聚类算法1——K-Means(算法步骤及matlab代码)

看了西关书的聚类算法，算法原理很容易明白，接下来就是整理成自己的理解思路，然后一步一步来实现算法，那么就来做吧。

k-means cluster 算法步骤

输入：txt样本数据；输出：样本的归属类

step1:初始化工作

输入，样本，k；输出：初始均值向量

step1.1：载入数据，选择k值，初始化存放每个类的容器

repeat

step2：根据距离相似度划分所有样本对应的类 (for 样本数量)

step2.1：更新样本所属类：计算样本xi与各均值向量的距离，根据距离最近原则将样本xi放入对应类容器中

step2.2：计算更新后的均值向量：求均值即可

step3判断是否应该停止更新（for k的数量）

step3.1若ui’!=ui,将ui更新为ui’，否则保持当前均值向量不变

step3.2若所有的均值向量均为更新,停止程序

end repeat

ok啦，接下来不废话上代码（Matlab发布形式）

K-Means代码及样本下载链接

function Main()
%funtion:k-means cluster
%Reference: Machine leaning by zhihua zhou
clc
clear
close all
%1.0initial
melon_data = load('melon4.0.txt');
plot(melon_data(:,1),melon_data(:,2),'+')
xlabel('density');ylabel('sugar rate');
melon_data(:,1)=[];
[m,n] = size(melon_data);k=3;
sample_vector = zeros(m,n,k);
cluster_vector = zeros(k,n);
cluster_vector = init_cluster_vector(k,melon_data);
%
%show original
plot(melon_data(:,1),melon_data(:,2),'+');hold on
plot(cluster_vector(:,1),cluster_vector(:,2),'pg');
xlabel('density');ylabel('sugar rate');
%}
%2.0 main loop
max_loop = 10;
for loop = 1:max_loop
    sample_vector = cluster(melon_data,cluster_vector);%2.1
    cur_cluster_vector = update_cluter_vector(sample_vector);%2.2
    show(sample_vector,cur_cluster_vector);
    %is_stop = find(pdist(cur_cluster_vector-cluster_vector)>0.1);
    if is_stop(cur_cluster_vector,cluster_vector)
        fprintf('循环次数为：%d\n',loop);
        break;
    else
        cluster_vector= cur_cluster_vector;
    end
end
end

subfunction

%1初始均值向量
function cluster_vector = init_cluster_vector(k,melon_data)
    cluster_vector =[];
    k_rows = ceil( size(melon_data,1)/k);
    start = 1;
    for i =1:k
        if i<k
            tem_vector = sum( melon_data(start:i*k_rows,:))/(i*k_rows-start+1);
            start = i*k_rows;
        else
            tem_vector = sum( melon_data(start:end,:))/(i*k_rows-start+1);
        end
        cluster_vector = [cluster_vector;tem_vector];
    end
end
%2.1更新样本所属类
function sample_vector = cluster(melon_data,cluster_vector)
    % return : m,n,k
    [m,n]= size(melon_data);
    k = size(cluster_vector,1);
    sample_vector =zeros(m,n,k);
    for i=1:length(melon_data)
        dist = pdist2(cluster_vector,melon_data(i,:));
        [~,min_ind]=min(dist);
        append_row =find(sample_vector(:,:,min_ind)==0);
        sample_vector(append_row(1),:,min_ind) =melon_data(i,:);
    end
end
%2.2更新均值向量
function cur_cluster_vector = update_cluter_vector(sample_vector)
    [m,n,k]=size(sample_vector);
    cur_cluster_vector = zeros(k,n);
    for i = 1:k
        zero_ind = find(sample_vector(:,:,i)==0);
        nums = zero_ind(1)-1;
        cur_cluster_vector(i,:) = sum(sample_vector(:,:,i),1)/nums;
    end
end

function show(sample_vector,cur_cluster_vector)
    close
    plot(sample_vector(:,1,1),sample_vector(:,2,1),'+r');hold on
    plot(sample_vector(:,1,2),sample_vector(:,2,2),'+g');
    plot(sample_vector(:,1,3),sample_vector(:,2,3),'+b');
    plot(cur_cluster_vector(1,1),cur_cluster_vector(1,2),'pr');
    plot(cur_cluster_vector(2,1),cur_cluster_vector(2,2),'pg');
    plot(cur_cluster_vector(3,1),cur_cluster_vector(3,2),'pb');
    xlabel('density');ylabel('sugar rate');
end

function isStop =is_stop(cur_cluster_vector,cluster_vector)
    isStop = 0;
    if cur_cluster_vector(1,1)==cluster_vector(1,1)&&...
            cur_cluster_vector(1,2)==cluster_vector(1,2)&&...
            cur_cluster_vector(2,1)==cluster_vector(2,1)&&...
            cur_cluster_vector(2,2)==cluster_vector(2,2)&&...
            cur_cluster_vector(3,1)==cluster_vector(3,1)&&...
            cur_cluster_vector(3,2)==cluster_vector(3,2)
        isStop = 1;
    end
end

循环次数为：3