som.h

 /*
 				som.h

*%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
*
*	Part of:	A program using neural networks.
*
*	Author:		E.BERTIN, IAP & Leiden observatory.
*
*	Contents:	Include for Kohonen's Self Organizing Map (V2.0).
*
*	Last modify:	17/12/97
*
*%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
*/

/*--------------------------------- constants ------------------------------*/

#define	INPUT_MAXDIM		9	/* Maximum dimensionality of input */
#define	SOM_MAXDIM		6	/* Maximum dimensionality of the SOM */

/*------------------------------- SOM flags --------------------------------*/

#define		SOM_NODE	0x01	/* Compute at some exact node pos */
#define		SOM_PHOTOM	0x02	/* Do photometry */
#define		SOM_GRADIENT	0x04	/* Compute interpolated SOM gradient */
#define		SOM_LINE	0x08	/* Proceed along a specific line */

/*--------------------------- structure definitions -------------------------*/

typedef struct
  {
  int		inputdim;		/* Dimensionality of input vector */
  int		*inputsize;		/* Dimensions of the input vector */
  int		ninput;			/* Total number of inputs */
  int		nextrainput;		/* Number of extra inputs */
  int		neurdim;		/* Dimensionality of the SOM */
  int		*neursize;		/* Dimensions of the SOM */
  int		nneur;			/* Total number of neurons */
  int		*neurstep;		/* Stepping through the SOM */
  float		*weight;		/* Weights */
  int		nweight;		/* Total number of weights */
  float		*input;			/* Input data */
  float		*inputw;		/* Input data weighting */
  float		*proto;			/* Current composite prototype */
  float		*dproto;		/* Current composite gradients */
  float		*vector;		/* Current SOM coordinates */
  float		*dvector;		/* Current SOM search direction */
  float		learnrate, clearnrate;	/* Starting and current learn. rates */
  float		learndecay;		/* Learning decay rate */
  float		kernw, ckernw;		/* Starting and current kernel width */
  float		kernwdecay;		/* Kernel width decay rate */
  float		xy_stiff;		/* Stiffness of the X/Y mapping */
  int		*freq;			/* Number of winning times per node */
  int		ntrain;			/* # of training examples so far */
  int		nsweep;			/* # of sweeps through the whole set */
  float		amp, sigamp;		/* Best fitting amplitude and error */
  float		stderror;		/* Global reduced error */
  }	somstruct;

somstruct	*thesom;

/*---------------------------------- protos --------------------------------*/

extern somstruct	*som_load(char *filename);

extern float		som_err(somstruct *som, float dist, int flag),
			som_linmin(somstruct *som);

extern int		som_mkweight(somstruct *som,float back,float backnoise,
				float gain);

extern void		som_conjgrad(somstruct *som, float ftol),
			som_end(somstruct *som),
			som_phot(somstruct *som, float back,float backnoise,
				float gain, float dx, float dy,
				float *vector, float clip),
			som_start(somstruct *som, float *context,
				int ncontext, float x, float y);