代码：

# -*- coding: utf-8 -*- 
# * @author Ming_H
# * @date 2019/05/16
# * Usage: Tree methods with LR method

import numpy as np
import pandas as pd
from sklearn.linear_model import LogisticRegression
from sklearn.preprocessing import OneHotEncoder
from sklearn.model_selection import train_test_split
from sklearn import metrics
import logging
import xgboost as xgb
import time
from sklearn.datasets import load_iris

logging.basicConfig(format='%(asctime)s : %(levelname)s: %(message)s', level=logging.INFO)


def XGBoost_LR(df_train):
    X_train = df_train.values[:, :-1]
    y_train = df_train.values[:, -1]
    X_train_xgb, X_train_lr, y_train_xgb, y_train_lr = train_test_split(X_train, 
                                                y_train, test_size=0.75)
    XGB = xgb.XGBClassifier(scale_pos_weight=1,
                            objective="binary:logistic",
                            eval_metric = 'logloss',
                            booster = 'gbtree',               # gbtree, gblinear or dart
                            learning_rate = 0.01,             # 如同学习率
                            n_estimators = 6,                 # 树的个数
                            max_depth = 5,                    # 构建树的深度，越大越容易过拟合
                            min_child_weight = 3,             # 这个参数默认是 1，是每个叶子里面 h 的和至少是多少
                            gamma = 0.3,                      # 树的叶子节点上作进一步分区所需的最小损失减少,越大越保守，一般0.1、0.2这样子。
                            # subsample = 0.5,                # 随机采样训练样本 训练实例的子采样比
                            # colsample_bytree = 0.8,         # 生成树时进行的列采样 
                            # reg_alpha = 1,                  # L1 正则项参数
                            # reg_lambda = 4,                 # 控制模型复杂度的权重值的L2正则化项参数，参数越大，模型越不容易过拟合。
                            # max_delta_step = 0,             # 最大增量步长，我们允许每个树的权重估计。
                            # seed = 100,                     # 随机种子
                            missing = -999.0                  # 为缺失值设置默认值
                            
    )
    XGB.fit(X_train_xgb, y_train_xgb)
    
    logging.info("训练集特征数据为： \n%s" % X_train_xgb)
    logging.info("训练集标签数据为： \n%s" % y_train_xgb)
    logging.info("转化为叶子节点后的特征为：\n%s" % XGB.apply(X_train_xgb, ntree_limit=0))

    XGB_enc = OneHotEncoder()
    XGB_enc.fit(XGB.apply(X_train_xgb, ntree_limit=0)) # ntree_limit 预测时使用的树的数量

    XGB_LR = LogisticRegression(class_weight= "balanced",
                                penalty='l2',
                                dual=False, 
                                tol=0.001,
                                C=0.001, 
                                # fit_intercept=True, 
                                # intercept_scaling=1, 
                                # random_state=None, 
                                # solver='liblinear',   # newton-cg, lbfgs, liblinear, sag, saga}, default: liblinear.
                                # max_iter=100, 
                                # multi_class='ovr',  # ovr, multinomial, auto
                                # verbose=0, 
                                # warm_start=False, 
                                # n_jobs=None
                                )

    XGB_LR.fit(XGB_enc.transform(XGB.apply(X_train_lr)), y_train_lr.astype('int'))

    X_predict = XGB_LR.predict_proba(XGB_enc.transform(XGB.apply(X_train)))[:, 1]
    
    AUC_train = metrics.roc_auc_score(y_train.astype('int'), X_predict)
    logging.info("AUC of train data is %f" % AUC_train)



if __name__ == "__main__":
    start = time.clock()
    #加载数据集
    iris=load_iris()
    df_data = pd.concat([pd.DataFrame(iris.data), pd.DataFrame(iris.target)], axis=1)
    df_data.columns = ["x1","x2", "x3","x4", "y"]
    df_new = df_data[df_data["y"]<2]

    logging.info("Train model begine...")
    XGBoost_LR(df_new)
    end = time.clock()
    logging.info("Program over, time cost is %s" % (end-start))