projekt_NER_Dominik_Zalewski.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "toc_visible": true,
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "gpuClass": "standard",
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/doxenix/INL/blob/main/projekt_NER_Dominik_Zalewski.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Projekt z NLP: rozpoznawanie nazw własnych w języku polskim (NER)\n",
        "autor: Dominik Zalewski\n",
        "\n",
        "indeks: s24216\n",
        "\n",
        "W ninijeszym projekcie zaprezentuje moje rozwiązanie w temacie rozpoznawania nazw własnych w języku polskim. Nasze dane treningowe to NKJP_org. Moje rozwiązanie opieram o wytrenowanie dwóch modeli opartych o sieci neuronowe. Zrobimy to od \"zera\", nie używając wytrenowanych już wcześniej modeli (tzw. \"transfer learing\")\n",
        "\n",
        "Środowiskiem w którym należy uruchomić notatnik jest Google Colab."
      ],
      "metadata": {
        "id": "GzH0KZx-wDQk"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Przygotowanie środowiska w Colab"
      ],
      "metadata": {
        "id": "XhAHuzUjE10u"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Zacznijmy od uruchomienia poniższej komórki. Potrzebujemy dwóch rzeczy:\n",
        "\n",
        "* Zainstalować bibliotekę `stop_words`\n",
        "* Doinstalować model `pl_core_news_sm` do pakietu spacy. Będzie nam to potrzebne w procesie lematyzacji\n",
        "* a także moduł `keras-tuner` skorzystamy z niego przy hypertuningu parametrów\n",
        "\n",
        "I to właśnie zrobimy uruchamiając komórke poniżej:\n"
      ],
      "metadata": {
        "id": "Aetjk_er8d8G"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install stop_words\n",
        "!python -m spacy download pl_core_news_sm"
      ],
      "metadata": {
        "id": "u7WghVnAKeTt"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Kolejnym krokiem jest oczywiście import bibliotek:"
      ],
      "metadata": {
        "id": "N5aoAz2_-JFA"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "c7AnNPN4lAtv",
        "outputId": "cbba3f9d-8e63-423a-cad3-37343f1b685c"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "131072"
            ]
          },
          "metadata": {},
          "execution_count": 2
        }
      ],
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "import csv\n",
        "import warnings\n",
        "\n",
        "from stop_words import get_stop_words\n",
        "import spacy\n",
        "\n",
        "import tensorflow as tf\n",
        "from tensorflow.keras.preprocessing.text import Tokenizer\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "from tensorflow.keras import Model, Input\n",
        "from tensorflow.keras.layers import (\n",
        "    LSTM,\n",
        "    Embedding,\n",
        "    Dense,\n",
        "    TimeDistributed,\n",
        "    SpatialDropout1D,\n",
        "    Bidirectional\n",
        ")\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "from tensorflow.keras.utils import to_categorical\n",
        "\n",
        "from sklearn.metrics import classification_report\n",
        "from sklearn.model_selection import train_test_split\n",
        "\n",
        "from matplotlib import pyplot as plt\n",
        "\n",
        "import ast\n",
        "\n",
        "warnings.filterwarnings(\"ignore\")\n",
        "\n",
        "# increase limit for reading csv data\n",
        "csv.field_size_limit(100000000)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Przygotowanie danych"
      ],
      "metadata": {
        "id": "3G1Ctdi1E9K2"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Zacznijmy od pobrania danych NKJP_org.csv\n",
        "\n",
        "Po pobraniu, wczytamy dane za pomocą funkcji `read_csv_data()` z odpowiednimi parametrami. Wyjątkowo dużo parametrów ze względu na specifikę pliku csv, to nam umożliwi poprawne wczytanie danych.\n",
        "\n"
      ],
      "metadata": {
        "id": "JpmWn4bu-RyQ"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!wget https://raw.githubusercontent.com/doxenix/INL/main/NKJP_org.csv"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "B77ruiSRv-x5",
        "outputId": "c6effb40-7b50-418e-e169-5d16c3295f60"
      },
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "--2022-12-24 12:45:20--  https://raw.githubusercontent.com/doxenix/INL/main/NKJP_org.csv\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.109.133, 185.199.110.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 24288720 (23M) [text/plain]\n",
            "Saving to: ‘NKJP_org.csv’\n",
            "\n",
            "NKJP_org.csv        100%[===================>]  23.16M  --.-KB/s    in 0.1s    \n",
            "\n",
            "2022-12-24 12:45:21 (211 MB/s) - ‘NKJP_org.csv’ saved [24288720/24288720]\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def read_csv_data(path, col_names):\n",
        "    raw_data = pd.read_csv(\n",
        "        path,\n",
        "        sep=\"\\t\",\n",
        "        encoding=\"UTF-8\",\n",
        "        header=None,\n",
        "        names=col_names,\n",
        "        engine=\"python\",\n",
        "        quotechar='\"',\n",
        "        error_bad_lines=False,\n",
        "        skip_blank_lines=False,\n",
        "    )\n",
        "\n",
        "    return raw_data\n",
        "\n",
        "\n",
        "col_names = [\"forma\", \"lemat\", \"tag\", \"etykieta\"]\n",
        "raw_data = read_csv_data(\"/content/NKJP_org.csv\", col_names)\n",
        "raw_data"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 476
        },
        "id": "w3dyc8dAxioY",
        "outputId": "8f0400f0-1d40-4065-c550-72bf0cc7270c"
      },
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Skipping line 476905: '\t' expected after '\"'\n",
            "Skipping line 481586: '\t' expected after '\"'\n",
            "Skipping line 587296: unexpected end of data\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                                                    forma  \\\n",
              "0                                              Zatrzasnął   \n",
              "1                                                   drzwi   \n",
              "2                                                      od   \n",
              "3                                              mieszkania   \n",
              "4                                                       ,   \n",
              "...                                                   ...   \n",
              "587288                                            Szalone   \n",
              "587289                                              życie   \n",
              "587290                                            Rudolfa   \n",
              "587291  \\tinterp\\n\\nExtra\\tadj:sg:nom:n:pos\\n!\\tinterp...   \n",
              "587292                                             mlaska   \n",
              "\n",
              "                      lemat       tag  etykieta  \n",
              "0          praet:sg:m1:perf      None      None  \n",
              "1         subst:pl:acc:n:pt      None      None  \n",
              "2             prep:gen:nwok      None      None  \n",
              "3       subst:sg:gen:n:ncol      None      None  \n",
              "4                    interp      None      None  \n",
              "...                     ...       ...       ...  \n",
              "587288     adj:sg:acc:n:pos      None      None  \n",
              "587289  subst:sg:acc:n:ncol      None      None  \n",
              "587290      subst:sg:gen:m1  persName  forename  \n",
              "587291               interp      None      None  \n",
              "587292    fin:sg:ter:imperf      None      None  \n",
              "\n",
              "[587293 rows x 4 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-dd6bd1e6-e70f-4e2c-8d72-6e7434ff2a85\">\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>forma</th>\n",
              "      <th>lemat</th>\n",
              "      <th>tag</th>\n",
              "      <th>etykieta</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>Zatrzasnął</td>\n",
              "      <td>praet:sg:m1:perf</td>\n",
              "      <td>None</td>\n",
              "      <td>None</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>drzwi</td>\n",
              "      <td>subst:pl:acc:n:pt</td>\n",
              "      <td>None</td>\n",
              "      <td>None</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>od</td>\n",
              "      <td>prep:gen:nwok</td>\n",
              "      <td>None</td>\n",
              "      <td>None</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>mieszkania</td>\n",
              "      <td>subst:sg:gen:n:ncol</td>\n",
              "      <td>None</td>\n",
              "      <td>None</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>,</td>\n",
              "      <td>interp</td>\n",
              "      <td>None</td>\n",
              "      <td>None</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>587288</th>\n",
              "      <td>Szalone</td>\n",
              "      <td>adj:sg:acc:n:pos</td>\n",
              "      <td>None</td>\n",
              "      <td>None</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>587289</th>\n",
              "      <td>życie</td>\n",
              "      <td>subst:sg:acc:n:ncol</td>\n",
              "      <td>None</td>\n",
              "      <td>None</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>587290</th>\n",
              "      <td>Rudolfa</td>\n",
              "      <td>subst:sg:gen:m1</td>\n",
              "      <td>persName</td>\n",
              "      <td>forename</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>587291</th>\n",
              "      <td>\\tinterp\\n\\nExtra\\tadj:sg:nom:n:pos\\n!\\tinterp...</td>\n",
              "      <td>interp</td>\n",
              "      <td>None</td>\n",
              "      <td>None</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>587292</th>\n",
              "      <td>mlaska</td>\n",
              "      <td>fin:sg:ter:imperf</td>\n",
              "      <td>None</td>\n",
              "      <td>None</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>587293 rows × 4 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-dd6bd1e6-e70f-4e2c-8d72-6e7434ff2a85')\"\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-dd6bd1e6-e70f-4e2c-8d72-6e7434ff2a85 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-dd6bd1e6-e70f-4e2c-8d72-6e7434ff2a85');\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",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Tak wyglądaja nasze surowe dane. Przed nami sporo pracy - dane musimy odpowiednio przygotować."
      ],
      "metadata": {
        "id": "t1b7Zox7-7pr"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Zaczniemy od lematyzacji każdego słowa. Wykorzystamy bibliotekę spacy i gotowy model `pl_core_news_sm`. Niestety odaplając funkcję poniżej jesteśmy zmuszeni długo poczekać. Ja oczywiście już to zrobiłem wcześniej, z tej racji można przejść kilka komórek niżej - czyli pominąć proces lematyzacji i obróbki danych. Gotowe dane zapisałem do nowego pliku i poźniej jest pobierany. Aby przejść do tego momentu w Spisie treści kliknąć podroździał `OPCJONALNIE - pobranie obrobionych danych`\n",
        "\n",
        "Aby przygotować dane zostały wykonane następujące kroki:\n",
        "\n",
        "* lematyzacja \n",
        "* usunięcie słów tzw. stop words\n",
        "* usunięcie znaków interpunkcyjnych \n",
        "* stworzenie poprawnej kolumny z tagiem dla każdego słowa za pomocą funkcji `create_clean_tag_column`\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "Ba_LmgXS_HeT"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "nlp = spacy.load(\"pl_core_news_sm\")\n",
        "\n",
        "\n",
        "def lemmatize_word(word):\n",
        "    if not pd.isna(word):\n",
        "        doc = nlp(word)\n",
        "        for token in doc:\n",
        "            lemmatized_token = token.lemma_\n",
        "        return lemmatized_token\n",
        "\n",
        "\n",
        "raw_data[\"forma\"] = raw_data[\"forma\"].map(lemmatize_word)"
      ],
      "metadata": {
        "id": "8qtxRtnbUUno"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def remove_stop_words(df):\n",
        "\n",
        "    stop_words = get_stop_words(\"polish\")\n",
        "    df = df[~df[\"forma\"].isin(stop_words)]\n",
        "\n",
        "    return df\n",
        "\n",
        "\n",
        "raw_data = remove_stop_words(raw_data)"
      ],
      "metadata": {
        "id": "UDxCFoQlyjbu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def remove_interp_chars(df):\n",
        "    df = df.loc[~(df[\"lemat\"] == \"interp\")]\n",
        "\n",
        "    return df\n",
        "\n",
        "\n",
        "def create_clean_tag_column(df):\n",
        "\n",
        "    df.etykieta.fillna(df.tag, inplace=True)\n",
        "    df[[\"tag\", \"etykieta\"]] = df[[\"tag\", \"etykieta\"]].applymap(\n",
        "        lambda x: \"O\" if pd.isnull(x) else x\n",
        "    )\n",
        "\n",
        "    df[[\"tag\", \"etykieta\"]] = df[[\"tag\", \"etykieta\"]].applymap(\n",
        "        lambda x: \"O\" if pd.isnull(x) else x\n",
        "    )\n",
        "\n",
        "    df.drop(\"lemat\", axis=1, inplace=True)\n",
        "    df.drop(\"tag\", axis=1, inplace=True)\n",
        "\n",
        "    return df\n",
        "\n",
        "\n",
        "clean_data = remove_interp_chars(raw_data)\n",
        "clean_data = create_clean_tag_column(clean_data)"
      ],
      "metadata": {
        "id": "3_n5FuEe2hxd"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "clean_data['etykieta'].value_counts()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "btApsEU14eBZ",
        "outputId": "22444673-ae4f-41cc-c0fa-c5e703e62657"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "O             382308\n",
              "surname         6647\n",
              "forename        6613\n",
              "orgName         4960\n",
              "country         3975\n",
              "settlement      3813\n",
              "geogName        1994\n",
              "date            1973\n",
              "persName         571\n",
              "addName          447\n",
              "region           290\n",
              "time             193\n",
              "district         182\n",
              "placeName        140\n",
              "bloc              59\n",
              "Name: etykieta, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "clean_data['etykieta'].value_counts(normalize=True)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "xwkkLmtdBrmB",
        "outputId": "2738caf5-39d4-4667-a407-98e89a5a94f4"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "O             0.923081\n",
              "surname       0.016049\n",
              "forename      0.015967\n",
              "orgName       0.011976\n",
              "country       0.009598\n",
              "settlement    0.009206\n",
              "geogName      0.004815\n",
              "date          0.004764\n",
              "persName      0.001379\n",
              "addName       0.001079\n",
              "region        0.000700\n",
              "time          0.000466\n",
              "district      0.000439\n",
              "placeName     0.000338\n",
              "bloc          0.000142\n",
              "Name: etykieta, dtype: float64"
            ]
          },
          "metadata": {},
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Powyżej możemy wstępnie podejrzeć jak wygląda rozkład naszych tagów. Zdecydowanie przeważa tutaj tag `O` - czyli ten który wskazuje na zwykłe słowo niebędące nazwą własną - to jest aż 92% wszystkich słów w naszym zbiorze.\n",
        "\n",
        "Łącznie mamy 15 unikalnych tagów -  to sporo każdy poza `O` symbolizuje nazwę własną jednak o innym charaketerze. Przykładowo tagi których stosunek do wszystkich wystąpień tagów przekracza 1% to:\n",
        "\n",
        "* `surname` - czyli nazwisko\n",
        "* `forename` - czyli imię\n",
        "* `orgName` - nazwa własna np. nazwa organziacji/firmy"
      ],
      "metadata": {
        "id": "ebMRYuSLBJYY"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Na koniec musimy pogrupować nasze dane tak aby z rozdzielonych słów zbudować wiersze z pełnymi zdaniami. Umożliwi to nam pusta przerwa między nowymi zdaniami. To samo zrobimy dla kolumny z tagami i zamienimy do listy. Sprawdźmy poniżej jak będą wyglądać nasze dane po tym zabiegu."
      ],
      "metadata": {
        "id": "qk1DdMUYDL6g"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def group_sentences(df):\n",
        "\n",
        "    df[\"zdanie\"] = 1\n",
        "    df[\"zdanie\"] = df[\"forma\"].isnull().cumsum()\n",
        "\n",
        "    df = df[df[\"forma\"].notna()]\n",
        "\n",
        "    aggr_sentences = (\n",
        "        df.groupby([\"zdanie\"])[\"forma\"].apply(lambda x: \" \".join(x)).reset_index()\n",
        "    )\n",
        "    return aggr_sentences\n",
        "\n",
        "\n",
        "def group_tags(df):\n",
        "    aggr_tokens = (\n",
        "        df.groupby([\"zdanie\"])[\"etykieta\"].apply(lambda x: \" \".join(x)).reset_index()\n",
        "    )\n",
        "    aggr_tokens[\"etykieta\"] = aggr_tokens[\"etykieta\"].str.split(\" \")\n",
        "\n",
        "    return aggr_tokens\n",
        "\n",
        "\n",
        "def merge_dfs(df_1, df_2, on, how):\n",
        "\n",
        "    sentences_df = df_1.merge(df_2, on=on, how=how)\n",
        "\n",
        "    return sentences_df\n",
        "\n",
        "\n",
        "aggr_sentences = group_sentences(clean_data)\n",
        "aggr_tokens = group_tags(clean_data)\n",
        "sentences_df = merge_dfs(df_1 = aggr_sentences, df_2 = aggr_tokens, on = 'zdanie', how = 'left')\n",
        "\n",
        "sentences_df"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 424
        },
        "id": "vRGgNt6c4gyu",
        "outputId": "796ac1ad-cb83-4eb6-9b95-0cf76e529e22"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "       zdanie                                              forma  \\\n",
              "0           0  Zatrzasnął drzwi mieszkania razy przekręcił kl...   \n",
              "1           1  Bohaterem powieści Paźniewskiego miasto Krzemi...   \n",
              "2           2  Jak za czasów Słowackiego funkcjonuje Liceum i...   \n",
              "3           3  Krzemieniec powieściowy tamtym Krzemieńcem ale...   \n",
              "4           4      Swoją drogę tego miasta autor zaczął z daleka   \n",
              "...       ...                                                ...   \n",
              "38085   38218  Przedstaw podstawie wybranych utworów literack...   \n",
              "38086   38219  Rozważ problem wybranych przykładach literacki...   \n",
              "38087   38220  Wyróżniającą cechą szamana niezależnie kultury...   \n",
              "38088   38221  Ani wiedza ani zrozumienie ani też bierna akce...   \n",
              "38089   38222  Szaman zagłębia w doświadczenia życiowe przy p...   \n",
              "\n",
              "                                                etykieta  \n",
              "0      [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "1                      [O, O, O, surname, O, settlement]  \n",
              "2      [O, O, O, O, surname, O, orgName, O, O, geogName]  \n",
              "3      [O, settlement, O, O, settlement, O, O, O, O, ...  \n",
              "4                            [O, O, O, O, O, O, O, O, O]  \n",
              "...                                                  ...  \n",
              "38085                        [O, O, O, O, O, O, O, O, O]  \n",
              "38086  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "38087                     [O, O, O, O, O, O, O, O, O, O]  \n",
              "38088            [O, O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "38089  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "\n",
              "[38090 rows x 3 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-c99180b1-007b-4170-a08e-6116d21595bd\">\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>zdanie</th>\n",
              "      <th>forma</th>\n",
              "      <th>etykieta</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>Zatrzasnął drzwi mieszkania razy przekręcił kl...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>Bohaterem powieści Paźniewskiego miasto Krzemi...</td>\n",
              "      <td>[O, O, O, surname, O, settlement]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2</td>\n",
              "      <td>Jak za czasów Słowackiego funkcjonuje Liceum i...</td>\n",
              "      <td>[O, O, O, O, surname, O, orgName, O, O, geogName]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3</td>\n",
              "      <td>Krzemieniec powieściowy tamtym Krzemieńcem ale...</td>\n",
              "      <td>[O, settlement, O, O, settlement, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4</td>\n",
              "      <td>Swoją drogę tego miasta autor zaczął z daleka</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38085</th>\n",
              "      <td>38218</td>\n",
              "      <td>Przedstaw podstawie wybranych utworów literack...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38086</th>\n",
              "      <td>38219</td>\n",
              "      <td>Rozważ problem wybranych przykładach literacki...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38087</th>\n",
              "      <td>38220</td>\n",
              "      <td>Wyróżniającą cechą szamana niezależnie kultury...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38088</th>\n",
              "      <td>38221</td>\n",
              "      <td>Ani wiedza ani zrozumienie ani też bierna akce...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38089</th>\n",
              "      <td>38222</td>\n",
              "      <td>Szaman zagłębia w doświadczenia życiowe przy p...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>38090 rows × 3 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-c99180b1-007b-4170-a08e-6116d21595bd')\"\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-c99180b1-007b-4170-a08e-6116d21595bd 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-c99180b1-007b-4170-a08e-6116d21595bd');\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",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## OPCJONALNIE - pobranie obrobionych danych\n",
        "\n",
        "W tym miejscu kontynuacja jeśli nie wykonaliśmy powyższych kroków. Tak naprawdę możemy zacząć od tego miejsca od zera (wcześniej tylko importy bibliotek i instalacja - 1 i 2 komórka)\n",
        "\n",
        "Pobieramy oczyszczone i przygotowane dane, a następnie je wczytujemy. Biblioteka ast umozliwi nam konwersję string do listy."
      ],
      "metadata": {
        "id": "O6gd8COzDwVZ"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!wget https://raw.githubusercontent.com/doxenix/INL/main/clean_data_with_lemma.csv"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "M9Cx3gWn69Ig",
        "outputId": "a961f1a2-40f9-477f-e167-13eb4ff615fb"
      },
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "--2022-12-24 12:45:54--  https://raw.githubusercontent.com/doxenix/INL/main/clean_data_with_lemma.csv\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.109.133, 185.199.110.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 5179828 (4.9M) [text/plain]\n",
            "Saving to: ‘clean_data_with_lemma.csv’\n",
            "\n",
            "clean_data_with_lem 100%[===================>]   4.94M  --.-KB/s    in 0.05s   \n",
            "\n",
            "2022-12-24 12:45:55 (102 MB/s) - ‘clean_data_with_lemma.csv’ saved [5179828/5179828]\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def read_preprocessed_data(path):\n",
        "    df = pd.read_csv(path)\n",
        "    df[\"etykieta\"] = df[\"etykieta\"].apply(lambda x: ast.literal_eval(x))\n",
        "\n",
        "    return df\n",
        "\n",
        "\n",
        "sentences_df = read_preprocessed_data(\"clean_data_with_lemma.csv\")\n",
        "sentences_df"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 424
        },
        "id": "W9CXFUfnlq8a",
        "outputId": "3ac18253-340b-48fa-e1f4-32ed25464dba"
      },
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "       zdanie                                              forma  \\\n",
              "0           0  zatrzasnąć drzwi mieszkanie raz przekręcić klu...   \n",
              "1           1     Bohater powieść paźniewskiy miasto krzemieniec   \n",
              "2           2  jak za czas Słowacki funkcjonować liceum i pły...   \n",
              "3           3  krzemieniec powieściowy tamto Krzemieńcem ale ...   \n",
              "4           4         swój droga to miasto autor zacząć z daleki   \n",
              "...       ...                                                ...   \n",
              "38085   38218  Przedstaw podstawa wybrać utwór literacki czło...   \n",
              "38086   38219  Rozważ problem wybrać przykład literacki wizja...   \n",
              "38087   38220  Wyróżniającą cecha szamać niezależnie kultura ...   \n",
              "38088   38221  ani wiedza ani zrozumienie ani też bierny akce...   \n",
              "38089   38222  szaman zagłębia w doświadczenie życiowy przy p...   \n",
              "\n",
              "                                                etykieta  \n",
              "0      [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "1                         [O, O, surname, O, settlement]  \n",
              "2         [O, O, O, surname, O, orgName, O, O, geogName]  \n",
              "3      [settlement, O, O, settlement, O, O, O, O, O, ...  \n",
              "4                               [O, O, O, O, O, O, O, O]  \n",
              "...                                                  ...  \n",
              "38085                           [O, O, O, O, O, O, O, O]  \n",
              "38086  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "38087                        [O, O, O, O, O, O, O, O, O]  \n",
              "38088               [O, O, O, O, O, O, O, O, O, O, O, O]  \n",
              "38089  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...  \n",
              "\n",
              "[38090 rows x 3 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-224b93fb-0bfe-49ba-867c-fd695f1953c3\">\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>zdanie</th>\n",
              "      <th>forma</th>\n",
              "      <th>etykieta</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>zatrzasnąć drzwi mieszkanie raz przekręcić klu...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>Bohater powieść paźniewskiy miasto krzemieniec</td>\n",
              "      <td>[O, O, surname, O, settlement]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2</td>\n",
              "      <td>jak za czas Słowacki funkcjonować liceum i pły...</td>\n",
              "      <td>[O, O, O, surname, O, orgName, O, O, geogName]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3</td>\n",
              "      <td>krzemieniec powieściowy tamto Krzemieńcem ale ...</td>\n",
              "      <td>[settlement, O, O, settlement, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4</td>\n",
              "      <td>swój droga to miasto autor zacząć z daleki</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38085</th>\n",
              "      <td>38218</td>\n",
              "      <td>Przedstaw podstawa wybrać utwór literacki czło...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38086</th>\n",
              "      <td>38219</td>\n",
              "      <td>Rozważ problem wybrać przykład literacki wizja...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38087</th>\n",
              "      <td>38220</td>\n",
              "      <td>Wyróżniającą cecha szamać niezależnie kultura ...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38088</th>\n",
              "      <td>38221</td>\n",
              "      <td>ani wiedza ani zrozumienie ani też bierny akce...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38089</th>\n",
              "      <td>38222</td>\n",
              "      <td>szaman zagłębia w doświadczenie życiowy przy p...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>38090 rows × 3 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-224b93fb-0bfe-49ba-867c-fd695f1953c3')\"\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-224b93fb-0bfe-49ba-867c-fd695f1953c3 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-224b93fb-0bfe-49ba-867c-fd695f1953c3');\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",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Tak prezentują się nasze gotowe dane. To jednak nie koniec. Zależy nam aby dane były jak najlepszej jakości. Zweryfikujemy teraz poprawność dopasowań słów do tagu w każdym zdaniu czyli każdym wierszu:"
      ],
      "metadata": {
        "id": "kJHMcmiyVTEp"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def check_num_equal_rows(df):\n",
        "\n",
        "    df[\"list_sentence\"] = df[\"forma\"].apply(lambda x: x.strip().split(\" \"))\n",
        "    df[\"len_sentence\"] = df[\"list_sentence\"].apply(lambda x: len(x))\n",
        "    df[\"len_tag\"] = df[\"etykieta\"].apply(lambda x: len(x))\n",
        "    df[\"is_equal\"] = df.apply(\n",
        "        lambda row: True if row[\"len_sentence\"] == row[\"len_tag\"] else False, axis=1\n",
        "    )\n",
        "    print(df[\"is_equal\"].value_counts())\n",
        "\n",
        "check_num_equal_rows(sentences_df)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "xRBa2H9pJNFj",
        "outputId": "9ce7f69b-c8ec-4345-dd00-545ec6883585"
      },
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "True     37436\n",
            "False      654\n",
            "Name: is_equal, dtype: int64\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Niestety w `654` wierszach liczba słów do liczby tagów się nie zgadzała. To spora ilość, jednak nie możemy pozwolić sobie na użycie tych wierszy. To co prawda będzie spory ubytek w naszych danych, ale kierujemy się tutaj przestrogą `garbage in = garbage out`. Spróbujemy to poźniej zrekompensować.\n",
        "\n",
        "Usuwamy wiersze:"
      ],
      "metadata": {
        "id": "G_1PJ0chVtYn"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "sentences_df = sentences_df[sentences_df['is_equal']!=False]\n",
        "sentences_df"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 424
        },
        "id": "Ye3qNsbSJa-z",
        "outputId": "c70b11b9-3aad-4250-e5b8-209e2ab12887"
      },
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "       zdanie                                              forma  \\\n",
              "0           0  zatrzasnąć drzwi mieszkanie raz przekręcić klu...   \n",
              "1           1     Bohater powieść paźniewskiy miasto krzemieniec   \n",
              "2           2  jak za czas Słowacki funkcjonować liceum i pły...   \n",
              "3           3  krzemieniec powieściowy tamto Krzemieńcem ale ...   \n",
              "4           4         swój droga to miasto autor zacząć z daleki   \n",
              "...       ...                                                ...   \n",
              "38085   38218  Przedstaw podstawa wybrać utwór literacki czło...   \n",
              "38086   38219  Rozważ problem wybrać przykład literacki wizja...   \n",
              "38087   38220  Wyróżniającą cecha szamać niezależnie kultura ...   \n",
              "38088   38221  ani wiedza ani zrozumienie ani też bierny akce...   \n",
              "38089   38222  szaman zagłębia w doświadczenie życiowy przy p...   \n",
              "\n",
              "                                                etykieta  \\\n",
              "0      [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "1                         [O, O, surname, O, settlement]   \n",
              "2         [O, O, O, surname, O, orgName, O, O, geogName]   \n",
              "3      [settlement, O, O, settlement, O, O, O, O, O, ...   \n",
              "4                               [O, O, O, O, O, O, O, O]   \n",
              "...                                                  ...   \n",
              "38085                           [O, O, O, O, O, O, O, O]   \n",
              "38086  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "38087                        [O, O, O, O, O, O, O, O, O]   \n",
              "38088               [O, O, O, O, O, O, O, O, O, O, O, O]   \n",
              "38089  [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "\n",
              "                                           list_sentence  len_sentence  \\\n",
              "0      [zatrzasnąć, drzwi, mieszkanie, raz, przekręci...            35   \n",
              "1      [Bohater, powieść, paźniewskiy, miasto, krzemi...             5   \n",
              "2      [jak, za, czas, Słowacki, funkcjonować, liceum...             9   \n",
              "3      [krzemieniec, powieściowy, tamto, Krzemieńcem,...            12   \n",
              "4      [swój, droga, to, miasto, autor, zacząć, z, da...             8   \n",
              "...                                                  ...           ...   \n",
              "38085  [Przedstaw, podstawa, wybrać, utwór, literacki...             8   \n",
              "38086  [Rozważ, problem, wybrać, przykład, literacki,...           132   \n",
              "38087  [Wyróżniającą, cecha, szamać, niezależnie, kul...             9   \n",
              "38088  [ani, wiedza, ani, zrozumienie, ani, też, bier...            12   \n",
              "38089  [szaman, zagłębia, w, doświadczenie, życiowy, ...            35   \n",
              "\n",
              "       len_tag  is_equal  \n",
              "0           35      True  \n",
              "1            5      True  \n",
              "2            9      True  \n",
              "3           12      True  \n",
              "4            8      True  \n",
              "...        ...       ...  \n",
              "38085        8      True  \n",
              "38086      132      True  \n",
              "38087        9      True  \n",
              "38088       12      True  \n",
              "38089       35      True  \n",
              "\n",
              "[37436 rows x 7 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-f09a27fb-eb66-4f45-882f-7e9671f62a7c\">\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>zdanie</th>\n",
              "      <th>forma</th>\n",
              "      <th>etykieta</th>\n",
              "      <th>list_sentence</th>\n",
              "      <th>len_sentence</th>\n",
              "      <th>len_tag</th>\n",
              "      <th>is_equal</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>zatrzasnąć drzwi mieszkanie raz przekręcić klu...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>[zatrzasnąć, drzwi, mieszkanie, raz, przekręci...</td>\n",
              "      <td>35</td>\n",
              "      <td>35</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>Bohater powieść paźniewskiy miasto krzemieniec</td>\n",
              "      <td>[O, O, surname, O, settlement]</td>\n",
              "      <td>[Bohater, powieść, paźniewskiy, miasto, krzemi...</td>\n",
              "      <td>5</td>\n",
              "      <td>5</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2</td>\n",
              "      <td>jak za czas Słowacki funkcjonować liceum i pły...</td>\n",
              "      <td>[O, O, O, surname, O, orgName, O, O, geogName]</td>\n",
              "      <td>[jak, za, czas, Słowacki, funkcjonować, liceum...</td>\n",
              "      <td>9</td>\n",
              "      <td>9</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3</td>\n",
              "      <td>krzemieniec powieściowy tamto Krzemieńcem ale ...</td>\n",
              "      <td>[settlement, O, O, settlement, O, O, O, O, O, ...</td>\n",
              "      <td>[krzemieniec, powieściowy, tamto, Krzemieńcem,...</td>\n",
              "      <td>12</td>\n",
              "      <td>12</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4</td>\n",
              "      <td>swój droga to miasto autor zacząć z daleki</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "      <td>[swój, droga, to, miasto, autor, zacząć, z, da...</td>\n",
              "      <td>8</td>\n",
              "      <td>8</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38085</th>\n",
              "      <td>38218</td>\n",
              "      <td>Przedstaw podstawa wybrać utwór literacki czło...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "      <td>[Przedstaw, podstawa, wybrać, utwór, literacki...</td>\n",
              "      <td>8</td>\n",
              "      <td>8</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38086</th>\n",
              "      <td>38219</td>\n",
              "      <td>Rozważ problem wybrać przykład literacki wizja...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>[Rozważ, problem, wybrać, przykład, literacki,...</td>\n",
              "      <td>132</td>\n",
              "      <td>132</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38087</th>\n",
              "      <td>38220</td>\n",
              "      <td>Wyróżniającą cecha szamać niezależnie kultura ...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O]</td>\n",
              "      <td>[Wyróżniającą, cecha, szamać, niezależnie, kul...</td>\n",
              "      <td>9</td>\n",
              "      <td>9</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38088</th>\n",
              "      <td>38221</td>\n",
              "      <td>ani wiedza ani zrozumienie ani też bierny akce...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "      <td>[ani, wiedza, ani, zrozumienie, ani, też, bier...</td>\n",
              "      <td>12</td>\n",
              "      <td>12</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>38089</th>\n",
              "      <td>38222</td>\n",
              "      <td>szaman zagłębia w doświadczenie życiowy przy p...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>[szaman, zagłębia, w, doświadczenie, życiowy, ...</td>\n",
              "      <td>35</td>\n",
              "      <td>35</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>37436 rows × 7 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-f09a27fb-eb66-4f45-882f-7e9671f62a7c')\"\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-f09a27fb-eb66-4f45-882f-7e9671f62a7c 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-f09a27fb-eb66-4f45-882f-7e9671f62a7c');\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",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Przyjrzyjmy się dokładnie naszym zdaniom. W ilu zdaniach dla każego słowa występuje tylko tag `O` a w ilu jest już jakaś nazwa własna? Odpowiedź na to pytanie umożliwi nam funckja "
      ],
      "metadata": {
        "id": "QoyMvKRsYolN"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "ner_tags = [\n",
        "    \"surname\",\n",
        "    \"settlement\",\n",
        "    \"orgName\",\n",
        "    \"geogName\",\n",
        "    \"addName\",\n",
        "    \"forename\",\n",
        "    \"persName\",\n",
        "    \"country\",\n",
        "    \"placeName\",\n",
        "    \"region\",\n",
        "    \"district\",\n",
        "    \"bloc\",\n",
        "    \"date\",\n",
        "    \"time\",\n",
        "]\n",
        "\n",
        "\n",
        "def check_data_diff(df, ner_tags):\n",
        "\n",
        "    idxs_ner = []\n",
        "    idxs_o = []\n",
        "    for idx, tags in df[\"etykieta\"].items():\n",
        "        if any(ner_tag in tags for ner_tag in ner_tags):\n",
        "            idxs_ner.append(idx)\n",
        "        else:\n",
        "            idxs_o.append(idx)\n",
        "\n",
        "    l_idxs_ner = len(idxs_ner)\n",
        "    l_idxs_o = len(idxs_o)\n",
        "\n",
        "    diff = l_idxs_o - l_idxs_ner\n",
        "\n",
        "    print(f\"Number of sentences with name entities {l_idxs_ner}\")\n",
        "    print(f\"Number of sentences without name entities {l_idxs_o}\")\n",
        "    print(f\"Difference is: {diff}\")\n",
        "\n",
        "\n",
        "check_data_diff(sentences_df, ner_tags)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "P2bHCxW6l2Pa",
        "outputId": "7fa98ef4-1c8a-4020-f972-0ae428994ec9"
      },
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of sentences with name entities 14659\n",
            "Number of sentences without name entities 22777\n",
            "Difference is: 8118\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Różnica jest spora, mamy tutaj znacznie więcej zdań w kórych nie ma nazw własnych. Spróbjemy to zrekompensować tak jak było wspomniane wcześniej i do naszego zbioru dodamy teraz zdania zawierające nazwy własne. To dobra praktyka - zależy nam na wykrywaniu dokładnie tych przypadków. Nie możemy też tutaj przesadzać dlatego pozwolimy na dodanie `4000` zduplikowanych wierszy (czyli około połowy rożnicy) "
      ],
      "metadata": {
        "id": "ClDNNJsuao4L"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def append_duplicate_ner_sentences(df, num, ner_tags):\n",
        "\n",
        "    idxs = []\n",
        "    for idx, tags in sentences_df[\"etykieta\"].items():\n",
        "        if any(ner_tag in tags for ner_tag in ner_tags):\n",
        "            idxs.append(idx)\n",
        "\n",
        "    df_to_append = df.iloc[idxs[:num]]\n",
        "    df = df.append(df_to_append)\n",
        "\n",
        "    return df\n",
        "\n",
        "\n",
        "sentences_df = append_duplicate_ner_sentences(sentences_df, 4000, ner_tags)\n",
        "sentences_df"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 424
        },
        "id": "Q6FL0FH6oZAP",
        "outputId": "1b5fb676-56b5-4605-f83f-79c244ecb197"
      },
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "       zdanie                                              forma  \\\n",
              "0           0  zatrzasnąć drzwi mieszkanie raz przekręcić klu...   \n",
              "1           1     Bohater powieść paźniewskiy miasto krzemieniec   \n",
              "2           2  jak za czas Słowacki funkcjonować liceum i pły...   \n",
              "3           3  krzemieniec powieściowy tamto Krzemieńcem ale ...   \n",
              "4           4         swój droga to miasto autor zacząć z daleki   \n",
              "...       ...                                                ...   \n",
              "10283   10311  po dotarciu miejsce praca dwa pogranicznik prz...   \n",
              "10284   10312  w czas obserwacja z tylny część dach wysuwać w...   \n",
              "10285   10313     na szczyć zamontować głowica optoelektroniczny   \n",
              "10295   10323  samochód pijany kierowca trafić opieka współma...   \n",
              "10296   10324  w to wypadek dowiedzieć śmy w prokuratura mąż ...   \n",
              "\n",
              "                                                etykieta  \\\n",
              "0      [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...   \n",
              "1                         [O, O, surname, O, settlement]   \n",
              "2         [O, O, O, surname, O, orgName, O, O, geogName]   \n",
              "3      [settlement, O, O, settlement, O, O, O, O, O, ...   \n",
              "4                               [O, O, O, O, O, O, O, O]   \n",
              "...                                                  ...   \n",
              "10283               [O, O, O, O, O, O, O, O, O, O, O, O]   \n",
              "10284      [O, O, O, O, O, O, O, O, O, O, O, O, O, O, O]   \n",
              "10285                                    [O, O, O, O, O]   \n",
              "10295                                 [O, O, O, O, O, O]   \n",
              "10296            [O, O, O, O, O, O, O, O, O, O, O, O, O]   \n",
              "\n",
              "                                           list_sentence  len_sentence  \\\n",
              "0      [zatrzasnąć, drzwi, mieszkanie, raz, przekręci...            35   \n",
              "1      [Bohater, powieść, paźniewskiy, miasto, krzemi...             5   \n",
              "2      [jak, za, czas, Słowacki, funkcjonować, liceum...             9   \n",
              "3      [krzemieniec, powieściowy, tamto, Krzemieńcem,...            12   \n",
              "4      [swój, droga, to, miasto, autor, zacząć, z, da...             8   \n",
              "...                                                  ...           ...   \n",
              "10283  [po, dotarciu, miejsce, praca, dwa, pograniczn...            12   \n",
              "10284  [w, czas, obserwacja, z, tylny, część, dach, w...            15   \n",
              "10285  [na, szczyć, zamontować, głowica, optoelektron...             5   \n",
              "10295  [samochód, pijany, kierowca, trafić, opieka, w...             6   \n",
              "10296  [w, to, wypadek, dowiedzieć, śmy, w, prokuratu...            13   \n",
              "\n",
              "       len_tag  is_equal  \n",
              "0           35      True  \n",
              "1            5      True  \n",
              "2            9      True  \n",
              "3           12      True  \n",
              "4            8      True  \n",
              "...        ...       ...  \n",
              "10283       12      True  \n",
              "10284       15      True  \n",
              "10285        5      True  \n",
              "10295        6      True  \n",
              "10296       13      True  \n",
              "\n",
              "[41436 rows x 7 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-f707c619-1a3b-4f52-9b2a-b08441a12487\">\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>zdanie</th>\n",
              "      <th>forma</th>\n",
              "      <th>etykieta</th>\n",
              "      <th>list_sentence</th>\n",
              "      <th>len_sentence</th>\n",
              "      <th>len_tag</th>\n",
              "      <th>is_equal</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>zatrzasnąć drzwi mieszkanie raz przekręcić klu...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O, ...</td>\n",
              "      <td>[zatrzasnąć, drzwi, mieszkanie, raz, przekręci...</td>\n",
              "      <td>35</td>\n",
              "      <td>35</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>Bohater powieść paźniewskiy miasto krzemieniec</td>\n",
              "      <td>[O, O, surname, O, settlement]</td>\n",
              "      <td>[Bohater, powieść, paźniewskiy, miasto, krzemi...</td>\n",
              "      <td>5</td>\n",
              "      <td>5</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2</td>\n",
              "      <td>jak za czas Słowacki funkcjonować liceum i pły...</td>\n",
              "      <td>[O, O, O, surname, O, orgName, O, O, geogName]</td>\n",
              "      <td>[jak, za, czas, Słowacki, funkcjonować, liceum...</td>\n",
              "      <td>9</td>\n",
              "      <td>9</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3</td>\n",
              "      <td>krzemieniec powieściowy tamto Krzemieńcem ale ...</td>\n",
              "      <td>[settlement, O, O, settlement, O, O, O, O, O, ...</td>\n",
              "      <td>[krzemieniec, powieściowy, tamto, Krzemieńcem,...</td>\n",
              "      <td>12</td>\n",
              "      <td>12</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4</td>\n",
              "      <td>swój droga to miasto autor zacząć z daleki</td>\n",
              "      <td>[O, O, O, O, O, O, O, O]</td>\n",
              "      <td>[swój, droga, to, miasto, autor, zacząć, z, da...</td>\n",
              "      <td>8</td>\n",
              "      <td>8</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>10283</th>\n",
              "      <td>10311</td>\n",
              "      <td>po dotarciu miejsce praca dwa pogranicznik prz...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "      <td>[po, dotarciu, miejsce, praca, dwa, pograniczn...</td>\n",
              "      <td>12</td>\n",
              "      <td>12</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>10284</th>\n",
              "      <td>10312</td>\n",
              "      <td>w czas obserwacja z tylny część dach wysuwać w...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "      <td>[w, czas, obserwacja, z, tylny, część, dach, w...</td>\n",
              "      <td>15</td>\n",
              "      <td>15</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>10285</th>\n",
              "      <td>10313</td>\n",
              "      <td>na szczyć zamontować głowica optoelektroniczny</td>\n",
              "      <td>[O, O, O, O, O]</td>\n",
              "      <td>[na, szczyć, zamontować, głowica, optoelektron...</td>\n",
              "      <td>5</td>\n",
              "      <td>5</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>10295</th>\n",
              "      <td>10323</td>\n",
              "      <td>samochód pijany kierowca trafić opieka współma...</td>\n",
              "      <td>[O, O, O, O, O, O]</td>\n",
              "      <td>[samochód, pijany, kierowca, trafić, opieka, w...</td>\n",
              "      <td>6</td>\n",
              "      <td>6</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>10296</th>\n",
              "      <td>10324</td>\n",
              "      <td>w to wypadek dowiedzieć śmy w prokuratura mąż ...</td>\n",
              "      <td>[O, O, O, O, O, O, O, O, O, O, O, O, O]</td>\n",
              "      <td>[w, to, wypadek, dowiedzieć, śmy, w, prokuratu...</td>\n",
              "      <td>13</td>\n",
              "      <td>13</td>\n",
              "      <td>True</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>41436 rows × 7 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-f707c619-1a3b-4f52-9b2a-b08441a12487')\"\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-f707c619-1a3b-4f52-9b2a-b08441a12487 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-f707c619-1a3b-4f52-9b2a-b08441a12487');\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",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "sentences_df[['len_sentence']].plot.hist(bins=250, xlim=(0,55), title='Lenght of sentences histogram')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 298
        },
        "id": "g2xSv9ygdOpI",
        "outputId": "4dacef81-d6b0-49ba-ccfb-16664f06c860"
      },
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f0eff49a0d0>"
            ]
          },
          "metadata": {},
          "execution_count": 11
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "sentences_df['len_sentence'].nlargest(5)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "J0-vD4gPSk9D",
        "outputId": "a55ed889-9a75-420c-c028-a1c2d6882cc6"
      },
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "26455    305\n",
              "28042    165\n",
              "25102    148\n",
              "30902    135\n",
              "38086    132\n",
              "Name: len_sentence, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Spoglądając na histogram powyżej możemy zauważyć, że większość zdań w naszym zbiorze to zdania stosunkowo krótkie – składające się najczęściej z 7-8 słów. Możemy też zaobserwować, że praktycznie większość zdań mieści się w zakresie do 50-55 słów. Natomiast patrząc na komórkę poniżej wykresu – aktualnie najdłuższym zdaniem jest wyrażenie składające się z 305 słów, później drugie składa się już z 165 a więc prawie dwa razy mniej. \n",
        "\n",
        "To dość ważna dla nas kwestia aby zdecydować jak maksymalnie długie zdanie będzie w naszym zbiorze, ponieważ to określi nam kształt próbki treningowej na wejściu do naszego modelu (czyli zdanie „stokeniozowane” w formie macierzy). Każda próbka, nawet jak będzie krótkim zdaniem będzie musiała mieć taki sam kształt. Większy kształt wydłuży nam czas treningu, a brakujące słowa zostaną potraktowane jako tag „O”. Zależy nam aby bardziej wykrywać tagi z nazwami własnymi, dlatego usuniemy najdłuższe zdanie z naszego zbioru i w ten sposób przyjmiemy maksymalna długość zdania jako `165`.\n"
      ],
      "metadata": {
        "id": "TIqA83YrpMI0"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "sentences_df = sentences_df[sentences_df['len_sentence'] != 305]\n",
        "sentences_df[sentences_df['len_sentence'] == 305]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 49
        },
        "id": "B9prqRNsoqw6",
        "outputId": "bce56a9a-3e1d-41e4-ccf5-70c8536355bb"
      },
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Empty DataFrame\n",
              "Columns: [zdanie, forma, etykieta, list_sentence, len_sentence, len_tag, is_equal]\n",
              "Index: []"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-3e075d81-9192-4a2f-96cf-ac723a6a56de\">\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>zdanie</th>\n",
              "      <th>forma</th>\n",
              "      <th>etykieta</th>\n",
              "      <th>list_sentence</th>\n",
              "      <th>len_sentence</th>\n",
              "      <th>len_tag</th>\n",
              "      <th>is_equal</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-3e075d81-9192-4a2f-96cf-ac723a6a56de')\"\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-3e075d81-9192-4a2f-96cf-ac723a6a56de 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-3e075d81-9192-4a2f-96cf-ac723a6a56de');\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",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "sentences_list = sentences_df['forma'].to_list()\n",
        "tags_list = sentences_df['etykieta'].to_list()"
      ],
      "metadata": {
        "id": "PRLg1uMDqMte"
      },
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#example sentence:\n",
        "print(sentences_list[12])\n",
        "print(tags_list[12])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "hIe9WjBgrdPd",
        "outputId": "6d0333bc-af7c-409c-95c5-f052e1efba84"
      },
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Halina Auderska wszyscy książka kazać swój bohater szukać tożsamość\n",
            "['forename', 'surname', 'O', 'O', 'O', 'O', 'O', 'O', 'O']\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Dokonamy teraz tokenizacji każdego zdania. Umożliwia nam to klasa `Tokenizer` w biblibotece tensorflow."
      ],
      "metadata": {
        "id": "Bqf_To4TsXzn"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "tokeniser= tf.keras.preprocessing.text.Tokenizer(lower=False,filters='')\n",
        "\n",
        "tokeniser.fit_on_texts(sentences_list)\n",
        "\n",
        "encoded_sentence=tokeniser.texts_to_sequences(sentences_list)"
      ],
      "metadata": {
        "id": "Vf42nCj9rwx1"
      },
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "tags = [\n",
        "    \"surname\",\n",
        "    \"settlement\",\n",
        "    \"orgName\",\n",
        "    \"geogName\",\n",
        "    \"addName\",\n",
        "    \"forename\",\n",
        "    \"persName\",\n",
        "    \"country\",\n",
        "    \"placeName\",\n",
        "    \"region\",\n",
        "    \"district\",\n",
        "    \"bloc\",\n",
        "    \"O\",\n",
        "    \"date\",\n",
        "    \"time\",\n",
        "]\n",
        "\n",
        "num_tags = len(tags)\n",
        "\n",
        "\n",
        "tags_map = {tag: i for i, tag in enumerate(tags)}\n",
        "print(\"Number of Tags \", num_tags)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bIbQj6KOsQkU",
        "outputId": "36fad6b9-15a6-4a04-9602-3d09ba4c9633"
      },
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of Tags  15\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "reverse_tag_map={v: k for k, v in tags_map.items()}\n",
        "reverse_tag_map"
      ],
      "metadata": {
        "id": "jR__zMFutYn8",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "452a4621-334a-4940-a860-e697edb00601"
      },
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{0: 'surname',\n",
              " 1: 'settlement',\n",
              " 2: 'orgName',\n",
              " 3: 'geogName',\n",
              " 4: 'addName',\n",
              " 5: 'forename',\n",
              " 6: 'persName',\n",
              " 7: 'country',\n",
              " 8: 'placeName',\n",
              " 9: 'region',\n",
              " 10: 'district',\n",
              " 11: 'bloc',\n",
              " 12: 'O',\n",
              " 13: 'date',\n",
              " 14: 'time'}"
            ]
          },
          "metadata": {},
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "encoded_tags=[[tags_map[w] for w in tag] for tag in tags_list]"
      ],
      "metadata": {
        "id": "HP2BAdWxtj8b"
      },
      "execution_count": 19,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Aby nasz trening był możliwy, muismy zrobić to o czym był już wspomniane - czyli każda nasza próbka danych musi mieć taki sam kształt:"
      ],
      "metadata": {
        "id": "7mC85SyUtHrf"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "max_len=165\n",
        "\n",
        "padded_encoded_sentences=pad_sequences(maxlen=max_len,sequences=encoded_sentence,padding=\"post\",value=0)\n",
        "padded_encoded_tags=pad_sequences(maxlen=max_len,sequences=encoded_tags,padding=\"post\",value=tags_map['O'])"
      ],
      "metadata": {
        "id": "O_elbeqjuMaj"
      },
      "execution_count": 20,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "target = [to_categorical(i,num_classes = num_tags) for i in  padded_encoded_tags]"
      ],
      "metadata": {
        "id": "WtdTRxrjuPvD"
      },
      "execution_count": 21,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Nasz dane są już przygotowane, możemy je teraz podzielić. Zgodnie z założeniem w zdaniu dzielimy je odpowiednio na ⁉\n",
        "\n",
        "* dane treningowe `50%`\n",
        "* dane walidacyjne `30%`\n",
        "* dane testowe `20%`"
      ],
      "metadata": {
        "id": "sKvDi8_htWzf"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "train_ratio = 0.5\n",
        "validation_ratio = 0.3\n",
        "test_ratio = 0.2\n",
        "\n",
        "x_train, x_test, y_train, y_test = train_test_split(\n",
        "    padded_encoded_sentences, target, test_size=1 - train_ratio, random_state=42\n",
        ")\n",
        "\n",
        "x_val, x_test, y_val, y_test = train_test_split(\n",
        "    x_test,\n",
        "    y_test,\n",
        "    test_size=test_ratio / (test_ratio + validation_ratio),\n",
        "    random_state=42,\n",
        ")"
      ],
      "metadata": {
        "id": "Jy60vsFaZ5bz"
      },
      "execution_count": 22,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(\"Shape of First Sentence -Train\",x_train[0].shape)\n",
        "print(\"Shape of First Sentence Label  -Train\",y_train[0].shape)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RJj9c4UNucvq",
        "outputId": "f6244fa8-3e1d-4f01-e90a-7bcc08ede803"
      },
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Shape of First Sentence -Train (165,)\n",
            "Shape of First Sentence Label  -Train (165, 15)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Trening\n",
        "\n",
        "Przechodzimy do przygotowania modeli. Wykorzystujemy sieci neuronowe i bibliotekę `Keras`. Inspiracja do modeli składających się z takich warstw została zaczerpnięta z różnych tutoriali. W pierwszym kroku użujemy gotowych ustawień. Następnie spóbujemy zmodyfikować nieco parametry naszej sieci. Zobaczmy jak wygląda nasza bazowa sieć:"
      ],
      "metadata": {
        "id": "j3_SHb5auJmH"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## model"
      ],
      "metadata": {
        "id": "r5uZTUPhBZZ0"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "embedding_dim = 165\n",
        "vocab_size = len(tokeniser.word_index) + 1\n",
        "lstm_units = 165\n",
        "max_len = 165\n",
        "\n",
        "input_word = Input(shape=(max_len,))\n",
        "model = Embedding(\n",
        "    input_dim=vocab_size + 1, output_dim=embedding_dim, input_length=max_len\n",
        ")(input_word)\n",
        "\n",
        "model = LSTM(units=lstm_units, return_sequences=True)(model)\n",
        "out = TimeDistributed(Dense(num_tags, activation=\"softmax\"))(model)\n",
        "model = Model(input_word, out)\n",
        "model.summary()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "OVYlkGJgvP7U",
        "outputId": "627369cc-8ff2-4eea-a9d8-675babecc2e9"
      },
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"model\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " input_1 (InputLayer)        [(None, 165)]             0         \n",
            "                                                                 \n",
            " embedding (Embedding)       (None, 165, 165)          8420940   \n",
            "                                                                 \n",
            " lstm (LSTM)                 (None, 165, 165)          218460    \n",
            "                                                                 \n",
            " time_distributed (TimeDistr  (None, 165, 15)          2490      \n",
            " ibuted)                                                         \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 8,641,890\n",
            "Trainable params: 8,641,890\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "model.compile(optimizer=\"adam\", loss=\"categorical_crossentropy\", metrics=[\"accuracy\"])\n",
        "\n",
        "history = model.fit(\n",
        "    x_train,\n",
        "    np.array(y_train),\n",
        "    validation_data=(x_val, np.array(y_val)),\n",
        "    batch_size=32,\n",
        "    epochs=8,\n",
        ")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "vHNNr1VbveM9",
        "outputId": "72c33b6a-4ee3-442a-c19f-9367a3082929"
      },
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/8\n",
            "648/648 [==============================] - 19s 21ms/step - loss: 0.0630 - accuracy: 0.9935 - val_loss: 0.0205 - val_accuracy: 0.9955\n",
            "Epoch 2/8\n",
            "648/648 [==============================] - 14s 21ms/step - loss: 0.0148 - accuracy: 0.9961 - val_loss: 0.0171 - val_accuracy: 0.9961\n",
            "Epoch 3/8\n",
            "648/648 [==============================] - 13s 20ms/step - loss: 0.0108 - accuracy: 0.9971 - val_loss: 0.0159 - val_accuracy: 0.9966\n",
            "Epoch 4/8\n",
            "648/648 [==============================] - 13s 20ms/step - loss: 0.0079 - accuracy: 0.9980 - val_loss: 0.0161 - val_accuracy: 0.9956\n",
            "Epoch 5/8\n",
            "648/648 [==============================] - 11s 17ms/step - loss: 0.0051 - accuracy: 0.9987 - val_loss: 0.0137 - val_accuracy: 0.9970\n",
            "Epoch 6/8\n",
            "648/648 [==============================] - 11s 17ms/step - loss: 0.0036 - accuracy: 0.9990 - val_loss: 0.0131 - val_accuracy: 0.9972\n",
            "Epoch 7/8\n",
            "648/648 [==============================] - 11s 17ms/step - loss: 0.0027 - accuracy: 0.9992 - val_loss: 0.0129 - val_accuracy: 0.9973\n",
            "Epoch 8/8\n",
            "648/648 [==============================] - 12s 18ms/step - loss: 0.0021 - accuracy: 0.9994 - val_loss: 0.0136 - val_accuracy: 0.9973\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Zakończyliśmy trening w 8 epoce z powodu rosnącej funkcji straty od 7 epoki w zbiorze walidacyjnym. Trening wyglądał poprawnie, accuracy na danych treningowych i walidacyjnych ciągle rosło, funkcja straty malała. \n",
        "\n",
        "Zmodyfikujemy powyższą sieć i sprawdzimy jak sobie poradzi:"
      ],
      "metadata": {
        "id": "RoHYvaoPxVsD"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## model_1"
      ],
      "metadata": {
        "id": "yeDg6hQMBVOe"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "embedding_dim = 300\n",
        "vocab_size = len(tokeniser.word_index) + 1\n",
        "lstm_units = 550\n",
        "max_len = 165\n",
        "\n",
        "input_word = Input(shape=(max_len,))\n",
        "model_1 = Embedding(\n",
        "    input_dim=vocab_size + 1, output_dim=embedding_dim, input_length=max_len\n",
        ")(input_word)\n",
        "\n",
        "model_1 = LSTM(units=550, return_sequences=True, dropout=0.5)(model_1)\n",
        "out = TimeDistributed(Dense(num_tags, activation=\"softmax\"))(model_1)\n",
        "model_1 = Model(input_word, out)\n",
        "model_1.summary()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-nroUdXPukcE",
        "outputId": "e28b680f-086b-4dd6-e1be-60ca308b8c25"
      },
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"model_1\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " input_2 (InputLayer)        [(None, 165)]             0         \n",
            "                                                                 \n",
            " embedding_1 (Embedding)     (None, 165, 300)          15310800  \n",
            "                                                                 \n",
            " lstm_1 (LSTM)               (None, 165, 550)          1872200   \n",
            "                                                                 \n",
            " time_distributed_1 (TimeDis  (None, 165, 15)          8265      \n",
            " tributed)                                                       \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 17,191,265\n",
            "Trainable params: 17,191,265\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "model_1.compile(optimizer = 'adam', loss = 'categorical_crossentropy',metrics = ['accuracy'])\n",
        "\n",
        "history_1 = model_1.fit(x_train,np.array(y_train),validation_data=(x_val,np.array(y_val)),batch_size = 32,epochs = 8)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fbu0BTSguqhL",
        "outputId": "d2e8dfa0-74a9-43c9-a2e5-d66706c118fd"
      },
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/8\n",
            "648/648 [==============================] - 34s 49ms/step - loss: 0.0551 - accuracy: 0.9932 - val_loss: 0.0225 - val_accuracy: 0.9953\n",
            "Epoch 2/8\n",
            "648/648 [==============================] - 31s 48ms/step - loss: 0.0146 - accuracy: 0.9963 - val_loss: 0.0155 - val_accuracy: 0.9964\n",
            "Epoch 3/8\n",
            "648/648 [==============================] - 31s 48ms/step - loss: 0.0093 - accuracy: 0.9974 - val_loss: 0.0135 - val_accuracy: 0.9971\n",
            "Epoch 4/8\n",
            "648/648 [==============================] - 31s 48ms/step - loss: 0.0062 - accuracy: 0.9983 - val_loss: 0.0124 - val_accuracy: 0.9973\n",
            "Epoch 5/8\n",
            "648/648 [==============================] - 31s 48ms/step - loss: 0.0042 - accuracy: 0.9989 - val_loss: 0.0122 - val_accuracy: 0.9971\n",
            "Epoch 6/8\n",
            "648/648 [==============================] - 32s 50ms/step - loss: 0.0030 - accuracy: 0.9992 - val_loss: 0.0118 - val_accuracy: 0.9977\n",
            "Epoch 7/8\n",
            "648/648 [==============================] - 31s 48ms/step - loss: 0.0023 - accuracy: 0.9994 - val_loss: 0.0121 - val_accuracy: 0.9977\n",
            "Epoch 8/8\n",
            "648/648 [==============================] - 31s 48ms/step - loss: 0.0018 - accuracy: 0.9995 - val_loss: 0.0124 - val_accuracy: 0.9976\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Zmodyfikowaliśmy warstwę LSTM, a w niej:\n",
        "\n",
        "* zwiększyliśmy liczbę jednostek w parametrze `units` do `550`\n",
        "* dodaliśmy parametr `dropout` ustawiony na `0.5`\n",
        "\n",
        "Na pierwszy rzut oka udało się nam osiągnąć nieco lepsze wyniki, jedank zweryfikujemy to na końcu."
      ],
      "metadata": {
        "id": "VE1EFBMSyCES"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def plot_trenning_results(model_name, model_history):\n",
        "    plt.plot(model_history.history[\"accuracy\"])\n",
        "    plt.plot(model_history.history[\"val_accuracy\"])\n",
        "    plt.title(f\"{model_name} accuracy\")\n",
        "    plt.ylabel(\"accuracy\")\n",
        "    plt.xlabel(\"epoch\")\n",
        "    plt.legend([\"train\", \"val\"], loc=\"upper left\")\n",
        "    plt.show()\n",
        "\n",
        "    plt.plot(model_history.history[\"loss\"])\n",
        "    plt.plot(model_history.history[\"val_loss\"])\n",
        "    plt.title(f\"{model_name} loss\")\n",
        "    plt.ylabel(\"loss\")\n",
        "    plt.xlabel(\"epoch\")\n",
        "    plt.legend([\"train\", \"val\"], loc=\"upper left\")\n",
        "    plt.show()"
      ],
      "metadata": {
        "id": "OmDaIEO4wBRD"
      },
      "execution_count": 33,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "plot_trenning_results('model', history)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 573
        },
        "id": "C1dSFjfhzpLI",
        "outputId": "6d932e93-bb95-43ae-8e96-230419198ea0"
      },
      "execution_count": 34,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plot_trenning_results('model_1', history_1)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 573
        },
        "id": "qbkY2LEF0K9I",
        "outputId": "8ae62233-d58b-412b-c214-5f30afab2dde"
      },
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Powyżej możemy dokładnie zobaczyć jak przebiegał trening obu modeli."
      ],
      "metadata": {
        "id": "Qnb6HLxT0Xtp"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "print('model:')\n",
        "model.evaluate(x_test, np.array(y_test))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qajKtpr7vK9y",
        "outputId": "e9fbed97-c971-45ab-df28-9ac20da6a99b"
      },
      "execution_count": 36,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "model:\n",
            "259/259 [==============================] - 2s 6ms/step - loss: 0.0138 - accuracy: 0.9973\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.013766448013484478, 0.9973046183586121]"
            ]
          },
          "metadata": {},
          "execution_count": 36
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print('model_1:')\n",
        "model_1.evaluate(x_test, np.array(y_test))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DKx774wt0nsI",
        "outputId": "0141e3cd-f03f-49ef-caf1-72edccb7a54d"
      },
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "model_1:\n",
            "259/259 [==============================] - 4s 14ms/step - loss: 0.0127 - accuracy: 0.9975\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.012656353414058685, 0.9975488185882568]"
            ]
          },
          "metadata": {},
          "execution_count": 37
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Po ewaluacji `model_1` osiąganł lepszy wynik. Accuracy jest większe, a funkcja straty mniejsza."
      ],
      "metadata": {
        "id": "aQ8ba6rI0sVw"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "preds=model.predict(x_test)\n",
        "preds_1=model_1.predict(x_test)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "17VoKJrmwG85",
        "outputId": "e57d8052-00a2-4d2c-bd1a-a8a767c56b26"
      },
      "execution_count": 38,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "259/259 [==============================] - 2s 5ms/step\n",
            "259/259 [==============================] - 3s 12ms/step\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Funkcja `evaluate_predictions` pomoże nam zamienić nasze predykcje na zbiory które będziemy mogli wykorzystać do stworzenia raportu z ostatecznymi wynikami."
      ],
      "metadata": {
        "id": "1JsFVIlR1c2I"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def evaluate_predictions(test_data, preds, actual_preds):    \n",
        "    print(\"Shape of Test Data Array\", test_data.shape)\n",
        "    y_actual = np.argmax(np.array(actual_preds), axis=2)\n",
        "    y_pred = np.argmax(preds, axis=2)\n",
        "    num_test_data = test_data.shape[0]\n",
        "    print(\"Number of Test Data Points \", num_test_data)\n",
        "    data = pd.DataFrame()\n",
        "    df_list = []\n",
        "    for i in range(num_test_data):\n",
        "        test_str = list(test_data[i])\n",
        "        df = pd.DataFrame()\n",
        "        df[\"test_tokens\"] = test_str\n",
        "        df[\"tokens\"] = df[\"test_tokens\"].apply(\n",
        "            lambda x: tokeniser.index_word[x] if x != 0 else \"<PAD>\"\n",
        "        )\n",
        "        df[\"actual_target_index\"] = list(y_actual[i])\n",
        "        df[\"pred_target_index\"] = list(y_pred[i])\n",
        "        df[\"actual_target_tag\"] = df[\"actual_target_index\"].apply(\n",
        "            lambda x: reverse_tag_map[x]\n",
        "        )\n",
        "        df[\"pred_target_tag\"] = df[\"pred_target_index\"].apply(\n",
        "            lambda x: reverse_tag_map[x]\n",
        "        )\n",
        "        df[\"id\"] = i + 1\n",
        "        df_list.append(df)\n",
        "    data = pd.concat(df_list)\n",
        "    pred_data = data[data[\"tokens\"] != \"<PAD>\"]\n",
        "    accuracy = (\n",
        "        pred_data[pred_data[\"actual_target_tag\"] == pred_data[\"pred_target_tag\"]].shape[\n",
        "            0\n",
        "        ]\n",
        "        / pred_data.shape[0]\n",
        "    )\n",
        "\n",
        "    return pred_data, accuracy"
      ],
      "metadata": {
        "id": "nkUZigK3vwc6"
      },
      "execution_count": 39,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "pred_data ,accuracy = evaluate_predictions(x_test,preds,y_test)\n",
        "pred_data_1 ,accuracy_1 = evaluate_predictions(x_test,preds_1,y_test)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "brJDh73VL6KD",
        "outputId": "7beb2d7a-4444-4114-9ef7-882f1826b1fb"
      },
      "execution_count": 40,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Shape of Test Data Array (8288, 165)\n",
            "Number of Test Data Points  8288\n",
            "Shape of Test Data Array (8288, 165)\n",
            "Number of Test Data Points  8288\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Stworzymy drugi model\n",
        "\n",
        "Tym razem podejdziemy bardzo kreatywnie i sprawdzimy czy więcej warstw w naszym modelu pomoże w osiągnięciu lepszych wyników.\n",
        "\n",
        "Dołożymy dodatkową warstwę `LSTM`\n",
        "\n",
        "Wiemy już że zwiększenie jednostek `units` a także dodanie `dropout` bardzo pomaga. "
      ],
      "metadata": {
        "id": "LbPY23Ip9HMi"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## model_2"
      ],
      "metadata": {
        "id": "-UYK2eoOBosl"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "embedding_dim = 165\n",
        "vocab_size=len(tokeniser.word_index)+1\n",
        "lstm_units=500\n",
        "max_len=165\n",
        "\n",
        "model_2 = tf.keras.models.Sequential([\n",
        "    tf.keras.layers.Embedding(vocab_size+1, embedding_dim, input_length=max_len),\n",
        "    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(units=lstm_units, activation='tanh', return_sequences=True, dropout=0.25)),    \n",
        "    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(units=lstm_units, activation='tanh', return_sequences=True, dropout=0.25)),\n",
        "    tf.keras.layers.TimeDistributed(tf.keras.layers.Dense(num_tags, activation='softmax'))\n",
        "])\n",
        "\n",
        "model_2.compile(loss='categorical_crossentropy',\n",
        "              optimizer='adam',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "history_2 = model_2.fit(x_train,np.array(y_train),validation_data=(x_val,np.array(y_val)),batch_size = 32,epochs = 4)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PjfwTAayeYFJ",
        "outputId": "efe001ad-4530-491f-9258-a59cd03708e4"
      },
      "execution_count": 62,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/4\n",
            "648/648 [==============================] - 114s 168ms/step - loss: 0.0333 - accuracy: 0.9937 - val_loss: 0.0162 - val_accuracy: 0.9959\n",
            "Epoch 2/4\n",
            "648/648 [==============================] - 107s 166ms/step - loss: 0.0120 - accuracy: 0.9967 - val_loss: 0.0139 - val_accuracy: 0.9965\n",
            "Epoch 3/4\n",
            "648/648 [==============================] - 107s 165ms/step - loss: 0.0082 - accuracy: 0.9976 - val_loss: 0.0130 - val_accuracy: 0.9969\n",
            "Epoch 4/4\n",
            "648/648 [==============================] - 107s 165ms/step - loss: 0.0055 - accuracy: 0.9984 - val_loss: 0.0135 - val_accuracy: 0.9968\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "model_2.evaluate(x_test, np.array(y_test))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "S3dkEWHXg8Ud",
        "outputId": "4ed1b95d-dc9e-482a-a06c-2fbe8832b54a"
      },
      "execution_count": 64,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "259/259 [==============================] - 13s 50ms/step - loss: 0.0136 - accuracy: 0.9968\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.01356450468301773, 0.9968336820602417]"
            ]
          },
          "metadata": {},
          "execution_count": 64
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "model_2.summary()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ly43smaXf8Mv",
        "outputId": "9042f2fb-bee7-4ee8-84ca-c5623b0c408a"
      },
      "execution_count": 65,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential_2\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " embedding_6 (Embedding)     (None, 165, 165)          8420940   \n",
            "                                                                 \n",
            " bidirectional_4 (Bidirectio  (None, 165, 1000)        2664000   \n",
            " nal)                                                            \n",
            "                                                                 \n",
            " bidirectional_5 (Bidirectio  (None, 165, 1000)        6004000   \n",
            " nal)                                                            \n",
            "                                                                 \n",
            " time_distributed_5 (TimeDis  (None, 165, 15)          15015     \n",
            " tributed)                                                       \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 17,103,955\n",
            "Trainable params: 17,103,955\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "preds_2=model_2.predict(x_test)\n",
        "\n",
        "pred_data_2 ,accuracy_2 =evaluate_predictions(x_test,preds_2,y_test)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Uz3IROkHHgEf",
        "outputId": "ba35f97d-4e4a-473d-e159-e8e3cb04c4f9"
      },
      "execution_count": 71,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "259/259 [==============================] - 13s 49ms/step\n",
            "Shape of Test Data Array (8288, 165)\n",
            "Number of Test Data Points  8288\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plot_trenning_results('model_2', history_2)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 573
        },
        "id": "lQzBiFoSCsCL",
        "outputId": "3a3d08b7-f436-4b5d-bc83-cb2273245f38"
      },
      "execution_count": 66,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Trening przebiagł znacznie słabiej niż we wcześnijszych modelach. Musieliśmy go już zakończyć w 4 epoce, ponieważ accuracy na zbiorze walidacyjnym bardzo szybko zaczęło spadać a loss rosnąć. Możemy już stwierdzić, że dołożenie kolejnych warstw nie jest dobrym pomysłem."
      ],
      "metadata": {
        "id": "yhErDPYGHiFH"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Wyniki wszystkich modeli - raport klasyfikacji\n",
        "\n",
        "Na końcu sprawdźmy dokładnie co udało się nam osiągnąć implementując każdy model. Wiemy już jak przebiagał trening każdego z modeli.\n",
        "\n",
        "Po ewaluacji każdy z modeli osiąganł:\n",
        "\n",
        "* `model` : loss `0.0138` - accuracy: `0.9973`\n",
        "* `model_1` : loss `0.0127` - accuracy: `0.9975`\n",
        "* `model_2` : loss `0.0136` - accuracy: `0.9968`\n",
        "\n",
        "Zobaczymy raport klasyfikacji dla każdego modelu:"
      ],
      "metadata": {
        "id": "uEt_-q9dFr95"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "y_pred = pred_data['pred_target_tag'].tolist()\n",
        "y_actual = pred_data['actual_target_tag'].tolist()\n",
        "\n",
        "print('classification_report for: model')\n",
        "print(classification_report(y_actual,y_pred))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6K4rOcR8FN4x",
        "outputId": "e36897af-9405-41f4-a542-8ef1b5d17ffe"
      },
      "execution_count": 68,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "classification_report for: model\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           O       0.97      0.99      0.98     75548\n",
            "     addName       0.53      0.16      0.25        97\n",
            "        bloc       0.00      0.00      0.00        16\n",
            "     country       0.90      0.83      0.86       908\n",
            "        date       0.68      0.51      0.59       402\n",
            "    district       0.58      0.19      0.29        37\n",
            "    forename       0.83      0.75      0.79      1437\n",
            "    geogName       0.57      0.47      0.52       447\n",
            "     orgName       0.53      0.51      0.52      1115\n",
            "    persName       0.76      0.54      0.64       101\n",
            "   placeName       1.00      0.24      0.39        29\n",
            "      region       0.08      0.35      0.12        49\n",
            "  settlement       0.87      0.59      0.70       801\n",
            "     surname       0.84      0.64      0.73      1462\n",
            "        time       1.00      0.05      0.09        44\n",
            "\n",
            "    accuracy                           0.96     82493\n",
            "   macro avg       0.68      0.45      0.50     82493\n",
            "weighted avg       0.95      0.96      0.95     82493\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "y_pred_1 = pred_data_1['pred_target_tag'].tolist()\n",
        "y_actual_1 = pred_data_1['actual_target_tag'].tolist()\n",
        "\n",
        "print('classification_report for: model_1')\n",
        "print(classification_report(y_actual_1,y_pred_1))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5kRBosYqFP8g",
        "outputId": "35a90d34-dcf5-435e-b8dd-6ac59c0c8e6e"
      },
      "execution_count": 69,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "classification_report for: model_1\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           O       0.97      0.99      0.98     75548\n",
            "     addName       0.50      0.21      0.29        97\n",
            "        bloc       0.33      0.12      0.18        16\n",
            "     country       0.94      0.83      0.88       908\n",
            "        date       0.67      0.58      0.62       402\n",
            "    district       0.50      0.16      0.24        37\n",
            "    forename       0.95      0.74      0.84      1437\n",
            "    geogName       0.39      0.56      0.46       447\n",
            "     orgName       0.65      0.49      0.56      1115\n",
            "    persName       0.72      0.63      0.67       101\n",
            "   placeName       0.65      0.52      0.58        29\n",
            "      region       0.50      0.29      0.36        49\n",
            "  settlement       0.92      0.57      0.70       801\n",
            "     surname       0.82      0.68      0.75      1462\n",
            "        time       0.61      0.39      0.47        44\n",
            "\n",
            "    accuracy                           0.96     82493\n",
            "   macro avg       0.68      0.52      0.57     82493\n",
            "weighted avg       0.96      0.96      0.96     82493\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "y_pred_2 = pred_data_2['pred_target_tag'].tolist()\n",
        "y_actual_2 = pred_data_2['actual_target_tag'].tolist()\n",
        "\n",
        "print('classification_report for: model_2')\n",
        "print(classification_report(y_actual_2,y_pred_2))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "aBBw3g1geo7h",
        "outputId": "bbc7acec-fe3c-42df-a339-cbc7d893952b"
      },
      "execution_count": 72,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "classification_report for: model_2\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           O       0.98      0.98      0.98     75548\n",
            "     addName       0.00      0.00      0.00        97\n",
            "        bloc       0.00      0.00      0.00        16\n",
            "     country       0.90      0.84      0.87       908\n",
            "        date       0.22      0.63      0.32       402\n",
            "    district       0.00      0.00      0.00        37\n",
            "    forename       0.92      0.73      0.81      1437\n",
            "    geogName       0.29      0.38      0.33       447\n",
            "     orgName       0.48      0.52      0.50      1115\n",
            "    persName       0.88      0.07      0.13       101\n",
            "   placeName       0.00      0.00      0.00        29\n",
            "      region       0.00      0.00      0.00        49\n",
            "  settlement       0.69      0.55      0.61       801\n",
            "     surname       0.86      0.54      0.66      1462\n",
            "        time       0.00      0.00      0.00        44\n",
            "\n",
            "    accuracy                           0.95     82493\n",
            "   macro avg       0.41      0.35      0.35     82493\n",
            "weighted avg       0.95      0.95      0.95     82493\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Ostatecznie raport też potwierdza, że `model_1` pordzaił sobie lepiej. Z racji 15 klas ciężko go dokładnie porównać, ale najlepszym odnisieniem będzie wynik w miarze `f1` macro avg. \n",
        "\n",
        "Dla każdego z modeli wygląda to następująco:\n",
        "\n",
        " * `model` z wynikiem `0.50`\n",
        " * `model_1` z wynikiem `0.57`\n",
        " * `model_1` z wynikiem `0.35`\n",
        "\n",
        "A więc najlepiej wygląda `model_1`\n",
        "\n",
        "Zobaczymy jeszcze jak poradził sobie `model_1`. Najlepiej poszło mu w predykcji klasy `forname`, najgorzej w klasie `bloc`. Zobaczmy jak wyglądała predykcja w klasie `bloc`."
      ],
      "metadata": {
        "id": "d6xvuuUZIXze"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "pred_data_1[(pred_data_1['actual_target_tag']==\"bloc\")].head(10)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 363
        },
        "id": "_ehGSJUbJNs9",
        "outputId": "587d3984-b6c1-4604-8256-433af8b29c20"
      },
      "execution_count": 73,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "    test_tokens      tokens  actual_target_index  pred_target_index  \\\n",
              "8           546        unia                   11                  2   \n",
              "2           336      Europa                   11                  3   \n",
              "6           414  europejski                   11                  3   \n",
              "11          414  europejski                   11                  3   \n",
              "13         1484        NATO                   11                  2   \n",
              "15          336      Europa                   11                  3   \n",
              "13         6288      Układu                   11                 11   \n",
              "27         6288      Układu                   11                 11   \n",
              "14          546        unia                   11                  2   \n",
              "15          414  europejski                   11                  3   \n",
              "\n",
              "   actual_target_tag pred_target_tag    id  \n",
              "8               bloc         orgName   285  \n",
              "2               bloc        geogName  2097  \n",
              "6               bloc        geogName  3389  \n",
              "11              bloc        geogName  4093  \n",
              "13              bloc         orgName  4093  \n",
              "15              bloc        geogName  4742  \n",
              "13              bloc            bloc  5455  \n",
              "27              bloc            bloc  5455  \n",
              "14              bloc         orgName  5502  \n",
              "15              bloc        geogName  5502  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-d7ecc12f-69e2-4065-b4e2-c4015aae8ee8\">\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>test_tokens</th>\n",
              "      <th>tokens</th>\n",
              "      <th>actual_target_index</th>\n",
              "      <th>pred_target_index</th>\n",
              "      <th>actual_target_tag</th>\n",
              "      <th>pred_target_tag</th>\n",
              "      <th>id</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>8</th>\n",
              "      <td>546</td>\n",
              "      <td>unia</td>\n",
              "      <td>11</td>\n",
              "      <td>2</td>\n",
              "      <td>bloc</td>\n",
              "      <td>orgName</td>\n",
              "      <td>285</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>336</td>\n",
              "      <td>Europa</td>\n",
              "      <td>11</td>\n",
              "      <td>3</td>\n",
              "      <td>bloc</td>\n",
              "      <td>geogName</td>\n",
              "      <td>2097</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>6</th>\n",
              "      <td>414</td>\n",
              "      <td>europejski</td>\n",
              "      <td>11</td>\n",
              "      <td>3</td>\n",
              "      <td>bloc</td>\n",
              "      <td>geogName</td>\n",
              "      <td>3389</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>11</th>\n",
              "      <td>414</td>\n",
              "      <td>europejski</td>\n",
              "      <td>11</td>\n",
              "      <td>3</td>\n",
              "      <td>bloc</td>\n",
              "      <td>geogName</td>\n",
              "      <td>4093</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>13</th>\n",
              "      <td>1484</td>\n",
              "      <td>NATO</td>\n",
              "      <td>11</td>\n",
              "      <td>2</td>\n",
              "      <td>bloc</td>\n",
              "      <td>orgName</td>\n",
              "      <td>4093</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>15</th>\n",
              "      <td>336</td>\n",
              "      <td>Europa</td>\n",
              "      <td>11</td>\n",
              "      <td>3</td>\n",
              "      <td>bloc</td>\n",
              "      <td>geogName</td>\n",
              "      <td>4742</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>13</th>\n",
              "      <td>6288</td>\n",
              "      <td>Układu</td>\n",
              "      <td>11</td>\n",
              "      <td>11</td>\n",
              "      <td>bloc</td>\n",
              "      <td>bloc</td>\n",
              "      <td>5455</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>27</th>\n",
              "      <td>6288</td>\n",
              "      <td>Układu</td>\n",
              "      <td>11</td>\n",
              "      <td>11</td>\n",
              "      <td>bloc</td>\n",
              "      <td>bloc</td>\n",
              "      <td>5455</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>14</th>\n",
              "      <td>546</td>\n",
              "      <td>unia</td>\n",
              "      <td>11</td>\n",
              "      <td>2</td>\n",
              "      <td>bloc</td>\n",
              "      <td>orgName</td>\n",
              "      <td>5502</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>15</th>\n",
              "      <td>414</td>\n",
              "      <td>europejski</td>\n",
              "      <td>11</td>\n",
              "      <td>3</td>\n",
              "      <td>bloc</td>\n",
              "      <td>geogName</td>\n",
              "      <td>5502</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-d7ecc12f-69e2-4065-b4e2-c4015aae8ee8')\"\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-d7ecc12f-69e2-4065-b4e2-c4015aae8ee8 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-d7ecc12f-69e2-4065-b4e2-c4015aae8ee8');\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",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 73
        }
      ]
    }
  ]
}