动态聚类中 C－均值算法 (K－均值算法)的C++实现

来源：互联网发布：淘宝优惠券采集api 编辑：程序博客网时间：2024/04/30 12:38

一：说明

动态聚类方法是模式识别中一种普遍采用的方法，它具有以下3个要点：

1：选定某种距离度量作为样本间的相似性度量

2：确定某个评价聚类结果质量的准则函数

3：给定某个初始分类，然后用迭代算法找出使准则函数取极值的最好的聚类结果

本文给出了 C－均值算法的 C++ 实现。

（算法描述参见边肇祺张学工等 << 模式识别 >> P237 清华大学出版社）

二：源码

2.1 头文件

#pragma once

#include <list>

#include <vector>

using namespace std;

#define DATANUM 19

#define MAXDIST 333333

struct CData

{

int x1;

int x2;

};

class CCMean

{

public:

CCMean(CData *pdata);

~CCMean(void);

void init();

void work(int InitClassNum);

private:

// calculate the mean of class i:

void CalcuMean( int i );

// calculate the ERROR of class i:

void CalcuJc(int i);

void CalcuJe();

// step 1 of C-Mean algorithm

void InitDeploy();

// step 4 and 5 of C-Mean algorithm

// da is now in class i,

// return ture when moving da from class to class k

// return false when we do not move da

bool MoveItoK( const CData& da, int i, int &k );

// calculate the distance of to data:

int dist( const CData& mean, const CData& da);

// print result:

void OutPut();

// iClassNum is the initial class number, in text book, iClassNum <==> C

int iClassNum;

// pointer to data array:

CData *pData;

// store the mean of all classes. just ueses 0 to iClassNum - 1;

// in text book is: m1, m2, m3, m4, ... , mc.

CData mean[DATANUM];

// store the ERROR of each class, just ueses 0 to iClassNum - 1;

// the sum of jc[0] to jc[iClassNum - 1] will be je defined following jc;

int jc[DATANUM];

//the sum of jc[0] to jc[iClassNum - 1]

int je;

// pcla[i] pointer class i which store in LIST

list< CData >* pcla[DATANUM];

};

2.2 实现文件

#include "assert.h"

#include "cmean.h"

CCMean::CCMean(CData *pdata)

{

pData = pdata;

for(int i = 0; i < DATANUM; i ++ )

{

pcla[i] = new list< CData >;

assert( pcla[i] != 0 );

}

je = 0;

}

CCMean::~CCMean()

{

for(int i = 0; i < DATANUM; i ++ )

delete pcla[i];

}

void CCMean::init()

{

for(int i = 0; i < DATANUM; i ++ )

{

pcla[i]->clear();

mean[i].x1 = 0;

mean[i].x2 = 0;

}

je = 0;

}

void CCMean::CalcuMean(int ii)

{

int sum1 = 0, sum2 = 0;

int si = (int)pcla[ii]->size();

list< CData >::iterator iter = pcla[ii]->begin();

for(int i = 0; i < si; i ++ )

{

sum1 += iter->x1;

sum2 += iter->x2;

iter++;

}

mean[ii].x1 = sum1 / si;

mean[ii].x2 = sum2 / si;

}

void CCMean::CalcuJe()

{

for( int i = 0; i < iClassNum ; i ++ )

{

CalcuJc( i );

je += jc[i];

}

void CCMean::CalcuJc( int index )

{

list< CData >::iterator iter = pcla[index]->begin();

int si = (int)pcla[index]->size();

jc[index] = 0;

for( int i = 0; i < si; i ++)

{

jc[index] += dist( mean[index], *iter );

iter ++;

}

int CCMean::dist(const CData& mean, const CData& da)

{

return (mean.x1 - da.x1)*(mean.x1 - da.x1) + (mean.x2 - da.x2)*(mean.x2 - da.x2);

}

void CCMean::InitDeploy()

{

CData *ptem = pData;

for( int i = 0; i < iClassNum; i ++ )

{

// choose the first iClassNum data as our initial class-center:

mean[i] = *ptem;

pcla[i]->push_back( *ptem );

ptem++;

}

// put other data to our initial classes:

for( int i = iClassNum; i < DATANUM; i ++ )

{

int mindis = MAXDIST;

int pos = 0;

// get the least distance between pData[i] and m1, m2, m3 ....

for( int j = 0; j < iClassNum; j ++ )

{

int curdis = dist( pData[i], mean[j] );

if( curdis < mindis )

{

mindis = curdis;

pos = j;

}

// add pData to class (pos):

pcla[pos]->push_back( pData[i] );

}

for( int i = 0; i < iClassNum ; i ++ )

CalcuMean( i );

CalcuJe();

}

bool CCMean::MoveItoK( const CData &da, int i , int& k )

{

// now da is in class i,if da is moved to another class, return true, else return false

int Pk = MAXDIST;

int Pj = 0;

int temk = 0;

for( int j = 0; j < iClassNum; j ++ )

{

int si = (int)pcla[j]->size();

if( j == i )

Pj = dist( mean[j], da ) * si/(si - 1);

else

Pj = dist( mean[j], da ) * si/(si + 1);

if( Pj < Pk )

{

Pk = Pj;

temk = j;

}

else if ( Pj == Pk && j == i )

{

// when Pj == Pk && j == i, we do not move (da) from class i to class j

temk = i;

}

if( i == temk )

return false; // we do NOT move da;

k = temk;

// add da to class k:

pcla[k]->push_back( da );

// delete da from class i, first find the positon of da in class i:

list< CData >::iterator iter = pcla[i]->begin();

while( iter != pcla[i]->end() )

{

if( iter->x1 == da.x1 && iter->x2 == da.x2 )

break;

iter++;

}

// now delete da from class i:

pcla[i]->erase( iter );

// we have move da from class i to class k;

return true;

}

void CCMean::OutPut()

{

for( int i = 0; i < iClassNum ; i ++ )

{

printf("class %d:/n", i );

list< CData >::iterator iter = pcla[i]->begin();

int j = 1;

while( iter != pcla[i]->end() )

{

printf( "(%d, %d) ", iter->x1, iter->x2 );

iter ++;

if( j++ % 5 == 0)

printf("/n");

}

printf("/n");

}

void CCMean::work(int InitClassNum)

{

iClassNum = InitClassNum;

// step 1 of C-Mean algorithm

InitDeploy();

int counter = 0;

Again:

//OutPut();

// step 2 of C-Mean algorithm: choose one sample y (here is da) from collection

for( int i = 0; i < iClassNum ; i ++ )

{

// step 3 of C-Mean algorithm:

int si = (int)pcla[i]->size();

if( si == 1 )

continue;

// step 4 of C-Mean algorithm:

list< CData >::iterator iter = pcla[i]->begin();

for(int j = 0; j < (int)pcla[i]->size(); j++)

{

int k = 0;

CData da = *iter;

iter ++;

// step 5 of C-Mean algorithm:

if( MoveItoK( da , i, k ) == true )

{

// step 6 of C-Mean algorithm:

int OldJe = je;

je -= jc[i];

je -= jc[k];

CalcuMean( i );

CalcuMean( k );

CalcuJc( i );

CalcuJc( k );

je += jc[i];

je += jc[k];

if( OldJe > je )

{

counter = 0;

goto Again;

}

counter++;

// step 7 of C-Mean algorithm:

if( counter == DATANUM )

goto end;

}

end:

printf(" current Je is: %d/n", je );

OutPut();

}

2.3 测试文件

#include "CMean.h"

#include "process.h"

CData yy[DATANUM] =

{

{0,0},{0,1},{1,0},{1,1},{1,2},{2,1},{2,2},{2,3}

,{6,6},{6,7},{6,8},{7,6},{7,7},{7,8},{7,9},{8,7}

,{8,8},{8,9},{9,8}

};

int main(int argc, char* argv[])

{

CCMean cmean( yy );

cmean.work(2);

system("pause");

return 0;

}