ePHt/1799xzj55++unEyyolSisBAAAAeKjRnTLprpU/+tGPtmz76Z/+af3VX/1V5N9plFxutmvXLi0vL3e1P1EI5AAAAAB4aXSokui4AZ9QWgkAAAAAniGQAwAAAADPEMgBAAAAgGcI5AAAAADAMwRyAAAAAOAZAjkAAAAAWGWM0Qc+8IG1P3/0ox9tmhn3qU99SoODg7r99tv12te+Vu95z3v0l3/5l9q/f7/279+vG264QXv37tX+/fv1cz/3c3rlK1+pxcXFpt/x/ve/Xx/+8Ie72k8COQAAAABYdc0112h6elrPPffclu/96Z/+qU6dOqXHHntMTz/9tJ566im96U1v0ste9jKdPXtWZ8+e1d13363JyUmdPXtWX/7yl3XvvffqkUceWfsZV65c0Wc+8xnde++9Xe0ngRwAAAAAPy1MSaf2ScfL9a8LU13/yKuuukr33XefTp06teV7H/rQh/TRj35UlUp9dl1PT49+/dd/Xf39/ZE/78iRI/r0pz+99uevfvWruvnmm3XzzTd3tZ8EcgAAAAD8szAlfeF+6dIzkoL61y/cn0gw9773vU9//Md/rEuXLjVtf/rpp/W6170u1s8aGBjQjh079M1vflOS9Mgjj+jIkSNd7yOBHAAAAAD/PH5CWq41b1uu1bd36Sd+4if0q7/6q3rwwQcjH3Pu3Dnt379fr3rVq5oybmGOHDmiRx55RC+++KJmZmZ0zz33dL2PBHIAAAAA/HPpfLztMb3//e/Xxz/+cf34xz9e23b77bfrqaeeklTPtJ09e1Z33nmnarVa1I+RJN17772amprSl7/8ZQ0ODuonf/Inu94/AjkAAAAA/tm5O972mG644QYdPnxYH//4x9e2jY+P67d+67d0/vx6sLhdECdJr3rVq7Rr1y4dPXo0kbJKiUAOAAAAgI8OHpN6S83bekv17Qn5wAc+0NS98hd+4Rd0//33684779Rtt92mN73pTerp6dHIyMi2P+vIkSP627/9Wx06dCiRfTNBECTyg5I2PDwczM3N2d4NAAAAABn5zne+o9e85jXt/4WFqfqauEvn65m4g8ekwcPp7WCKwv7txpgngyAYDnv8VZnsFRDDzHxVk7OLurBUU1+5pLGRfo0OVWzvFgAAAFwzeNjbwK1bBHJwysx8VePT51RbXpEkVZdqGp8+J0kEcwAAAMAq1sjBKZOzi2tBXENteUWTs4uW9ggAAABZcnXpV5o6+TcTyMEpF5bCu/5EbQcAAEB+XHvttbp48WKhgrkgCHTx4kVde+21sf4epZVwSl+5pGpI0NZXLoU8GgAAAHmye/dunT9/Xs8++6ztXcnUtddeq927441NIJCDU8ZG+pvWyElSqbdHYyP9FvcKAAAAWejt7dXevXtt74YXCOTglEZDE7pWAgAAANEI5OCc0aEKgRsAAADQAs1OAAAAAMAzBHIAAAAA4BkCOQAAAADwDIEcAAAAAHiGQA4AAAAAPEMgBwAAAACeIZADAAAAAM8QyAEAAACAZwjkAAAAAMAzBHIAAAAA4BkCOQAAAADwDIEcAAAAAHiGQA4AAAAAPEMgBwAAAACeIZADAAAAAM8QyAEAAACAZwjkAAAAAMAzBHIAAAAA4BkCOQAAAADwDIEcAAAAAHiGQA4AAAAAPEMgBwAAAACeIZADAAAAAM8QyAEAAACAZwjkAAAAAMAzBHIAAAAA4BkCOQAAAADwDIEcAAAAAHiGQA4AAAAAPEMgBwAAAACeIZADAAAAAM9cZXsHfDEzX9Xk7KIuLNXUVy5pbKRfo0MV27sFAAAAoIAI5NowM1/V+PQ51ZZXJEnVpZrGp89JEsEcMsdNBQAAAFBa2YbJ2cW1IK6htryiydlFS3uEomrcVKgu1RRo/abCzHzV9q4BAAAgQwRybbiwVIu1HUgLNxUAAAAgEci1pa9cirUdSAs3FQAAACARyLVlbKRfpd6epm2l3h6NjfRb2iMUFTcVAAAAIBHItWV0qKKThwZUKZdkJFXKJZ08NECDCWSOmwoAAACQ6FrZttGhCoEbrGucg3StBAAAKDYCOcAz3FQAAAAApZUAAAAA4BkCOQAAAADwDIEcAAAAAHiGQA4AAAAAPEMgBwAAAACeIZADAAAAAM8QyAEAAACAZwjkAAAAAMAzBHIAAAAA4BkCOQAAAADwDIEcAAAAAHjmKts7AAAA/DQzX9Xk7KIuLNXUVy5pbKRfo0MV27sFAIVAIAcAAGKbma9qfPqcassrkqTqUk3j0+ckiWAOADJAaSUAAIhtcnZxLYhrqC2vaHJ20dIeAUCxkJEDAHiN8j47LizVYm0HACSLjBwAwFuN8r7qUk2B1sv7Zuartnct9/rKpVjbAQDJIpADAHiL8j57xkb6VertadpW6u3R2Ei/pT0CgGKhtBIA4C3K++xplK9S1goAdhDIAQC81VcuqRoStFHel43RoQqBGwBYQmklAMBblPcBAIqKjBwAwFuU9wEAiopADgDgNcr7AABFRGklAAAAAHiGQA4AAAAAPEMgBwAAAACeIZADAAAAAM8QyAEAAACAZwjkAAAAAMAziYwfMMb8oaR/J+mHQRDsC/n+WyR9XtL3VzdNB0FwIonfDbhkZr7KPCsAAACkLqk5cp+Q9PuS/qjFY/4yCIJ/l9DvA5wzM1/V+PQ51ZZXJEnVpZrGp89JEsEcAAAAEpVIaWUQBF+V9I9J/CzAV5Ozi2tBXENteUWTs4uW9ggAAAB5leUauZ8xxnzTGPOYMeb2sAcYY+4zxswZY+aeffbZDHcN6N6FpVqs7QAAAECnsgrknpJ0cxAEr5X0/0iaCXtQEAQPBUEwHATB8E033ZTRrgHJ6CuXYm0HAAAAOpVJIBcEwT8HQfCj1f//kqReY8yuLH43kJWxkX6VenuatpV6ezQ20m9pjwAAAJBXSTU7ackY83JJ/18QBIEx5g2qB5AXs/jdQFYaDU3oWgkAAIC0JTV+4E8kvUXSLmPMeUm/K6lXkoIg+Jikfy/pfzfGvCipJuneIAiCJH434JLRoQqBGwAAAFKXSCAXBMGRbb7/+6qPJwAAAAAAdCnLrpUAAAAAgAQQyAEAAACAZwjkAAAAAMAzBHIAAAAA4BkCOQAAAADwDIEcAAAAAHiGQA4AAAAAPEMgBwAAAACeIZADAAAAAM8QyAEAAACAZwjkAAAAAMAzBHIAAAAA4JmrbO8AAHtm5quanF3UhaWa+soljY30a3SoYnu3AAAAsA0COaCgZuarGp8+p9ryiiSpulTT+PQ5SSKYAwAAcByllUBBTc4urgVxDbXlFU3OLlraIwAAALSLQA4oqAtLtVjbAQAA4A4COaCg+sqlWNsBAADgDgI5oKDGRvpV6u1p2lbq7dHYSL+lPQIAAEC7aHYCFFSjoYkrXSvpoAkAANA+AjmgwEaHKk4ES3TQBAAAiIfSSgDW0UETAAAgHgI5ANbRQRMAACAeAjkA1tFBEwAAIB4COQDW0UETAAAgHpqdALDOtQ6aAAAAriOQA+AEVzpoAgAA+IDSSgAAAADwDIEcAAAAAHiGQA4AAAAAPEMgBwAAAACeodlJgczMV+kKCAAAAOQAgVxBzMxXNT59TrXlFUlSdamm8elzkkQwBwAAAHiG0sqCmJxdXAviGmrLK5qcXbS0RwAAAAA6RSBXEBeWarG2AwAAAHAXgVxB9JVLsbYDAAAAcBeBXEGMjfSr1NvTtK3U26OxkX5LewQAAACgUzQ7KYhGQxO6VgIAAAD+I5ArkNGhCoEbAAAAkAOUVgIAAACAZwjkAAAAAMAzBHIAAAAA4BkCOQAAAADwDIEcAAAAAHiGQA4AAAAAPMP4AQAAkLqZ+SqzTAEgQQRyAAAgVTPzVY1Pn1NteUWSVF2qaXz6nCQRzAFAhyitBAAAqZqcXVwL4hpqyyuanF20tEcA4D8COQAAkKoLS7VY2wEA2yOQAwAAqeorl2JtBwBsj0AOAACkamykX6XenqZtpd4ejY30W9ojAPAfzU4AFAqd84DsNV5jvPYAIDkEcgAKg855gD2jQxVeZwCQIEorARQGnfMAAEBeEMgBKAw65wEAgLwgkANQGHTOAwAAeUEgB6Aw6JwHAADygmYnAAqDznkAACAvCOQAFAqd8wAAQB5QWgkAAAAAniGQAwAAAADPUFoJAEAOzMxXWf8JAAVCIAcAgOdm5qsanz63NvC+ulTT+PQ5SSKYA4CcorQSAADPTc4urgVxDbXlFU3OLlraIwBA2gjkAADw3IWlWqztAAD/EcgBAOC5vnIp1nYAgP8I5AAA8NzYSL9KvT1N20q9PRob6be0RwCAtNHsBAByjm6G+dd4PnmegZxbmJIePyFdOi/t3C0dPCYNHra9V7CEQA4AcoxuhsUxOlThOQXybGFK+sL90vLq2tdLz9T/LBHMFRSllQCQY3QzBICcePzEehDXsFyrb0chEcgBQI7RzRAAcuLS+XjbkXsEcgCQY3QzBICc2Lk73nbkHoEcAOQY3QwBICcOHpN6N92E6y3Vt6OQaHYCADlGN0NgE7r+wVeN85TzF6tMEAS29yHU8PBwMDc3Z3s3AACd4GIZLtrc9U+qZzTuepDzE4CTjDFPBkEwHPY9SisBAMlqXCxfekZSsN4ie2HK9p6h6Oj6ByBHCOQAAMniYhmuousfgBwhkAMAJIuLZbiKrn8AcoRADgCQLC6W4Sq6/gHIEQI5ALk0M1/VgYkz2nv0UR2YOKOZ+artXSoOLpbhqsHD9cYmO/dIMvWvNDoB4CnGDwDInZn5qsanz6m2vCJJqi7VND59TpJou58FWmTDZYOHORcB5AKBHIDcmZxdXAviGmrLK5qcXSSQywoXywAApIrSSgC5c2GpFms7AACAbwjkAOROX7kUazsAAIBvCOQA5M7YSL9KvT1N20q9PRob6be0RwAAAMlijRyA3Gmsg5ucXdSFpZr6yiWNjfSzPg4AAOQGgRxya2a+yoV8gY0OVXi+AQBAbhHIIZdoPw8AAIA8Y40ccqlV+3kAAADAdwRyyCXazwMAACDPCOSQS7SfBwAAQJ4RyCGXaD8PAACAPKPZCXKJ9vMAXEVHXQBAEgjkkFu0nwfgGjrqAgCSQmklAAAZoaMuACApBHIAAGSEjroAgKQQyAEAkBE66gIAkkIgBwBARuioCwBICs1OAADIyHYddeloCQBoF4EcAAAZiuqoS0dLAEAclFYCAOAAOloCAOIgkAMAwAF0tAQAxEEgBwCAA+hoCQCIg0AOAAAH0NESABAHzU4AAHDAdh0tAQDYiEAOAABHRHW0BABgMwI5AFjFDC8AAOALAjkAEDO8AACAXwjkAECtZ3hlGciRFQQAAO0gkAMAuTHDK09ZQQJSAADSxfgBAJAbM7xaZQV90ghIq0s1BVoPSGfmq7Z3rW0z81UdmDijvUcf1YGJM17tOwCgGAjkAEBuzPByISuYBN8D0jwEogCA/COQAwDVSxdPHhpQpVySkVQpl3Ty0ECm5YAuZAWT4HtA6nsgCgAoBtbIAcAq2zO8xkb6m9bISdlnBZPQVy6pGhK0+RKQ+h6IAgCKgYwcADjChaxgElwoU+2GS5lR1uoBAKKQkQMAh9jOCiahsf++dq10JTOapy6mAIDkEcgBABLnc0DqSiDqymxDAICbCOQAAM5wZf6cC4Eoa/UAAK0QyAEAnEApYTNXm8a4EmwDQNHR7AQA4ATa/jdzsWkMM/YAwB1k5ADAQ3nMilBK2CzVtXoLU9LjJ6RL56Wdu6WDx6TBw9v+NdbtAYA7COQAwDN5LUF0tZTQplTW6i1MSV+4X1pePdaXnqn/Wdo2mCPYBgB3UFoJAJ7Jawmii6WEufT4ifUgrmG5Vt++DZdm7AFA0RHIAYBn8poVcXog+sKUdGqfdLxc/7owZXuPOnfpfLztGxBsA4A7KK0EAM/kuQTRhbb/W3RRiuiknbvr/4aw7dtwZcYeAIBADgC8MzbS37RGTiIrkqpWpYg+BnIHjzUHppLUW6pvb4OTwTYAFBCBHAB4ZrusSB47WlrVRSmikxrBZwddKwEA7iCQAwAPRWVF8trR0qouShGdNXiYwA0APEezEwDIkbx2tLTq4LF66eFGMUoRAQBIA4EcAORIXjtaWjV4WLrrQWnnHkmm/vWuB8loAQCsorQSAHIkzx0traIUEQDgGDJyAJAjzPkCAKAYyMgBQI4w5wsAgGIgkAOAnGHOFwAA+ZdIaaUx5g+NMT80xnwr4vvGGPOgMeZ7xpgFY8zrkvi9AAAAAFBESa2R+4Skt7X4/p2Sbl397z5J/zWh3wsAAAAAhZNIaWUQBF81xtzS4iHvkPRHQRAEkv7GGFM2xvyrIAj+IYnfDwBAXDPzVdYSAgC8lVXXyoqkZzb8+fzqtibGmPuMMXPGmLlnn302o10DABTNzHxV49PnVF2qKZBUXappfPqcZuartncNAIC2ODV+IAiCh4IgGA6CYPimm26yvTsAgJyanF1UbXmlaVtteUWTs4uW9ggAgHiyCuSqkvZs+PPu1W0AAGTuQsjQ9FbbAQBwTVaB3GlJv7ravfLfSLrE+jgAgC195VKs7QAAuCap8QN/IumvJfUbY84bY95tjPkNY8xvrD7kS5L+XtL3JP13Se9N4vcCANCJsZF+lXp7mraVens0NtJvaY8AAIgnqa6VR7b5fiDpfUn8LgAAutXoTknXyg4tTEmPn5AunZd27pYOHpMGD9veKwAolEQCOQAp46IJSNzoUIXArRMLU9IX7peWV9cTXnqm/meJ9yUAyJBTXSsBhGhcNF16RlKwftG0MGV7zwAU0eMn1oO4huVafTsAIDNk5ADXRVw0Pf/YMf38l3ZRFgYgW5fOx9sOAEgFGTnAdREXR9c+/wOGGSN7C1PSqX3S8XL9K5nh4tm5O952AEAqCOQA10VcHF0Ibmz6M8OMkTrKfCHV1+j2bhrT0Fuqb/cZNykAeIZADnBdyEXT88HV+siLW5sKMMwYqWJtFKR6Q5O7HpR27pFk6l/vetDvRifcpADgIdbIAa5rXBxt6Fr5kR//kk7/yxu2PJRhxkgVa6PQMHjY78Bts1Y3KfL07wSQKwRygA82XTTtn6+qNH1OteWVtW0MM0bqdu5ezViEbAd8xk0KAB6itBLw0OhQRScPDahSLslIqpRLOnlogK6VSFde10YBNHAB4CEycoCnGGaMzIWU+W4ZTs/wevjo4LHmIecSNykAOI9ADgDQvlZroxoNIxoXw42GEY2/B7iqnZsUAOAYEwSB7X0INTw8HMzNzdneDWRoZr6qydlFBlwDvjq1L3QN3fkru/TO6/47r2kAAGIyxjwZBMFw2PfIyMEJM/NVjW9o3tEYcC2JCz/AFxGNIfrMRV7TAAAkjGYncMLk7GJTB0aJAdeAd7YZXs9rGgCA5BDIwQlRg6wZcA14pI3h9bymAQBIBoEcnBA1yJoB14BHBg9Ldz0o7dyjKzI6f2WXji6/R6evvHntIbymAQBIBoEcnDA20q9Sb0/TNgZcAx4aPCw98C2dfsfT+vngD5qCOF7TAAAkh2YncEKj+QFdK4F84DUNAEC6GD8AAAAAAA5i/AAAOISZiWiF8wMA0A4COQDIEDMT0QrnBwCgXTQ7AYAMMTMRrbR1fixMSaf2ScfL9a8LUxnvJQDABWTkACBDzExEK9ueHwtT0hful5ZX/3zpmfqfpXrHUABAYZCRA4AMMTMRrWx7fjx+Yj2Ia1iu1bcDAAqFQA4AMsTMxGKYma/qwMQZ7T36qA5MnNHMfLWtv7ft+XHpfPhfjNoOAMgtSisBIENOzFdbmKpncC6dl3bulg4eoywvQd00LNn2/Ni5u15OudnO3cn9AwAAXmCOHJAlLqBh2+Y1VpLUW5LuejCbc7EAr4EDE2dUDVnrVimX9MTRO7r74bafPwBAplrNkaO0EshK4wLs0jOSgvUmBXScQ5ZsrrEqyGsg1YY2g4frQdvOPZJM/StBHAAUEqWVQFZaXUBzEYas2FxjVZDXQF+5FJqRS6yhzeDhXB0vAEBnyMgBWaFJAVwQtZYqizVWBXkN0NAGAJAFAjkgKzYvoIGGg8fqa6o26i3Vt6etIK+B0aGKTh4aUKVcklF9bdzJQwPZNrQBAOQepZVAVg4eC29SkMUFNNDQKMmz0XCkQK+B0aEKgRsAIFUEckBWbF5AAxvZWmPFawAAgMQwfgAAAAAAHMT4AQAAAADIEQI5AAAAAPAMgRwAAAAAeIZADgAArFuYkk7tk46X618XpmzvEQAgBF0rAQBA3cJU84iIS8/U/yzRXRQAHENGDgCwFVmZYnr8RPOcP6n+58dP2NkfAEAkMnIAgGZkZYrr0vl42wEA1pCRAwA0IytTXDt3x9sOALCGQA4A8iKpckiyMsV18JjUW2re1luqbwcAOIVADgDyoFEOeekZScF6OWQnwRxZmeIaPCzd9aC0c48kU/9614OU1AKAg1gjBwB50KocMu5F+MFjzWvkJLIyRTJ4mMANADxARg4A8iDJckiyMgAAOI+MHADkwc7dq2WVIds7QVYGAACnkZEDgDygSQUAAIVCIAcAeUA5JHzBsHkASASllQCQF5RDwnUMmweAxJCRAwAA2WDYPAAkhkAOAABkg2HzAJAYSisBAPDEzHxVk7OLurBUU1+5pLGRfo0OVWzvVvuS7q4KAAVGIAcADvL+gh2Jm5mvanz6nGrLK5Kk6lJN49PnJMmfc4Nh8wCQGEorAcAxjQv26lJNgdYv2Gfmq7Z3DRZNzi6uBXENteUVTc4uWtqjDtBdFQASQ0YOABzT6oLdm8wLInWabb2wVIu13Vl0VwWARJCRAwDH5OaCHVt0k23tK5dibS8EZtIBKDACOQBwDBfs6ZqZr+rAxBntPfqoDkycybRktZvyyLGRfpV6e5q2lXp7NDbSn+g+eqMxk+7SM5KC9Zl0BHMACoJADgAcwwV7emyvP+wm2zo6VNHJQwOqlEsykirlkk4eGihuuS0z6QAUHGvkAMAxjQtzulYmz/b6w75ySdWQoK3dbOvoUIXzoIGZdAAKjkAOABzEBXs6bK8/HBvpbxohIJFt7Rgz6QAUHKWVAIDCsL3+0NnySB+bhhw8Vp9BtxEz6QAUCBk5AEBhuJARcy7b2mga0lhv1mgaIrk9JqCxb4+fqJdT7txdD+Jc3mcASBCBHACgMFh/GKJV0xDXgyJm0gEoMAI5AIDbFqYSzbo4lxGzjaYhAOAl1sgBANzFrLD0RTUHoWkIADiNQA4A4C5mhaWPpiEA4CUCOQDwjY8dBjtF2V/6Bg9Ldz0o7dwjydS/3vUga8/ghiK93wExsUYOAHzia4fBTjErLBs0DYGLivZ+B8RERg4AfFK0UkPK/oDiKtr7HRATgRwA+KRopYaU/QHFVbT3OyAmSisBwCdFLDWk7A8opiK+3wExkJEDAJ9Qagi0hyYZ/uP9DmiJjBwA+KSRmUpwQDaQOzTJyAfe74CWTBAEtvch1PDwcDA3N2d7NwAAiG9hiotPm07tiyjJ2yM98K3s9wcAOmSMeTIIguGw75GRAwAgSWSD7Bc7EDgAACAASURBVKNJBoACYI0cAPiGtT9uc7Bl+sx8VQcmzmjv0Ud1YOKMZuar1vYlE1HNMGiSkQ+8BwKSCOQAwC+NbM+lZyQF69keLmTc4Vg2aGa+qvHpc6ou1RRIqi7VND59Lt/BHE0y8ov3QGANgRwA+MTBbA82cSwbNDm7qNryStO22vKKJmcX0/ultjMmzB/ML94DgTWskUMmZuarmpxd1IWlmvrKJY2N9Gt0qGJ7twD/OJbtQYiDx5rXyElWs0EXlmqxtnfNlTWCzB/MJ94DgTVk5JC6Qpb1AGnpNNtjO0NSJI5lg/rKpVjbu0bGBGlyLOMN2EQgh9RZKesB8qqTtT+sKcne4OF6m/vjS/WvFjNDYyP9KvX2NG0r9fZobKQ/nV9IxgRpYv0jsIZADqnLvKwHyLNOsj1kSAptdKiik4cGVCmXZCRVyiWdPDSQXnl7nIwJmWLE5VjGG7CJNXJIXV+5pGpI0JZaWQ+Qd3HX/uQgQ8I62+6MDlWyO17trhF0ZS0d/MP6R0ASGTlkIPOyHiCHupoD5vmaktyus81rNqrdjAmZYgDoChk5pK5xF5i76UBnGoFMY61pI5CR1N7ryLEuinG1Wmfr7ftI3rNR7WRMcpApBgCbCOSQiUzLeoCc6TqQaVxQP36ifpG8c3c9iHMtYFiYCt3HXK6zbZWNcu15ScvO3asNeEK2d4DyWwBFQyAHAI5LJJBxfU1JiwxVX3lXvtbZLkyFBzBSsbJRCWaKu85aw00RN3cA1LFGDgAcl/kcMBtaZKhytc62EbBG8WTdYiIS7D7Y0ZibvK5RzAvGpgDbIiMHAI4bG+lvyjZIHgcym63dcY/OUOVmne3ClPS535CClfDve7RuMTEJZYpjZ63zvkYxDyg/BrZFIAcAjstNILPZ5ovpMKsZqtTW2WZVutX4t0YFcRKzsLqw7Zibzc/z5R8TJLiOZjjAtgjkAMADuWwYFHbHfaO0M1TdZmXiBIHb/Vt37iGA6ELLrHXY8xyliEGCq+vQEm6GA+QRa+QAAHa0umjuYr1U27qZYxZ3/U6rf2tKAWtXswc9MzpU0clDA6qUSzKSKuWSTh4aqN/82C6I3qhoQYLL69AOHqu/NjYqYvkx0AIZOQCAHZF33PdID3wr/d/fTelW3PU7Uf9W05NKwFrELo6RWet2s2xFDBJcXofmy9gUwCIycgAAO2zfcY/KvrSTlYkbBEb9W3/xY6lcmHbUxTGvop7P0g2JdMz0muvr0AYP12/qHF+qfy3a8wNsg4wccovhsMnhWCIVtu+4dzPHLO76nYz/rbkcot6pqOf5zg8TGLAODfAagRxyqYhlRWnhWCJV7bSfT6sZQzfBVSdBYIZD2bft4lgktm8YuCzBoewAsmeCILC9D6GGh4eDubk527sBTx2YOBN6EVMpl/TE0Tss7JG/OJZIQ9tZ3rARBb0lN8rgXO32p603YKR6F8e1BiBAg8PnMQDJGPNkEATDYd8jI4dcoqwoORxLJGphSs8/dkx3P/8DDQc36iM7Duv00pujs7yuN2OwvQ8RnJ49SODgFofPYwCtEcghlygrSg7HEolZza5dt1yTjLTbPKeJ3oelZen08ps1Obu4NdBwvRmDw5ycPdjt7D4g77jRgRjoWolcGhvpV6m3p2nb2nBYxOLysSzSnCwvLExJp/ZJx8v1r5tnUYVk164zl/XBq+qPC83ydtNZEu7pZnYfkHcuz/WDk8jIIZecLivyjKvHkiYsjnUTbSfTEpFF6zMX61/Dsrw0Y8iXDjKsTp3nQJpcLiWHkwjkkFtOlhWlLK0LHhePZas5Wa7taxqcC2TbuQCJaHVuFOiJa+7Xhds+KGlTAx06DuZLzHb3mZ7nlLTBNkrJEROllYCPQkrYGhc81aWaAq1f8OS13LDoTVicG/jczgVI2FBsScZIFfOcXn/ud8NLiBgKnB8xh8Bndp5T0gYXUEqOmAjkAN9EXHCcffQhty7sUxbVbKUoTVicC2TbuQAZPFwfG7BzT/hjWSuVCqfWkjadA6b+tcUoiczOc9tr97ZbX4piiHmjAyCQA3wTccHxnsufCn14XjNULjdhyUKqgWwnF5XtXoA0smsy4T+HEqJEOZmpj5FhzeyGjc2SNrKBaIh5owMgkAN8E9UwYsfF8O05zVCNDlV08tCAKuWSjOoDyos07Di1QLbTi8q4FyCUEGXCuRLcmDK7YWPzfLSdDUSyus2uUkqOGGh2AvgmolnAC6WXq7TS03TRlvcMlYtNWLKSWjfRbrqmxRksTDfKTDhXghtTZl1zbZ6PNLjID+YkImMEcoBvIi44rrvzhE6uDHjfpptW4+1LJZDN6qKSbpSZ6CuXVA0J2nzK1Gdyw8bm+RizkyccZmN8AN1WC41ADvBNiwuOUfk9Q81qS30+DOuyvKiMk8HLi4zPs7GR/qbXlJT/TH3HbJ2PZKfzI+vsKhnAwiOQA3yU0wtga7Ph+DBcx0VleiycZ5mVJqJzZKfzI+vsKgPEC49ADoAzrK3n4cNwHReV6bF0njm1lpTMd7ic3pwrnKxvhLG+svAI5AA4w9p6Hj4Mm3FRmY6inmdrwdszqo+dCOrbi5z5Rj5lfSOM9ZWFx/gBAM6wNhuOVvjIQhHPs6ZxFtJaENdAm33kTZbjAxggXngEcgCcYW02HB+GCDEzX9WBiTPae/RRHZg40/0Q7SKeZ2HlpJvlPSMJpIUB4oVHaSUAp1hZz8O6MGySSgfVIp5n7QRpec5IAmmjFL7QCOQAQOLDEE2S6qC6dS7iAY0+8K2kd9ddUWt4GnzNSNK0BYADCOQAIGEMNfdf7A6qIRf2MysH7M1FdEVYF79Gw5Ode/wMgBIeI8H7BQqFmyCJIpADgARZHWqOxMTqoBpxYX82+E+qLb+h6aGZzEV0SR7LSRMcI8H7RQYIHNzBzNbE0ewEABLUqiQP/ojVQTXiwv49lz8V+rNTn4vomiy7+GUhwTESvF+krKlrarAeOCxM2d6zYmp1EwQdIZADgARZG2qORMXqoBpxAd+342L49rTnIiJdCY6R4P0iZQQObinqLM0UUVoJAAmyNtQciWu7g2pEQ48XSi9XaaWnKeOSyVxEpCts3V+HTVvSfL9g7Z0IHFzDAPPEkZEDgARZG2oOeyLmw1135wk7cxGRrgRnd6X1ftFYe1ddqinQ+tq7rmch+iYqQChdL53aJx0v179SapmNIs7STJkJgsD2PoQaHh4O5ubmbO8GAMTmw51wH/bRKzRUQIfSeC0emDgTmumrlEt64ugdXf1sr2xuriFJO3olY6SVy+vbeksM0s4K75WxGWOeDIJgOPR7BHIAUCybO+VJ9SwA2SIgH/YefVRhV3dG0vcn3p717ti1OXC4/GOp9o9bH7dzT70ZD+CYVoEcpZUAUDB0ygPyLWqNXSHX6m7umlr7p/DHsW4OHiKQA4CCoVMevLQwxbqmNvm8VndmvqoDE2e09+ijOjBxJvl1fQl2HQVsI5ADgILx9W596hd4cBfzwGKJNT7DIRubtNy142v69PP/UXfP3K7nP/yvk3uuabiBHGH8AAAUzNhIf+gaOZfv1m9e19fowifJ+YtTJKDVPDAaJYRqe3yGQxpl33fv+Jomeh/WdabekOS62j/UA3ep++e78fdpuIEcIJADgIJpXNz51LWy1bo+l/fbe650mGMeWCE0yrs/eNXUWhC3JsnAffAwgRtygUAOADwU2bK8zQtv3+7Ws67Pgs2t2xvljFL2F8EMEi6ExoD0PvNc+AMI3IEmrJEDAM9EDfv9xun/ltt1RL6u6/Naq3LGrLGuyS8dNqZpNGm5EOwKfwCBO9CEQA4APBNVZrjnqUl3LrwT5nMXPm+5VM44eLg+sHnnHkmm/pUBzm7qojFNo0nLw1f/ip4Prm7+JoE7sEUipZXGmLdJ+r8l9Uh6OAiCiU3ff5ekSUmNFmO/HwTBw0n8bgAomqhywpcFz9Yn/m6Wg3IkH9f1Oa2dElzXyhlZ1+SHLhvT1Mu+f09auN3O+kxX1oUCbeg6kDPG9Ej6A0k/L+m8pG8YY04HQfDtTQ/9dBAEv9nt7wOAomusI9nsh+YmvVzPbv0LOSlH8m1dn7PaXft28Fjz4ySyItheUplcG4G7S+tCgTYkUVr5BknfC4Lg74MguCzpEUnvSODnAgBCRJUZPvO6MdYRYXvtrn1Lo5yRod755/PAbZfWhSI7Hr8vJVFaWZG0sfbivKQ3hjzul4wx/6ukv5P0QBAEW+o1jDH3SbpPkl7xilcksGsAEJMHZTVRZYavH3qbdMv1zu8/LIuTMUkyK0K2oxh8zuS6tC4U2fD8fSmr8QNfkPQnQRD8izHmP0n6pKQ7Nj8oCIKHJD0kScPDw0FG+wYAdR69oUeWGbKOCNuxtfaNod7F4PPAbdfWhSJ9nr8vJRHIVSXt2fDn3VpvaiJJCoLg4oY/PizpIwn8XgBIVsZv6JGz4IA02cqYkO0ojrRuKKVdMeFzNhGd8fx9KYk1ct+QdKsxZq8x5mpJ90o6vfEBxph/teGPd0v6TgK/FwCSleEbetQsuJn56rZ/F+hY40J4uSaZ1XWWWbXy93ntFOzrYqxB2xhzUTyevy91HcgFQfCipN+UNKt6gDYVBMHTxpgTxpi7Vx92vzHmaWPMNyXdL+ld3f5eAEhchm/oUbPgJmcXE/9dgKRNF8KSgpX1bEMWF6oM9UY3smpEMnhYeuBb0vGl+leCuHzz/H0pkYHgQRB8KQiCVwdB8KogCD60uu1YEASnV/9/PAiC24MgeG0QBP9bEAR/m8TvBYBEZfiGHjULLmo70DXbHfnIdqAbnpfAwVGevy9l1ewEANyX4SL9qFlwfeVSyKOBBLhwIUwznmx40H03NhqRIC0evy8RyAGb0ICi4DJ6Qx8b6df49Lmm8spSb4/GRvpT/90ohs3vZX9eermuq/3D1gdyIZwvHnXfjYVGJMAWBHLAqpn5qo6fflpLteW1bY0GFJII5pCoqFlwnGeO8izD0Wim07hRUF2q6djVv6SJ3od11coL6w/M4YVw5M04z57DjnneTj2Sz2MNgJSYIHBzXNvw8HAwNzdnezdQEJsvejarlEt64uiW0YcAimBzhkOqB0AOr6M4MHEmtHT3XS/5f3X8f/lsbi+Ew97LS709+qPX/0+9/tzvevUcdux4WVLYtZ2pN/AA4BVjzJNBEAyHfY+MHKDwDoIb0YACzitKtsEGDzMcUe9Zn/zRG3T8d34v473JTlQ32D1PTUry6znsWA7WkrHEAWhPIl0rAd9tF6jRgAJOy2K+UpG50CQkpqj3rNy9ly1MSaf21bNQp/Zp+J//PPRhLwueDf/7Dj+HHfO8nTozNoH2EcjBnk0fwDYvOltd3NCAAs6z3VY+7zwcGDs20q9Sb0/TtlJvj/6v277rzPtu10JuYExc/XHdveNrWx76Q3NT+M9w+Dns2Hbt1ON+9mb5Wb0wpX/z+X+rp3e8U1+7+v6155IZm0A4AjnY4VgGIeyiR5Kuv65XJw8NUNIBt3mYMfJKlhmOhC6aR4cqOnloQJVySUb1db5r68Qced/tWsgNjJL+Rb/d2/zvKfX26JnXjXmdpYotaqh13M/eLD+rV3/Xy/Wsdhhp947nNNH78FowxxIHYCvWyMEOx9ac0EEQXsvBmhinZdUtL+G28aNDleb3sFP3O/W+27WIGxV95qIq5VLTe/nrh94m3XI960jjfvZm+Vkd8ruuM5f1waumdPrym/NXFgwkgEAOdjiYQdhy0QP4gvlK6ctivmDaF80Ovu92JeIGhtm5W088ENJl2OOhv4mJew5kec60CMxZ4gCEo7QSdni45gRw1nZrYuCHtC+as3rfzWpNledNPayIew5k+Vkd8TN/aHaxxAGIQCAHO/gABpIVtSYG/kj7ojmL990s11RxAyO+uOdAlp/VEb/r5Yf+T4I4IAKllbAjqzUnAOCLtEtks3jfzXr9M+WS8cQ9B7L8rOa6AIjNBEFgex9CDQ8PB3Nzc7Z3AwAQhgHk6bB4XBMZwny8LCnsusLUs8UAgFiMMU8GQTAc9j0ycgCAeBLurui8LIMrSxmmxhDm2vKKpPUhzJLiBXN0UAWAzLBGDgAQT5EGkDs28zItk7OLa0FcQ0dDmFn/DACZIZADAMSTtzb2rRQkaI0athx7CDMNSAAgM5RWAgDiKVL5XEGC1r5ySdWQoK2jIcw0IAGATJCRA7qR1bwkwCVFKp8ryMzLsZF+lXp7mrYxhBkA3EYgB3SqIGtngC2KVD5XkKB1dKiik4cGVCmXZCRVyiWGMAOA4xg/AHTq1L6I8rI99YHMAPKh066VjGgAAHSJ8QNAGgqydgYovE7WfBVtRAMAIHOUVgKdKsjaGQAdSKvbJetyW+P45BPPKxCKQA7oVEHWzgDoQBoZe9bltsbxySeeVyASgRzQqZw3fJiZr+rAxBntPfqoDkyc0cx81fYuAf5II2NfkJl2HeP4+KedTBvPKxCJNXJAN3I6L2lmvqrx6XOqLa9IkqpLNY1Pn5MkutgB7Th4rHmNnNR9xj7tdbm+N2dh3XJnbD3v7a4j5XkFIpGRA7DF5OziWhDXUFte0eTsoqU9AjyTRsY+zXW5eShfS+L4FG0tls3nvd1MW9TzV7o+nf0CPEIgB2CLC0u1WNsBhBg8XB9Fcnyp/rXbLEea63LzUL7W7fHJQzAbl83nvd1M28Fj0o7erY+7/KN8PzdAGwjkAGzRVy7F2g4gA2muy81D+Vq3xycPwWxcNp/3djOog4ela1669XErl/P93LSjaBlkbMEaOQBbjI30N62Rk6RSb4/GRvot7hWA1Nbl7ty9mokK2e6Tbo6PpaBmZr6qydlFXViqqa9c0thIf3ZrkW0+73HWkdb+Kfxn+HSjIWnMqoTIyAEIMTpU0clDA6qUSzKSKuWSTh4aoNEJkFeMU4lec5ViUNNoLFVdqinQemOpzLoE23ze42RQmdu6VREzyNiCjByAUKNDFQI3F/neWTAtHJfuNI5VUY/hwlR9zdVmO3pTDWpaNZbK5P3X9vPebgY1jS6wvstDOTS6RiAHAL6glCYcxyUZOR2n0pbHT9TXXG12zUtTPSZONJby4Xm3HXC6KMmyWG6EeYvSSgDwBaU04Tgu6FZUFiNqbVZCaCwVQ9JdYH2XVFlsEbu15giBHPxGxyYUCaU04Tgu6MbClGQiLodSXoM1NtKvUm9P0zYaS6EtSXWx5UaY1yithL8op0LR5KWzYNI4LuhU43MkWNn6vQzWYDXWwVnrWgm/JVEWy40wrxHIwV+t7iIRyCGPWPAfzsZxYU1JPoR9jkiS6UluRt82aCwFq7gR5jUCOfiLu0goGhb8h8v6uNioBiBwXJPo3LWoz4vgSmGPb1FZnednEzcIvUYgB39xFwlF5EOHORuyPC5ZVwNQRr6mMXet0bK/MXdNUmcX3XyOQCmcVz7hBqHXaHYCfzHAFsgH35oWZV0NQDOCNa3mrnWEzxEohfPKN3QE9RaBHPyVVMcmAPb42Po6KluTVhaHMvI1ic9d43MkW47etHFinh/QAUor4TfKzJzUzlqDwq5HQDMfmxZlvaaE8r81feWSqiEX113NXeNzJBsOlwincl7lEWt1nUNGDkCiGmsNqks1BVpfazAzX431GBSEj9mmbrM4cbMSlP+tYe6ax+KWCGeYveO8aoOP1RMFQEYOQKJarTXYODNpu8egIHzNNnWaxekkK0EzgjXMXfNYnJs2GWfvOK/a4GP1RAEQyAFIVDtrDViPgDVFa33d6cUQ5X9rmLvmqTg3bSwEDUmeV7lcOuBj9UQBUFoJIFFRawo2bm/nMSiIojWb4GKoO442y0Ab4pQIe/w6ye3SgaybPKEtBHIAEtXOWgPWI6BJkVpfczHUOdbo+C3OTRuPXye5HWXAWl0nUVoJIFHtrDVgPQIKq2ilpElKsdzOy1K4OB0EXek22G6JsMevk7aXDrjynLSLtbpOIpADkLh21hqwzgWFxMVQ57Yrt+swsHm+9HJ97ce/pOrlN0laL4WT5O57VJxmIA63/Y/k8eukrVEGLjwnnQSSrNV1jgmCwPY+hBoeHg7m5uZs7waAHPDybjuAZqf2RTTL2BOdwQkr3dt8ES3p+eBqHV1+j05fefPatkq5pCeO3pH0vyIZrY7FA99q2vT8h/+1rqv9Q1uPzQ2L2a7GGrmN5ZWl3h6dPDSw/rkT4/lLRchrIPL1AuuMMU8GQTAc9j3WyAFFl/PmAbldeA4UTas1OnFmlIU89jpzWR+8qvm9z+kuum02A5mZr+ra538Q72f4bmFK+vz7mtdSfv59mX22jQ5VdPLQgCrlkozqNwSagjjJfjOXuDP94CxKK4Eic6G8I2XMrANyolW53fR94X8n7MI44mK5z1xs/rPLXXTbbOU/Obuo4eBG7TbPbfvY3Hjst6WVy83bVi7Xt2f0ubbt0gHb8zNtB5JIDBk5oMgKcFfOi5l1Oc+KAomJ6nAap8thxGMvBDeu/b/zXXTb7CB4Yammj7x4WM8HVzdtfz642ovGIR2p/WO87TbY7gDpcVdQNCOQA4qsAHflnJ9ZR0t1oHtxLoxDHvtiz7V6+OpfiS6Fc02brfz7yiWdvvJmHV1+j85f2aUrgdH5K7v0kd735qbqwku252faDiSRGJqdAEVme8F1BtpaeN7mz0mlYUoBngMgEz62409SyL9pZuVAIu9/Xvnw3vDsW+kG6be/n/3+uCqPr4GcatXshEAOKLKCdK7qNghLKhgMdbwsKex92NTLxwBgOy3ey2dWDhSra+/ClDTzXunK8vq2Hb3S6H+pf64RwMAzBHIAovGhtq0DE2dC5wIl0p6cjJxdnP/Z4Vinh/eRZlHnWkFuXiJfWgVydK0Eio4Bn9tKtWFK1Pwr1iqkrwBdW+NKrYSYY52uAqx3jiXqc61Vgy/OQ3iIZicAsI1UG6bYXvReZAXo2hpHqjMXOdbpKmIXwk66/RLwRqN7spfIyAHANsZG+kPXyHXVntyFMjMX9mEbqWWIJC7qNkl15iLHOl1Fy+x3muG1Pb/NVWTMvUVGDgC2MTpU0clDA6qUS8m0J3dh5IAL+7CNVDNEUjGzGC2kWkLMsU5X0TL7nWZ4absfrpuMOZk8q8jIAUAbRocqyWWCXFin4cI+bCPVDJFUvCzGNvrKpdCmPomUEHOs01ek9c6dZngbx6dRiVC6vv7n6fvq2xysSshEO8czrIJDIpNnGRk5wAEz81UdmDijvUcf1YGJM8llHOAmF8rMXNiHbaSaIZKKl8XYxthIv0q9PU3bui4hbuBY51vWWZluMryDh+udPA89JL1YW50552ZVQma2O55RFRyP/TZrXy0jIwdYtnlGWaN8TFK+Z/0UmQvrNFzYh22kmiFqyDKL4fiaxMb7TWprEouUMXJdkueijfVVSWR4k65KcPz13dJ2xzPqWG3e1uDQDcG8IyMHWNaqfAw55cI6DRf2YRupZoiy5sGaRKkezD1x9A59f+LteuLoHe7dTGI9TveSPhdtdCRNIsObZFWCJ6/vSNsdz7jHxKEbgnlHRg6wLPXyMdjR6u7s5nUaNu7e2tqHGHetU88QZcmDNYnOo7NeMpI+F22VaXeb4U2yKiEPr+9WxzPqWJVuqJensvbVGgI5wLJMyseQrXYuOF0oM8t6Hzq4EE+0yYxNHqxJdJ5PF8sul9klfS56UKYdKskGPHl/fUcdqzs/XP9/V8/1AqC0ErAsV+VjqGP4cbgiH5ectN+32pjJ1sVy3HJO18vskj4XPSjTDpVkA56cvL4jtTpWjeYxx5fqXwniMkVGDrAsV+VjqMv73dlOdXtcXM5ybCcH7fetN2aykfnppJwzoczhzHw1nc+FpM/FuGXaLr2Ok6pKyMHre1suVJFgCwI5wAG5KR9Dna+lRmnr5rj4vj7KhXWRXUp9rt92bFwsdxKUJXAjJyxo/trn/ove+mef1XW1H3R3/qRxLkZd5G8O2m59q/TN/+Hv6zhKDl7f8BOBHAAkrQh3ZzvRzXHxaX1UFM/vaHfdmKnbTIyNi+VOgrIEbuRsDprv3vE1nTAP67ra5dXf32UAlMW5GHbzZe4PJQXNj3PhdZxEltDz1zf8RCAHAEnj7my4bo5LUcpVXSo726SrxkxJZVSzvljuJChL4EbO5uD4g1dN6TpzuflBLgRArYTdfNkcxDXYfB37nu1HoRHIAUAauDsbrtPjUoRyVccvKMdG+pvK/aQYjZl8zah2EpQlcCNnc9DcZ54Lf6DLNzLi7JvN17Gv56ZvHL5J5TO6VgIA3OdrZ7w4HO/qOTpU0clDA6qUSzKSKuWSTh4aaG99nK8Z1U47G3bZyW9zN+MLwa7wB7p8IyNy30zzH22/jn09N33ieidXj5GRgzWpdeQCkD9FKFf14IKy48ZMPmdULWTXN3czfvjqX9HvBB/TVSsvrD/IdgC0nahs5mv/g/TdP3PndezzuekLsp6pIZCDFdbbWAPwT97LVfN8QdnpurECl2M1B81vlxZu9+tY+HLzheZU6fPgJpWvCORghfU21gDgmjxfUHZyUe/4msHM+Xgjw4d99iXg9Fmeb1JZRiAHK7puYw0AeZP3C8q4F/WUY9lTtEyojYCzSMc4zzepLCOQgxVdtbEGgE74cOHkQwYjK5Rj2UEmNH1FO8Z5v0llEYEcrOiqjTUAxNXBhRMNmSyjHMsOMqFbJP5eUMRjzE2qVDB+AFZ01cYaAOKK2dq/0ZCpulRToPWGTDPz1fT3FXVFGDnhIjKhTVJ5L+AYIyFk5GBNx22sgaLzoUTQNTEvnGjI5IDN5Vil6+t/nr6vvq0R0PFaSBaZ0CapvBdwjJEQMnIA4BMGq3Ym6gIpYjsNmRzRGKx96CHpxZpU+0etnfeff580815eC0kjE9oklfcCjjESQiAHAD6JWSKIVTEvnKIaL+WqIdPClHRqn3S8ZWbmTwAAIABJREFUXP/qcgAUdt6vXJauLDdvc/214MMxHzws3fWgtHOPJFP/eteDhc10pvJewDFGQiitBJAPRSk3ZG1FZ2J2Tct9QybfuubFOb9dfS34dMxpTLEmtfcCjjESQCAHwH8+XSB1i7UVnYtx4dRY+5LbrpW+dc2LOu+jHusi3445JEmjPU/orS85pmtrP9CFKzfq4at/Rfvffl9+3gvgNQI5AP4r0gVSTgerutjqP5WGTK5kjn3L7Iad9z1XS0HQXF7p8mvBt2OOtZuE162ed7t3PKfj5r9JPbdLsvjZ4sr7CKwjkAPgvyJdIOVwsGqjvXejdKnR3rvBtQCvYy5ljn3L7Ead92HbXH0t+HbMHZT5DR8XbxK69D4C60wQBLb3IdTw8HAwNzdnezcA+ODUvogLpD31jndw2oGJM6qGdIArl3r1Ly9e2bI2xduZky6dp5svBqV6NouGC+nhmHdl8w0fKYP3g+NlSWHXyUY6vpTO79yOS+8jyIQx5skgCIbDvkfXSgD+o5Wz16LaeC/VliPnN3nJpcwxXfOyxzHvSqt5bqmJObYkEy69j0TxoTtrTlBaCcB/OSw3LJK+cik0IxfF21lurpXW+dQ1Ly9rgnw65o6xMtvRxTXJrr2PbEbpZ6bIyAHIh8bg4ONL9a98YHhjbKRfpd6epm2l3h5df11v6OO9neVG5rgzjQtDBn8XmpXZji5mUV1/H2HWaabIyAEArIpq9S8pX7Pcip457jSr5mLDiYy42M3VFmuzHV3Lorr+PuJD6WeOEMgBAKxr1eo/Vxeyrl0UZqWbcquCXhi26ubq9WugQ7mf7RiHy+8jrpd+5gxdKwEAQLq66bRX0C59Ud1cK+WSnjh6h4U9AtpAd9bE0bUSAADY001WzfU1QSmx0twD6JaL6wpzjNJKAACQrm7KrVxfE5SSqG6u3jb7wfbozoqYCOQAAEC6um3jXsALQ2vNPWAHbfvRAQI5AP7Ly11MFIOD52vq3RELmlXrBs09CqbA3VnROZqdAPAbC6vhEwfP183dEaV65ufkoQGCBnTGwZsVoVzaz+NlSWHX5KY+HxWFRbMTAPnF8FH4xMHzdXJ2sSmIk6Ta8oomZxdT+X0z81UdmDijvUcf1YGJM5qZr6bye2CJLwPcM9jPWOd61HpR2vajBQI5AH4r6IwpeMrB8zXL7oiN7F91qaZA67PRCOZyxMGbFaFS3s/Y57rv3VkXpuqjQo6X619dC9xzikAOgN9cuovJB5l1zmd7XDpfV0V1QUyjO+Lx009nmv3zls/vJQ7erAjVaj8TOP6xM90+t+33JQubQwRyAPzmyl1MPsis8yLb48r5usHYSL9KvT1N29LojjgzX9VSbTn0e8xG28D39xIHb1aEitqf0vWJHP+OMt2Dh+tD7o8v1b/6EMRJ/mRhc4hADoDfXLmLyQeZdVmv9eqIK+frBqNDFZ08NKBKuSQjqVIupdLopNXzwGy0DXx/L3HwZkWoqP2UEjn+WWa6rfMlC5tDjB8A4D8XZkzxQWZdlmu9uuLC+brJ6FAl9Q6VrZ4HZqNt4Pt7yTajJlIfddHtfk7/x/DHxzz+hZoDuHP3agYzZDtSRSAHAEngg8y6vnJJ1ZBgwbk74K1anrvUDj1hUc/P9df1MuZgozy8l0TcrNg86qJR/izJXjC3cT8XpiQZhY4BiHn8O50D6EygG8fBY+FjVVzLwuYQgRwAJIEPMuu8uAO+eY5cY/1NQ9T3sgrmUgwko56f373r9kR+fm7k+L2kVfmzE8HK4ycUOcutg+MfN9PtXKDbrm2ysEgPA8EBICk5zqb4wvm72af2RWRb9tS/Rn3vgW+lu19S4sPKw54LKX6GopBC3ktmVg54f+z2Hn00KkzS9yfent2ORL1XRw7llnT8Uuq7dWDiTGjWulIu6Ymjd/AZU1CtBoKTkQOApKS19okP77ZlsdarK52sf8pqbVSrJhsxz7eozMLJQwP1C1K0tum9xNtMzSZOlD+3yopHlrXuyWTXWq7zbbXffB4UFl0rAcBlvrci93keVhpatWa33bY9wSYbXnQQ9YgTxzOB13JWoy5aanXDwnLHzZadLn3vZopUEMgBgMt8/vD2PQhNQ6sLRdtt2xMMJL3pIOoJ68czoddyIqMuug0oW92wsDwepGWg63s3U6SC0koAcJnPH94JlurlRjtNAWyV0SbYZMOJErocsX48E3wtd1X+nER54XZdQS2OB2nZ6fIvuuxmmucS/Tz/27ZBIAcALvO5FbnPQWiaWl0o2pwxF6Pz3HZNZbzoIJo0Cx0/MzuerryWkwgoHe8KGhnotrvfYeehlN/1dQVfO0ggBwAuc/yioyWfg9CiaiOQbKfxRqcztLyV8sWk9ePpyms5iYDS11b57ex31Hl4VSm/1REFr/xg/AAAuM7XspGE29nDDdu2SC+iVmMlshgdkba0X8vtvsfl/Th3K+r4RDLS8aXUdicTkSMjcvBvW8X4AQDwmc1yu274eucbLVlvvOEiV0oP05LmazlONjNuhYKvN8E6Ffd8SzKjautYu5IttoRADgCQHl+DUESy3ngjDd1ehBbhYjKt13Kc0rg4AWUR105FnYdhkizRt3msfV5+kADGDwAAgLY5MQssSUm01rc9OsJncbOZg4frZZTHl+pfowIFn0e3dCrsPNwihbEKNo+15ZERtpGRAwAAbbPeeCNpSTRLoIy4c2llM/Ne7hqmcb597jekYGXr99NaS2j7WBe48oNADgAAxNLVLDDXJHURWuCLya6kVRpXhHLXMI1zMMtyw6IeawdQWgkAAIor6mKTi9BspFUaV+Ry16zLDaOO9a1vrXfSPF6uf41Troy2kJEDAACF9Y1X/Wfte/J3VDKX17bVgqv1rVf9Z73e4n4VShrZzKKXu2aZIQ471re+Vfrm/yhWsxkLmCMHAAAK68DEGf30P/+5PnjVlPrMRV0IbtRHXjysJ3/i54s7Fw/oFjP/EsMcOQAAEGpmvpqfxiUduLBUU1Vv1unLb27aboo8Fy9vijZPzgW2G6AUBGvkAAAoqJn5qsanz6m6VFMgqbpU0/j0Oc3MV23vWmai5t95PRcP65IYL1FkC1OdrXNj7WkmCOQAACioydlF1Zab25TXllc0ObtoaY+yl7u5eGiWp3lynQZV3fy+ToPgIjebyRCBHAAABXUhonwwansejQ5VdPLQgCrlkoykSrmkk4cGClVemmt5KfGzkVnsJggu+KDurLBGDgCAguorl1QNCdpslRXaWq+Xq7l4aJbkjDOba+2SGFwfV7dBMLMVU0dGDgCAgnKprJD1ekhFUiV+ttfa2cgsss7NeQRyAAAUlEtlhazXQ6hu14UlVeJne62djaCKdW7Oo7QSAIACc6WskPV62KKRBet2qHQSJX6219odPNZ8LKT0g6qiD1X3AIEcAADoTgJrh1xbrwcH2FgXFiXJtXadsBVUsc7NaYmUVhpj3maMWTTGfM8YczTk+9cYYz69+v2vG2NuSeL3AgAAyxJaO+TSej2kJG6ZpO0s2EYulBkOHpYe+JZ0fKn+lQCr8LoO5IwxPZL+QNKdkm6TdMQYc9umh71b0j8FQfBTkk5J+nC3vxcA8mZmvqoDE2e09+ijOjBxhiYP8ENCa4dcWq+HFHQS8LvUbIN2+nBQEqWVb5D0vSAI/l6SjDGPSHqHpG9veMw7JB1f/f/PSPp9Y4wJgiBI4PcDgPcaHfsazR4aHfskFeNC1mZb75hstch3VoJZE1fW6yEFnZRJ2lgX1gplhnBMEoFcRdLGouHzkt4Y9ZggCF40xlySdKOk5zY+yBhzn6T7JOkVr3hFArsGAH5o1bEv9xe2STU0yEDhA+4wSa8d8iioRwydBPyuNtsIO0e72E9uDqFTTjU7CYLgIUkPSdLw8DDZOgCFUeiOfS41NNhGoQPuKG1kTdq+UPUoqEdMnQb8rmXBws7RmfdKxkgrl9e3tXnecnMI3Uii2UlV0p4Nf969ui30McaYqyTtlHQxgd8NALkQ1ZmvEB37XGposI1CB9xRtlk7FGvQd0RQf/4z46wb9Z0LzUKSEHaOXlleD+Ia2lwnyvxEdCOJQO4bkm41xuw1xlwt6V5Jpzc95rSkX1v9/38v6Qzr4wBgXaE79rnU0GAbhQ64W2nRTS/WhWpE8N5nLrYOAOE+G81Cuh0mHibODaY2HsvNIXSj60AuCIIXJf2mpFlJ35E0FQTB08aYE8aYu1cf9nFJNxpjvifp/5C0ZUQBABRZoTv2eXSnvtABd4diXahGBO8XghslbZOpSOOi3ebvyaMs2+e30yWzk+cyzg2mNh7rzM0hzmsvJbJGLgiCL0n60qZtxzb8/wuS7knidwFAXhW2Y5+rDQ1CNJ4fGhO0L9ag75D1ds8HV+sjL66fC6EBYFZr61jD54/t1t52+lyGrQnd0du8Rk5q+2bU2Eh/0xo5ycLNIc5rbxlXKxyHh4eDubk527sBAAC6sLmZg1S/UI3MOK92BLxy6bwuXLlRH3nxsE5fefPatyvlkp44ekfz3zm1L6KRxp565icpWf0edGZjN0lFXd+aekawm+eyna6Vt75V+u6ftXVzynrXSs5rpxljngyCYDjse051rQQAAPkSO4u52qXwdCMAvNJGpiKrhjkeNeYpnM1ZpSiNcsdunsuoTpqNbTEzXNarMTivvUUgBwAAUtXJhWqsADDpWXZRsvo9WcjbvL6wUsrNNpY7pvlcejRSRf9/e/cfInea5wX8/Zjp1eKU9C0nd5cfe66wBJdMY6ARJf9dVrOr5m4MGFSQEz3mH4/VQaIzLISw/0yOgIGgIIOK98ehNBizuyxH3Js5OVBPzJq1Z+724v0ASXpWT10SZW3Y7Pj1j+qapDvV3VVdP77fp+r1gqG6vt1T/VD5VvX3XZ/n+TzJYp3XS0aQAwA6aeQAOMJedlMxr98za4u4JurA6lF5OazO8t+ypgrX5kby/e+9fLzG83oJCXIAQN3m1TCnosY8B6qtYjSKfatK+6zzmuW/ZS0Vrv2mo/Y+mXzh5+s9F5aIIAcADNV6E4Zx7LduqWu/pwtTGmuqGI3qKBW2WZ0zk1T7Znl+7H3s739v+HTUT/yQEFcJQQ4AeMnebpODDbmTdDfMdV1XpjTWUjEaR5eqpUcdyyzPj2GPvZ+aA/2Ssf0AAPCS8zfeG7r/29D2/4ymK23eh02pW+kll26rxLRplufHfo89jG0HOuWg7Qd+37wHAwB039CNtw84zgi6MqVx7Uo/tB0/nX4jkNNCXBfM8vwY9TFqaXKyudEPp9dX+7ebG22PqBWmVgIALzmx2htakTux2mthNAuiS1Ma57WmkNHN8vzY77F7n+yviWt7Ouo4ujJFuQNU5ACAl1y9eCa9lWO7ju27ITejuXCtX/F4US0VkGXRZqVnlufHfo/9hZ/vT6O8/qR/W0MQOqjr6pJRkQMAXjLWhtyMpksNOXhZ25WeWZ4fi3TudWWKcgdodgIAAF1pRsPBluzfSbMTALrDInWgi1R66mCK8scEOQDmZzB16emjJM3zqUvCHDCOWXwgtF9TkZr311tEuq5+zBo5AObnoEXqS/hHGDiCWa1lu3Bt+P56S1jp6TxdV5OoyAEwT6YuAZOaVddClR4qoyIHwPx0aR8t6rS5sRid9zi6WX4gpNJDRVTkAJgfi9SZhDWWJNaywQ5BDoD5MXWJSdgImMQHQrDD1EoA5svUJY7KGsuZuPtgq/2N38eZMjtsc+vP/Jn+/Tuvm3LL0hDkAIA6WGM5dXcfbOWtO+9n+9lHSZKtJ9t56877STK/MHeULpQvfiA0qy6W0HGmVgIAdTClbupu3nv4cYgb2H72UW7eezi/QUw6ZdaUW5aUIAcA1GEaayxnsZF0xT58sj3W8ZmYdMqsKbcsKVMrAYB6TLLG0hS8l5xY7WVrSGg7sdob8tMzMumUWVNuWVIqcgDAcjAF7yVXL55Jb+XYrmO9lWO5evHM/AYx6ZRZU253uftgK+dvvJdPv/n1nL/xXu4+2Dr6g6lgd5qKHACwHEzBe8mgoUmrXSuHdaEcp+vkpP//Aplq8xoV7M4rTdO0PYah1tfXm/v377c9DACgAiO10L91dp8peKeTNz4Y7ReN0yYf5uz8jfeGTpU9udrLv33zJ8d7sGm8XphYKeWbTdOsD/ueqZUAQNUGVYitJ9tp8rwK8dKUskmn4A0qFE8fJWmeVyhMN6Mjptq8RgW78wQ5AKDbDlmnc2gL/cH/f+f15JVe0vtkjtT10ho7Om6/JjVHal6zX7MYTWQ6Q5ADALprhCrYgVWIvf//9neTH2wnl9/pTw8bZ1qkCgXDdKghyFSb12gi03mCHAB0SYcuCjthhCrYgVWIaVbRVCja1cXXRsem27527mTevvxqTq72UtJfG/f25VeP1rxmGvs2MlO6VgJAV+gS97IRqmBXL57Z1akveaEK8ZUpVtEuXNv975OoUOwYqdnMJLr62jjog4KWxvXauZPTe+4n2beRmVORA4CuWLA1WFPZz2qEKtiBVYhpVtFUKIYaudnMiI819Jzp6mvDdFtapCIHAF2xQBeFU9vPasQq2L5ViGlX0V6sUAy2Irjz+lJvRXBQs5lx/q0PPGe6+to4fmqfFv2m2zJ7KnIA0BULtAbr0E6So5q0CjarKlrH1ka1aVot7w88Z7r62lj0hiBdXJfIx1TkAKArZr0Ga46bWU91P6tJ1+nMYp1PB9dGteXEam/oJtTjtrw/8Jz5Kx1Yn3jQ62cRN4nv6rpEPibIAUBXzPKicM4XZdO6uO+srk71a8GBzWbGcOA5s/bn+nfaCkyHvX4WMdj4sKLzBDkA6JJZXRTO+aJsWhf3nTXq2qg5VkHbMlgHN2nXykPPmTYD0zKGGh9WdJ4gB1CDJbgYZMbmfFE2rYv7zhplGuwSTU2bRsv7Tp8zyxhqutjIxd/CXQQ5gK5bootBZqiFi7Kp7mfVNaNMg13GKs6EOnvOdDHUzFrX9k30t/AlulYCdF1X90+iLoveXa8Na1eSNz5Irj/p3+69mFzGKs6iWsbXT9f2TfS38CUqcgBdtwgXg6bDtG+Ru+t11TJWcRbVuK+fRXnPm8e6xFGfq0X4WzhlghxA19V+MWg6THcsane9rura1DQmM+rrx3ve6MZ5rmr/WzgDplYCdF3tU3pMh2FZdW1qGvPhPW904zxXtf8tnAEVOYCuq31KnOkwLDNV0OXjPW904zxXtf8tnAFBDqAGNV8Mmg6znBZljRCMy3ve6MZ9rmr+WzgDplYCMFumwyyfwbqXp4+SNM/XvWxutD0ymD3veaPzXE1EkANgtqwTWj7WCLHMvOeNznM1kdI0TdtjGGp9fb25f/9+28MAAMZ1fTXJsOuL0t9zDYCRlFK+2TTN+rDvqcgBANO13/oWa4QApkaQAwCmy7oXgJkT5ACA6bLuhcNsbiS3zvan4d46qxEOHIHtBwCA6dMmnP0MupoOGuIMupomzhkYg4ocAADzo6spTIUgBwDA/Dx9PN7xLjEllA4R5AAAmJ9au5ouwkb3tQbRWsc9Y4IcAADzU2tX09qnhNYaRGsd9xwIcgAAzE+tXU1rnhKa1BtEax33HOhaCQCMb3OjfyH19HF/StyFa92/EKc7auxqevzUTlVoyPEa1BpE9x33o/40yyV+D1KRAwDGY6oT41iU9U21TgkdqHVt4r7jK0v/HiTIAQDjMdWJUS1S6K91SuhArUF02LhTkjS7Dy3he5CplQDAeGqdosX8HRT6awlAL6pxSujAYNy1TYkeNu5hU1yTpXsPEuQAgPHUvlaI+akh9C/Tes9ag+jecd866z0oplYCAOOqdYoW89f1dVnDpn7eeT25frzu9XyLzntQEkEOABhX7WuFmJ+uX3APm/o5WHtV83q+Rec9KElSmqY5/KdasL6+3ty/f7/tYQAALLy7D7Zy897DfPhkOydWe7l68UxeO3dyOg/e5amL11fzUtOMvY6fTt74YC7DOZIuP79MrJTyzaZp1od9zxo5AIAldvfBVt668362n32UJNl6sp237ryfJNMJc11el3VQ44yBLq3n22swNXRQVRxUEZPuPudMjamVAABL7Oa9hx+HuIHtZx/l5r2HLY1ojoa2tt+jK+v5hrEVyFIT5AAAltiHT/auETv4+ELZtdYq6e9P9oIurecbpoauoMyMIAcAsMROrA6vSO13fOGsXemvgbv+NLn8Tl0NNLreFZSZskYOAGCJXb14ZtcauSTprRzL1YtnWhxVS7q8nm+YC9d2r5FLul9FZGoEOQCAJTZoaDKzrpXMziB06lq5lGw/AAAA0EG2HwAAqjbTfc4AKiTIAQCdNvN9zpg/m1jDxHStBAA6ban3OVtEg02snz5K0jzfxHpzo+2RQVUEOQCg05Z6n7NFZBNrmApBDgDotKXf52zR2MQapkKQAwA67erFM+mtHNt1bGn3ORtmcyO5dTa5vtq/7foURZtYw1QIcgBAp7127mTevvxqTq72UpKcXO3l7cuvanSS1Lne7MK1/qbVe33/e90eN3SMfeQAAGp16+xOiNvj+OnkjQ/mP55RbW4kv/T3ku3v7j6+0ksu3a6ng6Xum8zYQfvIqcgBANSq1vVma1eST/zQy8dranpSYzWUhSLIAQDUap7rzaa9Fq/WEDqg+yYtE+QAYFy1NZdgcQ1bb7bS6x+fpllUn2pvelJ7EKV6ghwAjMN0qpHcfbCV8zfey6ff/HrO33gvdx9stT2kxbR2pb+m7PjpJKV/O4s1ZrOoPs0rhM7KvIOoD5DY45W2BwAAVTnoglaTgyT9EPfWnfez/eyjJMnWk+28def9JNFpchbWrsz+3JtF9Wkw5lqbhVy41v8Q58X3g1kF0cEHSIPfNfgAKann+WLqBDkAGIfpVIe6ee/hxyFuYPvZR7l576EgV6vjp/bpjjlh9WkeIXRW5hlEfYDEEIIcAIxjVhe0C+TDJ9tjHacC86w+1WReQdQHSAxhjRwAjKP2dT1zcGJ1yGbPBxynAvNai8dwtTeGYSZU5ABgHLWv65mDqxfP7FojlyS9lWO5evFMi6NiYjVPg6ydiihDCHIAMC4XtAcarIO7ee9hPnyynROrvVy9eMb6ODgqHyAxRGmapu0xDLW+vt7cv3+/7WEAVOfugy0X0ACwAEop32yaZn3Y91TkABaItu/MzeaG6gBAiwQ5gAWi7TtzYU+rVqm6A4mulQALRdt35uKgPa2YqUHVfevJdpo8r7rffbDV9tCAORPkABaItu/MhT2tWnNQ1R1YLoIcwAK5evFMeivHdh3T9p2ps6dVa1TdgQFBDmCBvHbuZN6+/GpOrvZSkpxc7eXty69aP8N02RS9NaruwIBmJwAL5rVzJwU3ZsueVq2x2TpHptPswhHkAIDx2RS9FTZb50h0ml1INgQHAIBFdutsP7ztdfx08sYH8x8PIztoQ3Br5ABgmjY3+hdN11f7t5sbbY8IWHY6zS4kQQ4ApmUwfenpoyTN8+lLwhzQJp1mF5IgBwDTYqNsoIt0ml1IghwATIvpS0AXrV1JLt3ur4lL6d9euq3RSeV0rQSAaTl+ap+GAqYvAS3TaXbhqMgBLLi7D7Zy/sZ7+fSbX8/5G+/l7oOttoe0uExfAmBOVOQAFtjdB1u7Ng/eerKdt+68nyT2nZoFG2UDMCeCHMACu3nv4cchbmD72Ue5ee+hIDcrpi8BMAemVgIssA+fbI91HACogyAHsMBOrPbGOg4A1EGQA1hgVy+eSW/l2K5jvZVjuXrxTEsjAgCmwRo5gAU2WAd3897DfPhkOydWe7l68Yz1cQBQOUEOYMG9du6k4AYAC8bUSgAAgMoIcgAAAJUR5AAAACojyAEAAFRGkAMAAKiMIAcAAFAZQQ4AAKAyghwAAEBlBDkAAIDKCHIAAACVEeQAAAAqI8gBAABURpADAACojCAHAABQGUEOAACgMoIcAABAZQQ5AACAykwU5EopnyylfKOU8ls7tz+8z899VEr51s5/X53kdwIAACy7SStybyZ5t2mazyR5d+f+MNtN0/zxnf9+asLfCQAAsNQmDXI/neQXdr7+hSSvTfh4AAAAHGLSIPejTdN8Z+fr/5bkR/f5uT9QSrlfSvm1UoqwBwCwrDY3kltnk+ur/dvNjbZHBFV65bAfKKX8cpIfG/KtL714p2mappTS7PMwP9E0zVYp5Y8mea+U8n7TNL8z5He9nuT1JPnUpz516OABAKjI5kbytS8mz7b7958+6t9PkrUr7Y0LKnRokGua5nP7fa+U8t9LKT/eNM13Sik/nuT39nmMrZ3b3y2l/Jsk55K8FOSapnknyTtJsr6+vl8oBACgRu9++XmIG3i23T8uyMFYJp1a+dUkP7Pz9c8k+creHyil/HAp5ffvfP0jSc4n+Y0Jfy8AALV5+ni848C+Jg1yN5L86VLKbyX53M79lFLWSyn/eOdn/liS+6WU/5zkV5LcaJpGkAMAWDbHT413HNjXoVMrD9I0zf9KcmHI8ftJfnbn63+X5NVJfg8AAAvgwrXda+SSZKXXPw6MZdKKHAAAjGbtSnLpdnL8dJLSv7102/o4OIKJKnIAADCWtSuCG0yBihwAAEBlBDkAAIDKCHIAAACVEeQAAAAqI8gBAABURpADAJi2zY3k1tnk+mr/dnOj7REBC8b2AwAA07S5sXvT66eP+vcTbfeBqVGRAwCYpne//DzEDTzb7h8HmBJBDgBgmp4+Hu84wBEIcgAA03T81HjHAY5AkAMAmKYL15KV3u5jK73+cYApEeQAAKZp7Upy6XZy/HSS0r+9dLvuRie6cELn6FoJADBta1fqDm4v0oWTadnc6Df9efq4P9X4wjXn0ARU5AAA2J8unEzD4AOBp4+SNM8/EFDdPTJBDgCA/enCyTT4QGDqBDkAAPanCyfT4AOBqRPkAADYny6cTIMPBKYWhAimAAAGjUlEQVROkAMAYH+L2IXzRTpyzocPBKZO10oAAA62SF04X6Qj5/wMnk9dK6dGkAMAYDkd1IBDwJi+Rf1AoCWmVgIAsJw04KBighwAAMtJAw4qJsgBALCcNOCgYoIcAADLadE7crLQNDsBAGB5acBBpVTkAAAAKiPIAQAAVEaQAwCANmxuJLfOJtdX+7ebG22PiIpYIwcAAPO2uZF87YvPNyR/+qh/P7Fmj5GoyAEAwLy9++XnIW7g2Xb/OIxAkAMAgHl7+ni847CHIAcAAPN2/NR4x2EPQQ4AAObtwrVkpbf72EqvfxxGIMgBAMC8rV1JLt1Ojp9OUvq3l25rdMLIdK0EAIA2rF0R3DgyFTkAAIDKCHIAAACVEeQAAAAqI8gBAABURpADAACojCAHAABQGUEOAACgMoIcAABAZQQ5AACAyghyAAAAlRHkAAAAKiPIAQAAVEaQAwAAqIwgBwAAUBlBDgAAoDKCHAAAQGUEOQAAgMoIcgAAAJUR5AAAACojyAEAAFRGkAMAAKiMIAcAAFAZQQ4AAKAyghwAAEBlBDkAAIDKCHIAAACVEeQAAAAqI8gBAABURpADAACojCAHAABQGUEOAACgMoIcAABAZQQ5AACAyghyAAAAlRHkAAAAKiPIAQAAVEaQAwAAqIwgBwAAUBlBDgAAoDKCHAAAQGUEOQAAgMoIcgAAAJUR5AAAACojyAEAAFRGkAMAAKiMIAcAAFAZQQ4AAKAyghwAAEBlBDkAAIDKCHIAAACVEeQAAAAqI8gBAABURpADAACojCAHAADM3uZGcutscn21f7u50faIqvZK2wMAAAAW3OZG8rUvJs+2+/efPurfT5K1K+2Nq2IqcgAAwGy9++XnIW7g2Xb/OEciyAEAALP19PF4xzmUIAcAAMzW8VPjHedQghwAADBbF64lK73dx1Z6/eMciSAHAADM1tqV5NLt5PjpJKV/e+m2RicT0LUSAACYvbUrgtsUqcgBAABURpADAACojCAHAABQGUEOAACgMoIcAABAZQQ5AACAyghyAAAAlRHkAAAAKiPIAQAAVEaQAwAAqIwgBwAAUBlBDgAAoDKCHAAAQGUEOQAAgMoIcgAAAJUR5AAAACojyAEAAFRGkAMAAKiMIAcAAFAZQQ4AAKAyghwAAEBlBDkAAIDKCHIAAACVEeQAAAAqI8gBAABURpADAACojCAHAABQGUEOAACgMoIcAABAZQQ5AACAyghyAAAAlRHkAAAAKlOapml7DEOVUv5Hkv+65/CPJPmfLQwHDuPcpKucm3SR85Kucm7SNT/RNM0fHvaNzga5YUop95umWW97HLCXc5Oucm7SRc5Lusq5SU1MrQQAAKiMIAcAAFCZ2oLcO20PAPbh3KSrnJt0kfOSrnJuUo2q1sgBAABQX0UOAABg6QlyAAAAlakqyJVSbpZSfrOUsllK+VellNW2xwRJUkr5i6WUXy+l/L9SirbFtK6U8vlSysNSym+XUt5sezyQJKWUf1pK+b1SygdtjwVeVEo5XUr5lVLKb+z8Pf9bbY8JDlNVkEvyjSRnm6ZZS/JfkrzV8nhg4IMkl5P8atsDgVLKsST/MMkXknw2yV8upXy23VFBkuSfJfl824OAIX6Q5O80TfPZJH8yyd/0vknXVRXkmqb5103T/GDn7q8lOdXmeGCgaZpvN03zsO1xwI4/keS3m6b53aZpvp/kXyT56ZbHBGma5leTfLftccBeTdN8p2ma/7Tz9f9J8u0kJ9sdFRysqiC3x19P8kttDwKgg04mefTC/cdxQQIwklLKH0lyLsl/aHckcLBX2h7AXqWUX07yY0O+9aWmab6y8zNfSr8E/ovzHBvLbZRzEwCoVynlDyb5l0n+dtM0/7vt8cBBOhfkmqb53EHfL6X8tSR/PsmFxiZ4zNFh5yZ0yFaS0y/cP7VzDIB9lFJW0g9xv9g0zZ22xwOHqWpqZSnl80n+bpKfaprm/7Y9HoCO+o9JPlNK+XQp5RNJ/lKSr7Y8JoDOKqWUJP8kybebpvn7bY8HRlFVkEvyD5L8oSTfKKV8q5Tyj9oeECRJKeUvlFIeJ/lTSb5eSrnX9phYXjtNoX4uyb30F+xvNE3z6+2OCpJSyj9P8u+TnCmlPC6l/I22xwQ7zif5q0l+cuca81ullD/b9qDgIMXsRAAAgLrUVpEDAABYeoIcAABAZQQ5AACAyghyAAAAlRHkAAAAKiPIAQAAVEaQAwAAqMz/BxofC15SyG2ZAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 1080x1080 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "G0hCZhK9rTYd"
      },
      "source": [
        "## Работа с настоящими отзывами"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zBJX3N2lXqUi"
      },
      "source": [
        "navec_news = Navec.load('navec_news_v1_1B_250K_300d_100q.tar')\n",
        "syntax = Syntax.load('slovnet_syntax_news_v1.tar')\n",
        "syntax.navec(navec_news)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "S4UhkzWIm52i",
        "outputId": "8a24101a-e445-4f0f-cd27-5aead810198e"
      },
      "source": [
        "chunk = []\n",
        "r = data.iloc[np.random.choice(range(len(data)))][\"text\"]\n",
        "rewiew = [_.text for _ in tokenize(r)]\n",
        "for sent in sentenize(r):\n",
        "  tokens = [_.text for _ in tokenize(sent.text)]\n",
        "  chunk.append(tokens)\n",
        "\n",
        "print(r)\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "for markup in syntax.map(chunk):\n",
        "  for token in markup.tokens:\n",
        "      w = token.text\n",
        "      m = morph.parse(w)\n",
        "      if m[0].tag.POS in [\"ADJF\", \"ADJS\"]:\n",
        "        n_f = m[0].normal_form\n",
        "        if n_f in emb_words_adjs:\n",
        "          e = emb_words_adjs[n_f]\n",
        "          pred_class  = estimator.predict(e.reshape(1, -1))[0]\n",
        "          pred_word_token = markup.tokens[int(token.head_id)-1]\n",
        "          m = morph.parse(pred_word_token.text)\n",
        "          if not m[0].tag.POS in [\"NOUN\", \"NPRO\"] or m[0].normal_form in ['который', 'свой', 'тот', 'этот', 'весь', 'такой', 'такого', 'какой-то', 'самый']:\n",
        "              continue\n",
        "          # while True:\n",
        "          #   m = morph.parse(pred_word_token.text)\n",
        "          #   if m[0].tag.POS in [\"NOUN\", \"NPRO\"]:\n",
        "          #     break\n",
        "          #   if pred_word_token.head_id == '0' or pred_word_token.head_id == pred_id:\n",
        "          #     break\n",
        "          #   pred_id = pred_word_token.id\n",
        "          #   pred_word_token = markup.tokens[int(pred_word_token.head_id)-1]\n",
        "\n",
        "          template = pred_word_token.text\n",
        "          if pred_class == 0:\n",
        "            #print(f\"{w}: proba_good-{proba_good}\")\n",
        "            print(f\"{template} {Back.GREEN+w+Back.RESET}\")\n",
        "            pass\n",
        "          elif pred_class == 2:\n",
        "            #print(f\"{w}: proba_bad-{proba_bad}\")\n",
        "            print(f\"{template} {Back.RED+w+Back.RESET}\")\n",
        "            \n",
        "\n",
        "  "
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Начав свою карьеру еще в конце 1980-х, лишь к середине следующего десятилетия, в 1995 году, режиссер Брайан Сингер приобретает огромную популярность за успех ленты «Подозрительные лица». Брайан показал себя очень талантливым кинематографистом, который, не имея больших финансовых средств, может нагнетать напряжение и в обыкновенной разговорной сцене, что под силу мне кажется, далеко не всем режиссерам. Та лента, помимо неплохого кассового успеха (все-таки бюджет у фильма вышел очень маленький) получила также и две премии Оскар: лучший сценарий и Лучшая мужская роль второго плана (Кевин Спейси). \n",
            " \n",
            " После такого успеха, Сингер решил продолжить работу в этом жанре и следующей своей работой он выбирает экранизацию очередной повести одного из самых экранизируемых писателей мира – Стивена Кинга. Сама же лента получилась, на мой взгляд, несколько неровной и чуток отторгающей. Во многом благодаря самому сюжету, который вышел довольно грубоватым и очень уж давящим, особенно учитывая такой непривычный так скажем, финал. \n",
            " \n",
            " Главным минусом фильма лично для меня стало то, что напряжение здесь нагоняется вокруг вообщем-то обыденных и не особенно пугающих эпизодов. Да, напряжение есть, но сам сюжет почти весь первый час не особенно то и развивается, из-за чего попросту становится скучно. Ты жаждешь узнать, что там дальше будет, а вместо этого режиссер делает акцент на том, как актер сейчас справляется в школе и так далее. Но радует то, что все же дальше сюжет развивается по принципу снежного кома: каждое действие или опрометчивый поступок главного героя в конце-концов приводит его к мягко говоря, не самым предсказуемым последствиям.\n",
            " \n",
            " Брэд Ренфро – потрясающе отыграл роль главного героя. Ученика, который жаждет чего-то большего, однако из-за постоянного общения о том бесчеловечном насилии, которое в свое время устроил Третий Рейх, он и сам наполняется ненавистью, которую со временем просто не сможет контролировать. Йен МакКеллен воистину зловещ в данной роли бывшего офицера Гитлеровской Германии. Его взгляд холоден, а действия пугающие. Именно такими мне кажется и были те самые убийцы во время Второй Мировой. Кстати, удивило присутствие Дэвида Швиммера в роли зачуханного учителя. Тогда этот актер был уже популярен из-за сериала «Друзья» и было странно видеть его в подобной роли.\n",
            " \n",
            " Способный ученик – исключительно разговорный триллер, в котором первый час вообще практически ничего не происходит и взамен этого режиссер предлагает медленное и плавное раскрытие персонажей и погружение зрителя в атмосферу. Не скажу, что прошло крайне удачно, все-таки осечки были и нудноватых моментов хватало, но я думаю, фильм заслуживает хотя бы однократного просмотра. Насладитесь вполне себе неплохим триллером. Жаль только, лента не может предложить чего-то большего.\n",
            " \n",
            " 7 из 10\n",
            "лица \u001b[41mПодозрительные\u001b[49m\n",
            "кинематографистом \u001b[42mталантливым\u001b[49m\n",
            "успеха \u001b[42mнеплохого\u001b[49m\n",
            "сценарий \u001b[42mлучший\u001b[49m\n",
            "роль \u001b[42mЛучшая\u001b[49m\n",
            "действие \u001b[41mопрометчивый\u001b[49m\n",
            "последствиям \u001b[42mпредсказуемым\u001b[49m\n",
            "насилии \u001b[41mбесчеловечном\u001b[49m\n",
            "убийцы \u001b[42mсамые\u001b[49m\n",
            "учителя \u001b[41mзачуханного\u001b[49m\n",
            "ученик \u001b[42mСпособный\u001b[49m\n",
            "триллером \u001b[42mнеплохим\u001b[49m\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}