// Filtre avec une stratégie Online
void TwoObjectivesInstance::onlineFiltering()
{
for(int i = 0; i < SOLUTIONS; i++)
{
#if SHOW_DEBUGS
cout << endl << "Solution N°" << i+1 << endl;
#endif // SHOW_DEBUGS
for(int j = 0; j <= i; j++)
{
if( (m_solutions[i].Getdistance() < m_solutions[j].Getdistance()) && (m_solutions[i].Getcost() < m_solutions[j].Getcost()) )
this->m_solutions[j].SetOnlineDominated(true);
else if ((m_solutions[i].Getdistance() > m_solutions[j].Getdistance())&&(m_solutions[i].Getcost() > m_solutions[j].Getcost()))
this->m_solutions[i].SetOnlineDominated(true);
}
}
saveSolutions("online500_" + this->m_Name + ".txt");
savePareto("onlinePareto500_" + this->m_Name + ".txt", ONLINE);
}
// Filtre avec une stratégie Offline
void TwoObjectivesInstance::offlineFiltering()
{
for(int i = 0; i < SOLUTIONS; ++i)
{
for(int j = i+1; j < SOLUTIONS; ++j)
{
if((m_solutions[i].Getdistance() > m_solutions[j].Getdistance())&&(m_solutions[i].Getcost() > m_solutions[j].Getcost()))
{
m_solutions[i].SetOfflineDominated(true);
}
else if ((m_solutions[i].Getdistance() < m_solutions[j].Getdistance())&&(m_solutions[i].Getcost() < m_solutions[j].Getcost()))
{
m_solutions[j].SetOfflineDominated(true);
}
}
}
saveSolutions("offline500_" + this->m_Name + ".txt");
savePareto("offlinePareto500_" + this->m_Name + ".txt", OFFLINE);
}
///
/// Perform 1d Prob scan.
///
/// - Scan range defined through limit "scan".
/// - Will fill the hCL histogram with the 1-CL curve.
/// - Start at a scan value that is in the middle of the allowed
/// range, preferably a solution, and scan up and down from there.
/// - use the "probforce" command line flag to enable force minimum finding
///
/// \param fast This will scan each scanpoint only once.
/// \param reverse This will scan in reverse direction.
/// When using the drag mode, this can sometimes make a difference.
/// \return status: 2 = new global minimum found, 1 = error
///
int MethodProbScan::scan1d(bool fast, bool reverse)
{
if ( arg->debug ) cout << "MethodProbScan::scan1d() : starting ... " << endl;
nScansDone++;
// The "improve" method doesn't need multiple scans.
if ( arg->probforce || arg->probimprove ) fast = true;
if ( arg->probforce ) scanDisableDragMode = true;
// Save parameter values that were active at function call.
if ( startPars ) delete startPars;
startPars = new RooDataSet("startPars", "startPars", *w->set(parsName));
startPars->add(*w->set(parsName));
// // start scan from global minimum (not always a good idea as we need to set from other places as well)
// setParameters(w, parsName, globalMin);
// load scan parameter and scan range
setLimit(w, scanVar1, "scan");
RooRealVar *par = w->var(scanVar1);
assert(par);
float min = hCL->GetXaxis()->GetXmin();
float max = hCL->GetXaxis()->GetXmax();
if ( fabs(par->getMin()-min)>1e-6 || fabs(par->getMax()-max)>1e-6 ){
cout << "MethodProbScan::scan1d() : WARNING : Scan range was changed after initScan()" << endl;
cout << " was called so the old range will be used." << endl;
}
if ( arg->verbose ){
cout << "\nProb configuration:" << endl;
cout << " combination : " << title << endl;
cout << " scan variable : " << scanVar1 << endl;
cout << " scan range : " << min << " ... " << max << endl;
cout << " scan steps : " << nPoints1d << endl;
cout << " fast mode : " << fast << endl;
cout << endl;
}
// Set limit to all parameters.
combiner->loadParameterLimits();
// fix scan parameter
par->setConstant(true);
// j =
// 0 : start value -> upper limit
// 1 : upper limit -> start value
// 2 : start value -> lower limit
// 3 : lower limit -> start value
float startValue = par->getVal();
bool scanUp;
// for the status bar
float nTotalSteps = nPoints1d;
nTotalSteps *= fast ? 1 : 2;
float nStep = 0;
float printFreq = nTotalSteps>15 ? 10 : nTotalSteps;
// Report on the smallest new minimum we come across while scanning.
// Sometimes the scan doesn't find the minimum
// that was found before. Warn if this happens.
double bestMinOld = chi2minGlobal;
double bestMinFoundInScan = 100.;
for ( int jj=0; jj<4; jj++ )
{
int j = jj;
if ( reverse ) switch(jj)
{
case 0: j = 2; break;
case 1: j = 3; break;
case 2: j = 0; break;
case 3: j = 1; break;
}
float scanStart, scanStop;
switch(j)
{
case 0:
// UP
setParameters(w, parsName, startPars->get(0));
scanStart = startValue;
scanStop = par->getMax();
scanUp = true;
break;
case 1:
// DOWN
scanStart = par->getMax();
scanStop = startValue;
scanUp = false;
break;
case 2:
// DOWN
setParameters(w, parsName, startPars->get(0));
scanStart = startValue;
scanStop = par->getMin();
scanUp = false;
break;
case 3:
// UP
scanStart = par->getMin();
scanStop = startValue;
scanUp = true;
break;
}
if ( fast && ( j==1 || j==3 ) ) continue;
for ( int i=0; i<nPoints1d; i++ )
{
float scanvalue;
if ( scanUp )
{
scanvalue = min + (max-min)*(double)i/(double)nPoints1d + hCL->GetBinWidth(1)/2.;
if ( scanvalue < scanStart ) continue;
if ( scanvalue > scanStop ) break;
}
else
{
scanvalue = max - (max-min)*(double)(i+1)/(double)nPoints1d + hCL->GetBinWidth(1)/2.;
if ( scanvalue > scanStart ) continue;
if ( scanvalue < scanStop ) break;
}
// disable drag mode
// (the improve method doesn't work with drag mode as parameter run
// at their limits)
if ( scanDisableDragMode ) setParameters(w, parsName, startPars->get(0));
// set the parameter of interest to the scan point
par->setVal(scanvalue);
// don't scan in unphysical region
if ( scanvalue < par->getMin() || scanvalue > par->getMax() ) continue;
// status bar
if ( (((int)nStep % (int)(nTotalSteps/printFreq)) == 0))
cout << "MethodProbScan::scan1d() : scanning " << (float)nStep/(float)nTotalSteps*100. << "% \r" << flush;
// fit!
RooFitResult *fr = 0;
if ( arg->probforce ) fr = fitToMinForce(w, combiner->getPdfName());
else if ( arg->probimprove ) fr = fitToMinImprove(w, combiner->getPdfName());
else fr = fitToMinBringBackAngles(w->pdf(pdfName), false, -1);
double chi2minScan = fr->minNll();
if ( std::isinf(chi2minScan) ) chi2minScan=1e4; // else the toys in PDF_testConstraint don't work
RooSlimFitResult *r = new RooSlimFitResult(fr); // try to save memory by using the slim fit result
delete fr;
allResults.push_back(r);
bestMinFoundInScan = TMath::Min((double)chi2minScan, (double)bestMinFoundInScan);
TString warningChi2Neg;
if ( chi2minScan < 0 ){
float newChi2minScan = chi2minGlobal + 25.; // 5sigma more than best point
warningChi2Neg = "MethodProbScan::scan1d() : WARNING : " + title;
warningChi2Neg += TString(Form(" chi2 negative for scan point %i: %f",i,chi2minScan));
warningChi2Neg += " setting to: " + TString(Form("%f",newChi2minScan));
//cout << warningChi2Neg << "\r" << flush;
cout << warningChi2Neg << endl;
chi2minScan = newChi2minScan;
}
// If we find a minimum smaller than the old "global" minimum, this means that all
// previous 1-CL values are too high.
if ( chi2minScan<chi2minGlobal ){
if ( arg->verbose ) cout << "MethodProbScan::scan1d() : WARNING : '" << title << "' new global minimum found! "
<< " chi2minScan=" << chi2minScan << endl;
chi2minGlobal = chi2minScan;
// recompute previous 1-CL values
for ( int k=1; k<=hCL->GetNbinsX(); k++ ){
hCL->SetBinContent(k, TMath::Prob(hChi2min->GetBinContent(k)-chi2minGlobal, 1));
}
}
double deltaChi2 = chi2minScan - chi2minGlobal;
double oneMinusCL = TMath::Prob(deltaChi2, 1);
// Save the 1-CL value and the corresponding fit result.
// But only if better than before!
if ( hCL->GetBinContent(hCL->FindBin(scanvalue)) <= oneMinusCL ){
hCL->SetBinContent(hCL->FindBin(scanvalue), oneMinusCL);
hChi2min->SetBinContent(hCL->FindBin(scanvalue), chi2minScan);
int iRes = hCL->FindBin(scanvalue)-1;
curveResults[iRes] = r;
}
nStep++;
}
}
cout << "MethodProbScan::scan1d() : scan done. " << endl;
if ( bestMinFoundInScan-bestMinOld > 0.01 ){
cout << "MethodProbScan::scan1d() : WARNING: Scan didn't find similar minimum to what was found before!" << endl;
cout << "MethodProbScan::scan1d() : Too strict parameter limits? Too coarse scan steps? Didn't load global minimum?" << endl;
cout << "MethodProbScan::scan1d() : chi2 bestMinFoundInScan=" << bestMinFoundInScan << ", bestMinOld=" << bestMinOld << endl;
}
// attempt to correct for undercoverage
if (pvalueCorrectorSet) {
for ( int k=1; k<=hCL->GetNbinsX(); k++ ){
double pvalueProb = hCL->GetBinContent(k);
pvalueProb = pvalueCorrector->transform(pvalueProb);
hCL->SetBinContent(k, pvalueProb);
}
}
setParameters(w, parsName, startPars->get(0));
saveSolutions();
if (arg->confirmsols) confirmSolutions();
if ( (bestMinFoundInScan-bestMinOld)/bestMinOld > 0.01 ) return 1;
return 0;
}
