pca处理后建模 sklearn_sklearn—特征工程

项目合作QQ：231469242

变量筛选：(逻辑回归)

好处：

变量少，模型运行速度快，更容易解读和理解

坏处：

会牺牲掉少量精确性

变量不筛选：(random forest)

好处：

提高准确性

坏处：

变量多，运行速度慢

logistic模型为什么要考虑共线性问题？

共线性问题会导致估计结果不准确，系数方向都可能发生改变。不管是logistic回归模型，还是ols都要考虑。

官网

http://scikit-learn.org/stable/modules/feature_selection.html

测试结果随机森林特征工程效果最好，其次是支持向量l2模式

主成分一个就可以解释98%以上方差，和spss结果差距大，需要核实

# -*- coding: utf-8 -*-

"""

Created on Sat Apr 14 19:36:29 2018

@author: Administrator

#乳腺癌数据测试中，卡方检验效果不如随机森林,卡方筛选的2个最好因子在随机森林中权重不是最大

"""

from sklearn.decomposition import PCA

from sklearn.ensemble import ExtraTreesClassifier

from sklearn.feature_selection import SelectFromModel

from sklearn.datasets import load_breast_cancer

from sklearn.feature_selection import SelectKBest

from sklearn.feature_selection import chi2

from sklearn.svm import LinearSVC

cancer = load_breast_cancer()

X= cancer.data

y= cancer.target

#特征工程-卡方筛选K个最佳特征

#乳腺癌数据测试中，卡方检验效果不如随机森林,卡方筛选的2个最好因子在随机森林中权重不是最大

#we can perform a \chi^2 test to the samples to retrieve only the two best features as follows:

X_chi2 = SelectKBest(chi2, k=2).fit_transform(X, y)

print("X_chi2 shapes:",X_chi2.shape)

##特征工程-支持向量l1

#根据模型选择最佳特征，线性svc选出5个最佳特征

lsvc = LinearSVC(C=0.01, penalty="l1", dual=False).fit(X, y)

model = SelectFromModel(lsvc, prefit=True)

X_lsvc1 = model.transform(X)

print("LinearSVC l1 shapes:",X_lsvc1.shape)

#特征工程-支持向量l2

lsvc = LinearSVC(C=0.01, penalty="l2", dual=False).fit(X, y)

model = SelectFromModel(lsvc, prefit=True)

X_lsvc2 = model.transform(X)

print("LinearSVC l2 shapes:",X_lsvc2.shape)

#特征工程-forest of trees 随机森林

trees = ExtraTreesClassifier(n_estimators=10000)

trees = trees.fit(X, y)

trees.feature_importances_

model = SelectFromModel(trees, prefit=True)

X_trees = model.transform(X)

print("forest trees shapes:",X_trees.shape)

'''

测试结果随机森林特征工程效果最好，其次是支持向量l2模式

X_chi2 shapes: (569, 2)

LinearSVC l1 shapes: (569, 5)

LinearSVC l2 shapes: (569, 9)

forest trees shapes: (569, 10)

'''

#主成分分析

pca =PCA(n_components=0.98)

#pca= PCA(n_components='mle')

digits = load_breast_cancer()

X_digits = cancer.data

y_digits = digits.target

#scaled_x=preprocessing.scale(X_digits)

# Plot the PCA spectrum

#pca.fit(scaled_x)

pca.fit(X_digits)

print("PCA analysis:")

print (pca.explained_variance_ratio_)

print (pca.explained_variance_)

print (pca.n_components_)

#计算协方差

pca.get_covariance()

#Estimated precision of data.计算数据估计的准确性

pca.get_precision()

德国信用评分卡测试

数据必须整理为数字型，不接受分类变量

# -*- coding: utf-8 -*-

"""

Created on Sat Apr 14 19:36:29 2018

@author: Administrator

#乳腺癌数据测试中，卡方检验效果不如随机森林,卡方筛选的2个最好因子在随机森林中权重不是最大

"""

import pandas as pd

import numpy as np

from sklearn.decomposition import PCA

from sklearn.ensemble import ExtraTreesClassifier

from sklearn.feature_selection import SelectFromModel

from sklearn.feature_selection import SelectKBest

from sklearn.feature_selection import chi2

from sklearn.svm import LinearSVC

from sklearn.ensemble import RandomForestClassifier

trees=100

#读取文件

readFileName="GermanData_total.xlsx"

#读取excel

df=pd.read_excel(readFileName)

list_columns=list(df.columns[:-1])

X=df.ix[:,:-1]

y=df.ix[:,-1]

#X_dummy=pd.get_dummies(X)

#特征工程-卡方筛选K个最佳特征

#乳腺癌数据测试中，卡方检验效果不如随机森林,卡方筛选的2个最好因子在随机森林中权重不是最大

#we can perform a \chi^2 test to the samples to retrieve only the two best features as follows:

X_chi2 = SelectKBest(chi2, k=2).fit_transform(X, y)

print("X_chi2 shapes:",X_chi2.shape)

##特征工程-支持向量l1

#根据模型选择最佳特征，线性svc选出5个最佳特征

lsvc = LinearSVC(C=0.01, penalty="l1", dual=False).fit(X, y)

model = SelectFromModel(lsvc, prefit=True)

X_lsvc1 = model.transform(X)

print("LinearSVC l1 shapes:",X_lsvc1.shape)

#特征工程-支持向量l2

lsvc = LinearSVC(C=0.01, penalty="l2", dual=False).fit(X, y)

model = SelectFromModel(lsvc, prefit=True)

X_lsvc2 = model.transform(X)

print("LinearSVC l2 shapes:",X_lsvc2.shape)

#特征工程-forest of trees 随机森林

trees = ExtraTreesClassifier(n_estimators=10000)

trees = trees.fit(X, y)

trees.feature_importances_

model = SelectFromModel(trees, prefit=True)

X_trees = model.transform(X)

print("forest trees shapes:",X_trees.shape)

'''

测试结果随机森林特征工程效果最好，其次是支持向量l2模式

X_chi2 shapes: (569, 2)

LinearSVC l1 shapes: (569, 5)

LinearSVC l2 shapes: (569, 9)

forest trees shapes: (569, 10)

'''

#主成分分析

pca =PCA(n_components=0.98)

#pca= PCA(n_components='mle')

#scaled_x=preprocessing.scale(X_digits)

# Plot the PCA spectrum

#pca.fit(scaled_x)

pca.fit(X)

print("PCA analysis:")

print (pca.explained_variance_ratio_)

print (pca.ex