Part of speech tagger.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Тэггер частей речи дз",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.7.4"
    },
    "toc": {
      "base_numbering": 1,
      "nav_menu": {},
      "number_sections": false,
      "sideBar": true,
      "skip_h1_title": false,
      "title_cell": "Table of Contents",
      "title_sidebar": "Contents",
      "toc_cell": false,
      "toc_position": {},
      "toc_section_display": true,
      "toc_window_display": false
    },
    "varInspector": {
      "cols": {
        "lenName": 16,
        "lenType": 16,
        "lenVar": 40
      },
      "kernels_config": {
        "python": {
          "delete_cmd_postfix": "",
          "delete_cmd_prefix": "del ",
          "library": "var_list.py",
          "varRefreshCmd": "print(var_dic_list())"
        },
        "r": {
          "delete_cmd_postfix": ") ",
          "delete_cmd_prefix": "rm(",
          "library": "var_list.r",
          "varRefreshCmd": "cat(var_dic_list()) "
        }
      },
      "types_to_exclude": [
        "module",
        "function",
        "builtin_function_or_method",
        "instance",
        "_Feature"
      ],
      "window_display": false
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ot3c4fjZwC4T"
      },
      "source": [
        "<img src=\"https://s8.hostingkartinok.com/uploads/images/2018/08/308b49fcfbc619d629fe4604bceb67ac.jpg\" width=500, height=450>\n",
        "<h3 style=\"text-align: center;\"><b>Физтех-Школа Прикладной математики и информатики (ФПМИ) МФТИ</b></h3>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "P2JdzEXmwRU5"
      },
      "source": [
        "---"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oMohh_6CwC4W"
      },
      "source": [
        "### Задача определения частей речи, Part-Of-Speech Tagger (POS)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2Aad2tmBwC4Y"
      },
      "source": [
        "Мы будем решать задачу определения частей речи (POS-теггинга) с помощью скрытой марковской модели (HMM)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gYYV0mdmwC4f",
        "scrolled": false
      },
      "source": [
        "import nltk\n",
        "import pandas as pd\n",
        "import numpy as np\n",
        "from collections import OrderedDict, deque\n",
        "from nltk.corpus import brown\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FPgI52lRwC4n"
      },
      "source": [
        "Вам в помощь http://www.nltk.org/book/"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hxdJxMEAwC4o"
      },
      "source": [
        "Загрузим brown корпус"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZvhXAL_9wC4q",
        "scrolled": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "7b619108-e218-4967-9a4a-b9a914cd378d"
      },
      "source": [
        "nltk.download('brown')"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[nltk_data] Downloading package brown to /root/nltk_data...\n",
            "[nltk_data]   Package brown is already up-to-date!\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "True"
            ]
          },
          "metadata": {},
          "execution_count": 2
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ASm3Dpggs25b"
      },
      "source": [
        "Существует множество наборов грамматических тегов, или тегсетов, например:\n",
        "* НКРЯ\n",
        "* Mystem\n",
        "* UPenn\n",
        "* OpenCorpora (его использует pymorphy2)\n",
        "* Universal Dependencies"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wto8PSC6wC4v"
      },
      "source": [
        "<b>Существует не одна система тегирования, поэтому будьте внимательны, когда прогнозируете тег слов в тексте и вычисляете качество прогноза. Можете получить несправедливо низкое качество вашего решения."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eJ6tuHA_wC4z"
      },
      "source": [
        "На данный момент стандартом является **Universal Dependencies**. Подробнее про проект можно почитать [вот тут](http://universaldependencies.org/), а про теги — [вот тут](http://universaldependencies.org/u/pos/)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Cht7dImWwC42",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c4136b83-f8e4-4063-c7e7-1fea27af9cb9"
      },
      "source": [
        "nltk.download('universal_tagset')"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[nltk_data] Downloading package universal_tagset to /root/nltk_data...\n",
            "[nltk_data]   Package universal_tagset is already up-to-date!\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "True"
            ]
          },
          "metadata": {},
          "execution_count": 3
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IiTimRRywC47"
      },
      "source": [
        "<img src=\"https://4.bp.blogspot.com/-IcFli2wljs0/WrVCw3umY_I/AAAAAAAACYM/UJ_neoUAs3wF95dj2Ouf3BzxXzB_b2TbQCLcBGAs/s1600/postags.png\">\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "iyDBMcBSwC48"
      },
      "source": [
        "Мы имеем массив предложений пар (слово-тег)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BobflewQwC4-",
        "scrolled": false,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "0c63cdc5-cd32-4cff-952d-68b2d260145e"
      },
      "source": [
        "brown_tagged_sents = brown.tagged_sents(tagset=\"universal\")\n",
        "brown_tagged_sents"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[[('The', 'DET'), ('Fulton', 'NOUN'), ('County', 'NOUN'), ('Grand', 'ADJ'), ('Jury', 'NOUN'), ('said', 'VERB'), ('Friday', 'NOUN'), ('an', 'DET'), ('investigation', 'NOUN'), ('of', 'ADP'), (\"Atlanta's\", 'NOUN'), ('recent', 'ADJ'), ('primary', 'NOUN'), ('election', 'NOUN'), ('produced', 'VERB'), ('``', '.'), ('no', 'DET'), ('evidence', 'NOUN'), (\"''\", '.'), ('that', 'ADP'), ('any', 'DET'), ('irregularities', 'NOUN'), ('took', 'VERB'), ('place', 'NOUN'), ('.', '.')], [('The', 'DET'), ('jury', 'NOUN'), ('further', 'ADV'), ('said', 'VERB'), ('in', 'ADP'), ('term-end', 'NOUN'), ('presentments', 'NOUN'), ('that', 'ADP'), ('the', 'DET'), ('City', 'NOUN'), ('Executive', 'ADJ'), ('Committee', 'NOUN'), (',', '.'), ('which', 'DET'), ('had', 'VERB'), ('over-all', 'ADJ'), ('charge', 'NOUN'), ('of', 'ADP'), ('the', 'DET'), ('election', 'NOUN'), (',', '.'), ('``', '.'), ('deserves', 'VERB'), ('the', 'DET'), ('praise', 'NOUN'), ('and', 'CONJ'), ('thanks', 'NOUN'), ('of', 'ADP'), ('the', 'DET'), ('City', 'NOUN'), ('of', 'ADP'), ('Atlanta', 'NOUN'), (\"''\", '.'), ('for', 'ADP'), ('the', 'DET'), ('manner', 'NOUN'), ('in', 'ADP'), ('which', 'DET'), ('the', 'DET'), ('election', 'NOUN'), ('was', 'VERB'), ('conducted', 'VERB'), ('.', '.')], ...]"
            ]
          },
          "metadata": {},
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jSu1KqRrwC5L"
      },
      "source": [
        "Первое предложение"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zCHCZPlkwC5N",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c97fcefa-4c0f-4696-9b97-f97b7363c53b"
      },
      "source": [
        "brown_tagged_sents[0]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('The', 'DET'),\n",
              " ('Fulton', 'NOUN'),\n",
              " ('County', 'NOUN'),\n",
              " ('Grand', 'ADJ'),\n",
              " ('Jury', 'NOUN'),\n",
              " ('said', 'VERB'),\n",
              " ('Friday', 'NOUN'),\n",
              " ('an', 'DET'),\n",
              " ('investigation', 'NOUN'),\n",
              " ('of', 'ADP'),\n",
              " (\"Atlanta's\", 'NOUN'),\n",
              " ('recent', 'ADJ'),\n",
              " ('primary', 'NOUN'),\n",
              " ('election', 'NOUN'),\n",
              " ('produced', 'VERB'),\n",
              " ('``', '.'),\n",
              " ('no', 'DET'),\n",
              " ('evidence', 'NOUN'),\n",
              " (\"''\", '.'),\n",
              " ('that', 'ADP'),\n",
              " ('any', 'DET'),\n",
              " ('irregularities', 'NOUN'),\n",
              " ('took', 'VERB'),\n",
              " ('place', 'NOUN'),\n",
              " ('.', '.')]"
            ]
          },
          "metadata": {},
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SIV2MiRxwC5Q"
      },
      "source": [
        "Все пары (слово-тег)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "y-ADby6LwC5V"
      },
      "source": [
        "Проанализируйте данные, с которыми Вы работаете. Используйте `nltk.FreqDist()` для подсчета частоты встречаемости тега и слова в нашем корпусе. Под частой элемента подразумевается кол-во этого элемента в корпусе."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": false,
        "id": "JzRoXuKFcMZK"
      },
      "source": [
        "# Приведем слова к нижнему регистру\n",
        "brown_tagged_words = list(map(lambda x: (x[0].lower(), x[1]), brown_tagged_words))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4giWaqXjwC5W",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "cf9a4f89-0173-494e-bb7b-eefe679fb392"
      },
      "source": [
        "print('Кол-во предложений: ', len(brown_tagged_sents))\n",
        "tags = [tag for (word, tag) in brown_tagged_words] # наши теги\n",
        "words = [word for (word, tag) in brown_tagged_words] # наши слова\n",
        "\n",
        "tag_num = pd.Series(nltk.FreqDist(tags)).sort_values(ascending=False)# тег - кол-во тега в корпусе\n",
        "word_num = pd.Series(nltk.FreqDist(words)).sort_values(ascending=False) # слово - кол-во слова в корпусе"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Кол-во предложений:  57340\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yfiPpCcLwC5Z",
        "scrolled": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c0d3b667-6b1a-4358-fc40-c6bd3738d4b6"
      },
      "source": [
        "tag_num"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "NOUN    275558\n",
              "VERB    182750\n",
              ".       147565\n",
              "ADP     144766\n",
              "DET     137019\n",
              "ADJ      83721\n",
              "ADV      56239\n",
              "PRON     49334\n",
              "CONJ     38151\n",
              "PRT      29829\n",
              "NUM      14874\n",
              "X         1386\n",
              "dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1_OBArRIZxOV",
        "outputId": "0c2e8f82-92ed-4dfe-daa4-df877db80148"
      },
      "source": [
        "word_num"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "the             69971\n",
              ",               58334\n",
              ".               49346\n",
              "of              36412\n",
              "and             28853\n",
              "                ...  \n",
              "waist-length        1\n",
              "caveat              1\n",
              "patrician           1\n",
              "unfit               1\n",
              "stupefying          1\n",
              "Length: 49815, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8Y1huw7TwC5b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 336
        },
        "outputId": "7f104d6f-9f85-41d4-e635-4b2ace6e6b7d"
      },
      "source": [
        "plt.figure(figsize=(12, 5))\n",
        "plt.bar(tag_num.index, tag_num.values)\n",
        "plt.title(\"Tag_frequency\")\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gBbhnJsmwC5f",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ff61e229-f85e-4e6c-befb-f97b5f0e7e03"
      },
      "source": [
        "word_num[:5]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "the    69971\n",
              ",      58334\n",
              ".      49346\n",
              "of     36412\n",
              "and    28853\n",
              "dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1WmEOBMkwC5i",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 336
        },
        "outputId": "e8ef0341-803f-4abf-c761-2d215982dd9d"
      },
      "source": [
        "plt.figure(figsize=(12, 5))\n",
        "plt.bar(word_num.index[:10], word_num.values[:10])\n",
        "plt.title(\"Word_frequency\")\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "n08z2PjMwC5o"
      },
      "source": [
        "### Вопрос 1:\n",
        "* Кол-во слова `cat` в корпусе?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jhB7di3YwC5p",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "59c54239-a691-4c4f-9979-63af532e2ba7"
      },
      "source": [
        "word_num['cat']"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "23"
            ]
          },
          "metadata": {},
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UsCfVLsewC5s"
      },
      "source": [
        "### Вопрос 2:\n",
        "* Самое популярное слово с самым популярным тегом? <br>(*сначала выбираете слова с самым популярным тегом, а затем выбираете самое популярное слово из уже выбранных*)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oio-XBYkwC5t"
      },
      "source": [
        "# самый популярный тег NOUN. найдем самый популярное существительное\n",
        "noun_words = [word for (word, tag) in brown_tagged_words if tag == 'NOUN']\n",
        "noun_word_num = pd.Series(nltk.FreqDist(noun_words)).sort_values(ascending=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "E_UEUNG8ierf",
        "outputId": "c3a3c458-2d3c-4e0e-c83c-70901f8edaa0"
      },
      "source": [
        "noun_word_num[:10]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "time      1597\n",
              "man       1203\n",
              "af         995\n",
              "years      949\n",
              "way        899\n",
              "people     845\n",
              "mr.        844\n",
              "state      787\n",
              "world      787\n",
              "men        763\n",
              "dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HJ1PJr-Ga1pr"
      },
      "source": [
        "Итого значит ответ на второй вопрос слово \"time\""
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "K-OGc1rSwC5x"
      },
      "source": [
        "Впоследствии обучение моделей может занимать слишком много времени, работайте с подвыборкой, например, только текстами определенных категорий."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Eb7MhxVRwC5y"
      },
      "source": [
        "Категории нашего корпуса:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GSiVcP1TwC51",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8cc5c896-14b1-4df9-ee1c-0d62337ce520"
      },
      "source": [
        "brown.categories()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['adventure',\n",
              " 'belles_lettres',\n",
              " 'editorial',\n",
              " 'fiction',\n",
              " 'government',\n",
              " 'hobbies',\n",
              " 'humor',\n",
              " 'learned',\n",
              " 'lore',\n",
              " 'mystery',\n",
              " 'news',\n",
              " 'religion',\n",
              " 'reviews',\n",
              " 'romance',\n",
              " 'science_fiction']"
            ]
          },
          "metadata": {},
          "execution_count": 17
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MjSlFatJwC53"
      },
      "source": [
        "Будем работать с категорией humor"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_f1rl5x0wC55"
      },
      "source": [
        "Cделайте случайное разбиение выборки на обучение и контроль в отношении 9:1. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GX9t-1qowC58",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1b6e0a98-d7ff-4775-ec80-470c026595c5"
      },
      "source": [
        "brown_tagged_sents = brown.tagged_sents(tagset=\"universal\", categories='humor')\n",
        "# Приведем слова к нижнему регистру\n",
        "my_brown_tagged_sents = []\n",
        "for sent in brown_tagged_sents:\n",
        "    my_brown_tagged_sents.append(list(map(lambda x: (x[0].lower(), x[1]), sent)))\n",
        "my_brown_tagged_sents = np.array(my_brown_tagged_sents)\n",
        "\n",
        "from sklearn.model_selection import train_test_split\n",
        "train_sents, test_sents = train_test_split(my_brown_tagged_sents, random_state=0, test_size=0.1)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:6: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray\n",
            "  \n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pXkVwUjYwC5-",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "761d5d3f-de43-4a7a-fc62-4aefc63c4848"
      },
      "source": [
        "len(train_sents)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "947"
            ]
          },
          "metadata": {},
          "execution_count": 59
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JQMjzJ2YwC6C",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c962b6d5-b855-485b-8e37-16237dddf702"
      },
      "source": [
        "len(test_sents)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "106"
            ]
          },
          "metadata": {},
          "execution_count": 60
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Wdl0op_nUj_m",
        "outputId": "bb3eba3b-2d78-48fb-c0e6-017d9b05466e"
      },
      "source": [
        "test_sents"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([list([('``', '.'), ('imagine', 'VERB'), ('being', 'VERB'), ('able', 'ADJ'), ('to', 'PRT'), ('laugh', 'VERB'), ('and', 'CONJ'), ('wink', 'VERB'), ('when', 'ADV'), (\"you're\", 'PRT'), ('like', 'ADP'), ('the', 'DET'), ('top', 'ADJ'), ('part', 'NOUN'), ('of', 'ADP'), ('that', 'DET'), ('picture', 'NOUN'), ('frame', 'NOUN'), ('at', 'ADP'), ('home', 'NOUN'), (\"''\", '.'), (',', '.'), ('one', 'NUM'), ('of', 'ADP'), ('them', 'PRON'), ('said', 'VERB'), ('.', '.')]),\n",
              "       list([('``', '.'), ('just', 'ADV'), ('you', 'PRON'), ('wait', 'VERB'), (\"''\", '.'), (',', '.'), ('said', 'VERB'), ('arlene', 'NOUN'), ('again', 'ADV'), (',', '.'), ('as', 'ADP'), ('though', 'ADP'), ('she', 'PRON'), ('were', 'VERB'), ('discovering', 'VERB'), ('the', 'DET'), ('pleasantly', 'ADV'), ('tingling', 'VERB'), ('insinuations', 'NOUN'), ('of', 'ADP'), ('that', 'DET'), ('handy', 'ADJ'), ('little', 'ADJ'), ('sturdy', 'ADJ'), ('statement', 'NOUN'), ('.', '.')]),\n",
              "       list([('as', 'ADP'), ('we', 'PRON'), ('find', 'VERB'), ('out', 'PRT'), ('at', 'ADP'), ('the', 'DET'), ('end', 'NOUN'), (',', '.'), ('it', 'PRON'), ('is', 'VERB'), ('not', 'ADV'), ('the', 'DET'), ('stool', 'NOUN'), ('(', '.'), ('symbolizing', 'VERB'), ('doris', 'NOUN'), (',', '.'), ('therefore', 'ADV'), ('the', 'DET'), ('english', 'NOUN'), (')', '.'), ('that', 'ADP'), ('he', 'PRON'), ('is', 'VERB'), ('punishing', 'VERB'), ('but', 'CONJ'), ('the', 'DET'), ('piece', 'NOUN'), ('of', 'ADP'), ('venetian', 'ADJ'), ('blind', 'NOUN'), ('.', '.')]),\n",
              "       list([('as', 'ADP'), ('he', 'PRON'), ('leads', 'VERB'), ('the', 'DET'), ('neurenschatz', 'NOUN'), ('skolkau', 'NOUN'), ('orchestra', 'NOUN'), (',', '.'), ('schlek', 'NOUN'), ('gives', 'VERB'), ('a', 'DET'), ('tremendously', 'ADV'), ('inspired', 'VERB'), ('performance', 'NOUN'), ('of', 'ADP'), ('both', 'DET'), ('the', 'DET'), ('baslot', 'NOUN'), ('and', 'CONJ'), ('rattzhenfuut', 'NOUN'), ('concertos', 'NOUN'), (',', '.'), ('including', 'ADP'), ('the', 'DET'), ('controversial', 'ADJ'), ('tschilwyk', 'NOUN'), ('cadenza', 'NOUN'), (',', '.'), ('which', 'DET'), ('was', 'VERB'), ('included', 'VERB'), ('at', 'ADP'), ('the', 'DET'), (\"conductor's\", 'NOUN'), ('insistence', 'NOUN'), ('.', '.')]),\n",
              "       list([('not', 'ADV'), ('without', 'ADP'), ('good', 'ADJ'), ('reason', 'NOUN'), ('has', 'VERB'), ('the', 'DET'), ('anatomical', 'NOUN'), ('been', 'VERB'), ('called', 'VERB'), ('jocular', 'ADJ'), ('journalese', 'NOUN'), ('.', '.')]),\n",
              "       list([('at', 'ADP'), ('first', 'ADV'), ('barco', 'NOUN'), ('was', 'VERB'), ('evasive', 'ADJ'), ('and', 'CONJ'), ('shifty', 'ADJ'), ('.', '.')]),\n",
              "       list([('welch', 'NOUN'), ('summoned', 'VERB'), ('jail', 'NOUN'), ('officials', 'NOUN'), ('to', 'ADP'), (\"barco's\", 'NOUN'), ('cell', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), (\"who's\", 'PRT'), ('in', 'ADP'), ('it', 'PRON'), (\"''\", '.'), ('?', '.'), ('?', '.')]),\n",
              "       list([('``', '.'), ('the', 'DET'), ('world', 'NOUN'), ('that', 'PRON'), ('was', 'VERB'), ('once', 'ADV'), ('foot-loose', 'ADJ'), ('and', 'CONJ'), ('fancy-free', 'ADJ'), (\"''\", '.'), (',', '.'), ('he', 'PRON'), ('said', 'VERB'), (',', '.'), ('``', '.'), ('has', 'VERB'), ('now', 'ADV'), ('become', 'VERB'), ('screw-loose', 'ADJ'), ('and', 'CONJ'), ('frenzy-free', 'ADJ'), ('.', '.')]),\n",
              "       list([('he', 'PRON'), ('kissed', 'VERB'), ('her', 'PRON'), ('back', 'ADV'), ('.', '.')]),\n",
              "       list([('letch', 'NOUN'), ('had', 'VERB'), ('returned', 'VERB'), ('from', 'ADP'), ('his', 'DET'), ('debacle', 'NOUN'), ('unrepentant', 'ADJ'), ('and', 'CONJ'), ('more', 'ADV'), ('badly', 'ADV'), ('behaved', 'VERB'), ('than', 'ADP'), ('before', 'ADV'), ('.', '.')]),\n",
              "       list([('inscribed', 'VERB'), ('around', 'ADP'), ('its', 'DET'), ('base', 'NOUN'), ('is', 'VERB'), ('a', 'DET'), ('charm', 'NOUN'), ('in', 'ADP'), ('balinese', 'NOUN'), (',', '.'), ('a', 'DET'), ('dialect', 'NOUN'), ('i', 'PRON'), ('take', 'VERB'), ('it', 'PRON'), ('you', 'PRON'), (\"don't\", 'VERB'), ('comprehend', 'VERB'), ('.', '.')]),\n",
              "       list([('she', 'PRON'), ('patronized', 'VERB'), ('greenwich', 'NOUN'), ('village', 'NOUN'), ('artists', 'NOUN'), ('for', 'ADP'), ('awhile', 'ADV'), (',', '.'), ('then', 'ADV'), ('put', 'VERB'), ('some', 'DET'), ('money', 'NOUN'), ('into', 'ADP'), ('a', 'DET'), ('broadway', 'NOUN'), ('show', 'NOUN'), ('which', 'DET'), ('was', 'VERB'), ('successful', 'ADJ'), ('(', '.'), ('terrible', 'ADJ'), (',', '.'), ('but', 'CONJ'), ('successful', 'ADJ'), (')', '.'), ('.', '.')]),\n",
              "       list([('sometimes', 'ADV'), (',', '.'), ('as', 'ADP'), ('at', 'ADP'), ('this', 'DET'), ('juncture', 'NOUN'), (',', '.'), ('he', 'PRON'), ('did', 'VERB'), ('not', 'ADV'), ('.', '.')]),\n",
              "       list([('however', 'ADV'), (',', '.'), ('such', 'PRT'), ('a', 'DET'), ('hotbed', 'NOUN'), ('of', 'ADP'), ('gossip', 'NOUN'), ('had', 'VERB'), ('grown', 'VERB'), ('up', 'PRT'), ('during', 'ADP'), ('the', 'DET'), ('trial', 'NOUN'), (',', '.'), ('that', 'ADP'), ('every', 'DET'), ('precaution', 'NOUN'), ('had', 'VERB'), ('to', 'PRT'), ('be', 'VERB'), ('taken', 'VERB'), ('to', 'PRT'), ('keep', 'VERB'), ('my', 'DET'), ('visit', 'NOUN'), ('from', 'ADP'), ('being', 'VERB'), ('whispered', 'VERB'), ('to', 'ADP'), ('the', 'DET'), ('world', 'NOUN'), (',', '.'), ('society', 'NOUN'), (',', '.'), ('and', 'CONJ'), ('even', 'ADV'), (',', '.'), ('alas', 'PRT'), (',', '.'), ('to', 'ADP'), ('my', 'DET'), ('own', 'ADJ'), ('mother', 'NOUN'), ('.', '.')]),\n",
              "       list([('then', 'ADV'), (',', '.'), ('arlene', 'NOUN'), ('threw', 'VERB'), ('herself', 'PRON'), ('backwards', 'ADV'), ('and', 'CONJ'), ('wiggled', 'VERB'), ('in', 'ADP'), ('a', 'DET'), ('way', 'NOUN'), ('that', 'PRON'), ('was', 'VERB'), ('just', 'ADV'), ('wonderful', 'ADJ'), ('.', '.')]),\n",
              "       list([('she', 'PRON'), ('was', 'VERB'), ('rollickingly', 'ADV'), ('happy', 'ADJ'), ('.', '.')]),\n",
              "       list([('they', 'PRON'), ('wanted', 'VERB'), ('to', 'PRT'), ('touch', 'VERB'), ('the', 'DET'), ('mystery', 'NOUN'), ('.', '.')]),\n",
              "       list([('the', 'DET'), ('next', 'ADJ'), ('time', 'NOUN'), ('we', 'PRON'), ('saw', 'VERB'), ('him', 'PRON'), (',', '.'), ('he', 'PRON'), ('was', 'VERB'), ('a', 'DET'), ('changed', 'VERB'), ('person', 'NOUN'), (';', '.'), (';', '.')]),\n",
              "       list([('the', 'DET'), ('grocery', 'NOUN'), ('the', 'DET'), ('youth', 'NOUN'), ('worked', 'VERB'), ('for', 'ADP'), ('soon', 'ADV'), ('tired', 'VERB'), ('of', 'ADP'), ('his', 'DET'), ('depressing', 'ADJ'), ('effect', 'NOUN'), ('on', 'ADP'), ('customers', 'NOUN'), (',', '.'), ('most', 'ADJ'), ('of', 'ADP'), ('whom', 'PRON'), ('were', 'VERB'), ('sufficiently', 'ADV'), ('neurotic', 'ADJ'), ('without', 'ADP'), ('the', 'DET'), ('threat', 'NOUN'), ('of', 'ADP'), ('incubi', 'NOUN'), (',', '.'), ('and', 'CONJ'), ('let', 'VERB'), ('him', 'PRON'), ('go', 'VERB'), ('.', '.')]),\n",
              "       list([('my', 'DET'), ('object', 'NOUN'), (',', '.'), ('rather', 'ADV'), (',', '.'), ('is', 'VERB'), ('to', 'PRT'), ('alert', 'VERB'), ('you', 'PRON'), ('to', 'ADP'), ('an', 'DET'), ('aspect', 'NOUN'), ('or', 'CONJ'), ('two', 'NUM'), ('of', 'ADP'), ('the', 'DET'), ('affair', 'NOUN'), ('that', 'PRON'), ('could', 'VERB'), ('have', 'VERB'), ('the', 'DET'), ('gravest', 'ADJ'), ('implications', 'NOUN'), ('for', 'ADP'), ('you', 'PRON'), (',', '.'), ('far', 'ADV'), ('beyond', 'ADP'), ('the', 'DET'), ('legal', 'ADJ'), ('sanctions', 'NOUN'), ('society', 'NOUN'), ('might', 'VERB'), ('inflict', 'VERB'), ('.', '.')]),\n",
              "       list([('similarly', 'ADV'), (',', '.'), ('when', 'ADV'), ('a', 'DET'), ('reporter', 'NOUN'), ('once', 'ADV'), ('questioned', 'VERB'), ('lincoln', 'NOUN'), ('in', 'ADP'), ('cryptic', 'ADJ'), ('fashion', 'NOUN'), (',', '.'), ('lincoln', 'NOUN'), ('refused', 'VERB'), ('to', 'PRT'), ('make', 'VERB'), ('any', 'DET'), ('further', 'ADJ'), ('statement', 'NOUN'), ('.', '.')]),\n",
              "       list([('their', 'DET'), ('curiosity', 'NOUN'), ('went', 'VERB'), ('happily', 'ADV'), ('out', 'ADP'), ('of', 'ADP'), ('bounds', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), ('the', 'DET'), ('heart', 'NOUN'), (\"''\", '.'), (',', '.'), ('i', 'PRON'), ('said', 'VERB'), ('finally', 'ADV'), (',', '.'), ('``', '.'), ('is', 'VERB'), ('now', 'ADV'), ('either', 'CONJ'), ('in', 'ADP'), ('the', 'DET'), ('throat', 'NOUN'), ('or', 'CONJ'), ('the', 'DET'), ('mouth', 'NOUN'), ('or', 'CONJ'), ('the', 'DET'), ('stomach', 'NOUN'), ('or', 'CONJ'), ('the', 'DET'), ('shoes', 'NOUN'), ('.', '.')]),\n",
              "       list([('no', 'ADV'), (',', '.'), ('he', 'PRON'), (\"didn't\", 'VERB'), ('know', 'VERB'), ('of', 'ADP'), ('any', 'DET'), ('handyman-carpenter', 'NOUN'), ('.', '.')]),\n",
              "       list([('marcellus', 'NOUN'), (',', '.'), ('the', 'DET'), ('hero', 'NOUN'), (',', '.'), ('is', 'VERB'), ('beset', 'VERB'), ('from', 'ADP'), ('all', 'PRT'), ('sides', 'NOUN'), ('by', 'ADP'), ('the', 'DET'), ('problems', 'NOUN'), ('of', 'ADP'), ('approaching', 'VERB'), ('manhood', 'NOUN'), ('.', '.')]),\n",
              "       list([('but', 'CONJ'), ('it', 'PRON'), ('was', 'VERB'), ('small', 'ADJ'), ('consolation', 'NOUN'), ('.', '.')]),\n",
              "       list([('mr.', 'NOUN'), ('crombie', 'NOUN'), ('poked', 'VERB'), ('at', 'ADP'), ('it', 'PRON'), ('gingerly', 'ADV'), ('with', 'ADP'), ('his', 'DET'), ('foot', 'NOUN'), ('.', '.')]),\n",
              "       list([('but', 'CONJ'), ('they', 'PRON'), ('kept', 'VERB'), ('on', 'ADP'), ('clapping', 'VERB'), ('for', 'ADP'), ('a', 'DET'), ('long', 'ADJ'), (',', '.'), ('long', 'ADJ'), ('time', 'NOUN'), ('.', '.')]),\n",
              "       list([('the', 'DET'), ('first', 'ADJ'), ('one', 'NOUN'), ('said', 'VERB'), (',', '.'), ('``', '.'), ('he', 'PRON'), ('sure', 'ADV'), ('does', 'VERB'), ('shake', 'VERB'), ('.', '.')]),\n",
              "       list([('this', 'DET'), ('clergyman', 'NOUN'), ('should', 'VERB'), ('have', 'VERB'), ('referred', 'VERB'), ('to', 'ADP'), (\"shakespeare's\", 'NOUN'), ('dictum', 'NOUN'), (':', '.'), ('``', '.'), ('so-so', 'ADV'), ('is', 'VERB'), ('a', 'DET'), ('good', 'ADJ'), (',', '.'), ('very', 'ADV'), ('good', 'ADJ'), (',', '.'), ('very', 'ADV'), ('excellent', 'ADJ'), ('maxim', 'NOUN'), ('.', '.')]),\n",
              "       list([('evenings', 'NOUN'), (',', '.'), ('he', 'PRON'), ('was', 'VERB'), ('frequently', 'ADV'), ('to', 'PRT'), ('be', 'VERB'), ('seen', 'VERB'), ('at', 'ADP'), ('restaurants', 'NOUN'), ('like', 'ADP'), ('enrico', 'NOUN'), ('&', 'CONJ'), (\"paglieri's\", 'NOUN'), ('or', 'CONJ'), (\"peter's\", 'NOUN'), ('backyard', 'NOUN'), ('drunkenly', 'ADV'), ('donning', 'VERB'), (\"ladies'\", 'NOUN'), ('hats', 'NOUN'), ('and', 'CONJ'), ('singing', 'VERB'), ('``', '.'), ('o', 'X'), ('sole', 'X'), ('mio', 'X'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('the', 'DET'), ('girl', 'NOUN'), ('nodded', 'VERB'), ('understandingly', 'ADV'), ('.', '.')]),\n",
              "       list([('``', '.'), ('you', 'PRON'), ('what', 'DET'), (\"''\", '.'), ('?', '.'), ('?', '.')]),\n",
              "       list([('for', 'ADP'), ('it', 'PRON'), ('seems', 'VERB'), ('that', 'ADP'), ('barco', 'NOUN'), (',', '.'), ('fancying', 'VERB'), ('himself', 'PRON'), ('a', 'DET'), (\"ladies'\", 'NOUN'), ('man', 'NOUN'), ('(', '.'), ('and', 'CONJ'), ('why', 'ADV'), ('not', 'ADV'), (',', '.'), ('after', 'ADP'), ('seven', 'NUM'), ('marriages', 'NOUN'), ('?', '.'), ('?', '.')]),\n",
              "       list([('i', 'PRON'), ('studied', 'VERB'), ('welch', 'NOUN'), ('closely', 'ADV'), ('as', 'ADP'), ('the', 'DET'), ('trial', 'NOUN'), ('progressed', 'VERB'), ('for', 'ADP'), ('any', 'DET'), ('hint', 'NOUN'), ('which', 'DET'), ('might', 'VERB'), ('give', 'VERB'), ('me', 'PRON'), ('a', 'DET'), ('lead', 'NOUN'), ('as', 'ADP'), ('to', 'ADP'), ('how', 'ADV'), ('he', 'PRON'), ('might', 'VERB'), ('be', 'VERB'), ('thwarted', 'VERB'), ('.', '.')]),\n",
              "       list([('while', 'ADP'), ('mr.', 'NOUN'), ('blatz', 'NOUN'), ('was', 'VERB'), ('putting', 'VERB'), ('up', 'PRT'), ('the', 'DET'), ('pegboards', 'NOUN'), ('and', 'CONJ'), ('starting', 'VERB'), ('the', 'DET'), ('workbench', 'NOUN'), (',', '.'), ('mr.', 'NOUN'), ('crombie', 'NOUN'), ('told', 'VERB'), ('him', 'PRON'), ('of', 'ADP'), ('this', 'DET'), ('idea', 'NOUN'), ('about', 'ADP'), ('paneling', 'NOUN'), ('the', 'DET'), ('whole', 'ADJ'), ('end', 'NOUN'), ('of', 'ADP'), ('the', 'DET'), ('cellar', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), ('the', 'DET'), ('other', 'ADJ'), ('day', 'NOUN'), ('arnold', 'NOUN'), ('toynbee', 'NOUN'), ('spoke', 'VERB'), ('against', 'ADP'), ('the', 'DET'), ('inveterate', 'ADJ'), ('tendency', 'NOUN'), ('of', 'ADP'), ('our', 'DET'), ('species', 'NOUN'), ('to', 'PRT'), ('believe', 'VERB'), ('in', 'ADP'), ('the', 'DET'), ('uniqueness', 'NOUN'), ('of', 'ADP'), ('its', 'DET'), ('religions', 'NOUN'), (',', '.'), ('its', 'DET'), ('ideologies', 'NOUN'), (',', '.'), ('and', 'CONJ'), ('its', 'DET'), ('virtually', 'ADV'), ('everything', 'NOUN'), ('else', 'ADV'), ('.', '.')]),\n",
              "       list([('``', '.'), ('we', 'PRON'), ('have', 'VERB'), ('good', 'ADJ'), ('times', 'NOUN'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('another', 'DET'), ('case', 'NOUN'), ('involves', 'VERB'), ('a', 'DET'), ('newspaper', 'NOUN'), ('reporter', 'NOUN'), ('who', 'PRON'), ('tripped', 'VERB'), ('up', 'PRT'), ('a', 'DET'), ('politician', 'NOUN'), ('.', '.')]),\n",
              "       list([('her', 'DET'), ('reply', 'NOUN'), ('stung', 'VERB'), ('me', 'PRON'), (',', '.'), ('but', 'CONJ'), ('this', 'DET'), ('was', 'VERB'), ('too', 'ADV'), ('important', 'ADJ'), ('to', 'PRT'), ('let', 'VERB'), ('my', 'DET'), ('hurt', 'NOUN'), ('make', 'VERB'), ('any', 'DET'), ('difference', 'NOUN'), ('.', '.')]),\n",
              "       list([('like', 'ADP'), ('almost', 'ADV'), ('everybody', 'NOUN'), ('else', 'ADV'), (',', '.'), ('he', 'PRON'), ('confused', 'VERB'), ('the', 'DET'), ('heart', 'NOUN'), (',', '.'), ('both', 'DET'), ('as', 'ADP'), ('organ', 'NOUN'), ('and', 'CONJ'), ('as', 'ADP'), ('symbol', 'NOUN'), (',', '.'), ('with', 'ADP'), ('the', 'DET'), ('disturbed', 'VERB'), ('psyche', 'NOUN'), (',', '.'), ('the', 'DET'), ('deranged', 'VERB'), ('glands', 'NOUN'), (',', '.'), ('and', 'CONJ'), ('the', 'DET'), ('jumpy', 'ADJ'), ('central', 'ADJ'), ('nervous', 'ADJ'), ('system', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), ('there', 'ADV'), (\"''\", '.'), (',', '.'), ('said', 'VERB'), ('mr.', 'NOUN'), ('mills', 'NOUN'), ('.', '.')]),\n",
              "       list([('it', 'PRON'), ('read', 'VERB'), (':', '.'), ('``', '.'), ('shot', 'VERB'), ('in', 'ADP'), ('the', 'DET'), ('lumbar', 'ADJ'), ('region', 'NOUN'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('but', 'CONJ'), ('that', 'DET'), (\"wasn't\", 'VERB'), ('very', 'ADV'), ('funny', 'ADJ'), (',', '.'), ('either', 'ADV'), ('.', '.')]),\n",
              "       list([('a', 'DET'), (\"bee's\", 'NOUN'), ('place', 'NOUN'), ('is', 'VERB'), ('in', 'ADP'), ('the', 'DET'), ('hive', 'NOUN'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('a', 'DET'), ('man', 'NOUN'), ('of', 'ADP'), ('vaulting', 'VERB'), ('ambition', 'NOUN'), (',', '.'), ('with', 'ADP'), ('one', 'NUM'), ('eye', 'NOUN'), ('on', 'ADP'), ('the', 'DET'), ('mayorship', 'NOUN'), ('of', 'ADP'), ('los', 'NOUN'), ('angeles', 'NOUN'), (',', '.'), ('nothing', 'NOUN'), ('ever', 'ADV'), ('escaped', 'VERB'), ('him', 'PRON'), ('which', 'DET'), ('might', 'VERB'), ('possibly', 'ADV'), ('lead', 'VERB'), ('to', 'ADP'), ('personal', 'ADJ'), ('publicity', 'NOUN'), ('.', '.')]),\n",
              "       list([('the', 'DET'), ('attempt', 'NOUN'), ('had', 'VERB'), ('failed', 'VERB'), ('because', 'ADP'), (',', '.'), ('when', 'ADV'), ('endeavoring', 'VERB'), ('to', 'PRT'), ('cut', 'VERB'), ('his', 'DET'), ('wrists', 'NOUN'), (',', '.'), ('this', 'DET'), ('murderer', 'NOUN'), ('of', 'ADP'), ('seven', 'NUM'), ('women', 'NOUN'), ('had', 'VERB'), ('fainted', 'VERB'), ('at', 'ADP'), ('the', 'DET'), ('sight', 'NOUN'), ('of', 'ADP'), ('blood', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), (\"i've\", 'PRT'), ('been', 'VERB'), ('waiting', 'VERB'), ('to', 'PRT'), ('get', 'VERB'), ('these', 'DET'), ('things', 'NOUN'), ('done', 'VERB'), ('for', 'ADP'), ('months', 'NOUN'), (\"''\", '.'), (',', '.'), ('she', 'PRON'), ('said', 'VERB'), ('.', '.')]),\n",
              "       list([('she', 'PRON'), ('laughed', 'VERB'), ('a', 'DET'), ('little', 'ADJ'), ('and', 'CONJ'), ('tossed', 'VERB'), ('the', 'DET'), ('dregs', 'NOUN'), ('rakishly', 'ADV'), ('around', 'ADV'), ('in', 'ADP'), ('her', 'DET'), ('glass', 'NOUN'), ('.', '.')]),\n",
              "       list([('although', 'ADP'), ('the', 'DET'), ('play', 'NOUN'), ('does', 'VERB'), ('show', 'VERB'), ('a', 'DET'), ('certain', 'ADJ'), ('structural', 'ADJ'), ('amateurishness', 'NOUN'), ('(', '.'), ('there', 'PRT'), ('are', 'VERB'), ('eleven', 'NUM'), ('acts', 'NOUN'), ('varying', 'VERB'), ('in', 'ADP'), ('length', 'NOUN'), ('from', 'ADP'), ('twenty-five', 'NUM'), ('seconds', 'NOUN'), ('to', 'ADP'), ('an', 'DET'), ('hour', 'NOUN'), ('and', 'CONJ'), ('a', 'DET'), ('half', 'PRT'), (')', '.'), (',', '.'), ('the', 'DET'), ('statement', 'NOUN'), ('it', 'PRON'), ('makes', 'VERB'), ('concerning', 'ADP'), ('the', 'DET'), ('ceaseless', 'ADJ'), ('yearning', 'NOUN'), ('and', 'CONJ'), ('searching', 'VERB'), ('of', 'ADP'), ('youth', 'NOUN'), ('is', 'VERB'), ('profound', 'ADJ'), ('and', 'CONJ'), ('worthy', 'ADJ'), ('of', 'ADP'), ('our', 'DET'), ('attention', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), ('such', 'ADJ'), ('as', 'ADP'), (\"''\", '.'), ('?', '.'), ('?', '.')]),\n",
              "       list([('the', 'DET'), ('comic', 'ADJ'), ('indefinite', 'NOUN'), ('comprises', 'VERB'), ('an', 'DET'), ('extensive', 'ADJ'), ('class', 'NOUN'), ('of', 'ADP'), ('comedy', 'NOUN'), ('.', '.')]),\n",
              "       list([('i', 'PRON'), ('now', 'ADV'), ('felt', 'VERB'), ('it', 'PRON'), ('wiser', 'ADJ'), ('to', 'PRT'), ('keep', 'VERB'), ('baby-dear', 'NOUN'), ('in', 'ADP'), ('school', 'NOUN'), ('and', 'CONJ'), ('--', '.'), ('during', 'ADP'), ('the', 'DET'), ('summers', 'NOUN'), ('--', '.'), ('at', 'ADP'), ('a', 'DET'), ('camp', 'NOUN'), ('run', 'VERB'), ('by', 'ADP'), ('the', 'DET'), ('society', 'NOUN'), ('of', 'ADP'), ('friends', 'NOUN'), ('all', 'PRT'), ('year', 'NOUN'), ('around', 'ADV'), ('.', '.')]),\n",
              "       list([('i', 'PRON'), ('hated', 'VERB'), ('being', 'VERB'), ('dragged', 'VERB'), ('into', 'ADP'), ('the', 'DET'), ('salons', 'NOUN'), ('of', 'ADP'), ('these', 'DET'), ('aristocrats', 'NOUN'), ('.', '.')]),\n",
              "       list([('the', 'DET'), ('latest', 'ADJ'), ('and', 'CONJ'), (',', '.'), ('significantly', 'ADV'), (',', '.'), ('greatest', 'ADJ'), ('fruit', 'NOUN'), ('of', 'ADP'), ('this', 'DET'), ('theatrical', 'ADJ'), ('vine', 'NOUN'), ('is', 'VERB'), ('the', 'DET'), (',', '.'), ('an', 'DET'), ('adaptation', 'NOUN'), ('of', 'ADP'), (\"basho's\", 'NOUN'), ('classic', 'ADJ'), ('frog-haiku', 'NOUN'), ('by', 'ADP'), ('roger', 'NOUN'), ('entwhistle', 'NOUN'), (',', '.'), ('a', 'DET'), ('former', 'ADJ'), ('university', 'NOUN'), ('of', 'ADP'), ('maryland', 'NOUN'), ('chemistry', 'NOUN'), ('instructor', 'NOUN'), ('.', '.')]),\n",
              "       list([('for', 'ADP'), ('the', 'DET'), ('last', 'ADJ'), ('two', 'NUM'), ('years', 'NOUN'), (',', '.'), ('this', 'DET'), ('frontier', 'NOUN'), ('of', 'ADP'), ('the', 'DET'), ('arts', 'NOUN'), ('has', 'VERB'), ('produced', 'VERB'), ('a', 'DET'), ('number', 'NOUN'), ('of', 'ADP'), ('so-called', 'ADJ'), ('``', '.'), ('non-dramas', 'NOUN'), (\"''\", '.'), ('which', 'DET'), ('have', 'VERB'), ('left', 'VERB'), ('indelible', 'ADJ'), (',', '.'), ('bittersweet', 'ADJ'), ('impressions', 'NOUN'), ('on', 'ADP'), ('the', 'DET'), ('psyche', 'NOUN'), ('of', 'ADP'), ('this', 'DET'), ('veteran', 'NOUN'), ('theatregoer', 'NOUN'), ('.', '.')]),\n",
              "       list([('actually', 'ADV'), (',', '.'), ('it', 'PRON'), ('was', 'VERB'), ('inaudible', 'ADJ'), ('to', 'ADP'), ('anyone', 'NOUN'), ('not', 'ADV'), ('expecting', 'VERB'), ('it', 'PRON'), ('.', '.')]),\n",
              "       list([('one', 'NUM'), ('woman', 'NOUN'), ('--', '.'), ('she', 'PRON'), ('could', 'VERB'), ('have', 'VERB'), ('been', 'VERB'), ('either', 'CONJ'), ('english', 'ADJ'), ('or', 'CONJ'), ('american', 'ADJ'), ('--', '.'), ('went', 'VERB'), ('up', 'PRT'), ('to', 'ADP'), ('him', 'PRON'), ('and', 'CONJ'), ('said', 'VERB'), (',', '.'), (\"'\", '.'), ('but', 'CONJ'), ('you', 'PRON'), ('are', 'VERB'), ('the', 'DET'), ('foreigners', 'NOUN'), (\"'\", '.'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('for', 'ADP'), ('example', 'NOUN'), (',', '.'), ('when', 'ADV'), ('the', 'DET'), ('film', 'NOUN'), ('is', 'VERB'), ('only', 'ADV'), ('four', 'NUM'), ('minutes', 'NOUN'), ('old', 'ADJ'), (',', '.'), ('neitzbohr', 'NOUN'), ('refers', 'VERB'), ('to', 'ADP'), ('a', 'DET'), ('small', 'ADJ'), (',', '.'), ('victorian', 'ADJ'), ('piano', 'NOUN'), ('stool', 'NOUN'), ('as', 'ADP'), ('``', '.'), ('wilhelmina', 'NOUN'), (\"''\", '.'), (',', '.'), ('and', 'CONJ'), ('we', 'PRON'), ('are', 'VERB'), ('thereupon', 'ADV'), ('subjected', 'VERB'), ('to', 'ADP'), ('a', 'DET'), ('flashback', 'NOUN'), ('that', 'PRON'), ('informs', 'VERB'), ('us', 'PRON'), ('that', 'ADP'), ('this', 'DET'), ('very', 'ADJ'), ('piano', 'NOUN'), ('stool', 'NOUN'), ('was', 'VERB'), ('once', 'ADV'), ('used', 'VERB'), ('by', 'ADP'), ('an', 'DET'), ('epileptic', 'ADJ'), ('governess', 'NOUN'), ('whose', 'DET'), ('name', 'NOUN'), (',', '.'), ('of', 'ADP'), ('course', 'NOUN'), (',', '.'), ('was', 'VERB'), ('doris', 'NOUN'), ('(', '.'), ('the', 'DET'), ('english', 'NOUN'), ('equivalent', 'NOUN'), (',', '.'), ('when', 'ADV'), ('passed', 'VERB'), ('through', 'ADP'), ('middle-gaelic', 'NOUN'), ('derivations', 'NOUN'), (',', '.'), ('of', 'ADP'), ('wilhelmina', 'NOUN'), (')', '.'), ('.', '.')]),\n",
              "       list([('this', 'DET'), ('could', 'VERB'), ('be', 'VERB'), ('told', 'VERB'), ('chiefly', 'ADV'), ('from', 'ADP'), ('a', 'DET'), ('sort', 'NOUN'), ('of', 'ADP'), ('head-tossing', 'NOUN'), ('and', 'CONJ'), ('prancing', 'VERB'), (',', '.'), ('a', 'DET'), ('horselike', 'ADJ'), ('balkiness', 'NOUN'), ('of', 'ADP'), ('demeanor', 'NOUN'), ('.', '.')]),\n",
              "       list([('such', 'ADJ'), ('sweeping', 'VERB'), ('distractions', 'NOUN'), ('are', 'VERB'), ('hardly', 'ADV'), ('conducive', 'ADJ'), ('to', 'ADP'), ('``', '.'), ('oscar', 'NOUN'), (\"''\", '.'), ('winning', 'VERB'), ('performances', 'NOUN'), ('.', '.')]),\n",
              "       list([('it', 'PRON'), (\"isn't\", 'VERB'), ('what', 'DET'), ('the', 'DET'), ('ideologist', 'NOUN'), ('believes', 'VERB'), ('in', 'PRT'), (',', '.'), ('but', 'CONJ'), ('what', 'DET'), ('he', 'PRON'), ('hates', 'VERB'), (',', '.'), ('that', 'ADP'), ('puts', 'VERB'), ('the', 'DET'), ('world', 'NOUN'), ('in', 'ADP'), ('jeopardy', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), ('o', 'PRT'), ('!', '.'), ('!', '.')]),\n",
              "       list([('a', 'DET'), ('``', '.'), ('splash', 'NOUN'), ('party', 'NOUN'), (\"''\", '.'), ('at', 'ADP'), ('the', 'DET'), ('new', 'ADJ'), ('pool', 'NOUN'), (',', '.'), ('which', 'DET'), ('i', 'PRON'), ('had', 'VERB'), ('built', 'VERB'), ('in', 'ADP'), ('the', 'DET'), ('hope', 'NOUN'), ('of', 'ADP'), ('keeping', 'VERB'), ('letch', 'NOUN'), ('away', 'ADV'), ('from', 'ADP'), ('public', 'ADJ'), ('beaches', 'NOUN'), (',', '.'), ('when', 'ADV'), ('letch', 'NOUN'), ('and', 'CONJ'), ('a', 'DET'), ('certain', 'ADJ'), ('aquacutie', 'NOUN'), ('stayed', 'VERB'), ('underwater', 'ADV'), ('together', 'ADV'), ('for', 'ADP'), ('the', 'DET'), ('better', 'ADJ'), ('part', 'NOUN'), ('of', 'ADP'), ('an', 'DET'), ('hour', 'NOUN'), (';', '.'), (';', '.')]),\n",
              "       list([(\"''\", '.'), ('i', 'PRON'), ('thought', 'VERB'), (',', '.'), ('and', 'CONJ'), ('wondered', 'VERB'), ('what', 'DET'), ('kind', 'NOUN'), ('of', 'ADP'), ('homesteads', 'NOUN'), ('such', 'ADJ'), ('odd', 'ADJ'), ('pioneers', 'NOUN'), ('would', 'VERB'), ('establish', 'VERB'), ('in', 'ADP'), ('this', 'DET'), ('suburban', 'ADJ'), ('frontier', 'NOUN'), (';', '.'), (';', '.')]),\n",
              "       list([('mr.', 'NOUN'), ('mills', 'NOUN'), ('had', 'VERB'), ('done', 'VERB'), ('some', 'DET'), ('figuring', 'NOUN'), ('on', 'ADP'), ('a', 'DET'), ('scrap', 'NOUN'), ('of', 'ADP'), ('paper', 'NOUN'), ('and', 'CONJ'), ('given', 'VERB'), ('him', 'PRON'), ('the', 'DET'), ('various', 'ADJ'), ('kinds', 'NOUN'), ('of', 'ADP'), ('boards', 'NOUN'), ('and', 'CONJ'), ('two-by-fours', 'NOUN'), ('which', 'DET'), (',', '.'), ('properly', 'ADV'), ('handled', 'VERB'), (',', '.'), ('would', 'VERB'), (',', '.'), ('he', 'PRON'), ('had', 'VERB'), ('assured', 'VERB'), ('him', 'PRON'), (',', '.'), ('turn', 'VERB'), ('into', 'ADP'), ('a', 'DET'), ('workbench', 'NOUN'), ('.', '.')]),\n",
              "       list([('a', 'DET'), ('biting', 'VERB'), (',', '.'), ('pithy', 'ADJ'), ('parable', 'NOUN'), ('of', 'ADP'), ('the', 'DET'), ('all-pervading', 'ADJ'), ('hollowness', 'NOUN'), ('of', 'ADP'), ('modern', 'ADJ'), ('life', 'NOUN'), (',', '.'), ('the', 'DET'), ('piece', 'NOUN'), ('has', 'VERB'), ('been', 'VERB'), ('set', 'VERB'), ('by', 'ADP'), ('mlle', 'NOUN'), ('lagoon', 'NOUN'), ('to', 'ADP'), ('a', 'DET'), ('sumptuous', 'ADJ'), ('score', 'NOUN'), ('(', '.'), ('a', 'DET'), ('single', 'ADJ'), ('motif', 'NOUN'), ('played', 'VERB'), ('over', 'PRT'), ('and', 'CONJ'), ('over', 'PRT'), ('by', 'ADP'), ('four', 'NUM'), ('thousand', 'NUM'), ('french', 'ADJ'), ('horns', 'NOUN'), (')', '.'), ('by', 'ADP'), ('existentialist', 'NOUN'), ('hot-shot', 'NOUN'), ('jean-paul', 'NOUN'), ('sartre', 'NOUN'), ('.', '.')]),\n",
              "       list([('the', 'DET'), ('jail', 'NOUN'), ('authorities', 'NOUN'), ('--', '.'), ('attaching', 'VERB'), ('no', 'DET'), ('particular', 'ADJ'), ('significance', 'NOUN'), ('to', 'ADP'), ('the', 'DET'), ('episode', 'NOUN'), ('--', '.'), ('offered', 'VERB'), ('barco', 'NOUN'), ('whisky', 'NOUN'), ('to', 'PRT'), ('revive', 'VERB'), ('him', 'PRON'), (';', '.'), (';', '.')]),\n",
              "       list([('they', 'PRON'), ('also', 'ADV'), ('caught', 'VERB'), ('on', 'PRT'), ('a', 'DET'), ('little', 'ADJ'), ('bit', 'NOUN'), ('on', 'ADP'), ('how', 'ADV'), ('to', 'PRT'), ('smile', 'VERB'), ('a', 'DET'), ('lot', 'NOUN'), ('without', 'ADP'), ('your', 'DET'), ('lips', 'NOUN'), ('trembling', 'VERB'), ('.', '.')]),\n",
              "       list([('it', 'PRON'), ('was', 'VERB'), ('during', 'ADP'), ('her', 'DET'), ('``', '.'), ('writing', 'VERB'), (\"''\", '.'), ('period', 'NOUN'), ('that', 'ADP'), ('she', 'PRON'), ('and', 'CONJ'), ('herb', 'NOUN'), ('met', 'VERB'), ('and', 'CONJ'), ('decided', 'VERB'), ('that', 'ADP'), ('they', 'PRON'), ('were', 'VERB'), ('in', 'ADP'), ('love', 'NOUN'), ('.', '.')]),\n",
              "       list([('there', 'PRT'), ('was', 'VERB'), ('no', 'DET'), ('question', 'NOUN'), ('of', 'ADP'), ('that', 'DET'), ('.', '.')]),\n",
              "       list([('carpenters', 'NOUN'), ('all', 'PRT'), ('wanted', 'VERB'), ('steady', 'ADJ'), ('work', 'NOUN'), ('and', 'CONJ'), ('at', 'ADP'), ('the', 'DET'), ('moment', 'NOUN'), ('every', 'DET'), (\"mother's\", 'NOUN'), ('son', 'NOUN'), ('for', 'ADP'), ('twenty', 'NUM'), ('miles', 'NOUN'), ('around', 'ADV'), ('that', 'PRON'), ('could', 'VERB'), ('hammer', 'VERB'), ('nails', 'NOUN'), ('for', 'ADP'), ('twenty-five', 'NUM'), ('dollars', 'NOUN'), ('a', 'DET'), ('day', 'NOUN'), ('was', 'VERB'), ('working', 'VERB'), ('on', 'ADP'), ('the', 'DET'), ('school', 'NOUN'), ('job', 'NOUN'), ('.', '.')]),\n",
              "       list([('my', 'DET'), ('mother', 'NOUN'), ('constituted', 'VERB'), ('herself', 'PRON'), ('the', 'DET'), ('voice', 'NOUN'), ('of', 'ADP'), ('all', 'PRT'), ('of', 'ADP'), ('us', 'PRON'), ('.', '.')]),\n",
              "       list([('compassionately', 'ADV'), ('yours', 'PRON'), (',', '.')]),\n",
              "       list([('``', '.'), ('if', 'ADP'), ('you', 'PRON'), ('are', 'VERB'), ('trying', 'VERB'), ('to', 'PRT'), ('get', 'VERB'), ('us', 'PRON'), ('out', 'ADP'), ('of', 'ADP'), ('the', 'DET'), ('brothel', 'NOUN'), (',', '.'), ('the', 'DET'), ('dustbin', 'NOUN'), (',', '.'), ('the', 'DET'), ('kitchen', 'NOUN'), ('sink', 'NOUN'), (',', '.'), ('and', 'CONJ'), ('the', 'DET'), ('tawdry', 'ADJ'), ('living-room', 'NOUN'), (',', '.'), ('you', 'PRON'), ('are', 'VERB'), ('probably', 'ADV'), ('wasting', 'VERB'), ('your', 'DET'), ('time', 'NOUN'), (\"''\", '.'), (',', '.'), ('moreland', 'NOUN'), ('told', 'VERB'), ('me', 'PRON'), ('.', '.')]),\n",
              "       list([('``', '.'), ('i', 'PRON'), ('may', 'VERB'), ('be', 'VERB'), ('back', 'ADV'), (\"''\", '.'), (',', '.'), ('he', 'PRON'), ('explained', 'VERB'), (',', '.'), ('``', '.'), ('and', 'CONJ'), ('then', 'ADV'), ('again', 'ADV'), (',', '.'), ('i', 'PRON'), ('may', 'VERB'), ('not', 'ADV'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('but', 'CONJ'), ('the', 'DET'), ('supreme', 'ADJ'), ('court', 'NOUN'), (\"wouldn't\", 'VERB'), ('even', 'VERB'), ('hear', 'VERB'), ('my', 'DET'), ('case', 'NOUN'), ('!', '.'), ('!', '.')]),\n",
              "       list([('the', 'DET'), ('double', 'ADJ'), ('meaning', 'NOUN'), ('in', 'ADP'), ('the', 'DET'), ('anatomical', 'NOUN'), ('made', 'VERB'), ('it', 'PRON'), ('a', 'DET'), ('familiar', 'ADJ'), ('vaudeville', 'NOUN'), ('device', 'NOUN'), (',', '.'), ('as', 'ADP'), ('in', 'ADP'), ('the', 'DET'), ('gags', 'NOUN'), ('of', 'ADP'), ('weber', 'NOUN'), ('and', 'CONJ'), ('fields', 'NOUN'), ('.', '.')]),\n",
              "       list([(\"''\", '.')]),\n",
              "       list([('he', 'PRON'), ('asked', 'VERB'), (',', '.'), ('between', 'ADP'), ('wheezes', 'NOUN'), ('of', 'ADP'), ('laughter', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), ('i', 'PRON'), ('was', 'VERB'), ('just', 'ADV'), (\"sayin'\", 'VERB'), ('to', 'ADP'), ('him', 'PRON'), ('that', 'ADP'), (\"i'm\", 'PRT'), ('all', 'ADV'), ('ready', 'ADJ'), ('now', 'ADV'), ('for', 'ADP'), ('anything', 'NOUN'), ('else', 'ADV'), ('you', 'PRON'), ('want', 'VERB'), ('done', 'VERB'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('``', '.'), ('such', 'ADJ'), ('as', 'ADP'), (\"'\", '.'), ('sending', 'VERB'), ('the', 'DET'), ('cat', 'NOUN'), ('to', 'PRT'), ('guard', 'VERB'), ('the', 'DET'), ('mice', 'NOUN'), (\"'\", '.'), (',', '.'), ('or', 'CONJ'), (\"'\", '.'), ('the', 'DET'), ('falcon', 'NOUN'), ('to', 'PRT'), ('protect', 'VERB'), ('the', 'DET'), ('dove', 'NOUN'), (\"'\", '.'), (',', '.'), ('or', 'CONJ'), ('most', 'ADV'), ('terribly', 'ADV'), ('sharp', 'ADJ'), ('of', 'ADP'), ('all', 'PRT'), (',', '.'), (\"'\", '.'), ('the', 'DET'), ('human', 'NOUN'), ('being', 'NOUN'), ('to', 'PRT'), ('save', 'VERB'), ('humanity', 'NOUN'), (\"'\", '.'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('during', 'ADP'), ('these', 'DET'), ('first', 'ADJ'), ('days', 'NOUN'), ('of', 'ADP'), ('the', 'DET'), ('trial', 'NOUN'), ('i', 'PRON'), (\"didn't\", 'VERB'), ('have', 'VERB'), ('as', 'ADV'), ('much', 'ADJ'), ('time', 'NOUN'), ('to', 'PRT'), ('commiserate', 'VERB'), ('with', 'ADP'), ('viola', 'NOUN'), ('as', 'ADP'), ('i', 'PRON'), ('should', 'VERB'), ('have', 'VERB'), ('liked', 'VERB'), ('.', '.')]),\n",
              "       list([('``', '.'), ('what', 'DET'), ('a', 'DET'), ('nasty', 'ADJ'), ('fall', 'NOUN'), ('was', 'VERB'), ('there', 'ADV'), (\"''\", '.'), ('!', '.'), ('!', '.')]),\n",
              "       list([('the', 'DET'), ('police', 'NOUN'), ('captain', 'NOUN'), ('chortled', 'VERB'), ('.', '.')]),\n",
              "       list([('naturally', 'ADV'), (',', '.'), ('viola', 'NOUN'), ('had', 'VERB'), ('no', 'DET'), ('wish', 'NOUN'), ('to', 'PRT'), ('remain', 'VERB'), (',', '.'), ('but', 'CONJ'), ('she', 'PRON'), ('felt', 'VERB'), ('she', 'PRON'), (\"couldn't\", 'VERB'), ('leave', 'VERB'), ('so', 'ADV'), ('soon', 'ADV'), ('after', 'ADP'), ('her', 'DET'), ('arrival', 'NOUN'), (',', '.'), ('in', 'ADP'), ('all', 'PRT'), ('politeness', 'NOUN'), ('to', 'PRT'), ('her', 'DET'), ('host', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), ('by', 'ADP'), ('the', 'DET'), ('way', 'NOUN'), (\"''\", '.'), (',', '.'), ('said', 'VERB'), ('mr.', 'NOUN'), ('blatz', 'NOUN'), (',', '.'), ('packing', 'VERB'), ('his', 'DET'), ('tools', 'NOUN'), ('into', 'ADP'), ('a', 'DET'), ('battered', 'VERB'), ('carrier', 'NOUN'), (',', '.'), ('``', '.'), ('them', 'DET'), ('power', 'NOUN'), ('tools', 'NOUN'), ('needs', 'VERB'), ('extra', 'ADJ'), ('voltage', 'NOUN'), ('.', '.')]),\n",
              "       list([(\"hell's\", 'NOUN'), ('own', 'VERB'), ('amount', 'VERB'), ('of', 'ADP'), ('chaulmoogra', 'NOUN'), ('oil', 'NOUN'), ('did', 'VERB'), ('nothing', 'NOUN'), ('to', 'PRT'), ('alleviate', 'VERB'), ('their', 'DET'), ('torment', 'NOUN'), (';', '.'), (';', '.')]),\n",
              "       list([('the', 'DET'), ('doors', 'NOUN'), ('of', 'ADP'), ('the', 'DET'), ('d', 'NOUN'), ('train', 'NOUN'), ('slid', 'VERB'), ('shut', 'VERB'), (',', '.'), ('and', 'CONJ'), ('as', 'ADP'), ('i', 'PRON'), ('dropped', 'VERB'), ('into', 'ADP'), ('a', 'DET'), ('seat', 'NOUN'), ('and', 'CONJ'), (',', '.'), ('exhaling', 'VERB'), (',', '.'), ('looked', 'VERB'), ('up', 'PRT'), ('across', 'ADP'), ('the', 'DET'), ('aisle', 'NOUN'), (',', '.'), ('the', 'DET'), ('whole', 'ADJ'), ('aviary', 'NOUN'), ('in', 'ADP'), ('my', 'DET'), ('head', 'NOUN'), ('burst', 'VERB'), ('into', 'ADP'), ('song', 'NOUN'), ('.', '.')]),\n",
              "       list([('my', 'DET'), ('heart', 'NOUN'), ('was', 'VERB'), ('not', 'ADV'), ('in', 'ADP'), ('it', 'PRON'), (',', '.'), ('but', 'CONJ'), (',', '.'), ('oddly', 'ADV'), ('enough', 'ADV'), (',', '.'), ('it', 'PRON'), ('remains', 'VERB'), ('the', 'DET'), ('most', 'ADV'), ('financially', 'ADV'), ('successful', 'ADJ'), ('picture', 'NOUN'), ('of', 'ADP'), ('my', 'DET'), ('career', 'NOUN'), ('.', '.')]),\n",
              "       list([('in', 'ADP'), ('common', 'ADJ'), ('parlance', 'NOUN'), ('they', 'PRON'), ('would', 'VERB'), ('be', 'VERB'), ('described', 'VERB'), ('as', 'ADP'), ('misses', 'NOUN'), ('--', '.'), ('misinterpreters', 'NOUN'), (',', '.'), ('misunderstanders', 'NOUN'), (',', '.'), ('misdirectors', 'NOUN'), ('and', 'CONJ'), ('kindred', 'NOUN'), ('misdeeds', 'NOUN'), ('.', '.')]),\n",
              "       list([('they', 'PRON'), ('were', 'VERB'), ('``', '.'), ('sincere', 'ADJ'), (\"''\", '.'), ('--', '.'), ('men', 'NOUN'), ('of', 'ADP'), ('the', 'DET'), ('too-hearty', 'ADJ'), ('handclasp', 'NOUN'), ('and', 'CONJ'), ('the', 'DET'), ('urgent', 'ADJ'), ('smile', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), ('i', 'PRON'), ('see', 'VERB'), ('the', 'DET'), ('girl', 'NOUN'), ('but', 'CONJ'), ('i', 'PRON'), (\"don't\", 'VERB'), ('see', 'VERB'), ('the', 'DET'), ('car', 'NOUN'), ('with', 'ADP'), ('slacks', 'NOUN'), ('on', 'PRT'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('``', '.'), ('furiouser', 'ADJ'), ('and', 'CONJ'), ('furiouser', 'ADJ'), (\"''\", '.'), (',', '.'), ('i', 'PRON'), ('said', 'VERB'), ('.', '.')]),\n",
              "       list([('and', 'CONJ'), ('so', 'ADP'), ('they', 'PRON'), ('were', 'VERB'), ('consistently', 'ADV'), ('true', 'ADJ'), ('to', 'ADP'), ('their', 'DET'), ('principles', 'NOUN'), ('.', '.')]),\n",
              "       list([('moreland', 'NOUN'), ('sat', 'VERB'), ('brooding', 'VERB'), ('for', 'ADP'), ('a', 'DET'), ('full', 'ADJ'), ('minute', 'NOUN'), (',', '.'), ('during', 'ADP'), ('which', 'DET'), ('i', 'PRON'), ('made', 'VERB'), ('each', 'DET'), ('of', 'ADP'), ('us', 'PRON'), ('a', 'DET'), ('new', 'ADJ'), ('drink', 'NOUN'), ('.', '.')]),\n",
              "       list([('it', 'PRON'), ('all', 'PRT'), ('seemed', 'VERB'), ('rather', 'ADV'), ('hopeless', 'ADJ'), ('.', '.')]),\n",
              "       list([('he', 'PRON'), ('was', 'VERB'), ('bitter', 'ADJ'), ('and', 'CONJ'), ('resentful', 'ADJ'), ('toward', 'ADP'), ('her', 'PRON'), (',', '.'), ('personally', 'ADV'), ('resentful', 'ADJ'), ('.', '.')]),\n",
              "       list([('add', 'VERB'), ('things', 'NOUN'), ('as', 'ADP'), ('you', 'PRON'), ('find', 'VERB'), ('you', 'PRON'), ('need', 'VERB'), (\"'em\", 'PRON'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('my', 'DET'), ('plea', 'NOUN'), ('was', 'VERB'), ('unanimously', 'ADV'), ('voted', 'VERB'), ('down', 'PRT'), ('and', 'CONJ'), ('``', '.'), ('thrown', 'VERB'), ('out', 'PRT'), (\"''\", '.'), ('.', '.')]),\n",
              "       list([('moreland', 'NOUN'), ('asked', 'VERB'), (',', '.'), ('grimly', 'ADV'), ('.', '.')]),\n",
              "       list([('this', 'DET'), ('evening', 'NOUN'), (',', '.'), ('they', 'PRON'), ('were', 'VERB'), ('pushed', 'VERB'), ('in', 'PRT'), ('from', 'ADP'), ('the', 'DET'), ('breakfast', 'NOUN'), ('room', 'NOUN'), (',', '.'), ('with', 'ADP'), ('odds', 'NOUN'), ('and', 'CONJ'), ('ends', 'NOUN'), ('of', 'ADP'), ('dessert', 'NOUN'), ('distributed', 'VERB'), ('over', 'ADP'), ('them', 'PRON'), ('.', '.')]),\n",
              "       list([('``', '.'), ('there', 'PRT'), ('was', 'VERB'), ('also', 'ADV'), ('gott', 'X'), ('strafe', 'X'), ('angleterre', 'X'), (\"''\", '.'), (',', '.'), ('moreland', 'NOUN'), ('reminded', 'VERB'), ('me', 'PRON'), (',', '.'), ('``', '.'), ('and', 'CONJ'), ('carthago', 'X'), ('delenda', 'X'), ('est', 'X'), (',', '.'), ('or', 'CONJ'), ('if', 'ADP'), ('you', 'PRON'), ('will', 'VERB'), (',', '.'), ('deus', 'X'), ('strafe', 'X'), ('carthage', 'NOUN'), ('.', '.')]),\n",
              "       list([('``', '.'), ('mr.', 'NOUN'), ('jones', 'NOUN'), (',', '.'), ('you', 'PRON'), ('may', 'VERB'), ('recall', 'VERB'), ('that', 'ADP'), ('we', 'PRON'), ('printed', 'VERB'), ('last', 'ADJ'), ('week', 'NOUN'), ('your', 'DET'), ('denial', 'NOUN'), ('of', 'ADP'), ('having', 'VERB'), ('retracted', 'VERB'), ('the', 'DET'), ('contradiction', 'NOUN'), ('of', 'ADP'), ('your', 'DET'), ('original', 'ADJ'), ('statement', 'NOUN'), ('.', '.')]),\n",
              "       list([('she', 'PRON'), ('was', 'VERB'), ('biting', 'VERB'), ('into', 'ADP'), ('a', 'DET'), ('small', 'ADJ'), ('red', 'ADJ'), ('radish', 'NOUN'), (';', '.'), (';', '.')])],\n",
              "      dtype=object)"
            ]
          },
          "metadata": {},
          "execution_count": 81
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_rEasLVcwC6G"
      },
      "source": [
        "### Метод максимального правдоподобия для обучения модели\n",
        "\n",
        "* $\\normalsize S = s_0, s_1, ..., s_N$ - скрытые состояния, то есть различные теги\n",
        "* $\\normalsize O = o_0, o_1, ..., o_M$ - различные слова\n",
        "* $\\normalsize a_{i,j} = p(s_j|s_i)$ - вероятность того, что, находясь в скрытом состоянии $s_i$, мы попадем в состояние $s_j$ (элемент матрицы $A$)\n",
        "* $\\normalsize b_{k,j}=p(o_k|s_j)$ - вероятность того, что при скрытом состоянии $s_j$ находится слово $o_k$(элемент матрицы $B$)\n",
        "\n",
        "$$\\normalsize x_t \\in O, y_t \\in S$$\n",
        "$\\normalsize (x_t, y_t)$ - слово и тег, стоящие на месте $t$ $\\Rightarrow$ \n",
        "* $\\normalsize X$ - последовательность слов\n",
        "* $\\normalsize Y$ - последовательность тегов\n",
        "\n",
        "Требуется построить скрытую марковскую модель (class HiddenMarkovModel) и написать метод fit для настройки всех её параметров с помощью оценок максимального правдоподобия по размеченным данным (последовательности пар слово+тег):\n",
        "\n",
        "- Вероятности переходов между скрытыми состояниями $p(y_t | y_{t - 1})$ посчитайте на основе частот биграмм POS-тегов.\n",
        "\n",
        "\n",
        "- Вероятности эмиссий наблюдаемых состояний $p(x_t | y_t)$ посчитайте на основе частот \"POS-тег - слово\".\n",
        "\n",
        "\n",
        "- Распределение вероятностей начальных состояний $p(y_0)$ задайте равномерным.\n",
        "\n",
        "Пример $X = [x_0, x_1], Y = [y_0, y_1]$:<br><br>\n",
        "$$p(X, Y) = p(x_0, x_1, y_0, y_1) = p(y_0) \\cdot p(x_0, x_1, y_1 | y_0) = p(y_0) \\cdot p(x_0 | y_0) \\cdot\n",
        "p(x_1, y_1 | x_0, y_0) = \\\\ = p(y_0) \\cdot p(x_0 | y_0) \\cdot p(y_1 | x_0, y_0) \\cdot p(x_1 | x_0, y_0, y_1)\n",
        "= (\\text{в силу условий нашей модели}) = \\\\ = p(y_0) \\cdot p(x_0 | y_0) \\cdot p(y_1 | y_0) \\cdot p(x_1 | y_1) \\Rightarrow$$ <br>\n",
        "Для последовательности длины $n + 1$:<br>\n",
        "$$p(X, Y) = p(x_0 ... x_{n - 1}, y_0 ... y_{n - 1}) \\cdot p(y_n | y_{n - 1}) \\cdot p(x_n | y_n)$$"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tysPoe5rwC6I"
      },
      "source": [
        "#### Алгоритм Витерби для применения модели\n",
        "\n",
        "\n",
        "Требуется написать метод .predict для определения частей речи на тестовой выборке. Чтобы использовать обученную модель на новых данных, необходимо реализовать алгоритм Витерби. Это алгоритм динамиеского программирования, с помощью которого мы будем находить наиболее вероятную последовательность скрытых состояний модели для фиксированной последовательности слов:\n",
        "\n",
        "$$ \\hat{Y} = \\arg \\max_{Y} p(Y|X) = \\arg \\max_{Y} p(Y, X) $$\n",
        "\n",
        "Пусть $\\normalsize Q_{t,s}$ - самая вероятная последовательность скрытых состояний длины $t$ с окончанием в состоянии $s$. $\\normalsize q_{t, s}$ - вероятность этой последовательности.\n",
        "$$(1)\\: \\normalsize q_{t,s} = \\max_{s'} q_{t - 1, s'} \\cdot p(s | s') \\cdot p(o_t | s)$$\n",
        "$\\normalsize Q_{t,s}$ можно восстановить по argmax-ам."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QpEXdhOfwC6J"
      },
      "source": [
        "class HiddenMarkovModel:    \n",
        "    def __init__(self):\n",
        "    \n",
        "        pass\n",
        "        \n",
        "    def fit(self, train_tokens_tags_list):\n",
        "        \"\"\"\n",
        "        train_tokens_tags_list: массив предложений пар слово-тег (выборка для train) \n",
        "        \"\"\"\n",
        "        tags = [tag for sent in train_tokens_tags_list\n",
        "                for (word, tag) in sent]\n",
        "        words = [word for sent in train_tokens_tags_list\n",
        "                 for (word, tag) in sent]\n",
        "        \n",
        "        tag_num = pd.Series(nltk.FreqDist(tags)).sort_index()\n",
        "        word_num = pd.Series(nltk.FreqDist(words)).sort_values(ascending=False)\n",
        "         \n",
        "        self.tags = tag_num.index\n",
        "        self.words = word_num.index\n",
        "        \n",
        "        A = pd.DataFrame({'{}'.format(tag) : [0] * len(tag_num) for tag in tag_num.index}, index=tag_num.index)\n",
        "        B = pd.DataFrame({'{}'.format(tag) : [0] * len(word_num) for tag in tag_num.index}, index=word_num.index)\n",
        "        \n",
        "        # Вычисляем матрицу A и B по частотам слов и тегов\n",
        "        \n",
        "        # sent - предложение\n",
        "        # sent[i][0] - i слово в этом предложении, sent[i][1] - i тег в этом предложении\n",
        "        for sent in train_tokens_tags_list:\n",
        "            for i in range(len(sent)):\n",
        "                B.loc[sent[i][0],sent[i][1]] += 1 # текущая i-пара слово-тег (обновите матрицу B аналогично A)\n",
        "                if len(sent) - 1 != i: # для последнего тега нет следующего тега\n",
        "                    A.loc[sent[i][1], sent[i + 1][1]] += 1 # пара тег-тег\n",
        "        \n",
        "        # переходим к вероятностям\n",
        "        \n",
        "        # нормируем по строке, то есть по всем всевозможным следующим тегам\n",
        "        A = A.divide(A.sum(axis=1), axis=0)\n",
        "        \n",
        "        # нормируем по столбцу, то есть по всем всевозможным текущим словам\n",
        "        B = B / np.sum(B, axis=0)\n",
        "        \n",
        "        self.A = A\n",
        "        self.B = B\n",
        "        \n",
        "        return self\n",
        "        \n",
        "    \n",
        "    def predict(self, test_tokens_list):\n",
        "        \"\"\"\n",
        "        test_tokens_list : массив предложений пар слово-тег (выборка для test)\n",
        "        \"\"\"\n",
        "        predict_tags = OrderedDict({i : np.array([]) for i in range(len(test_tokens_list))})\n",
        "        print(self.A.keys())\n",
        "        print('words ', self.words)\n",
        "        for i_sent in range(len(test_tokens_list)):\n",
        "            \n",
        "            current_sent =  [sent_token[0] for sent_token in test_tokens_list[i_sent]]# текущее предложение\n",
        "            len_sent = len(current_sent) # длина предложения \n",
        "            q = np.zeros(shape=(len_sent + 1, len(self.tags)))\n",
        "            q[0] = 1 # нулевое состояние (равномерная инициализация по всем s)\n",
        "            back_point = np.zeros(shape=(len_sent + 1, len(self.tags))) # # argmax\n",
        "            back_point = back_point.astype('int')\n",
        "            for t in range(len_sent):\n",
        "                \n",
        "                # если мы не встречали такое слово в обучении, то вместо него будет \n",
        "                # самое популярное слово с самым популярным тегом (вопрос 2)\n",
        "                if current_sent[t] not in self.words:\n",
        "                    current_sent[t] = 'time'\n",
        "                    \n",
        "                # через max выбираем следующий тег\n",
        "                for i_s in range(len(self.tags)):\n",
        "                    \n",
        "                    s = self.tags[i_s]\n",
        "                    \n",
        "                    # формула (1)\n",
        "                    q[t + 1][i_s] = np.max(q[t, :] *\n",
        "                        self.A.loc[:, s] * \n",
        "                        self.B.loc[current_sent[t], s])\n",
        "                    \n",
        "                    # argmax формула(1)\n",
        "                    \n",
        "                    # argmax, чтобы восстановить последовательность тегов\n",
        "                    back_point[t + 1][i_s] = (q[t, :] * self.A.loc[:, s] * \n",
        "                        self.B.loc[current_sent[t],s]).reset_index()[s].idxmax() # индекс \n",
        "                    \n",
        "            back_point = back_point.astype('int')\n",
        "            \n",
        "            # выписываем теги, меняя порядок на реальный\n",
        "            back_tag = deque()\n",
        "            current_tag = np.argmax(q[len_sent])\n",
        "            for t in range(len_sent, 0, -1):\n",
        "                back_tag.appendleft(self.tags[current_tag])\n",
        "                current_tag = back_point[t, current_tag]\n",
        "             \n",
        "            predict_tags[i_sent] = np.array(back_tag)\n",
        "        \n",
        "        \n",
        "        return predict_tags                 "
      ],
      "execution_count": 33,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "y0BLgsWkwC6M"
      },
      "source": [
        "Обучите скрытую марковскую модель:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZcSoyUAxwC6M",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8ff21641-f349-4b98-f7f4-ffc97039cede"
      },
      "source": [
        "my_model = HiddenMarkovModel()\n",
        "my_model.fit(train_sents)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<__main__.HiddenMarkovModel at 0x7fa14cde5c50>"
            ]
          },
          "metadata": {},
          "execution_count": 93
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wWKE5RnL-_GD",
        "outputId": "43fa1499-7766-424e-d325-9371560271f1"
      },
      "source": [
        "predictions = my_model.predict(test_sents)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Index(['.', 'ADJ', 'ADP', 'ADV', 'CONJ', 'DET', 'NOUN', 'NUM', 'PRON', 'PRT',\n",
            "       'VERB', 'X'],\n",
            "      dtype='object')\n",
            "words  Index([',', 'the', '.', 'and', 'a', 'of', 'to', 'in', '''', '``',\n",
            "       ...\n",
            "       'ape', 'quavering', 'tapping', 'storming', 'occasions', 'sulks',\n",
            "       'blaming', 'profanity', 'temperament', 'meted'],\n",
            "      dtype='object', length=4431)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "BNiOYzUHWxj4",
        "outputId": "7e68843e-30b0-4185-a1cb-68a3f54ef60c"
      },
      "source": [
        "predictions"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "OrderedDict([(0,\n",
              "              array(['.', 'VERB', 'VERB', 'ADJ', 'ADP', 'NOUN', 'CONJ', 'VERB', 'ADV',\n",
              "                     'PRT', 'ADP', 'DET', 'NOUN', 'NOUN', 'ADP', 'DET', 'NOUN', 'NOUN',\n",
              "                     'ADP', 'NOUN', '.', '.', 'NUM', 'ADP', 'PRON', 'VERB', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (1,\n",
              "              array(['.', 'ADV', 'PRON', 'VERB', '.', '.', 'VERB', 'NOUN', 'ADV', '.',\n",
              "                     'ADP', 'ADP', 'PRON', 'VERB', 'NOUN', 'DET', 'NOUN', 'NOUN',\n",
              "                     'NOUN', 'ADP', 'DET', 'NOUN', 'ADV', 'ADJ', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (2,\n",
              "              array(['ADP', 'PRON', 'VERB', 'PRT', 'ADP', 'DET', 'NOUN', '.', 'PRON',\n",
              "                     'VERB', 'ADV', 'DET', 'NOUN', '.', 'VERB', 'NOUN', '.', 'ADV',\n",
              "                     'DET', 'NOUN', '.', 'ADP', 'PRON', 'VERB', 'NOUN', 'CONJ', 'DET',\n",
              "                     'NOUN', 'ADP', 'ADJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (3,\n",
              "              array(['ADP', 'PRON', 'NOUN', 'DET', 'NOUN', 'NOUN', 'NOUN', '.', 'NOUN',\n",
              "                     'NOUN', 'DET', 'NOUN', 'NOUN', 'NOUN', 'ADP', 'DET', 'DET', 'NOUN',\n",
              "                     'CONJ', 'NOUN', 'NOUN', '.', 'ADP', 'DET', 'NOUN', 'NOUN', 'NOUN',\n",
              "                     '.', 'DET', 'VERB', 'NOUN', 'ADP', 'DET', 'NOUN', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (4,\n",
              "              array(['ADV', 'ADP', 'ADJ', 'NOUN', 'VERB', 'DET', 'NOUN', 'VERB', 'VERB',\n",
              "                     'ADJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (5,\n",
              "              array(['ADP', 'ADJ', 'NOUN', 'VERB', 'ADJ', 'CONJ', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (6,\n",
              "              array(['NOUN', 'NOUN', 'NOUN', 'NOUN', 'ADP', 'NOUN', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (7,\n",
              "              array(['.', 'NOUN', 'ADP', 'PRON', '.', '.', '.'], dtype='<U4')),\n",
              "             (8,\n",
              "              array(['.', 'DET', 'NOUN', 'PRON', 'VERB', 'ADV', 'NOUN', 'CONJ', 'NOUN',\n",
              "                     '.', '.', 'PRON', 'VERB', '.', '.', 'VERB', 'ADV', 'VERB', 'NOUN',\n",
              "                     'CONJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (9, array(['PRON', 'VERB', 'DET', 'NOUN', '.'], dtype='<U4')),\n",
              "             (10,\n",
              "              array(['NOUN', 'VERB', 'VERB', 'ADP', 'DET', 'NOUN', 'NOUN', 'CONJ',\n",
              "                     'ADJ', 'NOUN', 'NOUN', 'ADP', 'ADV', '.'], dtype='<U4')),\n",
              "             (11,\n",
              "              array(['NOUN', 'ADP', 'DET', 'NOUN', 'VERB', 'DET', 'NOUN', 'ADP', 'NOUN',\n",
              "                     '.', 'DET', 'NOUN', 'PRON', 'VERB', 'PRON', 'PRON', 'VERB', 'NOUN',\n",
              "                     '.'], dtype='<U4')),\n",
              "             (12,\n",
              "              array(['PRON', 'NOUN', 'NOUN', 'NOUN', 'NOUN', 'ADP', 'NOUN', '.', 'ADV',\n",
              "                     'VERB', 'DET', 'NOUN', 'ADP', 'DET', 'NOUN', 'VERB', 'DET', 'VERB',\n",
              "                     'ADJ', '.', 'X', '.', 'CONJ', 'ADJ', '.', '.'], dtype='<U4')),\n",
              "             (13,\n",
              "              array(['ADV', '.', 'ADP', 'ADP', 'DET', 'NOUN', '.', 'PRON', 'VERB',\n",
              "                     'ADV', '.'], dtype='<U4')),\n",
              "             (14,\n",
              "              array(['ADV', '.', 'PRT', 'DET', 'NOUN', 'ADP', 'NOUN', 'VERB', 'VERB',\n",
              "                     'PRT', 'ADP', 'DET', 'NOUN', '.', 'ADP', 'DET', 'NOUN', 'VERB',\n",
              "                     'PRT', 'VERB', 'VERB', 'PRT', 'VERB', 'DET', 'NOUN', 'ADP', 'NOUN',\n",
              "                     'NOUN', 'ADP', 'DET', 'NOUN', '.', 'NOUN', '.', 'CONJ', 'ADV', '.',\n",
              "                     'NOUN', '.', 'ADP', 'DET', 'ADJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (15,\n",
              "              array(['ADV', '.', 'NOUN', 'VERB', 'PRON', 'NOUN', 'CONJ', 'NOUN', 'ADP',\n",
              "                     'DET', 'NOUN', 'PRON', 'VERB', 'ADV', 'ADJ', '.'], dtype='<U4')),\n",
              "             (16, array(['PRON', 'VERB', 'NOUN', 'ADJ', '.'], dtype='<U4')),\n",
              "             (17,\n",
              "              array(['PRON', 'VERB', 'PRT', 'VERB', 'DET', 'NOUN', '.'], dtype='<U4')),\n",
              "             (18,\n",
              "              array(['DET', 'ADJ', 'NOUN', 'PRON', 'VERB', 'PRON', '.', 'PRON', 'VERB',\n",
              "                     'DET', 'VERB', 'NOUN', '.', '.'], dtype='<U4')),\n",
              "             (19,\n",
              "              array(['DET', 'NOUN', 'DET', 'NOUN', 'VERB', 'ADP', 'ADV', 'VERB', 'ADP',\n",
              "                     'DET', 'NOUN', 'NOUN', 'ADP', 'NOUN', '.', 'ADV', 'ADP', 'PRON',\n",
              "                     'VERB', 'NOUN', 'NOUN', 'ADP', 'DET', 'NOUN', 'ADP', 'NOUN', '.',\n",
              "                     'CONJ', 'VERB', 'PRON', 'VERB', '.'], dtype='<U4')),\n",
              "             (20,\n",
              "              array(['DET', 'NOUN', '.', 'ADV', '.', 'VERB', 'ADP', 'NOUN', 'PRON',\n",
              "                     'ADP', 'DET', 'NOUN', 'CONJ', 'NUM', 'ADP', 'DET', 'NOUN', 'PRON',\n",
              "                     'VERB', 'VERB', 'DET', 'NOUN', 'NOUN', 'ADP', 'PRON', '.', 'ADJ',\n",
              "                     'NOUN', 'DET', 'NOUN', 'NOUN', 'NOUN', 'VERB', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (21,\n",
              "              array(['NOUN', '.', 'ADV', 'DET', 'NOUN', 'ADP', 'NOUN', 'NOUN', 'ADP',\n",
              "                     'ADJ', 'NOUN', '.', 'NOUN', 'VERB', 'PRT', 'VERB', 'DET', 'ADJ',\n",
              "                     'NOUN', '.'], dtype='<U4')),\n",
              "             (22,\n",
              "              array(['DET', 'NOUN', 'VERB', 'NOUN', 'PRT', 'ADP', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (23,\n",
              "              array(['.', 'DET', 'NOUN', '.', '.', 'PRON', 'VERB', 'ADV', '.', '.',\n",
              "                     'VERB', 'ADV', 'ADV', 'ADP', 'DET', 'NOUN', 'CONJ', 'DET', 'NOUN',\n",
              "                     'CONJ', 'DET', 'NOUN', 'CONJ', 'DET', 'NOUN', '.'], dtype='<U4')),\n",
              "             (24,\n",
              "              array(['ADV', '.', 'PRON', 'VERB', 'VERB', 'ADP', 'DET', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (25,\n",
              "              array(['NOUN', '.', 'DET', 'NOUN', '.', 'VERB', 'NOUN', 'ADP', 'PRT',\n",
              "                     'NOUN', 'ADP', 'DET', 'NOUN', 'ADP', 'NOUN', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (26,\n",
              "              array(['CONJ', 'PRON', 'VERB', 'ADJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (27,\n",
              "              array(['NOUN', 'NOUN', 'VERB', 'ADP', 'PRON', 'NOUN', 'ADP', 'DET',\n",
              "                     'NOUN', '.'], dtype='<U4')),\n",
              "             (28,\n",
              "              array(['CONJ', 'PRON', 'VERB', 'ADP', 'NOUN', 'ADP', 'DET', 'ADJ', '.',\n",
              "                     'ADJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (29,\n",
              "              array(['DET', 'ADJ', 'NOUN', 'VERB', '.', '.', 'PRON', 'ADJ', 'VERB',\n",
              "                     'VERB', '.'], dtype='<U4')),\n",
              "             (30,\n",
              "              array(['DET', 'NOUN', 'VERB', 'VERB', 'VERB', 'ADP', 'NOUN', 'NOUN', '.',\n",
              "                     '.', 'ADJ', 'VERB', 'DET', 'ADJ', '.', 'ADV', 'ADJ', '.', 'ADV',\n",
              "                     'ADJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (31,\n",
              "              array(['NOUN', '.', 'PRON', 'VERB', 'NOUN', 'PRT', 'VERB', 'VERB', 'ADP',\n",
              "                     'NOUN', 'ADP', 'NOUN', 'NOUN', 'NOUN', 'CONJ', 'NOUN', 'NOUN',\n",
              "                     'NOUN', 'NOUN', 'NOUN', 'NOUN', 'CONJ', 'NOUN', '.', 'NOUN', 'ADJ',\n",
              "                     'NOUN', '.', '.'], dtype='<U4')),\n",
              "             (32, array(['DET', 'NOUN', 'NOUN', 'NOUN', '.'], dtype='<U4')),\n",
              "             (33, array(['.', 'PRON', 'DET', '.', '.', '.'], dtype='<U4')),\n",
              "             (34,\n",
              "              array(['ADP', 'PRON', 'VERB', 'DET', 'NOUN', '.', 'NOUN', 'PRON', 'DET',\n",
              "                     'NOUN', 'NOUN', '.', 'CONJ', 'ADV', 'ADV', '.', 'ADP', 'NUM',\n",
              "                     'NOUN', '.', '.'], dtype='<U4')),\n",
              "             (35,\n",
              "              array(['PRON', 'NOUN', 'NOUN', 'ADV', 'ADP', 'DET', 'NOUN', 'VERB', 'ADP',\n",
              "                     'DET', 'NOUN', 'DET', 'VERB', 'VERB', 'PRON', 'DET', 'NOUN', 'ADP',\n",
              "                     'PRT', 'ADV', 'PRON', 'VERB', 'VERB', 'NOUN', '.'], dtype='<U4')),\n",
              "             (36,\n",
              "              array(['ADP', 'NOUN', 'NOUN', 'VERB', 'VERB', 'PRT', 'DET', 'NOUN',\n",
              "                     'CONJ', 'VERB', 'DET', 'NOUN', '.', 'NOUN', 'NOUN', 'VERB', 'PRON',\n",
              "                     'ADP', 'DET', 'NOUN', 'ADP', 'NOUN', 'DET', 'ADJ', 'NOUN', 'ADP',\n",
              "                     'DET', 'NOUN', '.'], dtype='<U4')),\n",
              "             (37,\n",
              "              array(['.', 'DET', 'ADJ', 'NOUN', 'NOUN', 'NOUN', 'VERB', 'ADP', 'DET',\n",
              "                     'NOUN', 'NOUN', 'ADP', 'DET', 'NOUN', 'PRT', 'VERB', 'ADP', 'DET',\n",
              "                     'NOUN', 'ADP', 'DET', 'NOUN', '.', 'DET', 'NOUN', '.', 'CONJ',\n",
              "                     'DET', 'ADV', 'NOUN', 'ADV', '.'], dtype='<U4')),\n",
              "             (38,\n",
              "              array(['.', 'PRON', 'VERB', 'ADJ', 'NOUN', '.', '.'], dtype='<U4')),\n",
              "             (39,\n",
              "              array(['DET', 'NOUN', 'VERB', 'DET', 'NOUN', 'NOUN', 'PRON', 'NOUN',\n",
              "                     'ADP', 'DET', 'NOUN', '.'], dtype='<U4')),\n",
              "             (40,\n",
              "              array(['DET', 'NOUN', 'VERB', 'PRON', '.', 'CONJ', 'DET', 'VERB', 'ADV',\n",
              "                     'ADJ', 'PRT', 'VERB', 'DET', 'NOUN', 'VERB', 'DET', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (41,\n",
              "              array(['VERB', 'ADV', 'NOUN', 'ADV', '.', 'PRON', 'VERB', 'DET', 'NOUN',\n",
              "                     '.', 'DET', 'ADP', 'NOUN', 'CONJ', 'ADP', 'NOUN', '.', 'ADP',\n",
              "                     'DET', 'NOUN', 'NOUN', '.', 'DET', 'NOUN', 'NOUN', '.', 'CONJ',\n",
              "                     'DET', 'NOUN', 'NOUN', 'ADJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (42,\n",
              "              array(['.', 'PRT', '.', '.', 'VERB', 'NOUN', 'NOUN', '.'], dtype='<U4')),\n",
              "             (43,\n",
              "              array(['PRON', 'VERB', '.', '.', 'VERB', 'ADP', 'DET', 'NOUN', 'NOUN',\n",
              "                     '.', '.'], dtype='<U4')),\n",
              "             (44,\n",
              "              array(['CONJ', 'PRON', 'VERB', 'ADV', 'ADJ', '.', 'CONJ', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (45,\n",
              "              array(['DET', 'NOUN', 'NOUN', 'VERB', 'ADP', 'DET', 'NOUN', '.', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (46,\n",
              "              array(['DET', 'NOUN', 'ADP', 'NOUN', 'NOUN', '.', 'ADP', 'NUM', 'NOUN',\n",
              "                     'ADP', 'DET', 'NOUN', 'ADP', 'NOUN', 'NOUN', '.', 'NOUN', 'ADV',\n",
              "                     'NOUN', 'PRON', 'DET', 'VERB', 'ADV', 'VERB', 'ADP', 'ADJ', 'NOUN',\n",
              "                     '.'], dtype='<U4')),\n",
              "             (47,\n",
              "              array(['DET', 'NOUN', 'VERB', 'NOUN', 'ADP', '.', 'ADV', 'NOUN', 'PRT',\n",
              "                     'VERB', 'DET', 'NOUN', '.', 'DET', 'NOUN', 'ADP', 'NUM', 'NOUN',\n",
              "                     'VERB', 'NOUN', 'ADP', 'DET', 'NOUN', 'ADP', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (48,\n",
              "              array(['.', 'PRT', 'VERB', 'VERB', 'PRT', 'VERB', 'DET', 'NOUN', 'VERB',\n",
              "                     'ADP', 'NOUN', '.', '.', 'PRON', 'VERB', '.'], dtype='<U4')),\n",
              "             (49,\n",
              "              array(['PRON', 'VERB', 'DET', 'ADJ', 'CONJ', 'VERB', 'DET', 'NOUN',\n",
              "                     'NOUN', 'ADV', 'ADP', 'DET', 'NOUN', '.'], dtype='<U4')),\n",
              "             (50,\n",
              "              array(['ADP', 'DET', 'NOUN', 'VERB', 'VERB', 'DET', 'ADJ', 'NOUN', 'NOUN',\n",
              "                     '.', 'PRT', 'VERB', 'NOUN', 'NOUN', 'NOUN', 'ADP', 'NOUN', 'ADP',\n",
              "                     'NOUN', 'NOUN', 'ADP', 'DET', 'NOUN', 'CONJ', 'DET', 'ADV', '.',\n",
              "                     '.', 'DET', 'NOUN', 'PRON', 'NOUN', 'ADP', 'DET', 'NOUN', 'NOUN',\n",
              "                     'CONJ', 'VERB', 'ADP', 'NOUN', 'VERB', 'ADJ', 'CONJ', 'ADJ', 'ADP',\n",
              "                     'DET', 'NOUN', '.'], dtype='<U4')),\n",
              "             (51, array(['.', 'ADJ', 'ADP', '.', '.', '.'], dtype='<U3')),\n",
              "             (52,\n",
              "              array(['DET', 'ADJ', 'ADJ', 'NOUN', 'DET', 'NOUN', 'NOUN', 'ADP', 'NOUN',\n",
              "                     '.'], dtype='<U4')),\n",
              "             (53,\n",
              "              array(['PRON', 'ADV', 'VERB', 'PRON', 'NOUN', 'PRT', 'VERB', 'NOUN',\n",
              "                     'ADP', 'NOUN', 'CONJ', '.', 'ADP', 'DET', 'NOUN', '.', 'ADP',\n",
              "                     'DET', 'NOUN', 'VERB', 'ADP', 'DET', 'NOUN', 'ADP', 'NOUN', 'PRT',\n",
              "                     'NOUN', 'ADV', '.'], dtype='<U4')),\n",
              "             (54,\n",
              "              array(['PRON', 'NOUN', 'VERB', 'VERB', 'ADP', 'DET', 'NOUN', 'ADP', 'DET',\n",
              "                     'NOUN', '.'], dtype='<U4')),\n",
              "             (55,\n",
              "              array(['DET', 'NOUN', 'CONJ', '.', 'NOUN', '.', 'ADJ', 'NOUN', 'ADP',\n",
              "                     'DET', 'NOUN', 'NOUN', 'VERB', 'DET', '.', 'DET', 'NOUN', 'ADP',\n",
              "                     'NOUN', 'ADJ', 'NOUN', 'ADP', 'NOUN', 'NOUN', '.', 'DET', 'ADJ',\n",
              "                     'NOUN', 'ADP', 'NOUN', 'NOUN', 'NOUN', '.'], dtype='<U4')),\n",
              "             (56,\n",
              "              array(['ADP', 'DET', 'ADJ', 'NUM', 'NOUN', '.', 'DET', 'NOUN', 'ADP',\n",
              "                     'DET', 'NOUN', 'VERB', 'NOUN', 'DET', 'NOUN', 'ADP', 'ADJ', '.',\n",
              "                     'NOUN', '.', 'DET', 'VERB', 'ADJ', 'NOUN', '.', 'NOUN', 'NOUN',\n",
              "                     'ADP', 'DET', 'NOUN', 'ADP', 'DET', 'NOUN', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (57,\n",
              "              array(['ADV', '.', 'PRON', 'VERB', 'NOUN', 'ADP', 'NOUN', 'ADV', 'NOUN',\n",
              "                     'PRON', '.'], dtype='<U4')),\n",
              "             (58,\n",
              "              array(['NUM', 'NOUN', '.', 'PRON', 'VERB', 'VERB', 'VERB', 'ADV', 'ADJ',\n",
              "                     'CONJ', 'ADJ', '.', 'VERB', 'PRT', 'ADP', 'PRON', 'CONJ', 'VERB',\n",
              "                     '.', '.', 'CONJ', 'PRON', 'VERB', 'DET', 'NOUN', '.', '.', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (59,\n",
              "              array(['ADP', 'NOUN', '.', 'ADV', 'DET', 'NOUN', 'VERB', 'ADV', 'NUM',\n",
              "                     'NOUN', 'ADJ', '.', 'NOUN', 'NOUN', 'ADP', 'DET', 'ADJ', '.',\n",
              "                     'NOUN', 'NOUN', 'NOUN', 'ADP', '.', 'NOUN', '.', '.', 'CONJ',\n",
              "                     'PRON', 'VERB', 'NOUN', 'VERB', 'ADP', 'DET', 'NOUN', 'PRON',\n",
              "                     'VERB', 'PRON', 'ADP', 'DET', 'ADJ', 'NOUN', 'NOUN', 'VERB', 'ADV',\n",
              "                     'VERB', 'ADP', 'DET', 'NOUN', 'NOUN', 'DET', 'NOUN', '.', 'ADP',\n",
              "                     'NOUN', '.', 'VERB', 'NOUN', '.', 'DET', 'ADJ', 'NOUN', '.', 'ADV',\n",
              "                     'VERB', 'ADP', 'NOUN', 'NOUN', '.', 'ADP', 'NOUN', '.', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (60,\n",
              "              array(['DET', 'VERB', 'VERB', 'VERB', 'NOUN', 'ADP', 'DET', 'NOUN', 'ADP',\n",
              "                     'NOUN', 'CONJ', 'NOUN', '.', 'DET', 'NOUN', 'NOUN', 'ADP', 'NOUN',\n",
              "                     '.'], dtype='<U4')),\n",
              "             (61,\n",
              "              array(['ADJ', 'NOUN', 'NOUN', 'VERB', 'ADV', 'NOUN', 'PRT', '.', 'NOUN',\n",
              "                     '.', 'NOUN', 'NOUN', '.'], dtype='<U4')),\n",
              "             (62,\n",
              "              array(['PRON', 'VERB', 'DET', 'DET', 'NOUN', 'VERB', 'ADP', '.', 'CONJ',\n",
              "                     'DET', 'PRON', 'NOUN', '.', 'DET', 'NOUN', 'DET', 'NOUN', 'ADP',\n",
              "                     'NOUN', '.'], dtype='<U4')),\n",
              "             (63, array(['.', 'NOUN', '.', '.'], dtype='<U4')),\n",
              "             (64,\n",
              "              array(['DET', '.', 'NOUN', 'NOUN', '.', 'ADP', 'DET', 'ADJ', 'NOUN', '.',\n",
              "                     'DET', 'PRON', 'VERB', 'VERB', 'ADP', 'DET', 'NOUN', 'ADP', 'VERB',\n",
              "                     'NOUN', 'ADV', 'ADP', 'ADJ', 'NOUN', '.', 'ADV', 'NOUN', 'CONJ',\n",
              "                     'DET', 'ADJ', 'NOUN', 'VERB', 'NOUN', 'ADV', 'ADP', 'DET', 'ADJ',\n",
              "                     'NOUN', 'ADP', 'DET', 'NOUN', '.', '.'], dtype='<U4')),\n",
              "             (65,\n",
              "              array(['.', 'PRON', 'VERB', '.', 'CONJ', 'NOUN', 'DET', 'NOUN', 'ADP',\n",
              "                     'NOUN', 'ADJ', 'NOUN', 'NOUN', 'VERB', 'NOUN', 'ADP', 'DET',\n",
              "                     'NOUN', 'NOUN', '.', '.'], dtype='<U4')),\n",
              "             (66,\n",
              "              array(['NOUN', 'NOUN', 'VERB', 'VERB', 'DET', 'NOUN', 'ADP', 'DET',\n",
              "                     'NOUN', 'ADP', 'NOUN', 'CONJ', 'VERB', 'PRON', 'DET', 'ADJ',\n",
              "                     'NOUN', 'ADP', 'NOUN', 'CONJ', 'NOUN', 'DET', '.', 'NOUN', 'NOUN',\n",
              "                     '.', 'VERB', '.', 'PRON', 'VERB', 'VERB', 'PRON', '.', 'VERB',\n",
              "                     'ADP', 'DET', 'NOUN', '.'], dtype='<U4')),\n",
              "             (67,\n",
              "              array(['DET', 'VERB', '.', 'NOUN', 'NOUN', 'ADP', 'DET', 'NOUN', 'NOUN',\n",
              "                     'ADP', 'ADJ', 'NOUN', '.', 'DET', 'NOUN', 'VERB', 'VERB', 'VERB',\n",
              "                     'ADP', 'NOUN', 'NOUN', 'ADP', 'DET', 'ADJ', 'NOUN', '.', 'DET',\n",
              "                     'ADJ', 'NOUN', 'VERB', 'PRT', 'CONJ', 'PRT', 'ADP', 'NUM', 'NUM',\n",
              "                     'ADJ', 'NOUN', '.', 'ADP', 'NOUN', 'NOUN', 'NOUN', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (68,\n",
              "              array(['DET', 'NOUN', 'NOUN', '.', 'VERB', 'DET', 'ADJ', 'NOUN', 'ADP',\n",
              "                     'DET', 'NOUN', '.', 'NOUN', 'NOUN', 'NOUN', 'PRT', 'VERB', 'PRON',\n",
              "                     '.', '.'], dtype='<U4')),\n",
              "             (69,\n",
              "              array(['PRON', 'ADV', 'VERB', 'ADP', 'DET', 'ADJ', 'NOUN', 'ADP', 'ADV',\n",
              "                     'ADP', 'NOUN', 'DET', 'NOUN', 'ADP', 'DET', 'NOUN', 'VERB', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (70,\n",
              "              array(['PRON', 'VERB', 'ADP', 'DET', '.', 'VERB', '.', 'NOUN', 'ADP',\n",
              "                     'PRON', 'CONJ', 'NOUN', 'VERB', 'CONJ', 'NOUN', 'ADP', 'PRON',\n",
              "                     'VERB', 'ADP', 'NOUN', '.'], dtype='<U4')),\n",
              "             (71,\n",
              "              array(['PRT', 'VERB', 'DET', 'NOUN', 'ADP', 'PRON', '.'], dtype='<U4')),\n",
              "             (72,\n",
              "              array(['NOUN', 'PRT', 'VERB', 'NOUN', 'NOUN', 'CONJ', 'ADP', 'DET',\n",
              "                     'NOUN', 'DET', 'NOUN', 'NOUN', 'ADP', 'NUM', 'NOUN', 'ADP', 'PRON',\n",
              "                     'VERB', 'NOUN', 'NOUN', 'ADP', 'NOUN', 'NOUN', 'DET', 'NOUN',\n",
              "                     'VERB', 'VERB', 'ADP', 'DET', 'NOUN', 'NOUN', '.'], dtype='<U4')),\n",
              "             (73,\n",
              "              array(['DET', 'NOUN', 'NOUN', 'PRON', 'DET', 'NOUN', 'ADP', 'PRT', 'ADP',\n",
              "                     'PRON', '.'], dtype='<U4')),\n",
              "             (74, array(['NOUN', 'PRON', '.'], dtype='<U4')),\n",
              "             (75,\n",
              "              array(['.', 'ADP', 'PRON', 'VERB', 'VERB', 'PRT', 'VERB', 'PRON', 'PRT',\n",
              "                     'ADP', 'DET', 'NOUN', '.', 'DET', 'NOUN', '.', 'DET', 'NOUN',\n",
              "                     'NOUN', '.', 'CONJ', 'DET', 'NOUN', 'NOUN', '.', 'PRON', 'VERB',\n",
              "                     'ADV', 'NOUN', 'DET', 'NOUN', '.', '.', 'NOUN', 'VERB', 'PRON',\n",
              "                     '.'], dtype='<U4')),\n",
              "             (76,\n",
              "              array(['.', 'PRON', 'VERB', 'VERB', 'ADV', '.', '.', 'PRON', 'VERB', '.',\n",
              "                     '.', 'CONJ', 'ADV', 'ADV', '.', 'PRON', 'VERB', 'ADV', '.', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (77,\n",
              "              array(['CONJ', 'DET', 'ADJ', 'NOUN', 'VERB', 'ADV', 'VERB', 'DET', 'NOUN',\n",
              "                     '.', '.'], dtype='<U4')),\n",
              "             (78,\n",
              "              array(['DET', 'ADJ', 'NOUN', 'ADP', 'DET', 'NOUN', 'VERB', 'PRON', 'DET',\n",
              "                     'ADJ', 'NOUN', 'NOUN', '.', 'ADP', 'ADP', 'DET', 'NOUN', 'ADP',\n",
              "                     'NOUN', 'CONJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (79, array(['.'], dtype='<U1')),\n",
              "             (80,\n",
              "              array(['PRON', 'VERB', '.', 'ADP', 'NOUN', 'ADP', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (81,\n",
              "              array(['.', 'PRON', 'VERB', 'ADV', 'NOUN', 'ADP', 'PRON', 'ADP', 'PRT',\n",
              "                     'ADV', 'ADJ', 'ADV', 'ADP', 'NOUN', 'ADV', 'PRON', 'VERB', 'VERB',\n",
              "                     '.', '.'], dtype='<U4')),\n",
              "             (82,\n",
              "              array(['.', 'ADJ', 'ADP', '.', 'VERB', 'DET', 'NOUN', 'ADP', 'NOUN',\n",
              "                     'DET', 'NOUN', '.', '.', 'CONJ', '.', 'DET', 'NOUN', 'ADP', 'NOUN',\n",
              "                     'DET', 'NOUN', '.', '.', 'CONJ', 'ADV', 'ADV', 'NOUN', 'ADP',\n",
              "                     'PRT', '.', '.', 'DET', 'NOUN', 'VERB', 'ADP', 'NOUN', 'NOUN', '.',\n",
              "                     '.', '.'], dtype='<U4')),\n",
              "             (83,\n",
              "              array(['ADP', 'DET', 'ADJ', 'NOUN', 'ADP', 'DET', 'NOUN', 'PRON', 'VERB',\n",
              "                     'VERB', 'ADP', 'ADJ', 'NOUN', 'ADP', 'NOUN', 'ADP', 'NOUN', 'ADP',\n",
              "                     'PRON', 'VERB', 'VERB', 'VERB', '.'], dtype='<U4')),\n",
              "             (84,\n",
              "              array(['.', 'DET', 'DET', 'NOUN', 'VERB', 'VERB', 'PRT', '.', '.', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (85, array(['DET', 'NOUN', 'NOUN', 'NOUN', '.'], dtype='<U4')),\n",
              "             (86,\n",
              "              array(['ADV', '.', 'NOUN', 'VERB', 'ADV', 'VERB', 'ADP', 'NOUN', '.',\n",
              "                     'CONJ', 'PRON', 'VERB', 'PRON', 'VERB', 'NOUN', 'ADV', 'ADV',\n",
              "                     'ADP', 'DET', 'NOUN', '.', 'ADP', 'PRT', 'NOUN', 'ADP', 'DET',\n",
              "                     'NOUN', '.'], dtype='<U4')),\n",
              "             (87,\n",
              "              array(['.', 'ADP', 'DET', 'NOUN', '.', '.', 'VERB', 'NOUN', 'NOUN', '.',\n",
              "                     'NOUN', 'DET', 'NOUN', 'ADP', 'DET', 'NOUN', 'NOUN', '.', '.',\n",
              "                     'DET', 'NOUN', 'NOUN', 'NOUN', 'ADJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (88,\n",
              "              array(['NOUN', 'ADJ', 'NOUN', 'ADP', 'NOUN', 'NOUN', 'VERB', 'NOUN',\n",
              "                     'PRT', 'VERB', 'DET', 'NOUN', '.', '.'], dtype='<U4')),\n",
              "             (89,\n",
              "              array(['DET', 'NOUN', 'ADP', 'DET', 'NOUN', 'NOUN', 'VERB', 'NOUN', '.',\n",
              "                     'CONJ', 'ADP', 'PRON', 'NOUN', 'ADP', 'DET', 'NOUN', 'CONJ', '.',\n",
              "                     'NOUN', '.', 'VERB', 'PRT', 'ADP', 'DET', 'NOUN', '.', 'DET',\n",
              "                     'ADJ', 'NOUN', 'ADP', 'DET', 'NOUN', 'NOUN', 'ADP', 'NOUN', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (90,\n",
              "              array(['DET', 'NOUN', 'VERB', 'ADV', 'ADP', 'PRON', '.', 'CONJ', '.',\n",
              "                     'ADV', 'ADV', '.', 'PRON', 'NOUN', 'DET', 'ADJ', 'NOUN', 'ADJ',\n",
              "                     'NOUN', 'ADP', 'DET', 'NOUN', '.'], dtype='<U4')),\n",
              "             (91,\n",
              "              array(['ADP', 'NOUN', 'NOUN', 'PRON', 'VERB', 'VERB', 'VERB', 'ADP',\n",
              "                     'NOUN', '.', 'NOUN', '.', 'NOUN', '.', 'NOUN', 'CONJ', 'NOUN',\n",
              "                     'NOUN', '.'], dtype='<U4')),\n",
              "             (92,\n",
              "              array(['PRON', 'VERB', '.', 'NOUN', '.', '.', 'NOUN', 'ADP', 'DET',\n",
              "                     'NOUN', 'NOUN', 'CONJ', 'DET', 'NOUN', 'NOUN', '.'], dtype='<U4')),\n",
              "             (93,\n",
              "              array(['.', 'PRON', 'VERB', 'DET', 'NOUN', 'CONJ', 'PRON', 'VERB', 'VERB',\n",
              "                     'DET', 'NOUN', 'ADP', 'NOUN', 'ADP', '.', '.'], dtype='<U4')),\n",
              "             (94,\n",
              "              array(['.', 'NOUN', 'CONJ', 'NOUN', '.', '.', 'PRON', 'VERB', '.'],\n",
              "                    dtype='<U4')),\n",
              "             (95,\n",
              "              array(['CONJ', 'ADV', 'PRON', 'VERB', 'NOUN', 'ADJ', 'ADP', 'DET', 'NOUN',\n",
              "                     '.'], dtype='<U4')),\n",
              "             (96,\n",
              "              array(['NOUN', 'VERB', 'NOUN', 'ADP', 'DET', 'ADJ', 'NOUN', '.', 'ADP',\n",
              "                     'DET', 'PRON', 'VERB', 'DET', 'ADP', 'PRON', 'DET', 'ADJ', 'NOUN',\n",
              "                     '.'], dtype='<U4')),\n",
              "             (97,\n",
              "              array(['PRON', 'PRT', 'VERB', 'ADV', 'NOUN', '.'], dtype='<U4')),\n",
              "             (98,\n",
              "              array(['PRON', 'VERB', 'ADJ', 'CONJ', 'NOUN', 'ADP', 'DET', '.', 'ADV',\n",
              "                     'NOUN', '.'], dtype='<U4')),\n",
              "             (99,\n",
              "              array(['NOUN', 'NOUN', 'ADP', 'PRON', 'VERB', 'PRON', 'VERB', 'PRON', '.',\n",
              "                     '.'], dtype='<U4')),\n",
              "             (100,\n",
              "              array(['DET', 'NOUN', 'VERB', 'NOUN', 'NOUN', 'PRT', 'CONJ', '.', 'VERB',\n",
              "                     'PRT', '.', '.'], dtype='<U4')),\n",
              "             (101, array(['NOUN', 'VERB', '.', 'NOUN', '.'], dtype='<U4')),\n",
              "             (102,\n",
              "              array(['DET', 'NOUN', '.', 'PRON', 'VERB', 'VERB', 'ADP', 'ADP', 'DET',\n",
              "                     'NOUN', 'NOUN', '.', 'ADP', 'NOUN', 'CONJ', 'NOUN', 'ADP', 'NOUN',\n",
              "                     'NOUN', 'ADP', 'PRON', '.'], dtype='<U4')),\n",
              "             (103,\n",
              "              array(['.', 'PRT', 'VERB', 'ADV', 'NOUN', 'NOUN', 'NOUN', '.', '.',\n",
              "                     'NOUN', 'NOUN', 'PRON', '.', '.', 'CONJ', 'NOUN', 'NOUN', 'NOUN',\n",
              "                     '.', 'CONJ', 'ADP', 'PRON', 'VERB', '.', 'NOUN', 'NOUN', 'NOUN',\n",
              "                     '.'], dtype='<U4')),\n",
              "             (104,\n",
              "              array(['.', 'NOUN', 'NOUN', '.', 'PRON', 'VERB', 'VERB', 'ADP', 'PRON',\n",
              "                     'NOUN', 'ADJ', 'NOUN', 'DET', 'NOUN', 'ADP', 'VERB', 'NOUN', 'DET',\n",
              "                     'NOUN', 'ADP', 'DET', 'ADJ', 'NOUN', '.'], dtype='<U4')),\n",
              "             (105,\n",
              "              array(['PRON', 'VERB', 'VERB', 'ADP', 'DET', 'ADJ', 'ADJ', 'NOUN', '.',\n",
              "                     '.'], dtype='<U4'))])"
            ]
          },
          "metadata": {},
          "execution_count": 95
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FeVNt19kwC6P"
      },
      "source": [
        "Проверьте работу реализованного алгоритма на следующих модельных примерах, проинтерпретируйте результат.\n",
        "\n",
        "- 'He can stay'\n",
        "- 'a cat and a dog'\n",
        "- 'I have a television'\n",
        "- 'My favourite character'"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cMJErf7NwC6Q",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5e98c2ad-2603-40df-902e-dcaacb7d6c40"
      },
      "source": [
        "sents = [['He', 'can', 'stay'], ['a', 'cat', 'and', 'a', 'dog'], ['I', 'have', 'a', 'television'],\n",
        "         ['My', 'favourite', 'character']]\n",
        "my_model.predict(sents)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Index(['.', 'ADJ', 'ADP', 'ADV', 'CONJ', 'DET', 'NOUN', 'NUM', 'PRON', 'PRT',\n",
            "       'VERB', 'X'],\n",
            "      dtype='object')\n",
            "words  Index([',', 'the', '.', 'and', 'a', 'of', 'to', 'in', '''', '``',\n",
            "       ...\n",
            "       'ape', 'quavering', 'tapping', 'storming', 'occasions', 'sulks',\n",
            "       'blaming', 'profanity', 'temperament', 'meted'],\n",
            "      dtype='object', length=4431)\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "OrderedDict([(0, array(['NOUN', 'NOUN', 'NOUN'], dtype='<U4')),\n",
              "             (1, array(['DET', 'NOUN', 'DET', 'DET', 'NOUN'], dtype='<U4')),\n",
              "             (2, array(['NOUN', 'NOUN', 'DET', 'NOUN'], dtype='<U4')),\n",
              "             (3, array(['NOUN', 'NOUN', 'NOUN'], dtype='<U4'))])"
            ]
          },
          "metadata": {},
          "execution_count": 101
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "mc3bpZVd4yfh",
        "outputId": "cb13757a-da6e-4c8b-a7a7-2b5bd12c35ab"
      },
      "source": [
        "sent = [['I', 'like', 'to', 'play', 'table', 'tennis']]\n",
        "my_model.predict(sent)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Index(['.', 'ADJ', 'ADP', 'ADV', 'CONJ', 'DET', 'NOUN', 'NUM', 'PRON', 'PRT',\n",
            "       'VERB', 'X'],\n",
            "      dtype='object')\n",
            "words  Index([',', 'the', '.', 'and', 'a', 'of', 'to', 'in', '''', '``',\n",
            "       ...\n",
            "       'ape', 'quavering', 'tapping', 'storming', 'occasions', 'sulks',\n",
            "       'blaming', 'profanity', 'temperament', 'meted'],\n",
            "      dtype='object', length=4431)\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "OrderedDict([(0,\n",
              "              array(['NOUN', 'NOUN', 'NOUN', 'NOUN', 'NOUN', 'NOUN'], dtype='<U4'))])"
            ]
          },
          "metadata": {},
          "execution_count": 102
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "suDCwbGMwC6T"
      },
      "source": [
        "### Вопрос 3:\n",
        "* Какой тег вы получили для слова `can`?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ReHeG3IjwC6U"
      },
      "source": [
        "\"NOUN\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ObAslurlwC6X"
      },
      "source": [
        "### Вопрос 4:\n",
        "* Какой тег вы получили для слова `favourite`?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "94crVrrXwC6Y"
      },
      "source": [
        "\"NOUN\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YPC4NZ4HwC6a"
      },
      "source": [
        "Примените модель к отложенной выборке Брауновского корпуса и подсчитайте точность определения тегов (accuracy). Сделайте выводы. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-7aioBc1wC6b"
      },
      "source": [
        "def accuracy_score(model, sents, predicts):\n",
        "    true_pred = 0\n",
        "    num_pred = 0\n",
        "\n",
        "    for i, sent in enumerate(sents):\n",
        "        for j, word_pair in enumerate(sent):\n",
        "            tag = word_pair[1]\n",
        "            word = word_pair[0]\n",
        "            if tag == predicts[i][j]:\n",
        "                true_pred += 1\n",
        "            num_pred += 1\n",
        "    \n",
        "    print(\"Accuracy:\", true_pred / num_pred * 100, '%')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "roesKrPCcMbp",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c6ebd9fa-8cdc-4fd7-bfeb-417df345bffd"
      },
      "source": [
        "accuracy_score(my_model, test_sents, predictions)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Accuracy: 88.82847256549678 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ff_W7J8XwC6e"
      },
      "source": [
        "### Вопрос 5:\n",
        "* Какое качество вы получили(округлите до одного знака после запятой)?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ptvlpc-6wC6f"
      },
      "source": [
        "88.8"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FpAgfZRTwC6h"
      },
      "source": [
        "## DefaultTagger"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9b4cPKyiwC6j"
      },
      "source": [
        "### Вопрос 6:\n",
        "* Какое качество вы бы получили, если бы предсказывали любой тег, как самый популярный тег на выборке train(округлите до одного знака после запятой)?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Td-0Pe0vwC6k"
      },
      "source": [
        "Вы можете испоьзовать DefaultTagger(метод tag для предсказания частей речи предложения)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NfZYlMxJwC6m"
      },
      "source": [
        "from nltk.tag import DefaultTagger"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_pF2mqrQCzF6"
      },
      "source": [
        "default_tagger = DefaultTagger('NOUN')\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-e5g3oJ6c1C8"
      },
      "source": [
        "def default_accuracy_score(sents):\n",
        "    true_pred = 0\n",
        "    num_pred = 0\n",
        "\n",
        "    for i, sent in enumerate(sents):\n",
        "        for j, word_pair in enumerate(sent):\n",
        "            tag = word_pair[1]\n",
        "            word = word_pair[0]\n",
        "            if tag == 'NOUN':\n",
        "                true_pred += 1\n",
        "            num_pred += 1\n",
        "    \n",
        "    print(\"Accuracy:\", true_pred / num_pred * 100, '%')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XeKNNGwVc6f8",
        "outputId": "4d5f9afa-6bd2-4d83-9540-cdb1e2393d45"
      },
      "source": [
        "default_accuracy_score(test_sents)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Accuracy: 20.217498764211566 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lz7Q3BfbwC6o"
      },
      "source": [
        "## NLTK, Rnnmorph"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PZvdB22Oyycz"
      },
      "source": [
        "Вспомним первый [семинар](https://colab.research.google.com/drive/1FHZVU6yJT61J8w1hALno0stD4VU36rit?usp=sharing) нашего курса. В том семинаре мы с вами работали c некоторыми библиотеками.\n",
        "\n",
        "Не забудьте преобразовать систему тэгов из `'en-ptb' в 'universal'` с помощью функции `map_tag` или используйте `tagset='universal'`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9bn1TGlGAfuL"
      },
      "source": [
        "from nltk.tag.mapping import map_tag"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UOMN3B3vqf_3"
      },
      "source": [
        "test_words = [[pair[0] for pair in sent] for sent in test_sents]\n",
        "test_tags = [[pair[1] for pair in sent] for sent in test_sents]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JJQFfbp8A_cj",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5ccc1c15-d3dd-4a4f-fffd-ccd8119c3e70"
      },
      "source": [
        "import nltk\n",
        "nltk.pos_tag(['I','like','eat','bananas'], tagset='universal')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[('I', 'PRON'), ('like', 'VERB'), ('eat', 'ADJ'), ('bananas', 'NOUN')]"
            ]
          },
          "metadata": {},
          "execution_count": 148
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JW-d91u1sLij"
      },
      "source": [
        "def nltk_accuracy_score(sentences, tags):\n",
        "    true_pred = 0\n",
        "    num_pred = 0\n",
        "    for i in range(len(sentences)):\n",
        "        preds = nltk.pos_tag(sentences[i],tagset='universal')\n",
        "        for j in range(len(sentences[i])):\n",
        "            if preds[j][1] == tags[i][j]:\n",
        "                true_pred += 1\n",
        "            num_pred += 1\n",
        "    \n",
        "    print(\"Accuracy:\", true_pred / num_pred * 100, '%')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "QcVm3e76tgrS",
        "outputId": "1948ec81-f36f-4cd8-f1bf-ce547e80d63e"
      },
      "source": [
        "nltk_accuracy_score(test_words, test_tags)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Accuracy: 89.22392486406328 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8LD_61W7N35q",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "2db7ca38-60c8-407c-e153-043b9d084e6e"
      },
      "source": [
        "from rnnmorph.predictor import RNNMorphPredictor\n",
        "predictor = RNNMorphPredictor(language=\"en\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[nltk_data] Downloading package wordnet to /root/nltk_data...\n",
            "[nltk_data]   Package wordnet is already up-to-date!\n",
            "[nltk_data] Downloading package averaged_perceptron_tagger to\n",
            "[nltk_data]     /root/nltk_data...\n",
            "[nltk_data]   Package averaged_perceptron_tagger is already up-to-\n",
            "[nltk_data]       date!\n",
            "[nltk_data] Downloading package universal_tagset to /root/nltk_data...\n",
            "[nltk_data]   Package universal_tagset is already up-to-date!\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DzrZyeLuvLNy",
        "outputId": "68ba3696-3a21-407d-8712-b93e000e7908"
      },
      "source": [
        "!pip install rnnmorph"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Collecting rnnmorph\n",
            "  Downloading rnnmorph-0.4.1.tar.gz (19.7 MB)\n",
            "\u001b[K     |████████████████████████████████| 19.7 MB 45.3 MB/s \n",
            "\u001b[?25hRequirement already satisfied: numpy>=1.12.1 in /usr/local/lib/python3.7/dist-packages (from rnnmorph) (1.19.5)\n",
            "Requirement already satisfied: scipy>=0.19.0 in /usr/local/lib/python3.7/dist-packages (from rnnmorph) (1.4.1)\n",
            "Requirement already satisfied: scikit-learn>=0.18.1 in /usr/local/lib/python3.7/dist-packages (from rnnmorph) (0.22.2.post1)\n",
            "Requirement already satisfied: keras>=2.1.4 in /usr/local/lib/python3.7/dist-packages (from rnnmorph) (2.6.0)\n",
            "Requirement already satisfied: h5py>=2.7.0 in /usr/local/lib/python3.7/dist-packages (from rnnmorph) (3.1.0)\n",
            "Collecting pymorphy2>=0.8\n",
            "  Downloading pymorphy2-0.9.1-py3-none-any.whl (55 kB)\n",
            "\u001b[K     |████████████████████████████████| 55 kB 4.1 MB/s \n",
            "\u001b[?25hCollecting russian-tagsets==0.6\n",
            "  Downloading russian-tagsets-0.6.tar.gz (23 kB)\n",
            "Requirement already satisfied: tqdm>=4.14.0 in /usr/local/lib/python3.7/dist-packages (from rnnmorph) (4.62.3)\n",
            "Collecting jsonpickle>=0.9.4\n",
            "  Downloading jsonpickle-2.0.0-py2.py3-none-any.whl (37 kB)\n",
            "Requirement already satisfied: nltk>=3.2.5 in /usr/local/lib/python3.7/dist-packages (from rnnmorph) (3.2.5)\n",
            "Requirement already satisfied: cached-property in /usr/local/lib/python3.7/dist-packages (from h5py>=2.7.0->rnnmorph) (1.5.2)\n",
            "Requirement already satisfied: importlib-metadata in /usr/local/lib/python3.7/dist-packages (from jsonpickle>=0.9.4->rnnmorph) (4.8.1)\n",
            "Requirement already satisfied: six in /usr/local/lib/python3.7/dist-packages (from nltk>=3.2.5->rnnmorph) (1.15.0)\n",
            "Collecting dawg-python>=0.7.1\n",
            "  Downloading DAWG_Python-0.7.2-py2.py3-none-any.whl (11 kB)\n",
            "Requirement already satisfied: docopt>=0.6 in /usr/local/lib/python3.7/dist-packages (from pymorphy2>=0.8->rnnmorph) (0.6.2)\n",
            "Collecting pymorphy2-dicts-ru<3.0,>=2.4\n",
            "  Downloading pymorphy2_dicts_ru-2.4.417127.4579844-py2.py3-none-any.whl (8.2 MB)\n",
            "\u001b[K     |████████████████████████████████| 8.2 MB 48.5 MB/s \n",
            "\u001b[?25hRequirement already satisfied: joblib>=0.11 in /usr/local/lib/python3.7/dist-packages (from scikit-learn>=0.18.1->rnnmorph) (1.0.1)\n",
            "Requirement already satisfied: typing-extensions>=3.6.4 in /usr/local/lib/python3.7/dist-packages (from importlib-metadata->jsonpickle>=0.9.4->rnnmorph) (3.7.4.3)\n",
            "Requirement already satisfied: zipp>=0.5 in /usr/local/lib/python3.7/dist-packages (from importlib-metadata->jsonpickle>=0.9.4->rnnmorph) (3.6.0)\n",
            "Building wheels for collected packages: rnnmorph, russian-tagsets\n",
            "  Building wheel for rnnmorph (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for rnnmorph: filename=rnnmorph-0.4.1-py3-none-any.whl size=19746379 sha256=21fd2b41d1a20e469acb7194b7bcc0eb4931a9f8ba50fcc719f065dd61c0ddf7\n",
            "  Stored in directory: /root/.cache/pip/wheels/b9/e8/8b/b4639d66ee5373f0db8dbe9a2ea94784c8e93fd52e9d2e811f\n",
            "  Building wheel for russian-tagsets (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for russian-tagsets: filename=russian_tagsets-0.6-py3-none-any.whl size=24637 sha256=813f69d7ae99f5014e339c40ac815794b9dd2573bd5a0b15ae16d7979a524072\n",
            "  Stored in directory: /root/.cache/pip/wheels/5e/b4/26/9c17a7cdcfc6b8cf43111312f3e7c6abb8e583599e37c422f8\n",
            "Successfully built rnnmorph russian-tagsets\n",
            "Installing collected packages: pymorphy2-dicts-ru, dawg-python, russian-tagsets, pymorphy2, jsonpickle, rnnmorph\n",
            "Successfully installed dawg-python-0.7.2 jsonpickle-2.0.0 pymorphy2-0.9.1 pymorphy2-dicts-ru-2.4.417127.4579844 rnnmorph-0.4.1 russian-tagsets-0.6\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "h6KCKuB1vEe4"
      },
      "source": [
        "smth = predictor.predict(['I','like','eat','bananas'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "LTpm8F9bv4HP",
        "outputId": "771fc9b8-5500-49f7-eed0-5cc449397c07"
      },
      "source": [
        "smth[1]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<normal_form=like; word=like; pos=VERB; tag=Mood=Ind|Tense=Pres|VerbForm=Fin; score=0.9968>"
            ]
          },
          "metadata": {},
          "execution_count": 193
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IceyOl4MxIAu"
      },
      "source": [
        "def rnnmorph_accuracy_score(sentences, tags):\n",
        "    true_pred = 0\n",
        "    num_pred = 0\n",
        "    for i in range(len(sentences)):\n",
        "        preds = predictor.predict(sentences[i])\n",
        "        for j in range(len(sentences[i])):\n",
        "            if preds[j].pos == tags[i][j]:\n",
        "                true_pred += 1\n",
        "            num_pred += 1\n",
        "    \n",
        "    print(\"Accuracy:\", true_pred / num_pred * 100, '%')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nA1xH__9xasP",
        "outputId": "26b126d7-5094-4021-98e3-baa9f6e94824"
      },
      "source": [
        "rnnmorph_accuracy_score(test_words, test_tags)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "WARNING:tensorflow:5 out of the last 8 calls to <function Model.make_predict_function.<locals>.predict_function at 0x7fa0faf99cb0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for  more details.\n",
            "Accuracy: 62.827483934750376 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "i1z8x4vvwC6s"
      },
      "source": [
        "### Вопрос 7:\n",
        "* Какое качество вы получили, используя каждую из двух библиотек? Сравните их результаты.\n",
        "\n",
        "* Качество с библиотекой rnnmorph должно быть хуже, так как там используется немного другая система тэгов. Какие здесь отличия?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GBd3RgqVwC6s"
      },
      "source": [
        "#nltk получен скор: Accuracy: 89.22392486406328 %\n",
        "#rnnmorph получен скор: Accuracy: 62.827483934750376 %\n",
        "#как и предполагалось, качество и вправду ниже у rnnmorph.predictor"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5w1W5hSkcMcV"
      },
      "source": [
        "## BiLSTMTagger"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mm1-S3t2cMcW"
      },
      "source": [
        "### Подготовка данных"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GayTl7mUcMcX"
      },
      "source": [
        "Изменим структуру данных"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cZsAN8oQUVbq"
      },
      "source": [
        "brown_tagged_sents = brown.tagged_sents(tagset=\"universal\", categories='humor')"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CnXcI64fxoj4",
        "scrolled": false,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "7c7c00e4-9ada-4f2e-e2b2-c804d631c479"
      },
      "source": [
        "pos_data = [list(zip(*sent)) for sent in brown_tagged_sents]\n",
        "print(pos_data[0])"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[('It', 'was', 'among', 'these', 'that', 'Hinkle', 'identified', 'a', 'photograph', 'of', 'Barco', '!', '!'), ('PRON', 'VERB', 'ADP', 'DET', 'ADP', 'NOUN', 'VERB', 'DET', 'NOUN', 'ADP', 'NOUN', '.', '.')]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DpRE3c-3cMcc"
      },
      "source": [
        "До этого мы писали много кода сами, теперь пора эксплуатировать pytorch"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gvFlzrYnxokE"
      },
      "source": [
        "from torchtext.legacy.data import Field, BucketIterator\n",
        "import torchtext\n",
        "\n",
        "# наши поля\n",
        "WORD = Field(lower=True)\n",
        "TAG = Field(unk_token=None) # все токены нам извсетны\n",
        "\n",
        "# создаем примеры\n",
        "examples = []\n",
        "for words, tags in pos_data:\n",
        "    examples.append(torchtext.legacy.data.Example.fromlist([list(words), list(tags)], fields=[('words', WORD), ('tags', TAG)]))"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tjl6u6cpOc1u"
      },
      "source": [
        "Вот один наш пример:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dnrzktytN9rL",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "14872865-9a23-45ad-a95a-6924d73b23a2"
      },
      "source": [
        "print(vars(examples[0]))"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "{'words': ['it', 'was', 'among', 'these', 'that', 'hinkle', 'identified', 'a', 'photograph', 'of', 'barco', '!', '!'], 'tags': ['PRON', 'VERB', 'ADP', 'DET', 'ADP', 'NOUN', 'VERB', 'DET', 'NOUN', 'ADP', 'NOUN', '.', '.']}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nUhTrWCWcMcj"
      },
      "source": [
        "Теперь формируем наш датасет"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LGKkbZUIxokO",
        "scrolled": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "fabc71c7-c32c-498c-b7ce-4f48f9999be7"
      },
      "source": [
        "# кладем примеры в наш датасет\n",
        "dataset = torchtext.legacy.data.Dataset(examples, fields=[('words', WORD), ('tags', TAG)])\n",
        "\n",
        "train_data, valid_data, test_data = dataset.split(split_ratio=[0.8, 0.1, 0.1])\n",
        "\n",
        "print(f\"Number of training examples: {len(train_data.examples)}\")\n",
        "print(f\"Number of validation examples: {len(valid_data.examples)}\")\n",
        "print(f\"Number of testing examples: {len(test_data.examples)}\")"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of training examples: 842\n",
            "Number of validation examples: 106\n",
            "Number of testing examples: 105\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "T89unpppcMcp"
      },
      "source": [
        "Построим словари. Параметр `min_freq` выберете сами. При построении словаря испольузем только **train**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tZwkwhlrxoka",
        "scrolled": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c451ae50-201a-42f5-9344-20eb85e6297c"
      },
      "source": [
        "WORD.build_vocab(train_data, min_freq=3)\n",
        "TAG.build_vocab(train_data)\n",
        "\n",
        "print(f\"Unique tokens in source (ru) vocabulary: {len(WORD.vocab)}\")\n",
        "print(f\"Unique tokens in target (en) vocabulary: {len(TAG.vocab)}\")\n",
        "\n",
        "print(WORD.vocab.itos[::200])\n",
        "print(TAG.vocab.itos)"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Unique tokens in source (ru) vocabulary: 732\n",
            "Unique tokens in target (en) vocabulary: 13\n",
            "['<unk>', 'woman', 'becomes', 'happened']\n",
            "['<pad>', 'NOUN', 'VERB', '.', 'DET', 'ADP', 'ADJ', 'PRON', 'ADV', 'CONJ', 'PRT', 'NUM', 'X']\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vjn07NP-xokl",
        "scrolled": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6aeab221-ca85-43aa-b772-c03142c47fbd"
      },
      "source": [
        "print(vars(train_data.examples[9]))"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "{'words': ['he', 'had', 'aged', 'thirty', 'years', ',', 'and', 'his', 'face', ',', 'the', 'color', 'of', 'tallow', ',', 'was', 'crisscrossed', 'with', 'wrinkles', ',', 'as', 'though', 'it', 'had', 'been', 'wrapped', 'in', 'chicken', 'wire', '.'], 'tags': ['PRON', 'VERB', 'VERB', 'NUM', 'NOUN', '.', 'CONJ', 'DET', 'NOUN', '.', 'DET', 'NOUN', 'ADP', 'NOUN', '.', 'VERB', 'VERB', 'ADP', 'NOUN', '.', 'ADP', 'ADP', 'PRON', 'VERB', 'VERB', 'VERB', 'ADP', 'NOUN', 'NOUN', '.']}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LxgkU4cZcMcz"
      },
      "source": [
        "Посмотрим с насколько большими предложениями мы имеем дело"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dVpMi1_0xoku",
        "scrolled": true,
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        },
        "outputId": "52ad6817-8950-46c6-b969-924bcb45b716"
      },
      "source": [
        "length = map(len, [vars(x)['words'] for x in train_data.examples])\n",
        "\n",
        "plt.figure(figsize=[8, 4])\n",
        "plt.title(\"Length distribution in Train data\")\n",
        "plt.hist(list(length), bins=20);"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAe0AAAEICAYAAAByPazKAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAYHklEQVR4nO3de5BmdX3n8fcnjIh4Gy4tgRl0Rh1x0YqR6kUsjItistzWoSoUgTU6KtlZDV6ji6AbwdRqcNeKl9KwmQAyGhalEGRWjBHRSKwVtPHKRddZrjMOTCsgiAYd/e4f54w+NN3T3c/zND1nnverqquf8zu3b//mTH/6/M55zpOqQpIk7fx+Z7ELkCRJc2NoS5LUEYa2JEkdYWhLktQRhrYkSR1haEuS1BGGtgQkWZGkkizpc/1XJvlKz/RPkzx1SLW9Pcm5w6hzmm0/ua11t2Fsb8q2h9YHO0MNSW5N8pJhbEvql6GtRbcYvwwXep9V9biqunmWGo5IsmkO23pPVf3ZMOqa+nNX1e1trb8axvZ7zaUPpqnvD9qg/WmSB9o/UH7a8/Xkha5hGNq6n/5I71e7vqH8tS5pYSRZUlXbFruOR0pV/QvwOGhGFYBbgKXT9cGo9Y0EnmlrJ5bkd5KcnuT/JflxkouT7N3O2z5MvCbJ7Ul+lOQdPes+Jsn6JPckuSnJadvPapN8HHgy8L/bs7fTenb7sum2N01t+yTZkOS+JF8DnjZl/m/OtJIck+TGJPcn2ZzkrUkeC/wjcEDPWeQBSc5KckmSf0hyH/DKtu0fppTw6iQ/TLIlyVt79ntBkv/WM/2bs/npfu6pw+1tDRuS3J1kY5L/1LOts9p/g4+1P8sNScZ30Ee9fXBBko8kuaJd99okT5tp3Rm2N13fHJrkq0nubfviw0l2H0YNSV6e5Lb22HvHlHkz7jfJ1e1i3277+U+S7JXkM0km22PyM0mWz+fnl8DQ1s7t9cDxwL8DDgDuAT4yZZkXAAcBRwLvTPJv2vYzgRXAU4E/BP50+wpV9XLgduA/tMOn/30O25vqI8C/AvsDr26/ZnIe8J+r6vHAs4EvVtUDwNHAD9saHldVP2yXXw1cAiwFLpxhmy8CVgF/BLwtcxjqn+Xn3u4TwCaa/j4BeE+SF/fMf2m7zFJgA/Dh2fbb4yTgXcBewEbg3fNYd7upffMr4M3AvsDzaf7d/nzQGpIcDJwDvJymL/YBekN2xv1W1QvbZZ7T9vMnaX7XfhR4Cs0fTj9nfn0nAYa2dm6vAd5RVZuq6kHgLOCEPPQmrHdV1c+r6tvAt4HntO0nAu+pqnuqahPwoTnuc6bt/Uaam7b+GHhnVT1QVdcD63ewzV8CByd5QlvPN2ap4atV9emq+nVV/XwHdT5QVd+lCYOTZ/3JZpHkQOBw4G1V9a9V9S3gXOAVPYt9pao+214D/zjT9M8OXFZVX2uHtC8Efr+PMh/SN1V1XVVdU1XbqupW4O9o/sgbtIYTgM9U1dXtsfeXwK+3z5zvfqvqx1X1qar6WVXdT/PHwo7qlKZlaGtn9hTgsnYI8l7gJpoznP16lrmz5/XPaK+H0pwd3dEzr/f1jsy0vV5jNPeD9G7zth1s84+BY4Dbknw5yfNnqWEutU7d9wFzWGc2BwB3t6HSu+1lPdNT+2ePzP1O9rn07Wwe0jdJntEONd/ZDpm/h+bsd9AaHnL8tCMjP+53v0n2TPJ37XD7fcDVwNIswF372rUZ2tqZ3QEcXVVLe772qKrNc1h3Cw8dzjxwyvxBPt5uEtg2ZZsz3tVcVV+vqtXAk4BPAxfPUsNcapu67+1D6w8Ae/bM+915bPuHwN5JHj9l23Pp70fK1PrPAb4HrKqqJwBvBzKE/Wyhp4+T7EkzRN7vft9Cc9nlee3y24fQh1GrRoihrZ3Fo5Ls0fO1BPifwLuTPAUgyViS1XPc3sXAGe0NQMuA102ZfxfN9e55a4eGLwXOas+gDgbWTLdskt2TvCzJE6vql8B9/HaY9S5gnyRP7KOMv2z3/SzgVcAn2/ZvAcck2TvJ7wJvmrLejD93Vd0B/B/gr9t/g98DTgGm3gS3M3k8TZ/+NMkzgdcOabuXAMcleUF7g9lf8dDfl7Ptd2o/P57mOva9aW6mPHNIdWrEGNraWXyW5pfa9q+zgA/S3Oz0+ST3A9cAz5vj9v6K5oaqW4Av0PwSfrBn/l8D/7Uden/rNOvP5nU0Q6t3AhfQXFeeycuBW9th0dcALwOoqu8BFwE3t3XMZ4j7yzQ3Ul0FvK+qPt+2f5zmWvytwOf5bZhvN9vPfTLNDXw/BC4DzqyqL8yjrkfaW4H/CNwP/D0P/3n7UlU3AKcC/4vmrPsemuNprvs9C1jf9vOJwAeAxwA/ojmOPzeMOjV6UjXIKKHUDUleC5xUVd78I6mzPNPWLinJ/kkOT/Ne74Noriletth1SdIgfCKadlW707wNZyVwL817i/92USuSpAE5PC5JUkc4PC5JUkfsFMPj++67b61YsWKxy5Ak6RFz3XXX/aiqxuazzqyhneR84Dhga1U9u6f99TRvifgVcEVVnda2n0Hz3s5fAW+oqn+abR8rVqxgYmJiPnVLktRpSXb0JMVpzeVM+wKaB9t/rGdHL6J5cP9zqurBJE9q2w+meSD/s2geA/iFJM9YiM/qlSRp1Mx6TbuqrgbuntL8WuDs9kH6VNXWtn018ImqerCqbqF5+MOhQ6xXkqSR1e+NaM8A/qD9PNovJ/m3bfsyHvpA/0089MMGfiPJ2iQTSSYmJyf7LEOSpNHRb2gvAfYGDgP+C3Bxknk9+L6q1lXVeFWNj43N6zq8JEkjqd/Q3gRcWo2v0XwAwr40nwbU++lDy9m5PiFIkqTO6je0Pw28CJrPlaV5+tSPaD7c4aQkj06yElgFfG0YhUqSNOrm8pavi4AjgH2TbKL5SLnzgfOTXA/8AlhTzaPVbkhyMXAjzecNn+qd45IkDcdO8RjT8fHx8n3akqRRkuS6qhqfzzo+xlSSpI7YKR5j2gUrTr9iqNu79exjh7o9SdKuzzNtSZI6wtCWJKkjDG1JkjrC0JYkqSMMbUmSOsLQliSpIwxtSZI6wtCWJKkjDG1JkjrC0JYkqSMMbUmSOsJnjy8Sn2UuSZovz7QlSeoIQ1uSpI4wtCVJ6ghDW5Kkjpg1tJOcn2RrkuunmfeWJJVk33Y6ST6UZGOS7yQ5ZCGKliRpFM3lTPsC4KipjUkOBP4IuL2n+WhgVfu1Fjhn8BIlSRLMIbSr6mrg7mlmvR84DaiettXAx6pxDbA0yf5DqVSSpBHX1zXtJKuBzVX17SmzlgF39Exvatum28baJBNJJiYnJ/spQ5KkkTLv0E6yJ/B24J2D7Liq1lXVeFWNj42NDbIpSZJGQj9PRHsasBL4dhKA5cA3khwKbAYO7Fl2edsmSZIGNO8z7ar6blU9qapWVNUKmiHwQ6rqTmAD8Ir2LvLDgJ9U1ZbhlixJ0miay1u+LgK+ChyUZFOSU3aw+GeBm4GNwN8Dfz6UKiVJ0uzD41V18izzV/S8LuDUwcuSJElT+UQ0SZI6wtCWJKkjDG1JkjrC0JYkqSMMbUmSOsLQliSpIwxtSZI6wtCWJKkjDG1JkjrC0JYkqSMMbUmSOsLQliSpIwxtSZI6wtCWJKkjDG1JkjrC0JYkqSMMbUmSOmLW0E5yfpKtSa7vafsfSb6X5DtJLkuytGfeGUk2Jvl+kn+/UIVLkjRq5nKmfQFw1JS2K4FnV9XvAf8XOAMgycHAScCz2nX+NsluQ6tWkqQRNmtoV9XVwN1T2j5fVdvayWuA5e3r1cAnqurBqroF2AgcOsR6JUkaWcO4pv1q4B/b18uAO3rmbWrbHibJ2iQTSSYmJyeHUIYkSbu2gUI7yTuAbcCF8123qtZV1XhVjY+NjQ1ShiRJI2FJvysmeSVwHHBkVVXbvBk4sGex5W2bJEkaUF9n2kmOAk4DXlpVP+uZtQE4Kcmjk6wEVgFfG7xMSZI065l2kouAI4B9k2wCzqS5W/zRwJVJAK6pqtdU1Q1JLgZupBk2P7WqfrVQxUuSNEpmDe2qOnma5vN2sPy7gXcPUpQkSXo4n4gmSVJHGNqSJHWEoS1JUkcY2pIkdUTf79PWzmXF6VcMfZu3nn3s0LcpSeqfoa0ZDfsPAf8IkKTBODwuSVJHGNqSJHWEoS1JUkcY2pIkdYShLUlSRxjakiR1hKEtSVJHGNqSJHWEoS1JUkcY2pIkdYShLUlSRxjakiR1xKyhneT8JFuTXN/TtneSK5P8oP2+V9ueJB9KsjHJd5IcspDFS5I0SuZypn0BcNSUttOBq6pqFXBVOw1wNLCq/VoLnDOcMiVJ0qyhXVVXA3dPaV4NrG9frweO72n/WDWuAZYm2X9YxUqSNMr6vaa9X1VtaV/fCezXvl4G3NGz3Ka27WGSrE0ykWRicnKyzzIkSRodA9+IVlUFVB/rrauq8aoaHxsbG7QMSZJ2ef2G9l3bh73b71vb9s3AgT3LLW/bJEnSgPoN7Q3Amvb1GuDynvZXtHeRHwb8pGcYXZIkDWDJbAskuQg4Atg3ySbgTOBs4OIkpwC3ASe2i38WOAbYCPwMeNUC1CxJ0kiaNbSr6uQZZh05zbIFnDpoUZIk6eF8IpokSR1haEuS1BGGtiRJHWFoS5LUEYa2JEkdYWhLktQRhrYkSR1haEuS1BGGtiRJHWFoS5LUEYa2JEkdYWhLktQRhrYkSR1haEuS1BGGtiRJHWFoS5LUEUsWuwCNjhWnXzHU7d169rFD3Z4k7ew805YkqSMGCu0kb05yQ5Lrk1yUZI8kK5Ncm2Rjkk8m2X1YxUqSNMr6Du0ky4A3AONV9WxgN+Ak4L3A+6vq6cA9wCnDKFSSpFE36PD4EuAxSZYAewJbgBcDl7Tz1wPHD7gPSZLEAKFdVZuB9wG304T1T4DrgHuralu72CZg2XTrJ1mbZCLJxOTkZL9lSJI0MgYZHt8LWA2sBA4AHgscNdf1q2pdVY1X1fjY2Fi/ZUiSNDIGGR5/CXBLVU1W1S+BS4HDgaXtcDnAcmDzgDVKkiQGC+3bgcOS7JkkwJHAjcCXgBPaZdYAlw9WoiRJgsGuaV9Lc8PZN4DvtttaB7wN+IskG4F9gPOGUKckSSNvoCeiVdWZwJlTmm8GDh1ku5Ik6eF8IpokSR1haEuS1BGGtiRJHWFoS5LUEYa2JEkdYWhLktQRhrYkSR1haEuS1BGGtiRJHWFoS5LUEYa2JEkdYWhLktQRhrYkSR1haEuS1BGGtiRJHWFoS5LUEYa2JEkdMVBoJ1ma5JIk30tyU5LnJ9k7yZVJftB+32tYxUqSNMoGPdP+IPC5qnom8BzgJuB04KqqWgVc1U5LkqQB9R3aSZ4IvBA4D6CqflFV9wKrgfXtYuuB4wctUpIkDXamvRKYBD6a5JtJzk3yWGC/qtrSLnMnsN+gRUqSpMFCewlwCHBOVT0XeIApQ+FVVUBNt3KStUkmkkxMTk4OUIYkSaNhkNDeBGyqqmvb6UtoQvyuJPsDtN+3TrdyVa2rqvGqGh8bGxugDEmSRkPfoV1VdwJ3JDmobToSuBHYAKxp29YAlw9UoSRJApoh7kG8Hrgwye7AzcCraP4QuDjJKcBtwIkD7kOSJDFgaFfVt4DxaWYdOch2JUnSw/lENEmSOsLQliSpIwxtSZI6wtCWJKkjBr17fKe04vQrFrsESZKGzjNtSZI6wtCWJKkjdsnhcY2GhbgMcuvZxw59m5I0LJ5pS5LUEYa2JEkdYWhLktQRhrYkSR1haEuS1BGGtiRJHWFoS5LUEYa2JEkdYWhLktQRPhFN6jHsp6z5hDVJwzTwmXaS3ZJ8M8ln2umVSa5NsjHJJ5PsPniZkiRpGMPjbwRu6pl+L/D+qno6cA9wyhD2IUnSyBsotJMsB44Fzm2nA7wYuKRdZD1w/CD7kCRJjUHPtD8AnAb8up3eB7i3qra105uAZdOtmGRtkokkE5OTkwOWIUnSrq/v0E5yHLC1qq7rZ/2qWldV41U1PjY21m8ZkiSNjEHuHj8ceGmSY4A9gCcAHwSWJlnSnm0vBzYPXqYkSer7TLuqzqiq5VW1AjgJ+GJVvQz4EnBCu9ga4PKBq5QkSQvycJW3AX+RZCPNNe7zFmAfkiSNnKE8XKWq/hn45/b1zcChw9iuJEn6LR9jKklSRxjakiR1hKEtSVJHGNqSJHWEoS1JUkcY2pIkdYSfpy0tID+fW9IweaYtSVJHGNqSJHWEoS1JUkcY2pIkdYShLUlSRxjakiR1hKEtSVJHGNqSJHWEoS1JUkcY2pIkdYShLUlSR/Qd2kkOTPKlJDcmuSHJG9v2vZNcmeQH7fe9hleuJEmja5Az7W3AW6rqYOAw4NQkBwOnA1dV1SrgqnZakiQNqO/QrqotVfWN9vX9wE3AMmA1sL5dbD1w/KBFSpKkIV3TTrICeC5wLbBfVW1pZ90J7DfDOmuTTCSZmJycHEYZkiTt0gb+PO0kjwM+Bbypqu5L8pt5VVVJarr1qmodsA5gfHx82mUkPZSfzy2NtoHOtJM8iiawL6yqS9vmu5Ls387fH9g6WImSJAkGu3s8wHnATVX1Nz2zNgBr2tdrgMv7L0+SJG03yPD44cDLge8m+Vbb9nbgbODiJKcAtwEnDlaiJEmCAUK7qr4CZIbZR/a7XUmSND2fiCZJUkcY2pIkdYShLUlSRxjakiR1hKEtSVJHGNqSJHWEoS1JUkcY2pIkdYShLUlSRwz8KV+SumvYnxoGfnKYtJA805YkqSM805Y0VH7mt7RwPNOWJKkjPNOWNFK8jq8uM7Ql7dQWImSHzUsCeqQ4PC5JUkcY2pIkdYShLUlSRyzYNe0kRwEfBHYDzq2qsxdqX5Ik9dpV7zNYkNBOshvwEeAPgU3A15NsqKobF2J/kqSZ7aoBNooWanj8UGBjVd1cVb8APgGsXqB9SZI0EhZqeHwZcEfP9Cbgeb0LJFkLrG0nf5rk+wPsb1/gRwOsr+nZr8Nnny6MXapf897FrgDYQZ/uJPU9oob0M0/t06fMdwOL9j7tqloHrBvGtpJMVNX4MLal37Jfh88+XRj26/DZp8M3jD5dqOHxzcCBPdPL2zZJktSnhQrtrwOrkqxMsjtwErBhgfYlSdJIWJDh8araluR1wD/RvOXr/Kq6YSH21RrKMLsexn4dPvt0Ydivw2efDt/AfZqqGkYhkiRpgflENEmSOsLQliSpIzof2kmOSvL9JBuTnL7Y9XRRkgOTfCnJjUluSPLGtn3vJFcm+UH7fa/FrrVrkuyW5JtJPtNOr0xybXu8frK9UVPzkGRpkkuSfC/JTUme77E6mCRvbv/vX5/koiR7eKzOX5Lzk2xNcn1P27THZhofavv3O0kOmcs+Oh3aPY9LPRo4GDg5ycGLW1UnbQPeUlUHA4cBp7b9eDpwVVWtAq5qpzU/bwRu6pl+L/D+qno6cA9wyqJU1W0fBD5XVc8EnkPTvx6rfUqyDHgDMF5Vz6a5efgkPFb7cQFw1JS2mY7No4FV7dda4Jy57KDToY2PSx2KqtpSVd9oX99P80twGU1frm8XWw8cvzgVdlOS5cCxwLntdIAXA5e0i9in85TkicALgfMAquoXVXUvHquDWgI8JskSYE9gCx6r81ZVVwN3T2me6dhcDXysGtcAS5PsP9s+uh7a0z0uddki1bJLSLICeC5wLbBfVW1pZ90J7LdIZXXVB4DTgF+30/sA91bVtnba43X+VgKTwEfbyw7nJnksHqt9q6rNwPuA22nC+ifAdXisDstMx2Zf+dX10NYQJXkc8CngTVV1X++8at4b6PsD5yjJccDWqrpusWvZxSwBDgHOqarnAg8wZSjcY3V+2musq2n+IDoAeCwPH+LVEAzj2Ox6aPu41CFJ8iiawL6wqi5tm+/aPlzTft+6WPV10OHAS5PcSnPZ5sU012KXtkOQ4PHaj03Apqq6tp2+hCbEPVb79xLglqqarKpfApfSHL8eq8Mx07HZV351PbR9XOoQtNdazwNuqqq/6Zm1AVjTvl4DXP5I19ZVVXVGVS2vqhU0x+UXq+plwJeAE9rF7NN5qqo7gTuSHNQ2HQnciMfqIG4HDkuyZ/u7YHufeqwOx0zH5gbgFe1d5IcBP+kZRp9R55+IluQYmmuH2x+X+u5FLqlzkrwA+Bfgu/z2+uvbaa5rXww8GbgNOLGqpt5koVkkOQJ4a1Udl+SpNGfeewPfBP60qh5czPq6Jsnv09zctztwM/AqmhMQj9U+JXkX8Cc07yT5JvBnNNdXPVbnIclFwBE0H8F5F3Am8GmmOTbbP5A+THMp4mfAq6pqYtZ9dD20JUkaFV0fHpckaWQY2pIkdYShLUlSRxjakiR1hKEtSVJHGNqSJHWEoS1JUkf8f+03S/+8DyNBAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 576x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yi28N2RBcMc5"
      },
      "source": [
        "Для обучения `BiLSTM` лучше использовать colab"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LAGSrqWsxok2",
        "scrolled": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "bdf5355a-844f-47ac-87e3-9cd8e5192f96"
      },
      "source": [
        "import torch\n",
        "from torch import nn\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim\n",
        "\n",
        "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
        "device"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "device(type='cuda')"
            ]
          },
          "metadata": {},
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2DSWm0UjcMc-"
      },
      "source": [
        "Для более быстрого и устойчивого обучения сгруппируем наши данные по батчам"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dmwAyhNgxok_"
      },
      "source": [
        "# бьем нашу выборку на батч, не забывая сначала отсортировать выборку по длине\n",
        "def _len_sort_key(x):\n",
        "    return len(x.words)\n",
        "\n",
        "BATCH_SIZE = 32\n",
        "\n",
        "train_iterator, valid_iterator, test_iterator = BucketIterator.splits(\n",
        "    (train_data, valid_data, test_data), \n",
        "    batch_size = BATCH_SIZE, \n",
        "    device = device,\n",
        "    sort_key=_len_sort_key\n",
        ")"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6aTjW00nxolI",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "bef766e4-1f1b-4a5f-aee8-45ea14d2e575"
      },
      "source": [
        "# посмотрим  на количество батчей\n",
        "list(map(len, [train_iterator, valid_iterator, test_iterator]))"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[27, 4, 4]"
            ]
          },
          "metadata": {},
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "vqFM_3NFdzGf",
        "outputId": "dfbf0836-eceb-4267-8dda-297ade5af726"
      },
      "source": [
        "len(WORD.vocab)"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "732"
            ]
          },
          "metadata": {},
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zyLQsizhcMdI"
      },
      "source": [
        "### Модель и её обучение"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-i9oHzcrcMdJ"
      },
      "source": [
        "Инициализируем нашу модель"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ff7BLWs_xolS",
        "scrolled": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3fabdc45-a814-41eb-d90a-187064ecb438"
      },
      "source": [
        "class LSTMTagger(nn.Module):\n",
        "\n",
        "    def __init__(self, input_dim, emb_dim, hid_dim, output_dim, dropout, bidirectional=False):\n",
        "        super().__init__()\n",
        "        \n",
        "        self.bidirectional = bidirectional\n",
        "        self.embeddings = nn.Embedding(input_dim, emb_dim)\n",
        "        self.dropout = nn.Dropout(p=dropout)\n",
        "\n",
        "        self.rnn = nn.LSTM(emb_dim, hid_dim, bidirectional=bidirectional)\n",
        "        \n",
        "        # если bidirectional, то предсказываем на основе конкатенации двух hidden\n",
        "        self.tag = nn.Linear((1 + bidirectional) * hid_dim, output_dim)\n",
        "\n",
        "    def forward(self, sent):\n",
        "        \n",
        "        #sent = [sent len, batch size] \n",
        "        \n",
        "        # не забываем применить dropout к embedding\n",
        "        embedded = self.embeddings(sent)\n",
        "\n",
        "        output, _ = self.rnn(embedded)\n",
        "        #output = [sent len, batch size, hid dim * n directions]\n",
        "        output = self.dropout(output)        \n",
        "        prediction = self.tag(output)    \n",
        "        return prediction\n",
        "        \n",
        "# параметры модели\n",
        "INPUT_DIM = len(WORD.vocab)\n",
        "OUTPUT_DIM = len(TAG.vocab)\n",
        "EMB_DIM = 300\n",
        "HID_DIM = 256\n",
        "DROPOUT = 0.2\n",
        "BIDIRECTIONAL = False\n",
        "\n",
        "model = LSTMTagger(INPUT_DIM, EMB_DIM, HID_DIM, OUTPUT_DIM, DROPOUT, BIDIRECTIONAL).to(device)\n",
        "\n",
        "# инициализируем веса\n",
        "def init_weights(m):\n",
        "    for name, param in m.named_parameters():\n",
        "        nn.init.uniform_(param, -0.08, 0.08)\n",
        "        \n",
        "model.apply(init_weights)"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "LSTMTagger(\n",
              "  (embeddings): Embedding(732, 300)\n",
              "  (dropout): Dropout(p=0.2, inplace=False)\n",
              "  (rnn): LSTM(300, 256)\n",
              "  (tag): Linear(in_features=256, out_features=13, bias=True)\n",
              ")"
            ]
          },
          "metadata": {},
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EJLqq8IHcMdQ"
      },
      "source": [
        "Подсчитаем количество обучаемых параметров нашей модели"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_Auu53Kdxolm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "bbedecc6-b45a-4709-f78b-49a57e4b0d9e"
      },
      "source": [
        "def count_parameters(model):\n",
        "    parameters_count = 0\n",
        "    for param in model.parameters():\n",
        "        temp = 1\n",
        "        for j in range(len(param.shape)):\n",
        "            temp *= param.shape[j]\n",
        "\n",
        "        parameters_count += temp\n",
        "        print(param.shape)\n",
        "    return parameters_count\n",
        "\n",
        "print(f'The model has {count_parameters(model):,} trainable parameters')"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "torch.Size([732, 300])\n",
            "torch.Size([1024, 300])\n",
            "torch.Size([1024, 256])\n",
            "torch.Size([1024])\n",
            "torch.Size([1024])\n",
            "torch.Size([13, 256])\n",
            "torch.Size([13])\n",
            "The model has 794,333 trainable parameters\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "w0p6pr86d8bH",
        "outputId": "841541fe-8f31-47ec-b11c-9735e73d02cf"
      },
      "source": [
        "# параметры в случае всего датасета\n",
        "print(f'The model has {count_parameters(model):,} trainable parameters')"
      ],
      "execution_count": 117,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "torch.Size([724, 300])\n",
            "torch.Size([1024, 300])\n",
            "torch.Size([1024, 256])\n",
            "torch.Size([1024])\n",
            "torch.Size([1024])\n",
            "torch.Size([13, 256])\n",
            "torch.Size([13])\n",
            "The model has 791,933 trainable parameters\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oSBfvf9HcMd9"
      },
      "source": [
        "Погнали обучать"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AjD1Y7Rmxolu",
        "scrolled": true
      },
      "source": [
        "PAD_IDX = TAG.vocab.stoi['<pad>']\n",
        "optimizer = optim.Adam(model.parameters(), lr=4e-4)\n",
        "criterion = nn.CrossEntropyLoss(ignore_index = PAD_IDX)\n",
        "\n",
        "def train(model, iterator, optimizer, criterion, clip, train_history=None, valid_history=None):\n",
        "    model.train()\n",
        "    \n",
        "    epoch_loss = 0\n",
        "    history = []\n",
        "    for i, batch in enumerate(iterator):\n",
        "        tags = batch.tags\n",
        "        words = batch.words\n",
        "        optimizer.zero_grad()\n",
        "        output = model(words)\n",
        "        tags = tags.view(-1)\n",
        "        output = output.view(output.shape[0]*output.shape[1],-1)\n",
        "        loss = criterion(output, tags)\n",
        "        epoch_loss += loss\n",
        "        #tags = [sent len, batch size]\n",
        "        #output = [sent len, batch size, output dim]\n",
        "        \n",
        "        \n",
        "        #tags = [sent len * batch size]\n",
        "        #output = [sent len * batch size, output dim]\n",
        "        \n",
        "        loss.backward()\n",
        "        \n",
        "        # Gradient clipping(решение проблемы взрыва граденты), clip - максимальная норма вектора\n",
        "        torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=clip)\n",
        "        \n",
        "        optimizer.step()\n",
        "        \n",
        "        epoch_loss += loss.item()\n",
        "        \n",
        "        history.append(loss.cpu().data.numpy())\n",
        "        if (i+1)%10==0:\n",
        "            fig, ax = plt.subplots(nrows=1, ncols=2, figsize=(12, 8))\n",
        "\n",
        "            clear_output(True)\n",
        "            ax[0].plot(history, label='train loss')\n",
        "            ax[0].set_xlabel('Batch')\n",
        "            ax[0].set_title('Train loss')\n",
        "            \n",
        "            if train_history is not None:\n",
        "                ax[1].plot(train_history, label='general train history')\n",
        "                ax[1].set_xlabel('Epoch')\n",
        "            if valid_history is not None:\n",
        "                ax[1].plot(valid_history, label='general valid history')\n",
        "            plt.legend()\n",
        "            \n",
        "            plt.show()\n",
        "\n",
        "        \n",
        "    return epoch_loss / len(iterator)\n",
        "\n",
        "def evaluate(model, iterator, criterion):\n",
        "    model.eval()\n",
        "    \n",
        "    epoch_loss = 0\n",
        "    \n",
        "    history = []\n",
        "    \n",
        "    with torch.no_grad():\n",
        "    \n",
        "        for i, batch in enumerate(iterator):\n",
        "            words = batch.words\n",
        "            tags = batch.tags\n",
        "            output = model(words)\n",
        "\n",
        "            #tags = [sent len, batch size]\n",
        "            #output = [sent len, batch size, output dim]\n",
        "\n",
        "            output = output.view(output.shape[0]*output.shape[1],-1)\n",
        "            tags = tags.view(-1)\n",
        "\n",
        "            #tags = [sent len * batch size]\n",
        "            #output = [sent len * batch size, output dim]\n",
        "\n",
        "            loss = criterion(output, tags)\n",
        "            \n",
        "            epoch_loss += loss.item()\n",
        "        \n",
        "    return epoch_loss / len(iterator)\n",
        "\n",
        "def epoch_time(start_time, end_time):\n",
        "    elapsed_time = end_time - start_time\n",
        "    elapsed_mins = int(elapsed_time / 60)\n",
        "    elapsed_secs = int(elapsed_time - (elapsed_mins * 60))\n",
        "    return elapsed_mins, elapsed_secs"
      ],
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TJdXIyTHxol2",
        "scrolled": false
      },
      "source": [
        "import time\n",
        "import math\n",
        "import matplotlib\n",
        "matplotlib.rcParams.update({'figure.figsize': (16, 12), 'font.size': 14})\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline\n",
        "from IPython.display import clear_output\n",
        "\n",
        "\n",
        "def train_model(model, epochs):\n",
        "    train_history = []\n",
        "    valid_history = []\n",
        "\n",
        "    N_EPOCHS = epochs\n",
        "    CLIP = 3\n",
        "\n",
        "    best_valid_loss = float('inf')\n",
        "\n",
        "    for epoch in range(N_EPOCHS):\n",
        "\n",
        "        start_time = time.time()\n",
        "\n",
        "        train_loss = train(model, train_iterator, optimizer, criterion, CLIP, train_history, valid_history)\n",
        "        valid_loss = evaluate(model, valid_iterator, criterion)\n",
        "\n",
        "        end_time = time.time()\n",
        "\n",
        "        epoch_mins, epoch_secs = epoch_time(start_time, end_time)\n",
        "\n",
        "        if valid_loss < best_valid_loss:\n",
        "            best_valid_loss = valid_loss\n",
        "            torch.save(model.state_dict(), 'best-val-model.pt')\n",
        "\n",
        "        train_history.append(train_loss)\n",
        "        valid_history.append(valid_loss)\n",
        "        print(f'Epoch: {epoch+1:02} | Time: {epoch_mins}m {epoch_secs}s')\n",
        "        print(f'\\tTrain Loss: {train_loss:.3f} | Train PPL: {math.exp(train_loss):7.3f}')\n",
        "        print(f'\\t Val. Loss: {valid_loss:.3f} |  Val. PPL: {math.exp(valid_loss):7.3f}')"
      ],
      "execution_count": 19,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 563
        },
        "id": "5h2ad4qliNxB",
        "outputId": "86487de9-c9b1-4da3-f327-6a4b5f3d4a6b"
      },
      "source": [
        "train_model(model, 35)"
      ],
      "execution_count": 120,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch: 35 | Time: 0m 1s\n",
            "\tTrain Loss: 0.318 | Train PPL:   1.375\n",
            "\t Val. Loss: 0.384 |  Val. PPL:   1.469\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Fr860UPacMeI"
      },
      "source": [
        "### Применение модели"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5sDAfAq9xol9"
      },
      "source": [
        "def accuracy_model(model, iterator):\n",
        "    model.eval()\n",
        "    \n",
        "    true_pred = 0\n",
        "    num_pred = 0\n",
        "    \n",
        "    with torch.no_grad():\n",
        "        for i, batch in enumerate(iterator):\n",
        "\n",
        "            words = batch.words\n",
        "            tags = batch.tags\n",
        "\n",
        "            output = model(words)\n",
        "            \n",
        "            #output = [sent len, batch size, output dim]\n",
        "            output = output.argmax(dim=2)\n",
        "            \n",
        "            #output = [sent len, batch size]\n",
        "            predict_tags = output.cpu().numpy()\n",
        "            true_tags = tags.cpu().numpy()\n",
        "\n",
        "            true_pred += np.sum((true_tags == predict_tags) & (true_tags != PAD_IDX))\n",
        "            num_pred += np.prod(true_tags.shape) - (true_tags == PAD_IDX).sum()\n",
        "        \n",
        "    return round(true_pred / num_pred * 100, 3)"
      ],
      "execution_count": 20,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V2n0H85mxomE",
        "scrolled": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6132befd-b446-4624-f687-bce2721b724c"
      },
      "source": [
        "print(\"Accuracy:\", accuracy_model(model, test_iterator), '%')"
      ],
      "execution_count": 122,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Accuracy: 85.259 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FacTKSPJcMeP"
      },
      "source": [
        "Вы можете улучшить качество, изменяя параметры модели. Но чтобы добиться нужного качества, вам неообходимо взять все выборку, а не только категорию `humor`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QXqXg0gbcMeR",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 435
        },
        "outputId": "b7d66d1a-53a2-4826-e9ce-66cb5f8be163"
      },
      "source": [
        "\n",
        "brown_tagged_sents = brown.tagged_sents(tagset=\"universal\")\n",
        "\n",
        "pos_data = [list(zip(*sent)) for sent in brown_tagged_sents]\n",
        "print(pos_data[0])\n",
        "\n",
        "# создаем примеры\n",
        "examples = []\n",
        "for words, tags in pos_data:\n",
        "    examples.append(torchtext.legacy.data.Example.fromlist([list(words), list(tags)], fields=[('words', WORD), ('tags', TAG)]))\n",
        "\n",
        "# кладем примеры в наш датасет\n",
        "dataset = torchtext.legacy.data.Dataset(examples, fields=[('words', WORD), ('tags', TAG)])\n",
        "\n",
        "train_data, valid_data, test_data = dataset.split(split_ratio=[0.8, 0.1, 0.1])\n",
        "\n",
        "print(f\"Number of training examples: {len(train_data.examples)}\")\n",
        "print(f\"Number of validation examples: {len(valid_data.examples)}\")\n",
        "print(f\"Number of testing examples: {len(test_data.examples)}\")\n",
        "\n",
        "\n",
        "WORD.build_vocab(train_data, min_freq=10)\n",
        "TAG.build_vocab(train_data)\n",
        "\n",
        "print(f\"Unique tokens in source (ru) vocabulary: {len(WORD.vocab)}\")\n",
        "print(f\"Unique tokens in target (en) vocabulary: {len(TAG.vocab)}\")\n",
        "\n",
        "print(WORD.vocab.itos[::200])\n",
        "print(TAG.vocab.itos)\n",
        "\n",
        "\n",
        "length = map(len, [vars(x)['words'] for x in train_data.examples])\n",
        "\n",
        "plt.figure(figsize=[8, 4])\n",
        "plt.title(\"Length distribution in Train data\")\n",
        "plt.hist(list(length), bins=20);\n",
        "\n",
        "\n",
        "INPUT_DIM = len(WORD.vocab)\n",
        "OUTPUT_DIM = len(TAG.vocab)\n",
        "EMB_DIM = 300\n",
        "HID_DIM = 256\n",
        "DROPOUT = 0.2\n",
        "BIDIRECTIONAL = False\n",
        "\n",
        "\n",
        "model_full = LSTMTagger(INPUT_DIM, EMB_DIM, HID_DIM, OUTPUT_DIM, DROPOUT, BIDIRECTIONAL).to(device)\n",
        "PAD_IDX = TAG.vocab.stoi['<pad>']\n",
        "optimizer = optim.Adam(model_full.parameters(), lr=4e-4)\n",
        "criterion = nn.CrossEntropyLoss(ignore_index = PAD_IDX)"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[('The', 'Fulton', 'County', 'Grand', 'Jury', 'said', 'Friday', 'an', 'investigation', 'of', \"Atlanta's\", 'recent', 'primary', 'election', 'produced', '``', 'no', 'evidence', \"''\", 'that', 'any', 'irregularities', 'took', 'place', '.'), ('DET', 'NOUN', 'NOUN', 'ADJ', 'NOUN', 'VERB', 'NOUN', 'DET', 'NOUN', 'ADP', 'NOUN', 'ADJ', 'NOUN', 'NOUN', 'VERB', '.', 'DET', 'NOUN', '.', 'ADP', 'DET', 'NOUN', 'VERB', 'NOUN', '.')]\n",
            "Number of training examples: 45872\n",
            "Number of validation examples: 5734\n",
            "Number of testing examples: 5734\n",
            "Unique tokens in source (ru) vocabulary: 7298\n",
            "Unique tokens in target (en) vocabulary: 13\n",
            "['<unk>', 'course', 'office', 'report', 'c', 'sides', 'units', 'selected', 'ancient', 'phil', 'p.', 'trained', 'incident', 'brings', 'normally', 'turns', 'camera', 'testing', 'partially', 'clarity', 'orange', 'shirt', 'widow', 'anticipated', \"who's\", 'revolutionary', 'milling', 'dangers', 'startling', 'humble', 'aqueous', 'orbit', 'cambridge', 'nam', 'assault', 'ideals', 'skies']\n",
            "['<pad>', 'NOUN', 'VERB', '.', 'ADP', 'DET', 'ADJ', 'ADV', 'PRON', 'CONJ', 'PRT', 'NUM', 'X']\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 576x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tuRClho8dRGS"
      },
      "source": [
        "train_iterator, valid_iterator, test_iterator = BucketIterator.splits(\n",
        "    (train_data, valid_data, test_data), \n",
        "    batch_size = 128, \n",
        "    device = device,\n",
        "    sort_key=_len_sort_key\n",
        ")"
      ],
      "execution_count": 22,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 563
        },
        "id": "MTdWGd1DdetK",
        "outputId": "3b9e7d9e-a66b-416e-edb4-eef941cd0d11"
      },
      "source": [
        "train_model(model_full, 20)"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch: 20 | Time: 0m 29s\n",
            "\tTrain Loss: 0.131 | Train PPL:   1.139\n",
            "\t Val. Loss: 0.219 |  Val. PPL:   1.245\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nUiWyVnuf_tZ",
        "outputId": "e36ce3cb-043f-487b-a342-fd4ca4a6b892"
      },
      "source": [
        "print(\"Accuracy:\", accuracy_model(model_full, test_iterator), '%')\n",
        "#тут чуть чуть переобучившись"
      ],
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Accuracy: 93.767 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QlNw7_Gnnf4U"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gnpi2R6rcMeU"
      },
      "source": [
        "Вам неоходимо добиться качества не меньше, чем `accuracy = 93 %` "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TqD1lZuwxomK",
        "scrolled": true
      },
      "source": [
        "best_model = LSTMTagger(INPUT_DIM, EMB_DIM, HID_DIM, OUTPUT_DIM, DROPOUT, BIDIRECTIONAL).to(device)\n",
        "best_model.load_state_dict(torch.load('best-val-model.pt'))\n",
        "assert accuracy_model(best_model, test_iterator) >= 93"
      ],
      "execution_count": 25,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "SAI_UD1enidS",
        "outputId": "4db194b6-efdb-451b-b32f-c591ceb19727"
      },
      "source": [
        "print(\"Accuracy:\", accuracy_model(best_model, test_iterator), '%')"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Accuracy: 94.348 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nVfdJM-lcMeZ"
      },
      "source": [
        "Пример решение нашей задачи:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W3GUbwldxomW"
      },
      "source": [
        "def print_tags(model, data):\n",
        "    model.eval()\n",
        "    \n",
        "    with torch.no_grad():\n",
        "        words, _ = data\n",
        "        example = torch.LongTensor([WORD.vocab.stoi[elem] for elem in words]).unsqueeze(1).to(device)\n",
        "        \n",
        "        output = model(example).argmax(dim=-1).cpu().numpy()\n",
        "        tags = [TAG.vocab.itos[int(elem)] for elem in output]\n",
        "\n",
        "        for token, tag in zip(words, tags):\n",
        "            print(f'{token:15s}{tag}')"
      ],
      "execution_count": 31,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6djqgrxZoL3G"
      },
      "source": [
        "print_tags(best_model, pos_data[-1])"
      ],
      "execution_count": 32,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "heading_collapsed": true,
        "id": "zMIJDOBmwC6v"
      },
      "source": [
        "## Сравните результаты моделей HiddenMarkov, LstmTagger:\n",
        "* при обучение на маленькой части корпуса, например, на категории humor\n",
        "* при обучении на всем корпусе"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uUjNMJyPoSx6"
      },
      "source": [
        "Итак у нас есть результаты: \n",
        "LstmTagger на Humor: 85.259\n",
        "LstmTagger на всем датасете: 94.348\n",
        "\n",
        "HiddenMarkov на Humor: 88.82\n",
        "Осталось только запустить HiddenMarkov на всем, это сделаем ниже для удобства\n",
        "Cпойлер: ниже мы узнаем что HiddenMarkov получила 96.26 на всем датасете"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uDdsG2AjO-sp",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "aa857aec-7ba3-4366-90db-d3d9bb39678d"
      },
      "source": [
        "brown_tagged_sents = brown.tagged_sents(tagset=\"universal\")\n",
        "# Приведем слова к нижнему регистру\n",
        "my_brown_tagged_sents = []\n",
        "for sent in brown_tagged_sents:\n",
        "    my_brown_tagged_sents.append(list(map(lambda x: (x[0].lower(), x[1]), sent)))\n",
        "my_brown_tagged_sents = np.array(my_brown_tagged_sents)\n",
        "\n",
        "from sklearn.model_selection import train_test_split\n",
        "train_sents, test_sents = train_test_split(my_brown_tagged_sents, random_state=0, test_size=0.1)"
      ],
      "execution_count": 34,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:6: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray\n",
            "  \n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2DpBHq4TpYl_",
        "outputId": "c17fc7b6-b582-487e-fe16-b3bf6dce9304"
      },
      "source": [
        "HMM = HiddenMarkovModel()\n",
        "HMM.fit(train_sents)"
      ],
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<__main__.HiddenMarkovModel at 0x7f74ce5c5590>"
            ]
          },
          "metadata": {},
          "execution_count": 35
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "focRIps5pgfx",
        "outputId": "3925c4ce-1883-4e4c-8aeb-3e4a50f86a67"
      },
      "source": [
        "predictions = HMM.predict(test_sents)"
      ],
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Index(['.', 'ADJ', 'ADP', 'ADV', 'CONJ', 'DET', 'NOUN', 'NUM', 'PRON', 'PRT',\n",
            "       'VERB', 'X'],\n",
            "      dtype='object')\n",
            "words  Index(['the', ',', '.', 'of', 'and', 'to', 'a', 'in', 'that', 'is',\n",
            "       ...\n",
            "       'second-stage', 'comas', 'electroshocks', 'hillsdale', 'oilman-rancher',\n",
            "       'scolatti', 'mcneil', 'ja', '11-year', '11,744'],\n",
            "      dtype='object', length=47597)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w6udOSmjpvWo"
      },
      "source": [
        "def accuracy_score(model, sents, predicts):\n",
        "    true_pred = 0\n",
        "    num_pred = 0\n",
        "\n",
        "    for i, sent in enumerate(sents):\n",
        "        for j, word_pair in enumerate(sent):\n",
        "            tag = word_pair[1]\n",
        "            word = word_pair[0]\n",
        "            if tag == predicts[i][j]:\n",
        "                true_pred += 1\n",
        "            num_pred += 1\n",
        "    \n",
        "    print(\"Accuracy:\", true_pred / num_pred * 100, '%')"
      ],
      "execution_count": 38,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "h0VeHh_hpv1R",
        "outputId": "4a658243-f268-44e1-a5e5-191444c22745"
      },
      "source": [
        "accuracy_score(HMM, test_sents, predictions)"
      ],
      "execution_count": 39,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Accuracy: 96.26295331104619 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ROAGQMo_45mJ",
        "outputId": "555fb383-6d36-48de-a5c8-d5f9985ff412"
      },
      "source": [
        "# проверим напоследок \n",
        "sents = [['He', 'can', 'stay'], ['a', 'cat', 'and', 'a', 'dog'], ['I', 'have', 'a', 'television'],\n",
        "         ['My', 'favourite', 'character']]\n",
        "HMM.predict(sents)"
      ],
      "execution_count": 40,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Index(['.', 'ADJ', 'ADP', 'ADV', 'CONJ', 'DET', 'NOUN', 'NUM', 'PRON', 'PRT',\n",
            "       'VERB', 'X'],\n",
            "      dtype='object')\n",
            "words  Index(['the', ',', '.', 'of', 'and', 'to', 'a', 'in', 'that', 'is',\n",
            "       ...\n",
            "       'second-stage', 'comas', 'electroshocks', 'hillsdale', 'oilman-rancher',\n",
            "       'scolatti', 'mcneil', 'ja', '11-year', '11,744'],\n",
            "      dtype='object', length=47597)\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "OrderedDict([(0, array(['NOUN', 'NOUN', 'NOUN'], dtype='<U4')),\n",
              "             (1, array(['DET', 'NOUN', 'DET', 'DET', 'NOUN'], dtype='<U4')),\n",
              "             (2, array(['NOUN', 'NOUN', 'DET', 'NOUN'], dtype='<U4')),\n",
              "             (3, array(['NOUN', 'NOUN', 'NOUN'], dtype='<U4'))])"
            ]
          },
          "metadata": {},
          "execution_count": 40
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "e8d5PlnF5Uuh"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}