机器学习--集成学习Stacking算法23

什么是Stacking

使用多个不同的分类器对训练集进预测，把预测 得到的结果作为一个次级分类器的输入。次级分 类器的输出是整个模型的预测结果。

Stacking需要训练两层分类器，第一层的初级分类器（比如：决策树 + KNN + 神经网络 + 逻辑回归）和第二层的次级分类器。

代码实现

from sklearn import datasets  
from sklearn import model_selection  
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier  
from sklearn.tree import DecisionTreeClassifier
from mlxtend.classifier import StackingClassifier # pip install mlxtend
import numpy as np  

# 载入数据集
iris = datasets.load_iris()  
# 只要第1,2列的特征
x_data, y_data = iris.data[:, 1:3], iris.target  

# 定义三个不同的分类器
clf1 = KNeighborsClassifier(n_neighbors=1)  
clf2 = DecisionTreeClassifier() 
clf3 = LogisticRegression()  
 
# 定义一个次级分类器
lr = LogisticRegression()  
sclf = StackingClassifier(classifiers=[clf1, clf2, clf3],   
                          meta_classifier=lr)
  
for clf,label in zip([clf1, clf2, clf3, sclf],
                      ['KNN','Decision Tree','LogisticRegression','StackingClassifier']):  
    scores = model_selection.cross_val_score(clf, x_data, y_data, cv=3, scoring='accuracy')  
    print("Accuracy: %0.2f [%s]" % (scores.mean(), label)) 

cross_val_score做交叉验证

Accuracy: 0.91 [KNN]
Accuracy: 0.93 [Decision Tree]
Accuracy: 0.95[LogisticRegression]
Accuracy: 0.95 [StackingClassifier]

Stacking一般都比单个的算法正确率要高。

下面介绍一种投票的集成学习

from sklearn import datasets  
from sklearn import model_selection  
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier  
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import VotingClassifier
import numpy as np

# 载入数据集
iris = datasets.load_iris()  
# 只要第1,2列的特征
x_data, y_data = iris.data[:, 1:3], iris.target  

# 定义三个不同的分类器
clf1 = KNeighborsClassifier(n_neighbors=1)  
clf2 = DecisionTreeClassifier() 
clf3 = LogisticRegression()  

sclf = VotingClassifier([('knn',clf1),('dtree',clf2), ('lr',clf3)])   
  
for clf, label in zip([clf1, clf2, clf3, sclf],
                      ['KNN','Decision Tree','LogisticRegression','VotingClassifier']):  
  
    scores = model_selection.cross_val_score(clf, x_data, y_data, cv=3, scoring='accuracy')  
    print("Accuracy: %0.2f [%s]" % (scores.mean(), label)) 

Accuracy: 0.91 [KNN]
Accuracy: 0.93 [Decision Tree]
Accuracy: 0.91 [LogisticRegression]
Accuracy: 0.93 [VotingClassifier]

Voting不需要训练次级分类器，按照投票的结果进行做交叉验证。

总结：集成学习

“人多力量大”，
好的集成学习：多样性+准确性