add preproc notebook

preprocessing/20NG-Preprocessing.ipynb

@@ -0,0 +1,699 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# 20NG (Twenty Newsgroups). Preprocessing\n",
+    "\n",
+    "Here goes an example of data preprocessing and converting it to TopicNet's Dataset format.\n",
+    "\n",
+    "* Example of a toy dataset: [test_dataset.csv](https://github.com/machine-intelligence-laboratory/TopicNet/blob/master/topicnet/tests/test_data/test_dataset.csv)\n",
+    "* Dataset source file (with some explanations in docstring): [dataset.py](https://github.com/machine-intelligence-laboratory/TopicNet/blob/master/topicnet/cooking_machine/dataset.py)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Contents<a id=\"contents\"></a>\n",
+    "\n",
+    "* [Loading data](#data-loading)\n",
+    "* [Preparing data](#data-preparation)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 45,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "import pandas as pd\n",
+    "import re\n",
+    "import shutil\n",
+    "import string\n",
+    "\n",
+    "from collections import Counter\n",
+    "from glob import glob\n",
+    "\n",
+    "from sklearn import datasets\n",
+    "from sklearn.datasets import fetch_20newsgroups"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import nltk\n",
+    "\n",
+    "from nltk.collocations import (\n",
+    "    BigramAssocMeasures,\n",
+    "    BigramCollocationFinder,\n",
+    ")\n",
+    "from nltk.corpus import (\n",
+    "    stopwords,\n",
+    "    wordnet,\n",
+    ")\n",
+    "from nltk.stem import WordNetLemmatizer"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import matplotlib.pyplot as plt\n",
+    "%matplotlib inline\n",
+    "\n",
+    "from matplotlib import cm"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Loading data<a id=\"data-loading\"></a>\n",
+    "\n",
+    "<div style=\"text-align: right\">Back to <a href=#contents>Contents</a></div>"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Let's download the dataset:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Downloading 20news dataset. This may take a few minutes.\n",
+      "Downloading dataset from https://ndownloader.figshare.com/files/5975967 (14 MB)\n"
+     ]
+    }
+   ],
+   "source": [
+    "train_20 = fetch_20newsgroups(\n",
+    "    subset='train',\n",
+    "    remove=('headers', 'footers', 'quotes'),\n",
+    ")\n",
+    "test_20 = fetch_20newsgroups(\n",
+    "    subset='test',\n",
+    "    remove=('headers', 'footers', 'quotes'),\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "11314 data\n",
+      "11314 filenames\n",
+      "11314 target\n"
+     ]
+    }
+   ],
+   "source": [
+    "train_20.pop('DESCR')\n",
+    "labels = train_20.pop('target_names')\n",
+    "\n",
+    "for k in train_20.keys():\n",
+    "    print(len(train_20[k]), k)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "7532 data\n",
+      "7532 filenames\n",
+      "7532 target\n"
+     ]
+    }
+   ],
+   "source": [
+    "test_20.pop('DESCR')\n",
+    "labels_test = test_20.pop('target_names')\n",
+    "\n",
+    "for k in test_20.keys():\n",
+    "    print(len(test_20[k]), k)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Preparing data (lemmatization, Vowpal Wabbit & TopicNet's format)<a id=\"data-preparation\"></a>\n",
+    "\n",
+    "<div style=\"text-align: right\">Back to <a href=#contents>Contents</a></div>"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Wrapping all in .csv files:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "train_pd = pd.DataFrame(train_20).rename(columns = {'data':'raw_text'},)\n",
+    "# train_pd['raw_text'] = train_pd['raw_text'].apply(lambda x: x.decode('windows-1252'))\n",
+    "train_pd['id'] = train_pd.filenames.apply( lambda x: '.'.join(x.split('/')[-2:]).replace('.','_'))\n",
+    "\n",
+    "test_pd = pd.DataFrame(test_20).rename(columns = {'data':'raw_text'})\n",
+    "# test_pd['raw_text'] = test_pd['raw_text'].apply(lambda x: x.decode('windows-1252'))\n",
+    "test_pd['id'] = test_pd.filenames.apply( lambda x: '.'.join(x.split('/')[-2:]))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Better to exclude these documents (one may look here [20-newsgroups-secrets](https://github.com/Alvant/20-newsgroups-secrets) for more details)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 36,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "bad_names = [9976, 9977, 9978, 9979, 9980, 9981, 9982, 9983, 9984, 9985, 9986, 9987, 9988, 9990]\n",
+    "bad_names = [f\"comp_os_ms-windows_misc_{i}\" for i in bad_names]\n",
+    "\n",
+    "bad_indices = train_pd.query(\"id in @bad_names\").index"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Below we define some functions for text preprocessing."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 24,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def nltk2wn_tag(nltk_tag):\n",
+    "    if nltk_tag.startswith('J'):\n",
+    "        return wordnet.ADJ\n",
+    "    elif nltk_tag.startswith('V'):\n",
+    "        return wordnet.VERB\n",
+    "    elif nltk_tag.startswith('N'):\n",
+    "        return wordnet.NOUN\n",
+    "    elif nltk_tag.startswith('R'):\n",
+    "        return wordnet.ADV\n",
+    "    else:          \n",
+    "        return ''"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "pattern = re.compile('\\S*@\\S*\\s?')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 26,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def vowpalize_sequence(sequence):\n",
+    "    word_2_frequency = Counter(sequence)\n",
+    "    \n",
+    "    del word_2_frequency['']\n",
+    "    \n",
+    "    vw_string = ''\n",
+    "    \n",
+    "    for word in word_2_frequency:\n",
+    "        vw_string += word + \":\" + str(word_2_frequency[word]) + ' '\n",
+    "    \n",
+    "    return vw_string\n",
+    "\n",
+    "def do_vw_for_me_please(dataframe):\n",
+    "    bad_entries = []\n",
+    "    tokenized_text = []\n",
+    "    \n",
+    "    for indx, text in enumerate(dataframe['raw_text'].values):\n",
+    "        try:\n",
+    "            text = str(pattern.sub('', text))\n",
+    "        except TypeError:\n",
+    "            text=''\n",
+    "        \n",
+    "        tokens = [tok for tok in nltk.wordpunct_tokenize(text.lower()) if len(tok) > 1]\n",
+    "        tokenized_text.append(nltk.pos_tag(tokens))\n",
+    "    \n",
+    "    dataframe['tokenized'] = tokenized_text\n",
+    "\n",
+    "    stop = set(stopwords.words('english'))\n",
+    "\n",
+    "    lemmatized_text = []\n",
+    "    wnl = WordNetLemmatizer()\n",
+    "    \n",
+    "    for text in dataframe['tokenized'].values:\n",
+    "        lemmatized = [wnl.lemmatize(word, nltk2wn_tag(pos))\n",
+    "                      if nltk2wn_tag(pos) != ''\n",
+    "                      else wnl.lemmatize(word)\n",
+    "                      for word, pos in text ]\n",
+    "        lemmatized = [word for word in lemmatized \n",
+    "                      if word not in stop and word.isalpha()]\n",
+    "        lemmatized_text.append(lemmatized)\n",
+    "    \n",
+    "    dataframe['lemmatized'] = lemmatized_text\n",
+    "\n",
+    "    bigram_measures = BigramAssocMeasures()\n",
+    "    finder = BigramCollocationFinder.from_documents(dataframe['lemmatized'])\n",
+    "    finder.apply_freq_filter(5)\n",
+    "    set_dict = set(finder.nbest(bigram_measures.pmi,32100)[100:])\n",
+    "    documents = dataframe['lemmatized']\n",
+    "    bigrams = []\n",
+    "\n",
+    "    for doc in documents:\n",
+    "        entry = ['_'.join([word_first, word_second])\n",
+    "                 for word_first, word_second in zip(doc[:-1],doc[1:])\n",
+    "                 if (word_first, word_second) in set_dict]\n",
+    "        bigrams.append(entry)\n",
+    "\n",
+    "    dataframe['bigram'] = bigrams\n",
+    "    \n",
+    "    vw_text = []\n",
+    "\n",
+    "    for index, data in dataframe.iterrows():\n",
+    "        vw_string = ''    \n",
+    "        doc_id = data.id\n",
+    "        lemmatized = '@lemmatized ' + vowpalize_sequence(data.lemmatized)\n",
+    "        bigram = '@bigram ' + vowpalize_sequence(data.bigram)\n",
+    "        vw_string = ' |'.join([doc_id, lemmatized, bigram])\n",
+    "        vw_text.append(vw_string)\n",
+    "\n",
+    "    dataframe['vw_text'] = vw_text\n",
+    "\n",
+    "    print('num bad entries ', len(bad_entries))\n",
+    "    print(bad_entries)\n",
+    "\n",
+    "    return dataframe"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "And here are the final datasets!\n",
+    "Each row represents a document.\n",
+    "Columns `id`, `raw_text` and `vw_text` are required (look at this [toy dataset](https://github.com/machine-intelligence-laboratory/TopicNet/blob/master/topicnet/tests/test_data/test_dataset.csv), for example)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 41,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "num bad entries  0\n",
+      "[]\n"
+     ]
+    },
+    {
+     "data": {
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>raw_text</th>\n",
+       "      <th>filenames</th>\n",
+       "      <th>target</th>\n",
+       "      <th>id</th>\n",
+       "      <th>tokenized</th>\n",
+       "      <th>lemmatized</th>\n",
+       "      <th>bigram</th>\n",
+       "      <th>vw_text</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>I was wondering if anyone out there could enli...</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>7</td>\n",
+       "      <td>rec_autos_102994</td>\n",
+       "      <td>[(was, VBD), (wondering, VBG), (if, IN), (anyo...</td>\n",
+       "      <td>[wonder, anyone, could, enlighten, car, saw, d...</td>\n",
+       "      <td>[wonder_anyone, anyone_could, sport_car, car_l...</td>\n",
+       "      <td>rec_autos_102994 |@lemmatized wonder:1 anyone:...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>A fair number of brave souls who upgraded thei...</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>4</td>\n",
+       "      <td>comp_sys_mac_hardware_51861</td>\n",
+       "      <td>[(fair, JJ), (number, NN), (of, IN), (brave, J...</td>\n",
+       "      <td>[fair, number, brave, soul, upgrade, si, clock...</td>\n",
+       "      <td>[clock_oscillator, please_send, top_speed, hea...</td>\n",
+       "      <td>comp_sys_mac_hardware_51861 |@lemmatized fair:...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>well folks, my mac plus finally gave up the gh...</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>4</td>\n",
+       "      <td>comp_sys_mac_hardware_51879</td>\n",
+       "      <td>[(well, RB), (folks, NNS), (my, PRP$), (mac, J...</td>\n",
+       "      <td>[well, folk, mac, plus, finally, give, ghost, ...</td>\n",
+       "      <td>[mac_plus, life_way, way_back, market_new, new...</td>\n",
+       "      <td>comp_sys_mac_hardware_51879 |@lemmatized well:...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>\\nDo you have Weitek's address/phone number?  ...</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>1</td>\n",
+       "      <td>comp_graphics_38242</td>\n",
+       "      <td>[(do, VBP), (you, PRP), (have, VB), (weitek, V...</td>\n",
+       "      <td>[weitek, address, phone, number, like, get, in...</td>\n",
+       "      <td>[address_phone, phone_number, number_like, lik...</td>\n",
+       "      <td>comp_graphics_38242 |@lemmatized weitek:1 addr...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>4</th>\n",
+       "      <td>From article &lt;[email protected]&gt;, by to...</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>14</td>\n",
+       "      <td>sci_space_60880</td>\n",
+       "      <td>[(from, IN), (article, NN), (by, IN), (tom, NN...</td>\n",
+       "      <td>[article, tom, baker, understanding, expected,...</td>\n",
+       "      <td>[system_software, thing_check, introduce_new, ...</td>\n",
+       "      <td>sci_space_60880 |@lemmatized article:1 tom:1 b...</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ],
+      "text/plain": [
+       "                                            raw_text  \\\n",
+       "0  I was wondering if anyone out there could enli...   \n",
+       "1  A fair number of brave souls who upgraded thei...   \n",
+       "2  well folks, my mac plus finally gave up the gh...   \n",
+       "3  \\nDo you have Weitek's address/phone number?  ...   \n",
+       "4  From article <[email protected]>, by to...   \n",
+       "\n",
+       "                                           filenames  target  \\\n",
+       "0  /home/bulatov/scikit_learn_data/20news_home/20...       7   \n",
+       "1  /home/bulatov/scikit_learn_data/20news_home/20...       4   \n",
+       "2  /home/bulatov/scikit_learn_data/20news_home/20...       4   \n",
+       "3  /home/bulatov/scikit_learn_data/20news_home/20...       1   \n",
+       "4  /home/bulatov/scikit_learn_data/20news_home/20...      14   \n",
+       "\n",
+       "                            id  \\\n",
+       "0             rec_autos_102994   \n",
+       "1  comp_sys_mac_hardware_51861   \n",
+       "2  comp_sys_mac_hardware_51879   \n",
+       "3          comp_graphics_38242   \n",
+       "4              sci_space_60880   \n",
+       "\n",
+       "                                           tokenized  \\\n",
+       "0  [(was, VBD), (wondering, VBG), (if, IN), (anyo...   \n",
+       "1  [(fair, JJ), (number, NN), (of, IN), (brave, J...   \n",
+       "2  [(well, RB), (folks, NNS), (my, PRP$), (mac, J...   \n",
+       "3  [(do, VBP), (you, PRP), (have, VB), (weitek, V...   \n",
+       "4  [(from, IN), (article, NN), (by, IN), (tom, NN...   \n",
+       "\n",
+       "                                          lemmatized  \\\n",
+       "0  [wonder, anyone, could, enlighten, car, saw, d...   \n",
+       "1  [fair, number, brave, soul, upgrade, si, clock...   \n",
+       "2  [well, folk, mac, plus, finally, give, ghost, ...   \n",
+       "3  [weitek, address, phone, number, like, get, in...   \n",
+       "4  [article, tom, baker, understanding, expected,...   \n",
+       "\n",
+       "                                              bigram  \\\n",
+       "0  [wonder_anyone, anyone_could, sport_car, car_l...   \n",
+       "1  [clock_oscillator, please_send, top_speed, hea...   \n",
+       "2  [mac_plus, life_way, way_back, market_new, new...   \n",
+       "3  [address_phone, phone_number, number_like, lik...   \n",
+       "4  [system_software, thing_check, introduce_new, ...   \n",
+       "\n",
+       "                                             vw_text  \n",
+       "0  rec_autos_102994 |@lemmatized wonder:1 anyone:...  \n",
+       "1  comp_sys_mac_hardware_51861 |@lemmatized fair:...  \n",
+       "2  comp_sys_mac_hardware_51879 |@lemmatized well:...  \n",
+       "3  comp_graphics_38242 |@lemmatized weitek:1 addr...  \n",
+       "4  sci_space_60880 |@lemmatized article:1 tom:1 b...  "
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "num bad entries  0\n",
+      "[]\n"
+     ]
+    },
+    {
+     "data": {
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>raw_text</th>\n",
+       "      <th>filenames</th>\n",
+       "      <th>target</th>\n",
+       "      <th>id</th>\n",
+       "      <th>tokenized</th>\n",
+       "      <th>lemmatized</th>\n",
+       "      <th>bigram</th>\n",
+       "      <th>vw_text</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>I am a little confused on all of the models of...</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>7</td>\n",
+       "      <td>rec.autos.103343</td>\n",
+       "      <td>[(am, VBP), (little, JJ), (confused, VBN), (on...</td>\n",
+       "      <td>[little, confuse, model, bonnevilles, hear, le...</td>\n",
+       "      <td>[could_someone, someone_tell, tell_difference,...</td>\n",
+       "      <td>rec.autos.103343 |@lemmatized little:1 confuse...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>I'm not familiar at all with the format of the...</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>5</td>\n",
+       "      <td>comp.windows.x.67445</td>\n",
+       "      <td>[(not, RB), (familiar, JJ), (at, IN), (all, DT...</td>\n",
+       "      <td>[familiar, format, face, thingies, see, folk, ...</td>\n",
+       "      <td>[get_see, make_one, one_get, seem_find, could_...</td>\n",
+       "      <td>comp.windows.x.67445 |@lemmatized familiar:1 f...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>\\nIn a word, yes.\\n</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>0</td>\n",
+       "      <td>alt.atheism.53603</td>\n",
+       "      <td>[(in, IN), (word, NN), (yes, NN)]</td>\n",
+       "      <td>[word, yes]</td>\n",
+       "      <td>[]</td>\n",
+       "      <td>alt.atheism.53603 |@lemmatized word:1 yes:1  |...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>\\nThey were attacking the Iraqis to drive them...</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>17</td>\n",
+       "      <td>talk.politics.mideast.77355</td>\n",
+       "      <td>[(they, PRP), (were, VBD), (attacking, VBG), (...</td>\n",
+       "      <td>[attack, iraqi, drive, kuwait, country, whose,...</td>\n",
+       "      <td>[think_u, saudi_arabia, much_anything, saudi_a...</td>\n",
+       "      <td>talk.politics.mideast.77355 |@lemmatized attac...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>4</th>\n",
+       "      <td>\\nI've just spent two solid months arguing tha...</td>\n",
+       "      <td>/home/bulatov/scikit_learn_data/20news_home/20...</td>\n",
+       "      <td>19</td>\n",
+       "      <td>talk.religion.misc.84194</td>\n",
+       "      <td>[(ve, NN), (just, RB), (spent, VBN), (two, CD)...</td>\n",
+       "      <td>[spend, two, solid, month, argue, thing, objec...</td>\n",
+       "      <td>[moral_system]</td>\n",
+       "      <td>talk.religion.misc.84194 |@lemmatized spend:1 ...</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ],
+      "text/plain": [
+       "                                            raw_text  \\\n",
+       "0  I am a little confused on all of the models of...   \n",
+       "1  I'm not familiar at all with the format of the...   \n",
+       "2                                \\nIn a word, yes.\\n   \n",
+       "3  \\nThey were attacking the Iraqis to drive them...   \n",
+       "4  \\nI've just spent two solid months arguing tha...   \n",
+       "\n",
+       "                                           filenames  target  \\\n",
+       "0  /home/bulatov/scikit_learn_data/20news_home/20...       7   \n",
+       "1  /home/bulatov/scikit_learn_data/20news_home/20...       5   \n",
+       "2  /home/bulatov/scikit_learn_data/20news_home/20...       0   \n",
+       "3  /home/bulatov/scikit_learn_data/20news_home/20...      17   \n",
+       "4  /home/bulatov/scikit_learn_data/20news_home/20...      19   \n",
+       "\n",
+       "                            id  \\\n",
+       "0             rec.autos.103343   \n",
+       "1         comp.windows.x.67445   \n",
+       "2            alt.atheism.53603   \n",
+       "3  talk.politics.mideast.77355   \n",
+       "4     talk.religion.misc.84194   \n",
+       "\n",
+       "                                           tokenized  \\\n",
+       "0  [(am, VBP), (little, JJ), (confused, VBN), (on...   \n",
+       "1  [(not, RB), (familiar, JJ), (at, IN), (all, DT...   \n",
+       "2                  [(in, IN), (word, NN), (yes, NN)]   \n",
+       "3  [(they, PRP), (were, VBD), (attacking, VBG), (...   \n",
+       "4  [(ve, NN), (just, RB), (spent, VBN), (two, CD)...   \n",
+       "\n",
+       "                                          lemmatized  \\\n",
+       "0  [little, confuse, model, bonnevilles, hear, le...   \n",
+       "1  [familiar, format, face, thingies, see, folk, ...   \n",
+       "2                                        [word, yes]   \n",
+       "3  [attack, iraqi, drive, kuwait, country, whose,...   \n",
+       "4  [spend, two, solid, month, argue, thing, objec...   \n",
+       "\n",
+       "                                              bigram  \\\n",
+       "0  [could_someone, someone_tell, tell_difference,...   \n",
+       "1  [get_see, make_one, one_get, seem_find, could_...   \n",
+       "2                                                 []   \n",
+       "3  [think_u, saudi_arabia, much_anything, saudi_a...   \n",
+       "4                                     [moral_system]   \n",
+       "\n",
+       "                                             vw_text  \n",
+       "0  rec.autos.103343 |@lemmatized little:1 confuse...  \n",
+       "1  comp.windows.x.67445 |@lemmatized familiar:1 f...  \n",
+       "2  alt.atheism.53603 |@lemmatized word:1 yes:1  |...  \n",
+       "3  talk.politics.mideast.77355 |@lemmatized attac...  \n",
+       "4  talk.religion.misc.84194 |@lemmatized spend:1 ...  "
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "train_pd = do_vw_for_me_please(train_pd)\n",
+    "display(train_pd.head())\n",
+    "\n",
+    "test_pd = do_vw_for_me_please(test_pd)\n",
+    "display(test_pd.head())"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Saving to disk (TopicNet's [Dataset](https://github.com/machine-intelligence-laboratory/TopicNet/blob/master/topicnet/cooking_machine/dataset.py) can be constructed using saved .csv file with text data)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 44,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "! mkdir 20_News_dataset\n",
+    "\n",
+    "train_pd.drop(bad_indices).to_csv('/data/datasets/20_News_dataset/train_preprocessed.csv')\n",
+    "test_pd.to_csv('/data/datasets/20_News_dataset/test_preprocessed.csv')"
+   ]
+  }
+ ],
+ "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": "3.6.9"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}