Task_06_Financial_Data_Analysis—Model_Integration.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 一、模型融合\n",
    "\n",
    "- **模型融合（Model Integration）**\n",
    "\n",
    "> **对Task 05 调优后的模型, 进行模型融合。例如, 用目前评分最高的模型作为基准模型, 和其他模型进行Stacking 模型融合, 得到最终模型和评分**\n",
    "\n",
    "> **模型融合，通常可以在各种不同的机器学习任务中使结果获得提升，即综合考虑不同模型的情况，并将它们的结果融合到一起**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### 多数表决融合: 模型融合能提高准确率以及对低相关的结果进行融合可以获得更好结果"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> **多数表决的融合方式默认了所有模型的重要度是一样的，但通常情况下我们会更重视表现较好的模型而需要赋予更大的权值**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### 加权表决融合: 在加权表决的情况下，表现较差的模型只能通过与其他模型获得一样的结果来增强自己的说服力"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### 对结果取平均"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> **对结果取平均在很多机器学习问题上以及不同的评估准则上都获得很不错的结果**\n",
    "\n",
    "> **取均值的做法常常可以减少过拟合现象**\n",
    "\n",
    "> **在机器学习的应用上，过拟合现象是很普遍的，根本问题是训练数据量不足以支撑复杂的模型，导致模型学习到数据集上的噪音，这样产生的问题使得模型很难泛化，因为模型“考虑”的过分片面**\n",
    "\n",
    "> **如果对结果取平均，可以在一定程度上减轻过拟合现象**\n",
    "\n",
    "> ### 机器学习的目的并不是让模型**记住**训练数据，而是对**未知数据**有更好的**推广**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### Stacking"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> **Stacking的基本思想是用一些 基分类器 进行分类，然后使用另一个分类器对结果进行整合**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### Blending"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> **Blending 与 Stacking 的主要区别在于训练集不是通过 K-Fold 的 交叉验证策略 来获得 预测值 从而生成 模型的特征 ，而是建立一个 Holdout集 **\n",
    "\n",
    "> **即把 Stacking 流程中的 K-Fold 交叉验证 改成 HoldOut 交叉验证**\n",
    "\n",
    "> **Blending优点在于：**\n",
    "\n",
    "> 1.比stacking简单（因为不用进行k次的交叉验证来获得stacker feature）\n",
    "\n",
    "> 2.避开了一个信息泄露问题：generlizers和stacker使用了不一样的数据集\n",
    "\n",
    ">3.在团队建模过程中，不需要给队友分享自己的随机种子\n",
    "\n",
    "> **Blending 缺点在于：**\n",
    "\n",
    "> 1.使用了很少的数据\n",
    "\n",
    "> 2.blender可能会过拟合\n",
    "\n",
    "> 3.Stacking使用多次的交叉验证会比较稳健对于实践中的结果而言，Stacking和Blending的效果是差不多的，所以使用哪种方法都没什么所谓，完全取决于个人爱好"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 二、任务说明：\n",
    "\n",
    "- 利用 **金融数据** 来预测贷款用户是否会逾期\n",
    "- 其中，**“status”** 是结果标签：**0** 表示未逾期，**1** 表示逾期"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 三、任务要求：\n",
    "\n",
    "> **模型融合（Model Integration）**\n",
    "\n",
    "> **模型融合，模型融合方式任意，并结合 Task 05 给出最优结果**\n",
    "\n",
    "> **Stacking 或 Blending 融合，用目前评分最高的模型作为基准模型，和其他模型进行 Stacking 融合，得到最终模型及评分结果**\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 四、任务实现（模型融合）：（2 Days）"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 6.1 数据加载"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(4754, 90)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Unnamed: 0</th>\n",
       "      <th>custid</th>\n",
       "      <th>trade_no</th>\n",
       "      <th>bank_card_no</th>\n",
       "      <th>low_volume_percent</th>\n",
       "      <th>middle_volume_percent</th>\n",
       "      <th>take_amount_in_later_12_month_highest</th>\n",
       "      <th>trans_amount_increase_rate_lately</th>\n",
       "      <th>trans_activity_month</th>\n",
       "      <th>trans_activity_day</th>\n",
       "      <th>...</th>\n",
       "      <th>loans_max_limit</th>\n",
       "      <th>loans_avg_limit</th>\n",
       "      <th>consfin_credit_limit</th>\n",
       "      <th>consfin_credibility</th>\n",
       "      <th>consfin_org_count_current</th>\n",
       "      <th>consfin_product_count</th>\n",
       "      <th>consfin_max_limit</th>\n",
       "      <th>consfin_avg_limit</th>\n",
       "      <th>latest_query_day</th>\n",
       "      <th>loans_latest_day</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>5</td>\n",
       "      <td>2791858</td>\n",
       "      <td>20180507115231274000000023057383</td>\n",
       "      <td>卡号1</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.99</td>\n",
       "      <td>0</td>\n",
       "      <td>0.90</td>\n",
       "      <td>0.55</td>\n",
       "      <td>0.313</td>\n",
       "      <td>...</td>\n",
       "      <td>2900.0</td>\n",
       "      <td>1688.0</td>\n",
       "      <td>1200.0</td>\n",
       "      <td>75.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1200.0</td>\n",
       "      <td>1200.0</td>\n",
       "      <td>12.0</td>\n",
       "      <td>18.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>10</td>\n",
       "      <td>534047</td>\n",
       "      <td>20180507121002192000000023073000</td>\n",
       "      <td>卡号1</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.94</td>\n",
       "      <td>2000</td>\n",
       "      <td>1.28</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0.458</td>\n",
       "      <td>...</td>\n",
       "      <td>3500.0</td>\n",
       "      <td>1758.0</td>\n",
       "      <td>15100.0</td>\n",
       "      <td>80.0</td>\n",
       "      <td>5.0</td>\n",
       "      <td>6.0</td>\n",
       "      <td>22800.0</td>\n",
       "      <td>9360.0</td>\n",
       "      <td>4.0</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>12</td>\n",
       "      <td>2849787</td>\n",
       "      <td>20180507125159718000000023114911</td>\n",
       "      <td>卡号1</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.96</td>\n",
       "      <td>0</td>\n",
       "      <td>1.00</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0.114</td>\n",
       "      <td>...</td>\n",
       "      <td>1600.0</td>\n",
       "      <td>1250.0</td>\n",
       "      <td>4200.0</td>\n",
       "      <td>87.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>4200.0</td>\n",
       "      <td>4200.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>6.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>13</td>\n",
       "      <td>1809708</td>\n",
       "      <td>20180507121358683000000388283484</td>\n",
       "      <td>卡号1</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.96</td>\n",
       "      <td>2000</td>\n",
       "      <td>0.13</td>\n",
       "      <td>0.57</td>\n",
       "      <td>0.777</td>\n",
       "      <td>...</td>\n",
       "      <td>3200.0</td>\n",
       "      <td>1541.0</td>\n",
       "      <td>16300.0</td>\n",
       "      <td>80.0</td>\n",
       "      <td>5.0</td>\n",
       "      <td>5.0</td>\n",
       "      <td>30000.0</td>\n",
       "      <td>12180.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>14</td>\n",
       "      <td>2499829</td>\n",
       "      <td>20180507115448545000000388205844</td>\n",
       "      <td>卡号1</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.99</td>\n",
       "      <td>0</td>\n",
       "      <td>0.46</td>\n",
       "      <td>1.00</td>\n",
       "      <td>0.175</td>\n",
       "      <td>...</td>\n",
       "      <td>2300.0</td>\n",
       "      <td>1630.0</td>\n",
       "      <td>8300.0</td>\n",
       "      <td>79.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>2.0</td>\n",
       "      <td>8400.0</td>\n",
       "      <td>8250.0</td>\n",
       "      <td>22.0</td>\n",
       "      <td>120.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 90 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   Unnamed: 0   custid                          trade_no bank_card_no  \\\n",
       "0           5  2791858  20180507115231274000000023057383          卡号1   \n",
       "1          10   534047  20180507121002192000000023073000          卡号1   \n",
       "2          12  2849787  20180507125159718000000023114911          卡号1   \n",
       "3          13  1809708  20180507121358683000000388283484          卡号1   \n",
       "4          14  2499829  20180507115448545000000388205844          卡号1   \n",
       "\n",
       "   low_volume_percent  middle_volume_percent  \\\n",
       "0                0.01                   0.99   \n",
       "1                0.02                   0.94   \n",
       "2                0.04                   0.96   \n",
       "3                0.00                   0.96   \n",
       "4                0.01                   0.99   \n",
       "\n",
       "   take_amount_in_later_12_month_highest  trans_amount_increase_rate_lately  \\\n",
       "0                                      0                               0.90   \n",
       "1                                   2000                               1.28   \n",
       "2                                      0                               1.00   \n",
       "3                                   2000                               0.13   \n",
       "4                                      0                               0.46   \n",
       "\n",
       "   trans_activity_month  trans_activity_day        ...         \\\n",
       "0                  0.55               0.313        ...          \n",
       "1                  1.00               0.458        ...          \n",
       "2                  1.00               0.114        ...          \n",
       "3                  0.57               0.777        ...          \n",
       "4                  1.00               0.175        ...          \n",
       "\n",
       "   loans_max_limit  loans_avg_limit  consfin_credit_limit  \\\n",
       "0           2900.0           1688.0                1200.0   \n",
       "1           3500.0           1758.0               15100.0   \n",
       "2           1600.0           1250.0                4200.0   \n",
       "3           3200.0           1541.0               16300.0   \n",
       "4           2300.0           1630.0                8300.0   \n",
       "\n",
       "   consfin_credibility  consfin_org_count_current  consfin_product_count  \\\n",
       "0                 75.0                        1.0                    2.0   \n",
       "1                 80.0                        5.0                    6.0   \n",
       "2                 87.0                        1.0                    1.0   \n",
       "3                 80.0                        5.0                    5.0   \n",
       "4                 79.0                        2.0                    2.0   \n",
       "\n",
       "   consfin_max_limit  consfin_avg_limit  latest_query_day  loans_latest_day  \n",
       "0             1200.0             1200.0              12.0              18.0  \n",
       "1            22800.0             9360.0               4.0               2.0  \n",
       "2             4200.0             4200.0               2.0               6.0  \n",
       "3            30000.0            12180.0               2.0               4.0  \n",
       "4             8400.0             8250.0              22.0             120.0  \n",
       "\n",
       "[5 rows x 90 columns]"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 数据集的加载\n",
    "data = pd.read_csv('Dataset/data.csv')\n",
    "# 剔除重复特征\n",
    "data.drop_duplicates(inplace=True)\n",
    "print(data.shape)\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    3561\n",
       "1    1193\n",
       "Name: status, dtype: int64"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pickle\n",
    "\n",
    "# 载入特征 获取label\n",
    "with open('feature.pkl', 'rb') as f:\n",
    "    X = pickle.load(f)\n",
    "\n",
    "# 观测正负样本是否均衡\n",
    "y = data.status\n",
    "y.value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 6.2 数据集划分"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 划分数据集（训练集和测试集7:3 无验证集）\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3,random_state=2333)\n",
    "\n",
    "# 特征归一化处理\n",
    "std = StandardScaler()\n",
    "X_train = std.fit_transform(X_train.values)\n",
    "X_test = std.transform(X_test.values)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 6.3 模型评估"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score\n",
    "from sklearn.metrics import roc_auc_score,roc_curve, auc\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "def model_metrics(clf, X_train, X_test, y_train, y_test):\n",
    "    # 预测训练集和测试集\n",
    "    y_train_pred = clf.predict(X_train)\n",
    "    y_test_pred = clf.predict(X_test)\n",
    "    \n",
    "    y_train_proba = clf.predict_proba(X_train)[:,1]\n",
    "    y_test_proba = clf.predict_proba(X_test)[:,1]\n",
    "    \n",
    "    # 准确率Accuracy\n",
    "    print('【Accuracy准确率】', end = ' ')\n",
    "    print('训练集：', '%.4f'%accuracy_score(y_train, y_train_pred), end = ' ')\n",
    "    print('测试集：', '%.4f'%accuracy_score(y_test, y_test_pred))\n",
    "    \n",
    "    # 精准率Precision\n",
    "    print('【Prediction精准率】', end = ' ')\n",
    "    print('训练集：', '%.4f'%precision_score(y_train, y_train_pred), end = ' ')\n",
    "    print('测试集：', '%.4f'%precision_score(y_test, y_test_pred))\n",
    "\n",
    "    # 召回率Recall\n",
    "    print('【Recall召回率】', end = ' ')\n",
    "    print('训练集：', '%.4f'%recall_score(y_train, y_train_pred), end = ' ')\n",
    "    print('测试集：', '%.4f'%recall_score(y_test, y_test_pred))\n",
    "    \n",
    "    # f1-score\n",
    "    print('【F1-score】', end = ' ')\n",
    "    print('训练集：', '%.4f'%f1_score(y_train, y_train_pred), end = ' ')\n",
    "    print('测试集：', '%.4f'%f1_score(y_test, y_test_pred))\n",
    "    \n",
    "    # AUC取值\n",
    "    print('【AUC取值】', end = ' ')\n",
    "    print('训练集：', '%.4f'%roc_auc_score(y_train, y_train_proba), end = ' ')\n",
    "    print('测试集：', '%.4f'%roc_auc_score(y_test, y_test_proba))\n",
    "    \n",
    "    # ROC曲线\n",
    "    fpr_train, tpr_train, thresholds_train = roc_curve(y_train, y_train_proba, pos_label = 1)\n",
    "    fpr_test, tpr_test, thresholds_test = roc_curve(y_test, y_test_proba, pos_label = 1)\n",
    "    \n",
    "    label = [\"Train - AUC:{:.4f}\".format(auc(fpr_train, tpr_train)), \n",
    "             \"Test - AUC:{:.4f}\".format(auc(fpr_test, tpr_test))]\n",
    "    plt.plot(fpr_train,tpr_train)\n",
    "    plt.plot(fpr_test,tpr_test)\n",
    "    plt.plot([0, 1], [0, 1], 'd--')\n",
    "    plt.xlabel('False Positive Rate')\n",
    "    plt.ylabel('True Positive Rate')\n",
    "    plt.legend(label, loc = 4)\n",
    "    plt.title(\"ROC Curve\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 6.4 模型导入"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### 导入不同分类器模型__LR, SVM, DT, RF, XGBoost, MLXtend"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn import svm\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from xgboost.sklearn import XGBClassifier"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 6.5 模型融合"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### 6.5.1 利用 MLXtend 库中的 StackingClassifier 分类器进行 模型融合"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "from mlxtend.classifier import StackingClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 模型调优后得到的参数\n",
    "lr = LogisticRegression(C = 0.1, penalty = 'l1')\n",
    "svm_linear = svm.SVC(C = 0.01, kernel = 'linear', probability=True)\n",
    "svm_poly =  svm.SVC(C = 0.01, kernel = 'poly', probability=True)\n",
    "svm_rbf =  svm.SVC(gamma = 0.01, C =0.01 , probability=True)\n",
    "svm_sigmoid =  svm.SVC(C = 0.01, kernel = 'sigmoid',probability=True)\n",
    "dt = DecisionTreeClassifier(max_depth=11,min_samples_split=550,min_samples_leaf=80,max_features=19)\n",
    "rf = RandomForestClassifier(n_estimators=120, max_depth=9, min_samples_split=50,\\\n",
    "                            min_samples_leaf=20, max_features = 9,oob_score=True, random_state=2333)\n",
    "xgb = XGBClassifier(learning_rate =0.01, n_estimators=180, max_depth=3, min_child_weight=5, \\\n",
    "                    gamma=0.4, subsample=0.5, colsample_bytree=0.9, reg_alpha=1,\\\n",
    "                    objective= 'binary:logistic', nthread=4,scale_pos_weight=1, seed=27)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### 6.5.2 手动实现 Stacking 存在问题"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np \n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "from sklearn.svm import SVC\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "# import xgboost as xgb\n",
    "from xgboost.sklearn import XGBClassifier\n",
    "from sklearn.ensemble import ExtraTreesClassifier\n",
    "from sklearn.linear_model import LogisticRegression"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def stacking(train_x, train_y, test):\n",
    "    \"\"\" stacking\n",
    "    input: train_x, train_y, test\n",
    "    output: test的预测值\n",
    "    clfs: 5个一级分类器\n",
    "    dataset_blend_train: 一级分类器的prediction, 二级分类器的train_x\n",
    "    dataset_blend_test: 二级分类器的test\n",
    "    \"\"\"\n",
    "    # 5个一级分类器\n",
    "    clfs = [SVC(C = 3, kernel=\"rbf\"),\n",
    "            RandomForestClassifier(n_estimators=100, max_features=\"log2\", max_depth=10, min_samples_leaf=1, bootstrap=True, n_jobs=-1, random_state=1),\n",
    "            KNeighborsClassifier(n_neighbors=15, n_jobs=-1),\n",
    "            xgb.XGBClassifier(n_estimators=100, objective=\"binary:logistic\", gamma=1, max_depth=10, subsample=0.8, nthread=-1, seed=1),\n",
    "            ExtraTreesClassifier(n_estimators=100, criterion=\"gini\", max_features=\"log2\", max_depth=10, min_samples_split=2, min_samples_leaf=1,bootstrap=True, n_jobs=-1, random_state=1)]\n",
    "    \n",
    "    # 二级分类器的train_x, test\n",
    "    dataset_blend_train = np.zeros((train_x.shape[0], len(clfs)), dtype=np.int)\n",
    "    dataset_blend_test = np.zeros((test.shape[0], len(clfs)), dtype=np.int)\n",
    "    \n",
    "    # 5个分类器进行8_folds预测\n",
    "    n_folds = 8\n",
    "    skf = StratifiedKFold(n_splits=n_folds, shuffle=True, random_state=1)\n",
    "    for i,clf in enumerate(clfs):\n",
    "        dataset_blend_test_j = np.zeros((test.shape[0], n_folds))  # 每个分类器的单次fold预测结果\n",
    "        for j,(train_index,test_index) in enumerate(skf.split(train_x, train_y)):\n",
    "            tr_x = train_x[train_index]\n",
    "            tr_y = train_y[train_index]\n",
    "            clf.fit(tr_x, tr_y)\n",
    "            dataset_blend_train[test_index, i] = clf.predict(train_x[test_index])\n",
    "            dataset_blend_test_j[:, j] = clf.predict(test)\n",
    "        dataset_blend_test[:, i] = dataset_blend_test_j.sum(axis=1) // (n_folds//2 + 1)\n",
    "    \n",
    "    # 二级分类器进行预测\n",
    "    clf = LogisticRegression(penalty=\"l1\", tol=1e-6, C=1.0, random_state=1, n_jobs=-1)\n",
    "    clf.fit(dataset_blend_train, train_y)\n",
    "    prediction = clf.predict(dataset_blend_test)\n",
    "    return prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# prediction = stacking(X_train, y_train, X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> ### 6.5.3 自动调用实现 模型融合"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> **6.5.3.1 利用5种SVM、决策树、随机森林以及XGBoost作为初级分类器，LR作为次级分类器**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "sclf = StackingClassifier(classifiers=[svm_linear, svm_poly, svm_rbf, svm_sigmoid, dt, rf, xgb], meta_classifier=lr)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> **6.5.3.2 将初级分类器产生的 类别标签  sclf 作为新特征**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "【Accuracy准确率】 训练集： 0.8118 测试集： 0.7940\n",
      "【Prediction精准率】 训练集： 0.8794 测试集： 0.8242\n",
      "【Recall召回率】 训练集： 0.2952 测试集： 0.2125\n",
      "【F1-score】 训练集： 0.4421 测试集： 0.3378\n",
      "【AUC取值】 训练集： 0.6652 测试集： 0.6403\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x241bbcf8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sclf.fit(X_train, y_train.values)\n",
    "model_metrics(sclf, X_train, X_test, y_train, y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> **6.5.3.3 将初级分类器产生的输出 类概率 use_probas 作为新特征**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "sclf = StackingClassifier(classifiers=[svm_linear, svm_poly, svm_rbf, svm_sigmoid, dt, rf, xgb], meta_classifier=lr, use_probas=True,average_probas=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "【Accuracy准确率】 训练集： 0.8518 测试集： 0.7919\n",
      "【Prediction精准率】 训练集： 0.7887 测试集： 0.6176\n",
      "【Recall召回率】 训练集： 0.5643 测试集： 0.4164\n",
      "【F1-score】 训练集： 0.6579 测试集： 0.4975\n",
      "【AUC取值】 训练集： 0.9027 测试集： 0.7823\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x24725cf8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sclf.fit(X_train, y_train.values)\n",
    "model_metrics(sclf, X_train, X_test, y_train, y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 五、讨论与问题（Discussion & Question）"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> #### 比较模型融合之前和融合之后的效果差异，分析导致产生差异的关键因素/原因在哪里？"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 六、总结（Conclusion）"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> #### Task01 Data Analysis (数据分析)\n",
    "\n",
    "> #### Task02 Feature Engineering (特征工程)\n",
    "\n",
    "> #### Task03  Mdel Construction(模型构建)\n",
    "\n",
    "> #### Task04  Model Evaluation(模型评估)\n",
    "\n",
    "> #### Task05  Model Tuning(模型调优)\n",
    "\n",
    "> #### Task06  Model Integration(模型融合)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Conference"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> **[模型融合方法总结](https://blog.csdn.net/hust_tsb/article/details/76577599)**\n",
    "\n",
    "> **[模型融合方法](https://blog.csdn.net/sinat_29819401/article/details/71191219)**\n",
    "\n",
    "> **[KAGGLE ENSEMBLING GUIDE](https://mlwave.com/kaggle-ensembling-guide/)**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}