Training model for classifcation using Word2Vec & Linear SVM.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Classification using Word2Vec and Extra Trees Classifiers\n",
    "\n",
    "This script describes how text data can be embedded using Word2Vec and used to train Extra Tree classiifers for classification into categories. Here, the text data example used are noise complaints."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "from tabulate import tabulate\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import gensim\n",
    "from gensim.models.word2vec import Word2Vec\n",
    "from sklearn import utils\n",
    "from sklearn.model_selection import train_test_split\n",
    "import regex as re\n",
    "import string\n",
    "from collections import Counter, defaultdict\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sklearn.ensemble import ExtraTreesClassifier\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.metrics import accuracy_score, f1_score\n",
    "from sklearn.metrics import classification_report\n",
    "from sklearn.model_selection import cross_val_score\n",
    "import dill as pickle\n",
    "\n",
    "GLOVE_6B_50D_PATH = \"glove.6B.50d.txt\"\n",
    "encoding=\"utf-8\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "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>CONTENT</th>\n",
       "      <th>CONTENT_label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>10am to 11am - resd f/b that the many prc ever...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>email i would like to share another feedback o...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>every week from monday to friday about 7pm to ...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>children shouting &amp; playing at v/deck blk 127 ...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>blk 703 and 704 - kids still playing at v/deck...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>grp of youth playing basket ball &amp; making alot...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>youth playing skateboard at mph. very noisy._</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>group of people playing music very loud at ope...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>kids playing &amp; shouting @ v/deck.  very noisy....</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>few people playing scoccer @ v/deck, very nois...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                             CONTENT CONTENT_label\n",
       "0  10am to 11am - resd f/b that the many prc ever...  Congregation\n",
       "1  email i would like to share another feedback o...  Congregation\n",
       "2  every week from monday to friday about 7pm to ...  Congregation\n",
       "3  children shouting & playing at v/deck blk 127 ...  Congregation\n",
       "4  blk 703 and 704 - kids still playing at v/deck...  Congregation\n",
       "5  grp of youth playing basket ball & making alot...  Congregation\n",
       "6      youth playing skateboard at mph. very noisy._  Congregation\n",
       "7  group of people playing music very loud at ope...  Congregation\n",
       "8  kids playing & shouting @ v/deck.  very noisy....  Congregation\n",
       "9  few people playing scoccer @ v/deck, very nois...  Congregation"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.read_csv('./LabelledData.csv', encoding='latin1')\n",
    "df.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "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>CONTENT</th>\n",
       "      <th>CONTENT_label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>10am to 11am - resd f/b that the many prc ever...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>email i would like to share another feedback o...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>every week from monday to friday about 7pm to ...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>children shouting &amp; playing at v/deck blk 127 ...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>blk 703 and 704 - kids still playing at v/deck...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>grp of youth playing basket ball &amp; making alot...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>youth playing skateboard at mph. very noisy._</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>group of people playing music very loud at ope...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>kids playing &amp; shouting @ v/deck.  very noisy....</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>few people playing scoccer @ v/deck, very nois...</td>\n",
       "      <td>Congregation</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                             CONTENT CONTENT_label\n",
       "0  10am to 11am - resd f/b that the many prc ever...  Congregation\n",
       "1  email i would like to share another feedback o...  Congregation\n",
       "2  every week from monday to friday about 7pm to ...  Congregation\n",
       "3  children shouting & playing at v/deck blk 127 ...  Congregation\n",
       "4  blk 703 and 704 - kids still playing at v/deck...  Congregation\n",
       "5  grp of youth playing basket ball & making alot...  Congregation\n",
       "6      youth playing skateboard at mph. very noisy._  Congregation\n",
       "7  group of people playing music very loud at ope...  Congregation\n",
       "8  kids playing & shouting @ v/deck.  very noisy....  Congregation\n",
       "9  few people playing scoccer @ v/deck, very nois...  Congregation"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = df[['CONTENT','CONTENT_label']]\n",
    "df = df[pd.notnull(df['CONTENT'])]\n",
    "df.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.index = range(df.shape[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "330064"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['CONTENT'].apply(lambda x: len(x.split(' '))).sum() # Number of words in all documents"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1728x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cnt_pro = df['CONTENT_label'].value_counts()\n",
    "\n",
    "plt.figure(figsize=(24,8))\n",
    "sns.barplot(cnt_pro.index, cnt_pro.values, alpha=0.8)\n",
    "plt.ylabel('Number of Occurrences', fontsize=24)\n",
    "# Get current axis on current figure\n",
    "ax = plt.gca()\n",
    "\n",
    "# Iterate through the list of axes' patches\n",
    "for p in ax.patches:\n",
    "    ax.text(p.get_x() + p.get_width()/2., p.get_height(), '%d' % int(p.get_height()), \n",
    "            fontsize=24, color='black', ha='center', va='bottom')\n",
    "plt.xticks(rotation=90, fontsize=24)\n",
    "plt.yticks(fontsize=24)\n",
    "plt.show();"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def print_complaint(index):\n",
    "    example = df[df.index == index][['CONTENT', 'CONTENT_label']].values[0]\n",
    "    if len(example) > 0:\n",
    "        print(example[0])\n",
    "        print('Label:', example[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "bunch of teenagers playing at badminton court creating noise and littering the place. req to stop them_\n",
      "Label: Congregation\n"
     ]
    }
   ],
   "source": [
    "print_complaint(12)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Text preprocessing\n",
    "\n",
    "Below we define a function to convert text to lower-case and strip punctuation/symbols from words and so on."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def cleanText(text):\n",
    "    text = re.sub(r'#', r'apartment unit ', text)\n",
    "    text = re.sub(r'['+string.punctuation+']', r\" \", text)\n",
    "    text = re.sub(r'[0-9]+', r' ', text)\n",
    "    text = re.sub('\\s+', \" \", text).strip()\n",
    "    text = text.lower()\n",
    "    text = re.sub(r\"\\bv deck\\b\", r\"voiddeck\", text)\n",
    "    text = re.sub(r\"\\bvd\\b\", r\"voiddeck\", text)\n",
    "    text = re.sub(r\"\\bv d\\b\", r\"voiddeck\", text)\n",
    "    text = re.sub(r\"\\bvdeck\\b\", r\"voiddeck\", text)\n",
    "    text = re.sub(r\"\\bvoid deck\\b\", r\"voiddeck\", text)\n",
    "    text = re.sub(r\"\\bfbck\\b\", r\"feedback\", text)\n",
    "    text = re.sub(r\"\\bfbk\\b\", r\"feedback\", text)\n",
    "    text = re.sub(r\"\\bf back\\b\", r\"feedback\", text)\n",
    "    text = re.sub(r\"\\bfback\\b\", r\"feedback\", text)\n",
    "    text = re.sub(r\"\\bfup\\b\", r\"feedback\", text)\n",
    "    text = re.sub(r\"\\bfb\\b\", r\"feedback\", text)\n",
    "    text = re.sub(r\"\\bres\\b\", r\"resident\", text)\n",
    "    text = re.sub(r\"\\bresd\\b\", r\"resident\", text)\n",
    "    text = re.sub(r\"\\bresid\\b\", r\"resident\", text)\n",
    "    text = re.sub(r\"\\breno\\b\", r\"renovation\", text)\n",
    "    text = re.sub(r\"\\bneighbour\\b\", r\"neighbor\", text)\n",
    "    text = re.sub(r\"\\bneighbours\\b\", r\"neighbor\", text)\n",
    "    text = re.sub(r\"\\bneighbors\\b\", r\"neighbor\", text)\n",
    "    text = re.sub(r\"\\breq\\b\", r\"request\", text)\n",
    "    text = re.sub(r\"\\brq\\b\", r\"request\", text)\n",
    "    text = re.sub(r\"\\binfd\\b\", r\"informed\", text)\n",
    "    text = re.sub(r\"\\badv\\b\", r\"advice\", text)\n",
    "    text = re.sub(r\"\\badvic\\b\", r\"advice\", text)\n",
    "    text = re.sub(r\"\\badvis\\b\", r\"advice\", text)\n",
    "    text = re.sub(r\"\\badvise\\b\", r\"advice\", text)\n",
    "    text = re.sub(r\"\\badviceeise\\b\", r\"advice\", text)\n",
    "    text = re.sub(r\"\\bppl\\b\", r\"people\", text)\n",
    "    text = re.sub(r\"\\bindians\\b\", r\"indian\", text)\n",
    "    text = re.sub(r\"\\bmalays\\b\", r\"malay\", text)\n",
    "    text = re.sub(r\"\\bment\\b\", r\"apartment\", text)\n",
    "    text = re.sub(r\"\\btcil\\b\", r\"towncouncil\", text)\n",
    "    text = re.sub(r\"\\btown council\\b\", r\"towncouncil\", text)\n",
    "    text = re.sub(r\"\\bcoffee shop\\b\", r\"coffeeshop\", text)\n",
    "    text = re.sub(r\"\\breno\\b\", r\"renovation\", text)\n",
    "    text = re.sub(r\"\\brenov\\b\", r\"renovation\", text)\n",
    "    text = re.sub(r\"\\bcpark\\b\", r\"carpark\", text)\n",
    "    text = re.sub(r\"\\bcar park\\b\", r\"carpark\", text)\n",
    "    text = re.sub(r\"\\bmscp\\b\", r\"multi-storey carpark\", text)\n",
    "    text = re.sub(r\"\\bmulti storey\\b\", r\"multi-storey\", text)\n",
    "    text = re.sub(r\"\\bmultistorey\\b\", r\"multi-storey\", text)\n",
    "    text = re.sub(r\"\\broad work\\b\", r\"roadworks\", text)\n",
    "    text = re.sub(r\"\\broad works\\b\", r\"roadworks\", text)\n",
    "    text = re.sub(r\"\\bfrm\\b\", r\"from\", text)\n",
    "    text = re.sub(r\"\\bgrp\\b\", r\"group\", text)\n",
    "    text = re.sub(r\"\\bhse\\b\", r\"house\", text)\n",
    "    text = re.sub(r\"\\bcan t\\b\", r\"cannot\", text)\n",
    "    text = re.sub(r\"\\bgrp\\b\", r\"group\", text)\n",
    "    text = re.sub(r\"\\bsaket bord\\b\", r\"skateboard\", text)\n",
    "    text = re.sub(r\"\\bw out\\b\", r\"without\", text)\n",
    "    text = re.sub(r\"\\bongg\\b\", r\"ongoing\", text)\n",
    "    text = re.sub(r\"\\bon gg\\b\", r\"ongoing\", text)\n",
    "    text = re.sub(r\"\\bfwake\\b\", r\"funeral wake\", text)\n",
    "    text = re.sub(r\"\\bcorr\\b\", r\"corridor\", text)\n",
    "    text = re.sub(r\"\\bcorri\\b\", r\"corridor\", text)\n",
    "    text = re.sub(r\"\\bcorridors\\b\", r\"corridor\", text)\n",
    "    text = re.sub(r\"\\bpg\\b\", r\"playground\", text)\n",
    "    text = re.sub(r\"\\bpgd\\b\", r\"playground\", text)\n",
    "    text = re.sub(r\"\\bpgrd\\b\", r\"playground\", text)\n",
    "    text = re.sub(r\"\\blway\\b\", r\"linkway\", text)\n",
    "    text = re.sub(r\"\\bl way\\b\", r\"linkway\", text)\n",
    "    text = re.sub(r\"\\bcontratcor\\b\", r\"linkway\", text)\n",
    "    text = re.sub(r\"\\bos\\b\", r\"outside\", text) \n",
    "    text = re.sub(r\"\\bo s\\b\", r\"outside\", text) \n",
    "    text = re.sub(r\"\\binfd\\b\", r\"informed\", text)\n",
    "    text = re.sub(r\"\\bfball\\b\", r\"football\", text)\n",
    "    text = re.sub(r\"\\bfoot ball\\b\", r\"football\", text)\n",
    "    text = re.sub(r\"\\brtop\\b\", r\"rooftop\", text)\n",
    "    text = re.sub(r\"\\br top\\b\", r\"rooftop\", text)\n",
    "    text = re.sub(r\"\\bramping\\b\", r\"revving\", text)\n",
    "    text = re.sub(r\"\\bramming\\b\", r\"revving\", text)\n",
    "    text = re.sub(r\"\\bcolln\\b\", r\"collection\", text)\n",
    "    text = re.sub(r\"\\bcontruction\\b\", r\"construction\", text)\n",
    "    text = re.sub(r\"\\bconctruction\\b\", r\"construction\", text)\n",
    "    text = re.sub(r\"\\bpurposed\\b\", r\"purpose\", text)\n",
    "    text = re.sub(r\"\\bmph\\b\", r\"multi-purpose hall\", text)\n",
    "    text = re.sub(r\"\\bmphall\\b\", r\"multi-purpose hall\", text)\n",
    "    text = re.sub(r\"\\bmulti purpose\\b\", r\"multi-purpose\", text)\n",
    "    text = re.sub(r\"\\bmultipurpose\\b\", r\"multi-purpose\", text)\n",
    "    text = re.sub(r\"\\bmulti purpos\\b\", r\"multi-purpose\", text)\n",
    "    text = re.sub(r\"\\bfurnitures\\b\", r\"furniture\", text)\n",
    "    text = re.sub(r\"\\bchairs\\b\", r\"chair\", text)\n",
    "    text = re.sub(r\"\\btables\\b\", r\"table\", text)\n",
    "    text = re.sub(r\"\\bcheers\\b\", r\"cheer\", text)\n",
    "    text = re.sub(r\"\\bsongs\\b\", r\"song\", text)\n",
    "    text = re.sub(r\"\\bsepak takraw\\b\", r\"sepaktakraw\", text)\n",
    "    text = re.sub(r\"\\bbadmointon\\b\", r\"badminton\", text)\n",
    "    text = re.sub(r\"\\bbadminton court\\b\", r\"badmintoncourt\", text)\n",
    "    text = re.sub(r\"\\bb minton\\b\", r\"badminton\", text)\n",
    "    text = re.sub(r\"\\bbasket ball\\b\", r\"basketball\", text)\n",
    "    text = re.sub(r\"\\bbasketball court\\b\", r\"basketballcourt\", text)\n",
    "    text = re.sub(r\"\\btakewondo\\b\", r\"taekwondo\", text)\n",
    "    text = re.sub(r\"\\btaichi\\b\", r\"tai chi\", text)\n",
    "    text = re.sub(r\"\\bgathing\\b\", r\"gathering\", text)\n",
    "    text = re.sub(r\"\\bsoccer court\\b\", r\"soccercourt\", text)\n",
    "    text = re.sub(r\"\\bfootball court\\b\", r\"footballcourt\", text)\n",
    "    return text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'group of boys playing soccer basketballcourt making lot of noise again and very dangerous reported nd times'"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['CONTENT_clean'] = df['CONTENT'].apply(cleanText)\n",
    "df['CONTENT_clean'].iloc[29]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Define locations of noise and the activities causing it."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Tag locations and activity to complaints where possible\n",
    "location = [\"void deck\",\n",
    "            \"multi-purpose hall\",\n",
    "            \"multi-purpose court\",\n",
    "            \"multi-purpose plaza\",\n",
    "            \"corridor\",\n",
    "            \"basketballcourt\", \n",
    "            \"badmintoncourt\",\n",
    "            \"soccercourt\",\n",
    "            \"footballcourt\",\n",
    "            \"playground\",\n",
    "            \"carpark\",\n",
    "            \"multi-storey carpark\",\n",
    "            \"coffeeshop\"]\n",
    "\n",
    "activity = [\"mahjong\",\n",
    "            \"babies\", \n",
    "            \"crying\",\n",
    "            \"dismantling\",\n",
    "            \"dismantle\",\n",
    "            \"furniture\", \n",
    "            \"drag\",\n",
    "            \"hack\",\n",
    "            \"push\", \n",
    "            \"drag\",\n",
    "            \"kick\",            \n",
    "            \"bang\",\n",
    "            \"table\",\n",
    "            \"chairs\", \n",
    "            \"pull\",\n",
    "            \"shout\",\n",
    "            \"talk\",\n",
    "            \"cheer\",\n",
    "            \"play\",\n",
    "            \"dance\",\n",
    "            \"dancing\"\n",
    "            \"music\",\n",
    "            \"song\",\n",
    "            \"dance\",\n",
    "            \"sepaktakraw\",\n",
    "            \"skateboard\", \n",
    "            \"wedding\", \n",
    "            \"funeral\", \n",
    "            \"party\", \n",
    "            \"taekwondo\",\n",
    "            \"exercise\",\n",
    "            \"exercising\",\n",
    "            \"wushu\",\n",
    "            \"tai chi\"]\n",
    "\n",
    "activity2 = ['basketball',\n",
    "             'football',\n",
    "             'soccer',\n",
    "             'badminton']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0                       \n",
       "1         badmintoncourt\n",
       "2                       \n",
       "3                       \n",
       "4             playground\n",
       "5        basketballcourt\n",
       "6     multi-purpose hall\n",
       "7                       \n",
       "8                       \n",
       "9                       \n",
       "10            playground\n",
       "11                      \n",
       "12        badmintoncourt\n",
       "13    multi-purpose hall\n",
       "14       basketballcourt\n",
       "15                      \n",
       "16                      \n",
       "17                      \n",
       "18                      \n",
       "19                      \n",
       "20                      \n",
       "21                      \n",
       "22                      \n",
       "23                      \n",
       "24                      \n",
       "25                      \n",
       "26    multi-purpose hall\n",
       "27                      \n",
       "28                      \n",
       "29       basketballcourt\n",
       "Name: Location, dtype: object"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['Location'] = \"\"\n",
    "\n",
    "location_lst = \"|\".join(location)\n",
    "\n",
    "ind = df.index[df['CONTENT_clean'].str.contains(location_lst)]\n",
    "\n",
    "for i in ind:\n",
    "    tmp = []\n",
    "    for word in location:\n",
    "        if word in df['CONTENT_clean'].loc[i]:\n",
    "            tmp.append(word)\n",
    "    if tmp:\n",
    "        df['Location'].loc[i] = ','.join(tmp)\n",
    "\n",
    "df['Location'].head(30)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0                          \n",
       "1            play,badminton\n",
       "2                      talk\n",
       "3                shout,play\n",
       "4                      play\n",
       "5           play,basketball\n",
       "6           play,skateboard\n",
       "7                      play\n",
       "8                shout,play\n",
       "9                      play\n",
       "10                     play\n",
       "11                     play\n",
       "12                     play\n",
       "13              play,soccer\n",
       "14          play,basketball\n",
       "15                  funeral\n",
       "16                         \n",
       "17                         \n",
       "18                     talk\n",
       "19                         \n",
       "20               shout,play\n",
       "21                         \n",
       "22                     play\n",
       "23                     play\n",
       "24    shout,play,basketball\n",
       "25            play,football\n",
       "26                     play\n",
       "27              play,soccer\n",
       "28            play,football\n",
       "29              play,soccer\n",
       "Name: Activity, dtype: object"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['Activity'] = \"\"\n",
    "\n",
    "activity_lst = \"|\".join(activity)\n",
    "\n",
    "ind = df.index[df['CONTENT_clean'].str.contains(activity_lst)]\n",
    "\n",
    "for i in ind:\n",
    "    tmp = []\n",
    "    for word in activity:\n",
    "        if word in df['CONTENT_clean'].loc[i]:\n",
    "            tmp.append(word)\n",
    "    for word in activity2:\n",
    "        if re.search(r'\\b' + word + r'\\b', df['CONTENT_clean'].loc[i]):\n",
    "            tmp.append(word)\n",
    "    if tmp:\n",
    "        df['Activity'].loc[i] = ','.join(tmp)\n",
    "\n",
    "df['Activity'].head(30)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Train/test split of 80/20."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(df['CONTENT_clean'],df['CONTENT_label'], test_size=0.2, random_state=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'group of boys playing soccer at voiddeck very noisy and kicking balls at the wall and dirty the place very dangerous for resident'"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.iloc[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The function below tokenizes the text data in each row (document) into individual words. Sentences smaller than 3 words and words that correspond to stopwords are discarded."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "import nltk\n",
    "from nltk.corpus import stopwords\n",
    "\n",
    "def tokenize_text(text, stop_words):\n",
    "    tokens = []\n",
    "    for sent in nltk.sent_tokenize(text):\n",
    "        for word in nltk.word_tokenize(sent):\n",
    "            if len(word) < 3 or word in stop_words:\n",
    "                continue\n",
    "            tokens.append(word.lower())\n",
    "    return tokens\n",
    "\n",
    "stop_words = stopwords.words('english')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tokenize the training and test data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "for i in range(X_train.shape[0]):\n",
    "    X_train.iloc[i] = tokenize_text(X_train.iloc[i], stop_words)\n",
    "\n",
    "for i in range(X_test.shape[0]):\n",
    "    X_test.iloc[i] = tokenize_text(X_test.iloc[i], stop_words)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is what a training entry looks like."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['group',\n",
       " 'boys',\n",
       " 'playing',\n",
       " 'soccer',\n",
       " 'voiddeck',\n",
       " 'noisy',\n",
       " 'kicking',\n",
       " 'balls',\n",
       " 'wall',\n",
       " 'dirty',\n",
       " 'place',\n",
       " 'dangerous',\n",
       " 'resident']"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.iloc[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(['elderly',\n",
       "  'man',\n",
       "  'playing',\n",
       "  'radio',\n",
       "  'loudly',\n",
       "  'function',\n",
       "  'hall',\n",
       "  'refer',\n",
       "  'lim',\n",
       "  'tanglin',\n",
       "  'police',\n",
       "  'station'],\n",
       " 'Neighbourly Disturbance')"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.iloc[30], y_train.iloc[30]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training the model\n",
    "\n",
    "We'll instantiate a Word2Vec model using Distributed Bag of Words (DBOW). Another approach possible is Skip-gram method. However, for short text data, DBOW performs better."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Prepare word embeddings and train a new one from scratch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "import multiprocessing\n",
    "\n",
    "cores = multiprocessing.cpu_count()\n",
    "\n",
    "# train word2vec on all the training text\n",
    "# we're not using test labels, just texts so this is fine\n",
    "model = Word2Vec(X_train, size=25, window=3, min_count=10, workers=cores)\n",
    "w2v = {w: vec for w, vec in zip(model.wv.index2word, model.wv.vectors)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# reading glove files, this may take a while\n",
    "# we're reading line by line and only saving vectors\n",
    "# that correspond to words from our training set\n",
    "# if you wan't to play around with the vectors and have \n",
    "# enough RAM - remove the 'if' line and load everything\n",
    "\n",
    "import struct \n",
    "\n",
    "glove_small = {}\n",
    "all_words = set(w for words in X_train for w in words)\n",
    "with open(GLOVE_6B_50D_PATH, \"rb\") as infile:\n",
    "    for line in infile:\n",
    "        parts = line.split()\n",
    "        word = parts[0].decode(encoding)\n",
    "        if (word in all_words):\n",
    "            nums=np.array(parts[1:], dtype=np.float32)\n",
    "            glove_small[word] = nums"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Model definitions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Classifiers using vector embeddings. We will implement an embedding vectorizer - a counterpart of CountVectorizer and TfidfVectorizer - that is given a word -> vector mapping and vectorizes texts by taking the mean of all the vectors corresponding to individual words."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MeanEmbeddingVectorizer(object):\n",
    "    def __init__(self, word2vec):\n",
    "        self.word2vec = word2vec\n",
    "        if len(word2vec)>0:\n",
    "            self.dim=len(word2vec[next(iter(glove_small))])\n",
    "        else:\n",
    "            self.dim=0\n",
    "            \n",
    "    def fit(self, X, y):\n",
    "        return self \n",
    "\n",
    "    def transform(self, X):\n",
    "        return np.array([\n",
    "            np.mean([self.word2vec[w] for w in words if w in self.word2vec] \n",
    "                    or [np.zeros(self.dim)], axis=0)\n",
    "            for words in X\n",
    "        ])\n",
    "\n",
    "    \n",
    "# and a tf-idf version of the same\n",
    "class TfidfEmbeddingVectorizer(object):\n",
    "    def __init__(self, word2vec):\n",
    "        self.word2vec = word2vec\n",
    "        self.word2weight = None\n",
    "        if len(word2vec)>0:\n",
    "            self.dim=len(word2vec[next(iter(glove_small))])\n",
    "        else:\n",
    "            self.dim=0\n",
    "        \n",
    "    def fit(self, X, y):\n",
    "        tfidf = TfidfVectorizer(analyzer=lambda x: x)\n",
    "        tfidf.fit(X)\n",
    "        # if a word was never seen - it must be at least as infrequent\n",
    "        # as any of the known words - so the default idf is the max of \n",
    "        # known idf's\n",
    "        max_idf = max(tfidf.idf_)\n",
    "        self.word2weight = defaultdict(\n",
    "            lambda: max_idf, \n",
    "            [(w, tfidf.idf_[i]) for w, i in tfidf.vocabulary_.items()])\n",
    "    \n",
    "        return self\n",
    "    \n",
    "    def transform(self, X):\n",
    "        return np.array([\n",
    "                np.mean([self.word2vec[w] * self.word2weight[w]\n",
    "                         for w in words if w in self.word2vec] or\n",
    "                        [np.zeros(self.dim)], axis=0)\n",
    "                for words in X\n",
    "            ])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here, we initialize the Extra Tree Classifiers using the Word2Vec model and train them using 2 different vector embeddings defined above. Extra Tree Classifiers are an extension and generalization of Random Forest classifier. Unlike Random Forest, Extra Tree Classifiers do not use bootstrap copies of the training data and selects a cut-point at random. This provides 2 benefits - 1) increased accuracy and 2) reduces computational burden which leads to faster training and classification."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "etree_w2v = Pipeline([(\"word2vec vectorizer\", MeanEmbeddingVectorizer(w2v)), \n",
    "                        (\"extra trees\", ExtraTreesClassifier(n_estimators=250))])\n",
    "etree_w2v_tfidf = Pipeline([(\"word2vec vectorizer\", TfidfEmbeddingVectorizer(w2v)), \n",
    "                        (\"extra trees\", ExtraTreesClassifier(n_estimators=250))])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Run 10-fold cross validation on all the models on the training set to find the model with the best accuracy."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "model        score\n",
      "---------  -------\n",
      "w2v         0.7951\n",
      "w2v_tfidf   0.7909\n"
     ]
    }
   ],
   "source": [
    "all_models = [\n",
    "    (\"w2v\", etree_w2v),\n",
    "    (\"w2v_tfidf\", etree_w2v_tfidf),\n",
    "]\n",
    "\n",
    "\n",
    "unsorted_scores = [(name, cross_val_score(model, X_train, y_train, cv=10).mean()) for name, model in all_models]\n",
    "scores = sorted(unsorted_scores, key=lambda x: -x[1])\n",
    "\n",
    "\n",
    "print (tabulate(scores, floatfmt=\".4f\", headers=(\"model\", 'score')))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Train the classifer with the best cross validation score."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "etree_w2v.fit(X_train,y_train)\n",
    "\n",
    "# save the model to disk\n",
    "pickle.dump(etree_w2v, open('noise_categories_model_etree_w2v.pkl', 'wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                         precision    recall  f1-score   support\n",
      "\n",
      "           Congregation       0.99      1.00      1.00      1880\n",
      "           Construction       1.00      1.00      1.00      2434\n",
      "Neighbourly Disturbance       1.00      0.99      1.00      1362\n",
      "                 Others       1.00      0.99      1.00      1047\n",
      "               Vehicles       1.00      1.00      1.00       643\n",
      "\n",
      "              micro avg       1.00      1.00      1.00      7366\n",
      "              macro avg       1.00      1.00      1.00      7366\n",
      "           weighted avg       1.00      1.00      1.00      7366\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# load the model from disk\n",
    "etree_w2v = pickle.load(open('noise_categories_model_etree_w2v.pkl', 'rb'))\n",
    "\n",
    "y_pred = etree_w2v.predict(X_train)\n",
    "\n",
    "print(classification_report(y_train, y_pred))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So, in general, when trained on the whole data, the accuracy is almost 100% for all categories. This clearly highlights that the model is __overfiiting__ the data as when cross-validated, the accuracy is almost 20% less. This is a known problem with tree-based classifiers. One way to mitigate this is to reduce the number of trees (at the moment, it has been set to 250)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## Running classifer on test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load the model from disk\n",
    "etree_w2v = pickle.load(open('noise_categories_model_etree_w2v.pkl', 'rb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_score = etree_w2v.predict_proba(X_test)\n",
    "y_pred = etree_w2v.predict(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                         precision    recall  f1-score   support\n",
      "\n",
      "           Congregation       0.78      0.84      0.81       483\n",
      "           Construction       0.88      0.96      0.92       581\n",
      "Neighbourly Disturbance       0.81      0.82      0.82       347\n",
      "                 Others       0.56      0.37      0.45       276\n",
      "               Vehicles       0.66      0.66      0.66       155\n",
      "\n",
      "              micro avg       0.79      0.79      0.79      1842\n",
      "              macro avg       0.74      0.73      0.73      1842\n",
      "           weighted avg       0.78      0.79      0.78      1842\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(classification_report(y_test, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}