K-means的简单实现

这篇文章中的K-means是简单的实现，没有调初始质心点和K的值，k的值设为3，初始质心点是测试数据的前三个。

并且没有对数据进行归一化处理，只是用原始数据进行聚类。

下面是代码：

#include <iostream>
#include<sstream>
#include<fstream>
#include<vector>
#include<math.h>
#include<stdlib.h>
#define k 3
using namespace std;

typedef vector<double> Train;

int dataNum;
int dimNum;

int Min(double d1,double d2,double d3){
     if(d1<=d2 && d1<=d3)
        return 1;
     if(d2<=d1 && d2<=d3)
        return 2;
     if(d3<=d1 && d3<=d2)
        return 3;
}
void getMinDistance(vector<Train>& trains,vector<Train>Cen){
   double d1=0;
   double d2=0;
   double d3=0;
   for(int i=0;i<dataNum;i++){

    for(int j=1;j<dimNum;j++){
        d1 += (trains[i][j]-Cen[0][j])*(trains[i][j]-Cen[0][j]);
        d2 += (trains[i][j]-Cen[1][j])*(trains[i][j]-Cen[1][j]);
        d3 += (trains[i][j]-Cen[2][j])*(trains[i][j]-Cen[2][j]);
    }
     switch(Min(d1,d2,d3)){
    case 1:
        trains[i][0]=1; break;
    case 2:
        trains[i][0]=2; break;
    case 3:
        trains[i][0]=3; break;
    default :
        break;
        }
    d1=0;d2=0;d3=0;
   }
}

vector<Train> getNewCenter(vector<Train> trains){
     //重新计算质心点
     vector<Train> newCenter;

     Train centerdata1(dimNum+1,0);
     Train centerdata2(dimNum+1,0);
     Train centerdata3(dimNum+1,0);
     centerdata1[0]=1;
     centerdata2[0]=2;
     centerdata3[0]=3;

     int i1,i2,i3;
     i1=0;i2=0;i3=0;

     for(int i=0;i<dataNum;i++){
            int Cases = (int)(trains[i][0]);
        switch(Cases){
        case 1:
            i1++;
            for(int j=1;j<dimNum;j++){
                centerdata1[j] += trains[i][j];
            }
            break;
        case 2:
            i2++;
            for(int j=1;j<dimNum;j++){
                centerdata2[j] += trains[i][j];
            }
            break;
        case 3:
            i3++;
            for(int j=1;j<dimNum;j++){
                centerdata3[j] += trains[i][j];
            }
            break;
        default:
            break;
        }
     }
     for(int j=1;j<dimNum;j++){
        centerdata1[j]=centerdata1[j]/i1;
        centerdata2[j]=centerdata2[j]/i2;
        centerdata3[j]=centerdata3[j]/i3;
     }
     newCenter.push_back(centerdata1);
     newCenter.push_back(centerdata2);
     newCenter.push_back(centerdata3);

     return newCenter;
}


vector<Train> Sort(vector<Train>& trains,vector<Train>Cen){
    vector<Train>newCenter;
    //数据分类
    getMinDistance(trains,Cen);
    //重新计算质心点
    newCenter = getNewCenter(trains);
    return newCenter;
}


void KMeans(vector<Train>& trains){
    //确定初始质心点 just use the 1,2,3,data(cannot ensure they are randable)
    //the first sort
    int isNotChange=1;

    //init centers
    vector<Train>lastCen;
    vector<Train>Centers;
    for(int i=0;i<3;i++){
   //  Centers.push_back(trains[i]);
     lastCen.push_back(trains[i]);
    }

    while(isNotChange){
        Centers = Sort(trains,lastCen);
        if(Centers != lastCen ){
            lastCen = Centers;
        }
        else{
            isNotChange = 0;
        }
    }
    cout<<"聚类结果："<<endl;
    for(int i=0;i<dataNum;i++){
        for(int j=0;j<=dimNum;j++){
            cout<<trains[i][j]<<" ";
        }
        cout<<endl;
    }



}

int main()
{
   char fname[256];
   cout<<"请输入存放数据的文件名："<<endl;
   cin>>fname;
   cout<<"维数："<<endl;
   cin>>dimNum;
   cout<<"样本数目："<<endl;
   cin>>dataNum;
   ifstream infile(fname);
   if(!infile){
    cout<<"cannot open the choosen file!"<<fname<<endl;
    return 0;
   }
   vector<Train> trains;
   for(int i=0;i<dataNum&&!infile.eof();i++){
    string str;

    getline(infile,str); //read data from infile to str;  only one line

    istringstream istr(str); //read data from str to istr;

    Train train(dimNum+1,0);
    train[0]=i+1;
    for(int j=1;j<=dimNum;j++){
         istr>>train[j];
    }
    trains.push_back(train);
   }

   cout<<"开始聚类"<<endl;
   KMeans(trains);
   return 0;
}

测试结果如下图：

 

第一列即为聚类结果，1表示第一类，2表示第二类，3表示第3类

只是最简单的K-means过程，，

如果有好的调K-means的方法，欢迎提出~