std::vector<bool> Line::getProblematic(){
std::vector<bool> prblm(m_Points.size());
for (std::vector<Point>::size_type i = 0; i < m_Points.size(); i++)
prblm[i] = m_Points[i].problematic;
return(prblm);
}
double CalibrationHelperCmsCorrelations::calibrate() {
CmsCorrelationCostFunction costfct(this,&(*corrTS_),lambda_,MINPRICE_,ONLYPOSITIVECORR_,MAXCORR_);
CmsCorrelationConstraint constraint;
int n = corrTS_->numberOfPillars();
Array guess(n);
for(int i=0;i<n;i++) {
double corr=corrTS_->pillarCorrelation(i);
if(corr>MAXCORR_) corr=MAXCORR_;
if(corr<-MAXCORR_) corr=-MAXCORR_;
if(ONLYPOSITIVECORR_) {
if(corr<0.0) corr=0.0; // only allow positive correlations
guess[i]=tan(M_PI/2.0*sqrt(corr)); // only allow positive correlations
}
else {
guess[i]=tan(M_PI/2.0*corr);
}
}
Problem prblm(costfct,constraint,guess);
EndCriteria ec(MAXIT_, MAXST_, OPTACCURACY_, OPTACCURACY_, OPTACCURACY_);
EndCriteria::Type ret;
if(optimizer_==0) {
Simplex method(0.05);
ret=method.minimize(prblm,ec);
}
if(optimizer_==1) {
LevenbergMarquardt method;
ret=method.minimize(prblm,ec);
}
if(optimizer_==2) {
SimulatedAnnealing method(0.1,0,50,500,2.5,0.2);
ret=method.minimize(prblm,ec);
}
//FILE* out=fopen("cost.log","a");
//fprintf(out,"Start Optimization with maxit=%d, maxst=%d, acc=%f, maxCorr=%f, minPrice=%f, allowPosOnly=%d",
// MAXIT,MAXST,OPTACCURACY,MAXCORR,MINPRICE,ONLYPOSITIVECORR);
//fclose(out);
Array x=prblm.currentValue();
double y=costfct.value(x);//prblm.functionValue(); to make sure, the model is set to optimal value
if(ret==EndCriteria::None) QL_FAIL("Optimizer returns status none");
if(ret==EndCriteria::MaxIterations) QL_FAIL("Optimizer returns status max iterations");
if(ret==EndCriteria::Unknown) QL_FAIL("Optimizer returns status unknown");
return y;
}
double CalibrationHelperCms::calibrate() {
double sumRmse=0.0;
//for(int i=0;i<optPillarIndex_.size();i++) {
CmsCostFunction* costfct;
if(!useStdCube_) costfct = new CmsCostFunction(this,&(*volCube_),noOptPillars_,optPillarTimes_,undPillarIndex_,maxMu_,maxNu_,minBeta_,maxBeta_,mode_);
else costfct = new CmsCostFunction(this,&(*volCube2_),noOptPillars_,optPillarTimes_,skipOptPillars_,underlying_,maxMu_,maxNu_,minBeta_,maxBeta_,mode_,weights_);
NoConstraint constraint;
int nop;
if(mode_==0 || mode_==4) nop=2;
if(mode_==1 || mode_==5) nop=3;
if(mode_==2 || mode_==6) nop=noOptPillars_;
if(mode_==3 || mode_==7) nop=4;
Array guess(nop);
if(!useStdCube_) {
if(mode_==0 || mode_==1) {
guess[0]=tan(M_PI*(volCube_->pillarBeta(0,undPillarIndex_)) - M_PI/2.0);
guess[1]=tan(M_PI*(volCube_->pillarBeta(noOptPillars_-1,undPillarIndex_)) - M_PI/2.0);
}
if(mode_==4 || mode_==5) {
guess[0]=sqrt(volCube_->pillarNu(0,undPillarIndex_));
guess[1]=sqrt(volCube_->pillarNu(noOptPillars_-1,undPillarIndex_));
}
if(mode_==1 || mode_==5) {
guess[2]=1.0; // decay factor
}
if(mode_==2) {
for(int i=0;i<noOptPillars_;i++)
guess[i]=tan(M_PI*(volCube_->pillarBeta(i,undPillarIndex_)) - M_PI/2.0);
}
if(mode_==6) {
for(int i=0;i<noOptPillars_;i++)
guess[i]=sqrt(volCube_->pillarNu(i,undPillarIndex_));
}
if(mode_==3) {
guess[3]=volCube_->pillarBeta(noOptPillars_-1,undPillarIndex_);
guess[0]=volCube_->pillarBeta(0,undPillarIndex_)-guess[3];
guess[1]=0.0;
guess[2]=1.0;
}
if(mode_==7) {
guess[3]=volCube_->pillarNu(noOptPillars_-1,undPillarIndex_);
guess[0]=volCube_->pillarNu(0,undPillarIndex_)-guess[3];
guess[1]=0.0;
guess[2]=1.0;
}
}
else {
if(mode_==0) {
{guess[0]=0.5; guess[1]=0.5;} // default guess for beta calibration
}
if(mode_==4) {
{guess[0]=1.0; guess[1]=1.0;} // default guess for nu calibration
}
if(mode_==2) {
for(int i=skipOptPillars_;i<noOptPillars_;i++)
guess[i]=0.5; // default guess for beta calibration
}
if(mode_==6) {
for(int i=skipOptPillars_;i<noOptPillars_;i++)
guess[i]=1.0; // default guess for nu calibration
}
}
Problem prblm(*costfct,constraint,guess);
EndCriteria ec(MAXIT_, MAXST_, OPTACCURACY_, OPTACCURACY_, OPTACCURACY_);
EndCriteria::Type ret;
if(optimizer_==0) {
Simplex method(0.5);
ret=method.minimize(prblm,ec);
}
if(optimizer_==1) {
LevenbergMarquardt method;
ret=method.minimize(prblm,ec);
}
if(optimizer_==2) {
SimulatedAnnealing method(0.1,0,50,500,2.5,0.2);
ret=method.minimize(prblm,ec);
}
//SteepestDescent method;
//LevenbergMarquardt method;
//SimulatedAnnealing method(0.1,0,50,500,2.5,0.2);
//FILE *out=fopen("costCms.log","a");
//fprintf(out,"Opt Settings: MAXIT=%d, MAXST=%d, OPTACCURACY=%1.5e\n",MAXIT,MAXST,OPTACCURACY);
//fclose(out);
Array x=prblm.currentValue();
double y=costfct->value(x);//prblm.functionValue(); to make sure, the model is set to optimal value
if(ret==EndCriteria::None) QL_FAIL("Optimizer returns status none");
if(ret==EndCriteria::MaxIterations) QL_FAIL("Optimizer returns status max iterations");
if(ret==EndCriteria::Unknown) QL_FAIL("Optimizer returns status unknown");
//sumRmse+=y;
//}
if(useStdCube_) volCube2_->updateAfterRecalibration(); // recalibrate
return y; //sumRmse;
}
