01.pre-make-full.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "pre_make_full.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "machine_shape": "hm"
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QDPTBPTD11t2"
      },
      "source": [
        "# Fake News Detection with Adversarial Normalization\n",
        "\n",
        "Notebook with implemenation and testing for applying Adversarial Neural Networks to the task of Fake News Detection. Part of the project for Northeastern University's Khoury College course on Practical Neural Networks (DS 4440), during Fall 2020.\n",
        "\n",
        "This notebook preprocess the data generating usable datasets as `.csv` files\n",
        "\n",
        "\n",
        "Note: this notebook expects the data to be used to be located on a Google Drive account on a specific path. This *can* be modified.\n",
        "\n",
        "See README at https://github.com/pablokvitca/ds4440-project for more details."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "su80ZP7Umk24"
      },
      "source": [
        "# Imports\n",
        "import pandas as pd\n",
        "import numpy as np\n",
        "import ast\n",
        "from time import time\n",
        "from sklearn.model_selection import train_test_split\n",
        "from collections import Counter\n",
        "import seaborn as sns\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rBP9OY5SniV9",
        "outputId": "0fca532d-d870-4347-c106-102036e4e3e0"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RbhDkOQ0njvz"
      },
      "source": [
        "source_folder = '/content/drive/My Drive/Projects/Academic/NN/data/'"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pDMHhEfinscz"
      },
      "source": [
        "raw_data_path = source_folder + \"news_wkta_all.csv.zip\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "81sQTEcCn4mW",
        "outputId": "6abc1c01-ebd5-47fb-eb13-91aee1908d8c"
      },
      "source": [
        "df_raw = pd.read_csv(\n",
        "    raw_data_path, \n",
        "    nrows=2_000_000,\n",
        "    encoding=\"utf-8\", quotechar='\"',\n",
        "    low_memory=True, memory_map=True,\n",
        "    engine=\"c\", compression=\"zip\", \n",
        "    error_bad_lines=False, warn_bad_lines=True\n",
        ")\n",
        "df_raw.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(2000000, 307)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wtTecj2GoIwj"
      },
      "source": [
        "metadata_counts_path = source_folder + \"metadata_counts.python_output\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1iWoMV02oZ7O"
      },
      "source": [
        "counts_keywords = Counter()\n",
        "counts_tags = Counter()\n",
        "counts_authors = Counter()\n",
        "with open(metadata_counts_path, 'r') as file:\n",
        "    data = file.read().split(\"Counter(\")\n",
        "    \n",
        "    counts_keywords.update(ast.literal_eval(data[1][:-2]))\n",
        "    counts_tags.update(ast.literal_eval(data[2][:-2]))\n",
        "    counts_authors.update(ast.literal_eval(data[3][:-2]))\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9x6Se7E9wIYt",
        "outputId": "9ec89d1c-2c92-45ef-fa9b-2b27479f8cae"
      },
      "source": [
        "counts_keywords.most_common(25)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('climate change', 313727),\n",
              " ('business', 258123),\n",
              " ('energy', 220365),\n",
              " ('renewables', 192477),\n",
              " ('health', 105349),\n",
              " ('food', 100328),\n",
              " ('biodiversity', 98606),\n",
              " ('sustainable business', 97040),\n",
              " ('coal', 75628),\n",
              " ('fracking', 74660),\n",
              " ('politics', 73086),\n",
              " ('science', 66711),\n",
              " ('china', 42220),\n",
              " ('ebola', 33690),\n",
              " ('cuba', 33528),\n",
              " ('cop21', 32877),\n",
              " ('pipelines', 32852),\n",
              " ('russia', 32836),\n",
              " ('tips', 32300),\n",
              " ('transportation', 31966),\n",
              " ('donald trump', 26632),\n",
              " ('united states', 23972),\n",
              " ('barack obama', 20173),\n",
              " ('hillary clinton', 17902),\n",
              " ('economics', 17598)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "oZHUaLc8wN0g",
        "outputId": "d9ea839d-3e5e-4db3-85f3-6c96d17742e6"
      },
      "source": [
        "counts_tags.most_common(25)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('russia', 65416),\n",
              " ('donald trump', 60004),\n",
              " ('hillary clinton', 42836),\n",
              " ('syria', 35145),\n",
              " ('9/11 articles', 31523),\n",
              " ('race/crime articles', 31523),\n",
              " ('american pravda articles', 31523),\n",
              " ('race/iq articles', 31523),\n",
              " ('israel', 31399),\n",
              " ('libya', 30813),\n",
              " ('united states', 30101),\n",
              " ('immigration', 29121),\n",
              " ('politics', 28905),\n",
              " ('barack obama', 25986),\n",
              " ('trump', 25812),\n",
              " ('obama', 24607),\n",
              " ('isis', 24055),\n",
              " ('europe', 23574),\n",
              " ('iran', 23016),\n",
              " ('china', 22821),\n",
              " ('cia', 22379),\n",
              " ('benghazi', 22064),\n",
              " ('children', 21172),\n",
              " ('terrorism', 20555),\n",
              " ('islam', 20424)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "TWfeibmfwOHF",
        "outputId": "57c77065-0a2f-45f0-92d4-6a44386aa9a3"
      },
      "source": [
        "counts_authors.most_common(25)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('lorraine chow', 746451),\n",
              " ('common dreams', 638516),\n",
              " ('guest contributor', 507058),\n",
              " ('the conversation', 384011),\n",
              " ('david leestma', 362631),\n",
              " ('environmental working group', 254739),\n",
              " ('ronnie cummins', 254659),\n",
              " ('the climate reality project', 254658),\n",
              " ('sierra club', 254636),\n",
              " ('rainforest action network', 254636),\n",
              " ('backgroundurl avatar_large', 224956),\n",
              " ('organic consumers association', 156531),\n",
              " (\"our children's trust\", 107978),\n",
              " ('american wild horse campaign', 107975),\n",
              " ('anastasia pantsios', 101821),\n",
              " ('stefanie spear', 98031),\n",
              " ('climate nexus', 82670),\n",
              " ('daily kos', 76786),\n",
              " ('happy cog studios - http', 76783),\n",
              " ('www.happycog.com', 76783),\n",
              " ('david codrea', 70085),\n",
              " ('tom mchale', 70085),\n",
              " ('jeff knox', 70085),\n",
              " ('the revelator', 49938),\n",
              " ('climate group', 49922)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OVVG8I8aqN3u"
      },
      "source": [
        "def filter_data_by_meta(df, prefix, meta_counts, n):\n",
        "    req_cols = [prefix + meta for meta, _ in meta_counts.most_common(n)]\n",
        "    \n",
        "    all_criteria = [df[col] == 1 for col in req_cols]\n",
        "    criteria = df[req_cols[0]] == 1\n",
        "    for col in req_cols[1:]:\n",
        "        criteria = criteria | df[col] == 1\n",
        "        \n",
        "    return df.loc[criteria]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "kYfZg43wkqS5",
        "outputId": "5e57504a-f18d-402a-c354-ae2b95dcfcd8"
      },
      "source": [
        "df_raw.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(2000000, 307)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rv9l_UT9hj7c",
        "outputId": "d4f05205-c428-4c88-92c6-3011ff974c5d"
      },
      "source": [
        "df_raw.drop_duplicates(inplace=True)\n",
        "df_raw.reset_index()\n",
        "df_raw.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1302734, 307)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "34fSPwYfw4U1"
      },
      "source": [
        "filtered_data = filter_data_by_meta(df_raw, \"has_mk_\", counts_keywords, 20)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8cEbZZGCw7Ex",
        "outputId": "fc081365-d266-41fe-9c7e-9c6f2baa38ce"
      },
      "source": [
        "filtered_data.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(39063, 307)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Gq4Aa82u3ImR"
      },
      "source": [
        "filtered_data = filtered_data.copy()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "r0en3TqMxPH2",
        "outputId": "06170f94-746b-4e6c-84df-3e835071395a"
      },
      "source": [
        "len(df_raw.domain.unique())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "379"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "FPSb-QOnyBi-",
        "outputId": "e007b39d-49c5-4a9a-890b-de269a7d254c"
      },
      "source": [
        "len(filtered_data.domain.unique())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "66"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eMmTYGq7hgIl"
      },
      "source": [
        "del df_raw"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yStmnHWKyLgD",
        "outputId": "1eff4953-2744-4f11-89c8-b5875a862c60"
      },
      "source": [
        "print(filtered_data.fake.value_counts(normalize=False))\n",
        "print(filtered_data.fake.value_counts(normalize=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1    23837\n",
            "0    15226\n",
            "Name: fake, dtype: int64\n",
            "1    0.610219\n",
            "0    0.389781\n",
            "Name: fake, dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3kMV89JHzIDX"
      },
      "source": [
        "filtered_data.to_csv(source_folder + \"news_filtered.csv\", index=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K7B0fcKT2Z3u"
      },
      "source": [
        "def clean_join_text_cols(df, col_a, col_b, out_col):\n",
        "    def clean_text_col(d):\n",
        "        return \" \\n \".join(d.splitlines())\n",
        "    df[out_col] = df[[col_a, col_b]].apply(lambda r: clean_text_col(r[col_a]) + \" <ENDTITLE> \\n \" + clean_text_col(r[col_b]), axis=1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RlgcmD402d7W"
      },
      "source": [
        "clean_join_text_cols(filtered_data, \"title\", \"content\", \"title_content\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3rpoZOgL3LA7"
      },
      "source": [
        "filtered_data[\"content_length\"] = filtered_data.content.apply(lambda d: len(d.split()))\n",
        "filtered_data[\"title_length\"] = filtered_data.title.apply(lambda d: len(d.split()))\n",
        "filtered_data[\"total_length\"] = filtered_data.content.apply(lambda d: len(d.split()))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XDGzhouIhPNy",
        "outputId": "1da75593-0284-40f1-b6c1-6c6c35c10135"
      },
      "source": [
        "filtered_data.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(39063, 311)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4n1Ycw_phH2w",
        "outputId": "29684354-cef5-470b-ee9a-83900ffa06d1"
      },
      "source": [
        "no_dups = filtered_data.drop_duplicates()\n",
        "no_dups.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(39063, 311)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "z3wUFu7g3q6E",
        "outputId": "9a19b92e-2227-46e2-dc58-5f3f2a4cb43c"
      },
      "source": [
        "filtered_data[\"total_length\"].describe()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "count    39063.000000\n",
              "mean       467.514707\n",
              "std        623.874759\n",
              "min          7.000000\n",
              "25%        149.000000\n",
              "50%        312.000000\n",
              "75%        603.000000\n",
              "max      17642.000000\n",
              "Name: total_length, dtype: float64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 28
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 349
        },
        "id": "NsrFK7kL4wZk",
        "outputId": "c0a67605-72b5-402f-9b89-11b0c56e7b66"
      },
      "source": [
        "sns.distplot(filtered_data[\"total_length\"])\n",
        "# plt.xlim([0, 1024]);\n",
        "plt.xlabel('Token count');\n",
        "plt.ylabel('Document count')\n",
        "plt.title(\"Distribution of Total Token Lengths\")\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/seaborn/distributions.py:2551: FutureWarning: `distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).\n",
            "  warnings.warn(msg, FutureWarning)\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZQAAAEWCAYAAABBvWFzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3deZxkZX3v8c+319lXGpABnFEGdHCLjqiJO5HtGia5Lgx6FRQlUdC4JAoxlxASEtGor2uEGBQCIjggbm2C4oIJbgwMCsiAg+0My7AOw+zM0lX1u3+cp2Zqaqq6q7uquqervu/Xq19T9ZxznvOc0z31q2c9igjMzMzq1THeBTAzs9bggGJmZg3hgGJmZg3hgGJmZg3hgGJmZg3hgGJmZg3hgGLDkvRFSf+3QXkdLmmrpM70/r8lvacReaf8vifptEblN4Lz/qOkJyU9NtbnTuc/XdLPGpjf+ZK+2qj89meSXitp7XiXoxU4oLQ5SfdL2i5pi6SNkn4h6S8k7f7biIi/iIh/qDGvPx5qn4h4MCKmRUS+AWXf50MvIk6MiCvrzXuE5Tgc+CiwKCIOLtv29hRAt6b7XCh5v3WIPK+Q9I8NKNurSs63TVKUnj+Vfdyk8hzR6udsFw4oBvAnETEdeCbwSeDjwGWNPomkrkbnuZ84HFgfEU+Ub4iIq1MAnQacCDxSfJ/SmioiflpyrqNT8qySMjzY7DJY+3BAsd0iYlNE9AOnAKdJeh7s/W1Z0gGS/jPVZp6S9FNJHZKuIvtg/W765vsxSfPTt8EzJD0I3FSSVhpcni3pVkmbJX1H0px0rn2aIoq1IEknAH8DnJLOd2favrsJLZXrbyU9IOkJSV+RNDNtK5bjNEkPpuaqT1S7N5JmpuPXpfz+NuX/x8APgUNSOa6o9X5Lem4q70ZJKyWdnNLPBN4OfCzl+d2Ufo6k36fa5D2S/qzWc1U5/yGS+tPvcUDSe6vs1y3pa5K+IaknHfeNdC/WSPpgyb7nS7ou3ast6boWj6JsvZL+Jf1uHlfW7Do5bXutpLWSPpp+r49KelfJsXMlfTf9Pd2mrDnyZ2nbzWm3O9O9PaXkuGr5nZTu9xZJD0v6q5FeT7twQLF9RMStwFrgVRU2fzRt6wMOIvtQj4h4B/AgWW1nWkR8quSY1wDPBY6vcsp3Au8GngHkgM/XUMbvA/8EXJvO98IKu52efl4HPAuYBnyhbJ9XAkcBxwLnSXpulVP+KzAz5fOaVOZ3RcSP2LvmcfpwZYfsQxr4LvAD4EDgA8DVko6KiEuBq4FPpTz/JB32e7LfyUzg74GvSnpGLeerYhnZ7/IQ4M3AP0l6fVk5JwPfBnYCbyX7/XwXuBOYR3bfPiSp9Hd7csp7FtDPvve8Fp8EjgReBByRznVeyfaDye7DPOAM4GJJs9O2i4FtaZ/T0g8AEfHq9PKF6d5eW0N+lwF/nmrxzwNuGsX1tAUHFKvmEWBOhfRBsg/+Z0bEYGpSGW5BuPMjYltEbK+y/aqIuDsitgH/F3irUqd9nd4OfDYiVkfEVuBcYGlZ7ejvI2J7RNxJ9iG5T2BKZVkKnBsRWyLifuAzwDvqKNvLyQLcJyNiV0TcBPwncGq1AyLi6xHxSEQU0gfh74BjRnNySYcBfwR8PCJ2RMQdwJfJAmXRDOD7ZIHsXanf66VAX0RckMq9GvgS2f0p+llE3JD2v4oK93SYsgk4E/hwRDwVEVvIvjyUnmMQuCD9Dd4AbAWOSr+rNwF/FxFPR8Q9QC19ahXzK9m2SNKMiNgQEb8ayfW0EwcUq2Ye8FSF9E8DA8APJK2WdE4NeT00gu0PAN3AATWVcmiHpPxK8+4iq1kVlY7KeprsQ77cAalM5XnNq7NsD0VEodY8Jb1T0h2piWwj2bfl0d6nQ4Dih3W1878ceAFZ0Ct+aXgmWfPexpJy/A1D39NJGln/WR8wBbi95BzfT+lF6yMiV3aeaWmfLvb+mxru72+o/CALUCcBD0j6H0mvGMG1tBUHFNuHpJeSfbDsMww1fUP/aEQ8i6xp4yOSji1urpLlcDWYw0peH072jfBJsmaLKSXl6mTvD5Xh8n2E7AOwNO8c8Pgwx5V7MpWpPK+HR5hPedkOU8lourI897o2Sc8kqwmcDcyNiFnA3YDqOP8cSdOrnB+y5rh/Bn4sqRgwHgLWRMSskp/pEXHSKMtRyZPAduDoknPMrHEQwzqy3/GhJWmHVdm3JhFxW0QsIWua/DZwXT35tTIHFNtN0gxJbyRr//5qRPymwj5vlHREapbYBOSB4rfsx8n6GEbq/0haJGkKcAFwfWouuY/s2+3/Sn0Ofwv0lhz3ODC/7EO51NeAD0taIGkae/pcclX2ryiV5TrgQknT04f7R4B65mksJ/sW/LHU6f1a4E/I7j3sey+nkgWZdQCp0/h5oz15RDwE/AL4Z0mTJL2ArO+gfBj2p4BryILKAcCtwBZJH5c0WVKnpOelLyGj1ZPKMEnSJLIg+SXgc5IOBJA0r6yfptp15YFvAudLmiLpOezdjAcj+DtNgxDeLmlmRAwCm9nz925lHFAMspFZW8i+fX4C+Czwrir7LgR+RNbG/Evgkoj4Sdr2z8DfpmaKkYyEuQq4gqypZBLwQchGnQHvJ2vbf5isxlI66uvr6d/1kiq1a1+e8r4ZWAPsIOv8Ho0PpPOvJqu5XZPyH5WI2EUWQE4k+0Z+CfDOiPht2uUysnb7jZK+nfoCPkN2zx8Hng/8fLTnT04F5pPVVr5F1u/wowpl/Qeyb+Y/Iuu4fiNZZ/maVPYvp/TRWklWIyn+vIts6PoAcIukzencR1XNYW9np/I8Rvb7/xrZoIKi84Er0719aw35vQO4P5XjL8j65qwC+QFbZtbKJF0EHBwRY76CQrtxDcXMWoqk50h6gTLHkDXlfWu8y9UOWnXmspm1r+lkzVyHkDUPfgb4zriWqE24ycvMzBrCTV5mZtYQbd3kdcABB8T8+fPHuxhmZhPK7bff/mRE9JWnt3VAmT9/PitWrBjvYpiZTSiSHqiU7iYvMzNriKYGFEknSFqlbGnsfdZ8UrZE9bVp+3JJ80u2nZvSV5XOkJV0eVpi+u4K+X1A0m+VLZn9qfLtZmbWPE0LKGndpYvJZgIvAk6VtKhstzOADRFxBPA54KJ07CKylUWPBk4ALilZffaKlFZ+vtcBS8iWpT4a+JdGX5OZmVXXzBrKMcBAWjp8F9kaRUvK9lnCnqWlrweOTWtELQGWRcTOiFhDtgTDMQARcTOVV8F9H9mqqDvTfvs8Pc/MzJqnmQFlHnsvG72WfZfm3r1PWrBvEzC3xmPLHQm8KjWd/U+1xeoknSlphaQV69atq/lizMxsaK3UKd9F9kColwN/DVyXajt7iYhLI2JxRCzu69tn1JuZmY1SMwPKw+z9HIJD2ff5Ebv3SQ/gmQmsr/HYcmuBb0bmVrIlphvxkCYzM6tBMwPKbcDC9CyKHrJO9v6yffrZ87znNwM3pSfD9ZM9qrVX0gKyJdNvHeZ83yZ7djiSjgR6yJbWNjOzMdC0gJL6RM4GbgTuBa6LiJWSLpB0ctrtMmCupAGyBxadk45dSfZAo3vIHv15VnpwDpK+RvZMiKMkrZV0RsrrcuBZaTjxMuC0Gp51bmZmDdLWi0MuXrw4GjVT/prlD1ZMf9vLDm9I/mZm+wtJt0fE4vL0VuqUNzOzceSAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGAYmZmDeGA0mCFCL7xq7Xc++jm8S6KmdmYckBpsOVrnuL2Bzaw6rEt410UM7Mx1dSAIukESaskDUg6p8L2XknXpu3LJc0v2XZuSl8l6fiS9MslPZGeHV/pnB+VFJIOaMY1DWXz9kF+sPIxAHbm8mN9ejOzcdW0gCKpE7gYOBFYBJwqaVHZbmcAGyLiCOBzwEXp2EXAUuBo4ATgkpQfwBUprdI5DwOOAyo/4L3Jfv3gBnbmCkzr7WJnrjAeRTAzGzfNrKEcAwxExOqI2AUsA5aU7bMEuDK9vh44VpJS+rKI2BkRa4CBlB8RcTPwVJVzfg74GBANvZIabdg+yJSeTuZO7WGXA4qZtZlmBpR5wEMl79emtIr7REQO2ATMrfHYvUhaAjwcEXcOs9+ZklZIWrFu3bparqNmm7cPMnNyN73dHa6hmFnbaYlOeUlTgL8Bzhtu34i4NCIWR8Tivr6+hpZj8/ZBZkzqpqer0zUUM2s7zQwoDwOHlbw/NKVV3EdSFzATWF/jsaWeDSwA7pR0f9r/V5IOrqP8I7Zp+yAzJnfT29nhTnkzazvNDCi3AQslLZDUQ9bJ3l+2Tz9wWnr9ZuCmiIiUvjSNAlsALARurXaiiPhNRBwYEfMjYj5ZE9mLI+Kxxl5Sdbl8gW278syc3EWPm7zMrA01LaCkPpGzgRuBe4HrImKlpAsknZx2uwyYK2kA+AhwTjp2JXAdcA/wfeCsiMgDSPoa8EvgKElrJZ3RrGsYic07cgBZH0pXB7tyBbLYaGbWHrqamXlE3ADcUJZ2XsnrHcBbqhx7IXBhhfRTazjv/JGWtV6btg8CMGNSN1t35AhgMO+AYmbtoyU65fcHm3ekgDK5m57ubMqM+1HMrJ04oDTI5lRDKTZ5AR7pZWZtxQGlQTZvH6Snq4Perg56OrPb6o55M2snDigNsmn7IDMndSOJ3m7XUMys/TigNMjmHTlmTM7GOPR2FftQHFDMrH04oDTIprTsCkBPV7HJy53yZtY+HFAaICLYsmOQ6ZOygOJOeTNrRw4oDVCI7Ke7U8CegOImLzNrJw4oDTCYzwJHp7KA0uOAYmZtyAGlAXKFbEZ8R0cWULo6OujskJu8zKytOKA0QK5YQ0kBBbJmL3fKm1k7cUBpgOKaXR3aE1B60gKRZmbtwgGlAfKpyatT5TUUBxQzax8OKA1Q7JTv2KvJy09tNLP24oDSAMVO+c6Su9njPhQzazMOKA1Q7JTvcJOXmbUxB5QG2D1suCyg7Mo7oJhZ+2hqQJF0gqRVkgYknVNhe6+ka9P25ZLml2w7N6WvknR8Sfrlkp6QdHdZXp+W9FtJd0n6lqRZzby2Url8sclr71FeOwcdUMysfTQtoEjqBC4GTgQWAadKWlS22xnAhog4AvgccFE6dhGwFDgaOAG4JOUHcEVKK/dD4HkR8QLgPuDchl7QEAYLleahuFPezNpLM2soxwADEbE6InYBy4AlZfssAa5Mr68HjpWklL4sInZGxBpgIOVHRNwMPFV+soj4QUTk0ttbgEMbfUHV5CrMQ+nt6iAf4aBiZm2jmQFlHvBQyfu1Ka3iPikYbALm1njsUN4NfK/SBklnSlohacW6detGkGV1uzvly0Z5AWzbmat0iJlZy2m5TnlJnwBywNWVtkfEpRGxOCIW9/X1NeScuQoTG4uPAd62ywHFzNpDMwPKw8BhJe8PTWkV95HUBcwE1td47D4knQ68EXh7RMRoCz5SuQp9KMUayvZdnotiZu2hmQHlNmChpAWSesg62fvL9ukHTkuv3wzclAJBP7A0jQJbACwEbh3qZJJOAD4GnBwRTzfwOoZVbS0vgKcdUMysTTQtoKQ+kbOBG4F7gesiYqWkCySdnHa7DJgraQD4CHBOOnYlcB1wD/B94KyIyANI+hrwS+AoSWslnZHy+gIwHfihpDskfbFZ11au4rBhN3mZWZvpambmEXEDcENZ2nklr3cAb6ly7IXAhRXST62y/xF1FbYOxSavSjUUN3mZWbtouU758TBUDcVNXmbWLhxQGmBPDWVP2p4+FDd5mVl7cEBpgEHXUMzMHFAaodJqw90e5WVmbWbYgCLpolrS2tme56HsCShdHaJDbvIys/ZRSw3lDRXSTmx0QSaySsvXS6K7s8M1FDNrG1WHDUt6H/B+4FmS7irZNB34ebMLNpEUm7xKayiQdcx72LCZtYuh5qFcQ7bA4j+TJhwmWyJin9V+29memfJ7p/d0drDNAcXM2kTVgBIRm8hW/z01PYvkoLT/NEnTIuLBMSrjfi9XKNChrJmrVFZDcR+KmbWHYWfKSzobOB94HCg+3COAFzSvWBNLLh979Z8U9bgPxczaSC1Lr3wIOCoi1je7MBNVrhD79J9AVkNxQDGzdlHLKK+HyJq+rIpcvlC5htLV4WHDZtY2aqmhrAb+W9J/ATuLiRHx2aaVaoIZLAQdlWoonR1s2j44DiUyMxt7tQSUB9NPT/qxMrl8gc594wndHjZsZm1k2IASEX8/FgWZyHL5Kn0onR1+HoqZtY1aRnn9hGxU114i4vVNKdEElCtUGeXV1cGOwQKFKk1iZmatpJYmr78qeT0JeBPgr90lcoVC1RoKwPbBPFN7m/osMzOzcVdLk9ftZUk/lzTk893bzWC1eShdex4D7IBiZq2ultWG55T8HCDpeGBmLZlLOkHSKkkDks6psL1X0rVp+3JJ80u2nZvSV6VzFtMvl/SEpLsrlPOHkn6X/p1dSxkbIZevUkPxY4DNrI3UMg/ldmBF+veXwEeBM4Y7KC3XcjHZysSLyJZwWVS22xnAhvQ8+M8BF6VjFwFLgaOBE4BLUn4AV6S0cucAP46IhcCP2Xv9sabK+lD2TfdDtsysnQwbUCJiQUQ8K/27MCKOi4if1ZD3McBARKyOiF3AMmBJ2T5LgCvT6+uBY5UtiLUEWBYROyNiDTCQ8iMibgYqLU5ZmteVwJ/WUMaGqDrKy48BNrM2UkuTV7ekD0q6Pv2cLam7hrznkc2yL1qb0iruExE5shn5c2s8ttxBEfFoev0Y2WKWla7nTEkrJK1Yt25dDZcxvFyhUHEUV7drKGbWRmpp8vo34CXAJennJSltvxURQYWhzmnbpRGxOCIW9/X1NeR8g/mgs0KnfK8fA2xmbaSWoUcvjYgXlry/SdKdNRz3MHBYyftDU1qlfdZK6iLr7F9f47HlHpf0jIh4VNIzgCdqKGNDZMvXDzFs2AHFzNpALTWUvKRnF99IehZQyyfkbcBCSQsk9ZB1sveX7dMPnJZevxm4KdUu+oGlaRTYAmAhMNxQ5dK8TgO+U0MZG6JaH0p3ybBhM7NWV0sN5a+Bn0haDQh4JvCu4Q6KiFx6lsqNQCdweUSslHQBsCIi+oHLgKskDZB1tC9Nx66UdB1wD9kkyrMiIg8g6WvAa4EDJK0F/i4iLgM+CVwn6QzgAeCttd6EeuUKwaTufWOzayhm1k5qmdj4Y0kLgaNS0qqI2DnUMSXH3gDcUJZ2XsnrHcBbqhx7IXBhhfRTq+y/Hji2lnI1Wi5foKOnc5/0HvehmFkbqWWU11nA5Ii4KyLuAqZIen/zizZxVOuU7+yQF4g0s7ZRSx/KeyNiY/FNRGwA3tu8Ik081dbyApjc0+kmLzNrC7UElM402RDYPQPez0UpkR9iNeGpPZ1s2+mAYmatr5ZO+e8D10r69/T+z1OaJdWavABmTulh0/ZdY1wiM7OxV0tA+ThwJvC+9P6HwJebVqIJKHumfOVtc6Z289Q2BxQza321jPIqAF9MP1bBYKHyPBSAWVN6eGTj5jEukZnZ2KulD8WGkctXXssLYM6UHjY87RqKmbU+B5Q6FQpBIajahzJ7ag+btg+SL1RcWszMrGXUMg9ln4mHldLaVS4FimpNXrOndBMBm7YPjmWxzMzGXC01lHNrTGtLuUIBoOLikABzpmYjrN0xb2atrmqnvKQTgZOAeZI+X7JpBtn6WkY2ZBio2ocye0oWUNyPYmatbqhRXo+QPfr3ZLLH/xZtAT7czEJNJLl8VkPprDJseHdAcQ3FzFpc1YASEXcCd0q6JiLcAVBFsbO9ag1lavZwS9dQzKzV1TKx8RhJ55MtW99FtoR9RMSzmlmwiWKw2ClfbZTXlGIfimOymbW2WgLKZWRNXLdT24O12kqxyataDWVKTyc9XR1sdA3FzFpcLQFlU0R8r+klmaCKnfLVaiiSmDOlx6O8zKzl1RJQfiLp08A3gd0P1oqIXzWtVBPI7mHD1RbzAmZN6XYfipm1vFrmobwMWAz8E/CZ9PMvtWQu6QRJqyQNSDqnwvZeSdem7cslzS/Zdm5KXyXp+OHylHSspF9JukPSzyQdUUsZ65UbpoYC2VyUDU+7D8XMWlsti0O+bjQZp+emXAy8AVgL3CapPyLuKdntDGBDRBwhaSlwEXCKpEVkz5c/GjgE+JGkI9Mx1fL8N2BJRNybnij5t8Dpoyn7SOR2j/Kqvs/sqT3c+4gXiDSz1lbL0isHSbpM0vfS+0WSzqgh72OAgYhYHRG7gGXAkrJ9lgBXptfXA8emh3ktAZZFxM6IWAMMpPyGyjPIJl0CzCSbR9N0e+ahVK+hzHaTl5m1gVqavK4AbiSrKQDcB3yohuPmAQ+VvF+b0iruExE5YBMwd4hjh8rzPcANktYC7wA+WalQks6UtELSinXr1tVwGUMbbqY8ZCsOb/QCkWbW4moJKAdExHVAAXZ/8O+Pw4c/DJwUEYcC/wF8ttJOEXFpRCyOiMV9fX11nzQ/zDwUyJq8vECkmbW6WgLKNklzyZqUkPRysprEcB4GDit5f2hKq7iPpC6ypqr1QxxbMV1SH/DCiFie0q8F/rCGMtZtsIZRXnsWiNxZdR8zs4muloDyEaAfeLaknwNfAT5Qw3G3AQslLZDUQ9bJ3l+2Tz9wWnr9ZuCmiIiUvjSNAlsALARuHSLPDcDMko77NwD31lDGutUyyuugGZMAeHTTjrEokpnZuKhllNevJL0GOIps2ZVVtaztFRE5SWeT9b90ApdHxEpJFwArIqKfbBb+VZIGgKfIAgRpv+uAe8hWNj4rIvIAlfJM6e8FviGpQBZg3j2SGzFae2bKV99n3qzJADyycftYFMnMbFwMG1DS8N+TgPlp/+MkEREV+yhKRcQNwA1laeeVvN4BVHxYV0RcCFxYS54p/VvAt4YrU6MNt5YXZDUUCR7Z6BqKmbWuWmbKfxfYAfyG1DFve+RTH0q1JzZes/xBAKb1dvGzgSd3N3+97WWHj00BzczGSC0B5dCIeEHTSzJB7R42PEQNBWDm5G42eba8mbWwWjrlvyfpuKaXZILK1TAPBWDW5G42etiwmbWwWgLKLcC3JG2XtFnSFkleRyTJDdPkVTRzcjebtu8iG8RmZtZ6agkonwVeAUyJiBkRMT0iZgx3ULsYbvn6ollTehjMB9t37Y9zQs3M6ldLQHkIuDv81bqifGH4YcOQ1VAAN3uZWcuqpVN+NfDfaXHI0uehDDtsuB3U2ik/a0oWUDZtH+SQNC/FzKyV1BJQ1qSfnvRjJXbPlK+hDwVcQzGz1lXLTPm/H4uCTFS5QgFp+BrK1N4uOjvEJi9jb2YtqpaZ8j8hLQxZKiJe35QSTTCD+aB7uA4UsoAz00OHzayF1dLk9VclrycBbyJbX8vI1vLq6hy6dlI0fVIXW3f41plZa6qlyev2sqSfS7q1SeWZcHKFoGuY/pOiab1drNviJezNrDXV0uQ1p+RtB/ASsueWGDCYL9DdWcvo66wfZc2T25pcIjOz8VFLk9ftZH0oImvqWgPU8kz5tpDLR81NXtN6u9i+K+9HAZtZS6qlyWvBWBRkohos1F5DmdbbRQBP73I/ipm1nmE/CSWdJWlWyfvZkt7f3GJNHIP5GFGTF8DWnQ4oZtZ6avkkfG9EbCy+iYgNwHubV6SJJZcvjKhTHmDbTq/nZWatp5aA0intmbWXnuBY04x5SSdIWiVpQNI5Fbb3Sro2bV8uaX7JtnNT+ipJxw+XpzIXSrpP0r2SPlhLGes1mA+6RtDkBbB1p+eimFnrqaVT/vvAtZL+Pb3/85Q2pBR4LgbeAKwFbpPUHxH3lOx2BrAhIo6QtBS4CDhF0iKy58sfDRwC/EjSkemYanmeDhwGPCciCpIOrOHa6pYrFOgeQac8wFbXUMysBdUSUD5OFkTel97/EPhyDccdAwxExGoAScuAJUBpQFkCnJ9eXw98IdWGlgDLImInsEbSQMqPIfJ8H/C2iCgARMQTNZSxbrl87fNQJnV30CmxzX0oZtaCahnlVZB0GfAzsuHDqyKilq/Y88iWvi9aC7ys2j4RkZO0CZib0m8pO3Zeel0tz2eT1W7+DFgHfDAifldDOesyknkokpja2+nZ8mbWkmoZ5fVa4HfAF4BLgPskvbrJ5RqNXmBHRCwGvgRcXmknSWdKWiFpxbp16+o+6UgCCmTNXh7lZWatqJZPws8Ax0XEayLi1cDxwOdqOO5hsj6NokNTWsV9JHWRzcBfP8SxQ+W5Fvhmev0t4AWVChURl0bE4ohY3NfXV8NlDC1XqH1iI2RDh7d5HoqZtaBaAkp3RKwqvomI+4DuGo67DVgoaYGkHrJO9v6yffqB09LrNwM3pSdD9gNL0yiwBcBC4NZh8vw28Lr0+jXAfTWUsW6D+aCrhtWGi6b1eoFIM2tNtXTKr5D0ZeCr6f3bgRXDHZT6RM4GbgQ6gcsjYqWkC4AVEdEPXAZclTrdnyILEKT9riPrbM8BZxX7bSrlmU75SeBqSR8GtgLvqeHa6pbL1z7KC/Y0eUUEGuYZKmZmE0ktAeV9wFlAcV7HT8n6UoYVETcAN5SlnVfyegfwlirHXghcWEueKX0j8L9qKVcjZU1etddQpvZ2kSsE23bldw8jNjNrBbWM8top6Srgqoiovxe7xQyOtIYyKbvlT27Z6YBiZi2l6lfrNPP8fElPAquAVZLWSTqv2jHtaDBfqOmJjUVTe7Igsn6bn4tiZq1lqE/CDwN/BLw0IuZExByyOR9/lPopjJEtXw8wpacTgI1Pe/kVM2stQwWUdwCnRsSaYkKaof5/gHc2u2ATxUjnoRQDygYHFDNrMUN9EnZHxJPliakfpZZhw21hJI8ABpiSmrw2Pr2rWUUyMxsXQwWUoT7x/GmY5Eaw2jBk63l1CDY4oJhZixlqmNELJW2ukC5gUpPKM6FEBIOFAj0j6EORxOTuTjd5mVnLqRpQIqJzLAsyEeULQQQjqqFA1uzlJi8zazUj+yS0veQKATCiUV6Qdcxv2OYaipm1FmNmk7cAABQOSURBVAeUOgzmCwAjmocCKaC4hmJmLcYBpQ65/GhrKF2eh2JmLccBpQ6DhVRDGXEfimsoZtZ6HFDqUKyhjGQtL8gCys5cge27/Gx5M2sdDih1KPahjOR5KLBncqNrKWbWShxQ6jA4yj6UybuXX3FAMbPW4YBSh9xo+1B6vUCkmbUeB5Q67B7lNYK1vMBNXmbWmhxQ6rB7HkrXyEd5gWsoZtZamhpQJJ0gaZWkAUnnVNjeK+natH25pPkl285N6askHT+CPD8vaWuzrqlUcab8aCY2glccNrPW0rSAIqkTuBg4EVgEnCppUdluZwAbIuII4HPARenYRcBS4GjgBOASSZ3D5SlpMTC7WddUbjCXRnmNsFO+q6ODqT1eINLMWkszayjHAAMRsToidgHLgCVl+ywBrkyvrweOlaSUviwidqYHfA2k/KrmmYLNp4GPNfGa9jJYGN08FIBZU3rch2JmLaWZAWUe8FDJ+7UpreI+EZEDNgFzhzh2qDzPBvoj4tGhCiXpTEkrJK1Yt27diC6oXG6U81AAZk/tdh+KmbWUluiUl3QI8BbgX4fbNyIujYjFEbG4r6+vrvOOdh4KwJypvazfurOu85uZ7U+aGVAeBg4reX9oSqu4j6QuYCawfohjq6X/AXAEMCDpfmCKpIFGXUg1xXkoPSOchwJw4PRentjigGJmraOZAeU2YKGkBZJ6yDrZ+8v26QdOS6/fDNwUEZHSl6ZRYAuAhcCt1fKMiP+KiIMjYn5EzAeeTh39TbVnteHRBZQnt+6kkPphzMwmuqEeAVyXiMhJOhu4EegELo+IlZIuAFZERD9wGXBVqk08RRYgSPtdB9wD5ICzIiIPUCnPZl3DcHbt7kMZeZNX3/ReBvPBxu2DzJna0+iimZmNuaYFFICIuAG4oSztvJLXO8j6PiodeyFwYS15Vthn2mjKO1J7VhseTQ1lEgBPbNnhgGJmLaElOuXHS7EPZTSd8n3TewFY534UM2sRDih1GKyrhpIFlCc2O6CYWWtwQKlDcR7KaCY2FmsoHullZq3CAaUOxbW8RjOxcWpvF1N7Ot3kZWYtwwGlDrtyo6+hABw4YxJPbNnRyCKZmY0bB5Q65AoFOjtEtvzYyPVN8+RGM2sdDih1yOVjVHNQivpm9PKkA4qZtQgHlDoM5mNUy64UefkVM2slDih1yBUKo5qDUtQ3vZetO3M8vSvXwFKZmY0PB5Q6DOZjVOt4FRVny3ukl5m1AgeUOgzmC3TX0YdyoOeimFkLcUCpQy5fqKuGcujsyQA8sP7pRhXJzGzcNHVxyFY3WIhR96Fcs/xB8oWgU6L/jkd2z2l528sOb2QRzczGjGsodcjlC3WN8ursEHOn9bDOT240sxbggFKHXH70NZSivum9rPNseTNrAQ4odRgsxKjW8SrVN72Xp7bt2r0UvpnZROWAUofBXGHU63gV9U3rpRCwfuuuBpXKzGx8NDWgSDpB0ipJA5LOqbC9V9K1aftySfNLtp2b0ldJOn64PCVdndLvlnS5pO5mXhukiY111lA8F8XMWkXTAoqkTuBi4ERgEXCqpEVlu50BbIiII4DPARelYxeRPV/+aOAE4BJJncPkeTXwHOD5wGTgPc26tqLBBvShHDA9e/yvO+bNbKJrZg3lGGAgIlZHxC5gGbCkbJ8lwJXp9fXAscqW7l0CLIuInRGxBhhI+VXNMyJuiAS4FTi0idcGZDWUekZ5AfR2dTJzcrdrKGY24TUzoMwDHip5vzalVdwnInLAJmDuEMcOm2dq6noH8P1KhZJ0pqQVklasW7duhJe0t0aM8oJsxvwjG7fXnY+Z2XhqxU75S4CbI+KnlTZGxKURsTgiFvf19dV1osE6Z8oXHXnQdJ7YspP1bvYyswmsmQHlYeCwkveHprSK+0jqAmYC64c4dsg8Jf0d0Ad8pCFXMIzBfNS1llfR0YfMAODuRzbXnZeZ2XhpZkC5DVgoaYGkHrJO9v6yffqB09LrNwM3pT6QfmBpGgW2AFhI1i9SNU9J7wGOB06NiDGZ1FHvWl5Fs6b0cOjsydz98KYGlMrMbHw0LaCkPpGzgRuBe4HrImKlpAsknZx2uwyYK2mArFZxTjp2JXAdcA9ZX8hZEZGvlmfK64vAQcAvJd0h6bxmXVvRYCHobkBAAXjeITN5eON2HnrKC0Wa2cTU1MUhI+IG4IaytPNKXu8A3lLl2AuBC2vJM6WP+UKXuXz9ExuLnjdvJjeufIyvLn+Ac098bkPyNDMbS63YKT9msmfKN+YWzpnawwsPm8VXfvGAO+fNbEJyQKnDYKFxNRSA1x7Vx45cnkt/urpheZqZjRUHlDo0YqZ8qQOnT+KNLziEq295kG07/Zx5M5tYHFBGKSLIN2C14XKn/+Ez2bozR/+djzQ0XzOzZnNAGaXBfADQ09XYW/jiw2fznIOn89VbHiAbQW1mNjE4oIxS8fklXQ2Y2FhKEm9/2eGsfGQzdzy0saF5m5k1kwPKKG3ZkfVxTO1t7Gjla5Y/yGA+mNzdyce/cRfXLH+Qa5Y/2NBzmJk1gwPKKD26KXts7zNmTmp43pO6O3nNkX3c9/hWVj+5teH5m5k1gwPKKD22KVsd+OAmBBSAVzx7LjMmdfFfdz3KjsF8U85hZtZIDiijVKyhHDyjOQGlu7ODk184j8c37+DSm1fz+OYdTTmPmVmjOKCM0mObd9DT2cGcqT1NO8eiQ2Zw+h8u4Kmnd3Hql27xQ7jMbL/mgDJKj23awcEzJ5E9YLJ5jjhwGqe9Yj6PbNzOOy+/1c1fZrbfckAZpUdTQBkLCw6Yyr+9/SXc++hmPvm9347JOc3MRsoBZZQe27Sjaf0nlbzuOQfyrj+azxW/uJ/v3/3YmJ3XzKxWDiijEBE8tmlHU4YMV3PN8geZP3cq82ZN5v1X386Hr73D81PMbL/igDIKT23bxa58YcyavIq6Ozt4zysX8Oy+aXzr1w/z1Vse4MH1fiCXme0fxvyhVK3gsc3Nm9Q4nN7uTt75ivn8fOBJfnTv47z60z/hZQvm8MfPPYjjjj6IZ86dOuZlMjMDB5RReSzNQTloDPtQSnV2iFcf2ccLD5vF7Q9s4K61G7nwhnu58IZ7OWz2ZBYcMJW3vexwnn/oLA6eMYnOBq83ZmZWSVMDiqQTgP8HdAJfjohPlm3vBb4CvARYD5wSEfenbecCZwB54IMRceNQeUpaACwD5gK3A++IiF3NuK49y65Mbkb2NZs5uZvXP+dAXv+cA9nw9C7uWruJlY9s4ucD67n5d08C0CHom97LwTMn8/IFc3j5s+dyRN80pGyl5L5pvU0f+mxm7aFpAUVSJ3Ax8AZgLXCbpP6IuKdktzOADRFxhKSlwEXAKZIWAUuBo4FDgB9JOjIdUy3Pi4DPRcQySV9Mef9bM67tsU076OwQfdN7m5H9qMye0sNrjuzjNUf2MZgv8OimHTy6aTubtw+yeUeO9Vt38eWfruHfb977aZC9XR3Mmz2ZA6f3Mqm7k96ujt3/dnZ00NkBnRIdHSICNj69i8FC0NUhpk/qYubk7r1+pk/qZsakbib3dABCgg4Jwe7XlLyWQBT/hZ25rOy/+P2TrF63ja07cxyzYA7PnzeTmZO7mT21hyndnXR1iu7ODjokgqC40n8xzw6JDuFgaTaGmllDOQYYiIjVAJKWAUuA0oCyBDg/vb4e+IKyT4AlwLKI2AmskTSQ8qNSnpLuBV4PvC3tc2XKtykBZcdgnkNm7b9NSd2dHRw+ZwqHz5myV/rOwTyPbNrBk1t3ImAwX2Dj04NseHoXj23aQa4QDOYL5PLZv4WAQmQf1kH2iT25u5Oujg7yEewYzLNjME+hCY9tETB7ag+dEjf99on68ioJavuD0hi3V6nKClj6tuox+2wzq80X3/ESXrWwr6F5NjOgzAMeKnm/FnhZtX0iIidpE1mT1TzglrJj56XXlfKcC2yMiFyF/fci6UzgzPR2q6RVI7imvfP6+F5vDwCeHG1eY2jClPP+CVJOXM5Gcjkbq2o5X/0PdeX7zEqJbdcpHxGXApc2Ol9JKyJicaPzbTSXs7FczsZyORtrrMvZzHkoDwOHlbw/NKVV3EdSFzCTrHO+2rHV0tcDs1Ie1c5lZmZN1MyAchuwUNICST1knez9Zfv0A6el128GborsQer9wFJJvWn01kLg1mp5pmN+kvIg5fmdJl6bmZmVaVqTV+oTORu4kWyI7+URsVLSBcCKiOgHLgOuSp3uT5EFCNJ+15F14OeAsyIiD1Apz3TKjwPLJP0j8OuU91hqeDNak7icjeVyNpbL2VhjWk5FNGGIjpmZtR2v5WVmZg3hgGJmZg3hgFInSSdIWiVpQNI543D+wyT9RNI9klZK+suUfr6khyXdkX5OKjnm3FTeVZKOH6trkXS/pN+k8qxIaXMk/VDS79K/s1O6JH0+leUuSS8uyee0tP/vJJ1W7XyjLONRJffsDkmbJX1of7mfki6X9ISku0vSGnYPJb0k/Y4G0rGjmitZpZyflvTbVJZvSZqV0udL2l5yb784XHmqXXODytmw37WyAUTLU/q1ygYTNaqc15aU8X5Jd6T0cbufRIR/RvlDNjDg98CzgB7gTmDRGJfhGcCL0+vpwH3AIrKVAv6qwv6LUjl7gQWp/J1jcS3A/cABZWmfAs5Jr88BLkqvTwK+Rzb5++XA8pQ+B1id/p2dXs9u4u/3MbJJXPvF/QReDbwYuLsZ95BsNOXL0zHfA05sYDmPA7rS64tKyjm/dL+yfCqWp9o1N6icDftdA9cBS9PrLwLva1Q5y7Z/BjhvvO+nayj12b28TGQLURaXlxkzEfFoRPwqvd4C3EuVVQKS3cvaRMQaoLiszXhdyxKypXJI//5pSfpXInML2TyjZwDHAz+MiKciYgPwQ+CEJpXtWOD3EfHAEPuM6f2MiJvJRkSWl6Hue5i2zYiIWyL7ZPlKSV51lzMifhB7VrO4hWy+WFXDlKfaNdddziGM6Hedvv2/nmxZqaaVM53nrcDXhspjLO6nA0p9Ki0vM9SHeVNJmg/8AbA8JZ2dmhcuL6nCVivzWFxLAD+QdLuyJXAADoqIR9Prx4CD9oNyFi1l7/+k+9v9LGrUPZyXXpenN8O7yb4hFy2Q9GtJ/yPpVSltqPJUu+ZGacTvuuYloer0KuDxiPhdSdq43E8HlBYhaRrwDeBDEbGZbGHMZwMvAh4lqxKPt1dGxIuBE4GzJL26dGP61rRfjGNPbd0nA19PSfvj/dzH/nQPq5H0CbL5ZVenpEeBwyPiD4CPANdImlFrfk245gnxuy5xKnt/8Rm3++mAUp9alpdpOkndZMHk6oj4JkBEPB4R+YgoAF9iz2rNI13WpmEi4uH07xPAt1KZHk9V8WKVvLi08LiVMzkR+FVEPJ7KvN/dzxKNuocPs3czVMPLLOl04I3A29MHF6kJaX16fTtZf8SRw5Sn2jXXrYG/66YvCZXy/t/AtSXlH7f76YBSn1qWl2mq1H56GXBvRHy2JP0ZJbv9GVAcHTKiZW0aWM6pkqYXX5N10N7N3svvlC6Z0w+8U5mXA5tSlfxG4DhJs1NTxHEprdH2+ta3v93PMg25h2nbZkkvT39X76SBSxgpezjex4CTI+LpkvQ+Zc9PQtKzyO7h6mHKU+2aG1HOhvyuU8Bs9pJQfwz8NiJ2N2WN6/0cTU++f/YaNXES2ciq3wOfGIfzv5KsenoXcEf6OQm4CvhNSu8HnlFyzCdSeVdRMoqnmddCNgLmzvSzspg/WTvzj4HfAT8C5qR0kT1M7ffpOhaX5PVusg7RAeBdTbinU8m+Xc4sSdsv7idZkHsUGCRrAz+jkfcQWEz2Afp74Auk1TQaVM4Bsr6G4t/pF9O+b0p/E3cAvwL+ZLjyVLvmBpWzYb/r9Hd/a7r2rwO9jSpnSr8C+IuyfcftfnrpFTMzawg3eZmZWUM4oJiZWUM4oJiZWUM4oJiZWUM4oJiZWUM07YmNZq1CUnFIJcDBQB5Yl94fE9n6TcV97ycbnvvkmBayDpL+FLgvIu4Z77LYxOaAYjaMyGYdvwiypc2BrRHxL+NaqMb6U+A/yR65bTZqbvIyGwVJx6bF936TFhDsLds+WdL3JL03rRJwuaRb0zFL0j6nS/qmpO8rew7Fp6qc66WSfiHpzpTHdEmTJP1HOv+vJb2uJM8vlBz7n5Jem15vlXRhyucWSQdJ+kOyNcs+rezZGc9u0i2zNuCAYjZyk8hmKJ8SEc8nq+m/r2T7NOC7wNci4ktks6tviohjgNeRfXhPTfu+CDgFeD5wiqTSNaGKi1ReC/xlRLyQbKmN7cBZZOv4PZ9smZgrJU0aptxTgVtSPjcD742IX5DNBv/riHhRRPx+5LfDLOOAYjZyncCaiLgvvb+S7AFIRd8B/iMivpLeHweco+yJev9NFpAOT9t+HBGbImIHWZPTM8vOdRTwaETcBhARmyNbDv2VwFdT2m+BB8gWABzKLrKmLYDbyR7EZNYwDihmjfdzsgdWFR+fK+BNqQbwoog4PCLuTdt2lhyXp/5+zRx7/78urbUMxp61lhpxLrO9OKCYjVwemC/piPT+HcD/lGw/D9hAtjAjZKv7fqAYYCT9wQjOtQp4hqSXpmOnpyXLfwq8PaUdSVbjWUX2mOUXSepIzWfHVMx1b1vIHh9tVhcHFLOR2wG8C/i6pN8ABbLnhZf6S2By6mj/B6AbuEvSyvS+JmlI8inAv0q6k+xxvZOAS4COdP5rgdMjYidZ7WgNWfPZ58lWmx3OMuCvU+e+O+Vt1LzasJmZNYRrKGZm1hAOKGZm1hAOKGZm1hAOKGZm1hAOKGZm1hAOKGZm1hAOKGZm1hD/H/wuimVlEQZ1AAAAAElFTkSuQmCC\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": 349
        },
        "id": "k9VGjoOv5Sgz",
        "outputId": "ce4ca302-0f0f-4334-b88b-f9a02ab38f82"
      },
      "source": [
        "sns.distplot(filtered_data[\"total_length\"])\n",
        "plt.xlim([0, 512]);\n",
        "plt.xlabel('Token count');\n",
        "plt.ylabel('Document count')\n",
        "plt.title(\"Distribution of Total Token Lengths\")\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/seaborn/distributions.py:2551: FutureWarning: `distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).\n",
            "  warnings.warn(msg, FutureWarning)\n"
          ],
          "name": "stderr"
        },
        {
          "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": "fJ77tjeI3qER"
      },
      "source": [
        "def get_keyword_col_names(df, prefix=\"has_mk_\"):\n",
        "    return [col for col in df.columns if \"has_mk_\" in col]\n",
        "keyword_cols = get_keyword_col_names(filtered_data)\n",
        "prefix = \"has_mk_\"\n",
        "most_common_keywords = [prefix + k[0] for k in counts_keywords.most_common(20)]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "htLE0w0O35QP"
      },
      "source": [
        "def get_most_common_keyword_for_news_item(item):\n",
        "    for col in most_common_keywords:\n",
        "        if item[col] == 1:\n",
        "            return col[len(prefix):]\n",
        "    return 0"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Wq_9Nw1G4W7B"
      },
      "source": [
        "filtered_data[\"keyword\"] = filtered_data.apply(get_most_common_keyword_for_news_item, axis=1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "QDE2_JmZ4dcK",
        "outputId": "cdff020d-1589-4866-a128-cb6776d2ca13"
      },
      "source": [
        "print(filtered_data.keyword.value_counts(normalize=False))\n",
        "print(filtered_data.keyword.value_counts(normalize=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "politics                12110\n",
            "russia                  11084\n",
            "china                    3427\n",
            "health                   3285\n",
            "energy                   2056\n",
            "business                 1640\n",
            "climate change           1606\n",
            "science                  1374\n",
            "food                     1054\n",
            "cuba                      430\n",
            "ebola                     205\n",
            "fracking                  198\n",
            "biodiversity              196\n",
            "transportation            146\n",
            "coal                      121\n",
            "tips                       98\n",
            "pipelines                  18\n",
            "renewables                  8\n",
            "cop21                       6\n",
            "sustainable business        1\n",
            "Name: keyword, dtype: int64\n",
            "politics                0.310012\n",
            "russia                  0.283747\n",
            "china                   0.087730\n",
            "health                  0.084095\n",
            "energy                  0.052633\n",
            "business                0.041983\n",
            "climate change          0.041113\n",
            "science                 0.035174\n",
            "food                    0.026982\n",
            "cuba                    0.011008\n",
            "ebola                   0.005248\n",
            "fracking                0.005069\n",
            "biodiversity            0.005018\n",
            "transportation          0.003738\n",
            "coal                    0.003098\n",
            "tips                    0.002509\n",
            "pipelines               0.000461\n",
            "renewables              0.000205\n",
            "cop21                   0.000154\n",
            "sustainable business    0.000026\n",
            "Name: keyword, dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4EnMsEH8ySDS"
      },
      "source": [
        "unbalanced_train_set, unbalanced_test_set = train_test_split(filtered_data, test_size = 0.1, random_state = 1001)\n",
        "unbalanced_train_set, unbalanced_validation_set = train_test_split(unbalanced_train_set, test_size = 0.1, random_state = 1001)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2YWqciZZzCw2"
      },
      "source": [
        "unbalanced_train_set.to_csv(source_folder + \"train_unbalanced.csv\", index=False)\n",
        "unbalanced_test_set.to_csv(source_folder + \"test_unbalanced.csv\", index=False)\n",
        "unbalanced_validation_set.to_csv(source_folder + \"validation_unbalanced.csv\", index=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "jdQzBLlxy5k2",
        "outputId": "eb938668-c0a3-47fb-a268-f401370c8e0b"
      },
      "source": [
        "print(unbalanced_train_set.fake.value_counts(normalize=False))\n",
        "print(unbalanced_train_set.fake.value_counts(normalize=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1    19315\n",
            "0    12325\n",
            "Name: fake, dtype: int64\n",
            "1    0.610461\n",
            "0    0.389539\n",
            "Name: fake, dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "oB88I529137e",
        "outputId": "6c3216b7-6afa-45c6-ff79-0028451fd41c"
      },
      "source": [
        "print(unbalanced_test_set.fake.value_counts(normalize=False))\n",
        "print(unbalanced_test_set.fake.value_counts(normalize=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1    2362\n",
            "0    1545\n",
            "Name: fake, dtype: int64\n",
            "1    0.604556\n",
            "0    0.395444\n",
            "Name: fake, dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "HnfZ0pKp13ua",
        "outputId": "20d1a8d8-6618-40f9-9960-2145a9566ba6"
      },
      "source": [
        "print(unbalanced_validation_set.fake.value_counts(normalize=False))\n",
        "print(unbalanced_validation_set.fake.value_counts(normalize=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1    2160\n",
            "0    1356\n",
            "Name: fake, dtype: int64\n",
            "1    0.614334\n",
            "0    0.385666\n",
            "Name: fake, dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_S7uHYDg1Brb"
      },
      "source": [
        "BALANCED_N = 10_000"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Klgpk1Ny07QZ"
      },
      "source": [
        "balanced_real = filtered_data[filtered_data[\"fake\"] == 0].sample(BALANCED_N, random_state = 1001)\n",
        "balanced_fake = filtered_data[filtered_data[\"fake\"] == 1].sample(BALANCED_N, random_state = 1001)\n",
        "balanced_data = balanced_real.append(balanced_fake)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V3Ae4e-S06Vq"
      },
      "source": [
        "balanced_train_set, balanced_test_set = train_test_split(balanced_data, test_size = 0.1, random_state = 1001)\n",
        "balanced_train_set, balanced_validation_set = train_test_split(balanced_train_set, test_size = 0.1, random_state = 1001)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UWiaO8MQzBgx"
      },
      "source": [
        "balanced_train_set.to_csv(source_folder + \"train_balanced.csv\", index=False)\n",
        "balanced_test_set.to_csv(source_folder + \"test_balanced.csv\", index=False)\n",
        "balanced_validation_set.to_csv(source_folder + \"validation_balanced.csv\", index=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_oHubqTN1pu5",
        "outputId": "c4247e08-36b3-47c1-be42-63414ff4dcb0"
      },
      "source": [
        "print(balanced_train_set.fake.value_counts(normalize=False))\n",
        "print(balanced_train_set.fake.value_counts(normalize=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1    8100\n",
            "0    8100\n",
            "Name: fake, dtype: int64\n",
            "1    0.5\n",
            "0    0.5\n",
            "Name: fake, dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bIvMYr1t1uXj",
        "outputId": "fb8c1e61-07b1-46b4-e7e2-2a1f29dc6ad9"
      },
      "source": [
        "print(balanced_test_set.fake.value_counts(normalize=False))\n",
        "print(balanced_test_set.fake.value_counts(normalize=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1    1015\n",
            "0     985\n",
            "Name: fake, dtype: int64\n",
            "1    0.5075\n",
            "0    0.4925\n",
            "Name: fake, dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "naEfzTU81utE",
        "outputId": "3d06bd39-7293-4988-fd7a-6b80042163a8"
      },
      "source": [
        "print(balanced_validation_set.fake.value_counts(normalize=False))\n",
        "print(balanced_validation_set.fake.value_counts(normalize=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0    915\n",
            "1    885\n",
            "Name: fake, dtype: int64\n",
            "0    0.508333\n",
            "1    0.491667\n",
            "Name: fake, dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1kfAPogR-lvN"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}