不同分类算法对比

不归一化情况下不同算法对比

对不同分类算法在数据归一化，和不归一化的情况下，对比算法预测得分。总结不同算法的特点。

逻辑斯蒂回归算法分类效果

# 导包，一次把所有的包都导进来
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.tree import DecisionTreeClassifier
from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import warnings
warnings.filterwarnings('ignore')

# 加载数据
data=datasets.load_wine()
X=data['data']
y=data['target']
feature_names=data['feature_names']

%%time
score=0
for i in range(1000):
    # 拆分数据
    X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2)
    # 构建模型
    model=LogisticRegression()
    # 训练模型
    model.fit(X_train,y_train)
    # 预测得分
    score += model.score(X_test,y_test)/1000
score

Wall time: 23.6 s
0.9484722222222285

支持向量机

%%time
score=0
for i in range(1000):
    # 拆分数据
    X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2)
    # 构建模型
    model=SVC()
    # 训练模型
    model.fit(X_train,y_train)
    # 预测得分
    score += model.score(X_test,y_test)/1000
score

Wall time: 2.21 s
0.6722499999999996

决策树

%%time
score=0
for i in range(1000):
    # 拆分数据
    X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2)
    # 构建模型
    model=DecisionTreeClassifier(criterion='entropy')
    # 训练模型
    model.fit(X_train,y_train)
    # 预测得分
    score += model.score(X_test,y_test)/1000
score

Wall time: 1.57 s
0.9227500000000051

归一化情况下不同算法对比

# 归一化X
standard=StandardScaler()
X=standard.fit_transform(X)

逻辑斯蒂回归

%%time
score=0
for i in range(1000):
    # 拆分数据
    X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2)
    # 构建模型
    model=LogisticRegression()
    # 训练模型
    model.fit(X_train,y_train)
    # 预测得分
    score += model.score(X_test,y_test)/1000
score

Wall time: 8.21 s
0.981000000000007

支持向量机

%%time
score=0
for i in range(1000):
    # 拆分数据
    X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2)
    # 构建模型
    model=SVC()
    # 训练模型
    model.fit(X_train,y_train)
    # 预测得分
    score += model.score(X_test,y_test)/1000
score

Wall time: 1.85 s

0.9836388888888954

决策树

%%time
score=0
for i in range(1000):
    # 拆分数据
    X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2)
    # 构建模型
    model=DecisionTreeClassifier(criterion='entropy')
    # 训练模型
    model.fit(X_train,y_train)
    # 预测得分
    score += model.score(X_test,y_test)/1000
score

Wall time: 1.42 s

0.925611111111115

结论

• 归一化会提高逻辑斯蒂回归和支持向量机的预测分数
• 决策树预测结果对归一化不敏感
• 归一化会缩短逻辑斯蒂回归的运行时间