winsvmdlg.cpp

采用c++编写的有关svm交叉验证的程序。

		{
			MessageBox("嘻嘻，训练还没结束呢！");
			return;
		}
	}
	m_yuce=new Cyuce;
	if(m_yuce->DoModal()==IDOK)
	{
     //   char* line;
      //  int max_line_len = 1024;
        
        max_nr_attr = 64;
        char input0[1024],output0[1024],model0[1024];


	    FILE *input, *output;
	    strcpy(input0, m_yuce->m_input);
	    strcpy(model0, m_yuce->m_model);
		strcpy(output0, m_yuce->m_output);
        predict_probability = m_yuce->predict_probability;

	    input = fopen(input0,"r");//预测数据文件
	    if(input == NULL)
	    {
		    //fprintf(stderr,"can't open input file %s\n",argv[i]);
		    //exit(1);
		    CString strs;
		    strs.Format("不能打开文件 %s\n",input0);
		    MessageBox(strs);
		    return;
	    }

	    output = fopen(output0,"w");//预测结果文件保存地址
	    if(output == NULL)
	    {
		    //fprintf(stderr,"can't open output file %s\n",argv[i+2]);
		    //exit(1);
		    CString strs;
		    strs.Format("不能保存文件 %s\n",output0);
		    MessageBox(strs);
		    return;
	    }

	    if((trainmodel=svm_load_model(model0))==0)//模型文件地址选择
	    {
		    //fprintf(stderr,"can't open model file %s\n",argv[i+1]);
		    //exit(1);
		    //CString strs;//不需要了 因为svm_load_model(model0))里面有消息弹出，更说明问题
		    //strs.Format("不能打开文件 %s\n",model0);
		    //MessageBox(strs);
			return;
	    }
	  //  line = (char *) malloc(max_line_len*sizeof(char));
	    x = (struct svm_node *) malloc(max_nr_attr*sizeof(struct svm_node));
	    if(predict_probability)
		    if(svm_check_probability_model(trainmodel)==0)
		    {
			    //fprintf(stderr,"Model does not support probabiliy estimates\n");
			    //exit(1);
                MessageBox("训练好的model文件不支持概率预报");
				return;
		    }
	    predict(input,output);
	    svm_destroy_model(trainmodel);
	//    free(line);
	    free(x);
	    fclose(input);
	    fclose(output);

		delete m_yuce;
		m_yuce=NULL;
		return;
	};
	delete m_yuce;
	m_yuce=NULL;
}

void CwinsvmDlg::predict(FILE *input, FILE *output)
{
	int correct = 0;
	int total = 0;
	double error = 0;
	double sumv = 0, sumy = 0, sumvv = 0, sumyy = 0, sumvy = 0;

//    CTime  ts1 = CTime::GetCurrentTime();

	int svm_type=svm_get_svm_type(trainmodel);
	int nr_class=svm_get_nr_class(trainmodel);
	int *labels=(int *) malloc(nr_class*sizeof(int));
	double *prob_estimates=NULL;
	int j;
    CString str;//添加
	if(predict_probability)
	{
		if (svm_type==NU_SVR || svm_type==EPSILON_SVR)
		{
			//printf("Prob. model for test data: target value = predicted value + z,\nz: Laplace distribution e^(-|z|/sigma)/(2sigma),sigma=%g\n",svm_get_svr_probability(model));
		    str.Format("测试数据的样本: 目标值 = 预测值 + z,\nz: 拉普拉斯分布 e^(-|z|/sigma)/(2sigma),sigma=%g\n",svm_get_svr_probability(trainmodel));
		    info+=str;
		}
		else
		{
			svm_get_labels(trainmodel,labels);
			prob_estimates = (double *) malloc(nr_class*sizeof(double));
			fprintf(output,"labels");		
			for(j=0;j<nr_class;j++)
				fprintf(output," %d",labels[j]);
			fprintf(output,"\n");
		}
	}
	while(1)
	{
		int i = 0;
		int c;
		double target,v;

		if (fscanf(input,"%lf",&target)==EOF)
			break;

		while(1)
		{
			if(i>=max_nr_attr-1)	// need one more for index = -1
			{
				max_nr_attr *= 2;
				x = (struct svm_node *) realloc(x,max_nr_attr*sizeof(struct svm_node));
			}

			do {
				c = getc(input);
				if(c=='\n' || c==EOF) goto out2;
			} while(isspace(c));
			ungetc(c,input);
			if (fscanf(input,"%d:%lf",&x[i].index,&x[i].value) < 2)
			{
				//fprintf(stderr,"Wrong input format at line %d\n", total+1);
				//exit(1);
		        CString strs;
		        strs.Format("错误的输入格式在 %d行\n",total+1);
		        MessageBox(strs);info.Empty();
		        return;
			}
			++i;
		}	

out2:
		x[i++].index = -1;

		if (predict_probability && (svm_type==C_SVC || svm_type==NU_SVC))
		{
			v = svm_predict_probability(trainmodel,x,prob_estimates);
			fprintf(output,"%g ",v);
			for(j=0;j<nr_class;j++)
				fprintf(output,"%g ",prob_estimates[j]);
			fprintf(output,"\n");
		}
		else
		{
			v = svm_predict(trainmodel,x);
			fprintf(output,"%g\n",v);
		}

		if(v == target)
			++correct;
		error += (v-target)*(v-target);
		sumv += v;
		sumy += target;
		sumvv += v*v;
		sumyy += target*target;
		sumvy += v*target;
		++total;
	}
	//printf("Accuracy = %g%% (%d/%d) (classification)\n",
	//	(double)correct/total*100,correct,total);
	//printf("Mean squared error = %g (regression)\n",error/total);
	//printf("Squared correlation coefficient = %g (regression)\n",
	//	((total*sumvy-sumv*sumy)*(total*sumvy-sumv*sumy))/
	//	((total*sumvv-sumv*sumv)*(total*sumyy-sumy*sumy))
	//	);

	if (svm_type==NU_SVR || svm_type==EPSILON_SVR)
	{
	str.Format("方差 = %g \n",error/total);
	info+=str;
	str.Format("平方相关系数(相似性) = %g \n",
		((total*sumvy-sumv*sumy)*(total*sumvy-sumv*sumy))/
		((total*sumvv-sumv*sumv)*(total*sumyy-sumy*sumy))
		);
	info+=str+"\n";
	}
	else
	{
	str.Format("精确度 = %g%% (%d/%d)\n",
		(double)correct/total*100,correct,total);
	info+=str;
	}
 //   CTimeSpan  ts = CTime::GetCurrentTime()-ts1;
	//str.Format("预测时间为%d日%d小时%d分钟%d秒",ts.GetDays(),ts.GetHours() ,ts.GetMinutes(),ts.GetSeconds());
 //   info+=str;
	if(predict_probability)
	{
		free(prob_estimates);
		free(labels);
	}
    MessageBox(info);
	info.Empty();

}

void CwinsvmDlg::OnBnClickedButton3()
{
	if (m_guyihua==NULL)
	{
	    m_guyihua=new Cguyihua;
		m_guyihua->Create(IDD_DIALOG3,this);
		m_guyihua->ShowWindow(1);
	}
	else
	{
	    delete m_guyihua;
	    m_guyihua=NULL;
	}

	//if(m_guyihua->DoModal()==IDOK)
	//{	   
	//	//MessageBox(info);
	//	delete m_guyihua;
	//	m_guyihua=NULL;
	//}
	//delete m_guyihua;
	//m_guyihua=NULL;

}

//遗传求参数
void CwinsvmDlg::yichuan(int i,int a,int aend,double b,double bend,double bstep,double c,double cend,double cstep)
{
	//int j=aend-a;//度的步数
	int k=(bend-b)/bstep;//coef0步数
	int l=(cend-c)/cstep;//C步数
	double zy;//预测效率
	int m=0;//递归参数
//临时变量
	int degreetemp(0);
	double coef0temp(0);
	double Ctemp(0);
	for( int ii=a;ii<aend;ii++)
	{
		param.degree=ii;
		for( double jj=b;jj<bend;jj+=bstep)
		{
			param.coef0=jj;
			for( double kk=c;kk<cend;kk+=cstep)
			{
				param.C=kk;
				info.Empty();//放这里增加运算
				zy = do_cross_validationy();
				if( zy > (limit+limitstep) )
				{
					m++;
					limit=zy;
					degreetemp=param.degree;
				    coef0temp=param.coef0;
				    Ctemp=param.C;
				    infoy=info;
				    info.Format("度是 %d,coef0是 %f, 惩罚因子C是 %f\n",degreetemp,coef0temp,Ctemp);
				    infoy+=info;
 				}
	        }
		}
	}
	if(m)
	{
		limitstep=limitstep/3;	
		a=degreetemp;aend=degreetemp+1;
	    b=coef0temp-bstep+4*bstep/(k+4),bend=coef0temp+bstep-4*bstep/(k+4),bstep=4*bstep/(k+4);//可以为负数//步宽先改变//步数减半-2处理，再改变步长
	    c=Ctemp-cstep+4*cstep/(l+4);
		if(c<=0)
			c=0.00000001,cend=Ctemp+cstep-2*cstep/(l+2),cstep=2*cstep/(l+2);
		else
			cend=Ctemp+cstep-4*cstep/(l+4),cstep=4*cstep/(l+4);
		yichuan(++i,a,aend, b, bend, bstep, c, cend, cstep);
		i--;
	}
	if(i==0)
	{
	yichuanend();
	}
}

void CwinsvmDlg::yichuan(int i,double a,double aend,double astep)
{	
	int l=(aend-a)/astep;//C步数
	double zy;//预测效率
	int m=0;//递归参数
//临时变量
	double Ctemp(0);
	for(double ii=a;ii<aend;ii+=astep)
	{
		param.C=ii;
		info.Empty();
		zy = do_cross_validationy();
		if( zy > (limit+limitstep) )
		{
			m++;
			limit=zy;
			Ctemp=param.C;
			infoy=info;
			info.Format("惩罚因子C是 %f\n",Ctemp);
			infoy+=info;
 		}
	}
	if(m)
	{
		limitstep=limitstep/3;	
	    a=Ctemp-astep+4*astep/(l+4);
		if(a<=0)
			a=0.00000001,aend=Ctemp+astep-2*astep/(l+2),astep=2*astep/(l+2);
		else
			aend=Ctemp+astep-4*astep/(l+4),astep=4*astep/(l+4);	 
		yichuan(++i,a,aend,astep);
		i--;
	}
	if(i==0)
	{
	yichuanend();
	}
}

void CwinsvmDlg::yichuan(int i,double b,double bend,double bstep,double c,double cend,double cstep)
{
	int k=(bend-b)/bstep;//coef0步数
	int l=(cend-c)/cstep;//C步数
	double zy;//预测效率
	int m=0;//递归参数
//临时变量
	double coef0temp(0);
	double Ctemp(0);	
	for(double ii=b ;ii<bend;ii+=bstep)
	{
		param.coef0=ii;
		for(double jj=c ;jj<cend;jj+=cstep)
		{
			param.C=jj;
			info.Empty();
			zy = do_cross_validationy();
			if( zy > (limit+limitstep) )
			{
				m++;
				limit=zy;
				coef0temp=param.coef0;
				Ctemp=param.C;
				infoy=info;
				info.Format("coef0是 %f, 惩罚因子C是 %f\n",coef0temp,Ctemp);
				infoy+=info;
 			}
	    }
	}
	if(m)
	{
		limitstep=limitstep/3;	
	    b=coef0temp-bstep+4*bstep/(k+4),bend=coef0temp+bstep-4*bstep/(k+4),bstep=4*bstep/(k+4);//可以为负数
	    c=Ctemp-cstep+4*cstep/(l+4);
		if(c<=0)
			c=0.00000001,cend=Ctemp+cstep-2*cstep/(l+2),cstep=2*cstep/(l+2);
		else
			cend=Ctemp+cstep-4*cstep/(l+4),cstep=4*cstep/(l+4);
		yichuan(++i, b, bend, bstep, c, cend, cstep);
		i--;
	}
	if(i==0)
	{
	yichuanend();
	}
}

void CwinsvmDlg::yichuan1(int i,int a,int aend,double b,double bend,double bstep,double c,double cend,double cstep)
{
	//int j=aend-a;//度的步数
	int k=(bend-b)/bstep;//coef0步数
	int l=(cend-c)/cstep;//nu步数
	double zy;//预测效率
	int m=0;//递归参数
//临时变量
	int degreetemp(0);
	double coef0temp(0);
	double nutemp(0);	
	for( double kk=c;kk<cend;kk+=cstep)
	{
		param.nu=kk;
	    for(int ii=a ;ii<aend;ii++)
	    {
		    param.degree=ii;
		    for(double jj=b ;jj<bend;jj+=bstep)
		    {
			    param.coef0=jj;
				info.Empty();
				zy = do_cross_validationy();
				if( zy > (limit+limitstep) )
				{
					m++;
					limit=zy;
					degreetemp=param.degree;
				    coef0temp=param.coef0;
				    nutemp=param.nu;
				    infoy=info;
				    info.Format("度是 %d,coef0是 %f, 参数v（nu）是 %f\n",degreetemp,coef0temp,nutemp);
				    infoy+=info;
 				}
	        }
		}
	}
	if(m)
	{
		limitstep=limitstep/3;
		a=degreetemp;aend=degreetemp+1;
	    b=coef0temp-bstep+4*bstep/(k+4),bend=coef0temp+bstep-4*bstep/(k+4),bstep=4*bstep/(k+4);//可以为负数
		c=nutemp-cstep+4*cstep/(l+4);
		if(c<=0)
			c=0.00000001,cend=nutemp+cstep-2*cstep/(l+2),cstep=2*cstep/(l+2);
		else
		{
			cend=nutemp+cstep-4*cstep/(l+4),cstep=4*cstep/(l+4);
			if(cend>1)
				cend=1;
		}		
		yichuan1(++i, a,aend, b, bend, bstep, c, cend, cstep);
		i--;
	}
	if(i==0)
	{
	yichuanend();
	}
}
void CwinsvmDlg::yichuan1(int i,double c,double cend,double cstep)
{
	int l=(cend-c)/cstep;//nu步数
	double zy;//预测效率
	int m=0;//递归参数
//临时变量
	double nutemp(0);
	for( double ii=c;ii<cend;ii+=cstep)
		{
			param.nu=ii;
			info.Empty();
			zy = do_cross_validationy();
			if( zy > (limit+limitstep) )
			{
				m++;
				limit=zy;
				nutemp=param.nu;
				infoy=info;
				info.Format("参数v（nu）是 %f\n",nutemp);
				infoy+=info;
 			}
	    }
	if(m)
	{
		limitstep=limitstep/3;
		c=nutemp-cstep+4*cstep/(l+4);
		if(c<=0)
			c=0.00000001,cend=nutemp+cstep-2*cstep/(l+2),cstep=2*cstep/(l+2);
		else
		{
			cend=nutemp+cstep-4*cstep/(l+4),cstep=4*cstep/(l+4);
			if(cend>1)
				cend=1;
		}
		yichuan1(++i, c, cend, cstep);
		i--;
	}
	if(i==0)
	{
	yichuanend();
	}
}
void CwinsvmDlg::yichuan1(int i,double b,double bend,double bstep,double c,double cend,double cstep)
{
	//int j=aend-a;//度的步数
	int k=(bend-b)/bstep;//coef0步数
	int l=(cend-c)/cstep;//nu步数
	double zy;//预测效率
	int m=0;//递归参数
//临时变量
	double coef0temp(0);
	double nutemp(0);
	for( double jj=c;jj<cend;jj+=cstep)
	{
		param.nu=jj;
	    for(double ii=b ;ii<bend;ii+=bstep)
	    {
		    param.coef0=ii;
			info.Empty();
			zy = do_cross_validationy();
			if( zy > (limit+limitstep) )
			{
				m++;
				limit=zy;
				coef0temp=param.coef0;
				nutemp=param.nu;
				infoy=info;
				info.Format("coef0是 %f, 参数v（nu）是 %f\n",coef0temp,nutemp);
				infoy+=info;

 			}
	    }
	}
	if(m)
	{