1-feature-extraction-3features.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd,gensim,collections,numpy as np,math,matplotlib.pyplot as plt,re, string,os\n",
    "from collections import defaultdict\n",
    "from gensim import corpora, models, similarities\n",
    "from collections import Counter\n",
    "from utils.dataset import DataSet\n",
    "from nltk import word_tokenize, sent_tokenize\n",
    "import nltk, pickle\n",
    "%matplotlib inline\n",
    "\n",
    "stopwords = set(nltk.corpus.stopwords.words('english'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def clean(s):\n",
    "    return \" \".join(re.findall(r'[\\w\\']+', s, flags=re.UNICODE)).lower()\n",
    "\n",
    "def tfidf_vectorization(train_stance, dict_articles):\n",
    "    # panda dataframe which has columns: 'Headline', 'Body ID', 'Stance'\n",
    "    bodyids =  train_stance[\"Body ID\"].values\n",
    "    bodies = [ dict_articles[bodyid] for bodyid in dict_articles.keys() ]\n",
    "    stances = train_stance[\"Stance\"].values\n",
    "\n",
    "    # tokenize each body\n",
    "    words_per_body = []\n",
    "    vocab_per_body = [] # for idf\n",
    "    for body in bodies:\n",
    "        words = word_tokenize(clean(body))\n",
    "        words = [w for w in words if w not in string.punctuation and w not in stopwords]\n",
    "        words_per_body.append(words)\n",
    "        vocab_per_body += list(set(words))\n",
    "    \n",
    "    # tokenize each headlline\n",
    "    headlines = train_stance[\"Headline\"].values\n",
    "    tokens_per_headline = []\n",
    "\n",
    "    for head in headlines:\n",
    "        words = word_tokenize(clean(head))\n",
    "        words = [w for w in words if w not in string.punctuation and w not in stopwords]\n",
    "        tokens_per_headline.append(words)\n",
    "\n",
    "    # compute idf, only needs words set\n",
    "    body_corpus_count = Counter(vocab_per_body)\n",
    "    idf = {}\n",
    "    \n",
    "    for vocab in body_corpus_count.keys():\n",
    "        D = len(dict_articles.keys())\n",
    "        idf[vocab] = math.log( D / body_corpus_count[vocab])\n",
    "        \n",
    "    # calculate body tfidf and save in dict(body id)-dict(word) \n",
    "    tfidf_body_dict = {}\n",
    "    \n",
    "    for i, bodyid in enumerate(dict_articles.keys()):\n",
    "        tf_body = Counter(words_per_body[i])\n",
    "        total_words_in_body = len(words_per_body[i])\n",
    "        \n",
    "        tfidf_per_word_dict = {}\n",
    "        for word in tf_body.keys():\n",
    "            tf = tf_body[word]/total_words_in_body\n",
    "            tfidf = tf * idf[word]\n",
    "            tfidf_per_word_dict[word] = tfidf\n",
    "            \n",
    "        tfidf_body_dict[bodyid] = tfidf_per_word_dict\n",
    "        \n",
    "    # compute headline tfidf and save in dict - dict\n",
    "    tfidf_headline_dict = {}\n",
    "    \n",
    "    for i,(head,bodyID) in enumerate(zip(headlines, bodyids)):\n",
    "        current_headline_tfidf_dict = {}\n",
    "        tokens = tokens_per_headline[i]\n",
    "        token_counts = Counter(tokens)\n",
    "        \n",
    "        for word in tokens:\n",
    "            if word in idf:\n",
    "                tfidf = token_counts[word]/len(tokens)*idf[word]\n",
    "                current_headline_tfidf_dict[word] = tfidf\n",
    "        tfidf_headline_dict[i] = current_headline_tfidf_dict\n",
    "        \n",
    "    return tfidf_headline_dict, tfidf_body_dict, tokens_per_headline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def extract_features(headline_tfidf_dict, body_tfidf_dict, tokens_per_headline, bodyIDs):\n",
    "    feature_cosine, feature_dis, feature_kl = [], [], []\n",
    "    \n",
    "    # loop through each headline \n",
    "    for i, bodyid in enumerate(bodyIDs):\n",
    "        headline_tokens = tokens_per_headline[i]\n",
    "        current_head_tfidf_dict = headline_tfidf_dict[i]\n",
    "        current_body_tfidf_dict = body_tfidf_dict[bodyid]\n",
    "        \n",
    "        # loop through each word\n",
    "        cosine = 0\n",
    "        distance = 0\n",
    "        kl = 0\n",
    "        \n",
    "        for word in headline_tokens:\n",
    "            if word in current_body_tfidf_dict:\n",
    "                vec1 = current_head_tfidf_dict[word]\n",
    "                vec2 = current_body_tfidf_dict[word]\n",
    "                cosine += vec1*vec2\n",
    "                distance += (vec1-vec2)**2\n",
    "                kl += vec1 * np.log(vec1 / vec2)\n",
    "        # normalize vector dot        \n",
    "        norm_headline = np.linalg.norm(list(current_head_tfidf_dict.values()))\n",
    "        norm_body = np.linalg.norm(list(current_body_tfidf_dict.values()))\n",
    "        cosine = cosine / norm_headline / norm_body\n",
    "        feature_cosine.append(cosine)\n",
    "        \n",
    "        # square root distance\n",
    "        distance = math.sqrt(distance)\n",
    "        feature_dis.append(distance)\n",
    "        feature_kl.append(kl)\n",
    "        \n",
    "    return feature_cosine, feature_dis, feature_kl"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def feature_extraction_pipeline(train_stance, articles, file_save_path):\n",
    "        \n",
    "    headline_tfidf_dict, body_tfidf_dict, tokens_per_headline = tfidf_vectorization(train_stance, articles)\n",
    "    bodyids = train_stance['Body ID'].values\n",
    "    stances = train_stance['Stance'].values\n",
    "\n",
    "    feature_cos, feature_dis, feature_kl = extract_features(headline_tfidf_dict, body_tfidf_dict, tokens_per_headline, bodyids)\n",
    "    \n",
    "    train_df = pd.DataFrame(np.column_stack([feature_cos, feature_dis, feature_kl, stances]), columns=['cos','dis', 'kl', 'Stance'])\n",
    "    train_df.to_csv(file_save_path)\n",
    "    return feature_cos, feature_dis, feature_kl"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Reading dataset\n",
      "Total stances: 49972\n",
      "Total bodies: 1683\n"
     ]
    }
   ],
   "source": [
    "# process train data\n",
    "train = DataSet('train')\n",
    "train_stance = pd.read_csv('fnc-1/train_stance_pd.csv')\n",
    "cosine, dis, kl = feature_extraction_pipeline( train_stance, train.articles, 'train_processed_feat.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Reading dataset\n",
      "Total stances: 25413\n",
      "Total bodies: 904\n"
     ]
    }
   ],
   "source": [
    "test = DataSet('competition_test')\n",
    "# process test data\n",
    "test_stance = pd.read_csv('fnc-1/competition_test_stances.csv')\n",
    "cos2, dis2, kl2 = feature_extraction_pipeline( test_stance, test.articles, 'test_processed_feat.csv')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 验证 feature importance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "class_index = {'agree':[], 'disagree':[], 'discuss':[], 'unrelated':[]}\n",
    "\n",
    "for i,data in enumerate(train_stance.Stance):\n",
    "    class_index[data].append(i)\n",
    "\n",
    "f = kl\n",
    "agree_cosine = [ f[index] for index in class_index['agree']]\n",
    "disagree_cosine = [ f[index] for index in class_index['disagree']]\n",
    "discuss_cosine = [ f[index] for index in class_index['discuss']]\n",
    "unrelated_cosine = [ f[index] for index in class_index['unrelated']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,0.98,'kl')"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAagAAAEdCAYAAABZtfMGAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\npP8PR7GVuDIvxeQAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1205a74a8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "a = 0\n",
    "b = 15\n",
    "ax = plt.figure()\n",
    "plt.subplot(2, 2, 1)\n",
    "plt.hist(agree_cosine)\n",
    "plt.xlim(a,b)\n",
    "plt.xlabel('agree', fontsize = 15)\n",
    "plt.subplot(2, 2, 2)\n",
    "plt.hist(disagree_cosine)\n",
    "plt.xlim(a,b)\n",
    "\n",
    "plt.xlabel('disagree', fontsize = 15)\n",
    "plt.subplot(2, 2, 3)\n",
    "plt.hist(discuss_cosine)\n",
    "plt.xlim(a,b)\n",
    "\n",
    "plt.xlabel('discuss', fontsize = 15)\n",
    "plt.subplot(2, 2, 4)\n",
    "plt.hist(unrelated_cosine)\n",
    "plt.xlim(a,b)\n",
    "\n",
    "plt.xlabel('unrelated', fontsize = 15)\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.suptitle('kl', fontsize = 20)\n",
    "# plt.savefig('/Desktop/class_distribution_in_compound.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# stances = test_stance['Stance'].values\n",
    "# plt.figure()\n",
    "# plot_cos = []\n",
    "# feature = cos2\n",
    "# for (cos,stance) in zip(feature, stances):\n",
    "#     if stance != 'unrelated':\n",
    "#         plot_cos.append(cos)\n",
    "\n",
    "# for cos,stance in zip(feature, stances):\n",
    "#     if stance == 'unrelated':\n",
    "#         plot_cos.append(cos)\n",
    "\n",
    "# plt.plot(range(len(plot_cos)), plot_cos, 'g*')\n",
    "\n",
    "# plt.figure()\n",
    "# plot_cos = []\n",
    "# feature = cosine\n",
    "# stances = train_stance['Stance'].values\n",
    "# for cos,stance in zip(feature, stances):\n",
    "#     if stance != 'unrelated':\n",
    "#         plot_cos.append(cos)\n",
    "\n",
    "# for cos,stance in zip(feature, stances):\n",
    "#     if stance == 'unrelated':\n",
    "#         plot_cos.append(cos)\n",
    "\n",
    "# plt.plot(range(len(plot_cos)), plot_cos, 'g*')\n",
    "\n",
    "# plt.figure()\n",
    "# plot_cos = []\n",
    "# feature = kl\n",
    "# for cos,stance in zip(feature, stances):\n",
    "#     if stance != 'unrelated':\n",
    "#         plot_cos.append(cos)\n",
    "\n",
    "# for cos,stance in zip(feature, stances):\n",
    "#     if stance == 'unrelated':\n",
    "#         plot_cos.append(cos)\n",
    "\n",
    "# plt.plot(range(len(plot_cos)), plot_cos, 'g*')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "os.system('say \"you are very beautiful\"')"
   ]
  }
 ],
 "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": "3.6.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}