wiki_movie_year&genre_predictor.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5j03GkclkXOf"
      },
      "source": [
        "# Predicting Movie Release Year and Genre\n",
        "### By: Matthew Fernandez & Andrew Cline"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "In this notebook we will dive into a dataset regarding movies made from the early 1900s to 2017, each movie has features like plot synposis, Release year, Genre, cast, title, Orgin/Ethnicity, Director, and Wikipedia Page. We are intrested in predicting the top five most occuring genre labels and the release year of a movie given all other features. We will build one RNN model to tackle both of these tasks accurately. Lets begin with some needed imports."
      ],
      "metadata": {
        "id": "60SXcq8UG6wN"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "zB7yfGjIs4OC"
      },
      "outputs": [],
      "source": [
        "import string\n",
        "import re\n",
        "import random\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import tensorflow as tf\n",
        "import seaborn as sns \n",
        "import matplotlib.pyplot as plt\n",
        "import keras.layers\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras import layers, models\n",
        "from tensorflow.keras.callbacks import EarlyStopping\n",
        "from sklearn.model_selection import train_test_split\n",
        "from sklearn.preprocessing import LabelEncoder, OneHotEncoder\n",
        "from sklearn.utils import resample\n",
        "from scipy import stats"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3UHjlUWpstVY"
      },
      "source": [
        "## Read in Data\n",
        "This dataset was taken from Kaggle, https://www.kaggle.com/jrobischon/wikipedia-movie-plots/version/1 on November 18th, 2021. There are roughly 33K instances with 8 total features, six of which will be used to predict the title and genre."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "O8Lq6b6ntzJq",
        "outputId": "5426db46-fd39-45ab-db89-460a4a5b4539"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /content/gdrive/\n"
          ]
        }
      ],
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/gdrive/')\n",
        "df = pd.read_csv(\"/content/gdrive/MyDrive/Machine Learning Colab Notebooks/data/wiki_movie_plots_deduped.csv\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "m4PPA71auXlI"
      },
      "source": [
        "## Initial Data Exploration \n",
        "<br> Lets take a look at the first 20 rows of our dataset."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 677
        },
        "id": "zJxszy6Hus5L",
        "outputId": "642ac1a9-3df8-4a60-a100-545a53bbbc65"
      },
      "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>Release Year</th>\n",
              "      <th>Title</th>\n",
              "      <th>Origin/Ethnicity</th>\n",
              "      <th>Director</th>\n",
              "      <th>Cast</th>\n",
              "      <th>Genre</th>\n",
              "      <th>Wiki Page</th>\n",
              "      <th>Plot</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>1901</td>\n",
              "      <td>Kansas Saloon Smashers</td>\n",
              "      <td>American</td>\n",
              "      <td>Unknown</td>\n",
              "      <td>NaN</td>\n",
              "      <td>unknown</td>\n",
              "      <td>https://en.wikipedia.org/wiki/Kansas_Saloon_Sm...</td>\n",
              "      <td>A bartender is working at a saloon, serving dr...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1901</td>\n",
              "      <td>Love by the Light of the Moon</td>\n",
              "      <td>American</td>\n",
              "      <td>Unknown</td>\n",
              "      <td>NaN</td>\n",
              "      <td>unknown</td>\n",
              "      <td>https://en.wikipedia.org/wiki/Love_by_the_Ligh...</td>\n",
              "      <td>The moon, painted with a smiling face hangs ov...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1901</td>\n",
              "      <td>The Martyred Presidents</td>\n",
              "      <td>American</td>\n",
              "      <td>Unknown</td>\n",
              "      <td>NaN</td>\n",
              "      <td>unknown</td>\n",
              "      <td>https://en.wikipedia.org/wiki/The_Martyred_Pre...</td>\n",
              "      <td>The film, just over a minute long, is composed...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>1901</td>\n",
              "      <td>Terrible Teddy, the Grizzly King</td>\n",
              "      <td>American</td>\n",
              "      <td>Unknown</td>\n",
              "      <td>NaN</td>\n",
              "      <td>unknown</td>\n",
              "      <td>https://en.wikipedia.org/wiki/Terrible_Teddy,_...</td>\n",
              "      <td>Lasting just 61 seconds and consisting of two ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>1902</td>\n",
              "      <td>Jack and the Beanstalk</td>\n",
              "      <td>American</td>\n",
              "      <td>George S. Fleming, Edwin S. Porter</td>\n",
              "      <td>NaN</td>\n",
              "      <td>unknown</td>\n",
              "      <td>https://en.wikipedia.org/wiki/Jack_and_the_Bea...</td>\n",
              "      <td>The earliest known adaptation of the classic f...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5</th>\n",
              "      <td>1903</td>\n",
              "      <td>Alice in Wonderland</td>\n",
              "      <td>American</td>\n",
              "      <td>Cecil Hepworth</td>\n",
              "      <td>May Clark</td>\n",
              "      <td>unknown</td>\n",
              "      <td>https://en.wikipedia.org/wiki/Alice_in_Wonderl...</td>\n",
              "      <td>Alice follows a large white rabbit down a \"Rab...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6</th>\n",
              "      <td>1903</td>\n",
              "      <td>The Great Train Robbery</td>\n",
              "      <td>American</td>\n",
              "      <td>Edwin S. Porter</td>\n",
              "      <td>NaN</td>\n",
              "      <td>western</td>\n",
              "      <td>https://en.wikipedia.org/wiki/The_Great_Train_...</td>\n",
              "      <td>The film opens with two bandits breaking into ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7</th>\n",
              "      <td>1904</td>\n",
              "      <td>The Suburbanite</td>\n",
              "      <td>American</td>\n",
              "      <td>Wallace McCutcheon</td>\n",
              "      <td>NaN</td>\n",
              "      <td>comedy</td>\n",
              "      <td>https://en.wikipedia.org/wiki/The_Suburbanite</td>\n",
              "      <td>The film is about a family who move to the sub...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>8</th>\n",
              "      <td>1905</td>\n",
              "      <td>The Little Train Robbery</td>\n",
              "      <td>American</td>\n",
              "      <td>Edwin Stanton Porter</td>\n",
              "      <td>NaN</td>\n",
              "      <td>unknown</td>\n",
              "      <td>https://en.wikipedia.org/wiki/The_Little_Train...</td>\n",
              "      <td>The opening scene shows the interior of the ro...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>9</th>\n",
              "      <td>1905</td>\n",
              "      <td>The Night Before Christmas</td>\n",
              "      <td>American</td>\n",
              "      <td>Edwin Stanton Porter</td>\n",
              "      <td>NaN</td>\n",
              "      <td>unknown</td>\n",
              "      <td>https://en.wikipedia.org/wiki/The_Night_Before...</td>\n",
              "      <td>Scenes are introduced using lines of the poem....</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>10</th>\n",
              "      <td>1906</td>\n",
              "      <td>Dream of a Rarebit Fiend</td>\n",
              "      <td>American</td>\n",
              "      <td>Wallace McCutcheon and Edwin S. Porter</td>\n",
              "      <td>NaN</td>\n",
              "      <td>short</td>\n",
              "      <td>https://en.wikipedia.org/wiki/Dream_of_a_Rareb...</td>\n",
              "      <td>The Rarebit Fiend gorges on Welsh rarebit at a...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>11</th>\n",
              "      <td>1906</td>\n",
              "      <td>From Leadville to Aspen: A Hold-Up in the Rockies</td>\n",
              "      <td>American</td>\n",
              "      <td>Francis J. Marion and Wallace McCutcheon</td>\n",
              "      <td>NaN</td>\n",
              "      <td>short action/crime western</td>\n",
              "      <td>https://en.wikipedia.org/wiki/From_Leadville_t...</td>\n",
              "      <td>The film features a train traveling through th...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>12</th>\n",
              "      <td>1906</td>\n",
              "      <td>Kathleen Mavourneen</td>\n",
              "      <td>American</td>\n",
              "      <td>Edwin S. Porter</td>\n",
              "      <td>NaN</td>\n",
              "      <td>short film</td>\n",
              "      <td>https://en.wikipedia.org/wiki/Kathleen_Mavourn...</td>\n",
              "      <td>Irish villager Kathleen is a tenant of Captain...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>13</th>\n",
              "      <td>1907</td>\n",
              "      <td>Daniel Boone</td>\n",
              "      <td>American</td>\n",
              "      <td>Wallace McCutcheon and Ediwin S. Porter</td>\n",
              "      <td>William Craven, Florence Lawrence</td>\n",
              "      <td>biographical</td>\n",
              "      <td>https://en.wikipedia.org/wiki/Daniel_Boone_(19...</td>\n",
              "      <td>Boone's daughter befriends an Indian maiden as...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>14</th>\n",
              "      <td>1907</td>\n",
              "      <td>How Brown Saw the Baseball Game</td>\n",
              "      <td>American</td>\n",
              "      <td>Unknown</td>\n",
              "      <td>Unknown</td>\n",
              "      <td>comedy</td>\n",
              "      <td>https://en.wikipedia.org/wiki/How_Brown_Saw_th...</td>\n",
              "      <td>Before heading out to a baseball game at a nea...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>15</th>\n",
              "      <td>1907</td>\n",
              "      <td>Laughing Gas</td>\n",
              "      <td>American</td>\n",
              "      <td>Edwin Stanton Porter</td>\n",
              "      <td>Bertha Regustus, Edward Boulden</td>\n",
              "      <td>comedy</td>\n",
              "      <td>https://en.wikipedia.org/wiki/Laughing_Gas_(fi...</td>\n",
              "      <td>The plot is that of a black woman going to the...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>16</th>\n",
              "      <td>1908</td>\n",
              "      <td>The Adventures of Dollie</td>\n",
              "      <td>American</td>\n",
              "      <td>D. W. Griffith</td>\n",
              "      <td>Arthur V. Johnson, Linda Arvidson</td>\n",
              "      <td>drama</td>\n",
              "      <td>https://en.wikipedia.org/wiki/The_Adventures_o...</td>\n",
              "      <td>On a beautiful summer day a father and mother ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>17</th>\n",
              "      <td>1908</td>\n",
              "      <td>The Black Viper</td>\n",
              "      <td>American</td>\n",
              "      <td>D. W. Griffith</td>\n",
              "      <td>D. W. Griffith</td>\n",
              "      <td>drama</td>\n",
              "      <td>https://en.wikipedia.org/wiki/The_Black_Viper</td>\n",
              "      <td>A thug accosts a girl as she leaves her workpl...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>18</th>\n",
              "      <td>1908</td>\n",
              "      <td>A Calamitous Elopement</td>\n",
              "      <td>American</td>\n",
              "      <td>D.W. Griffith</td>\n",
              "      <td>Harry Solter, Linda Arvidson</td>\n",
              "      <td>comedy</td>\n",
              "      <td>https://en.wikipedia.org/wiki/A_Calamitous_Elo...</td>\n",
              "      <td>A young couple decides to elope after being ca...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>19</th>\n",
              "      <td>1908</td>\n",
              "      <td>The Call of the Wild</td>\n",
              "      <td>American</td>\n",
              "      <td>D. W. Griffith</td>\n",
              "      <td>Charles Inslee</td>\n",
              "      <td>adventure</td>\n",
              "      <td>https://en.wikipedia.org/wiki/The_Call_of_the_...</td>\n",
              "      <td>A white girl (Florence Lawrence) rejects a pro...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "    Release Year  ...                                               Plot\n",
              "0           1901  ...  A bartender is working at a saloon, serving dr...\n",
              "1           1901  ...  The moon, painted with a smiling face hangs ov...\n",
              "2           1901  ...  The film, just over a minute long, is composed...\n",
              "3           1901  ...  Lasting just 61 seconds and consisting of two ...\n",
              "4           1902  ...  The earliest known adaptation of the classic f...\n",
              "5           1903  ...  Alice follows a large white rabbit down a \"Rab...\n",
              "6           1903  ...  The film opens with two bandits breaking into ...\n",
              "7           1904  ...  The film is about a family who move to the sub...\n",
              "8           1905  ...  The opening scene shows the interior of the ro...\n",
              "9           1905  ...  Scenes are introduced using lines of the poem....\n",
              "10          1906  ...  The Rarebit Fiend gorges on Welsh rarebit at a...\n",
              "11          1906  ...  The film features a train traveling through th...\n",
              "12          1906  ...  Irish villager Kathleen is a tenant of Captain...\n",
              "13          1907  ...  Boone's daughter befriends an Indian maiden as...\n",
              "14          1907  ...  Before heading out to a baseball game at a nea...\n",
              "15          1907  ...  The plot is that of a black woman going to the...\n",
              "16          1908  ...  On a beautiful summer day a father and mother ...\n",
              "17          1908  ...  A thug accosts a girl as she leaves her workpl...\n",
              "18          1908  ...  A young couple decides to elope after being ca...\n",
              "19          1908  ...  A white girl (Florence Lawrence) rejects a pro...\n",
              "\n",
              "[20 rows x 8 columns]"
            ]
          },
          "metadata": {},
          "execution_count": 4
        }
      ],
      "source": [
        "df.head(20)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "As we can see, the dataset is in ascending order by release year, and there are a handful of nan values within the cast feature. There is also a handful of Unknown genre types, we intent to remove these movies but before we do, we need to make sure that we are not dropping too many movies. Most of the data in this data set is text, but the type of text varies in each feature. For example Plot is a long string of text that describes the plot of the movie, where as Origin/Ethnicity is just label corresponding the country of origin. We will later drop the Wiki Page column since this will be of no help for our task.<br><br>Lets now take a look at a summary of the dataframe. "
      ],
      "metadata": {
        "id": "R9C9JaB9V0Wd"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ZuOBJUiWualN",
        "outputId": "6b28263d-52fc-4537-8f5b-ac88fbfcf6db"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "<class 'pandas.core.frame.DataFrame'>\n",
            "RangeIndex: 34886 entries, 0 to 34885\n",
            "Data columns (total 8 columns):\n",
            " #   Column            Non-Null Count  Dtype \n",
            "---  ------            --------------  ----- \n",
            " 0   Release Year      34886 non-null  int64 \n",
            " 1   Title             34886 non-null  object\n",
            " 2   Origin/Ethnicity  34886 non-null  object\n",
            " 3   Director          34886 non-null  object\n",
            " 4   Cast              33464 non-null  object\n",
            " 5   Genre             34886 non-null  object\n",
            " 6   Wiki Page         34886 non-null  object\n",
            " 7   Plot              34886 non-null  object\n",
            "dtypes: int64(1), object(7)\n",
            "memory usage: 2.1+ MB\n"
          ]
        }
      ],
      "source": [
        "df.info()"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Looks like there are a few na values in Cast, around 1422 values to be exact, all other columns are not na values.<br><br> What is the range of values in release year?"
      ],
      "metadata": {
        "id": "7xgqqqV0HqOc"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 300
        },
        "id": "DLVVG3VPjAbU",
        "outputId": "83a0871e-ecc5-425d-96d9-16e948c0a070"
      },
      "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>Release Year</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>count</th>\n",
              "      <td>34886.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>mean</th>\n",
              "      <td>1981.314252</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>std</th>\n",
              "      <td>27.815174</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>min</th>\n",
              "      <td>1901.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>25%</th>\n",
              "      <td>1957.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>50%</th>\n",
              "      <td>1988.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>75%</th>\n",
              "      <td>2007.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>max</th>\n",
              "      <td>2017.000000</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "       Release Year\n",
              "count  34886.000000\n",
              "mean    1981.314252\n",
              "std       27.815174\n",
              "min     1901.000000\n",
              "25%     1957.000000\n",
              "50%     1988.000000\n",
              "75%     2007.000000\n",
              "max     2017.000000"
            ]
          },
          "metadata": {},
          "execution_count": 6
        }
      ],
      "source": [
        "df.describe()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8TMjnZGfj7Zh"
      },
      "source": [
        "Looks like the midpoint of the release year is 1981, the largest value is 2017 and the minimum value is in 1901.<br><br>Let us now take a look at the distribution of the release year."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 354
        },
        "id": "_G9kmkEit6pg",
        "outputId": "86306b13-81ee-4835-c88c-136b2d4c06df"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/seaborn/distributions.py:2619: 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"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ],
      "source": [
        "sns.distplot(df['Release Year'])\n",
        "plt.title('Distribution of Release Year', fontsize=15)\n",
        "plt.xlabel('Release Year', fontsize=13)\n",
        "plt.ylabel('Density', fontsize=13);"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LbER23Fegafz"
      },
      "source": [
        "From the above distribution plot, we can see that a majority of our data consists of movies that took place after the 1980s.  There is a maximum in the plot that shows a majority of release year ranges from 2010 to 2017. This is interesting, if we are to blindly guess a year, the best guess would be from 2010-2017. If we are to use the release year as is, our model would make two predictions, release year would be a regression task, and predicting the genre would be a classification task. This is a model we can assume will either do well on one task and bad on the other or altogether perform poorly.  We will predict the decade rather than the release year to avoid said problems. "
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Data Cleaning"
      ],
      "metadata": {
        "id": "AD0c1kNrIZ4U"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let us now visualize the top five genre occurrences. We will also create the new feature 'decade' from which the recurrent neural network will predict."
      ],
      "metadata": {
        "id": "4y81rktBf7ne"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 330
        },
        "id": "I3eSkv1K05g5",
        "outputId": "2c083a8a-2760-45bd-9945-d856ed80689a"
      },
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ],
      "source": [
        "# drop na values that occur in Cast\n",
        "df.dropna(inplace=True)\n",
        "df2 = df['Genre'].value_counts()[df['Genre'].value_counts() > 150]\n",
        "genres = df2.keys()\n",
        "df3 = df[df[\"Genre\"].isin(genres)] # dataframe with the genres we have counts > 150\n",
        "\n",
        "df_genre = df3['Genre'].value_counts()[:6].keys()\n",
        "df3 = df3[df3['Genre'].isin(df_genre)]\n",
        "df4 = df3[df3['Genre'] != 'unknown']\n",
        "df5 = df4.dropna(axis=0)\n",
        "df5['decade'] = pd.cut(df5['Release Year'], bins=[1900, 1909,1919,1929,1939,1949,1959,1969,1979,1989,1999,2009,2029], labels=['1900s','10s','20s', '30s','40s',\n",
        "                                                                                                        '50s','60s','70s',\n",
        "                                                                                                        '80s', '90s', '2000s',\n",
        "                                                                                                        '2010s'])\n",
        "df5 = df5[df5['decade'].isin(['1900s', '10s', '20s']) == False]\n",
        "sns.histplot(df5['Genre'])\n",
        "plt.title('Counts of Top 5 Genres',fontsize=15)\n",
        "plt.ylabel('Count',fontsize=13)\n",
        "plt.xlabel('Genre',fontsize=13)\n",
        "plt.xticks(rotation=90);"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "We can see here that drama is the most occurring genre followed by comedy. horror and action are very similar in counts and lastly thriller."
      ],
      "metadata": {
        "id": "6ClSRW0UvK4B"
      }
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_1eAYtRpX_uQ"
      },
      "source": [
        "## Pre-processing\n",
        "Let us now prepare the data for Machine Learning. We will apply word embedding to our features that will be passed into our model. We have to vectorize each feature independently. The Text Vectorization layer code was obtained from the Keras documentation on text classification."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "f8UvOJoCYDWh"
      },
      "outputs": [],
      "source": [
        "# remove feature Wiki Page, not useful for this application\n",
        "X = df5[['Plot', 'Origin/Ethnicity', 'Director', 'Cast']].values\n",
        "le = LabelEncoder()\n",
        "y1 = df5['Genre'].values\n",
        "y2 = df5['decade'].values\n",
        "le.fit(y1)\n",
        "y1 = le.transform(y1)\n",
        "le2 = LabelEncoder()\n",
        "le2.fit(y2)\n",
        "y2 = le2.transform(y2)\n",
        "y = np.vstack((y1,y2)).T\n",
        "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)\n",
        "\n",
        "genre_labels = df5['Genre'].unique() \n",
        "year_labels = df5['decade'].unique()\n",
        "\n",
        "def standardize(input_data):\n",
        "  lower = tf.strings.lower(input_data)\n",
        "  return tf.strings.regex_replace(lower, f\"[{re.escape(string.punctuation)}]\",\"\")\n",
        "\n",
        "plot_vectorized_layer = layers.TextVectorization(\n",
        "    standardize=standardize,\n",
        "    max_tokens=10000,\n",
        "    output_mode='int',\n",
        "    output_sequence_length=1000\n",
        ")\n",
        "plot_vectorized_layer.adapt(X_train[0:,0])\n",
        "origin_vectorized_layer = layers.TextVectorization(\n",
        "    standardize=standardize,\n",
        "    max_tokens=10000,\n",
        "    output_mode='int',\n",
        "    output_sequence_length=1000\n",
        ")\n",
        "origin_vectorized_layer.adapt(X_train[0:,1])\n",
        "director_vectorized_layer = layers.TextVectorization(\n",
        "    standardize=standardize,\n",
        "    max_tokens=10000,\n",
        "    output_mode='int',\n",
        "    output_sequence_length=1000\n",
        ")\n",
        "director_vectorized_layer.adapt(X_train[0:,2])\n",
        "cast_vectorized_layer = layers.TextVectorization(\n",
        "    standardize=standardize,\n",
        "    max_tokens=10000,\n",
        "    output_mode='int',\n",
        "    output_sequence_length=1000\n",
        ")\n",
        "cast_vectorized_layer.adapt(X_train[0:,3])"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Above we also implemented the y values to be encoded from 0 to n."
      ],
      "metadata": {
        "id": "1C_lm86ENHps"
      }
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "818Eeir-pIoM"
      },
      "source": [
        "## Machine Learning\n",
        "<br>\n",
        "Let us now build a basic baseline by always predicting the most common decade along with genre. \n",
        "<br> \n",
        "\n",
        "### Baseline Model"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "y_baseline_genre = stats.mode(y_test[0:,0], axis=None).mode[0]\n",
        "genre_test_accuracy = sum(y_baseline_genre == y_test[0:,0]) / len(y_test[0:,0])\n",
        "print('test accuracy genre: {:0.2f}'.format(genre_test_accuracy))\n",
        "y_baseline_year = stats.mode(y_test[0:,1], axis=None).mode[0]\n",
        "year_test_accuracy = sum(y_baseline_year == y_test[0:,1]) / len(y_test[0:,1])\n",
        "print('test accuracy genre: {:0.2f}'.format(year_test_accuracy))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fwvnwe6KELsO",
        "outputId": "1e505392-7486-4f77-e19c-489787446dac"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "test accuracy Decade: 0.43\n",
            "test accuracy Genre: 0.17\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "For genre, since the most common value 'Drama' represents a high proportion of the data the baseline is high. For year, that low value is to be expected since there is a relatively good spread of data."
      ],
      "metadata": {
        "id": "dgu5hgM3NXQA"
      }
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "us2uyALWpSjV"
      },
      "source": [
        "# Actual Model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TZvFUmfRpccN"
      },
      "source": [
        "Building the RNN Model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "-isCxN5l5Gxv"
      },
      "outputs": [],
      "source": [
        "plot_input = tf.keras.Input(shape=(1,), dtype=tf.string, name='plot_input')\n",
        "cast_input = tf.keras.Input(shape=(1,), dtype=tf.string, name='cast_input')\n",
        "director_input = tf.keras.Input(shape=(1,), dtype=tf.string, name='director_input')\n",
        "origin_input = tf.keras.Input(shape=(1,), dtype=tf.string, name='origin_input')\n",
        "plot_out = plot_vectorized_layer(plot_input)\n",
        "plot_out = layers.Embedding(10000, 32)(plot_out)\n",
        "plot_out = layers.BatchNormalization()(plot_out)\n",
        "plot_out = layers.Bidirectional(layers.LSTM(32, return_sequences=True))(plot_out)\n",
        "cast_out = cast_vectorized_layer(cast_input)\n",
        "cast_out = layers.Embedding(10000, 32)(cast_out)\n",
        "cast_out = layers.BatchNormalization()(cast_out)\n",
        "cast_out = layers.Bidirectional(layers.LSTM(32, return_sequences=True))(cast_out)\n",
        "director_out = director_vectorized_layer(director_input)\n",
        "director_out = layers.Embedding(10000, 32)(director_out)\n",
        "director_out = layers.BatchNormalization()(director_out)\n",
        "director_out = layers.Bidirectional(layers.LSTM(32, return_sequences=True))(director_out)\n",
        "origin_out = origin_vectorized_layer(origin_input)\n",
        "origin_out = layers.Embedding(10000, 32)(origin_out)\n",
        "origin_out = layers.BatchNormalization()(origin_out)\n",
        "origin_out = layers.Bidirectional(layers.LSTM(32, return_sequences=True))(origin_out)\n",
        "concat = layers.concatenate([plot_out, cast_out, director_out, origin_out])\n",
        "# x = layers.GRU(64)(concat)\n",
        "x = layers.Flatten()(concat)\n",
        "year_output = layers.Dense(len(year_labels), activation='softmax', name = 'year_output')(x)\n",
        "genre_output = layers.Dense(len(genre_labels), activation='softmax', name='genre_output')(x)\n",
        "model = keras.Model(inputs=[plot_input, cast_input, director_input, origin_input], outputs=[year_output, genre_output])"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "So for this model there is 4 different input layers to account for 4 different embedding spaces for each input. Each input gets its own Batch Normalization layer and birdirectional LSTM layer. After all the models tried, giving each input its own LSTM layers instead of after the concatenation has proved to be the best method.<br><br>Compiling the RNN model"
      ],
      "metadata": {
        "id": "1Q9Q3KPWNsrc"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Vzh0_7k-pXdB",
        "outputId": "fc606ea7-9aaa-4dca-9198-c45387957311"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/100\n",
            "141/141 [==============================] - 202s 1s/step - loss: 3.1344 - year_output_loss: 1.8801 - genre_output_loss: 1.2543 - year_output_acc: 0.3198 - genre_output_acc: 0.5062 - val_loss: 4.5071 - val_year_output_loss: 2.5004 - val_genre_output_loss: 2.0067 - val_year_output_acc: 0.0842 - val_genre_output_acc: 0.0798\n",
            "Epoch 2/100\n",
            "141/141 [==============================] - 152s 1s/step - loss: 0.9134 - year_output_loss: 0.5379 - genre_output_loss: 0.3755 - year_output_acc: 0.8232 - genre_output_acc: 0.8716 - val_loss: 3.2695 - val_year_output_loss: 1.9597 - val_genre_output_loss: 1.3099 - val_year_output_acc: 0.2132 - val_genre_output_acc: 0.5025\n",
            "Epoch 3/100\n",
            "141/141 [==============================] - 164s 1s/step - loss: 0.1963 - year_output_loss: 0.1142 - genre_output_loss: 0.0821 - year_output_acc: 0.9744 - genre_output_acc: 0.9815 - val_loss: 2.4387 - val_year_output_loss: 1.4554 - val_genre_output_loss: 0.9833 - val_year_output_acc: 0.4172 - val_genre_output_acc: 0.6218\n",
            "Epoch 4/100\n",
            "141/141 [==============================] - 156s 1s/step - loss: 0.0476 - year_output_loss: 0.0257 - genre_output_loss: 0.0218 - year_output_acc: 0.9981 - genre_output_acc: 0.9978 - val_loss: 2.6717 - val_year_output_loss: 1.5064 - val_genre_output_loss: 1.1653 - val_year_output_acc: 0.4806 - val_genre_output_acc: 0.6291\n",
            "Epoch 5/100\n",
            "141/141 [==============================] - 148s 1s/step - loss: 0.0298 - year_output_loss: 0.0129 - genre_output_loss: 0.0169 - year_output_acc: 0.9994 - genre_output_acc: 0.9992 - val_loss: 3.4064 - val_year_output_loss: 1.8777 - val_genre_output_loss: 1.5287 - val_year_output_acc: 0.4791 - val_genre_output_acc: 0.6301\n",
            "Epoch 6/100\n",
            "141/141 [==============================] - 164s 1s/step - loss: 0.0283 - year_output_loss: 0.0159 - genre_output_loss: 0.0124 - year_output_acc: 0.9992 - genre_output_acc: 0.9991 - val_loss: 3.8446 - val_year_output_loss: 2.2069 - val_genre_output_loss: 1.6376 - val_year_output_acc: 0.4741 - val_genre_output_acc: 0.6311\n",
            "Epoch 7/100\n",
            "141/141 [==============================] - 169s 1s/step - loss: 0.0188 - year_output_loss: 0.0069 - genre_output_loss: 0.0119 - year_output_acc: 0.9994 - genre_output_acc: 0.9990 - val_loss: 4.2326 - val_year_output_loss: 2.4065 - val_genre_output_loss: 1.8261 - val_year_output_acc: 0.4663 - val_genre_output_acc: 0.6283\n",
            "Epoch 8/100\n",
            "141/141 [==============================] - 169s 1s/step - loss: 0.0165 - year_output_loss: 0.0081 - genre_output_loss: 0.0084 - year_output_acc: 0.9994 - genre_output_acc: 0.9993 - val_loss: 4.0961 - val_year_output_loss: 2.3448 - val_genre_output_loss: 1.7513 - val_year_output_acc: 0.4796 - val_genre_output_acc: 0.6114\n",
            "Epoch 9/100\n",
            "141/141 [==============================] - 167s 1s/step - loss: 0.0152 - year_output_loss: 0.0060 - genre_output_loss: 0.0092 - year_output_acc: 0.9997 - genre_output_acc: 0.9991 - val_loss: 4.2957 - val_year_output_loss: 2.3902 - val_genre_output_loss: 1.9055 - val_year_output_acc: 0.4752 - val_genre_output_acc: 0.6202\n",
            "Epoch 10/100\n",
            "141/141 [==============================] - 166s 1s/step - loss: 0.0152 - year_output_loss: 0.0048 - genre_output_loss: 0.0105 - year_output_acc: 0.9998 - genre_output_acc: 0.9991 - val_loss: 4.3643 - val_year_output_loss: 2.4292 - val_genre_output_loss: 1.9350 - val_year_output_acc: 0.4882 - val_genre_output_acc: 0.6101\n",
            "Epoch 11/100\n",
            "141/141 [==============================] - 169s 1s/step - loss: 0.0124 - year_output_loss: 0.0045 - genre_output_loss: 0.0079 - year_output_acc: 0.9998 - genre_output_acc: 0.9993 - val_loss: 4.8531 - val_year_output_loss: 2.6200 - val_genre_output_loss: 2.2331 - val_year_output_acc: 0.4760 - val_genre_output_acc: 0.6179\n",
            "Epoch 12/100\n",
            "141/141 [==============================] - 169s 1s/step - loss: 0.0116 - year_output_loss: 0.0048 - genre_output_loss: 0.0069 - year_output_acc: 0.9998 - genre_output_acc: 0.9996 - val_loss: 4.5476 - val_year_output_loss: 2.5371 - val_genre_output_loss: 2.0105 - val_year_output_acc: 0.4825 - val_genre_output_acc: 0.6205\n",
            "Epoch 13/100\n",
            "141/141 [==============================] - 169s 1s/step - loss: 0.0090 - year_output_loss: 0.0055 - genre_output_loss: 0.0035 - year_output_acc: 0.9997 - genre_output_acc: 0.9996 - val_loss: 5.0014 - val_year_output_loss: 2.7362 - val_genre_output_loss: 2.2652 - val_year_output_acc: 0.4687 - val_genre_output_acc: 0.6140\n"
          ]
        }
      ],
      "source": [
        "callback_year  = EarlyStopping(patience=4, monitor='val_year_output_loss')\n",
        "callback_genre = EarlyStopping(patience=4, monitor='val_genre_output_loss')# stop after overfitting on the 4th epoch\n",
        "\n",
        "model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['acc'])\n",
        "\n",
        "# indexing the X_train list of np arrays\n",
        "input_train = {'plot_input' : X_train[0:,0],\n",
        "    'origin_input' : X_train[0:,1],\n",
        "    'director_input' : X_train[0:,2],\n",
        "    'cast_input' : X_train[0:,3]}\n",
        "input_test = {'plot_input' : X_test[0:,0],\n",
        "    'origin_input' : X_test[0:,1],\n",
        "    'director_input' : X_test[0:,2],\n",
        "    'cast_input' : X_test[0:,3]}\n",
        "\n",
        "history = model.fit(input_train, \n",
        "                    {'genre_output': y_train[0:,0], 'year_output' : y_train[0:,1]},\n",
        "                    epochs=100, batch_size=64,\n",
        "                    validation_data=(input_test, {'genre_output' : y_test[0:,0], 'year_output' : y_test[0:,1]}),\n",
        "                   callbacks=[callback_year, callback_genre])"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "In order to prevent over fitting, Early Stopping is used on both the validation loss of year and genre. Also since each input goes to a different layer a dictionary for input had to be created so that each input can explicity be associated with its layers.<br><br>Lets now plot the validation loss given from predicting both genre and decade "
      ],
      "metadata": {
        "id": "p6NVHvM4OJsH"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "cA-7DWm9DvK3",
        "outputId": "96507063-79a1-44a7-c7d2-3b54ff27a5e7",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 573
        }
      },
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEWCAYAAABrDZDcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3dd3xUVfrH8c+TEAglBOmhgyBSEoKEJqIiKE1RAUUUUEHFlbWtsuqqq7uru7qWnw0LKIKCgIIuKCBNEQuChN6bIKGGDoGQdn5/nJswhCSkzOTOZJ736zUvZu7cufeZJMx37rnnnCvGGJRSSgWvELcLUEop5S4NAqWUCnIaBEopFeQ0CJRSKshpECilVJDTIFBKqSCnQaBKNBEZJyIv5HPdHSLSzdc1eZuIGBFp7HYdKnBpECilVJDTIFDKj4hIKbdrUMFHg0C5zmmSGSkiq0UkSUQ+EpEaIjJbRE6IyHwRuchj/T4isk5EjorIQhFp5vFcaxFZ7rxuChCebV/Xi8hK57W/iEhMPmusIiJfi8hxEflNRF4QkZ88nr9UROaJyGER2SQit3o8N05ERonITKeuJSJyscfzRkRGiMgWYEsR64wUkU9EJFFEdorIMyIS4jzXWER+EJFjInLQ+fkg1v+JyAHn/a0RkZb52Z8qIYwxetObqzdgB/ArUAOoDRwAlgOtsR/k3wHPOeteAiQB1wJhwF+BrUBp57YTeNR5rj+QCrzgvLa1s+32QChwp7PvMh51dMulxsnOrRzQHNgF/OQ8V955fDdQytnPQaC58/w44BDQznl+IjDZY9sGmAdUBspeqM4cajNAY+f+J8B0IAJoAGwGhjnPTQKexn4BDAeucJZ3B+KBSoAAzYAot/8u9FZ8Nz0iUP7ibWPMfmPMbuBHYIkxZoUxJhn4CvvhCDAAmGmMmWeMSQVexX54Xg50wAbAG8aYVGPMVOA3j33cB3xgjFlijEk3xowHzjivy5WIhAL9sGF0yhizHhjvscr1wA5jzMfGmDRjzApgGnCLxzpfGWOWGmPSsEEQm203/zHGHDbGnC5inbcBTxljThhjdgCvAYOdVVKB+kAtY0yyMeYnj+URwKWAGGM2GGP25rUvVbJoECh/sd/j/ukcHldw7tfCfusHwBiTgf02Xtt5brcxxnMmxZ0e9+sDjznNLUdF5ChQ13ldXqphv8nv8ljmeb8+0D7bdu8Aanqss8/j/imP95Pb9gpTZ1VsEHq+553Ynw3YoycBljpNa0MBjDHfAe8Ao4ADIjJaRCpeYF+qBNEgUIFmD/aDErDt29gPyd3AXqC2syxTPY/7u4AXjTGVPG7ljDGTLrDPRCANqOOxrG627f6QbbsVjDF/KsD78gyvwtZ5kLPf+jPVw/5sMMbsM8bca4ypBQwH3s3sdmqMecsY0wbb7HUJMLIAtasAp0GgAs3nQG8R6SoiYcBj2GaTX4DF2A/sh0QkTET6YtvlM40B7heR9s4J0vIi0ltEIvLaoTEmHfgSeF5EyonIpcAQj1W+AS4RkcHOfsNEpK3nSewCKkqdnwMvikiEiNQH/gJMABCRW0QkM8yOYMMnw6m1vfPzTAKSgYxC1q4CkAaBCijGmE3AIOBt7DfgG4AbjDEpxpgUoC9wF3AYez7hS4/XLgPuxTaDHMGeZL4rn7v+MxCJbeL5FHvi9Yyz3RPAddj2+T3OOi8DZQr5HotS54PYD/PtwE/AZ8BY57m2wBIROQnMAB42xmwHKmLD5wi2KekQ8EphaleBSc5tTlVK5YeIvAzUNMbc6XYtShWVHhEolQ/OOIEYp6mmHTAM25tJqYCnoxiVyp8IbHNQLWyPptew/fWVCnjaNKSUUkFOm4aUUirIBVzTUNWqVU2DBg3cLkMppQJKfHz8QWNMtZyeC7ggaNCgAcuWLXO7DKWUCigisjO357RpSCmlgpwGgVJKBTkNAqWUCnIaBEopFeQ0CJRSKshpECilVJDTIFBKqSCnQaCUUv7OGPjhv7BvjU82H3ADypRSKqikpcCMB2H1ZEg9BTWjvb4LDQKllPJXycdgymD4/Qfo8gxc+bhPdqNBoJQquZZ9DCf3Q+fHIDTM7WoK5lgCTLwFDm6Gm96H2IE+25UGgVKqZFo8Cub8zd7f9h30/xgia7tbU37tW2tD4MwJuGMqXNzFp7vTk8VKqZJn6RgbAs36QL+PYP86+KAzbPve7coubNt3MLaHvT/0W5+HAGgQKKVKmvhxMOtxuKSnDYHo/nDfQihfHT692fa+ychwuchcrJhojwQq1YN75kPNlsWyWw0CpVTJsfIz+PoRaNwNbh0PpUrb5VWbwL0LIOZW+P5F+OwWOHXY3Vo9GQMLX4bpD0D9TjB0drE2Y2kQKKVKhjVTYfoIaHQVDJgApcqc+3zp8nDzB3D9G/D7Ini/MyT4wbVN0lNt99CF/4ZWA+05gfDIYi1Bg0ApFfjWT4cv74N6HeG2SRBWNuf1RCDubhg2F0JCbFv8ktH2G7kbzpyAzwbAik/hqifgpvfOHsUUIw0CpVRg2zQbpg6F2m3g9ilQutyFX1OrNdz3AzTuCrNHwrRhcOak72v1dHwvfNwTti+EPm9Dl7/ZoHKBBoFSKnBtmQ+fD4GaMTBoKpSJyP9ry1W2Rw9dn4N1X8GYLnBgo+9q9XRgA3x0LRz+HW7/HC4bUjz7zYUGgVIqMG1fCJNvh2pNYfCXhWtXDwmBzn+BIdPh9BEbBqu/8Hqp5/h9EXzUHdJT4O5Z0KSbb/eXDxoESqnAs+Nn+Ow2qNIYBk+HshcVbXsNr4ThP0JULHx5D8x8DNLOeKdWT6u/gE/7QkRN2z00qpX391EIGgRKqcCyayl8ditUqmu/yZev4p3tVoyCO7+Gyx+C3z6Esd3hyE7vbNsY+PF1GzJ128OwOXasgJ/QIFBKBY7d8TChH1SoDkNmQIVq3t1+aCm47l8wYCIc2g4fXAmb5xRtm+lpMPMvsOAf0LK/bcYq6hGMl2kQKKUCw95VdmRw2YvsN/eKUb7bV7PrYfhCe9Tx2a2w4F+QkV7w7aQkwZQ7YNlYuOJR6Dvm/PENfkCDQCnl//avg09ugjIVbQhE1vH9Pis3gmHzbI+eH1+FT2+Ck4n5f/3JAzCuN2yZC71fg27P25PTfsg/q1JKqUyJm+GTG+036TtnwEX1i2/fYWVtH/8b37XnJj7oDDsXX/h1B7fAh90gcZPtotr2Ht/XWgQaBEop/3VoG4y/ARB7JFC5kTt1tL7D9vIJK2u/5f/ydu6jkXcutmMEUk/BXd9A0x7FW2shaBAopfzTkR02BDJS7ZFA1Sbu1lMz2s5iemkvmPsMTBlkryDmad1X9uilXBXbrFS7jRuVFpjPgkBE6orI9yKyXkTWicjDOawjIvKWiGwVkdUicpmv6lFKBZCju2DcDfZk65DpUL2Z2xVZ4ZFw66fQ/d+w+VsYfbW9oLwx8Ms78MVddvqKYfOgckO3q803X16hLA14zBizXEQigHgRmWeMWe+xTk+giXNrD7zn/OsbqckQFu6zzSulvOD4HnskkHwM7pzuk4u1F4kIdBxhv+1/cZc9F3BxV9g0E5rfCDePDrjPGZ8dERhj9hpjljv3TwAbgOwTbN8IfGKsX4FKIuKbPmHrZ8Brl9g/MqWUfzqxH8b3gaREGDTNfrv2V/U62NHIddvbEOj4Z+g/LuBCAIrpmsUi0gBoDSzJ9lRtYJfH4wRn2d5sr78PuA+gXr1CjsarGQ3Jx+3Vi7r8rXDbUEr5TtJB275+fDcM+hLqtnW7ogurUA0GfwWHtto5jwKUz08Wi0gFYBrwiDHmeGG2YYwZbYyJM8bEVatWyJGElRtCk2ttEKSnFm4bSinfOHXYjhM48rudSrp+R7cryr+Q0IAOAfBxEIhIGDYEJhpjvsxhld1AXY/HdZxlvtH2Xji5HzZ87bNdKKUK6PRRO2L44Ca47TM7AZwqVr7sNSTAR8AGY8zruaw2Axji9B7qABwzxuzNZd2ia9wVKtW3E0oppdx35gRM7G9HDg+YYP+PqmLnyyOCTsBg4BoRWenceonI/SJyv7POLGA7sBUYAzzgw3rsIVzbYbDzZ9i//sLrK6V8JyUJJt4Cu5fDLR/DJd3driho+exksTHmJyDP664ZYwwwwlc15Kj1YPjuRXtUcH1uBypKKa9JOwMn9jm3vc6/e+wFWvaugn4fQbMb3K4yqBVLryG/Uq4ytOwHq6fYSaDCK7pdkVKBKT3VTqyW9QG/N4cP/L1w+vD5rw0tDRVr2dk4W/Yt/trVOYIvCADa3QOrPrNh0O5et6tRyv+cTLTf2rN/qJ/YZ8finNhn+/qTbb4dCYUKNewVuC5qYPvaR0TZx5n/Vqxlp5J26ULt6nzBGQS129iBKkvH2FkB9Q9SKWvPCpj/D9j+/fnPla/mfJhH2f8/2T/gI6KgfFV7Lk4FlOAMArBdSac/ADt+goad3a5GKXcd3ALfvQDr/2cnTOvyDFS/9OyHfIUaEBrmdpXKR4I3CFr2hblPw29jNAhU8Dq2G354GVZMsFMsX/WknUdHz50FleANgrCy0HoQLH7XtnlWrOV2RUoVn1OH4af/g6Wj7SUY290LnR/3/jWAVUAI7usRxA0DkwHx492uRKnikZIEi16FN2PtxVVa9IUH46HnyxoCQSx4jwjg3PmHrnxc20BVyZWWAsvHww//haQD0LQ3XPMM1GjudmXKDwT3EQHYXkMn9+n8Q6rozpyAjAy3qzhXRgas/gJGtYVZj0OVxjB0Lgz8TENAZQnuIwKAxt2c+Yc+0oEtqnDOnLQ9bpa8bwcsNuhsJ05reBVUudid7snGwJZ5sOCfsH8N1IiGO6bav3ftLq2y0SAICYW4oTD/OTv/kH5LUgWx7Tv4+mE4+oftfJCRAb//YLthAkTUckLBuVWqm/f2vOGPX+1YgD9+sYO6+n1kzwWEaAOAypkGAdj5h77/t84/pPLv9BF7AfMVE2xzy92zof7l9jlj4PB2O5fO74tg63xYPdk+d1HDc4OhQnXv1bR/HSz4F2yebfv9934NWg+BUqW9tw9VIomd9y1wxMXFmWXLlnl/w1/9CTbMgL9s0D7UKm8bvoaZj9kranV6GK56Iu/LExoDBzacDYYdP8GZY/a5as3OhkKDTnbqhYI6sgO+/4+dMqVMRbjiYWh/P5QuX6i3p0omEYk3xsTl+JwGgSMhHj68Bnq9qvMPqZyd2A+zR8J654Lqfd6BWrEF305Gup11MzMY/lgMqacAgahWZ88v1OsAZSrkvp2TB2xX0GVjbRNn++HQ6RF7nkKpbDQI8mv01ZB6Gh74VU+oqbOMgVWT4dsn7d/H1U/A5Q95r7txWgrsjj8bDAlLIT0FQkpB7bizRwx12tojj+TjdgzA4lGQlgyXDbZHJTooUuVBgyC/Vky08w/d+Y1OO6Gso3/A14/AtgVQt709Cqh2iW/3mXIKdi05Gwx7ltuBj6XCbRjsX2endm5xs50TqGpj39ajSoS8gkBPFnvS+YdUpowMWPYRzH/eHhH0fMWOOSmOnjely8HFXewNIPkY7FzsnF9YBHXioMvf7AygSnmBBoGnc+Yf2gsVo9yuSLnh4BaY/mfY9StcfA1c/wZcVN+9esIjoWkPe1PKB7RjcXZxQ535h8a5XYkqbump8ONr8F4nSNwIN70Hg750NwSUKgYaBNlVbmRHX8aPsx8MKjjsXQVjrrEjcZv2gBFLIfZ27TSggoIGQU7a3WvnH9r4jduVKF9LTbajcEd3sZdfvPVTuPUTiKjhdmVKFRs9R5CTxt2gUj1Y+qHtmaFKpp2LYcaf4dBWiB0E3V8o3IAupQKcHhHkJCTUXqtg5092RKgqWc6cgJmPw8c9bB/+QV/CTaM0BFTQ0iDITevBEFrGzj+kSo6t8+Hdjvb32v5+eGAxNO7qdlVKuUqDIDflq9hxBasm25GcKrCdOgxf3Q8T+tluwkPn2Kty5TWFg1JBQoMgL23vhZSTdjIvFZiSDtpxIaPawZov7HV5h/8I9dq7XZlSfkNPFuel9mUQFWubEdreo10JA0V6KmyZCys/g83fQkaanR6i16sQFeN2dUr5HQ2CvIjYrqTTR9ipg3XaCf+2b4398F/9OZw6COWr2/MAsXfoBYeUyoMGwYW07AdznrZHBRoE/ifpoG3yWTnRBkFIGDTtaT/8G3f13gyhSpVgGgQXkjn/0JL3df4hf5Geaq/Hu3IibJ4DGam2Ca/nKxDdX+fjV6qANAjyo+0wO/d7/Djo8pTb1QSvfWudpp8pTtNPNXsxltjboUYLt6tTKmBpEOSH5/xDVz6uzQ3FKekQrJ1qv/3vXeU0/fSwI4G16Ucpr9AgyK+298CkAXb+IZ12wrfSU+3Ar5UTYdO3TtNPK+j5X2jZ347xUEp5jQZBfjW51s4/9NtHGgS+sn/d2aafpMSzTT+tBkLNlm5Xp1SJpUGQXyGh9loF85+38w9Vb+Z2RSXDqcOwJrPpZ6VH088dtjlOm36U8jmfjSwWkbEickBE1uby/NUickxEVjq3v/uqFq9pPUTnH/Km43vhvcth9kh7MaAeL8Njm2DABNsFVENAqWLhyyOCccA7wCd5rPOjMeZ6H9bgXZ7zD3V7HspEuF1R4Eo7A58PsfM4DZ0D9Tq4XZFSQctnRwTGmEXAYV9t3zVt77HzD62a7HYlgW32E5CwFG56V0NAKZe5PelcRxFZJSKzRSTXjuAicp+ILBORZYmJicVZ3/lqtzk7/5Ax7tYSqOLHQfzHcMWj0OImt6tRKui5GQTLgfrGmFbA28D/clvRGDPaGBNnjImrVq1asRWYo8z5hxI3ws6f3a0lEO1aai8Kc3FXuOZZt6tRSuFiEBhjjhtjTjr3ZwFhIlLVrXoKpEVfCK8ES8e4XUlgObEPpgyGyNrQ70PbE0sp5TrXgkBEaorYeZ1FpJ1TyyG36imQ0uXs/EMbv7E9X9SFpaXYk8NnjsNtn+l8QEr5EV92H50ELAaaikiCiAwTkftF5H5nlf7AWhFZBbwF3GZMADW6xw2189wvH+92JYHh2ydg1xK4cZTOC6SUn/FZ91FjzMALPP8OtntpYKpysR3wtOxj6PyY9nnPS/x4WDYWOj1su98qpfyK272GAlvbe+HkPtg40+1K/Neu32DW43DxNdD1OberUUrlQIOgKJpcC5H1dKRxbk7sh88HQ0QU9PtITw4r5ac0CIoiJBTaDoUdP9r5h9RZmSeHk4/pyWGl/JwGQVFlzT/0kduV+Jc5T8GuX6HP2zpzqFJ+ToOgqMpXsdNSr5oMZ064XY1/WP6pbS67/CF76UillF/TIPCGdvdCygmdfwggIR5m/gUaXa0nh5UKEBoE3lC7jb2C1m8fBff8QycPwJRBEFET+n8MoXq5C6UCgQaBN4jYrqSJG4J3/qG0FPj8Tjh9BAZM1JPDSgUQDQJvadnPzj8UrF1J5z4Nf/wCN74DUTFuV6OUKgANAm/JnH9ow9d2crVgsmIiLB0NHf+sJ4eVCkDaiOtNcUNh8Tvw/b/hhjdtk1FJtzsevnkUGl4F3f7hdjXKj6WmppKQkEBycrLbpZRo4eHh1KlTh7Cw/E97o0HgTVUutvPp/PwmVG4EVzzidkW+dfKAnVa6Qg09OawuKCEhgYiICBo0aIAEw5ckFxhjOHToEAkJCTRs2DDfr9OmIW/r+rw9XzD/OVj9udvV+E56KnxxF5w6DLdNsOMplMpDcnIyVapU0RDwIRGhSpUqBT7q0q9w3hYSAje9Z78t/+8BKF8NLu7idlXeN/cZ20Oq74e266xS+aAh4HuF+RkHzRFBRoZhdcJRiuWSB6XKwG0Toeoltulk72rf77M4rZwES96HDiMg5ha3q1EqX44ePcq7775bqNf26tWLo0ePerki/xE0QTB1eQJ93vmZzftPFs8OwyNh0FT778Rb4OgfxbNfX9uzAr5+GBp0hmv/6XY1SuVbXkGQlpaW52tnzZpFpUqVfFFWgVyozsIKmiDo0rQ6IQIz1xTjpSUr1rJhkHYaJvSz7emB7GQiTB4EFarDLeP05LAKKE8++STbtm0jNjaWkSNHsnDhQjp37kyfPn1o3rw5ADfddBNt2rShRYsWjB49Ouu1DRo04ODBg+zYsYNmzZpx77330qJFC6677jpOnz593r62bdtGhw4diI6O5plnnqFChQpZz73yyiu0bduWmJgYnnvOTsOS13avvvpqHnnkEeLi4njzzTeJj4/nqquuok2bNnTv3p29e4v+mRY0/5OrRZShXcPKzFqzl79ce0nx7bh6M7htEnx6E0waCEP+B2Fli2//3pJ1cvggDJ0D5au6XZEKYP/4eh3r9xz36jab16rIczfkfhnUl156ibVr17Jy5UoAFi5cyPLly1m7dm1WD5uxY8dSuXJlTp8+Tdu2benXrx9VqpzbEWLLli1MmjSJMWPGcOuttzJt2jQGDRp0zjoPP/wwDz/8MAMHDuT999/PWj537ly2bNnC0qVLMcbQp08fFi1aRL169fLcbkpKCsuWLSM1NZWrrrqK6dOnU61aNaZMmcLTTz/N2LFji/SzC5ojAoDe0VFsPXCSLfuLeZbQBp2g72h7zd4v74WM9OLdvzfMfRZ2/mTHR9SKdbsapbyiXbt253SzfOutt2jVqhUdOnRg165dbNmy5bzXNGzYkNhY+3+gTZs27Nix47x1Fi9ezC232PNnt99+e9byuXPnMnfuXFq3bs1ll13Gxo0bs/aR13YHDBgAwKZNm1i7di3XXnstsbGxvPDCCyQkJBTth0AQHREAdG9Zk7/PWMfMNXt5pEZE8e68xc12xPG3T9pbz/8GzoCzVZNhyXvQ/k/Q6ja3q1ElQF7f3ItT+fLls+4vXLiQ+fPns3jxYsqVK8fVV1+dYzfMMmXKZN0PDQ3NsWkoN8YYnnrqKYYPH37O8h07duS53cw6jTG0aNGCxYsX53uf+ZGvIwIReVhEKor1kYgsF5HrvFpJMageEU7bBpWZvcalKSA6/MlOw7B0tB10Fgj2rDx7cvi6f7ldjVKFFhERwYkTubcGHDt2jIsuuohy5cqxceNGfv3110Lvq0OHDkybNg2AyZPPTk/fvXt3xo4dy8mTttPK7t27OXDgQL6327RpUxITE7OCIDU1lXXr1hW6zkz5bRoaaow5DlwHXAQMBl4q8t5d0KtlTTbtP8HWA8XUeyi7a/8VOAPOkg7aaaXLVXVGDud/yLpS/qZKlSp06tSJli1bMnLkyPOe79GjB2lpaTRr1ownn3ySDh06FHpfb7zxBq+//joxMTFs3bqVyMhIAK677jpuv/12OnbsSHR0NP37988znLIrXbo0U6dO5YknnqBVq1bExsbyyy+/FLrOTJKffvUistoYEyMibwILjTFficgKY0zrIldQQHFxcWbZsmWFfv3+48l0+M8C/tLtEh7s2sSLlRVA2hnbi+iPX+GOL/xzwFlqMkzsD7uWwrA5UKvYf9WqhNmwYQPNmjVzu4xicerUKcqWLYuIMHnyZCZNmsT06dOLbf85/axFJN4YE5fT+vk9IogXkblAL2COiEQAGUWq1CU1KoYTV/+i4u1Gml2pMjBgAlRt4p8Dzrb/AO9dDjt+dE4OawgoVRDx8fHExsYSExPDu+++y2uvveZ2SXnKbxAMA54E2hpjTgFhwN0+q8rHeraMYuO+E2xPdKl5CKBsJbhjKoRX9J8BZ0kH4av74ZM+gIHB/4PYgW5XpVTA6dy5M6tWrWL16tUsWrSIxo0bu11SnvIbBB2BTcaYoyIyCHgGOOa7snyrZ3RNAGavdfm6AZG1YdA0SHV5wJkx9poC77SFNVOh8+Pwp1/8s8lKKeV1+Q2C94BTItIKeAzYBnzis6p8LCqyLJfVq8QsN5uHMlVvBgM/gyM77ICz1Px3RfOKg1tg/A0w/QE7N9L9P0LXZwNz0JtSqlDyGwRpxp5VvhF4xxgzCijmjvje1Ss6inV7jrPzUJLbpUCDK+DmD2DXr8U34CztDCx82Z4L2Lfangu4e7YNJqVUUMlvEJwQkaew3UZnikgI9jxBwOoZHQXALLfGFGTXsi90/4+91OW3T9rmGl/Z8RO81wkW/hua3QAjfoM2d9kptJVSQSe///MHAGew4wn2AXWAV3xWVTGoXakssXX9pHkoU8cHfDvg7NRhmD4CxvWG9BS4Yxr0HwsRNby/L6X8TFGmoQY7NuDUqVNerMh/5CsInA//iUCkiFwPJBtjAvYcQaZe0TVZs/sYfxzyo1/utf+CFn29O+DMGFg1xZ4MXjkJOj0CD/wKTbp5Z/tKBYBADIL09OKZlyy/U0zcCiwFbgFuBZaISH9fFlYcera0zUOz1/rRUUFICNz8vp3S4X8PwPaFRdveoW125tOv7oOLGsDwRXDtP6B0OW9Uq1TAyD4NNeQ8JXRSUhK9e/emVatWtGzZkilTpvDWW2+xZ88eunTpQpcu5/emmzVrFpdeeilt2rThoYce4vrrr8/a1tChQ2nXrh2tW7fOGlQ2btw4+vbtS48ePWjSpAl//etfs7ZVoUIFHnvsMVq1asXixYuZMGEC7dq1IzY2luHDh/skHPI76dzT2DEEBwBEpBowH5jq9YqKUd3K5YipE8msNXsZftXFbpdzVuaAs4972vn/h86GmtEF20ZaCvzyJvzwit1er1chbiiEhPqmZqUKYvaTsG+Nd7dZMxp65j7zTfZpqHObEjoxMZFatWoxc+ZMwM5BFBkZyeuvv873339P1arnTsGenJzM8OHDWbRoEQ0bNmTgwLNjb1588UWuueYaxo4dy9GjR2nXrh3dutkj8ZUrV7JixQrKlClD06ZNefDBB6lbty5JSUm0b9+e1157jQ0bNvDyyy/z888/ExYWxgMPPMDEiRMZMmSIV390+T1HEJIZAo5DBXitX+sVHcWqhGMkHPGj5iE4d8DZhP4FG3D2x6/wwZXw3QvQtAeMWArt7tUQUMpDblNCR0dHM2/ePJ544gl+/PHHrHmCcrNx40YaNWqUNZ21ZxDMnTuXl156idjY2KzZTP/4w/5f7tq1K5GRkYSHh9O8eXN27twJ2JlH+/XrB8CCBQuIj4+nbdu2xMbGsmDBArZv3+71n0TUEekAABUwSURBVEV+jwi+FZE5wCTn8QBgltercUGvllG8NHsjs9fs494rG7ldzrkia9swGNvDhsHQb6Fc5dzXP30E5j8P8eMgsi4MnGKDQCl/k8c39+KS25TQAMuXL2fWrFk888wzdO3alb///e+F3se0adNo2rTpOcuXLFly3rTTmZehDA8PJzQ0NOv1d955J//5z38Ktf/8yu/J4pHAaCDGuY02xjyR12tEZKyIHBCRtbk8LyLylohsFZHVInJZQYv3hnpVytGydkVm+dN5Ak81mjsDzn6HybfbyeCyM8aOCH6nHSz/xPY8euBXDQGlPGSfhjq3KaH37NlDuXLlGDRoECNHjmT58uU5vj5T06ZN2b59e9aFZKZMmXLOPt5++20yJ/dcsWJFgWru2rUrU6dOzZqq+vDhw1lHDt6U7wvTGGOmAdMKsO1xwDvkPgK5J9DEubXHjl5uX4Dte02v6Cj+++0m9hw9Ta1KfjiiNnPA2dS77YCzW8adbeY5sgNmPgZb59vJ4QZNhahWblarlF/ynIa6Z8+evPLKK2zYsIGOHTsC9iTthAkT2Lp1KyNHjiQkJISwsDDee+89AO677z569OhBrVq1+P7777O2W7ZsWd5991169OhB+fLladu2bdZzzz77LI888ggxMTFkZGTQsGFDvvnmm3zX3Lx5c1544QWuu+46MjIyCAsLY9SoUdSvX99LPxUrz2moReQEkNMKAhhjTMU8Ny7SAPjGGNMyh+c+wE5pPcl5vAm42hiT51fzok5DnZMdB5O4+tWFPHt9c4Zd0fDCL3DL4lEw52/Qbjh0f9E+XviSDYVrntXzAMqvleRpqE+ePEmFChUwxjBixAiaNGnCo48+6lo9BZ2GOs8jAmOML6eRqA3s8nic4Cw7LwhE5D7gPoB69ep5vZAGVcvTPKois9bs9e8g6DgCju+Bxe/Axm/g+G5o2ht6/Rci67hdnVJBa8yYMYwfP56UlBRat26d43kHfxYQ1yw2xozGnqMgLi7OJ3Mv9IquyatzN7PvWDI1I8N9sQvvuPZfdrronT/DgInQ7Hq3K1Iq6D366KOuHgEUlZtdQHcDdT0e13GWuaJXtB8OLstJSAj0/QAeWaMhoJTyCjeDYAYwxOk91AE4dqHzA77UqFoFLq0Z4d6F7QtKxO0KlCqw/FwaVxVNYX7GPgsCEZkELAaaikiCiAwTkftF5H5nlVnAdmArMAZ4wFe15Fev6Ch+23mYA8dz6KKplCqS8PBwDh06pGHgQ8YYDh06RHh4wZq3fXaOwBiT5zUOnesbjPDV/gujV3RNXp+3mdlr93Hn5Q3cLkepEqVOnTokJCSQmJjodiklWnh4OHXqFKzzSECcLC4ujatHcEmNCsxas1eDQCkvCwsLy5qGQfmXEjFfkDf1bBnF0h2HOXBCm4eUUsFBgyCb3jFRGANz1u13uxSllCoWGgTZNKlegYurlWfWaj/vRqqUUl6iQZCNiNA7Ooolvx/i4MkzbpejlFI+p0GQg14xUWQYmLMuQMYUKKVUEWgQ5KBpjQgaVS0fOIPLlFKqCDQIciAi9IqOYvH2QxxOSnG7HKWU8ikNglz0jK5JeoZhrjYPKaVKOA2CXDSPqkiDKuWYuUZ7DymlSjYNglyICD2jo/hl2yGOaPOQUqoE0yDIQ+/oKNIzDPPW6+AypVTJpUGQhxa1KlK3clltHlJKlWgaBHnI7D3089aDHDuV6nY5SinlExoEF9CrZRRpGYa567X3kFKqZNIguICYOpHUrlSW2Ws1CJRSJZMGwQXY5qGa/LglkWOntXlIKVXyaBDkQ6/oKFLTDQs2aO8hpVTJo0GQD7F1K1ErMpxZ2ntIKVUCaRDkQ+bgskWbD3IiWZuHlFIliwZBPvWKjiIlPYMFGw64XYpSSnmVBkE+ta5biZoVtXlIKVXyaBDkU0iI0DO6Jgs3J3LyTJrb5SillNdoEBRAr+goUtIytPeQUqpE0SAogDb1LqJ6RBm9cplSqkTRICiAkBChZ8uafL/pAEnaPKSUKiE0CAqoV3QUZ9Iy+H6T9h5SSpUMGgQFFNegMlUrlNHeQ0qpEkODoIBCM5uHNiZyKkWbh5RSgU+DoBB6RtfkdGo6Czclul2KUkoVmQZBIbRvWIWqFUpr85BSqkTQICiE0BChe4uafLfxAMmp6W6Xo5RSRaJBUEi9oqM4laLNQ0qpwKdBUEjtG1amcnltHlJKBT6fBoGI9BCRTSKyVUSezOH5u0QkUURWOrd7fFmPN5UKDaF7ixos2LBfm4eUUgHNZ0EgIqHAKKAn0BwYKCLNc1h1ijEm1rl96Kt6fKFXdBRJKeks2qzNQ0qpwOXLI4J2wFZjzHZjTAowGbjRh/srdh0aVaFSuTBtHlJKBTRfBkFtYJfH4wRnWXb9RGS1iEwVkbo+rMfrwkJD6N68JvM3HOBMmjYPKaUCk9sni78GGhhjYoB5wPicVhKR+0RkmYgsS0z0r2aYntE1OXkmjR83H3S7FKWUKhRfBsFuwPMbfh1nWRZjzCFjzBnn4YdAm5w2ZIwZbYyJM8bEVatWzSfFFlanxlWJLBvGrLXaPKSUCky+DILfgCYi0lBESgO3ATM8VxCRKI+HfYANPqzHJ8JCQ7i2eQ3mrd+vzUNKqYDksyAwxqQBfwbmYD/gPzfGrBORf4pIH2e1h0RknYisAh4C7vJVPb7UOzqKE8lp/LL1kNulKKVUgZXy5caNMbOAWdmW/d3j/lPAU76soTh0alyViPBSzFyzly6XVne7HKWUKhC3TxaXCKVL2eahuev2kZKW4XY5SilVIBoEXtI7OorjyWn8sk17DymlAosGgZdc0aQqEWVK6YXtlVIBR4PAS8qUCqVb8xrMWb+P1HRtHlJKBQ4NAi/q2bImR0+lsnib9h5SSgUODQIvuvKSalQqF8bzX6/jwIlkt8tRSql80SDwovCwUEYPjmPfsWRuH7OExBNnLvwipZRymQaBl7VrWJmP72rL7iOnuePDXzl4UsNAKeXfNAh8oH2jKoy9qy1/HD7FoA+XcEjDQCnlxzQIfKTjxVX46M62/H4wiTs+XMLhpBS3S1JKqRxpEPhQp8ZVs8Jg0IdLOHpKw0Ap5X80CHzsiiZVGTMkjq2JJ7njwyUcO5XqdklKKXUODYJicOUl1Rg9uA1b9p9k0EdLOHZaw0Ap5T80CIrJ1U2r8/7gy9i47zhDNAyUUn5Eg6AYXXNpDd67ow3r9x7nzrFLOZGsYaCUcp8GQTHr1rwGo26/jLW7j2kYKKX8ggaBC65rUZN3br+M1QnHuOvj3zh5Js3tkpRSQUyDwCU9Wtbk7YGtWbnrKHd/vJQkDQOllEs0CFzUMzqKt25rzfI/jnL3uN84laJhoJQqfhoELusdE8UbA2JZtuMwQ8f9xumUdLdLUkoFGQ0CP3BDq1r834BYlv5+mGHjNQyUUsVLg8BP3Bhbm9dubcXi7Ye495NlJKdqGCiliocGgR+5uXUdXu3fip+3HdQwUEoVGw0CP9OvTR1e7hfDT1sPMvzTeA0DpZTPaRD4oVvj6vJS32h+2JzInybEcyZNw0Ap5TsaBH5qQNt6/PvmaL7flMiIictJSctwuySlVAmlQeDHbm9fjxduasn8DQcY8ZmGgVLKNzQI/NygDvX5540tmLd+Pw9OWk5quoaBUsq7NAgCwJCODXj+hubMWbefhyat0DBQSnmVBkGAuKtTQ/5+fXNmr93HI5NXkqZhoJTyklJuF6Dyb+gVDckwhhdmbuBQ0hm6XlqDVnUr0bJ2RcqV1l+lUqpw9NMjwNzTuRFhoSGMXrSdF2dtACBEoEn1CGLqRBJTtxKxdSrRtGYEpUvpAZ9S6sLEGON2DQUSFxdnli1b5nYZfiHxxBnW7D7Kyl3HWJ1wlNUJxziclAJA6dAQmtWqSKs6kcTUqURs3UgaVa1ASIi4XLVSyg0iEm+MicvxOQ2CksMYQ8KR06xOOMaqhKOs2nWUtbuPkeRMYlehTCla1q5IqzqViKlTiVZ1I6ldqSwiGg5KlXR5BYE2DZUgIkLdyuWoW7kcvWOiAEjPMGxPPMmqBHvUsGrXUT7+eQcpzsnmKuVL2yYlJxhi6lSiaoUybr4NpVQx82kQiEgP4E0gFPjQGPNStufLAJ8AbYBDwABjzA5f1hRsQkOEJjUiaFIjgv5t6gBwJi2dTftO2HDYdZRVCUdZuDmRzIPD2pXK0qpuJC1qRVK+dCilQkMoFSKEhgilQoVSIWcfh4WG2OUhQqlz7tv17DrOa3N4bNcTPSpRykU+CwIRCQVGAdcCCcBvIjLDGLPeY7VhwBFjTGMRuQ14GRjgq5qUVaZUKDFO8xAd6gOQdCaNtbuPZTUrrU44xqw1+4qtphAhKyhKhQihTkB4Bkjm48xACvUIpHP/DTlnG6VCQggRyMwa4WzoZC07J4fkvGXC+evltJ0QOVujvQ+hYmsKDYGQEHEeS9a6mctKZd4P8diOyDmvCQ0RRMAYyGzUNcZk3ceAwWSFek7rGY+Vs9bLWtec87rMpuMMY9f1XG6y7c94rudsJGu7HvczsjVHZ/0cxB7Vet7P/BmGiP2yECrnr2efI+tnKkLWz82eErPLBPs6cX5f4iz3/B1mLsu8f3Z55u/43G1l/m141uD5u7V1cM4yf/zi48sjgnbAVmPMdgARmQzcCHgGwY3A8879qcA7IiIm0E5clADly5SifaMqtG9UJWtZ0pk0UtIySMswpGVkkJZuSM+8n2FISzekZRjSnefsevZxata6hrT0DGf5ufdT0+3jdHN23XRnu+kZHq8551/POs7Wk5qewenUs69P83x9usn68PH8y/L80Du7jPOWkcd6drnJWpaeYcjIMKQbQ0YGWe9NKU+eYRXqEWrZg+TcLxYwsF097uncyOv1+DIIagO7PB4nAO1zW8cYkyYix4AqwEHPlUTkPuA+gHr16vmqXpVN+TKlKK+nC7wiwwmuDCcYbFB43ndCI91khUfWutnuZ8r8hpp5/+xRi3h8g839m2/W/WzrnXv0c/YbeE7fpMXjucxvxZllSQ7fxEOcBZnLM5yjhQznSCEj4+z99AyTdQRx9pbTcziPz7+f+brzjl5yPVIy2Z4794gnK/SzjqQy6zhb/7m/W7KWZf7+0nP8OyCPv42z2/HV+buAOFlsjBkNjAbba8jlcpQqsJAQobR23VV+ypcjjnYDdT0e13GW5biOiJQCIrEnjZVSShUTXwbBb0ATEWkoIqWB24AZ2daZAdzp3O8PfKfnB5RSqnj5rGnIafP/MzAH2310rDFmnYj8E1hmjJkBfAR8KiJbgcPYsFBKKVWMfHqOwBgzC5iVbdnfPe4nA7f4sgallFJ501nJlFIqyGkQKKVUkNMgUEqpIKdBoJRSQS7gpqEWkURgZyFfXpVso5YDmL4X/1RS3ktJeR+g7yVTfWNMtZyeCLggKAoRWZbbfNyBRt+Lfyop76WkvA/Q95If2jSklFJBToNAKaWCXLAFwWi3C/AifS/+qaS8l5LyPkDfywUF1TkCpZRS5wu2IwKllFLZaBAopVSQC5ogEJEeIrJJRLaKyJNu11NYIlJXRL4XkfUisk5EHna7pqIQkVARWSEi37hdS1GISCURmSoiG0Vkg4h0dLumwhKRR52/rbUiMklEwt2uKb9EZKyIHBCRtR7LKovIPBHZ4vx7kZs15lcu7+UV529stYh8JSKVvLGvoAgCEQkFRgE9gebAQBFp7m5VhZYGPGaMaQ50AEYE8HsBeBjY4HYRXvAm8K0x5lKgFQH6nkSkNvAQEGeMaYmdQj6QpocfB/TItuxJYIExpgmwwHkcCMZx/nuZB7Q0xsQAm4GnvLGjoAgCoB2w1Riz3RiTAkwGbnS5pkIxxuw1xix37p/AfuDUdreqwhGROkBv4EO3aykKEYkErsReXwNjTIox5qi7VRVJKaCsc9XAcsAel+vJN2PMIuy1TTzdCIx37o8HbirWogopp/dijJlrjElzHv6KvfJjkQVLENQGdnk8TiBAPzw9iUgDoDWwxN1KCu0N4K9AhtuFFFFDIBH42Gnm+lBEyrtdVGEYY3YDrwJ/AHuBY8aYue5WVWQ1jDF7nfv7gBpuFuNFQ4HZ3thQsARBiSMiFYBpwCPGmONu11NQInI9cMAYE+92LV5QCrgMeM8Y0xpIInCaH87htJ/fiA23WkB5ERnkblXe41wKN+D7zIvI09hm4one2F6wBMFuoK7H4zrOsoAkImHYEJhojPnS7XoKqRPQR0R2YJvqrhGRCe6WVGgJQIIxJvPIbCo2GAJRN+B3Y0yiMSYV+BK43OWaimq/iEQBOP8ecLmeIhGRu4DrgTu8dY33YAmC34AmItJQREpjT37NcLmmQhERwbZFbzDGvO52PYVljHnKGFPHGNMA+/v4zhgTkN88jTH7gF0i0tRZ1BVY72JJRfEH0EFEyjl/a10J0BPfHmYAdzr37wSmu1hLkYhID2xzah9jzClvbTcogsA5ufJnYA72j/pzY8w6d6sqtE7AYOw36JXOrZfbRSkeBCaKyGogFvi3y/UUinNUMxVYDqzBfkYEzBQNIjIJWAw0FZEEERkGvARcKyJbsEc8L7lZY37l8l7eASKAec7//fe9si+dYkIppYJbUBwRKKWUyp0GgVJKBTkNAqWUCnIaBEopFeQ0CJRSKshpEChVjETk6kCfaVWVPBoESikV5DQIlMqBiAwSkaXOoJ0PnOsmnBSR/3Pm6l8gItWcdWNF5FePOeIvcpY3FpH5IrJKRJaLyMXO5it4XLtgojOCVynXaBAolY2INAMGAJ2MMbFAOnAHUB5YZoxpAfwAPOe85BPgCWeO+DUeyycCo4wxrbDz9WTOgNkaeAR7bYxG2NHiSrmmlNsFKOWHugJtgN+cL+tlsROVZQBTnHUmAF861yKoZIz5wVk+HvhCRCKA2saYrwCMMckAzvaWGmMSnMcrgQbAT75/W0rlTINAqfMJMN4Yc87Vn0Tk2WzrFXZ+ljMe99PR/4fKZdo0pNT5FgD9RaQ6ZF3ztj72/0t/Z53bgZ+MMceAIyLS2Vk+GPjBuXpcgojc5GyjjIiUK9Z3oVQ+6TcRpbIxxqwXkWeAuSISAqQCI7AXnGnnPHcAex4B7NTG7zsf9NuBu53lg4EPROSfzjZuKca3oVS+6eyjSuWTiJw0xlRwuw6lvE2bhpRSKsjpEYFSSgU5PSJQSqkgp0GglFJBToNAKaWCnAaBUkoFOQ0CpZQKcv8Pd8plRZGSG0gAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEWCAYAAABrDZDcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3deXwV5dn/8c+VjRASAklYZIcEURYBCWvqI4oKbmDViguurWjrho/VatXa+lNrH6tV64pKXaBIC+5iRa2IQlDCKosKhC3sBAhryHb9/rgncAghhCQnk5NzvV+v88o5M3NmrhPCfM/cM3PfoqoYY4wJXxF+F2CMMcZfFgTGGBPmLAiMMSbMWRAYY0yYsyAwxpgwZ0FgjDFhzoLAhBUReV1EHqnksqtF5Kxg13S8RGS6iPzK7zpM/WFBYIwxYc6CwJg6TEQi/a7B1H8WBKbO8Zpk7haRRSKyV0ReE5EWIvKJiOwWkc9FpGnA8sNFZImI7PSaTU4OmNdbROZ575sExJbZ1gUissB77ywROaUS9fUVkc2BO2kRuVhEFnrPI0TkXhFZKSK5IvIvEUkKWPbfIrJJRPJEZIaIdAuY97qIvCgiU0VkL3DGMWqJEJEHRGSNiGwRkTdFJNGbFysi470adorIHBFp4c27TkSyvd/LKhG56lif29RfFgSmrroEOBs4EbgQ+AT4PdAM93d7O4CInAhMBMZ486YCH4pIjIjEAO8BbwFJwL+99eK9tzcwDrgJSAZeBj4QkQYVFaaqc4Bc4JyAyVcDb3rPbwMuAk4HWgE7gOcDlv0E6Aw0B+YBE8ps4krgUSAB+KaiWoDrvMcZQCcgHnjOm3ctkAi09T7fzcB+EWkEPAucq6oJwCBgwTG2Y+oxCwJTV/1dVTer6nrga+BbVZ2vqvnAu0Bvb7mRwMeq+pmqFgJ/BRridm4DgGjgaVUtVNXJwJyAbYwGXlbVb1W1WFXfAA547zuWN4BRAN63/aHAP715NwP3q2qOqh4A/ghcKiJRAKo6TlV3B8zrWfot3vO+qs5U1RLv81bkKuApVc1W1T3AfcDl3rYKcQGQ5n2+uaq6y3tfCdBdRBqq6kZVXVKJz2zqKQsCU1dtDni+v5zX8d7zVsCa0hmqWgKsA1p789br4T0rrgl43h64y2s22SkiO3HfnltVor7xwIXet+vLgK9VdWPAet8NWOcyoBhoISKRIvK412y0C1jtvSclYN3rKrH9Uod9fu95FNACdyT0KfC2iGwQkf8TkWhV3YsL0JuBjSLysYicdBzbNPWMBYEJdRtwO14ARERwO/P1wEagtTetVLuA5+uAR1W1ScAjTlUnHmuj3pFKJnAxrlnorTLrPbfMemO991wJjADOwjXbdCgtPXD1lfngnsM+v/f5ioDN3lHQn1S1K+4I6QLgGq/+T1X1bOAE4AfglePYpqlnLAhMqPsXcL6IDBGRaOAuXPPOLNyOugi4XUSiReRioF/Ae18BbhaR/uI0EpHzRSShktt+E7gH6AG8EzD9JeBREWkPICLNRGSENy/Bqy8XiAMeq8JnDjQRuFNEOopIvLe+SapaJCJniEgP76T2LlxTUYl34n2EdzRzANiDayoyYcqCwIQ0Vf0R11b/d2Ab7sTyhapaoKoFuG/s1wHbcc0h7wS8Nwu4EXdydQewwlu2st7FawZS1X0B058BPgCmichuYDbQ35v3Jq75Zj2w1JtXHeNwRyMzgFVAPu5kNUBLYDIuBJYBX3nLRgD/izua2I47qf3ratZhQpjYwDTGVJ2IrARuUtXP/a7FmKqyIwJjqkhELsG15//X71qMqY4ovwswJhSJyHSgK3C1d6WSMSHLmoaMMSbMWdOQMcaEuZBrGkpJSdEOHTr4XYYxxoSUuXPnblPVZuXNC7kg6NChA1lZWX6XYYwxIUVE1hxtnjUNGWNMmLMgMMaYMGdBYIwxYc6CwBhjwpwFgTHGhDkLAmOMCXMWBMYYE+YsCIwxpq4rOgDfPA3rvgvK6i0IjDGmrlKFpe/Dc33h84fgx6lB2UzI3VlsjDFhYcN8+PR+WDMTmneDq9+D1DOCsikLAmOMqUt2bYT//j9Y8E+IS4YLnoZTr4GIyKBt0oLAGGPqgsL9MOs5+OZvUFIIGbfDaXdBbGLQN21BYIwxflKFxVPgs4dgVw6cPBzOfhiSOtZaCRYExhjjl3Vz4NP7IGcOtDwFLh4LHTJqvQwLAmOMqW15OfD5H+H7f0N8CxjxAvS8AiL8uZDTgsAYU3+VlLj29qgGflfiHNgDM5+BWc+61/9zN2SMgQbxvpZlQWCMqX9KSty37f8+Ars3Qqte0G6g9xgAcUm1X8+it+HzP8GeTdD9Ujjrj9Ckbe3WcRQWBMaY+mXlf92J102LXLt7t4vcHbnfvnTom3izk1wotB/kgqFJu+DVs2YW/Oc+2LgAWqfDyLegbb/gba8KLAiMMfXDxoUuALK/dDv2i1+F7pccancvzIcN89yOee1sd6XO3H+4eY3buEBo7x01NDu5+u31O1bDZ39wdwY3bg0Xv+KOBHw6D1ARCwJjTGjbudY1AS2aBA2bwtDHoO+vjjwvEB3rjgDaD3KvS4phy1JYkwlrM90dvIsnu3mxidA2IBha9a78eYb8XfD1kzD7BYiIgjPuh4G3QkxczX3mGmZBYIwJTfu2ux3ud2NBIuBnd7oTrw2bVO79EZHQsod79B/truffsdodLaz1jhqWf+qWjWwArft4wTAI2vY98kavkmKY/5YLpb1boeeVMORBaNyqRj92MIiq+l3DcUlPT9esrCy/yzDG+KVwP3z7MnzzlPv23esqOOP3kNi65re1d5sXDN5Rw4YFoMUueFp0O3QCOjrOdQuxebF7PfQxaH1qzddTDSIyV1XTy50XrCAQkbbAm0ALQIGxqvpMmWUGA+8Dq7xJ76jqwxWt14LAmDBVUgwL34YvH3N34HYe6q68adG19moo2Otu/lo7251ryJkDhfvcvCbt4Oz/B11HgEjt1VRJFQVBMJuGioC7VHWeiCQAc0XkM1VdWma5r1X1giDWYYwJZaqw4nN3InjLEmh1Kvz8Jeh4Wu3XEtMIOg12D4DiQnd10s51cOIwdx4iBAUtCFR1I7DRe75bRJYBrYGyQVA79m5zbYmn/y6ovfgZY2rQ+nnuypvVX0PTjnDpP6Dbz+vON+7IaHfuoHUfvyupllo5WSwiHYDewLflzB4oIguBDcBvVXVJOe8fDYwGaNeuitf7rvoKvvoLxDaBgb+p2jqMMbVj+yrX5r54iuuK+dz/gz7XQ1SM35XVS0E/WSwi8cBXwKOq+k6ZeY2BElXdIyLnAc+oaueK1lflcwSqMPFyWDUDfj2rVnv2M8ZU0t5cmPEEzHnVXXo58BbIuANiG/tdWcir6BxBUO9sEJFoYAowoWwIAKjqLlXd4z2fCkSLSEqQioHznwKJhA/vcMFgjKkbCvbBjL/Cs73gu5eh15Vw+3x3+aWFQNAFLQhERIDXgGWq+tRRlmnpLYeI9PPqyQ1WTSS2hnMeds1E88cHbTPGmEoqLoJ5b8LfT3VNQR1+Br/OhOHPQuMT/K4ubATzHEEGcDXwvYgs8Kb9HmgHoKovAZcCvxaRImA/cLkGu63q1Ovg+8luLNC0s+yPzZjalp8HuzbA5iWuGWjrD9CmL1w67tBdv6ZWhecNZdtWwEsZLghGjq87VyAYE+ryd8Gu9e6Rt97t8HfluJ+lrwt2H1o+KRXOesiNymX/D4PKr/sI6q6UNBh8H3z+kOsQqttFfldkQt3Wn9zlyXFJ0Lyru+s0qVP9ulQ5f1c5O/bSh/c6cCcPgLiBVxq3gmYnQuoZrgO2xq0gsY277DIy2pePYw4JzyAA1wnUkndg6m+h4//Ufv/kpn7Ytx2mP37oKpeSQtASNy8qFpp1gebd3N2vpQER36LuffstKXb99u9YAzvXuI7c8rwdfumO/sCuMm8SiG/uduzJae4mq8atvB19a3dOLr6lXfIZAsI3CCKjYMTzMHawO1/w8xf9rsiEkuJCmPMaTP+z20H2ud71dxPTyLV5b17qerbcvARWfgEL/3novQ2TXCC06HYoHJqdFNxRqlRdaO1cfWhnv2ON62Rt5xp3Z2xJYcAbjrGTb9wKEk6wnXw9Eb5BAK7XwYwx8PVfoccl7pyBMRVRheXT3JeH3OXQ6QzXwVhgfzetertHoL25rnuEzUtdx2RblrqrZUr7qQFo2uHIo4ekVPelpTIK9rpv8jtWH76zL/1ZttmmYZLb5gk9XRt90/bQpL2bltjWdvJhJDxPFgcqzIeXT3M9Gv5mtu9jh5o6bMsy+PT3bgSs5DQXAJ3PqXozT0mJ+4YeePSwZSnkrjjUvBTZwLWtHwyIbi4YAr/Nl+7s9249fP3Rcd6OPWAHf/B5e2iQUI1fhgk1vvQ+GixB6X107WwYNwz63wTn/qVm121C395cmP4YZI1zO8/B90H6L4P3jbkwH7b96IKhNBw2L3Vj3QaSSHfC9bAdfIdDPxul1L1zEcY3dtXQsbQbAP1Guz7Ou10M7fr7XZGpC4oK3JVAX/0fFOxxo14Nvi/4FxZEx7rmmhN6Hj5933YXDKjb2TduXflmI2MqYH9FpYb8AX6cCh/cCjd9HbLdyZoaoAo/fgLT7oft2ZB2NpzzCDQ/yd+64pL86XrZ1Ht1bxRlvzSIhwufhm0/uZPHJjxtWgxvjoC3r4CIaLhqMoya7H8IGBNEdkQQKO0sN87oN39zowy17OF3Raa27NkKXz7iruSJTYRzn4D06+1mJxMW7IigrKGPQsOm8P6trkMsU78VHYCZz8CzvV1HhP1vdr1e9h9tIWDChgVBWXFJcN4TsHEBzH7e72pMsKjC0g/g+X5uBKwOGe7y4WF/dl8EjAkj1jRUnq4XwUkXuEGyT7oAklP9rsjUpI0L4T+/hzXfQLOT4ep3IfVMv6syxjd2RFAeETjvr+5mng9udzf+mNC3ezO8fwu8fDpsXeYGKrr5GwsBE/YsCI6m8Qkw9BH3rXHe635XY6qjMB++ftINfrJwkhv+8LZ50PeXdh2+MVjTUMV6Xw3f/xum/QE6D3W9KZrQkp8H/xwJazNdM9/ZD1tTnzFl2BFBRUTgwmehpAg+utPGOQ41e7fB6xdAThZc8hpcPsFCwJhyWBAcS1JHN4D28k9h8RS/qzGVlbce/nGuu0HwionQ41K/KzKmzrIgqIz+N7uRlD65x3VAZuq23JWuE8FdG2HUO9D5bL8rMqZOsyCojIhIGP6cG6rvP/f6XY2pyOYl7kigYA9c96G7P8AYUyELgspq0RX+57fw/b/gp0/9rsaUJycL/nEeSARc/8mRg8MYY8plQXA8fva/buSoj+50Rwem7lg1A94YDg2bwA3/sU7ijDkOFgTHIyoGhv/dDeT9+R/9rsaU+mEqjL8UmrSDGz51g7IYYyrNguB4tUmHAb+BrNdg9Uy/qzGL/g2TRrnxfa+fCgkt/a7ImJBjQVAVZ97vRoj64DY31rHxx5xX4Z0bof0guPaD4I8cZkw9ZUFQFTGNYPizsH0lTH/c72rC09dPwcd3wYlD4ap/20DsxlSDBUFVdRrsuqCY9XfYMN/vasKHqjs/88WfoPulMHI8RDf0uypjQlrQgkBE2orIlyKyVESWiMgd5SwjIvKsiKwQkUUicmqw6gmKcx6BRs3g/duguNDvauq/khJ3FPDN3yD9Brh4rA0eY0wNCOYRQRFwl6p2BQYAt4hI1zLLnAt09h6jgReDWE/Na9gEzn8SNn/vRrkywVNcCO/e5E7SZ9zhupCOiPS7KmPqhaAFgapuVNV53vPdwDKgbPedI4A31ZkNNBGRE4JVU1CcfIEbyOarv8DWn/yupn4qzId/XeNu5hvykOtBVMTvqoypN2rlHIGIdAB6A9+WmdUaWBfwOocjw6LuO+8JiI5zVxHZIDY168Bu+Ocv4MepbrCg0/7X74qMqXeCHgQiEg9MAcaoapVuxxWR0SKSJSJZW7durdkCa0J8cxj2OKyb7ZouTM3Ytx3evMjdr/HzsdDvRr8rMqZeCmoQiEg0LgQmqOo75SyyHmgb8LqNN+0wqjpWVdNVNb1Zs2bBKba6el4OqUPcFS071/pdTejbvQlePx82LYKRb0HPkX5XZEy9FcyrhgR4DVimqk8dZbEPgGu8q4cGAHmqujFYNQWVCFz4tLu88f1brYmoOnascd1I71jj7hE46Xy/KzKmXgvmEUEGcDVwpogs8B7nicjNInKzt8xUIBtYAbwC/CaI9QRfk3Yw7DFY9RV897Lf1YSmrT+5ENi/Ha55392vYYwJqqCNWayq3wAVXtqhqgrcEqwafHHqtfDjJ/DZQ9DpDOsF83hsWADjLwaJhOumQsvufldkTFiwO4trmojrobRBvOsHp6jA74pCw5pMeONCd/XVDf+xEDCmFlkQBEN8czfo/aZF8JX1RXRMyz+Ht34O8S1cCNgA88bUqrAJgsXr87jvnUXsLyiunQ2efAH0HuW6Q1g7u3a2GYqWvAcTL4eUNDeqWGIbvysyJuyETRBs23OAid+tY87q7bW30WGPQ2Jb1zXCgd21t91Q8cPHMPkGaN0Hrv0I4uvopcHG1HNhEwT9OiYRHSnMXLmt9jbaIAF+/rK7DPLT+2tvu6Egezr8+zpo1QtGTXb9NhljfBE2QRAXE0Xvtk3JXJlbuxtuPxB+NgbmveGuJjKw7juYeCUkd4arJttYAsb4LGyCAGBQWjLfr88jb18tdxk9+PfQoofri2hPHewiozZtXAQTLoWEFnD1uzaqmDF1QFgFQUZaCqqQmV3LRwVRMa7v/Pxd8OEd7u7jcLRtubs6KCbB3SyW0MLviowxhFkQ9GzThLiYSGbV5nmCUi26wlkPwY8fw/zxtb99v+1cC2+OcPdZXPO+uwvbGFMnhFUQxERF0K9jEjNX+BAEAP1/DR1Og//cC9tX+VODH3ZvgjeGQ8Ee1xyUkuZ3RcaYAGEVBACDUpNZuXUvm/Lya3/jERFw0YuuC4V3b4aSWrqnwU/7trvmoD1b4Kop0LKH3xUZY8oIwyBIASAz26ejgiZt3UA262bX/+EtD+yG8ZdA7kq4YiK07et3RcaYcoRdEHQ9oTFN46KZuaKWTxgHOuUyN7zll4/BxoX+1RFMhfvhn5e7z3fZG9DpdL8rMsYcRdgFQUSEMDA1mVkrtqF+Xb0jAhf8DeKS4Z3Rbkze+qSowI0xvGamu1qqy7l+V2SMqUDYBQG45qENefmszt3nXxFxSXDR87D1B/jiYf/qqGklxfDuaFg+zYVdj0v9rsgYcwxhGQQZae48gW9XD5VKOwv63gizn4fsr/ytpSaUlMCHt8OSd+GcRyD9er8rMsZUQlgGQYfkOE5IjPXnfoKyzn4YktPgvd/A/p1+V1N1qjDtfnePxOm/g0G3+V2RMaaSwjIIRIRBqSlkrsylpMTnu3xj4lw7+u6N8Mk9/tZSHdMfh9kvuHslBt/ndzXGmOMQlkEAkJGWzI59hSzbtMvvUlw3zKf/DhZNgsXv+F3N8Zv1dzcAT+9RMPQxdzLcGBMywjgI3HmCWX5eRhrotLtcIHx0J+za4Hc1lTf3dZj2gLsc9sJn3U1zxpiQErb/a1s0jiW1WaPaHZ+gIpFR8POxUFwA798SGh3TfT8ZPhwDaWfDxa9ARKTfFRljqiBsgwDcUcF3q7ZTUFTidylOSpq72mblf2HOq35XU7EfP3Ejr7XPgJFvuR5WjTEhKayDYFBqCvsKilmYU4eu1km/wX3DnvYgbP3J72rKl/0V/OtaaHmK6zoiuqHfFRljqiGsg2BApyRE6sD9BIFEYMRzbuf67mgoruVBdI5l3RyYeAUkp8KoKRDb2O+KjDHVFOV3AX5qEhdD91aJzFqRy5iz/K4mQEJLuPAZ+NfVMOMJOOP3flfkbFoMEy6B+OY2upg5boWFheTk5JCfX8+6VKljYmNjadOmDdHR0ZV+T1gHAbjhK8d9s4p9BUXExdShX0fX4dDzSpjxV9dU5HfPndtWwFsXQUy8N7pYS3/rMSEnJyeHhIQEOnTogNglxkGhquTm5pKTk0PHjh0r/b6wbhoCyEhNobBYmbN6h9+lHOncx6Fxa9dEVLDXvzp2rnOji6m6EGja3r9aTMjKz88nOTnZQiCIRITk5OTjPuoK+yDo2yGJmMgIZtWl8wSlYhPh5y+60cymPehPDXu2uBA4sNsbXayzP3WYesFCIPiq8jsOWhCIyDgR2SIii48yf7CI5InIAu/xh2DVUpGGMZH0btek7txPUFaHn7l+e7Jeg+Wf1e62922HNy9yQ02OmgwnnFK72zemBu3cuZMXXnihSu8977zz2LmzDl1dWMOCeUTwOjDsGMt8raq9vIdvfTEPSk1hyYZd7NxX4FcJFTvzAWjezd1otjeId0Krwp6tsCbTdR43/hLIXQ6XT4C2/YK3XWNqQUVBUFRUVOF7p06dSpMmTYJR1nE5Vp1VFbQgUNUZwPZgrb8mZaQlowqZK+tIdxNlRTVwHdPt3wEf3VH9u47374CcLFg4yY2SNvmX8PLp8Hg7+Gsa/GOYC52tP8IvXofUM2rkYxjjp3vvvZeVK1fSq1cv7r77bqZPn85pp53G8OHD6dq1KwAXXXQRffr0oVu3bowdO/bgezt06MC2bdtYvXo1J598MjfeeCPdunXjnHPOYf/+/Udsa+XKlQwYMIAePXrwwAMPEB8ff3DeE088Qd++fTnllFN46KGHACpc7+DBgxkzZgzp6ek888wzzJ07l9NPP50+ffowdOhQNm7cWO3fjd+XyQwUkYXABuC3qrqkvIVEZDQwGqBdu3Y1XkTPtk1oFBPJzJXbOLfHCTW+/hrRsrs7MvjsD7Dwbeh1RcXLH9jtxgrevhJys72fK9y0/YH5LG4c5eQ0aNPX3R+QlOp+NmkHkZW/BM2YyvrTh0tYuqFmO3zs2qoxD13Y7ajzH3/8cRYvXsyCBQsAmD59OvPmzWPx4sUHr7AZN24cSUlJ7N+/n759+3LJJZeQnJx82HqWL1/OxIkTeeWVV7jsssuYMmUKo0aNOmyZO+64gzvuuIMrrriCl1566eD0adOmsXz5cr777jtUleHDhzNjxgzatWtX4XoLCgrIysqisLCQ008/nffff59mzZoxadIk7r//fsaNG1et352fQTAPaK+qe0TkPOA9oNwzkao6FhgLkJ6eXuOd8ERHRtCvYxKz6uoRQamBt8JPn8LUu6H9IHc9//bsgB3+ikM7/T2bD39v49aQ1MldlpqU6nb8yanQtIM74jAmDPXr1++wyyyfffZZ3n33XQDWrVvH8uXLjwiCjh070qtXLwD69OnD6tWrj1hvZmYm7733HgBXXnklv/3tbwEXBNOmTaN3794A7Nmzh+XLl9OuXbsK1zty5EgAfvzxRxYvXszZZ58NQHFxMSecUP0vr74FgaruCng+VUReEJEUVfXlrG1GWgpffryMTXn5tEyM9aOEY4uIhItehBcz4IWBUFjmktJGzd3OPe1sSO7kdvZJqS4AYuL8qdmYclT0zb02NWrU6ODz6dOn8/nnn5OZmUlcXByDBw8u9zLMBg0OfXGKjIwst2noaFSV++67j5tuuumw6atXr65wvaV1qirdunUjMzOz0tusjEoFgYjcAfwD2A28CvQG7lXVaVXdsIi0BDarqopIP9z5Ct++kg9KPTR85SV92vhVxrE1bQ+XvApL33M7+KROh5pzrLsHY44qISGB3bt3H3V+Xl4eTZs2JS4ujh9++IHZs2dXeVsDBgxgypQpjBw5krfffvvg9KFDh/Lggw9y1VVXER8fz/r164/rDuAuXbqwdetWMjMzGThwIIWFhfz0009061a9YK3sEcENqvqMiAwFmgJXA28BRw0CEZkIDAZSRCQHeAiIBlDVl4BLgV+LSBGwH7hc1b++l09qmUBSoxhmrqzjQQDQZZh7GGMqLTk5mYyMDLp37865557L+eeff9j8YcOG8dJLL3HyySfTpUsXBgwYUOVtPf3004waNYpHH32UYcOGkZiYCMA555zDsmXLGDhwIADx8fGMHz+eyMjKdeEeExPD5MmTuf3228nLy6OoqIgxY8ZUOwikMvteEVmkqqeIyDPAdFV9V0Tmq2rvam29CtLT0zUrKyso675lwjzmrtlB5n1n2o0vxtSwZcuWcfLJJ/tdRq3Yt28fDRs2RER4++23mThxIu+//36tbb+837WIzFXV9PKWr+wRwVwRmQZ0BO4TkQSgjnTiX3MGpSXz8fcbyd62l9Rm8cd+gzHGlGPu3LnceuutqCpNmjSp9lU9wVbZIPgl0AvIVtV9IpIEXB+8svyRkVo6fOU2CwJjTJWddtppLFy40O8yKq2yN5QNBH5U1Z0iMgp4AMgLXln+aJ8cR+smDev+ZaTGGFODKhsELwL7RKQncBewEngzaFX5REQYlJpMZnYuJSUhMGawMcbUgMoGQZF3Rc8I4DlVfR5ICF5Z/slIS2HnvkKWbqzZux6NMaauqmwQ7BaR+3CXjX4sIhF4l4LWNwNT3V2EdWr4SmOMCaLKBsFI4ADufoJNQBvgiaBV5aMWjWNJax7PTDtPYEy9Up1uqMHdG7Bv374arKjuqFQQeDv/CUCiiFwA5KtqvTtHUCojNZk5q7ZTUFTvrpA1JmyFYhAUFxfXynYqFQQichnwHfAL4DLgWxG5NJiF+WlQWgr7C4uZv7YODl9pjKmSst1QQ/ldQu/du5fzzz+fnj170r17dyZNmsSzzz7Lhg0bOOOMMzjjjCO7ZZ86dSonnXQSffr04fbbb+eCCy44uK4bbriBfv360bt374M3lb3++utcfPHFDBs2jM6dO3PPPfccXFd8fDx33XUXPXv2JDMzk/Hjx9OvXz969erFTTfdFJRwqOx9BPcDfVV1C4CINAM+BybXeEV1wIBOyUQIzFqZS/9Oycd+gzHm+HxyL2z6vmbX2bKHG+f7KMp2Q320LqG3bt1Kq1at+PjjjwHXB1FiYiJPPfUUX375JSkpKYetNz8/n5tuuokZM2bQsWNHrrjiUBfxjz76KGeeeSbjxo1j586d9OvXj7POOguABQsWMH/+fBo0aECXLiS7hSkAABURSURBVF247bbbaNu2LXv37qV///48+eSTLFu2jL/85S/MnDmT6OhofvOb3zBhwgSuueaaGv3VVfYcQURpCHhyj+O9ISexYTQ9Wicyq64OX2mMqbbALqFPPfVUfvjhB5YvX06PHj347LPP+N3vfsfXX399sJ+go/nhhx/o1KnTwe6sA4Ng2rRpPP744/Tq1etgb6Zr164FYMiQISQmJhIbG0vXrl1Zs2YN4HoeveSSSwD44osvmDt3Ln379qVXr1588cUXZGdn1/jvorJHBP8RkU+Bid7rkcDUGq+mDhmUlsIrM7LZe6CIRg38Hr/HmHqmgm/uteVoXUIDzJs3j6lTp/LAAw8wZMgQ/vCHqg2prqpMmTKFLl26HDb922+/PaLb6dJhKGNjYw92QqeqXHvttfz5z3+u0vYrq7Ini+/GDQxzivcYq6q/C2ZhfhuUmkxRifLd6pAYbdMYcwxlu6EeOnQo48aNY8+ePQCsX7+eLVu2sGHDBuLi4hg1ahR333038+bNK/f9pbp06UJ2dvbBgWQmTZp02Db+/ve/U9q55/z584+r5iFDhjB58mS2bHENMtu3bz945FCTKv1VV1WnAFNqvII6Kr19EjGREcxasY0zujT3uxxjTDWV7Yb6iSeeKLdL6BUrVnD33XcTERFBdHQ0L774IgCjR49m2LBhtGrVii+//PLgehs2bMgLL7zAsGHDaNSoEX379j0478EHH2TMmDGccsoplJSU0LFjRz766KNK19y1a1ceeeQRzjnnHEpKSoiOjub555+nffv2NfRbcSrshlpEdgPlLSCAqmqtj4QSzG6oy7p8bCa79hcx9Y7TamV7xtRn9bkb6j179hAfH4+qcsstt9C5c2fuvPNO3+o53m6oK2waUtUEVW1cziPBjxCobRmpKSzduIsdewv8LsUYU4e98sor9OrVi27dupGXl1fueYe6rN5e+VMTBqW5y8Qys+0uY2PM0d15550sWLCApUuXMmHCBOLiQmuMcAuCCvRsk0h8gyjrd8gYU69ZEFQgKjKCfh2TbHwCY2qIj8OSh42q/I4tCI5hUGoyq7btZcPO/X6XYkxIi42NJTc318IgiFSV3NxcYmNjj+t9dqfUMWR45wlmrtjGL9Lb+lyNMaGrTZs25OTksHXrVr9LqddiY2Np06bNcb3HguAYurRIILlRDLNW5loQGFMN0dHRB7thMHWLNQ0dQ0SEMDA1mVkrt9khrTGmXrIgqISMtBQ27zrAyq17/S7FGGNqnAVBJWSkuvME1hupMaY+siCohLZJDWndpKHdT2CMqZcsCCpBRMhISyZzZS7FJXaewBhTvwQtCERknIhsEZHFR5kvIvKsiKwQkUUicmqwaqkJGWkp7MovYsmGPL9LMcaYGhXMI4LXgWEVzD8X6Ow9RgMvBrGWahuY6oasnLnC7jI2xtQvQQsCVZ0BVDSqywjgTXVmA01E5IRg1VNdzRNiObFFvJ0wNsbUO36eI2gNrAt4neNNO4KIjBaRLBHJ8vOuxEGpKcxZvZ0DRcW+1WCMMTUtJE4Wq+pYVU1X1fRmzZr5Vseg1GTyC0uYv3anbzUYY0xN8zMI1gOBfTa08abVWf07JRMhMMsuIzXG1CN+BsEHwDXe1UMDgDxV3ehjPceU2DCaHm2aMNO6pTbG1CNB63RORCYCg4EUEckBHgKiAVT1JWAqcB6wAtgHXB+sWmpSRmoyY2dks+dAEfENrM8+Y0zoC9qeTFWvOMZ8BW4J1vaDJSMthRemr+S7VbmceVILv8sxxphqC4mTxXVJn/ZNiYmKYJbdT2CMqScsCI5TbHQk6e2b2nkCY0y9YUFQBYNSk1m2cRe5ew74XYoxxlSbBUEVDPKGr8zMtqMCY0zosyCoglNaJ5LQIMr6HTLG1AsWBFUQFRlB/05J1u+QMaZesCCookGpKazJ3UfOjn1+l2KMMdViQVBFGWmlw1da85AxJrRZEFTRiS3iSYmPsX6HjDEhz4KgikSEgakpzFyZi7tJ2hhjQpMFQTVkpCazdfcBVmzZ43cpxhhTZRYE1VB6nmCmNQ8ZY0KYBUE1tE2Ko21SQ+tuwhgT0iwIqikjNYXZ2bkUl9h5AmNMaLIgqKZBaSnszi9i8fo8v0sxxpgqsSCopoGdkgGYaXcZG2NClAVBNTVLaECXFgk2PoExJmRZENSAQWnJzFm9nfzCYr9LMcaY42ZBUAMyUlM4UFTCvLU7/C7FGGOOmwVBDejfKYnICLHmIWNMSLIgqAEJsdGc0ibRuqU2xoQkC4IaMig1mYU5eezOL/S7FGOMOS4WBDUkIzWF4hLlu1Xb/S7FGGOOiwVBDTm1fVMaREXY8JXGmJBjQVBDYqMjSe/Q1M4TGGNCjgVBDRqUmsIPm3azbc8Bv0sxxphKsyCoQTZ8pTEmFAU1CERkmIj8KCIrROTecuZfJyJbRWSB9/hVMOsJth6tE2mW0ICnP/+JvP129ZAxJjQELQhEJBJ4HjgX6ApcISJdy1l0kqr28h6vBque2hAZITx3RW/W5u7jtonzKSou8bskY4w5pmAeEfQDVqhqtqoWAG8DI4K4vTqhf6dkHrmoOzN+2spjU3/wuxxjjDmmYAZBa2BdwOscb1pZl4jIIhGZLCJty1uRiIwWkSwRydq6dWswaq1Rl/drx/UZHRg3cxWT5qz1uxxjjKmQ3yeLPwQ6qOopwGfAG+UtpKpjVTVdVdObNWtWqwVW1f3nncxpnVN44L3FdpOZMaZOC2YQrAcCv+G38aYdpKq5qlp6reWrQJ8g1lOroiIjeO6KU2nbNI6bx89l3fZ9fpdkjDHlCmYQzAE6i0hHEYkBLgc+CFxARE4IeDkcWBbEempdYlw0r16bTlFxCTe+mcWeA0V+l2SMMUcIWhCoahFwK/Apbgf/L1VdIiIPi8hwb7HbRWSJiCwEbgeuC1Y9funULJ7nrzqV5Vv2cOekBZTYIPfGmDpGVENrx5Senq5ZWVl+l3Hc/jFzFX/6cCm3nJHK3UNP8rscY0yYEZG5qppe3ryo2i4mXF03qAM/bd7N81+u5MQWCYzoVd4FVMYYU/v8vmoobIgIfxrenX4dk7h78iIWrNvpd0nGGANYENSqmKgIXhrVh+YJDRj9Zhab8vL9LskYYywIaltSoxheu7Yvew8UMfqtLPYXFPtdkjEmzFkQ+KBLywSevrw336/P4+7JCwm1E/bGmPrFgsAnZ3dtwT1DT+KjRRt57r8r/C7HGBPG7KohH918eid+2rybJz/7ic4tEhjWvaXfJRljwpAdEfhIRPjzxT3o1bYJd05awJINeX6XZIwJQxYEPouNjmTs1X1IbBjNjW9ksXW3DXNpjKldFgR1QPPGsbxyTTrb9xVw8/i5HCiyK4mMMbXHgqCO6NEmkSd/0Yu5a3bwwLuL7UoiY0ytsSCoQ84/5QRuH9KZf8/N4bVvVvldjjEmTFgQ1DFjhnTm3O4teWzqMr78YYvf5RhjwoAFQR0TESE8eVlPTmrZmNsnzmfFlt1+l2SMqecsCOqguJgoXrk2nQbRkfzyjSx27C3wuyRjTD1mQVBHtW7SkJev7sPGnfnc8s95FBaX+F2SMaaesiCow/q0b8pjF/dg1spcHv5wqd/lGGPqKetioo67tE8blm/ezcszsjmxZQJXD2jvd0nGmHrGjghCwD3DTuLMk5rzxw+WMGvFNr/LMcbUMxYEISAyQnjm8l50SmnEryfMY/W2vX6XZIypRywIQkRCbDSvXpuOCPzqzSx25Rf6XZIxpp6wIAgh7ZMb8cJVp7J6216ufGU2r8zI5vucPIpLrDsKY0zV2cniEDMoNYUnL+vJM58v59GpywBIiI2if8ckBnRKZmBqMie3bExEhPhcqTEmVFgQhKARvVozoldrNu/KZ3Z2LrOzc8lcmcvny1yXFIkNow8Lhi4tEiwYjDFHJaHWy2V6erpmZWX5XUadtDFv/8FQmJ29nbXb9wHQNC6a/h1dKAzolMyJLeIRsWAwJpyIyFxVTS93ngVB/ZWzYx+zs7cfDIf1O/cDkNwohv6dkhjoHTGkNrNgMKa+syAwAKzbvo9Mrylp9spcNuTlA5AS34ABnQ41JXVKaWTBYEw9U1EQBPUcgYgMA54BIoFXVfXxMvMbAG8CfYBcYKSqrg5mTeGsbVIcbZPiuCy9LarK2u37Dh4tZGbn8tGijQA0T2jAgE7J9O2YRNO4aBpGRxIbHUlsdIT30z0alk6LirRzEMaEsKAFgYhEAs8DZwM5wBwR+UBVAzvN+SWwQ1XTRORy4C/AyGDVZA4REdonN6J9ciNG9m2HqrI69/Bg+GDhhkqvLyYqgtioCBrGeEERFUlsTCSxURGHh0ZAkJS+jhShRJUSxf0sOfRcVSnWwNdQUuKmaenyqhSXgHrPy1tPaY0xkREHf0aXeR0T5R7RB6eJ9zOS6NLnR13WPQAUDo4w556D4uqlzOvylqWc+QfneSLE/RtGCAiCRIAAESKIuJ94rw9b1o70TDmCeUTQD1ihqtkAIvI2MAIIDIIRwB+955OB50RENNTaq+oBEaFjSiM6pjTiin4uGDbk5bP3QBH7C4rJLywmv6iE/QXFHChyr/cXuGn5hcXsLyzmQOGh5/mFxeQXlrC/sJid+wrY5D3PD5hXUIkeVUt3apEBO7gIceM2HHwugogQGVE6/9CykRHuOQoFxSUUFLntFpb+LA6/PzWRw0NDRI4IEQHwlpPS5Ti0bOl6IHCeF0oByx62zXLmVxhLR5l5tPccT8iVt4sp9y+hnInlLRcY5u516bKHfwE47veUmXftwPbcNqRzeZVWSzCDoDWwLuB1DtD/aMuoapGI5AHJwGEd6ojIaGA0QLt27YJVrwkgIrRu0jCo2yguUfILi1EI2KFDZMDOPNjfYEtKlMISLyCKXDC4sCimoEgPhkeh9/NAwHMXJIdPh8N3dsBhn+PIneGh1xxcXgLmHb4jRgS8I4SSEu+nekcN3pGEe33oSKj0yKh0Hnr4a9VDRx2H1hm4Azt0RKLerunQ0QpQevRyxJGM9353iHPE0U1FEXy074JHfU8FK1OUciOncpPK/Rssf7nD5x38Nw+YWFrHkcseOf/QZg9N69wivpwtV19I3EegqmOBseBOFvtcjqkhkRFCowb+/glGRAgNIiJpEBXpax3G+CmYXUysB9oGvG7jTSt3GRGJAhJxJ42NMcbUkmAGwRygs4h0FJEY4HLggzLLfABc6z2/FPivnR8wxpjaFbTjcq/N/1bgU9zlo+NUdYmIPAxkqeoHwGvAWyKyAtiOCwtjjDG1KKgNtKo6FZhaZtofAp7nA78IZg3GGGMqZt1QG2NMmLMgMMaYMGdBYIwxYc6CwBhjwlzI9T4qIluBNVV8ewpl7loOYfZZ6qb68lnqy+cA+yyl2qtqs/JmhFwQVIeIZB2tG9ZQY5+lbqovn6W+fA6wz1IZ1jRkjDFhzoLAGGPCXLgFwVi/C6hB9lnqpvryWerL5wD7LMcUVucIjDHGHCncjgiMMcaUYUFgjDFhLmyCQESGiciPIrJCRO71u56qEpG2IvKliCwVkSUicoffNVWHiESKyHwR+cjvWqpDRJqIyGQR+UFElonIQL9rqioRudP721osIhNFJNbvmipLRMaJyBYRWRwwLUlEPhOR5d7Ppn7WWFlH+SxPeH9ji0TkXRFpUhPbCosgEJFI4HngXKArcIWIdPW3qiorAu5S1a7AAOCWEP4sAHcAy/wuogY8A/xHVU8CehKin0lEWgO3A+mq2h3XhXwodQ//OjCszLR7gS9UtTPwhfc6FLzOkZ/lM6C7qp4C/ATcVxMbCosgAPoBK1Q1W1ULgLeBET7XVCWqulFV53nPd+N2OK39rapqRKQNcD7wqt+1VIeIJAL/gxtfA1UtUNWd/lZVLVFAQ2/UwDhgg8/1VJqqzsCNbRJoBPCG9/wN4KJaLaqKyvssqjpNVYu8l7NxIz9WW7gEQWtgXcDrHEJ05xlIRDoAvYFv/a2kyp4G7gFK/C6kmjoCW4F/eM1cr4pII7+LqgpVXQ/8FVgLbATyVHWav1VVWwtV3eg93wS08LOYGnQD8ElNrChcgqDeEZF4YAowRlV3+V3P8RKRC4AtqjrX71pqQBRwKvCiqvYG9hI6zQ+H8drPR+DCrRXQSERG+VtVzfGGwg35a+ZF5H5cM/GEmlhfuATBeqBtwOs23rSQJCLRuBCYoKrv+F1PFWUAw0VkNa6p7kwRGe9vSVWWA+SoaumR2WRcMISis4BVqrpVVQuBd4BBPtdUXZtF5AQA7+cWn+upFhG5DrgAuKqmxngPlyCYA3QWkY4iEoM7+fWBzzVViYgIri16mao+5Xc9VaWq96lqG1XtgPv3+K+qhuQ3T1XdBKwTkS7epCHAUh9Lqo61wAARifP+1oYQoie+A3wAXOs9vxZ438daqkVEhuGaU4er6r6aWm9YBIF3cuVW4FPcH/W/VHWJv1VVWQZwNe4b9ALvcZ7fRRluAyaIyCKgF/CYz/VUiXdUMxmYB3yP20eETBcNIjIRyAS6iEiOiPwSeBw4W0SW4454Hvezxso6ymd5DkgAPvP+779UI9uyLiaMMSa8hcURgTHGmKOzIDDGmDBnQWCMMWHOgsAYY8KcBYExxoQ5CwJjapGIDA71nlZN/WNBYIwxYc6CwJhyiMgoEfnOu2nnZW/chD0i8jevr/4vRKSZt2wvEZkd0Ed8U296moh8LiILRWSeiKR6q48PGLtggncHrzG+sSAwpgwRORkYCWSoai+gGLgKaARkqWo34CvgIe8tbwK/8/qI/z5g+gTgeVXtieuvp7QHzN7AGNzYGJ1wd4sb45sovwswpg4aAvQB5nhf1hviOiorASZ5y4wH3vHGImiiql95098A/i0iCUBrVX0XQFXzAbz1faeqOd7rBUAH4JvgfyxjymdBYMyRBHhDVQ8b/UlEHiyzXFX7ZzkQ8LwY+39ofGZNQ8Yc6QvgUhFpDgfHvG2P+/9yqbfMlcA3qpoH7BCR07zpVwNfeaPH5YjIRd46GohIXK1+CmMqyb6JGFOGqi4VkQeAaSISARQCt+AGnOnnzduCO48Armvjl7wdfTZwvTf9auBlEXnYW8cvavFjGFNp1vuoMZUkIntUNd7vOoypadY0ZIwxYc6OCIwxJszZEYExxoQ5CwJjjAlzFgTGGBPmLAiMMSbMWRAYY0yY+/89QEbc2zLxygAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ],
      "source": [
        "def plot_metric(history, output, metric='loss'):\n",
        "    \"\"\" Plot training and test values for a metric. \"\"\"\n",
        "    val_metric = 'val_{}_output_loss'.format(output)\n",
        "    train_metric = '{}_output_loss'.format(output)\n",
        "    plt.plot(history.history[train_metric])\n",
        "    plt.plot(history.history[val_metric])\n",
        "    plt.title('model {} '.format(output) + metric)\n",
        "    plt.ylabel(metric)\n",
        "    plt.xlabel('epoch')\n",
        "    plt.legend(['train genre', 'test genre', 'train year', 'test year'])\n",
        "    plt.show();\n",
        "\n",
        "plot_metric(history, 'genre')\n",
        "plot_metric(history, 'year')"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "We can see that the Model overfits after the 3rd epoch for both prediction tasks.<br><br> What is the accuracy found with our model?"
      ],
      "metadata": {
        "id": "5pD3B4huF5j1"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "FZOiQPV3DvK3",
        "outputId": "08718e87-5cb2-42bd-b22a-a46778e99d1c",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "test accuracy genre: 0.61\n",
            "test accuracy year: 0.47\n"
          ]
        }
      ],
      "source": [
        "y_predict_genre = model.predict(input_test)[1].argmax(axis=1)\n",
        "genre_test_accuracy = sum(y_predict_genre == y_test[0:,0]) / len(y_test[0:,0])\n",
        "print('test accuracy genre: {:0.2f}'.format(genre_test_accuracy))\n",
        "y_predict_year = model.predict(input_test)[0].argmax(axis=1)\n",
        "year_test_accuracy = sum(y_predict_year == y_test[0:,1]) / len(y_test[0:,1])\n",
        "print('test accuracy year: {:0.2f}'.format(year_test_accuracy))"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Conclusion"
      ],
      "metadata": {
        "id": "gMkDPlT4LkHp"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "The baseline found when predicting the most common decade and genre gave us an accuracy of 43% and 17% respectively. The optimal neural network that gave us the best results for the task of predicting the top 5 most genres of the given movie, along with the decade resulted in 4 different inputs each with its own word embedding. The output of the genre needed to be encoded from 0 to n. The architecture we found best was an embedding layer followed by a batch normalization layer, and finally a bidirectional LSTM layer. Also, we implemented early stopping and were able to avoid the model from overfitting dramatically. We obtained a very well-performing model that outperformed the baseline, for the decade we got 30% better accuracy and for the genre, the task performed 18% greater. It must be noted that when running the notebook on an Nvidia RTX 3080 GPU the accuracy achieved was better than when run on the NVIDIA Tesla K80 GPU that Colab provides. To improve this model in the future there would need to be more movie data and probably more concise labels for the genre on some movies. For example, there was a lot of data lost when removing all genres with a low amount of movies within."
      ],
      "metadata": {
        "id": "xN2D8CI6LmBr"
      }
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [],
      "name": "wiki_movie_year&genre_predictor.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3 (ipykernel)",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.9.7"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}