void in_Rsockwrite(int *sockp, char **buf, int *start, int *end, int *len)
{
ssize_t n;
if (*end > *len)
*end = *len;
if (*start < 0)
*start = 0;
if (*end < *start) {
*len = -1;
return;
}
check_init();
#ifdef DEBUG
printf("writing %s to %d", *buf, *sockp);
#endif
perr.error = 0;
n = Sock_write(*sockp, *buf + *start, *end - *start, &perr);
*len = (int) n;
if(perr.error) REprintf("socket error: %s\n", strerror(perr.error));
}
SEXP dbExistsTable(SEXP dbi_conn_sexp, SEXP tableName_sexp) {
SEXP ans;
if(TYPEOF(dbi_conn_sexp) != EXTPTRSXP || dbi_conn_sexp == R_NilValue) {
return R_NilValue;
}
DatabaseConnection* conn = reinterpret_cast<DatabaseConnection*>(R_ExternalPtrAddr(dbi_conn_sexp));
if(!conn) {
// throw bad_connection_object
REprintf("bad database connection.\n");
return R_NilValue;
}
const char* tableName = CHAR(STRING_ELT(tableName_sexp,0));
PROTECT(ans = allocVector(LGLSXP,1));
LOGICAL(ans)[0] = static_cast<int>(conn->existsTable(tableName));
UNPROTECT(1);
return ans;
}
/* This does *not* work: gives *empty* .Data slot [bug in NEW_OBJECT()? ] */
SEXP d2mpfr(SEXP x, SEXP prec)
{
int i_prec = asInteger(prec),
nx = LENGTH(x), np = LENGTH(prec),
n = (nx == 0 || np == 0) ? 0 : imax2(nx, np),
nprot = 1;
SEXP val = PROTECT(NEW_OBJECT(MAKE_CLASS("mpfr"))),
lis = ALLOC_SLOT(val, Rmpfr_Data_Sym, VECSXP, n);
double *dx;
if(!isReal(x)) { PROTECT(x = coerceVector(x, REALSXP)); nprot++; }
REprintf("d2mpfr(x, prec): length(x) = %d, prec = %d -> length(lis) = %d\n",
nx, i_prec, LENGTH(lis));
dx = REAL(x);
for(int i = 0; i < n; i++) {
SET_VECTOR_ELT(lis, i, duplicate(d2mpfr1_(dx [i % nx],
i_prec [i % np])));
}
UNPROTECT(nprot);
return val;
}
void nvimcom_Stop()
{
#ifndef WIN32
if(ih){
removeInputHandler(&R_InputHandlers, ih);
close(ifd);
close(ofd);
}
#endif
if(nvimcom_initialized){
Rf_removeTaskCallbackByName("NVimComHandler");
#ifdef WIN32
closesocket(sfd);
WSACleanup();
#else
close(sfd);
pthread_cancel(tid);
pthread_join(tid, NULL);
#endif
ListStatus *tmp = firstList;
while(tmp){
firstList = tmp->next;
free(tmp->key);
free(tmp);
tmp = firstList;
}
for(int i = 0; i < 64; i++){
free(loadedlibs[i]);
loadedlibs[i] = NULL;
}
if(obbrbuf1)
free(obbrbuf1);
if(obbrbuf2)
free(obbrbuf2);
if(verbose)
REprintf("nvimcom stopped\n");
}
nvimcom_initialized = 0;
}
void setColnames(const std::vector<std::string> &cnames) {
int protect_count(0);
if (static_cast<R_len_t>(cnames.size()) != Rf_ncols(Robject)) {
REprintf("setColnames: colnames size does not match ncols(Robject).");
return;
}
// check if we have existing dimnames
SEXP dimnames = Rf_getAttrib(Robject, R_DimNamesSymbol);
if (dimnames == R_NilValue) {
PROTECT(dimnames = Rf_allocVector(VECSXP, 2));
++protect_count;
SET_VECTOR_ELT(dimnames, 0, R_NilValue);
}
SEXP cnames_sexp = PROTECT(Rf_allocVector(STRSXP, cnames.size()));
++protect_count;
for (size_t i = 0; i < cnames.size(); ++i) { SET_STRING_ELT(cnames_sexp, i, Rf_mkChar(cnames[i].c_str())); }
SET_VECTOR_ELT(dimnames, 1, cnames_sexp);
Rf_setAttrib(Robject, R_DimNamesSymbol, dimnames);
UNPROTECT(protect_count);
}
/* Poll related. */
SEXP R_zmq_poll(SEXP R_socket, SEXP R_type, SEXP R_timeout){
int C_ret = -1, C_errno, i;
PBD_POLLITEM_LENGTH = LENGTH(R_socket);
if(PBD_POLLITEM_LENGTH > PBD_POLLITEM_MAXSIZE){
REprintf("Too many sockets (%d) are asked.\n", PBD_POLLITEM_LENGTH);
}
PBD_POLLITEM = (zmq_pollitem_t *) malloc(PBD_POLLITEM_LENGTH * sizeof(zmq_pollitem_t));
for(i = 0; i < PBD_POLLITEM_LENGTH; i++){
PBD_POLLITEM[i].socket = R_ExternalPtrAddr(VECTOR_ELT(R_socket, i));
PBD_POLLITEM[i].events = (short) INTEGER(R_type)[i];
}
C_ret = zmq_poll(PBD_POLLITEM, PBD_POLLITEM_LENGTH, (long) INTEGER(R_timeout)[0]);
if(C_ret == -1){
C_errno = zmq_errno();
warning("R_zmq_poll: %d strerror: %s\n",
C_errno, zmq_strerror(C_errno));
}
return(AsInt(C_ret));
} /* End of R_zmq_poll(). */
void RangeList::filterGeneName(const char* inclusionGeneFileName, const char* geneTableFileName){
// require user input gene list file
if (strlen(geneTableFileName) == 0 && strlen(inclusionGeneFileName) != 0) {
REprintf("Please provide gene list file (e.g. refFlat) until we are able to process gene\n");
return;
//exit(1);
}
// if not specify any gene, return whole range.
if (strlen(inclusionGeneFileName) == 0) {
return;
}
// store which gene do we want if specified
std::set< std::string > inclusionSet;
LineReader lr(inclusionGeneFileName);
std::string gene;
while(lr.readLine(&gene)) {
inclusionSet.insert(gene);
}
std::vector<std::string> fields;
std::string chr;
std::string geneNameTbl;
LineReader geneTable(geneTableFileName);
while (geneTable.readLineBySep(&fields, "\t ")) {
geneNameTbl = fields[0];
if (inclusionSet.find(geneNameTbl) != inclusionSet.end()){ // store gene range
chr = chopChr(fields[2].c_str());
this->rangeCollection.addRange(chr,
atoi(fields[4].c_str()), // start
atoi(fields[5].c_str())); // end
}
}
if (this->rangeCollection.size() == 0){
Rprintf("We cannot find given gene in your geneListFile, so all sites will be outputed\n");
}
}
void R_nc_get_vara_double( int *ncid, int *varid, int *start,
int *count, double *data, int *retval )
{
int i, err, ndims;
size_t s_start[MAX_NC_DIMS], s_count[MAX_NC_DIMS];
char vn[2048];
err = nc_inq_varndims(*ncid, *varid, &ndims );
if( err != NC_NOERR )
REprintf( "Error in R_nc_get_vara_double while getting ndims: %s\n",
nc_strerror(*retval) );
for( i=0; i<ndims; i++ ) {
s_start[i] = (size_t)start[i];
s_count[i] = (size_t)count[i];
}
*retval = nc_get_vara_double(*ncid, *varid, s_start, s_count, data );
if( *retval != NC_NOERR ) {
nc_inq_varname( *ncid, *varid, vn );
REprintf( "Error in R_nc_get_vara_double: %s\n",
nc_strerror(*retval) );
REprintf( "Var: %s Ndims: %d Start: ", vn, ndims );
for( i=0; i<ndims; i++ ) {
REprintf( "%u", (unsigned int)s_start[i] );
if( i < ndims-1 )
REprintf( "," );
}
REprintf( "Count: " );
for( i=0; i<ndims; i++ ) {
REprintf( "%u", (unsigned int)s_count[i] );
if( i < ndims-1 )
REprintf( "," );
}
}
}
/* This is not used in R and in no header */
void
Rf_callToplevelHandlers(SEXP expr, SEXP value, Rboolean succeeded,
Rboolean visible)
{
R_ToplevelCallbackEl *h, *prev = NULL;
Rboolean again;
if(Rf_RunningToplevelHandlers == TRUE)
return;
h = Rf_ToplevelTaskHandlers;
Rf_RunningToplevelHandlers = TRUE;
while(h) {
again = (h->cb)(expr, value, succeeded, visible, h->data);
if(R_CollectWarnings) {
REprintf(_("warning messages from top-level task callback '%s'\n"),
h->name);
PrintWarnings();
}
if(again) {
prev = h;
h = h->next;
} else {
R_ToplevelCallbackEl *tmp;
tmp = h;
if(prev)
prev->next = h->next;
h = h->next;
if(tmp == Rf_ToplevelTaskHandlers)
Rf_ToplevelTaskHandlers = h;
if(tmp->finalizer)
tmp->finalizer(tmp->data);
free(tmp);
}
}
Rf_RunningToplevelHandlers = FALSE;
}
void show_family(Family *f) {
Offspring *child;
if (f) {
REprintf(" %d: %d + %d / ",
f->pedigree, f->father_id, f->mother_id);
for (child=f->children; child; child=child->next) {
REprintf(" %d", child->id);
if (child->affected==2)
REprintf("*");
if (child->next)
REprintf(",");
}
REprintf("\n");
}
else {
REprintf("*** empty family ***\n");
}
}
attribute_hidden
double R_pretty(double *lo, double *up, int *ndiv, int min_n,
double shrink_sml, double high_u_fact[],
int eps_correction, int return_bounds)
{
/* From version 0.65 on, we had rounding_eps := 1e-5, before, r..eps = 0
* then, 1e-7 was consistent with seq.default() and seq.int() till 2010-02-03,
* where it was changed to 1e-10 for seq*(), and in 2017-08-14 for pretty(): */
#define rounding_eps 1e-10
#define h high_u_fact[0]
#define h5 high_u_fact[1]
double dx, cell, unit, base, U;
double ns, nu;
int k;
Rboolean i_small;
dx = *up - *lo;
/* cell := "scale" here */
if(dx == 0 && *up == 0) { /* up == lo == 0 */
cell = 1;
i_small = TRUE;
} else {
cell = fmax2(fabs(*lo),fabs(*up));
/* U = upper bound on cell/unit */
U = 1 + ((h5 >= 1.5*h+.5) ? 1/(1+h) : 1.5/(1+h5));
U *= imax2(1,*ndiv) * DBL_EPSILON; // avoid overflow for large ndiv
/* added times 3, as several calculations here */
i_small = dx < cell * U * 3;
}
/*OLD: cell = FLT_EPSILON+ dx / *ndiv; FLT_EPSILON = 1.192e-07 */
if(i_small) {
if(cell > 10)
cell = 9 + cell/10;
cell *= shrink_sml;
if(min_n > 1) cell /= min_n;
} else {
cell = dx;
if(*ndiv > 1) cell /= *ndiv;
}
if(cell < 20*DBL_MIN) {
warning(_("Internal(pretty()): very small range.. corrected"));
cell = 20*DBL_MIN;
} else if(cell * 10 > DBL_MAX) {
warning(_("Internal(pretty()): very large range.. corrected"));
cell = .1*DBL_MAX;
}
/* NB: the power can be negative and this relies on exact
calculation, which glibc's exp10 does not achieve */
base = pow(10.0, floor(log10(cell))); /* base <= cell < 10*base */
/* unit : from { 1,2,5,10 } * base
* such that |u - cell| is small,
* favoring larger (if h > 1, else smaller) u values;
* favor '5' more than '2' if h5 > h (default h5 = .5 + 1.5 h) */
unit = base;
if((U = 2*base)-cell < h*(cell-unit)) { unit = U;
if((U = 5*base)-cell < h5*(cell-unit)) { unit = U;
if((U =10*base)-cell < h*(cell-unit)) unit = U; }}
/* Result: c := cell, u := unit, b := base
* c in [ 1, (2+ h) /(1+h) ] b ==> u= b
* c in ( (2+ h)/(1+h), (5+2h5)/(1+h5)] b ==> u= 2b
* c in ( (5+2h)/(1+h), (10+5h) /(1+h) ] b ==> u= 5b
* c in ((10+5h)/(1+h), 10 ) b ==> u=10b
*
* ===> 2/5 *(2+h)/(1+h) <= c/u <= (2+h)/(1+h) */
ns = floor(*lo/unit+rounding_eps);
nu = ceil (*up/unit-rounding_eps);
#ifdef DEBUGpr
REprintf("pretty(lo=%g,up=%g,ndiv=%d,min_n=%d,shrink=%g,high_u=(%g,%g),"
"eps=%d)\n\t dx=%g; is.small:%d. ==> cell=%g; unit=%g\n",
*lo, *up, *ndiv, min_n, shrink_sml, h, h5,
eps_correction, dx, (int)i_small, cell, unit);
#endif
if(eps_correction && (eps_correction > 1 || !i_small)) {
if(*lo != 0.) *lo *= (1- DBL_EPSILON); else *lo = -DBL_MIN;
if(*up != 0.) *up *= (1+ DBL_EPSILON); else *up = +DBL_MIN;
}
#ifdef DEBUGpr
if(ns*unit > *lo)
REprintf("\t ns= %.0f -- while(ns*unit > *lo) ns--;\n", ns);
#endif
while(ns*unit > *lo + rounding_eps*unit) ns--;
#ifdef DEBUGpr
if(nu*unit < *up)
REprintf("\t nu= %.0f -- while(nu*unit < *up) nu++;\n", nu);
#endif
while(nu*unit < *up - rounding_eps*unit) nu++;
k = (int)(0.5 + nu - ns);
if(k < min_n) {
/* ensure that nu - ns == min_n */
#ifdef DEBUGpr
REprintf("\tnu-ns=%.0f-%.0f=%d SMALL -> ensure nu-ns= min_n=%d\n",
nu,ns, k, min_n);
#endif
k = min_n - k;
if(ns >= 0.) {
nu += k/2;
ns -= k/2 + k%2;/* ==> nu-ns = old(nu-ns) + min_n -k = min_n */
} else {
ns -= k/2;
nu += k/2 + k%2;
}
*ndiv = min_n;
}
else {
*ndiv = k;
}
if(return_bounds) { /* if()'s to ensure that result covers original range */
if(ns * unit < *lo) *lo = ns * unit;
if(nu * unit > *up) *up = nu * unit;
} else {
*lo = ns;
*up = nu;
}
#ifdef DEBUGpr
REprintf("\t ns=%.0f ==> lo=%g\n", ns, *lo);
REprintf("\t nu=%.0f ==> up=%g ==> ndiv = %d\n", nu, *up, *ndiv);
#endif
return unit;
#undef h
#undef h5
}
/*
* Public
*/
SEXP do_mrdwt(SEXP vntX, SEXP vntH, SEXP vntL)
{
SEXP vntOut;
SEXP vntYl;
SEXP vntYh;
SEXP vntLr;
double *x, *h, *yl, *yh;
int m, n, lh, L;
#ifdef DEBUG_RWT
REprintf("In do_mrdwt(x, h, L)...\n");
#endif
/*
* Handle first parameter (numeric matrix)
*/
#ifdef DEBUG_RWT
REprintf("\tfirst param 'x'\n");
#endif
if (GetMatrixDimen(vntX, &m, &n) != 2)
{
error("'x' is not a two dimensional matrix");
/*NOTREACHED*/
}
PROTECT(vntX = AS_NUMERIC(vntX));
x = NUMERIC_POINTER(vntX);
#ifdef DEBUG_RWT
REprintf("x[%d][%d] = 0x%p\n", m, n, x);
#endif
/*
* Handle second parameter (numeric vector)
*/
#ifdef DEBUG_RWT
REprintf("\tsecond param 'h'\n");
#endif
PROTECT(vntH = AS_NUMERIC(vntH));
h = NUMERIC_POINTER(vntH);
lh = GET_LENGTH(vntH);
#ifdef DEBUG_RWT
REprintf("h[%d] = 0x%p\n", GET_LENGTH(vntH), h);
#endif
/*
* Handle third parameter (integer scalar)
*/
#ifdef DEBUG_RWT
REprintf("\tthird param 'L'\n");
#endif
{
PROTECT(vntL = AS_INTEGER(vntL));
{
int *piL = INTEGER_POINTER(vntL);
L = piL[0];
}
UNPROTECT(1);
}
#ifdef DEBUG_RWT
REprintf("L = %d\n", L);
#endif
#ifdef DEBUG_RWT
REprintf("\tcheck number of levels\n");
#endif
if (L < 0)
{
error("The number of levels, L, must be a non-negative integer");
/*NOTREACHED*/
}
#ifdef DEBUG_RWT
REprintf("\tcheck dimen prereqs\n");
#endif
/* Check the ROW dimension of input */
if (m > 1)
{
double mtest = (double) m / pow(2.0, (double) L);
if (!isint(mtest))
{
error("The matrix row dimension must be of size m*2^(L)");
/*NOTREACHED*/
}
}
/* Check the COLUMN dimension of input */
if (n > 1)
{
double ntest = (double) n / pow(2.0, (double) L);
if (!isint(ntest))
{
error("The matrix column dimension must be of size n*2^(L)");
/*NOTREACHED*/
}
}
#ifdef DEBUG_RWT
REprintf("\tcreating value objects\n");
#endif
/* Create yl value object */
{
#ifdef DEBUG_RWT
REprintf("\tcreating 'yl' value object\n");
#endif
PROTECT(vntYl = NEW_NUMERIC(m*n));
yl = NUMERIC_POINTER(vntYl);
/* Add dimension attribute to value object */
#ifdef DEBUG_RWT
REprintf("\tconvert 'yl' value object to matrix\n");
#endif
{
SEXP vntDim;
PROTECT(vntDim = NEW_INTEGER(2));
INTEGER(vntDim)[0] = m;
INTEGER(vntDim)[1] = n;
SET_DIM(vntYl, vntDim);
UNPROTECT(1);
}
}
/* Create yh value object */
{
int cols = (min(m,n) == 1) ? (L * n) : (3 * L * n);
#ifdef DEBUG_RWT
REprintf("\tcreating 'yh' value object\n");
#endif
PROTECT(vntYh = NEW_NUMERIC(m*cols));
yh = NUMERIC_POINTER(vntYh);
/* Add dimension attribute to value object */
#ifdef DEBUG_RWT
REprintf("\tconvert 'yh' value object to matrix\n");
#endif
{
SEXP vntDim;
PROTECT(vntDim = NEW_INTEGER(2));
INTEGER(vntDim)[0] = m;
INTEGER(vntDim)[1] = cols;
SET_DIM(vntYh, vntDim);
UNPROTECT(1);
}
}
/* Create Lr value object */
{
#ifdef DEBUG_RWT
REprintf("\tcreating 'Lr' value object\n");
#endif
PROTECT(vntLr = NEW_INTEGER(1));
INTEGER_POINTER(vntLr)[0] = L;
}
#ifdef DEBUG_RWT
REprintf("\tcompute redundant discrete wavelet transform\n");
#endif
MRDWT(x, m, n, h, lh, L, yl, yh);
/* Unprotect params */
UNPROTECT(2);
#ifdef DEBUG_RWT
REprintf("\tcreate list output object\n");
#endif
PROTECT(vntOut = NEW_LIST(3));
#ifdef DEBUG_RWT
REprintf("\tassigning value objects to list\n");
#endif
SET_VECTOR_ELT(vntOut, 0, vntYl);
SET_VECTOR_ELT(vntOut, 1, vntYh);
SET_VECTOR_ELT(vntOut, 2, vntLr);
/* Unprotect value objects */
UNPROTECT(3);
{
SEXP vntNames;
#ifdef DEBUG_RWT
REprintf("\tassigning names to value objects in list\n");
#endif
PROTECT(vntNames = NEW_CHARACTER(3));
SET_STRING_ELT(vntNames, 0, CREATE_STRING_VECTOR("yl"));
SET_STRING_ELT(vntNames, 1, CREATE_STRING_VECTOR("yh"));
SET_STRING_ELT(vntNames, 2, CREATE_STRING_VECTOR("L"));
SET_NAMES(vntOut, vntNames);
UNPROTECT(1);
}
/* Unprotect output object */
UNPROTECT(1);
#ifdef DEBUG_RWT
REprintf("\treturning output...\n");
#endif
return vntOut;
}
double qnorm5(double p, double mu, double sigma, int lower_tail, int log_p)
{
double p_, q, r, val;
#ifdef IEEE_754
if (ISNAN(p) || ISNAN(mu) || ISNAN(sigma))
return p + mu + sigma;
#endif
R_Q_P01_boundaries(p, ML_NEGINF, ML_POSINF);
if(sigma < 0) ML_ERR_return_NAN;
if(sigma == 0) return mu;
p_ = R_DT_qIv(p);/* real lower_tail prob. p */
q = p_ - 0.5;
#ifdef DEBUG_qnorm
REprintf("qnorm(p=%10.7g, m=%g, s=%g, l.t.= %d, log= %d): q = %g\n",
p,mu,sigma, lower_tail, log_p, q);
#endif
/*-- use AS 241 --- */
/* double ppnd16_(double *p, long *ifault)*/
/* ALGORITHM AS241 APPL. STATIST. (1988) VOL. 37, NO. 3
Produces the normal deviate Z corresponding to a given lower
tail area of P; Z is accurate to about 1 part in 10**16.
(original fortran code used PARAMETER(..) for the coefficients
and provided hash codes for checking them...)
*/
if (fabs(q) <= .425) {/* 0.075 <= p <= 0.925 */
r = .180625 - q * q;
val =
q * (((((((r * 2509.0809287301226727 +
33430.575583588128105) * r + 67265.770927008700853) * r +
45921.953931549871457) * r + 13731.693765509461125) * r +
1971.5909503065514427) * r + 133.14166789178437745) * r +
3.387132872796366608)
/ (((((((r * 5226.495278852854561 +
28729.085735721942674) * r + 39307.89580009271061) * r +
21213.794301586595867) * r + 5394.1960214247511077) * r +
687.1870074920579083) * r + 42.313330701600911252) * r + 1.);
}
else { /* closer than 0.075 from {0,1} boundary */
/* r = min(p, 1-p) < 0.075 */
if (q > 0)
r = R_DT_CIv(p);/* 1-p */
else
r = p_;/* = R_DT_Iv(p) ^= p */
r = sqrt(- ((log_p &&
((lower_tail && q <= 0) || (!lower_tail && q > 0))) ?
p : /* else */ log(r)));
/* r = sqrt(-log(r)) <==> min(p, 1-p) = exp( - r^2 ) */
#ifdef DEBUG_qnorm
REprintf("\t close to 0 or 1: r = %7g\n", r);
#endif
if (r <= 5.) { /* <==> min(p,1-p) >= exp(-25) ~= 1.3888e-11 */
r += -1.6;
val = (((((((r * 7.7454501427834140764e-4 +
.0227238449892691845833) * r + .24178072517745061177) *
r + 1.27045825245236838258) * r +
3.64784832476320460504) * r + 5.7694972214606914055) *
r + 4.6303378461565452959) * r +
1.42343711074968357734)
/ (((((((r *
1.05075007164441684324e-9 + 5.475938084995344946e-4) *
r + .0151986665636164571966) * r +
.14810397642748007459) * r + .68976733498510000455) *
r + 1.6763848301838038494) * r +
2.05319162663775882187) * r + 1.);
}
else { /* very close to 0 or 1 */
r += -5.;
val = (((((((r * 2.01033439929228813265e-7 +
2.71155556874348757815e-5) * r +
.0012426609473880784386) * r + .026532189526576123093) *
r + .29656057182850489123) * r +
1.7848265399172913358) * r + 5.4637849111641143699) *
r + 6.6579046435011037772)
/ (((((((r *
2.04426310338993978564e-15 + 1.4215117583164458887e-7)*
r + 1.8463183175100546818e-5) * r +
7.868691311456132591e-4) * r + .0148753612908506148525)
* r + .13692988092273580531) * r +
.59983220655588793769) * r + 1.);
}
if(q < 0.0)
val = -val;
/* return (q >= 0.)? r : -r ;*/
}
return mu + sigma * val;
}
/* Convert an R value to a GenericValue based on the type expected, given by type. */
bool
convertRToGenericValue(llvm::GenericValue *rv, SEXP rval, const llvm::Type *type)
{
llvm::Type::TypeID ty;
if(!type) {
REprintf("var arg %d\n", TYPEOF(rval));
rv->IntVal = INTEGER(rval)[0];
// rv->IntVal = llvm::APInt((unsigned) 32, INTEGER(rval)[0]);
return(true);
}
// FIX - enhance to cover more situations.
if(type->isPointerTy()) {
const llvm::Type *elType = ((const llvm::PointerType*) type)->getElementType();
ty = elType->getTypeID();
bool ok = true;
switch(ty) {
case llvm::Type::IntegerTyID:
if(elType->isIntegerTy(8)) {
if(TYPEOF(rval) == STRSXP) {
rv->PointerVal = Rf_length(rval) ? (void*) CHAR(STRING_ELT(rval, 0)) : (void *) NULL;
} else if(TYPEOF(rval) == NILSXP) {
rv->PointerVal = (void*) NULL;
} else
ok = false;
} else if(TYPEOF(rval) == INTSXP)
rv->PointerVal = INTEGER(rval);
else
ok = false;
break;
case llvm::Type::DoubleTyID:
if(TYPEOF(rval) == REALSXP)
rv->PointerVal = REAL(rval);
else
ok = false;
break;
case llvm::Type::PointerTyID:
if(TYPEOF(rval) == STRSXP) {
rv->PointerVal = Rf_length(rval) ? (void*) CHAR(STRING_ELT(rval, 0)) : (void *) NULL;
} if(TYPEOF(rval) == NILSXP || rval == R_NilValue) {
rv->PointerVal = (void*) NULL;
} else if(TYPEOF(rval) == RAWSXP)
rv->PointerVal = (void*) RAW(rval);
else
ok = false;
break;
case llvm::Type::VoidTyID:
if(rval == R_NilValue)
rv->PointerVal = (void*) NULL;
else if(TYPEOF(rval) == RAWSXP)
rv->PointerVal = (void*) RAW(rval);
break;
default:
ok = false;
}
if(ok == false) {
int rtype = isSEXPType(type);
if(rtype > 0) {
rv->PointerVal = rval;
ok = true;
}
}
if(ok == false && TYPEOF(rval) == EXTPTRSXP) {
rv->PointerVal = R_ExternalPtrAddr(rval);
ok = true;
}
/* See if this is an S4 object with a "ref" slot that is an external pointer */
SEXP refRVal = NULL;
if(ok == false && IS_S4_OBJECT(rval) && (refRVal = GET_SLOT(rval, Rf_install("ref")))
&& refRVal != R_NilValue && TYPEOF(refRVal) == EXTPTRSXP) {
rv->PointerVal = R_ExternalPtrAddr(refRVal);
ok = true;
}
if(ok == false) {
PROBLEM "no method to convert R object of R type %d to LLVM pointer to type %d", TYPEOF(rval), ty
WARN;
}
return(ok);
}
ty = type->getTypeID();
switch(ty) {
case llvm::Type::IntegerTyID: {
uint64_t val = asInteger(rval);
unsigned BitWidth = llvm::cast<llvm::IntegerType>(type)->getBitWidth();
rv->IntVal = llvm::APInt(BitWidth, val);
return rv;
}
break;
case llvm::Type::DoubleTyID: {
rv->DoubleVal = Rf_asReal(rval);
}
break;
case llvm::Type::FloatTyID: {
rv->FloatVal = Rf_asReal(rval);
}
break;
default:
PROBLEM "no code yet for converting R to GV for type %d", (int) ty
ERROR;
}
return(true);
}
void mypause() {
REprintf("--------------------------------------------------\n");
}
void attribute_hidden Rstd_ShowMessage(const char *s)
{
REprintf("%s\n", s);
}
void attribute_hidden Rstd_Suicide(const char *s)
{
REprintf("Fatal error: %s\n", s);
/* Might be called before translation is running */
R_CleanUp(SA_SUICIDE, 2, 0);
}
double qgamma(double p, double alpha, double scale, int lower_tail, int log_p)
/* shape = alpha */
{
#define C7 4.67
#define C8 6.66
#define C9 6.73
#define C10 13.32
#define EPS1 1e-2
#define EPS2 5e-7/* final precision */
#define MAXIT 1000/* was 20 */
#define pMIN 1e-100 /* was 0.000002 = 2e-6 */
#define pMAX (1-1e-12)/* was 0.999998 = 1 - 2e-6 */
const double
i420 = 1./ 420.,
i2520 = 1./ 2520.,
i5040 = 1./ 5040;
double p_, a, b, c, ch, g, p1, v;
double p2, q, s1, s2, s3, s4, s5, s6, t, x;
int i;
/* test arguments and initialise */
#ifdef IEEE_754
if (ISNAN(p) || ISNAN(alpha) || ISNAN(scale))
return p + alpha + scale;
#endif
R_Q_P01_check(p);
if (alpha <= 0) ML_ERR_return_NAN;
/* FIXME: This (cutoff to {0, +Inf}) is far from optimal when log_p: */
p_ = R_DT_qIv(p);/* lower_tail prob (in any case) */
if (/* 0 <= */ p_ < pMIN) return 0;
if (/* 1 >= */ p_ > pMAX) return BOOM::infinity();
v = 2*alpha;
c = alpha-1;
g = lgammafn(alpha);/* log Gamma(v/2) */
/*----- Phase I : Starting Approximation */
#ifdef DEBUG_qgamma
REprintf("qgamma(p=%7g, alpha=%7g, scale=%7g, l.t.=%2d, log_p=%2d): ",
p,alpha,scale, lower_tail, log_p);
#endif
if(v < (-1.24)*R_DT_log(p)) { /* for small chi-squared */
#ifdef DEBUG_qgamma
REprintf(" small chi-sq.\n");
#endif
/* FIXME: Improve this "if (log_p)" :
* (A*exp(b)) ^ 1/al */
ch = pow(p_* alpha*exp(g+alpha*M_LN2), 1/alpha);
if(ch < EPS2) {/* Corrected according to AS 91; MM, May 25, 1999 */
goto END;
}
} else if(v > 0.32) { /* using Wilson and Hilferty estimate */
x = qnorm(p, 0, 1, lower_tail, log_p);
p1 = 0.222222/v;
ch = v*pow(x*sqrt(p1)+1-p1, 3);
#ifdef DEBUG_qgamma
REprintf(" v > .32: Wilson-Hilferty; x = %7g\n", x);
#endif
/* starting approximation for p tending to 1 */
if( ch > 2.2*v + 6 )
ch = -2*(R_DT_Clog(p) - c*log(0.5*ch) + g);
} else { /* for v <= 0.32 */
ch = 0.4;
a = R_DT_Clog(p) + g + c*M_LN2;
#ifdef DEBUG_qgamma
REprintf(" v <= .32: a = %7g\n", a);
#endif
do {
q = ch;
p1 = 1. / (1+ch*(C7+ch));
p2 = ch*(C9+ch*(C8+ch));
t = -0.5 +(C7+2*ch)*p1 - (C9+ch*(C10+3*ch))/p2;
ch -= (1- exp(a+0.5*ch)*p2*p1)/t;
} while(fabs(q - ch) > EPS1*fabs(ch));
}
#ifdef DEBUG_qgamma
REprintf("\t==> ch = %10g:", ch);
#endif
/*----- Phase II: Iteration
* Call pgamma() [AS 239] and calculate seven term taylor series
*/
for( i=1 ; i <= MAXIT ; i++ ) {
q = ch;
p1 = 0.5*ch;
p2 = p_ - pgamma(p1, alpha, 1, /*lower_tail*/true, /*log_p*/false);
#ifdef IEEE_754
if(!R_FINITE(p2))
#else
if(errno != 0)
#endif
return numeric_limits<double>::quiet_NaN();
t = p2*exp(alpha*M_LN2+g+p1-c*log(ch));
b = t/ch;
a = 0.5*t - b*c;
s1 = (210+a*(140+a*(105+a*(84+a*(70+60*a))))) * i420;
s2 = (420+a*(735+a*(966+a*(1141+1278*a)))) * i2520;
s3 = (210+a*(462+a*(707+932*a))) * i2520;
s4 = (252+a*(672+1182*a)+c*(294+a*(889+1740*a))) * i5040;
s5 = (84+2264*a+c*(1175+606*a)) * i2520;
s6 = (120+c*(346+127*c)) * i5040;
ch += t*(1+0.5*t*s1-b*c*(s1-b*(s2-b*(s3-b*(s4-b*(s5-b*s6))))));
if(fabs(q - ch) < EPS2*ch)
goto END;
}
ML_ERROR(ME_PRECISION);/* no convergence in MAXIT iterations */
END:
return 0.5*scale*ch;
}
SEXP readBGEN2List(BGenFile* bin) {
// Rprintf("vcfColumn.size() = %u\n", FLAG_vcfColumn.size());
// Rprintf("vcfInfo.size() = %u\n", FLAG_infoTag.size());
// Rprintf("vcfIndv.size() = %u\n", FLAG_indvTag.size());
// also append sample names at the end
// 7: chrom, pos, varId, rsId, alleles, isPhased, prob, sampleId
int retListLen = 8;
if (retListLen == 0) {
return R_NilValue;
}
int numAllocated =
0; // record how many times we allocate (using PROTECT in R);
SEXP ret;
PROTECT(ret = allocVector(VECSXP, retListLen));
numAllocated++;
// store results
std::vector<std::string> idVec;
std::vector<std::string> chrom;
std::vector<int> pos;
std::vector<std::string> varId;
std::vector<std::string> rsId;
std::vector<std::string> alleles;
// std::vector<std::vector<bool> > missing;
std::vector<bool> isPhased;
std::vector<std::vector<double> >
prob; // prob[variant][each_sample * (prob1, prob2, ...)]
// std::map<std::string, std::vector<std::string> > infoMap;
// std::map<std::string, std::vector<std::string> > indvMap;
/// int nRow = 0; // # of positions that will be outputed
// get effective sample names
const int N = bin->getNumSample();
std::vector<std::string> sm = bin->getSampleIdentifier(); // all sample names
std::vector<std::string>& names = idVec;
if (!sm.size()) {
char buf[1024];
for (int i = 0; i < N; ++i) {
sprintf(buf, "sample_%d", i);
sm.push_back(buf);
}
}
const size_t sampleSize = bin->getNumEffectiveSample();
for (size_t i = 0; i != sampleSize; ++i) {
names.push_back(sm[bin->getEffectiveIndex(i)]);
}
// real working part
int nRecord = 0;
const int numProbValues =
3; // if multi-allelic/multi-haploid, this value can be different
int maxProbValues = -1;
while (bin->readRecord()) {
// REprintf("read a record\n");
const BGenVariant& var = bin->getVariant();
const size_t sampleSize = bin->getNumEffectiveSample();
// store results here
nRecord++;
chrom.push_back(var.chrom);
pos.push_back(var.pos);
varId.push_back(var.varid);
rsId.push_back(var.rsid);
alleles.push_back(toString(var.alleles, ","));
isPhased.push_back(var.isPhased);
prob.resize(nRecord);
std::vector<double>& p = prob[nRecord - 1];
p.reserve(sampleSize * numProbValues);
for (size_t i = 0; i != sampleSize; ++i) {
int beg = var.index[bin->getEffectiveIndex(i)];
int end = var.index[bin->getEffectiveIndex(i) + 1];
if (end - beg > maxProbValues) {
maxProbValues = end - beg;
}
for (int j = 0; j < numProbValues; ++j) {
if (j < numProbValues) {
p.push_back(var.prob[beg + j]);
} else {
p.push_back(-9);
}
}
// REprintf("beg = %d, end = %d, prob[%d][%d] len = %d\n", beg,end,
// nRecord - 1, i, p[i].size());
}
// Rprintf("Done add indv\n");
} // end while
if (maxProbValues > numProbValues) {
REprintf("some sample has more than %d > %d probabilities per variant!\n",
maxProbValues, numProbValues);
}
// pass value back to R (see Manual Chapter 5)
std::vector<std::string> listNames;
int retListIdx = 0;
storeResult(chrom, ret, retListIdx++);
storeResult(pos, ret, retListIdx++);
storeResult(varId, ret, retListIdx++);
storeResult(rsId, ret, retListIdx++);
storeResult(alleles, ret, retListIdx++);
storeResult(isPhased, ret, retListIdx++);
storeResult(prob, ret, retListIdx);
for (size_t i = 0; i != prob.size(); ++i) {
SEXP s = VECTOR_ELT(VECTOR_ELT(ret, retListIdx), i);
setDim(numProbValues, sampleSize, s);
}
retListIdx++;
listNames.push_back("chrom");
listNames.push_back("pos");
listNames.push_back("varid");
listNames.push_back("rsid");
listNames.push_back("alleles");
listNames.push_back("isPhased");
listNames.push_back("probability");
// store sample ids
// Rprintf("set sample id");
listNames.push_back("sampleId");
storeResult(idVec, ret, retListIdx++);
// Rprintf("set list names\n");
SEXP sListNames;
PROTECT(sListNames = allocVector(STRSXP, listNames.size()));
numAllocated++;
for (unsigned int i = 0; i != listNames.size(); ++i) {
SET_STRING_ELT(sListNames, i, mkChar(listNames[i].c_str()));
}
setAttrib(ret, R_NamesSymbol, sListNames);
// finish up
UNPROTECT(numAllocated);
// Rprintf("Unprotected: %d\n", (retListLen + 1));
return (ret);
}
// for now still *export* M_Matrix_check_class_etc()
int M_Matrix_check_class_etc(SEXP x, const char **valid)
{
REprintf("M_Matrix_check_class_etc() is deprecated; use R_check_class_etc() instead");
return R_check_class_etc(x, valid);
}
double pnchisq_raw(double x, double f, double theta,
double errmax, double reltol, int itrmax)
{
double ans, lam, u, v, x2, f2, t, term, bound, f_x_2n, f_2n, lt;
double lu = -1., l_lam = -1., l_x = -1.; /* initialized for -Wall */
int n;
Rboolean lamSml, tSml, is_r, is_b, is_it;
static const double _dbl_min_exp = M_LN2 * DBL_MIN_EXP;
/*= -708.3964 for IEEE double precision */
if (x <= 0.) return 0.;
if(!R_FINITE(x)) return 1.;
#ifdef DEBUG_pnch
REprintf("pnchisq(x=%g, f=%g, theta=%g): ",x,f,theta);
#endif
lam = .5 * theta;
lamSml = (-lam < _dbl_min_exp);
if(lamSml) {
/* MATHLIB_ERROR(
"non centrality parameter (= %g) too large for current algorithm",
theta) */
u = 0;
lu = -lam;/* == ln(u) */
l_lam = log(lam);
} else {
u = exp(-lam);
}
/* evaluate the first term */
v = u;
x2 = .5 * x;
f2 = .5 * f;
f_x_2n = f - x;
#ifdef DEBUG_pnch
REprintf("-- v=exp(-th/2)=%g, x/2= %g, f/2= %g\n",v,x2,f2);
#endif
if(f2 * DBL_EPSILON > 0.125 && /* very large f and x ~= f: probably needs */
fabs(t = x2 - f2) < /* other algorithm anyway */
sqrt(DBL_EPSILON) * f2) {
/* evade cancellation error */
/* t = exp((1 - t)*(2 - t/(f2 + 1))) / sqrt(2*M_PI*(f2 + 1));*/
lt = (1 - t)*(2 - t/(f2 + 1)) - 0.5 * log(2*M_PI*(f2 + 1));
#ifdef DEBUG_pnch
REprintf(" (case I) ==> ");
#endif
}
else {
/* Usual case 2: careful not to overflow .. : */
lt = f2*log(x2) -x2 - lgammafn(f2 + 1);
}
#ifdef DEBUG_pnch
REprintf(" lt= %g", lt);
#endif
tSml = (lt < _dbl_min_exp);
if(tSml) {
if (x > f + theta + 5* sqrt( 2*(f + 2*theta))) {
/* x > E[X] + 5* sigma(X) */
return 1.; /* better than 0 --- but definitely "FIXME" */
} /* else */
l_x = log(x);
ans = term = t = 0.;
}
else {
t = exp(lt);
#ifdef DEBUG_pnch
REprintf(", t=exp(lt)= %g\n", t);
#endif
ans = term = v * t;
}
for (n = 1, f_2n = f + 2., f_x_2n += 2.; ; n++, f_2n += 2, f_x_2n += 2) {
#ifdef DEBUG_pnch
REprintf("\n _OL_: n=%d",n);
#endif
/* f_2n === f + 2*n
* f_x_2n === f - x + 2*n > 0 <==> (f+2n) > x */
if (f_x_2n > 0) {
/* find the error bound and check for convergence */
bound = t * x / f_x_2n;
#ifdef DEBUG_pnch
REprintf("\n L10: n=%d; term= %g; bound= %g",n,term,bound);
#endif
is_r = is_it = FALSE;
/* convergence only if BOTH absolute and relative error < 'bnd' */
if (((is_b = (bound <= errmax)) &&
(is_r = (term <= reltol * ans))) || (is_it = (n > itrmax)))
{
#ifdef DEBUG_pnch
REprintf("BREAK n=%d %s; bound= %g %s, rel.err= %g %s\n",
n, (is_it ? "> itrmax" : ""),
bound, (is_b ? "<= errmax" : ""),
term/ans, (is_r ? "<= reltol" : ""));
#endif
break; /* out completely */
}
}
/* evaluate the next term of the */
/* expansion and then the partial sum */
if(lamSml) {
lu += l_lam - log(n); /* u = u* lam / n */
if(lu >= _dbl_min_exp) {
/* no underflow anymore ==> change regime */
#ifdef DEBUG_pnch
REprintf(" n=%d; nomore underflow in u = exp(lu) ==> change\n",
n);
#endif
v = u = exp(lu); /* the first non-0 'u' */
lamSml = FALSE;
}
} else {
u *= lam / n;
v += u;
}
if(tSml) {
lt += l_x - log(f_2n);/* t <- t * (x / f2n) */
if(lt >= _dbl_min_exp) {
/* no underflow anymore ==> change regime */
#ifdef DEBUG_pnch
REprintf(" n=%d; nomore underflow in t = exp(lt) ==> change\n",
n);
#endif
t = exp(lt); /* the first non-0 't' */
tSml = FALSE;
}
} else {
t *= x / f_2n;
}
if(!lamSml && !tSml) {
term = v * t;
ans += term;
}
} /* for(n ...) */
if (is_it) {
MATHLIB_WARNING2(_("pnchisq(x=%g, ..): not converged in %d iter."),
x, itrmax);
}
#ifdef DEBUG_pnch
REprintf("\n == L_End: n=%d; term= %g; bound=%g\n",n,term,bound);
#endif
return (ans);
}
////////////////////////////////////////
// getHists - Get Histograms
SEXP getHistsR(SEXP fileForHists, SEXP histNames, SEXP directoryR)
{
TFile* f = checkForFileForHistsWrapper(fileForHists);
const char* oldDirectory = setFileDirectory(f, directoryR);
// Make the return list
SEXP ans, ansNames;
PROTECT( ans = NEW_LIST( GET_LENGTH(histNames) ) );
PROTECT( ansNames = NEW_STRING( GET_LENGTH(histNames) ) );
// Loop over the names
for ( unsigned int i = 0; i < GET_LENGTH(histNames); ++i ) {
// Get the name of the object
std::string name = CHAR( STRING_ELT(histNames, i) );
// Set the list name
SET_STRING_ELT( ansNames, i, mkChar(name.c_str()) );
// What is this thing?
TKey* key = gDirectory->FindKey(name.c_str());
// If not found, then skip
if ( ! key ) continue;
// The object is real! - Get the class name
const char* className = key->GetClassName();
// Do certain things depending on the class name
// For right now, we'll just do TH1F's and TH1D's. Maybe do more later.
if ( strcmp(className, "TH1F") == 0 || strcmp(className, "TH1D") == 0 ) {
// Get the histogram
TH1* hist;
gDirectory->GetObject(name.c_str(), hist);
// For TH1F, there are 6 elements to the list (name, type, title,
// breaks, counts,
// uncert,
// mids, xname, underOverFlows, mean, rms, yname)
SET_ELEMENT( ans, i, NEW_LIST(12) );
SEXP data = VECTOR_ELT(ans, i);
SEXP dataNames;
PROTECT( dataNames = NEW_STRING(12) );
// Add the type
unsigned int j = 0;
SEXP type = addCharVector(data, dataNames, j++, 1, "type");
SET_STRING_ELT(type, 0, mkChar(className));
// Add the name
SEXP theName = addCharVector(data, dataNames, j++, 1, "name");
SET_STRING_ELT(theName, 0, mkChar(name.c_str()));
// Add basic histogram information
j = addHistInfo(data, dataNames, j, hist);
// Add the x-axis information
j = addXAxis(data, dataNames, j, hist);
// Add the x-contents information
j = addXContents(data, dataNames, j, hist);
// Add Y-axis name
SEXP yname = addCharVector(data, dataNames, j++, 1, "yname");
SET_STRING_ELT(yname, 0, mkChar( hist->GetYaxis()->GetTitle() ) );
// Done
SET_NAMES(data, dataNames);
UNPROTECT(1); // unprotect dataNames
}
else {
REprintf("!! Do not know how to handle %s of class %s !!\n",
name.c_str(), className);
continue;
}
} // For over names
// Restore the root old directory
if ( ! f->cd(oldDirectory) ) {
error("namesMatchingClass: cd to old directory failed");
}
// return the results
SET_NAMES(ans, ansNames);
UNPROTECT(2);
return ans;
}
// rhyper(NR, NB, n) -- NR 'red', NB 'blue', n drawn, how many are 'red'
double rhyper(double nn1in, double nn2in, double kkin)
{
/* extern double afc(int); */
int nn1, nn2, kk;
int ix; // return value (coerced to double at the very end)
Rboolean setup1, setup2;
/* These should become 'thread_local globals' : */
static int ks = -1, n1s = -1, n2s = -1;
static int m, minjx, maxjx;
static int k, n1, n2; // <- not allowing larger integer par
static double tn;
// II :
static double w;
// III:
static double a, d, s, xl, xr, kl, kr, lamdl, lamdr, p1, p2, p3;
/* check parameter validity */
if(!R_FINITE(nn1in) || !R_FINITE(nn2in) || !R_FINITE(kkin))
ML_ERR_return_NAN;
nn1in = R_forceint(nn1in);
nn2in = R_forceint(nn2in);
kkin = R_forceint(kkin);
if (nn1in < 0 || nn2in < 0 || kkin < 0 || kkin > nn1in + nn2in)
ML_ERR_return_NAN;
if (nn1in >= INT_MAX || nn2in >= INT_MAX || kkin >= INT_MAX) {
/* large n -- evade integer overflow (and inappropriate algorithms)
-------- */
// FIXME: Much faster to give rbinom() approx when appropriate; -> see Kuensch(1989)
// Johnson, Kotz,.. p.258 (top) mention the *four* different binomial approximations
if(kkin == 1.) { // Bernoulli
return rbinom(kkin, nn1in / (nn1in + nn2in));
}
// Slow, but safe: return F^{-1}(U) where F(.) = phyper(.) and U ~ U[0,1]
return qhyper(unif_rand(), nn1in, nn2in, kkin, FALSE, FALSE);
}
nn1 = (int)nn1in;
nn2 = (int)nn2in;
kk = (int)kkin;
/* if new parameter values, initialize */
if (nn1 != n1s || nn2 != n2s) {
setup1 = TRUE; setup2 = TRUE;
} else if (kk != ks) {
setup1 = FALSE; setup2 = TRUE;
} else {
setup1 = FALSE; setup2 = FALSE;
}
if (setup1) {
n1s = nn1;
n2s = nn2;
tn = nn1 + nn2;
if (nn1 <= nn2) {
n1 = nn1;
n2 = nn2;
} else {
n1 = nn2;
n2 = nn1;
}
}
if (setup2) {
ks = kk;
if (kk + kk >= tn) {
k = (int)(tn - kk);
} else {
k = kk;
}
}
if (setup1 || setup2) {
m = (int) ((k + 1.) * (n1 + 1.) / (tn + 2.));
minjx = imax2(0, k - n2);
maxjx = imin2(n1, k);
#ifdef DEBUG_rhyper
REprintf("rhyper(nn1=%d, nn2=%d, kk=%d), setup: floor(mean)= m=%d, jx in (%d..%d)\n",
nn1, nn2, kk, m, minjx, maxjx);
#endif
}
/* generate random variate --- Three basic cases */
if (minjx == maxjx) { /* I: degenerate distribution ---------------- */
#ifdef DEBUG_rhyper
REprintf("rhyper(), branch I (degenerate)\n");
#endif
ix = maxjx;
goto L_finis; // return appropriate variate
} else if (m - minjx < 10) { // II: (Scaled) algorithm HIN (inverse transformation) ----
const static double scale = 1e25; // scaling factor against (early) underflow
const static double con = 57.5646273248511421;
// 25*log(10) = log(scale) { <==> exp(con) == scale }
if (setup1 || setup2) {
double lw; // log(w); w = exp(lw) * scale = exp(lw + log(scale)) = exp(lw + con)
if (k < n2) {
lw = afc(n2) + afc(n1 + n2 - k) - afc(n2 - k) - afc(n1 + n2);
} else {
lw = afc(n1) + afc( k ) - afc(k - n2) - afc(n1 + n2);
}
w = exp(lw + con);
}
double p, u;
#ifdef DEBUG_rhyper
REprintf("rhyper(), branch II; w = %g > 0\n", w);
#endif
L10:
p = w;
ix = minjx;
u = unif_rand() * scale;
#ifdef DEBUG_rhyper
REprintf(" _new_ u = %g\n", u);
#endif
while (u > p) {
u -= p;
p *= ((double) n1 - ix) * (k - ix);
ix++;
p = p / ix / (n2 - k + ix);
#ifdef DEBUG_rhyper
REprintf(" ix=%3d, u=%11g, p=%20.14g (u-p=%g)\n", ix, u, p, u-p);
#endif
if (ix > maxjx)
goto L10;
// FIXME if(p == 0.) we also "have lost" => goto L10
}
} else { /* III : H2PE Algorithm --------------------------------------- */
double u,v;
if (setup1 || setup2) {
s = sqrt((tn - k) * k * n1 * n2 / (tn - 1) / tn / tn);
/* remark: d is defined in reference without int. */
/* the truncation centers the cell boundaries at 0.5 */
d = (int) (1.5 * s) + .5;
xl = m - d + .5;
xr = m + d + .5;
a = afc(m) + afc(n1 - m) + afc(k - m) + afc(n2 - k + m);
kl = exp(a - afc((int) (xl)) - afc((int) (n1 - xl))
- afc((int) (k - xl))
- afc((int) (n2 - k + xl)));
kr = exp(a - afc((int) (xr - 1))
- afc((int) (n1 - xr + 1))
- afc((int) (k - xr + 1))
- afc((int) (n2 - k + xr - 1)));
lamdl = -log(xl * (n2 - k + xl) / (n1 - xl + 1) / (k - xl + 1));
lamdr = -log((n1 - xr + 1) * (k - xr + 1) / xr / (n2 - k + xr));
p1 = d + d;
p2 = p1 + kl / lamdl;
p3 = p2 + kr / lamdr;
}
#ifdef DEBUG_rhyper
REprintf("rhyper(), branch III {accept/reject}: (xl,xr)= (%g,%g); (lamdl,lamdr)= (%g,%g)\n",
xl, xr, lamdl,lamdr);
REprintf("-------- p123= c(%g,%g,%g)\n", p1,p2, p3);
#endif
int n_uv = 0;
L30:
u = unif_rand() * p3;
v = unif_rand();
n_uv++;
if(n_uv >= 10000) {
REprintf("rhyper() branch III: giving up after %d rejections", n_uv);
ML_ERR_return_NAN;
}
#ifdef DEBUG_rhyper
REprintf(" ... L30: new (u=%g, v ~ U[0,1])[%d]\n", u, n_uv);
#endif
if (u < p1) { /* rectangular region */
ix = (int) (xl + u);
} else if (u <= p2) { /* left tail */
ix = (int) (xl + log(v) / lamdl);
if (ix < minjx)
goto L30;
v = v * (u - p1) * lamdl;
} else { /* right tail */
ix = (int) (xr - log(v) / lamdr);
if (ix > maxjx)
goto L30;
v = v * (u - p2) * lamdr;
}
/* acceptance/rejection test */
Rboolean reject = TRUE;
if (m < 100 || ix <= 50) {
/* explicit evaluation */
/* The original algorithm (and TOMS 668) have
f = f * i * (n2 - k + i) / (n1 - i) / (k - i);
in the (m > ix) case, but the definition of the
recurrence relation on p134 shows that the +1 is
needed. */
int i;
double f = 1.0;
if (m < ix) {
for (i = m + 1; i <= ix; i++)
f = f * (n1 - i + 1) * (k - i + 1) / (n2 - k + i) / i;
} else if (m > ix) {
for (i = ix + 1; i <= m; i++)
f = f * i * (n2 - k + i) / (n1 - i + 1) / (k - i + 1);
}
if (v <= f) {
reject = FALSE;
}
} else {
const static double deltal = 0.0078;
const static double deltau = 0.0034;
double e, g, r, t, y;
double de, dg, dr, ds, dt, gl, gu, nk, nm, ub;
double xk, xm, xn, y1, ym, yn, yk, alv;
#ifdef DEBUG_rhyper
REprintf(" ... accept/reject 'large' case v=%g\n", v);
#endif
/* squeeze using upper and lower bounds */
y = ix;
y1 = y + 1.0;
ym = y - m;
yn = n1 - y + 1.0;
yk = k - y + 1.0;
nk = n2 - k + y1;
r = -ym / y1;
s = ym / yn;
t = ym / yk;
e = -ym / nk;
g = yn * yk / (y1 * nk) - 1.0;
dg = 1.0;
if (g < 0.0)
dg = 1.0 + g;
gu = g * (1.0 + g * (-0.5 + g / 3.0));
gl = gu - .25 * (g * g * g * g) / dg;
xm = m + 0.5;
xn = n1 - m + 0.5;
xk = k - m + 0.5;
nm = n2 - k + xm;
ub = y * gu - m * gl + deltau
+ xm * r * (1. + r * (-0.5 + r / 3.0))
+ xn * s * (1. + s * (-0.5 + s / 3.0))
+ xk * t * (1. + t * (-0.5 + t / 3.0))
+ nm * e * (1. + e * (-0.5 + e / 3.0));
/* test against upper bound */
alv = log(v);
if (alv > ub) {
reject = TRUE;
} else {
/* test against lower bound */
dr = xm * (r * r * r * r);
if (r < 0.0)
dr /= (1.0 + r);
ds = xn * (s * s * s * s);
if (s < 0.0)
ds /= (1.0 + s);
dt = xk * (t * t * t * t);
if (t < 0.0)
dt /= (1.0 + t);
de = nm * (e * e * e * e);
if (e < 0.0)
de /= (1.0 + e);
if (alv < ub - 0.25 * (dr + ds + dt + de)
+ (y + m) * (gl - gu) - deltal) {
reject = FALSE;
}
else {
/* * Stirling's formula to machine accuracy
*/
if (alv <= (a - afc(ix) - afc(n1 - ix)
- afc(k - ix) - afc(n2 - k + ix))) {
reject = FALSE;
} else {
reject = TRUE;
}
}
}
} // else
if (reject)
goto L30;
}
L_finis:
/* return appropriate variate */
if (kk + kk >= tn) {
if (nn1 > nn2) {
ix = kk - nn2 + ix;
} else {
ix = nn1 - ix;
}
} else {
if (nn1 > nn2)
ix = kk - ix;
}
return ix;
}
static void *DBFReadAttribute(DBFHandle psDBF, int hEntity, int iField,
char chReqType )
{
int nRecordOffset;
unsigned char *pabyRec;
void *pReturnField = NULL;
static double dDoubleField;
/* -------------------------------------------------------------------- */
/* Verify selection. */
/* -------------------------------------------------------------------- */
if( hEntity < 0 || hEntity >= psDBF->nRecords )
return( NULL );
if( iField < 0 || iField >= psDBF->nFields )
return( NULL );
/* -------------------------------------------------------------------- */
/* Have we read the record? */
/* -------------------------------------------------------------------- */
if( psDBF->nCurrentRecord != hEntity )
{
DBFFlushRecord( psDBF );
nRecordOffset = psDBF->nRecordLength * hEntity + psDBF->nHeaderLength;
if( fseek( psDBF->fp, nRecordOffset, 0 ) != 0 )
{
REprintf("fseek(%d) failed on DBF file", nRecordOffset);
return NULL;
}
if( fread( psDBF->pszCurrentRecord, psDBF->nRecordLength,
1, psDBF->fp ) != 1 )
{
REprintf("fread(%d) failed on DBF file", psDBF->nRecordLength );
return NULL;
}
psDBF->nCurrentRecord = hEntity;
}
pabyRec = (unsigned char *) psDBF->pszCurrentRecord;
/* -------------------------------------------------------------------- */
/* Ensure our field buffer is large enough to hold this buffer. */
/* -------------------------------------------------------------------- */
if( psDBF->panFieldSize[iField]+1 > nStringFieldLen )
{
nStringFieldLen = psDBF->panFieldSize[iField]*2 + 10;
pszStringField = (char *) SfRealloc(pszStringField,nStringFieldLen);
}
/* -------------------------------------------------------------------- */
/* Extract the requested field. */
/* -------------------------------------------------------------------- */
strncpy( pszStringField,
((const char *) pabyRec) + psDBF->panFieldOffset[iField],
psDBF->panFieldSize[iField] );
pszStringField[psDBF->panFieldSize[iField]] = '\0';
pReturnField = pszStringField;
/* -------------------------------------------------------------------- */
/* Decode the field. */
/* -------------------------------------------------------------------- */
if( chReqType == 'N' )
{
dDoubleField = R_atof(pszStringField);
pReturnField = &dDoubleField;
}
/* -------------------------------------------------------------------- */
/* Should we trim white space off the string attribute value? */
/* -------------------------------------------------------------------- */
#ifdef TRIM_DBF_WHITESPACE
else
{
char *pchSrc, *pchDst;
pchDst = pchSrc = pszStringField;
while( *pchSrc == ' ' )
pchSrc++;
while( *pchSrc != '\0' )
*(pchDst++) = *(pchSrc++);
*pchDst = '\0';
while( pchDst != pszStringField && *(--pchDst) == ' ' )
*pchDst = '\0';
}
#endif
return( pReturnField );
}
double qbinom(double p, double n, double pr, int lower_tail, int log_p)
{
double q, mu, sigma, gamma, z, y;
#ifdef IEEE_754
if (ISNAN(p) || ISNAN(n) || ISNAN(pr))
return p + n + pr;
#endif
if(!R_FINITE(n) || !R_FINITE(pr))
ML_ERR_return_NAN;
/* if log_p is true, p = -Inf is a legitimate value */
if(!R_FINITE(p) && !log_p)
ML_ERR_return_NAN;
if(n != floor(n + 0.5)) ML_ERR_return_NAN;
if (pr < 0 || pr > 1 || n < 0)
ML_ERR_return_NAN;
R_Q_P01_boundaries(p, 0, n);
if (pr == 0. || n == 0) return 0.;
q = 1 - pr;
if(q == 0.) return n; /* covers the full range of the distribution */
mu = n * pr;
sigma = sqrt(n * pr * q);
gamma = (q - pr) / sigma;
#ifdef DEBUG_qbinom
REprintf("qbinom(p=%7g, n=%g, pr=%7g, l.t.=%d, log=%d): sigm=%g, gam=%g\n",
p,n,pr, lower_tail, log_p, sigma, gamma);
#endif
/* Note : "same" code in qpois.c, qbinom.c, qnbinom.c --
* FIXME: This is far from optimal [cancellation for p ~= 1, etc]: */
if(!lower_tail || log_p) {
p = R_DT_qIv(p); /* need check again (cancellation!): */
if (p == 0.) return 0.;
if (p == 1.) return n;
}
/* temporary hack --- FIXME --- */
if (p + 1.01*DBL_EPSILON >= 1.) return n;
/* y := approx.value (Cornish-Fisher expansion) : */
z = qnorm(p, 0., 1., /*lower_tail*/TRUE, /*log_p*/FALSE);
y = floor(mu + sigma * (z + gamma * (z*z - 1) / 6) + 0.5);
if(y > n) /* way off */ y = n;
#ifdef DEBUG_qbinom
REprintf(" new (p,1-p)=(%7g,%7g), z=qnorm(..)=%7g, y=%5g\n", p, 1-p, z, y);
#endif
z = pbinom(y, n, pr, /*lower_tail*/TRUE, /*log_p*/FALSE);
/* fuzz to ensure left continuity: */
p *= 1 - 64*DBL_EPSILON;
if(n < 1e5) return do_search(y, &z, p, n, pr, 1);
/* Otherwise be a bit cleverer in the search */
{
double incr = floor(n * 0.001), oldincr;
do {
oldincr = incr;
y = do_search(y, &z, p, n, pr, incr);
incr = fmax2(1, floor(incr/100));
} while(oldincr > 1 && incr > n*1e-15);
return y;
}
}
/* Continued fraction for calculation of
* scaled upper-tail F_{gamma}
* ~= (y / d) * [1 + (1-y)/d + O( ((1-y)/d)^2 ) ]
*/
static double
pd_lower_cf (double y, double d)
{
double f= 0.0 /* -Wall */, of, f0;
double i, c2, c3, c4, a1, b1, a2, b2;
#define NEEDED_SCALE \
(b2 > scalefactor) { \
a1 /= scalefactor; \
b1 /= scalefactor; \
a2 /= scalefactor; \
b2 /= scalefactor; \
}
#define max_it 200000
#ifdef DEBUG_p
REprintf("pd_lower_cf(y=%.14g, d=%.14g)", y, d);
#endif
if (y == 0) return 0;
f0 = y/d;
/* Needed, e.g. for pgamma(10^c(100,295), shape= 1.1, log=TRUE): */
if(fabs(y - 1) < fabs(d) * DBL_EPSILON) { /* includes y < d = Inf */
#ifdef DEBUG_p
REprintf(" very small 'y' -> returning (y/d)\n");
#endif
return (f0);
}
if(f0 > 1.) f0 = 1.;
c2 = y;
c4 = d; /* original (y,d), *not* potentially scaled ones!*/
a1 = 0; b1 = 1;
a2 = y; b2 = d;
while NEEDED_SCALE
i = 0; of = -1.; /* far away */
while (i < max_it) {
i++; c2--; c3 = i * c2; c4 += 2;
/* c2 = y - i, c3 = i(y - i), c4 = d + 2i, for i odd */
a1 = c4 * a2 + c3 * a1;
b1 = c4 * b2 + c3 * b1;
i++; c2--; c3 = i * c2; c4 += 2;
/* c2 = y - i, c3 = i(y - i), c4 = d + 2i, for i even */
a2 = c4 * a1 + c3 * a2;
b2 = c4 * b1 + c3 * b2;
if NEEDED_SCALE
if (b2 != 0) {
f = a2 / b2;
/* convergence check: relative; "absolute" for very small f : */
if (fabs (f - of) <= DBL_EPSILON * fmax2(f0, fabs(f))) {
#ifdef DEBUG_p
REprintf(" %g iter.\n", i);
#endif
return f;
}
of = f;
}
}
MATHLIB_WARNING(" ** NON-convergence in pgamma()'s pd_lower_cf() f= %g.\n",
f);
return f;/* should not happen ... */
} /* pd_lower_cf() */
double attribute_hidden
pnchisq_raw(double x, double f, double theta,
double errmax, double reltol, int itrmax, Rboolean lower_tail)
{
double lam, x2, f2, term, bound, f_x_2n, f_2n;
double l_lam = -1., l_x = -1.; /* initialized for -Wall */
int n;
Rboolean lamSml, tSml, is_r, is_b, is_it;
LDOUBLE ans, u, v, t, lt, lu =-1;
static const double _dbl_min_exp = M_LN2 * DBL_MIN_EXP;
/*= -708.3964 for IEEE double precision */
if (x <= 0.) {
if(x == 0. && f == 0.)
return lower_tail ? exp(-0.5*theta) : -expm1(-0.5*theta);
/* x < 0 or {x==0, f > 0} */
return lower_tail ? 0. : 1.;
}
if(!R_FINITE(x)) return lower_tail ? 1. : 0.;
/* This is principally for use from qnchisq */
#ifndef MATHLIB_STANDALONE
R_CheckUserInterrupt();
#endif
if(theta < 80) { /* use 110 for Inf, as ppois(110, 80/2, lower.tail=FALSE) is 2e-20 */
LDOUBLE sum = 0, sum2 = 0, lambda = 0.5*theta,
pr = EXP(-lambda); // does this need a feature test?
double ans;
int i;
/* we need to renormalize here: the result could be very close to 1 */
for(i = 0; i < 110; pr *= lambda/++i) {
sum2 += pr;
sum += pr * pchisq(x, f+2*i, lower_tail, FALSE);
if (sum2 >= 1-1e-15) break;
}
ans = (double) (sum/sum2);
return ans;
}
#ifdef DEBUG_pnch
REprintf("pnchisq(x=%g, f=%g, theta=%g): ",x,f,theta);
#endif
lam = .5 * theta;
lamSml = (-lam < _dbl_min_exp);
if(lamSml) {
/* MATHLIB_ERROR(
"non centrality parameter (= %g) too large for current algorithm",
theta) */
u = 0;
lu = -lam;/* == ln(u) */
l_lam = log(lam);
} else {
u = exp(-lam);
}
/* evaluate the first term */
v = u;
x2 = .5 * x;
f2 = .5 * f;
f_x_2n = f - x;
#ifdef DEBUG_pnch
REprintf("-- v=exp(-th/2)=%g, x/2= %g, f/2= %g\n",v,x2,f2);
#endif
if(f2 * DBL_EPSILON > 0.125 && /* very large f and x ~= f: probably needs */
FABS(t = x2 - f2) < /* another algorithm anyway */
sqrt(DBL_EPSILON) * f2) {
/* evade cancellation error */
/* t = exp((1 - t)*(2 - t/(f2 + 1))) / sqrt(2*M_PI*(f2 + 1));*/
lt = (1 - t)*(2 - t/(f2 + 1)) - 0.5 * log(2*M_PI*(f2 + 1));
#ifdef DEBUG_pnch
REprintf(" (case I) ==> ");
#endif
}
else {
/* Usual case 2: careful not to overflow .. : */
lt = f2*log(x2) -x2 - lgammafn(f2 + 1);
}
#ifdef DEBUG_pnch
REprintf(" lt= %g", lt);
#endif
tSml = (lt < _dbl_min_exp);
if(tSml) {
if (x > f + theta + 5* sqrt( 2*(f + 2*theta))) {
/* x > E[X] + 5* sigma(X) */
return lower_tail ? 1. : 0.; /* FIXME: We could be more accurate than 0. */
} /* else */
l_x = log(x);
ans = term = 0.; t = 0;
}
else {
t = EXP(lt);
#ifdef DEBUG_pnch
REprintf(", t=exp(lt)= %g\n", t);
#endif
ans = term = (double) (v * t);
}
for (n = 1, f_2n = f + 2., f_x_2n += 2.; ; n++, f_2n += 2, f_x_2n += 2) {
#ifdef DEBUG_pnch
REprintf("\n _OL_: n=%d",n);
#endif
#ifndef MATHLIB_STANDALONE
if(n % 1000) R_CheckUserInterrupt();
#endif
/* f_2n === f + 2*n
* f_x_2n === f - x + 2*n > 0 <==> (f+2n) > x */
if (f_x_2n > 0) {
/* find the error bound and check for convergence */
bound = (double) (t * x / f_x_2n);
#ifdef DEBUG_pnch
REprintf("\n L10: n=%d; term= %g; bound= %g",n,term,bound);
#endif
is_r = is_it = FALSE;
/* convergence only if BOTH absolute and relative error < 'bnd' */
if (((is_b = (bound <= errmax)) &&
(is_r = (term <= reltol * ans))) || (is_it = (n > itrmax)))
{
#ifdef DEBUG_pnch
REprintf("BREAK n=%d %s; bound= %g %s, rel.err= %g %s\n",
n, (is_it ? "> itrmax" : ""),
bound, (is_b ? "<= errmax" : ""),
term/ans, (is_r ? "<= reltol" : ""));
#endif
break; /* out completely */
}
}
/* evaluate the next term of the */
/* expansion and then the partial sum */
if(lamSml) {
lu += l_lam - log(n); /* u = u* lam / n */
if(lu >= _dbl_min_exp) {
/* no underflow anymore ==> change regime */
#ifdef DEBUG_pnch
REprintf(" n=%d; nomore underflow in u = exp(lu) ==> change\n",
n);
#endif
v = u = EXP(lu); /* the first non-0 'u' */
lamSml = FALSE;
}
} else {
u *= lam / n;
v += u;
}
if(tSml) {
lt += l_x - log(f_2n);/* t <- t * (x / f2n) */
if(lt >= _dbl_min_exp) {
/* no underflow anymore ==> change regime */
#ifdef DEBUG_pnch
REprintf(" n=%d; nomore underflow in t = exp(lt) ==> change\n",
n);
#endif
t = EXP(lt); /* the first non-0 't' */
tSml = FALSE;
}
} else {
t *= x / f_2n;
}
if(!lamSml && !tSml) {
term = (double) (v * t);
ans += term;
}
} /* for(n ...) */
if (is_it) {
MATHLIB_WARNING2(_("pnchisq(x=%g, ..): not converged in %d iter."),
x, itrmax);
}
#ifdef DEBUG_pnch
REprintf("\n == L_End: n=%d; term= %g; bound=%g\n",n,term,bound);
#endif
return (double) (lower_tail ? ans : 1 - ans);
}
/*
* Asymptotic expansion to calculate the probability that Poisson variate
* has value <= x.
* Various assertions about this are made (without proof) at
* http://members.aol.com/iandjmsmith/PoissonApprox.htm
*/
static double
ppois_asymp (double x, double lambda, int lower_tail, int log_p)
{
static const double coefs_a[8] = {
-1e99, /* placeholder used for 1-indexing */
2/3.,
-4/135.,
8/2835.,
16/8505.,
-8992/12629925.,
-334144/492567075.,
698752/1477701225.
};
static const double coefs_b[8] = {
-1e99, /* placeholder */
1/12.,
1/288.,
-139/51840.,
-571/2488320.,
163879/209018880.,
5246819/75246796800.,
-534703531/902961561600.
};
double elfb, elfb_term;
double res12, res1_term, res1_ig, res2_term, res2_ig;
double dfm, pt_, s2pt, f, np;
int i;
dfm = lambda - x;
/* If lambda is large, the distribution is highly concentrated
about lambda. So representation error in x or lambda can lead
to arbitrarily large values of pt_ and hence divergence of the
coefficients of this approximation.
*/
pt_ = - log1pmx (dfm / x);
s2pt = sqrt (2 * x * pt_);
if (dfm < 0) s2pt = -s2pt;
res12 = 0;
res1_ig = res1_term = sqrt (x);
res2_ig = res2_term = s2pt;
for (i = 1; i < 8; i++) {
res12 += res1_ig * coefs_a[i];
res12 += res2_ig * coefs_b[i];
res1_term *= pt_ / i ;
res2_term *= 2 * pt_ / (2 * i + 1);
res1_ig = res1_ig / x + res1_term;
res2_ig = res2_ig / x + res2_term;
}
elfb = x;
elfb_term = 1;
for (i = 1; i < 8; i++) {
elfb += elfb_term * coefs_b[i];
elfb_term /= x;
}
if (!lower_tail) elfb = -elfb;
#ifdef DEBUG_p
REprintf ("res12 = %.14g elfb=%.14g\n", elfb, res12);
#endif
f = res12 / elfb;
np = pnorm (s2pt, 0.0, 1.0, !lower_tail, log_p);
if (log_p) {
double n_d_over_p = dpnorm (s2pt, !lower_tail, np);
#ifdef DEBUG_p
REprintf ("pp*_asymp(): f=%.14g np=e^%.14g nd/np=%.14g f*nd/np=%.14g\n",
f, np, n_d_over_p, f * n_d_over_p);
#endif
return np + log1p (f * n_d_over_p);
} else {
double nd = dnorm (s2pt, 0., 1., log_p);
#ifdef DEBUG_p
REprintf ("pp*_asymp(): f=%.14g np=%.14g nd=%.14g f*nd=%.14g\n",
f, np, nd, f * nd);
#endif
return np + f * nd;
}
} /* ppois_asymp() */
double qn0(double *x, int n)
{
/*--------------------------------------------------------------------
Efficient algorithm for the scale estimator:
Q*_n = { |x_i - x_j|; i<j }_(k) [= Qn without scaling ]
i.e. the k-th order statistic of the |x_i - x_j|
Parameters of the function Qn :
x : double array containing the observations
n : number of observations (n >=2)
*/
double *y = (double *)R_alloc(n, sizeof(double));
double *work = (double *)R_alloc(n, sizeof(double));
double *a_srt = (double *)R_alloc(n, sizeof(double));
double *a_cand = (double *)R_alloc(n, sizeof(double));
int *left = (int *)R_alloc(n, sizeof(int));
int *right = (int *)R_alloc(n, sizeof(int));
int *p = (int *)R_alloc(n, sizeof(int));
int *q = (int *)R_alloc(n, sizeof(int));
int *weight = (int *)R_alloc(n, sizeof(int));
double trial = R_NaReal;/* -Wall */
Rboolean found;
int h, i, j,jj,jh;
/* Following should be `long long int' : they can be of order n^2 */
int64_t k, knew, nl,nr, sump,sumq;
h = n / 2 + 1;
k = (int64_t)h * (h - 1) / 2;
for (i = 0; i < n; ++i) {
y[i] = x[i];
left [i] = n - i + 1;
right[i] = (i <= h) ? n : n - (i - h);
/* the n - (i-h) is from the paper; original code had `n' */
}
R_qsort(y, 1, n); /* y := sort(x) */
nl = (int64_t)n * (n + 1) / 2;
nr = (int64_t)n * n;
knew = k + nl;/* = k + (n+1 \over 2) */
found = FALSE;
#ifdef DEBUG_qn
REprintf("qn0(): h,k= %2d,%2d; nl,nr= %d,%d\n", h,k, nl,nr);
#endif
/* L200: */
while(!found && nr - nl > n) {
j = 0;
/* Truncation to float :
try to make sure that the same values are got later (guard bits !) */
for (i = 1; i < n; ++i) {
if (left[i] <= right[i]) {
weight[j] = right[i] - left[i] + 1;
jh = left[i] + weight[j] / 2;
work[j] = (float)(y[i] - y[n - jh]);
++j;
}
}
trial = whimed_i(work, weight, j, a_cand, a_srt, /*iw_cand*/ p);
#ifdef DEBUG_qn
REprintf(" ..!found: whimed(");
# ifdef DEBUG_long
REprintf("wrk=c(");
for(i=0; i < j; i++) REprintf("%s%g", (i>0)? ", " : "", work[i]);
REprintf("),\n wgt=c(");
for(i=0; i < j; i++) REprintf("%s%d", (i>0)? ", " : "", weight[i]);
REprintf("), j= %3d) -> trial= %7g\n", j, trial);
# else
REprintf("j=%3d) -> trial= %g:", j, trial);
# endif
#endif
j = 0;
for (i = n - 1; i >= 0; --i) {
while (j < n && ((float)(y[i] - y[n - j - 1])) < trial)
++j;
p[i] = j;
}
#ifdef DEBUG_qn
REprintf(" f_1: j=%2d", j);
#endif
j = n + 1;
for (i = 0; i < n; ++i) {
while ((float)(y[i] - y[n - j + 1]) > trial)
--j;
q[i] = j;
}
sump = 0;
sumq = 0;
for (i = 0; i < n; ++i) {
sump += p[i];
sumq += q[i] - 1;
}
#ifdef DEBUG_qn
REprintf(" f_2 -> j=%2d, sump|q= %lld,%lld", j, sump,sumq);
#endif
if (knew <= sump) {
for (i = 0; i < n; ++i)
right[i] = p[i];
nr = sump;
#ifdef DEBUG_qn
REprintf("knew <= sump =: nr , new right[]\n");
#endif
} else if (knew > sumq) {
for (i = 0; i < n; ++i)
left[i] = q[i];
nl = sumq;
#ifdef DEBUG_qn
REprintf("knew > sumq =: nl , new left[]\n");
#endif
} else { /* sump < knew <= sumq */
found = TRUE;
#ifdef DEBUG_qn
REprintf("sump < knew <= sumq ---> FOUND\n");
#endif
}
} /* while */
if (found)
return trial;
else {
#ifdef DEBUG_qn
REprintf(".. not fnd -> new work[]");
#endif
j = 0;
for (i = 1; i < n; ++i) {
for (jj = left[i]; jj <= right[i]; ++jj) {
work[j] = y[i] - y[n - jj];
j++;
}/* j will be = sum_{i=2}^n (right[i] - left[i] + 1)_{+} */
}
#ifdef DEBUG_qn
REprintf(" of length %d; knew-nl=%d\n", j, knew-nl);
#endif
/* return pull(work, j - 1, knew - nl) : */
knew -= (nl + 1); /* -1: 0-indexing */
rPsort(work, j, knew);
return(work[knew]);
}
} /* qn0 */
double pgamma_raw (double x, double alph, int lower_tail, int log_p)
{
/* Here, assume that (x,alph) are not NA & alph > 0 . */
double res;
#ifdef DEBUG_p
REprintf("pgamma_raw(x=%.14g, alph=%.14g, low=%d, log=%d)\n",
x, alph, lower_tail, log_p);
#endif
R_P_bounds_01(x, 0., ML_POSINF);
if (x < 1) {
res = pgamma_smallx (x, alph, lower_tail, log_p);
} else if (x <= alph - 1 && x < 0.8 * (alph + 50)) {
/* incl. large alph compared to x */
double sum = pd_upper_series (x, alph, log_p);/* = x/alph + o(x/alph) */
double d = dpois_wrap (alph, x, log_p);
#ifdef DEBUG_p
REprintf(" alph 'large': sum=pd_upper*()= %.12g, d=dpois_w(*)= %.12g\n",
sum, d);
#endif
if (!lower_tail)
res = log_p
? R_Log1_Exp (d + sum)
: 1 - d * sum;
else
res = log_p ? sum + d : sum * d;
} else if (alph - 1 < x && alph < 0.8 * (x + 50)) {
/* incl. large x compared to alph */
double sum;
double d = dpois_wrap (alph, x, log_p);
#ifdef DEBUG_p
REprintf(" x 'large': d=dpois_w(*)= %.14g ", d);
#endif
if (alph < 1) {
if (x * DBL_EPSILON > 1 - alph)
sum = R_D__1;
else {
double f = pd_lower_cf (alph, x - (alph - 1)) * x / alph;
/* = [alph/(x - alph+1) + o(alph/(x-alph+1))] * x/alph = 1 + o(1) */
sum = log_p ? log (f) : f;
}
} else {
sum = pd_lower_series (x, alph - 1);/* = (alph-1)/x + o((alph-1)/x) */
sum = log_p ? log1p (sum) : 1 + sum;
}
#ifdef DEBUG_p
REprintf(", sum= %.14g\n", sum);
#endif
if (!lower_tail)
res = log_p ? sum + d : sum * d;
else
res = log_p
? R_Log1_Exp (d + sum)
: 1 - d * sum;
} else { /* x >= 1 and x fairly near alph. */
#ifdef DEBUG_p
REprintf(" using ppois_asymp()\n");
#endif
res = ppois_asymp (alph - 1, x, !lower_tail, log_p);
}
/*
* We lose a fair amount of accuracy to underflow in the cases
* where the final result is very close to DBL_MIN. In those
* cases, simply redo via log space.
*/
if (!log_p && res < DBL_MIN / DBL_EPSILON) {
/* with(.Machine, double.xmin / double.eps) #|-> 1.002084e-292 */
#ifdef DEBUG_p
REprintf(" very small res=%.14g; -> recompute via log\n", res);
#endif
return exp (pgamma_raw (x, alph, lower_tail, 1));
} else
return res;
}