stack模型融合

#coding:utf-8
import numpy as np
import pandas as pd
from sklearn.metrics import roc_auc_score
from sklearn.cross_validation import cross_val_score
from sklearn.svm import SVC
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import AdaBoostClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.cross_validation import train_test_split
from random import shuffle
import copy
import random
random.seed(42)

class Essemble(object):
    def __init__(self):
        self.trainDf = pd.DataFrame()
        self.testDf = pd.DataFrame()

    def loadTrainData(self):
        traindf = pd.read_csv("E:\\horseColicTraining2.csv")
        label_train = traindf["label"]
        label_train.ix[label_train==-1]=0
        attr_train = traindf.ix[:,:-1]
        self.label_train = label_train
        self.attr_train = attr_train

    def loadTestData(self):
        testdf = pd.read_csv("E:\\horseColicTest2.csv")
        label_test = testdf["label"]
        label_test.ix[label_test==-1]=0
        attr_test = testdf.ix[:,:-1]
        self.label_test = label_test
        self.attr_test = attr_test

    def Kfold(self,model,folds=5):
        span = self.attr_train.shape[0]/folds+1   #跨度，每折的样本数量
        sampleIndex = range(self.attr_train.shape[0])
        shuffle(sampleIndex)
        sp = 0
        predResult = [-1]*self.attr_train.shape[0]   #预测的结果记录
        rtestLst = []
        while sp < self.attr_train.shape[0]:
            testIndex = sampleIndex[sp:(sp+span)]    #测试集的索引
            trainIndex = copy.deepcopy(sampleIndex)
            del trainIndex[sp:(sp+span)]             #训练集的索引
            attr_train_sample = self.attr_train.ix[trainIndex,:]   #训练集的属性
            label_train_sample = self.label_train.ix[trainIndex]   #训练集的标签
            attr_test_sample = self.attr_train.ix[testIndex,:]    #测试集的属性
            label_test_sample = self.label_train.ix[testIndex]    #测试集的属性
            model.fit(attr_train_sample,label_train_sample)
            testpred = model.predict(attr_test_sample)
            rtestpred = model.predict(self.attr_test)               #真实的外部测试集的估计
            rtestLst.append(rtestpred)
            for i in range(len(testIndex)):
                predResult[testIndex[i]] = testpred[i]
            sp = sp+span
            if (sp+span) >= self.attr_train.shape[0]:
                span = self.attr_train.shape[0]-sp
        rtestByModel = np.mat(rtestLst).mean(axis=0).tolist()[0]         #模型估计的测试集的结果的均值
        return predResult,rtestByModel

    def package(self,model,key):
        predResult,rtestByModel = self.Kfold(model)
        self.trainDf[key] = pd.Series(predResult)
        self.testDf[key] = pd.Series(rtestByModel)
        self.singleModel(model,key)

    def singleModel(self,model,key):
        model.fit(self.attr_train,self.label_train)
        pred = model.predict(self.attr_test)
        auc = roc_auc_score(self.label_test,pred)
        print "single model of " + str(key) + ":" + str(auc)

    def console(self):
        self.loadTrainData()
        self.loadTestData()

        ada = AdaBoostClassifier(n_estimators=500)
        self.package(ada,"ada")

        dt_entropy = DecisionTreeClassifier(criterion="entropy")
        self.package(dt_entropy,"dt_entropy")

        dt_gini = DecisionTreeClassifier(criterion="gini")
        self.package(dt_gini,"dt_gini")

        gb = GradientBoostingClassifier(n_estimators=500)
        self.package(gb,"gb")

        rf = RandomForestClassifier(n_estimators=500)
        self.package(rf,"rf")

        svc = SVC(kernel="rbf")
        self.package(svc,"svc")

    def LRstack(self):
        logist = LogisticRegression()
        logist.fit(self.trainDf,self.label_train)
        pred = logist.predict(self.testDf)
        auc = roc_auc_score(self.label_test,pred)
        print auc

###############################
Eobject = Essemble()
Eobject.console()
Eobject.LRstack()

数据是机器学习实战中的马的疾病数据，原理知乎上的简介是https://zhuanlan.zhihu.com/p/25836678