改写的C＋＋BP类未测试

devil1980 发表于 2006-3-28 17:28

// BpNetWork.cpp: implementation of the CBpNetWork class. // //////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// // Construction/Destruction ////////////////////////////////////////////////////////////////////// #include "stdafx.h" #include "BpNetWork.h" #include "iostream.h"CBpNetWork::CBpNetWork() {  m_npBeta = new double;//Òþ²ãÖÁÊäÈë²ãÑ§Ï°Ð§ÂÊ  *m_npBeta = 0.2;  m_npAlpha = new double;//Êä³ö²ãÖÁÒþ²ãµÄÑ§Ï°Ð§ÂÊ  *m_npAlpha = 0.2;  m_npTotalErrofM = NULL;//µÚm¸öÑù±¾µÄ×ÜÎó²î  m_npOutputLayerThreshold = NULL;//Êä³ö²ãµÄãÐÖµ  m_npHideLayerThreshold = NULL;//Òþ²ãµÄãÐÖµ  m_npOutputLayerOutputValue = NULL;//Êä³ö²ãµÄÊä³ö  m_npHideOutputValue = NULL;//Òþ²ãµÄÊä³ö  m_npOutputLayerInputValue = NULL;//Êä³ö²ãµÄÊäÈë  m_npHideInputValue;//Òþ²ãµÄÊäÈë  m_npHideToOutputWeightMatrix = NULL;//Òþ²ãÖÁÊä³ö²ãÈ¨Öµ  m_npInputToHideWeightMatrix = NULL;//ÊäÈë²ãÖÁÒþ²ãÈ¨Öµ  m_fpSingleTrainingSample = NULL;//µ¥¸öÑù±¾½ÌÊ¦Êý¾Ý  m_fpSingleInputSample = NULL;//µ¥¸öÑù±¾ÊäÈëÊý¾Ý  m_fpE_Err = NULL;  m_fpD_Err = NULL;  m_fpInput = NULL;  m_fpTeach = NULL;  m_spStudy_Data = NULL;  m_fpAbsErrValBefore =NULL;  m_nLearningSampleNumber = 0;//Ñ§Ï°Ñù±¾¸öÊý  m_nOutputNetNumber = 0;//ÊäÈë²ãÉñ¾ÔªÊýÄ¿  m_nHideNetNumber = 0;//Òþ²ãÉñ¾ÔªÊýÄ¿  m_nInputNetNumber = 0;//Êä³ö²ãÉñ¾ÔªÊýÄ¿  Init(); }CBpNetWork::~CBpNetWork() {  if(!m_npBeta) delete m_npBeta,m_npBeta = NULL;  if(!m_npAlpha) delete m_npAlpha,m_npAlpha = NULL;  if(!m_npTotalErrofM) delete m_npTotalErrofM, m_npTotalErrofM = NULL;  if(!m_npOutputLayerThreshold) delete []m_npOutputLayerThreshold, m_npOutputLayerThreshold= NULL;  if(!m_npHideLayerThreshold) delete[]m_npHideLayerThreshold, m_npHideLayerThreshold =NULL;  if(!m_npOutputLayerOutputValue) delete[]m_npOutputLayerOutputValue,m_npOutputLayerOutputValue = NULL;  if(!m_npHideOutputValue) delete []m_npHideOutputValue,m_npHideOutputValue=NULL;  if(!m_npOutputLayerInputValue) delete []m_npOutputLayerInputValue,m_npOutputLayerInputValue= NULL;  if(!m_npHideInputValue) delete []m_npHideInputValue, m_npHideInputValue = NULL;  if(!m_npHideToOutputWeightMatrix) delete []m_npHideToOutputWeightMatrix, m_npHideToOutputWeightMatrix = NULL;  if(!m_npInputToHideWeightMatrix) delete []m_npInputToHideWeightMatrix,m_npInputToHideWeightMatrix = NULL;  if(!m_fpSingleTrainingSample) delete []m_fpSingleTrainingSample,m_fpSingleTrainingSample = NULL;  if(!m_fpSingleInputSample) delete []m_fpSingleInputSample, m_fpSingleInputSample =NULL;  if(!m_fpE_Err) delete []m_fpE_Err, m_fpE_Err =NULL;  if(!m_fpD_Err) delete []m_fpD_Err, m_fpD_Err = NULL;  if(!m_fpAbsErrValBefore) delete []m_fpAbsErrValBefore, m_fpAbsErrValBefore= NULL;  Study_Data_T *p = m_spStudy_Data; for(int i = 0;i< m_nLearningSampleNumber;i++)   for(int j = 0; j< m_nInputNetNumber; j++)  {    m_fpInput = m_spStudy_Data->input;  m_fpTeach = m_spStudy_Data->teach;  if(!m_fpInput) delete []m_fpInput, m_fpInput = NULL;  if(!m_fpTeach) delete []m_fpTeach, m_fpTeach = NULL;  m_spStudy_Data++;  }  delete []p; } CBpNetWork::CBpNetWork(int LearningSplNum, int OutLayNumber, int HideLayNumber, int inLayNum) {  m_nLearningSampleNumber = LearningSplNum;  m_nOutputNetNumber = OutLayNumber;  m_nHideNetNumber = HideLayNumber;  m_nInputNetNumber = inLayNum;  Init(); } //initial weight and threshold bool CBpNetWork::Init() {  if(!m_nLearningSampleNumber||!m_nOutputNetNumber||!m_nHideNetNumber||!m_nInputNetNumber)   return false;  float sgn;  float rnd;  int i,j;  m_npInputToHideWeightMatrix = new double ;  m_npHideLayerThreshold = new double;  m_npHideToOutputWeightMatrix = new double;  m_fpSingleInputSample = new double;  m_fpSingleTrainingSample = new double ;  m_npHideInputValue = new double;  m_npOutputLayerOutputValue = new double;  m_npOutputLayerInputValue = new double;  m_npHideLayerThreshold = new double;  m_npOutputLayerThreshold = new double ;  m_npTotalErrofM = new double;  m_fpE_Err = new double;  m_fpD_Err = new double;  m_fpAbsErrValBefore = new double; m_spStudy_Data = (Study_Data_T *)new double;  for( i = 0;i< m_nLearningSampleNumber;i++)   for( j = 0; j< m_nInputNetNumber; j++)   {   m_fpInput = new double;   m_fpTeach = new double;   m_spStudy_Data->input = m_fpInput;   m_spStudy_Data->teach =m_fpTeach;   m_spStudy_Data++;   } //Òþ²ãÈ¨¡¢ãÐÖµ³õÊ¼»¯//  for(j=0;j<m_nHideNetNumber;j++)   for(i=0;i<m_nInputNetNumber;i++)        {          //sgn=pow((-1),random(100));          sgn=(float)rand()/RAND_MAX ;          rnd=sgn*(rand()%10);          m_npInputToHideWeightMatrix= rnd/100;//Òþ²ãÈ¨Öµ³õÊ¼»¯¡£          } //randomize();  for(j=0;j<m_nHideNetNumber;j++)     {      //sgn=pow((-1),random(1000));      sgn=(float)rand()/RAND_MAX ;  rnd=sgn*(rand()%10);  m_npHideLayerThreshold= rnd/1000;//ÖÐ¼ä²ããÐÖµ³õÊ¼»¯      }       //Êä³ö²ãÈ¨¡¢ãÐÖµ³õÊ¼»¯//       //randomize();       for (int k=0;k<m_nOutputNetNumber;k++)        for (int j=0;j<m_nHideNetNumber;j++)        {        //sgn=pow((-1),random(1000));          sgn=(float)rand()/RAND_MAX ;          rnd=sgn*(rand()%10);          m_npHideToOutputWeightMatrix=rnd/100;//µÚm¸öÑù±¾Êä³ö²ãÈ¨Öµ³õÊ¼»¯        }       //randomize();     for(k =0; k<m_nOutputNetNumber;k++)         {         //sgn=pow((-1),random(10));         sgn=(float)rand()/RAND_MAX ;         rnd=sgn*(rand()%10);         m_npOutputLayerThreshold=rnd/10;//Êä³ö²ããÐÖµ³õÊ¼»¯         }         return true; } //input No. m Tarinning data void CBpNetWork::InputOfNoMTrainingData(int m,Study_Data_T * tnspl) {  for (int k=0;k<m;k++)        m_fpSingleTrainingSample=tnspl.teach;//Study_Data.m_fpTeach; } ////µÚm¸öÑ§Ï°Ñù±¾ÊäÈë×Ó³ÌÐò/// void CBpNetWork::InputOfNoMSampleData(int m, Study_Data_T * lnspl) {  for (int i=0;i<m_nInputNetNumber;i++)        m_fpSingleInputSample= lnspl.input;//Study_Data.input;    } //Òþ²ã¸÷µ¥ÔªÊäÈë¡¢Êä³öÖµ×Ó³ÌÐò/// void CBpNetWork::HideLayInOut() {       double sigma;       int i,j;       for (j=0;j<m_nHideNetNumber;j++)    {        sigma=0.0;       for (i=0;i<m_nInputNetNumber;i++)        sigma+=m_npInputToHideWeightMatrix*m_fpSingleInputSample;//ÇóÒþ²ãÄÚ»ý            m_npHideInputValue=sigma - m_npHideLayerThreshold;//ÇóÒþ²ã¾»ÊäÈë       m_npHideOutputValue=1.0/(1.0+exp(-m_npHideInputValue));//ÇóÒþ²ãÊä³ö       }       } //Êä³ö²ã¸÷µ¥ÔªÊäÈë¡¢Êä³öÖµ×Ó³ÌÐò/// void CBpNetWork::OutLayInOut() {       double sigma;       for (int k=0;k<m_nOutputNetNumber;k++)    {        sigma=0.0;       for (int j=0;j<m_nHideNetNumber;j++)        sigma+=m_npHideToOutputWeightMatrix*m_npHideOutputValue;//ÇóÊä³ö²ãÄÚ»ý             m_npOutputLayerInputValue=sigma-m_npOutputLayerThreshold; //ÇóÊä³ö²ã¾»ÊäÈë       m_npOutputLayerOutputValue=1.0/(1.0+exp(-m_npOutputLayerInputValue));//ÇóÊä³ö²ãÊä³ö       }    } //Êä³ö²ãÖÁÒþ²ãµÄÒ»°ã»¯Îó²î×Ó³ÌÐò//// void CBpNetWork::NormalizeErrOutLayToHideLay(int m) {  double *abs_err = new double;  //double abs_err;//Ã¿¸öÑù±¾µÄ¾ø¶ÔÎó²î¶¼ÊÇ´Ó0¿ªÊ¼µÄ  double sqr_err=0;//Ã¿¸öÑù±¾µÄÆ½·½Îó²î¼ÆËã¶¼ÊÇ´Ó0¿ªÊ¼µÄ  //for (int output=0;output<m_nOutputNetNumber;output++)  //output???  for (int k=0;k<m_nOutputNetNumber;k++)  {  abs_err=m_fpSingleTrainingSample-m_npOutputLayerOutputValue;  if(abs_err == 0)   abs_err = 0.00001;  //¸ù¾ÝÎó²î¸Ä±äÑ§Ï°Ð§ÂÊ  if(abs_err>m_fpAbsErrValBefore)  {   *m_npAlpha += (abs_err/(abs_err+m_fpAbsErrValBefore))**m_npAlpha;   *m_npBeta += (abs_err/(abs_err+m_fpAbsErrValBefore))**m_npBeta;  }  else  {   *m_npAlpha -= (abs_err/(abs_err+m_fpAbsErrValBefore))**m_npAlpha;   *m_npBeta -= (abs_err/(abs_err+m_fpAbsErrValBefore))**m_npBeta;  }  m_fpAbsErrValBefore = abs_err;  //ÇóµÚm¸öÑù±¾ÏÂµÄµÚk¸öÉñ¾ÔªµÄ¾ø¶ÔÎó²î  sqr_err+=(abs_err)*(abs_err);//ÇóµÚm¸öÑù±¾ÏÂÊä³ö²ãµÄÆ½·½Îó²î  m_fpD_Err=abs_err*m_npOutputLayerOutputValue*(1.0-m_npOutputLayerOutputValue);//d_errÊä³ö²ã¸÷Éñ¾ÔªµÄÒ»°ã»¯Îó²î  }  m_npTotalErrofM=sqr_err/2;//µÚm¸öÑù±¾ÏÂÊä³ö²ãµÄÆ½·½Îó²î/2=µÚm¸öÑù±¾µÄ¾ù·½Îó²î  delete []abs_err; }void CBpNetWork::NormalizeErrHideLayToOutLay() {  double sigma;         for (int j=0;j<m_nHideNetNumber;j++)   {         sigma=0.0;            for (int k=0;k<m_nOutputNetNumber;k++)             sigma=m_fpD_Err*m_npHideToOutputWeightMatrix;          m_fpE_Err=sigma*exp(m_npHideOutputValue)/((1-exp(m_npHideOutputValue))*(1-exp(m_npHideOutputValue)));//Òþ²ã¸÷Éñ¾ÔªµÄÒ»°ã»¯Îó²î   } }   //Êä³ö²ãÖÁÒþ²ãµÄÈ¨Öµµ÷Õû¡¢Êä³ö²ããÐÖµµ÷Õû¼ÆËã×Ó³ÌÐò////// void CBpNetWork::AdjustWeightAndThresholdOutLayToHideLay(int m, FILE * fp) {  for (int k=0;k<m_nOutputNetNumber;k++)  {      for (int j=0;j<m_nHideNetNumber;j++)      {     m_npHideToOutputWeightMatrix+=*m_npAlpha*m_fpD_Err*m_npHideOutputValue;//Êä³ö²ãÖÁÒþ²ãµÄÈ¨Öµµ÷Õû      }     m_npOutputLayerThreshold+=*m_npAlpha*m_fpD_Err;//Êä³ö²ãÖÁÒþ²ãµÄãÐÖµµ÷Õû  } } //m_nLearningSampleNumber¸öÑù±¾µÄÈ«¾ÖÎó²î¼ÆËã×Ó³ÌÐò//// double CBpNetWork::GlobalErrSum() {  double total_err=0;       for (int m=0;m<m_nLearningSampleNumber;m++)         total_err+=m_npTotalErrofM;//Ã¿¸öÑù±¾µÄ¾ù·½Îó²î¼ÓÆðÀ´¾Í³ÉÁËÈ«¾ÖÎó²î         return total_err; }  //Òþ²ãÖÁÊäÈë²ãµÄÈ¨Öµµ÷Õû¡¢Òþ²ããÐÖµµ÷Õû¼ÆËã×Ó³ÌÐò///// void CBpNetWork::AdjustWeightAndThresholdHideLayToOutLay(int m,FILE * fp) { for (int j=0;j<m_nHideNetNumber;j++)  {         for (int i=0;i<m_nInputNetNumber;i++)   {        //cout<<"µÚ"< fprintf(fp,"µÚ%d¸öÑù±¾Ê±µÄÊä³ö²ãÖÁÒþ²ãµÄÈ¨ÖµÎª£º%f\n",m,m_npHideToOutputWeightMatrix);       m_npInputToHideWeightMatrix+=*m_npBeta*m_fpE_Err*m_fpSingleInputSample;//Òþ²ãÖÁÊäÈë²ãµÄÈ¨Öµµ÷Õû   }       m_npHideLayerThreshold+=*m_npBeta*m_fpE_Err;  } }void CBpNetWork::Excute(Study_Data_T * spl, double err, int count) {  double temp_err = 0;  int local_count =0;  if(count == 0)   count = 1000;  if(!spl)   return;  do  {  Tarning(spl);  temp_err = GlobalErrSum();  local_count++;  }while(temp_err>err&&local_count<count); }void CBpNetWork::Tarning(Study_Data_T * spl) {   int i = 0;   Study_Data_T * tlnspl =spl;   Study_Data_T * ttnspl =spl;  for(;tlnspl != NULL&& ttnspl != NULL && i<m_nLearningSampleNumber*m_nInputNetNumber;tlnspl++,ttnspl++,i++)  {   InputOfNoMSampleData(i, tlnspl);   InputOfNoMTrainingData(i, ttnspl);   HideLayInOut();   OutLayInOut();   NormalizeErrOutLayToHideLay(i);   NormalizeErrHideLayToOutLay();   AdjustWeightAndThresholdOutLayToHideLay(i,NULL);   AdjustWeightAndThresholdHideLayToOutLay(i,NULL);  } }  // head file// BpNetWork.h: interface for the CBpNetWork class. // //////////////////////////////////////////////////////////////////////#if !defined(AFX_BPNETWORK_H__2FDDE314_11F6_4D49_A046_C42B29C3754D__INCLUDED_) #define AFX_BPNETWORK_H__2FDDE314_11F6_4D49_A046_C42B29C3754D__INCLUDED_#if _MSC_VER > 1000 #pragma once #endif // _MSC_VER > 1000  #include "math.h" typedef struct Study_Data_S {    double *input;    double *teach;   } Study_Data_T; class CBpNetWork  { public://Êä³ö²ãÖÁÒþ²ãµÄÒ»°ã»¯Îó²î×Ó³ÌÐò//// void Tarning(Study_Data_T * spl);  void Excute(Study_Data_T * spl, double err, int count);  void AdjustWeightAndThresholdHideLayToOutLay(int m, FILE * fp);  double GlobalErrSum();  void AdjustWeightAndThresholdOutLayToHideLay(int m, FILE * fp);  void NormalizeErrHideLayToOutLay();  void NormalizeErrOutLayToHideLay(int m);  void OutLayInOut();  void HideLayInOut();  void InputOfNoMSampleData(int m,Study_Data_T * lnspl);  void InputOfNoMTrainingData(int m,Study_Data_T * lnspl);  bool Init();  CBpNetWork(int LearningSplNum, int OutLayNumber,int HideLayNumber,int inLayNum); double* m_npBeta;//Òþ²ãÖÁÊäÈë²ãÑ§Ï°Ð§ÂÊ  double* m_npAlpha;//Êä³ö²ãÖÁÒþ²ãµÄÑ§Ï°Ð§ÂÊ  double* m_npTotalErrofM;//µÚm¸öÑù±¾µÄ×ÜÎó²î  double* m_npOutputLayerThreshold;//Êä³ö²ãµÄãÐÖµ  double* m_npHideLayerThreshold;//Òþ²ãµÄãÐÖµ  double* m_npOutputLayerOutputValue;//Êä³ö²ãµÄÊä³ö  double* m_npHideOutputValue;//Òþ²ãµÄÊä³ö  double* m_npOutputLayerInputValue;//Êä³ö²ãµÄÊäÈë  double* m_npHideInputValue;//Òþ²ãµÄÊäÈë  double* m_npHideToOutputWeightMatrix;//Òþ²ãÖÁÊä³ö²ãÈ¨Öµ  double* m_npInputToHideWeightMatrix;//ÊäÈë²ãÖÁÒþ²ãÈ¨Öµ  double* m_fpSingleTrainingSample;//µ¥¸öÑù±¾½ÌÊ¦Êý¾Ý  double* m_fpSingleInputSample;//µ¥¸öÑù±¾ÊäÈëÊý¾Ý  double* m_fpInput;  double* m_fpTeach;  double* m_fpD_Err;  double* m_fpE_Err;  double * m_fpAbsErrValBefore;  Study_Data_T *m_spStudy_Data; CBpNetWork();  virtual ~CBpNetWork();private:  int m_nLearningSampleNumber;//Ñ§Ï°Ñù±¾¸öÊý  int m_nOutputNetNumber;//ÊäÈë²ãÉñ¾ÔªÊýÄ¿  int m_nHideNetNumber;//Òþ²ãÉñ¾ÔªÊýÄ¿  int m_nInputNetNumber;//Êä³ö²ãÉñ¾ÔªÊýÄ¿ };#endif // !defined(AFX_BPNETWORK_H__2FDDE314_11F6_4D49_A046_C42B29C3754D__INCLUDED_)

devil1980 发表于 2006-3-30 17:01

测试通过的BP类，可以收敛Total误差

// BpNetWork.cpp: implementation of the CBpNetWork class. // ////////////////////////////////////////////////////////////////////// //Author mail: <a href="mailto:os.win2k@163.com">os.win2k@163.com</a> ////////////////////////////////////////////////////////////////////// // Construction/Destruction ////////////////////////////////////////////////////////////////////// #include "stdafx.h" #include "BpNetWork.h" #include "iostream.h"CBpNetWork::CBpNetWork() {  m_npBeta = new double;//Òþ²ãÖÁÊäÈë²ãÑ§Ï°Ð§ÂÊ  *m_npBeta = 0.002;  m_npAlpha = new double;//Êä³ö²ãÖÁÒþ²ãµÄÑ§Ï°Ð§ÂÊ  *m_npAlpha = 0.002;  m_npTotalErrofM = NULL;//µÚm¸öÑù±¾µÄ×ÜÎó²î  m_npOutputLayerThreshold = NULL;//Êä³ö²ãµÄãÐÖµ  m_npHideLayerThreshold = NULL;//Òþ²ãµÄãÐÖµ  m_npOutputLayerOutputValue = NULL;//Êä³ö²ãµÄÊä³ö  m_npHideOutputValue = NULL;//Òþ²ãµÄÊä³ö  m_npOutputLayerInputValue = NULL;//Êä³ö²ãµÄÊäÈë  m_npHideInputValue;//Òþ²ãµÄÊäÈë  m_npHideToOutputWeightMatrix = NULL;//Òþ²ãÖÁÊä³ö²ãÈ¨Öµ  m_npInputToHideWeightMatrix = NULL;//ÊäÈë²ãÖÁÒþ²ãÈ¨Öµ  m_fpSingleTrainingSample = NULL;//µ¥¸öÑù±¾½ÌÊ¦Êý¾Ý  m_fpSingleInputSample = NULL;//µ¥¸öÑù±¾ÊäÈëÊý¾Ý  m_fpE_Err = NULL;  m_fpD_Err = NULL;  m_fpInput = NULL;  m_fpTeach = NULL;  m_spStudy_Data = NULL;  m_fpAbsErrValBefore =NULL;  m_nLearningSampleNumber = 10;//Ñ§Ï°Ñù±¾¸öÊý  m_nOutputNetNumber = 1;//ÊäÈë²ãÉñ¾ÔªÊýÄ¿  m_nHideNetNumber = 10;//Òþ²ãÉñ¾ÔªÊýÄ¿  m_nInputNetNumber = 1;//Êä³ö²ãÉñ¾ÔªÊýÄ¿  Init(); }CBpNetWork::~CBpNetWork() {  if(m_npBeta!=NULL) delete m_npBeta,m_npBeta = NULL;  if(NULL != m_npAlpha) delete m_npAlpha,m_npAlpha = NULL;  if(NULL != m_npTotalErrofM) delete m_npTotalErrofM, m_npTotalErrofM = NULL;  if(NULL != m_npOutputLayerThreshold) delete []m_npOutputLayerThreshold, m_npOutputLayerThreshold= NULL;  if(NULL != m_npHideLayerThreshold) delete[]m_npHideLayerThreshold, m_npHideLayerThreshold =NULL;  if(NULL != m_npOutputLayerOutputValue) delete[]m_npOutputLayerOutputValue,m_npOutputLayerOutputValue = NULL;  if(NULL != m_npHideOutputValue) delete []m_npHideOutputValue,m_npHideOutputValue=NULL;  if(NULL != m_npOutputLayerInputValue) delete []m_npOutputLayerInputValue,m_npOutputLayerInputValue= NULL;  if(NULL != m_npHideInputValue) delete []m_npHideInputValue, m_npHideInputValue = NULL;  if(NULL != m_npHideToOutputWeightMatrix) delete []m_npHideToOutputWeightMatrix, m_npHideToOutputWeightMatrix = NULL;  if(NULL != m_npInputToHideWeightMatrix) delete []m_npInputToHideWeightMatrix,m_npInputToHideWeightMatrix = NULL;  if(NULL != m_fpSingleTrainingSample) delete []m_fpSingleTrainingSample,m_fpSingleTrainingSample = NULL;  if(NULL != m_fpSingleInputSample) delete []m_fpSingleInputSample, m_fpSingleInputSample =NULL;  if(NULL != m_fpE_Err) delete []m_fpE_Err, m_fpE_Err =NULL;  if(NULL != m_fpD_Err) delete []m_fpD_Err, m_fpD_Err = NULL;  if(NULL != m_fpAbsErrValBefore) delete []m_fpAbsErrValBefore, m_fpAbsErrValBefore= NULL;  Study_Data_T *p = m_spStudy_Data;   for(int i = 0;i< m_nLearningSampleNumber;i++)   for(int j = 0; j< m_nInputNetNumber; j++)  {    m_fpInput = m_spStudy_Data->input;  m_fpTeach = m_spStudy_Data->teach;  if(NULL != m_fpInput) delete []m_fpInput, m_fpInput = NULL;  if(NULL != m_fpTeach) delete []m_fpTeach, m_fpTeach = NULL;  m_spStudy_Data++;  }  delete []p; } CBpNetWork::CBpNetWork(int LearningSplNum, int OutLayNumber, int HideLayNumber, int inLayNum) {  m_nLearningSampleNumber = LearningSplNum;  m_nOutputNetNumber = OutLayNumber;  m_nHideNetNumber = HideLayNumber;  m_nInputNetNumber = inLayNum;  Init(); } //initial weight and threshold bool CBpNetWork::Init() {  if(!m_nLearningSampleNumber||!m_nOutputNetNumber||!m_nHideNetNumber||!m_nInputNetNumber)   return false;  float sgn;  float rnd;  int i,j;  m_npInputToHideWeightMatrix = new double ;  Set(m_npInputToHideWeightMatrix,m_nHideNetNumber*m_nInputNetNumber); m_npHideLayerThreshold = new double;  Set(m_npHideLayerThreshold,m_nHideNetNumber); m_npHideToOutputWeightMatrix = new double;  Set(m_npHideToOutputWeightMatrix,m_nOutputNetNumber*m_nHideNetNumber); m_fpSingleInputSample = new double;  Set(m_fpSingleInputSample,m_nLearningSampleNumber); m_fpSingleTrainingSample = new double ;  Set(m_fpSingleTrainingSample,m_nOutputNetNumber);  m_npHideInputValue = new double;  Set(m_npHideInputValue,m_nHideNetNumber); m_npOutputLayerOutputValue = new double;  Set(m_npOutputLayerOutputValue,m_nOutputNetNumber); m_npOutputLayerInputValue = new double;  Set(m_npOutputLayerInputValue,m_nOutputNetNumber); m_npHideLayerThreshold = new double;  Set(m_npHideLayerThreshold,m_nHideNetNumber); m_npOutputLayerThreshold = new double ;  Set(m_npOutputLayerThreshold,m_nOutputNetNumber); m_npTotalErrofM = new double;  Set(m_npTotalErrofM,m_nLearningSampleNumber); m_fpE_Err = new double;  Set(m_fpE_Err,m_nHideNetNumber); m_fpD_Err = new double;  Set(m_fpD_Err,m_nOutputNetNumber); m_fpAbsErrValBefore = new double;  Set(m_fpAbsErrValBefore,m_nOutputNetNumber); m_npHideOutputValue = new double;  Set(m_npHideOutputValue,m_nHideNetNumber); m_spStudy_Data = (Study_Data_T *)new double;  for( i = 0;i< m_nLearningSampleNumber;i++)   for( j = 0; j< m_nInputNetNumber; j++)   {   m_fpInput = new double;   m_fpTeach = new double;   m_spStudy_Data->input = m_fpInput;   m_spStudy_Data->teach =m_fpTeach;   m_spStudy_Data++;   } //Òþ²ãÈ¨¡¢ãÐÖµ³õÊ¼»¯//  for(j=0;j<m_nHideNetNumber;j++)   for(i=0;i<m_nInputNetNumber;i++)        {          //sgn=pow((-1),random(100));          sgn=(float)rand()/RAND_MAX ;          rnd=sgn*(rand()%10);          m_npInputToHideWeightMatrix= rnd/100;//Òþ²ãÈ¨Öµ³õÊ¼»¯¡£          } //randomize();  for(j=0;j<m_nHideNetNumber;j++)     {      //sgn=pow((-1),random(1000));      sgn=(float)rand()/RAND_MAX;  rnd=sgn*(rand()%10);  m_npHideLayerThreshold= 0;//ÖÐ¼ä²ããÐÖµ³õÊ¼»¯      }       //Êä³ö²ãÈ¨¡¢ãÐÖµ³õÊ¼»¯//       //randomize();       for (int k=0;k<m_nOutputNetNumber;k++)        for (int j=0;j<m_nHideNetNumber;j++)        {        //sgn=pow((-1),random(1000));          sgn=(float)rand()/RAND_MAX ;          rnd=sgn*(rand()%10);          m_npHideToOutputWeightMatrix=rnd/100;//µÚm¸öÑù±¾Êä³ö²ãÈ¨Öµ³õÊ¼»¯        }       //randomize();     for(k =0; k<m_nOutputNetNumber;k++)         {         //sgn=pow((-1),random(10));         sgn=(float)rand()/RAND_MAX ;         rnd=sgn*(rand()%10);         m_npOutputLayerThreshold=0;//Êä³ö²ããÐÖµ³õÊ¼»¯         }         return true; } //input No. m Tarinning data void CBpNetWork::InputOfNoMTrainingData(int m,Study_Data_T * tnspl) {  for (int k=0;k<m_nInputNetNumber;k++)        m_fpSingleTrainingSample=tnspl.teach;//Study_Data.m_fpTeach; } ////µÚm¸öÑ§Ï°Ñù±¾ÊäÈë×Ó³ÌÐò/// void CBpNetWork::InputOfNoMSampleData(int m, Study_Data_T * lnspl) {  for (int i=0;i<m_nInputNetNumber;i++)        m_fpSingleInputSample= lnspl.input;//Study_Data.input;    } //Òþ²ã¸÷µ¥ÔªÊäÈë¡¢Êä³öÖµ×Ó³ÌÐò/// void CBpNetWork::HideLayInOut() {       double sigma;       int i,j;       for (j=0;j<m_nHideNetNumber;j++)    {        sigma=0.0;       for (i=0;i<m_nInputNetNumber;i++)        sigma+=m_npInputToHideWeightMatrix*m_fpSingleInputSample;//ÇóÒþ²ãÄÚ»ý            m_npHideInputValue=sigma - m_npHideLayerThreshold;//ÇóÒþ²ã¾»ÊäÈë       m_npHideOutputValue=1.0/(1.0+exp(-m_npHideInputValue));//ÇóÒþ²ãÊä³ö       }       } //Êä³ö²ã¸÷µ¥ÔªÊäÈë¡¢Êä³öÖµ×Ó³ÌÐò/// void CBpNetWork::OutLayInOut() {       double sigma;       for (int k=0;k<m_nOutputNetNumber;k++)    {        sigma=0.0;       for (int j=0;j<m_nHideNetNumber;j++)       sigma+=m_npHideToOutputWeightMatrix*m_npHideOutputValue;//ÇóÊä³ö²ãÄÚ»ý             m_npOutputLayerInputValue=sigma-m_npOutputLayerThreshold; //ÇóÊä³ö²ã¾»ÊäÈë       m_npOutputLayerOutputValue=1.0/(1.0+exp(-m_npOutputLayerInputValue));//ÇóÊä³ö²ãÊä³ö       }    } //Êä³ö²ãÖÁÒþ²ãµÄÒ»°ã»¯Îó²î×Ó³ÌÐò//// void CBpNetWork::NormalizeErrOutLayToHideLay(int m) {  double *abs_err = new double;  //double abs_err;//Ã¿¸öÑù±¾µÄ¾ø¶ÔÎó²î¶¼ÊÇ´Ó0¿ªÊ¼µÄ  double sqr_err=0;//Ã¿¸öÑù±¾µÄÆ½·½Îó²î¼ÆËã¶¼ÊÇ´Ó0¿ªÊ¼µÄ  //for (int output=0;output<m_nOutputNetNumber;output++)  //output???  for (int k=0;k<m_nOutputNetNumber;k++)  {  abs_err=m_fpSingleTrainingSample-m_npOutputLayerOutputValue;  if(abs_err == 0)   abs_err = 0.0000001;  //¸ù¾ÝÎó²î¸Ä±äÑ§Ï°Ð§ÂÊ /* if(abs_err>m_fpAbsErrValBefore)  {   *m_npAlpha += (abs_err/(abs_err+m_fpAbsErrValBefore))**m_npAlpha;   *m_npBeta += (abs_err/(abs_err+m_fpAbsErrValBefore))**m_npBeta;  }  else  {   *m_npAlpha -= (abs_err/(abs_err+m_fpAbsErrValBefore))**m_npAlpha;   *m_npBeta -= (abs_err/(abs_err+m_fpAbsErrValBefore))**m_npBeta;  }*/  m_fpAbsErrValBefore = abs_err;  //ÇóµÚm¸öÑù±¾ÏÂµÄµÚk¸öÉñ¾ÔªµÄ¾ø¶ÔÎó²î  sqr_err+=(abs_err)*(abs_err);//ÇóµÚm¸öÑù±¾ÏÂÊä³ö²ãµÄÆ½·½Îó²î  m_fpD_Err=abs_err*m_npOutputLayerOutputValue*(1.0-m_npOutputLayerOutputValue);//d_errÊä³ö²ã¸÷Éñ¾ÔªµÄÒ»°ã»¯Îó²î  }  m_npTotalErrofM=sqr_err/2;//µÚm¸öÑù±¾ÏÂÊä³ö²ãµÄÆ½·½Îó²î/2=µÚm¸öÑù±¾µÄ¾ù·½Îó²î  delete []abs_err; }void CBpNetWork::NormalizeErrHideLayToOutLay() {  double sigma;         for (int j=0;j<m_nHideNetNumber;j++)   {         sigma=0.0;            for (int k=0;k<m_nOutputNetNumber;k++)             sigma=m_fpD_Err*m_npHideToOutputWeightMatrix;          m_fpE_Err=sigma*exp(m_npHideOutputValue)/((1-exp(m_npHideOutputValue))*(1-exp(m_npHideOutputValue)));//Òþ²ã¸÷Éñ¾ÔªµÄÒ»°ã»¯Îó²î   } }   //Êä³ö²ãÖÁÒþ²ãµÄÈ¨Öµµ÷Õû¡¢Êä³ö²ããÐÖµµ÷Õû¼ÆËã×Ó³ÌÐò////// void CBpNetWork::AdjustWeightAndThresholdOutLayToHideLay(int m, FILE * fp) {  for (int k=0;k<m_nOutputNetNumber;k++)  {      for (int j=0;j<m_nHideNetNumber;j++)      {     m_npHideToOutputWeightMatrix+=*m_npAlpha*m_fpD_Err*m_npHideOutputValue;//Êä³ö²ãÖÁÒþ²ãµÄÈ¨Öµµ÷Õû      }     m_npOutputLayerThreshold+=*m_npAlpha*m_fpD_Err;//Êä³ö²ãÖÁÒþ²ãµÄãÐÖµµ÷Õû  } } //m_nLearningSampleNumber¸öÑù±¾µÄÈ«¾ÖÎó²î¼ÆËã×Ó³ÌÐò//// double CBpNetWork::GlobalErrSum() {  double total_err=0;       for (int m=0;m<m_nLearningSampleNumber;m++)         total_err+=m_npTotalErrofM;//Ã¿¸öÑù±¾µÄ¾ù·½Îó²î¼ÓÆðÀ´¾Í³ÉÁËÈ«¾ÖÎó²î         return total_err; }  //Òþ²ãÖÁÊäÈë²ãµÄÈ¨Öµµ÷Õû¡¢Òþ²ããÐÖµµ÷Õû¼ÆËã×Ó³ÌÐò///// void CBpNetWork::AdjustWeightAndThresholdHideLayToOutLay(int m,FILE * fp) { for (int j=0;j<m_nHideNetNumber;j++)  {         for (int i=0;i<m_nInputNetNumber;i++)   {        //cout<<"µÚ"< fprintf(fp,"µÚ%d¸öÑù±¾Ê±µÄÊä³ö²ãÖÁÒþ²ãµÄÈ¨ÖµÎª£º%f\n",m,m_npHideToOutputWeightMatrix);       m_npInputToHideWeightMatrix+=*m_npBeta*m_fpE_Err*m_fpSingleInputSample;//Òþ²ãÖÁÊäÈë²ãµÄÈ¨Öµµ÷Õû   }       m_npHideLayerThreshold+=*m_npBeta*m_fpE_Err;  } }void CBpNetWork::Excute(Study_Data_T * spl, double err, int count) {  double temp_err = 0;  int local_count =0;  FILE * f;  if(count == 0)   count = 1000;  if(!spl)   return;  f = fopen("c:\\ err.txt","a+");  do  {  Tarning(spl);  temp_err = GlobalErrSum(); fprintf(f," %d, %f ",local_count, temp_err);  temp_err = sqrt(temp_err/m_nLearningSampleNumber);  local_count++;  }while(temp_err>err);//&&local_count<count);  fprintf(f,"\n");  fclose(f);  f = fopen("c:\\ weight.txt","a+");  for(int i =0;i<m_nHideNetNumber*m_nInputNetNumber;i++)  fprintf(f," %f ",m_npInputToHideWeightMatrix);  fprintf(f,"\n\n");  for (i = 0; i<m_nOutputNetNumber*m_nHideNetNumber;i++)  fprintf(f,"%f",m_npHideToOutputWeightMatrix);  fprintf(f,"\n\n");  fclose(f); }void CBpNetWork::Tarning(Study_Data_T * spl) {   int i = 0;   Study_Data_T * tlnspl =spl;   Study_Data_T * ttnspl =spl;  for(;tlnspl != NULL&& ttnspl != NULL && i<m_nLearningSampleNumber*m_nInputNetNumber;i++)  {   InputOfNoMSampleData(0, tlnspl);   InputOfNoMTrainingData(0, ttnspl);   HideLayInOut();   OutLayInOut();   NormalizeErrOutLayToHideLay(0);   NormalizeErrHideLayToOutLay();   AdjustWeightAndThresholdOutLayToHideLay(0,NULL);   AdjustWeightAndThresholdHideLayToOutLay(0,NULL);  } }void CBpNetWork::Set(double * pointer, int num) { int i =0; for(;i<num;i++) pointer = 0; }void CBpNetWork::Predict(int m, Study_Data_T *PresentData) { InputOfNoMSampleData(m,  PresentData); HideLayInOut(); OutLayInOut(); }   //head file// BpNetWork.h: interface for the CBpNetWork class. // //////////////////////////////////////////////////////////////////////#if !defined(AFX_BPNETWORK_H__2FDDE314_11F6_4D49_A046_C42B29C3754D__INCLUDED_) #define AFX_BPNETWORK_H__2FDDE314_11F6_4D49_A046_C42B29C3754D__INCLUDED_#if _MSC_VER > 1000 #pragma once #endif // _MSC_VER > 1000  #include "math.h" typedef struct Study_Data_S     {     double *input;     double *teach;     } Study_Data_T; class CBpNetWork  { public://Êä³ö²ãÖÁÒþ²ãµÄÒ»°ã»¯Îó²î×Ó³ÌÐò////  void Predict(int m, Study_Data_T *PresentData); void Tarning(Study_Data_T * spl);  void Excute(Study_Data_T * spl, double err, int count);  void AdjustWeightAndThresholdHideLayToOutLay(int m, FILE * fp);  double GlobalErrSum();  void AdjustWeightAndThresholdOutLayToHideLay(int m, FILE * fp);  void NormalizeErrHideLayToOutLay();  void NormalizeErrOutLayToHideLay(int m);  void OutLayInOut();  void HideLayInOut();  void InputOfNoMSampleData(int m,Study_Data_T * lnspl);  void InputOfNoMTrainingData(int m,Study_Data_T * lnspl);  bool Init();  CBpNetWork(int LearningSplNum, int OutLayNumber,int HideLayNumber,int inLayNum); double* m_npBeta;//Òþ²ãÖÁÊäÈë²ãÑ§Ï°Ð§ÂÊ  double* m_npAlpha;//Êä³ö²ãÖÁÒþ²ãµÄÑ§Ï°Ð§ÂÊ  double* m_npTotalErrofM;//µÚm¸öÑù±¾µÄ×ÜÎó²î  double* m_npOutputLayerThreshold;//Êä³ö²ãµÄãÐÖµ  double* m_npHideLayerThreshold;//Òþ²ãµÄãÐÖµ  double* m_npOutputLayerOutputValue;//Êä³ö²ãµÄÊä³ö  double* m_npHideOutputValue;//Òþ²ãµÄÊä³ö  double* m_npOutputLayerInputValue;//Êä³ö²ãµÄÊäÈë  double* m_npHideInputValue;//Òþ²ãµÄÊäÈë  double* m_npHideToOutputWeightMatrix;//Òþ²ãÖÁÊä³ö²ãÈ¨Öµ  double* m_npInputToHideWeightMatrix;//ÊäÈë²ãÖÁÒþ²ãÈ¨Öµ  double* m_fpSingleTrainingSample;//µ¥¸öÑù±¾½ÌÊ¦Êý¾Ý  double* m_fpSingleInputSample;//µ¥¸öÑù±¾ÊäÈëÊý¾Ý  double* m_fpInput;  double* m_fpTeach;  double* m_fpD_Err;  double* m_fpE_Err;  double* m_fpAbsErrValBefore;  Study_Data_T *m_spStudy_Data; CBpNetWork();  virtual ~CBpNetWork();private:  void Set(double*,int );  int m_nLearningSampleNumber;//Ñ§Ï°Ñù±¾¸öÊý  int m_nOutputNetNumber;//ÊäÈë²ãÉñ¾ÔªÊýÄ¿  int m_nHideNetNumber;//Òþ²ãÉñ¾ÔªÊýÄ¿  int m_nInputNetNumber;//Êä³ö²ãÉñ¾ÔªÊýÄ¿ };#endif // !defined(AFX_BPNETWORK_H__2FDDE314_11F6_4D49_A046_C42B29C3754D__INCLUDED_)

页: [1]

数学建模社区-数学中国's Archiver

改写的C＋＋BP类 未测试

测试通过的BP类，可以收敛Total误差

改写的C＋＋BP类未测试