notebook_assignment-two.ipynb

{
  "cells": [
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "CRWCzaDnyW1G"
      },
      "outputs": [],
      "source": [
        "'''\n",
        "Marco Costante, Alex Costanzino, Alessandra Stramiglio, Xiaowei Wen\n",
        "MSc students in Artificial Intelligence\n",
        "@ Alma Mater Studiorum, University of Bologna\n",
        "December, 2021\n",
        "'''"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "_RN0OFNS38wr"
      },
      "outputs": [],
      "source": [
        "from google.colab import drive\n",
        "drive.mount(\"/content/drive\", force_remount = True)\n",
        "%cd drive/MyDrive/[NLP]/assignment-two"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HCav8XDzDhrf"
      },
      "source": [
        "# Assignment No. 2: fact checking, natural language inference\n",
        "In the last few years, disinformation and fake news have been spreading throughout the world at rates never seen before, creating the need for fact-checking organizations. \n",
        "Despite many human-based fact-checking organizations that are currently in operation, disinformation continues to briskly spread throughout the internet, and the existing organizations are unable to keep up.\n",
        "\n",
        "Automated fact checking refers to the process that seeks to verify factual information with the aim to promote veracity and correctness of the text. \n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "yLx-JAcKFANt"
      },
      "outputs": [],
      "source": [
        "import os\n",
        "import requests\n",
        "import zipfile\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "import tensorflow as tf\n",
        "from tensorflow.keras.models import load_model, save_model\n",
        "\n",
        "!pip install focal-loss --quiet\n",
        "import focal_loss"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zw98yz7vaUQ1"
      },
      "source": [
        "## Data processing"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8ifJWcks9U4M"
      },
      "source": [
        "### Data extraction"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "XGb38ul1aa75"
      },
      "outputs": [],
      "source": [
        "def save_response_content(response, destination):\n",
        "    CHUNK_SIZE = 32768\n",
        "\n",
        "    with open(destination, \"wb\") as f:\n",
        "        for chunk in response.iter_content(CHUNK_SIZE):\n",
        "            if chunk: # Filter out keep-alive new chunks.\n",
        "                f.write(chunk)\n",
        "\n",
        "def download_data(data_path):\n",
        "    toy_data_path = os.path.join(data_path, 'fever_data.zip')\n",
        "    toy_data_url_id = \"1wArZhF9_SHW17WKNGeLmX-QTYw9Zscl1\"\n",
        "    toy_url = \"https://docs.google.com/uc?export=download\"\n",
        "\n",
        "    if not os.path.exists(data_path):\n",
        "        os.makedirs(data_path)\n",
        "\n",
        "    if not os.path.exists(toy_data_path):\n",
        "        print(\"Downloading FEVER data splits...\")\n",
        "        with requests.Session() as current_session:\n",
        "            response = current_session.get(toy_url,\n",
        "                                   params = {'id': toy_data_url_id},\n",
        "                                   stream = True)\n",
        "        save_response_content(response, toy_data_path)\n",
        "        print(\"Download completed!\")\n",
        "\n",
        "        print(\"Extracting dataset...\")\n",
        "        with zipfile.ZipFile(toy_data_path) as loaded_zip:\n",
        "            loaded_zip.extractall(data_path)\n",
        "        print(\"Extraction completed!\")\n",
        "\n",
        "download_data('dataset')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W84c95pHbQ0w"
      },
      "source": [
        "### Data loading\n",
        "We loaded the splits into separate dataframes composed by three columns: `Claim`, `Evidence`, `Label`, dropping everything else."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "qF-3L8LVbYZM"
      },
      "outputs": [],
      "source": [
        "import pandas as pd\n",
        "\n",
        "df_train = pd.read_csv('dataset/train_pairs.csv', sep = ',')\n",
        "df_train.drop(['Unnamed: 0', 'ID'], axis = 1, inplace = True)\n",
        "\n",
        "df_val = pd.read_csv('dataset/val_pairs.csv', sep = ',')\n",
        "df_val.drop(['Unnamed: 0', 'ID'], axis = 1, inplace = True)\n",
        "\n",
        "df_test = pd.read_csv('dataset/test_pairs.csv', sep = ',')\n",
        "df_test.drop(['Unnamed: 0', 'ID'], axis = 1, inplace = True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "fd7rFvoZc70t",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "outputId": "e3308117-e369-436b-f3da-364f8147a0b8"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "\n",
              "  <div id=\"df-ba239e78-d442-44c9-b73c-afc5ff04ed44\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <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>Claim</th>\n",
              "      <th>Evidence</th>\n",
              "      <th>Label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>Chris Hemsworth appeared in A Perfect Getaway.</td>\n",
              "      <td>2\\tHemsworth has also appeared in the science ...</td>\n",
              "      <td>SUPPORTS</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>Roald Dahl is a writer.</td>\n",
              "      <td>0\\tRoald Dahl -LRB- -LSB- langpronˈroʊ.əld _ ˈ...</td>\n",
              "      <td>SUPPORTS</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>Roald Dahl is a governor.</td>\n",
              "      <td>0\\tRoald Dahl -LRB- -LSB- langpronˈroʊ.əld _ ˈ...</td>\n",
              "      <td>REFUTES</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>Ireland has relatively low-lying mountains.</td>\n",
              "      <td>10\\tThe island 's geography comprises relative...</td>\n",
              "      <td>SUPPORTS</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>Ireland does not have relatively low-lying mou...</td>\n",
              "      <td>10\\tThe island 's geography comprises relative...</td>\n",
              "      <td>REFUTES</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-ba239e78-d442-44c9-b73c-afc5ff04ed44')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-ba239e78-d442-44c9-b73c-afc5ff04ed44 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-ba239e78-d442-44c9-b73c-afc5ff04ed44');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ],
            "text/plain": [
              "                                               Claim  ...     Label\n",
              "0     Chris Hemsworth appeared in A Perfect Getaway.  ...  SUPPORTS\n",
              "1                            Roald Dahl is a writer.  ...  SUPPORTS\n",
              "2                          Roald Dahl is a governor.  ...   REFUTES\n",
              "3        Ireland has relatively low-lying mountains.  ...  SUPPORTS\n",
              "4  Ireland does not have relatively low-lying mou...  ...   REFUTES\n",
              "\n",
              "[5 rows x 3 columns]"
            ]
          },
          "metadata": {},
          "execution_count": 7
        }
      ],
      "source": [
        "# Check.\n",
        "df_train.head()\n",
        "# df_val.head()\n",
        "# df_test.tail()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "w_f3DQJZ9ahX"
      },
      "source": [
        "### Text filtering\n",
        "We have to remove some spurious words from the sentences:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "pKqMhuRDfKlF",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c0866583-fde1-4705-9759-750021732b28"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Before: \n",
            "\n",
            "                                               Claim  ...     Label\n",
            "0     Chris Hemsworth appeared in A Perfect Getaway.  ...  SUPPORTS\n",
            "1                            Roald Dahl is a writer.  ...  SUPPORTS\n",
            "2                          Roald Dahl is a governor.  ...   REFUTES\n",
            "3        Ireland has relatively low-lying mountains.  ...  SUPPORTS\n",
            "4  Ireland does not have relatively low-lying mou...  ...   REFUTES\n",
            "\n",
            "[5 rows x 3 columns]\n",
            "After: \n",
            "\n",
            "                                               Claim  ...     Label\n",
            "0      chris hemsworth appeared in a perfect getaway  ...  supports\n",
            "1                             roald dahl is a writer  ...  supports\n",
            "2                           roald dahl is a governor  ...   refutes\n",
            "3        ireland has relatively low  lying mountains  ...  supports\n",
            "4  ireland does not have relatively low  lying mo...  ...   refutes\n",
            "\n",
            "[5 rows x 3 columns]\n",
            "Before: \n",
            "\n",
            "                                               Claim  ...     Label\n",
            "0  The Indian Army comprises part of the country'...  ...  SUPPORTS\n",
            "1                         Recovery features Rihanna.  ...  SUPPORTS\n",
            "2                            Rihanna is on Recovery.  ...  SUPPORTS\n",
            "3                           Philomena was nominated.  ...  SUPPORTS\n",
            "4                               Vietnam is a mascot.  ...   REFUTES\n",
            "\n",
            "[5 rows x 3 columns]\n",
            "After: \n",
            "\n",
            "                                               Claim  ...     Label\n",
            "0  the indian army comprises part of the countrys...  ...  supports\n",
            "1                          recovery features rihanna  ...  supports\n",
            "2                             rihanna is on recovery  ...  supports\n",
            "3                            philomena was nominated  ...  supports\n",
            "4                                vietnam is a mascot  ...   refutes\n",
            "\n",
            "[5 rows x 3 columns]\n",
            "Before: \n",
            "\n",
            "                                               Claim  ...     Label\n",
            "0    Anxiety has been linked with physical symptoms.  ...  SUPPORTS\n",
            "1                         Firefox is an application.  ...  SUPPORTS\n",
            "2  Keegan-Michael Key played President Barack Oba...  ...  SUPPORTS\n",
            "3               Google Search can find stock quotes.  ...  SUPPORTS\n",
            "4  A Good Day to Die Hard was directed solely by ...  ...   REFUTES\n",
            "\n",
            "[5 rows x 3 columns]\n",
            "After: \n",
            "\n",
            "                                               Claim  ...     Label\n",
            "0     anxiety has been linked with physical symptoms  ...  supports\n",
            "1                          firefox is an application  ...  supports\n",
            "2  keegan  michael key played president barack ob...  ...  supports\n",
            "3                google search can find stock quotes  ...  supports\n",
            "4  a good day to die hard was directed solely by ...  ...   refutes\n",
            "\n",
            "[5 rows x 3 columns]\n"
          ]
        }
      ],
      "source": [
        "import re\n",
        "from functools import reduce\n",
        "import nltk\n",
        "from typing import List, Callable\n",
        " \n",
        "re_remove_number= re.compile('([0-9]\\\\t|[0-9])')\n",
        "replace_symbols = re.compile('[/(){}\\[\\]\\|@,;]')\n",
        "good_symbols = re.compile('[^0-9a-z #+_]')\n",
        " \n",
        "# All function names are self-explainatory.\n",
        "def remove_leading_tabs(text: str) -> str:\n",
        "   pattern = r'[0-9]+?\\t'\n",
        "   return re.sub(pattern, '', text)\n",
        " \n",
        "def remove_pronunciations(text: str) -> str:\n",
        "   pattern = r'-LSB-.*?-RSB-(\\s;)*?'\n",
        "   return re.sub(pattern, '', text)\n",
        " \n",
        "def remove_brackets(text: str) -> str:\n",
        "   pattern = re.compile('(\\-RSB\\-|\\-RRB\\-|\\-LSB\\-|\\-LRB\\-)')\n",
        "   return re.sub(pattern, '', text)\n",
        " \n",
        "def fix_double_dashes(text: str) -> str:\n",
        "   pattern = r'\\-\\-'\n",
        "   return re.sub(pattern, '-', text)\n",
        " \n",
        "def remove_trailing_words(text) :\n",
        "   pattern = r'.\\t.*?$'\n",
        "   return re.sub(pattern, '.', text)\n",
        " \n",
        "def split_genitive(text: str) -> str:\n",
        "   pattern = r\"(\\s.+?)'s\"\n",
        "   return re.sub(pattern, r\"\\1's\", text)\n",
        " \n",
        "def separate_non_words(text) :\n",
        "   pattern = r'([a-zA-Z]+?)\\-([a-zA-Z]+?)'\n",
        "   return re.sub(pattern, r'\\1 - \\2', text)\n",
        " \n",
        "def lower(text: str) -> str:\n",
        "   return text.lower()\n",
        " \n",
        "def remove_initial_no(str:str) -> str:\n",
        "   return re.sub(re_remove_number,\"\",str)\n",
        " \n",
        "def replace_special_characters(text: str) -> str:\n",
        "   return replace_symbols.sub(' ', text)\n",
        " \n",
        "def replace_br(text: str) -> str:\n",
        "   return text.replace('br', '')\n",
        " \n",
        "def filter_out_uncommon_symbols(text: str) -> str:\n",
        "   return re.sub(good_symbols, '',text)\n",
        " \n",
        "def remove_stopwords(text: str) -> str:\n",
        "   return ' '.join([x for x in text.split() if x and x not in stopwords_set])\n",
        " \n",
        "def strip_text(text: str) -> str:\n",
        "   return text.strip()\n",
        " \n",
        "def remove_langpron(text: str)-> str:\n",
        "   return text.replace('langpron',\"\")\n",
        " \n",
        "def remove_dl(text:str)-> str:\n",
        "   pattern = re.compile('( dl | ul )')\n",
        "   return re.sub(pattern, '', text)\n",
        " \n",
        "pipeline = [remove_leading_tabs, remove_pronunciations, remove_brackets,\n",
        "           fix_double_dashes, remove_trailing_words, split_genitive,\n",
        "           separate_non_words,\n",
        "           lower, remove_initial_no, replace_special_characters,\n",
        "           replace_br, filter_out_uncommon_symbols,\n",
        "           strip_text, remove_langpron]\n",
        " \n",
        "def text_prepare(text: str,\n",
        "                filter_methods: List[Callable[[str], str]] = None) -> str:\n",
        "   '''\n",
        "   Applies a list of pre-processing functions in sequence (reduce).\n",
        "   Note that the order is important here!\n",
        "   '''\n",
        " \n",
        "   filter_methods = filter_methods if filter_methods is not None else pipeline\n",
        " \n",
        "   return reduce(lambda txt, f: f(txt), filter_methods, text)\n",
        " \n",
        "def cleaning(dataframes: list, labels: list):\n",
        " for dataframe in dataframes:\n",
        "   print('Before: \\n')\n",
        "   print(dataframe.head(5))\n",
        "  \n",
        "   for label in labels:\n",
        "     dataframe[label] = dataframe[label].apply(lambda txt: text_prepare(txt))\n",
        "  \n",
        "   print('After: \\n')\n",
        "   print(dataframe.head(5))\n",
        " \n",
        "datasets_list = [df_train, df_val, df_test]\n",
        "labels_list = ['Claim', 'Evidence', 'Label']\n",
        " \n",
        "cleaning(datasets_list, labels_list)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Wj-mV3Y4jJQ5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "outputId": "9a1357e8-db06-4aa4-8b2f-be69d291d9c5"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "\n",
              "  <div id=\"df-d4cf87ca-42f5-4ebc-a83c-df2f6e12bfac\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <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>Claim</th>\n",
              "      <th>Evidence</th>\n",
              "      <th>Label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>chris hemsworth appeared in a perfect getaway</td>\n",
              "      <td>hemsworth has also appeared in the science fic...</td>\n",
              "      <td>supports</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>roald dahl is a writer</td>\n",
              "      <td>roald dahl         september    november   was...</td>\n",
              "      <td>supports</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>roald dahl is a governor</td>\n",
              "      <td>roald dahl         september    november   was...</td>\n",
              "      <td>refutes</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>ireland has relatively low  lying mountains</td>\n",
              "      <td>the island s geography comprises relatively lo...</td>\n",
              "      <td>supports</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>ireland does not have relatively low  lying mo...</td>\n",
              "      <td>the island s geography comprises relatively lo...</td>\n",
              "      <td>refutes</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-d4cf87ca-42f5-4ebc-a83c-df2f6e12bfac')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-d4cf87ca-42f5-4ebc-a83c-df2f6e12bfac button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-d4cf87ca-42f5-4ebc-a83c-df2f6e12bfac');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ],
            "text/plain": [
              "                                               Claim  ...     Label\n",
              "0      chris hemsworth appeared in a perfect getaway  ...  supports\n",
              "1                             roald dahl is a writer  ...  supports\n",
              "2                           roald dahl is a governor  ...   refutes\n",
              "3        ireland has relatively low  lying mountains  ...  supports\n",
              "4  ireland does not have relatively low  lying mo...  ...   refutes\n",
              "\n",
              "[5 rows x 3 columns]"
            ]
          },
          "metadata": {},
          "execution_count": 9
        }
      ],
      "source": [
        "# Check.\n",
        "df_train.head()\n",
        "# df_val.head()\n",
        "# df_test.head()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "FGfyr-whiFxi"
      },
      "outputs": [],
      "source": [
        "# Copy needed later for the majority vote.\n",
        "df_test_m = pd.DataFrame.copy(df_test, deep = True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bY94gYdw9t3z"
      },
      "source": [
        "### Text tokenization\n",
        "Text tokenization consists into split a whole sentencences into meaningful pieces of it. It is achieved with the punkt tokenizer by NLTK."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "8cIie76IuHwD"
      },
      "outputs": [],
      "source": [
        "# Note: it takes a while.\n",
        "from nltk.tokenize import wordpunct_tokenize\n",
        "\n",
        "def tokenize_dataset(dataframes: list, labels: list):\n",
        "  for df in dataframes:\n",
        "    for lab in labels:\n",
        "      df[lab] = df[lab].apply(wordpunct_tokenize)\n",
        "\n",
        "# Note: pass just 'Claim' and 'Evidence' otherwise also 'Label' will be turned into a list and following code will fail!\n",
        "tokenize_dataset(datasets_list, labels = ['Claim', 'Evidence'])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "NIYByqYZickb",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "outputId": "074b0d66-e087-49bf-90db-bfb963d6b4e1"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "\n",
              "  <div id=\"df-6b7a1027-ab3d-4134-8b8d-f1f6c0f0a91e\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <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>Claim</th>\n",
              "      <th>Evidence</th>\n",
              "      <th>Label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>[chris, hemsworth, appeared, in, a, perfect, g...</td>\n",
              "      <td>[hemsworth, has, also, appeared, in, the, scie...</td>\n",
              "      <td>supports</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>[roald, dahl, is, a, writer]</td>\n",
              "      <td>[roald, dahl, september, november, was, a, iti...</td>\n",
              "      <td>supports</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>[roald, dahl, is, a, governor]</td>\n",
              "      <td>[roald, dahl, september, november, was, a, iti...</td>\n",
              "      <td>refutes</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>[ireland, has, relatively, low, lying, mountains]</td>\n",
              "      <td>[the, island, s, geography, comprises, relativ...</td>\n",
              "      <td>supports</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>[ireland, does, not, have, relatively, low, ly...</td>\n",
              "      <td>[the, island, s, geography, comprises, relativ...</td>\n",
              "      <td>refutes</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-6b7a1027-ab3d-4134-8b8d-f1f6c0f0a91e')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-6b7a1027-ab3d-4134-8b8d-f1f6c0f0a91e button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-6b7a1027-ab3d-4134-8b8d-f1f6c0f0a91e');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ],
            "text/plain": [
              "                                               Claim  ...     Label\n",
              "0  [chris, hemsworth, appeared, in, a, perfect, g...  ...  supports\n",
              "1                       [roald, dahl, is, a, writer]  ...  supports\n",
              "2                     [roald, dahl, is, a, governor]  ...   refutes\n",
              "3  [ireland, has, relatively, low, lying, mountains]  ...  supports\n",
              "4  [ireland, does, not, have, relatively, low, ly...  ...   refutes\n",
              "\n",
              "[5 rows x 3 columns]"
            ]
          },
          "metadata": {},
          "execution_count": 12
        }
      ],
      "source": [
        "# Check.\n",
        "df_train.head()\n",
        "# df_val.head()\n",
        "# df_test.head()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SzJKKfPGCzQK"
      },
      "source": [
        "## Out-of-vocaboulary terms & embedding\n",
        "At first some functions to handle OOVs terms and token embeddings are define. In particular:\n",
        "* `build_vocaboulary`: create a double vocaboulary from a dataframe to mantain encode and decode tokens;\n",
        "* `load_glove_embedding`: given a chosen embedding dimension it downloads the corresponding pre-trained corpus;\n",
        "* `get_OOV_terms`: given a vocaboulary and a word listings returns the set of OOV terms;\n",
        "* `build_embedding_matrix`: returns the matrix to pass to the embedding layer as kernel initializer."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "WnqffCXUDkEj",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ac59208d-a1ed-4dd4-88e2-941fda014ced"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[==================================================] 100.0% 66.0/66.0MB downloaded\n"
          ]
        }
      ],
      "source": [
        "import gensim\n",
        "import gensim.downloader as gloader\n",
        "\n",
        "from collections import OrderedDict\n",
        "from typing import OrderedDict\n",
        "from tqdm import tqdm\n",
        "import numpy as np\n",
        "\n",
        "def build_vocabulary(dataset: pd.DataFrame, labels: list) -> (OrderedDict[int, str], OrderedDict[str, int], list):\n",
        "    idx2word = OrderedDict()\n",
        "    word2idx = OrderedDict()\n",
        "    \n",
        "    curr_idx = 1 # Start at 1 since 0 is reserved for padding.\n",
        "\n",
        "    for label in labels:\n",
        "      for row in tqdm(dataset[label].values):\n",
        "        for token in row:\n",
        "            if token not in word2idx:\n",
        "                word2idx[token] = curr_idx\n",
        "                idx2word[curr_idx] = token\n",
        "                curr_idx += 1\n",
        "\n",
        "    word_listing = list(idx2word.values())\n",
        "\n",
        "    return idx2word, word2idx, word_listing\n",
        "\n",
        "def load_glove_embedding(embedding_dimension: int) -> gensim.models.keyedvectors.KeyedVectors:\n",
        "    download_path = \"glove-wiki-gigaword-{}\".format(embedding_dimension)\n",
        "\n",
        "    # Sanity check.\n",
        "    try:\n",
        "        embedded_model = gloader.load(download_path)\n",
        "    except ValueError as e:\n",
        "        print(\"Invalid embedding dimension.\")\n",
        "        raise e\n",
        "\n",
        "    return embedded_model\n",
        "\n",
        "def get_OOV_terms(embedding_model: gensim.models.keyedvectors.KeyedVectors, word_listing: List[str]) -> list:\n",
        "    embedding_vocabulary = set(embedding_model.vocab.keys())\n",
        "    oov = set(word_listing).difference(embedding_vocabulary)\n",
        "    return list(oov)\n",
        "\n",
        "def build_embedding_matrix(embedding_model: gensim.models.keyedvectors.KeyedVectors,\n",
        "                           embedding_dimension: int,\n",
        "                           word2idx: OrderedDict[str, int],\n",
        "                           vocab_size: int) -> np.ndarray:\n",
        "\n",
        "    # Note: we take (vocab_size) + 1 since we start counting by 1, having reserved 0 for the padding.\n",
        "    embedding_matrix = np.zeros((vocab_size + 1, embedding_dimension), dtype = np.float32) \n",
        "\n",
        "    for word, idx in tqdm(word2idx.items()):\n",
        "        try:\n",
        "            embedding_vector = embedding_model[word]\n",
        "        except (KeyError, TypeError):\n",
        "            embedding_vector = np.random.uniform(low = -0.05, high = 0.05, size = embedding_dimension)\n",
        "\n",
        "        embedding_matrix[idx] = embedding_vector\n",
        "\n",
        "    return embedding_matrix\n",
        "\n",
        "emb_dim = 50\n",
        "embedded_model = load_glove_embedding(embedding_dimension = emb_dim)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-JIzI7GXhAba"
      },
      "source": [
        "At first we created the vocabularies of words and tags for the training set, we extracted the out-of-vocabularies (OOV) words for it and then its embedding matrix. We assigned random vectors for the OOV words. We repeated the same process for the validation set, taking into account the fact that the OOV words already considered for the training set, are not OOV words for the validation set. The same was done for the test set.\n",
        "\n",
        "Then, we concatenated all embedding matrixes into a single one, and merged the various vocabularies into a single one."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "g6OCAUVQ1kar",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "76867356-38a5-4134-9143-10b4eb5185aa"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 121740/121740 [00:00<00:00, 630987.41it/s]\n",
            "100%|██████████| 121740/121740 [00:00<00:00, 266103.36it/s]\n",
            "100%|██████████| 121740/121740 [00:00<00:00, 1222715.54it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "There are 3028 OOVs in the training set.\n",
            "\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 7165/7165 [00:00<00:00, 481590.14it/s]\n",
            "100%|██████████| 7165/7165 [00:00<00:00, 276695.62it/s]\n",
            "100%|██████████| 7165/7165 [00:00<00:00, 1322253.97it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "There are 237 new OOVs in the validation set.\n",
            "\n",
            "\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 7189/7189 [00:00<00:00, 381474.03it/s]\n",
            "100%|██████████| 7189/7189 [00:00<00:00, 249184.76it/s]\n",
            "100%|██████████| 7189/7189 [00:00<00:00, 862594.45it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "There are 288 new OOVs in the test set.\n",
            "\n",
            "\n"
          ]
        }
      ],
      "source": [
        "# Step 0: load GloVe vocaboulary.\n",
        "vocaboulary = list(embedded_model.vocab.keys()) \n",
        "\n",
        "# Step 1: creation of vocaboulary for the training set and extraction of his OOVs.\n",
        "idx2word_train, word2idx_train, word_listing_train = build_vocabulary(df_train, labels_list)\n",
        "\n",
        "oov_terms_train = get_OOV_terms(embedded_model, word_listing_train)\n",
        "print(f'There are {len(oov_terms_train)} OOVs in the training set.')\n",
        "\n",
        "print('\\n')\n",
        "\n",
        "# Step 2: creation of vocaboulary for the validation set and extraction of his OOVs, taking into account the ones already from training set.\n",
        "idx2word_val, word2idx_val, word_listing_val = build_vocabulary(df_val, labels_list)\n",
        "\n",
        "oov_terms_val = get_OOV_terms(embedded_model, word_listing_val)\n",
        "oov_terms_val = list(set(oov_terms_val).difference(set(oov_terms_train)))\n",
        "print(f'There are {len(oov_terms_val)} new OOVs in the validation set.')\n",
        "\n",
        "print('\\n')\n",
        "\n",
        "# Step 3: creation of vocaboulary for the test set and extraction of his OOVs.\n",
        "idx2word_test, word2idx_test, word_listing_test = build_vocabulary(df_test, labels_list)\n",
        "\n",
        "oov_terms_test = get_OOV_terms(embedded_model, word_listing_test)\n",
        "oov_terms_test = list(set(oov_terms_test).difference((set(oov_terms_train).union(set(oov_terms_val)))))\n",
        "print(f'There are {len(oov_terms_test)} new OOVs in the test set.')\n",
        "\n",
        "print('\\n')\n",
        "\n",
        "# Step 4: merge vocaboulary and OOVs.\n",
        "vocaboulary = np.concatenate((vocaboulary, oov_terms_train, oov_terms_val, oov_terms_test))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "uNsrlXx94nQB",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "df82e751-da66-47ac-f505-074fd7a11059"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 403553/403553 [00:01<00:00, 331179.88it/s]"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Embedding matrix shape: (403554, 50).\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\n"
          ]
        }
      ],
      "source": [
        "word2idx = dict(zip(vocaboulary, range(1, len(vocaboulary) + 1))) # Start from 1 since we reserve 0 for padding.\n",
        "\n",
        "# Whole embedding matrix.\n",
        "embedding_matrix = build_embedding_matrix(embedded_model, emb_dim, word2idx, len(word2idx))\n",
        "print(\"Embedding matrix shape: {}.\".format(embedding_matrix.shape))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GgCUXWRe99gF"
      },
      "source": [
        "### Padding and generator creation\n",
        "We will pad up to the maximum number of token since we need to preserve informative content.\n",
        "\n",
        "Note that we shouldn't have info on the maximum number of tokens on the test set, but for sake of simplicity we will calculate the number of tokens on the test set as well, in order to avoid splitting and concatenation at inference time."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "wFYVTnhA1Y3p",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 544
        },
        "outputId": "947acbef-f996-43aa-c63a-28a975d206c7"
      },
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ],
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "def extract_info(dataframe: pd.DataFrame) -> int:\n",
        "  plt.figure()\n",
        "  plt.title('Claim length distribution vs. Evidence length distribution ')\n",
        "  \n",
        "  no_token_claim = [len(sentence) for sentence in dataframe['Claim']]\n",
        "  n_c, _, _ = plt.hist(no_token_claim, bins = [i for i in range(1, 120)], cumulative = False)\n",
        "\n",
        "  no_token_evidence = [len(sentence) for sentence in dataframe['Evidence']]\n",
        "  n_e, _, _ = plt.hist(no_token_evidence, bins = [i for i in range(1, 120)], cumulative = False)\n",
        "\n",
        "  return max(max(no_token_claim), max(no_token_evidence))\n",
        "\n",
        "max_tokens = max(extract_info(df_train), extract_info(df_val))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jHudXOw3hIkK"
      },
      "source": [
        "Finally, we put our dataframes into matrixes, padding and truncating the sentences, using the aforementioned vocabulary to encode the words into integers."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "TJmUAjtLqE9u"
      },
      "outputs": [],
      "source": [
        "binary_map = {'refutes': 0, 'supports': 1}\n",
        "\n",
        "def encode_sentence(sentence, word2idx):\n",
        "  return [word2idx[words] for words in sentence]\n",
        "\n",
        "def encode_dataframe(dataframe, word2idx):\n",
        "  encoded_dataframe = pd.DataFrame(columns = ['Claim', 'Evidence', 'Label'])\n",
        "\n",
        "  for columns in ['Claim', 'Evidence'] :\n",
        "    encoded_dataframe[columns] = dataframe[columns].apply(lambda sentence: encode_sentence(sentence, word2idx))\n",
        "\n",
        "  encoded_dataframe['Label'] = dataframe['Label'].map(binary_map)\n",
        "\n",
        "  return encoded_dataframe\n",
        "\n",
        "train = encode_dataframe(df_train, word2idx)\n",
        "val = encode_dataframe(df_val, word2idx)\n",
        "test = encode_dataframe(df_test, word2idx)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "y2JfixYSrCnl"
      },
      "outputs": [],
      "source": [
        "from keras.preprocessing.sequence import pad_sequences\n",
        "\n",
        "train_claim = pad_sequences(train['Claim'], maxlen = max_tokens, padding = 'post')\n",
        "train_evidence = pad_sequences(train['Evidence'], maxlen = max_tokens, padding = 'post')\n",
        "\n",
        "val_claim = pad_sequences(val['Claim'], maxlen = max_tokens, padding = 'post')\n",
        "val_evidence = pad_sequences(val['Evidence'], maxlen = max_tokens, padding = 'post')\n",
        "\n",
        "test_claim = pad_sequences(test['Claim'], maxlen = max_tokens, padding = 'post')\n",
        "test_evidence = pad_sequences(test['Evidence'], maxlen = max_tokens, padding = 'post')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-OPMAgBKtKTB"
      },
      "source": [
        "## Neural architectures"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Rcw6n4TKj5N-"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras.layers import SimpleRNN, Bidirectional, Dense, TimeDistributed, Input, Reshape, Embedding, concatenate, add, average, Average, Dropout, dot, maximum\n",
        "from tensorflow.keras.models import Model\n",
        "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint\n",
        "from tensorflow.keras.initializers import Constant, glorot_normal\n",
        "from tensorflow.keras.optimizers import RMSprop\n",
        "from tensorflow.keras.optimizers.schedules import ExponentialDecay\n",
        "from tensorflow.keras import regularizers\n",
        "from tensorflow.keras.losses import BinaryCrossentropy \n",
        "\n",
        "\n",
        "# Note: the learning rate is not parametrized since we use a scheduler.\n",
        "def fact_checker(max_tokens: int, embedding_dimension: int, embedding_matrix: np.ndarray, merge_strategy: str, embedding_strategy: str, cos_similarity: bool, \n",
        "                 dropout: float = 0.40, recurrent_dropout: float = 0.25, initial_learning_rate: float = 0.0001) -> Model:\n",
        "  hidden_nodes = emb_dim\n",
        "\n",
        "  ''' Network structure '''\n",
        "  claim_input = Input(shape = (max_tokens, ), name = 'claim_input')\n",
        "\n",
        "  evidence_input = Input(shape = (max_tokens, ), name = 'evidence_input')\n",
        "\n",
        "  shared_embedding = Embedding(input_dim = embedding_matrix.shape[0],\n",
        "                               output_dim = embedding_dimension,\n",
        "                               mask_zero = True,\n",
        "                               embeddings_initializer = Constant(embedding_matrix),\n",
        "                               trainable = False,\n",
        "                               name = 'embedding_layer')\n",
        "  \n",
        "  embedded_claim = shared_embedding(claim_input)\n",
        "  embedded_evidence = shared_embedding(evidence_input)\n",
        "\n",
        "  # Sentence encoding\n",
        "  if embedding_strategy == 'last_state':\n",
        "    shared_rnn = Bidirectional(SimpleRNN(hidden_nodes,\n",
        "                                    # return_sequences: whether to return the last output in the output sequence, or the full sequence. \n",
        "                                    return_sequences = False, return_state = False,\n",
        "                                    dropout = dropout, recurrent_dropout = recurrent_dropout, \n",
        "                                    name = 'rnn_layer_claim'))\n",
        "\n",
        "    rnn_layer_claim = shared_rnn(embedded_claim)\n",
        "    rnn_layer_evidence = shared_rnn(embedded_evidence)\n",
        "\n",
        "  elif embedding_strategy == 'average_state':\n",
        "    shared_rnn = Bidirectional(SimpleRNN(hidden_nodes,\n",
        "                                         # return_sequences: whether to return the last output in the output sequence, or the full sequence. \n",
        "                                         return_sequences = True, return_state = False,\n",
        "                                         dropout = dropout, recurrent_dropout = recurrent_dropout, \n",
        "                                         name = 'rnn_layer_claim'))\n",
        "\n",
        "    rnn_layer_claim = shared_rnn(embedded_claim)\n",
        "    rnn_layer_evidence = shared_rnn(embedded_evidence)\n",
        "\n",
        "    rnn_layer_claim = tf.reduce_mean(rnn_layer_claim, axis = 1)\n",
        "    rnn_layer_evidence = tf.reduce_mean(rnn_layer_evidence, axis = 1)\n",
        "\n",
        "  elif embedding_strategy == 'mlp':\n",
        "    reshaper = Reshape((max_tokens * emb_dim, ))\n",
        "    mlp0 = Dense(units = max_tokens, activation = 'selu', name = 'mpl1_claim')\n",
        "    mlp1 = Dropout(rate = dropout, name = 'claim_dropout')\n",
        "    mlp2 = Dense(units = emb_dim, activation = 'selu', name = 'mpl2_claim')\n",
        "\n",
        "    claim_reshape = reshaper(embedded_claim)\n",
        "    mlp0_claim = mlp0(claim_reshape)\n",
        "    mlp1_claim = mlp1(mlp0_claim)\n",
        "    rnn_layer_claim = mlp2(mlp1_claim)\n",
        "\n",
        "    evidence_reshape = reshaper(embedded_evidence)\n",
        "    mlp0_evidence = mlp0(evidence_reshape)\n",
        "    mlp1_evidence = mlp1(mlp0_evidence)\n",
        "    rnn_layer_evidence = mlp2(mlp1_evidence)\n",
        "\n",
        "  elif embedding_strategy == 'bov':\n",
        "    # Mean on axis 1 since we want to average all embeddings, obtaining a single one with the same dimension.\n",
        "    rnn_layer_claim = tf.reduce_mean(embedded_claim, axis = 1)\n",
        "    rnn_layer_evidence = tf.reduce_mean(embedded_evidence, axis = 1)\n",
        "\n",
        "  ''' Merge strategy '''\n",
        "  if merge_strategy == 'add':\n",
        "    merge = add([rnn_layer_claim, rnn_layer_evidence])\n",
        "  elif merge_strategy == 'average':\n",
        "    merge = average([rnn_layer_claim, rnn_layer_evidence])\n",
        "  elif merge_strategy == 'concatenate':\n",
        "    merge = concatenate([rnn_layer_claim, rnn_layer_evidence])\n",
        "  elif merge_strategy == 'maximum':\n",
        "    merge = maximum([rnn_layer_claim, rnn_layer_evidence])\n",
        "\n",
        "  if cos_similarity:\n",
        "    # Since the cosine similarity it is just a normalized dot product between vector it suffices to do as follows.\n",
        "    cos_sim = dot(inputs = [rnn_layer_claim, rnn_layer_evidence], axes = -1, normalize = True)\n",
        "    merge = concatenate([merge, cos_sim])\n",
        "\n",
        "  classifier = Dense(units = max_tokens/2, activation = 'selu', \n",
        "                     kernel_regularizer = regularizers.l1_l2(l1 = 1e-5, l2 = 1e-4),\n",
        "                     bias_regularizer = regularizers.l2(1e-4),\n",
        "                     activity_regularizer = regularizers.l2(1e-5),\n",
        "                     kernel_initializer = 'glorot_normal', name = 'classifier1')(merge)\n",
        "  classifier = Dropout(rate = dropout, name = 'classifier_dropout1')(classifier)\n",
        "\n",
        "  classifier = Dense(units = max_tokens/4, activation = 'selu', \n",
        "                     kernel_regularizer = regularizers.l1_l2(l1 = 1e-5, l2 = 1e-4),\n",
        "                     bias_regularizer = regularizers.l2(1e-4),\n",
        "                     activity_regularizer = regularizers.l2(1e-5),\n",
        "                      kernel_initializer = 'glorot_normal', name = 'classifier2')(classifier)\n",
        "  classifier = Dropout(rate = dropout, name = 'classifier_dropout2')(classifier)\n",
        "  \n",
        "  classifier = Dense(units = max_tokens/8, activation = 'selu', \n",
        "                     kernel_regularizer = regularizers.l1_l2(l1 = 1e-5, l2 = 1e-4),\n",
        "                     bias_regularizer = regularizers.l2(1e-4),\n",
        "                     activity_regularizer = regularizers.l2(1e-5),\n",
        "                      kernel_initializer = 'glorot_normal', name = 'classifier3')(classifier)\n",
        "  classifier = Dropout(rate = dropout, name = 'classifier_dropout3')(classifier)\n",
        "\n",
        "  classifier = Dense(units = 1, activation = 'sigmoid', name = 'classifier4')(classifier)\n",
        "\n",
        "  end_to_end = Model(inputs = (claim_input, evidence_input), outputs = classifier)\n",
        "\n",
        "  # Compile the model.\n",
        "  scheduler = ExponentialDecay(initial_learning_rate,\n",
        "                              decay_steps = 100000,\n",
        "                              decay_rate = 0.96,\n",
        "                              staircase = True)\n",
        "\n",
        "  end_to_end.compile(loss = focal_loss.BinaryFocalLoss(gamma = 2),\n",
        "                     optimizer = 'adam', #RMSprop(learning_rate = scheduler),\n",
        "                     metrics = ['binary_accuracy'])\n",
        "\n",
        "  # Check if the structure is correct.\n",
        "  end_to_end.summary()\n",
        "\n",
        "  return end_to_end"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8vcvRNHvlbO8",
        "outputId": "1657f454-667d-482a-e637-7fdf2bd5940e"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"model\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " claim_input (InputLayer)       [(None, 121)]        0           []                               \n",
            "                                                                                                  \n",
            " evidence_input (InputLayer)    [(None, 121)]        0           []                               \n",
            "                                                                                                  \n",
            " embedding_layer (Embedding)    (None, 121, 50)      20177700    ['claim_input[0][0]',            \n",
            "                                                                  'evidence_input[0][0]']         \n",
            "                                                                                                  \n",
            " bidirectional (Bidirectional)  (None, 100)          10100       ['embedding_layer[0][0]',        \n",
            "                                                                  'embedding_layer[1][0]']        \n",
            "                                                                                                  \n",
            " add (Add)                      (None, 100)          0           ['bidirectional[0][0]',          \n",
            "                                                                  'bidirectional[1][0]']          \n",
            "                                                                                                  \n",
            " dot (Dot)                      (None, 1)            0           ['bidirectional[0][0]',          \n",
            "                                                                  'bidirectional[1][0]']          \n",
            "                                                                                                  \n",
            " concatenate (Concatenate)      (None, 101)          0           ['add[0][0]',                    \n",
            "                                                                  'dot[0][0]']                    \n",
            "                                                                                                  \n",
            " classifier1 (Dense)            (None, 60)           6120        ['concatenate[0][0]']            \n",
            "                                                                                                  \n",
            " classifier_dropout1 (Dropout)  (None, 60)           0           ['classifier1[0][0]']            \n",
            "                                                                                                  \n",
            " classifier2 (Dense)            (None, 30)           1830        ['classifier_dropout1[0][0]']    \n",
            "                                                                                                  \n",
            " classifier_dropout2 (Dropout)  (None, 30)           0           ['classifier2[0][0]']            \n",
            "                                                                                                  \n",
            " classifier3 (Dense)            (None, 15)           465         ['classifier_dropout2[0][0]']    \n",
            "                                                                                                  \n",
            " classifier_dropout3 (Dropout)  (None, 15)           0           ['classifier3[0][0]']            \n",
            "                                                                                                  \n",
            " classifier4 (Dense)            (None, 1)            16          ['classifier_dropout3[0][0]']    \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 20,196,231\n",
            "Trainable params: 18,531\n",
            "Non-trainable params: 20,177,700\n",
            "__________________________________________________________________________________________________\n",
            "Model: \"model_1\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " claim_input (InputLayer)       [(None, 121)]        0           []                               \n",
            "                                                                                                  \n",
            " evidence_input (InputLayer)    [(None, 121)]        0           []                               \n",
            "                                                                                                  \n",
            " embedding_layer (Embedding)    (None, 121, 50)      20177700    ['claim_input[0][0]',            \n",
            "                                                                  'evidence_input[0][0]']         \n",
            "                                                                                                  \n",
            " bidirectional_1 (Bidirectional  (None, 100)         10100       ['embedding_layer[0][0]',        \n",
            " )                                                                'embedding_layer[1][0]']        \n",
            "                                                                                                  \n",
            " average (Average)              (None, 100)          0           ['bidirectional_1[0][0]',        \n",
            "                                                                  'bidirectional_1[1][0]']        \n",
            "                                                                                                  \n",
            " dot_1 (Dot)                    (None, 1)            0           ['bidirectional_1[0][0]',        \n",
            "                                                                  'bidirectional_1[1][0]']        \n",
            "                                                                                                  \n",
            " concatenate_1 (Concatenate)    (None, 101)          0           ['average[0][0]',                \n",
            "                                                                  'dot_1[0][0]']                  \n",
            "                                                                                                  \n",
            " classifier1 (Dense)            (None, 60)           6120        ['concatenate_1[0][0]']          \n",
            "                                                                                                  \n",
            " classifier_dropout1 (Dropout)  (None, 60)           0           ['classifier1[0][0]']            \n",
            "                                                                                                  \n",
            " classifier2 (Dense)            (None, 30)           1830        ['classifier_dropout1[0][0]']    \n",
            "                                                                                                  \n",
            " classifier_dropout2 (Dropout)  (None, 30)           0           ['classifier2[0][0]']            \n",
            "                                                                                                  \n",
            " classifier3 (Dense)            (None, 15)           465         ['classifier_dropout2[0][0]']    \n",
            "                                                                                                  \n",
            " classifier_dropout3 (Dropout)  (None, 15)           0           ['classifier3[0][0]']            \n",
            "                                                                                                  \n",
            " classifier4 (Dense)            (None, 1)            16          ['classifier_dropout3[0][0]']    \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 20,196,231\n",
            "Trainable params: 18,531\n",
            "Non-trainable params: 20,177,700\n",
            "__________________________________________________________________________________________________\n",
            "Model: \"model_2\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " claim_input (InputLayer)       [(None, 121)]        0           []                               \n",
            "                                                                                                  \n",
            " evidence_input (InputLayer)    [(None, 121)]        0           []                               \n",
            "                                                                                                  \n",
            " embedding_layer (Embedding)    (None, 121, 50)      20177700    ['claim_input[0][0]',            \n",
            "                                                                  'evidence_input[0][0]']         \n",
            "                                                                                                  \n",
            " bidirectional_2 (Bidirectional  (None, 100)         10100       ['embedding_layer[0][0]',        \n",
            " )                                                                'embedding_layer[1][0]']        \n",
            "                                                                                                  \n",
            " concatenate_2 (Concatenate)    (None, 200)          0           ['bidirectional_2[0][0]',        \n",
            "                                                                  'bidirectional_2[1][0]']        \n",
            "                                                                                                  \n",
            " dot_2 (Dot)                    (None, 1)            0           ['bidirectional_2[0][0]',        \n",
            "                                                                  'bidirectional_2[1][0]']        \n",
            "                                                                                                  \n",
            " concatenate_3 (Concatenate)    (None, 201)          0           ['concatenate_2[0][0]',          \n",
            "                                                                  'dot_2[0][0]']                  \n",
            "                                                                                                  \n",
            " classifier1 (Dense)            (None, 60)           12120       ['concatenate_3[0][0]']          \n",
            "                                                                                                  \n",
            " classifier_dropout1 (Dropout)  (None, 60)           0           ['classifier1[0][0]']            \n",
            "                                                                                                  \n",
            " classifier2 (Dense)            (None, 30)           1830        ['classifier_dropout1[0][0]']    \n",
            "                                                                                                  \n",
            " classifier_dropout2 (Dropout)  (None, 30)           0           ['classifier2[0][0]']            \n",
            "                                                                                                  \n",
            " classifier3 (Dense)            (None, 15)           465         ['classifier_dropout2[0][0]']    \n",
            "                                                                                                  \n",
            " classifier_dropout3 (Dropout)  (None, 15)           0           ['classifier3[0][0]']            \n",
            "                                                                                                  \n",
            " classifier4 (Dense)            (None, 1)            16          ['classifier_dropout3[0][0]']    \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 20,202,231\n",
            "Trainable params: 24,531\n",
            "Non-trainable params: 20,177,700\n",
            "__________________________________________________________________________________________________\n",
            "Model: \"model_3\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " claim_input (InputLayer)       [(None, 121)]        0           []                               \n",
            "                                                                                                  \n",
            " evidence_input (InputLayer)    [(None, 121)]        0           []                               \n",
            "                                                                                                  \n",
            " embedding_layer (Embedding)    (None, 121, 50)      20177700    ['claim_input[0][0]',            \n",
            "                                                                  'evidence_input[0][0]']         \n",
            "                                                                                                  \n",
            " bidirectional_3 (Bidirectional  (None, 100)         10100       ['embedding_layer[0][0]',        \n",
            " )                                                                'embedding_layer[1][0]']        \n",
            "                                                                                                  \n",
            " maximum (Maximum)              (None, 100)          0           ['bidirectional_3[0][0]',        \n",
            "                                                                  'bidirectional_3[1][0]']        \n",
            "                                                                                                  \n",
            " dot_3 (Dot)                    (None, 1)            0           ['bidirectional_3[0][0]',        \n",
            "                                                                  'bidirectional_3[1][0]']        \n",
            "                                                                                                  \n",
            " concatenate_4 (Concatenate)    (None, 101)          0           ['maximum[0][0]',                \n",
            "                                                                  'dot_3[0][0]']                  \n",
            "                                                                                                  \n",
            " classifier1 (Dense)            (None, 60)           6120        ['concatenate_4[0][0]']          \n",
            "                                                                                                  \n",
            " classifier_dropout1 (Dropout)  (None, 60)           0           ['classifier1[0][0]']            \n",
            "                                                                                                  \n",
            " classifier2 (Dense)            (None, 30)           1830        ['classifier_dropout1[0][0]']    \n",
            "                                                                                                  \n",
            " classifier_dropout2 (Dropout)  (None, 30)           0           ['classifier2[0][0]']            \n",
            "                                                                                                  \n",
            " classifier3 (Dense)            (None, 15)           465         ['classifier_dropout2[0][0]']    \n",
            "                                                                                                  \n",
            " classifier_dropout3 (Dropout)  (None, 15)           0           ['classifier3[0][0]']            \n",
            "                                                                                                  \n",
            " classifier4 (Dense)            (None, 1)            16          ['classifier_dropout3[0][0]']    \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 20,196,231\n",
            "Trainable params: 18,531\n",
            "Non-trainable params: 20,177,700\n",
            "__________________________________________________________________________________________________\n"
          ]
        }
      ],
      "source": [
        "end_to_end_add = fact_checker(max_tokens = max_tokens,\n",
        "                              embedding_dimension = emb_dim,\n",
        "                              embedding_matrix = embedding_matrix,\n",
        "                              merge_strategy = 'add',\n",
        "                              embedding_strategy = 'last_state',\n",
        "                              cos_similarity = True)  \n",
        "\n",
        "end_to_end_average = fact_checker(max_tokens = max_tokens,\n",
        "                                  embedding_dimension = emb_dim,\n",
        "                                  embedding_matrix = embedding_matrix,\n",
        "                                  merge_strategy = 'average',\n",
        "                                  embedding_strategy = 'last_state',\n",
        "                                  cos_similarity = True)\n",
        "\n",
        "end_to_end_concat = fact_checker(max_tokens = max_tokens,\n",
        "                                 embedding_dimension = emb_dim,\n",
        "                                 embedding_matrix = embedding_matrix,\n",
        "                                 merge_strategy = 'concatenate',\n",
        "                                 embedding_strategy = 'last_state',\n",
        "                                 cos_similarity = True)\n",
        "\n",
        "end_to_end_maximum = fact_checker(max_tokens = max_tokens,\n",
        "                                  embedding_dimension = emb_dim,\n",
        "                                  embedding_matrix = embedding_matrix,\n",
        "                                  merge_strategy = 'maximum',\n",
        "                                  embedding_strategy = 'last_state',\n",
        "                                  cos_similarity = True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "LsQbeFfHV9Dn",
        "outputId": "12cab11a-7a74-4cb0-c467-4a9e5696f46c"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {},
          "execution_count": 21
        }
      ],
      "source": [
        "from tensorflow.keras.utils import plot_model\n",
        "plot_model(end_to_end_add, show_shapes = True)\n",
        "plot_model(end_to_end_average, show_shapes = True)\n",
        "plot_model(end_to_end_concat, show_shapes = True)\n",
        "plot_model(end_to_end_maximum, show_shapes = True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "V4aH0a7OixrX"
      },
      "outputs": [],
      "source": [
        "checkpoint = [EarlyStopping(monitor = 'val_loss', \n",
        "                            patience = 5,\n",
        "                            verbose = 1,\n",
        "                            mode = 'auto'),\n",
        "              ModelCheckpoint(filepath = './models/end_to_end_add.h5',\n",
        "                              monitor = 'val_binary_accuracy',\n",
        "                              verbose = 1,\n",
        "                              save_best_only = True)]"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "The following code is just to handle the loading and training of a previously trained model."
      ],
      "metadata": {
        "id": "HVvS7KjYDrnr"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "LOAD_MODEL = [True, True, True, False]\n",
        "FIT_MODEL = [False, False, False, True]"
      ],
      "metadata": {
        "id": "aJCl08iSMltw"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "0RZHL9haVG69"
      },
      "outputs": [],
      "source": [
        "if LOAD_MODEL[0]:\n",
        "  end_to_end_add = load_model(filepath = './models/end_to_end_add.h5',\n",
        "                            # custom_objects = {'focal_loss': focal_loss.BinaryFocalLoss(gamma = 2), }\n",
        "                            ) \n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "j3NUbc5x-C8j",
        "outputId": "584f77f0-a5ca-4387-98af-1970387e5ae1"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1319 - binary_accuracy: 0.7777\n",
            "Epoch 00001: val_binary_accuracy improved from -inf to 0.61703, saving model to ./models/end_to_end_add.h5\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/keras/engine/functional.py:1410: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.\n",
            "  layer_config = serialize_layer_fn(layer)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r1903/1903 [==============================] - 935s 489ms/step - loss: 0.1319 - binary_accuracy: 0.7777 - val_loss: 0.1750 - val_binary_accuracy: 0.6170\n",
            "Epoch 2/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1314 - binary_accuracy: 0.7787\n",
            "Epoch 00002: val_binary_accuracy did not improve from 0.61703\n",
            "1903/1903 [==============================] - 872s 458ms/step - loss: 0.1314 - binary_accuracy: 0.7787 - val_loss: 0.1728 - val_binary_accuracy: 0.6131\n",
            "Epoch 3/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1307 - binary_accuracy: 0.7811\n",
            "Epoch 00003: val_binary_accuracy improved from 0.61703 to 0.62610, saving model to ./models/end_to_end_add.h5\n",
            "1903/1903 [==============================] - 936s 492ms/step - loss: 0.1307 - binary_accuracy: 0.7811 - val_loss: 0.1670 - val_binary_accuracy: 0.6261\n",
            "Epoch 4/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1307 - binary_accuracy: 0.7807\n",
            "Epoch 00004: val_binary_accuracy improved from 0.62610 to 0.63545, saving model to ./models/end_to_end_add.h5\n",
            "1903/1903 [==============================] - 944s 496ms/step - loss: 0.1307 - binary_accuracy: 0.7807 - val_loss: 0.1630 - val_binary_accuracy: 0.6355\n",
            "Epoch 5/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1304 - binary_accuracy: 0.7809\n",
            "Epoch 00005: val_binary_accuracy did not improve from 0.63545\n",
            "1903/1903 [==============================] - 876s 460ms/step - loss: 0.1304 - binary_accuracy: 0.7809 - val_loss: 0.1719 - val_binary_accuracy: 0.6294\n",
            "Epoch 6/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1304 - binary_accuracy: 0.7812\n",
            "Epoch 00006: val_binary_accuracy did not improve from 0.63545\n",
            "1903/1903 [==============================] - 880s 462ms/step - loss: 0.1304 - binary_accuracy: 0.7812 - val_loss: 0.1668 - val_binary_accuracy: 0.6294\n",
            "Epoch 7/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1296 - binary_accuracy: 0.7835\n",
            "Epoch 00007: val_binary_accuracy did not improve from 0.63545\n",
            "1903/1903 [==============================] - 886s 466ms/step - loss: 0.1296 - binary_accuracy: 0.7835 - val_loss: 0.1786 - val_binary_accuracy: 0.6289\n",
            "Epoch 8/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1295 - binary_accuracy: 0.7827\n",
            "Epoch 00008: val_binary_accuracy improved from 0.63545 to 0.63671, saving model to ./models/end_to_end_add.h5\n",
            "1903/1903 [==============================] - 946s 497ms/step - loss: 0.1295 - binary_accuracy: 0.7827 - val_loss: 0.1610 - val_binary_accuracy: 0.6367\n",
            "Epoch 9/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1293 - binary_accuracy: 0.7838\n",
            "Epoch 00009: val_binary_accuracy did not improve from 0.63671\n",
            "1903/1903 [==============================] - 883s 464ms/step - loss: 0.1293 - binary_accuracy: 0.7838 - val_loss: 0.1625 - val_binary_accuracy: 0.6331\n",
            "Epoch 10/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1292 - binary_accuracy: 0.7834\n",
            "Epoch 00010: val_binary_accuracy did not improve from 0.63671\n",
            "1903/1903 [==============================] - 873s 459ms/step - loss: 0.1292 - binary_accuracy: 0.7834 - val_loss: 0.1626 - val_binary_accuracy: 0.6181\n",
            "Epoch 11/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1293 - binary_accuracy: 0.7849\n",
            "Epoch 00011: val_binary_accuracy improved from 0.63671 to 0.64201, saving model to ./models/end_to_end_add.h5\n",
            "1903/1903 [==============================] - 933s 491ms/step - loss: 0.1293 - binary_accuracy: 0.7849 - val_loss: 0.1656 - val_binary_accuracy: 0.6420\n",
            "Epoch 12/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1287 - binary_accuracy: 0.7850\n",
            "Epoch 00012: val_binary_accuracy did not improve from 0.64201\n",
            "1903/1903 [==============================] - 868s 456ms/step - loss: 0.1287 - binary_accuracy: 0.7850 - val_loss: 0.1854 - val_binary_accuracy: 0.6204\n",
            "Epoch 13/100\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1289 - binary_accuracy: 0.7848\n",
            "Epoch 00013: val_binary_accuracy did not improve from 0.64201\n",
            "1903/1903 [==============================] - 880s 463ms/step - loss: 0.1289 - binary_accuracy: 0.7848 - val_loss: 0.1752 - val_binary_accuracy: 0.6293\n",
            "Epoch 00013: early stopping\n"
          ]
        }
      ],
      "source": [
        "# Train the model.\n",
        "if FIT_MODEL[0]:\n",
        "  end_to_end_add.fit(x = (train_claim, train_evidence), y = np.array(train['Label']),\n",
        "                   validation_data = ((val_claim, val_evidence), np.array(val['Label'])),\n",
        "                   batch_size = 64,\n",
        "                   epochs = 100, callbacks = checkpoint)"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "if LOAD_MODEL[1]:\n",
        "  end_to_end_average = load_model(filepath = './models/end_to_end_average.h5')"
      ],
      "metadata": {
        "id": "jZ0kJA4RlLwf"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "lR0Rfii0ZmPz",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a711d061-b246-4be9-a6ec-b9f94495fbb7"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.6182 - binary_accuracy: 0.5006\n",
            "Epoch 00001: val_binary_accuracy improved from -inf to 0.50998, saving model to ./models/end_to_end_add.h5\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/keras/engine/functional.py:1410: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.\n",
            "  layer_config = serialize_layer_fn(layer)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r113/113 [==============================] - 90s 738ms/step - loss: 0.6182 - binary_accuracy: 0.5006 - val_loss: 0.2148 - val_binary_accuracy: 0.5100\n",
            "Epoch 2/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.3965 - binary_accuracy: 0.5073\n",
            "Epoch 00002: val_binary_accuracy did not improve from 0.50998\n",
            "113/113 [==============================] - 59s 522ms/step - loss: 0.3965 - binary_accuracy: 0.5073 - val_loss: 0.2009 - val_binary_accuracy: 0.5084\n",
            "Epoch 3/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.3057 - binary_accuracy: 0.5042\n",
            "Epoch 00003: val_binary_accuracy improved from 0.50998 to 0.52338, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 84s 747ms/step - loss: 0.3057 - binary_accuracy: 0.5042 - val_loss: 0.1943 - val_binary_accuracy: 0.5234\n",
            "Epoch 4/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.2565 - binary_accuracy: 0.5016\n",
            "Epoch 00004: val_binary_accuracy improved from 0.52338 to 0.52408, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 80s 707ms/step - loss: 0.2565 - binary_accuracy: 0.5016 - val_loss: 0.1929 - val_binary_accuracy: 0.5241\n",
            "Epoch 5/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.2313 - binary_accuracy: 0.5024\n",
            "Epoch 00005: val_binary_accuracy did not improve from 0.52408\n",
            "113/113 [==============================] - 59s 518ms/step - loss: 0.2313 - binary_accuracy: 0.5024 - val_loss: 0.1926 - val_binary_accuracy: 0.5171\n",
            "Epoch 6/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.2187 - binary_accuracy: 0.4914\n",
            "Epoch 00006: val_binary_accuracy improved from 0.52408 to 0.52729, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 82s 726ms/step - loss: 0.2187 - binary_accuracy: 0.4914 - val_loss: 0.1922 - val_binary_accuracy: 0.5273\n",
            "Epoch 7/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.2077 - binary_accuracy: 0.4978\n",
            "Epoch 00007: val_binary_accuracy improved from 0.52729 to 0.52994, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 86s 763ms/step - loss: 0.2077 - binary_accuracy: 0.4978 - val_loss: 0.1917 - val_binary_accuracy: 0.5299\n",
            "Epoch 8/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.2020 - binary_accuracy: 0.5067\n",
            "Epoch 00008: val_binary_accuracy improved from 0.52994 to 0.53510, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 78s 696ms/step - loss: 0.2020 - binary_accuracy: 0.5067 - val_loss: 0.1914 - val_binary_accuracy: 0.5351\n",
            "Epoch 9/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1988 - binary_accuracy: 0.4933\n",
            "Epoch 00009: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 59s 518ms/step - loss: 0.1988 - binary_accuracy: 0.4933 - val_loss: 0.1914 - val_binary_accuracy: 0.5206\n",
            "Epoch 10/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1959 - binary_accuracy: 0.5080\n",
            "Epoch 00010: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 58s 512ms/step - loss: 0.1959 - binary_accuracy: 0.5080 - val_loss: 0.1912 - val_binary_accuracy: 0.5153\n",
            "Epoch 11/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1953 - binary_accuracy: 0.4948\n",
            "Epoch 00011: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 58s 511ms/step - loss: 0.1953 - binary_accuracy: 0.4948 - val_loss: 0.1908 - val_binary_accuracy: 0.5274\n",
            "Epoch 12/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1935 - binary_accuracy: 0.5002\n",
            "Epoch 00012: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 57s 508ms/step - loss: 0.1935 - binary_accuracy: 0.5002 - val_loss: 0.1907 - val_binary_accuracy: 0.5246\n",
            "Epoch 13/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1927 - binary_accuracy: 0.5061\n",
            "Epoch 00013: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 58s 513ms/step - loss: 0.1927 - binary_accuracy: 0.5061 - val_loss: 0.1905 - val_binary_accuracy: 0.5228\n",
            "Epoch 14/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1917 - binary_accuracy: 0.5067\n",
            "Epoch 00014: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 58s 514ms/step - loss: 0.1917 - binary_accuracy: 0.5067 - val_loss: 0.1902 - val_binary_accuracy: 0.5280\n",
            "Epoch 15/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1911 - binary_accuracy: 0.5016\n",
            "Epoch 00015: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 58s 511ms/step - loss: 0.1911 - binary_accuracy: 0.5016 - val_loss: 0.1900 - val_binary_accuracy: 0.5312\n",
            "Epoch 16/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1909 - binary_accuracy: 0.5037\n",
            "Epoch 00016: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 58s 513ms/step - loss: 0.1909 - binary_accuracy: 0.5037 - val_loss: 0.1897 - val_binary_accuracy: 0.5312\n",
            "Epoch 17/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1904 - binary_accuracy: 0.5061\n",
            "Epoch 00017: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 58s 512ms/step - loss: 0.1904 - binary_accuracy: 0.5061 - val_loss: 0.1896 - val_binary_accuracy: 0.5255\n",
            "Epoch 18/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1899 - binary_accuracy: 0.5033\n",
            "Epoch 00018: val_binary_accuracy did not improve from 0.53510\n",
            "113/113 [==============================] - 58s 511ms/step - loss: 0.1899 - binary_accuracy: 0.5033 - val_loss: 0.1893 - val_binary_accuracy: 0.5302\n",
            "Epoch 19/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1894 - binary_accuracy: 0.5183\n",
            "Epoch 00019: val_binary_accuracy improved from 0.53510 to 0.53580, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 81s 717ms/step - loss: 0.1894 - binary_accuracy: 0.5183 - val_loss: 0.1889 - val_binary_accuracy: 0.5358\n",
            "Epoch 20/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1891 - binary_accuracy: 0.5198\n",
            "Epoch 00020: val_binary_accuracy improved from 0.53580 to 0.53747, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 82s 725ms/step - loss: 0.1891 - binary_accuracy: 0.5198 - val_loss: 0.1885 - val_binary_accuracy: 0.5375\n",
            "Epoch 21/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1891 - binary_accuracy: 0.5084\n",
            "Epoch 00021: val_binary_accuracy did not improve from 0.53747\n",
            "113/113 [==============================] - 59s 523ms/step - loss: 0.1891 - binary_accuracy: 0.5084 - val_loss: 0.1884 - val_binary_accuracy: 0.5351\n",
            "Epoch 22/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1887 - binary_accuracy: 0.5155\n",
            "Epoch 00022: val_binary_accuracy did not improve from 0.53747\n",
            "113/113 [==============================] - 58s 517ms/step - loss: 0.1887 - binary_accuracy: 0.5155 - val_loss: 0.1880 - val_binary_accuracy: 0.5352\n",
            "Epoch 23/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1882 - binary_accuracy: 0.5173\n",
            "Epoch 00023: val_binary_accuracy did not improve from 0.53747\n",
            "113/113 [==============================] - 58s 516ms/step - loss: 0.1882 - binary_accuracy: 0.5173 - val_loss: 0.1876 - val_binary_accuracy: 0.5368\n",
            "Epoch 24/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1879 - binary_accuracy: 0.5190\n",
            "Epoch 00024: val_binary_accuracy improved from 0.53747 to 0.53775, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 79s 703ms/step - loss: 0.1879 - binary_accuracy: 0.5190 - val_loss: 0.1872 - val_binary_accuracy: 0.5378\n",
            "Epoch 25/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1877 - binary_accuracy: 0.5172\n",
            "Epoch 00025: val_binary_accuracy did not improve from 0.53775\n",
            "113/113 [==============================] - 59s 519ms/step - loss: 0.1877 - binary_accuracy: 0.5172 - val_loss: 0.1870 - val_binary_accuracy: 0.5333\n",
            "Epoch 26/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1873 - binary_accuracy: 0.5138\n",
            "Epoch 00026: val_binary_accuracy did not improve from 0.53775\n",
            "113/113 [==============================] - 58s 514ms/step - loss: 0.1873 - binary_accuracy: 0.5138 - val_loss: 0.1866 - val_binary_accuracy: 0.5375\n",
            "Epoch 27/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1868 - binary_accuracy: 0.5219\n",
            "Epoch 00027: val_binary_accuracy did not improve from 0.53775\n",
            "113/113 [==============================] - 58s 512ms/step - loss: 0.1868 - binary_accuracy: 0.5219 - val_loss: 0.1861 - val_binary_accuracy: 0.5378\n",
            "Epoch 28/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1863 - binary_accuracy: 0.5269\n",
            "Epoch 00028: val_binary_accuracy improved from 0.53775 to 0.54389, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 81s 715ms/step - loss: 0.1863 - binary_accuracy: 0.5269 - val_loss: 0.1856 - val_binary_accuracy: 0.5439\n",
            "Epoch 29/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1861 - binary_accuracy: 0.5209\n",
            "Epoch 00029: val_binary_accuracy improved from 0.54389 to 0.54431, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 82s 730ms/step - loss: 0.1861 - binary_accuracy: 0.5209 - val_loss: 0.1851 - val_binary_accuracy: 0.5443\n",
            "Epoch 30/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1851 - binary_accuracy: 0.5294\n",
            "Epoch 00030: val_binary_accuracy improved from 0.54431 to 0.54529, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 81s 720ms/step - loss: 0.1851 - binary_accuracy: 0.5294 - val_loss: 0.1844 - val_binary_accuracy: 0.5453\n",
            "Epoch 31/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1849 - binary_accuracy: 0.5339\n",
            "Epoch 00031: val_binary_accuracy did not improve from 0.54529\n",
            "113/113 [==============================] - 59s 520ms/step - loss: 0.1849 - binary_accuracy: 0.5339 - val_loss: 0.1841 - val_binary_accuracy: 0.5398\n",
            "Epoch 32/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1842 - binary_accuracy: 0.5435\n",
            "Epoch 00032: val_binary_accuracy improved from 0.54529 to 0.54696, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 87s 769ms/step - loss: 0.1842 - binary_accuracy: 0.5435 - val_loss: 0.1834 - val_binary_accuracy: 0.5470\n",
            "Epoch 33/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1835 - binary_accuracy: 0.5454\n",
            "Epoch 00033: val_binary_accuracy improved from 0.54696 to 0.55646, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 78s 689ms/step - loss: 0.1835 - binary_accuracy: 0.5454 - val_loss: 0.1828 - val_binary_accuracy: 0.5565\n",
            "Epoch 34/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1835 - binary_accuracy: 0.5431\n",
            "Epoch 00034: val_binary_accuracy improved from 0.55646 to 0.56190, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 87s 772ms/step - loss: 0.1835 - binary_accuracy: 0.5431 - val_loss: 0.1820 - val_binary_accuracy: 0.5619\n",
            "Epoch 35/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1820 - binary_accuracy: 0.5564\n",
            "Epoch 00035: val_binary_accuracy improved from 0.56190 to 0.56762, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 79s 701ms/step - loss: 0.1820 - binary_accuracy: 0.5564 - val_loss: 0.1811 - val_binary_accuracy: 0.5676\n",
            "Epoch 36/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1816 - binary_accuracy: 0.5614\n",
            "Epoch 00036: val_binary_accuracy improved from 0.56762 to 0.56916, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 81s 718ms/step - loss: 0.1816 - binary_accuracy: 0.5614 - val_loss: 0.1804 - val_binary_accuracy: 0.5692\n",
            "Epoch 37/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1808 - binary_accuracy: 0.5674\n",
            "Epoch 00037: val_binary_accuracy improved from 0.56916 to 0.57627, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 89s 787ms/step - loss: 0.1808 - binary_accuracy: 0.5674 - val_loss: 0.1795 - val_binary_accuracy: 0.5763\n",
            "Epoch 38/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1798 - binary_accuracy: 0.5741\n",
            "Epoch 00038: val_binary_accuracy improved from 0.57627 to 0.58576, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 85s 752ms/step - loss: 0.1798 - binary_accuracy: 0.5741 - val_loss: 0.1779 - val_binary_accuracy: 0.5858\n",
            "Epoch 39/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1792 - binary_accuracy: 0.5705\n",
            "Epoch 00039: val_binary_accuracy did not improve from 0.58576\n",
            "113/113 [==============================] - 59s 520ms/step - loss: 0.1792 - binary_accuracy: 0.5705 - val_loss: 0.1781 - val_binary_accuracy: 0.5803\n",
            "Epoch 40/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1790 - binary_accuracy: 0.5714\n",
            "Epoch 00040: val_binary_accuracy improved from 0.58576 to 0.59498, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 78s 690ms/step - loss: 0.1790 - binary_accuracy: 0.5714 - val_loss: 0.1768 - val_binary_accuracy: 0.5950\n",
            "Epoch 41/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1778 - binary_accuracy: 0.5820\n",
            "Epoch 00041: val_binary_accuracy improved from 0.59498 to 0.59763, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 84s 742ms/step - loss: 0.1778 - binary_accuracy: 0.5820 - val_loss: 0.1759 - val_binary_accuracy: 0.5976\n",
            "Epoch 42/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1778 - binary_accuracy: 0.5788\n",
            "Epoch 00042: val_binary_accuracy improved from 0.59763 to 0.59874, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 92s 816ms/step - loss: 0.1778 - binary_accuracy: 0.5788 - val_loss: 0.1754 - val_binary_accuracy: 0.5987\n",
            "Epoch 43/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1768 - binary_accuracy: 0.5848\n",
            "Epoch 00043: val_binary_accuracy did not improve from 0.59874\n",
            "113/113 [==============================] - 59s 523ms/step - loss: 0.1768 - binary_accuracy: 0.5848 - val_loss: 0.1749 - val_binary_accuracy: 0.5939\n",
            "Epoch 44/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1754 - binary_accuracy: 0.5948\n",
            "Epoch 00044: val_binary_accuracy did not improve from 0.59874\n",
            "113/113 [==============================] - 59s 518ms/step - loss: 0.1754 - binary_accuracy: 0.5948 - val_loss: 0.1758 - val_binary_accuracy: 0.5844\n",
            "Epoch 45/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1747 - binary_accuracy: 0.5970\n",
            "Epoch 00045: val_binary_accuracy improved from 0.59874 to 0.59972, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 93s 829ms/step - loss: 0.1747 - binary_accuracy: 0.5970 - val_loss: 0.1740 - val_binary_accuracy: 0.5997\n",
            "Epoch 46/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1744 - binary_accuracy: 0.5922\n",
            "Epoch 00046: val_binary_accuracy did not improve from 0.59972\n",
            "113/113 [==============================] - 60s 528ms/step - loss: 0.1744 - binary_accuracy: 0.5922 - val_loss: 0.1745 - val_binary_accuracy: 0.5925\n",
            "Epoch 47/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1732 - binary_accuracy: 0.5981\n",
            "Epoch 00047: val_binary_accuracy did not improve from 0.59972\n",
            "113/113 [==============================] - 59s 522ms/step - loss: 0.1732 - binary_accuracy: 0.5981 - val_loss: 0.1735 - val_binary_accuracy: 0.5973\n",
            "Epoch 48/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1735 - binary_accuracy: 0.5899\n",
            "Epoch 00048: val_binary_accuracy did not improve from 0.59972\n",
            "113/113 [==============================] - 59s 522ms/step - loss: 0.1735 - binary_accuracy: 0.5899 - val_loss: 0.1737 - val_binary_accuracy: 0.5872\n",
            "Epoch 49/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1733 - binary_accuracy: 0.5917\n",
            "Epoch 00049: val_binary_accuracy did not improve from 0.59972\n",
            "113/113 [==============================] - 59s 521ms/step - loss: 0.1733 - binary_accuracy: 0.5917 - val_loss: 0.1721 - val_binary_accuracy: 0.5976\n",
            "Epoch 50/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1722 - binary_accuracy: 0.6015\n",
            "Epoch 00050: val_binary_accuracy did not improve from 0.59972\n",
            "113/113 [==============================] - 59s 518ms/step - loss: 0.1722 - binary_accuracy: 0.6015 - val_loss: 0.1722 - val_binary_accuracy: 0.5946\n",
            "Epoch 51/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1728 - binary_accuracy: 0.5977\n",
            "Epoch 00051: val_binary_accuracy did not improve from 0.59972\n",
            "113/113 [==============================] - 59s 519ms/step - loss: 0.1728 - binary_accuracy: 0.5977 - val_loss: 0.1730 - val_binary_accuracy: 0.5912\n",
            "Epoch 52/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1716 - binary_accuracy: 0.6005\n",
            "Epoch 00052: val_binary_accuracy did not improve from 0.59972\n",
            "113/113 [==============================] - 59s 521ms/step - loss: 0.1716 - binary_accuracy: 0.6005 - val_loss: 0.1714 - val_binary_accuracy: 0.5925\n",
            "Epoch 53/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1712 - binary_accuracy: 0.6056\n",
            "Epoch 00053: val_binary_accuracy improved from 0.59972 to 0.60167, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 78s 690ms/step - loss: 0.1712 - binary_accuracy: 0.6056 - val_loss: 0.1700 - val_binary_accuracy: 0.6017\n",
            "Epoch 54/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1706 - binary_accuracy: 0.6019\n",
            "Epoch 00054: val_binary_accuracy did not improve from 0.60167\n",
            "113/113 [==============================] - 60s 529ms/step - loss: 0.1706 - binary_accuracy: 0.6019 - val_loss: 0.1700 - val_binary_accuracy: 0.6011\n",
            "Epoch 55/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1696 - binary_accuracy: 0.6080\n",
            "Epoch 00055: val_binary_accuracy did not improve from 0.60167\n",
            "113/113 [==============================] - 58s 518ms/step - loss: 0.1696 - binary_accuracy: 0.6080 - val_loss: 0.1702 - val_binary_accuracy: 0.6017\n",
            "Epoch 56/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1705 - binary_accuracy: 0.6045\n",
            "Epoch 00056: val_binary_accuracy improved from 0.60167 to 0.60377, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 92s 813ms/step - loss: 0.1705 - binary_accuracy: 0.6045 - val_loss: 0.1693 - val_binary_accuracy: 0.6038\n",
            "Epoch 57/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1694 - binary_accuracy: 0.6132\n",
            "Epoch 00057: val_binary_accuracy did not improve from 0.60377\n",
            "113/113 [==============================] - 59s 522ms/step - loss: 0.1694 - binary_accuracy: 0.6132 - val_loss: 0.1690 - val_binary_accuracy: 0.6032\n",
            "Epoch 58/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1690 - binary_accuracy: 0.6133\n",
            "Epoch 00058: val_binary_accuracy improved from 0.60377 to 0.60740, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 83s 739ms/step - loss: 0.1690 - binary_accuracy: 0.6133 - val_loss: 0.1688 - val_binary_accuracy: 0.6074\n",
            "Epoch 59/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1695 - binary_accuracy: 0.6022\n",
            "Epoch 00059: val_binary_accuracy did not improve from 0.60740\n",
            "113/113 [==============================] - 59s 522ms/step - loss: 0.1695 - binary_accuracy: 0.6022 - val_loss: 0.1689 - val_binary_accuracy: 0.6029\n",
            "Epoch 60/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1683 - binary_accuracy: 0.6158\n",
            "Epoch 00060: val_binary_accuracy did not improve from 0.60740\n",
            "113/113 [==============================] - 58s 514ms/step - loss: 0.1683 - binary_accuracy: 0.6158 - val_loss: 0.1704 - val_binary_accuracy: 0.6015\n",
            "Epoch 61/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1684 - binary_accuracy: 0.6097\n",
            "Epoch 00061: val_binary_accuracy did not improve from 0.60740\n",
            "113/113 [==============================] - 58s 512ms/step - loss: 0.1684 - binary_accuracy: 0.6097 - val_loss: 0.1688 - val_binary_accuracy: 0.6025\n",
            "Epoch 62/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1685 - binary_accuracy: 0.6082\n",
            "Epoch 00062: val_binary_accuracy did not improve from 0.60740\n",
            "113/113 [==============================] - 58s 511ms/step - loss: 0.1685 - binary_accuracy: 0.6082 - val_loss: 0.1686 - val_binary_accuracy: 0.6029\n",
            "Epoch 63/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1680 - binary_accuracy: 0.6168\n",
            "Epoch 00063: val_binary_accuracy did not improve from 0.60740\n",
            "113/113 [==============================] - 58s 514ms/step - loss: 0.1680 - binary_accuracy: 0.6168 - val_loss: 0.1687 - val_binary_accuracy: 0.5957\n",
            "Epoch 64/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1672 - binary_accuracy: 0.6168\n",
            "Epoch 00064: val_binary_accuracy did not improve from 0.60740\n",
            "113/113 [==============================] - 58s 515ms/step - loss: 0.1672 - binary_accuracy: 0.6168 - val_loss: 0.1683 - val_binary_accuracy: 0.6043\n",
            "Epoch 65/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1676 - binary_accuracy: 0.6136\n",
            "Epoch 00065: val_binary_accuracy did not improve from 0.60740\n",
            "113/113 [==============================] - 58s 514ms/step - loss: 0.1676 - binary_accuracy: 0.6136 - val_loss: 0.1709 - val_binary_accuracy: 0.5954\n",
            "Epoch 66/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1676 - binary_accuracy: 0.6122\n",
            "Epoch 00066: val_binary_accuracy did not improve from 0.60740\n",
            "113/113 [==============================] - 58s 514ms/step - loss: 0.1676 - binary_accuracy: 0.6122 - val_loss: 0.1678 - val_binary_accuracy: 0.6053\n",
            "Epoch 67/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1666 - binary_accuracy: 0.6172\n",
            "Epoch 00067: val_binary_accuracy did not improve from 0.60740\n",
            "113/113 [==============================] - 58s 513ms/step - loss: 0.1666 - binary_accuracy: 0.6172 - val_loss: 0.1674 - val_binary_accuracy: 0.6066\n",
            "Epoch 68/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1670 - binary_accuracy: 0.6150\n",
            "Epoch 00068: val_binary_accuracy improved from 0.60740 to 0.60893, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 91s 805ms/step - loss: 0.1670 - binary_accuracy: 0.6150 - val_loss: 0.1668 - val_binary_accuracy: 0.6089\n",
            "Epoch 69/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1671 - binary_accuracy: 0.6095\n",
            "Epoch 00069: val_binary_accuracy did not improve from 0.60893\n",
            "113/113 [==============================] - 59s 521ms/step - loss: 0.1671 - binary_accuracy: 0.6095 - val_loss: 0.1670 - val_binary_accuracy: 0.6059\n",
            "Epoch 70/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1671 - binary_accuracy: 0.6165\n",
            "Epoch 00070: val_binary_accuracy did not improve from 0.60893\n",
            "113/113 [==============================] - 58s 516ms/step - loss: 0.1671 - binary_accuracy: 0.6165 - val_loss: 0.1666 - val_binary_accuracy: 0.6082\n",
            "Epoch 71/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1665 - binary_accuracy: 0.6269\n",
            "Epoch 00071: val_binary_accuracy improved from 0.60893 to 0.60921, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 91s 808ms/step - loss: 0.1665 - binary_accuracy: 0.6269 - val_loss: 0.1665 - val_binary_accuracy: 0.6092\n",
            "Epoch 72/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1657 - binary_accuracy: 0.6257\n",
            "Epoch 00072: val_binary_accuracy improved from 0.60921 to 0.60935, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 80s 706ms/step - loss: 0.1657 - binary_accuracy: 0.6257 - val_loss: 0.1665 - val_binary_accuracy: 0.6094\n",
            "Epoch 73/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1662 - binary_accuracy: 0.6179\n",
            "Epoch 00073: val_binary_accuracy did not improve from 0.60935\n",
            "113/113 [==============================] - 59s 524ms/step - loss: 0.1662 - binary_accuracy: 0.6179 - val_loss: 0.1671 - val_binary_accuracy: 0.6071\n",
            "Epoch 74/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1657 - binary_accuracy: 0.6258\n",
            "Epoch 00074: val_binary_accuracy did not improve from 0.60935\n",
            "113/113 [==============================] - 58s 515ms/step - loss: 0.1657 - binary_accuracy: 0.6258 - val_loss: 0.1665 - val_binary_accuracy: 0.6091\n",
            "Epoch 75/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1660 - binary_accuracy: 0.6294\n",
            "Epoch 00075: val_binary_accuracy did not improve from 0.60935\n",
            "113/113 [==============================] - 59s 520ms/step - loss: 0.1660 - binary_accuracy: 0.6294 - val_loss: 0.1673 - val_binary_accuracy: 0.6042\n",
            "Epoch 76/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1664 - binary_accuracy: 0.6171\n",
            "Epoch 00076: val_binary_accuracy improved from 0.60935 to 0.61228, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 82s 731ms/step - loss: 0.1664 - binary_accuracy: 0.6171 - val_loss: 0.1668 - val_binary_accuracy: 0.6123\n",
            "Epoch 77/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1654 - binary_accuracy: 0.6193\n",
            "Epoch 00077: val_binary_accuracy improved from 0.61228 to 0.61507, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 92s 821ms/step - loss: 0.1654 - binary_accuracy: 0.6193 - val_loss: 0.1661 - val_binary_accuracy: 0.6151\n",
            "Epoch 78/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1643 - binary_accuracy: 0.6322\n",
            "Epoch 00078: val_binary_accuracy did not improve from 0.61507\n",
            "113/113 [==============================] - 59s 523ms/step - loss: 0.1643 - binary_accuracy: 0.6322 - val_loss: 0.1648 - val_binary_accuracy: 0.6149\n",
            "Epoch 79/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1649 - binary_accuracy: 0.6226\n",
            "Epoch 00079: val_binary_accuracy did not improve from 0.61507\n",
            "113/113 [==============================] - 58s 512ms/step - loss: 0.1649 - binary_accuracy: 0.6226 - val_loss: 0.1668 - val_binary_accuracy: 0.6003\n",
            "Epoch 80/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1655 - binary_accuracy: 0.6268\n",
            "Epoch 00080: val_binary_accuracy did not improve from 0.61507\n",
            "113/113 [==============================] - 58s 515ms/step - loss: 0.1655 - binary_accuracy: 0.6268 - val_loss: 0.1650 - val_binary_accuracy: 0.6148\n",
            "Epoch 81/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1651 - binary_accuracy: 0.6289\n",
            "Epoch 00081: val_binary_accuracy did not improve from 0.61507\n",
            "113/113 [==============================] - 58s 515ms/step - loss: 0.1651 - binary_accuracy: 0.6289 - val_loss: 0.1652 - val_binary_accuracy: 0.6137\n",
            "Epoch 82/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1653 - binary_accuracy: 0.6214\n",
            "Epoch 00082: val_binary_accuracy improved from 0.61507 to 0.61856, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 78s 696ms/step - loss: 0.1653 - binary_accuracy: 0.6214 - val_loss: 0.1643 - val_binary_accuracy: 0.6186\n",
            "Epoch 83/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1653 - binary_accuracy: 0.6207\n",
            "Epoch 00083: val_binary_accuracy did not improve from 0.61856\n",
            "113/113 [==============================] - 59s 522ms/step - loss: 0.1653 - binary_accuracy: 0.6207 - val_loss: 0.1649 - val_binary_accuracy: 0.6130\n",
            "Epoch 84/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1646 - binary_accuracy: 0.6282\n",
            "Epoch 00084: val_binary_accuracy did not improve from 0.61856\n",
            "113/113 [==============================] - 58s 516ms/step - loss: 0.1646 - binary_accuracy: 0.6282 - val_loss: 0.1651 - val_binary_accuracy: 0.6134\n",
            "Epoch 85/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1645 - binary_accuracy: 0.6244\n",
            "Epoch 00085: val_binary_accuracy did not improve from 0.61856\n",
            "113/113 [==============================] - 58s 514ms/step - loss: 0.1645 - binary_accuracy: 0.6244 - val_loss: 0.1656 - val_binary_accuracy: 0.6085\n",
            "Epoch 86/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1642 - binary_accuracy: 0.6329\n",
            "Epoch 00086: val_binary_accuracy improved from 0.61856 to 0.62163, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 78s 694ms/step - loss: 0.1642 - binary_accuracy: 0.6329 - val_loss: 0.1640 - val_binary_accuracy: 0.6216\n",
            "Epoch 87/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1633 - binary_accuracy: 0.6314\n",
            "Epoch 00087: val_binary_accuracy did not improve from 0.62163\n",
            "113/113 [==============================] - 59s 519ms/step - loss: 0.1633 - binary_accuracy: 0.6314 - val_loss: 0.1642 - val_binary_accuracy: 0.6137\n",
            "Epoch 88/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1639 - binary_accuracy: 0.6264\n",
            "Epoch 00088: val_binary_accuracy did not improve from 0.62163\n",
            "113/113 [==============================] - 58s 513ms/step - loss: 0.1639 - binary_accuracy: 0.6264 - val_loss: 0.1659 - val_binary_accuracy: 0.6209\n",
            "Epoch 89/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1635 - binary_accuracy: 0.6317\n",
            "Epoch 00089: val_binary_accuracy did not improve from 0.62163\n",
            "113/113 [==============================] - 58s 516ms/step - loss: 0.1635 - binary_accuracy: 0.6317 - val_loss: 0.1637 - val_binary_accuracy: 0.6191\n",
            "Epoch 90/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1628 - binary_accuracy: 0.6375\n",
            "Epoch 00090: val_binary_accuracy improved from 0.62163 to 0.62177, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 82s 729ms/step - loss: 0.1628 - binary_accuracy: 0.6375 - val_loss: 0.1634 - val_binary_accuracy: 0.6218\n",
            "Epoch 91/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1641 - binary_accuracy: 0.6223\n",
            "Epoch 00091: val_binary_accuracy did not improve from 0.62177\n",
            "113/113 [==============================] - 59s 522ms/step - loss: 0.1641 - binary_accuracy: 0.6223 - val_loss: 0.1634 - val_binary_accuracy: 0.6169\n",
            "Epoch 92/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1633 - binary_accuracy: 0.6392\n",
            "Epoch 00092: val_binary_accuracy did not improve from 0.62177\n",
            "113/113 [==============================] - 58s 517ms/step - loss: 0.1633 - binary_accuracy: 0.6392 - val_loss: 0.1626 - val_binary_accuracy: 0.6218\n",
            "Epoch 93/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1622 - binary_accuracy: 0.6371\n",
            "Epoch 00093: val_binary_accuracy improved from 0.62177 to 0.62359, saving model to ./models/end_to_end_add.h5\n",
            "113/113 [==============================] - 80s 712ms/step - loss: 0.1622 - binary_accuracy: 0.6371 - val_loss: 0.1632 - val_binary_accuracy: 0.6236\n",
            "Epoch 94/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1625 - binary_accuracy: 0.6396\n",
            "Epoch 00094: val_binary_accuracy did not improve from 0.62359\n",
            "113/113 [==============================] - 58s 518ms/step - loss: 0.1625 - binary_accuracy: 0.6396 - val_loss: 0.1637 - val_binary_accuracy: 0.6209\n",
            "Epoch 95/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1630 - binary_accuracy: 0.6331\n",
            "Epoch 00095: val_binary_accuracy did not improve from 0.62359\n",
            "113/113 [==============================] - 58s 514ms/step - loss: 0.1630 - binary_accuracy: 0.6331 - val_loss: 0.1659 - val_binary_accuracy: 0.6197\n",
            "Epoch 96/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1630 - binary_accuracy: 0.6390\n",
            "Epoch 00096: val_binary_accuracy did not improve from 0.62359\n",
            "113/113 [==============================] - 59s 518ms/step - loss: 0.1630 - binary_accuracy: 0.6390 - val_loss: 0.1634 - val_binary_accuracy: 0.6221\n",
            "Epoch 97/100\n",
            "113/113 [==============================] - ETA: 0s - loss: 0.1638 - binary_accuracy: 0.6329\n",
            "Epoch 00097: val_binary_accuracy did not improve from 0.62359\n",
            "113/113 [==============================] - 58s 518ms/step - loss: 0.1638 - binary_accuracy: 0.6329 - val_loss: 0.1646 - val_binary_accuracy: 0.6135\n",
            "Epoch 00097: early stopping\n"
          ]
        }
      ],
      "source": [
        "if FIT_MODEL[1]:\n",
        "  checkpoint[-1]=ModelCheckpoint(filepath = './models/end_to_end_average.h5',\n",
        "                              monitor = 'val_binary_accuracy',\n",
        "                              verbose = 1,\n",
        "                              save_best_only = True)\n",
        "  end_to_end_average.fit(x = (test_claim, test_evidence), y = np.array(test['Label']),\n",
        "                   validation_data = ((val_claim, val_evidence), np.array(val['Label'])),\n",
        "                   batch_size = 64,\n",
        "                   epochs = 100, callbacks = checkpoint)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "SaJldzlN3sX8"
      },
      "outputs": [],
      "source": [
        "if LOAD_MODEL[2]: \n",
        "  end_to_end_concat =  load_model(filepath = './models/end_to_end_concat.h5')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8snOJtjsjMhr",
        "outputId": "5b278acd-afbf-4051-f840-d9407ac00980"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.2219 - binary_accuracy: 0.6890\n",
            "Epoch 00001: val_binary_accuracy improved from -inf to 0.50398, saving model to ./models/end_to_end_concat.h5\n",
            "1903/1903 [==============================] - 939s 491ms/step - loss: 0.2219 - binary_accuracy: 0.6890 - val_loss: 0.2209 - val_binary_accuracy: 0.5040\n",
            "Epoch 2/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1634 - binary_accuracy: 0.7342\n",
            "Epoch 00002: val_binary_accuracy improved from 0.50398 to 0.51975, saving model to ./models/end_to_end_concat.h5\n",
            "1903/1903 [==============================] - 937s 492ms/step - loss: 0.1634 - binary_accuracy: 0.7342 - val_loss: 0.2072 - val_binary_accuracy: 0.5197\n",
            "Epoch 3/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1480 - binary_accuracy: 0.7426\n",
            "Epoch 00003: val_binary_accuracy improved from 0.51975 to 0.59093, saving model to ./models/end_to_end_concat.h5\n",
            "1903/1903 [==============================] - 943s 495ms/step - loss: 0.1480 - binary_accuracy: 0.7426 - val_loss: 0.1992 - val_binary_accuracy: 0.5909\n",
            "Epoch 4/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1391 - binary_accuracy: 0.7591\n",
            "Epoch 00004: val_binary_accuracy improved from 0.59093 to 0.61982, saving model to ./models/end_to_end_concat.h5\n",
            "1903/1903 [==============================] - 943s 496ms/step - loss: 0.1391 - binary_accuracy: 0.7591 - val_loss: 0.1833 - val_binary_accuracy: 0.6198\n",
            "Epoch 5/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1355 - binary_accuracy: 0.7680\n",
            "Epoch 00005: val_binary_accuracy did not improve from 0.61982\n",
            "1903/1903 [==============================] - 915s 481ms/step - loss: 0.1355 - binary_accuracy: 0.7680 - val_loss: 0.1827 - val_binary_accuracy: 0.6039\n",
            "Epoch 6/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1331 - binary_accuracy: 0.7740\n",
            "Epoch 00006: val_binary_accuracy did not improve from 0.61982\n",
            "1903/1903 [==============================] - 914s 480ms/step - loss: 0.1331 - binary_accuracy: 0.7740 - val_loss: 0.1748 - val_binary_accuracy: 0.6135\n",
            "Epoch 7/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1323 - binary_accuracy: 0.7756\n",
            "Epoch 00007: val_binary_accuracy improved from 0.61982 to 0.65806, saving model to ./models/end_to_end_concat.h5\n",
            "1903/1903 [==============================] - 916s 482ms/step - loss: 0.1323 - binary_accuracy: 0.7756 - val_loss: 0.1594 - val_binary_accuracy: 0.6581\n",
            "Epoch 8/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1315 - binary_accuracy: 0.7801\n",
            "Epoch 00008: val_binary_accuracy did not improve from 0.65806\n",
            "1903/1903 [==============================] - 892s 469ms/step - loss: 0.1315 - binary_accuracy: 0.7801 - val_loss: 0.1717 - val_binary_accuracy: 0.6341\n",
            "Epoch 9/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1303 - binary_accuracy: 0.7817\n",
            "Epoch 00009: val_binary_accuracy did not improve from 0.65806\n",
            "1903/1903 [==============================] - 898s 472ms/step - loss: 0.1303 - binary_accuracy: 0.7817 - val_loss: 0.1685 - val_binary_accuracy: 0.6294\n",
            "Epoch 10/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1292 - binary_accuracy: 0.7829\n",
            "Epoch 00010: val_binary_accuracy did not improve from 0.65806\n",
            "1903/1903 [==============================] - 907s 477ms/step - loss: 0.1292 - binary_accuracy: 0.7829 - val_loss: 0.1686 - val_binary_accuracy: 0.6345\n",
            "Epoch 11/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1286 - binary_accuracy: 0.7861\n",
            "Epoch 00011: val_binary_accuracy did not improve from 0.65806\n",
            "1903/1903 [==============================] - 920s 484ms/step - loss: 0.1286 - binary_accuracy: 0.7861 - val_loss: 0.1684 - val_binary_accuracy: 0.6487\n",
            "Epoch 12/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1287 - binary_accuracy: 0.7862\n",
            "Epoch 00012: val_binary_accuracy did not improve from 0.65806\n",
            "1903/1903 [==============================] - 923s 485ms/step - loss: 0.1287 - binary_accuracy: 0.7862 - val_loss: 0.1590 - val_binary_accuracy: 0.6537\n",
            "Epoch 13/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1278 - binary_accuracy: 0.7886\n",
            "Epoch 00013: val_binary_accuracy did not improve from 0.65806\n",
            "1903/1903 [==============================] - 918s 482ms/step - loss: 0.1278 - binary_accuracy: 0.7886 - val_loss: 0.1612 - val_binary_accuracy: 0.6569\n",
            "Epoch 14/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1273 - binary_accuracy: 0.7892\n",
            "Epoch 00014: val_binary_accuracy did not improve from 0.65806\n",
            "1903/1903 [==============================] - 920s 484ms/step - loss: 0.1273 - binary_accuracy: 0.7892 - val_loss: 0.1696 - val_binary_accuracy: 0.6497\n",
            "Epoch 15/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1274 - binary_accuracy: 0.7897\n",
            "Epoch 00015: val_binary_accuracy improved from 0.65806 to 0.68960, saving model to ./models/end_to_end_concat.h5\n",
            "1903/1903 [==============================] - 946s 497ms/step - loss: 0.1274 - binary_accuracy: 0.7897 - val_loss: 0.1502 - val_binary_accuracy: 0.6896\n",
            "Epoch 16/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1271 - binary_accuracy: 0.7901\n",
            "Epoch 00016: val_binary_accuracy did not improve from 0.68960\n",
            "1903/1903 [==============================] - 924s 485ms/step - loss: 0.1271 - binary_accuracy: 0.7901 - val_loss: 0.1675 - val_binary_accuracy: 0.6537\n",
            "Epoch 17/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1266 - binary_accuracy: 0.7917\n",
            "Epoch 00017: val_binary_accuracy did not improve from 0.68960\n",
            "1903/1903 [==============================] - 927s 487ms/step - loss: 0.1266 - binary_accuracy: 0.7917 - val_loss: 0.1624 - val_binary_accuracy: 0.6514\n",
            "Epoch 18/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1265 - binary_accuracy: 0.7916\n",
            "Epoch 00018: val_binary_accuracy did not improve from 0.68960\n",
            "1903/1903 [==============================] - 908s 477ms/step - loss: 0.1265 - binary_accuracy: 0.7916 - val_loss: 0.1679 - val_binary_accuracy: 0.6548\n",
            "Epoch 19/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1262 - binary_accuracy: 0.7928\n",
            "Epoch 00019: val_binary_accuracy did not improve from 0.68960\n",
            "1903/1903 [==============================] - 891s 468ms/step - loss: 0.1262 - binary_accuracy: 0.7928 - val_loss: 0.1603 - val_binary_accuracy: 0.6586\n",
            "Epoch 20/50\n",
            "1903/1903 [==============================] - ETA: 0s - loss: 0.1262 - binary_accuracy: 0.7921\n",
            "Epoch 00020: val_binary_accuracy did not improve from 0.68960\n",
            "1903/1903 [==============================] - 901s 473ms/step - loss: 0.1262 - binary_accuracy: 0.7921 - val_loss: 0.1651 - val_binary_accuracy: 0.6572\n",
            "Epoch 00020: early stopping\n"
          ]
        }
      ],
      "source": [
        "# Train the model.\n",
        "if FIT_MODEL[2]:\n",
        "  checkpoint[-1]=ModelCheckpoint(filepath = './models/end_to_end_concat.h5',\n",
        "                              monitor = 'val_binary_accuracy',\n",
        "                              verbose = 1,\n",
        "                              save_best_only = True)\n",
        "  end_to_end_concat.fit(x = (train_claim, train_evidence), y = np.array(train['Label']),\n",
        "                      validation_data = ((val_claim, val_evidence), np.array(val['Label'])), \n",
        "                      batch_size = 64,\n",
        "                      epochs = 50, callbacks = checkpoint)"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "if LOAD_MODEL[3]:\n",
        "  end_to_end_maximum = load_model(filepath = './models/end_to_end_max.h5')"
      ],
      "metadata": {
        "id": "sxO9qqPHeEDR"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "if FIT_MODEL[3]:\n",
        "  checkpoint[-1] = ModelCheckpoint(filepath = './models/end_to_end_max.h5',\n",
        "                              monitor = 'val_binary_accuracy',\n",
        "                              verbose = 1,\n",
        "                              save_best_only = True)\n",
        "  end_to_end_maximum.fit(x = (train_claim, train_evidence), y = np.array(train['Label']),\n",
        "                       validation_data = ((val_claim, val_evidence), np.array(val['Label'])),\n",
        "                       batch_size = 64,\n",
        "                       epochs = 100, callbacks = checkpoint)"
      ],
      "metadata": {
        "id": "7cdHXpxxeI3N"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JeASOfIQYLCt"
      },
      "source": [
        "## Evaluation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ngtcxu8VgbcB"
      },
      "source": [
        "### A. Classification metrics"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "XxMvV7BKwgvm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "81dd40e9-9f1e-4167-8f78-26f7a529c3c7"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "         0.0       0.35      0.84      0.49      1501\n",
            "         1.0       0.93      0.59      0.72      5688\n",
            "\n",
            "    accuracy                           0.64      7189\n",
            "   macro avg       0.64      0.71      0.61      7189\n",
            "weighted avg       0.81      0.64      0.68      7189\n",
            "\n",
            "The binary accuracy is: 0.6425093893448324.\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.6425093893448324"
            ]
          },
          "metadata": {},
          "execution_count": 25
        }
      ],
      "source": [
        "from sklearn.metrics import accuracy_score, classification_report, f1_score, recall_score, precision_score\n",
        "\n",
        "def evaluate_model_A(model: tf.keras.Model, test_set: tuple, test_labels):\n",
        "    pred = model.predict(test_set)\n",
        "\n",
        "    pred_thr = np.zeros(pred.shape)\n",
        "    pred_thr[pred >= 0.5] = 1\n",
        "    pred_thr[pred < 0.5] = 0\n",
        "\n",
        "    print(classification_report(pred_thr, test_labels))\n",
        "\n",
        "    bin_acc = accuracy_score(pred_thr, test_labels)\n",
        "    print(f'The binary accuracy is: {bin_acc}.')\n",
        "    return bin_acc\n",
        "\n",
        "evaluate_model_A(end_to_end_add, (test_claim, test_evidence), np.array(test['Label']))\n",
        "evaluate_model_A(end_to_end_average, (test_claim, test_evidence), np.array(test['Label']))\n",
        "evaluate_model_A(end_to_end_concat, (test_claim, test_evidence), np.array(test['Label']))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "j0H-X422gRjQ"
      },
      "source": [
        "### B. Majority voting metrics\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "qY8tq9ueYm6B"
      },
      "outputs": [],
      "source": [
        "def build_dataframe_from_model(model: tf.keras.Model, dataframe: pd.DataFrame, claim: np.ndarray, evidence: np.ndarray) -> pd.DataFrame:\n",
        "  # Get predictions from trained model.\n",
        "  predictions = model.predict((claim, evidence))\n",
        "\n",
        "  # Create a new dataframe.\n",
        "  predictions_df = pd.DataFrame()\n",
        "  # The claims are the same.\n",
        "  predictions_df[\"Claim\"] = dataframe[\"Claim\"]\n",
        "\n",
        "  # The predictions are taken from the model and mapped to supports/refutes.\n",
        "  predictions_df[\"Label\"] = np.rint(predictions)\n",
        "  predictions_df[\"Label\"] = predictions_df[\"Label\"].apply(lambda x: \"supports\" if x == 1 else \"refutes\")\n",
        "  \n",
        "  return predictions_df\n",
        "\n",
        "def majority_voting(dataframe: pd.DataFrame) -> np.ndarray:\n",
        "  # Create a new dataframe with results column.\n",
        "  unique_claims = pd.DataFrame(columns = [\"Claim\", \"Result\"])\n",
        "  unique_claims[\"Claim\"] = pd.unique(dataframe[\"Claim\"]) \n",
        "\n",
        "  for claim in unique_claims[\"Claim\"]:\n",
        "    labels = dataframe[dataframe[\"Claim\"] == claim]\n",
        "    supports = len(labels[labels[\"Label\"] == \"supports\"])\n",
        "    refutes = len(labels[labels[\"Label\"] == \"refutes\"])\n",
        "    \n",
        "    # Count the evidences that support or refutes the claim to cast a vote.\n",
        "    if supports > refutes :\n",
        "      unique_claims[\"Result\"][unique_claims[\"Claim\"] == claim] = 1\n",
        "    else :\n",
        "      unique_claims[\"Result\"][unique_claims[\"Claim\"] == claim] = 0\n",
        "\n",
        "  return unique_claims\n",
        "\n",
        "def compute_metrics_majority_vote(true_values, predicted_values):\n",
        "  print(\"Accuracy: \", accuracy_score(true_values, predicted_values))\n",
        "  print(\"Precision: \", precision_score(true_values, predicted_values))\n",
        "  print(\"Recall: \", recall_score(true_values, predicted_values))\n",
        "  print(\"F1 score: \", f1_score(true_values, predicted_values))\n",
        "\n",
        "true_values = majority_voting(df_test_m)[\"Result\"].to_numpy(int)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "bce4fmVLSJez",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 183
        },
        "outputId": "f1119bbf-bf35-4545-e44a-5570c26934d6"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "end_to_end_add\n",
            "Accuracy:  0.6419868791002812\n",
            "Precision:  0.5886201427438541\n",
            "Recall:  0.9318895166352793\n",
            "F1 score:  0.7215066828675577\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'\\nprint(\\'end_to_end_average\\')\\n# Model \\'average\\'.\\npredicted_values = build_dataframe_from_model(end_to_end_average, df_test_m, test_claim, test_evidence)\\npredicted_values = majority_voting(predicted_values)[\"Result\"].to_numpy(int)\\ncompute_metrics_majority_vote(true_values, predicted_values)\\n\\nprint(\\'end_to_end_concat\\')\\n# Model \\'concat\\'.\\npredicted_values = build_dataframe_from_model(end_to_end_concat, df_test_m, test_claim, test_evidence)\\npredicted_values = majority_voting(predicted_values)[\"Result\"].to_numpy(int)\\ncompute_metrics_majority_vote(true_values, predicted_values)\\n'"
            ]
          },
          "metadata": {},
          "execution_count": 28
        }
      ],
      "source": [
        "# Model 'add'.\n",
        "print('end_to_end_add')\n",
        "predicted_values = build_dataframe_from_model(end_to_end_add, df_test_m, test_claim, test_evidence)\n",
        "predicted_values = majority_voting(predicted_values)[\"Result\"].to_numpy(int)\n",
        "compute_metrics_majority_vote(true_values, predicted_values)\n",
        "\n",
        "print('end_to_end_average')\n",
        "# Model 'average'.\n",
        "predicted_values = build_dataframe_from_model(end_to_end_average, df_test_m, test_claim, test_evidence)\n",
        "predicted_values = majority_voting(predicted_values)[\"Result\"].to_numpy(int)\n",
        "compute_metrics_majority_vote(true_values, predicted_values)\n",
        "\n",
        "print('end_to_end_concat')\n",
        "# Model 'concat'.\n",
        "predicted_values = build_dataframe_from_model(end_to_end_concat, df_test_m, test_claim, test_evidence)\n",
        "predicted_values = majority_voting(predicted_values)[\"Result\"].to_numpy(int)\n",
        "compute_metrics_majority_vote(true_values, predicted_values)"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        ""
      ],
      "metadata": {
        "id": "mf23-PuSDklL"
      },
      "execution_count": null,
      "outputs": []
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [],
      "machine_shape": "hm",
      "name": "assignment_two.ipynb",
      "provenance": [],
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}