转：社区发现评估指标-NMI

版权声明：本文为博主原创文章，遵循 CC 4.0 by-sa 版权协议，转载请附上原文出处链接和本声明。
              
                    本文链接：https://blog.csdn.net/kan2281123066/article/details/73825009

                   
                                            1、介绍
NMI(Normalized Mutual Information)标准化互信息，常用在聚类中，度量两个聚类结果的相近程度。是社区发现(community detection)的重要衡量指标，基本可以比较客观地评价出一个社区划分与标准划分之间相比的准确度。NMI的值域是0到1，越高代表划分得越准。

2、python代码

# coding=utf-8
import numpy as np
import math
def NMI(A,B):
    # len(A) should be equal to len(B)
    total = len(A)
    A_ids = set(A)
    B_ids = set(B)
    #Mutual information
    MI = 0
    eps = 1.4e-45
    for idA in A_ids:
        for idB in B_ids:
            idAOccur = np.where(A==idA)
            idBOccur = np.where(B==idB)
            idABOccur = np.intersect1d(idAOccur,idBOccur)
            px = 1.0*len(idAOccur[0])/total
            py = 1.0*len(idBOccur[0])/total
            pxy = 1.0*len(idABOccur)/total
            MI = MI + pxy*math.log(pxy/(px*py)+eps,2)
    # Normalized Mutual information
    Hx = 0
    for idA in A_ids:
        idAOccurCount = 1.0*len(np.where(A==idA)[0])
        Hx = Hx - (idAOccurCount/total)*math.log(idAOccurCount/total+eps,2)
    Hy = 0
    for idB in B_ids:
        idBOccurCount = 1.0*len(np.where(B==idB)[0])
        Hy = Hy - (idBOccurCount/total)*math.log(idBOccurCount/total+eps,2)
    MIhat = 2.0*MI/(Hx+Hy)
    return MIhat

if __name__ == '__main__':
    A = np.array([1,1,1,1,1,1,2,2,2,2,2,2,3,3,3,3,3])
    B = np.array([1,2,1,1,1,1,1,2,2,2,2,3,1,1,3,3,3])
    print (NMI(A,B))
12345678910111213141516171819202122232425262728293031323334353637

结果：0.36456
---------------------
版权声明：本文为CSDN博主「开开_王子」的原创文章，遵循CC 4.0 by-sa版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/kan2281123066/article/details/73825009