bkneuro.h

類神經網路的BK演算法

#if !defined(BKNEURO_H)
#include "array.h"
#include "randgen.h"
#include <math.h>
#include <time.h>
#include <iomanip.h>
#include <fstream.h>

class BKNeuroNet
{
protected:
//	Nombres de neurones pour chaque couche
	int nb_dim;
	int nb_out;
	int nb_hid;
// Poids de connextion au seins du R.N.
	f2D poid_xh, poid_hy;
	f2D dPoid_xh, dPoid_hy;
	f1D theta_h, theta_y;
	f1D dTheta_h, dTheta_y;
	f1D x, hid, y;
	f1D desire_y;
	f1D delta_h, delta_y;
    f1D max,min,scaling;
//	Parametres d'apprentissage
	double eta,eta_rate,eta_min;
	double alpha,alpha_rate,alpha_min;
//	Initialisation des poids de connextion
	void MemoryAlloc();
	void MemoryFree();
// Autres
	float sum;
// Batch data
	f2D input,output;
	int nb_record;
   bool vide;
public:
//	vecteurs de normalisation
	BKNeuroNet(){nb_dim=0;nb_hid=0;nb_out=0;vide=true;}
	void Initialize(int n1, int n2, int n3);
  	void Initialize(int nn_hidden);
	void Initialize(char* Filename);
	void Initialize(ifstream& in);
   void Finish(){MemoryFree();}
	~BKNeuroNet();
	void SetRandomPoid();
	void SetRandomPoid(unsigned long seed);
	void SetNbOfHidden(int nn){nb_hid=nn;}
	void SetupPara(double& e1,double& e2,double& e3,double& r1,double& r2,double& r3);
	void GetPara(double& e1,double& e2,double& e3,double& r1,double& r2,double& r3);
	void AutoSetPara(int nb_example, int nb_cycle);
	void AutoSetPara(int nb){AutoSetPara(nb_record, nb);}
	void ModifyPara();
	void LoadTrainingData(ifstream& in);
	void LoadTrainingData(int n1,int n2,int n3, f2D& XV, f2D& YV);
	void ApplyToTestingData(ifstream& in, ofstream& out);
	void ApplyToTestingData(ifstream& in, f2D& observed, f2D& infered);
	void ApplyToData(f2D& Input, f2D& Output);
	void Save(char *FileName);
	void Save(ofstream& out);
	void SetData(f1D &xi, f1D &s){x=xi;desire_y=s;}
	void SetData(f1D &xi){x=xi;}
	float Learning();
	float Learning(f1D &xi, f1D &s);
	float LearningFromBatchData();
	f1D& Using();
	f1D& Using(f1D &xi);
	int Get_nb_dim(){return nb_dim;}
	int Get_nb_out(){return nb_out;}
	f2D TransDependantTable();
};

void BKNeuroNet::MemoryAlloc()
{
	poid_xh.Initialize(nb_dim,nb_hid);
	poid_hy.Initialize(nb_hid,nb_out);
	dPoid_xh.Initialize(nb_dim,nb_hid);
	dPoid_hy.Initialize(nb_hid,nb_out);
	theta_h.Initialize(nb_hid);
	theta_y.Initialize(nb_out);
	dTheta_h.Initialize(nb_hid);
	dTheta_y.Initialize(nb_out);
	x.Initialize(nb_dim);
	hid.Initialize(nb_hid);
	y.Initialize(nb_out);
	desire_y.Initialize(nb_out);
	delta_h.Initialize(nb_hid);
	delta_y.Initialize(nb_out);
	max.Initialize(nb_dim+nb_out);
	min.Initialize(nb_dim+nb_out);
	scaling.Initialize(nb_dim+nb_out);
   vide=false;
}

void BKNeuroNet::MemoryFree()
{
	poid_xh.Finish();
	poid_hy.Finish();
	dPoid_xh.Finish();
	dPoid_hy.Finish();
	theta_h.Finish();
	theta_y.Finish();
	dTheta_h.Finish();
	dTheta_y.Finish();
	x.Finish();
	hid.Finish();
	y.Finish();
	desire_y.Finish();
	delta_h.Finish();
	delta_y.Finish();
	max.Finish();
	min.Finish();
	scaling.Finish();
   vide=true;
}

void BKNeuroNet::Initialize(int n1, int n2, int n3)
{
	nb_dim=n1;
	nb_hid=n2;
	nb_out=n3;
	MemoryAlloc();
	SetRandomPoid();
	eta=5.0;eta_rate=0.9999;eta_min=0.00001;
	alpha=1.0;alpha_rate=0.9998;alpha_min=0.00001;
}

void BKNeuroNet::Initialize(char* FileName)
{
	ifstream f2(FileName);
	Initialize(f2);
}

void BKNeuroNet::Initialize(ifstream& f2)
{
	int h,i,j;
	f2>>ws>>setw(10)>>nb_dim>>setw(10)>>nb_hid>>setw(10)>>nb_out;
	MemoryAlloc();
	for(h=0;h<nb_hid;h++)for(i=0;i<nb_dim;i++)
		f2>>ws>>setw(12)>>setprecision(6)>>poid_xh.m[i][h];
	for(j=0;j<nb_out;j++)for(h=0;h<nb_hid;h++)
		f2>>ws>>setw(12)>>setprecision(6)>>poid_hy.m[h][j];
	for(h=0;h<nb_hid;h++)
		f2>>ws>>setw(12)>>setprecision(6)>>theta_h.m[h];
	for(j=0;j<nb_out;j++)
		f2>>ws>>setw(12)>>setprecision(6)>>theta_y.m[j];
	for(j=0;j<nb_dim+nb_out;j++)
		f2>>ws>>setw(12)>>setprecision(6)>>max.m[j];
	for(j=0;j<nb_dim+nb_out;j++)
		f2>>ws>>setw(12)>>setprecision(6)>>min.m[j];
	for(j=0;j<nb_dim+nb_out;j++)
		f2>>ws>>setw(12)>>setprecision(6)>>scaling.m[j];
	eta=5.0;eta_rate=0.99999;eta_min=0.0001;
	alpha=1.0;alpha_rate=0.99998;alpha_min=0.0001;
}

//	Destructeur
BKNeuroNet::~BKNeuroNet()
{
}

//	fonction membre BKNeuroNet::SetPoid
void BKNeuroNet::SetRandomPoid()
{
	int h,i,j;
	unsigned long seed = (unsigned long)time(0);
	fRan32 *rg=new fRan32(seed);
	for(h=0;h<nb_hid;h++)for(i=0;i<nb_dim;i++)poid_xh.m[i][h]=rg->Next(0.0,1.0);
	for(j=0;j<nb_out;j++)for(h=0;h<nb_hid;h++)poid_hy.m[h][j]=rg->Next(0.0,1.0);
	for(h=0;h<nb_hid;h++)theta_h.m[h]=rg->Next(0.0,1.0);
	for(j=0;j<nb_out;j++)theta_y.m[j]=rg->Next(0.0,1.0);
	delete rg;
}

//	fonction membre BKNeuroNet::SetPoid
void BKNeuroNet::SetRandomPoid(unsigned long seed)
{
	int h,i,j;
	fRan32 *rg=new fRan32(seed);
	for(h=0;h<nb_hid;h++)for(i=0;i<nb_dim;i++)poid_xh.m[i][h]=rg->Next(0.0,1.0);
	for(j=0;j<nb_out;j++)for(h=0;h<nb_hid;h++)poid_hy.m[h][j]=rg->Next(0.0,1.0);
	for(h=0;h<nb_hid;h++)theta_h.m[h]=rg->Next(0.0,1.0);
	for(j=0;j<nb_out;j++)theta_y.m[j]=rg->Next(0.0,1.0);
	delete rg;
}
//	Fonction membre BKNeuroNet::Save
void BKNeuroNet::Save(char* FileName)
{
	ofstream out(FileName);
	Save(out);
}

void BKNeuroNet::Save(ofstream& out)
{
	int h,i,j;
	out<<setw(10)<<nb_dim<<setw(10)<<nb_hid<<setw(10)<<nb_out<<endl;
	for(h=0;h<nb_hid;h++){for(i=0;i<nb_dim;i++){
		out<<setw(12)<<setprecision(6)<<poid_xh.m[i][h];}out<<endl;}
	for(j=0;j<nb_out;j++){for(h=0;h<nb_hid;h++){
		out<<setw(12)<<setprecision(6)<<poid_hy.m[h][j];}out<<endl;}
	for(h=0;h<nb_hid;h++){
		out<<setw(12)<<setprecision(6)<<theta_h.m[h];}out<<endl;
	for(j=0;j<nb_out;j++){
		out<<setw(12)<<setprecision(6)<<theta_y.m[j];}out<<endl;
	for(j=0;j<nb_dim+nb_out;j++){
		out<<setw(12)<<setprecision(6)<<max.m[j];}out<<endl;
  	for(j=0;j<nb_dim+nb_out;j++){
		out<<setw(12)<<setprecision(6)<<min.m[j];}out<<endl;
	for(j=0;j<nb_dim+nb_out;j++){
		out<<setw(12)<<setprecision(6)<<scaling.m[j];}out<<endl;
}

//	Fonction membre BKNeuroNet::SetupPara
void BKNeuroNet::SetupPara(double& e1,double& e2,double& e3,double& r1,double& r2,double& r3)
{
	eta=e1;
	eta_rate=e2;
	eta_min=e3;
	alpha=r1;
	alpha_rate=r2;
	alpha_min=r3;
}

//	Fonction membre BKNeuroNet::GetPara
void BKNeuroNet::GetPara(double& e1,double& e2,double& e3,double& r1,double& r2,double& r3)
{
	e1=eta;
	e2=eta_rate;
	e3=eta_min;
	r1=alpha;
	r2=alpha_rate;
	r3=alpha_min;
}

void BKNeuroNet::AutoSetPara(int nb_example, int nb_cycle)
{
	eta=8.0;eta_rate=0.99999;eta_min=0.001;
	alpha=1.5;alpha_rate=0.99997;alpha_min=0.001;
	float n=1.0/float(nb_cycle);
	eta_rate=pow(eta_min/eta,n);
	alpha_rate=pow(alpha_min/alpha,n);
}

//	Fonction membre BKNeuroNet::Learning
float BKNeuroNet::Learning()
{
	int h,i,j;
	for(h=0; h<nb_hid;h++)
	{
		sum=0.0;
		for(i=0;i<nb_dim;i++)sum=sum+x.m[i]*poid_xh.m[i][h];
		hid.m[h]=1.0/(1.0+exp(-(sum-theta_h.m[h])));
	}
	for(j=0;j<nb_out;j++)
	{
		sum=0.0;
		for(h=0;h<nb_hid;h++)sum=sum+hid.m[h]*poid_hy.m[h][j];
		y.m[j]=1.0/(1.0+exp(-(sum-theta_y.m[j])));
	}
	for(j=0;j<nb_out;j++)delta_y.m[j]=y.m[j]*(1.0-y.m[j])*(desire_y.m[j]-y.m[j]);
	for(h=0;h<nb_hid;h++)
	{
		sum=0.0;
		for(j=0;j<nb_out;j++)sum=sum+poid_hy.m[h][j]*delta_y.m[j];
		delta_h.m[h]=hid.m[h]*(1.0-hid.m[h])*sum;
	}
	for(j=0;j<nb_out;j++)for(h=0;h<nb_hid;h++)dPoid_hy.m[h][j]=0.0;
	for(h=0;h<nb_hid;h++)for(i=0;i<nb_dim;i++)dPoid_xh.m[i][h]=0.0;
	for(j=0;j<nb_out;j++)dTheta_y.m[j]=0.0;
	for(h=0;h<nb_hid;h++)dTheta_h.m[h]=0.0;
	for(j=0;j<nb_out;j++){for(h=0;h<nb_hid;h++)
		{dPoid_hy.m[h][j]=eta*delta_y.m[j]*hid.m[h]+alpha*dPoid_hy.m[h][j];}}
	for(j=0;j<nb_out;j++)dTheta_y.m[j]=-eta*delta_y.m[j]+alpha*dTheta_y.m[j];
	for(h=0;h<nb_hid;h++){for(i=0;i<nb_dim;i++)
		{dPoid_xh.m[i][h]=eta*delta_h.m[h]*x.m[i]+alpha*dPoid_xh.m[i][h];}}
	for(h=0;h<nb_hid;h++)dTheta_h.m[h]=-eta*delta_h.m[h]+alpha*dTheta_h.m[h];
	for(j=0;j<nb_out;j++){for(h=0;h<nb_hid;h++)
		{poid_hy.m[h][j]=poid_hy.m[h][j]+dPoid_hy.m[h][j];}}
	for(j=0;j<nb_out;j++)theta_y.m[j]=theta_y.m[j]+dTheta_y.m[j];
	for(h=0;h<nb_hid;h++){for(i=0;i<nb_dim;i++)
		{poid_xh.m[i][h]=poid_xh.m[i][h]+dPoid_xh.m[i][h];}}
	for(h=0;h<nb_hid;h++)theta_h.m[h]=theta_h.m[h]+dTheta_h.m[h];
	alpha=alpha*alpha_rate;
	if(alpha<alpha_min)alpha=alpha_min;
	eta=eta*eta_rate;
	if(eta<eta_min)eta=eta_min;
	float err=0.0;
    for(j=0;j<nb_out;j++)err+=fabs(desire_y.m[j]-y.m[j]);
	return err/nb_out;
}

//	Fonction membre BKNeuroNet::Learning
float BKNeuroNet::Learning(f1D &xi, f1D &s)
{
	int h,i,j;
	for(i=0;i<nb_dim;i++)x.m[i]=(xi.m[i]-min.m[i])*scaling.m[i];
	for(j=0;j<nb_out;j++)desire_y.m[j]=(s.m[j]-min.m[j+nb_dim])*scaling.m[j+nb_dim];
	for(h=0; h<nb_hid;h++)
	{
		sum=0.0;
		for(i=0;i<nb_dim;i++)sum=sum+x.m[i]*poid_xh.m[i][h];