eca.cpp

LCL分群法 改善cure kmean.......等

// ECA.cpp : Defines the entry point for the console application.
//

#include "stdafx.h"
#include "ECA.h"
#include "Globals.h"
#include "CCluster.h"
#include "CRecord.h"
#include "float.h"
#include "math.h"
#include "sort.h"
#include "CVerify.h"
#include "mmsystem.h"

#ifdef _DEBUG
#define new DEBUG_NEW
#endif


// The one and only application object

CWinApp theApp;

using namespace std;
extern float cluster_decomposition_algorithm(long density,float ratio,CPtrArray &data,CPtrArray &array,CPtrArray &outline);
extern void  competitive_cluster_algorithm(CPtrArray &array,long count);
extern void  cluster_nerge_algorithm(CPtrArray &array,CPtrArray &outline);
extern void  sort_cluster(CPtrArray &array);
extern BOOL  is_merge_cluster(CCluster *c1,CCluster *c2,long threshold);
extern BOOL  have_boundary_points(CCluster *c1,CCluster *c2,float r);
extern void  normalize(CPtrArray &data);
extern float cal_davies_bouldin_validity_index(CPtrArray &array);

int _tmain(int argc, TCHAR* argv[], TCHAR* envp[])
{
	int nRetCode = 0;

	// initialize MFC and print and error on failure
	if (!AfxWinInit(::GetModuleHandle(NULL), NULL, ::GetCommandLine(), 0))
	{
		// TODO: change error code to suit your needs
		_tprintf(_T("Fatal Error: MFC initialization failed\n"));
		nRetCode = 1;
	}
	else
	{
		// TODO: code your application's behavior here.
		CString dataset = "D:\\cluster.dat"; if(argc >= 2){dataset = argv[1];}
		CString output  = "D:\\output.dat";  if(argc >= 3){output  = argv[2];}
		long    test    = 10;                if(argc >= 4){test    = max(1,atoi(argv[3]));}
		float   ratio   = 2.0;               if(argc >= 5){ratio   = max(1,atof(argv[4]));}
		long    density = 5;                 if(argc >= 6){density = max(1,atoi(argv[5]));}
		CString nor     = "N";               if(argc >= 7){nor     = argv[6];}
		CPtrArray data;
		if(read_data(dataset,data) == FALSE){
			return nRetCode;
		}
		if(nor == "Y" || nor == "y"){
			normalize(data);
		}
		long start_time = timeGetTime();
			CPtrArray array,outline;
			float r = cluster_decomposition_algorithm(density,ratio,data,array,outline);
			long t = 0;
			while(t < test){
				long count = (long)array.GetCount();
				competitive_cluster_algorithm(array,(long)data.GetSize());
				if(array.GetSize() == count) {
					t++;
				} else {
					t = 0;
				}
			}
			cluster_nerge_algorithm(array,outline);
		long end_time = timeGetTime();

		printf("Input file : %s\n",dataset);
		printf("Output file : %s\n",output);
		printf("Runtime : %.3lf\n",(end_time-start_time)/1000.0);
		printf("Radius : %.3lf\n",r);
		printf("Within-cluster distance : nearest neighor distance\n");
		printf("Between-clusters distance : single linkage\n");
		printf("Davies Bouldin validity index : %.3lf\n",cal_davies_bouldin_validity_index(array));

	CPtrList verify;
	for(long i=0;i<array.GetSize();i++){
		CCluster *c = (CCluster *)array[i];
		for(long j=0;j<c->get_record_array().GetSize();j++){
			CRecord *r = (CRecord *)c->get_record_array().GetAt(j);
			CString str;
			str.Format("%s-%s",c->get_label(),r->get_label());
			BOOL ok = FALSE;
			POSITION pos = verify.GetHeadPosition();
			while(pos){
				CVerify *v = (CVerify *)verify.GetNext(pos);
				if(v->get_label() == str){
					v->add_count();
					ok = TRUE;
					break;
				}
			}
			if(ok == FALSE){
				CVerify *v = new CVerify(str,c->get_label(),r->get_label());
				verify.AddTail(v);
			}
		}
	}

	CPtrList list;
	float diff = 0;
	while(!verify.IsEmpty()){
		CVerify *v = (CVerify *)verify.RemoveHead();
		POSITION pos = verify.GetHeadPosition();
		while(pos){
			POSITION prev = pos;
			CVerify *vv = (CVerify *)verify.GetNext(pos);
			if(strcmp(v->get_name1(),vv->get_name1()) != 0){
				continue;
			}
			verify.RemoveAt(prev);
			if(v->get_count() > vv->get_count()){
				diff += vv->get_count();
				delete vv;
			} else {
				diff += v->get_count();
				delete v;
				v = vv;
			}
		}
		list.AddTail(v);
	}
	verify.AddTail(&list);
	list.RemoveAll();
	while(!verify.IsEmpty()){
		CVerify *v = (CVerify *)verify.RemoveHead();
		POSITION pos = verify.GetHeadPosition();
		while(pos){
			POSITION prev = pos;
			CVerify *vv = (CVerify *)verify.GetNext(pos);
			if(strcmp(v->get_name2(),vv->get_name2()) != 0){
				continue;
			}
			verify.RemoveAt(prev);
			if(v->get_count() > vv->get_count()){
				diff += vv->get_count();
				delete vv;
			} else {
				diff += v->get_count();
				delete v;
				v = vv;
			}
		}
		list.AddTail(v);
	}
	verify.AddTail(&list);
	list.RemoveAll();

//	printf("Criterion function : %.3lf\n",mse);
	printf("Accuracy : %.3lf\n",1.0-diff/data.GetSize());


	while(!verify.IsEmpty()){
		CVerify *v = (CVerify *)verify.RemoveHead();
		delete v;
	}
		write_data(output,array);

		for(long i=0;i<outline.GetSize();i++){
			CCluster *c = (CCluster *)outline[i];
			delete c;
		}

		for(long i=0;i<array.GetSize();i++){
			CCluster *c = (CCluster *)array[i];
			delete c;
		}
		for(long i=0;i<data.GetSize();i++){
			CRecord *r = (CRecord *)data[i];
			delete r;
		}
	}

	return nRetCode;
}

float cluster_decomposition_algorithm(long density,float ratio,CPtrArray &data,CPtrArray &array,CPtrArray &outline)
{
/*
	//衡