swh:1:snp:8c443ee102a7f3a9fe0e834b4caec2f704d4d442
Tip revision: e10243fbd4eb0cbeaf518e67fbc5b8ad44889954 authored by Martin Schlather on 12 December 2019, 13:40:13 UTC
version 3.3.7
version 3.3.7
Tip revision: e10243f
Brown.cc
/*
Authors
Marco Oesting,
Martin Schlather, schlather@math.uni-mannheim.de
simulation of Brown-Resnick processes
Copyright (C) 2009 -- 2010 Martin Schlather
Copyright (C) 2011 -- 2017 Marco Oesting & 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 PURSE. 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 "questions.h"
#include "Processes.h"
#include "operator.h"
#include "rf_interfaces.h"
#include "variogramAndCo.h"
#define BR_MESHSIZE (LAST_MAXSTABLE + 1)
#define BR_VERTNUMBER (LAST_MAXSTABLE + 2)
#define BR_OPTIM (LAST_MAXSTABLE + 3)
#define BR_OPTIMTOL (LAST_MAXSTABLE + 4)
#define BR_LAMBDA (LAST_MAXSTABLE + 5)
#define BR_OPTIMAREA (LAST_MAXSTABLE + 6)
#define BR_VARIOBOUND (LAST_MAXSTABLE + 7)
int newmodel_covcpy(model **localcov, int modelnr, model *cov, //err
double *x, double *y, double *T,
int spatialdim, /* spatial dim only ! */
int xdimOZ, long lx, long ly, bool Time, bool grid,
bool distances) {
Types type = SYSTYPE(DefList[modelnr].systems[0], 0);
assert(type == InterfaceType && DefList[modelnr].variants == 1); // anderes z.zt nicht benutzt
int i, err;
assert(*localcov == NULL);
assert(cov->calling != NULL);
addModel(localcov, modelnr, NULL, true);
model *neu = *localcov;
neu->base = cov->base;
neu->root = neu;
// PMI(neu);
assert(neu->ownloc==NULL && neu->prevloc==NULL);
assert(type == InterfaceType); // braucht prevloc !!
neu->prevloc = LOCLIST_CREATE(1, xdimOZ + (int) Time); // locown
assert(cov->prevloc != NULL);
assert(PLoc(cov) == cov->prevloc);
assert(!((y==NULL) xor (ly == 0)));
loc_set(x, y, T, spatialdim, xdimOZ, lx, ly, Time, grid, distances, neu);
if ((err = covcpy(neu->sub + 0, cov)) != NOERROR) RETURN_ERR(err);
SET_CALLING(neu->sub[0], neu);
for (i=0; i<2; i++) {
if ((err = CHECK(neu, OWNLOGDIM(0), PREVXDIM(0),
// DefList[COVNR].t ype,Martin:changed 27.11.13
//co v->ty pus,// auch nicht 20.5.14
type,
type == InterfaceType ? XONLY : PREVDOM(0),
type == InterfaceType ? CARTESIAN_COORD : PREVISO(0),
cov->vdim, EvaluationType)) != NOERROR) {
RETURN_ERR(err);
}
if (i==0 && (err = STRUCT(neu, NULL)) != NOERROR) RETURN_ERR(err);
}
RETURN_NOERROR;
}
int newmodel_covcpy(model **localcov, int modelnr, model *cov) {//err
int err,
store = GLOBAL.general.set;
GLOBAL.general.set = 0;
location_type *loc = Loc(cov);
err = newmodel_covcpy(localcov, modelnr, cov,
loc->grid ? loc->xgr[0] : loc->x,
loc->grid ? loc->ygr[0] : loc->y,
loc->grid ? loc->xgr[0] + 3 * loc->spatialdim : loc->T,
loc->spatialdim, loc->xdimOZ,
loc->grid ? 3 : loc->totalpoints,
loc->ly == 0 ? 0 : loc->grid ? 3 : loc->totalpoints,
loc->Time, loc->grid, loc->distances);
GLOBAL.general.set = store;
RETURN_ERR(err);
}
// **********************************************************************
// Brown Resnick
int checkBrownResnickProc(model *cov) {
//NotProgrammedYet("at the moment Brown-Resnick processes");
model
*key = cov->key,
*sub = key != NULL ? key :
cov->sub[cov->sub[MPP_SHAPE] != NULL ? MPP_SHAPE: MPP_TCF];
int err;
isotropy_type isoprev;
Types type, frame;
ASSERT_ONESYSTEM;
ASSERT_CARTESIAN;
ASSERT_ONE_SUBMODEL(cov);
if ((err = SetGEVetc(cov)) != NOERROR) RETURN_ERR(err);
type = isProcess(sub) || isPointShape(sub)
? SYSTYPE(DEFSYS(sub), 0) : VariogramType;
frame = isVariogram(type) ? EvaluationType : BrMethodType;
isoprev = equalsVariogram(frame) ? SYMMETRIC : CARTESIAN_COORD;
if ((err = CHECK(sub, OWNLOGDIM(0), OWNXDIM(0),
type,//Martin: changed 27.11.13 DefList[COVNR].Type,
XONLY, isoprev, SCALAR, frame)) != NOERROR) {
RETURN_ERR(err);
}
setbackward(cov, sub);
RETURN_NOERROR;
}
int general_init(model* cov, int trendlen, gen_storage *s) {
model *key=cov->key;
br_storage *sBR = cov->Sbr;
int err,
total = Gettotalpoints(key),
dim = ANYDIM; // ownxdim(0)
bool keygrid = Getgrid(key);
key->simu.expected_number_simu = cov->simu.expected_number_simu;
if ((err = INIT(key, COVNR != BRNORMED, s)) != NOERROR) {
RETURN_ERR(err);
}
// PMI(cov);
assert(key->simu.active);
assert(sBR->trend == NULL);
sBR->trendlen = trendlen;
if ( (sBR->trend = (double**) CALLOC(trendlen, sizeof(double*))) == NULL)
RETURN_ERR(ERRORMEMORYALLOCATION);
for (int j=0; j < trendlen; j++) {
if ((sBR->trend[j] = (double*) MALLOC(total * sizeof(double)))==NULL)
RETURN_ERR(ERRORMEMORYALLOCATION);
}
// PMI(key);
double *x;
int len;
if (keygrid) {
x = Getxgr(key)[0];
len = 3;
} else {
x = Getx(key);
len = Gettotalpoints(key);
}
if ((err = loc_set(x, NULL, NULL, dim, dim, len, 0, false, keygrid,
DistancesGiven(key), sBR->vario)) > NOERROR)
RETURN_ERR(err);
// printf("A\n");
if (sBR->vario->sub[0] != NULL)
SetLoc2NewLoc(sBR->vario->sub[0], PLoc(sBR->vario));
// printf("AB\n");
cov->loggiven = wahr;
pgs_storage *pgs = cov->Spgs;
for (int d=0; d<dim; d++) {
pgs->supportmin[d] = RF_NEGINF; // 4 * for debugging...
pgs->supportmax[d] = RF_INF;
pgs->supportcentre[d] = RF_NA;
}
// printf("done\n");
RETURN_NOERROR;
}
#define ORIG_IDX 0
int init_BRorig(model *cov, gen_storage *s){
assert(cov->mpp.moments >= 1);
model *key = cov->key;
if (key == NULL) BUG;
int err,
dim = OWNXDIM(0);
br_storage *sBR = cov->Sbr;
pgs_storage *pgs = NULL;
//PMI(cov->calling->calling);
if ((err = alloc_cov(cov, dim, 1, 1)) != NOERROR) RETURN_ERR(err);
pgs = cov->Spgs;
if ((err = general_init(cov, 1, s)) != NOERROR) goto ErrorHandling;
Variogram(NULL, sBR->vario, sBR->trend[ORIG_IDX]);
assert(MODELNR(key) == GAUSSPROC);
assert(key->mpp.moments >= 1);
cov->mpp.mM[0] = cov->mpp.mMplus[0] = 1.0;
cov->mpp.mM[1] = cov->mpp.mMplus[1] = 1.0;
cov->mpp.maxheights[0] = EXP(GLOBAL.extreme.standardmax);
pgs->zhou_c = 1.0;
if ((err = ReturnOwnField(cov)) != NOERROR) // must be later than INIT !
goto ErrorHandling;
ErrorHandling:
if (err != NOERROR) br_DELETE(&(cov->Sbr), cov);
cov->simu.active = cov->initialised = err == NOERROR;
RETURN_ERR(err);
}
void do_BRorig(model *cov, gen_storage *s) {
model *key = cov->key;
assert(key != NULL);
br_storage *sBR = cov->Sbr;
assert(sBR != NULL && sBR->trend != NULL);
double *res = cov->rf,
*trend = sBR->trend[ORIG_IDX];
int
zeropos = sBR->zeropos;
long
totalpoints = Gettotalpoints(cov);
assert(totalpoints > 0);
assert(zeropos >= 0 && zeropos <totalpoints);
assert(cov->origrf);
DO(key, s); // nicht gatternr,
double *lgres = key->rf;
double lgreszeropos = (double) lgres[zeropos]; // wird durch DO veraendert!
for (int i=0; i<totalpoints; i++) {
res[i] = lgres[i] - lgreszeropos - trend[i];
//printf("res = %10g %10g\n", lgres[i], res[i]);
}
}
int init_BRshifted(model *cov, gen_storage *s) {
model *key=cov->key;
if (key == NULL) RETURN_NOERROR;
int err = NOERROR,
dim = ANYDIM;
bool keygrid = Getgrid(key);
long
keytotal = Gettotalpoints(key),
shiftedloclen = keygrid ? 3 : keytotal,
trendlenmax = (int) CEIL((double) GLOBAL.br.BRmaxmem / keytotal),
trendlenneeded = MIN(keytotal, cov->simu.expected_number_simu);
br_storage *sBR = cov->Sbr;
pgs_storage *pgs = NULL;
if ((err = alloc_cov(cov, dim, 1, 1)) != NOERROR) goto ErrorHandling;
pgs = cov->Spgs;
if ((err = general_init(cov, MIN(trendlenmax, trendlenneeded), s))
!= NOERROR) goto ErrorHandling;
assert(cov->mpp.moments >= 1);
assert(MODELNR(key) == GAUSSPROC);
assert(key->mpp.moments >= 1);
cov->mpp.mM[0] = cov->mpp.mMplus[0] = 1.0;
cov->mpp.mM[1] = cov->mpp.mMplus[1] = 1.0;
cov->mpp.maxheights[0] = EXP(GLOBAL.extreme.standardmax);
pgs->zhou_c = 1.0;
if ((sBR->shift.loc = (double*)
MALLOC(dim*shiftedloclen*sizeof(double))) == NULL ||
(sBR->shift.locindex = (int*) MALLOC(sizeof(int) * dim))==NULL
) {
err=ERRORMEMORYALLOCATION; goto ErrorHandling;
}
trendlenmax = (int) CEIL((double) GLOBAL.br.BRmaxmem / keytotal);
trendlenneeded = MIN(keytotal, cov->simu.expected_number_simu);
sBR->shift.memcounter = 0;
if ((sBR->shift.loc2mem=(int*) MALLOC(sizeof(int)*keytotal))==NULL) {
err = ERRORMEMORYALLOCATION; goto ErrorHandling;
}
for (long j=0; j<keytotal; j++) sBR->shift.loc2mem[j] = UNSET;
if ((sBR->shift.mem2loc=(int*) MALLOC(sizeof(int)*sBR->trendlen))==NULL) {
err = ERRORMEMORYALLOCATION; goto ErrorHandling;
}
for (long j=0; j < sBR->trendlen; j++) sBR->shift.mem2loc[j] = UNSET;
err = ReturnOwnField(cov);
ErrorHandling:
if (err != NOERROR) br_DELETE(&(cov->Sbr), cov);
cov->simu.active = cov->initialised = err == NOERROR;
RETURN_ERR(err);
}
void indextrafo(long onedimindex, double ** xgr, int dim, int *multidimindex) {
int d;
for (d=0; d<dim; d++) {
multidimindex[d] = onedimindex % (int) xgr[d][XLENGTH];
onedimindex = onedimindex / (int) xgr[d][XLENGTH];
}
}
void do_BRshifted(model *cov, gen_storage *s) {
br_storage *sBR = cov->Sbr;
model *key = cov->key;
assert(cov->key != NULL);
location_type *keyloc = Loc(key);
long i, k, zeropos, zeroposMdim,
keytotal = Gettotalpoints(key);
int d, trendindex,
dim = ANYDIM,
trendlen = sBR->trendlen,
*locindex = sBR->shift.locindex,
*mem2loc = sBR->shift.mem2loc,
*loc2mem = sBR->shift.loc2mem;
bool keygrid = Getgrid(key);
double *shiftedloc = sBR->shift.loc,
**xgr = keyloc->xgr,
**trend = sBR->trend,
*res = cov->rf,
*lgres = cov->key->rf;
assert(cov->origrf);
DO(key, s);
zeropos = (long) FLOOR(UNIFORM_RANDOM * keytotal);
if (loc2mem[zeropos] != UNSET) {
trendindex = loc2mem[zeropos];
if (mem2loc[trendindex] != zeropos) BUG;
} else {
if (sBR->shift.memcounter<trendlen) {
trendindex = sBR->shift.memcounter;
sBR->shift.memcounter++;
} else {
trendindex = trendlen - 1;
loc2mem[mem2loc[trendlen-1]] = UNSET;
mem2loc[trendlen-1] = UNSET;
}
if (keygrid) {
indextrafo(zeropos, keyloc->xgr, dim, locindex); // to do: ersetzen
for (d=0; d<dim; d++) {
shiftedloc[3*d+XSTART] = -locindex[d] * xgr[d][XSTEP];
shiftedloc[3*d+XLENGTH] = xgr[d][XLENGTH];
shiftedloc[3*d+XSTEP] = xgr[d][XSTEP];
}
} else {
zeroposMdim = zeropos*dim;
for (k=i=0; i<keytotal; i++) {
for (d=0; d<dim; d++, k++) {
shiftedloc[k] = keyloc->x[k] - keyloc->x[zeroposMdim+d];
}
}
}
partial_loc_set(Loc(sBR->vario), shiftedloc, NULL,
keygrid ? 3: keytotal, 0, keyloc->distances,
dim, NULL, keygrid, true);
if (sBR->vario->sub[0] != NULL)
SetLoc2NewLoc(sBR->vario->sub[0], PLoc(sBR->vario));
Variogram(NULL, sBR->vario, sBR->trend[trendindex]);
mem2loc[trendindex] = zeropos; // todo ? long instead of int
loc2mem[zeropos] = trendindex;
}
for (i=0; i<keytotal; i++) {
res[i] = lgres[i] - lgres[zeropos] - trend[trendindex][i];
}
}
int check_BRmixed(model *cov) {
//NotProgrammedYet("at the moment Brown-Resnick processes");
int err;
br_param *bp = &(GLOBAL.br);
if (!cov->logspeed) SERR("BrownResnick requires a variogram model as submodel that tends to infinity [t rate of at least 4log(h) for being compatible with BRmixed");
kdefault(cov, BR_MESHSIZE, bp->BRmeshsize);
kdefault(cov, BR_VERTNUMBER, bp->BRvertnumber);
kdefault(cov, BR_OPTIM, bp->BRoptim);
kdefault(cov, BR_OPTIMTOL, bp->BRoptimtol);
kdefault(cov, BR_VARIOBOUND, bp->variobound);
if (COVNR == BRMIXED_USER && cov->key == NULL && P0INT(BR_OPTIM) > 0) {
if (!PisNULL(BR_LAMBDA)) {
if (PisNULL(BR_OPTIMAREA)) SERR1("'%.50s' not given", KNAME(BR_OPTIMAREA));
if (PL > 0) { PRINTF("'%.50s' set to '0'", KNAME(BR_OPTIM));}
PINT(BR_OPTIM)[0] = 0;
} else if (P0INT(BR_OPTIM) == 2 && !PisNULL(BR_OPTIMAREA))
if (PL > 0) {PRINTF("'%.50s' set to '1'", KNAME(BR_OPTIM));}
}
if (cov->key != NULL && P0INT(BR_OPTIM) == 2) {
if (!isIsotropic(SYSOF(cov->key))) {
// PMI(cov->key);
SERR("area optimisation implemented for the isotropic case only"); //@MARTIN: das scheint nicht zu funktionieren, wenn ich ein Variogramm eingebe
}
}
kdefault(cov, BR_LAMBDA, RF_NA);
if (PisNULL(BR_OPTIMAREA)) kdefault(cov, BR_OPTIMAREA, 0.0);
if ((err = checkBrownResnickProc(cov)) != NOERROR) RETURN_ERR(err);
if ((err = checkkappas(cov, true)) != NOERROR) RETURN_ERR(err);
if (VDIM0 != 1) SERR("BR only works in the univariate case");
RETURN_NOERROR;
}
void kappaBRmixed(int i, model VARIABLE_IS_NOT_USED *cov, int *nr, int *nc){
// i nummer des parameters
switch(i) {
case GEV_XI: case GEV_MU: case GEV_S: case BR_MESHSIZE:
case BR_VERTNUMBER: case BR_OPTIM: case BR_OPTIMTOL:
case BR_LAMBDA: case BR_VARIOBOUND:
*nr = 1;
*nc = 1;
break;
case BR_OPTIMAREA:
*nr = 1;
*nc = SIZE_NOT_DETERMINED;
break;
default:
*nr = *nc = OUT_OF_RANGE;
}
}
double IdxDistance(int maxind, int zeropos, double **xgr, int dim) {
int d,
delta = 0,
x = maxind,
y = zeropos;
for (d=0; d<dim; d++) {
delta += ( (x % (int) xgr[d][XLENGTH]) - (y % (int) xgr[d][XLENGTH]))* ( (x % (int) xgr[d][XLENGTH]) - (y % (int) xgr[d][XLENGTH]));
x /= xgr[d][XLENGTH];
y /= xgr[d][XLENGTH];
}
return SQRT(delta);
}
void OptimArea(model *cov) {
// side effect auf P(BR_OPTIMAREA) !!
br_storage *sBR = cov->Sbr;
pgs_storage *pgs = cov->Spgs;
model *key = sBR->m3.sub[0];
location_type *keyloc = Loc(key);
double **xgr = keyloc->xgr,
*optimarea, dummy, Error, lambda,
maxErrorbound, Errorbound, Errorboundtmp,
Errortol = P0(BR_OPTIMTOL),
step = P0(BR_MESHSIZE),
invstepdim, **am=NULL;
int d, j, k, cellnumber,
*cellcounter, idxdist,
n_zhou_c = pgs->n_zhou_c,
vertnumber = P0INT(BR_VERTNUMBER),
minradius = (int) (sBR->m3.minradius/step),
**countvector = sBR->m3.countvector,
zeropos = sBR->zeropos,
dim = ANYDIM,
keytotal = Gettotalpoints(key);
if (minradius==0) return;
if ((cellcounter = (int*) MALLOC((minradius+1) * sizeof(int))) == NULL)
goto ErrorHandling;
for (k=0; k<=minradius; k++) cellcounter[k]=0;
if ((am = (double**) CALLOC(vertnumber, sizeof(double*)))==NULL)
goto ErrorHandling;
for (j=0; j<vertnumber; j++) {
if ((am[j] = (double*) MALLOC((minradius+1) * sizeof(double))) == NULL)
goto ErrorHandling;
}
for (k=0; k<keytotal; k++) {
idxdist = (int) CEIL(IdxDistance(k, zeropos, xgr, dim));
if (idxdist<=minradius) (cellcounter[idxdist])++;
}
assert(cellcounter[0]==1)
invstepdim = intpow(step, -dim);
lambda = 0.0;
for (j=0; j<vertnumber; j++) lambda += countvector[j][0]*invstepdim / n_zhou_c;
for (d=0; d<=minradius; d++) { //monotonic in each column
for (j=vertnumber-1; j>=0; j--) {
am[j][d] = countvector[j][d] * invstepdim / (n_zhou_c * cellcounter[d]);
for (k=j+1; k<vertnumber; k++) {
if (am[j][d] < am[k][d]) {
dummy = am[j][d];
am[j][d] = am[k][d];
am[k][d] = dummy;
}
}
am[j][d] = FMIN(am[j][d], lambda/vertnumber); //cutoff
}
}
for (j=0; j<vertnumber; j++) { //monotonic in each row
for (d=0; d<=minradius; d++) {
for (k=d+1; k<=minradius; k++) {
if (am[j][d] < am[j][k]) {
dummy = am[j][d];
am[j][d] = am[j][k];
am[j][k] = dummy;
}
}
}
}
maxErrorbound = 0.0;
for (d=0; d<=minradius; d++) { //areamatrix => Error matrix
for (j=0; j<vertnumber; j++) {
am[j][d] = lambda/vertnumber - am[j][d];
maxErrorbound = FMAX(maxErrorbound, am[j][d]);
}
}
cellnumber = 0;
Error = 0.0;
Errorboundtmp = Errorbound = 0.0;
while (Errorboundtmp < maxErrorbound && Error < Errortol) {
Errorbound = Errorboundtmp;
Errorboundtmp = maxErrorbound;
for (d=0; d<=minradius; d++) {
for (j=0; j<vertnumber; j++) {
if (am[j][d] > Errorbound) {
Errorboundtmp = FMIN(Errorboundtmp, am[j][d]);
}
}
}
Error = 0.0;
for (d=0; d<=minradius; d++) {
for (j=0; j<vertnumber; j++) {
if (am[j][d] <= Errorboundtmp + 1e-6) {
Error += (double) cellcounter[d] * am[j][d];
cellnumber += cellcounter[d];
}
}
}
Error = Error / (cellnumber*lambda/vertnumber);
}
PFREE(BR_OPTIMAREA);
PALLOC(BR_OPTIMAREA, 1, minradius+1);
optimarea = P(BR_OPTIMAREA);
for (d=0; d<=minradius; d++) {
optimarea[d] = 0.0;
for (j=0; j<vertnumber; j++) {
if (am[j][d] <= Errorbound + 1e-6) {
optimarea[d] += 1.0/vertnumber;
}
}
//printf("%10g ", optimarea[d]);
}
//printf("\n");
ErrorHandling:
if (cellcounter != NULL) FREE(cellcounter);
if (am != NULL) {
for (j=0; j<vertnumber; j++) {
if (am[j] != 0) FREE(am[j]);
}
FREE(am);
}
return;
}
void set_lowerbounds(model *cov) {
br_storage *sBR = cov->Sbr;
assert(sBR != NULL);
assert(sBR->m3.sub[0] != NULL);
double step = P0(BR_MESHSIZE),
*optimarea = P(BR_OPTIMAREA);
int j, k,
dim = ANYDIM,
minradius = (int) (sBR->m3.minradius / step);
model *key = sBR->m3.sub[0];
location_type *keyloc = Loc(key);
double **xgr = keyloc->xgr;
long keytotal = Gettotalpoints(key);
for (j=0; j<keytotal; j++) {
sBR->m3.lowerbounds[j] = RF_INF;
k = (int) CEIL(IdxDistance(j, sBR->zeropos, xgr, dim));
if (k <= minradius && optimarea[k]>1e-5) {
sBR->m3.lowerbounds[j] = -LOG(optimarea[k]);
}
//printf("%10g ", sBR->m3.lowerbounds[j]);
}
//printf("\n");
}
int prepareBRoptim(model *cov) {
br_storage *sBR = cov->Sbr;
model *key = sBR->m3.sub[0];
location_type *keyloc = Loc(key);
double step = P0(BR_MESHSIZE),
**xgr = keyloc->xgr;
int i, j, d,
vertnumber = P0INT(BR_VERTNUMBER),
dim = ANYDIM,
maxradius = 1,
minradius = (int) (sBR->m3.minradius/step);
for (d=0; d<dim; d++) {
maxradius *= (int) FLOOR(xgr[d][XLENGTH] / 2.0) + 1;
}
switch(P0INT(BR_OPTIM)) {
case 0:
if (ISNAN(P0(BR_LAMBDA))) P(BR_LAMBDA)[0] = 1.0;
break;
case 1: // nothing to do
break;
case 2:
sBR->m3.vertnumber = vertnumber;
if (sBR->m3.countvector != NULL || sBR->m3.areamatrix != NULL) BUG;
if ((sBR->m3.countvector = (int**) CALLOC(vertnumber, sizeof(int*)))==NULL ||
(sBR->m3.logvertnumber = (double *) MALLOC(vertnumber * sizeof(double)))
== NULL)
RETURN_ERR(ERRORMEMORYALLOCATION);
for (j=0; j<vertnumber; j++) {
if ((sBR->m3.countvector[j] = (int*) CALLOC(minradius + 1, sizeof(int)))
== NULL)
RETURN_ERR(ERRORMEMORYALLOCATION);
for (i=0; i<=minradius; i++) sBR->m3.countvector[j][i] = 0;
}
for (j=0; j<vertnumber; j++)
sBR->m3.logvertnumber[j] = - LOG((double) (j+1)/vertnumber);
break;
default:
SERR("optimization might not be used here\n");
}
if ((sBR->m3.areamatrix = (double *) MALLOC((minradius + 1)* sizeof(double)))
== NULL) {
RETURN_ERR(ERRORMEMORYALLOCATION);
}
sBR->m3.areamatrix[0] = 1.0;
if (minradius > 0) {
for (i=1; i<=minradius; i++) {
if (i <= cov->ncol[BR_OPTIMAREA]) {
sBR->m3.areamatrix[i] = P(BR_OPTIMAREA)[i-1];
}
else sBR->m3.areamatrix[i] = 0.0;
}
}
PFREE(BR_OPTIMAREA);
PALLOC(BR_OPTIMAREA, 1, minradius + 1);
double *optimarea = P(BR_OPTIMAREA);
for (i=0; i<=minradius; i++) {
optimarea[i] = sBR->m3.areamatrix[i];
}
set_lowerbounds(cov);
if (PL >= PL_STRUCTURE) {PRINTF("BR optimisation finished...\n");}
RETURN_NOERROR;
}
int init_BRmixed(model *cov, gen_storage *s) {
location_type *loc = Loc(cov);
br_storage *sBR = cov->Sbr;
assert(sBR != NULL);
assert(sBR->m3.sub[0] != NULL);
model *key = sBR->m3.sub[0];
location_type *keyloc = Loc(key);
int d, err = NOERROR,
dim = ANYDIM,
bytes = sizeof(double) * dim,
keytotal = Gettotalpoints(key);
double area = 1.0,
step = P0(BR_MESHSIZE);
pgs_storage *pgs = NULL;
assert(isPointShape(cov));
assert(dim > 0);
if ((err = alloc_cov(cov, dim, 1, 1)) != NOERROR) goto ErrorHandling;//nur pgs
pgs = cov->Spgs; // nach alloc_cov !!
if ((err = general_init(cov, 1, s)) != NOERROR) goto ErrorHandling;
Variogram(NULL, sBR->vario, sBR->trend[0]);
if ((sBR->m3.suppmin = (double*) MALLOC(bytes))==NULL ||
(sBR->m3.suppmax = (double*) MALLOC(bytes))==NULL ||
(sBR->m3.loccentre = (double*) MALLOC(bytes))==NULL
// (sBR->shift.locindex = (int*) MALLOC(sizeof(int) * dim))==NULL ||
) {
err = ERRORMEMORYALLOCATION; goto ErrorHandling;
}
assert(Getcaniso(cov) == NULL);
GetDiameter(loc, sBR->m3.suppmin, sBR->m3.suppmax, sBR->m3.loccentre);
assert(pgs->supportmin != NULL);
for (d=0; d<dim; d++) {
pgs->own_grid_cumsum[d] =
d==0 ? 1 : pgs->own_grid_cumsum[d-1] * pgs->own_grid_len[d-1];
sBR->m3.suppmin[d] = FLOOR((sBR->m3.suppmin[d] - sBR->m3.radius -
sBR->m3.minradius) / step) * step - step / 2;
sBR->m3.suppmax[d] = CEIL((sBR->m3.suppmax[d] + sBR->m3.radius +
sBR->m3.minradius) / step) * step + step / 2;
area *= sBR->m3.suppmax[d] - sBR->m3.suppmin[d];
pgs->own_grid_start[d] = RF_NEGINF;
pgs->own_grid_step[d] = keyloc->xgr[d][XSTEP];
pgs->own_grid_len[d] = keyloc->xgr[d][XLENGTH];
}
for (d=0; d<=cov->mpp.moments; d++) {
cov->mpp.mM[d] = cov->mpp.mMplus[d] = 1.0;
}
cov->mpp.maxheights[0] = EXP(0.0);
assert(MODELNR(key) == GAUSSPROC);
assert(cov->mpp.moments >= 1);
assert(Getgrid(key));
assert(key->mpp.moments >= 1);
key->mpp.mM[0] = key->mpp.mMplus[0] = 1.0;
key->mpp.mM[1] = key->mpp.mMplus[1] = 1.0;
if ((sBR->m3.lowerbounds = (double*) MALLOC(keytotal*sizeof(double))) == NULL) {
err=ERRORMEMORYALLOCATION;
goto ErrorHandling;
}
if ((err = prepareBRoptim(cov)) != NOERROR) {
goto ErrorHandling;
}
assert(keyloc != NULL);
key->simu.active = true;
set_lowerbounds(cov);
cov->rf = sBR->m3.sub[0]->rf;
cov->origrf = false;
cov->fieldreturn = Huetchenownsize;
pgs->estimated_zhou_c = (bool) ISNAN(P0(BR_LAMBDA));
pgs->zhou_c = pgs->estimated_zhou_c ? 0.0 : P0(BR_LAMBDA)*area; //@MARTIN: area kann weg, falls in logdens
pgs->logmean = false;
pgs->sq_zhou_c = pgs->sum_zhou_c = 0.0;
pgs->n_zhou_c = 0;
sBR->m3.next_am_check = GLOBAL.br.deltaAM;
ErrorHandling:
if (err != NOERROR) br_DELETE(&(cov->Sbr), cov);
cov->simu.active = cov->initialised = err == NOERROR;
RETURN_ERR(err);
}
void do_BRmixed(model *cov, gen_storage *s) {
// to do: improve simulation speed by dynamic sizes
assert(cov->key!=NULL);
br_storage *sBR = cov->Sbr;
model *key = sBR->m3.sub[0];
assert(cov->rf == key->rf);
location_type *keyloc = Loc(key);
assert(Getgrid(key));
pgs_storage *pgs = cov->Spgs;
assert(pgs != NULL);
int d,
i, j, maxind, idxdist,
dim = ANYDIM,
hatnumber=0,
lgtotalpoints = Gettotalpoints(key),
zeropos = sBR->zeropos,
vertnumber = P0INT(BR_VERTNUMBER);
double step = P0(BR_MESHSIZE),
invstepdim = intpow(step, -dim),
uplusmaxval , maxval, u[MAXMPPDIM],
** xgr = keyloc->xgr,
*lowerbounds = sBR->m3.lowerbounds,
area=1.0, *lgres = key->rf, //@MARTIN: area obsolet, falls in logdens
*trend = sBR->trend[0];
int minradius = (int) (sBR->m3.minradius/step);
if (P0INT(BR_OPTIM) == 2 && pgs->n_zhou_c >= sBR->m3.next_am_check) {
sBR->m3.next_am_check += GLOBAL.br.deltaAM;
OptimArea(cov);
set_lowerbounds(cov);
}
for (d=0; d<dim; d++) {
u[d] = ROUND((UNIFORM_RANDOM*(sBR->m3.suppmax[d] - sBR->m3.suppmin[d]) +
sBR->m3.suppmin[d]) / step) * step;
area *= sBR->m3.suppmax[d] - sBR->m3.suppmin[d];
pgs->supportmin[d] = u[d] - sBR->m3.minradius - sBR->m3.radius;
pgs->supportmax[d] = u[d] + sBR->m3.minradius + sBR->m3.radius;
pgs->supportcentre[d] = u[d];
pgs->own_grid_start[d] = keyloc->xgr[d][XSTART] + u[d];
}
while(true) {
DO(key, s);
hatnumber++;
maxval = RF_NEGINF;
maxind = 0;
for (i=0; i<lgtotalpoints; i++) {
lgres[i] -= trend[i];
if (lgres[i] > maxval) {
maxval = lgres[i];
maxind = i;
}
}
if (maxind == zeropos) {
pgs->sq_zhou_c += area * invstepdim * area * invstepdim; // @MARTIN: s.o
pgs->sum_zhou_c += area * invstepdim; //
}
uplusmaxval = maxval - lgres[zeropos] - LOG(UNIFORM_RANDOM);
if (P0INT(BR_OPTIM) == 2) {
for (j=0; j<vertnumber; j++) {
if (uplusmaxval > sBR->m3.logvertnumber[j]) {
idxdist = (int) CEIL(IdxDistance(maxind, zeropos, xgr, dim));
if (idxdist<=minradius) (sBR->m3.countvector[j][idxdist])++;
break;
}
}
}
if (uplusmaxval > lowerbounds[maxind]) break;
}
pgs->n_zhou_c += hatnumber;
if (PL >= PL_STRUCTURE && hatnumber > 300) {
PRINTF("note: large hat number (%d) might indicate numerically suboptimal framework\n",
hatnumber);
}
//shifting maximum to origin is not necessary because of stationarity
//(conditional on T=t, the "correct" shape function is shifted which yields
//the same stationary max-stable process; OK because there is a finite number
//of values for t!)
for (i=0; i<lgtotalpoints; i++) {
lgres[i] -= maxval;
}
return;
}
void range_BRmixed(model *cov, range_type *range) {
range_mpp(cov, range);
range->min[BR_MESHSIZE] = 0;
range->max[BR_MESHSIZE] = RF_INF;
range->pmin[BR_MESHSIZE] = 0;
range->pmax[BR_MESHSIZE] = RF_INF;
range->openmin[BR_MESHSIZE] = true;
range->openmax[BR_MESHSIZE] = true;
range->min[BR_VERTNUMBER] = 1;
range->max[BR_VERTNUMBER] = RF_INF;
range->pmin[BR_VERTNUMBER] = 1;
range->pmax[BR_VERTNUMBER] = 50;
range->openmin[BR_VERTNUMBER] = false;
range->openmax[BR_VERTNUMBER] = false;
range->min[BR_OPTIM] = 0;
range->max[BR_OPTIM] = 2;
range->pmin[BR_OPTIM] = 0;
range->pmax[BR_OPTIM] = 2;
range->openmin[BR_OPTIM] = false;
range->openmax[BR_OPTIM] = false;
range->min[BR_OPTIMTOL] = 0;
range->max[BR_OPTIMTOL] = 1;
range->pmin[BR_OPTIMTOL] = 0;
range->pmax[BR_OPTIMTOL] = 0.1;
range->openmin[BR_OPTIMTOL] = true;
range->openmax[BR_OPTIMTOL] = true;
range->min[BR_LAMBDA] = 0;
range->max[BR_LAMBDA] = RF_INF;
range->pmin[BR_LAMBDA] = 0;
range->pmax[BR_LAMBDA] = RF_INF;
range->openmin[BR_LAMBDA] = true;
range->openmax[BR_LAMBDA] = true;
range->min[BR_OPTIMAREA] = 0;
range->max[BR_OPTIMAREA] = 1;
range->pmin[BR_OPTIMAREA] = 0;
range->pmax[BR_OPTIMAREA] = 1;
range->openmin[BR_OPTIMAREA] = false;
range->openmax[BR_OPTIMAREA] = false;
range->min[BR_VARIOBOUND] = 0;
range->max[BR_VARIOBOUND] = RF_INF;
range->pmin[BR_VARIOBOUND] = 2;
range->pmax[BR_VARIOBOUND] = 25;
range->openmin[BR_VARIOBOUND] = false;
range->openmax[BR_VARIOBOUND] = true;
}
int structBRuser(model *cov, model **newmodel) {
location_type *loc = Loc(cov);
model *sub = cov->sub[cov->sub[MPP_SHAPE] != NULL ? MPP_SHAPE: MPP_TCF];
int i, d, err, model_intern,
dim = ANYDIMOF(sub),
newxlen;
bool grid;
double centreloc[MAXMPPDIM], minloc[MAXMPPDIM], maxloc[MAXMPPDIM],
*newx= NULL,
**xgr = loc->xgr;
ASSERT_NEWMODEL_NULL;
assert(isBrMethod(cov));
if (loc->Time || (Getgrid(cov) && loc->caniso != NULL)) {
TransformLoc(cov, false, GRIDEXPAND_AVOID, false); // changes loc !
SetLoc2NewLoc(sub, PLoc(cov));
}
loc = Loc(cov);
grid = Getgrid(cov);
model_intern = (COVNR == BRORIGINAL_USER) ? BRORIGINAL_INTERN
: (COVNR == BRMIXED_USER) ? BRMIXED_INTERN
: (COVNR == BRSHIFTED_USER) ? BRSHIFTED_INTERN
: BRORIGINAL_USER;
if (cov->key != NULL) COV_DELETE(&(cov->key), cov);// should be ok
ONCE_NEW_STORAGE(gen);
assert(Getcaniso(cov) == NULL);
GetDiameter(loc, minloc, maxloc, centreloc);
int totalpoints = loc->totalpoints;
newxlen = grid ? 3 : totalpoints;
if ((newx = (double*) MALLOC(dim * newxlen * sizeof(double))) == NULL) {
GERR("Memory allocation failed.\n");
}
if (grid) {
for (d=0; d<dim; d++) {
newx[3 * d + XSTART] = xgr[d][XSTART] - centreloc[d]
+ (((int) xgr[d][XLENGTH]) % 2 == 0) * xgr[d][XSTEP]/2;
newx[3 * d + XSTEP] = xgr[d][XSTEP];
newx[3 * d + XLENGTH] = xgr[d][XLENGTH];
}
} else {
for (i=0; i<totalpoints; i++)
for(d=0; d<dim; d++) newx[i*dim+d] = loc->x[i*dim+d] - centreloc[d];
}
if ((err = loc_set(newx, NULL, dim, dim, newxlen, false, grid, // loc->grid
loc->distances, cov)) > NOERROR)
goto ErrorHandling;
SetLoc2NewLoc(sub, PLoc(cov));
if ((err=covcpy(&(cov->key), sub)) > NOERROR) goto ErrorHandling;
if (cov->sub[MPP_TCF] != NULL) {
if ((err = STRUCT(sub, &(cov->key))) > NOERROR) goto ErrorHandling;
assert(cov->key->calling == cov);
}
addModel(&(cov->key), model_intern);
assert(cov->key->calling == cov);
kdefault(cov->key, GEV_XI, P0(GEV_XI));
kdefault(cov->key, GEV_MU, P0(GEV_MU));
kdefault(cov->key, GEV_S, P0(GEV_S));
if (COVNR == BRMIXED_USER) {
kdefault(cov->key, BR_MESHSIZE, P0(BR_MESHSIZE));
kdefault(cov->key, BR_VERTNUMBER, P0INT(BR_VERTNUMBER));
kdefault(cov->key, BR_OPTIM, P0INT(BR_OPTIM));
kdefault(cov->key, BR_OPTIMTOL, P0(BR_OPTIMTOL));
kdefault(cov->key, BR_VARIOBOUND, P0(BR_VARIOBOUND));
kdefault(cov->key, BR_LAMBDA, P0(BR_LAMBDA));
if (!PisNULL(BR_OPTIMAREA)) {
PARAMALLOC(cov->key, BR_OPTIMAREA, cov->nrow[BR_OPTIMAREA],
cov->ncol[BR_OPTIMAREA]);
PCOPY(cov->key, cov, BR_OPTIMAREA);
}
}
if ((err = CHECK(cov->key, OWNLOGDIM(0), OWNXDIM(0), PointShapeType,
OWNDOM(0), OWNISO(0), 1, BrMethodType)) == NOERROR) {
if ((err = STRUCT(cov->key, NULL)) <= NOERROR) {
err = CHECK(cov->key, OWNLOGDIM(0), OWNXDIM(0),
PointShapeType, OWNDOM(0), OWNISO(0), 1, BrMethodType);
}
}
ErrorHandling:
FREE(newx);
RETURN_ERR(err);
}
// new
int structBRintern(model *cov, model **newmodel) {
model *sub = cov->sub[cov->sub[MPP_SHAPE] != NULL ? MPP_SHAPE: MPP_TCF];
location_type *loc = Loc(cov);
int i, d, j, err,
dim = ANYDIM, // sub->ts dim,
totaldim = Gettotalpoints(cov) * dim,
zeropos = 0, // default, mostly overwritten
newxlen = 3; // default (grid)
bool grid = Getgrid(cov);
double step, mindist, dist,
*newx = NULL,
**xgr = loc->xgr;
br_storage *sBR = NULL;
model *submodel = NULL;
ASSERT_NEWMODEL_NULL;
assert(isPointShape(cov));
if (cov->key != NULL) COV_DELETE(&(cov->key), cov);// should be ok
ONCE_NEW_STORAGE(gen);
NEW_STORAGE_WITH_SAVE(br);
sBR = cov->Sbr;
sBR->nr = COVNR;
if (cov->sub[MPP_TCF] != NULL) {
if ((err = STRUCT(sub, &(cov->key))) > NOERROR) goto ErrorHandling;
assert(cov->key->calling == cov);
// SET_CALLING(cov->key, cov);
} else if ((err=covcpy(&(cov->key), sub)) > NOERROR) goto ErrorHandling;
if ((err = CHECK(cov->key,OWNLOGDIM(0), OWNXDIM(0), VariogramType, OWNDOM(0),
SYMMETRIC, 1, EvaluationType)) != NOERROR)
goto ErrorHandling;
if ((err = covcpy(&submodel, cov->key)) != NOERROR)
goto ErrorHandling;
if ((err = CHECK(submodel, 1, 1, VariogramType, XONLY,
ISOTROPIC, 1, EvaluationType)) != NOERROR)
goto ErrorHandling;
if ((err = newmodel_covcpy(&(sBR->vario), VARIOGRAM_CALL, cov->key))!=NOERROR)
goto ErrorHandling;
if ((err = alloc_cov(sBR->vario, dim, 1, 1)) != NOERROR) goto ErrorHandling;
addModel(&(cov->key), GAUSSPROC, cov);
assert(cov->key->ownloc == NULL);
if (COVNR == BRORIGINAL_INTERN) {
if (!grid) {
int totalpoints = loc->totalpoints;
zeropos = totalpoints;
for (i=0; i<totalpoints; i++) {
double norm = 0.0;
for (d=0; d<dim; d++) {
double dummy = loc->x[i*dim+d];
norm += dummy * dummy;
}
if (norm < 1e-8) zeropos = i;
}
int newtotpoints = totalpoints + (zeropos == totalpoints);
if ((newx = (double*) MALLOC(dim*newtotpoints*sizeof(double))) == NULL) {
err=ERRORMEMORYALLOCATION; goto ErrorHandling;
}
for(d=0; d<dim; d++) {
for (i=0; i<totalpoints; i++) newx[i*dim+d] = loc->x[i*dim+d];
if (zeropos == totalpoints) newx[totalpoints*dim+d] = 0.0;
}
}
} else if (COVNR == BRMIXED_INTERN) {
step = P0(BR_MESHSIZE);
mindist = RF_INF;
if (grid) {
for (d=0; d<dim; d++)
if (xgr[d][XSTEP] < mindist)
mindist = xgr[d][XSTEP];
} else {
int maxtotal = totaldim;
if (maxtotal > 1000) {
warning("minimal distance is only estimated");
maxtotal = 1000; // to do
}
for (i=0; i<maxtotal; i+=dim)
for (j=i+dim; j<maxtotal; )
for (d=0; d<dim; d++, j++) {
dist = FABS(loc->x[i+d] - loc->x[j]);
if (dist > 1e-15 && dist < mindist) mindist = dist;
}
grid = true;
}
newxlen = 3;
if (mindist < step) {
PRINTF("Warning! meshsize is larger than resolution of simulation! meshsize is automatically decreased to %10g.\n", mindist);
P(BR_MESHSIZE)[0] = step = mindist;
}
if (!PisNULL(BR_OPTIMAREA)) {
sBR->m3.minradius = cov->ncol[BR_OPTIMAREA] * step;
} else {
sBR->m3.minradius = 0;
}
double yy, C0, gammamin,
alpha,
xx=step * 1e-6;
COV(ZERO(submodel), submodel, &C0);
COV(&xx, submodel, &yy);
alpha = LOG(C0 - yy) / LOG(xx);
if (alpha > 2.0) alpha = 2.0;
gammamin = 4.0 - 1.5 * alpha;
INVERSE(&gammamin, submodel, &xx);
xx = CEIL(xx/step) * step;
sBR->m3.minradius = FMAX(sBR->m3.minradius, xx);
yy = P0(BR_VARIOBOUND);
INVERSE(&yy, submodel, &(sBR->m3.radius));
sBR->m3.radius = CEIL(sBR->m3.radius / step) * step;
if ((newx = (double*) MALLOC(newxlen*dim*sizeof(double))) == NULL) {
err = ERRORMEMORYALLOCATION; goto ErrorHandling;
}
if ((err = covcpy(sBR->m3.sub, cov->key)) != NOERROR) goto ErrorHandling;
for (d=0; d<dim; d++) {
newx[3*d+XSTART] = -sBR->m3.radius - sBR->m3.minradius;
newx[3*d+XLENGTH] =
2 * ((int) ROUND((sBR->m3.radius + sBR->m3.minradius) / step)) + 1;
newx[3*d+XSTEP] = step;
}
err = loc_set(newx, NULL, dim, dim, newxlen, false, grid,
false, sBR->m3.sub[0]);
double **subxgr = Loc(sBR->m3.sub[0])->xgr;
zeropos = 0;
for (d = dim; d > 0; d--) {
double len = subxgr[d-1][XLENGTH];
zeropos = zeropos * (int) len + (int) CEIL(len * 0.5) - 1;
}
sBR->zeropos = zeropos;
if ((err = CHECK(sBR->m3.sub[0], OWNLOGDIM(0), OWNXDIM(0), ProcessType,
OWNDOM(0), OWNISO(0), 1, GaussMethodType)) == NOERROR) {
if ((err = STRUCT(sBR->m3.sub[0], NULL)) <= NOERROR) {
err = CHECK(sBR->m3.sub[0], OWNLOGDIM(0), OWNXDIM(0), ProcessType,
OWNDOM(0), OWNISO(0), 1, GaussMethodType);
}
}
if (err > NOERROR) goto ErrorHandling;
assert(sBR->m3.sub[0]->calling == cov);
} else { // END BRMIXED; START SHIFTED
assert(COVNR == BRSHIFTED_INTERN);
}
if (newx == NULL) {
if ((err = loc_set(grid ? * loc->xgr : loc->x, NULL, dim, dim,
grid ? 3 : Gettotalpoints(cov), false, grid,
loc->distances, cov->key)) != NOERROR)
goto ErrorHandling;
} else {
if ((err = loc_set(newx, NULL, dim, dim, newxlen, false, grid,
false, cov->key)) != NOERROR) goto ErrorHandling;
}
xgr = loc->xgr;
if (COVNR != BRMIXED_INTERN && grid) {
double **subxgr = Loc(cov->key)->xgr;
for (d=dim; d>0; d--) {
double len = subxgr[d-1][XLENGTH];
zeropos = zeropos * len + (int) CEIL(len * 0.5) - 1;
}
sBR->zeropos = zeropos;
}
if ((err = CHECK(cov->key, OWNLOGDIM(0), OWNXDIM(0), ProcessType, OWNDOM(0),
OWNISO(0), 1, GaussMethodType)) == NOERROR) {
if ((err = STRUCT(cov->key, NULL)) <= NOERROR) {
err = CHECK(cov->key, OWNLOGDIM(0), OWNXDIM(0), ProcessType, OWNDOM(0),
OWNISO(0), 1, GaussMethodType);
}
}
if (err > NOERROR) goto ErrorHandling;
assert(cov->key->calling == cov);
ErrorHandling:
if (submodel != NULL) COV_DELETE(&submodel, cov); // ok
FREE(newx);
RETURN_ERR(err);
}
int structBrownResnick(model *cov, model **newmodel) {
int d, err, meth,
dim = ANYDIM;
double maxcov,
minloc[MAXMPPDIM], maxloc[MAXMPPDIM],
centreloc[MAXMPPDIM], maxdist[MAXMPPDIM];
model *next = cov->sub[0];
location_type *loc = Loc(cov);
if (loc->Time || (Getgrid(cov) && loc->caniso != NULL)) {
TransformLoc(cov, false, GRIDEXPAND_AVOID, false);
SetLoc2NewLoc(next, PLoc(cov));
}
loc = Loc(cov);
if (cov->key != NULL) COV_DELETE(&(cov->key), cov);// should be ok
if (hasSmithFrame(cov)) {
if (!cov->logspeed)
SERR2("'%.50s' requires a variogram model as submodel that tends to infinity with rate of at least 4log(h) for being compatible with '%.50s'", NICK(cov), DefList[SMITHPROC].nick);
model *calling = cov->calling;
double newscale,
factor = INVSQRTTWO;
ASSERT_NEWMODEL_NULL;
if (next->full_derivs < 0) SERR("given submodel does not make sense");
while (isDollar(next)) {
addModel(&(cov->key), DOLLAR, cov);
newscale = 1.0;
if (!PARAMisNULL(next, DSCALE) ) newscale *= PARAM0(next, DSCALE);
if (!PARAMisNULL(next, DVAR) ) newscale /= SQRT(PARAM0(next, DVAR));
if (factor != 1.0) {
newscale *= factor;
factor = 1.0;
}
RETURN_ERR(ERRORNOTPROGRAMMEDYET);
kdefault(calling, DSCALE, newscale);
next = next->sub[0]; // ok
}
if (cov->sub[MPP_TCF] != NULL) {
RETURN_ERR(ERRORNOTPROGRAMMEDYET);
}
if (NEXTNR == BROWNIAN && PARAM0(next, BROWN_ALPHA) == 2.0) {
addModel(&(cov->key), GAUSS, cov); // ??
if (factor != 1.0) {
addModel(&(cov->key), DOLLAR, cov);
kdefault(cov->key, DSCALE, factor);
}
} else {
SERR("Smith process with BrownResnick tcf only possible for fractal Brownian motion with alpha=2");
}
} else if (hasBrMethodFrame(cov) || hasInterfaceFrame(cov) ||
hasNormedProcessFrame(cov)) {
if (hasBrMethodFrame(next)) {
SERR1("submodel of '%.50s' must be a covariance model or tcf",
NICK(cov));
} else {
ASSERT_FRAME_DEFINED(next);
Types frame = isnowVariogram(next) ? EvaluationType : BadType;
if (((err = covcpy(&(cov->key), next)) != NOERROR)
|| ((err = CHECK(cov->key, OWNLOGDIM(0), OWNXDIM(0),
VariogramType, XONLY,
SYMMETRIC, 1, frame)) != NOERROR)) {
RETURN_ERR(err);
}
assert(Getcaniso(cov) == NULL);
GetDiameter(loc, minloc, maxloc, centreloc);
for (d=0; d<MAXMPPDIM; d++) maxdist[d] = 0.5*(maxloc[d] - minloc[d]);
model *K = NULL;
if ((err = newmodel_covcpy(&K, VARIOGRAM_CALL, cov->key, maxdist, NULL,
NULL, dim, dim, 1, 0, false, false, false))
!= NOERROR) RETURN_ERR(err);
if ((err = alloc_cov(K, dim, 1, 1)) != NOERROR) RETURN_ERR(err);
if (K->sub[0] != NULL) SetLoc2NewLoc(K->sub[0], PLoc(K));
Variogram(NULL, K, &maxcov);
COV_DELETE(&K, cov);
if (isnowPosDef(next) || maxcov <= 4.0) {
meth = BRORIGINAL_USER;
} else if (!next->logspeed || next->logspeed <= 4.0 || maxcov <= 10.0) {
meth = BRSHIFTED_USER;
} else {
meth = BRMIXED_USER;
}
addModel(&(cov->key), meth, cov);
model *key = cov->key;
key->prevloc = PLoc(cov);
kdefault(key, GEV_XI, P0(GEV_XI));
kdefault(key, GEV_MU, P0(GEV_MU));
kdefault(key, GEV_S, P0(GEV_S));
if ((err = CHECK(key, OWNLOGDIM(0), OWNXDIM(0), BrMethodType,
OWNDOM(0), OWNISO(0),
1, BrMethodType)) == NOERROR) {
if ((err = STRUCT(key, NULL)) <= NOERROR) {
err = CHECK(key, OWNLOGDIM(0), OWNXDIM(0), BrMethodType,
OWNDOM(0), OWNISO(0),
1, BrMethodType);
}
}
if (err > NOERROR) RETURN_ERR(err);
}
} else {
ILLEGAL_FRAME;
}
// need check to be called?
RETURN_NOERROR;
}
int initBrownResnick (model *cov, gen_storage *S) {
model *sub = cov->key;
if (sub == NULL)
sub = cov->sub[cov->sub[MPP_SHAPE] != NULL ? MPP_SHAPE: MPP_TCF];
int err;
assert(COVNR == BROWNRESNICKPROC);
if (cov->key != NULL) {
sub->simu.active = true;
sub->simu.expected_number_simu = cov->simu.expected_number_simu;
if ((err = INIT(sub, 0, S)) != NOERROR) RETURN_ERR(err);
ReturnOtherField(cov, sub);
// cov->fieldreturn = wahr;
//cov->origrf = false;
//cov->rf = sub->rf;
}
cov->simu.active = cov->initialised = true;
RETURN_NOERROR;
}
void doBrownResnick(model *cov, gen_storage *s) {
assert(!cov->origrf);
assert(cov->key != NULL);
model *key = cov->key;
PL++;
DO(key, s); // nicht gatternr
PL--;
}
void finaldoBrownResnick(model *cov, double *res, int n, gen_storage *s) {
model *key = cov->key;
finalmaxstable(key, res, n, s);
}
int initBRuser (model *cov, gen_storage *S) {
location_type *loc = Loc(cov);
model *sub = cov->key;
if (sub == NULL)
sub = cov->sub[cov->sub[MPP_SHAPE] != NULL ? MPP_SHAPE: MPP_TCF];
int err, maxpoints = GLOBAL.extreme.maxpoints;
assert(isBrMethod(cov));
if (loc->distances) RETURN_ERR(ERRORFAILED);
if (cov->key != NULL) {
sub->simu.active = true;
double ens = ((double) cov->simu.expected_number_simu) * maxpoints;
sub->simu.expected_number_simu = (int) MIN(ens, (double) MAXINT);
if ((err = INIT(sub, 1, S)) != NOERROR) RETURN_ERR(err);
ReturnOwnField(cov);
}
cov->simu.active = cov->initialised = true;
RETURN_NOERROR;
}
#define ABS std::abs
#define NORMED_PROB 0
#define NORMED_OPTIMP 1 // true/false
#define NORMED_NTH 2
#define NORMED_BURNIN 3
#define NORMED_REJECTION 4
#define NORMED_MAX 0
#define NORMED_VALUES 4
void kappabrnormed(int i, model VARIABLE_IS_NOT_USED *cov, int *nr, int *nc){
// i nummer des parameters
*nc = *nr = 1;
if (i == NORMED_PROB) *nr = SIZE_NOT_DETERMINED;
else if (i > NORMED_REJECTION) *nr = *nc = OUT_OF_RANGE;
// printf("kappa%d = %d %d\n", i,*nr ,*nc);
}
void brnormed() {
BUG;
}
int check_brnormed(model *cov) {
model
*key = cov->key,
*next = cov->sub[0],
*sub = key != NULL ? key : next;
ASSERT_ONESYSTEM;
ASSERT_CARTESIAN;
// printf("start\n");
kdefault(cov, NORMED_REJECTION, true);
kdefault(cov, NORMED_OPTIMP, false);
kdefault(cov, NORMED_NTH, NA_INTEGER); // i.e. adaptive
kdefault(cov, NORMED_BURNIN, NA_INTEGER);// i.e. adaptive
int
err = NOERROR,
total = Gettotalpoints(cov);
if (total <= 1)
SERR1("'%.50s' only works with at least 2 locations.", NICK(cov));
if (!PisNULL(NORMED_PROB)) SERR1("'%.50s' must be given.", KNAME(NORMED_PROB));
if (NROW(NORMED_PROB) !=1 && NROW(NORMED_PROB) != total)
SERR1("length of '%.50s' must equal either 1 or the number of locations",
KNAME(NORMED_PROB));
if (NROW(NORMED_PROB) == total) {
double sum = 0.0;
double *p = P(NORMED_PROB);
for (int k=1; k<total; sum += p[k++]);
sum = 1.0 - sum;
if ((!ISNA(p[0]) && ABS(sum - p[0]) > 1e-5) || sum < 0.0)
SERR1("Values of '%.50s' do not sum up to 1.", KNAME(NORMED_PROB));
p[0] = sum;
}
// printf("normed null%d\n",PisNULL(NORMED_PROB));
if ((err = checkkappas(cov, false)) != NOERROR) RETURN_ERR(err);
Types type = isProcess(sub) ? SYSTYPE(DEFSYS(sub), 0) : VariogramType,
frame = isVariogram(type) ? EvaluationType : BrMethodType;
isotropy_type isoprev = isVariogram(type) ? SYMMETRIC : CARTESIAN_COORD;
assert( OWNXDIM(0) * OWNLOGDIM(0) ==1);
cov->mpp.maxheights[0] = 1.0;
if ((err = CHECK(sub, OWNLOGDIM(0), OWNXDIM(0),
type,//Martin: changed 27.11.13 DefList[COVNR].Type,
XONLY, isoprev, SCALAR, frame)) != NOERROR) {
RETURN_ERR(err);
}
setbackward(cov, sub);
// printf("start ende\n");
RETURN_ERR(err);
}
int struct_brnormed(model *cov, model **newmodel) {
model *next = cov->sub[0];
ASSERT_NEWMODEL_NULL;
if (!hasSchlatherFrame(cov)) SERR2("'%.50s' may only called by '%.50s'",
NAME(cov), DefList[SCHLATHERPROC].name);
if (GetTime(cov) || (Getgrid(cov) && Getcaniso(cov) != NULL)) {
TransformLoc(cov, false, GRIDEXPAND_AVOID, false); // changes loc !
SetLoc2NewLoc(next, PLoc(cov));
}
location_type *loc = Loc(cov);
int i, d, err,
centreidx[MAXMPPDIM],
totalpoints = loc->totalpoints,
newxlen = loc->grid ? 3 : totalpoints,
dim = ANYDIMOF(next);
bool grid = Getgrid(cov);
double centreloc[MAXMPPDIM], minloc[MAXMPPDIM], maxloc[MAXMPPDIM],
*newx= NULL,
**xgr = Getxgr(cov);
if (cov->key != NULL) COV_DELETE(&(cov->key), cov); // should be ok
NEW_STORAGE_WITH_SAVE(br);
br_storage *sBR = cov->Sbr;
assert(Getcaniso(cov) == NULL);
GetDiameter(loc, minloc, maxloc, centreloc, false, true, centreidx);
if ((newx = (double*) MALLOC(dim * newxlen * sizeof(double))) == NULL) {
GERR("Memory allocation failed.\n");
}
if (grid) {
for (d=0; d<dim; d++) {
newx[3 * d + XSTART] = xgr[d][XSTART] - centreloc[d];
newx[3 * d + XSTEP] = xgr[d][XSTEP];
newx[3 * d + XLENGTH] = xgr[d][XLENGTH];
}
} else {
int endfor = totalpoints * dim;
for (i=0; i<endfor; i+=dim)
for (d=0; d<dim; d++) newx[i + d] = loc->x[i + d] - centreloc[d];
}
if ((err = loc_set(newx, NULL, dim, dim, newxlen, false, grid, // loc->grid
loc->distances, cov)) > NOERROR)
goto ErrorHandling;
SetLoc2NewLoc(next, PLoc(cov));
if ((err=covcpy(&(cov->key), next)) > NOERROR) goto ErrorHandling;
if ((err = newmodel_covcpy(&(sBR->vario), VARIOGRAM_CALL, cov->key))!=NOERROR)
goto ErrorHandling;
if ((err = alloc_cov(sBR->vario, dim, 1, 1)) != NOERROR) goto ErrorHandling;
if (isnowVariogram(next)) addModel(&(cov->key), GAUSSPROC, cov);
assert(cov->key->calling == cov);
// PMI(cov);
if ((err = CHECK(sBR->vario->sub[0], 1, 1, VariogramType, XONLY,
ISOTROPIC, 1, EvaluationType)) != NOERROR) goto ErrorHandling;
if ((err = CHECK_PASSTF(cov->key, ProcessType, cov->vdim[0],
GaussMethodType)) != NOERROR)
goto ErrorHandling;
if ((err = STRUCT(cov->key, NULL)) > NOERROR) goto ErrorHandling;
ErrorHandling:
FREE(newx);
RETURN_ERR(err);
}
double* getCi(model *cov, int i) {
br_storage *sBR = cov->Sbr;
if (sBR->normed.C[i] != NULL) return sBR->normed.C[i];
double **Ci = &(sBR->normed.dummyCi);
if (sBR->normed.nCis < sBR->normed.maxCi) {
Ci = sBR->normed.C + i;
sBR->normed.nCis++;
}
bool null;
if ((null = *Ci == NULL))
*Ci = (double*) MALLOC(sBR->normed.total * sizeof(double));
// printf("%d\n", null);
if (null ||
(sBR->normed.nCis >= sBR->normed.maxCi && i != sBR->normed.current_i)) {
CovarianceMatrixCol(sBR->vario->sub[0], i, *Ci);
sBR->normed.current_i = i;
}
return *Ci;
}
void NormedSimulation(model *cov, gen_storage *s) {
br_storage *sBR = cov->Sbr;
model *key = cov->key;
double *field = key->rf,
*res = cov->rf,
*p = P(NORMED_PROB),
*trend = sBR->trend[ORIG_IDX];
pgs_storage *pgs = cov->Spgs;
if (P0INT(NORMED_REJECTION)) {
BUG;
} else { // mcmc
int
total = sBR->normed.total,
endfor = sBR->normed.nth,
zeropos = sBR->zeropos;
for (int n=0; n<endfor; n++) {
sBR->normed.zaehler++;
double u = UNIFORM_RANDOM;
int koben,
i = sBR->normed.total / 2; // kunten
while (p[i] >= u && i > 0) i /= 2;
koben = i * 2 + 1;
if (koben >= n) koben = n - 1;
while (i <= koben) {
int kneu = (koben + i) / 2;
if (p[kneu] >= u) koben = kneu;
else i = kneu + 1;
}
double
sum = 0.0,
max = RF_NEGINF,
*Ci = getCi(cov, i);
DO(key, s); // nicht gatternr,
double fieldzeropos = (double) field[zeropos];// wird durch DO veraendert!
for (int k=0; k<total; k++) {
// printf("%d %d %d %10g\n", k, i, n, field[k] );
// printf("%10g\n", Ci[k]);
// printf("%10g\n", fieldzeropos );
// printf("%10g\n", trend[k] );
double
dummy = field[k] = EXP(field[k] + Ci[k] - fieldzeropos - trend[k]);
if (dummy > max) max = dummy;
sum += p[k] * dummy;
}
// eigentlich nur das Feld nach den endfor Simulationen verwenden,
// da innerhalb abhaengig. Deshalb n_zhou nur um eins erhoeht und
// der Mittelwert ueber alle endfor Fehlder zu sum_zhou addiert.
pgs->sum_zhou_c += max / (double) endfor;
double fmaxDfprop = max / sum,
ratio = fmaxDfprop / sBR->normed.fmaxDfprop;
if (ratio >= 1.0 || UNIFORM_RANDOM < ratio) {
for (int k=0; k<total; k++) res[k] = field[k] / max;
sBR->normed.fmaxDfprop = fmaxDfprop;
sBR->normed.max = max;
sBR->normed.accepted++;
}
} // end for
}
pgs->n_zhou_c ++;
}
// total, nth, zeropos, zaehler, fmaxDfprop, max, accepted, C, dummyCi, maxCi
int init_brnormed(model *cov, gen_storage *s){
br_storage *sBR = cov->Sbr;
pgs_storage *pgs = NULL;
int err,
total = Gettotalpoints(cov),
dim = ANYDIM,
burnin = P0INT(NORMED_BURNIN),
nth = P0INT(NORMED_NTH),
zeropos = sBR->zeropos;
double *trend = sBR->trend[ORIG_IDX];
if ((err = alloc_cov(cov, dim, 1, 1)) != NOERROR) RETURN_ERR(err);
pgs = cov->Spgs;
if ((err = general_init(cov, 1, s)) != NOERROR) RETURN_ERR(err);
Variogram(NULL, sBR->vario, trend);
sBR->normed.total = Gettotalpoints(cov);
assert(sBR->normed.current_prob == NULL);
sBR->normed.current_prob =
(double*) MALLOC(sBR->normed.total * sizeof(double));
assert(sBR->normed.C == NULL);
sBR->normed.C = (double**) CALLOC(sBR->normed.total, sizeof(double*));
// assert(sBR->normed.field == NULL);
// sBR->normed.field = (double*) MALLOC(sBR->normed.total * sizeof(double));
assert(sBR->normed.dummyCi == NULL);
if (NROW(NORMED_PROB) == 1) {
int start = (int) P0(NORMED_PROB);
PFREE(NORMED_PROB);
PALLOC(NORMED_PROB, total, 1);
if (start == BRP_COV && total > 20000) start = BRP_UNIF;
else if (start == BRP_KRIG && total > 1000) start = BRP_UNIF;
if (start != P0INT(NORMED_PROB) && PL >= PL_IMPORTANT ) {
PRINTF("value of '%.50s' has changed from %d to %d",
KNAME(NORMED_PROB), (int) P0(NORMED_PROB), start);
}
if (start == BRP_UNIF) {
double unif = 1.0 / (double) total;
for (int i=0; i<total; P(NORMED_PROB)[i++] = unif);
} else {
assert(start == BRP_KRIG || start == BRP_COV);
int totalSq = total * total;
sBR->normed.do_not_delete_C = true;
double
*p = P(NORMED_PROB),
*c = sBR->normed.C[0] = (double*) MALLOC(totalSq * sizeof(double));
double *one = (double *) MALLOC(total * sizeof(double));
one[0] = 1.0;
for (int i=1; i<total; i++) {
sBR->normed.C[i] = c + i * total;
one[i] = 1.0;
}
if (start == BRP_COV) {
CovarianceMatrix(sBR->vario->sub[0], c);
for (int i=0; i<totalSq; i++) c[i] = EXP(c[i]);
} else { // start == BRP_KRIG
#define nSigma 100
model *key = cov->key;
if (key == NULL) BUG;
double *field = key->rf,
*neu = (double*) CALLOC(totalSq, sizeof(double));
for (int m=0; m<nSigma; m++) {
DO(key, s); // nicht gatternr,
double fieldzeropos = (double) field[zeropos];// d. DO veraendert!
for (int k=0; k<total; k++) field[k] -= fieldzeropos + trend[k];
for(int i=0; i<total; i++) {
double
*sigma_ik = c + i,
*sigma_ki = c + i * total,
max = RF_NEGINF,
*Ci = getCi(cov, i);
for (int k=0; k<total; k++) {
double dummy = neu[k] = field[k] + Ci[k];
if (dummy > max) max = dummy;
}
// untere dreiecksmatrix
for (int k=0; k<=i; k++) sigma_ik[k * total] += EXP(neu[k] - max);
for (int k=i+1; k<total; k++) sigma_ki[k] += EXP(neu[k] - max);
}
}
double minv = 1.0 / (double) nSigma;
for (int i=0; i<total; i++) {
for (int k=0; k<=i; k++) {
c[k + i * total] = (c[i + k * total] *= minv);
}
}
FREE(neu);
}
err = Ext_solvePosDef(c, total, true, one, 1, p, NULL);
FREE(one);
if (err != NOERROR) RETURN_ERR(err);
double sum = 0.0;
for (int i=0; i<total; i++) if (p[i] > 0.0) sum += p[i]; else p[i] = 0.0;
for (int i=0; i<total; i++) p[i] /= sum;
}
}
pgs->zhou_c = 1.0;
pgs->n_zhou_c = 0;
pgs->sum_zhou_c = 0.0;
pgs->estimated_zhou_c = true;
sBR->normed.maxCi = (int) CEIL((double) GLOBAL.br.BRmaxmem / dim);
cov->mpp.mM[0] = cov->mpp.mMplus[0] = 1.0; // 1.0 is import dummy value
cov->mpp.mM[1] = cov->mpp.mMplus[1] = 1.0;
cov->mpp.maxheights[0] = 1.0;
if ((err = ReturnOwnField(cov)) != NOERROR) // must be later than INIT !
RETURN_ERR(err);
#define min_nth 2
#define burninDnth 50
#define burnin_min_rejections 5
#define factor 3
sBR->normed.zaehler = 0;
sBR->normed.accepted = 0;
sBR->normed.nth = 1;
sBR->normed.current_i = UNSET;
if (P0(NORMED_REJECTION)) {
} else {
if (burnin == NA_INTEGER) { // adaptive burnin
burnin = min_nth * burninDnth;
int zaehler = 0;
while (zaehler < burnin) {
// (1 - q - k sigma) n = burnin_min_rejections
// with q the estimated acceptance rate
//sigma^2 = q(1-q) / m variance of q based on the simulation up to now
// and k an arbitrary factor, at least or about 2
NormedSimulation(cov, s);
if (++zaehler == burnin) {
double
q = sBR->normed.accepted / sBR->normed.zaehler,
N = burnin_min_rejections /
(q - factor * SQRT(q * (1 - q) / sBR->normed.zaehler));
if (!R_finite(N) || N < 0) burnin *= 2;
else if (N > burnin) burnin = N;
else break;
}
}
} else {// fixed burnin
for (int zaehler=0; zaehler<burnin; zaehler++) NormedSimulation(cov, s);
}
if ((sBR->normed.adaptive_nth = nth == NA_INTEGER)) {
nth = ROUND((double) burnin / burninDnth);
if (nth <= 0)
SERR1("'%.50s' cannot be determined without burnin", KNAME(NORMED_NTH));
if (nth > 10000)
WARN2("'%.50s'=%d is very large", KNAME(NORMED_NTH), nth);
}
sBR->normed.burnin = burnin;
sBR->normed.nth = nth;
}
double *p = P(NORMED_PROB);
int totalM1 = sBR->normed.total - 1;
sBR->normed.current_cumprob[0] = p[0];
for (int k=1; k<totalM1; k++)
sBR->normed.current_cumprob[k] = sBR->normed.current_cumprob[k-1] + p[k];
sBR->normed.current_cumprob[totalM1] = 1.0;
cov->simu.active = cov->initialised = err == NOERROR;
RETURN_ERR(err);
}
#define NORMED_EACH 100
void do_brnormed(model *cov, gen_storage *s) {
assert(cov->key != NULL);
br_storage *sBR = cov->Sbr;
assert(sBR != NULL && sBR->trend != NULL);
// double *res = cov->rf,
// *trend = sBR->trend[ORIG_IDX];
unsigned int each = NORMED_EACH * sBR->normed.nth;
assert(sBR->normed.total > 0);
assert(sBR->zeropos >= 0 && sBR->zeropos < sBR->normed.total);
assert(cov->origrf);
NormedSimulation(cov, s);
if (sBR->normed.zaehler % each == 0) {
if (P0INT(NORMED_OPTIMP) && true) {
// update the probabilities
BUG;
// sBR->normed.current_prob
}
if (sBR->normed.adaptive_nth) {
double q = sBR->normed.accepted / sBR->normed.zaehler,
N = burnin_min_rejections /
(q - factor * SQRT(q * (1 - q) / sBR->normed.zaehler));
sBR->normed.nth = ROUND(N / burninDnth);
}
}
}
void range_brnormed(model VARIABLE_IS_NOT_USED *cov,
range_type *range){
range->min[NORMED_PROB] = 0;
range->max[NORMED_PROB] = 1;
range->pmin[NORMED_PROB] = 0;
range->pmax[NORMED_PROB] = 1;
range->openmin[NORMED_PROB] = true;
range->openmax[NORMED_PROB] = true;
booleanRange(NORMED_OPTIMP);
range->min[NORMED_NTH] = 1;
range->max[NORMED_NTH] = RF_INF;
range->pmin[NORMED_NTH] = 1;
range->pmax[NORMED_NTH] = 1e4;
range->openmin[NORMED_NTH] = false;
range->openmax[NORMED_NTH] = true;
range->min[NORMED_BURNIN] = 0;
range->max[NORMED_BURNIN] = RF_INF;
range->pmin[NORMED_BURNIN] = 1;
range->pmax[NORMED_BURNIN] = 1e6;
range->openmin[NORMED_BURNIN] = false;
range->openmax[NORMED_BURNIN] = true;
booleanRange(NORMED_REJECTION);
}