08_convert_raw_data.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import gzip\n",
    "import json\n",
    "import jsonlines\n",
    "from pprint import pprint\n",
    "from typing import List, Pattern, Optional\n",
    "import re\n",
    "import random\n",
    "from mltb2.text import clean_all_invisible_chars_and_whitespaces"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "LINE_PREFIX: Pattern = re.compile(r\"^[1-6][\\.)][ \\t]\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_clean_text_set = set()\n",
    "duplicate_clean_text_set = set()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def extract_content(content: str) -> Optional[List[str]]:\n",
    "    global all_clean_text_set, duplicate_clean_text_set\n",
    "\n",
    "    # split content by line breaks\n",
    "    content = content.replace(\"\\n\\n\", \"\\n\")\n",
    "    lines = content.split(\"\\n\")\n",
    "    assert len(lines) >= 6\n",
    "    lines = lines[:6]\n",
    "    for idx, line in enumerate(lines):\n",
    "        if re.search(LINE_PREFIX, line) is None:\n",
    "            return None\n",
    "        else:\n",
    "            # clean the text\n",
    "            clean_text = re.sub(LINE_PREFIX, \"\", line)\n",
    "            clean_text = clean_all_invisible_chars_and_whitespaces(clean_text)\n",
    "            lines[idx] = clean_text\n",
    "\n",
    "            # collect duplicates - we clean them later\n",
    "            if clean_text in all_clean_text_set:\n",
    "                duplicate_clean_text_set.add(clean_text)\n",
    "            else:\n",
    "                all_clean_text_set.add(clean_text)\n",
    "\n",
    "    return lines"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def select_data(data: dict):\n",
    "    result = data.copy()\n",
    "\n",
    "    # check finish_reason in llm_response_question\n",
    "    llm_response_question = result[\"llm_response_question\"]\n",
    "    if llm_response_question[\"finish_reason\"] != \"stop\":\n",
    "        return None\n",
    "\n",
    "    # check finish_reason in llm_response_imperative\n",
    "    llm_response_imperative = result[\"llm_response_imperative\"]\n",
    "    if llm_response_imperative[\"finish_reason\"] != \"stop\":\n",
    "        return None\n",
    "\n",
    "    # handle llm_response_question\n",
    "    del result[\"llm_response_question\"]\n",
    "    lines = extract_content(llm_response_question[\"content\"])\n",
    "    if lines is not None:\n",
    "        result[\"questions\"] = lines\n",
    "\n",
    "    # handle llm_response_imperative\n",
    "    del result[\"llm_response_imperative\"]\n",
    "    lines = extract_content(llm_response_imperative[\"content\"])\n",
    "    if lines is not None:\n",
    "        result[\"imperatives\"] = lines\n",
    "\n",
    "    # check if either questions or imperatives is available\n",
    "    if \"questions\" not in result and \"imperatives\" not in result:\n",
    "        return None\n",
    "\n",
    "    del result[\"_id\"]\n",
    "    del result[\"_key\"]\n",
    "    del result[\"_rev\"]\n",
    "    del result[\"context_token_count\"]\n",
    "    del result[\"meta_data_imperative\"]\n",
    "    del result[\"meta_data_question\"]\n",
    "    return result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "extracted_data = []\n",
    "\n",
    "with gzip.open(\"./data/wikipedia-22-12-de-raw-results.jsonl.gz\", \"r\") as gzip_in:\n",
    "    jsonlines_reader = jsonlines.Reader(gzip_in)\n",
    "    for data in jsonlines_reader:\n",
    "        selected_data = select_data(data)\n",
    "        if selected_data is not None:\n",
    "            extracted_data.append(selected_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "9663"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(duplicate_clean_text_set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1589170"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(all_clean_text_set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'context': 'Georg Marius, auch Georg Mayer (und \"Georg Meyer\"), latinisiert '\n",
      "            'Georgius Marius (oder \"Mayerus\"), auch \"Georgius Marius '\n",
      "            'Herbipolensis\" (und \"… Wirtzeburgensis\") sowie \"Jorg Mayr\", sich '\n",
      "            'selbst bezeichnend auch Mayer von Würzburg – (* 1533 in Würzburg; '\n",
      "            '† 5. März 1606 in Heidelberg) war ein deutscher Mediziner. Er war '\n",
      "            'Hochschullehrer, Lehrstuhlinhaber und Dekan in Heidelberg und an '\n",
      "            'der protestantischen Universität Marburg sowie Leibarzt dreier, '\n",
      "            'in der Geschichte der Reformation bedeutsamer, Fürsten. Er '\n",
      "            'betrieb auch eine große Privatpraxis. Marius studierte in '\n",
      "            'Heidelberg wo er sich im Juni 1548 als Georgius Marius (bzw. '\n",
      "            'Mayer) immatrikuliert hatte und, geprüft unter anderem von Jakob '\n",
      "            'Curio und Stephan Feyerabend, \"Baccalaureus artium\" geworden war '\n",
      "            'und danach in Marburg Medizin.',\n",
      " 'context_uuid': 'd82a3c4b-f73c-41c6-8424-43ee9ff6c3eb',\n",
      " 'imperatives': ['Nennen Sie die verschiedenen Namen, unter denen Georg Marius '\n",
      "                 'bekannt war.',\n",
      "                 'Geben Sie das Geburts- und Todesjahr von Georg Marius an.',\n",
      "                 'Erklären Sie, welche Positionen Georg Marius innehatte.',\n",
      "                 'Beschreiben Sie die Rolle von Georg Marius in der Geschichte '\n",
      "                 'der Reformation.',\n",
      "                 'Zeigen Sie auf, wo Georg Marius studiert hat und welchen '\n",
      "                 'Abschluss er erworben hat.',\n",
      "                 'Nennen Sie die Personen, die Georg Marius geprüft haben.'],\n",
      " 'questions': ['Wo wurde Georg Marius geboren?',\n",
      "               'Welche Positionen hatte Georg Marius an der Universität '\n",
      "               'Heidelberg und der Universität Marburg inne?',\n",
      "               'Welche Studien hat Georg Marius in Heidelberg absolviert?',\n",
      "               'Wann hat sich Georg Marius in Heidelberg immatrikuliert?',\n",
      "               'Wer waren einige der Personen, die Georg Marius geprüft haben?',\n",
      "               \"Was war Georg Marius' Beruf?\"],\n",
      " 'title': 'Georg Marius',\n",
      " 'wiki_id': 8339522}\n"
     ]
    }
   ],
   "source": [
    "pprint(extracted_data[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# clean duplicate texts\n",
    "for data in extracted_data:\n",
    "    if \"questions\" in data:\n",
    "        for question in data[\"questions\"].copy():\n",
    "            if question in duplicate_clean_text_set:\n",
    "                data[\"questions\"].remove(question)\n",
    "        if len(data[\"questions\"]) == 0:\n",
    "            del data[\"questions\"]\n",
    "\n",
    "    if \"imperatives\" in data:\n",
    "        for imperative in data[\"imperatives\"].copy():\n",
    "            if imperative in duplicate_clean_text_set:\n",
    "                data[\"imperatives\"].remove(imperative)\n",
    "        if len(data[\"imperatives\"]) == 0:\n",
    "            del data[\"imperatives\"]\n",
    "\n",
    "    assert \"questions\" in data or \"imperatives\" in data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "question_counts = []\n",
    "imperative_counts = []\n",
    "for data in extracted_data:\n",
    "    if \"questions\" in data:\n",
    "        question_counts.append(len(data[\"questions\"]))\n",
    "    if \"imperatives\" in data:\n",
    "        imperative_counts.append(len(data[\"imperatives\"]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'context': 'Georg Marius, auch Georg Mayer (und \"Georg Meyer\"), latinisiert '\n",
      "            'Georgius Marius (oder \"Mayerus\"), auch \"Georgius Marius '\n",
      "            'Herbipolensis\" (und \"… Wirtzeburgensis\") sowie \"Jorg Mayr\", sich '\n",
      "            'selbst bezeichnend auch Mayer von Würzburg – (* 1533 in Würzburg; '\n",
      "            '† 5. März 1606 in Heidelberg) war ein deutscher Mediziner. Er war '\n",
      "            'Hochschullehrer, Lehrstuhlinhaber und Dekan in Heidelberg und an '\n",
      "            'der protestantischen Universität Marburg sowie Leibarzt dreier, '\n",
      "            'in der Geschichte der Reformation bedeutsamer, Fürsten. Er '\n",
      "            'betrieb auch eine große Privatpraxis. Marius studierte in '\n",
      "            'Heidelberg wo er sich im Juni 1548 als Georgius Marius (bzw. '\n",
      "            'Mayer) immatrikuliert hatte und, geprüft unter anderem von Jakob '\n",
      "            'Curio und Stephan Feyerabend, \"Baccalaureus artium\" geworden war '\n",
      "            'und danach in Marburg Medizin.',\n",
      " 'context_uuid': 'd82a3c4b-f73c-41c6-8424-43ee9ff6c3eb',\n",
      " 'imperatives': ['Nennen Sie die verschiedenen Namen, unter denen Georg Marius '\n",
      "                 'bekannt war.',\n",
      "                 'Geben Sie das Geburts- und Todesjahr von Georg Marius an.',\n",
      "                 'Erklären Sie, welche Positionen Georg Marius innehatte.',\n",
      "                 'Beschreiben Sie die Rolle von Georg Marius in der Geschichte '\n",
      "                 'der Reformation.',\n",
      "                 'Zeigen Sie auf, wo Georg Marius studiert hat und welchen '\n",
      "                 'Abschluss er erworben hat.',\n",
      "                 'Nennen Sie die Personen, die Georg Marius geprüft haben.'],\n",
      " 'questions': ['Wo wurde Georg Marius geboren?',\n",
      "               'Welche Positionen hatte Georg Marius an der Universität '\n",
      "               'Heidelberg und der Universität Marburg inne?',\n",
      "               'Welche Studien hat Georg Marius in Heidelberg absolviert?',\n",
      "               'Wann hat sich Georg Marius in Heidelberg immatrikuliert?',\n",
      "               'Wer waren einige der Personen, die Georg Marius geprüft haben?',\n",
      "               \"Was war Georg Marius' Beruf?\"],\n",
      " 'title': 'Georg Marius',\n",
      " 'wiki_id': 8339522}\n"
     ]
    }
   ],
   "source": [
    "pprint(extracted_data[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "_ = pd.Series(question_counts).hist(bins=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "_ = pd.Series(imperative_counts).hist(bins=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "134591"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(extracted_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "786353"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sum(question_counts)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "793143"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sum(imperative_counts)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# split to train and test data\n",
    "random.shuffle(extracted_data)\n",
    "\n",
    "test_data = extracted_data[:5000]\n",
    "train_data = extracted_data[5000:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save train data\n",
    "with gzip.GzipFile(\"./data/wikipedia-22-12-de-train-data.json.gz\", \"w\") as f:\n",
    "    f.write(json.dumps(train_data, indent=1).encode(\"utf-8\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save test data\n",
    "with gzip.GzipFile(\"./data/wikipedia-22-12-de-test-data.json.gz\", \"w\") as f:\n",
    "    f.write(json.dumps(test_data, indent=1).encode(\"utf-8\"))"
   ]
  }
 ],
 "metadata": {
  "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": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}