Python 相关系数矩阵如何筛选高相关系数变量 python求相关系数矩阵

1 # -*- coding: utf-8 -*-
 2 """
 3 Created on Mon Jan  8 19:36:48 2018
 4 
 5 @author: markli
 6 """
 7 
 8 import numpy as np;
 9 import math;
10 
11 '''
12 计算矩阵A的相关系数矩阵
13 '''
14 def Correlation(A):
15     #得到A的形状 m 是行数 n 是列数
16     m,n = A.shape;
17     #存放每一列的均值
18     means = [];
19     #存放每一列的方差
20     var = [];
21     for i in range(n):
22         me = np.mean(A[:,i]);
23         means.append(me);
24         temp = A[:,i] - me;
25         #计算方差，除以m-1 与np.corrcoef有误差，除以m则基本没有误差
26         v = np.sum([p**2 for p in temp]) / m  ;
27         var.append(v);
28         
29     #存放相关系数
30     r = np.ones((n,n));
31     #离差矩阵
32     deviation = A - np.atleast_2d(means);
33     
34     for i in range(n):
35         for j in range(n):
36             cov = np.dot(np.atleast_2d(deviation[:,i]),np.atleast_2d(deviation[:,j]).T)/ (m);
37             va = math.sqrt(var[i] * var[j]);
38             r[i,j] = cov / va;
39             if(i==j):
40                 r[i,j]=1;
41             
42     return r;
43 #coeffs = np.ployfit(X,Y,degree) #得到一组回归方程的系数 X为一维，Y一般为一维，最多二维 degree指定自变量的次数
44 #np.ploy1d(coeffs) 产生一个多项式 多项式次数由高到低
45 A = np.random.random((10,3));
46 r1 = Correlation(A);
47 r2 = np.corrcoef(A,rowvar=False);
48 print(r1);
49 print('numpy 计算值：');
50 print(r2);
51             
52