Neural-Network-Based.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Importing libraries"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 41,
      "metadata": {
        "id": "cR2HOAU4LLvf"
      },
      "outputs": [],
      "source": [
        "from sklearn.preprocessing import LabelEncoder, MultiLabelBinarizer, OneHotEncoder, StandardScaler\n",
        "import numpy as np\n",
        "import tensorflow as tf\n",
        "from sklearn.model_selection import train_test_split\n",
        "import pandas as pd\n",
        "import json\n",
        "import matplotlib.pyplot as plt\n",
        "import pandas as pd"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Loading data"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 42,
      "metadata": {
        "id": "2_4cAoIALMqM"
      },
      "outputs": [],
      "source": [
        "with open('steam_games.json', 'r') as f:\n",
        "    data = [json.loads(line) for line in f]\n",
        "df = pd.DataFrame(data)\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 43,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 625
        },
        "id": "-gO33luXLOX0",
        "outputId": "e4db8d66-7f62-493d-a723-a4bc8d43cd4b"
      },
      "outputs": [
        {
          "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>publisher</th>\n",
              "      <th>genres</th>\n",
              "      <th>app_name</th>\n",
              "      <th>title</th>\n",
              "      <th>url</th>\n",
              "      <th>release_date</th>\n",
              "      <th>tags</th>\n",
              "      <th>discount_price</th>\n",
              "      <th>reviews_url</th>\n",
              "      <th>specs</th>\n",
              "      <th>price</th>\n",
              "      <th>early_access</th>\n",
              "      <th>id</th>\n",
              "      <th>developer</th>\n",
              "      <th>sentiment</th>\n",
              "      <th>metascore</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>Kotoshiro</td>\n",
              "      <td>[Action, Casual, Indie, Simulation, Strategy]</td>\n",
              "      <td>Lost Summoner Kitty</td>\n",
              "      <td>Lost Summoner Kitty</td>\n",
              "      <td>http://store.steampowered.com/app/761140/Lost_...</td>\n",
              "      <td>2018-01-04</td>\n",
              "      <td>[Strategy, Action, Indie, Casual, Simulation]</td>\n",
              "      <td>4.49</td>\n",
              "      <td>http://steamcommunity.com/app/761140/reviews/?...</td>\n",
              "      <td>[Single-player]</td>\n",
              "      <td>4.99</td>\n",
              "      <td>False</td>\n",
              "      <td>761140</td>\n",
              "      <td>Kotoshiro</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>Making Fun, Inc.</td>\n",
              "      <td>[Free to Play, Indie, RPG, Strategy]</td>\n",
              "      <td>Ironbound</td>\n",
              "      <td>Ironbound</td>\n",
              "      <td>http://store.steampowered.com/app/643980/Ironb...</td>\n",
              "      <td>2018-01-04</td>\n",
              "      <td>[Free to Play, Strategy, Indie, RPG, Card Game...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>http://steamcommunity.com/app/643980/reviews/?...</td>\n",
              "      <td>[Single-player, Multi-player, Online Multi-Pla...</td>\n",
              "      <td>Free To Play</td>\n",
              "      <td>False</td>\n",
              "      <td>643980</td>\n",
              "      <td>Secret Level SRL</td>\n",
              "      <td>Mostly Positive</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>Poolians.com</td>\n",
              "      <td>[Casual, Free to Play, Indie, Simulation, Sports]</td>\n",
              "      <td>Real Pool 3D - Poolians</td>\n",
              "      <td>Real Pool 3D - Poolians</td>\n",
              "      <td>http://store.steampowered.com/app/670290/Real_...</td>\n",
              "      <td>2017-07-24</td>\n",
              "      <td>[Free to Play, Simulation, Sports, Casual, Ind...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>http://steamcommunity.com/app/670290/reviews/?...</td>\n",
              "      <td>[Single-player, Multi-player, Online Multi-Pla...</td>\n",
              "      <td>Free to Play</td>\n",
              "      <td>False</td>\n",
              "      <td>670290</td>\n",
              "      <td>Poolians.com</td>\n",
              "      <td>Mostly Positive</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>f7ccb8df</td>\n",
              "      <td>[Action, Adventure, Casual]</td>\n",
              "      <td>f39b8eba2222</td>\n",
              "      <td>f39b8eba2222</td>\n",
              "      <td>http://store.steampowered.com/app/767400/2222/</td>\n",
              "      <td>2017-12-07</td>\n",
              "      <td>[Action, Adventure, Casual]</td>\n",
              "      <td>0.83</td>\n",
              "      <td>http://steamcommunity.com/app/767400/reviews/?...</td>\n",
              "      <td>[Single-player]</td>\n",
              "      <td>0.99</td>\n",
              "      <td>False</td>\n",
              "      <td>767400</td>\n",
              "      <td>f7ccb8df</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>Log Challenge</td>\n",
              "      <td>NaN</td>\n",
              "      <td>http://store.steampowered.com/app/773570/Log_C...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>[Action, Indie, Casual, Sports]</td>\n",
              "      <td>1.79</td>\n",
              "      <td>http://steamcommunity.com/app/773570/reviews/?...</td>\n",
              "      <td>[Single-player, Full controller support, HTC V...</td>\n",
              "      <td>2.99</td>\n",
              "      <td>False</td>\n",
              "      <td>773570</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "          publisher                                             genres  \\\n",
              "0         Kotoshiro      [Action, Casual, Indie, Simulation, Strategy]   \n",
              "1  Making Fun, Inc.               [Free to Play, Indie, RPG, Strategy]   \n",
              "2      Poolians.com  [Casual, Free to Play, Indie, Simulation, Sports]   \n",
              "3          f7ccb8df                        [Action, Adventure, Casual]   \n",
              "4               NaN                                                NaN   \n",
              "\n",
              "                  app_name                    title  \\\n",
              "0      Lost Summoner Kitty      Lost Summoner Kitty   \n",
              "1                Ironbound                Ironbound   \n",
              "2  Real Pool 3D - Poolians  Real Pool 3D - Poolians   \n",
              "3             f39b8eba2222             f39b8eba2222   \n",
              "4            Log Challenge                      NaN   \n",
              "\n",
              "                                                 url release_date  \\\n",
              "0  http://store.steampowered.com/app/761140/Lost_...   2018-01-04   \n",
              "1  http://store.steampowered.com/app/643980/Ironb...   2018-01-04   \n",
              "2  http://store.steampowered.com/app/670290/Real_...   2017-07-24   \n",
              "3     http://store.steampowered.com/app/767400/2222/   2017-12-07   \n",
              "4  http://store.steampowered.com/app/773570/Log_C...          NaN   \n",
              "\n",
              "                                                tags  discount_price  \\\n",
              "0      [Strategy, Action, Indie, Casual, Simulation]            4.49   \n",
              "1  [Free to Play, Strategy, Indie, RPG, Card Game...             NaN   \n",
              "2  [Free to Play, Simulation, Sports, Casual, Ind...             NaN   \n",
              "3                        [Action, Adventure, Casual]            0.83   \n",
              "4                    [Action, Indie, Casual, Sports]            1.79   \n",
              "\n",
              "                                         reviews_url  \\\n",
              "0  http://steamcommunity.com/app/761140/reviews/?...   \n",
              "1  http://steamcommunity.com/app/643980/reviews/?...   \n",
              "2  http://steamcommunity.com/app/670290/reviews/?...   \n",
              "3  http://steamcommunity.com/app/767400/reviews/?...   \n",
              "4  http://steamcommunity.com/app/773570/reviews/?...   \n",
              "\n",
              "                                               specs         price  \\\n",
              "0                                    [Single-player]          4.99   \n",
              "1  [Single-player, Multi-player, Online Multi-Pla...  Free To Play   \n",
              "2  [Single-player, Multi-player, Online Multi-Pla...  Free to Play   \n",
              "3                                    [Single-player]          0.99   \n",
              "4  [Single-player, Full controller support, HTC V...          2.99   \n",
              "\n",
              "   early_access      id         developer        sentiment metascore  \n",
              "0         False  761140         Kotoshiro              NaN       NaN  \n",
              "1         False  643980  Secret Level SRL  Mostly Positive       NaN  \n",
              "2         False  670290      Poolians.com  Mostly Positive       NaN  \n",
              "3         False  767400          f7ccb8df              NaN       NaN  \n",
              "4         False  773570               NaN              NaN       NaN  "
            ]
          },
          "execution_count": 43,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "df.head()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Discarding rows with too few Steam reviews to have a sentinment score"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 44,
      "metadata": {
        "id": "pPfsZ_AJLPKz"
      },
      "outputs": [],
      "source": [
        "discarded_sentiments = ['1 user reviews', '2 user reviews', '3 user reviews',\n",
        "                        '4 user reviews', '5 user reviews', '6 user reviews',\n",
        "                        '7 user reviews', '8 user reviews', '9 user reviews']\n",
        "\n",
        "df = df[~df['sentiment'].isin(discarded_sentiments)]"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Discarding null rows because this is the target variable"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 45,
      "metadata": {
        "id": "g_daWkB7LQLH"
      },
      "outputs": [],
      "source": [
        "df = df.dropna(subset='sentiment')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Converting the sentiment names to numerical values from 1-9. This is the \"rating\" that the neural network will learn to predict"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 46,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zwcKslVZLRJf",
        "outputId": "61da6995-fa01-434c-b65b-758db4d6b76c"
      },
      "outputs": [],
      "source": [
        "sentiment_rating_mapping = {\n",
        "    'Overwhelmingly Negative': 1,\n",
        "    'Very Negative': 2,\n",
        "    'Mostly Negative': 3,\n",
        "    'Negative': 4,\n",
        "    'Mixed': 5,\n",
        "    'Mostly Positive': 6,\n",
        "    'Positive': 7,\n",
        "    'Very Positive': 8,\n",
        "    'Overwhelmingly Positive': 9,\n",
        "}\n",
        "\n",
        "df['sentiment'] = df['sentiment'].map(sentiment_rating_mapping)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "These features won't be used so dropping them"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 47,
      "metadata": {
        "id": "BIDZDFqcLTAU"
      },
      "outputs": [],
      "source": [
        "df = df.drop(['discount_price', 'url', 'reviews_url'], axis=1)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Preparing null metascore values. We will handle these null values soon"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 48,
      "metadata": {
        "id": "q2WLtDWrLTzq"
      },
      "outputs": [],
      "source": [
        "df['metascore'] = df['metascore'].replace('NA', float('nan'))\n",
        "df['metascore'] = df['metascore'].replace(pd.NA, float('nan'))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Converting the release date feature to year, month and day_of_week features"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 49,
      "metadata": {
        "id": "shKrL9yKLUZE"
      },
      "outputs": [],
      "source": [
        "df['release_date'] = pd.to_datetime(df['release_date'], errors='coerce')\n",
        "df['year'] = df['release_date'].dt.year\n",
        "df['month'] = df['release_date'].dt.month\n",
        "df['day_of_week'] = df['release_date'].dt.dayofweek\n",
        "df = df.drop('release_date', axis=1)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "The dataset had strings like 'Free' written in the Price column in some areas. Replace all that with 0 price"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 50,
      "metadata": {
        "id": "O7SQ8FB6LVMz"
      },
      "outputs": [],
      "source": [
        "df['price'] = df['price'].replace(to_replace=r'.*Free.*', value=0, regex=True)\n",
        "df['price'] = df['price'].replace(to_replace=r'.*Play.*', value=0, regex=True)\n",
        "df['price'] = df['price'].replace(to_replace=r'.*Install.*', value=0, regex=True)\n",
        "df = df[df['price'] != 'Third-party']\n",
        "df['price'] = df['price'].astype('float')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "If the specs, tags of genres rows are null replace them with an empty list"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 51,
      "metadata": {
        "id": "TD3klhkvLWP_"
      },
      "outputs": [],
      "source": [
        "df['specs'] = df['specs'].apply(lambda x: x if isinstance(x, list) else [])\n",
        "df['tags'] = df['tags'].apply(lambda x: x if isinstance(x, list) else [])\n",
        "df['genres'] = df['genres'].apply(lambda x: x if isinstance(x, list) else [])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Handling null values for each feature"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 52,
      "metadata": {
        "id": "dWQWO37_LW8U"
      },
      "outputs": [],
      "source": [
        "df['publisher'] = df['publisher'].fillna('NA')\n",
        "df['developer'] = df['developer'].fillna('NA')\n",
        "df['app_name'] = df['app_name'].fillna('NA')\n",
        "\n",
        "# Title can be inferred from app_name\n",
        "df['title'] = df['title'].fillna(df['app_name'])\n",
        "df['id'] = df['id'].fillna('NA')\n",
        "\n",
        "df['early_access'] = df['early_access'].fillna(False)\n",
        "\n",
        "df['year'] = df['year'].fillna(-1)\n",
        "df['month'] = df['month'].fillna(-1)\n",
        "df['day_of_week'] = df['day_of_week'].fillna(-1)\n",
        "\n",
        "\n",
        "df['price'].fillna(-1, inplace=True)\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Since metascore is an important feature that can tell us a lot about how good the game is, we will train a metalearner to perform Regression Imputation and predict the mising value using the other features"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 53,
      "metadata": {},
      "outputs": [],
      "source": [
        "from sklearn.ensemble import RandomForestRegressor\n",
        "\n",
        "# Select features that will be used for prediction\n",
        "predictor_vars = ['price', 'sentiment', 'early_access', 'year', 'month', 'day_of_week']\n",
        "\n",
        "# Split data into two parts\n",
        "train_data = df.dropna(subset=['metascore'])\n",
        "predict_data = df[df['metascore'].isnull()]\n",
        "\n",
        "# Train a regression model\n",
        "model = RandomForestRegressor(n_estimators=100)\n",
        "model.fit(train_data[predictor_vars], train_data['metascore'])\n",
        "\n",
        "# Predict the missing metascores\n",
        "predicted_scores = model.predict(predict_data[predictor_vars])\n",
        "\n",
        "# Replace the missing metascores in the original DataFrame\n",
        "df.loc[df['metascore'].isnull(), 'metascore'] = predicted_scores"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Splitting features into numerical and categorical columns"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 54,
      "metadata": {},
      "outputs": [],
      "source": [
        "categorical_features = ['publisher', 'developer']\n",
        "multi_label_features = ['specs', 'tags', 'genres']\n",
        "numerical_features = ['price', 'year', 'metascore', 'month', 'day_of_week', 'early_access']"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Sampling 100 random rows of the original dataset for demonstration later"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 69,
      "metadata": {},
      "outputs": [
        {
          "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>publisher</th>\n",
              "      <th>genres</th>\n",
              "      <th>app_name</th>\n",
              "      <th>title</th>\n",
              "      <th>tags</th>\n",
              "      <th>specs</th>\n",
              "      <th>price</th>\n",
              "      <th>early_access</th>\n",
              "      <th>id</th>\n",
              "      <th>developer</th>\n",
              "      <th>...</th>\n",
              "      <th>genres_Photo Editing</th>\n",
              "      <th>genres_RPG</th>\n",
              "      <th>genres_Racing</th>\n",
              "      <th>genres_Simulation</th>\n",
              "      <th>genres_Software Training</th>\n",
              "      <th>genres_Sports</th>\n",
              "      <th>genres_Strategy</th>\n",
              "      <th>genres_Utilities</th>\n",
              "      <th>genres_Video Production</th>\n",
              "      <th>genres_Web Publishing</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>14573</th>\n",
              "      <td>556</td>\n",
              "      <td>[Action]</td>\n",
              "      <td>Wolfenstein II: The New Colossus</td>\n",
              "      <td>Wolfenstein II: The New Colossus</td>\n",
              "      <td>[Action, FPS, Gore, Violent, Alternate History...</td>\n",
              "      <td>[Single-player, Steam Achievements, Full contr...</td>\n",
              "      <td>3.617664</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>612880</td>\n",
              "      <td>3815</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>30855</th>\n",
              "      <td>4982</td>\n",
              "      <td>[Adventure, Indie]</td>\n",
              "      <td>Resonance</td>\n",
              "      <td>Resonance</td>\n",
              "      <td>[Adventure, Point &amp; Click, Indie, Sci-fi, Retr...</td>\n",
              "      <td>[Single-player, Steam Achievements, Steam Trad...</td>\n",
              "      <td>0.022143</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>212050</td>\n",
              "      <td>7009</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6077</th>\n",
              "      <td>2018</td>\n",
              "      <td>[Action, Indie]</td>\n",
              "      <td>Share</td>\n",
              "      <td>Share</td>\n",
              "      <td>[Indie, Action, Horror]</td>\n",
              "      <td>[Single-player, Steam Trading Cards]</td>\n",
              "      <td>-0.337409</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>448020</td>\n",
              "      <td>2779</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>21537</th>\n",
              "      <td>5005</td>\n",
              "      <td>[Action, Adventure]</td>\n",
              "      <td>Rebels Character Pack</td>\n",
              "      <td>Rebels Character Pack</td>\n",
              "      <td>[Action, Adventure]</td>\n",
              "      <td>[Single-player, Shared/Split Screen, Downloada...</td>\n",
              "      <td>-0.553140</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>466549</td>\n",
              "      <td>6501</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7571</th>\n",
              "      <td>4357</td>\n",
              "      <td>[Strategy]</td>\n",
              "      <td>Hidden Dimensions 3</td>\n",
              "      <td>Hidden Dimensions 3</td>\n",
              "      <td>[Strategy, Free to Play, Card Game, Indie]</td>\n",
              "      <td>[Single-player]</td>\n",
              "      <td>-0.696242</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>531540</td>\n",
              "      <td>5948</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>31063</th>\n",
              "      <td>3947</td>\n",
              "      <td>[]</td>\n",
              "      <td>Sonic Generations Demo</td>\n",
              "      <td>Sonic Generations Demo</td>\n",
              "      <td>[Free to Play, Adventure, 3D Platformer, Actio...</td>\n",
              "      <td>[Game demo]</td>\n",
              "      <td>-0.768152</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>202290</td>\n",
              "      <td>1732</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5742</th>\n",
              "      <td>3056</td>\n",
              "      <td>[]</td>\n",
              "      <td>Carpe Lucem - Seize The Light VR</td>\n",
              "      <td>Carpe Lucem - Seize The Light VR</td>\n",
              "      <td>[Casual, Puzzle, Strategy, Indie, VR]</td>\n",
              "      <td>[Single-player, Full controller support, Steam...</td>\n",
              "      <td>0.381695</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>433700</td>\n",
              "      <td>4195</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>30072</th>\n",
              "      <td>3402</td>\n",
              "      <td>[Action, Indie]</td>\n",
              "      <td>Teleglitch: Die More Edition</td>\n",
              "      <td>Teleglitch: Die More Edition</td>\n",
              "      <td>[Rogue-like, Indie, Top-Down Shooter, Difficul...</td>\n",
              "      <td>[Single-player, Steam Achievements, Full contr...</td>\n",
              "      <td>0.237874</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>234390</td>\n",
              "      <td>6281</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>9399</th>\n",
              "      <td>569</td>\n",
              "      <td>[Action]</td>\n",
              "      <td>The Signal From Tlva</td>\n",
              "      <td>The Signal From Tlva</td>\n",
              "      <td>[Action, Robots, Open World, Sci-fi, FPS, Firs...</td>\n",
              "      <td>[Single-player, Steam Achievements, Full contr...</td>\n",
              "      <td>0.741247</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>457760</td>\n",
              "      <td>756</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7629</th>\n",
              "      <td>5442</td>\n",
              "      <td>[Free to Play, Indie, Massively Multiplayer, R...</td>\n",
              "      <td>The Infinite Black</td>\n",
              "      <td>The Infinite Black</td>\n",
              "      <td>[Free to Play, Strategy, Massively Multiplayer...</td>\n",
              "      <td>[Multi-player, Online Multi-Player, MMO, Onlin...</td>\n",
              "      <td>-0.696242</td>\n",
              "      <td>-0.279955</td>\n",
              "      <td>489000</td>\n",
              "      <td>5831</td>\n",
              "      <td>...</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>100 rows × 415 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "       publisher                                             genres  \\\n",
              "14573        556                                           [Action]   \n",
              "30855       4982                                 [Adventure, Indie]   \n",
              "6077        2018                                    [Action, Indie]   \n",
              "21537       5005                                [Action, Adventure]   \n",
              "7571        4357                                         [Strategy]   \n",
              "...          ...                                                ...   \n",
              "31063       3947                                                 []   \n",
              "5742        3056                                                 []   \n",
              "30072       3402                                    [Action, Indie]   \n",
              "9399         569                                           [Action]   \n",
              "7629        5442  [Free to Play, Indie, Massively Multiplayer, R...   \n",
              "\n",
              "                               app_name                             title  \\\n",
              "14573  Wolfenstein II: The New Colossus  Wolfenstein II: The New Colossus   \n",
              "30855                         Resonance                         Resonance   \n",
              "6077                              Share                             Share   \n",
              "21537             Rebels Character Pack             Rebels Character Pack   \n",
              "7571                Hidden Dimensions 3               Hidden Dimensions 3   \n",
              "...                                 ...                               ...   \n",
              "31063            Sonic Generations Demo            Sonic Generations Demo   \n",
              "5742   Carpe Lucem - Seize The Light VR  Carpe Lucem - Seize The Light VR   \n",
              "30072      Teleglitch: Die More Edition      Teleglitch: Die More Edition   \n",
              "9399               The Signal From Tlva              The Signal From Tlva   \n",
              "7629                 The Infinite Black                The Infinite Black   \n",
              "\n",
              "                                                    tags  \\\n",
              "14573  [Action, FPS, Gore, Violent, Alternate History...   \n",
              "30855  [Adventure, Point & Click, Indie, Sci-fi, Retr...   \n",
              "6077                             [Indie, Action, Horror]   \n",
              "21537                                [Action, Adventure]   \n",
              "7571          [Strategy, Free to Play, Card Game, Indie]   \n",
              "...                                                  ...   \n",
              "31063  [Free to Play, Adventure, 3D Platformer, Actio...   \n",
              "5742               [Casual, Puzzle, Strategy, Indie, VR]   \n",
              "30072  [Rogue-like, Indie, Top-Down Shooter, Difficul...   \n",
              "9399   [Action, Robots, Open World, Sci-fi, FPS, Firs...   \n",
              "7629   [Free to Play, Strategy, Massively Multiplayer...   \n",
              "\n",
              "                                                   specs     price  \\\n",
              "14573  [Single-player, Steam Achievements, Full contr...  3.617664   \n",
              "30855  [Single-player, Steam Achievements, Steam Trad...  0.022143   \n",
              "6077                [Single-player, Steam Trading Cards] -0.337409   \n",
              "21537  [Single-player, Shared/Split Screen, Downloada... -0.553140   \n",
              "7571                                     [Single-player] -0.696242   \n",
              "...                                                  ...       ...   \n",
              "31063                                        [Game demo] -0.768152   \n",
              "5742   [Single-player, Full controller support, Steam...  0.381695   \n",
              "30072  [Single-player, Steam Achievements, Full contr...  0.237874   \n",
              "9399   [Single-player, Steam Achievements, Full contr...  0.741247   \n",
              "7629   [Multi-player, Online Multi-Player, MMO, Onlin... -0.696242   \n",
              "\n",
              "       early_access      id  developer  ...  genres_Photo Editing  genres_RPG  \\\n",
              "14573     -0.279955  612880       3815  ...                     0           0   \n",
              "30855     -0.279955  212050       7009  ...                     0           0   \n",
              "6077      -0.279955  448020       2779  ...                     0           0   \n",
              "21537     -0.279955  466549       6501  ...                     0           0   \n",
              "7571      -0.279955  531540       5948  ...                     0           0   \n",
              "...             ...     ...        ...  ...                   ...         ...   \n",
              "31063     -0.279955  202290       1732  ...                     0           0   \n",
              "5742      -0.279955  433700       4195  ...                     0           0   \n",
              "30072     -0.279955  234390       6281  ...                     0           0   \n",
              "9399      -0.279955  457760        756  ...                     0           0   \n",
              "7629      -0.279955  489000       5831  ...                     0           1   \n",
              "\n",
              "       genres_Racing  genres_Simulation  genres_Software Training  \\\n",
              "14573              0                  0                         0   \n",
              "30855              0                  0                         0   \n",
              "6077               0                  0                         0   \n",
              "21537              0                  0                         0   \n",
              "7571               0                  0                         0   \n",
              "...              ...                ...                       ...   \n",
              "31063              0                  0                         0   \n",
              "5742               0                  0                         0   \n",
              "30072              0                  0                         0   \n",
              "9399               0                  0                         0   \n",
              "7629               0                  0                         0   \n",
              "\n",
              "       genres_Sports  genres_Strategy  genres_Utilities  \\\n",
              "14573              0                0                 0   \n",
              "30855              0                0                 0   \n",
              "6077               0                0                 0   \n",
              "21537              0                0                 0   \n",
              "7571               0                1                 0   \n",
              "...              ...              ...               ...   \n",
              "31063              0                0                 0   \n",
              "5742               0                0                 0   \n",
              "30072              0                0                 0   \n",
              "9399               0                0                 0   \n",
              "7629               0                1                 0   \n",
              "\n",
              "       genres_Video Production  genres_Web Publishing  \n",
              "14573                        0                      0  \n",
              "30855                        0                      0  \n",
              "6077                         0                      0  \n",
              "21537                        0                      0  \n",
              "7571                         0                      0  \n",
              "...                        ...                    ...  \n",
              "31063                        0                      0  \n",
              "5742                         0                      0  \n",
              "30072                        0                      0  \n",
              "9399                         0                      0  \n",
              "7629                         0                      0  \n",
              "\n",
              "[100 rows x 415 columns]"
            ]
          },
          "execution_count": 69,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "# Split the original dataset first\n",
        "df_test = df.sample(n=100)\n",
        "df_test"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "We will now encode the features into embeddings to train the neural network. There are 3 types of features:\n",
        "- Multi-Label features (specs, genres, tags)\n",
        "- Numerical Features (price, etc)\n",
        "- Categorical Features (developer, publisher) "
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 56,
      "metadata": {},
      "outputs": [
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n",
            "C:\\Users\\Ayaan Danish\\AppData\\Local\\Temp\\ipykernel_17808\\13245007.py:6: PerformanceWarning: DataFrame is highly fragmented.  This is usually the result of calling `frame.insert` many times, which has poor performance.  Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()`\n",
            "  df[f\"{col}_{feature}\"] = binarized[:, i]\n"
          ]
        }
      ],
      "source": [
        "# Creating MultiLabelBinarizers for multi-label categorical features\n",
        "multi_label_binarizers = {col: MultiLabelBinarizer().fit(df[col]) for col in multi_label_features}\n",
        "for col, mlb in multi_label_binarizers.items():\n",
        "    binarized = mlb.transform(df[col])\n",
        "    for i, feature in enumerate(mlb.classes_):\n",
        "        df[f\"{col}_{feature}\"] = binarized[:, i]\n",
        "        \n",
        "# Standardizing numerical features\n",
        "scaler = StandardScaler()\n",
        "df[numerical_features] = scaler.fit_transform(df[numerical_features])\n",
        "\n",
        "# Creating Label Encoders for single-label categorical features\n",
        "label_encoders = {col: LabelEncoder().fit(df[col]) for col in categorical_features}\n",
        "for col, le in label_encoders.items():\n",
        "    df[col] = le.transform(df[col])\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Extracting target and features"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 57,
      "metadata": {},
      "outputs": [],
      "source": [
        "X_categorical = df[categorical_features]\n",
        "X_multi_label = df[[f\"{col}_{feature}\" for col in multi_label_features for feature in multi_label_binarizers[col].classes_]]\n",
        "X_numerical = df[numerical_features]\n",
        "\n",
        "y = df['sentiment'].values"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Splitting the data into training and testing sets"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 58,
      "metadata": {},
      "outputs": [],
      "source": [
        "X_train_categorical, X_test_categorical, X_train_multi, X_test_multi, X_train_num, X_test_num, y_train, y_test = train_test_split(\n",
        "    X_categorical, X_multi_label, X_numerical, y, test_size=0.2, random_state=42)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Defining input layers for categorical features and their embeddings, as well as the hidden layers of the model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 59,
      "metadata": {},
      "outputs": [],
      "source": [
        "embedding_layers = []\n",
        "categorical_inputs = []\n",
        "for col in categorical_features:\n",
        "    num_classes = df[col].nunique()\n",
        "    input_layer = tf.keras.layers.Input(shape=(1,), name=col)\n",
        "    embedding_layer = tf.keras.layers.Embedding(input_dim=num_classes, output_dim=100)(input_layer)\n",
        "    embedding_layer = tf.keras.layers.Flatten()(embedding_layer)\n",
        "    embedding_layers.append(embedding_layer)\n",
        "    categorical_inputs.append(input_layer)\n",
        "\n",
        "# Input for multi-label features\n",
        "multi_input = tf.keras.layers.Input(shape=(X_train_multi.shape[1],), name='multi_label_input')\n",
        "multi_dense = tf.keras.layers.Dense(64, activation='relu')(multi_input)\n",
        "\n",
        "# Input for numerical features\n",
        "num_input = tf.keras.layers.Input(shape=(X_train_num.shape[1],), name='numerical_input')\n",
        "num_dense = tf.keras.layers.Dense(64, activation='relu')(num_input)\n",
        "\n",
        "# Concatenate all embeddings and inputs\n",
        "all_features = tf.keras.layers.Concatenate(axis=1)(embedding_layers + [multi_dense, num_dense])\n",
        "\n",
        "# Add hidden layers\n",
        "x = tf.keras.layers.Dense(128, activation='relu', kernel_regularizer='l1')(all_features)\n",
        "x = tf.keras.layers.Dropout(0.1)(x)\n",
        "x = tf.keras.layers.Dense(64, activation='relu')(x)\n",
        "\n",
        "# Output layer\n",
        "output = tf.keras.layers.Dense(1, activation='linear')(x)\n",
        "\n",
        "# Building the model\n",
        "model = tf.keras.Model(inputs=[categorical_inputs, multi_input, num_input], outputs=output)\n",
        "model.compile(optimizer='adam', loss='mse', metrics=['mae'])\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Training the model. This has been treated as a regression problem instead of a classification problem"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 60,
      "metadata": {},
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "10/10 [==============================] - 1s 30ms/step - loss: 58.1033 - mae: 5.7290 - val_loss: 45.8391 - val_mae: 4.6965\n",
            "Epoch 2/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 34.5403 - mae: 3.3208 - val_loss: 24.1960 - val_mae: 1.8795\n",
            "Epoch 3/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 24.4605 - mae: 1.9545 - val_loss: 20.4595 - val_mae: 1.6117\n",
            "Epoch 4/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 19.3133 - mae: 1.5833 - val_loss: 17.7009 - val_mae: 1.5707\n",
            "Epoch 5/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 16.1684 - mae: 1.3462 - val_loss: 14.5378 - val_mae: 1.2264\n",
            "Epoch 6/20\n",
            "10/10 [==============================] - 0s 14ms/step - loss: 13.6256 - mae: 1.1708 - val_loss: 12.2739 - val_mae: 1.1258\n",
            "Epoch 7/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 11.5275 - mae: 1.0937 - val_loss: 10.4449 - val_mae: 1.0442\n",
            "Epoch 8/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 9.8165 - mae: 1.0241 - val_loss: 8.9604 - val_mae: 1.0034\n",
            "Epoch 9/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 8.4632 - mae: 0.9953 - val_loss: 7.7767 - val_mae: 0.9805\n",
            "Epoch 10/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 7.3688 - mae: 0.9573 - val_loss: 6.8445 - val_mae: 0.9536\n",
            "Epoch 11/20\n",
            "10/10 [==============================] - 0s 13ms/step - loss: 6.5593 - mae: 0.9432 - val_loss: 6.1832 - val_mae: 0.9362\n",
            "Epoch 12/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 5.9888 - mae: 0.9290 - val_loss: 5.7167 - val_mae: 0.9186\n",
            "Epoch 13/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 5.5401 - mae: 0.9105 - val_loss: 5.3237 - val_mae: 0.9016\n",
            "Epoch 14/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 5.1928 - mae: 0.8964 - val_loss: 5.0015 - val_mae: 0.8903\n",
            "Epoch 15/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 4.8864 - mae: 0.8831 - val_loss: 4.7158 - val_mae: 0.8756\n",
            "Epoch 16/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 4.6090 - mae: 0.8673 - val_loss: 4.4644 - val_mae: 0.8633\n",
            "Epoch 17/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 4.3583 - mae: 0.8545 - val_loss: 4.2367 - val_mae: 0.8528\n",
            "Epoch 18/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 4.1310 - mae: 0.8437 - val_loss: 4.0243 - val_mae: 0.8430\n",
            "Epoch 19/20\n",
            "10/10 [==============================] - 0s 16ms/step - loss: 3.9252 - mae: 0.8297 - val_loss: 3.8286 - val_mae: 0.8340\n",
            "Epoch 20/20\n",
            "10/10 [==============================] - 0s 12ms/step - loss: 3.7261 - mae: 0.8174 - val_loss: 3.6499 - val_mae: 0.8265\n"
          ]
        }
      ],
      "source": [
        "num_epochs = 20\n",
        "history = model.fit(\n",
        "    [X_train_categorical[col] for col in categorical_features] + [X_train_multi, X_train_num],\n",
        "    y_train,\n",
        "    epochs=num_epochs,\n",
        "    batch_size=1024,\n",
        "    validation_split=0.2\n",
        ")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "The model gives a Mean Absolute Error of 0.85, which is a good result. It means it can predict ratings/sentiment scores within a margin of < 1 point"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 61,
      "metadata": {},
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "96/96 [==============================] - 0s 3ms/step - loss: 3.7211 - mae: 0.8497\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "[3.721102476119995, 0.8496538400650024]"
            ]
          },
          "execution_count": 61,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "model.evaluate([X_test_categorical[col] for col in categorical_features] + [X_test_multi, X_test_num],  y_test)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 62,
      "metadata": {},
      "outputs": [],
      "source": [
        "def plot_model_performance(history):\n",
        "    acc = history.history['mae']\n",
        "    val_acc = history.history['val_mae']\n",
        "\n",
        "    loss = history.history['loss']\n",
        "    val_loss = history.history['val_loss']\n",
        "\n",
        "    epochs_range = range(num_epochs)\n",
        "\n",
        "    plt.figure(figsize=(20, 8))\n",
        "    plt.subplot(1, 2, 1)\n",
        "    plt.plot(epochs_range, acc, label='Training Accuracy')\n",
        "    plt.plot(epochs_range, val_acc, label='Validation Accuracy')\n",
        "    plt.legend(loc='upper right')\n",
        "    plt.title('Training and Validation Accuracy')\n",
        "\n",
        "    plt.subplot(1, 2, 2)\n",
        "    plt.plot(epochs_range, loss, label='Training Loss')\n",
        "    plt.plot(epochs_range, val_loss, label='Validation Loss')\n",
        "    plt.legend(loc='upper right')\n",
        "    plt.title('Training and Validation Loss')\n",
        "    plt.suptitle(f'Model Results')\n",
        "    plt.savefig(f'training_history.png')\n",
        "    plt.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Plotting the training history we can see that it is not overfitting (took me ages to make it not overfit)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 63,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 2000x800 with 2 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "plot_model_performance(history)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Demonstrating the model predictions on those 100 rows we sampled earlier. The model calculates predicted sentiment scores for all rows and sorts them to generate recommendations. These recommendations are not personalized to one specific user but instead represent the general public sentiment surrounding that game, as obtained from its reviews on Steam"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 67,
      "metadata": {},
      "outputs": [
        {
          "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>publisher</th>\n",
              "      <th>developer</th>\n",
              "      <th>app_name</th>\n",
              "      <th>specs</th>\n",
              "      <th>tags</th>\n",
              "      <th>genres</th>\n",
              "      <th>price</th>\n",
              "      <th>metascore</th>\n",
              "      <th>year</th>\n",
              "      <th>month</th>\n",
              "      <th>day_of_week</th>\n",
              "      <th>early_access</th>\n",
              "      <th>true_sentiment</th>\n",
              "      <th>predicted_sentiment</th>\n",
              "      <th>sentiment_name</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>84</th>\n",
              "      <td>Brace Yourself Games, Klei Entertainment</td>\n",
              "      <td>Brace Yourself Games</td>\n",
              "      <td>Crypt of the NecroDancer</td>\n",
              "      <td>[Single-player, Local Co-op, Steam Achievement...</td>\n",
              "      <td>[Rhythm, Rogue-like, Great Soundtrack, Pixel G...</td>\n",
              "      <td>[Action, Indie, RPG]</td>\n",
              "      <td>14.99</td>\n",
              "      <td>87.000000</td>\n",
              "      <td>2015.0</td>\n",
              "      <td>4.0</td>\n",
              "      <td>3.0</td>\n",
              "      <td>False</td>\n",
              "      <td>9</td>\n",
              "      <td>9</td>\n",
              "      <td>Overwhelmingly Positive</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>99</th>\n",
              "      <td>Devolver Digital, Croteam</td>\n",
              "      <td>Mommy's Best Games</td>\n",
              "      <td>Serious Sam Double D XXL</td>\n",
              "      <td>[Single-player, Local Co-op, Steam Achievement...</td>\n",
              "      <td>[Action, Indie, Platformer, Comedy, Shooter, 2...</td>\n",
              "      <td>[Action, Indie]</td>\n",
              "      <td>9.99</td>\n",
              "      <td>83.370000</td>\n",
              "      <td>2011.0</td>\n",
              "      <td>8.0</td>\n",
              "      <td>1.0</td>\n",
              "      <td>False</td>\n",
              "      <td>8</td>\n",
              "      <td>8</td>\n",
              "      <td>Very Positive</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>44</th>\n",
              "      <td>Electronic Arts</td>\n",
              "      <td>BioWare</td>\n",
              "      <td>Dragon Age: Origins</td>\n",
              "      <td>[Single-player]</td>\n",
              "      <td>[RPG, Fantasy, Story Rich, Multiple Endings, S...</td>\n",
              "      <td>[RPG]</td>\n",
              "      <td>19.99</td>\n",
              "      <td>91.000000</td>\n",
              "      <td>2009.0</td>\n",
              "      <td>11.0</td>\n",
              "      <td>4.0</td>\n",
              "      <td>False</td>\n",
              "      <td>8</td>\n",
              "      <td>8</td>\n",
              "      <td>Very Positive</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>75</th>\n",
              "      <td>Square Enix</td>\n",
              "      <td>Square Enix</td>\n",
              "      <td>WORLD OF FINAL FANTASY</td>\n",
              "      <td>[Single-player, Steam Achievements, Full contr...</td>\n",
              "      <td>[RPG, JRPG, Anime, Cute, Fantasy, Turn-Based C...</td>\n",
              "      <td>[RPG]</td>\n",
              "      <td>39.99</td>\n",
              "      <td>84.020000</td>\n",
              "      <td>2017.0</td>\n",
              "      <td>11.0</td>\n",
              "      <td>1.0</td>\n",
              "      <td>False</td>\n",
              "      <td>8</td>\n",
              "      <td>8</td>\n",
              "      <td>Very Positive</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>26</th>\n",
              "      <td>Sekai Project</td>\n",
              "      <td>Vifth Floor</td>\n",
              "      <td>Just Deserts</td>\n",
              "      <td>[Single-player, Steam Achievements, Steam Trad...</td>\n",
              "      <td>[Anime, Strategy, Visual Novel, Dating Sim, In...</td>\n",
              "      <td>[Casual, Indie, Simulation, Strategy]</td>\n",
              "      <td>9.99</td>\n",
              "      <td>76.836667</td>\n",
              "      <td>2016.0</td>\n",
              "      <td>7.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>False</td>\n",
              "      <td>8</td>\n",
              "      <td>8</td>\n",
              "      <td>Very Positive</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>Dovetail Games- Flight</td>\n",
              "      <td>TakeFlight Interactive</td>\n",
              "      <td>FSX Steam Edition: Final Approach Add-On</td>\n",
              "      <td>[Single-player, Multi-player, Downloadable Con...</td>\n",
              "      <td>[Simulation]</td>\n",
              "      <td>[Simulation]</td>\n",
              "      <td>5.99</td>\n",
              "      <td>57.890000</td>\n",
              "      <td>2016.0</td>\n",
              "      <td>10.0</td>\n",
              "      <td>2.0</td>\n",
              "      <td>False</td>\n",
              "      <td>5</td>\n",
              "      <td>4</td>\n",
              "      <td>Negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>61</th>\n",
              "      <td>WB Games</td>\n",
              "      <td>NetherRealm Studios,QLOC</td>\n",
              "      <td>Injustice 2 - Fighter Pack 2</td>\n",
              "      <td>[Single-player, Multi-player, Online Multi-Pla...</td>\n",
              "      <td>[Action]</td>\n",
              "      <td>[Action]</td>\n",
              "      <td>19.99</td>\n",
              "      <td>56.450000</td>\n",
              "      <td>2017.0</td>\n",
              "      <td>11.0</td>\n",
              "      <td>1.0</td>\n",
              "      <td>False</td>\n",
              "      <td>3</td>\n",
              "      <td>4</td>\n",
              "      <td>Negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>79</th>\n",
              "      <td>The_Kiddy</td>\n",
              "      <td>The_Kiddy</td>\n",
              "      <td>Polterheist</td>\n",
              "      <td>[Single-player, Steam Trading Cards]</td>\n",
              "      <td>[Action, Indie, Simulation, Survival, Horror, ...</td>\n",
              "      <td>[Action, Indie, Simulation]</td>\n",
              "      <td>4.99</td>\n",
              "      <td>47.640000</td>\n",
              "      <td>2017.0</td>\n",
              "      <td>2.0</td>\n",
              "      <td>1.0</td>\n",
              "      <td>False</td>\n",
              "      <td>3</td>\n",
              "      <td>3</td>\n",
              "      <td>Mostly Negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>63</th>\n",
              "      <td>FireFly Studios</td>\n",
              "      <td>FireFly Studios</td>\n",
              "      <td>Stronghold 3 Gold</td>\n",
              "      <td>[Single-player, Multi-player, Cross-Platform M...</td>\n",
              "      <td>[Strategy, Medieval, City Builder, Simulation,...</td>\n",
              "      <td>[Simulation, Strategy]</td>\n",
              "      <td>29.99</td>\n",
              "      <td>46.960000</td>\n",
              "      <td>2012.0</td>\n",
              "      <td>5.0</td>\n",
              "      <td>4.0</td>\n",
              "      <td>False</td>\n",
              "      <td>3</td>\n",
              "      <td>3</td>\n",
              "      <td>Mostly Negative</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>57</th>\n",
              "      <td>Siberian Digital</td>\n",
              "      <td>Rusty Knights</td>\n",
              "      <td>The Atomy</td>\n",
              "      <td>[Single-player, Steam Trading Cards]</td>\n",
              "      <td>[Action, Adventure, Indie]</td>\n",
              "      <td>[Action, Adventure, Indie]</td>\n",
              "      <td>3.99</td>\n",
              "      <td>47.290000</td>\n",
              "      <td>2016.0</td>\n",
              "      <td>10.0</td>\n",
              "      <td>2.0</td>\n",
              "      <td>False</td>\n",
              "      <td>3</td>\n",
              "      <td>3</td>\n",
              "      <td>Mostly Negative</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>100 rows × 15 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "                                   publisher                 developer  \\\n",
              "84  Brace Yourself Games, Klei Entertainment      Brace Yourself Games   \n",
              "99                 Devolver Digital, Croteam        Mommy's Best Games   \n",
              "44                           Electronic Arts                   BioWare   \n",
              "75                               Square Enix               Square Enix   \n",
              "26                             Sekai Project               Vifth Floor   \n",
              "..                                       ...                       ...   \n",
              "1                     Dovetail Games- Flight    TakeFlight Interactive   \n",
              "61                                  WB Games  NetherRealm Studios,QLOC   \n",
              "79                                 The_Kiddy                 The_Kiddy   \n",
              "63                           FireFly Studios           FireFly Studios   \n",
              "57                          Siberian Digital             Rusty Knights   \n",
              "\n",
              "                                    app_name  \\\n",
              "84                  Crypt of the NecroDancer   \n",
              "99                  Serious Sam Double D XXL   \n",
              "44                       Dragon Age: Origins   \n",
              "75                    WORLD OF FINAL FANTASY   \n",
              "26                              Just Deserts   \n",
              "..                                       ...   \n",
              "1   FSX Steam Edition: Final Approach Add-On   \n",
              "61              Injustice 2 - Fighter Pack 2   \n",
              "79                               Polterheist   \n",
              "63                         Stronghold 3 Gold   \n",
              "57                                 The Atomy   \n",
              "\n",
              "                                                specs  \\\n",
              "84  [Single-player, Local Co-op, Steam Achievement...   \n",
              "99  [Single-player, Local Co-op, Steam Achievement...   \n",
              "44                                    [Single-player]   \n",
              "75  [Single-player, Steam Achievements, Full contr...   \n",
              "26  [Single-player, Steam Achievements, Steam Trad...   \n",
              "..                                                ...   \n",
              "1   [Single-player, Multi-player, Downloadable Con...   \n",
              "61  [Single-player, Multi-player, Online Multi-Pla...   \n",
              "79               [Single-player, Steam Trading Cards]   \n",
              "63  [Single-player, Multi-player, Cross-Platform M...   \n",
              "57               [Single-player, Steam Trading Cards]   \n",
              "\n",
              "                                                 tags  \\\n",
              "84  [Rhythm, Rogue-like, Great Soundtrack, Pixel G...   \n",
              "99  [Action, Indie, Platformer, Comedy, Shooter, 2...   \n",
              "44  [RPG, Fantasy, Story Rich, Multiple Endings, S...   \n",
              "75  [RPG, JRPG, Anime, Cute, Fantasy, Turn-Based C...   \n",
              "26  [Anime, Strategy, Visual Novel, Dating Sim, In...   \n",
              "..                                                ...   \n",
              "1                                        [Simulation]   \n",
              "61                                           [Action]   \n",
              "79  [Action, Indie, Simulation, Survival, Horror, ...   \n",
              "63  [Strategy, Medieval, City Builder, Simulation,...   \n",
              "57                         [Action, Adventure, Indie]   \n",
              "\n",
              "                                   genres  price  metascore    year  month  \\\n",
              "84                   [Action, Indie, RPG]  14.99  87.000000  2015.0    4.0   \n",
              "99                        [Action, Indie]   9.99  83.370000  2011.0    8.0   \n",
              "44                                  [RPG]  19.99  91.000000  2009.0   11.0   \n",
              "75                                  [RPG]  39.99  84.020000  2017.0   11.0   \n",
              "26  [Casual, Indie, Simulation, Strategy]   9.99  76.836667  2016.0    7.0   \n",
              "..                                    ...    ...        ...     ...    ...   \n",
              "1                            [Simulation]   5.99  57.890000  2016.0   10.0   \n",
              "61                               [Action]  19.99  56.450000  2017.0   11.0   \n",
              "79            [Action, Indie, Simulation]   4.99  47.640000  2017.0    2.0   \n",
              "63                 [Simulation, Strategy]  29.99  46.960000  2012.0    5.0   \n",
              "57             [Action, Adventure, Indie]   3.99  47.290000  2016.0   10.0   \n",
              "\n",
              "    day_of_week  early_access  true_sentiment  predicted_sentiment  \\\n",
              "84          3.0         False               9                    9   \n",
              "99          1.0         False               8                    8   \n",
              "44          4.0         False               8                    8   \n",
              "75          1.0         False               8                    8   \n",
              "26          0.0         False               8                    8   \n",
              "..          ...           ...             ...                  ...   \n",
              "1           2.0         False               5                    4   \n",
              "61          1.0         False               3                    4   \n",
              "79          1.0         False               3                    3   \n",
              "63          4.0         False               3                    3   \n",
              "57          2.0         False               3                    3   \n",
              "\n",
              "             sentiment_name  \n",
              "84  Overwhelmingly Positive  \n",
              "99            Very Positive  \n",
              "44            Very Positive  \n",
              "75            Very Positive  \n",
              "26            Very Positive  \n",
              "..                      ...  \n",
              "1                  Negative  \n",
              "61                 Negative  \n",
              "79          Mostly Negative  \n",
              "63          Mostly Negative  \n",
              "57          Mostly Negative  \n",
              "\n",
              "[100 rows x 15 columns]"
            ]
          },
          "execution_count": 67,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "import math\n",
        "\n",
        "# Adjusted function to preprocess new data for prediction\n",
        "def preprocess_input(\n",
        "    new_data, label_encoders, multi_label_binarizers, numerical_scaler\n",
        "):\n",
        "\n",
        "    # Encoding single-label categorical features\n",
        "    encoded_categorical = []\n",
        "    for col, le in label_encoders.items():\n",
        "        # Transform and reshape each categorical feature to (1, 1)\n",
        "        encoded_value = le.transform([new_data[col]]).reshape(1, 1)\n",
        "        encoded_categorical.append(encoded_value)\n",
        "\n",
        "    # Binarizing multi-label features\n",
        "    encoded_multi = []\n",
        "    for col, mlb in multi_label_binarizers.items():\n",
        "        binarized = mlb.transform([new_data[col]]).reshape(1, -1)\n",
        "        encoded_multi.append(binarized)\n",
        "\n",
        "    # Normalizing numerical features\n",
        "    numerical_df = pd.DataFrame([new_data], columns=numerical_features)\n",
        "    scaled_numerical = numerical_scaler.transform(numerical_df).reshape(1, -1)\n",
        "\n",
        "    return encoded_categorical, np.concatenate(encoded_multi, axis=1), scaled_numerical\n",
        "\n",
        "\n",
        "reverse_sentiment_mapping = {\n",
        "    1: \"Overwhelmingly Negative\",\n",
        "    2: \"Very Negative\",\n",
        "    3: \"Mostly Negative\",\n",
        "    4: \"Negative\",\n",
        "    5: \"Mixed\",\n",
        "    6: \"Mostly Positive\",\n",
        "    7: \"Positive\",\n",
        "    8: \"Very Positive\",\n",
        "    9: \"Overwhelmingly Positive\",\n",
        "}\n",
        "\n",
        "true_labels = []\n",
        "games_info = []\n",
        "predicted_classes = []\n",
        "predicted_labels = []\n",
        "\n",
        "for _, row in df_test.iterrows():\n",
        "    new_game = {\n",
        "        \"publisher\": row[\"publisher\"],\n",
        "        \"developer\": row[\"developer\"],\n",
        "        \"app_name\": row[\"app_name\"],\n",
        "        \"title\": row[\"title\"],\n",
        "        \"id\": row[\"id\"],\n",
        "        \"specs\": row[\"specs\"],\n",
        "        \"tags\": row[\"tags\"],\n",
        "        \"genres\": row[\"genres\"],\n",
        "        \"price\": row[\"price\"],\n",
        "        \"metascore\": row[\"metascore\"],\n",
        "        \"year\": row[\"year\"],\n",
        "        \"month\": row[\"month\"],\n",
        "        \"day_of_week\": row[\"day_of_week\"],\n",
        "        \"early_access\": row[\"early_access\"],\n",
        "    }\n",
        "\n",
        "    true_labels.append(row[\"sentiment\"])\n",
        "    games_info.append(new_game)\n",
        "\n",
        "    encoded_categorical, encoded_multi, scaled_numerical = preprocess_input(\n",
        "        new_game, label_encoders, multi_label_binarizers, scaler\n",
        "    )\n",
        "\n",
        "    inputs = encoded_categorical + [encoded_multi] + [scaled_numerical]\n",
        "\n",
        "    predictions = model.predict(inputs, verbose=0)\n",
        "    predicted_class = math.floor(predictions[0][0])\n",
        "    predicted_class = min(predicted_class, 9)\n",
        "    predicted_class = max(predicted_class, 1)\n",
        "    predicted_classes.append(predicted_class)\n",
        "    predicted_label = reverse_sentiment_mapping[predicted_class]\n",
        "    predicted_labels.append(predicted_label)\n",
        "\n",
        "pd.concat(\n",
        "    [\n",
        "        pd.DataFrame(games_info).drop([\"title\", \"id\"], axis=1),\n",
        "        pd.Series(true_labels, name=\"true_sentiment\"),\n",
        "        pd.Series(predicted_classes, name=\"predicted_sentiment\"),\n",
        "        pd.Series(predicted_labels, name=\"sentiment_name\"),\n",
        "    ],\n",
        "    axis=1,\n",
        ").sort_values(by='predicted_sentiment', ascending=False)"
      ]
    }
  ],
  "metadata": {
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "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": "3.8.17"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}