advanced.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import os,sys\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Загрузим данные\n",
    " * При первом запуске скачаем train.csv и evaluation.csv\n",
    " * При повторных запусках файлы уже будут на месте"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from datanight import download_datasets\n",
    "download_datasets()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "data = pd.DataFrame.from_csv(\"./train.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#Если у вас слабый компьютер - можно взять подвыборку данных\n",
    "#data = data[::2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "x_col = [ 'dist','due', 'lat','lon','f_class','s_class','t_class',]\n",
    "X_raw = data[x_col]\n",
    "y = data['burned'].values\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Используем время и внешние данные\n",
    "* признаки - \n",
    " * день недели (1-7 от понедельника до воскресенья категориальными признаками) \n",
    " * время дня (вещественное число от 0 до 1 - от 00.00 до 23.59)\n",
    "* абсолютное время не используем т.к. обучение и тест не пересекаются по времени\n",
    "* отдельно посмотрим список праздников в этом году и сделаем из этого отдельный признак"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#Используем данные о праздниках отсюда - http://eduscan.net/help/calendar2014 http://newslab.ru/article/559455\n",
    "\n",
    "holidays = \"1.01,2.01,3.01,4.01,5.01,6.01,7.01,8.01,\"\\\n",
    "\"23.02,8.03,9.03,10.03,1.05,2.05,3.05,4.05,9.05,10.05,11.05,12.06,13.06\".split(',')\n",
    "\n",
    "holidays = map(lambda s: tuple(map(int,s.split('.'))),holidays)\n",
    "\n",
    "#вытаскиватель категориальных фичей\n",
    "from sklearn.feature_extraction import DictVectorizer\n",
    "vectorizer = DictVectorizer(sparse=False,dtype=np.bool)\n",
    "\n",
    "# используем встроенный питоновый парсер времени\n",
    "from dateutil import parser\n",
    "\n",
    "def preprocess_data(X_raw):\n",
    "\n",
    "    #распарсим время\n",
    "    datetimes = list(X_raw.due.apply(lambda x: parser.parse(x)))\n",
    "\n",
    "    #время дня от 0 до 1\n",
    "    rel_times = map(lambda dt: (dt.hour*60 + dt.minute)/(24.*60), datetimes)\n",
    "\n",
    "    #день недели\n",
    "    week_days = map( lambda x: x.isoweekday() , datetimes)\n",
    "\n",
    "    rel_times = np.array(rel_times)\n",
    "    week_days = np.array(week_days)\n",
    "    \n",
    "    is_holiday = map(lambda dt: (dt.day,dt.month) in holidays,\n",
    "                 datetimes)\n",
    "\n",
    "    dow_names = ['mon','tue','wed','thu','fri','sat','sun']\n",
    "    \n",
    "    data_dict = ( {'f_class':f,'s_class':s,'t_class':t,'day_of_week':dow_names[day_of_week-1]}\n",
    "               for (f,s,t),day_of_week in zip(X_raw[['f_class','s_class','t_class']].values, week_days) )\n",
    "\n",
    "    Xcat = vectorizer.fit_transform(data_dict)\n",
    "\n",
    "    other_features = [\"dist\",\"lat\",\"lon\"]\n",
    "    \n",
    "    Xreal = X_raw[other_features].values\n",
    "    \n",
    "    \n",
    "    Xfull = np.concatenate([\n",
    "            Xreal,\n",
    "            Xcat            \n",
    "        ],axis=1)\n",
    "    \n",
    "    Xfull = pd.DataFrame(Xfull,columns=other_features+vectorizer.feature_names_)\n",
    "    \n",
    "    Xfull[\"time_of_day_rel\"] = rel_times\n",
    "    Xfull[\"is_holiday\"] = is_holiday\n",
    "\n",
    "    \n",
    "    return Xfull"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#обработаем данные\n",
    "x_col = [ 'dist','due', 'lat','lon','f_class','s_class','t_class',]\n",
    "X = data[x_col]\n",
    "y = data['burned'].values\n",
    "\n",
    "X = preprocess_data(X)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Поделим выборку на обучение и тест в хронологическом порядке"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#обучение и тест\n",
    "pivot = int(len(X) * 0.75)\n",
    "Xtr = X[:pivot]\n",
    "Xval = X[pivot:]\n",
    "Ytr = y[:pivot]\n",
    "Yval = y[pivot:]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Обучим модель\n",
    "* Всё тот же случайный лес, только с немного изменёнными параметрами"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini',\n",
       "            max_depth=None, max_features='auto', max_leaf_nodes=None,\n",
       "            min_samples_leaf=20, min_samples_split=75,\n",
       "            min_weight_fraction_leaf=0.0, n_estimators=100, n_jobs=-1,\n",
       "            oob_score=False, random_state=None, verbose=0,\n",
       "            warm_start=False)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "\n",
    "#Также обучим более \"случайную\" версию RandomForest\n",
    "model = RandomForestClassifier(n_estimators=100,n_jobs=-1,min_samples_split=75,min_samples_leaf=20)\n",
    "\n",
    "model.fit(Xtr,Ytr)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train: \t AUC = 0.75770 \t Accuracy = 0.28933\n",
      "val: \t AUC = 0.53951 \t Accuracy = 0.27600\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f7a1263a390>"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEACAYAAABI5zaHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7a1260c8d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import roc_auc_score,accuracy_score,precision_score,recall_score\n",
    "from sklearn.metrics import roc_curve\n",
    "\n",
    "\n",
    "for source_i, Xi,Yi in [\n",
    "                            [\"train\",Xtr,Ytr],\n",
    "                            [\"val\",Xval,Yval]\n",
    "                                ]:\n",
    "    \n",
    "    # Предскажем вероятность сгорания\n",
    "    Yi_pred_proba = model.predict_proba(Xi)[:,1]\n",
    "    \n",
    "    #Поделим предсказание на сгоревшие и не сгоревшие\n",
    "    Yi_pred_class = np.argsort(Yi_pred_proba) < 10000\n",
    "    \n",
    "    auc = roc_auc_score(Yi,Yi_pred_proba)\n",
    "    acc = accuracy_score(Yi,Yi_pred_class)\n",
    "    \n",
    "    print '%s: \\t AUC = %.5f \\t Accuracy = %.5f'%(source_i, auc, acc)\n",
    "    \n",
    "    fpr,tpr,_ = roc_curve(Yi,Yi_pred_proba)\n",
    "    \n",
    "    plt.plot(fpr,tpr,label = source_i)\n",
    "    \n",
    "plt.legend()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Feature ranking:\n",
      "1. dist (0.299162)\n",
      "2. lat (0.246287)\n",
      "3. lon (0.241503)\n",
      "4. day_of_week=wed (0.116396)\n",
      "5. f_class=business (0.020024)\n",
      "6. f_class=econom (0.015453)\n",
      "7. f_class=vip (0.014647)\n",
      "8. s_class (0.006529)\n",
      "9. s_class=business (0.000000)\n",
      "10. s_class=econom (0.000000)\n",
      "11. s_class=vip (0.000000)\n",
      "12. t_class (0.000000)\n",
      "13. t_class=business (0.000000)\n",
      "14. t_class=econom (0.000000)\n",
      "15. t_class=vip (0.000000)\n",
      "16. time_of_day_rel (0.000000)\n",
      "17. is_holiday (0.000000)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/lib/pymodules/python2.7/matplotlib/collections.py:548: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison\n",
      "  if self._edgecolors == 'face':\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAEKCAYAAAAGvn7fAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nrkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7a125b7a50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from datanight import plot_forest_feature_importances\n",
    "plot_forest_feature_importances(model,X.columns)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from datanight import generate_submission\n",
    "generate_submission(preprocess_data,model,\"submission.csv\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Советы по улучшению модели:\n",
    "\n",
    "* Смотри глазами на данные - строй графики, исследуй аномалия \n",
    " * это может принести новые идеи и отбросить нежизнеспособные намного раньше \n",
    " * Туториал по библиотеке, в которой можно строить графики \n",
    "   * http://matplotlib.org/users/pyplot_tutorial.html\n",
    " * Ключевой вопрос самому себе - __\"От чего ещё может зависеть, сгорит ли заказ?\"__\n",
    "\n",
    "\n",
    "* Попробуй более точно настроить модель или выбрать другую\n",
    " * Random Forest с текущими параметрами можно улучшить\n",
    "   * Документация по нему - http://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html\n",
    " * А можно использовать другие, более подходящие методы\n",
    "   * Документация по методам - http://scikit-learn.org/stable/supervised_learning.html\n",
    " * Наконец, модели можно комбинировать\n",
    "   * Например, усреднять с весами\n",
    "   * В итоге качество может быть лучше, чем у каждой модели по отдельности\n",
    " * Важно не увлечься - зачастую качественные изменения решения намного лучше подкручивания параметров\n",
    "\n",
    "\n",
    "* Подумай, какие ещё дополнительные данные можно провязать с выборкой? \n",
    " * насколько реально применить их за отведённое время?\n",
    "\n",
    "\n",
    "* __Главное__ - пытайся понять, дадут ли твои улучшения прирост на новых данных\n",
    " * Тестовая выборка находится по времени дальше, чем обучающая.\n",
    "   * Например, нельзя использовать номер дня в году, потому что мы обучаемся на выборке __ДО__ начала контрольной\n",
    " * Это не значит, что нужно отправлять миллион решений на проверку и выбирать лучшее - это приведёт к переобучению\n",
    " "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}