xgboost-v2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# xgboost"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline\n",
    "import datetime\n",
    "import re"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "###データ読込(結合したデータ)\n",
    "train = pd.read_csv(\"~/repo/study/JLeague/data/train_new3.csv\")\n",
    "test = pd.read_csv(\"~/repo/study/JLeague/data/test_new3.csv\")\n",
    "player = pd.read_csv(\"~/repo/study/JLeague/data/player.csv\")\n",
    "test_id = pd.read_csv(\"~/repo/study/JLeague/motodata/test.csv\")[\"id\"]\n",
    "test_id_all = test[\"id\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature Enginiering"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 1952 entries, 0 to 1951\n",
      "Data columns (total 43 columns):\n",
      "id             1952 non-null int64\n",
      "y              1952 non-null int64\n",
      "year           1952 non-null int64\n",
      "stage          1952 non-null object\n",
      "match          1952 non-null object\n",
      "gameday        1952 non-null object\n",
      "time           1952 non-null object\n",
      "home           1952 non-null object\n",
      "away           1952 non-null object\n",
      "stadium        1952 non-null object\n",
      "tv             1952 non-null object\n",
      "home_score     1952 non-null int64\n",
      "away_score     1952 non-null int64\n",
      "weather        1952 non-null object\n",
      "temperature    1952 non-null float64\n",
      "humidity       1952 non-null object\n",
      "referee        1952 non-null object\n",
      "home_team      1952 non-null object\n",
      "home_01        1952 non-null object\n",
      "home_02        1952 non-null object\n",
      "home_03        1952 non-null object\n",
      "home_04        1952 non-null object\n",
      "home_05        1952 non-null object\n",
      "home_06        1952 non-null object\n",
      "home_07        1952 non-null object\n",
      "home_08        1952 non-null object\n",
      "home_09        1952 non-null object\n",
      "home_10        1952 non-null object\n",
      "home_11        1952 non-null object\n",
      "away_team      1952 non-null object\n",
      "away_01        1952 non-null object\n",
      "away_02        1952 non-null object\n",
      "away_03        1952 non-null object\n",
      "away_04        1952 non-null object\n",
      "away_05        1952 non-null object\n",
      "away_06        1952 non-null object\n",
      "away_07        1952 non-null object\n",
      "away_08        1952 non-null object\n",
      "away_09        1952 non-null object\n",
      "away_10        1952 non-null object\n",
      "away_11        1952 non-null object\n",
      "address        1952 non-null object\n",
      "capa           1952 non-null int64\n",
      "dtypes: float64(1), int64(6), object(36)\n",
      "memory usage: 655.8+ KB\n"
     ]
    }
   ],
   "source": [
    "train.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def j2num(j):\n",
    "    return 1 if j == 'Ｊ１' else 0\n",
    "train[\"stage\"] = train[\"stage\"].map(j2num)\n",
    "test[\"stage\"] = test[\"stage\"].map(j2num)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/tsu-nera/anaconda3/envs/opt/lib/python3.6/site-packages/ipykernel_launcher.py:4: DeprecationWarning: \n",
      ".ix is deprecated. Please use\n",
      ".loc for label based indexing or\n",
      ".iloc for positional indexing\n",
      "\n",
      "See the documentation here:\n",
      "http://pandas.pydata.org/pandas-docs/stable/indexing.html#ix-indexer-is-deprecated\n",
      "  after removing the cwd from sys.path.\n",
      "/home/tsu-nera/anaconda3/envs/opt/lib/python3.6/site-packages/ipykernel_launcher.py:12: DeprecationWarning: \n",
      ".ix is deprecated. Please use\n",
      ".loc for label based indexing or\n",
      ".iloc for positional indexing\n",
      "\n",
      "See the documentation here:\n",
      "http://pandas.pydata.org/pandas-docs/stable/indexing.html#ix-indexer-is-deprecated\n",
      "  if sys.path[0] == '':\n"
     ]
    }
   ],
   "source": [
    "for i, weather in enumerate(train[\"weather\"]):\n",
    "    weather = str(weather)\n",
    "    if re.match(\"雨\", weather):\n",
    "        train.ix[i, \"rain\"] = 1\n",
    "    else:\n",
    "        train.ix[i, \"rain\"] = 0\n",
    "for i, weather in enumerate(test[\"weather\"]):\n",
    "    weather = str(weather)\n",
    "    if re.match(\"雨\", weather):\n",
    "        test.ix[i, \"rain\"] = 1\n",
    "    else:\n",
    "        test.ix[i, \"rain\"] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for i, m in enumerate(train[\"match\"]):\n",
    "    match = re.findall(\"第(\\w+)節第(\\w+)日\", str(m))[0]\n",
    "    train.ix[i, \"match2\"] = int(match[0])\n",
    "    train.ix[i, \"match3\"] = int(match[1])\n",
    "for i, m in enumerate(test[\"match\"]):\n",
    "    match = re.findall(\"第(\\w+)節第(\\w+)日\", str(m))[0]\n",
    "    test.ix[i, \"match2\"] = int(match[0])\n",
    "    test.ix[i, \"match3\"] = int(match[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for i, tv in enumerate(train[\"tv\"]):\n",
    "    tv = str(tv)\n",
    "    tvlist = tv.split(\"／\")\n",
    "    train.ix[i, \"media\"] = int(len(tvlist))\n",
    "for i, tv in enumerate(test[\"tv\"]):\n",
    "    tv = str(tv)\n",
    "    tvlist = tv.split(\"／\")\n",
    "    test.ix[i, \"media\"] = int(len(tvlist))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "for i, day in enumerate(train[\"gameday\"]):\n",
    "    day = str(day)\n",
    "    train.ix[i, \"month\"] = int(day[:2])\n",
    "    train.ix[i, \"day\"] = int(day[3:5])\n",
    "    weekday = day[5:]\n",
    "    if re.match(\"\\(\\w・祝\\)\", weekday):\n",
    "        train.ix[i, \"weekday\"] = 0\n",
    "    elif re.match(\"\\(\\w・休\\)\", weekday):\n",
    "        train.ix[i, \"weekday\"] = 0\n",
    "    elif weekday == \"(土)\":\n",
    "        train.ix[i, \"weekday\"] = 1\n",
    "    elif weekday == \"(日)\":\n",
    "        train.ix[i, \"weekday\"] = 2\n",
    "    else:\n",
    "        train.ix[i, \"weekday\"] = 3\n",
    "for i, day in enumerate(test[\"gameday\"]):\n",
    "    day = str(day)\n",
    "    test.ix[i, \"month\"] = int(day[:2])\n",
    "    test.ix[i, \"day\"] = int(day[3:5])\n",
    "    weekday = day[5:]\n",
    "    if re.match(\"\\(\\w・祝\\)\", weekday):\n",
    "        test.ix[i, \"weekday\"] = 0\n",
    "    elif re.match(\"\\(\\w・休\\)\", weekday):\n",
    "        test.ix[i, \"weekday\"] = 0\n",
    "    elif weekday == \"(土)\":\n",
    "        test.ix[i, \"weekday\"] = 1\n",
    "    elif weekday == \"(日)\":\n",
    "        test.ix[i, \"weekday\"] = 2\n",
    "    else:\n",
    "        test.ix[i, \"weekday\"] = 3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from dateutil.parser import parse\n",
    "for i, t in enumerate(train[\"time\"]):\n",
    "    train.ix[i, \"hour\"] = int(str(t)[:2])\n",
    "for i, t in enumerate(test[\"time\"]):\n",
    "    test.ix[i, \"hour\"] = int(str(t)[:2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import LabelEncoder\n",
    "le = LabelEncoder()\n",
    "le.fit(train[\"home\"])\n",
    "train[\"home\"] = le.transform(train[\"home\"])\n",
    "train[\"away\"] = le.transform(train[\"away\"])\n",
    "test[\"home\"] = le.transform(test[\"home\"])\n",
    "test[\"away\"] = le.transform(test[\"away\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for i, t in enumerate(train[\"humidity\"]):\n",
    "    train.ix[i, \"humid\"] = int(str(t)[:2])\n",
    "for i, t in enumerate(test[\"humidity\"]):\n",
    "    test.ix[i, \"humid\"] = int(str(t)[:2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def discomfort_index(T, H):\n",
    "    return 0.81*T+0.01*H*(0.99*T - 14.3) + 46.3\n",
    "train[\"discomfort\"] = discomfort_index(train[\"temperature\"], train[\"humid\"])\n",
    "test[\"discomfort\"] = discomfort_index(test[\"temperature\"], test[\"humid\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for i in range(1,12):\n",
    "    column = \"home_%02d\" % i\n",
    "    for row, name in enumerate(train[column]):\n",
    "        salary = player[player.player_name==name].annual_salary.values\n",
    "        train.ix[row, column] = 2500000 if salary.size == 0 else salary\n",
    "    column = \"away_%02d\" % i\n",
    "    for row, name in enumerate(train[column]):\n",
    "        salary = player[player.player_name==name].annual_salary.values\n",
    "        train.ix[row, column] = 2500000 if salary.size == 0 else salary\n",
    "for i in range(1,12):\n",
    "    column = \"home_%02d\" % i\n",
    "    for row, name in enumerate(test[column]):\n",
    "        salary = player[player.player_name==name].annual_salary.values\n",
    "        test.ix[row, column] = 2500000 if salary.size == 0 else salary\n",
    "    column = \"away_%02d\" % i\n",
    "    for row, name in enumerate(test[column]):\n",
    "        salary = player[player.player_name==name].annual_salary.values\n",
    "        test.ix[row, column] = 2500000 if salary.size == 0 else salary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train[\"home_all\"] = train[[\"home_01\",\"home_02\",\"home_03\",\"home_04\",\"home_05\",\"home_06\",\"home_07\",\"home_08\",\"home_09\",\"home_10\",\"home_11\"]].sum(axis=1) \n",
    "train[\"away_all\"] = train[[\"away_01\",\"away_02\",\"away_03\",\"away_04\",\"away_05\",\"away_06\",\"away_07\",\"away_08\",\"away_09\",\"away_10\",\"away_11\"]].sum(axis=1) \n",
    "test[\"home_all\"] = test[[\"home_01\",\"home_02\",\"home_03\",\"home_04\",\"home_05\",\"home_06\",\"home_07\",\"home_08\",\"home_09\",\"home_10\",\"home_11\"]].sum(axis=1) \n",
    "test[\"away_all\"] = test[[\"away_01\",\"away_02\",\"away_03\",\"away_04\",\"away_05\",\"away_06\",\"away_07\",\"away_08\",\"away_09\",\"away_10\",\"away_11\"]].sum(axis=1) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_team = pd.get_dummies(train[\"home_team\"]) + pd.get_dummies(train[\"away_team\"])\n",
    "test_team = pd.get_dummies(test[\"home_team\"]) + pd.get_dummies(test[\"away_team\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['アビスパ福岡', 'アルビレックス新潟', 'カターレ富山', 'カマタマーレ讃岐', 'ガイナーレ鳥取', 'ガンバ大阪',\n",
       "       'ギラヴァンツ北九州', 'コンサドーレ札幌', 'サガン鳥栖', 'サンフレッチェ広島', 'ザスパ草津', 'ジェフユナイテッド千葉',\n",
       "       'ジュビロ磐田', 'セレッソ大阪', 'ファジアーノ岡山', 'ベガルタ仙台', 'モンテディオ山形', 'ロアッソ熊本',\n",
       "       'ヴァンフォーレ甲府', 'ヴィッセル神戸', '京都サンガF.C.', '名古屋グランパス', '大分トリニータ', '大宮アルディージャ',\n",
       "       '川崎フロンターレ', '徳島ヴォルティス', '愛媛ＦＣ', '東京ヴェルディ', '松本山雅ＦＣ', '柏レイソル', '栃木ＳＣ',\n",
       "       '横浜Ｆ・マリノス', '横浜ＦＣ', '水戸ホーリーホック', '浦和レッズ', '清水エスパルス', '湘南ベルマーレ',\n",
       "       '鹿島アントラーズ', 'ＦＣ岐阜', 'ＦＣ東京', 'ＦＣ町田ゼルビア', 'Ｖ・ファーレン長崎'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_team.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['アビスパ福岡', 'アルビレックス新潟', 'カターレ富山', 'カマタマーレ讃岐', 'ガンバ大阪', 'ギラヴァンツ北九州',\n",
       "       'コンサドーレ札幌', 'サガン鳥栖', 'サンフレッチェ広島', 'ザスパ草津', 'ジェフユナイテッド千葉', 'ジュビロ磐田',\n",
       "       'セレッソ大阪', 'ファジアーノ岡山', 'ベガルタ仙台', 'モンテディオ山形', 'ロアッソ熊本', 'ヴァンフォーレ甲府',\n",
       "       'ヴィッセル神戸', '京都サンガF.C.', '名古屋グランパス', '大分トリニータ', '大宮アルディージャ', '川崎フロンターレ',\n",
       "       '徳島ヴォルティス', '愛媛ＦＣ', '東京ヴェルディ', '松本山雅ＦＣ', '柏レイソル', '栃木ＳＣ', '横浜Ｆ・マリノス',\n",
       "       '横浜ＦＣ', '水戸ホーリーホック', '浦和レッズ', '清水エスパルス', '湘南ベルマーレ', '鹿島アントラーズ', 'ＦＣ岐阜',\n",
       "       'ＦＣ東京', 'Ｖ・ファーレン長崎'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 100,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_team.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_team['ガイナーレ鳥取'] = 0\n",
    "test_team['ＦＣ町田ゼルビア'] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import PolynomialFeatures\n",
    "poly = PolynomialFeatures(degree=2, include_bias=False)\n",
    "poly.fit(train_team)\n",
    "train_poly = poly.transform(train_team)\n",
    "train_poly_df = pd.DataFrame(columns=poly.get_feature_names(),data=train_poly)\n",
    "test_poly = poly.transform(test_team)\n",
    "test_poly_df = pd.DataFrame(columns=poly.get_feature_names(),data=test_poly)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train1 = train[[\"y\", \"year\", \"capa\", \"home_score\", \"away_score\", \"month\", \"day\", \"weekday\", \"home\", \"away\", \"media\", \"discomfort\", \"temperature\", \"humid\", \"home_all\", \"away_all\", \"stage\"]]\n",
    "test1 = test[[\"capa\", \"home_score\", \"away_score\",  \"month\", \"day\", \"weekday\", \"home\", \"away\", \"media\", \"discomfort\", \"temperature\", \"humid\", \"home_all\", \"away_all\", \"stage\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y1 = train1[\"y\"]\n",
    "train1 = train1.drop([\"y\", \"year\"], axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_df = pd.concat([train1, train_poly_df], axis=1)\n",
    "test_df = pd.concat([test1, test_poly_df], axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "X_train, X_test, y_train, y_test = train_test_split(train_df, y1, test_size=0.2, random_state=17)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## build model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import xgboost\n",
    "from xgboost import XGBRegressor\n",
    "from numpy import sort\n",
    "from sklearn.feature_selection import SelectFromModel\n",
    "from sklearn.metrics import mean_squared_error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "XGBRegressor(base_score=0.5, booster='gbtree', colsample_bylevel=1,\n",
       "       colsample_bytree=1, eval_metrics='rmse', gamma=0, learning_rate=0.1,\n",
       "       max_delta_step=0, max_depth=3, min_child_weight=1, missing=None,\n",
       "       n_estimators=100, n_jobs=1, nthread=None, objective='reg:linear',\n",
       "       random_state=17, reg_alpha=0, reg_lambda=1, scale_pos_weight=1,\n",
       "       seed=None, silent=True, subsample=1)"
      ]
     },
     "execution_count": 108,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = XGBRegressor(eval_metrics='rmse', random_state=17)\n",
    "model.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pred = model.predict(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3204.6047864664692"
      ]
     },
     "execution_count": 110,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.sqrt(mean_squared_error(y_test, pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f04e715fb38>"
      ]
     },
     "execution_count": 111,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAbQAAAEaCAYAAAB5IqLYAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\naII0NCGEEJogDU0IIYQmSEMTQgihCf8DiiMvg4GkDKUAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f04e71d4fd0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.style.use('ggplot')\n",
    "xgboost.plot_importance(model, max_num_features=20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## submit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred = model.predict(test_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/tsu-nera/anaconda3/envs/opt/lib/python3.6/site-packages/ipykernel_launcher.py:2: DeprecationWarning: \n",
      ".ix is deprecated. Please use\n",
      ".loc for label based indexing or\n",
      ".iloc for positional indexing\n",
      "\n",
      "See the documentation here:\n",
      "http://pandas.pydata.org/pandas-docs/stable/indexing.html#ix-indexer-is-deprecated\n",
      "  \n"
     ]
    }
   ],
   "source": [
    "for i, p in enumerate(pred):\n",
    "    if p > test1.ix[i, \"capa\"]:\n",
    "        pred[i] = test1.ix[i, \"capa\"]\n",
    "    else:\n",
    "        pred[i] = p"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for i, p in enumerate(pred):\n",
    "    if p < min(y1):\n",
    "        print(p, min(y1))\n",
    "        pred[i] = min(y1)\n",
    "    else:\n",
    "        pred[i] = p"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pred = pd.DataFrame(pred, columns={\"pred\"})\n",
    "pred[\"id\"] = test_id_all"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "test_id = pd.DataFrame(test_id)\n",
    "test_id[\"id\"] = test_id"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pred = pd.merge(pred, test_id, how=\"right\", on=[\"id\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pred = pred.loc[:, [\"id\", \"pred\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pred.to_csv(\"~/repo/study/JLeague/submit/submit_1126_1_xgboost.csv\", index=False, header=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.6.1"
  },
  "toc": {
   "colors": {
    "hover_highlight": "#DAA520",
    "navigate_num": "#000000",
    "navigate_text": "#333333",
    "running_highlight": "#FF0000",
    "selected_highlight": "#FFD700",
    "sidebar_border": "#EEEEEE",
    "wrapper_background": "#FFFFFF"
   },
   "moveMenuLeft": true,
   "nav_menu": {
    "height": "38px",
    "width": "253px"
   },
   "navigate_menu": true,
   "number_sections": false,
   "sideBar": true,
   "threshold": 4,
   "toc_cell": false,
   "toc_section_display": "block",
   "toc_window_display": false,
   "widenNotebook": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}