ADAMS_NLP_SS20_AFBothe.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "ADAMS-NLP-SS20-AFBothe.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "2SgVg7SVynW1",
        "qqa4fMcB5JKT",
        "38UCz9885Trp",
        "_d1yCdXiDfNT",
        "drqNlQ-xHfF6",
        "u2xxU01oFrQa",
        "GVnYGDgSkdko",
        "OUnBgMMlBDhg",
        "WoNWaNBwA6Ct",
        "c2A43Kv4kmsN",
        "IMBlYVuUi_Aw"
      ],
      "toc_visible": true,
      "authorship_tag": "ABX9TyPYTcoB4HILPoBFQoc4PpEs",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/AnFrBo/clap_predictions/blob/main/ADAMS_NLP_SS20_AFBothe.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "__inXi-9oHsk"
      },
      "source": [
        "#Assignment ADAMS SS 2020\n",
        "\n",
        "by Anna Franziska Bothe (576309)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MxogqT7x1Nha"
      },
      "source": [
        "##Introduction\n",
        "\n",
        "The underlying data are retrieved from the blog Medium and contain variables such as the text, author, published date, comments and the total number of claps (equivalent to a like) of many posts. The task is to predict the number of claps on a given data set. Since we have rich textual information like the header, texts and several tags (tag_name, slug, name), we will mainly focus on natural language processing (NLP). Nevertheless, other categorical and numerical features will be implemented. The first part of the assignment consists of setting up the environment and exploring the data, followed by data preparation and text cleaning. Moreover, we will have a look at different methods and constructions of word embedding and compare the performance of \"self\"-pretrained word embeddings to pretrained word embeddings. Least but not least, the best performing word embedding will be apply to different neural network architecture which will be compared to a baseline models. The model with the lowest MSE will be used for the final prediction.<br>\n",
        "To give an overview over all chapters and subchapters, Google Colab as well as Jupyter Notebook provide an excellent, responsive table of contents located on the top of the bar on the left."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xEtPDqmR51an"
      },
      "source": [
        "###Set-up System "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KUHyESHuoNfm",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "outputId": "10127b57-814f-48a4-cfcd-2ccb9fd18ab1"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount(\"/content/gdrive\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/gdrive; to attempt to forcibly remount, call drive.mount(\"/content/gdrive\", force_remount=True).\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4YC9gQK8zqjH",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 153
        },
        "outputId": "7243cc86-1b69-43a1-bdcf-3f60636c6ec4"
      },
      "source": [
        "#load packages/libraries\n",
        "import pandas as pd\n",
        "import numpy as np\n",
        "import seaborn as sns\n",
        "import time\n",
        "import os\n",
        "import pickle\n",
        "\n",
        "#plots\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline  \n",
        "plt.rcParams[\"figure.figsize\"] = (18,5)\n",
        "\n",
        "#text cleaning\n",
        "import nltk\n",
        "from nltk.tokenize import word_tokenize\n",
        "from nltk.corpus import stopwords\n",
        "from nltk.stem import WordNetLemmatizer\n",
        "nltk.download(\"punkt\")\n",
        "nltk.download(\"stopwords\")\n",
        "nltk.download(\"wordnet\")\n",
        "import re\n",
        "from bs4 import BeautifulSoup\n",
        "!pip install unidecode\n",
        "import unidecode\n",
        "import spacy #lemmatize and stopwords\n",
        "from nltk.tokenize.treebank import TreebankWordDetokenizer\n",
        "\n",
        "#sentiment analysis\n",
        "!pip install afinn\n",
        "from afinn import Afinn\n",
        "\n",
        "#embeddings/NN\n",
        "from sklearn.model_selection import train_test_split\n",
        "import collections\n",
        "import gensim\n",
        "from gensim.models import Word2Vec \n",
        "from gensim.models import Phrases\n",
        "from gensim import models\n",
        "from gensim.models.phrases import Phraser\n",
        "from gensim.models import KeyedVectors\n",
        "import unicodedata\n",
        "\n",
        "from sklearn.linear_model import ElasticNet\n",
        "from sklearn.model_selection import GridSearchCV\n",
        "from sklearn.metrics import mean_squared_error\n",
        "from sklearn.model_selection import train_test_split\n",
        "from keras.preprocessing.text import Tokenizer\n",
        "from keras.preprocessing.sequence import pad_sequences\n",
        "from sklearn.preprocessing import StandardScaler\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Dense, Embedding, GRU, Dropout, Input\n",
        "from keras.layers.embeddings import Embedding\n",
        "from keras.initializers import Constant\n",
        "from tensorflow import keras\n",
        "from keras.layers import Bidirectional\n",
        "from keras.models import load_model\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n",
            "[nltk_data] Downloading package stopwords to /root/nltk_data...\n",
            "[nltk_data]   Package stopwords is already up-to-date!\n",
            "[nltk_data] Downloading package wordnet to /root/nltk_data...\n",
            "[nltk_data]   Package wordnet is already up-to-date!\n",
            "Requirement already satisfied: unidecode in /usr/local/lib/python3.6/dist-packages (1.1.1)\n",
            "Requirement already satisfied: afinn in /usr/local/lib/python3.6/dist-packages (0.1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LjvvP-vpK4lS"
      },
      "source": [
        "###Set-up Data\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "a1KB_FubK23N"
      },
      "source": [
        "#set working directory\n",
        "os.chdir(\"/content/gdrive/My Drive/ADAMS/Assignment\") #TO DO: please change to your corresponding wd"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jZ9jQR01oaNc"
      },
      "source": [
        "#load data file\n",
        "data_file = \"Train.csv\"\n",
        "train = pd.read_csv(data_file, index_col = False, encoding = \"utf-8\")\n",
        "\n",
        "data_file = \"Test.csv\"\n",
        "test = pd.read_csv(data_file, index_col = False, encoding = \"utf-8\")\n",
        "\n",
        "#train.info() #102.9+ MB"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vevfcRXk3upM"
      },
      "source": [
        "##EDA/Descriptive Statistics"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Y0q826Kh9TyE"
      },
      "source": [
        "The training data set consists of 279.577 entries and contains 50 different variables representing the meta data retrieved from the scrapped blog posts. All datat was scrapped at November 04, 2018. The testing data set have only 8 different variables and 514 entries. The variable are all in a similar form represented in the training data set.<br>\n",
        "As the boxplot below shows, the target variable \"totalClapCount\" has a wide range and extreme outliers. We manually checked some of the extreme outliers to make sure that it is not a scrapping mistake. 50% of the blog posts have 8 claps or less and 75% have 72 claps or less. However, the maximum clap count is 291.706 claps. <br>\n",
        "In the following subchapter we will concentrate on the descriptives of the train data set. All measurements that are applied to the train set will be applied analogously to the test set before the final prediction (if necessary)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sYy2RMKNlx8h",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 534
        },
        "outputId": "8a8c8b62-867e-4435-b860-b129a34919ef"
      },
      "source": [
        "train.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>audioVersionDurationSec</th>\n",
              "      <th>codeBlock</th>\n",
              "      <th>codeBlockCount</th>\n",
              "      <th>collectionId</th>\n",
              "      <th>createdDate</th>\n",
              "      <th>createdDatetime</th>\n",
              "      <th>firstPublishedDate</th>\n",
              "      <th>firstPublishedDatetime</th>\n",
              "      <th>imageCount</th>\n",
              "      <th>isSubscriptionLocked</th>\n",
              "      <th>language</th>\n",
              "      <th>latestPublishedDate</th>\n",
              "      <th>latestPublishedDatetime</th>\n",
              "      <th>linksCount</th>\n",
              "      <th>postId</th>\n",
              "      <th>readingTime</th>\n",
              "      <th>recommends</th>\n",
              "      <th>responsesCreatedCount</th>\n",
              "      <th>socialRecommendsCount</th>\n",
              "      <th>subTitle</th>\n",
              "      <th>tagsCount</th>\n",
              "      <th>text</th>\n",
              "      <th>title</th>\n",
              "      <th>totalClapCount</th>\n",
              "      <th>uniqueSlug</th>\n",
              "      <th>updatedDate</th>\n",
              "      <th>updatedDatetime</th>\n",
              "      <th>url</th>\n",
              "      <th>vote</th>\n",
              "      <th>wordCount</th>\n",
              "      <th>publicationdescription</th>\n",
              "      <th>publicationdomain</th>\n",
              "      <th>publicationfacebookPageName</th>\n",
              "      <th>publicationfollowerCount</th>\n",
              "      <th>publicationname</th>\n",
              "      <th>publicationpublicEmail</th>\n",
              "      <th>publicationslug</th>\n",
              "      <th>publicationtags</th>\n",
              "      <th>publicationtwitterUsername</th>\n",
              "      <th>tag_name</th>\n",
              "      <th>slug</th>\n",
              "      <th>name</th>\n",
              "      <th>postCount</th>\n",
              "      <th>author</th>\n",
              "      <th>bio</th>\n",
              "      <th>userId</th>\n",
              "      <th>userName</th>\n",
              "      <th>usersFollowedByCount</th>\n",
              "      <th>usersFollowedCount</th>\n",
              "      <th>scrappedDate</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>638f418c8464</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:55:34</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:57:03</td>\n",
              "      <td>1</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:58:20</td>\n",
              "      <td>1</td>\n",
              "      <td>10007d3018fe</td>\n",
              "      <td>0.958491</td>\n",
              "      <td>2</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>A major private IT company implements blockcha...</td>\n",
              "      <td>5</td>\n",
              "      <td>Private Business, Government and Blockchain\\n\\...</td>\n",
              "      <td>Private Business, Government and Blockchain</td>\n",
              "      <td>100</td>\n",
              "      <td>private-business-government-and-blockchain-100...</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:58:20</td>\n",
              "      <td>https://medium.com/s/story/private-business-go...</td>\n",
              "      <td>False</td>\n",
              "      <td>201</td>\n",
              "      <td>ICOBox is the first and the biggest new genera...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>icobox.io</td>\n",
              "      <td>NaN</td>\n",
              "      <td>ICOBox</td>\n",
              "      <td>support@icobox.io</td>\n",
              "      <td>icobox-io</td>\n",
              "      <td>BLOCKCHAIN,ICO,ETHEREUM,ETHEREUM BLOCKCHAIN,TO...</td>\n",
              "      <td>icobox_io</td>\n",
              "      <td>Blockchain</td>\n",
              "      <td>blockchain</td>\n",
              "      <td>Blockchain</td>\n",
              "      <td>265164.0</td>\n",
              "      <td>Anar Babaev</td>\n",
              "      <td>NaN</td>\n",
              "      <td>f1ad85af0169</td>\n",
              "      <td>babaevanar</td>\n",
              "      <td>450.0</td>\n",
              "      <td>404.0</td>\n",
              "      <td>20181104</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>638f418c8464</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:55:34</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:57:03</td>\n",
              "      <td>1</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:58:20</td>\n",
              "      <td>1</td>\n",
              "      <td>10007d3018fe</td>\n",
              "      <td>0.958491</td>\n",
              "      <td>2</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>A major private IT company implements blockcha...</td>\n",
              "      <td>5</td>\n",
              "      <td>Private Business, Government and Blockchain\\n\\...</td>\n",
              "      <td>Private Business, Government and Blockchain</td>\n",
              "      <td>100</td>\n",
              "      <td>private-business-government-and-blockchain-100...</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:58:20</td>\n",
              "      <td>https://medium.com/s/story/private-business-go...</td>\n",
              "      <td>False</td>\n",
              "      <td>201</td>\n",
              "      <td>ICOBox is the first and the biggest new genera...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>icobox.io</td>\n",
              "      <td>NaN</td>\n",
              "      <td>ICOBox</td>\n",
              "      <td>support@icobox.io</td>\n",
              "      <td>icobox-io</td>\n",
              "      <td>BLOCKCHAIN,ICO,ETHEREUM,ETHEREUM BLOCKCHAIN,TO...</td>\n",
              "      <td>icobox_io</td>\n",
              "      <td>Samsung</td>\n",
              "      <td>samsung</td>\n",
              "      <td>Samsung</td>\n",
              "      <td>5708.0</td>\n",
              "      <td>Anar Babaev</td>\n",
              "      <td>NaN</td>\n",
              "      <td>f1ad85af0169</td>\n",
              "      <td>babaevanar</td>\n",
              "      <td>450.0</td>\n",
              "      <td>404.0</td>\n",
              "      <td>20181104</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>638f418c8464</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:55:34</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:57:03</td>\n",
              "      <td>1</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:58:20</td>\n",
              "      <td>1</td>\n",
              "      <td>10007d3018fe</td>\n",
              "      <td>0.958491</td>\n",
              "      <td>2</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>A major private IT company implements blockcha...</td>\n",
              "      <td>5</td>\n",
              "      <td>Private Business, Government and Blockchain\\n\\...</td>\n",
              "      <td>Private Business, Government and Blockchain</td>\n",
              "      <td>100</td>\n",
              "      <td>private-business-government-and-blockchain-100...</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:58:20</td>\n",
              "      <td>https://medium.com/s/story/private-business-go...</td>\n",
              "      <td>False</td>\n",
              "      <td>201</td>\n",
              "      <td>ICOBox is the first and the biggest new genera...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>icobox.io</td>\n",
              "      <td>NaN</td>\n",
              "      <td>ICOBox</td>\n",
              "      <td>support@icobox.io</td>\n",
              "      <td>icobox-io</td>\n",
              "      <td>BLOCKCHAIN,ICO,ETHEREUM,ETHEREUM BLOCKCHAIN,TO...</td>\n",
              "      <td>icobox_io</td>\n",
              "      <td>It</td>\n",
              "      <td>it</td>\n",
              "      <td>It</td>\n",
              "      <td>3720.0</td>\n",
              "      <td>Anar Babaev</td>\n",
              "      <td>NaN</td>\n",
              "      <td>f1ad85af0169</td>\n",
              "      <td>babaevanar</td>\n",
              "      <td>450.0</td>\n",
              "      <td>404.0</td>\n",
              "      <td>20181104</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:04:37</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:06:29</td>\n",
              "      <td>13</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:18:38</td>\n",
              "      <td>24</td>\n",
              "      <td>1000c43bcb97</td>\n",
              "      <td>19.716981</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>Introduction</td>\n",
              "      <td>5</td>\n",
              "      <td>EPQ draft 1 (4844 words)\\nhttps://upload.wikim...</td>\n",
              "      <td>EPQ draft 1 (4844 words)</td>\n",
              "      <td>0</td>\n",
              "      <td>introduction-3-1000c43bcb97</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:18:39</td>\n",
              "      <td>https://medium.com/s/story/introduction-3-1000...</td>\n",
              "      <td>False</td>\n",
              "      <td>4854</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>Technology</td>\n",
              "      <td>technology</td>\n",
              "      <td>Technology</td>\n",
              "      <td>166125.0</td>\n",
              "      <td>George Sykes</td>\n",
              "      <td>NaN</td>\n",
              "      <td>93b9e94f08ca</td>\n",
              "      <td>tasty231</td>\n",
              "      <td>6.0</td>\n",
              "      <td>22.0</td>\n",
              "      <td>20181104</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:04:37</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:06:29</td>\n",
              "      <td>13</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:18:38</td>\n",
              "      <td>24</td>\n",
              "      <td>1000c43bcb97</td>\n",
              "      <td>19.716981</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>Introduction</td>\n",
              "      <td>5</td>\n",
              "      <td>EPQ draft 1 (4844 words)\\nhttps://upload.wikim...</td>\n",
              "      <td>EPQ draft 1 (4844 words)</td>\n",
              "      <td>0</td>\n",
              "      <td>introduction-3-1000c43bcb97</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:18:39</td>\n",
              "      <td>https://medium.com/s/story/introduction-3-1000...</td>\n",
              "      <td>False</td>\n",
              "      <td>4854</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>Robotics</td>\n",
              "      <td>robotics</td>\n",
              "      <td>Robotics</td>\n",
              "      <td>9103.0</td>\n",
              "      <td>George Sykes</td>\n",
              "      <td>NaN</td>\n",
              "      <td>93b9e94f08ca</td>\n",
              "      <td>tasty231</td>\n",
              "      <td>6.0</td>\n",
              "      <td>22.0</td>\n",
              "      <td>20181104</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "   audioVersionDurationSec codeBlock  ...  usersFollowedCount scrappedDate\n",
              "0                        0       NaN  ...               404.0     20181104\n",
              "1                        0       NaN  ...               404.0     20181104\n",
              "2                        0       NaN  ...               404.0     20181104\n",
              "3                        0       NaN  ...                22.0     20181104\n",
              "4                        0       NaN  ...                22.0     20181104\n",
              "\n",
              "[5 rows x 50 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 172
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N1H9pfAdn-6Q",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 167
        },
        "outputId": "a1f9a8b9-394a-4b61-e47b-d2ce610a4ca4"
      },
      "source": [
        "print(train[\"totalClapCount\"].describe())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "count    279577.000000\n",
            "mean        184.963473\n",
            "std        1985.353004\n",
            "min           0.000000\n",
            "25%           0.000000\n",
            "50%           8.000000\n",
            "75%          72.000000\n",
            "max      291706.000000\n",
            "Name: totalClapCount, dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LcqNpIjxwX-1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 486
        },
        "outputId": "b2d69729-b5dc-493e-c25a-086487f0708d"
      },
      "source": [
        "print(train[\"totalClapCount\"].describe())\n",
        "print(sns.boxplot(train[\"totalClapCount\"]))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "count    279577.000000\n",
            "mean        184.932983\n",
            "std        1985.685059\n",
            "min           0.000000\n",
            "25%           0.000000\n",
            "50%           8.000000\n",
            "75%          72.000000\n",
            "max      291706.000000\n",
            "Name: totalClapCount, dtype: float64\n",
            "AxesSubplot(0.125,0.125;0.775x0.755)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1296x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7ZqgtUNuMd_C"
      },
      "source": [
        "##Data Preparation and Cleaning"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JIiW39cHqiAy"
      },
      "source": [
        "###Transformation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-kbRYXwDqmdn"
      },
      "source": [
        "All data types will be transformed to more reasonable dtype with less memory consumption and for better and more convenient processing the further analysis."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "y-ELLF1SpnzV",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 886
        },
        "outputId": "c8db5023-671e-4495-afe5-0bd93c07d71b"
      },
      "source": [
        "#transform all int64 and float64 to float32\n",
        "num_vars = train.select_dtypes(include=[np.float64, np.int64 ]).columns\n",
        "train[num_vars] = train[num_vars].astype(np.float32)\n",
        "\n",
        "#transform all objects to categories\n",
        "object_vars = train.select_dtypes(exclude=np.float32).columns\n",
        "train[object_vars] = train[object_vars].astype(\"category\")\n",
        "\n",
        "train.info() #80.3 MB"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "<class 'pandas.core.frame.DataFrame'>\n",
            "RangeIndex: 279577 entries, 0 to 279576\n",
            "Data columns (total 45 columns):\n",
            " #   Column                       Non-Null Count   Dtype   \n",
            "---  ------                       --------------   -----   \n",
            " 0   audioVersionDurationSec      279577 non-null  float32 \n",
            " 1   codeBlock                    25179 non-null   category\n",
            " 2   codeBlockCount               279577 non-null  float32 \n",
            " 3   collectionId                 137878 non-null  category\n",
            " 4   createdDate                  279577 non-null  category\n",
            " 5   createdDatetime              279577 non-null  category\n",
            " 6   firstPublishedDate           279577 non-null  category\n",
            " 7   firstPublishedDatetime       279577 non-null  category\n",
            " 8   imageCount                   279577 non-null  float32 \n",
            " 9   isSubscriptionLocked         279577 non-null  category\n",
            " 10  language                     279577 non-null  category\n",
            " 11  latestPublishedDate          279577 non-null  category\n",
            " 12  latestPublishedDatetime      279577 non-null  category\n",
            " 13  linksCount                   279577 non-null  float32 \n",
            " 14  postId                       279577 non-null  category\n",
            " 15  readingTime                  279577 non-null  float32 \n",
            " 16  recommends                   279577 non-null  float32 \n",
            " 17  responsesCreatedCount        279577 non-null  float32 \n",
            " 18  socialRecommendsCount        279577 non-null  float32 \n",
            " 19  subTitle                     271217 non-null  category\n",
            " 20  tagsCount                    279577 non-null  float32 \n",
            " 21  text                         279577 non-null  category\n",
            " 22  title                        279572 non-null  category\n",
            " 23  totalClapCount               279577 non-null  float32 \n",
            " 24  uniqueSlug                   279577 non-null  category\n",
            " 25  updatedDate                  279577 non-null  category\n",
            " 26  updatedDatetime              279577 non-null  category\n",
            " 27  url                          279577 non-null  category\n",
            " 28  vote                         279577 non-null  category\n",
            " 29  wordCount                    279577 non-null  float32 \n",
            " 30  publicationdescription       137231 non-null  category\n",
            " 31  publicationdomain            53972 non-null   category\n",
            " 32  publicationfacebookPageName  100874 non-null  category\n",
            " 33  publicationname              137231 non-null  category\n",
            " 34  publicationpublicEmail       101982 non-null  category\n",
            " 35  publicationslug              137231 non-null  category\n",
            " 36  publicationtags              130298 non-null  category\n",
            " 37  publicationtwitterUsername   119851 non-null  category\n",
            " 38  author                       279577 non-null  category\n",
            " 39  bio                          225480 non-null  category\n",
            " 40  userId                       279577 non-null  category\n",
            " 41  userName                     279577 non-null  category\n",
            " 42  usersFollowedByCount         279577 non-null  float32 \n",
            " 43  usersFollowedCount           279577 non-null  float32 \n",
            " 44  scrappedDate                 279577 non-null  float32 \n",
            "dtypes: category(31), float32(14)\n",
            "memory usage: 74.3 MB\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "d1NTUGE5GwxY"
      },
      "source": [
        "###Missing Values"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kKe1wdch-FIc"
      },
      "source": [
        "The plot below shows that the data set does not contain many variables with missing data but if they have missing entries than mostly a lot. 10 of 13 of the variable with missing variables have more than 50% NAs. The highest number of missing values -- besides the variable publication follower count with 100% NAs -- is 254398 (approx. 91 %). The corresponding variable is called codeBlock. In contrast, the variable title has the lowest number of missing values with 5 (approx. 0.002 %) missing entries. <br>\n",
        "\n",
        "Since all missing values miss textual data it is not possible to replace them with numbers such as median, mode or mean. For textual variables the three possibilities are at hand: attempt to scrap the missing data (not allowed for this assignment), replacing them manually or to dummy code them as \"available - yes, no (or unknown)\". The second approach is only feasible with a low number of missing variables such as title in our case. Also the information has to be available in general. The last approach suits variables such as social media account. They might not be missing entries but rather non-existent.<br>\n",
        "\n",
        "Spoiler: Out of all variables with missing data only the variable text will be used. The missing value will be imputed by hand.\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oITBs4M5yZ3B",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 430
        },
        "outputId": "4f3c7e2f-278f-4cf7-8049-2911ac4f9946"
      },
      "source": [
        "#visualisation of missing values\n",
        "\n",
        "sns.heatmap(train.isnull(), cbar = False); #\";\" gets rid of the message \"<matplotlib.axes._subplots.AxesSubplot at 0x7fe969b4bcf8>\""
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1296x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hdXPsaV-5H5k",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 251
        },
        "outputId": "8afae12e-ef93-4de7-d6e7-020bd66dabc1"
      },
      "source": [
        "for z in train:\n",
        "  if train.loc[:,z].isnull().sum() > 0:\n",
        "    print('Missing values of {}: {} ({:.3f}%)'.format(z, train.loc[:,z].isnull().sum(), 100*train.loc[:,z].isnull().sum()/train.shape[0]))\n",
        "    #print(z, train.loc[:,z].isnull().sum())  "
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Missing values of codeBlock: 254398 (90.994%)\n",
            "Missing values of collectionId: 141699 (50.683%)\n",
            "Missing values of subTitle: 8360 (2.990%)\n",
            "Missing values of title: 5 (0.002%)\n",
            "Missing values of publicationdescription: 142346 (50.915%)\n",
            "Missing values of publicationdomain: 225605 (80.695%)\n",
            "Missing values of publicationfacebookPageName: 178703 (63.919%)\n",
            "Missing values of publicationfollowerCount: 279577 (100.000%)\n",
            "Missing values of publicationname: 142346 (50.915%)\n",
            "Missing values of publicationpublicEmail: 177595 (63.523%)\n",
            "Missing values of publicationslug: 142346 (50.915%)\n",
            "Missing values of publicationtags: 149279 (53.395%)\n",
            "Missing values of publicationtwitterUsername: 159726 (57.131%)\n",
            "Missing values of bio: 54097 (19.350%)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Tm_Z9DvP4B4T"
      },
      "source": [
        "###Removing Duplicates"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jcPFheu0G4r2"
      },
      "source": [
        "As shown in the table below no variable has more than 72.337 unique items. Not even variables like postID, uniqueSlug or url which are unique by nature. This proves that the training data set contains some duplicates that have to be removed. The challenge is that the rows are not fully equal because they contain different keywords (see variables tag_name, slug and name) with corresponding postCounts. Especially the keywords provide valuable information since they sum up the main content of a text. In order to prevent the train set from loosing them, the keywords are merged and the postCounts are summed up. Afterwards, the duplicate rows are removed."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BSDrAsZzBedq",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 870
        },
        "outputId": "7ef2c7b1-b701-4dbb-c551-1e6f298ca03a"
      },
      "source": [
        "#unique value per col\n",
        "train.astype('object').describe(include='all').loc['unique', :]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "audioVersionDurationSec           51\n",
              "codeBlock                       6230\n",
              "codeBlockCount                   129\n",
              "collectionId                    6527\n",
              "createdDate                      741\n",
              "createdDatetime                72209\n",
              "firstPublishedDate               610\n",
              "firstPublishedDatetime         71406\n",
              "imageCount                        89\n",
              "isSubscriptionLocked               2\n",
              "language                          52\n",
              "latestPublishedDate              577\n",
              "latestPublishedDatetime        72158\n",
              "linksCount                       162\n",
              "postId                         72337\n",
              "readingTime                    16433\n",
              "recommends                       887\n",
              "responsesCreatedCount            139\n",
              "socialRecommendsCount              4\n",
              "subTitle                       65673\n",
              "tagsCount                          6\n",
              "text                           72024\n",
              "title                          70275\n",
              "totalClapCount                  2321\n",
              "uniqueSlug                     72337\n",
              "updatedDate                      431\n",
              "updatedDatetime                71827\n",
              "url                            72337\n",
              "vote                               2\n",
              "wordCount                       4193\n",
              "publicationdescription          6506\n",
              "publicationdomain               1062\n",
              "publicationfacebookPageName     3836\n",
              "publicationfollowerCount           0\n",
              "publicationname                 6507\n",
              "publicationpublicEmail          5035\n",
              "publicationslug                 6512\n",
              "publicationtags                 5697\n",
              "publicationtwitterUsername      4861\n",
              "tag_name                       19278\n",
              "slug                           19278\n",
              "name                           19278\n",
              "postCount                       2945\n",
              "author                         31021\n",
              "bio                            22352\n",
              "userId                         31317\n",
              "userName                       31317\n",
              "usersFollowedByCount            2569\n",
              "usersFollowedCount              1584\n",
              "scrappedDate                       1\n",
              "Name: unique, dtype: object"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 260
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rOD5YRamVw-e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 213
        },
        "outputId": "e7d6f05c-60a4-4040-da7f-468b9c0c77c3"
      },
      "source": [
        "#sum up postCounts (pC) per article before removing duplicates of train data set; otherwise, information gets lost\n",
        "pC = train.groupby([\"postId\"])[\"postCount\"].apply(sum).reset_index()\n",
        "#validation check: the difference between train.postCounts and pC.postCount should be 0\n",
        "print(pC.postCount.sum()-train.postCount.sum())\n",
        "pC.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "762.0\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>postId</th>\n",
              "      <th>postCount</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>10007d3018fe</td>\n",
              "      <td>274592.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1000b3d0d2</td>\n",
              "      <td>475794.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1000c43bcb97</td>\n",
              "      <td>184243.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>100139913e4c</td>\n",
              "      <td>33657.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>1002a55eca89</td>\n",
              "      <td>108641.0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "         postId  postCount\n",
              "0  10007d3018fe   274592.0\n",
              "1    1000b3d0d2   475794.0\n",
              "2  1000c43bcb97   184243.0\n",
              "3  100139913e4c    33657.0\n",
              "4  1002a55eca89   108641.0"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 261
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JFm-_01PNDGm",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 196
        },
        "outputId": "bf0b2591-c4ee-4284-e78a-ecbacd7d6a57"
      },
      "source": [
        "#merging different names \"tags\" (equal to slug and tag_name) seperated by comma before removing the duplicate entries from data set\n",
        "tags = train.groupby([\"postId\"])[\"name\"].apply(', '.join).reset_index()\n",
        "tags.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>postId</th>\n",
              "      <th>name</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>10007d3018fe</td>\n",
              "      <td>Blockchain, Samsung, It</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1000b3d0d2</td>\n",
              "      <td>Blockchain, Machine Learning, Synapse, Cryptoc...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1000c43bcb97</td>\n",
              "      <td>Technology, Robotics, Automation, Epq</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>100139913e4c</td>\n",
              "      <td>Data Science, Sas Training In Pune, Hadoop Tra...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>1002a55eca89</td>\n",
              "      <td>Robotics, Meditation, Therapy, Artificial Inte...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "         postId                                               name\n",
              "0  10007d3018fe                            Blockchain, Samsung, It\n",
              "1    1000b3d0d2  Blockchain, Machine Learning, Synapse, Cryptoc...\n",
              "2  1000c43bcb97              Technology, Robotics, Automation, Epq\n",
              "3  100139913e4c  Data Science, Sas Training In Pune, Hadoop Tra...\n",
              "4  1002a55eca89  Robotics, Meditation, Therapy, Artificial Inte..."
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 262
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iBmV_Ox8dSu7"
      },
      "source": [
        "#deleted columns that have already been processed or that are constants\n",
        "\n",
        "#already used\n",
        "del train[\"tag_name\"]\n",
        "del train[\"slug\"]\n",
        "del train[\"name\"] \n",
        "del train[\"postCount\"]\n",
        "\n",
        "#variable is always null\n",
        "del train[\"publicationfollowerCount\"] \n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gF-TFfVGSIem",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 584
        },
        "outputId": "c46565f4-2925-4677-b5d0-2288a1b494ed"
      },
      "source": [
        "#merge all\n",
        "train_new = pd.merge(pC, train, on='postId', how='left')\n",
        "train_new = pd.merge(train_new, tags, on='postId', how='left').drop_duplicates()\n",
        "train_new.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>postId</th>\n",
              "      <th>postCount</th>\n",
              "      <th>audioVersionDurationSec</th>\n",
              "      <th>codeBlock</th>\n",
              "      <th>codeBlockCount</th>\n",
              "      <th>collectionId</th>\n",
              "      <th>createdDate</th>\n",
              "      <th>createdDatetime</th>\n",
              "      <th>firstPublishedDate</th>\n",
              "      <th>firstPublishedDatetime</th>\n",
              "      <th>imageCount</th>\n",
              "      <th>isSubscriptionLocked</th>\n",
              "      <th>language</th>\n",
              "      <th>latestPublishedDate</th>\n",
              "      <th>latestPublishedDatetime</th>\n",
              "      <th>linksCount</th>\n",
              "      <th>readingTime</th>\n",
              "      <th>recommends</th>\n",
              "      <th>responsesCreatedCount</th>\n",
              "      <th>socialRecommendsCount</th>\n",
              "      <th>subTitle</th>\n",
              "      <th>tagsCount</th>\n",
              "      <th>text</th>\n",
              "      <th>title</th>\n",
              "      <th>totalClapCount</th>\n",
              "      <th>uniqueSlug</th>\n",
              "      <th>updatedDate</th>\n",
              "      <th>updatedDatetime</th>\n",
              "      <th>url</th>\n",
              "      <th>vote</th>\n",
              "      <th>wordCount</th>\n",
              "      <th>publicationdescription</th>\n",
              "      <th>publicationdomain</th>\n",
              "      <th>publicationfacebookPageName</th>\n",
              "      <th>publicationname</th>\n",
              "      <th>publicationpublicEmail</th>\n",
              "      <th>publicationslug</th>\n",
              "      <th>publicationtags</th>\n",
              "      <th>publicationtwitterUsername</th>\n",
              "      <th>author</th>\n",
              "      <th>bio</th>\n",
              "      <th>userId</th>\n",
              "      <th>userName</th>\n",
              "      <th>usersFollowedByCount</th>\n",
              "      <th>usersFollowedCount</th>\n",
              "      <th>scrappedDate</th>\n",
              "      <th>name</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>10007d3018fe</td>\n",
              "      <td>274592.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>638f418c8464</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:55:34</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:57:03</td>\n",
              "      <td>1.0</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:58:20</td>\n",
              "      <td>1.0</td>\n",
              "      <td>0.958491</td>\n",
              "      <td>2.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>A major private IT company implements blockcha...</td>\n",
              "      <td>5.0</td>\n",
              "      <td>Private Business, Government and Blockchain\\n\\...</td>\n",
              "      <td>Private Business, Government and Blockchain</td>\n",
              "      <td>100.0</td>\n",
              "      <td>private-business-government-and-blockchain-100...</td>\n",
              "      <td>2018-09-18</td>\n",
              "      <td>2018-09-18 20:58:20</td>\n",
              "      <td>https://medium.com/s/story/private-business-go...</td>\n",
              "      <td>False</td>\n",
              "      <td>201.0</td>\n",
              "      <td>ICOBox is the first and the biggest new genera...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>icobox.io</td>\n",
              "      <td>ICOBox</td>\n",
              "      <td>support@icobox.io</td>\n",
              "      <td>icobox-io</td>\n",
              "      <td>BLOCKCHAIN,ICO,ETHEREUM,ETHEREUM BLOCKCHAIN,TO...</td>\n",
              "      <td>icobox_io</td>\n",
              "      <td>Anar Babaev</td>\n",
              "      <td>NaN</td>\n",
              "      <td>f1ad85af0169</td>\n",
              "      <td>babaevanar</td>\n",
              "      <td>450.0</td>\n",
              "      <td>404.0</td>\n",
              "      <td>20181104.0</td>\n",
              "      <td>Blockchain, Samsung, It</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>1000b3d0d2</td>\n",
              "      <td>475794.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>ff54a0aaacd7</td>\n",
              "      <td>2017-12-06</td>\n",
              "      <td>2017-12-06 22:23:06</td>\n",
              "      <td>2018-02-24</td>\n",
              "      <td>2018-02-24 19:36:01</td>\n",
              "      <td>2.0</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-03-01</td>\n",
              "      <td>2018-03-01 07:44:04</td>\n",
              "      <td>7.0</td>\n",
              "      <td>18.387108</td>\n",
              "      <td>8.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>Synapse CEO Dan Gailey is interviewed by Gabe ...</td>\n",
              "      <td>4.0</td>\n",
              "      <td>Democratizing Artificial intelligence by Democ...</td>\n",
              "      <td>Democratizing Artificial intelligence by Democ...</td>\n",
              "      <td>205.0</td>\n",
              "      <td>democratizing-artificial-intelligence-by-democ...</td>\n",
              "      <td>2018-03-27</td>\n",
              "      <td>2018-03-27 20:09:42</td>\n",
              "      <td>https://blog.synapse.ai/democratizing-artifici...</td>\n",
              "      <td>False</td>\n",
              "      <td>4771.0</td>\n",
              "      <td>The Official Synapse Blog</td>\n",
              "      <td>blog.synapse.ai</td>\n",
              "      <td>syntoken</td>\n",
              "      <td>Synapse AI</td>\n",
              "      <td>hello@synapse.ai</td>\n",
              "      <td>synapseai</td>\n",
              "      <td>ETHEREUM,BLOCKCHAIN,ICO,AI,CRYPTOCURRENCY</td>\n",
              "      <td>aisynapse</td>\n",
              "      <td>Dan Gailey</td>\n",
              "      <td>NaN</td>\n",
              "      <td>112dfc533722</td>\n",
              "      <td>dangailey</td>\n",
              "      <td>913.0</td>\n",
              "      <td>508.0</td>\n",
              "      <td>20181104.0</td>\n",
              "      <td>Blockchain, Machine Learning, Synapse, Cryptoc...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7</th>\n",
              "      <td>1000c43bcb97</td>\n",
              "      <td>184243.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:04:37</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:06:29</td>\n",
              "      <td>13.0</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:18:38</td>\n",
              "      <td>24.0</td>\n",
              "      <td>19.716982</td>\n",
              "      <td>0.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>Introduction</td>\n",
              "      <td>5.0</td>\n",
              "      <td>EPQ draft 1 (4844 words)\\nhttps://upload.wikim...</td>\n",
              "      <td>EPQ draft 1 (4844 words)</td>\n",
              "      <td>0.0</td>\n",
              "      <td>introduction-3-1000c43bcb97</td>\n",
              "      <td>2018-01-07</td>\n",
              "      <td>2018-01-07 17:18:39</td>\n",
              "      <td>https://medium.com/s/story/introduction-3-1000...</td>\n",
              "      <td>False</td>\n",
              "      <td>4854.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>George Sykes</td>\n",
              "      <td>NaN</td>\n",
              "      <td>93b9e94f08ca</td>\n",
              "      <td>tasty231</td>\n",
              "      <td>6.0</td>\n",
              "      <td>22.0</td>\n",
              "      <td>20181104.0</td>\n",
              "      <td>Technology, Robotics, Automation, Epq</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>11</th>\n",
              "      <td>100139913e4c</td>\n",
              "      <td>33657.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>2018-03-08</td>\n",
              "      <td>2018-03-08 07:04:31</td>\n",
              "      <td>2018-03-08</td>\n",
              "      <td>2018-03-08 07:07:42</td>\n",
              "      <td>1.0</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-03-08</td>\n",
              "      <td>2018-03-08 07:07:42</td>\n",
              "      <td>3.0</td>\n",
              "      <td>2.211321</td>\n",
              "      <td>0.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>Various associations in the present days are o...</td>\n",
              "      <td>4.0</td>\n",
              "      <td>Ascent of data Science, SAS and Big data Analy...</td>\n",
              "      <td>Ascent of data Science, SAS and Big data Analy...</td>\n",
              "      <td>0.0</td>\n",
              "      <td>ascent-of-data-science-sas-and-big-data-analys...</td>\n",
              "      <td>2018-03-08</td>\n",
              "      <td>2018-03-08 07:07:42</td>\n",
              "      <td>https://medium.com/s/story/ascent-of-data-scie...</td>\n",
              "      <td>False</td>\n",
              "      <td>533.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>NaN</td>\n",
              "      <td>tech data</td>\n",
              "      <td>Tech data Providing Courses Like SaS Training ...</td>\n",
              "      <td>60a3bfd83742</td>\n",
              "      <td>techdatasolutions18</td>\n",
              "      <td>4.0</td>\n",
              "      <td>1.0</td>\n",
              "      <td>20181104.0</td>\n",
              "      <td>Data Science, Sas Training In Pune, Hadoop Tra...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>15</th>\n",
              "      <td>1002a55eca89</td>\n",
              "      <td>108641.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>NaN</td>\n",
              "      <td>0.0</td>\n",
              "      <td>a8fc5dd0676e</td>\n",
              "      <td>2018-04-16</td>\n",
              "      <td>2018-04-16 22:49:09</td>\n",
              "      <td>2018-04-16</td>\n",
              "      <td>2018-04-16 22:58:07</td>\n",
              "      <td>0.0</td>\n",
              "      <td>False</td>\n",
              "      <td>en</td>\n",
              "      <td>2018-04-16</td>\n",
              "      <td>2018-04-16 22:58:07</td>\n",
              "      <td>2.0</td>\n",
              "      <td>0.652830</td>\n",
              "      <td>1.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>0.0</td>\n",
              "      <td>I discussed this with Michelle Tsng on my Podc...</td>\n",
              "      <td>5.0</td>\n",
              "      <td>Can a robot love us better than another human ...</td>\n",
              "      <td>Can a robot love us better than another human ...</td>\n",
              "      <td>50.0</td>\n",
              "      <td>can-a-robot-love-us-better-than-another-human-...</td>\n",
              "      <td>2018-04-20</td>\n",
              "      <td>2018-04-20 00:47:51</td>\n",
              "      <td>https://medium.com/s/story/can-a-robot-love-us...</td>\n",
              "      <td>False</td>\n",
              "      <td>173.0</td>\n",
              "      <td>Non-obvious meditation advice from people on t...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>yogastew</td>\n",
              "      <td>Crazy Wisdom</td>\n",
              "      <td>salsop@yogastew.org</td>\n",
              "      <td>crazy-wisdom</td>\n",
              "      <td>MEDITATION,MINDFULNESS,YOGA,CREATIVITY,BUSINESS</td>\n",
              "      <td>stewartalsopIII</td>\n",
              "      <td>Stewart Alsop</td>\n",
              "      <td>NaN</td>\n",
              "      <td>d0481dc55f0e</td>\n",
              "      <td>stewartalsop</td>\n",
              "      <td>512.0</td>\n",
              "      <td>531.0</td>\n",
              "      <td>20181104.0</td>\n",
              "      <td>Robotics, Meditation, Therapy, Artificial Inte...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "          postId  ...                                               name\n",
              "0   10007d3018fe  ...                            Blockchain, Samsung, It\n",
              "3     1000b3d0d2  ...  Blockchain, Machine Learning, Synapse, Cryptoc...\n",
              "7   1000c43bcb97  ...              Technology, Robotics, Automation, Epq\n",
              "11  100139913e4c  ...  Data Science, Sas Training In Pune, Hadoop Tra...\n",
              "15  1002a55eca89  ...  Robotics, Meditation, Therapy, Artificial Inte...\n",
              "\n",
              "[5 rows x 47 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 264
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SDwSA_pCPrUv",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 67
        },
        "outputId": "c79cf573-59b5-47ba-9624-76153d0b1b6d"
      },
      "source": [
        "#validation check: in the end new data set should have the length of the tags or respectively pC \n",
        "print(len(pC))\n",
        "print(len(tags))\n",
        "print(len(train_new))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "72337\n",
            "72337\n",
            "72337\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1x_C9M4G5s8c"
      },
      "source": [
        "###Language Cleaning"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bo0T5eZaM6BK"
      },
      "source": [
        "The list below shows that 66380 of 72337 blog posts are written in English which is almost 92%. The second most used language is Spanish with only around 1% of the texts being in this language. It is a trade-off between introducing noise with rare data that is difficult to handle but also with holding back information not at random which might also introduce bias. Moreover, the applied models have no possibility to pick up patterns that are connected with non English posts. <br> We decided that the introduced noise due to different languages weights much heavier and therefore we excluded all languages except English from the further analysis. On top, it is possible to check the plausibility of results of the text after text cleaning. <br>\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iacCGHKi6N5F",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 920
        },
        "outputId": "5248c537-96b4-4bb7-8cbe-bdcaf279ae34"
      },
      "source": [
        "#language handling\n",
        "print(len(train_new))\n",
        "train_new['language'].value_counts()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "72337\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "en         66380\n",
              "es           890\n",
              "zh-Hant      880\n",
              "pt           873\n",
              "fr           551\n",
              "tr           378\n",
              "th           350\n",
              "ru           347\n",
              "ja           334\n",
              "it           218\n",
              "id           200\n",
              "ko           200\n",
              "zh           105\n",
              "de           102\n",
              "un            93\n",
              "vi            88\n",
              "nl            71\n",
              "sv            57\n",
              "pl            25\n",
              "cs            22\n",
              "bn            20\n",
              "no            15\n",
              "is            15\n",
              "da            13\n",
              "my            12\n",
              "ar            10\n",
              "uk            10\n",
              "mn             9\n",
              "el             8\n",
              "lo             7\n",
              "hi             6\n",
              "la             5\n",
              "si             5\n",
              "bg             4\n",
              "sk             4\n",
              "sr             3\n",
              "fa             3\n",
              "ka             3\n",
              "ro             3\n",
              "ca             2\n",
              "lv             2\n",
              "nn             2\n",
              "ms             2\n",
              "hu             2\n",
              "te             1\n",
              "az             1\n",
              "mk             1\n",
              "sl             1\n",
              "lt             1\n",
              "mr             1\n",
              "ml             1\n",
              "fi             1\n",
              "Name: language, dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 266
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "085mWPUpjfzC",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "726bfee1-de7e-48db-d5ad-829c0b864dec"
      },
      "source": [
        "train_new = train_new[train_new.language == \"en\"]\n",
        "train_new.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(66380, 47)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 267
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XdxCy_hRo0eO"
      },
      "source": [
        "66.380 (started with: 279.577) data points will be processed in the text cleaning. About 30% of all claps are 0."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gpM036_G59J5"
      },
      "source": [
        "###Correlation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kGa9gmcTpQDO"
      },
      "source": [
        "The upcoming chapter function as a pre-selection of non-textual variables. <br>\n",
        "The variable \"responsesCreatedCount\" in the training set is the same as the variable \"responses\" in the testing data set. The variable has no missing values and at least 75% of the variable is 0. Nevertheless, the variable has a medium correlation of about 0.58 with the target variable and\n",
        "will be used for training. The variable does not need any processing in the train set. However, in the test set the words \"responses\" and \"response\" have been removed. <br>\n",
        "Another interesting finding was the very high eta coefficient between the variable author and the target variable. An eta coefficient of 0.76 indicates a medium to strong association (depending on the source, see e.g. [2]). Thus, the variable will be integrate in the prediction task. Since it is a nominal-scaled variable, target encoding will be applied (and discussed) after train-test-splitting. This transformation is need to be able to process the data via neural networds. <br>\n",
        "Surprisingly, variables such as length of texts/word count and postCount have no strong association with the target variable. The correlations were between -0.1 and 0.1."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jWo1m1OLdM25",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 230
        },
        "outputId": "0a3dc82b-6947-417f-e1d1-47417eaf5873"
      },
      "source": [
        "print(train_new.responsesCreatedCount.describe())\n",
        "print(train_new.responsesCreatedCount.isnull().sum())\n",
        "print(np.corrcoef(train_new.responsesCreatedCount, train_new.totalClapCount))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "count    66380.000000\n",
            "mean         0.902757\n",
            "std         14.940658\n",
            "min          0.000000\n",
            "25%          0.000000\n",
            "50%          0.000000\n",
            "75%          0.000000\n",
            "max       3076.000000\n",
            "Name: responsesCreatedCount, dtype: float64\n",
            "0\n",
            "[[1.         0.57761543]\n",
            " [0.57761543 1.        ]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "r-RyEm7qnypU",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "8f5f645f-88c2-4e98-b847-9c718a4a907a"
      },
      "source": [
        "#correlation between authors and claps (eta-coefficient)\n",
        "#code source: https://towardsdatascience.com/the-search-for-categorical-correlation-a1cf7f1888c9\n",
        "\n",
        "train_new = train_new.reset_index(drop = True)\n",
        "\n",
        "def correlation_ratio(categories, measurements):\n",
        "    fcat, _ = pd.factorize(categories)\n",
        "    cat_num = np.max(fcat)+1\n",
        "    y_avg_array = np.zeros(cat_num)\n",
        "    n_array = np.zeros(cat_num)\n",
        "    for i in range(0,cat_num):\n",
        "        cat_measures = measurements[np.argwhere(fcat == i).flatten()]\n",
        "        n_array[i] = len(cat_measures)\n",
        "        y_avg_array[i] = np.average(cat_measures)\n",
        "        y_total_avg = np.sum(np.multiply(y_avg_array,n_array))/np.sum(n_array)\n",
        "        numerator = np.sum(np.multiply(n_array,np.power(np.subtract(y_avg_array,y_total_avg),2)))\n",
        "        denominator = np.sum(np.power(np.subtract(measurements,y_total_avg),2))\n",
        "        \n",
        "        if numerator == 0:\n",
        "          eta = 0.0\n",
        "        else:\n",
        "          eta = np.sqrt(numerator/denominator)\n",
        "\n",
        "    return eta\n",
        "\n",
        "correlation_ratio(train_new[\"author\"], train_new[\"totalClapCount\"])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.7575183107222517"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 270
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MDEaJoXWCGA8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 217
        },
        "outputId": "9b9df67b-252e-4cde-e9c1-86b18f592e85"
      },
      "source": [
        "#most frequent authors\n",
        "print(train_new.author.value_counts().head(10))\n",
        "train_new.author.isnull().sum()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Synced                  377\n",
            "DEEP AERO DRONES        279\n",
            "Yves Mulkers            279\n",
            "AI Hawk                 245\n",
            "Jae Duk Seo             173\n",
            "C Gavilanes             164\n",
            "WELTARE Strategies      141\n",
            "TWiML & AI              136\n",
            "Michael K. Spencer      124\n",
            "Corsair's Publishing    121\n",
            "Name: author, dtype: int64\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 20
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9gzkIANjXDYl",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "7718c569-5560-43f0-90e4-9b1aae9efc92"
      },
      "source": [
        "print(np.corrcoef(tr_cl.recommends, tr_cl.totalClapCount)) \n",
        "#very strong relationship: 0.96; no corresponding variable in test set"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[1.         0.96155606]\n",
            " [0.96155606 1.        ]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_d1yCdXiDfNT"
      },
      "source": [
        "###Text Cleaning\n",
        "\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mmivvOHwxlCH"
      },
      "source": [
        "The last step of the data preparation is the text cleaning. As stated above, the variables text, header and the newly created variable tags will be used for the NLP analysis. <br> \n",
        "\n",
        "Before applying the text cleaning, we put the focus on a few lacks in content (= texts with \\n, NaN/Na or almost no content). Some texts are very short or contain no content (such as \\n, NaN etc.); however, they have claps. I randomly checked about 50 webpages that all complied with the following reason: The reason is that the original articles contain images (mostly with texts or code) that could not be scraped. The reconstruction is not possible (in case of the non-existing webpage, e.g. error code 110 (the author deleted his Medium story)) or exceeds/fails the scope of the course (extracting text from images). Thus, the representation of the text length relatively to their number of claps is not representative here and might introduce noisy and wrong indications. Anyways, the correlation of the text length and the number of claps is with 0.09 very weak (compare [3]). All texts were kept. The \"tags\" (var: name) contain the most important keywords regarding the texts; thus, they are merge with the texts. The tags are placed at the beginning of the sentence to make sure they are not lost during the cleaning process (last step: truncating the number of tokens). Also the headers are added to the text. <br> If reconstruction was possible, we imputed some missing values manually. As an example: one text got somehow stuck with the code block during the scraping process; another text with many likes existed and we just copied it to the data frame.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UbZCDfRNrrr5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 916
        },
        "outputId": "7ceefcbe-4009-4741-a3e0-9f79506e2edb"
      },
      "source": [
        "train_new[\"text\"] = train_new[\"text\"].astype(\"object\")\n",
        "\n",
        "#correlation text length and number of claps\n",
        "train_new[\"len_text\"] = [len(x) for x in train_new.text.values]\n",
        "print(np.corrcoef(train_new.len_text, train_new.totalClapCount))\n",
        "#merge name with text (includes title already) to one column; the variable name is in the beginning because it contains the most important key words of the text\n",
        "train_new[\"text\"] = train_new.name + \" \" + train_new.text\n",
        "train_new[\"len_text\"] = [len(x) for x in train_new.text.values]\n",
        "\n",
        "train_new.loc[(train_new.len_text < 100) & (train_new.totalClapCount > 0 ), [\"postId\", \"totalClapCount\", \"text\"]]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[1.         0.09353421]\n",
            " [0.09353421 1.        ]]\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>postId</th>\n",
              "      <th>totalClapCount</th>\n",
              "      <th>text</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>1209</th>\n",
              "      <td>146c91e8eff3</td>\n",
              "      <td>1.0</td>\n",
              "      <td>Data Science Ilona’s Test\\nThis is simply a te...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6980</th>\n",
              "      <td>29b7ab1c99b0</td>\n",
              "      <td>1.0</td>\n",
              "      <td>Philosophy, Future McKenna on Super AI\\n\\n</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7598</th>\n",
              "      <td>2bf3c93d8f85</td>\n",
              "      <td>66.0</td>\n",
              "      <td>Machine Learning, Westworld Some thoughts on v...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>9878</th>\n",
              "      <td>345a3015e1a0</td>\n",
              "      <td>1.0</td>\n",
              "      <td>Social Media, Artificial Intelligence, Wellnes...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>11224</th>\n",
              "      <td>3916fd0334d8</td>\n",
              "      <td>632.0</td>\n",
              "      <td>Bottos, Blockchain, Artificial Intelligence, C...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>12685</th>\n",
              "      <td>3e691fa8543b</td>\n",
              "      <td>321.0</td>\n",
              "      <td>Burning Man, Machine Learning, Blockchain, Blo...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>13565</th>\n",
              "      <td>41b7d4739d80</td>\n",
              "      <td>10.0</td>\n",
              "      <td>Automotive, Innovation, Robotics Autonomous Ve...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>15565</th>\n",
              "      <td>48a0523224e5</td>\n",
              "      <td>1.0</td>\n",
              "      <td>Data Science, Learning, Tableau Homework #4-Ji...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>16069</th>\n",
              "      <td>4a852c39df75</td>\n",
              "      <td>60.0</td>\n",
              "      <td>Chatbots, Bots, Artificial Intelligence, Machi...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>19089</th>\n",
              "      <td>559f7443f052</td>\n",
              "      <td>2.0</td>\n",
              "      <td>Blockchain, Deepbrain DeepBrain Chain Weekly R...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>23425</th>\n",
              "      <td>653e1c6519c0</td>\n",
              "      <td>1.0</td>\n",
              "      <td>Blockchain, Technology, Artificial Intelligenc...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>24016</th>\n",
              "      <td>675119c12a5d</td>\n",
              "      <td>1.0</td>\n",
              "      <td>Data Analysis, Data Science, Data Analytics, D...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>24059</th>\n",
              "      <td>677f0bafb7c7</td>\n",
              "      <td>30.0</td>\n",
              "      <td>Artificial Intelligence, Web Development MOCK ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>25292</th>\n",
              "      <td>6bbd49b10dd6</td>\n",
              "      <td>30.0</td>\n",
              "      <td>Science Fiction, Writing, Science, Artificial ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>28280</th>\n",
              "      <td>76ac1342a614</td>\n",
              "      <td>6.0</td>\n",
              "      <td>Coding, Competition, Streaming, Rustlang Eveni...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>29136</th>\n",
              "      <td>79c62bb729c6</td>\n",
              "      <td>162.0</td>\n",
              "      <td>Bottos, Blockchain, Artificial Intelligence, C...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>39888</th>\n",
              "      <td>a0baa2ee571a</td>\n",
              "      <td>4.0</td>\n",
              "      <td>Data Science Knowledge Extraction Projects\\nRe...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>41029</th>\n",
              "      <td>a4d30a895a93</td>\n",
              "      <td>612.0</td>\n",
              "      <td>Information Retrieval, Linear Algebra, Machine...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>45570</th>\n",
              "      <td>b57ba9d4862e</td>\n",
              "      <td>129.0</td>\n",
              "      <td>Art, Crowdfunding, Startup, Investing, Artific...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>50585</th>\n",
              "      <td>c76309b17d1</td>\n",
              "      <td>1.0</td>\n",
              "      <td>Xmc, Xmedchain, Blockchain Famous ICO Review C...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>53217</th>\n",
              "      <td>d0c58b5f2ab7</td>\n",
              "      <td>51.0</td>\n",
              "      <td>Datecoin, Denim, Blockchain, Dating What is Da...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>53274</th>\n",
              "      <td>d0efa707315e</td>\n",
              "      <td>6.0</td>\n",
              "      <td>Artificial Intelligence, Robots 6 Scariest Thi...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>55674</th>\n",
              "      <td>d982685ae4f7</td>\n",
              "      <td>1.0</td>\n",
              "      <td>Machine Learning, Python Build a basic ML mode...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>55872</th>\n",
              "      <td>da33297b668c</td>\n",
              "      <td>185.0</td>\n",
              "      <td>Blockchain, Matrix, Crypto, Bitcoin Why MATRIX...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>60653</th>\n",
              "      <td>eb484d556</td>\n",
              "      <td>22.0</td>\n",
              "      <td>Artificial Intelligence, Education, Inspiratio...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>66042</th>\n",
              "      <td>fecb4e514c4b</td>\n",
              "      <td>6.0</td>\n",
              "      <td>Machine Learning, Numpy, Python, Arrays, Deep ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>66054</th>\n",
              "      <td>fed2e269ae58</td>\n",
              "      <td>229.0</td>\n",
              "      <td>Bottos, Blockchain, Artificial Intelligence, C...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>66216</th>\n",
              "      <td>ff6589097809</td>\n",
              "      <td>1.0</td>\n",
              "      <td>Data Science, Python, Data Visualization, Rese...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "             postId  ...                                               text\n",
              "1209   146c91e8eff3  ...  Data Science Ilona’s Test\\nThis is simply a te...\n",
              "6980   29b7ab1c99b0  ...         Philosophy, Future McKenna on Super AI\\n\\n\n",
              "7598   2bf3c93d8f85  ...  Machine Learning, Westworld Some thoughts on v...\n",
              "9878   345a3015e1a0  ...  Social Media, Artificial Intelligence, Wellnes...\n",
              "11224  3916fd0334d8  ...  Bottos, Blockchain, Artificial Intelligence, C...\n",
              "12685  3e691fa8543b  ...  Burning Man, Machine Learning, Blockchain, Blo...\n",
              "13565  41b7d4739d80  ...  Automotive, Innovation, Robotics Autonomous Ve...\n",
              "15565  48a0523224e5  ...  Data Science, Learning, Tableau Homework #4-Ji...\n",
              "16069  4a852c39df75  ...  Chatbots, Bots, Artificial Intelligence, Machi...\n",
              "19089  559f7443f052  ...  Blockchain, Deepbrain DeepBrain Chain Weekly R...\n",
              "23425  653e1c6519c0  ...  Blockchain, Technology, Artificial Intelligenc...\n",
              "24016  675119c12a5d  ...  Data Analysis, Data Science, Data Analytics, D...\n",
              "24059  677f0bafb7c7  ...  Artificial Intelligence, Web Development MOCK ...\n",
              "25292  6bbd49b10dd6  ...  Science Fiction, Writing, Science, Artificial ...\n",
              "28280  76ac1342a614  ...  Coding, Competition, Streaming, Rustlang Eveni...\n",
              "29136  79c62bb729c6  ...  Bottos, Blockchain, Artificial Intelligence, C...\n",
              "39888  a0baa2ee571a  ...  Data Science Knowledge Extraction Projects\\nRe...\n",
              "41029  a4d30a895a93  ...  Information Retrieval, Linear Algebra, Machine...\n",
              "45570  b57ba9d4862e  ...  Art, Crowdfunding, Startup, Investing, Artific...\n",
              "50585   c76309b17d1  ...  Xmc, Xmedchain, Blockchain Famous ICO Review C...\n",
              "53217  d0c58b5f2ab7  ...  Datecoin, Denim, Blockchain, Dating What is Da...\n",
              "53274  d0efa707315e  ...  Artificial Intelligence, Robots 6 Scariest Thi...\n",
              "55674  d982685ae4f7  ...  Machine Learning, Python Build a basic ML mode...\n",
              "55872  da33297b668c  ...  Blockchain, Matrix, Crypto, Bitcoin Why MATRIX...\n",
              "60653     eb484d556  ...  Artificial Intelligence, Education, Inspiratio...\n",
              "66042  fecb4e514c4b  ...  Machine Learning, Numpy, Python, Arrays, Deep ...\n",
              "66054  fed2e269ae58  ...  Bottos, Blockchain, Artificial Intelligence, C...\n",
              "66216  ff6589097809  ...  Data Science, Python, Data Visualization, Rese...\n",
              "\n",
              "[28 rows x 3 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 279
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YbM_unEqRkNw"
      },
      "source": [
        "#in this case the code block contains the original text; the variable text is empty, thus, the empty text will be replace by the text of the codeBlock\n",
        "train_new.text[train_new.postId == \"6bbd49b10dd6\"] = train_new.codeBlock[train_new.postId == \"6bbd49b10dd6\"]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DOJTHRlWyxIF"
      },
      "source": [
        "#this text was not scraped properly, however it contains many claps and the text exists; thus, I decided to copy and implement the text manually\n",
        "print(train_new.url[53531])\n",
        "text_3e691fa8543b = \"one of my earliest memories where I am alone, I am writing in a little notebook. at the time I carried around with me inventor notebooks that contained different types of inventions. I had one for safety inventions, one for toys I invented, and my favorite one, the one with the most designs and specs was my flying car notebook. It was filled with every way I could think of that could make a “car” fly. At the time I was ~7 and only drove one of these: When I got older I started to throw out ideas to solve social problems, they ranged from //saving dogs from kill shelters and giving them to families with mental health problems like my own// to //creating spaces for community members to come together and cook//. Unfortunately at the time I was relatively shy and had a hard time telling anyone my ideas beyond my Mother and Father. Me, Kirsten (my american girl doll), my mother Sharon Reddick, my father Edward Chilmonik ll, and Cissy my first Dog just deeep in the 90's My house sat just across the river from the Colville Reservation in Omak, Washington. I had lived on the reservation side for a while when I was younger but as non-tribal members my parents couldn't by property on tribal lands. Omak is 2 hours from a mall and 4 hours from a commercial airport. The biggest event of the year there is the Suicide Race, the steepest horse race in the world. During my senior year my science teacher, Nancy Ridenour, asked my class to do a report on a STEM job. I chose civil/environmental engineering nearly at random out of a list. After presenting the report, I thought, “I guess this is my destiny.” I had done ballet for 16 years, competed in California and had won 1st place in a National Point Solo competition last year, but had broken my foot this year and realized how uncertain a career in the arts could be for me. Mrs. Ridenour was an inspiration to me as a young woman who had talents in math and science and so it felt simply <right> I was extremely supported in Omak. I was a cheerleader, a dancer, a church goer, close to my family, had lots of friends, was prom princess, and valedictorian. My first step out of that fishbowl was college. ~engineering school~is harrrddd. I wasn’t a studier. I had trouble concentrating in long classes. I realized that school was competitive and people weren’t always nice. I tore my ACL after my first month of school and felt utterly alone. My biggest bright spot was my roommate Pooja. We had met at orientation and our mothers liked each other immediately & they set us up as roommates. She was a brilliant young woman with high ambitions and endless empathy. She was studying Biomedical Engineering with a Father in Computer Science and through her I learned a lot more about the other options in engineering. Because I couldn’t dance at my previous level anymore I started to paint as well. This combination pushed me toward the intersection of software and art. Taking a class in transportation engineering was my final straw. I thought this might draw me closer to my *flying~car~dreams* but I ended up on a street corner counting cars for homework. I was done. I decided to switch my major to Systems Engineering. It had much more to do with math and programming. After school and a job as a developer in DC at a startup called Routeam I moved to NYC to accept a job at Thicket Labs. Since then I have learned and done a lot of things. During a period of depression being separated again from friends and family I asked myself what will make me get up each morning? What can I promise myself that will motivate me? What do I think is worth living for? Machine Learning was my answer the podcast Linear Digressions probably kept me alive. I spent most of my free time learning about ML algorithms and trying them out on my computer. I avoided any classes because they would take all the fun out of it for me. My boss Deepthi Welaratna suggested I apply for NASA datanauts. This was one of the best decisions for my career and social life. I was invited to present at NASA Headquarters in DC, that year I spoke at SXSW on behalf of Thicket Labs about self organizing maps (a kind of neural net), spoke to young data scientists at the American Museum of Natural History in NYC and taught a class at CoLab-Factory on Machine Learning for Business Application. Coming up on this last winter of 2017 I took time off work to explore how I could really support this new habit through my work in the most impactful way. ~Enter Byzantine~ Sam Hutch had been my roommate when I first moved to NYC and we worked together at Thicket Labs for a while before she left. There she has always astonished me with the quality and quantity of her work. She was amazing at design, UX, UI, and understanding complexity but most of all was a person I believed to be good at heart & quite cunning. She had reached out to me about a new company she was forming with another one of our former coworkers and introduced me to the new-new world of blockchain! back in college I had dabbled in crypto investing a small amount into Bitcoin (HODL) and had read early white papers at that time, but hadn’t been an avid follower of the industry for the last two years. They contacted me about their vision and hinted that they were interested in working with me. Immediately I learned everything about the industry I could. I fell in love once again with a tech revolution that gave me life. I have deep respect for Sam and luckily for me they saw that spark of an inventor, tinkerer, polymath and experimentalist in me and asked me to become Chief Discovery Officer. I SAID YES! At Byzantine we aren’t just stopping with tech innovation, we are blurring the lines between the physical and digital worlds. Byzantine has a space in Brooklyn over 92,000 sq. ft and our footprint will be expanding fast. We are managing a community called Liminal (our cute name that means “the place in between” in this case — digital and physical worlds) in our spaces that provides digital creators, technologists, and artists a place to co-create with access to resources like — Motion capture suits from Rokoko, Sound production booths, Live streaming studios, Maker Equipment, Mixed Reality Cubes, High speed computing, even higher speed internet (necessary for blockchain & game development), gallery space and much more. You may say ~WHOA there, that sounds EXpExnSive! Well it is! But thats why Byzantine has a two pronged approach — providing resources and building resources. We are in development of a Mobile IOS game with live player interactions built on the blockchain. The idea for the game came from Sam who brilliantly came up with a deeply engaging strategy game thats a mix between Farmville/Animal Crossing and Burning Man culture called *PLAYA*. The in game currencies are NFTs or unique encrypted tokens built on the blockchain. If you don’t know how transformative that is, I don’t blame you. This industry is moving fast. but it basically turns the gaming industry on its head because now game players can actually extract monetary IRL/Fiat/USD (etc.) value out of their in-game assets. Playa will push the boundaries between digital and physical environments even more by interacting through beacons with the physical spaces Byzantine and partners own, and unlocking events, rooms, and more though in-game achievements. Our first step toward releasing this game is #Space Quest# it is the first round of tokens for sale for use in the game! This is a product sale for in-game and IRL perks but happily can also be traded/sold on the market once they are all bought. After the last Space Quest Tile is sold there will be a “Flip Event” where the tiles will expose their metadata and the unique art that has meaning in Playa lore will be revealed. “Imagine if every dollar was a unique piece of art” The tiles are valued in Ethereum (ETH) which at the time of publishing was worth $ USD and climbing. The tiles are sold at increasing rates according to a step function (SALE is over since time of publishing) I hope that the Liminal spaces and similar innovative spaces we are building today will be the “Open Source Bell Labs” that will finally make my 10 types of flying cars a reality. My inner child is most excited about supporting all types of artists, technologists, philosophers, and makers to move past incremental innovations into a new age of radical innovation through cross-pollination of industries and radical inclusion.\"\n",
        "train_new.text[train_new.postId == \"3e691fa8543b\"] = text_3e691fa8543b"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pX8OjBMYEOIk"
      },
      "source": [
        "As already stated, the variable title has only 5 missing values. Since the title represents a crucial information and is needed before text cleaning, the missing values will be looked up and imputed manually."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "b8Po1ZYM1PBS"
      },
      "source": [
        "#add the missing titles manually; where all the same\n",
        "train_new.title[train_new.title.isnull()] = \"How I Became Chief Discovery Officer of a Blockchain Company by age 25\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XG_I2E9B226d"
      },
      "source": [
        "In order to make the text cleaning more efficient, we cut the max. sentence length at 2000. Since we will cut the max. length after processing to 400 anyways, setting it to 2000 reduces the computational time but ensures that no information that is needed it lost. <br>\n",
        "Of course, we applied a stop word removal; however, we decided that words such as no, not and nothing might have a contextual impact. Moreover, we used contraction mapping to keep certain sequences such as \"does not\" which would otherwised be removed during the cleaning process. <br>\n",
        "Lemmatization transforms a word into its lexicographically correct word and thus is used to reduce the amount of words but minimizes minor difference such as a word in singluar and plural which can be analysed as being the same word. The package nltk as well as spaCy both have excellent functions for the lemmatization of words. <br>\n",
        "Furthermore, the following cleaning steps have been applied:\n",
        "\n",
        "* removal of hmtl tags\n",
        "* removal of non alphanumeric letters (numbers, punctuation, hypens)\n",
        "* lowering text\n",
        "* removal white spaces if there are more than one between words\n",
        "* replace \"data\" to \"datum\" (falsely change during lemmatization and costs are higher for loosing the word data then for falsely misreplacing datum)\n",
        "* removal of single letters \n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UM_QgGNMiPgc"
      },
      "source": [
        "#general setting for preparation of the cleaning process such as:\n",
        "\n",
        "max_len_sentence = 2000\n",
        "\n",
        "  #loading stop words and excluding no, not and nothing since they might have a strong contextual impact and the meaning of a\n",
        "  #sequence in order to be able to detect all see next step: contractionmapping\n",
        "\n",
        "nlp = spacy.load(\"en\", disable=['parser', 'ner']) #keeping only tagger component needed for lemmatization and contains of list of english stop words (see nlp.Defaults.stop_words)\n",
        "\n",
        "#keep\n",
        "keep = {\"no\", \"not\", \"nothing\"} \n",
        "\n",
        "#remove\n",
        "remove = {\"www\", \"http\", \"https\", \"com\", \"html\"} #these tokens should be removed\n",
        "stop_words = set([re.sub(r\"\\W\", \"\", word) for word in nlp.Defaults.stop_words if word not in keep])\n",
        "  \n",
        "  #constraction mapper --> especially used to be able to keep the \"not\" which contain an important contextual message and should\n",
        "  #not be lost during the cleaning process; they will also be excluded from removing the stopwords -- as well as the words \"no\" and \"nothing\".\n",
        "  #source: https://github.com/dipanjanS/practical-machine-learning-with-python/blob/master/bonus%20content/nlp%20proven%20approach/contractions.py\n",
        "  \n",
        "contraction_mapping = { \n",
        "    \"ain't\": \"is not\", \n",
        "    \"aren't\": \"are not\", \n",
        "    \"can't\": \"cannot\", \n",
        "    \"'cause\": \"because\", \n",
        "    \"could've\": \"could have\", \n",
        "    \"couldn't\": \"could not\",\n",
        "    \"didn't\": \"did not\", \n",
        "    \"doesn't\": \"does not\", \n",
        "    \"don't\": \"do not\", \n",
        "    \"hadn't\": \"had not\", \n",
        "    \"hasn't\": \"has not\", \n",
        "    \"haven't\": \"have not\",\n",
        "    \"he'd\": \"he would\", \n",
        "    \"he'll\": \"he will\", \n",
        "    \"he's\": \"he is\", \n",
        "    \"how'd\": \"how did\", \n",
        "    \"how'd'y\": \"how do you\", \n",
        "    \"how'll\": \"how will\", \n",
        "    \"how's\": \"how is\",\n",
        "    \"I'd\": \"I would\", \n",
        "    \"I'd've\": \"I would have\", \n",
        "    \"I'll\": \"I will\", \n",
        "    \"I'll've\": \"I will have\", \n",
        "    \"I'm\": \"I am\", \n",
        "    \"I've\": \"I have\", \n",
        "    \"i'd\": \"i would\",\n",
        "    \"i'd've\": \"i would have\", \n",
        "    \"i'll\": \"i will\", \n",
        "    \"i'll've\": \"i will have\", \n",
        "    \"i'm\": \"i am\", \n",
        "    \"i've\": \"i have\", \n",
        "    \"isn't\": \"is not\", \n",
        "    \"it'd\": \"it would\",\n",
        "    \"it'd've\": \"it would have\", \n",
        "    \"it'll\": \"it will\", \n",
        "    \"it'll've\": \"it will have\", \n",
        "    \"it's\": \"it is\", \n",
        "    \"let's\": \"let us\", \n",
        "    \"ma'am\": \"madam\",\n",
        "    \"mayn't\": \"may not\", \n",
        "    \"might've\": \"might have\", \n",
        "    \"mightn't\": \"might not\", \n",
        "    \"mightn't've\": \"might not have\", \n",
        "    \"must've\": \"must have\",\n",
        "    \"mustn't\": \"must not\", \n",
        "    \"mustn't've\": \"must not have\", \n",
        "    \"needn't\": \"need not\", \n",
        "    \"needn't've\": \"need not have\", \n",
        "    \"o'clock\": \"of the clock\",\n",
        "    \"oughtn't\": \"ought not\", \n",
        "    \"oughtn't've\": \"ought not have\", \n",
        "    \"shan't\": \"shall not\", \n",
        "    \"sha'n't\": \"shall not\", \n",
        "    \"shan't've\": \"shall not have\",\n",
        "    \"she'd\": \"she would\", \n",
        "    \"she'd've\": \"she would have\", \n",
        "    \"she'll\": \"she will\", \n",
        "    \"she'll've\": \"she will have\", \n",
        "    \"she's\": \"she is\",\n",
        "    \"should've\": \"should have\", \n",
        "    \"shouldn't\": \"should not\", \n",
        "    \"shouldn't've\": \"should not have\", \n",
        "    \"so've\": \"so have\", \n",
        "    \"so's\": \"so as\",\n",
        "    \"this's\": \"this is\", \n",
        "    \"that'd\": \"that would\", \n",
        "    \"that'd've\": \"that would have\", \n",
        "    \"that's\": \"that is\", \n",
        "    \"there'd\": \"there would\",\n",
        "    \"there'd've\": \"there would have\", \n",
        "    \"there's\": \"there is\", \n",
        "    \"here's\": \"here is\", \n",
        "    \"they'd\": \"they would\", \n",
        "    \"they'd've\": \"they would have\",\n",
        "    \"they'll\": \"they will\", \n",
        "    \"they'll've\": \"they will have\", \n",
        "    \"they're\": \"they are\", \n",
        "    \"they've\": \"they have\", \n",
        "    \"to've\": \"to have\",\n",
        "    \"wasn't\": \"was not\", \n",
        "    \"we'd\": \"we would\", \n",
        "    \"we'd've\": \"we would have\", \n",
        "    \"we'll\": \"we will\", \n",
        "    \"we'll've\": \"we will have\", \n",
        "    \"we're\": \"we are\",\n",
        "    \"we've\": \"we have\", \n",
        "    \"weren't\": \"were not\", \n",
        "    \"what'll\": \"what will\", \n",
        "    \"what'll've\": \"what will have\", \n",
        "    \"what're\": \"what are\",\n",
        "    \"what's\": \"what is\", \n",
        "    \"what've\": \"what have\", \n",
        "    \"when's\": \"when is\", \n",
        "    \"when've\": \"when have\", \n",
        "    \"where'd\": \"where did\", \n",
        "    \"where's\": \"where is\",\n",
        "    \"where've\": \"where have\", \n",
        "    \"who'll\": \"who will\", \n",
        "    \"who'll've\": \"who will have\", \n",
        "    \"who's\": \"who is\", \n",
        "    \"who've\": \"who have\",\n",
        "    \"why's\": \"why is\", \n",
        "    \"why've\": \"why have\", \n",
        "    \"will've\": \"will have\", \n",
        "    \"won't\": \"will not\", \n",
        "    \"won't've\": \"will not have\",\n",
        "    \"would've\": \"would have\", \n",
        "    \"wouldn't\": \"would not\", \n",
        "    \"wouldn't've\": \"would not have\", \n",
        "    \"y'all\": \"you all\",\n",
        "    \"y'all'd\": \"you all would\", \n",
        "    \"y'all'd've\": \"you all would have\", \n",
        "    \"y'all're\": \"you all are\", \n",
        "    \"y'all've\": \"you all have\",\n",
        "    \"you'd\": \"you would\", \n",
        "    \"you'd've\": \"you would have\", \n",
        "    \"you'll\": \"you will\", \n",
        "    \"you'll've\": \"you will have\",\n",
        "    \"you're\": \"you are\", \n",
        "    \"you've\": \"you have\"}"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "E3-JCmHPO82E"
      },
      "source": [
        "#cleaning functions\n",
        "#source https://www.kdnuggets.com/2018/08/practitioners-guide-processing-understanding-text-2.html\n",
        "\n",
        "def lemmatize_function(text):\n",
        "\n",
        "    text = nlp(text)\n",
        "    text = \" \".join([word.lemma_ if word.lemma_ != \"-PRON-\" else word.text for word in text])\n",
        "    \n",
        "    return text\n",
        "   \n",
        "def cleaning_function(text):\n",
        "\n",
        "    text = \" \".join([contraction_mapping[word] if word in contraction_mapping else word for word in text.split(\" \")]) #split up words that are listed in the contraction map in order to not lose the meaning during text cleaning\n",
        "    text = BeautifulSoup(text).text #doing hmtl cleaning etc\n",
        "    \n",
        "    text = re.sub(\"[^a-zA-Z]\", \" \", text) #removes non alphanumeric letters (numbers, punctuation, hypens)\n",
        "    text = text.lower() #lowers text\n",
        "  \n",
        "    text = re.sub(\"\\s\\s+\" , \" \", text) #removes white spaces if there are more than one between words\n",
        "\n",
        "    text = text.replace(\"datum\", \" data \") #lemma changes data to datum quite often (of course in some case this correction might be wrong; however, it costs more to misinterpret the word \"data\" than the word \"datum\")\n",
        "    \n",
        "    text = re.sub(\"(\\\\b[A-Za-z] \\\\b|\\\\b [A-Za-z]\\\\b)\", \"\", text) #removes single letters \n",
        "    \n",
        "    text = word_tokenize(text)\n",
        "    text = [token for token in text if token.lower() not in stop_words | remove] #kicks out all the stopwords\n",
        "\n",
        "    text = text[:max_len_sentence] #sets text to max. length of 2000 tokens\n",
        "    \n",
        "    text = TreebankWordDetokenizer().detokenize(text)\n",
        "    text = lemmatize_function(text)\n",
        "    text = text.replace(\"datum\", \"data\")\n",
        "    \n",
        "    text = word_tokenize(text)\n",
        "\n",
        "    return text"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-men8E23S39w"
      },
      "source": [
        "#apply all functions to every row of the col \"text\"\n",
        "\n",
        "def clean_all(row, df_length):\n",
        "  \n",
        "  row[\"text\"] = cleaning_function(row[\"text\"])\n",
        "\n",
        "  print(f\"\\rcompleted: {row.name + 1} of {df_length}\", end=\"\")\n",
        "  return row\n",
        "\n",
        "t1 = time.perf_counter()\n",
        "#train_cleaned = tr_new.reset_index()\n",
        "#train_cleaned = train_cleaned.drop(\"index\", axis=1)\n",
        "train_cleaned = tr_new.apply(lambda row: clean_all(row, len(tr_new)), axis=1)\n",
        "t2 = time.perf_counter()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N7bMzydxTYaz"
      },
      "source": [
        "t3 = (t2-t1)/60\n",
        "print(\"It took {:.2f} minutes.\".format(t3))\n",
        "#full data (old cleaning): It took 154.10 minutes.\n",
        "#full data (new cleaning): It took 139.45 minutes.\n",
        "#full data (updated cleaning): It took 9.12 minutes (without lemmatization)\n",
        "#data (updated cleaning): It took ~40 (6.43 for 9999) minutes (with lemmatization)\n",
        "#data (updated cleaning): It took ~30 (4.59 for 9999) minutes (with tokenize, detokenize, lemmatize, tokenize)\n",
        "#full data (final cleaning): It took 31.93 minutes."
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4cAGD2PvlJrh"
      },
      "source": [
        "#clean dates regarding the first time the articles where published\n",
        "#remove \"-\"\" and all white spaces\n",
        "#the corresponding date to the var \"PublicationDetails\" is not the created but rather a published date\n",
        "train_cleaned[\"firstPublishedDate\"] = [re.sub(\"[^0-9]\", \" \", x) for x in train_cleaned.firstPublishedDate]\n",
        "train_cleaned[\"firstPublishedDate\"] = [re.sub(\" \", \"\", x) for x in train_cleaned.firstPublishedDate]\n",
        "train_cleaned[\"firstPublishedDate\"] = train_cleaned[\"firstPublishedDate\"].astype(\"int64\")\n",
        "#only year\n",
        "train_cleaned[\"firstPublishedYear\"] = [int(str(x)[:4]) for x in train_cleaned.firstPublishedDate]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ROfON_QhGdHM",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        },
        "outputId": "7970766b-7971-43cc-cd17-d8b60945d40b"
      },
      "source": [
        "print(np.corrcoef(tr_cl.firstPublishedYear, tr_cl.totalClapCount)) \n",
        "#correlation of only year is: -0.024 (very weak)\n",
        "print(np.corrcoef(tr_cl.firstPublishedDate, tr_cl.totalClapCount)) \n",
        "#correlation of year-month-day is: -0.026 (very weak)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[ 1.         -0.02415034]\n",
            " [-0.02415034  1.        ]]\n",
            "[[ 1.         -0.02595821]\n",
            " [-0.02595821  1.        ]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bY6huUZH8mj8"
      },
      "source": [
        "Statistics"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LijVsdrV6uqD",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 50
        },
        "outputId": "5678102d-34e0-464b-9412-812e2c7234b4"
      },
      "source": [
        "len_text = [len(x) for x in tr_cl.text.values]\n",
        "import statistics\n",
        "statistics.mean(len_text) #average length of tokens after cleaning\n",
        "print(max(len_text))\n",
        "print(min(len_text))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "2028\n",
            "3\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "T0YIEue8-iYG",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 50
        },
        "outputId": "4d5dfca6-7337-4b02-d986-19c0ed6f5cb9"
      },
      "source": [
        "tr_cl[\"len_text\"] = [len(x) for x in tr_cl.text.values]\n",
        "print(np.corrcoef(tr_cl.len_text, tr_cl.totalClapCount))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[1.         0.09367424]\n",
            " [0.09367424 1.        ]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7o0YCo916wcR",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 404
        },
        "outputId": "43390541-7fc9-42a1-af0f-227cee9e0a65"
      },
      "source": [
        "plt.figure(figsize=(10, 6))\n",
        "plt.subplot(1, 2, 1)\n",
        "\n",
        "#Tokens Length Plot\n",
        "plt.hist(tr_cl.len_text, 50, color= \"green\")\n",
        "plt.title(\"Text Length\")\n",
        "plt.xlabel(\"Number of Words after Text Cleaning\")\n",
        "plt.ylabel(\"No of articles\")\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 720x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WDfWUVehiwoW",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 187
        },
        "outputId": "a691290f-0a69-4896-9ec6-2202febc9475"
      },
      "source": [
        "#post-cleaning\n",
        "print(tr_cl.text[4])\n",
        "#pre-cleaning\n",
        "tr_new = tr_new.reset_index(drop = True)\n",
        "print(tr_new.text[4])\n",
        "\n",
        "##changes better to well but best to good"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['robotic', 'meditation', 'therapy', 'artificial', 'intelligence', 'feeling', 'robot', 'love', 'well', 'human', 'discuss', 'michelle', 'tsng', 'podcast', 'crazy', 'wisdom', 'say', 'robot', 'love', 'well', 'human', 'no', 'judgement', 'human', 'being', 'particularly', 'one', 'traumatize', 'subconsciously', 'detect', 'judge', 'know', 'true', 'feeling', 'hide', 'people', 'judge', 'good', 'guide', 'find', 'withhold', 'judgmental', 'thought', 'express', 'safe', 'warm', 'loving', 'connection', 'robot', 'sophisticated', 'able', 'provide', 'love', 'warm', 'connection', 'audio', 'clip', 'michelle', 'discuss', 'experience', 'talk', 'sofia', 'robotic', 'companion', 'human', 'being', 'say', 'soon', 'build', 'robot', 'well', 'love', 'human', 'think', 'feel', 'comfortable', 'share', 'intimate', 'experience', 'seek', 'therapeutic', 'treatment', 'robot', 'check', 'interview', 'website']\n",
            "Robotics, Meditation, Therapy, Artificial Intelligence, Feelings Can a robot love us better than another human can?\n",
            "\n",
            "I discussed this with Michelle Tsng on my Podcast “Crazy Wisdom”.\n",
            "She says that a robot can love us better than a human being because there is no judgement. Human beings, particularly ones who have been traumatized can subconsciously detect when someone is judging them. They know when to keep their true feelings hidden from people who judge them and thus the best guide they can find is someone who can withhold judgmental thoughts and just express a safe, warm, and loving connection.\n",
            "As robots become more sophisticated they might be able to provide this loving and warm connection. In this audio clip, Michelle discusses her experiences talking with Sofia, a robotic companion to human beings. She says that soon we will build robots who are better at love than humans are.\n",
            "What do you think? Would you ever feel comfortable sharing your most intimate experiences or seeking therapeutic treatment from a robot?\n",
            "You can check out the full interview on my website.\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "u-hwW4Z97mOB"
      },
      "source": [
        "Short texts (less than ten tokens) seem to have less claps than longer texts with a few exceptions. However, those exceptions are the mentioned blog posts that contain images with text rather than \"scrappable\" text."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zV7U6LT38Owq",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 286
        },
        "outputId": "0add5f92-1c63-4b27-9a23-89f7ce048a47"
      },
      "source": [
        "tr_cl[\"text_len\"] = [len(x) for x in tr_cl.text.values]\n",
        "text_short = tr_cl.loc[tr_cl.text_len < 10, [\"totalClapCount\"]]\n",
        "text_short.describe()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>totalClapCount</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>count</th>\n",
              "      <td>102.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>mean</th>\n",
              "      <td>23.784314</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>std</th>\n",
              "      <td>94.378281</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>min</th>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>25%</th>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>50%</th>\n",
              "      <td>0.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>75%</th>\n",
              "      <td>1.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>max</th>\n",
              "      <td>632.000000</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "       totalClapCount\n",
              "count      102.000000\n",
              "mean        23.784314\n",
              "std         94.378281\n",
              "min          0.000000\n",
              "25%          0.000000\n",
              "50%          0.000000\n",
              "75%          1.000000\n",
              "max        632.000000"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 300
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4j6fux7p6TPD"
      },
      "source": [
        "**Checkpoint**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3zXff84IgfaG"
      },
      "source": [
        "#SAVE cleaned training data file\n",
        "import pickle\n",
        "with open(\"train_cleaned\", \"wb\") as path_name:\n",
        "  pickle.dump(train_cleaned, path_name)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8_y4FRMkguX9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 53
        },
        "outputId": "e7ba7159-0ea8-4d4d-a599-ea4b21f14e80"
      },
      "source": [
        "#LOAD cleaned training data file\n",
        "import pickle\n",
        "with open(\"train_cleaned\", \"rb\") as path_name:\n",
        "  tr_cl = pickle.load(path_name)\n",
        "\n",
        "print(tr_cl.loc[10, \"text\"])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['artificial', 'intelligence', 'ai', 'impact', 'world', 'actually', 'ai', 'efiicte', 'world', 'good', 'way', 'efficient', 'way', 'go', 'to', 'kill', 'no', 'different', 'job', 'world', 'today', 'go', 'to', 'automate', 'open', 'open', 'no', 'opportunity', 'different', 'job', 'jobs', 'people', 'like', 'current', 'world', 'save', 'complex', 'statistic', 'mathematics', 'didn', 'ai', 'ai', 'get', 'new', 'feature', 'like', 'machine', 'talk', 'like', 'human', 'someday', 'machine', 'not', 'machine', 'human', 'like', 'solve', 'no', 'current', 'problem', 'fear', 'ai', 'apoocylapse', 'not', 'usedsny', 'command', 'line', 'server', 'language', 'programmer', 'know', 'machine', 'way', 'dumb', 'human']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xEHImqADtYdx"
      },
      "source": [
        "from nltk.tokenize.treebank import TreebankWordDetokenizer\n",
        "tr_cl[\"text_cl\"] = [TreebankWordDetokenizer().detokenize(token) for token in tr_cl.text] #needed as an input for afinn and for one-hot-encoding"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "drqNlQ-xHfF6"
      },
      "source": [
        "###Sentiment Analysis"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mbWjbmDdxRP8"
      },
      "source": [
        "The idea behind conducting a sentiment analysis, even though the scrapped articles were mainly tech related article which are not exactly famous for their emotionality, was that we observed that the very high claps were blog posts with emotional topic such as chronic depression or groundbreaking skills. However, the results were not very promising and thus were not examined any further. For the sake of completeness the brief sentiment is still in the assignment post since I might be an interesting topic for further research."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eNEgvnLfXyjK",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 67
        },
        "outputId": "2e3277b4-428a-4941-b36d-9561513ced91"
      },
      "source": [
        "a = \"totalClapCount > 150000\"\n",
        "highclaps = tr_cl.query(a)\n",
        "highclaps[\"title\"]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "35112    My name is Wil Wheaton. I Live With Chronic De...\n",
              "38329           The Most Important Skill Nobody Taught You\n",
              "Name: title, dtype: object"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 303
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HAO8AaGZuF5a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "outputId": "12236d80-057a-4dd8-f966-a02298a531fd"
      },
      "source": [
        "af = Afinn()\n",
        "#Afinn - dictionary based results might be biased towards positive articles as the result showed in the tut #11 regarding the reviews of movies."
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting afinn\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/86/e5/ffbb7ee3cca21ac6d310ac01944fb163c20030b45bda25421d725d8a859a/afinn-0.1.tar.gz (52kB)\n",
            "\r\u001b[K     |██████▎                         | 10kB 11.2MB/s eta 0:00:01\r\u001b[K     |████████████▌                   | 20kB 1.8MB/s eta 0:00:01\r\u001b[K     |██████████████████▊             | 30kB 2.3MB/s eta 0:00:01\r\u001b[K     |█████████████████████████       | 40kB 2.6MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████▏| 51kB 2.1MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 61kB 1.8MB/s \n",
            "\u001b[?25hBuilding wheels for collected packages: afinn\n",
            "  Building wheel for afinn (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for afinn: filename=afinn-0.1-cp36-none-any.whl size=53453 sha256=03608037d68ca8d78fcfbbbc805ec57cbfbf9befd00315a00abda1824dbd53db\n",
            "  Stored in directory: /root/.cache/pip/wheels/b5/1c/de/428301f3333ca509dcf20ff358690eb23a1388fbcbbde008b2\n",
            "Successfully built afinn\n",
            "Installing collected packages: afinn\n",
            "Successfully installed afinn-0.1\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XtZu3syGn1mm"
      },
      "source": [
        "tr_cl[\"text_cl\"] = tr_cl[\"text_cl\"][:400] "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yBcOqe511x2B"
      },
      "source": [
        "start = time.time()\n",
        "tr_cl[\"sentiment\"] = tr_cl[\"text_cl\"].apply(af.score)\n",
        "end = time.time()\n",
        "\n",
        "print(\"Processed {} articles in {:.0f} sec.\".format(tr_cl.shape[0], end-start))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Cph0lb39yGI0"
      },
      "source": [
        "tr_cl[\"text_len\"] = tr_cl.text.apply(len)\n",
        "#tr_cl[\"rel_sentiment\"] = tr_cl.apply(lambda row: row[\"sentiment\"] / row[\"text_len\"], axis=1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jI0eP-h7igM8"
      },
      "source": [
        "tr_cl[[\"totalClapCount\", \"sentiment\", \"text_len\", \"emotionality\"]]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "h61NQ9hycgYS"
      },
      "source": [
        "tr_cl[\"emotionality\"] = np.where((tr_cl[\"sentiment\"] > 70) | (tr_cl[\"sentiment\"] < -70), 1, 0) #1 = very emotional"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KWhMC2CTdaLJ",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 68
        },
        "outputId": "45af097b-098f-46ff-f360-6ed338c56a64"
      },
      "source": [
        "tr_cl.emotionality.value_counts()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0    17101\n",
              "1     2898\n",
              "Name: emotionality, dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 69
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pPp40-j0gW4g"
      },
      "source": [
        "emotional = tr_cl.loc[tr_cl.emotionality == 1, [\"totalClapCount\"]]\n",
        "unemotional = tr_cl.loc[tr_cl.emotionality == 0, [\"totalClapCount\"]]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RcI_8yICf2px"
      },
      "source": [
        "#total clap counts of emotional texts\n",
        "plt.figure(figsize=(10, 6))\n",
        "plt.subplot(1, 2, 1)\n",
        "plt.hist(emotional.totalClapCount, 200, color = \"green\", density = True , range = (10,400))\n",
        "plt.title(\"Total Clap Counts of emotional articles\")\n",
        "plt.xlabel(\"# of Claps\")\n",
        "#total clap counts of less emotional texts\n",
        "plt.subplot(1, 2, 2)\n",
        "plt.hist(unemotional.totalClapCount, 200,color = \"darkorange\", density = True , range = (10,400))\n",
        "plt.title(\"Total clap counts of less emotional texts\")\n",
        "plt.xlabel(\"# of Claps\")\n",
        "plt.subplots_adjust(bottom=0.25, top=0.75)\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Uhub9WAKkDBd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 328
        },
        "outputId": "f41eb9bc-60df-450f-e025-6573d5ae9dc9"
      },
      "source": [
        "highclaps = tr_cl.loc[tr_cl.totalClapCount > 30000, [\"totalClapCount\", \"emotionality\", \"sentiment\", \"title\"]]\n",
        "highclaps"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>totalClapCount</th>\n",
              "      <th>emotionality</th>\n",
              "      <th>sentiment</th>\n",
              "      <th>title</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>590</th>\n",
              "      <td>47445.0</td>\n",
              "      <td>0</td>\n",
              "      <td>23.0</td>\n",
              "      <td>Learning Python: From Zero to Hero</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2930</th>\n",
              "      <td>39376.0</td>\n",
              "      <td>0</td>\n",
              "      <td>13.0</td>\n",
              "      <td>Want to know how Deep Learning works? Here’s a...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5729</th>\n",
              "      <td>49838.0</td>\n",
              "      <td>0</td>\n",
              "      <td>21.0</td>\n",
              "      <td>I interviewed at five top companies in Silicon...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>12640</th>\n",
              "      <td>79766.0</td>\n",
              "      <td>0</td>\n",
              "      <td>9.0</td>\n",
              "      <td>Towards a Bra-free Instagram Experience</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>14560</th>\n",
              "      <td>33856.0</td>\n",
              "      <td>0</td>\n",
              "      <td>12.0</td>\n",
              "      <td>The Snips AIR Token Sale</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>16485</th>\n",
              "      <td>112595.0</td>\n",
              "      <td>0</td>\n",
              "      <td>-2.0</td>\n",
              "      <td>We fired our top talent. Best decision we ever...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>17956</th>\n",
              "      <td>32693.0</td>\n",
              "      <td>0</td>\n",
              "      <td>30.0</td>\n",
              "      <td>What exactly can you do with Python? Here are ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>18499</th>\n",
              "      <td>42710.0</td>\n",
              "      <td>0</td>\n",
              "      <td>27.0</td>\n",
              "      <td>Fast-forward twenty years with Anything App.</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>19092</th>\n",
              "      <td>35192.0</td>\n",
              "      <td>0</td>\n",
              "      <td>-14.0</td>\n",
              "      <td>Cognitive bias cheat sheet</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "       totalClapCount  ...                                              title\n",
              "590           47445.0  ...                 Learning Python: From Zero to Hero\n",
              "2930          39376.0  ...  Want to know how Deep Learning works? Here’s a...\n",
              "5729          49838.0  ...  I interviewed at five top companies in Silicon...\n",
              "12640         79766.0  ...            Towards a Bra-free Instagram Experience\n",
              "14560         33856.0  ...                           The Snips AIR Token Sale\n",
              "16485        112595.0  ...  We fired our top talent. Best decision we ever...\n",
              "17956         32693.0  ...  What exactly can you do with Python? Here are ...\n",
              "18499         42710.0  ...       Fast-forward twenty years with Anything App.\n",
              "19092         35192.0  ...                         Cognitive bias cheat sheet\n",
              "\n",
              "[9 rows x 4 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 88
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0JxjjJN0kqrp",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 235
        },
        "outputId": "248b9e67-af3e-4035-87fc-8e83c980f6d8"
      },
      "source": [
        "neg_senti = tr_cl.loc[tr_cl.sentiment < -80, [\"totalClapCount\", \"emotionality\", \"sentiment\", \"title\"]]\n",
        "neg_senti #very negative sentiment has barely any claps"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>totalClapCount</th>\n",
              "      <th>emotionality</th>\n",
              "      <th>sentiment</th>\n",
              "      <th>title</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>663</th>\n",
              "      <td>2.0</td>\n",
              "      <td>1</td>\n",
              "      <td>-100.0</td>\n",
              "      <td>Fraud Detection with ML</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2936</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>-156.0</td>\n",
              "      <td>Pdf Download eBook Free Fraud Analytics Using ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>7065</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>-160.0</td>\n",
              "      <td>Download pdf Online No bullshit guide to linea...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>9759</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>-111.0</td>\n",
              "      <td>Fraud Management: Detection and Prevention in ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>14490</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>-84.0</td>\n",
              "      <td>Google is Using its AI to Track U.S. Hate Crimes</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>18784</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>-87.0</td>\n",
              "      <td>Why Every Business Needs Fraud Analytics</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "       totalClapCount  ...                                              title\n",
              "663               2.0  ...                            Fraud Detection with ML\n",
              "2936              0.0  ...  Pdf Download eBook Free Fraud Analytics Using ...\n",
              "7065              0.0  ...  Download pdf Online No bullshit guide to linea...\n",
              "9759              0.0  ...  Fraud Management: Detection and Prevention in ...\n",
              "14490             0.0  ...   Google is Using its AI to Track U.S. Hate Crimes\n",
              "18784             0.0  ...           Why Every Business Needs Fraud Analytics\n",
              "\n",
              "[6 rows x 4 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 95
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "446zLedtlWlH",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 545
        },
        "outputId": "ce604369-7a94-4c10-f7dd-a9e74e1ad6e9"
      },
      "source": [
        "pos_senti = tr_cl.loc[tr_cl.sentiment > 80, [\"totalClapCount\", \"emotionality\", \"sentiment\", \"title\"]]\n",
        "pos_senti #very negative sentiment has barely any claps"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>totalClapCount</th>\n",
              "      <th>emotionality</th>\n",
              "      <th>sentiment</th>\n",
              "      <th>title</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>2225</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>81.0</td>\n",
              "      <td>Introducing “NEVA” by NICE — The First Robotic...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3706</th>\n",
              "      <td>5.0</td>\n",
              "      <td>1</td>\n",
              "      <td>114.0</td>\n",
              "      <td>We analyzed historical Super Bowl data and fou...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4776</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>110.0</td>\n",
              "      <td>Pdf download Journey to Data Scientist: Interv...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4996</th>\n",
              "      <td>234.0</td>\n",
              "      <td>1</td>\n",
              "      <td>88.0</td>\n",
              "      <td>Artificial Intelligence (AI) In Beauty Industry</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>8480</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>96.0</td>\n",
              "      <td>Download pdf Read Online Howl’s Moving Castle ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>9405</th>\n",
              "      <td>526.0</td>\n",
              "      <td>1</td>\n",
              "      <td>87.0</td>\n",
              "      <td>Free game-analytics on moreMMR.com VS DotA Plus</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>11582</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>96.0</td>\n",
              "      <td>Read Artificial Intelligence: A Guide To Intel...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>12240</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>90.0</td>\n",
              "      <td>buy elo boost</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>13038</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>99.0</td>\n",
              "      <td>Tips on How to Hire an Artificial Intelligence...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>14387</th>\n",
              "      <td>106.0</td>\n",
              "      <td>1</td>\n",
              "      <td>83.0</td>\n",
              "      <td>2017 in review — startups, bots &amp; AI</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>15427</th>\n",
              "      <td>15.0</td>\n",
              "      <td>1</td>\n",
              "      <td>83.0</td>\n",
              "      <td>KDD 2018 Announces Best Paper &amp; Other Awards</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>17515</th>\n",
              "      <td>0.0</td>\n",
              "      <td>1</td>\n",
              "      <td>97.0</td>\n",
              "      <td>Download Robot Ethics: The Ethical and Social ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>18422</th>\n",
              "      <td>105.0</td>\n",
              "      <td>1</td>\n",
              "      <td>85.0</td>\n",
              "      <td>Basketball Analytics: Predicting Win Shares</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>18661</th>\n",
              "      <td>1322.0</td>\n",
              "      <td>1</td>\n",
              "      <td>83.0</td>\n",
              "      <td>Sustainable Product Growth</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>19480</th>\n",
              "      <td>92.0</td>\n",
              "      <td>1</td>\n",
              "      <td>93.0</td>\n",
              "      <td>IJCAI 2018 Kicks Off; DeepMind AlphaGo Wins Ma...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>19608</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1</td>\n",
              "      <td>85.0</td>\n",
              "      <td>Why Beauty Is Essential to Human Survival</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "       totalClapCount  ...                                              title\n",
              "2225              0.0  ...  Introducing “NEVA” by NICE — The First Robotic...\n",
              "3706              5.0  ...  We analyzed historical Super Bowl data and fou...\n",
              "4776              0.0  ...  Pdf download Journey to Data Scientist: Interv...\n",
              "4996            234.0  ...    Artificial Intelligence (AI) In Beauty Industry\n",
              "8480              0.0  ...  Download pdf Read Online Howl’s Moving Castle ...\n",
              "9405            526.0  ...    Free game-analytics on moreMMR.com VS DotA Plus\n",
              "11582             0.0  ...  Read Artificial Intelligence: A Guide To Intel...\n",
              "12240             0.0  ...                                      buy elo boost\n",
              "13038             0.0  ...  Tips on How to Hire an Artificial Intelligence...\n",
              "14387           106.0  ...               2017 in review — startups, bots & AI\n",
              "15427            15.0  ...       KDD 2018 Announces Best Paper & Other Awards\n",
              "17515             0.0  ...  Download Robot Ethics: The Ethical and Social ...\n",
              "18422           105.0  ...        Basketball Analytics: Predicting Win Shares\n",
              "18661          1322.0  ...                         Sustainable Product Growth\n",
              "19480            92.0  ...  IJCAI 2018 Kicks Off; DeepMind AlphaGo Wins Ma...\n",
              "19608             1.0  ...          Why Beauty Is Essential to Human Survival\n",
              "\n",
              "[16 rows x 4 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 97
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tp9B5eK79hZf",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 119
        },
        "outputId": "e28534ea-cca7-4ec8-d8aa-bcd3248d6cb0"
      },
      "source": [
        "print(np.corrcoef(tr_cl.totalClapCount, tr_cl.sentiment))\n",
        "#print(np.corrcoef(tr_cl.totalClapCount, tr_cl.rel_sentiment))\n",
        "print(np.corrcoef(tr_cl.totalClapCount, tr_cl.text_len))\n",
        "#validation check: should be somewhat positive correlated \n",
        "print(np.corrcoef(tr_cl.text_len, tr_cl.sentiment))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[ 1.         -0.00419232]\n",
            " [-0.00419232  1.        ]]\n",
            "[[1.         0.02545214]\n",
            " [0.02545214 1.        ]]\n",
            "[[1.         0.21701534]\n",
            " [0.21701534 1.        ]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "u2xxU01oFrQa"
      },
      "source": [
        "## Embeddings\n",
        "\n",
        "Before feeding the words to a neural network with the target of predicting the clap scores, the words of the blog posts can be be transformed into number in order to keep information about the relation between words, e.g. how similar a word is to another or the co-occurence of words. This is either possible with pre-trained embedding as well as with \"self-trained\" embeddings. The goal is to catch the semantic as well as syntatic relationship between words. Both option will be applied and compared. <br>\n",
        "Training own embeddings is computationally very costly and can take up a long time. Furthermore, it is dependend on the amount and quality of the training data, the used algorithm and the size of the vectors. The underlying hardware has a significant impact and sets limitation on the choice of the parameters of the algorithm such as the dimensions of the vector. However, the upside of this approach is that the word embeddings are tailored to solve a task with exactly the underlying kind of data. In contrast, pre-trained embeddings contain huge amounts of data in their embeddings. But, they might be trained on different kinds of data and therefore, might be outperformed by the self-trained embeddings. It is possible to fine-tune them with the target corpus. [4] <br>\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "emgcBCO0zBYS"
      },
      "source": [
        "X = tr_cl.loc[:,[\"text\"]] \n",
        "y = tr_cl.loc[:,[\"totalClapCount\"]]\n",
        "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=123) #random state for comparability"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SegAP5BNLbNh"
      },
      "source": [
        "word_counter = collections.Counter()\n",
        "for r in X_train[\"text\"]:\n",
        "  for w in r:\n",
        "    word_counter.update({w: 1})"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0Lzw2hUYMID5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 442
        },
        "outputId": "0a33471e-efa6-4a95-ff20-6829a8bf0d1f"
      },
      "source": [
        "#most common words occuring in the text corpus of X_train\n",
        "word_counter.most_common(25)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('data', 360322),\n",
              " ('ai', 199470),\n",
              " ('not', 175814),\n",
              " ('learn', 152833),\n",
              " ('machine', 149704),\n",
              " ('model', 131426),\n",
              " ('time', 116843),\n",
              " ('use', 107613),\n",
              " ('like', 106578),\n",
              " ('intelligence', 100385),\n",
              " ('work', 98651),\n",
              " ('learning', 93021),\n",
              " ('new', 92458),\n",
              " ('human', 90804),\n",
              " ('technology', 87100),\n",
              " ('artificial', 85584),\n",
              " ('need', 84517),\n",
              " ('way', 73486),\n",
              " ('network', 72830),\n",
              " ('create', 70346),\n",
              " ('people', 69399),\n",
              " ('good', 68987),\n",
              " ('system', 68262),\n",
              " ('company', 66180),\n",
              " ('user', 66115)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 48
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GVnYGDgSkdko"
      },
      "source": [
        "### Self-trained Embeddings"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6ivZWScgBKlu"
      },
      "source": [
        "A famous method for creating word embeddings is Word2Vec. Word2Vec inherits two methods that work slightly different. They are called Skip-Gram (SG) and Common Bag of Words (CBOW). Both are based on neural networks. [4] <br> Skip-gram is slower than cbow but it is more suitable for infrequent words. Both algorithms will be tested with the similar parameter (except window size) in order to be able to compare them. Setting the hyper-parameter is essential for the performance and the accuracy of the resulting embedding. Google suggests to set the window size for cbow to 5 and for skip-gram to 10. The hierarchical softmax function provides better results for infrequent words. In contrast, the negative sampling is more suitable for frequent words but also for vectors with low dimensions. In general, the most dimensions the better. [1] Unfortunately, the computing PC has computational limitations.\n",
        "<br> Often only unigram approaches are applied. However, in the contest of the analysis many important bigrams occur such as artificial intelligence and data science. In order to obtain those bigrams we include the bigram tokens in our analysis. <br>\n",
        "Both models are trained 25 epochs with an embedding dimension with 150 and 300. In the following both models will explained in more detail. <br> \n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QFodU78EGpTm"
      },
      "source": [
        "####Preparation of Bigram Tokens"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9DcnlExdGuOV"
      },
      "source": [
        "#including bigrams to word embedding; some unneccessary bigram which do not contribute anything to the word embedding quality\n",
        "#however, very important bigrams such as artificial intelligent etc. are included which contribute a lot; as it can be observed\n",
        "#in the most frequent words the first 10 bigrams are very meaningful regarding the context\n",
        "bigram_model = Phrases(X_train[\"text\"], min_count=100, threshold=2, delimiter=b\"_\")\n",
        "bigram_phraser = Phraser(bigram_model)\n",
        "\n",
        "bigram_token = []\n",
        "for x in X_train[\"text\"]:\n",
        "    bigram_token.append(bigram_phraser[x])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z_yBlICVY9Zt"
      },
      "source": [
        "#count bigrams\n",
        "bigram_counter = collections.Counter()\n",
        "for key in bigram_model.vocab.keys():\n",
        "  if key.decode().find(\"_\")>-1:\n",
        "    bigram_counter[key] += bigram_model.vocab[key]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AFZtYcH9ZSuu",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 442
        },
        "outputId": "22220396-43a7-45c1-a3af-8e2b9cdb9af4"
      },
      "source": [
        "bigram_counter.most_common(25)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[(b'artificial_intelligence', 78673),\n",
              " (b'machine_learning', 54116),\n",
              " (b'machine_learn', 52847),\n",
              " (b'data_science', 44161),\n",
              " (b'neural_network', 33346),\n",
              " (b'deep_learn', 19734),\n",
              " (b'data_scientist', 17342),\n",
              " (b'deep_learning', 16678),\n",
              " (b'big_data', 14431),\n",
              " (b'data_set', 11296),\n",
              " (b'natural_language', 9291),\n",
              " (b'real_time', 8935),\n",
              " (b'look_like', 8196),\n",
              " (b'social_medium', 7770),\n",
              " (b'use_case', 6998),\n",
              " (b'open_source', 6602),\n",
              " (b'computer_vision', 6536),\n",
              " (b'learning_model', 6529),\n",
              " (b'training_data', 6520),\n",
              " (b'data_analysis', 6514),\n",
              " (b'real_world', 6071),\n",
              " (b'intelligence_ai', 5886),\n",
              " (b'language_processing', 5423),\n",
              " (b'linear_regression', 5410),\n",
              " (b'learn_artificial', 5309)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 58
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OUnBgMMlBDhg"
      },
      "source": [
        "####Skip-gram Model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rSRd8nijAzvc"
      },
      "source": [
        "The skip-gram approach loops over the each word of the corpus aka the input sequence and tries to predict the neighboring words which are the representation of the context. How many neighboring words is dependend on the parameter \"window_size\". If the window size is e.g. 5, the model tries to predict the two words in front and behind the target word. The model is trained on a one layer neural network. [5]"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IBHR0H8S-g6r",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "159a310a-3bac-477e-91d8-856ce00e37e6"
      },
      "source": [
        "epochs = 25\n",
        "\n",
        "t1 = time.perf_counter()\n",
        "model_sg_bigram = Word2Vec(bigram_token,\n",
        "                 min_count = 5,## min_count means the frequency benchmark, if =2 and word is used only once - it's not included\n",
        "                 sg = 1, #trains for skip-gram (tut: CBOW by default, we have to add it to specify for skipgram)\n",
        "                 window = 10, #the size of context\n",
        "                 iter = epochs, #how many times the training code will run through the data set, same as epochs (first pass is to create dict)\n",
        "                 size = 150, #size of embedding\n",
        "                 workers = 4, #for parallel computing\n",
        "                 hs = 0) #0 for negative sampling; 1 for softmax function\n",
        "    \n",
        "t2 = time.perf_counter()\n",
        "t3_sg_bigram = (t2-t1)/60\n",
        "print(model_sg_bigram)\n",
        "print(\"The computation of the embedding for the skip-gram model took {:.2f} minutes.\".format(t3_sg_bigram))\n",
        "\n",
        "#with epoch = 15, size = 150\n",
        "#Word2Vec(vocab=58073, size=150, alpha=0.025)\n",
        "#The computation of the embedding for the skip-gram model took 81.55 minutes.\n",
        "\n",
        "#with epoch = 25, size = 300\n",
        "#Word2Vec(vocab=58073, size=300, alpha=0.025)\n",
        "#The computation of the embedding for the skip-gram model took 159.33 minutes.\n",
        "\n",
        "#with epoch = 25, size = 300\n",
        "#Word2Vec(vocab=64334, size=300, alpha=0.025)\n",
        "#The computation of the embedding for the skip-gram model took 179.85 minutes.\n",
        "\n",
        "#with epoch = 25, size = 300\n",
        "#Word2Vec(vocab=64334, size=150, alpha=0.025)\n",
        "#The computation of the embedding for the skip-gram model took 150.50 minutes."
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Word2Vec(vocab=64334, size=150, alpha=0.025)\n",
            "The computation of the embedding for the skip-gram model took 150.50 minutes.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "P7kEDZJJ7QBc",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 258
        },
        "outputId": "a2f5a7ff-af47-474d-dca5-40765307c7d7"
      },
      "source": [
        "#results: 150\n",
        "print(model_sg_bigram.wv.vectors.shape)\n",
        "model_sg_bigram.wv.most_similar(positive=[\"good\"], topn = 10) "
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(64334, 150)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/gensim/matutils.py:737: FutureWarning: Conversion of the second argument of issubdtype from `int` to `np.signedinteger` is deprecated. In future, it will be treated as `np.int64 == np.dtype(int).type`.\n",
            "  if np.issubdtype(vec.dtype, np.int):\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('best', 0.7949308156967163),\n",
              " ('well', 0.7241239547729492),\n",
              " ('better', 0.6603259444236755),\n",
              " ('bad', 0.6512195467948914),\n",
              " ('great', 0.646959125995636),\n",
              " ('perfect', 0.6388470530509949),\n",
              " ('excellent', 0.6254705786705017),\n",
              " ('decent', 0.6192342042922974),\n",
              " ('not', 0.6097713112831116),\n",
              " ('want', 0.6087747812271118)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 60
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-5FgA9am53sI",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 258
        },
        "outputId": "f3c542ce-ae15-4ced-eae7-976817cc5dbc"
      },
      "source": [
        "#results: 300\n",
        "print(model_sg_bigram.wv.vectors.shape)\n",
        "model_sg_bigram.wv.most_similar(positive=[\"good\"], topn = 10) #model seems to catch the meaning very well"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(64334, 300)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/gensim/matutils.py:737: FutureWarning: Conversion of the second argument of issubdtype from `int` to `np.signedinteger` is deprecated. In future, it will be treated as `np.int64 == np.dtype(int).type`.\n",
            "  if np.issubdtype(vec.dtype, np.int):\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('best', 0.6863915324211121),\n",
              " ('well', 0.6473300457000732),\n",
              " ('bad', 0.596216082572937),\n",
              " ('excellent', 0.5689988732337952),\n",
              " ('great', 0.5610009431838989),\n",
              " ('better', 0.5605751872062683),\n",
              " ('perfect', 0.533531665802002),\n",
              " ('find_good', 0.5205310583114624),\n",
              " ('decent', 0.5025161504745483),\n",
              " ('want', 0.49737104773521423)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 57
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YaXdF1N2BnlH"
      },
      "source": [
        "Saving"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "talkVL_maeTg",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 71
        },
        "outputId": "f2a6f6fc-94d1-45e1-9bc7-a7f795818d9f"
      },
      "source": [
        "#saving\n",
        "embs=\"sg_150_bigram_embeddings.txt\"\n",
        "model_sg_bigram.wv.save_word2vec_format(embs, binary=False)\n",
        "#embs=\"sg_300_bigram_embeddings.txt\"\n",
        "#model_sg_bigram.wv.save_word2vec_format(embs, binary=False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/smart_open/smart_open_lib.py:254: UserWarning: This function is deprecated, use smart_open.open instead. See the migration notes for details: https://github.com/RaRe-Technologies/smart_open/blob/master/README.rst#migrating-to-the-new-open-function\n",
            "  'See the migration notes for details: %s' % _MIGRATION_NOTES_URL\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WoNWaNBwA6Ct"
      },
      "source": [
        "####Cbow Model (Word2Vec)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ri4dy672FgU4"
      },
      "source": [
        "The CBOW model works similar to the skip-gram model just the other way around. Instead of predicting the neighboring words, it uses the neighboring words in order to predict the target word. [5]"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UqF_Xa4w2eZD",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "ff9e8582-fa0b-4b2e-e3cc-512d0c246201"
      },
      "source": [
        "epochs = 25\n",
        "\n",
        "t1 = time.perf_counter()\n",
        "model_cbow_bigram = Word2Vec(bigram_token,\n",
        "                 min_count = 5, # min_count means the frequency benchmark, if =2 and word is used only once - it's not included\n",
        "                 sg = 0, #trains for skip-gram (tut: CBOW by default, we have to add it to specify for skipgram)\n",
        "                 window = 5, #the size of context\n",
        "                 iter = epochs, #how many times the training code will run through the data set, same as epochs (first pass is to create dict)\n",
        "                 size = 300, #size of embedding\n",
        "                 workers = 4, #for parallel computing\n",
        "                 hs = 0) #0 for negative sampling; 1 for softmax function\n",
        "    \n",
        "t2 = time.perf_counter()\n",
        "t3_cbow_bigram = (t2-t1)/60\n",
        "print(model_cbow_bigram)\n",
        "print(\"The computation of the embedding for the cbow model took {:.2f} minutes.\".format(t3_cbow_bigram))\n",
        "\n",
        "#Word2Vec(vocab=56943, size=150, alpha=0.025)\n",
        "#The computation of the embedding for the cbow model took 20.29 minutes.\n",
        "\n",
        "#Word2Vec(vocab=56943, size=300, alpha=0.025)\n",
        "#The computation of the embedding for the cbow model took 22.09 minutes.\n",
        "\n",
        "#Word2Vec(vocab=64334, size=300, alpha=0.025)\n",
        "#The computation of the embedding for the cbow model took 27.93 minutes."
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Word2Vec(vocab=64334, size=300, alpha=0.025)\n",
            "The computation of the embedding for the cbow model took 27.93 minutes.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Vgy7ZBKJ4Der",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 71
        },
        "outputId": "345f6f5c-1319-4e00-9583-646a0de1e439"
      },
      "source": [
        "#saving (depending on the further use both types of saving can be useful)\n",
        "\n",
        "embs=\"cbow_300_bigram_embeddings.txt\"\n",
        "model_cbow_bigram.wv.save_word2vec_format(embs, binary=False)\n",
        "\n",
        "embs_cbow_bigram = \"w2v_article_text_embeddings_cbow_e25_s300.model\"\n",
        "model_cbow_bigram.wv.save_word2vec_format(embs_cbow_bigram, binary = False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/smart_open/smart_open_lib.py:254: UserWarning: This function is deprecated, use smart_open.open instead. See the migration notes for details: https://github.com/RaRe-Technologies/smart_open/blob/master/README.rst#migrating-to-the-new-open-function\n",
            "  'See the migration notes for details: %s' % _MIGRATION_NOTES_URL\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pY6msUFZ4C5K",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 258
        },
        "outputId": "71d0c7b4-8585-4286-e956-355f6a73d4c6"
      },
      "source": [
        "print(model_cbow_bigram.wv.vectors.shape)\n",
        "model_cbow_bigram.wv.most_similar(positive=[\"good\"], topn = 10) "
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(64334, 300)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/gensim/matutils.py:737: FutureWarning: Conversion of the second argument of issubdtype from `int` to `np.signedinteger` is deprecated. In future, it will be treated as `np.int64 == np.dtype(int).type`.\n",
            "  if np.issubdtype(vec.dtype, np.int):\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('well', 0.5801551342010498),\n",
              " ('best', 0.5796145796775818),\n",
              " ('pretty_good', 0.5109511613845825),\n",
              " ('perfect', 0.5035198926925659),\n",
              " ('find_good', 0.48832467198371887),\n",
              " ('decent', 0.4844015836715698),\n",
              " ('great', 0.4740249812602997),\n",
              " ('bad', 0.4659029543399811),\n",
              " ('want', 0.45844513177871704),\n",
              " ('good_way', 0.4515240788459778)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 67
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "A64ylFdkuQNw",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 258
        },
        "outputId": "e01ab2aa-9fb8-4fd4-c55f-1b74d91f298b"
      },
      "source": [
        "print(model_cbow_bigram.wv.vectors.shape)\n",
        "model_cbow_bigram.wv.most_similar(positive=[\"ai\"], topn = 10) "
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(64334, 300)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/gensim/matutils.py:737: FutureWarning: Conversion of the second argument of issubdtype from `int` to `np.signedinteger` is deprecated. In future, it will be treated as `np.int64 == np.dtype(int).type`.\n",
            "  if np.issubdtype(vec.dtype, np.int):\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('artificial_intelligence', 0.7699136734008789),\n",
              " ('technology', 0.6802035570144653),\n",
              " ('ai_ml', 0.5772320032119751),\n",
              " ('new_technology', 0.531817615032196),\n",
              " ('cognitive_technology', 0.5301433801651001),\n",
              " ('advance_ai', 0.5288519859313965),\n",
              " ('automation', 0.5155724287033081),\n",
              " ('intelligent_machine', 0.510002613067627),\n",
              " ('ai_technologies', 0.4944221079349518),\n",
              " ('narrow_ai', 0.4883231222629547)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 66
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wGAbpJflyHsk",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "c3ef8f79-2072-4ac0-ebed-8077cac2e46c"
      },
      "source": [
        "epochs = 25\n",
        "\n",
        "t1 = time.perf_counter()\n",
        "model_cbow_bigram = Word2Vec(bigram_token,\n",
        "                 min_count = 5, # min_count means the frequency benchmark, if =2 and word is used only once - it's not included\n",
        "                 sg = 0, #trains for skip-gram (tut: CBOW by default, we have to add it to specify for skipgram)\n",
        "                 window = 5, #the size of context\n",
        "                 iter = epochs, #how many times the training code will run through the data set, same as epochs (first pass is to create dict)\n",
        "                 size = 150, #size of embedding\n",
        "                 workers = 4, #for parallel computing\n",
        "                 hs = 0) #0 for negative sampling; 1 for softmax function\n",
        "    \n",
        "t2 = time.perf_counter()\n",
        "t3_cbow_bigram = (t2-t1)/60\n",
        "print(model_cbow_bigram)\n",
        "print(\"The computation of the embedding for the cbow model took {:.2f} minutes.\".format(t3_cbow_bigram))\n",
        "\n",
        "#Word2Vec(vocab=56943, size=150, alpha=0.025)\n",
        "#The computation of the embedding for the cbow model took 20.29 minutes.\n",
        "\n",
        "#Word2Vec(vocab=56943, size=300, alpha=0.025)\n",
        "#The computation of the embedding for the cbow model took 22.09 minutes.\n",
        "\n",
        "#Word2Vec(vocab=64334, size=150, alpha=0.025)\n",
        "#The computation of the embedding for the cbow model took 23.69 minutes."
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Word2Vec(vocab=64334, size=150, alpha=0.025)\n",
            "The computation of the embedding for the cbow model took 23.69 minutes.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uZBGlzT9yHso",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 71
        },
        "outputId": "2931393f-9a97-4e9a-f177-c76808070af7"
      },
      "source": [
        "#saving\n",
        "embs=\"cbow_150_bigram_embeddings.txt\"\n",
        "model_cbow_bigram.wv.save_word2vec_format(embs, binary=False)\n",
        "\n",
        "embs_cbow_bigram = \"w2v_article_text_embeddings_cbow_e25_s150.model\"\n",
        "model_cbow_bigram.wv.save_word2vec_format(embs_cbow_bigram, binary = False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/smart_open/smart_open_lib.py:254: UserWarning: This function is deprecated, use smart_open.open instead. See the migration notes for details: https://github.com/RaRe-Technologies/smart_open/blob/master/README.rst#migrating-to-the-new-open-function\n",
            "  'See the migration notes for details: %s' % _MIGRATION_NOTES_URL\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6tqOiKuKyHfZ",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 258
        },
        "outputId": "c4759381-1506-47b8-e7ab-99660d9fb24c"
      },
      "source": [
        "print(model_cbow_bigram.wv.vectors.shape)\n",
        "model_cbow_bigram.wv.most_similar(positive=[\"good\"], topn = 10) "
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(64334, 150)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/gensim/matutils.py:737: FutureWarning: Conversion of the second argument of issubdtype from `int` to `np.signedinteger` is deprecated. In future, it will be treated as `np.int64 == np.dtype(int).type`.\n",
            "  if np.issubdtype(vec.dtype, np.int):\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('well', 0.6915244460105896),\n",
              " ('best', 0.6755548715591431),\n",
              " ('perfect', 0.6312260627746582),\n",
              " ('pretty_good', 0.6138462424278259),\n",
              " ('great', 0.5810209512710571),\n",
              " ('decent', 0.5731098651885986),\n",
              " ('better', 0.5647061467170715),\n",
              " ('bad', 0.5459110140800476),\n",
              " ('find_good', 0.5434296727180481),\n",
              " ('want', 0.5409783124923706)]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 70
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "c2A43Kv4kmsN"
      },
      "source": [
        "### Pre-trained Embeddings"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KNKT644sJj2K"
      },
      "source": [
        "As a pre-trained embedding we used GloVe which is trained by Stanford and publicly accessible. The training of GloVe is performed with an unsupervised machine model on \"aggregated global word-word co-occurrence statistics from a corpus, and the resulting representations showcase interesting linear substructures of the word vector space\" [6].\n",
        "<br>The corpus that we downloaded and used was trained on Wikipedia 2014 + Gigaword 5. It contains 400.000 tokens. We decided on the Wikipedia trained version since it is the most similar to the underlying data. It is a public blog where professionals and non-professionals published condensed information. The word embedding has 300 dimensions which is needed for comparability.\n",
        "<br> \n",
        "A drawback of this model might be that it is not trained on bigrams.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HhCTOtjYMrkK"
      },
      "source": [
        "glove_emb_index = {}\n",
        "\n",
        "file = open(os.getcwd() + \"/\" + \"glove.6B.300d.txt\", encoding = \"utf-8\")\n",
        "for line in file:\n",
        "  values = line.split()\n",
        "  word = values[0]\n",
        "  coefs = np.asarray(values[1:], dtype = \"float32\")\n",
        "  embeddings_index[word] = coefs\n",
        "file.close()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "u3sQB9dt7o_D"
      },
      "source": [
        "##Implementation of Prediction Models"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4vWrgmhEO1PW"
      },
      "source": [
        "In the following chapter we will implement an Elastic Net as a baseline model as a benchmark. Afterwards all word embeddings will be evaluate applied to a GRU architecture in order to decide with which word embedding we want to train our prediction models."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bkcGVEXaqP5I"
      },
      "source": [
        "**Checkpoint**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z5fUYTDmqMM-",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "outputId": "ccbd8f97-20cb-461e-ef2d-e9a5658232bc"
      },
      "source": [
        "#load new training data file\n",
        "with open(\"train_cleaned\", \"rb\") as path_name:\n",
        "  tr_cl = pickle.load(path_name)\n",
        "\n",
        "print(tr_cl.loc[10, \"text\"])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['artificial', 'intelligence', 'ai', 'impact', 'world', 'actually', 'ai', 'efiicte', 'world', 'good', 'way', 'efficient', 'way', 'go', 'to', 'kill', 'no', 'different', 'job', 'world', 'today', 'go', 'to', 'automate', 'open', 'open', 'no', 'opportunity', 'different', 'job', 'jobs', 'people', 'like', 'current', 'world', 'save', 'complex', 'statistic', 'mathematics', 'didn', 'ai', 'ai', 'get', 'new', 'feature', 'like', 'machine', 'talk', 'like', 'human', 'someday', 'machine', 'not', 'machine', 'human', 'like', 'solve', 'no', 'current', 'problem', 'fear', 'ai', 'apoocylapse', 'not', 'usedsny', 'command', 'line', 'server', 'language', 'programmer', 'know', 'machine', 'way', 'dumb', 'human']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8rKhTe5xQf7z"
      },
      "source": [
        "###Data Preparation for Prediction"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MJmTepsc7wf3"
      },
      "source": [
        "#LOAD data\n",
        "X = tr_cl.loc[:,[\"text\"]] \n",
        "#X = tr_cl.drop([\"totalClapCount\"], axis = 1) \n",
        "y = tr_cl.loc[:,[\"totalClapCount\"]]\n",
        "#X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)#, random_state=123) \n",
        "#test data are 20 % of the full data set\n",
        "#the seed 123 is provided for comparability of the results since I will test different approaches"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IHU5AZk18XS1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 71
        },
        "outputId": "97cff159-e62b-40b2-d994-76a86aa1de13"
      },
      "source": [
        "print(X.shape)\n",
        "print(y.shape)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(66380, 1)\n",
            "(66380, 1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "i9BoH9jvnKvD"
      },
      "source": [
        "#split data\n",
        "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=123) #random state for comparability"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0oYFKI4OL6uu",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 71
        },
        "outputId": "98cd4118-6397-40e6-94b1-5fa6dc8dea3e"
      },
      "source": [
        "#train\n",
        "\n",
        "#tokenize texts to sequence (NNs only take numerical values as inputs and no strings)\n",
        "tokenizer_obj_train = Tokenizer() #vectorizes the text corpus\n",
        "tokenizer_obj_train.fit_on_texts(X_train[\"text\"])\n",
        "seq_train = tokenizer_obj_train.texts_to_sequences(X_train[\"text\"])\n",
        "\n",
        "max_text_len = 400\n",
        "\n",
        "#the length of the word index holds the number of unique tokens in the data set\n",
        "word_index = tokenizer_obj_train.word_index\n",
        "print(\"Unique tokens: {:,.0f} \".format(len(word_index)))\n",
        "word_index_train = word_index\n",
        "\n",
        "#\"filling up\" embeddings with a text length less than the max text length of 427 with 0s\n",
        "X_train = pad_sequences(seq_train, max_text_len) \n",
        "print(X_train.shape) #(number of text (rows), number of max length of a sequence)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Unique tokens: 205,741 \n",
            "(53104, 400)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BBuZ3U6poQ7e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 71
        },
        "outputId": "f06172a3-a44c-42a6-a68c-96838ef393ce"
      },
      "source": [
        "#test\n",
        "\n",
        "#tokenize texts to sequence (NNs only take numerical values as inputs and no strings)\n",
        "tokenizer_obj_test = Tokenizer() #vectorizes the text corpus\n",
        "tokenizer_obj_test.fit_on_texts(X_test[\"text\"])\n",
        "seq_test = tokenizer_obj_test.texts_to_sequences(X_test[\"text\"])\n",
        "\n",
        "#the length of the word index holds the number of unique tokens in the data set\n",
        "word_index_test = tokenizer_obj_test.word_index  #word_index contains vector words (shape: 1 x number of unique tokens)\n",
        "print(\"Unique tokens: {:,.0f} \".format(len(word_index_test)))\n",
        "\n",
        "#\"filling up\" embeddings with a text length less than the max text length of 427 with 0s\n",
        "X_test = pad_sequences(seq_test, max_text_len) \n",
        "print(X_test.shape) #(number of text (rows), number of max length of a sequence)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Unique tokens: 93,945 \n",
            "(13276, 400)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pd_DMQszP1P4"
      },
      "source": [
        "As already stated in the first part of the analysis, the target variable contains extreme values. Therefore, we decided to make a commom outlier cut at 3*Interquartile-Range (IQR). Hence, our model will not be able to calculate a huge number of claps but will train better and more common observations."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kGsEtLhA43LG",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 292
        },
        "outputId": "2c14f72b-7bcd-42d9-c4c3-944706438103"
      },
      "source": [
        "Q1 = np.percentile(y_train[\"totalClapCount\"], 25, interpolation = 'midpoint') \n",
        "Q3 = np.percentile(y_train[\"totalClapCount\"], 75, interpolation = 'midpoint') \n",
        "IQR = Q3 - Q1\n",
        "cutoff = IQR * 3\n",
        "claps = np.clip(y_train[\"totalClapCount\"], 0, cutoff)\n",
        "y_train[\"totalClapCount\"] = claps\n",
        "y_train[\"totalClapCount\"].describe()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:6: SettingWithCopyWarning: \n",
            "A value is trying to be set on a copy of a slice from a DataFrame.\n",
            "Try using .loc[row_indexer,col_indexer] = value instead\n",
            "\n",
            "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
            "  \n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "count    53104.000000\n",
              "mean        46.330521\n",
              "std         68.312569\n",
              "min          0.000000\n",
              "25%          0.000000\n",
              "50%          6.000000\n",
              "75%         65.000000\n",
              "max        195.000000\n",
              "Name: totalClapCount, dtype: float64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "X6qMQhx5rGGr"
      },
      "source": [
        "**Other Variables**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VOUv0XAUrE-K"
      },
      "source": [
        "X = tr_cl.loc[:,[\"author\", \"responsesCreatedCount\", \"totalClapCount\"]] #the target variable totalClapCount is needed in order to compute the target encoding and will be removed later on"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cBh7jQFEsRxk"
      },
      "source": [
        "X_train_var, X_test_var = train_test_split(X, test_size=0.2, random_state=123)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "x1OQNDKV8S9I",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 241
        },
        "outputId": "d20a4f48-a507-41f2-defc-755406716a26"
      },
      "source": [
        "#target encoding\n",
        "#https://maxhalford.github.io/blog/target-encoding/\n",
        "\n",
        "X_train_var[\"author\"] = X_train_var[\"author\"].map(X_train_var.groupby(\"author\")[\"totalClapCount\"].mean())\n",
        "X_test_var[\"author\"] = X_test_var[\"author\"].map(X_test_var.groupby(\"author\")[\"totalClapCount\"].mean())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:4: SettingWithCopyWarning: \n",
            "A value is trying to be set on a copy of a slice from a DataFrame.\n",
            "Try using .loc[row_indexer,col_indexer] = value instead\n",
            "\n",
            "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
            "  after removing the cwd from sys.path.\n",
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:5: SettingWithCopyWarning: \n",
            "A value is trying to be set on a copy of a slice from a DataFrame.\n",
            "Try using .loc[row_indexer,col_indexer] = value instead\n",
            "\n",
            "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
            "  \"\"\"\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "o6BEKEiu9OmS"
      },
      "source": [
        "X_train_var = X_train_var.drop(columns = \"totalClapCount\")\n",
        "X_test_var = X_test_var.drop(columns = \"totalClapCount\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KETMyddmxY0e"
      },
      "source": [
        "X_train_var = X_train_var[[\"author\", \"responsesCreatedCount\"]].values\n",
        "X_test_var = X_test_var[[\"author\", \"responsesCreatedCount\"]].values"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mIOHUAKo-ey9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "outputId": "52ff05a5-1cf2-419e-94ce-78060bfd45ff"
      },
      "source": [
        "X_train_var.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(53104, 2)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1bIMM8aIuyeY"
      },
      "source": [
        "scale = StandardScaler().fit(X_train_var)\n",
        "\n",
        "X_train_var_scaled = scale.transform(X_train_var)\n",
        "X_test_var_scaled = scale.transform(X_test_var)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-T3-aGhu3vRN"
      },
      "source": [
        "#Input for models\n",
        "X_train = np.concatenate((X_train, X_train_var_scaled), axis=1)\n",
        "X_test = np.concatenate((X_test, X_test_var_scaled), axis=1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "s6z99J3vxNXi",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "outputId": "f5b6afc9-4d3f-44c3-fdf1-bc7de64a4a56"
      },
      "source": [
        "print(X_train.shape, X_test.shape)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(53104, 402) (13276, 402)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7Mzcf_8oOrFR"
      },
      "source": [
        "###Benchmark Model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bDWKjoJSRh7w"
      },
      "source": [
        "As a model to benchmark the performance of the other models we choose an elastic net. Elastic Net is linear model but it also applies regularization. It can be a mixture of Lasso and Ridge depending on the parameter l1_ratio. If it is 1 than an Elastic Net is equal to Lasso and with a l1_ratio equal to 0, it is a Ridge Model. Elastic Net is suitable if the dimension are high and the assumption of unnecessary features. [7]"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7-smdfZ_JoXD"
      },
      "source": [
        "models = {}"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KA7r8M7MTvzV"
      },
      "source": [
        "We applied parameter tuning with grid search in order to optimize the basic model since it will compete with more sophisticated models later on. To reduce the chance of overfitting, a 10-fold crossvalidation is performed."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_C2E-oF5CUKz"
      },
      "source": [
        "#tune parameters of model with grid search\n",
        "parametersGrid = {\"max_iter\": [1, 5, 10],\n",
        "                  \"alpha\": [0.0001, 0.001, 0.01, 0.1, 1, 10, 100],\n",
        "                  \"l1_ratio\": np.arange(0.0, 1.0, 0.1)} #l1-ration: 0 = ridge, 1 = lasso\n",
        "\n",
        "model = ElasticNet(tol = 1) #otherwise: \"ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations\"\n",
        "grid = GridSearchCV(model, parametersGrid, cv=10)\n",
        "grid = grid.fit(X_train, y_train)\n",
        "y_pred = grid.predict(X_test)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DwcYDqEidG-S",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 297
        },
        "outputId": "3b3ae761-48d5-4572-97e9-c4ac38d2ba25"
      },
      "source": [
        "x_ax = range(len(X_test))\n",
        "plt.scatter(x_ax, y_test, s=5, color=\"blue\", label=\"original\")\n",
        "plt.plot(x_ax, y_pred, lw=0.8, color=\"red\", label=\"predicted\")\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1296x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SM3Xx4wLTOOt",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "dbc40fba-3bef-4c6c-a091-11522c7e6a48"
      },
      "source": [
        "grid.best_params_"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'alpha': 1, 'l1_ratio': 0.7000000000000001, 'max_iter': 5}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 345
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nvfnrc3PE95Y"
      },
      "source": [
        "mse_en_grid = mean_squared_error(y_test, y_pred)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yb8PA_Z8RV5Y"
      },
      "source": [
        "models.update({\"ElastricNet_Grid\" : mse_en_grid})"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vmz91zVrTb17",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "6d77b3e8-cfdf-49f2-d1da-91faf035ec00"
      },
      "source": [
        "models"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'ElastricNet_Grid': 1243992.2881074175}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 350
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IMBlYVuUi_Aw"
      },
      "source": [
        "###Evaluation of Word Embeddings\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0U1gjmjgWwoV"
      },
      "source": [
        "####Preparation for Evaluation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HpheW0dtPeK2"
      },
      "source": [
        "**Loading Functions**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "l_BJDTKdQ9V-"
      },
      "source": [
        "def load_embedding(file):\n",
        "  \n",
        "    embeddings_index = {}\n",
        "    file = open(file, encoding = \"utf-8\")\n",
        "    for line in file:\n",
        "        values = line.split()\n",
        "        word = values[0]\n",
        "        coefs = np.asarray(values[1:], dtype = \"float32\")\n",
        "        embeddings_index[word] = coefs\n",
        "    file.close()\n",
        "\n",
        "    return embeddings_index"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Mg64w5DSlXrc"
      },
      "source": [
        "An embedding matrix functions as a lookup table for input words. The embedding of a word , e.g. \"data\", has an index i which can be \"looked up\" in the $i^{th}$ row of the embedding matrix. Hence, we match the word indexes with the embedding matrix. An embedding matrix can be used as input for the training of a neural network. <p>The code of the following function is inspired by the [code]({https://github.com/stefanlessmann/adams/blob/master/exercises/Ex11_GRU_GloVe/Ex11%20sentiment_analysis.ipynb}) given in lecture 11 of the ADAM course by Prof. Lessmann (HU zu Berlin). "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BgnZexYTR0KB"
      },
      "source": [
        "#word_index = word_index from tokenizer (it the same for all following NN-models)\n",
        "#embedding_index = loaded word embedding\n",
        "#dim = size of word embedding\n",
        "#num_words should be the length of the word index + 1 for collection the unknown words which will be represented by 0s in the embedding matrix\n",
        "\n",
        "def generate_emb_mat(word_index, embedding_index, dim):\n",
        "    \n",
        "    num_words = len(word_index) + 1\n",
        "    embedding_matrix = np.zeros((num_words, dim))\n",
        "\n",
        "    for word, i in word_index.items():\n",
        "        if i > num_words:\n",
        "            continue\n",
        "      \n",
        "        try:\n",
        "            embedding_vector = embedding_index.get(word)\n",
        "        except:\n",
        "            embedding_vector = embedding_index.word_vectors[embedding_index.dictionary[word]]\n",
        "        if embedding_vector is not None:\n",
        "            embedding_matrix[i] = embedding_vector\n",
        "        \n",
        "    return embedding_matrix"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LIAs94YKTjQx"
      },
      "source": [
        "def plot_hist(model_hist):\n",
        "    plt.figure(figsize=(15, 8))\n",
        "    plt.subplot(1, 2, 1)\n",
        "    \n",
        "    #show training loss\n",
        "    plt.plot(model_hist.history[\"loss\"], color=\"green\")\n",
        "    plt.title(\"model loss\")\n",
        "    plt.ylabel(\"MSE\")\n",
        "    plt.xlabel(\"epoch\")\n",
        "    plt.legend([\"train\"], loc=\"upper right\")\n",
        "\n",
        "    #show validation loss\n",
        "    plt.subplot(1, 2, 2)\n",
        "    plt.plot(model_hist.history[\"val_loss\"], color = \"orange\")\n",
        "    plt.title(\"validation loss\")\n",
        "    plt.ylabel(\"MSE\")\n",
        "    plt.xlabel(\"epoch\")\n",
        "    plt.legend([\"val\"], loc=\"upper right\")\n",
        "    plt.subplots_adjust(bottom=0.25, top=0.75)\n",
        "\n",
        "    return plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qZi_OQB_7-h5"
      },
      "source": [
        "#Method/Function for training a GRU model and making a prediction (besides calling the function for generating the embedding matrix)\n",
        "#No trainable features\n",
        "\n",
        "#dim = size of the embedding (150 or 300)\n",
        "#emb_mat = generated embedding matrix\n",
        "#name = for naming the produced scores etc. (e.g. cbow150)\n",
        "#word_index = needed for the gen_embed_matrix function\n",
        "\n",
        "def GRU_embed_models(embedding_index, name, word_index, dim):\n",
        "  \n",
        "  emb_mat = generate_emb_mat(word_index, embedding_index, dim) \n",
        "  \n",
        "  print(\"Training the GRU with a pretrained {}-embedding\".format(name))\n",
        "\n",
        "  #embedding layer\n",
        "  embedding_layer = Embedding(input_dim = no_of_words,\n",
        "                            output_dim = dim,\n",
        "                            input_length = max_text_len,\n",
        "                            embeddings_initializer = Constant(emb_mat),\n",
        "                            trainable = False)\n",
        "\n",
        "  #GRU set up\n",
        "  model = Sequential()\n",
        "  model.add(embedding_layer)\n",
        "  model.add(GRU(100))\n",
        "  model.add(Dense(1, activation = \"relu\"))\n",
        "  model.compile(loss = \"mse\", optimizer = \"adam\")#, metrics = [\"mae\"])\n",
        "  model.summary()\n",
        "\n",
        "  #train\n",
        "  model_hist_GRU = model.fit(X_train, y_train, validation_split = val_split, epochs= epochs, batch_size = batch_size) #validation_data=(x_val, y_val)\n",
        "\n",
        "  #save weights\n",
        "  GRU_w_up = model.layers[1].get_weights()\n",
        "  model.save_weights(\"GRU_\" + name + \"_10ep_64ba.h5\")\n",
        "  print(\"Saving successful\")\n",
        "\n",
        "  #show\n",
        "  plot_hist(model_hist_GRU)\n",
        "\n",
        "  #evaluate and predict\n",
        "  scores = model.evaluate(X_test, y_test, verbose = 0)\n",
        "  #print(\"The loss is: {}\".format(scores[0])) #loss and mae\n",
        "  #print(\"The MAE is: {:.2f}\".format(scores[1])) \n",
        "  y_pred = model.predict(X_test)\n",
        "\n",
        "  GRU_models.update({name + \" MSE\": mean_squared_error(y_test, y_pred)})\n",
        "  GRU_models_hist.update({name: model_hist_GRU.history})\n",
        "\n",
        "  return scores, model_hist_GRU.history"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ugt_7sQtRCRD"
      },
      "source": [
        "#loading pretrained embedding matrices\n",
        "cbow150_emb_index = load_embedding(os.getcwd() + \"/\" + \"cbow_150_bigram_embeddings.txt\")\n",
        "cbow300_emb_index = load_embedding(os.getcwd() + \"/\" + \"cbow_300_bigram_embeddings.txt\")\n",
        "sg150_emb_index = load_embedding(os.getcwd() + \"/\" + \"sg_150_bigram_embeddings.txt\")\n",
        "sg300_emb_index = load_embedding(os.getcwd() + \"/\" + \"sg_300_bigram_embeddings.txt\")\n",
        "\n",
        "glove_emb_index = load_embedding(os.getcwd() + \"/\" + \"glove.6B.300d.txt\")\n",
        "#not working this way: news_emb_index = load_embedding(os.getcwd() + \"/\" + \"GoogleNews-vectors-negative300.bin\")\n",
        "\n",
        "#print(\"The embedding contains {} word vectors\".format(len(cbow_emb_index)))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "A6B9nWW94PDm",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "0aa3d213-b1f1-47ca-9dcf-cba424b7062b"
      },
      "source": [
        "print(len(cbow150_emb_index))\n",
        "print(len(cbow300_emb_index))\n",
        "print(len(sg150_emb_index))\n",
        "print(len(sg300_emb_index))\n",
        "print(len(glove_emb_index))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "64335\n",
            "64335\n",
            "64335\n",
            "64335\n",
            "400000\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yryNtV8xmCRE"
      },
      "source": [
        "With help of the parameter \"embeddings_initializer\" and the keras function \"Constant\", an pretrained embedding matrix can be used as an input for the embedding layer. If not, embedding weights are initialized. The function \"embedding\" assigns by default randomly weights to the given words by using an uniform distribution. The parameter \"trainable\" controlls which weights will be trained and which will be frozen. Setting this parameter to false speeds up training since less parameters have to be trained (see model summary non-trainable parameters) and the problem of a vanishing gradient is less present. Nevertheless, the trained models performed better with updating the weights during training.<p>\n",
        "In preparation for this model, the weights of the pretrained cbow word embedding matrix were trained on the model. This updated weights are used in the following model in order to fine-tune the results of the GRU model."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "K0V9dP8YWMMu"
      },
      "source": [
        "**Set Parameters**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "l76Q02_JMEhB"
      },
      "source": [
        "#set parameters\n",
        "\n",
        "no_of_words = len(tokenizer_obj_train.word_index)+1 #number of unique words in the embedding plus one for unknown tokens\n",
        "max_text_len = 400 #take only the first 400 words per article\n",
        "nb_hidden = 100\n",
        "epochs = 10\n",
        "batch_size = 128\n",
        "val_split = 0.25"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-jEZ5354LxLd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "fa42db09-6d1c-464c-f5bf-c893d813bb2c"
      },
      "source": [
        "no_of_words"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "205742"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "795sbLNiMQ2j"
      },
      "source": [
        "####Application of Word Embeddings"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7PYtcfWdXnmc"
      },
      "source": [
        "All word embeddings will be trained on a non-trainable GRU architecture with 10 epochs. Even though 10 epochs is not very big, it is sufficient for the task of evaluation the different performances of the word embeddings because the validation losses stop dropping after this period for all word embeddings<br> The MSE of each model is summed up in the subchapter results."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QMju3r97BtJh"
      },
      "source": [
        "GRU_models = {}\n",
        "GRU_models_hist = {}"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XxanS8sK8n0D",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "f4107cb7-9383-4c87-8e57-089e7be586ce"
      },
      "source": [
        "#cbow150\n",
        "GRU_embed_models(cbow150_emb_index, \"cbow150\", word_index, 150)\n",
        "#pretrained w2v\n",
        "#embed_models_notrain(cbow150_emb_index, \"cbow150\", word_index, 150)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training the GRU with a pretrained cbow150-embedding\n",
            "Model: \"sequential_10\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_10 (Embedding)     (None, 400, 150)          34813350  \n",
            "_________________________________________________________________\n",
            "gru_8 (GRU)                  (None, 100)               75600     \n",
            "_________________________________________________________________\n",
            "dense_8 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 34,889,051\n",
            "Trainable params: 75,701\n",
            "Non-trainable params: 34,813,350\n",
            "_________________________________________________________________\n",
            "Epoch 1/10\n",
            "623/623 [==============================] - 303s 487ms/step - loss: 4966.9668 - val_loss: 4407.6982\n",
            "Epoch 2/10\n",
            "623/623 [==============================] - 307s 493ms/step - loss: 4072.6179 - val_loss: 3767.0737\n",
            "Epoch 3/10\n",
            "623/623 [==============================] - 304s 488ms/step - loss: 3691.6062 - val_loss: 3620.1265\n",
            "Epoch 4/10\n",
            "623/623 [==============================] - 303s 487ms/step - loss: 3479.7461 - val_loss: 3600.7480\n",
            "Epoch 5/10\n",
            "623/623 [==============================] - 304s 488ms/step - loss: 3316.1521 - val_loss: 3488.9421\n",
            "Epoch 6/10\n",
            "623/623 [==============================] - 304s 489ms/step - loss: 3166.0056 - val_loss: 3477.9316\n",
            "Epoch 7/10\n",
            "623/623 [==============================] - 307s 493ms/step - loss: 3003.7610 - val_loss: 3502.8896\n",
            "Epoch 8/10\n",
            "623/623 [==============================] - 309s 496ms/step - loss: 2835.7971 - val_loss: 3440.2314\n",
            "Epoch 9/10\n",
            "623/623 [==============================] - 313s 503ms/step - loss: 2700.7927 - val_loss: 3522.3511\n",
            "Epoch 10/10\n",
            "623/623 [==============================] - 310s 497ms/step - loss: 2544.5322 - val_loss: 3572.9309\n",
            "Saving successful\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1246689.75,\n",
              " {'loss': [4966.966796875,\n",
              "   4072.617919921875,\n",
              "   3691.606201171875,\n",
              "   3479.74609375,\n",
              "   3316.152099609375,\n",
              "   3166.005615234375,\n",
              "   3003.760986328125,\n",
              "   2835.797119140625,\n",
              "   2700.792724609375,\n",
              "   2544.5322265625],\n",
              "  'val_loss': [4407.6982421875,\n",
              "   3767.07373046875,\n",
              "   3620.12646484375,\n",
              "   3600.748046875,\n",
              "   3488.942138671875,\n",
              "   3477.931640625,\n",
              "   3502.8896484375,\n",
              "   3440.2314453125,\n",
              "   3522.35107421875,\n",
              "   3572.930908203125]})"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 54
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DPN4hgLHdM60",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "3d57dfc7-1e7f-4111-84ac-059cd63e7fce"
      },
      "source": [
        "#cbow300\n",
        "GRU_embed_models(cbow300_emb_index, \"cbow300\", word_index, 300)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training the GRU with a pretrained cbow300-embedding\n",
            "Model: \"sequential_11\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_11 (Embedding)     (None, 400, 300)          69626700  \n",
            "_________________________________________________________________\n",
            "gru_9 (GRU)                  (None, 100)               120600    \n",
            "_________________________________________________________________\n",
            "dense_9 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 69,747,401\n",
            "Trainable params: 120,701\n",
            "Non-trainable params: 69,626,700\n",
            "_________________________________________________________________\n",
            "Epoch 1/10\n",
            "623/623 [==============================] - 415s 666ms/step - loss: 4949.6353 - val_loss: 4275.0791\n",
            "Epoch 2/10\n",
            "623/623 [==============================] - 419s 673ms/step - loss: 4027.7305 - val_loss: 3720.2024\n",
            "Epoch 3/10\n",
            "623/623 [==============================] - 416s 667ms/step - loss: 3604.8821 - val_loss: 3595.9099\n",
            "Epoch 4/10\n",
            "623/623 [==============================] - 416s 668ms/step - loss: 3342.7695 - val_loss: 3496.8911\n",
            "Epoch 5/10\n",
            "623/623 [==============================] - 413s 663ms/step - loss: 3110.2046 - val_loss: 3480.9011\n",
            "Epoch 6/10\n",
            "623/623 [==============================] - 416s 667ms/step - loss: 2879.8953 - val_loss: 3465.9194\n",
            "Epoch 7/10\n",
            "623/623 [==============================] - 416s 667ms/step - loss: 2649.1748 - val_loss: 3522.8516\n",
            "Epoch 8/10\n",
            "623/623 [==============================] - 413s 662ms/step - loss: 2446.5684 - val_loss: 3710.0400\n",
            "Epoch 9/10\n",
            "623/623 [==============================] - 412s 661ms/step - loss: 2270.5903 - val_loss: 3677.7524\n",
            "Epoch 10/10\n",
            "623/623 [==============================] - 413s 664ms/step - loss: 2101.2161 - val_loss: 3682.2151\n",
            "Saving successful\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1244447.625,\n",
              " {'loss': [4949.63525390625,\n",
              "   4027.73046875,\n",
              "   3604.882080078125,\n",
              "   3342.76953125,\n",
              "   3110.20458984375,\n",
              "   2879.895263671875,\n",
              "   2649.1748046875,\n",
              "   2446.568359375,\n",
              "   2270.59033203125,\n",
              "   2101.216064453125],\n",
              "  'val_loss': [4275.0791015625,\n",
              "   3720.202392578125,\n",
              "   3595.909912109375,\n",
              "   3496.89111328125,\n",
              "   3480.901123046875,\n",
              "   3465.91943359375,\n",
              "   3522.8515625,\n",
              "   3710.0400390625,\n",
              "   3677.75244140625,\n",
              "   3682.215087890625]})"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 55
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8GsIGkROdOw0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "b365ac22-01c3-463d-efab-44d004a62a3f"
      },
      "source": [
        "#sg150\n",
        "GRU_embed_models(sg150_emb_index, \"sg150\", word_index, 150)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training the GRU with a pretrained sg150-embedding\n",
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding (Embedding)        (None, 400, 150)          34813350  \n",
            "_________________________________________________________________\n",
            "gru (GRU)                    (None, 100)               75600     \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 34,889,051\n",
            "Trainable params: 75,701\n",
            "Non-trainable params: 34,813,350\n",
            "_________________________________________________________________\n",
            "Epoch 1/10\n",
            "623/623 [==============================] - 300s 481ms/step - loss: 5012.7129 - val_loss: 4568.1396\n",
            "Epoch 2/10\n",
            "623/623 [==============================] - 305s 490ms/step - loss: 4700.3833 - val_loss: 4564.0327\n",
            "Epoch 3/10\n",
            "623/623 [==============================] - 300s 482ms/step - loss: 5289.3315 - val_loss: 6609.2632\n",
            "Epoch 4/10\n",
            "623/623 [==============================] - 306s 491ms/step - loss: 6127.4756 - val_loss: 5966.9014\n",
            "Epoch 5/10\n",
            "623/623 [==============================] - 305s 490ms/step - loss: 6125.5317 - val_loss: 6154.1743\n",
            "Epoch 6/10\n",
            "623/623 [==============================] - 306s 492ms/step - loss: 6832.3984 - val_loss: 6622.2310\n",
            "Epoch 7/10\n",
            "623/623 [==============================] - 306s 492ms/step - loss: 6872.3276 - val_loss: 6621.4883\n",
            "Epoch 8/10\n",
            "623/623 [==============================] - 308s 494ms/step - loss: 6407.4746 - val_loss: 5773.8091\n",
            "Epoch 9/10\n",
            "623/623 [==============================] - 309s 496ms/step - loss: 6834.3467 - val_loss: 6618.2573\n",
            "Epoch 10/10\n",
            "623/623 [==============================] - 309s 496ms/step - loss: 6270.8364 - val_loss: 5740.6426\n",
            "Saving successful\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1263030.0,\n",
              " {'loss': [5012.712890625,\n",
              "   4700.38330078125,\n",
              "   5289.33154296875,\n",
              "   6127.4755859375,\n",
              "   6125.53173828125,\n",
              "   6832.3984375,\n",
              "   6872.32763671875,\n",
              "   6407.474609375,\n",
              "   6834.3466796875,\n",
              "   6270.83642578125],\n",
              "  'val_loss': [4568.1396484375,\n",
              "   4564.03271484375,\n",
              "   6609.26318359375,\n",
              "   5966.9013671875,\n",
              "   6154.17431640625,\n",
              "   6622.23095703125,\n",
              "   6621.48828125,\n",
              "   5773.80908203125,\n",
              "   6618.25732421875,\n",
              "   5740.642578125]})"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 25
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "75WFMuHbdQd2",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "4238968d-38f9-4733-fc83-d708c438f1ea"
      },
      "source": [
        "#sg300\n",
        "GRU_embed_models(sg300_emb_index, \"sg300\", word_index, 300)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training the GRU with a pretrained sg300-embedding\n",
            "Model: \"sequential_1\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_1 (Embedding)      (None, 400, 300)          69626700  \n",
            "_________________________________________________________________\n",
            "gru_1 (GRU)                  (None, 100)               120600    \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 69,747,401\n",
            "Trainable params: 120,701\n",
            "Non-trainable params: 69,626,700\n",
            "_________________________________________________________________\n",
            "Epoch 1/10\n",
            "623/623 [==============================] - 424s 680ms/step - loss: 4977.9497 - val_loss: 4567.2534\n",
            "Epoch 2/10\n",
            "623/623 [==============================] - 416s 668ms/step - loss: 4694.2432 - val_loss: 4514.7583\n",
            "Epoch 3/10\n",
            "623/623 [==============================] - 416s 667ms/step - loss: 4222.0552 - val_loss: 3750.3000\n",
            "Epoch 4/10\n",
            "623/623 [==============================] - 414s 665ms/step - loss: 3661.6394 - val_loss: 3642.4980\n",
            "Epoch 5/10\n",
            "623/623 [==============================] - 415s 666ms/step - loss: 3464.2727 - val_loss: 3469.9023\n",
            "Epoch 6/10\n",
            "623/623 [==============================] - 415s 666ms/step - loss: 3311.0901 - val_loss: 3421.1968\n",
            "Epoch 7/10\n",
            "623/623 [==============================] - 415s 666ms/step - loss: 3172.8096 - val_loss: 3409.3667\n",
            "Epoch 8/10\n",
            "623/623 [==============================] - 414s 664ms/step - loss: 3023.7729 - val_loss: 3461.3772\n",
            "Epoch 9/10\n",
            "623/623 [==============================] - 417s 669ms/step - loss: 2873.4075 - val_loss: 3447.5159\n",
            "Epoch 10/10\n",
            "623/623 [==============================] - 412s 661ms/step - loss: 2721.0952 - val_loss: 3453.1865\n",
            "Saving successful\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1247335.875,\n",
              " {'loss': [4977.94970703125,\n",
              "   4694.2431640625,\n",
              "   4222.05517578125,\n",
              "   3661.639404296875,\n",
              "   3464.272705078125,\n",
              "   3311.090087890625,\n",
              "   3172.8095703125,\n",
              "   3023.77294921875,\n",
              "   2873.407470703125,\n",
              "   2721.09521484375],\n",
              "  'val_loss': [4567.25341796875,\n",
              "   4514.75830078125,\n",
              "   3750.300048828125,\n",
              "   3642.498046875,\n",
              "   3469.90234375,\n",
              "   3421.19677734375,\n",
              "   3409.36669921875,\n",
              "   3461.377197265625,\n",
              "   3447.515869140625,\n",
              "   3453.1865234375]})"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tLRhkK2xdRw0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "a532271f-74b3-42c7-decc-7f063851cf41"
      },
      "source": [
        "#glove300\n",
        "GRU_embed_models(glove_emb_index, \"glove300\", word_index, 300)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training the GRU with a pretrained glove300-embedding\n",
            "Model: \"sequential_2\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_2 (Embedding)      (None, 400, 300)          69626700  \n",
            "_________________________________________________________________\n",
            "gru_2 (GRU)                  (None, 100)               120600    \n",
            "_________________________________________________________________\n",
            "dense_2 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 69,747,401\n",
            "Trainable params: 120,701\n",
            "Non-trainable params: 69,626,700\n",
            "_________________________________________________________________\n",
            "Epoch 1/10\n",
            "623/623 [==============================] - 415s 667ms/step - loss: 4979.5283 - val_loss: 4557.9258\n",
            "Epoch 2/10\n",
            "623/623 [==============================] - 414s 664ms/step - loss: 4574.3379 - val_loss: 4127.1904\n",
            "Epoch 3/10\n",
            "623/623 [==============================] - 414s 664ms/step - loss: 4099.1084 - val_loss: 3938.1995\n",
            "Epoch 4/10\n",
            "623/623 [==============================] - 413s 663ms/step - loss: 3889.6648 - val_loss: 3795.8240\n",
            "Epoch 5/10\n",
            "623/623 [==============================] - 412s 661ms/step - loss: 3726.0806 - val_loss: 3699.2007\n",
            "Epoch 6/10\n",
            "623/623 [==============================] - 411s 660ms/step - loss: 3583.2593 - val_loss: 3638.4941\n",
            "Epoch 7/10\n",
            "623/623 [==============================] - 412s 661ms/step - loss: 3439.6392 - val_loss: 3625.0691\n",
            "Epoch 8/10\n",
            "623/623 [==============================] - 411s 660ms/step - loss: 3286.0818 - val_loss: 3573.7734\n",
            "Epoch 9/10\n",
            "623/623 [==============================] - 412s 662ms/step - loss: 3128.7041 - val_loss: 3584.5615\n",
            "Epoch 10/10\n",
            "623/623 [==============================] - 415s 667ms/step - loss: 2964.7910 - val_loss: 3807.4805\n",
            "Saving successful\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1245252.625,\n",
              " {'loss': [4979.5283203125,\n",
              "   4574.337890625,\n",
              "   4099.1083984375,\n",
              "   3889.664794921875,\n",
              "   3726.08056640625,\n",
              "   3583.25927734375,\n",
              "   3439.63916015625,\n",
              "   3286.081787109375,\n",
              "   3128.7041015625,\n",
              "   2964.791015625],\n",
              "  'val_loss': [4557.92578125,\n",
              "   4127.1904296875,\n",
              "   3938.199462890625,\n",
              "   3795.823974609375,\n",
              "   3699.20068359375,\n",
              "   3638.494140625,\n",
              "   3625.069091796875,\n",
              "   3573.7734375,\n",
              "   3584.5615234375,\n",
              "   3807.48046875]})"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "n7VOporhXA0X"
      },
      "source": [
        "####Results"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6rO5qOo1XJrN"
      },
      "source": [
        "The word embedding model Cbow with 300 dimension performs the best. The GloVe is the second best. Even if the GloVe was trained on more data, the Cbow is more fitted on the underlying data. Moreover, it can be assumed that the implementation of bigram has a positive effect. The skip-gram model with 150 dimension performs by far the worst. It also shows big jumps in the train and validation history."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Xvr7Fw0v84H0"
      },
      "source": [
        "\\begin{array}{lll}\n",
        "Word \\: Embedding \\: Model & Dimension & MSE \\: on \\: Test Set\\\\\n",
        "\\hline\n",
        "Cbow & 150 & 1,246,689.75 \\\\\n",
        "Cbow & 300 & \\bf{1,244,447.63} \\\\\n",
        "Skip-gram & 150 & 1,263,030.00 \\\\\n",
        "Skip-gram & 300 & 1,247,335.88 \\\\\n",
        "GloVe & 300 & 1,245,252.63 \\\\\n",
        "\\end{array}"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5XNii85RjWxk",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 491
        },
        "outputId": "e745ec23-73c4-4372-890f-b5af98e62b1f"
      },
      "source": [
        "plt.figure(figsize=(15, 8))\n",
        "\n",
        "#show loss\n",
        "plt.subplot(1, 2, 1)\n",
        "plt.plot(GRU_models_hist['sg150']['loss'], color=\"green\")\n",
        "plt.plot(GRU_models_hist['sg300']['loss'], color=\"red\")\n",
        "plt.plot(GRU_models_hist['cbow300']['loss'], color=\"purple\")\n",
        "plt.plot(GRU_models_hist['cbow150']['loss'], color=\"orange\")\n",
        "plt.plot(GRU_models_hist['glove300']['loss'], color=\"blue\")\n",
        "plt.legend(labels = (\"Skip-gram 150\", \"Skip-gram 300\", \"Cbow 150\", \"Cbow 300\", \"GloVe 300\"), loc = \"upper left\")\n",
        "plt.title(\"model loss\")\n",
        "plt.ylabel(\"loss\")\n",
        "plt.xlabel(\"epoch\")\n",
        "\n",
        "#show validation loss\n",
        "plt.subplot(1, 2, 2)\n",
        "plt.plot(GRU_models_hist['sg150']['val_loss'], color=\"green\")\n",
        "plt.plot(GRU_models_hist['sg300']['val_loss'], color=\"red\")\n",
        "plt.plot(GRU_models_hist['cbow300']['val_loss'], color=\"purple\")\n",
        "plt.plot(GRU_models_hist['cbow150']['val_loss'], color=\"orange\")\n",
        "plt.plot(GRU_models_hist['glove300']['val_loss'], color=\"blue\")\n",
        "plt.legend(labels = (\"Skip-gram 150\", \"Skip-gram 300\", \"Cbow 150\", \"Cbow 300\", \"GloVe 300\"), loc = \"upper left\")\n",
        "plt.title(\"model validation loss\")\n",
        "plt.ylabel(\"validation loss\")\n",
        "plt.xlabel(\"epoch\")\n",
        "#plt.subplots_adjust(bottom=0.25, top=0.75)\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FkqSdUsnUvRl"
      },
      "source": [
        "###Prediction Models"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pqnnBeOHU-V_"
      },
      "source": [
        "We will predict the number of claps with help of a Multi-Layer-Perceptron (MLP), a Gated Recurrent Unit (GRU) and a Bidirectional GRU. <br>For all models the activation function \"rectified linear\" (relu) is applied since it ensures that the output is positive. All negative outputs are set to zero. This behaviour is favorable because the number of claps can never be negative -- as least not on the blog posts of our data set. Moreover, the relu activation functions overcome the problem of a vanishing gradient which is especially important for bidirectional architectures or neural networks with many layers. Additionally, the computation is very efficiently [8]<br>\n",
        "Also one classifier is applied to all models. It is called \"Adam\". It can be used instead of the classical stochastic gradient descent. In contrast to the classical stochastic gradient descent, Adam updates the weights during training in an individual and adaptive manne. [9] As the original paper states, it combines the benefits are the classifer Adaptive Gradient Algorithm (AdaGrad) and Root Mean Square Propagation (RMSProp) -- handling problem with sparse gradients but also dealing with e.g. noice. Further benefits that are presented in the orginal paper are an efficient computation and the good performance on big data. [10] <br>\n",
        "As presented in the results of the word embedding evaluation, the embedding matrix of the Cbow model with 300 dimensions is used. To increase performance in the next models, we use the already pretrained embedding weights that were saved during evaluation or rather training process of the word embeddings."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MVSlXiCoiy7U"
      },
      "source": [
        "####Multi-Layer-Perceptron (MLP)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2p3XGZwN1JMa"
      },
      "source": [
        "Single neurons cannot solve complex task because they are restricted to a linear function. In order to overcome this problem, non-linearity is needed. A MLP is one way to do that. It is a finite directed acyclic graph with neurons that are activated with some activaton function which is in our case relu. The architecture consists of an input layer which is not a target to any connection, an output layer which are, in contrast to the input layer, no source of any connection and a hidden layer which is a layer that is not the input nor output layer. MLPs are also called vanila neural network and belong to the group of feedforward artificial neural networks. [11]"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DmLuaM2TrKDO"
      },
      "source": [
        "cbow300_emb_index = load_embedding(os.getcwd() + \"/\" + \"cbow_300_bigram_embeddings.txt\")\n",
        "cbow300_emb_mat = generate_emb_mat(word_index_train, cbow300_emb_index, 300) #dim = 300"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bJmylXo5_Nd4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 289
        },
        "outputId": "a93ead28-739a-4abe-99fa-9cc62c8b1a9d"
      },
      "source": [
        "embedding_layer = Embedding(input_dim = no_of_words,\n",
        "                            output_dim = 300, #embed_dim,\n",
        "                            input_length = 402,\n",
        "                            embeddings_initializer=Constant(cbow300_emb_mat),\n",
        "                            trainable= True)\n",
        "\n",
        "#MLP set up\n",
        "model = Sequential()\n",
        "model.add(embedding_layer)\n",
        "model.add(Dense(100))\n",
        "model.add(Dropout(0.4))\n",
        "model.add(Dense(1, activation = \"relu\"))\n",
        "model.compile(loss = \"mse\", optimizer = \"adam\")\n",
        "model.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding (Embedding)        (None, 402, 300)          61722600  \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 402, 100)          30100     \n",
            "_________________________________________________________________\n",
            "dropout (Dropout)            (None, 402, 100)          0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 402, 1)            101       \n",
            "=================================================================\n",
            "Total params: 61,752,801\n",
            "Trainable params: 61,752,801\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EotcmGJnR-D1"
      },
      "source": [
        "model.load_weights(\"GRU_cbow300_10ep_64ba.h5\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UVlXZWXHbTmQ",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 374
        },
        "outputId": "c0a6bfd2-f937-4f2f-b30f-12c1d5d85add"
      },
      "source": [
        "model.fit(X_train, y_train, validation_split = val_split, epochs= epochs, batch_size = batch_size, verbose = 1)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/10\n",
            "312/312 [==============================] - 285s 913ms/step - loss: 5061.3452 - val_loss: 4614.2632\n",
            "Epoch 2/10\n",
            "312/312 [==============================] - 285s 915ms/step - loss: 4716.9189 - val_loss: 4555.1558\n",
            "Epoch 3/10\n",
            "312/312 [==============================] - 286s 916ms/step - loss: 4674.0293 - val_loss: 4538.6680\n",
            "Epoch 4/10\n",
            "312/312 [==============================] - 286s 917ms/step - loss: 4643.7314 - val_loss: 4525.1450\n",
            "Epoch 5/10\n",
            "312/312 [==============================] - 287s 920ms/step - loss: 4624.7217 - val_loss: 4506.7173\n",
            "Epoch 6/10\n",
            "312/312 [==============================] - 287s 919ms/step - loss: 4606.0815 - val_loss: 4514.4492\n",
            "Epoch 7/10\n",
            "312/312 [==============================] - 287s 921ms/step - loss: 4595.4150 - val_loss: 4494.6211\n",
            "Epoch 8/10\n",
            "312/312 [==============================] - 285s 913ms/step - loss: 4582.2612 - val_loss: 4494.1226\n",
            "Epoch 9/10\n",
            "312/312 [==============================] - 283s 906ms/step - loss: 4573.1211 - val_loss: 4492.7319\n",
            "Epoch 10/10\n",
            "312/312 [==============================] - 282s 903ms/step - loss: 4566.4653 - val_loss: 4488.6157\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f25682dc908>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 31
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uu5zKqyqa1Cr"
      },
      "source": [
        "MLP_MSE = model.evaluate(X_test, y_test, verbose = 0)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J06D7jISitpD"
      },
      "source": [
        "y_pred = model.predict(X_test)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eONZlBy3blE8"
      },
      "source": [
        "models.update({MLP_MSE})"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3siw9FcJMV-o"
      },
      "source": [
        "####GRU with trained weights"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HPctRB6BDznb"
      },
      "source": [
        "GRUs and LSTMs have very similar recurrent neural network architectures that aim to solve the vanishing gradient problem which is typically introduced by standard recurrent neural networks (RNN) [12]. However, GRUs are computationally less costly to implement. Moreover, LSTM does not perform better in general. [13] Therefore, we decided to implement a GRU and run e.g. more epochs. <br> In contrast to an standard RNN, the architecture of GRU contain a update and reset gate. These two types of gates decides on which information is passed through to the next layer. However, they can still keep long term information which does not vanish over time like it would in a standard RNN. The update gate controlls if information is copied for instance or if only a little bit of information is passed through. Additionally, the reset gate supports in that way that it decides how much of the previous information can be forgotten. [12] <br>GRUs work in an online mode which means that they do not know the whole text in advance but rather learn word by word. In contrast bidirectional GRU work in an offline mode. This means that they know the text sequence as a whole. In some context it is crucial to know the whole context beforehand like in text summarization. [14] <br>\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UWBN_N5BJsgA"
      },
      "source": [
        "#load generated embedding matrix\n",
        "cbow300_emb_index = load_embedding(os.getcwd() + \"/\" + \"cbow_300_bigram_embeddings.txt\")\n",
        "cbow300_emb_mat = generate_emb_mat(word_index_train, cbow300_emb_index, 300) #dim = 300"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8VFh0ZLddL2U",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 289
        },
        "outputId": "09cc12ae-a262-4ee0-a40e-3945cd581b5a"
      },
      "source": [
        "#Size of RNN hidden state. Rule of thumb for RNN hidden state size: 'the optimal size of the hidden layer is usually between the size of the input and size of the output layers'. -Jeff Heaton, author of Introduction to Neural Networks in Java\n",
        "#embedding layer\n",
        "embedding_layer = Embedding(input_dim = no_of_words,\n",
        "                            output_dim = 300, #embed_dim,\n",
        "                            input_length = 402, #max_len_text,\n",
        "                            embeddings_initializer=Constant(cbow300_emb_mat),\n",
        "                            trainable=True)\n",
        "                            #embeddings_regularizer=l2_reg)\n",
        "\n",
        "\n",
        "#GRU set up\n",
        "model = Sequential()\n",
        "model.add(embedding_layer)\n",
        "model.add(GRU(100)) #, weights = pre_weights)\n",
        "model.add(Dropout(0.2))\n",
        "model.add(Dense(1, activation = \"relu\"))\n",
        "model.compile(loss = \"mse\", optimizer = \"adam\")\n",
        "model.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_1\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_1 (Embedding)      (None, 402, 300)          69626700  \n",
            "_________________________________________________________________\n",
            "gru_1 (GRU)                  (None, 100)               120600    \n",
            "_________________________________________________________________\n",
            "dropout_1 (Dropout)          (None, 100)               0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 69,747,401\n",
            "Trainable params: 69,747,401\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PKFyxJRIcZyu"
      },
      "source": [
        "model.load_weights(\"GRU_cbow300_10ep_64ba.h5\") #load pretrained weights"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EEblmlxtHT7g",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 697
        },
        "outputId": "5261ecb0-75e7-4816-a71a-7c987b6b89d3"
      },
      "source": [
        "model_hist_GRU_w = model.fit(X_train, y_train, validation_split= val_split, epochs= epochs, batch_size = batch_size) #validation_data=(X_test, y_test)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "312/312 [==============================] - 759s 2s/step - loss: 2062.7976 - val_loss: 3704.5308\n",
            "Epoch 2/20\n",
            "312/312 [==============================] - 759s 2s/step - loss: 1653.6630 - val_loss: 3828.9692\n",
            "Epoch 3/20\n",
            "312/312 [==============================] - 750s 2s/step - loss: 1362.3687 - val_loss: 3915.8696\n",
            "Epoch 4/20\n",
            "312/312 [==============================] - 765s 2s/step - loss: 1086.8198 - val_loss: 3997.7192\n",
            "Epoch 5/20\n",
            "312/312 [==============================] - 767s 2s/step - loss: 895.4716 - val_loss: 4064.7925\n",
            "Epoch 6/20\n",
            "312/312 [==============================] - 767s 2s/step - loss: 767.1475 - val_loss: 4242.9126\n",
            "Epoch 7/20\n",
            "312/312 [==============================] - 760s 2s/step - loss: 614.1409 - val_loss: 4158.6689\n",
            "Epoch 8/20\n",
            "312/312 [==============================] - 766s 2s/step - loss: 533.3627 - val_loss: 4212.9863\n",
            "Epoch 9/20\n",
            "312/312 [==============================] - 750s 2s/step - loss: 452.2273 - val_loss: 4286.1270\n",
            "Epoch 10/20\n",
            "312/312 [==============================] - 759s 2s/step - loss: 396.7936 - val_loss: 4300.3623\n",
            "Epoch 11/20\n",
            "312/312 [==============================] - 753s 2s/step - loss: 345.0874 - val_loss: 4260.9487\n",
            "Epoch 12/20\n",
            "312/312 [==============================] - 757s 2s/step - loss: 308.3256 - val_loss: 4107.3423\n",
            "Epoch 13/20\n",
            "312/312 [==============================] - 769s 2s/step - loss: 337.5449 - val_loss: 4400.8623\n",
            "Epoch 14/20\n",
            "312/312 [==============================] - 763s 2s/step - loss: 324.3201 - val_loss: 4155.4443\n",
            "Epoch 15/20\n",
            "312/312 [==============================] - 766s 2s/step - loss: 294.5098 - val_loss: 4334.2026\n",
            "Epoch 16/20\n",
            "312/312 [==============================] - 765s 2s/step - loss: 283.1808 - val_loss: 4117.9722\n",
            "Epoch 17/20\n",
            "312/312 [==============================] - 761s 2s/step - loss: 262.8714 - val_loss: 4098.8613\n",
            "Epoch 18/20\n",
            "312/312 [==============================] - 754s 2s/step - loss: 227.2899 - val_loss: 4168.6123\n",
            "Epoch 19/20\n",
            "312/312 [==============================] - 751s 2s/step - loss: 188.8043 - val_loss: 4054.0435\n",
            "Epoch 20/20\n",
            "312/312 [==============================] - 754s 2s/step - loss: 179.4255 - val_loss: 4044.2725\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eVC5HymtTph5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 350
        },
        "outputId": "f78a0944-e697-4200-e084-d6e8ed8e37f8"
      },
      "source": [
        "plot_hist(model_hist_GRU_w)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PPkleWkyYS9h"
      },
      "source": [
        "y_pred = model.predict(X_test)\n",
        "mse_GRU = mean_squared_error(y_test, y_pred)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ouGKH1w9CgBf",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "a406895e-1d57-43e4-92f5-b43cc585b5ff"
      },
      "source": [
        "mse_GRU"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1238598.8923287229"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 52
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DPAoDhFXcg9A"
      },
      "source": [
        "####Bidirectional GRU"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Hnuj7ZxXKuTU",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 289
        },
        "outputId": "cc47b4df-4fba-4a3f-877a-f34d3530c73c"
      },
      "source": [
        "#size of RNN hidden state. Rule of thumb for RNN hidden state size: 'the optimal size of the hidden layer is usually between the size of the input and size of the output layers'. -Jeff Heaton, author of Introduction to Neural Networks in Java\n",
        "\n",
        "#embedding layer\n",
        "embedding_layer = Embedding(input_dim = no_of_words,\n",
        "                            output_dim = 300, #embed_dim,\n",
        "                            input_length = 402, #max_len_text,\n",
        "                            embeddings_initializer=Constant(cbow300_emb_mat),\n",
        "                            trainable=True)\n",
        "                            #embeddings_regularizer=l2_reg)\n",
        "\n",
        "\n",
        "#GRU set up\n",
        "model = Sequential()\n",
        "model.add(embedding_layer)\n",
        "model.add(Bidirectional(GRU(100), merge_mode = \"concat\")) #, weights = pre_weights)\n",
        "model.add(Dropout(0.4))\n",
        "model.add(Dense(1, activation = \"relu\"))\n",
        "model.compile(loss = \"mse\", optimizer = \"adam\")\n",
        "model.summary()\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_1\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_1 (Embedding)      (None, 402, 300)          69626700  \n",
            "_________________________________________________________________\n",
            "bidirectional_1 (Bidirection (None, 200)               241200    \n",
            "_________________________________________________________________\n",
            "dropout_1 (Dropout)          (None, 200)               0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 1)                 201       \n",
            "=================================================================\n",
            "Total params: 69,868,101\n",
            "Trainable params: 69,868,101\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FkBjK6Ckc-oc",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 357
        },
        "outputId": "99a30ede-43bd-43e5-c1e0-ba9baeaaf1cd"
      },
      "source": [
        "model_hist_BiGRU = model.fit(X_train, y_train, validation_split= val_split, epochs= epochs, batch_size = batch_size) #validation_data=(X_test, y_test)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/10\n",
            "312/312 [==============================] - 1224s 4s/step - loss: 4922.8496 - val_loss: 4353.7422\n",
            "Epoch 2/10\n",
            "312/312 [==============================] - 1274s 4s/step - loss: 4034.5369 - val_loss: 3667.1392\n",
            "Epoch 3/10\n",
            "312/312 [==============================] - 1225s 4s/step - loss: 3312.4834 - val_loss: 3395.3215\n",
            "Epoch 4/10\n",
            "312/312 [==============================] - 1266s 4s/step - loss: 2406.7710 - val_loss: 3559.4353\n",
            "Epoch 5/10\n",
            "312/312 [==============================] - 1306s 4s/step - loss: 1594.3810 - val_loss: 3720.5793\n",
            "Epoch 6/10\n",
            "312/312 [==============================] - 1304s 4s/step - loss: 1040.1787 - val_loss: 3861.9666\n",
            "Epoch 7/10\n",
            "312/312 [==============================] - 1299s 4s/step - loss: 721.3184 - val_loss: 3832.3081\n",
            "Epoch 8/10\n",
            "312/312 [==============================] - 1322s 4s/step - loss: 524.5369 - val_loss: 3907.1851\n",
            "Epoch 9/10\n",
            "312/312 [==============================] - 1354s 4s/step - loss: 393.2196 - val_loss: 3867.0562\n",
            "Epoch 10/10\n",
            "312/312 [==============================] - 1357s 4s/step - loss: 323.2965 - val_loss: 3854.4802\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4hhjODidKmQG"
      },
      "source": [
        "model.save(\"BiGRU_10.h5\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ENdn4vFsB7R9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 366
        },
        "outputId": "8ea0a2e1-4d37-4c87-be64-73da02c8a390"
      },
      "source": [
        "plot_hist(model_hist_BiGRU)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Eq3j6RDbfObz"
      },
      "source": [
        "from keras import models    \n",
        "model = models.load_model('BiGRU_10.h5')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "osN5ryFqgQjG",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 33
        },
        "outputId": "2f4fdee4-c239-4681-8ced-d339147006c6"
      },
      "source": [
        "BiGru_mse = model.evaluate(X_test, y_test, verbose = 0)\n",
        "BiGru_mse"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1248442.625"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GLEczsMyh09Q"
      },
      "source": [
        "?model.evaluate"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vYRa7SUPd15I"
      },
      "source": [
        "####Results"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ytMMfsCJL9eu"
      },
      "source": [
        "The models are compared with their scalar test loss which is in our case equal to the mean-squared-error since we specified loss as \"mse\" in all models set-ups. The function model.evaluate(X_test, y_test) returns two scores. The loss and the accuracy (not specified in our analysis).\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "H6TOx0jWgRWs"
      },
      "source": [
        "\\begin{array}{ll}\n",
        "Model & MSE \\\\\n",
        "\\hline\n",
        "Elastic Net & 1,243,992.29 \\\\\n",
        "MLP \\:10 & 1,257,236.75 \\\\\n",
        "Gru \\:20 & 1,238,598.89 \\\\\n",
        "BiGru \\:10 & 1,248,442.63 \\\\\n",
        "\\end{array}"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nhC-jKFKlW_p"
      },
      "source": [
        "Even though the dropout rates of the GRU and BiGRU are with 0.4 quite high and should prevent the model from overfitting, both models seem to suffer from overfitting. Nevertheless, the forward GRU that trained 20 epochs performed the best in spite of overfitting and therefore will be used for the final prediction."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "l7kYQq0qd5-G"
      },
      "source": [
        "models"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hf0DJ-i8ymkQ"
      },
      "source": [
        "##Test Set Prediction"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u6STMkPgiS1h"
      },
      "source": [
        "from keras import models    \n",
        "model = models.load_model('GRU_test.h5')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZkD4Oomlyp99"
      },
      "source": [
        "#combine, header und text, clean text\n",
        "#clean author, clean zum numbers, check for empty authors\n",
        "\n",
        "data_file = \"Test.csv\"\n",
        "test = pd.read_csv(data_file, encoding='utf-8')\n",
        "del(test[\"Unnamed: 0\"])\n",
        "#test = test.drop(\"index\", axis=1)\n",
        "#test = test.reset_index()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ufvt96iBzsLe",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 156
        },
        "outputId": "14542dd8-3cd9-4db3-83f9-2ec212be5cf1"
      },
      "source": [
        "#extract number of responses, remove all characters which is not numeric, impute missing value with value 0\n",
        "\n",
        "print(test.Responses.isnull().sum())\n",
        "test.Responses[test.Responses.isnull()] = 0\n",
        "print(test.Responses.isnull().sum())\n",
        "\n",
        "def clean_respond(row, df_length):\n",
        "  \n",
        "  row[\"Responses\"] = str(row[\"Responses\"])\n",
        "  row[\"Responses\"] = re.sub(\"[^0-9]\", \" \", row[\"Responses\"])\n",
        "\n",
        "  return row\n",
        "\n",
        "test = test.apply(lambda row: clean_respond(row, len(test)), axis=1)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "82\n",
            "0\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:4: SettingWithCopyWarning: \n",
            "A value is trying to be set on a copy of a slice from a DataFrame\n",
            "\n",
            "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
            "  after removing the cwd from sys.path.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5UdPmb0m_aCE"
      },
      "source": [
        "#extract date zum PublicationDetails, remove all characters which is not numeric\n",
        "\n",
        "def clean_publ(row, df_length):\n",
        "  \n",
        "  #row[\"Responses\"] = str(row[\"Responses\"])\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Dec\", \"12\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Nov\", \"11\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Oct\", \"10\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Sep\", \"09\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Aug\", \"08\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Jul\", \"07\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Jun\", \"06\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"May\", \"05\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Apr\", \"04\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Mar\", \"03\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Feb\", \"02\")\n",
        "  row[\"PublicationDetails\"] = row[\"PublicationDetails\"].replace(\"Jan\", \"01\")\n",
        "  \n",
        "  return row\n",
        "\n",
        "test = test.apply(lambda row: clean_publ(row, len(test)), axis=1) \n",
        "\n",
        "test[\"date\"] = [re.sub(\"[^0-9]\", \"\", x) for x in test[\"PublicationDetails\"]]\n",
        "test[\"year\"] = [int(str(x)[-4:]) for x in test[\"date\"]]\n",
        "test[\"author_info\"] = [re.sub(\"[0-9]\", \"\", x) for x in test[\"PublicationDetails\"]]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "slLCwzGjDexo"
      },
      "source": [
        "#author cleaning: remove digits/date from var\n",
        "\n",
        "def clean_author(row, df_length):\n",
        "  \n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Dec\", \"12\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Nov\", \"11\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Oct\", \"10\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Sep\", \"09\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Aug\", \"08\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Jul\", \"07\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Jun\", \"06\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"May\", \"05\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Apr\", \"04\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Mar\", \"03\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Feb\", \"02\")\n",
        "  row[\"Author\"] = row[\"Author\"].replace(\"Jan\", \"01\")\n",
        "  \n",
        "  return row\n",
        "\n",
        "test = test.apply(lambda row: clean_author(row, len(test)), axis=1) \n",
        "test[\"Author\"] = [re.sub(\"[0-9]\", \"\", x) for x in test[\"Author\"]]\n",
        "test[\"Author\"] = test[\"Author\"].replace(\" , \", \"Unknown\")\n",
        "test[\"Author\"] = test[\"Author\"].replace(\"-\", \"Unknown\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kfWKiLTuKVr4"
      },
      "source": [
        "def clean_split(row, df_length):\n",
        "\n",
        "  #row[\"author_info\"] = str(row[\"author_info\"])\n",
        "  #row[\"author_info\"] = row[\"author_info\"].split(\"in\")\n",
        "  row[\"author_info\"] = row[\"author_info\"].partition('in')[0]\n",
        "  row[\"author_info\"] = re.sub(\"[^a-zA-Z]\", \" \", row[\"author_info\"])\n",
        "  row[\"author_info\"] = ' '.join( [w for w in row[\"author_info\"].split() if len(w)>1] )\n",
        "  \n",
        "  return row\n",
        "\n",
        "test = test.apply(lambda row: clean_split(row, len(test)), axis=1) "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WEwxuIX4AzIl"
      },
      "source": [
        "test[\"Author\"] = np.where((test[\"Author\"] == \"Unknown\") & (test[\"author_info\"] != \"\"), test[\"author_info\"], np.where((test[\"Author\"] == \"Unknown\") & (test[\"author_info\"] == \"\"), \"Unknown\", test[\"Author\"]))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oJkqoyXuR51U",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 397
        },
        "outputId": "006f2918-046a-4509-8ffd-c685ec4c121d"
      },
      "source": [
        "test.tail()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>index</th>\n",
              "      <th>Author</th>\n",
              "      <th>PublicationDetails</th>\n",
              "      <th>Responses</th>\n",
              "      <th>Header</th>\n",
              "      <th>Text</th>\n",
              "      <th>Length</th>\n",
              "      <th>date</th>\n",
              "      <th>year</th>\n",
              "      <th>author_info</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>509</th>\n",
              "      <td>598</td>\n",
              "      <td>Tim Romero</td>\n",
              "      <td>Tim Romero in Startup Lessons Learned06 9, 2016</td>\n",
              "      <td>181</td>\n",
              "      <td>Why I turned down $500K, Pissed off my investo...</td>\n",
              "      <td>I just did what no startup founder is ever sup...</td>\n",
              "      <td>9025</td>\n",
              "      <td>0692016</td>\n",
              "      <td>2016</td>\n",
              "      <td>Tim Romero</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>510</th>\n",
              "      <td>599</td>\n",
              "      <td>Product Hunt</td>\n",
              "      <td>Product Hunt06 10, 2016</td>\n",
              "      <td>24</td>\n",
              "      <td>These Tools Will Help You Launch Your Startup</td>\n",
              "      <td>If you’re embarking on the startup journey and...</td>\n",
              "      <td>5571</td>\n",
              "      <td>06102016</td>\n",
              "      <td>2016</td>\n",
              "      <td>Product Hunt</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>511</th>\n",
              "      <td>600</td>\n",
              "      <td>Joe Procopio</td>\n",
              "      <td>Joe Procopio03 19</td>\n",
              "      <td>24</td>\n",
              "      <td>Why Startups Fall Apart at 50 Employees</td>\n",
              "      <td>Fuck you startups with your extravagant partie...</td>\n",
              "      <td>475</td>\n",
              "      <td>0319</td>\n",
              "      <td>319</td>\n",
              "      <td>Joe Procopio</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>512</th>\n",
              "      <td>601</td>\n",
              "      <td>Unknown</td>\n",
              "      <td>Winerist in Be Yourself03 18, 2016</td>\n",
              "      <td>116</td>\n",
              "      <td>Beware of Startup Prostitution</td>\n",
              "      <td>Note: This is not a post to glorify prostituti...</td>\n",
              "      <td>13483</td>\n",
              "      <td>03182016</td>\n",
              "      <td>2016</td>\n",
              "      <td></td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>513</th>\n",
              "      <td>602</td>\n",
              "      <td>Aytek</td>\n",
              "      <td>Aytekin Tank in The Startup01 11, 2018</td>\n",
              "      <td>34</td>\n",
              "      <td>How to build a startup — without quitting your...</td>\n",
              "      <td>Question: What do Craigslist and Albert Einste...</td>\n",
              "      <td>8736</td>\n",
              "      <td>01112018</td>\n",
              "      <td>2018</td>\n",
              "      <td>Aytek</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "     index        Author  ...  year   author_info\n",
              "509    598    Tim Romero  ...  2016    Tim Romero\n",
              "510    599  Product Hunt  ...  2016  Product Hunt\n",
              "511    600  Joe Procopio  ...   319  Joe Procopio\n",
              "512    601       Unknown  ...  2016              \n",
              "513    602         Aytek  ...  2018         Aytek\n",
              "\n",
              "[5 rows x 10 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "087wmtPFCdS4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 111
        },
        "outputId": "67f61b90-84a3-4a8f-d450-c61118a40507"
      },
      "source": [
        "test.loc[test.Author == \"Product Hunt\", [\"Author\"]]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Author</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>505</th>\n",
              "      <td>Product Hunt</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>510</th>\n",
              "      <td>Product Hunt</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "           Author\n",
              "505  Product Hunt\n",
              "510  Product Hunt"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 56
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JMLGEfiokl1p",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "2457dfb3-a11d-4cab-b9cc-f98848f5ef81"
      },
      "source": [
        "#Reset missing values in Header, otherwise, the next processing step does not work properly, 0 is implement because it is easy to apply, store and remove in the text cleaning\n",
        "test[\"Header\"] = np.where(test.Header.isnull(), 0, test[\"Header\"])\n",
        "\n",
        "#Merge headlines and texts \n",
        "test[\"text_full\"] = test[\"Header\"].astype(str) + \" \" + test[\"Text\"]\n",
        "\n",
        "#Clean text like the training set\n",
        "def clean_all(row, df_length):\n",
        "  \n",
        "  row[\"text_full\"] = cleaning_function(row[\"text_full\"])\n",
        "\n",
        "  print(f\"\\rcompleted: {row.name + 1} of {df_length}\", end=\"\")\n",
        "  return row\n",
        "\n",
        "t1 = time.perf_counter()\n",
        "#für später: tr_new anstatt df (2x)\n",
        "#train_cleaned = tr_new.reset_index()\n",
        "#train_cleaned = train_cleaned.drop(\"index\", axis=1)\n",
        "test_cleaned = test.apply(lambda row: clean_all(row, len(test)), axis=1)\n",
        "t2 = time.perf_counter()\n",
        "\n",
        "print(\"It took {:.2f} minutes to compile\".format((t2-t1)/60))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "completed: 514 of 514It took 0.38 minutes to compile\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "skq7yAvFu8FP",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "03a0020d-74d3-4d88-9bff-2abac2dd06d2"
      },
      "source": [
        "X_test = test_cleaned.loc[:,[\"text_full\"]]\n",
        "\n",
        "tokenizer_obj = Tokenizer() #vectorizes the text corpus\n",
        "tokenizer_obj.fit_on_texts(X_test[\"text_full\"])\n",
        "seq = tokenizer_obj.texts_to_sequences(X_test[\"text_full\"])\n",
        "\n",
        "from nltk.tokenize.treebank import TreebankWordDetokenizer\n",
        "test_cleaned[\"text_full_detoken\"] = [TreebankWordDetokenizer().detokenize(token) for token in test_cleaned[\"text_full\"]] \n",
        "\n",
        "#max. length of text (of the training data)\n",
        "max_len_text = max([len(x.split()) for x in test_cleaned[\"text_full_detoken\"]]) #\"text_cl\" because tokens are not countable this way\n",
        "print(\"Max. text length: {}\".format(max_len_text))\n",
        "\n",
        "X_test = pad_sequences(seq, 402) \n",
        "print(X_test.shape)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Max. text length: 2026\n",
            "(514, 402)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DkkgVKvdxdDh"
      },
      "source": [
        "claps_pred = model.predict(X_test)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K6Xo0Ghix2LG"
      },
      "source": [
        "data_file = \"Sample_submission.csv\"\n",
        "sample = pd.read_csv(data_file, encoding='utf-8')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ESt2gGGdzjS-"
      },
      "source": [
        "test[\"Claps\"] = claps_pred"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YiG5sM7g0zHh",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 37
        },
        "outputId": "f83cc765-28ad-4158-d1bf-57a511490307"
      },
      "source": [
        "path = os.getcwd()\n",
        "path"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'/content/gdrive/My Drive/ADAMS/Assignment'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 74
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Soz4ZvE7sjQg"
      },
      "source": [
        "test[[\"index\",\"Claps\"]].to_csv(path + \"/finalprediction2.csv\",index=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kXO-2dMuWd7x"
      },
      "source": [
        "##Recap and Potential Next Steps"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Xen8ZnoQWthA"
      },
      "source": [
        "1) Tune the number of claps<br> \n",
        "Right now we choose a cutoff at 3*IQR. This cutoff effected about the top 12% of the observations. The finalprediction.csv file shows that the model has difficulties to predict a number of claps that was more than 150. Instead of taking a cutoff related to the IQR, one could cutoff the top x%. <p>\n",
        "2) Focus on the sentiment analysis<br>\n",
        "Use different dictionaries, form clusters or similar. <p>\n",
        "3) Scrapping more data for the test set<br>\n",
        "If allowed, one could also scrap more data for the test set to be able to use variable with very high correlations such as the train variable recommends with a correlation towards the target variable of 0.96. <p> \n",
        "4) Focus on publishing websites<br>\n",
        "Extracting the websites on which the post first were published might show a correlation towards the number of claps.<p>\n",
        "5) Tune parameters<br>\n",
        "Tuning the parameter, more regularization to the GRUs etc."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "U2PF2chgsDjQ"
      },
      "source": [
        "## References\n",
        "\n",
        "[1] Google Code Archive, word2vec, project created on Jul 30, 2013, https://code.google.com/archive/p/word2vec/. <br>\n",
        "[2] Kennedy, J. J. (1970), The eta coefficient in complex ANOVA designs, Educational and Psychological Measurement, 30(4), 885-889. <br>\n",
        "[3] Zakaria Jaadi, Everything you need to know about interpreting correlations, Towards Data Science. Oct 15, 2019.<br>\n",
        "[4] Dhruvil Karani, Introduction to Word Embedding and Word2Vec, Towards Data Science, Sep 1, 2018.<br>\n",
        "[5] Manish Chablani, Word2Vec (skip-gram model): PART 1 - Intuition. Jun 14, 2017.<br>\n",
        "[6] Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. GloVe: Global Vectors for Word Representation. <br>\n",
        "[7] Amar Mandal, Ridge, Lasso and Elastic Net Regularization methods. Jun 07, 2020.<br>\n",
        "[8] Jason Brownlee, A Gentle Introduction to the Rectified Linear Unit (ReLU), Deep Learning Performance. Jan 09, 2019 (last update: Aug 20, 2020), https://machinelearningmastery.com/rectified-linear-activation-function-for-deep-learning-neural-networks/.<br>\n",
        "[9] Jason Brownlee, Gentle Introduction to the Adam Optimization Algorithm for Deep Learning,  Deep Learning Performance. Jul 03, 2017 (last update: Aug 20, 2020) https://machinelearningmastery.com/adam-optimization-algorithm-for-deep-learning/. <br> \n",
        "[10] Kingma, D. P., & Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.<br>\n",
        "[11] Prof. Dr. Martin Riedmiller, Machine Learning: Multi Layer Perceptrons, Albert-Ludwigs-University Freiburg AG Maschinelles Lernen, http://ml.informatik.uni-freiburg.de/former/_media/documents/teaching/ss09/ml/mlps.pdf. (last accessed: Aug 31, 2020).<br>\n",
        "[12] Simeon Kostadinov, Understanding GRU Networks, Towards Data Science. Dec 16, 2017. https://towardsdatascience.com/understanding-gru-networks-2ef37df6c9be.<br>\n",
        "[13] Chung, J., Gulcehre, C., Cho, K., & Bengio, Y. (2014). Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint arXiv:1412.3555.<br>\n",
        "[14] Madhu Ramiah, Bi-directional RNN & Basics of LSTM and GRU. Jul 9, 2019. https://medium.com/@madhuramiah/bi-directional-rnn-basics-of-lstm-and-gru-e114aa4779bb.\n",
        "\n"
      ]
    }
  ]
}