01_EDA.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import re\n",
    "\n",
    "import numpy as np\n",
    "\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import seaborn as sns\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "MAIN_DIR = os.path.relpath(os.path.dirname(os.path.abspath('__file__'))) # relative path of the main directory\n",
    "data_dir = DEFAULT_DATA_DIR = os.path.join(MAIN_DIR, \"data\") # relative path of data dir\n",
    "\n",
    "train_context_path = os.path.join(data_dir, \"train.context\")\n",
    "train_qn_path = os.path.join(data_dir, \"train.question\")\n",
    "train_ans_path = os.path.join(data_dir, \"train.span\")\n",
    "dev_context_path = os.path.join(data_dir, \"dev.context\")\n",
    "dev_qn_path = os.path.join(data_dir, \"dev.question\")\n",
    "dev_ans_path = os.path.join(data_dir, \"dev.span\")\n",
    "\n",
    "def split_by_whitespace(sentence):\n",
    "    words = []\n",
    "    for space_separated_fragment in sentence.strip().split():\n",
    "        words.extend(re.split(\" \", space_separated_fragment))\n",
    "    return [w for w in words if w]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_context = list(open(train_context_path))\n",
    "train_qn = list(open(train_qn_path))\n",
    "train_ans = list(open(train_ans_path))\n",
    "train_ans_span = list(map(lambda x:[int(x[0]), int(x[1])], map(split_by_whitespace, train_ans)))\n",
    "\n",
    "dev_context = list(open(dev_context_path))\n",
    "dev_qn = list(open(dev_qn_path))\n",
    "dev_ans = list(open(dev_ans_path))\n",
    "dev_ans_span = list(map(lambda x:[int(x[0]), int(x[1])], map(split_by_whitespace, dev_ans)))\n",
    "\n",
    "train_context_len = list(map(len, map(split_by_whitespace, train_context)))\n",
    "train_qn_len = list(map(len, map(split_by_whitespace, train_qn)))\n",
    "train_ans_len = list(map(lambda x: int(x[1]) - int(x[0]) + 1, map(split_by_whitespace, train_ans)))\n",
    "\n",
    "dev_context_len = list(map(len, map(split_by_whitespace, dev_context)))\n",
    "dev_qn_len = list(map(len, map(split_by_whitespace, dev_qn)))\n",
    "dev_ans_len = list(map(lambda x: int(x[1]) - int(x[0]) + 1, map(split_by_whitespace, dev_ans)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Size of training set: 86303\n",
      "Size of dev set: 10391\n"
     ]
    }
   ],
   "source": [
    "print('Size of training set: {}'.format(len(train_context)))\n",
    "print('Size of dev set: {}'.format(len(dev_context)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Size of training set: 86303\n",
      "Train context: min len 22; max len 766; >600: 19\n",
      "Train question: min len 1; max len 60; >30: 86\n",
      "Train answer: min len 1; max len 46; >30: 57\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x288 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('Size of training set: {}'.format(len(train_context)))\n",
    "\n",
    "plt.figure(figsize=(15, 4))\n",
    "\n",
    "plt.subplot(1, 3, 1)\n",
    "plt.hist(train_context_len);\n",
    "plt.title('train context');\n",
    "plt.xlabel('length');\n",
    "thres = 600\n",
    "print('Train context: min len {}; max len {}; >{}: {}'.\\\n",
    "      format(min(train_context_len), max(train_context_len), \\\n",
    "             thres, sum(map(lambda x: x >= thres, train_context_len))))\n",
    "\n",
    "plt.subplot(1, 3, 2)\n",
    "plt.hist(train_qn_len);\n",
    "plt.title('train question');\n",
    "plt.xlabel('length');\n",
    "thres = 30\n",
    "print('Train question: min len {}; max len {}; >{}: {}'.\\\n",
    "      format(min(train_qn_len), max(train_qn_len), \\\n",
    "             thres, sum(map(lambda x: x >= thres, train_qn_len))))\n",
    "\n",
    "\n",
    "plt.subplot(1, 3, 3)\n",
    "plt.hist(train_ans_len);\n",
    "plt.title('train answer');\n",
    "plt.xlabel('length');\n",
    "thres = 30\n",
    "print('Train answer: min len {}; max len {}; >{}: {}'.\\\n",
    "      format(min(train_ans_len), max(train_ans_len), \\\n",
    "             thres, sum(map(lambda x: x >= thres, train_ans_len))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(9, 4))\n",
    "\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.hist([x[0] / train_context_len[i] for i, x in enumerate(train_ans_span)]);\n",
    "plt.title('train answer start position in context');\n",
    "\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.hist([x[1] / train_context_len[i] for i, x in enumerate(train_ans_span)]);\n",
    "plt.title('train answer end position in context');"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The answer is more likely to appear at the beginning of the context."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Size of dev set: 10391\n",
      "Dev context: min len 24; max len 700; >600: 9\n",
      "Dev question: min len 3; max len 34; >30: 9\n",
      "Dev answer: min len 1; max len 37; >30: 1\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x288 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('Size of dev set: {}'.format(len(dev_context)))\n",
    "\n",
    "plt.figure(figsize=(15, 4))\n",
    "\n",
    "plt.subplot(1, 3, 1)\n",
    "plt.hist(dev_context_len);\n",
    "plt.title('dev context');\n",
    "plt.xlabel('length');\n",
    "thres = 600\n",
    "print('Dev context: min len {}; max len {}; >{}: {}'.\\\n",
    "      format(min(dev_context_len), max(dev_context_len), \\\n",
    "             thres, sum(map(lambda x: x >= thres, dev_context_len))))\n",
    "\n",
    "plt.subplot(1, 3, 2)\n",
    "plt.hist(dev_qn_len);\n",
    "plt.title('dev question');\n",
    "plt.xlabel('length');\n",
    "thres = 30\n",
    "print('Dev question: min len {}; max len {}; >{}: {}'.\\\n",
    "      format(min(dev_qn_len), max(dev_qn_len), \\\n",
    "             thres, sum(map(lambda x: x >= thres, dev_qn_len))))\n",
    "\n",
    "\n",
    "plt.subplot(1, 3, 3)\n",
    "plt.hist(dev_ans_len);\n",
    "plt.title('dev answer');\n",
    "plt.xlabel('length');\n",
    "thres = 30\n",
    "print('Dev answer: min len {}; max len {}; >{}: {}'.\\\n",
    "      format(min(dev_ans_len), max(dev_ans_len), \\\n",
    "             thres, sum(map(lambda x: x >= thres, dev_ans_len))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(9, 4))\n",
    "\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.hist([x[0] / dev_context_len[i] for i, x in enumerate(dev_ans_span)]);\n",
    "plt.title('dev answer start position in context');\n",
    "\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.hist([x[1] / dev_context_len[i] for i, x in enumerate(dev_ans_span)]);\n",
    "plt.title('dev answer end position in context');"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The statistics of dev set are similar to those of training set."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "py36",
   "language": "python",
   "name": "py36"
  },
  "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
}