swh:1:snp:e520bf41b0e99213acde680a9d87fadac1aee079
Tip revision: cb8a32f1ddcfbcce003da5312030ddb4d1d04dc7 authored by Martin Schlather on 10 July 2012, 00:00:00 UTC
version 2.0.57
version 2.0.57
Tip revision: cb8a32f
MLE.cc
/*
Authors
Martin Schlather, martin.schlather@math.uni-goettingen.de
library for simulation of random fields
Copyright (C) 2001 -- 2011 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 2
of the License, or (at your option) any later version.
RO
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.
*/
extern "C" {
#include <R.h>
#include <Rdefines.h>
}
#include <R_ext/Linpack.h>
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "RF.h"
#include "primitive.h"
// #include "Covariance.h"
/*
operator "#" that has been modified such that explanatory
variables can be used instead of the spatial coordinates
*/
void iso2iso_MLE(double *x, cov_model *cov, double *v) {
cov_model *next = cov->sub[0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
}
void spiso2spiso_MLE(double *x, cov_model *cov, double *v) {
cov_model *next = cov->sub[0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
}
void spacetime2iso_MLE(double *x, cov_model *cov, double *v) {
cov_model *next = cov->sub[0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
}
void Stat2iso_MLE(double *x, cov_model *cov, double *v) {
cov_model *next = cov->sub[0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
}
void Nonstat2iso_MLE(double *x, double *y, cov_model *cov, double *v) {
cov_model *next = cov->sub[0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
}
void Stat2spacetime_MLE(double *x, cov_model *cov, double *v) {
cov_model *next = cov->sub[0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
}
void Nonstat2spacetime_MLE(double *x, double *y, cov_model *cov, double *v) {
//assert(false);
cov_model *next = cov->sub[0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
}
void Stat2Stat_MLE(double *x, cov_model *cov, double *v) {
cov_model *next = cov->sub[0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
}
void Nonstat2Stat_MLE(double *x, double *y, cov_model *cov, double *v) {
cov_model *next = cov->sub[0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
}
/*
*/
// simple transformations as variance, scale, anisotropy matrix, etc.
void Siso_MLE(double *x, cov_model *cov, double *v){
cov_model *next = cov->sub[0];
int i,
vdimSq = cov->vdim * cov->vdim;
double **p = cov->p,
var = p[DVAR][0];
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);//x wird nicht verwendet
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
for (i=0; i<vdimSq; i++) v[i] *= var;
}
void Sstat_MLE(double *x, cov_model *cov, double *v){
cov_model *next = cov->sub[0];
double
**p = cov->p,
var = p[DVAR][0];
int i,
vdimSq = cov->vdim * cov->vdim;
if (cov->MLE == NULL) CovList[next->nr].cov(x, next, v);//x wird nicht verwendet
else {
CovList[next->nr].cov(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
for (i=0; i<vdimSq; i++) v[i] *= var;
}
void Snonstat_MLE(double *x, double *y, cov_model *cov, double *v){
cov_model *next = cov->sub[0];
double
**p = cov->p,
var = p[DVAR][0];
int i,
vdimSq = cov->vdim * cov->vdim;
if (cov->MLE == NULL) CovList[next->nr].nonstat_cov(x, y, next, v);
else {
double *xx;
xx = cov->MLE + CovMatrixTotal * next->xdim * 2;
CovList[next->nr].nonstat_cov(xx, xx + next->xdim, next, v);
}
for (i=0; i<vdimSq; i++) v[i] *= var;
}
void DS_MLE(double *x, cov_model *cov, double *v){
cov_model *next = cov->sub[0];
int i,
vdimSq = cov->vdim * cov->vdim;
double
**p = cov->p,
spinvscale = (p[DANISO] == NULL) ? 1.0 / p[DSCALE][0] : p[DANISO][0],
varSc = p[DVAR][0] * spinvscale;
if (cov->MLE == NULL) CovList[next->nr].D(x, next, v);//x wird nicht verwendet
else {
CovList[next->nr].D(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
for (i=0; i<vdimSq; i++) v[i] *= varSc;
}
void DDS_MLE(double *x, cov_model *cov, double *v){
cov_model *next = cov->sub[0];
int i,
vdimSq = cov->vdim * cov->vdim;
double
**p = cov->p,
spinvscale = (p[DANISO] == NULL) ? 1.0 / p[DSCALE][0] : p[DANISO][0],
varScSq = p[DVAR][0] * spinvscale * spinvscale;
if (cov->MLE == NULL) CovList[next->nr].D2(x, next, v);//x wird nicht verwendet
else {
CovList[next->nr].D2(cov->MLE + CovMatrixTotal * next->xdim, next, v);
}
for (i=0; i<vdimSq; i++) v[i] *= varScSq;
}
// nachfolgende Funktion sowie anyNAscaleup, anyNAdow, manipulate_x
// nur fuer speed in MLE verwendet, wenn die x-Werte direkt den
// knoten vorliegen sollen
void SetPrevToTriMaxFalse(cov_model *calling) {
// bislang war TriFalse; jetzt nicht mehr;
// setze letzes DOLLAR auf TriMaxFalse;
if (calling != NULL && calling->anyNAscaleup == TriFalse) {
cov_model *prev = calling;
while (prev != NULL &&
(prev->nr < DOLLAR || prev->nr > LASTDOLLAR)) {
// prev->anyNAscaleup = TriBeyond;
prev = prev->calling;
}
if (prev != NULL) prev->anyNAscaleup = TriMaxFalse;
}
}
void GetNAPosition(cov_model *cov, int *usern, int *internn,
naptr_type mem, naptr_type address,
NAname_type names, sortsofparam *sorts,
sortsofparam *internsorts, bool *isnan,
int *covzaehler,
int printing, int depth) {
/*
printing <= 0 nothing
1 only NA, NaN
2 others as "*"
3 others as value
s
mem[0..internn] : addresse wo (neue) NA position im Speicher
address[0..internn]
== NULL : address in mem the value should be put;
== mem[internn] :
* internsort=ANISOFROMSCALE : point to scale (using 1/value)
sorts[0..usern] : see sortsofparam in RF.h
internsorts[0..internn] : see sortsofparam in RF.h
isnan[0..usern] : NA or NaN
*/
int i, c, r,
namenr = 1,
*nrow = cov->nrow,
*ncol = cov->ncol;
// double *lastmem = NULL;
#define SHORTlen 4
char shortname[SHORTlen+2], shortD[SHORTlen+2];
bool user;
cov_fct *C = CovList + cov->nr,
*CC = C;
SEXPTYPE *type = C->kappatype;
// C is pointer to true model; CC points to the first model passed
// in initNerror.cc by IncludeModel or IncludePrim
while(strncmp(CC->name, InternalName, strlen(InternalName)) ==0) CC--;
covzaehler[cov->nr]++;
strcopyN(shortname, CC->name, SHORTlen);
if (covzaehler[cov->nr] >= 2) {
char dummy[SHORTlen];
strcopyN(dummy, shortname, SHORTlen-1);
sprintf(shortname, "%s%d", dummy, covzaehler[cov->nr]);
}
if (printing>0) PRINTF("%s\n", CC->name);
// CC needed below for the kappa.names which are given
// printf("%d:%s %d %s\n", cov->nr, CC->name, covzaehler[cov->nr], shortname);
if (cov->manipulating_x > 0) {
// any scale==NA in the calling function
cov->anyNAscaleup = cov->anyNAdown = TriTrue;
SetPrevToTriMaxFalse(cov->calling);
} else {
cov->anyNAscaleup = cov->anyNAdown = TriFalse;
if (cov->calling != NULL) cov->anyNAscaleup = cov->calling->anyNAscaleup;
}
depth++;
for (i=0; i<C->kappas; i++) {
if (nrow[i] == 0 || ncol[i] == 0) continue;
if (printing > 0) {
leer(depth); PRINTF("%s\n", C->kappanames[i]);
}
for (r=0; r<nrow[i]; r++) {
int nv = 0; // anzahl NA in aktuellem parameter
for (c=0; c<ncol[i]; c++) {
double v = RF_NAN; // value in aktuellem parameter
int idx = c * nrow[i] + r;
if (*internn >= MAX_NA) error("maximum number of NA reached");
if (type[i] == REALSXP) {
v = cov->p[i][idx];
mem[*internn] = &(cov->p[i][idx]);
} else if (type[i] == INTSXP) {
v = ((int *) cov->p[i])[idx] == NA_INTEGER
? NA_REAL : (double) ((int *) cov->p[i])[idx];
if (ISNA(v) || ISNAN(v))
error("integer variables currently do not allowed for NA"); // !!!
// mem[*internn] = (*double) &(((int *) cov->p[i])[idx]);
} else if (type[i] == LISTOF + REALSXP) {
listoftype *q;
int j, end;
double *p;
q=(listoftype*) cov->p[i];
p = q->p[r];
end = q->nrow[r] * q->ncol[r];
for (j=0; j<end; j++)
if (ISNA(p[j]) || ISNAN(p[j]))
error("no NAs allowed in regression ");
v = 0.0; // dummy
} else {
error("unknown SXP type");
}
user = true;
isnan[*usern] = ISNAN(v) && !ISNA(v);
if (ISNA(v) || ISNAN(v)) { // entgegen Arith.h gibt ISNA nur NA an !!
cov->anyNAdown = TriTrue;
if (printing > 0) {
if (nv>1 || (c>0 && printing > 1)) PRINTF(", "); else leer(depth+1);
}
if (cov->nr >= DOLLAR && cov->nr <= LASTDOLLAR) {
// shortD partial name for R level
cov_model *next = cov->sub[0];
while((next->nr >= GATTER && next->nr <= LASTGATTER) ||
(next->nr >= DOLLAR && next->nr <= LASTDOLLAR) ||
(next->nr == NATSC))
next = next->sub[0];
if (covzaehler[next->nr] == 0) { // next wurde noch nicht
// untersucht, somit ist covzaehler[next->nr] um 1 niedriger
// als covzaehler[cov->nr] !
strcopyN(shortD, CovList[next->nr].name, SHORTlen);
} else {
char dummy[SHORTlen];
strcopyN(dummy, CovList[next->nr].name, SHORTlen-1);
sprintf(shortD, "%s%d", dummy, covzaehler[next->nr]+1);
// printf("$$ > %d:%s %d %s\n", next->nr, CovList[next->nr].name,
// covzaehler[next->nr], shortD);
}
// assert(DANISO == DSCALE + 1);
address[*internn] = NULL;
if (i==DVAR) {
internsorts[*internn] = sorts[*usern] =
// zur sicherheit beide:
(next->nr == MIXEDEFFECT || next->nr == MLEMIXEDEFFECT) ?
MIXEDVAR : next->nr == NUGGET ? NUGGETVAR : VARPARAM;
sprintf(names[*usern], "%s.var", shortD); // for R level only
} else {
cov->anyNAscaleup = TriTrue;
SetPrevToTriMaxFalse(cov->calling);
if (i==DSCALE) {
// lastmem = mem[*internn]; // used for all diagonal elements of
// aniso
sorts[*usern] = internsorts[*internn] = SCALEPARAM;
sprintf(names[*usern], "%s.s", shortD);// for R level only
} else {
assert(i == DANISO);
// no lastmem here !
if (cov->ncol[DSCALE] != 0 &&
(ISNA(cov->p[DSCALE][0]) || ISNAN(cov->p[DSCALE][0]))) {
assert(false);
/*
// internal
address[*internn] = lastmem; // points to scale which
// was just be before and had NA
internsorts[*internn] = ANISOFROMSCALE;
sprintf(names[*usern], "%s.aniso", shortD);// R level only
// no setting of sorts[users]
if (printing > 1) {
if (printing > 2) PRINTF("(given by scale)");
else PRINTF("*");
}
user = false;
*/
} else {
if (cov->q != NULL && cov->q[0] != 0.0) {
// printf("sdfasdf\n");
ERR("naturalscaling only allowed for isotropic setting");
}
if (r==c) {
internsorts[*internn] = sorts[*usern] = DIAGPARAM;
sprintf(names[*usern], "%s.d.%d", shortD, r);// R level only
} else {
internsorts[*internn] = sorts[*usern] = ANISOPARAM;
sprintf(names[*usern], "%s.d.%d.%d", shortD, r, c);// R level
}
}
}
}
} else {
// standard setting !
address[*internn] = NULL;
if (cov->nr==MIXEDEFFECT || cov->nr==MLEMIXEDEFFECT)
continue;
else
sorts[*usern] = internsorts[*internn] = ANYPARAM;
char kappashort[SHORTlen];
strcopyN(kappashort, CC->kappanames[i], SHORTlen);
sprintf(names[*usern], "%s.%s", shortname, kappashort);
if (*usern > 0 && 0==
strncmp(names[*usern], names[*usern-1], strlen(names[*usern]))){
if (namenr == 1) {
sprintf(names[*usern-1], "%s.%s.%d",
shortname, kappashort, namenr);
}
namenr++;
sprintf(names[*usern], "%s.%s.%d", shortname, kappashort,namenr);
} else namenr=1;
}
if (PL > 7) {
// PRINTF(" %d mem %ld; addr %ld; %s, %d %d\n", *usern,
// (POINTER) mem[*internn], (POINTER)address[*internn],
// names[*usern], sorts[*usern], internsorts[*internn]);
}
if (user) {
if (printing > 0) {
if (printing <= 2) PRINTF("%d",*usern + 1);
else PRINTF("!%d", *usern + 1);
}
(*usern)++;
}
(*internn)++;
nv++;
} else {
// kein NA an der Position; somit nur Anzeige
if (printing > 1) {
if (c>0) PRINTF(", "); else leer(depth+1);
if (printing <= 2) PRINTF("*");
else {
if (type[i]== REALSXP) PRINTF("%6.2e", v);
else PRINTF("%5d", (int) v);
}
}
} // else kein NA
} //c
if ((printing > 0 && nv > 0) || (printing > 1 && ncol[i] > 0))
PRINTF("\n");
} // r
} // i
cov_model *sub;
for (i=0; i<MAXSUB; i++) {
sub = cov->sub[i];
if (sub != NULL) {
if (printing > 0) {
leer(depth);
PRINTF("%s = ", C->subnames[i]);
}
GetNAPosition(sub, usern, internn, mem, address,
names, sorts, internsorts, isnan,
covzaehler, printing, depth);
if (cov->anyNAdown==TriFalse && sub->anyNAdown==TriTrue)
cov->anyNAdown = TriTrue;
}
}
if (cov->anyNAdown == TriTrue && cov->anyNAscaleup != TriTrue) {
for (i=0; i<MAXSUB; i++) {
sub = cov->sub[i];
if (sub != NULL) {
if (sub->anyNAdown==TriFalse && sub->anyNAscaleup != TriTrue)
sub->anyNAdown=TriMaxFalse;
}
}
}
}
SEXP GetNAPositions(SEXP model, SEXP tsdim, SEXP xdim, SEXP stationary,
SEXP Print) {
int i, usern, internn, covzaehler[MAXNRCOVFCTS];
naptr_type mem, address;
sortsofparam sorts[MAX_NA], internsorts[MAX_NA];
bool isnan[MAX_NA];
NAname_type names;
bool skipchecks = GLOBAL.general.skipchecks;
GLOBAL.general.skipchecks = true;
CheckModel(model, tsdim, xdim, stationary, STORED_MODEL + MODEL_USER,
MAXMLEDIM);
GLOBAL.general.skipchecks = skipchecks;
usern = internn = 0;
for (i=0; i<MAXNRCOVFCTS; covzaehler[i++]=0);
GetNAPosition(STORED_MODEL[MODEL_USER], &usern, &internn, mem, address,
names,
sorts, internsorts, isnan,
covzaehler, INTEGER(Print)[0], 0);
SEXP ans;
PROTECT (ans = allocVector(INTSXP, 1));
INTEGER(ans)[0] = usern;
UNPROTECT(1);
return ans;
}
int countnas(cov_model *cov) {
int i, r, count, endfor,
*nrow = cov->nrow,
*ncol = cov->ncol;
cov_fct *C = CovList + cov->nr;
SEXPTYPE *type = C->kappatype;
if (( cov->nr == MIXEDEFFECT || cov->nr == MLEMIXEDEFFECT )
&& cov->nrow[1] > 0) // b is given
return 0;
count= 0;
for (i=0; i<C->kappas; i++) {
if (nrow[i] == 0 || ncol[i] == 0) continue;
endfor = nrow[i] * ncol[i];
if (type[i] == REALSXP) {
double *p = cov->p[i];
for (r=0; r<endfor; r++) if (ISNAN(p[r]) || ISNA(p[r])) count++;
} else if (type[i] == INTSXP) {
int *p = (int *) cov->p[i];
for (r=0; r<endfor; r++) if (p[r] == NA_INTEGER) count++;
} else if (type[i] == LISTOF + REALSXP) {
continue; // no NAs allowed
} else assert(false);
}
cov_model *sub;
for (i=0; i<MAXSUB; i++) {
sub = cov->sub[i];
if (sub == NULL) continue;
count += countnas(sub);
}
return count;
}
void Take21internal(cov_model *cov, cov_model *cov_bound,
double **bounds_pointer, int *NBOUNDS) {
/* determine the bounds for the parameters that
are to be estimated
*/
// PrintModelInfo(cov_bound);
// assert(false);
int i, c, r, nv=0,
*nrow = cov->nrow,
*ncol = cov->ncol,
*nrow2 = cov_bound->nrow,
*ncol2 = cov_bound->ncol;
cov_fct *C = CovList + cov->nr;
SEXPTYPE *type = C->kappatype;
// printf("\n\n\n ======================= \n");
// PrintModelInfo(cov);
// PrintModelInfo(cov_bound);
for (i=0; i<C->kappas; i++) {
// printf("%s %s \n", C->name, C->kappanames[i]);
if (C->kappatype[i] >= LISTOF) continue;
if (nrow[i] != nrow2[i] || ncol[i] != ncol2[i]) {
PRINTF("%s i: %d, nrow1=%d, nrow2=%d, ncol1=%d, ncol2=%d\n",
C->name, i, nrow[i], nrow2[i], ncol[i], ncol2[i]);
error("lower/upper/user does not fit to model (size of matrix)");
}
for (r=0; r<nrow[i]; r++) {
for (c=0; c<ncol[i]; c++) {
double v=RF_NAN,
w=RF_NAN; // value in aktuellem parameter
int idx = c * nrow[i] + r;
// printf("%d %d idx=%d %d %d %d\n", r,c, idx, type[i], REALSXP, INTSXP);
if (type[i] == REALSXP) {
v = cov->p[i][idx];
w = cov_bound->p[i][idx];
}
else if (type[i] == INTSXP) {
v = ((int *) cov->p[i])[idx] == NA_INTEGER
? NA_REAL : (double) ((int *) cov->p[i])[idx];;
w = ((int *) cov_bound->p[i])[idx] == NA_INTEGER
? NA_REAL : (double) ((int *) cov_bound->p[i])[idx];
}
// if (cov->nr >= DOLLAR && cov->nr <= LASTDOLLAR)
// printf("%s ", CovList[cov->sub[0]->nr].name);
// printf("%s %s, r=%d, c=%d: %d %d %f\n",
// C->name, C->kappanames[i], r, c, nrow[i], ncol[i], v);
if (ISNA(v) || ISNAN(v)) { // entgegen Arith.h gibt ISNA nur NA an !!
if ((cov->nr < DOLLAR || cov->nr > LASTDOLLAR ||
i == DVAR ||
(i== DSCALE && cov->q == NULL) || // ! natscaling
i == DANISO) &&
((cov->nr != MIXEDEFFECT && cov->nr != MLEMIXEDEFFECT) ||
i != BETAMIXED)
)// aniso ?? ABfrage OK ??
{
// printf("%s %s, r=%d, c=%d: %d <? %d\n",
// C->name, C->kappanames[i], r, c, nv, *NBOUNDS);
if (nv >= *NBOUNDS) {
// PRINTF("%s %s, r=%d, c=%d: %d >= %d\n",
// C->name, C->kappanames[i], r, c, nv, *NBOUNDS);
error("lower/upper/user does not fit to model (number parameters)");
}
(*bounds_pointer)[nv] = w;
nv++;
}
} //ISNA
} //c
} // r
} // i
*NBOUNDS = *NBOUNDS - nv;
(*bounds_pointer) += nv;
for (i=0; i<MAXSUB; i++) {
if (cov->sub[i] != NULL) {
Take21internal(cov->sub[i], cov_bound->sub[i], bounds_pointer, NBOUNDS);
}
}
}
SEXP Take2ndAtNaOf1st(SEXP model, SEXP model_bound, SEXP tsdim, SEXP xdim,
SEXP stationary, SEXP nbounds, SEXP skipchecks) {
// MAXMLEDIM
double *bounds_pointer;
int NBOUNDS_INT =INTEGER(nbounds)[0],
*NBOUNDS= &NBOUNDS_INT;
bool oldskipchecks = GLOBAL.general.skipchecks;
NAOK_RANGE = true;
if (LOGICAL(skipchecks)[0]) GLOBAL.general.skipchecks = true;
CheckModel(model_bound, tsdim, xdim, stationary, STORED_MODEL + MODEL_BOUNDS,
MAXMLEDIM);
GLOBAL.general.skipchecks = oldskipchecks;
// PrintModelInfo(STORED_MODEL[MODEL_BOUNDS]); assert(false);
NAOK_RANGE = true;
CheckModel(model, tsdim, xdim, stationary, STORED_MODEL + MODEL_INTERN,
MAXMLEDIM);
// print("take\n");
// PrintModelInfo(STORED_MODEL[MODEL_INTERN]);
NAOK_RANGE = false;
SEXP bounds;
PROTECT(bounds = allocVector(REALSXP, *NBOUNDS));
bounds_pointer = NUMERIC_POINTER(bounds);
// printf("%f %f %f\n", bounds_pointer[0], bounds_pointer[1],
// bounds_pointer[2]);
// assert(false);
// printf("s %ld\n", bounds_pointer);
Take21internal(STORED_MODEL[MODEL_INTERN], STORED_MODEL[MODEL_BOUNDS],
&bounds_pointer, NBOUNDS);
// printf("e %ld\n", bounds_pointer);
// assert(false);
if (*NBOUNDS != 0) error("lower/upper does not fit to model");
UNPROTECT(1);
return bounds;
}
void GetNARanges(cov_model *cov, cov_model *min, cov_model *max,
double *minpile, double *maxpile, int *usern) {
/*
determine the ranges of the parameters to be estimated
*/
int i, r,
*nrow = cov->nrow,
*ncol = cov->ncol;
cov_fct *C = CovList + cov->nr;
SEXPTYPE *type = C->kappatype;
double
v = RF_NAN,
dmin = RF_NAN,
dmax = RF_NAN;
for (i=0; i<C->kappas; i++) {
int end = nrow[i] * ncol[i];
if (end == 0) continue;
if (type[i] == REALSXP) {
dmin = min->p[i][0];
dmax = max->p[i][0];
} else if (type[i] == INTSXP) {
dmin = ((int *) min->p[i])[0] == NA_INTEGER
? NA_REAL : (double) ((int *) min->p[i])[0];
dmax = ((int *) max->p[i])[0] == NA_INTEGER
? NA_REAL : (double) ((int *) max->p[i])[0];
} else if (type[i] == LISTOF + REALSXP) {
dmin = min->p[i][0];
dmax = max->p[i][0];
} else {
error("unknown SXP type in GetRanges.");
}
for (r=0; r<end; r++) {
if (type[i] == REALSXP) {
v = cov->p[i][r];
}
else if (type[i] == INTSXP) {
v = ((int *) cov->p[i])[r] == NA_INTEGER
? NA_REAL : (double) ((int *) cov->p[i])[r];
} else if (type[i] == LISTOF + REALSXP) {
continue; // !!!!!!!!!!!
} else assert(false);
if ( (ISNA(v) || ISNAN(v))
&& cov->nr!=MIXEDEFFECT && cov->nr!=MLEMIXEDEFFECT) {
if (cov->nr < DOLLAR || cov->nr > LASTDOLLAR ||
i == DVAR ||
(i== DSCALE && cov->q == NULL) || // ! natscaling
i == DANISO // aniso ?? ABfrage OK ??
) {
minpile[*usern] = dmin;
maxpile[*usern] = dmax;
// printf("%f %f\n", dmin, dmax);
(*usern)++;
} else {
// maybe internn and internpile should be set here
continue;
}
} // isna
} // r
} // kappas
for (i=0; i<MAXSUB; i++) {
if (cov->sub[i] != NULL) {
GetNARanges(cov->sub[i], min->sub[i], max->sub[i], minpile, maxpile,
usern);
}
}
}
void CovMatrixMLE(double *x, int *dist, int *lx, int *set, int *submodel, // k
double *value) {
LIST_ELEMENT = *set - 1;
ELEMENTNR_PLUS = *submodel - 1;
CovarianceMatrix(x, (bool) *dist, *lx, STORED_MODEL[MODEL_MLE], value);
ELEMENTNR_PLUS = LIST_ELEMENT = - 1;
}
int check_recursive_range(cov_model *cov, bool NAOK) {
/*
called by MLEGetModelInfo
*/
int i, err;
if ((err = check_within_range(cov, NAOK)) != NOERROR) return err;
for (i=0; i<MAXSUB; i++) {
if (cov->sub[i] != NULL) {
if ((err = check_recursive_range(cov->sub[i], NAOK)) != NOERROR)
return err;
}
}
return NOERROR;
}
int CheckEffect(cov_model *cov) {
int i,j,end;
bool na_var = false;
double *p;
cov_model *next;
if (cov->nr >= GATTER && cov->nr <= LASTGATTER) return CheckEffect(cov->sub[0]);
else if (cov->nr == MIXEDEFFECT) {
cov->nr = MLEMIXEDEFFECT;
if (cov->nrow[1] == 0 && cov->nsub == 0) return deteffect;
if (cov->nsub == 0) return fixedeffect;
next = cov->sub[0];
if (next->nr >= GATTER && next->nr <= LASTGATTER) next = next->sub[0];
if (next->nr >= DOLLAR && next->nr <= LASTDOLLAR) {
if (next->ncol[DVAR] == 1 && next->nrow[DVAR] == 1) {
na_var = (ISNA(next->p[DVAR][0]) || ISNAN(next->p[DVAR][0]));
}
for (i=0; i<=DMAX; i++) {
if (i!=DVAR) {
end = next->ncol[i] * next->nrow[i];
p = next->p[i];
for (j=0; j<end; j++) {
if (ISNA(p[j]) || ISNAN(p[j])) {
return (next->nr == CONSTANT) ? eff_error : spaceeffect;
// NA in e.g. scale for constant matrix -- does not make sense
}
}
}
}
next = next->sub[0];
if (next->nr >= GATTER && next->nr <= LASTGATTER) next = next->sub[0];
}
if (next->nr == CONSTANT) {
return (cov->nrow[0] > cov->ncol[0])
? (na_var ? rvareffect : randomeffect)
: (na_var ? lvareffect : largeeffect);
} else return spaceeffect;
} else return remaining;
}
static naptr_type MLE_MEM, MLE_ADDRESS;
static int MLE_USERN=-1, MLE_INTERNN = -1;
static sortsofparam MLE_SORTS[MAX_NA], MLE_INTERNSORTS[MAX_NA];
static bool MLE_ISNAN[MAX_NA];
SEXP MLEGetModelInfo(SEXP model, SEXP tsdim, SEXP xdim) { // ex MLEGetNAPos
/*
basic set up of covariance model, determination of the NA's
and the corresponding ranges.
*/
cov_model *cov, *min=NULL,
*max=NULL,
*openmin=NULL,
*openmax=NULL;
// bool skipchecks = GLOBAL.general.skipchecks;
NAname_type names;
double mle_min[MAX_NA], mle_max[MAX_NA];
int i, covzaehler[MAXNRCOVFCTS],
usern = 0, effect[MAXSUB], nas[MAXSUB];
bool anyeffect;
SEXP stationary, isotropy, Effect, Nas;
#define total 5
PROTECT(stationary=allocVector(LGLSXP, 1));
LOGICAL(stationary)[0] = false;
PROTECT(isotropy=allocVector(LGLSXP, 1));
LOGICAL(isotropy)[0] = false;
NAOK_RANGE = true;
// printf("here\n");
CheckModel(model, tsdim, xdim, stationary, STORED_MODEL + MODEL_MLE,
MAXMLEDIM);
NAOK_RANGE = false;
// printf("here\n");
cov = STORED_MODEL[MODEL_MLE];
if (cov->nr >= GATTER && cov->nr <= LASTGATTER) {
cov_model *next = cov->sub[0];
if (next->statIn < COVARIANCE) {
LOGICAL(stationary)[0] = true;
if (next->isoIn==ISOTROPIC) {
LOGICAL(isotropy)[0] = true;
INTEGER(xdim)[0] = 1;
// removeOnly(STORED_MODEL + MODEL_MLE); nicht wegnehmen
// da derzeit negative distanczen in CovMatrixMult auftreten
}
}
}
// GLOBAL.general.skipchecks = skipchecks;
check_recursive_range(STORED_MODEL[MODEL_MLE], true);
get_ranges(STORED_MODEL[MODEL_MLE], &min, &max, &openmin, &openmax,
true);
// printf("min/max\n");
// PrintModelInfo(min);
// PrintModelInfo(max);
//x PrintModelInfo(cov); assert(false);
MLE_USERN = MLE_INTERNN = 0;
for (i=0; i<MAXNRCOVFCTS; covzaehler[i++]=0);
GetNAPosition(STORED_MODEL[MODEL_MLE], &MLE_USERN, &MLE_INTERNN,
MLE_MEM, MLE_ADDRESS, names,
MLE_SORTS, MLE_INTERNSORTS, MLE_ISNAN,
covzaehler,
(PL > 2) + (PL > 4) + (PL > 6), 0);
GetNARanges(STORED_MODEL[MODEL_MLE], min, max, mle_min, mle_max,
&usern);
cov = STORED_MODEL[MODEL_MLE];
if (cov->nr >= GATTER && cov->nr <= LASTGATTER) cov = cov->sub[0];
anyeffect=false;
if (cov->nr == PLUS) {
for (i=0; i<cov->nsub; i++) {
effect[i] = CheckEffect(cov->sub[i]);
nas[i] = countnas(cov->sub[i]);
// printf("%d %d %d\n", i, effect[i], remaining);
if (effect[i] == eff_error)
ERR("scaling parameter appears with constant matrix in the mixed effect part");
if (effect[i] != remaining) anyeffect = true;
}
}
PROTECT(Effect = allocVector(INTSXP, anyeffect ? cov->nsub : 0));
PROTECT(Nas = allocVector(INTSXP, anyeffect ? cov->nsub : 0));
if (anyeffect) {
for (i=0; i<cov->nsub;i++) INTEGER(Effect)[i] = effect[i];
for (i=0; i<cov->nsub;i++) INTEGER(Nas)[i] = nas[i];
assert(cov->nr == PLUS); // zur sicherheit
cov->nr = MLEPLUS;
}
// printf("xx %d %d %d %d\n", effect[0], effect[1], anyeffect, LENGTH(Effect));
COV_DELETE(&min);
COV_DELETE(&max);
COV_DELETE(&openmin);
COV_DELETE(&openmax);
SEXP ans, matrix, nameAns, nameMatrix;
PROTECT(matrix = allocMatrix(REALSXP, MLE_USERN, 4));
PROTECT(nameMatrix = allocVector(STRSXP, MLE_USERN));
for (i=0; i<MLE_USERN; i++) {
REAL(matrix)[i] = mle_min[i];
REAL(matrix)[i + MLE_USERN] = mle_max[i];
REAL(matrix)[i + 2 * MLE_USERN] = MLE_SORTS[i];
REAL(matrix)[i + 3 * MLE_USERN] = (int) MLE_ISNAN[i];
SET_STRING_ELT(nameMatrix, i, mkChar(names[i]));
}
setAttrib(matrix, R_RowNamesSymbol, nameMatrix);
PROTECT(ans = allocVector(VECSXP, total));
PROTECT(nameAns = allocVector(STRSXP, total));
i = 0;
SET_STRING_ELT(nameAns, i, mkChar("minmax"));
SET_VECTOR_ELT(ans, i++, matrix);
SET_STRING_ELT(nameAns, i, mkChar("stationary"));
SET_VECTOR_ELT(ans, i++, stationary);
SET_STRING_ELT(nameAns, i, mkChar("isotropy"));
SET_VECTOR_ELT(ans, i++, isotropy);
SET_STRING_ELT(nameAns, i, mkChar("effect"));
SET_VECTOR_ELT(ans, i++, Effect);
SET_STRING_ELT(nameAns, i, mkChar("NAs"));
SET_VECTOR_ELT(ans, i++, Nas);
setAttrib(ans, R_NamesSymbol, nameAns);
assert(i==total);
UNPROTECT(8);
return ans;
}
bool anymixed(cov_model *cov) {
int i;
if (cov->nr == MIXEDEFFECT || cov->nr == MLEMIXEDEFFECT) return true;
for (i=0; i<cov->nsub; i++) {
if (anymixed(cov->sub[i])) return true;
}
return false;
}
void MLEanymixed(int *anymix) {
*anymix = (int) anymixed(STORED_MODEL[MODEL_MLE]);
}
void iexplDollar(cov_model *cov) {
/*
get the naturalscaling values and devide the preceeding scale model
by this value
*/
double *p, scale;
if (cov->nr == NATSC) {
cov_model
*next = cov->sub[0],
*calling = cov->calling;
if (calling->nr >= GATTER && calling->nr <= LASTGATTER)
calling = calling->calling;
assert(calling!=NULL && calling->nr >= DOLLAR && calling->nr <= LASTDOLLAR);
scale = CovList[next->nr].naturalscale(next);
p = calling->p[DSCALE];
if (p != NULL) p[0] /= scale;
else {
int i,
n = calling->nrow[DANISO] * calling->ncol[DANISO];
p = calling->p[DANISO];
for (i=0; i<n; i++) p[i] /= scale;
}
} else {
int i;
for (i=0; i<MAXSUB; i++) { // cov->sub[i]: luecken erlaubt bei PSgen !
if (cov->sub[i] != NULL) iexplDollar(cov->sub[i]);
}
}
}
void expliciteDollarMLE(double *values) {
// userinterfaces.cc
int i, un;
// first get the naturalscaling values and devide the preceeding
// scale model by this value
if (NS==NATSCALE_MLE) {
iexplDollar(STORED_MODEL[MODEL_MLE]);
}
// Then get the values out of the model
for (un=i=0; un<MLE_USERN; i++) {
values[un++] = MLE_MEM[i][0];
MLE_MEM[i][0] = RF_NAN;
}
}
void PutValuesAtNA(double *values){
int i, un;
// set ordinary parameters for all (sub)models
for (un=i=0; i<MLE_INTERNN; i++) MLE_MEM[i][0] = values[un++];
if (un != MLE_USERN) ERR("severe error in PutValuesAtNA.");
// finally set all aniso from scale
//for (i=0; i<MLE_INTERNN; i++) {
// if (MLE_INTERNSORTS[i] == ANISOFROMSCALE) {
// MLE_MEM[i][0] = 1.0 / MLE_ADDRESS[i][0];
// }
// }
// PrintModelInfo(STORED_MODEL[MODEL_INTERN]);
}
#define MAX_INT 2000000
SEXP IGetModel(cov_model *cov, int modus) {
// # modus: 1 : Modell wie gespeichert
// # 0 : Modell unter Annahme PracticalRange>0
// # [ 2 : natscale soweit wie moeglich zusammengezogen ]
SEXP model, nameMvec, dummy;
int i, nmodelinfo,
k = 0;
long int mem=0;
cov_fct *C = CovList + cov->nr;
if ((modus == 0 && cov->nr == NATSC)
|| (cov->nr>=GATTER && cov->nr<=LASTGATTER)) {
return IGetModel(cov->sub[0], modus);
}
nmodelinfo = C->kappas + 1;
for (i=0; i<MAXSUB; i++) if (cov->sub[i] != NULL) nmodelinfo++;
for (i=0; i<C->kappas; i++) if (cov->p[i] == NULL) nmodelinfo--;
PROTECT(model = allocVector(VECSXP, nmodelinfo));
PROTECT(nameMvec = allocVector(STRSXP, nmodelinfo));
SET_STRING_ELT(nameMvec, k, mkChar("")); // name
cov_fct *CC = CovList + cov->nr;
while(strncmp(CC->name, InternalName, strlen(InternalName)) ==0) CC--;
SET_VECTOR_ELT(model, k++, mkString(CC->name));
for(i=0; i<C->kappas; i++) {
if (cov->p[i] == NULL) {
// k++; // 9.1.09, needed for "$" (proj can be NULL) -- 19.1.09 k++
// deleted since otherwise outcome does not fit
// model definition anymore (arbitrary NULL in the definition)
// instead: nmodelinfo--; above wihtin loop
continue;
} else dummy = Param((void*) cov->p[i], cov->nrow[i], cov->ncol[i],
C->kappatype[i], true, &mem);
SET_STRING_ELT(nameMvec, k, mkChar(C->kappanames[i]));
SET_VECTOR_ELT(model, k++, dummy);
}
// vielleicht mal spaeter !
// SET_STRING_ELT(nameMvec, k, mkChar("user"));
// SET_VECTOR_ELT(model, k++, Int(cov->user, Nothing + 1, MAXINT, mem));
// printf("a %d %d %d %s\n",cov->nsub, k, nmodelinfo, CovList[cov->nr].name);
// PrintModelInfo(cov);
int zaehler = 0;
for (i=0; i<MAXSUB; i++) {
if (cov->sub[i] != NULL) {
// printf("i=%d %d %d %s\n", i,cov->nsub,zaehler,CovList[cov->sub[i]->nr].name);
SET_VECTOR_ELT(model, k++, IGetModel(cov->sub[i], modus));
if (++zaehler >= cov->nsub) break;
}
}
// PrintModelInfo(cov);
// printf("b %d %d %s\n", k, nmodelinfo, CovList[cov->nr].name);
assert(k == nmodelinfo);
setAttrib(model, R_NamesSymbol, nameMvec);
UNPROTECT(2); // model + namemodelvec
return model;
}
SEXP GetModel(SEXP keynr, SEXP Modus) {
// # modus: 1 : Modell wie gespeichert
// # 0 : Modell unter Annahme PracticalRange>0 (natsc werden geloescht)
// # 2 : natscale soweit wie moeglich zusammengezogen (natsc werden
// drauf multipliziert)
// Nutzer kann 3 Modifikationen des Models in MLE laufen lassen:
// * keinen Praktikal range oder individuell angeben
// * extern practical range definieren
// * intern practical range verwenden lassen (use.naturalscaling)
int knr = INTEGER(keynr)[0],
modus = INTEGER(Modus)[0];
cov_model *cov;
if (knr>=0) {
if (knr < MAXKEYS) {
key_type *key;
key = &(KEY[knr]);
if (key->cov == NULL) return allocVector(VECSXP, 0);
cov = key->cov;
} else return allocVector(VECSXP, 0);
} else {
knr = -knr-1;
if (knr < MODEL_MAX && STORED_MODEL[knr] != NULL) cov = STORED_MODEL[knr];
else return allocVector(VECSXP, 0);
}
if (modus == 0 || modus == 1) return IGetModel(cov, modus);
else {
cov_model *dummy = NULL; //ACHTUNG: "=NULL" hinzugefuegt
SEXP value;
covcpy(&dummy, cov, false, true);
iexplDollar(dummy);
value = IGetModel(dummy, 0);
COV_DELETE(&dummy);
return(value);
}
}