/*
	Self-Organizing Map:  A demo using C language and gnuplot !!!

	 2005. 4.15 by A.Date
	 Last modified 1 May 2009

*/
#include <stdlib.h>
#include <stdio.h>
#include <math.h>


#define N_XUNITS 20
#define N_YUNITS  1
#define DIM_INPUT 3 /* dimension of input */

#define Z_RANGE 1.0

// #define SLEEP 1000
#define SLEEP 10

double X_RANGE = 0.01;
double Y_RANGE = 0.01;

void som2d();
void init_reference();
void write_data();

double ALPHA=0.01;
double SIGMA=0.8;
int N_LEARNING = 10000;
int N_SUB_LEARNING = 1000;

double S[DIM_INPUT];    /* input */

double RV[N_XUNITS][N_YUNITS][DIM_INPUT];
int T[N_XUNITS][N_YUNITS]; 

FILE *gp;


typedef struct _Position Position;
struct _Position {
	int  x;
	int  y;
};



int main (int argc, char *argv[] )
{
	int i,j;
	int r1, r2;
	double x,y;
	long seed = 1234567;

        for (i=1; i<argc; i++) {
          switch (*(argv[i]+1)) {
          case 'r':
            seed = atoi(argv[++i]);
            break;
          default:
            fprintf(stderr, "Usage : %s\n",argv[0]);
            fprintf(stderr, "\t-r : random-seed(%ld)\n",seed) ;
            exit(0);
            break;
          }
        }

	srand48(seed);

	gp = popen("gnuplot -geometry 640x480","w");
	fprintf(gp, "set term x11\n");
	fprintf(gp, "set parametric\n");
	fprintf(gp, "set data style linespoints\n");
	fprintf(gp, "set xlabel \"x\"\n");
	fprintf(gp, "set ylabel \"y\"\n");
	fprintf(gp, "set zlabel \"z\"\n");

	fprintf(gp, "set xrange[-1:1]\n");
	fprintf(gp, "set yrange[-1:1]\n");
	fprintf(gp, "set zrange[0:1]\n");

	fprintf(gp, "set nokey\n");
//	fprintf(gp, "set view 0,0\n");
	fflush(gp);

	while(1){

		init_reference();
		for(i=0;i<N_LEARNING;i++){
			fprintf(gp, "splot '-'\n");
			fflush(gp); 

			for(j=0;j<N_SUB_LEARNING;j++){
				som2d();
			}
			X_RANGE = X_RANGE + 0.001;

			write_data();
			fprintf(gp,"e\n");
			fprintf(gp, "set title \"t=%d\"\n",i*j+j);
			fflush(gp); 


			r1 = i % 360;
			r2 = j % 20;
			fprintf(gp, "set view %d,%d\n",r2+50, r1);
			fflush(gp); 

			usleep(SLEEP*100);
		}
	}
	fclose(gp);
}


void
init_reference(){ 

	int i,j,k; 
	for (i=0; i < N_XUNITS; i++){
		for (j=0; j < N_YUNITS; j++){
			for (k=0; k<DIM_INPUT; k++){
				RV[i][j][k] = drand48();
			}
		}       
	}       
}


void
som2d()
{

/*
	1.  Choose an input x randomly.  
	2.  Find the winnder, and apply neighborhood learning rule.
*/ 

	int i,j,k;
	double tmp;
	double min;
	double denom;
	double hci;
	double dist;
	double sum = 0.0;

	Position argmin; 
	
	/* 	Choose an input x randomly.  */


	S[0] = X_RANGE;
	S[1] = X_RANGE;
	while ( sqrt(S[0]*S[0] + S[1]*S[1]) > X_RANGE ){
		S[0] = drand48()*X_RANGE;
		S[1] = drand48()*X_RANGE;
	}
	S[0] = S[0] - 0.5*X_RANGE;
	S[1] = S[1] - 0.5*X_RANGE;
	S[2] = drand48()*Z_RANGE;

/*
	for (k=0; k<DIM_INPUT; k++){
		S[k] = drand48();
	}
*/

	/* find closest ref. vector */

	for (k=0; k<DIM_INPUT; k++){
		tmp = RV[0][0][k] - S[k];
		sum += tmp*tmp;
	}
	argmin.x = 0; 
	argmin.y = 0; 
	min = sum; 
	
	for (i=0; i<N_XUNITS; i++){
		for (j=0; j<N_YUNITS; j++){
		sum = 0.0;
			for (k=0; k<DIM_INPUT; k++){
				tmp = RV[i][j][k] - S[k];
				sum += tmp*tmp;
			}
			if ( sum < min ){
				min = sum;
				argmin.x = i; 
				argmin.y = j; 
			} 
		} 
	}

	/* learning,  c = argmin */

	for (i=0; i<N_XUNITS; i++){
		for (j=0; j<N_YUNITS; j++){
			dist = (i-argmin.x)*(i-argmin.x) +(j-argmin.y)*(j-argmin.y);
			denom = 2.0*SIGMA*SIGMA;
			hci = ALPHA*exp(-(double)dist/denom);

			for (k=0; k<DIM_INPUT; k++){ 
				RV[i][j][k] =  RV[i][j][k] + hci* (S[k] -RV[i][j][k]);
			} 
		}
	} 
	
}


void
write_data(){

	int i,j;

	/* write data for gnuplot */

	for (i=0; i<N_XUNITS; i++){
		for (j=0; j<N_YUNITS; j++){
			fprintf(gp, "%.5lf %.5lf %.5lf\n", RV[i][j][0], RV[i][j][1], RV[i][j][2]); 
		}
		fprintf(gp, "\n"); 
	}		

	for (j=0; j<N_YUNITS; j++){
		for (i=0; i<N_XUNITS; i++){
			fprintf(gp, "%.5lf %.5lf %.5lf\n", RV[i][j][0], RV[i][j][1], RV[i][j][2]); 
		}
		fprintf(gp, "\n"); 
	}

}