|
#include
#include
#include
#include
#include
//----------------------------------------------------------------------------
// DEFINES
#define MAXNEURONS 64
#define MAXPATTERNS 10
#define MAXITER 600000
#define TRUE 1
#define FALSE 0
//#undef SIDESHOW
#define SIDESHOW
char *TRAINFILE = "H7X8N4.TRN";
char *TESTFILE = "H7X8D5.TST";
//----------------------------------------------------------------------------
// FUNCTION PROTOTYPES
// network fns
void InitNetRand(void); // 随机初始化用户数据
void LoadTrainingSet(char *Fname); // 从文件加载训练数据
void TrainNet(); // 训练NN
void RunNet(void); // 联想记忆
int UpdateNeuron(int j); // 更新神经元，如果状态改变了返回true
void LoadUserPattern(char *Fname);
int QueryAllClean(void); // 如果所有神经元至少被访问一次，返回true
void SetAllToDirty(void); // 设置所有神经元为未被访问
void Initialize(int cnt, char *Name); // 初始化数据
void DisplayPatVect(int V[MAXNEURONS]); // 显示模式向量
void SavePatVect(int V[MAXNEURONS], // 将模式向量保存于文件中
char *txt, int i);
void DisplayWeights(void); // 显示权重向量
void DisplayPatterns(void); // 显示所有训练模式
int QueryUserInt(char *msg);
void KeyWait(void);
void KeyWait2(char *txt);
int random(int N);
//----------------------------------------------------------------------------
// GLOBALS
int PatVec[MAXNEURONS]; // 模式向量（缓存）
int PatMatrix[MAXPATTERNS][MAXNEURONS]; // 模式向量
int NEURON[MAXNEURONS]; // NN
int T[MAXNEURONS][MAXNEURONS]; // 权重矩阵
int NumNeurons; // 神经元数量
int NumPatterns; // 模式数量
int PatternX; // 显示维数x
int PatternY; // 显示维数y
int Dirty[MAXNEURONS]; // 如果神经元未被更新则返回true
FILE *ARCHIVE;
//----------------------------------------------------------------------------
int random(int N){
int x;
x=N*rand();
return (x/RAND_MAX);
}
void DisplayPatVect(int V[MAXNEURONS]){
int x,y,indx;
indx=0;
for (y=0; y for (x=0; x if (V[indx]==1) {
printf("X");
} else {
printf(".");
} /* endif */
indx++;
} /* endfor */
printf("\n");
} /* endfor */
printf("\n");
}
void SavePatVect(int V[MAXNEURONS], char *txt, int i) {
int x,y,indx;
indx=0;
fprintf(ARCHIVE,"\n");
for (y=0; y for (x=0; x if (V[indx]==1) {
fprintf(ARCHIVE,"X");
} else {
fprintf(ARCHIVE,".");
} /* endif */
indx++;
} /* endfor */
fprintf(ARCHIVE,"\n");
} /* endfor */
fprintf(ARCHIVE,"\n%s ",txt);
if (i>=0) fprintf(ARCHIVE,"%d ",i);
fprintf(ARCHIVE,"\n\n ");
}
void DisplayWeights(){
int i,j;
fprintf(ARCHIVE,"WEIGHTS:\n");
for (i=0; i fprintf(ARCHIVE,"[");
for (j=0; j fprintf(ARCHIVE, " %d",T[j][i]);
} /* endfor */
fprintf(ARCHIVE,"]\n");
} /* endfor */
}
void DisplayPatterns() {
int i,p;
for (p=0; p for (i=0; i PatVec[i] =PatMatrix[p][i];
} /* endfor */
DisplayPatVect(PatVec); // show 1st training pattern
SavePatVect(PatVec, "训练模式", p+1);
printf("\n\n训练模式 %d of %d\n\n",p+1,NumPatterns);
KeyWait();
} /* endfor */
}
int QueryUserInt(char *msg){
int rv;
printf("Enter %s ==> ",msg);
scanf("%d",&rv);
return rv;
}
void KeyWait(void){
printf("Press any key to continue.\n");
while (!kbhit()) { } /* endwhile */
getch();
system("cls");
}
void KeyWait2(char *txt){
printf("\n\n%s\n",txt);
KeyWait();
}
void InitNetRand() {
int i,r;
fprintf(ARCHIVE,"随机创建用户模式\n");
srand(5);
for (i=0; i r=random(100);
if (r >= 50) {
NEURON[i]=0;
}
else {
NEURON[i]=1;
} /* endif */
} /* endfor */
}
void LoadTrainingSet(char *Fname) {
int pat,j, InVal;
FILE *PATTERNFILE;
printf("从文件: %s 加载训练模式\n",Fname);
fprintf(ARCHIVE,"从文件: %s 加载训练模式\n",Fname);
PATTERNFILE = fopen(Fname,"r");
if (PATTERNFILE==NULL){
printf("不能打开训练模式文件: %s",Fname);
exit(0);
}
fscanf(PATTERNFILE,"%d",&NumNeurons); // 获得神经元数
fscanf(PATTERNFILE,"%d",&NumPatterns); // 获得模式数
fscanf(PATTERNFILE,"%d",&PatternX); // 矩阵宽度
fscanf(PATTERNFILE,"%d",&PatternY); // 矩阵长度
printf("%d 模式被加载\n",NumPatterns);
fprintf(ARCHIVE,"%d 模式被加载\n",NumPatterns);
for (pat=0; pat for (j=0; j fscanf(PATTERNFILE,"%d",&InVal);
PatMatrix[pat][j] =InVal;
} // endfor
} // endfor
fclose(PATTERNFILE);
}
void LoadUserPattern(char *Fname) {
int j, InVal;
FILE *PATTERNFILE;
printf("从文件: %s 加载用户模式\n", Fname);
fprintf(ARCHIVE,"从文件: %s 加载用户模式\n", Fname);
PATTERNFILE = fopen(Fname,"r");
if (PATTERNFILE==NULL){
printf("不能打开用户模式文件: %s",Fname);
exit(0);
}
printf("\n");
for (j=0; j fscanf(PATTERNFILE,"%d",&InVal);
NEURON[j] =InVal;
} // endfor
fclose(PATTERNFILE);
}
void TrainNet(){
int i,j,pat;
int Sum;
printf("训练开始。。。\n");
for (i=0; i for (j=0; j Sum=0;
for (pat=0; pat Sum += (2*PatMatrix[pat][i]-1) * (2*PatMatrix[pat][j]-1);
} /* endfor */
T[j][i] = T[i][j] = Sum;
} /* endfor */
} /* endfor */
for (i=0; i T[i][i]=0;
} /* endfor */
printf("训练结束。。。");
}
int QueryAllClean() {
int i;
for (i=0; i if (Dirty[i]==TRUE) return FALSE;
} // endfor
return TRUE;
}
void SetAllToDirty() {
int i;
for (i=0; i Dirty[i]=TRUE;
} // endfor
}
int UpdateNeuron(int j) {
int i;
int Sum = 0;
int OldState = NEURON[j];
for (i=0; i Sum += T[j][i] * NEURON[i];
} /* endfor */
if (Sum < 0) {
NEURON[j] = 0;
}
else {
if (Sum>0)
NEURON[j] = 1;
} /* endif */
if (NEURON[j] == OldState) {
return 0;
}
else {
return 1;
} /* endif */
}
void RunNet(void) {
int j;
int Converged = FALSE;
int ChngCount = 0;
unsigned long int IterCount = 0;
int ArchCnt=0;
SetAllToDirty();
while ( (!Converged) && (IterCount < MAXITER) ) {
j = random(NumNeurons);
ChngCount += UpdateNeuron(j); // 如果神经元被改变则被改变的神经元数量加1
DisplayPatVect(NEURON);
printf("运行... 次数=%d \n",IterCount);
if (ArchCnt>=9) {
SavePatVect(NEURON, "NN 输出 =", IterCount+1);
ArchCnt=0;
} else {
ArchCnt++;
} /* endif */
Dirty[j] = FALSE; // 记录此神经元被访问
if (QueryAllClean()) { //测试时候所有神经元被访问
// 如果每个神经元至少被访问一次
if (ChngCount == 0) { // 测试本次循环是否有神经元被更新
// 没有被更新的神经元
Converged = TRUE;
printf("\n识别结束");
SavePatVect(NEURON, "识别结束于次数=", IterCount);
}
else {
// 如果有被更新的神经元，则进行新的循环
SetAllToDirty();
ChngCount=0;
} /* endif */
} /* endif */
IterCount++;
} /* endwhile */
}
void Initialize(char *Name1, char *Name2) {
char TrnName[50];
char TstName[50];
ARCHIVE = fopen("HOPFIELD.RST","w");
if (ARCHIVE==NULL){
printf("不能打开默认归档文件: HOPFIELD.RST");
exit(0);
}
strcpy(TrnName,Name1);
strcpy(TstName,Name2);
LoadTrainingSet(TrnName);
if (strcmpi("RANDOM",TstName) )
LoadUserPattern(TstName);
else
InitNetRand(); // 随机产生用户模式
KeyWait();
}
int main(int argc, char *argv[]) {
int i;
char *f1, *f2;
if(argc == 3){
TRAINFILE = argv[1];
TESTFILE = argv[2];
}
Initialize(TRAINFILE, TESTFILE);
DisplayPatVect(NEURON); // 显示网络用来测试
SavePatVect(NEURON, "用户模式", -1);
KeyWait2("用户模式");
DisplayPatterns();
TrainNet();
DisplayWeights();
RunNet();
fclose(ARCHIVE);
return 0;
}
| 