LMU/musrfit
LMU/musrfit
5
0
Fork 0
Code Issues 6 Pull Requests Actions Packages Projects Releases 3 Activity

Removed some IF clauses from the chisq loops in order to enhance the fitting performance

Browse Source
This commit is contained in:
Bastian M. Wojek 2011-05-22 07:54:03 +00:00
parent 708cb0115b
commit 8fbadeb506
2 changed files with 76 additions and 49 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
src/classes/PRunAsymmetry.cpp
View File

@ -166,6 +166,14 @@ Double_t PRunAsymmetry::CalcChiSquare(const std::vector<Double_t>& par)
Double_t time(1.0);
Int_t i, N(static_cast<Int_t>(fData.GetValue()->size()));
// In order not to have an IF in the next loop, determine the start and end bins for the fit range now
Int_t startTimeBin = static_cast<Int_t>(ceil((fFitStartTime - fData.GetDataTimeStart())/fData.GetDataTimeStep()));
if (startTimeBin < 0)
startTimeBin = 0;
Int_t endTimeBin = static_cast<Int_t>(floor((fFitEndTime - fData.GetDataTimeStart())/fData.GetDataTimeStep())) + 1;
if (endTimeBin >= N)
endTimeBin = N;
// Calculate the theory function once to ensure one function evaluation for the current set of parameters.
// This is needed for the LF and user functions where some non-thread-save calculations only need to be calculated once
// for a given set of parameters---which should be done outside of the parallelized loop.
@ -173,37 +181,38 @@ Double_t PRunAsymmetry::CalcChiSquare(const std::vector<Double_t>& par)
asymFcnValue = fTheory->Func(time, par, fFuncValues);
#ifdef HAVE_GOMP
#pragma omp parallel for default(shared) private(i,time,diff,asymFcnValue) schedule(dynamic,N/(2*omp_get_num_procs())) reduction(+:chisq)
Int_t chunk = (endTimeBin - startTimeBin)/omp_get_num_procs();
if (chunk < 10)
chunk = 10;
#pragma omp parallel for default(shared) private(i,time,diff,asymFcnValue,a,b,f) schedule(dynamic,chunk) reduction(+:chisq)
#endif
for (i=0; i < N; ++i) {
for (i=startTimeBin; i < endTimeBin; ++i) {
time = fData.GetDataTimeStart() + (Double_t)i*fData.GetDataTimeStep();
if ((time>=fFitStartTime) && (time<=fFitEndTime)) {
switch (fAlphaBetaTag) {
case 1: // alpha == 1, beta == 1
asymFcnValue = fTheory->Func(time, par, fFuncValues);
break;
case 2: // alpha != 1, beta == 1
a = par[fRunInfo->GetAlphaParamNo()-1];
f = fTheory->Func(time, par, fFuncValues);
asymFcnValue = (f*(a+1.0)-(a-1.0))/((a+1.0)-f*(a-1.0));
break;
case 3: // alpha == 1, beta != 1
b = par[fRunInfo->GetBetaParamNo()-1];
f = fTheory->Func(time, par, fFuncValues);
asymFcnValue = f*(b+1.0)/(2.0-f*(b-1.0));
break;
case 4: // alpha != 1, beta != 1
a = par[fRunInfo->GetAlphaParamNo()-1];
b = par[fRunInfo->GetBetaParamNo()-1];
f = fTheory->Func(time, par, fFuncValues);
asymFcnValue = (f*(a*b+1.0)-(a-1.0))/((a+1.0)-f*(a*b-1.0));
break;
default:
break;
}
diff = fData.GetValue()->at(i) - asymFcnValue;
chisq += diff*diff / (fData.GetError()->at(i)*fData.GetError()->at(i));
switch (fAlphaBetaTag) {
case 1: // alpha == 1, beta == 1
asymFcnValue = fTheory->Func(time, par, fFuncValues);
break;
case 2: // alpha != 1, beta == 1
a = par[fRunInfo->GetAlphaParamNo()-1];
f = fTheory->Func(time, par, fFuncValues);
asymFcnValue = (f*(a+1.0)-(a-1.0))/((a+1.0)-f*(a-1.0));
break;
case 3: // alpha == 1, beta != 1
b = par[fRunInfo->GetBetaParamNo()-1];
f = fTheory->Func(time, par, fFuncValues);
asymFcnValue = f*(b+1.0)/(2.0-f*(b-1.0));
break;
case 4: // alpha != 1, beta != 1
a = par[fRunInfo->GetAlphaParamNo()-1];
b = par[fRunInfo->GetBetaParamNo()-1];
f = fTheory->Func(time, par, fFuncValues);
asymFcnValue = (f*(a*b+1.0)-(a-1.0))/((a+1.0)-f*(a*b-1.0));
break;
default:
break;
}
diff = fData.GetValue()->at(i) - asymFcnValue;
chisq += diff*diff / (fData.GetError()->at(i)*fData.GetError()->at(i));
}
return chisq;

60
src/classes/PRunSingleHisto.cpp
View File

@ -150,6 +150,14 @@ Double_t PRunSingleHisto::CalcChiSquare(const std::vector<Double_t>& par)
Double_t time(1.0);
Int_t i, N(static_cast<Int_t>(fData.GetValue()->size()));
// In order not to have an IF in the next loop, determine the start and end bins for the fit range now
Int_t startTimeBin = static_cast<Int_t>(ceil((fFitStartTime - fData.GetDataTimeStart())/fData.GetDataTimeStep()));
if (startTimeBin < 0)
startTimeBin = 0;
Int_t endTimeBin = static_cast<Int_t>(floor((fFitEndTime - fData.GetDataTimeStart())/fData.GetDataTimeStep())) + 1;
if (endTimeBin >= N)
endTimeBin = N;
// Calculate the theory function once to ensure one function evaluation for the current set of parameters.
// This is needed for the LF and user functions where some non-thread-save calculations only need to be calculated once
// for a given set of parameters---which should be done outside of the parallelized loop.
@ -157,15 +165,16 @@ Double_t PRunSingleHisto::CalcChiSquare(const std::vector<Double_t>& par)
time = fTheory->Func(time, par, fFuncValues);
#ifdef HAVE_GOMP
#pragma omp parallel for default(shared) private(i,time,diff) schedule(dynamic,N/(2*omp_get_num_procs())) reduction(+:chisq)
Int_t chunk = (endTimeBin - startTimeBin)/omp_get_num_procs();
if (chunk < 10)
chunk = 10;
#pragma omp parallel for default(shared) private(i,time,diff) schedule(dynamic,chunk) reduction(+:chisq)
#endif
for (i=0; i < N; ++i) {
for (i=startTimeBin; i < endTimeBin; ++i) {
time = fData.GetDataTimeStart() + (Double_t)i*fData.GetDataTimeStep();
if ((time>=fFitStartTime) && (time<=fFitEndTime)) {
diff = fData.GetValue()->at(i) -
(N0*TMath::Exp(-time/tau)*(1.0+fTheory->Func(time, par, fFuncValues))+bkg);
chisq += diff*diff / (fData.GetError()->at(i)*fData.GetError()->at(i));
}
diff = fData.GetValue()->at(i) -
(N0*TMath::Exp(-time/tau)*(1.0+fTheory->Func(time, par, fFuncValues))+bkg);
chisq += diff*diff / (fData.GetError()->at(i)*fData.GetError()->at(i));
}
// the correction factor is need since the data scales like pack*t_res,
@ -240,6 +249,14 @@ Double_t PRunSingleHisto::CalcMaxLikelihood(const std::vector<Double_t>& par)
if (fScaleN0AndBkg)
normalizer = fRunInfo->GetPacking() * (fTimeResolution * 1.0e3);
// In order not to have an IF in the next loop, determine the start and end bins for the fit range now
Int_t startTimeBin = static_cast<Int_t>(ceil((fFitStartTime - fData.GetDataTimeStart())/fData.GetDataTimeStep()));
if (startTimeBin < 0)
startTimeBin = 0;
Int_t endTimeBin = static_cast<Int_t>(floor((fFitEndTime - fData.GetDataTimeStart())/fData.GetDataTimeStep())) + 1;
if (endTimeBin >= N)
endTimeBin = N;
// Calculate the theory function once to ensure one function evaluation for the current set of parameters.
// This is needed for the LF and user functions where some non-thread-save calculations only need to be calculated once
// for a given set of parameters---which should be done outside of the parallelized loop.
@ -247,22 +264,23 @@ Double_t PRunSingleHisto::CalcMaxLikelihood(const std::vector<Double_t>& par)
time = fTheory->Func(time, par, fFuncValues);
#ifdef HAVE_GOMP
#pragma omp parallel for default(shared) private(i,time,theo,data) schedule(dynamic,N/(2*omp_get_num_procs())) reduction(-:mllh)
Int_t chunk = (endTimeBin - startTimeBin)/omp_get_num_procs();
if (chunk < 10)
chunk = 10;
#pragma omp parallel for default(shared) private(i,time,theo,data) schedule(dynamic,chunk) reduction(-:mllh)
#endif
for (i=0; i < N; ++i) {
for (i=startTimeBin; i < endTimeBin; ++i) {
time = fData.GetDataTimeStart() + (Double_t)i*fData.GetDataTimeStep();
if ((time>=fFitStartTime) && (time<=fFitEndTime)) {
// calculate theory for the given parameter set
theo = N0*TMath::Exp(-time/tau)*(1+fTheory->Func(time, par, fFuncValues))+bkg;
theo *= normalizer;
// check if data value is not too small
if (fData.GetValue()->at(i) > 1.0e-9)
data = normalizer*fData.GetValue()->at(i);
else
data = 1.0e-9;
// add maximum log likelihood contribution of bin i
mllh -= data*TMath::Log(theo) - theo - TMath::LnGamma(data+1);
}
// calculate theory for the given parameter set
theo = N0*TMath::Exp(-time/tau)*(1+fTheory->Func(time, par, fFuncValues))+bkg;
theo *= normalizer;
// check if data value is not too small
if (fData.GetValue()->at(i) > 1.0e-9)
data = normalizer*fData.GetValue()->at(i);
else
data = 1.0e-9;
// add maximum log likelihood contribution of bin i
mllh -= data*TMath::Log(theo) - theo - TMath::LnGamma(data+1);
}
return mllh;