凯鲁嘎吉
用书写铭记日常，最迷人的不在远方
posts - 292,comments - 409,views - 138万
ISODATA聚类算法的matlab程序
回到顶部(go to top)

ISODATA聚类算法的matlab程序

作者：凯鲁嘎吉 - 博客园 http://www.cnblogs.com/kailugaji/

参考：Kmeans及ISODATA算法的matlab实现

算法简介：聚类算法：ISODATA算法

数据见：MATLAB实例：PCA降维中的iris数据集，保存为：iris.data，最后一列是类标签。

demo_isodata.m

?
1
2
3
4
5
6
7
8
9
10
11
12
clear
clc
data_load=dlmread('iris.data');
[~,dim]=size(data_load);
x=data_load(:,1:dim-1);
K=3;
theta_N=1;
theta_S=1;
theta_c=4;
L=1;
I=5;
ISODATA(x,K,theta_N,theta_S,theta_c,L,I)

ISODATA.m

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
function ISODATA(x,K,theta_N,theta_S,theta_c,L,I)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    %%%%%%%%input parameters%%%%%%
% x : data
% K : 预期的聚类中心数
% theta_N : 每一聚类中心中最少的样本数，少于此数就不作为一个独立的聚类
% theta_S ：一个聚类中样本距离分布的标准差
% theta_c : 两聚类中心之间的最小距离，如小于此数，两个聚类进行合并
% L : 在一次迭代运算中可以和并的聚类中心的最多对数
% I ：迭代运算的次数序号
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% step1
n = size(x,1);
N_c = K;
mean = cell(K,1);
for i=1:K
    mean{i} = x(i,:);
end
ite = 1;
while ite<I
    flag = 1;
    while flag
    %% step2
    class = cell(size(mean));
    for i=1:n
        num = Belong2(x(i,:),mean);
        class{num} =  [class{num};x(i,:)];
    end
    %% step3
    for i=1:N_c
        size_i = size(class{i},1);
        if size_i<theta_N
          class_i = class{i};
          mean = DeleteRow(mean,i);
          class = DeleteRow(class,i);
          N_c = N_c-1;
          for j=1:size_i
            class_ij = class_i(j,:);%the j'th row of class{i}
            num = Belong2(class_ij,mean);
            class{num} = [class{num};class_ij];
          end
        end
    end
 
    %% step4
    for i=1:N_c
        if ~isempty(mean{i})
            mean{i} = sum(class{i})./size(class{i},1);
        end
    end
    %% step5
    Dis = zeros(N_c,1);
    for i=1:N_c
        if ~isempty(class{i})
            N_i =size(class{i},1);
            tmp = bsxfun(@minus,class{i},mean{i});
            Dis(i) = sum(arrayfun(@(x)norm(tmp(x,:)),1:N_i))/N_i;
        end
    end
    %% step6
    D = 0;
    for i=1:N_c
        if ~isempty(class{i})
            N_i =size(class{i},1);
            D = D + N_i*Dis(i);
        end
    end
    D = D/n;
    %% step7
    flag = 0;
    if ite == I
        theta_c = 0;
        flag = 0;
    elseif ~(N_c > K/2)
        flag = 1;
    elseif mod(ite,2)==0 || ~(N_c<2*K)
        flag = 0;
    end
    %% 分裂处理
    %% step8
    if flag
        flag = 0;
        delta = cell(N_c,1);
        for i=1:N_c
            if ~isempty(class{i})
                 N_i =size(class{i},1);
                 tmp = bsxfun(@minus,class{i},mean{i});
                 delta{i} = arrayfun(@(x)norm(tmp(:,x)),1:size(tmp,2))/N_i;
            end
        end
 
    %% step9
    delta_max = cell(N_c,1);
    for i=1:N_c
        if ~isempty(class{i})
            max_i = max(delta{i});
            sub = find(delta{i}==max_i,1);
            delta_max{i} = [max_i,sub];
        end
    end
    %% step10  
    for i=1:N_c
        if delta_max{i}(1) > theta_S
            N_i =size(class{i},1);
            con1 = (Dis(i)>D && N_i>2*(theta_N + 1));
            con2 = ~(N_c>K/2);
            if con1 || con2
               %%%%这里分裂%%%%%
               flag = 1;%一旦发生分裂，那么分裂一次后就返回第二步；若没发生分裂，则直接进入合并处理步
               lamda = 0.5;
               max_sub = delta_max{i}(2);
               mean{i}(max_sub) = mean{i}(max_sub) + lamda * delta_max{i}(1);
               addOneMean =  mean{i};
               addOneMean(max_sub) = addOneMean(max_sub) - lamda * delta_max{i}(1);
               mean = [mean;addOneMean];
               N_c = N_c+1;
               break;
            end
        end
     end
 
    end
 
    end
    %% 合并处理
    if L
    %% step11
    Distance = zeros(N_c,N_c);
    for i=1:N_c-1
        for j=i:N_c
            Distance(i,j) = norm(mean{i}-mean{j});
        end
    end
    %% step12
    index = find(-Distance>theta_c);
    keepIndex = [Distance(index),index];
    [~, index] = sort(keepIndex(:,1));
    if size(index,1) > L
        index = index(1:L,:);
    end
    %% step13
    if size(index,1) ~= 0
        for id=1:size(index,1)
            [m_i m_j]= seq2idx(index(id),N_c);
            %%%%%这里合并%%%%%
            N_mi = size(class{m_i},1);
            N_mj = size(class{m_j},1);
            mean{m_i} = (N_mi*mean{m_i} + N_mj*mean{m_j})/(N_mi+N_mj);
            mean = DeleteRow(mean,m_j);
            class{m_i} = [class{m_i};class{m_j}];
            class = DeleteRow(class,m_j);
        end  
    end
    end
    %% step14
    ite=ite+1;
end
   for  i=1:N_c
       fprintf('第%d类聚类中心为\n',i);
       disp(mean{i});
       fprintf('第%d类中元素为\n',i);
       disp(class{i});
   end
end
 
 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function number = Belong2(x_i,means)
    INF = 10000;
    min = INF;
    kk = size(means,1);
    number = 1;
    for i=1:kk
        if ~isempty(means{i})
            if norm(x_i - means{i}) < min
                min = norm(x_i - means{i});
                number = i;
            end
        end
    end
end
 
function A_del = DeleteRow(A,r)
    n = size(A,1);
    if r == 1
        A_del = A(2:n,:);
    elseif r == n
        A_del = A(1:n-1,:);
    else
        A_del = [A(1:r-1,:);A(r+1:n,:)];
    end
end
 
 
function [row col] = seq2idx(id,n)
    if mod(id,n)==0
        row = n;
        col = id/n;
    else
        row = mod(id,n);
        col = ceil(id/n);
    end
end

结果

?
1
>> demo_isodata

2019-10-10 16:10:14

posted on   凯鲁嘎吉  阅读(5223)  评论(19)  编辑  收藏  举报
编辑推荐：
· 10年+ .NET Coder 心语，封装的思维：从隐藏、稳定开始理解其本质意义
· .NET Core 中如何实现缓存的预热？
· 从 HTTP 原因短语缺失研究 HTTP/2 和 HTTP/3 的设计差异
· AI与.NET技术实操系列：向量存储与相似性搜索在 .NET 中的实现
· 基于Microsoft.Extensions.AI核心库实现RAG应用
阅读排行：
· 10年+ .NET Coder 心语 ── 封装的思维：从隐藏、稳定开始理解其本质意义
· 地球OL攻略 —— 某应届生求职总结
· 提示词工程——AI应用必不可少的技术
· Open-Sora 2.0 重磅开源！
· 字符编码：从基础到乱码解决

Copyright © 2025 凯鲁嘎吉
Powered by .NET 9.0 on Kubernetes Powered By博客园
点击右上角即可分享
微信分享提示
AI IDE Trae