A simple Model of Neurone Network
Here is source code of a Neurone. You can know the principle of Neuron Network after you
understand it. It is basically a complete Neuron Network because it
includes Back-Propogation training Algorithm though it's so short.
Function menu() is to show a menu, including Train, Use, and Demo.
The weight values (AND1,AND2, OR1,OR2) in Demo are generated from training.
/*Neurone Model*/
#include < math.h>
#include < stdio.h>
#include < time.h>
#include < stdlib.h>
#include < conio.h>
#define AND1 .571388
#define AND2 .560999
#define OR1 1.149045
#define OR2 1.131896
#define SITA 1
#define NIU .1
#define OMIGA 50
float w[2];
float terg[4][3];
main()
{int m_result = 1, s;
float w1[2];
while(m_result != 5){
m_result = menu(m_result);
date();
switch(m_result){
case(1): /* Train*/
case(3):randomize(); /*ReTrain*/
w[0] = random(80)+10; w[0] = w[0]/100.0;
w[1] = random(80)+10; w[1] = w[1]/100.0;
study();
break;
case(2):practic(w); /*Use new Model*/
break;
case(4):printf(" what do you want to see, OR or AND?(1/2) " );
scanf(" %d" , & s);
if(s==1){w1[0] = OR1; w1[1] = OR2; }
else if(s==2){w1[0] = AND1; w1[1] = AND2; }
else break;
practic(w1);
break; /*Demo, show trained model*/
default:{}
} /*End Switch*/
} /*End While */
clrscr();
printf(" Good Bye!" );
}
menu(int r)
{int r1;
clrscr();
printf(" \n\n\n" );
printf(" \t\t1\tStudy a Model\n" ); /*Train */
printf(" \t\t2\tPratice the model\n" ); /*Retrain */
printf(" \t\t3\tDefail the MOdel\n" ); /*Use Model trained */
printf(" \t\t4\tPrepared Model\n" ); /*Demo*/
printf(" \t\t5\tEnd\n" ); /*End */
printf(" \n\n\t\tWhat do you want to choice?(%d)" , r);
r1 = getche();
if (r1!=13) r = r1-48;
return(r);
}
date() /*Show values.*/
{int i, j;
clrscr();
for(i = 0; i < 4; i++)
for(j = 0; j< 3; j++){
gotoxy(20+j*4, i+5);
printf("%.0f", terg[i][j]);
}
printf("\n\n\tThe quarities:%f, %f \n", w[0], w[1]);
}
study() /*function of training*/
{float x, y, width;
int wrong = 1, i, total = 0;
input(); /*andinput(); */ /*orinput(); */
while(wrong){
wrong = 0;
for(i = 0; i< 4; i++){
x = OMIGA*(w[0]*terg[i][0]+w[1]*terg[i][1]-SITA);
y = 1.0/(1.0+exp(-x));
width = terg[i][2]-y; /*Delta = "Teacher signal-Y" */
if(fabs(width)>=.01){
wrong = 1;
total++;
w[0] = w[0]*(1+width*NIU); /*Change Weight0*/
w[1] = w[1]*(1+width*NIU); /*Change Weight1*/
if(total> =200) /*Prevent Dead Lock*/
{error(); return 1; }
}
}
}/*If (fabs(width)>=0.01) then wrong = 0 and jump out of the loop*/
}
practic(quarity) /*Use trained model*/
float quarity[2];
{int k, y;
float x, width, x1, x2;
for(k = 0; k < 4; k++){
gotoxy(0, 4+k*3);
printf("\nPlease input :x1, x2 ");
scanf("%f, %f", &x1, &x2);
x = OMIGA*(quarity[0]*x1+quarity[1]*x2-SITA);
y = 1.0/(1.0+exp(-x))+.01; /*Accuracy is 0.01*/
printf("\t\t\ty = "%d" , y);
getch();
}
}
input() /*Input events when training*/
{int i;
printf("Please input the 4 groups data:\n\n");
for(i = 0; i< 4; i++){
printf("\tThe %d group x1, x2, teacher's point: ", i+1);
scanf("%f, %f, %f", &terg[i][0], &terg[i][1], &terg[i][2]);
}
clrscr();
}
andinput()
{terg[0][0] = 1; terg[0][1] = 1; terg[0][2] = 1;
terg[1][0] = 1;
terg[1][1] = 0;
terg[1][2] = 0;
terg[2][0] = 0;
terg[2][1] = 1;
terg[2][2] = 0;
terg[3][0] = 0;
terg[3][1] = 0;
terg[3][2] = 0;
}
orinput()
{terg[0][0] = 1;
terg[0][1] = 1;
terg[0][2] = 1;
terg[1][0] = 1;
terg[1][1] = 0;
terg[1][2] = 1;
terg[2][0] = 0;
terg[2][1] = 1;
terg[2][2] = 1;
terg[3][0] = 0;
terg[3][1] = 0;
terg[3][2] = 0;
}
error() /*Exit, maybe the Original weight value is not good,
or the model is not relizable (ex. XOR Model) */
{clrscr();
printf("\n\n\n\tBecause the firse QUARITIES's choise is unavailble, \n");
printf("\tIt can't get the right result, forveve!");
}
I think the comments in the source code are enought to read.
|
