https://github.com/cran/RandomFields
Raw File
Tip revision: e5a7a2f272b7834f96c925ced7acfa0c6456a87f authored by Martin Schlather on 17 April 2017, 22:09:51 UTC
version 3.1.50
Tip revision: e5a7a2f
Xempvario.cc
/*
  Authors 
  Martin Schlather, schlather@math.uni-mannheim.de 

  a second library for calculating the empirical variogram

  Copyright (C) 2002 -- 2017 Martin Schlather, 

  This program is free software; you can redistribute it and/or
  modify it under the terms of the GNU General Public License
  as published by the Free Software Foundation; either version 3
  of the License, or (at your option) any later version.

  This program is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  GNU General Public License for more details.

  You should have received a copy of the GNU General Public License
  along with this program; if not, write to the Free Software
  Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
*/


#include <Rmath.h>  
#include <stdio.h>  
//#include <stdlib.h>
#include "RF.h"
 

// z coordinate run the fastest in values, x the slowest   

#define TOOLS_MEMORYERROR 1
#define TOOLS_XERROR 2
#define TOOLS_BIN_ERROR 3
#define TOOLS_METHOD 4
#define NEARBY 1e15
#define METHOD_CROSS 0
#define METHOD_PSEUDO 1
#define METHOD_COVARIANCE 2
#define METHOD_PSEUDOMADOGRAM 3
#define METHOD_CROSSMADOGRAM 4

// naechste Zeile nur notwendig, weil atan2 in Windows nicht
// ordentlich programmiert ist
#define NEARBYINT(x)  (FLOOR((x) * (NEARBY) + 0.5) / (NEARBY))
// #define NEARBYINT(x)  x

SEXP empvarioXT(SEXP Xsexp, SEXP Tsexp, 
		SEXP Lx, 
	        SEXP Values, SEXP Repet, SEXP Grid,
		SEXP Bin, SEXP Nbin, 
		SEXP Phi,    // vector of a real and an integer
		SEXP Theta,  // vector of a real and an integer
		SEXP DT,   // in grid units, smallest positive value, for  
		//            the temporal lag of the emp. variogram
	        //            stepT * nstepT is the greatest time span
		SEXP Vdim, SEXP Method)
/*    X      : matrix of coordinates, ### rows define points
 *    lx     : length of x,y, and z
 *    values : (univariate) data for the points
 *    repet  : number of repetitions (the calculated emprical variogram will 
 *             be an overall average, see also EmpiricalVariogram in empvario.R)
 *    grid   : (boolean) if true lx must be 3 [ x==c(start,end,step) ]
 *    bin    : specification of the bins 
 *             sequence is checked whether it is strictly isotone
 *    nbin   : number of bins. i.e. length(bin)-1
 *    sum    : empirical variogram
 *    n      : number of pairs for each bin
 *    vdim   : dimension of data
 *    Method : 0 cross-variogram, 1 pseudo-variogram, 2 covariance
 */
{ 
  long rep;
  int i, halfnbin,gridpoints[4], err, totalbins, totalspatialbins,
    twoNphi, twoNphiNbin, Ntheta, nbinNphiNtheta, maxi[4], mini[4], ix, iy, iz,
    it, endX, startX, endY, startY, endZ, startZ, endT, low, cur, up, totaln,
    ktheta, kphi, nstepTstepT, vec[4], deltaT, x, y, z, t,  j, stepT, nstepT; 
  double *xx[4], *BinSq, maxbinsquare, step[4], delta[4], psq[4], dx[4],
    startphi, invsegphi, starttheta, invsegtheta, thetadata, phidata,
    zylinderradius,
    *sumhead = NULL,
    *sumtail = NULL;
  SEXP Res;
  
  // turning SEXP to c++ variable
  int lx = INTEGER(Lx)[0], repet = INTEGER(Repet)[0],
    grid = INTEGER(Grid)[0], nbin = INTEGER(Nbin)[0],
    *dT = INTEGER(DT), vdim = INTEGER(Vdim)[0], method = INTEGER(Method)[0];
  double *X = REAL(Xsexp), *T = REAL(Tsexp),
    *values = REAL(Values), *bin = REAL(Bin),
    *phi = REAL(Phi), *theta = REAL(Theta), *res = NULL;
  
  stepT = dT[0];
  nstepT = dT[1];
  nstepTstepT = stepT * nstepT;
  twoNphi =  phi[1]==0 ?  1 : (2 * (int) phi[1]);
  twoNphiNbin = twoNphi * nbin;
  Ntheta = theta[1]==0 ? 1 : (int) theta[1];
  startphi = phi[0] - PI / (double) twoNphi; // [0, 2 pi]
  invsegphi = phi[1] / PI; // note that phi[1] can be zero!
  //starttheta = theta[0] - PIHALF / (double) Ntheta;  // [0, pi]
  starttheta = theta[0] - PIHALF;  // [0, pi]
  invsegtheta = theta[1] / PI; // note that theta[1] can be zero
  nbinNphiNtheta = twoNphiNbin * Ntheta;
   
  BinSq = NULL;
  for (rep=i=0; i<3; i++, rep+= lx) xx[i]=&(X[rep]);
  xx[3] = T;

  if (xx[0]==NULL) {err=TOOLS_XERROR; goto ErrorHandling;}
  for (i=0; i< nbin;i++) {
    if (bin[i]>=bin[i+1])  {err=TOOLS_BIN_ERROR; goto ErrorHandling;}
  }
  if(vdim == 1) {
    if(method == METHOD_CROSS) method = METHOD_PSEUDO;
    if(method == METHOD_CROSSMADOGRAM) method = METHOD_PSEUDOMADOGRAM;
  }
  halfnbin = nbin / 2;
   
  if ((BinSq = (double *) MALLOC(sizeof(double)* (nbin + 1)))==NULL) {
    err=TOOLS_MEMORYERROR; goto ErrorHandling; 
  } 

  totalspatialbins =  twoNphiNbin * Ntheta;
  totalbins = totalspatialbins * (nstepT + 1);
  totaln = totalbins * vdim * vdim;
  if(method == METHOD_COVARIANCE) {
    if( (sumhead = (double *) CALLOC(totaln, sizeof(double))) == NULL) {
      err = TOOLS_MEMORYERROR; goto ErrorHandling;
    }
    if( (sumtail = (double *) CALLOC(totaln, sizeof(double))) == NULL) {
      err = TOOLS_MEMORYERROR; goto ErrorHandling;
    }
  }

  // res contains the variogram, sd and n.bin (res gets returned to function)
  // first column is of res is the variogram
  // second column is the sd and the third n.bin (number of data per bin)
  PROTECT(Res = allocMatrix(REALSXP, totaln, 3));
  res = REAL(Res);
  //printf("total %d %d %d %d \n", nstepT,  twoNphi, *nbin,  Ntheta);

  for(i = 0; i < totaln * 3; res[i++] = 0); 
  for (i=0; i<= nbin; i++){if (bin[i]>0) BinSq[i]=bin[i] * bin[i]; 
    else BinSq[i]=bin[i];
  }

  assert(NEARBYINT(atan2(-1.0, 0.0) + PIHALF) == 0.0);
  assert(atan2(0.0, 0.0) == 0.0);
  maxbinsquare = BinSq[nbin];
 

  //////////////////////////////////// GRID ////////////////////////////////////
  if (grid) {
    int segmentbase[6];
    long totalpoints = 1;
    segmentbase[0]=1; // here x runs the fastest;
    
    for (i=0; i<=3; i++) {
      step[i] = xx[i][XSTEP];
      gridpoints[i] = (int) xx[i][XLENGTH];
      totalpoints *= (gridpoints[i] > 0) ? gridpoints[i] : 1;
      maxi[i] = (int) (SQRT(maxbinsquare) / step[i] + 0.1);
      if (maxi[i] >= gridpoints[i]) maxi[i] = gridpoints[i] - 1;
      mini[i] = -maxi[i];
      //      if (maxi[i]==0)  maxi[i] = 1; wrong 22.11.03
      segmentbase[i+1] =  segmentbase[i] * gridpoints[i];
    }
    segmentbase[4] *= vdim;
    segmentbase[5] = segmentbase[4] * repet;

    //printf("\n");
    //for(int i=totalpoints;i<2*totalpoints;i+=100){
    //  printf("%f ", values[i]);
    //}
    //printf("\n");
    // sementbase: [0] : 1, [1] : length(x), [2] : length(x) * l(y),
    //  [3] : l(x) * l(y) * l(z), [4] : l(x) * l(y) * l(z) * l(T)
    //  [5] : l(x) * l(y) * l(z) * l(T) * repet

    // define needed so there is no if needed before each calculation (grid)
#define FOR(DO)								\
    for (ix=mini[0]; ix <= maxi[0]; ix++) {				\
      delta[0] = ix * step[0];						\
      if ((psq[0] = delta[0] * delta[0]) > maxbinsquare) continue;	\
									\
      vec[0] = ix;							\
      if (ix>=0) {							\
	endX = gridpoints[0] - ix;					\
	startX = 0;							\
      } else {								\
	endX = gridpoints[0];						\
	startX = -ix;							\
      }									\
      for (iy= mini[1]; iy <= maxi[1];  iy++) {				\
	delta[1] = iy * step[1];					\
	if ((psq[1] = delta[1] * delta[1] + psq[0]) > maxbinsquare) continue; \
	vec[1] = vec[0] + iy * segmentbase[1];				\
									\
        /* angles */							\
	phidata = NEARBYINT(atan2(delta[1], delta[0]) - startphi);	\
	while (phidata < 0) phidata += TWOPI;				\
	while (phidata >= TWOPI) phidata -= TWOPI;			\
	kphi = nbin * (int) (phidata * invsegphi);			\
									\
									\
	for (iz=mini[2]; iz <= maxi[2]; iz++) {				\
	  delta[2] = iz * step[2];					\
	  if ((psq[2] = delta[2] * delta[2] + psq[1]) > maxbinsquare) continue;	\
	  vec[2] = vec[1] + iz * segmentbase[2];			\
									\
	  {								\
	    low=0; up= nbin; /* */ cur= halfnbin;			\
	    while(low!=up){						\
	      if (psq[2] > BinSq[cur]) {low=cur;} else {up=cur-1;}/*( . ; . ]*/ \
	      cur=(up+low+1)/2;						\
	    }								\
	  }	/* low*/						\
									\
	  /* angles*/							\
	  thetadata = NEARBYINT(atan2(delta[2], SQRT(psq[1])) - starttheta); \
	  while (thetadata < 0) thetadata += PI;			\
	  while (thetadata >= PI) thetadata -= PI;			\
	  ktheta = (int) (thetadata * invsegtheta);			\
									\
	  low += kphi + ktheta * twoNphiNbin;				\
									\
	  for (deltaT=0, it=low; deltaT<=nstepTstepT;			\
	       deltaT+=stepT, it+=nbinNphiNtheta) {			\
	    vec[3] = vec[2] + deltaT * segmentbase[3];			\
									\
	    assert(startX>=0);						\
	    for (x=startX; x<endX; x++) {				\
	      if (iy>=0) {						\
		endY = gridpoints[1]  * segmentbase[1] - vec[1];	\
		startY = x;						\
	      } else {							\
		endY = gridpoints[1] * segmentbase[1];			\
		startY = x - iy * segmentbase[1];			\
	      }								\
	      assert(startY>=0);					\
	      for (y=startY; y<endY; y+=segmentbase[1]) {		\
		if (iz>=0) {						\
		  endZ = gridpoints[2] * segmentbase[2] - vec[2];	\
		  startZ = y;						\
		} else {						\
		  endZ = gridpoints[2] * segmentbase[2];		\
		  startZ = y - iz * segmentbase[2];			\
		}							\
                assert(startZ>=0);					\
		for (z=startZ; z<endZ; z+=segmentbase[2]) {		\
		  endT = gridpoints[3] * segmentbase[3] - vec[3];	\
		  for (t=z; t<endT; t+=segmentbase[3]) {		\
		    for(int row = 0; row < vdim; row++) {		\
		      for(int col = row; col < vdim; col++) {		\
			int curbin = it + totalbins * (vdim * row + col);	\
			for (rep = t; rep < segmentbase[5]; rep += segmentbase[4]) { \
			  int rv;					\
			  rv = rep + vec[3];				\
			  assert(rv < segmentbase[5] && rv >=0);	\
			  /* inserts calculation from switch below */	\
			  DO;						\
			  res[curbin] += x2;				\
			  res[curbin + totaln] += x2 * x2;		\
			  res[curbin + 2*totaln]++;			\
			}						\
		      }							\
		    } /* repeat*/					\
		  } /* t */						\
		} /* z */						\
	      } /* y */							\
	    } /* x */							\
	  } /* deltaT */						\
	} /* iz */							\
      } /* iy */							\
    } /* ix */

    // checks which method shoud be used (gets inserted above)
   
    switch(method) {
    case(METHOD_PSEUDO):
      // pseudo variogram
      FOR(double x2 = values[rep + totalpoints * row]
	  - values[rv + totalpoints * col];
	  x2 *= x2);
      break;
    case(METHOD_CROSS):
      // cross variogram
      FOR(double x2 = (values[rep + totalpoints * row]
		       - values[rv + totalpoints * row])
	  * (values[rep + totalpoints * col]
	     - values[rv + totalpoints * col]));	  
      break;
    case(METHOD_COVARIANCE):
      FOR(double x2 = (values[rep + totalpoints * row]
		       * values[rv + totalpoints * col]);
	  sumhead[curbin] += values[rep + totalpoints * row];
	  sumtail[curbin] += values[rv + totalpoints * col]);
      break;
    case(METHOD_PSEUDOMADOGRAM):
      // pseudo madogram
      FOR(double x2 = FABS(values[rep + totalpoints * row]
			   - values[rv + totalpoints * col]));
      break;
    case(METHOD_CROSSMADOGRAM):
      // cross madogram
      FOR(double x2 = FABS(values[rep + totalpoints * row]
			   - values[rv + totalpoints * row])
	  * FABS(values[rep + totalpoints * col]
		 - values[rv + totalpoints * col]); x2 = SQRT(x2));	  
      break;
    default:
      err = TOOLS_METHOD; goto ErrorHandling;
    }
    
  } else {
    ////////////////////////////////////  ARBITRARY /////////////////////////////
    // rows : x, columns : T 
    long totalpoints, totalpointsrepetvdim, totalpointsvdim, spatial,  jj;
    spatial = lx;
    i = 3;
    step[i] = xx[i][XSTEP];
    gridpoints[i] = (int) xx[i][XLENGTH];
    totalpoints = spatial * gridpoints[i];
    totalpointsvdim = totalpoints * vdim;
    totalpointsrepetvdim = totalpoints * repet;
    
    // define needed so there is no if needed before each calculation (arbitrary)
#define FORARB(DO)							\
    for (i=0;i<spatial;i++) { /* to have a better performance for large*/ \
      /*                  data sets, group the data first into blocks*/	\
      for (j=0; j<spatial; j++) {					\
	double distSq;							\
	dx[0] = xx[0][j] - xx[0][i];					\
	dx[1] = xx[1][j] - xx[1][i];					\
	dx[2] = xx[2][j] - xx[2][i];					\
	distSq = dx[0] * dx[0] + dx[1] * dx[1];				\
	zylinderradius = SQRT(distSq);					\
	distSq += dx[2] * dx[2];					\
									\
	if ((distSq>BinSq[0]) && (distSq<=BinSq[nbin])) {		\
	  low=0; up=nbin; cur=halfnbin;					\
	  while (low!=up) {						\
	    if (distSq> BinSq[cur]) {low=cur;} else {up=cur-1;} /* ( * ; * ]*/ \
	    cur=(up+low+1)/2;						\
	  }								\
									\
	  /* angles */							\
	  phidata = NEARBYINT(atan2(dx[1], dx[0]) - startphi);		\
	  while (phidata < 0) phidata += TWOPI;				\
	  while (phidata >= TWOPI) phidata -= TWOPI;			\
	  kphi = nbin * (int) (phidata * invsegphi);			\
									\
	  thetadata = NEARBYINT(atan2(dx[2], zylinderradius) - starttheta); \
	  while (thetadata < 0) thetadata += PI;			\
	  while (thetadata >= PI) thetadata -= PI;			\
	  ktheta = (int) (thetadata * invsegtheta);			\
									\
	  low += kphi + twoNphiNbin * ktheta;				\
	  assert(low>=0 && low<totalspatialbins);			\
									\
	  for (deltaT=0; deltaT<= nstepTstepT; deltaT+=stepT,		\
		 low+=totalspatialbins) {				\
	    jj = j + deltaT * spatial;					\
	    endT= totalpoints - deltaT * spatial;			\
	    for (t=0; t<endT; t+=spatial) {				\
	      for(int row = 0; row < vdim; row++) {			\
		for(int col = row; col < vdim; col++) {			\
		  for (rep = t; rep < totalpointsrepetvdim; rep += totalpointsvdim){ \
		    int curbin = 0;					\
		    /* inserts calculation from switch below */		\
		    curbin = low + totalbins * (vdim * row + col);	\
		    DO;							\
		    res[curbin] += x2;					\
		    res[curbin + totaln] += x2 * x2;			\
		    res[curbin + 2*totaln]++;				\
		  }							\
		}							\
	      }								\
	    } /* t*/							\
	  } /* deltat*/							\
	} /* if (distSq)*/						\
      } /* j */								\
    } /* i */
   
    // checks which method shoud be used (gets inserted above) 
    switch(method) {
    case(METHOD_PSEUDO):
      // pseudo variogram
      FORARB(double x2 = values[rep + jj + totalpoints * row]
	     - values[i + rep + totalpoints * col];
	     x2 *= x2);
      break;
    case(METHOD_CROSS):
      // cross variogram
      FORARB(double x2 = (values[rep + jj + totalpoints * row]
			  - values[i + rep + totalpoints * row])
	     * (values[rep + jj + totalpoints * col]
		- values[i + rep + totalpoints * col]));	  
      break;
    case(METHOD_COVARIANCE):
      FORARB(double x2 = (values[rep + jj + totalpoints * row]
			  * values[i + rep + totalpoints * col]);
	     sumhead[curbin] += values[rep + jj + totalpoints * row];
	     sumtail[curbin] += values[i + rep  + totalpoints * col]);
      break;
    case(METHOD_PSEUDOMADOGRAM):
      // pseudo madogram
      FORARB(double x2 = FABS(values[rep + jj + totalpoints * row]
			      - values[i + rep + totalpoints * col]));
      break;
    case(METHOD_CROSSMADOGRAM):
      // cross madogram
      FORARB(double x2 = FABS(values[rep + jj + totalpoints * row]
			      - values[i + rep + totalpoints * row])
	     * FABS(values[rep + jj + totalpoints * col]
		    - values[i + rep + totalpoints * col]); x2 = SQRT(x2));	  
      break;
    default:
      err = TOOLS_METHOD; goto ErrorHandling;
    }

    
  } /* arbitrary */

  if(method == METHOD_COVARIANCE) {
    for(int row = 0; row < vdim; row++) {
      for(int col = row; col < vdim; col++) {
	int curbin = totalbins * (vdim * row + col);
	for(i = 0; i < totalbins; i++) {
	  double n = res[i + curbin + 2*totaln];
	  double m0 = sumhead[i + curbin] / n;
	  double mh = sumtail[i + curbin] / n;	    
	  res[i + curbin] = res[i + curbin] / n - m0 * mh ;
	}
      }
    }
  } else if(method == METHOD_CROSS || method == METHOD_PSEUDO) {
    for(int row = 0; row < vdim; row++) {
      for(int col = row; col < vdim; col++) {
	int curbin = totalbins * (vdim * row + col);
	for(i=0; i < totalbins; i++) {
	  double n = res[i + curbin + 2*totaln];
	  res[i + curbin] /= (2.0 * n);
	  res[i + curbin + totaln] *= 0.25;
	}
      }
    }
  } else {
    for(int row = 0; row < vdim; row++) {
      for(int col = row; col < vdim; col++) {
	int curbin = totalbins * (vdim * row + col);
	for(i=0; i < totalbins; i++) {
	  double n = res[i + curbin + 2*totaln];
	  res[i + curbin] /= n;
	}
      }
    }
  }  

  for(int row = 1; row < vdim; row++) {
    for(int col = 0; col < row; col++) {
      int noncalcpos = totalbins * (vdim * row + col);
      int calcpos = totalbins * (row + vdim * col);
      res[noncalcpos] = RF_NAN;
      for(i = 1; i < totalbins; i++) {
	res[i + noncalcpos] = res[i + calcpos];
	res[i + noncalcpos + totaln] = res[i + calcpos + totaln];
	res[i + noncalcpos + 2*totaln] = res[i + calcpos + 2*totaln];
      }
    }
  }
  
  UNPROTECT(1); 
  FREE(BinSq);
  if(method == METHOD_COVARIANCE) {
    FREE(sumhead);
    FREE(sumtail);
  }
  return Res;

 ErrorHandling:
  FREE(BinSq);
  if(method == METHOD_COVARIANCE) {
    FREE(sumhead);
    FREE(sumtail);
  }
  UNPROTECT(1);
  PRINTF("Error: ");
  switch (err) {
  case TOOLS_MEMORYERROR :  
    PRINTF("Memory alloc failed in empiricalvariogram.\n"); break;
  case TOOLS_XERROR :  
    PRINTF("The x coordinate may not be NULL.\n"); break;
  case TOOLS_BIN_ERROR :
    PRINTF("Bin components not an increasing sequence.\n"); break;
  case TOOLS_METHOD:
    BUG; break;
  default : BUG;
  }
  return R_NilValue;  
} 
	 



back to top