Agenda.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# COMP9417 Group Project--Text Classification\n",
    "---"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import altair as alt\n",
    "import imblearn"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 0. Task\n",
    "1. Problem description\n",
    "    1. [x] 9000 train + 500 validating + 500 testing\n",
    "    - [x] 10 Classes + \"IRRELEVANT\"\n",
    "        1. How do we deal with \"IRRELEVANT\" in test set? Can we remove it and predict only 10 classes?\n",
    "    - [x] At least 2 classifiers\n",
    "    - [x] At most 10 suggestions above a threshold\n",
    "        1. Threshold may be relative to proportion of \"IRRELEVANT\" entries in test set        \n",
    "\n",
    "\n",
    "2. Exploratory Data Analysis\n",
    "    \n",
    "    1. [x] Data visualization\n",
    "        1. [x] Very unbalanced over classes -- Histogram\n",
    "        - [x] Some articles might have way larger vocabulary -- Box plots\n",
    "\n",
    "\n",
    "3. Feature Engineering\n",
    "    1. [x] Text representation\n",
    "        1. Word Count Vectors\n",
    "        - [x] TF-IDF Vectors -- balance word frequency and document size\n",
    "\n",
    "\n",
    "4. Label encoding\n",
    "    1. [x] Classes --> \\[0, 9\\]\n",
    "    \n",
    " \n",
    "5. Train/validation splitting\n",
    "\n",
    "\n",
    "6. GridSearchCV\n",
    "    1. [x] Models: **each with own parameters**\n",
    "        1. [x] Multinomial Naive Bayes\n",
    "        - [x] SVM\n",
    "        - KNN\n",
    "        - [x] Decision Tree / Random forest\n",
    "        - Grdient Boosting <p>...</p>\n",
    "    2. [x] Metrics\n",
    "        1. [x] Use accuracy for parameter tuning since False positive and False negative should be equally bad...\n",
    "        2. [x] Include others in report too\n",
    "        3. [x] Draw confusion matrix to visually illustrate\n",
    "        \n",
    "7. Predictions\n",
    "    1. [x] Model interpretation\n",
    "        1. [x] Need to explain misclassifications?\n",
    "    2. [x] Dimentionality reduction\n",
    "        1. [x] PCA\n",
    "        2. [x] t-SEN\n",
    "    3. [x] **Predicted Conditional Probabilities**\n",
    "        1. Set a threshold, suggest only those with a Conditional Probability above (at most 10)\n",
    "        - How to set a proper threshold? Has anything to do with \"IRRELEVANT\"?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Data Visulization\n",
    "Check the distribution of data over classes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "scrolled": true
   },
   "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>article_number</th>\n",
       "      <th>article_words</th>\n",
       "      <th>topic</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>open,absent,cent,cent,cent,stock,inflow,rate,k...</td>\n",
       "      <td>FOREX MARKETS</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>morn,stead,end,end,day,day,day,patch,patch,pat...</td>\n",
       "      <td>MONEY MARKETS</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>3</td>\n",
       "      <td>socc,socc,world,world,recent,law,fifa,fifa,fif...</td>\n",
       "      <td>SPORTS</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>4</td>\n",
       "      <td>open,forint,forint,forint,forint,cent,cent,ste...</td>\n",
       "      <td>FOREX MARKETS</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>5</td>\n",
       "      <td>morn,complet,weekend,minut,minut,minut,arrow,d...</td>\n",
       "      <td>IRRELEVANT</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   article_number                                      article_words  \\\n",
       "0               1  open,absent,cent,cent,cent,stock,inflow,rate,k...   \n",
       "1               2  morn,stead,end,end,day,day,day,patch,patch,pat...   \n",
       "2               3  socc,socc,world,world,recent,law,fifa,fifa,fif...   \n",
       "3               4  open,forint,forint,forint,forint,cent,cent,ste...   \n",
       "4               5  morn,complet,weekend,minut,minut,minut,arrow,d...   \n",
       "\n",
       "           topic  \n",
       "0  FOREX MARKETS  \n",
       "1  MONEY MARKETS  \n",
       "2         SPORTS  \n",
       "3  FOREX MARKETS  \n",
       "4     IRRELEVANT  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 9500 entries, 0 to 9499\n",
      "Data columns (total 3 columns):\n",
      "article_number    9500 non-null int64\n",
      "article_words     9500 non-null object\n",
      "topic             9500 non-null object\n",
      "dtypes: int64(1), object(2)\n",
      "memory usage: 222.8+ KB\n"
     ]
    }
   ],
   "source": [
    "df = pd.read_csv(\"training.csv\")#.set_index(\"article_number\")\n",
    "df.head()\n",
    "df.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "data = df[df.topic != \"IRRELEVANT\"].reset_index(drop=True)\n",
    "distribution = data.groupby(\"topic\").article_words.count().plot(kind=\"bar\", label=\"count\", legend=\"best\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. "
   ]
  }
 ],
 "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}