XGBoost_best.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 天池精准医疗大赛"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.linear_model import LinearRegression\n",
    "from sklearn import preprocessing\n",
    "import xgboost as xgb\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 导入数据"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train_df = pd.read_csv('data/d_train_20180102.csv', encoding='gbk')\n",
    "#train_df = pd.read_excel('data/d_train_20180120.xlsx', encoding='gbk')\n",
    "testA_df = pd.read_csv('data/d_test_A_20180102.csv', encoding='gbk')\n",
    "testB_df = pd.read_csv('data/d_test_B_20180128.csv', encoding ='gbk')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#训练集和测试集链接\n",
    "frames = [train_df, testA_df, testB_df]\n",
    "data_df = pd.concat(frames, join='outer',ignore_index = True)\n",
    "data_df = data_df.reindex_axis(train_df.columns, axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**增加7个特征**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "data_df['AST/ALT'] = data_df['*天门冬氨酸氨基转换酶']/data_df['*丙氨酸氨基转换酶']\n",
    "data_df['GFR'] = ((data_df['性别']=='男' )+ 0.742*(data_df['性别'] == '女')) * (186 * data_df['肌酐']/88.41 - 1.154*data_df['年龄']-0.203)\n",
    "data_df['中性粒细胞绝对值'] = data_df['白细胞计数'] * data_df['中性粒细胞%']/100\n",
    "data_df['淋巴细胞绝对值'] = data_df['白细胞计数'] * data_df['淋巴细胞%']/100\n",
    "data_df['单核细胞绝对值'] = data_df['白细胞计数'] * data_df['单核细胞%']/100\n",
    "data_df['嗜酸细胞绝对值'] = data_df['白细胞计数'] * data_df['嗜酸细胞%']/100\n",
    "data_df['嗜碱细胞绝对值'] = data_df['白细胞计数'] * data_df['嗜碱细胞%']/100"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**删除id, 体检日期两个特征，并且对性别特征one-hot编码**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\n",
    "onehot = pd.get_dummies(data_df['性别'])\n",
    "data_df = onehot.iloc[:,1:3].join(data_df) #由于id=580这一条数据的性别为：??,所以这里将其性别的one-hot编码为00\n",
    "#data_df['性别'] = data_df['性别'].map({'男':1, '女': 0})\n",
    "data_df.drop(['id','性别','体检日期'],axis=1,inplace=True)\n",
    "data_df.drop([ '乙肝表面抗原', '乙肝表面抗体', '乙肝e抗原', '乙肝e抗体', '乙肝核心抗体'],axis=1,inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#由于删除了血糖为38的一条样本，剩余5641条\n",
    "train_size = 5641  #原始数据：5642"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#重新排列列名，使得最后一列为血糖特征\n",
    "columns = ['女', '男', '年龄', '*天门冬氨酸氨基转换酶', '*丙氨酸氨基转换酶', '*碱性磷酸酶', '*r-谷氨酰基转换酶',\n",
    "       '*总蛋白', '白蛋白', '*球蛋白', '白球比例', '甘油三酯', '总胆固醇', '高密度脂蛋白胆固醇', '低密度脂蛋白胆固醇',\n",
    "       '尿素', '肌酐', '尿酸', '白细胞计数', '红细胞计数', '血红蛋白', '红细胞压积', '红细胞平均体积',\n",
    "       '红细胞平均血红蛋白量', '红细胞平均血红蛋白浓度', '红细胞体积分布宽度', '血小板计数', '血小板平均体积',\n",
    "       '血小板体积分布宽度', '血小板比积', '中性粒细胞%', '淋巴细胞%', '单核细胞%', '嗜酸细胞%', '嗜碱细胞%',\n",
    "     'AST/ALT', 'GFR', '中性粒细胞绝对值', '淋巴细胞绝对值', '单核细胞绝对值', '嗜酸细胞绝对值',\n",
    "       '嗜碱细胞绝对值','血糖']\n",
    "data_df = data_df.reindex_axis(columns, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "feat_generate = pd.DataFrame()\n",
    "feat_generate['22*25'] = data_df.iloc[:,22] * data_df.iloc[:,25]\n",
    "feat_generate['17*35'] = data_df.iloc[:,17] * data_df.iloc[:,35]\n",
    "feat_generate['2+30'] = data_df.iloc[:, 2] + data_df.iloc[:, 30]\n",
    "feat_generate['13*17'] = data_df.iloc[:,13] * data_df.iloc[:,17]\n",
    "feat_generate['2*19'] = data_df.iloc[:,2] * data_df.iloc[:,19]\n",
    "\n",
    "feat_generate['22divide24'] = data_df.iloc[:,22] / data_df.iloc[:,24]\n",
    "feat_generate['2divide35'] = data_df.iloc[:,2] / data_df.iloc[:,35]\n",
    "feat_generate['15divide17'] = data_df.iloc[:,15] / data_df.iloc[:,17]\n",
    "feat_generate['2divide25'] = data_df.iloc[:,2] / data_df.iloc[:,25]\n",
    "feat_generate['19divide25'] = data_df.iloc[:,19] / data_df.iloc[:,25]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "feat_generate[data_df.columns] = data_df \n",
    "data_df = feat_generate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "feat = ['2*19','甘油三酯','*天门冬氨酸氨基转换酶','*r-谷氨酰基转换酶','22*25', '2divide35','15divide17', '2+30',\\\n",
    "           '总胆固醇', '22divide24', '2divide25', '*碱性磷酸酶', '13*17', '高密度脂蛋白胆固醇', '17*35', 'AST/ALT',\\\n",
    "           '嗜碱细胞绝对值', '尿酸', '单核细胞绝对值', '嗜酸细胞绝对值', '*总蛋白', '血小板体积分布宽度', '血红蛋白',\\\n",
    "           'GFR', '红细胞压积', '白球比例', '血小板平均体积','中性粒细胞绝对值','低密度脂蛋白胆固醇', '红细胞平均血红蛋白浓度',\\\n",
    "          '*丙氨酸氨基转换酶', '中性粒细胞%', '19divide25', '白细胞计数', '*球蛋白', '尿素', '血小板比积', '肌酐', '血小板计数',\\\n",
    "          '淋巴细胞绝对值', '白蛋白', '淋巴细胞%', '单核细胞%', '红细胞平均体积', '红细胞计数', '嗜酸细胞%', '红细胞体积分布宽度', \\\n",
    "          '年龄', '红细胞平均血红蛋白量', '女', '嗜碱细胞%', '男']\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Program Files\\setup\\Anaconda\\lib\\site-packages\\ipykernel_launcher.py:2: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  \n"
     ]
    }
   ],
   "source": [
    "data_df1 = data_df[feat[0:32]]\n",
    "data_df1['血糖'] = data_df['血糖']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(7641, 53)"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#train_df = data_df.iloc[0:train_size,:]\n",
    "train_df = data_df1.iloc[0:6641,:]\n",
    "testA = data_df1.iloc[5641:6641,:]\n",
    "testB = data_df1.iloc[6641:,:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train: (6641, 33) \n",
      "testA： (1000, 33) \n",
      "testB: (1000, 33)\n"
     ]
    }
   ],
   "source": [
    "print('train:',train_df.shape,'\\ntestA：',testA.shape,'\\ntestB:',testB.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#将数据分开\n",
    "#训练\n",
    "x_train = train_df.iloc[:,0:-1]\n",
    "y_train = train_df.iloc[:,-1]\n",
    "#测试\n",
    "x_test = testB.iloc[:,0:-1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## XGBoost模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import mean_squared_error\n",
    "def xg_eval_mae(yhat, dtrain):\n",
    "    y = dtrain.get_label()\n",
    "    return 'mae', 1/2*mean_squared_error(y, yhat)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x_train:  (6641, 32) \n",
      "x_test: (1000, 32)\n"
     ]
    }
   ],
   "source": [
    "#将数据转化为xgboost对应的输入数据格式\n",
    "dtrain = xgb.DMatrix(x_train, y_train)\n",
    "dtest = xgb.DMatrix(x_test)\n",
    "print('x_train: ',x_train.shape,'\\nx_test:', x_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CV score: 0.852937111359\n",
      "train score: 0.600637373159\n"
     ]
    }
   ],
   "source": [
    "#xgboost 参数\n",
    "xgb_params = {\n",
    "    'seed': 0,\n",
    "    'eta': 0.02,\n",
    "    'learning_rate':0.1,\n",
    "    #'min_child_weight':3,\n",
    "    'colsample_bytree': 0.8,\n",
    "    'silent': 1,\n",
    "    'subsample': 0.6,\n",
    "    'objective': 'reg:linear',\n",
    "    'max_depth':3,     #避免过拟合，值越大，模型会学到更具体更局部的样本，一般3-5\n",
    "    'min_child_weight':3,\n",
    "    'reg_lambda':0.35,\n",
    "    'reg_alpha':10,\n",
    "    'gamma':0.2,\n",
    "}\n",
    "\n",
    "bst_cv1 = xgb.cv(xgb_params, dtrain, num_boost_round=100, nfold=5, seed=2, \n",
    "                    feval=xg_eval_mae, maximize=False)\n",
    "\n",
    "print ('CV score:', 1/2 * bst_cv1.iloc[-1,:]['test-rmse-mean']**2)\n",
    "print ('train score:', 1/2 * bst_cv1.iloc[-1,:]['train-rmse-mean']**2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "test-mae-mean      0.853745\n",
       "test-mae-std       0.052816\n",
       "test-rmse-mean     1.306091\n",
       "test-rmse-std      0.040203\n",
       "train-mae-mean     0.600697\n",
       "train-mae-std      0.012064\n",
       "train-rmse-mean    1.096027\n",
       "train-rmse-std     0.010948\n",
       "Name: 99, dtype: float64"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bst_cv1.iloc[-1,:]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**绘制误差曲线随着迭代次数的变化趋势**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x228c45b0908>"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.figure.Figure at 0x228c34dfba8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAD8CAYAAABn919SAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3Xl8VPW9//HXZ5asQAhJWAJIQBGBsO/luiCguKH21rZa\nq94uar3VLvdHxdveulz9tX301p/V3mq96MU+9KG23tpq8bZIhWqtiqCICAiiQcIatpCEbDPz/f0x\nkzAJAUIyzJlM3s/HI485c5bv9zMn5H0OZ85izjlERKTr83ldgIiIJIYCXUQkTSjQRUTShAJdRCRN\nKNBFRNKEAl1EJE0o0EVE0oQCXUQkTSjQRUTSRCCZnRUWFrqSkpJkdiki0uWtXr16r3Ou6ETzJTXQ\nS0pKWLVqVTK7FBHp8sxsa3vm0yEXEZE0ccJAN7PHzWyPma2LG/dTM9toZmvN7Hkz631qyxQRkRNp\nzx76YmBeq3EvA6XOubHAJuCOBNclIiIn6YTH0J1zr5pZSatxS+Pevgl8LrFlicixNDY2Ul5eTl1d\nndelSIJlZWUxaNAggsFgh5ZPxJeiXwGeTUA7ItIO5eXl9OzZk5KSEszM63IkQZxz7Nu3j/LycoYO\nHdqhNjr1paiZfR8IAU8dZ54bzWyVma2qqKjoTHciAtTV1VFQUKAwTzNmRkFBQaf+59XhQDezG4BL\ngS+54zz2yDn3qHNusnNuclHRCU+jFJF2UJinp87+Xjt0yMXM5gHfA851zh1u73KHK7WHLiJyqrTn\ntMWngTeAEWZWbmZfBX4B9AReNrM1ZvZIezqz2gOdKlZEvHfw4EF++ctfnvRyF198MQcPHjwFFUmT\n9pzlcnUbox/rSGc+F+rIYiKSQpoC/ZZbbmkxPhQKEQgcO1JeeumlDvfpnMM5h8+nayGPJ6lrx69A\nF+nyFi5cyJYtWxg/fjxTpkzh7LPPZv78+YwaNQqAK664gkmTJjF69GgeffTR5uVKSkrYu3cvZWVl\njBw5kq9//euMHj2aCy64gNra2qP6KSsrY8SIEVx33XWUlpaybds2evTowYIFCxg9ejRz5sxh5cqV\nnHfeeQwbNowXXngBgA8++ICpU6cyfvx4xo4dy+bNmwF48sknm8ffdNNNhMPho/pcvHgxV1xxBXPn\nzqWkpIRf/OIX3H///UyYMIHp06ezf/9+ALZs2cK8efOYNGkSZ599Nhs3bgTgxRdfZNq0aUyYMIE5\nc+awe/duAO666y6+8pWvNNf64IMPJvA3coQd5/vMhJtcHHBvbj1MIJiRtD5F0s2GDRsYOXIkAHe/\n+AHrdxxKaPujintx52Wjjzm9rKyMSy+9lHXr1rFixQouueQS1q1b13yq3f79++nTpw+1tbVMmTKF\nv/71rxQUFDTfy6m6upozzjiDVatWMX78eD7/+c8zf/58rr322qP6GTZsGH//+9+ZPn06EP3S8KWX\nXuKiiy7iyiuvpKamhiVLlrB+/Xquv/561qxZw6233sr06dP50pe+RENDA+FwmLKyMr73ve/xu9/9\njmAwyC233ML06dO57rrrWvS5ePFi7r33Xt59913q6uo444wz+MlPfsLNN9/Md77zHYYMGcK3v/1t\nZs+ezSOPPMLw4cN56623uOOOO3jllVc4cOAAvXv3xsxYtGgRGzZs4Gc/+xl33XUXS5cuZfny5VRV\nVTFixAh27drV5vnm8b/fJma22jk3+US/u6TenAscByt2Ulg8JLndisgpM3Xq1BbnTT/44IM8//zz\nAGzbto3NmzdTUFDQYpmhQ4cyfvx4ACZNmkRZWVmbbQ8ZMqQ5zAEyMjKYNy964fqYMWPIzMwkGAwy\nZsyY5jZmzJjBfffdR3l5OZ/97GcZPnw4f/nLX1i9ejVTpkwBoLa2lr59+7bZ56xZs+jZsyc9e/Yk\nLy+Pyy67rLm/tWvXUl1dzd///neuuuqq5mXq6+uB6DUCX/jCF9i5cycNDQ0t1ssll1xCZmYmmZmZ\n9O3bl927dzNo0KDjr9yTlORAh4MV5Qp0kQQ53p50suTm5jYPr1ixgmXLlvHGG2+Qk5PDeeed1+Z5\n1ZmZmc3Dfr+f2tpatm3b1hyeN998M/PmzWvRNkAwGGw+tc/n8zW34/P5CIWih3SvueYapk2bxpIl\nS7j44ov51a9+hXOO66+/nh/96Ect2nv++ee5++67AVi0aNFRtbXVRyQSoXfv3qxZs+aoz3Xrrbfy\n3e9+l/nz57NixQruuuuuY37mpnoTKenfMNTs35HsLkUkgXr27ElVVVWb0yorK8nPzycnJ4eNGzfy\n5ptvtrvdwYMHs2bNGtasWcPNN9/c4fo+/vhjhg0bxm233cbll1/O2rVrmT17Ns899xx79uwBooeF\ntm7dypVXXtnc5+TJJzyiAUCvXr0YOnQov/3tb4HoF7bvvfceEP38AwcOBOCJJ57o8GfoqKQHev3B\nncnuUkQSqKCggJkzZ1JaWsqCBQtaTJs3bx6hUIiRI0eycOHCFodLkuU3v/kNpaWljB8/nnXr1nHd\nddcxatQo7r33Xi644ALGjh3L3Llz2bmz41n01FNP8dhjjzFu3DhGjx7NH/7wByD65edVV13FpEmT\nKCwsTNRHarckfynqdw/96HZmXP9/k9anSLpp60szSR+d+VI0qXvoYXxY9Z5kdiki0m0kOdADBGt1\n+b+IyKmQ3EA3P9kNe5PZpYhIt5HUQHe+AD0b9yezSxGRbiO5gW5B8p1u0CUiciok97RFf4Ae1FJX\nk9hLlUVEJOmBHr1vwf4925ParYgkTkdvnwvwwAMPcPhwux+hICcpqYHuiwV61V4FukhXlYxAPxWX\nxXcHyQ30QDTQD+vyf5EuK/72uQsWLOCnP/0pU6ZMYezYsdx5550A1NTUcMkllzBu3DhKS0t59tln\nefDBB9mxYwezZs1i1qxZR7W7ePFi5s+fz/nnn8/s2bNZsWIF5557LpdffjnDhg1j4cKFPPXUU0yd\nOpUxY8awZcsWAH77299SWlrKuHHjOOeccwAIh8MsWLCgua5f/epXbX6WG264gW984xtMnz6dYcOG\nsWLFCr7yla8wcuRIbrjhhub5li5dyowZM5g4cSJXXXUV1dXVANxzzz1MmTKF0tJSbrzxRpou1Dzv\nvPO4/fbbmTp1KmeeeSavvfZawtb/8ST15lz+QPS2uY2VuvxfJCH+dyHsej+xbfYfAxf9+JiTf/zj\nH7Nu3TrWrFnD0qVLee6551i5ciXOOebPn8+rr75KRUUFxcXFLFmyBIje4yQvL4/777+f5cuXH/Oy\n+HfeeYe1a9fSp08fVqxYwXvvvceGDRvo06cPw4YN42tf+xorV67k5z//OQ899BAPPPAA99xzD3/+\n858ZOHBg8xORHnvsMfLy8nj77bepr69n5syZXHDBBS3uftjkwIEDvPHGG7zwwgvMnz+f119/nUWL\nFjFlyhTWrFnDoEGDuPfee1m2bBm5ubn85Cc/4f777+eHP/wh3/zmN/nhD38IwJe//GX++Mc/Nt9g\nLBQKsXLlSl566SXuvvtuli1b1qlfS3sk9wEXgSBhZ0QO7U5mtyJyiixdupSlS5cyYcIEJk6cyMaN\nG9m8eTNjxozh5Zdf5vbbb+e1114jLy+vXe3NnTuXPn36NL+fMmUKAwYMIDMzk9NPP50LLrgAoMXt\ncmfOnMkNN9zAf/3XfzU/tGLp0qX8+te/Zvz48UybNo19+/Y1P+iitcsuuwwzY8yYMfTr148xY8bg\n8/kYPXo0ZWVlvPnmm6xfv56ZM2cyfvx4nnjiCbZu3QrA8uXLmTZtGmPGjOGVV17hgw8+aG73s5/9\nLHD82wMnWlL30M2MA5aH77Au/xdJiOPsSSeDc4477riDm2666ahp77zzDi+99BI/+MEPmD17dvOe\nbJO2bl3b+na5J7qVLcAjjzzCW2+9xZIlS5g0aRKrV6/GOcdDDz3EhRde2KK973//+83/a2i6/W18\nm637C4VC+P1+5s6dy9NPP92irbq6Om655RZWrVrF4MGDueuuu1rcKriprVN1q9y2JP1ui5W+fDLq\ndLWoSFcVf/vcCy+8kMcff7z5mPL27dvZs2cPO3bsICcnh2uvvZYFCxbwzjvvHLVsR25d25YtW7Yw\nbdo07rnnHoqKiti2bRsXXnghDz/8MI2NjQBs2rSJmpoa7rvvvuY+22v69Om8/vrrfPTRR0D0+4FN\nmzY1h3dhYSHV1dU899xzHf4MiZL0B1zUBAvIbdiX7G5FJEHib5970UUXcc011zBjxgwAevTowZNP\nPslHH33EggUL8Pl8BINBHn74YQBuvPFG5s2bR3FxMcuXL09IPQsWLGDz5s0455g9ezbjxo1j7Nix\nlJWVMXHiRJxzFBUV8fvf/75D7RcVFbF48WKuvvrq5icT3XvvvZx55pl8/etfp7S0lP79+zc/DclL\nyb197uTJ7pdfOp0hh1bR784tSetXJJ3o9rnprcvcPhcglFNEfuQALhJJdtciImkt6YFuPfqRYWGq\nKnUcXUQkkZIe6IFe/QE4uKc82V2LpI1kHiqV5Ons7zXpgZ6VPwCAal3+L9IhWVlZ7Nu3T6GeZpxz\n7Nu3j6ysrA63ccKzXMzsceBSYI9zrjQ2rg/wLFAClAGfd65998XtURh9InbdAV0tKtIRgwYNory8\nnIoKPf0r3WRlZTFo0KAOL9+e0xYXA78Afh03biHwF+fcj81sYez97e3pML8oWmxj5a6TKlREooLB\nYJuXsIuc8JCLc+5VoPVjhi4HnogNPwFc0d4Oe/UuoN4FoVqX/4uIJFJHj6H3c841HTPZBfRrd4d+\nH/utN4HD+u+iiEgidfpLURf9ZuaY386Y2Y1mtsrMVjUd86sM9CGzXoEuIpJIHQ303WY2ACD2esy7\nbTnnHnXOTXbOTS4qKgLgcLCAXD0sWkQkoToa6C8A18eGrwf+cDILN2QVkRdWoIuIJNIJA93Mngbe\nAEaYWbmZfRX4MTDXzDYDc2Lv2y2SW0RvV0W4saEjNYuISBtOeNqic+7qY0ya3dFOrWd/fObYW1FO\nYfGwjjYjIiJxkn6lKEBmwWkAHNz5iRfdi4ikJU8CvWe/6EUR1XvKvOheRCQteRLoBQNPB6Bh36de\ndC8ikpY8CfT8/D5UuWzskG7QJSKSKJ4EuplR4e9LRs0OL7oXEUlLngQ6QGVGP3rU6wZdIiKJ4lmg\n12X3p09Il/+LiCSKZ4Ee7jmIfA7RWFftVQkiImnFs0AP5Efvi753+8delSAiklY8C/TswiEAVO4q\n86oEEZG04lmg5w2IXlx0uKLMqxJERNKKZ4FeVDyUiDPCB7Z5VYKISFrxLNBzc3LYa73xVelcdBGR\nRPAs0AH2+/uSdViBLiKSCJ4GelVmP/Ia9LBoEZFE8DTQG3KLKYzsAXfMR5KKiEg7eRroLm8gWTRS\nc/CYjyQVEZF28jTQM/pEz0Xfv2OLl2WIiKQFTwM9p28JAId0cZGISKd5Guj5/aMXF9Xt2+plGSIi\nacHTQO/bfyD1Log7WO5lGSIiacHTQA8G/Oy2QgLVenKRiEhneRroAAeDfcmp04MuREQ6y/NAr8ka\nQH6jTlsUEekszwM91LOYgsh+XKjB61JERLo0zwPd8gbhM0flHt11UUSkMzwP9MwCXVwkIpIInQp0\nM/uOmX1gZuvM7GkzyzrZNnr1LwGgek9ZZ0oREen2OhzoZjYQuA2Y7JwrBfzAF0+2nT7FwwBo3K+L\ni0REOqOzh1wCQLaZBYAc4KRvbl6Yn0+F643vwCedLEVEpHvrcKA757YD/wF8CuwEKp1zS1vPZ2Y3\nmtkqM1tVUVFxVDtmxq7AQHKrP+1oKSIiQucOueQDlwNDgWIg18yubT2fc+5R59xk59zkoqKiNts6\nlDOYggZd/i8i0hmdOeQyB/jEOVfhnGsEfgd8piMNNeYNpcAdIFxX1YlyRES6t84E+qfAdDPLMTMD\nZgMbOtJQsOh0ACq2dmhxERGhc8fQ3wKeA94B3o+19WhH2uo58CwADpRv7Gg5IiLdXqAzCzvn7gTu\n7GwR/YaMBKBu1+bONiUi0m15fqUoQN/CAva43nDgY69LERHpslIi0M2MPYFicqt0cZGISEelRKAD\nVOacRh+duigi0mEpE+iNeSUU6tRFEZEOS5lADxSdAcDeT3Wmi4hIR6RMoPcsjp66uH+bAl1EpCNS\nJtD7lUQDvW63Tl0UEemIlAn0vgWF7HG9sf160IWISEekTKD7fNFTF7N110URkQ5JmUAHqMw+jcJ6\nnbooItIRKRXojb1LKHD7idRVe12KiEiXk1KB7i+MnrpYoTNdREROWkoFeq/iMwE4oEAXETlpKRXo\nfUtGAVC7a5PHlYiIdD0pFej9CgupcHmY7rooInLSUirQfb7oA6OzdddFEZGTllKBDtG7LvatV6CL\niJyslAv0+j5nke8qaajc7XUpIiJdSsoFevbAUgB2bV7tcSUiIl1LygV60RkTAagse8/jSkREupaU\nC/Qhpw1ln+uF273e61JERLqUlAv0jICPTwMl9KjUuegiIicj5QId4GCv4fRvKINIxOtSRES6jJQM\ndFc4khzqqNmje6OLiLRXSgZ67mnjAJ3pIiJyMlIy0AcOnwBAzbb3Pa5ERKTr6FSgm1lvM3vOzDaa\n2QYzm5GIoor7FvKp64dvz4ZENCci0i0EOrn8z4E/Oec+Z2YZQE4CasLnM3ZmDmVgtR4YLSLSXh3e\nQzezPOAc4DEA51yDc+5gogqrzjuT/qFyCNUnqkkRkbTWmUMuQ4EK4L/N7F0zW2RmuQmqC1+/0QSI\nsH+rjqOLiLRHZwI9AEwEHnbOTQBqgIWtZzKzG81slZmtqqioaHfjeSXRM10qtqzpRIkiIt1HZwK9\nHCh3zr0Ve/8c0YBvwTn3qHNusnNuclFRUbsbP234GBqcn/rt2kMXEWmPDge6c24XsM3MRsRGzQYS\ndgOWwrwelNkgMvbpTBcRkfbo7FkutwJPxc5w+Rj4p86XdMSe7NM56/DaRDYpIpK2OhXozrk1wOQE\n1XKU2vwRFO54hUjNAXy5+aeqGxGRtJCSV4o2ySiOPuxizxbdAkBE5ERSOtALR0QvPD2w6U2PKxER\nSX0pHejDhw5lm+sL21d5XYqISMpL6UDPCPgoyxpJUaVOXRQROZGUDnSAmqLxFEb20nhwu9eliIik\ntJQP9Jxh0wDY+cHfPK5ERCS1pXygl4yeQYPzU7VFX4yKiBxPygf64L75bLYSMne963UpIiIpLeUD\n3czY1XM0xYc3QiTsdTkiIikr5QMdIDRgEjnUUlO+zutSRERSVpcI9PwzPwPoi1ERkePpEoE+YuR4\nDrpc6reu9LoUEZGU1SUCPS83g02BM8nb957XpYiIpKwuEegAB/LHMqBxK66+yutSRERSUpcJ9MBp\nU/ATYe+mt048s4hIN9RlAr3/qJkA7N34useViIikpi4T6MNLSvjYDcBfritGRUTa0mUCPSPg46Pc\nSQyufAfCjV6XIyKScrpMoAM0DjmHbOo49NEbXpciIpJyulSgD5xwIWFn7Hr3T16XIiKScrpUoJee\nfhofcDoZn77qdSkiIimnSwV6wO9jW++pDDr8AdQd8rocEZGU0qUCHcBOn0WACBXrXvG6FBGRlNLl\nAn34pFnUugwOrFvqdSkiIimlywX6GcWFrPGNpOcOXWAkIhKvywW6mbGncAYDGspwh3Z4XY6ISMro\ndKCbmd/M3jWzPyaioPbIPmsOADvX/DlZXYqIpLxE7KF/C9iQgHbarXTiZ9jnelK9flkyuxURSWmd\nCnQzGwRcAixKTDntU5yfy3vBcRRV/B2cS2bXIiIpq7N76A8A3wMiCajlpOwvPpf88H7qt76d7K5F\nRFJShwPdzC4F9jjnVp9gvhvNbJWZraqoqOhod0fpN+VKGpyfXW8+m7A2RUS6ss7soc8E5ptZGfAM\ncL6ZPdl6Jufco865yc65yUVFRZ3orqVpI09npY0hd8tLOuwiIkInAt05d4dzbpBzrgT4IvCKc+7a\nhFV2AhkBH9sHXEBh4w4at+tZoyIiXe489Hh9p1xJyPnY+cYzXpciIuK5hAS6c26Fc+7SRLR1MmaM\nGcFKRpOz+UUddhGRbq9L76FnBf1s7TeHwoZywrs+8LocERFPdelAByic/I9EnOmwi4h0e10+0D8z\nbiRvcxYZm5J25wERkZTU5QM9NzPAx4Vz6Fv3CZHdG70uR0TEM10+0AF6T/osYWfs/ttir0sREfFM\nWgT6zIljWMFEemx4BkINXpcjIuKJtAj0XllBPhnyBXqGDlC37gWvyxER8URaBDrAxPP/kW2RIg6+\n+iuvSxER8UTaBPqEIQUsy7mI/vtXQsUmr8sREUm6tAl0MyN3+g00OD8Vf33E63JERJIubQIdYN70\nsSxzU8nd8BtorPW6HBGRpEqrQO+VFeTTYV8kJ1xF7Zr/8bocEZGkSqtAB5h+/uVsiQzg8Gv/qRt2\niUi3knaBPm5wb17scRUFh9YT2bTU63JERJIm7QLdzBg256tsixRx6E/3ai9dRLqNtAt0gEvGD+E3\n2Z+n94G1uI+WeV2OiEhSpGWg+33G0Dlfo9wVUqm9dBHpJtIy0AHmTyzhmYzP0XvfGtyW5V6XIyJy\nyqVtoAf8Pkpm38gO10d76SLSLaRtoANcPmUoTwWvovfe1bgNL3pdjojIKZXWgR70+xg0+yY2RAZT\n9+ICqK/2uiQRkVMmrQMd4KqpQ1mUdxvZtbtoeOVHXpcjInLKpH2gB/w+rvncVTwdmkXgrYdh1zqv\nSxIROSXSPtABJg3J58PS73LQ5VD7+29BJOJ1SSIiCdctAh3g1kuncb9dR/auVbhVj3tdjohIwnWb\nQC/okcnIeTfyangMkT/dAbve97okEZGE6jaBDnD11CE80f9f2RvOpeHpL0PdIa9LEhFJmA4HupkN\nNrPlZrbezD4ws28lsrBTwecz/v1Ls7jD9238lVsJv3CbLjgSkbTRmT30EPAvzrlRwHTgn81sVGLK\nOnWKe2dz7Reu5meNV+Ff/zyseszrkkREEqLDge6c2+mceyc2XAVsAAYmqrBT6fyz+tE44zaWh8cR\neel22Kw7MopI15eQY+hmVgJMAN5qY9qNZrbKzFZVVFQkoruEWDBvFI/2/QEfRgYRefZa2Pa21yWJ\niHRKpwPdzHoA/wN82zl31LeMzrlHnXOTnXOTi4qKOttdwmQEfPy/685hQdadlIfyCD91FezZ6HVZ\nIiId1qlAN7Mg0TB/yjn3u8SUlDz987J44Ktzudl+wMF6CP/6Cti3xeuyREQ6pDNnuRjwGLDBOXd/\n4kpKrjP69uTfb7iEf2pcSE1NDZFFc6F8lddliYictM7soc8Evgycb2ZrYj8XJ6iupJo0pA+3XnMl\nV9bfxe76IG7xpfDh/3pdlojISenMWS5/c86Zc26sc2587OelRBaXTHNH9eMH11/G5xrvYWNkEO6Z\na+DNh3Weuoh0Gd3qStETmTWiL/9544V8lTv5q5sEf1oIz1wDNfu8Lk1E5IQU6K2MH9ybJ78xi3/L\nvoN7w9cR3vQy7pGZ8MmrXpcmInJcCvQ2DCvqwQvfPJuPT7+O+XV3s7suAE9cBs9/A6p2e12eiEib\nFOjHkJ+bwaLrJjN/3jzm1NzDk4Eribz/W3hoErz+IITqvS5RRKQFBfpx+HzGTeeezhM3ncfj2Tdw\nfu2PeT8wGl7+N3hwIry9SMEuIilDgd4Ok4b04X+/dTafm3sunzv0bb4a+T47XB9Y8i/w4AR445dw\neL/XZYpIN2cuiaflTZ482a1a1bUv2tm2/zD3LdnAnz7YydysDdzd648UH1oD/kwYdTlMvA6GzASf\ntpUikhhmtto5N/mE8ynQO2bd9koeWLaZZRt2MzmznO8VvcmkQ8vwNxyCHv1h5GXRgD9tBvgDXpcr\nIl2YAj1J1pYf5PG/fcKS93fiD9fxzeIPuSJjNQP3/g0L1UJmLyj5Bxh6Lgw9G4pGau9dRE6KAj3J\nKqrqefbtT3nm7W2UH6glz9/ATQM/4cKs9QypfJtAZVl0xsw8GDQZBk+FAeOgXynkDQIzT+sXkdSl\nQPeIc4412w7yx7U7WbJ2J7sO1QHwD4U1/GPBVib6NlNctZbg3o1AbN1n5UHRWdDndOgzDAqGQX4J\n5A+F7HyFvUg3p0BPAc45PtxdxWub9vLq5greLttPXWMEgOF5cGHRPqbm7OBMV0ZB3VaCBz+Bqp0t\nG8nsBXmDIW9gdE++ZzH0KILcvtCjL+T0gZyC6HwKfpG0pEBPQY3hCOt3HOLtsv288+kB1pZXUn6g\ntnl6YY9MxhQFmJJ3kBGZ+znNV0Hf0E561O/Cf6gcKrdD7TFOj/QFo3v6mT0hqxdk9ISMHAhmQzAX\nglkQbHqffWQ4kA2BjOhZOv4MCGRCICv2GhvXNN7nB/ODLxA37NeGROQUa2+g6/SLJAr6fYwb3Jtx\ng3s3jztQ08D72yv5cFcVm3ZXsWlPNb/ckE11fT+gH1AKQFHPTIp7ZzN4gI9hObUMzqimOFBNob+a\nfCrpGa4kK1yD1R+C+qroT/UeaKyFxsOx19gwCd6Imy8W8sHoqz8Qex84eiOAxTYAFlsutqz5Y+9j\nrxabbr7ovC36syPjj9qYxLcf30br2eLb8B3dZ/y4Fu0e1dCR/uLbbprW+r3FjYvvp0X7bfTVvH58\nsTuAuhP/Gi2urRaftdX6Od5nausztx7X1k7hUb+j9r42F96yn2Ot++ZprT9n0+/Ete+OqcfcKWnj\nd3jCZdpwMjvOrf89nQQFusfyczM458wizjnzyOP5nHPsq2lg674aPtl7mB0Ha9l+oJYdlbVsqKjl\nr1WNVNX5gbzYT/TZ3D6D/JwM+uRmkJ+TQa/sIHn5QfKyg/TKDtArK0ivrAB5GRF6+Rvp4W+kh6+B\nTAuT7QuTZY1kWAgLNUCoLvoTboRwQ/QnEoZICCKN0WEXiRt3jJ9wCFz4yHzxf2Dxy7tYe85Fx+GO\nTG/BxQVapNWk+KCLzeci0bZb/IG0asO1fo0cqSe+3daa23At2yZ+dtfGvHG1Edfn8foSaQcFegoy\nMwp7ZFLEKGuLAAAHIklEQVTYI5NJQ/q0Oc/hhhC7D9VTURX92VNVx/6aBvbVNHCgpoH9NQ2UHzjM\n+h2NVNY2UtPQOhiPLTvoJycjk+yMHDIDPrKCfrKCfjL8PjKDPjIDPjIDfrKC0WmZAR8ZAR8Zfj+Z\nWb64+fyx8T4yAhadHmx672ueFvT7CPqNYOx9wGf4fYbpUE6Ua9oAxDaMrfds217oGBuPNobb2uNu\na8MSP865VsudYIPZup6jXuP7oNX4Y21M2/icrfs84Xo6Rvut+2jx/jjLtNV0c9ft+fcct0MSP//d\nJzzaAijQu6ycjABDCwMMLcxt1/yhcITq+hCVtY1U1YWoqQ9R0xCipj5MbWOY2oYwhxuahkPNw/WN\nEeoaw9SFwjSEItTUhKhvjFAfClPXGKEuFKauMTotkuAdy6DfCMYCPuj3EYi9z4gbDvgMn8+aNwIB\nX3RaIDbsjxsO+Kz5vb95GR9+H9FXsyPLxtoO+A2/xeb3Gz6LtesjOhw3zuejeVpTe03L+X1H2vHF\nhn3GkWHfkXma2vFZdONO0w8+8AcTu5IlrSjQu4mA30fvnAx652Scsj5C4QgN4QgNoQj1oabXMA0h\n1zy+aVx9KEJjOEJj2MVeo9Maw46GUIRQJDocCkcIRaLzhJrmjTgaY8uHnSMccYTC0dfDoRDhiKMx\n9r4xEmmeFo44QhFHKBI58j7smttIRU0bLJ/RIvgDvuiGxB979fli080w48j4uI1IcxvxGw2f4Y/N\n3/RjFrfBsSPLNg03TbNYX764eS02HN+O3xedt2mDZXH9tZ7Hd6x+Y98HNL1vvWFs6rdpmfh5m9aJ\nGRjRz96in7h52myzefmj2/S1MY+n/1487V3SSsDvI+D3cQq3GaeMc46Ig1BsAxAKRzcGkdhGIH6D\nEHHR6RF39LhwJLqBOLJchHCEFhuRSKyvpuH48eEILdoNR6IbNBebP36Zphoi7sgyDoi06KeppqbP\nGJsWgdpwOFp73PxNn8XF1efi2o+v3cVNc3Cklrj5uqPWoe/3GQYtNhq+NjYarTcS8RvM9lKgixDb\ngzTw+/xk6q8iIVyrjU2LDVnThs9FNy5h13IDEZ0/2kY4fsPRqr1wpGU/TdObhh3R9ps2Ok3zOnek\nz3Ck5XJNbUXiamgaBxzpl9hykSM1Nm1QHUc2zvF9xr9vXiZ+Yxy3MW2q1zn4SzvXuf7pisgp0byR\nxAj6va6ma/vlte2bT3eJEhFJEwp0EZE0oUAXEUkTCnQRkTShQBcRSRMKdBGRNKFAFxFJEwp0EZE0\nkdQHXJhZFfBh0jpMfYXAXq+LSBFaFy1pfRyhdQFDnHNFJ5op2VeKftiep250F2a2SusjSuuiJa2P\nI7Qu2k+HXERE0oQCXUQkTSQ70B9Ncn+pTuvjCK2LlrQ+jtC6aKekfikqIiKnjg65iIikiaQEupnN\nM7MPzewjM1uYjD5TiZkNNrPlZrbezD4ws2/Fxvcxs5fNbHPsNd/rWpPFzPxm9q6Z/TH2vjuvi95m\n9pyZbTSzDWY2o5uvj+/E/k7WmdnTZpbVndfHyTjlgW5mfuA/gYuAUcDVZjbqVPebYkLAvzjnRgHT\ngX+OrYOFwF+cc8OJPpSkO23svgVsiHvfndfFz4E/OefOAsYRXS/dcn2Y2UDgNmCyc64U8ANfpJuu\nj5OVjD30qcBHzrmPnXMNwDPA5UnoN2U453Y6596JDVcR/YMdSHQ9PBGb7QngCm8qTC4zGwRcAiyK\nG91d10UecA7wGIBzrsE5d5Buuj5iAkC2mQWAHGAH3Xt9tFsyAn0gsC3ufXlsXLdkZiXABOAtoJ9z\nbmds0i6gn0dlJdsDwPeASNy47rouhgIVwH/HDkEtMrNcuun6cM5tB/4D+BTYCVQ655bSTdfHydKX\noklkZj2A/wG+7Zw7FD/NRU83SvtTjszsUmCPc271sebpLusiJgBMBB52zk0Aamh1OKE7rY/YsfHL\niW7oioFcM2vxRM3utD5OVjICfTswOO79oNi4bsXMgkTD/Cnn3O9io3eb2YDY9AHAHq/qS6KZwHwz\nKyN6+O18M3uS7rkuIPo/1nLn3Fux988RDfjuuj7mAJ845yqcc43A74DP0H3Xx0lJRqC/DQw3s6Fm\nlkH0C44XktBvyjAzI3qMdINz7v64SS8A18eGrwf+kOzaks05d4dzbpBzroTov4VXnHPX0g3XBYBz\nbhewzcxGxEbNBtbTTdcH0UMt080sJ/Z3M5vod07ddX2clKRcWGRmFxM9buoHHnfO3XfKO00hZvYP\nwGvA+xw5bvyvRI+j/wY4DdgKfN45t9+TIj1gZucB/8c5d6mZFdBN14WZjSf6BXEG8DHwT0R3trrr\n+rgb+ALRs8PeBb4G9KCbro+ToStFRUTShL4UFRFJEwp0EZE0oUAXEUkTCnQRkTShQBcRSRMKdBGR\nNKFAFxFJEwp0EZE08f8BbJzh7m3dgFQAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x228c459fe80>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import  matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "plt.figure()\n",
    "(bst_cv1[['train-rmse-mean', 'test-rmse-mean']]**2 * 0.5).plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 对测试集预测"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "bst = xgb.train(xgb_params, dtrain, num_boost_round=100) \n",
    "y_pred = bst.predict(dtest)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.figure.Figure at 0x228c4712cf8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAbcAAAJcCAYAAAB6/Oo3AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nkwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x228c47051d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import operator\n",
    "importance = bst.get_fscore()  \n",
    "importance = sorted(importance.items(), key=operator.itemgetter(1))  \n",
    "  \n",
    "df = pd.DataFrame(importance, columns=['feature', 'fscore'])  \n",
    "df['fscore'] = df['fscore'] / df['fscore'].sum()  \n",
    "df.to_csv(\"feat_combine_importance.csv\", index=False)  \n",
    "  \n",
    "plt.figure()  \n",
    "df.plot(kind='barh', x='feature', y='fscore', legend=False, figsize=(6, 10))  \n",
    "plt.title('XGBoost Feature Importance')  \n",
    "plt.xlabel('relative importance')  \n",
    "plt.show()  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "bst = xgb.train(xgb_params, dtrain, num_boost_round=100) \n",
    "y_pred = bst.predict(dtest)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "s = pd.Series(y_pred) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    1000.000000\n",
       "mean        5.590151\n",
       "std         0.691109\n",
       "min         4.555100\n",
       "25%         5.118604\n",
       "50%         5.417735\n",
       "75%         5.854177\n",
       "max        10.887362\n",
       "dtype: float64"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "s.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "#将预测值导出到文件\n",
    "s.to_csv('result/features_select_xgb.csv',index = 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"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}