程序实现:
上面的公式看似复杂,其实我们关心的就是最后的5个计算步骤,这里说明一下,有的书上以隶属度矩阵的某一范数小于一定值作为收敛的条件,这也可,不过计算量稍微要大一点了。
程序采用VB6.0编制,完全按照以上的步骤进行。
'程序实现功能:模糊聚类和硬聚类
'作 者: laviewpbt
'联系方式: laviewpbt@sina.com
'QQ:33184777
'版本:Version 2.3.1
'说明:复制请保留源作者信息,转载请说明,欢迎大家提出意见和建议
Private Declare Function GetTickCount Lib "kernel32" () As Long
Private Enum IniCenterMethod '初始中心的方法
CreateRandom '随机的中心点
CreateByHcm '由HCM创建的中心点
CreateByRandomZadeh '先随机创建隶属矩阵,然后计算得到的中心点
CreateByHand '手工确定初始中心点
End Enum
Private Enum AntiFuzzyMethod '反模糊的方法
Max '最大隶属度法
Middle '中位数法
Mean '加权均值法
End Enum
Private Type FcmInfo
Center() As Double '聚类中心
Degree() As Double '隶属度,为Double类型
Class() As Byte '记录数据属于那一类
TimeUse As Long '所用时间
Iterations As Long '迭带次数
ErrMsg As String '错误信息
End Type
Private Type HcmInfo
Center() As Double '聚类中心
Class() As Byte '记录数据属于那一类
TimeUse As Long '所用时间
Iterations As Long '迭带次数
ErrMsg As String '错误信息
End Type
'*************************************************************************************
'* 作 者 : laviewpbt
'* 函 数 名 : Fcm
'* 参 数 : Data - 待分类的样本,第一维的大小表示样本的个数,
'* 第二维的大小表示样本的维数
'* Cluster - 分类数
'* CreateIniCenter - 初始聚类中心的创建方法
'* AntiFuzzy - 反模糊化的方法
'* Exponent - 一个控制聚类效果的参数,一般取2
'* Maxiterations - 最大的迭代次数
'* MinImprovement - 最小的改进参数(两次迭代间聚类中心的距离)
'* 返回值 : FcmInfo结构,记录了相关的参数
'* 功能描述 : 利用模糊理论的聚类方法把数据分类
'* 日 期 : 2004-10-27 10.25.32
'* 修 改 人 : laviewpbt
'* 日 期 : 2006-11-7 19.25.31
'* 版 本 : Version 2.3.1
'**************************************************************************************
Private Function Fcm(ByRef Data() As Double, ByVal Cluster As Long, Optional ByVal CreateIniCenter As IniCenterMethod = IniCenterMethod.CreateByHcm, Optional AntiFuzzy As AntiFuzzyMethod = Max, Optional Exponent As Byte = 2, Optional Maxiterations As Long = 400, Optional MinImprovement As Double = 0.01, Optional ByRef CenterByHandle As Variant) As FcmInfo
If ArrayRange(Data) <> 2 Then
Fcm.ErrMsg = "数据只能为二维数组"
Exit Function
End If
Dim i As Long, j As Long, k As Long, l As Long, m As Long
Dim DataNumber As Long, DataSize As Long
Dim Temp As Double, Sum1 As Double, Sum2 As Double, Sum3 As Double, Index As Integer
Dim OldCenter() As Double
Fcm.TimeUse = GetTickCount
DataNumber = UBound(Data, 1): DataSize = UBound(Data, 2)
ReDim Fcm.center(1 To Cluster, 1 To DataSize) As Double
ReDim Fcm.Degree(1 To Cluster, 1 To DataNumber) As Double
ReDim Fcm.Class(1 To DataNumber) As Byte
ReDim OldCenter(1 To Cluster, 1 To DataSize) As Double
On Error GoTo ErrHandle:
Randomize
If CreateIniCenter = CreateRandom Then
For i = 1 To Cluster
For j = 1 To DataSize
OldCenter(i, j) = Data(Rnd * DataNumber, j) '产生随机初始中心点
Next
Next
ElseIf CreateIniCenter = CreateByHcm Then
Dim HcmCenter As HcmInfo
HcmCenter = Hcm(Data, Cluster)
For i = 1 To Cluster
For j = 1 To DataSize
OldCenter(i, j) = HcmCenter.center(i, j) '产生HCM初始中心点
Next
Next
ElseIf CreateIniCenter = CreateByRandomZadeh Then
ReDim RndDegree(1 To Cluster, 1 To DataNumber) As Double
Dim RndSum As Double
For i = 1 To Cluster
For j = 1 To DataNumber
RndDegree(i, j) = Rnd '创建随机的隶属度
Next
Next
For j = 1 To DataNumber
RndSum = 0
For i = 1 To Cluster
RndSum = RndSum + RndDegree(i, j)
Next
For i = 1 To Cluster
RndDegree(i, j) = RndDegree(i, j) / RndSum '隶属度矩阵每列之后必须为1
Next
Next
For i = 1 To Cluster
For j = 1 To DataSize
Sum1 = 0: Sum2 = 0
For k = 1 To DataNumber
Temp = Exp(Log(RndDegree(i, k)) * Exponent) '其实就是RndDegree(i, k)^Exponent
Sum1 = Sum1 + Temp * Data(k, j) '隶属度的平方乘以数值
Sum2 = Sum2 + Temp '隶属度的和
Next
OldCenter(i, j) = Sum1 / Sum2 '得到聚类中心
Next
Next
ElseIf CreateIniCenter = CreateByHand Then
If IsMissing(CenterByHandle) Then
Fcm.ErrMsg = "请提供初始聚类中心。."
Exit Function
ElseIf UBound(CenterByHandle, 1) <> Cluster Or UBound(CenterByHandle, 2) <> DataSize Then
Fcm.ErrMsg = "手工提供的初始聚类中心维数有错误."
Exit Function
End If
For i = 1 To Cluster
For j = 1 To DataSize
OldCenter(i, j) = CenterByHandle(i, j)
Next
Next
End If
Do
Fcm.Iterations = Fcm.Iterations + 1
For i = 1 To Cluster
本文转自feisky博客园博客,原文链接:http://www.cnblogs.com/feisky/archive/2009/03/14/1586312.html,如需转载请自行联系原作者