int main(void)
{
//
#ifdef T_addBigInt
string num1("123");
string num2("4567");
string sum = addBigInt(num1, num2);
if (sum == "")
{
cerr << "error in addBigInt" << endl;
return -1;
}
cout << num1 << " + " << num2 << " = " << sum << endl;
#endif
#ifdef T_mulOneBit
BigFloat bigNum;
cout << "enter a big float: ";
cin >> bigNum;
cout << "digits = " << bigNum.digits << endl;
cout << "exp = " << bigNum.exp << endl;
string resu(bigNum.mulOneBit('0'));
if ("" == resu)
{
cerr << "mulOneBit error" << endl;
return -1;
}
cout << "result is " << resu << endl;
#endif
#ifdef T_opMul
BigFloat num1(100);
BigFloat num2;
cout << "Enter a number : ";
cin >> num2;
BigFloat prdt = num1 * num2;
cout << "result is : \n" << prdt << endl;
#endif
return 0;
}
int BigFloat::ucmp(const BigFloat &x) const{
// Compare function that ignores the sign.
// This is needed to determine which direction subtractions will go.
// Magnitude
int64_t magA = exp + L;
int64_t magB = x.exp + x.L;
if (magA > magB)
return 1;
if (magA < magB)
return -1;
// Compare
int64_t mag = magA;
while (mag >= exp || mag >= x.exp){
uint32_t wordA = word_at(mag);
uint32_t wordB = x.word_at(mag);
if (wordA < wordB)
return -1;
if (wordA > wordB)
return 1;
mag--;
}
return 0;
}
BigFloat BigFloat::usub(const BigFloat &x,size_t p) const{
// Perform subtraction ignoring the sign of the two operands.
// "this" must be greater than or equal to x. Otherwise, the behavior
// is undefined.
// Magnitude
int64_t magA = exp + L;
int64_t magB = x.exp + x.L;
int64_t top = std::max(magA,magB);
int64_t bot = std::min(exp,x.exp);
// Truncate precision
int64_t TL = top - bot;
if (p == 0){
// Default value. No trunction.
p = (size_t)TL;
}else{
// Increase precision
p += YCL_BIGFLOAT_EXTRA_PRECISION;
}
if (TL > (int64_t)p){
bot = top - p;
TL = p;
}
// Compute basic fields.
BigFloat z;
z.sign = sign;
z.exp = bot;
z.L = (uint32_t)TL;
// Allocate mantissa
z.T = std::unique_ptr<uint32_t[]>(new uint32_t[z.L]);
// Subtract
int32_t carry = 0;
for (size_t c = 0; bot < top; bot++, c++){
int32_t word = (int32_t)word_at(bot) - (int32_t)x.word_at(bot) - carry;
carry = 0;
if (word < 0){
word += 1000000000;
carry = 1;
}
z.T[c] = word;
}
// Strip leading zeros
while (z.L > 0 && z.T[z.L - 1] == 0)
z.L--;
if (z.L == 0){
z.exp = 0;
z.sign = true;
z.T.reset();
}
return z;
}
inline BigFloat Epsilon<BigFloat>( const BigFloat& alpha )
{
// NOTE: This 'precision' is the number of bits in the mantissa
auto p = alpha.Precision();
// epsilon = b^{-(p-1)} / 2 = 2^{-p}
BigFloat one(1,p);
return one >> p;
}
inline BigFloat SafeMin( const BigFloat& alpha )
{
// TODO: Decide how to only recompute this when the precision changes
const BigFloat one = BigFloat(1,alpha.Precision());
const BigFloat eps = Epsilon(alpha);
const BigFloat minVal = Min(alpha);
const BigFloat invMax = one/Max(alpha);
return ( invMax>minVal ? invMax*(one+eps) : minVal );
}
BigFloat invsqrt(uint32_t x,size_t p,int tds){
// Compute inverse square root using Newton's Method.
// ( r0^2 * x - 1 )
// r1 = r0 - (----------------) * r0
// ( 2 )
if (x == 0)
throw "Divide by Zero";
// End of recursion. Generate starting point.
if (p == 0){
double val = 1. / sqrt((double)x);
int64_t exponent = 0;
// Scale
while (val < 1000000000.){
val *= 1000000000.;
exponent--;
}
// Rebuild a BigFloat.
uint64_t val64 = (uint64_t)val;
BigFloat out;
out.sign = true;
out.T = std::unique_ptr<uint32_t[]>(new uint32_t[2]);
out.T[0] = (uint32_t)(val64 % 1000000000);
out.T[1] = (uint32_t)(val64 / 1000000000);
out.L = 2;
out.exp = exponent;
return out;
}
// Half the precision
size_t s = p / 2 + 1;
if (p == 1) s = 0;
if (p == 2) s = 1;
// Recurse at half the precision
BigFloat T = invsqrt(x,s,tds);
BigFloat temp = T.mul(T,p); // r0^2
temp = temp.mul(x,p,tds); // r0^2 * x
temp = temp.sub(BigFloat(1),p); // r0^2 * x - 1
temp = temp.mul(500000000); // (r0^2 * x - 1) / 2
temp.exp--;
temp = temp.mul(T,p,tds); // (r0^2 * x - 1) / 2 * r0
return T.sub(temp,p); // r0 - (r0^2 * x - 1) / 2 * r0
}
BigFloat BigFloat::sub(const BigFloat &x,size_t p) const{
// Subtraction
// The target precision is p.
// If (p = 0), then no truncation is done. The entire operation is done
// at maximum precision with no data loss.
// Different sign. Add.
if (sign != x.sign)
return uadd(x,p);
// this > x
if (ucmp(x) > 0)
return usub(x,p);
// this < x
BigFloat z = x.usub(*this,p);
z.negate();
return z;
}
int main()
{
BigFloat *sum;
sum = new BigFloat(0.0);
for ( int i = 1; i <= 100 ; i ++)
{
BigFloat index(1.0);
BigFloat temp((double)i);
if ( 0 == index.divide(&index, &temp))
sum->add(sum,&index);
else
{
printf("error");
exit(1);
}
}
sum->printAll();
sum->print_fraction_portion_in_decimal(1000,NULL);
}
BigFloat BigFloat::uadd(const BigFloat &x,size_t p) const{
// Perform addition ignoring the sign of the two operands.
// Magnitude
int64_t magA = exp + L;
int64_t magB = x.exp + x.L;
int64_t top = std::max(magA,magB);
int64_t bot = std::min(exp,x.exp);
// Target length
int64_t TL = top - bot;
if (p == 0){
// Default value. No trunction.
p = (size_t)TL;
}else{
// Increase precision
p += YCL_BIGFLOAT_EXTRA_PRECISION;
}
// Perform precision truncation.
if (TL > (int64_t)p){
bot = top - p;
TL = p;
}
// Compute basic fields.
BigFloat z;
z.sign = sign;
z.exp = bot;
z.L = (uint32_t)TL;
// Allocate mantissa
z.T = std::unique_ptr<uint32_t[]>(new uint32_t[z.L + 1]);
// Add
uint32_t carry = 0;
for (size_t c = 0; bot < top; bot++, c++){
uint32_t word = word_at(bot) + x.word_at(bot) + carry;
carry = 0;
if (word >= 1000000000){
word -= 1000000000;
carry = 1;
}
z.T[c] = word;
}
// Carry out
if (carry != 0){
z.T[z.L++] = 1;
}
return z;
}
////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
// Pi
void Pi_BSR(BigFloat &P,BigFloat &Q,BigFloat &R,uint32_t a,uint32_t b,size_t p){
// Binary Splitting recursion for the Chudnovsky Formula.
if (b - a == 1){
// P = (13591409 + 545140134 b)(2b-1)(6b-5)(6b-1) (-1)^b
P = BigFloat(b).mul(545140134);
P = P.add(BigFloat(13591409));
P = P.mul(2*b - 1);
P = P.mul(6*b - 5);
P = P.mul(6*b - 1);
if (b % 2 == 1)
P.negate();
// Q = 10939058860032000 * b^3
Q = BigFloat(b);
Q = Q.mul(Q).mul(Q).mul(26726400).mul(409297880);
// R = (2b-1)(6b-5)(6b-1)
R = BigFloat(2*b - 1);
R = R.mul(6*b - 5);
R = R.mul(6*b - 1);
return;
}
uint32_t m = (a + b) / 2;
BigFloat P0,Q0,R0,P1,Q1,R1;
Pi_BSR(P0,Q0,R0,a,m,p);
Pi_BSR(P1,Q1,R1,m,b,p);
P = P0.mul(Q1,p).add(P1.mul(R0,p),p);
Q = Q0.mul(Q1,p);
R = R0.mul(R1,p);
}
bool GaussElimination::changeValuesInRowByBigFloat(vector<vector<BigFloat>>& matrix, int indexRow)
{
if( matrix[indexRow][indexRow].compare( &BigFloat(0.0) ) != 0 ) return true;
// wenn i.zeile zahl in spalte i = 0, tausche mit zeile darunter!!, wo zahl in i.spalte != 0 -> falls nicht da, gs nicht lösbar
for(unsigned int rowCounter = indexRow+1; rowCounter < row; ++rowCounter)
{
BigFloat value = matrix[rowCounter][indexRow];
if(value.compare( &BigFloat(0.0) ) != 0 )
{
// tausche reihen
for(unsigned int i = 0; i <= col; ++i) //zusätzliche spalte angefügt durch vector
{
BigFloat currowValue = matrix[indexRow][i];
BigFloat changeValue = matrix[rowCounter][i];
matrix[indexRow][i] = changeValue;
matrix[rowCounter][i] = currowValue;
}
return true;
}
}
return false;
}
// Zoom vers la cible du clic
static void clicGauche(SDL_Event& event, Affichage* disp)
{
// Calcul de la position du clic dans le plan complexe
float x = disp->center.x + disp->scale*(((float)event.motion.x) / WIDTH - 0.5f);
float y = disp->center.y + disp->scale*(((float)event.motion.y) / HEIGHT - 0.5f);
// MAJ de la position du nouveau centre dans le plan complexe
if (INTERACTIVE) {
disp->center.x = x + (disp->center.x - x) * ZOOM_FACTOR;
disp->center.y = y + (disp->center.y - y) * ZOOM_FACTOR;
updateBigCenter(event, true);
}
// MAJ de l'echelle
disp->scale *= ZOOM_FACTOR;
//BigFloat zoomFactor(true, 0, 0x80000000, 0, 0);
BigFloat temp, temp2;
BigFloat::mult(ZOOM_FACTOR, *bigScale, temp);
bigScale->reset();
temp2.copy(*bigScale);
BigFloat::add(temp, temp2, *bigScale);
//Nouveau calcul de la fractale avec chrono
disp->start = chrono::system_clock::now();
if (GPU && BIG_FLOAT_SIZE == 0)
affichageGPU(disp);
else if (GPU)
computeBigMandelGPU(disp, xCenter->pos, xCenter->decimals, yCenter->pos, yCenter->decimals, bigScale->decimals);
else
if (BIG_FLOAT_SIZE == 0)
computeMandel(disp->pixels, disp->center, disp->scale);
else {
computeBigMandel(disp->pixels, *xCenter, *yCenter, *bigScale);
}
disp->end = chrono::system_clock::now();
disp->duration = disp->end - disp->start;
cout << "Frame computing time : " << disp->duration.count() << endl;
//cout << "Frame computing scale : " << disp->scale << endl;
// Affichage de la fractale
disp->dessin();
}
////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////
// Pi
void Pi_BSR(BigFloat &P,BigFloat &Q,BigFloat &R,uint32_t a,uint32_t b,size_t p,int tds = 1){
// Binary Splitting recursion for the Chudnovsky Formula.
if (b - a == 1){
// P = (13591409 + 545140134 b)(2b-1)(6b-5)(6b-1) (-1)^b
P = BigFloat(b).mul(545140134);
P = P.add(BigFloat(13591409));
P = P.mul(2*b - 1);
P = P.mul(6*b - 5);
P = P.mul(6*b - 1);
if (b % 2 == 1)
P.negate();
// Q = 10939058860032000 * b^3
Q = BigFloat(b);
Q = Q.mul(Q).mul(Q).mul(26726400).mul(409297880);
// R = (2b-1)(6b-5)(6b-1)
R = BigFloat(2*b - 1);
R = R.mul(6*b - 5);
R = R.mul(6*b - 1);
return;
}
uint32_t m = (a + b) / 2;
BigFloat P0,Q0,R0,P1,Q1,R1;
if (b - a < 1000 || tds < 2){
// No more threads.
Pi_BSR(P0,Q0,R0,a,m,p);
Pi_BSR(P1,Q1,R1,m,b,p);
}else{
// Run sub-recursions in parallel.
int tds0 = tds / 2;
int tds1 = tds - tds0;
#pragma omp parallel num_threads(2)
{
int tid = omp_get_thread_num();
if (tid == 0){
Pi_BSR(P0,Q0,R0,a,m,p,tds0);
}
if (tid != 0 || omp_get_num_threads() < 2){
Pi_BSR(P1,Q1,R1,m,b,p,tds1);
}
}
}
P = P0.mul(Q1,p,tds).add(P1.mul(R0,p,tds),p);
Q = Q0.mul(Q1,p,tds);
R = R0.mul(R1,p,tds);
}
BigFloat BigFloat::add(const BigFloat &x,size_t p) const{
// Addition
// The target precision is p.
// If (p = 0), then no truncation is done. The entire operation is done
// at maximum precision with no data loss.
// Same sign. Add.
if (sign == x.sign)
return uadd(x,p);
// this > x
if (ucmp(x) > 0)
return usub(x,p);
// this < x
return x.usub(*this,p);
}
void ComputeFloatSession<wtype>::write_digits(const BigFloat<wtype>& x, const std::string& name, const std::string& algorithm){
std::string dec_name, hex_name;
if (algorithm.empty()){
dec_name = name + " - Dec.txt";
hex_name = name + " - Hex.txt";
}else{
dec_name = name + " - Dec - " + algorithm + ".txt";
hex_name = name + " - Hex - " + algorithm + ".txt";
}
// Special Case: If the number is close to one, don't count the digits in
// front of the decimal place. This keeps the behavior consistent with
// y-cruncher and other Pi programs that start counting digits after the
// decimal place.
upL_t dec_digits = m_dec;
upL_t hex_digits = m_hex;
if (x.get_mag() == 1){
wtype top_word = x[0];
dec_digits += std::to_string(top_word).size();
while (top_word != 0){
top_word >>= 4;
hex_digits++;
}
}
BigFloat BigFloat::rcp(size_t p,int tds) const{
// Compute reciprocal using Newton's Method.
// r1 = r0 - (r0 * x - 1) * r0
if (L == 0)
throw "Divide by Zero";
// Collect operand
int64_t Aexp = exp;
size_t AL = L;
uint32_t *AT = T.get();
// End of recursion. Generate starting point.
if (p == 0){
// Truncate precision to 3.
p = 3;
if (AL > p){
size_t chop = AL - p;
AL = p;
Aexp += chop;
AT += chop;
}
// Convert number to floating-point.
double val = AT[0];
if (AL >= 2)
val += AT[1] * 1000000000.;
if (AL >= 3)
val += AT[2] * 1000000000000000000.;
// Compute reciprocal.
val = 1. / val;
Aexp = -Aexp;
// Scale
while (val < 1000000000.){
val *= 1000000000.;
Aexp--;
}
// Rebuild a BigFloat.
uint64_t val64 = (uint64_t)val;
BigFloat out;
out.sign = sign;
out.T = std::unique_ptr<uint32_t[]>(new uint32_t[2]);
out.T[0] = (uint32_t)(val64 % 1000000000);
out.T[1] = (uint32_t)(val64 / 1000000000);
out.L = 2;
out.exp = Aexp;
return out;
}
// Half the precision
size_t s = p / 2 + 1;
if (p == 1) s = 0;
if (p == 2) s = 1;
// Recurse at half the precision
BigFloat T = rcp(s,tds);
// r1 = r0 - (r0 * x - 1) * r0
return T.sub(this->mul(T,p,tds).sub(BigFloat(1),p).mul(T,p,tds),p);
}
inline Int NumMantissaBits<BigFloat>( const BigFloat& alpha )
{ return alpha.Precision(); }
inline bool IsFinite( const BigFloat& alpha )
{ return mpfr_number_p( alpha.LockedPointer() ) != 0; }
inline BigFloat Infinity<BigFloat>( const BigFloat& alpha )
{
BigFloat inf(0,alpha.Precision());
mpfr_set_inf( inf.Pointer(), 1 );
return inf;
}
// freebayes main
int main (int argc, char *argv[]) {
// install segfault handler
signal(SIGSEGV, segfaultHandler);
AlleleParser* parser = new AlleleParser(argc, argv);
Parameters& parameters = parser->parameters;
list<Allele*> alleles;
Samples samples;
ostream& out = *(parser->output);
Bias observationBias;
if (!parameters.alleleObservationBiasFile.empty()) {
observationBias.open(parameters.alleleObservationBiasFile);
}
Contamination contaminationEstimates(0.5+parameters.probContamination, parameters.probContamination);
if (!parameters.contaminationEstimateFile.empty()) {
contaminationEstimates.open(parameters.contaminationEstimateFile);
}
// this can be uncommented to force operation on a specific set of genotypes
vector<Allele> allGenotypeAlleles;
allGenotypeAlleles.push_back(genotypeAllele(ALLELE_GENOTYPE, "A", 1));
allGenotypeAlleles.push_back(genotypeAllele(ALLELE_GENOTYPE, "T", 1));
allGenotypeAlleles.push_back(genotypeAllele(ALLELE_GENOTYPE, "G", 1));
allGenotypeAlleles.push_back(genotypeAllele(ALLELE_GENOTYPE, "C", 1));
int allowedAlleleTypes = ALLELE_REFERENCE;
if (parameters.allowSNPs) {
allowedAlleleTypes |= ALLELE_SNP;
}
if (parameters.allowIndels) {
allowedAlleleTypes |= ALLELE_INSERTION;
allowedAlleleTypes |= ALLELE_DELETION;
}
if (parameters.allowMNPs) {
allowedAlleleTypes |= ALLELE_MNP;
}
if (parameters.allowComplex) {
allowedAlleleTypes |= ALLELE_COMPLEX;
}
// output VCF header
if (parameters.output == "vcf") {
out << parser->variantCallFile.header << endl;
}
Allele nullAllele = genotypeAllele(ALLELE_NULL, "N", 1, "1N");
unsigned long total_sites = 0;
unsigned long processed_sites = 0;
while (parser->getNextAlleles(samples, allowedAlleleTypes)) {
++total_sites;
DEBUG2("at start of main loop");
// don't process non-ATGC's in the reference
string cb = parser->currentReferenceBaseString();
if (cb != "A" && cb != "T" && cb != "C" && cb != "G") {
DEBUG2("current reference base is N");
continue;
}
if (parameters.trace) {
for (Samples::iterator s = samples.begin(); s != samples.end(); ++s) {
const string& name = s->first;
for (Sample::iterator g = s->second.begin(); g != s->second.end(); ++g) {
vector<Allele*>& group = g->second;
for (vector<Allele*>::iterator a = group.begin(); a != group.end(); ++a) {
Allele& allele = **a;
parser->traceFile << parser->currentSequenceName << "," << (long unsigned int) parser->currentPosition + 1
<< ",allele," << name << "," << allele.readID << "," << allele.base() << ","
<< allele.currentQuality() << "," << allele.mapQuality << endl;
}
}
}
DEBUG2("after trace generation");
}
if (!parser->inTarget()) {
DEBUG("position: " << parser->currentSequenceName << ":" << (long unsigned int) parser->currentPosition + 1
<< " is not inside any targets, skipping");
continue;
}
int coverage = countAlleles(samples);
DEBUG("position: " << parser->currentSequenceName << ":" << (long unsigned int) parser->currentPosition + 1 << " coverage: " << coverage);
if (!parser->hasInputVariantAllelesAtCurrentPosition()) {
// skips 0-coverage regions
if (coverage == 0) {
DEBUG("no alleles left at this site after filtering");
continue;
} else if (coverage < parameters.minCoverage) {
DEBUG("post-filtering coverage of " << coverage << " is less than --min-coverage of " << parameters.minCoverage);
continue;
} else if (parameters.onlyUseInputAlleles) {
DEBUG("no input alleles, but using only input alleles for analysis, skipping position");
continue;
}
DEBUG2("coverage " << parser->currentSequenceName << ":" << parser->currentPosition << " == " << coverage);
// establish a set of possible alternate alleles to evaluate at this location
if (!parameters.reportMonomorphic
&& !sufficientAlternateObservations(samples, parameters.minAltCount, parameters.minAltFraction)) {
DEBUG("insufficient alternate observations");
continue;
}
if (parameters.reportMonomorphic) {
DEBUG("calling at site even though there are no alternate observations");
}
} else {
/*
cerr << "has input variants at " << parser->currentSequenceName << ":" << parser->currentPosition << endl;
vector<Allele>& inputs = parser->inputVariantAlleles[parser->currentPosition];
for (vector<Allele>::iterator a = inputs.begin(); a != inputs.end(); ++a) {
cerr << *a << endl;
}
*/
}
// to ensure proper ordering of output stream
vector<string> sampleListPlusRef;
for (vector<string>::iterator s = parser->sampleList.begin(); s != parser->sampleList.end(); ++s) {
sampleListPlusRef.push_back(*s);
}
if (parameters.useRefAllele) {
sampleListPlusRef.push_back(parser->currentSequenceName);
}
// establish genotype alleles using input filters
map<string, vector<Allele*> > alleleGroups;
groupAlleles(samples, alleleGroups);
DEBUG2("grouped alleles by equivalence");
vector<Allele> genotypeAlleles = parser->genotypeAlleles(alleleGroups, samples, parameters.onlyUseInputAlleles);
// always include the reference allele as a possible genotype, even when we don't include it by default
if (!parameters.useRefAllele) {
vector<Allele> refAlleleVector;
refAlleleVector.push_back(genotypeAllele(ALLELE_REFERENCE, string(1, parser->currentReferenceBase), 1, "1M"));
genotypeAlleles = alleleUnion(genotypeAlleles, refAlleleVector);
}
map<string, vector<Allele*> > partialObservationGroups;
map<Allele*, set<Allele*> > partialObservationSupport;
// build haplotype alleles matching the current longest allele (often will do nothing)
// this will adjust genotypeAlleles if changes are made
DEBUG("building haplotype alleles, currently there are " << genotypeAlleles.size() << " genotype alleles");
DEBUG(genotypeAlleles);
parser->buildHaplotypeAlleles(genotypeAlleles,
samples,
alleleGroups,
partialObservationGroups,
partialObservationSupport,
allowedAlleleTypes);
DEBUG("built haplotype alleles, now there are " << genotypeAlleles.size() << " genotype alleles");
DEBUG(genotypeAlleles);
string referenceBase = parser->currentReferenceHaplotype();
/* for debugging
for (Samples::iterator s = samples.begin(); s != samples.end(); ++s) {
string sampleName = s->first;
Sample& sample = s->second;
cerr << sampleName << ": " << sample << endl;
}
*/
// re-calculate coverage, as this could change now that we've built haplotype alleles
coverage = countAlleles(samples);
// estimate theta using the haplotype length
long double theta = parameters.TH * parser->lastHaplotypeLength;
// if we have only one viable allele, we don't have evidence for variation at this site
if (!parser->hasInputVariantAllelesAtCurrentPosition() && !parameters.reportMonomorphic && genotypeAlleles.size() <= 1 && genotypeAlleles.front().isReference()) {
DEBUG("no alternate genotype alleles passed filters at " << parser->currentSequenceName << ":" << parser->currentPosition);
continue;
}
DEBUG("genotype alleles: " << genotypeAlleles);
// add the null genotype
bool usingNull = false;
if (parameters.excludeUnobservedGenotypes && genotypeAlleles.size() > 2) {
genotypeAlleles.push_back(nullAllele);
usingNull = true;
}
++processed_sites;
// generate possible genotypes
// for each possible ploidy in the dataset, generate all possible genotypes
vector<int> ploidies = parser->currentPloidies(samples);
map<int, vector<Genotype> > genotypesByPloidy = getGenotypesByPloidy(ploidies, genotypeAlleles);
int numCopiesOfLocus = parser->copiesOfLocus(samples);
DEBUG2("generated all possible genotypes:");
if (parameters.debug2) {
for (map<int, vector<Genotype> >::iterator s = genotypesByPloidy.begin(); s != genotypesByPloidy.end(); ++s) {
vector<Genotype>& genotypes = s->second;
for (vector<Genotype>::iterator g = genotypes.begin(); g != genotypes.end(); ++g) {
DEBUG2(*g);
}
}
}
// get estimated allele frequencies using sum of estimated qualities
map<string, double> estimatedAlleleFrequencies = samples.estimatedAlleleFrequencies();
double estimatedMaxAlleleFrequency = 0;
double estimatedMaxAlleleCount = 0;
double estimatedMajorFrequency = estimatedAlleleFrequencies[referenceBase];
if (estimatedMajorFrequency < 0.5) estimatedMajorFrequency = 1-estimatedMajorFrequency;
double estimatedMinorFrequency = 1-estimatedMajorFrequency;
//cerr << "num copies of locus " << numCopiesOfLocus << endl;
int estimatedMinorAllelesAtLocus = max(1, (int) ceil((double) numCopiesOfLocus * estimatedMinorFrequency));
//cerr << "estimated minor frequency " << estimatedMinorFrequency << endl;
//cerr << "estimated minor count " << estimatedMinorAllelesAtLocus << endl;
Results results;
map<string, vector<vector<SampleDataLikelihood> > > sampleDataLikelihoodsByPopulation;
map<string, vector<vector<SampleDataLikelihood> > > variantSampleDataLikelihoodsByPopulation;
map<string, vector<vector<SampleDataLikelihood> > > invariantSampleDataLikelihoodsByPopulation;
map<string, int> inputAlleleCounts;
int inputLikelihoodCount = 0;
DEBUG2("calculating data likelihoods");
// calculate data likelihoods
//for (Samples::iterator s = samples.begin(); s != samples.end(); ++s) {
for (vector<string>::iterator n = parser->sampleList.begin(); n != parser->sampleList.end(); ++n) {
//string sampleName = s->first;
string& sampleName = *n;
//DEBUG2("sample: " << sampleName);
//Sample& sample = s->second;
if (samples.find(sampleName) == samples.end()
&& !(parser->hasInputVariantAllelesAtCurrentPosition()
|| parameters.reportMonomorphic)) {
continue;
}
Sample& sample = samples[sampleName];
vector<Genotype>& genotypes = genotypesByPloidy[parser->currentSamplePloidy(sampleName)];
vector<Genotype*> genotypesWithObs;
for (vector<Genotype>::iterator g = genotypes.begin(); g != genotypes.end(); ++g) {
if (parameters.excludePartiallyObservedGenotypes) {
if (g->sampleHasSupportingObservationsForAllAlleles(sample)) {
genotypesWithObs.push_back(&*g);
}
} else if (parameters.excludeUnobservedGenotypes && usingNull) {
if (g->sampleHasSupportingObservations(sample)) {
//cerr << sampleName << " has suppporting obs for " << *g << endl;
genotypesWithObs.push_back(&*g);
} else if (g->hasNullAllele() && g->homozygous) {
// this genotype will never be added if we are running in observed-only mode, but
// we still need it for consistency
genotypesWithObs.push_back(&*g);
}
} else {
genotypesWithObs.push_back(&*g);
}
}
// skip this sample if we have no observations supporting any of the genotypes we are going to evaluate
if (genotypesWithObs.empty()) {
continue;
}
vector<pair<Genotype*, long double> > probs
= probObservedAllelesGivenGenotypes(sample, genotypesWithObs,
parameters.RDF, parameters.useMappingQuality,
observationBias, parameters.standardGLs,
genotypeAlleles,
contaminationEstimates,
estimatedAlleleFrequencies);
#ifdef VERBOSE_DEBUG
if (parameters.debug2) {
for (vector<pair<Genotype*, long double> >::iterator p = probs.begin(); p != probs.end(); ++p) {
cerr << parser->currentSequenceName << "," << (long unsigned int) parser->currentPosition + 1 << ","
<< sampleName << ",likelihood," << *(p->first) << "," << p->second << endl;
}
}
#endif
Result& sampleData = results[sampleName];
sampleData.name = sampleName;
sampleData.observations = &sample;
for (vector<pair<Genotype*, long double> >::iterator p = probs.begin(); p != probs.end(); ++p) {
sampleData.push_back(SampleDataLikelihood(sampleName, &sample, p->first, p->second, 0));
}
sortSampleDataLikelihoods(sampleData);
string& population = parser->samplePopulation[sampleName];
vector<vector<SampleDataLikelihood> >& sampleDataLikelihoods = sampleDataLikelihoodsByPopulation[population];
vector<vector<SampleDataLikelihood> >& variantSampleDataLikelihoods = variantSampleDataLikelihoodsByPopulation[population];
vector<vector<SampleDataLikelihood> >& invariantSampleDataLikelihoods = invariantSampleDataLikelihoodsByPopulation[population];
if (parameters.genotypeVariantThreshold != 0) {
if (sampleData.size() > 1
&& abs(sampleData.at(1).prob - sampleData.front().prob)
< parameters.genotypeVariantThreshold) {
variantSampleDataLikelihoods.push_back(sampleData);
} else {
invariantSampleDataLikelihoods.push_back(sampleData);
}
} else {
variantSampleDataLikelihoods.push_back(sampleData);
}
sampleDataLikelihoods.push_back(sampleData);
DEBUG2("obtaining genotype likelihoods input from VCF");
int prevcount = sampleDataLikelihoods.size();
parser->addCurrentGenotypeLikelihoods(genotypesByPloidy, sampleDataLikelihoods);
// add these sample data likelihoods to 'invariant' likelihoods
inputLikelihoodCount += sampleDataLikelihoods.size() - prevcount;
parser->addCurrentGenotypeLikelihoods(genotypesByPloidy, invariantSampleDataLikelihoods);
}
// if there are not any input GLs, attempt to use the input ACs
if (inputLikelihoodCount == 0) {
parser->getInputAlleleCounts(genotypeAlleles, inputAlleleCounts);
}
DEBUG2("finished calculating data likelihoods");
// this section is a hack to make output of trace identical to BamBayes trace
// and also outputs the list of samples
vector<bool> samplesWithData;
if (parameters.trace) {
parser->traceFile << parser->currentSequenceName << "," << (long unsigned int) parser->currentPosition + 1 << ",samples,";
for (vector<string>::iterator s = sampleListPlusRef.begin(); s != sampleListPlusRef.end(); ++s) {
if (parameters.trace) parser->traceFile << *s << ":";
Results::iterator r = results.find(*s);
if (r != results.end()) {
samplesWithData.push_back(true);
} else {
samplesWithData.push_back(false);
}
}
parser->traceFile << endl;
}
// if somehow we get here without any possible sample genotype likelihoods, bail out
bool hasSampleLikelihoods = false;
for (map<string, vector<vector<SampleDataLikelihood> > >::iterator s = sampleDataLikelihoodsByPopulation.begin(); s != sampleDataLikelihoodsByPopulation.end(); ++s) {
if (!s->second.empty()) {
hasSampleLikelihoods = true;
break;
}
}
if (!hasSampleLikelihoods) {
continue;
}
DEBUG2("calulating combo posteriors over " << parser->populationSamples.size() << " populations");
// XXX
// TODO skip these steps in the case that there is only one population?
// we provide p(var|data), or the probability that the location has
// variation between individuals relative to the probability that it
// has no variation
//
// in other words:
// p(var|d) = 1 - p(AA|d) - p(TT|d) - P(GG|d) - P(CC|d)
//
// the approach is go through all the homozygous combos
// and then subtract this from 1... resolving p(var|d)
BigFloat pVar = 1.0;
BigFloat pHom = 0.0;
long double bestComboOddsRatio = 0;
GenotypeCombo bestCombo; // = NULL;
// what a hack...
/*
if (parameters.trace) {
for (list<GenotypeCombo>::iterator gc = genotypeCombos.begin(); gc != genotypeCombos.end(); ++gc) {
vector<Genotype*> comboGenotypes;
for (GenotypeCombo::iterator g = gc->begin(); g != gc->end(); ++g)
comboGenotypes.push_back((*g)->genotype);
long double posteriorProb = gc->posteriorProb;
long double dataLikelihoodln = gc->probObsGivenGenotypes;
long double priorln = gc->posteriorProb;
long double priorlnG_Af = gc->priorProbG_Af;
long double priorlnAf = gc->priorProbAf;
long double priorlnBin = gc->priorProbObservations;
parser->traceFile << parser->currentSequenceName << "," << (long unsigned int) parser->currentPosition + 1 << ",genotypecombo,";
int j = 0;
GenotypeCombo::iterator i = gc->begin();
for (vector<bool>::iterator d = samplesWithData.begin(); d != samplesWithData.end(); ++d) {
if (*d) {
parser->traceFile << IUPAC(*(*i)->genotype);
++i;
} else {
parser->traceFile << "?";
}
}
// TODO cleanup this and above
parser->traceFile
<< "," << dataLikelihoodln
<< "," << priorln
<< "," << priorlnG_Af
<< "," << priorlnAf
<< "," << priorlnBin
<< "," << posteriorProb
<< "," << safe_exp(posteriorProb - posteriorNormalizer)
<< endl;
}
}
*/
// the second clause guards against float underflow causing us not to output a position
// practically, parameters.PVL == 0 means "report all genotypes which pass our input filters"
GenotypeCombo bestGenotypeComboByMarginals;
vector<vector<SampleDataLikelihood> > allSampleDataLikelihoods;
DEBUG("searching genotype space");
// resample the posterior, this time without bounds on the
// samples we vary, ensuring that we can generate marginals for
// all sample/genotype combinations
//SampleDataLikelihoods marginalLikelihoods = sampleDataLikelihoods; // heavyweight copy...
map<string, list<GenotypeCombo> > genotypeCombosByPopulation;
int genotypingTotalIterations = 0; // tally total iterations required to reach convergence
map<string, list<GenotypeCombo> > glMaxCombos;
for (map<string, SampleDataLikelihoods>::iterator p = sampleDataLikelihoodsByPopulation.begin(); p != sampleDataLikelihoodsByPopulation.end(); ++p) {
const string& population = p->first;
SampleDataLikelihoods& sampleDataLikelihoods = p->second;
list<GenotypeCombo>& populationGenotypeCombos = genotypeCombosByPopulation[population];
DEBUG2("genqerating banded genotype combinations from " << sampleDataLikelihoods.size() << " sample genotypes in population " << population);
// cap the number of iterations at 2 x the number of alternate alleles
// max it at parameters.genotypingMaxIterations iterations, min at 10
int itermax = min(max(10, 2 * estimatedMinorAllelesAtLocus), parameters.genotypingMaxIterations);
//int itermax = parameters.genotypingMaxIterations;
// XXX HACK
// passing 0 for bandwidth and banddepth means "exhaustive local search"
// this produces properly normalized GQ's at polyallelic sites
int adjustedBandwidth = 0;
int adjustedBanddepth = 0;
// however, this can lead to huge performance problems at complex sites,
// so we implement this hack...
if (parameters.genotypingMaxBandDepth > 0 &&
genotypeAlleles.size() > parameters.genotypingMaxBandDepth) {
adjustedBandwidth = 1;
adjustedBanddepth = parameters.genotypingMaxBandDepth;
}
GenotypeCombo nullCombo;
SampleDataLikelihoods nullSampleDataLikelihoods;
// this is the genotype-likelihood maximum
if (parameters.reportGenotypeLikelihoodMax) {
GenotypeCombo comboKing;
vector<int> initialPosition;
initialPosition.assign(sampleDataLikelihoods.size(), 0);
SampleDataLikelihoods nullDataLikelihoods; // dummy variable
makeComboByDatalLikelihoodRank(comboKing,
initialPosition,
sampleDataLikelihoods,
nullDataLikelihoods,
inputAlleleCounts,
theta,
parameters.pooledDiscrete,
parameters.ewensPriors,
parameters.permute,
parameters.hwePriors,
parameters.obsBinomialPriors,
parameters.alleleBalancePriors,
parameters.diffusionPriorScalar);
glMaxCombos[population].push_back(comboKing);
}
// search much longer for convergence
convergentGenotypeComboSearch(
populationGenotypeCombos,
nullCombo,
sampleDataLikelihoods, // vary everything
sampleDataLikelihoods,
nullSampleDataLikelihoods,
samples,
genotypeAlleles,
inputAlleleCounts,
adjustedBandwidth,
adjustedBanddepth,
theta,
parameters.pooledDiscrete,
parameters.ewensPriors,
parameters.permute,
parameters.hwePriors,
parameters.obsBinomialPriors,
parameters.alleleBalancePriors,
parameters.diffusionPriorScalar,
itermax,
genotypingTotalIterations,
true); // add homozygous combos
// ^^ combo results are sorted by default
}
// generate the GL max combo
GenotypeCombo glMax;
if (parameters.reportGenotypeLikelihoodMax) {
list<GenotypeCombo> glMaxGenotypeCombos;
combinePopulationCombos(glMaxGenotypeCombos, glMaxCombos);
glMax = glMaxGenotypeCombos.front();
}
// accumulate combos from independently-calculated populations into the list of combos
list<GenotypeCombo> genotypeCombos; // build new combos into this list
combinePopulationCombos(genotypeCombos, genotypeCombosByPopulation);
// TODO factor out the following blocks as they are repeated from above
// re-get posterior normalizer
vector<long double> comboProbs;
for (list<GenotypeCombo>::iterator gc = genotypeCombos.begin(); gc != genotypeCombos.end(); ++gc) {
comboProbs.push_back(gc->posteriorProb);
}
long double posteriorNormalizer = logsumexp_probs(comboProbs);
// recalculate posterior normalizer
pVar = 1.0;
pHom = 0.0;
// calculates pvar and gets the best het combo
for (list<GenotypeCombo>::iterator gc = genotypeCombos.begin(); gc != genotypeCombos.end(); ++gc) {
if (gc->isHomozygous()
&& (parameters.useRefAllele
|| !parameters.useRefAllele && gc->alleles().front() == referenceBase)) {
pVar -= big_exp(gc->posteriorProb - posteriorNormalizer);
pHom += big_exp(gc->posteriorProb - posteriorNormalizer);
}
}
// report the maximum a posteriori estimate
// unless we're reporting the GL maximum
if (!parameters.reportGenotypeLikelihoodMax) {
bestCombo = genotypeCombos.front();
} else {
bestCombo = glMax;
}
DEBUG2("best combo: " << bestCombo);
// odds ratio between the first and second-best combinations
if (genotypeCombos.size() > 1) {
bestComboOddsRatio = genotypeCombos.front().posteriorProb - (++genotypeCombos.begin())->posteriorProb;
}
if (parameters.calculateMarginals) {
// make a combined, all-populations sample data likelihoods vector to accumulate marginals
SampleDataLikelihoods allSampleDataLikelihoods;
for (map<string, SampleDataLikelihoods>::iterator p = sampleDataLikelihoodsByPopulation.begin(); p != sampleDataLikelihoodsByPopulation.end(); ++p) {
SampleDataLikelihoods& sdls = p->second;
allSampleDataLikelihoods.reserve(allSampleDataLikelihoods.size() + distance(sdls.begin(), sdls.end()));
allSampleDataLikelihoods.insert(allSampleDataLikelihoods.end(), sdls.begin(), sdls.end());
}
// calculate the marginal likelihoods for this population
marginalGenotypeLikelihoods(genotypeCombos, allSampleDataLikelihoods);
// store the marginal data likelihoods in the results, for easy parsing
// like a vector -> map conversion...
results.update(allSampleDataLikelihoods);
}
map<string, int> repeats;
if (parameters.showReferenceRepeats) {
repeats = parser->repeatCounts(parser->currentSequencePosition(), parser->currentSequence, 12);
}
vector<Allele> alts;
if (parameters.onlyUseInputAlleles
|| parameters.reportAllHaplotypeAlleles
|| parameters.pooledContinuous) {
//alts = genotypeAlleles;
for (vector<Allele>::iterator a = genotypeAlleles.begin(); a != genotypeAlleles.end(); ++a) {
if (!a->isReference()) {
alts.push_back(*a);
}
}
} else {
// get the unique alternate alleles in this combo, sorted by frequency in the combo
vector<pair<Allele, int> > alternates = alternateAlleles(bestCombo, referenceBase);
for (vector<pair<Allele, int> >::iterator a = alternates.begin(); a != alternates.end(); ++a) {
Allele& alt = a->first;
if (!alt.isNull() && !alt.isReference())
alts.push_back(alt);
}
// if there are no alternate alleles in the best combo, use the genotype alleles
// XXX ...
if (alts.empty()) {
for (vector<Allele>::iterator a = genotypeAlleles.begin(); a != genotypeAlleles.end(); ++a) {
if (!a->isReference()) {
alts.push_back(*a);
}
}
}
}
//if (alts.empty()) alts = genotypeAlleles;
if (!alts.empty() && (1 - pHom.ToDouble()) >= parameters.PVL || parameters.PVL == 0) {
vcf::Variant var(parser->variantCallFile);
out << results.vcf(
var,
pHom,
bestComboOddsRatio,
samples,
referenceBase,
alts,
repeats,
genotypingTotalIterations,
parser->sampleList,
coverage,
bestCombo,
alleleGroups,
partialObservationGroups,
partialObservationSupport,
genotypesByPloidy,
parser->sequencingTechnologies,
parser)
<< endl;
} else if (!parameters.failedFile.empty()) {
// get the unique alternate alleles in this combo, sorted by frequency in the combo
long unsigned int position = parser->currentPosition;
for (vector<Allele>::iterator ga = genotypeAlleles.begin(); ga != genotypeAlleles.end(); ++ga) {
if (ga->type == ALLELE_REFERENCE)
continue;
parser->failedFile
<< parser->currentSequenceName << "\t"
<< position << "\t"
<< position + ga->length << "\t"
<< *ga << endl;
}
// BED format
}
DEBUG2("finished position");
}
DEBUG("total sites: " << total_sites << endl
<< "processed sites: " << processed_sites << endl
<< "ratio: " << (float) processed_sites / (float) total_sites);
delete parser;
return 0;
}
inline BigFloat Min<BigFloat>( const BigFloat& alpha )
{
BigFloat one(1,alpha.Precision());
return one << (MinExponent<BigFloat>()-1);
}
inline BigFloat Max<BigFloat>( const BigFloat& alpha )
{
BigFloat one(1,alpha.Precision());
return (one-Epsilon(one)) << MaxExponent<BigFloat>();
}
NxI32 CommandCallback(NxI32 token,NxI32 count,const char **arglist)
{
NxI32 ret = 0;
saveMenuState();
switch ( token )
{
case MC_MEMORY_REPORT:
break;
case MC_FLOATING_POINT_RESOLUTION:
if ( count == 2 )
{
MultiFloatType type = MFT_MEDIUM;
if ( stricmp(arglist[1],"SMALL") == 0 )
type = MFT_SMALL;
else if ( stricmp(arglist[1],"MEDIUM") == 0 )
type = MFT_MEDIUM;
else if ( stricmp(arglist[1],"BIGFLOAT") == 0 )
type = MFT_BIG;
else if ( stricmp(arglist[1],"FIXED32") == 0 )
type = MFT_FIXED32;
tf_setFloatingPointResolution(gTfrac,type);
gLog->Display("Setting Floating Point Resolution to: %s\r\n", arglist[1] );
}
break;
case MC_SHOW_NORMALS:
if ( count == 2 )
{
setShowNormals( getBool(arglist[1]) );
}
break;
case MC_ENVIRONMENT_TEXTURE:
if ( count == 2 )
{
const char *t = arglist[1];
if ( gFileSystem ) t = gFileSystem->FileOpenString(t,true);
gPd3d->setEnvironmentTexture(t);
}
break;
case MC_ROTATION_SPEED:
if ( count == 2 )
{
NxF32 rspeed = (NxF32)atof( arglist[1] );
setRotationSpeed(rspeed);
}
break;
case MC_OPTIMIZE_MESH:
if ( !mStartup )
{
tf_state(gTfrac,TS_OPTIMIZE_MESH);
}
break;
case MC_FILTER_FRACTAL:
if ( !mStartup )
{
tf_state(gTfrac,TS_FILTER_FRACTAL);
}
break;
case MC_DEFAULT_MANDELBROT:
if ( !mStartup )
{
BigFloat xleft;
BigFloat xright;
BigFloat ytop;
xleft.FromDouble(-2.5);
xright.FromDouble(0.75);
ytop.FromDouble(-1.5);
tf_setFractalCoordinates(gTfrac,xleft,xright,ytop);
tf_action(gTfrac,FA_MOUSE_CENTER,false,1024/2,768/2);
}
break;
case MC_CLAMP_LOW:
if ( count == 2 && gTfrac )
{
NxF32 c = (NxF32) atof( arglist[1] );
tf_state(gTfrac,TS_CLAMP_LOW,false,0,c);
}
break;
case MC_CLAMP_HIGH:
if ( count == 2 && gTfrac )
{
NxF32 c = (NxF32) atof( arglist[1] );
tf_state(gTfrac,TS_CLAMP_HIGH,false,0,c);
}
break;
case MC_CLAMP_SCALE:
if ( count == 2 && gTfrac )
{
NxF32 c = (NxF32) atof( arglist[1] );
tf_state(gTfrac,TS_CLAMP_SCALE,false,0,c);
}
break;
case MC_WIREFRAME_OVERLAY:
if ( count == 2 && gTfrac )
{
bool state = getBool(arglist[1]);
tf_state(gTfrac,TS_WIREFAME_OVERLAY,state);
}
break;
case MC_VIEW3D:
if ( count == 2 )
{
gView3d = getBool(arglist[1]);
if ( gView3d )
{
NxF32 eye[3];
NxF32 look[3];
look[0] = 0;
look[1] = 0;
look[2] = 0;
eye[0] = 200;
eye[1] = 250;
eye[2] = 200;
lookAt(eye,look);
}
}
break;
case MC_FRACTAL_COORDINATES:
if ( count == 4 && gTfrac )
{
BigFloat xleft = arglist[1];
BigFloat xright = arglist[2];
BigFloat ytop = arglist[3];
tf_setFractalCoordinates(gTfrac,xleft,xright,ytop);
}
break;
case MC_COLOR_PALETTE:
if ( count == 2 && gTfrac )
{
tf_setPal(gTfrac,arglist[1]);
}
break;
case MC_ITERATION_COUNT:
if ( count == 2 && gTfrac )
{
NxU32 icount = (NxU32)atoi(arglist[1]);
tf_state(gTfrac,TS_ITERATION_COUNT,false,icount);
}
break;
case MC_CLOCK_CYCLES:
if ( count == 2 && gTfrac )
{
NxU32 icount = (NxU32)atoi(arglist[1]);
tf_state(gTfrac,TS_CLOCK_CYCLES,false,icount);
}
break;
case MC_USE_THREADING:
if ( count == 2 && gTfrac )
{
bool state = getBool(arglist[1]);
tf_state(gTfrac,TS_THREADING,state);
}
break;
case MC_SMOOTH_COLOR:
if ( count == 2 && gTfrac )
{
NxI32 cscale = atoi(arglist[1]);
tf_state(gTfrac,TS_SMOOTH_COLOR,false,cscale);
}
break;
case MC_PREVIEW_ONLY:
if ( count == 2 && gTfrac )
{
bool state = getBool(arglist[1]);
tf_state(gTfrac,TS_PREVIEW_ONLY,state);
}
break;
case MC_RECTANGLE_SUBDIVISION:
if ( count == 2 && gTfrac )
{
bool state = getBool(arglist[1]);
tf_state(gTfrac,TS_RECTANGLE_SUBDIVISION,state);
}
break;
case MC_PSLOOKAT:
// 0 1 2 3 4 5 6
// Usage: PsLookAt <eyex> <eyey> <eyez> <lookx> <looky> <lookz>
if ( count == 7 )
{
NxF32 eye[3];
NxF32 look[3];
eye[0] = (NxF32) atof( arglist[1] );
eye[1] = (NxF32) atof( arglist[2] );
eye[2] = (NxF32) atof( arglist[3] );
look[0] = (NxF32) atof(arglist[4] );
look[1] = (NxF32) atof(arglist[5] );
look[2] = (NxF32) atof(arglist[6] );
lookAt(eye,look);
}
break;
case MC_PSSCRIPT:
{
const char *fname = 0;
if ( count >= 2 )
{
fname = arglist[1];
}
#if TODO
SoftFileInterface *sfi = gSoftBodySystem->getSoftFileInterface();
if ( sfi )
{
fname = sfi->getLoadFileName(".psc", "Select a demo script to run.");
}
if ( fname )
{
CPARSER.Parse("Run \"%s\"",fname);
}
#endif
}
break;
}
return ret;
}
vector<double> GaussElimination::solveLinearEquationByBigFloat(vector<vector<double>> matrix, vector<double> v)
{
row = matrix.size();
col = matrix[0].size();
if( row != col )
{
cout << "It has to be a linear equation (row != col)." << endl;
return vector<double>(0);
}
if( row == 0 )
{
cout << "There are no member in equation (row = col = 0)." << endl;
return vector<double>(0);
}
if( row != v.size() )
{
cout << "Length of vector matched not matrix size." << endl;
return vector<double>(0);
}
// umwandlung in bigfloats und anhängen des vectors
vector<vector<BigFloat>> matrixBF(row);
unsigned int newcol = col + 1;
for(unsigned int r = 0; r < row; ++r)
{
vector<BigFloat> vBF(newcol);
matrixBF[r] = vBF;
for(unsigned c = 0; c < newcol; ++c)
{
BigFloat bf;
// letzte spalte kommt vom vector
if( c == col )
bf.set_with_double( v[r] );
else
bf.set_with_double( matrix[r][c] );
matrixBF[r][c] = bf;
}
}
//writeMatrix(matrixBF);
// für alle reihen
for(unsigned int indexDiagonal = 0; indexDiagonal < row; ++indexDiagonal)
{
bool success = handleEliminatingValuesInRowByBigFloat(matrixBF, indexDiagonal);
if(!success)
{
cout << "Current row has problems. Row: " << indexDiagonal << endl;
return vector<double>(0);
}
}
vector<double> coefficients(row);
// letzte spalte = coefficients -> return letzte spalte als vector
for(unsigned int i = 0; i < row; ++i)
coefficients[i] = matrixBF[i][col].getdouble();
return coefficients;
}
BigFloat BigFloat::div(const BigFloat &x,size_t p) const{
// Division
return this->mul(x.rcp(p),p);
}
BigFloat BigFloat::div(const BigFloat &x,size_t p,int tds) const{
// Division
return this->mul(x.rcp(p,tds),p,tds);
}