Page Menu
Home
HEPForge
Search
Configure Global Search
Log In
Files
F10881329
No One
Temporary
Actions
View File
Edit File
Delete File
View Transforms
Subscribe
Mute Notifications
Award Token
Flag For Later
Size
59 KB
Subscribers
None
View Options
diff --git a/src/FitBase/Measurement1D.cxx b/src/FitBase/Measurement1D.cxx
index b60e459..d9f0444 100644
--- a/src/FitBase/Measurement1D.cxx
+++ b/src/FitBase/Measurement1D.cxx
@@ -1,1899 +1,1896 @@
// Copyright 2016 L. Pickering, P. Stowell, R. Terri, C. Wilkinson, C. Wret
/*******************************************************************************
* This ile is part of NUISANCE.
*
* NUISANCE is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* NUISANCE is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with NUISANCE. If not, see <http://www.gnu.org/licenses/>.
*******************************************************************************/
#include "Measurement1D.h"
//********************************************************************
Measurement1D::Measurement1D(void) {
//********************************************************************
// XSec Scalings
fScaleFactor = -1.0;
fCurrentNorm = 1.0;
// Histograms
fDataHist = NULL;
fDataTrue = NULL;
fMCHist = NULL;
fMCFine = NULL;
fMCWeighted = NULL;
fMaskHist = NULL;
// Covar
covar = NULL;
fFullCovar = NULL;
fShapeCovar = NULL;
fCovar = NULL;
fInvert = NULL;
fDecomp = NULL;
// Fake Data
fFakeDataInput = "";
fFakeDataFile = NULL;
// Options
fDefaultTypes = "FIX/FULL/CHI2";
fAllowedTypes =
"FIX,FREE,SHAPE/FULL,DIAG/CHI2/NORM/ENUCORR/Q2CORR/ENU1D/MASK/NOWIDTH";
fIsFix = false;
fIsShape = false;
fIsFree = false;
fIsDiag = false;
fIsFull = false;
fAddNormPen = false;
fIsMask = false;
fIsChi2SVD = false;
fIsRawEvents = false;
fIsNoWidth = false;
fIsDifXSec = false;
fIsEnu1D = false;
// Inputs
fInput = NULL;
fRW = NULL;
// Extra Histograms
fMCHist_Modes = NULL;
}
//********************************************************************
Measurement1D::~Measurement1D(void) {
//********************************************************************
if (fDataHist) delete fDataHist;
if (fDataTrue) delete fDataTrue;
if (fMCHist) delete fMCHist;
if (fMCFine) delete fMCFine;
if (fMCWeighted) delete fMCWeighted;
if (fMaskHist) delete fMaskHist;
if (covar) delete covar;
if (fFullCovar) delete fFullCovar;
if (fShapeCovar) delete fShapeCovar;
if (fCovar) delete fCovar;
if (fInvert) delete fInvert;
if (fDecomp) delete fDecomp;
}
//********************************************************************
void Measurement1D::FinaliseSampleSettings() {
//********************************************************************
MeasurementBase::FinaliseSampleSettings();
// Setup naming + renaming
fName = fSettings.GetName();
fSettings.SetS("originalname", fName);
if (fSettings.Has("rename")) {
fName = fSettings.GetS("rename");
fSettings.SetS("name", fName);
}
// Setup all other options
LOG(SAM) << "Finalising Sample Settings: " << fName << std::endl;
if ((fSettings.GetS("originalname").find("Evt") != std::string::npos)) {
fIsRawEvents = true;
LOG(SAM) << "Found event rate measurement but using poisson likelihoods."
<< std::endl;
}
if (fSettings.GetS("originalname").find("XSec_1DEnu") != std::string::npos) {
fIsEnu1D = true;
LOG(SAM) << "::" << fName << "::" << std::endl;
LOG(SAM) << "Found XSec Enu measurement, applying flux integrated scaling, "
<< "not flux averaged!" << std::endl;
}
if (fIsEnu1D && fIsRawEvents) {
- LOG(SAM) << "Found 1D Enu XSec distribution AND fIsRawEvents, is this "
+ ERR(FTL) << "Found 1D Enu XSec distribution AND fIsRawEvents, is this "
"really correct?!"
<< std::endl;
- LOG(SAM) << "Check experiment constructor for " << fName
+ ERR(FTL) << "Check experiment constructor for " << fName
<< " and correct this!" << std::endl;
- LOG(SAM) << "I live in " << __FILE__ << ":" << __LINE__ << std::endl;
- exit(-1);
+ ERR(FTL) << "I live in " << __FILE__ << ":" << __LINE__ << std::endl;
+ throw;
}
if (!fRW) fRW = FitBase::GetRW();
if (!fInput and !fIsJoint) SetupInputs(fSettings.GetS("input"));
// Setup options
SetFitOptions(fDefaultTypes); // defaults
SetFitOptions(fSettings.GetS("type")); // user specified
EnuMin = GeneralUtils::StrToDbl(fSettings.GetS("enu_min"));
EnuMax = GeneralUtils::StrToDbl(fSettings.GetS("enu_max"));
if (fAddNormPen) {
if (fNormError <= 0.0) {
ERR(WRN) << "Norm error for class " << fName << " is 0.0!" << std::endl;
ERR(WRN) << "If you want to use it please add fNormError=VAL" << std::endl;
throw;
}
}
}
//********************************************************************
void Measurement1D::CreateDataHistogram(int dimx, double* binx) {
//********************************************************************
if (fDataHist) delete fDataHist;
fDataHist = new TH1D( (fSettings.GetName() + "_data").c_str(), (fSettings.GetFullTitles()).c_str(),
dimx, binx) ;
}
//********************************************************************
void Measurement1D::SetDataFromTextFile(std::string datafile) {
//********************************************************************
LOG(SAM) << "Reading data from text file: " << datafile << std::endl;
fDataHist = PlotUtils::GetTH1DFromFile(datafile,
fSettings.GetName() + "_data",
fSettings.GetFullTitles());
}
//********************************************************************
void Measurement1D::SetDataFromRootFile(std::string datafile,
std::string histname) {
//********************************************************************
LOG(SAM) << "Reading data from root file: " << datafile << ";" << histname << std::endl;
fDataHist = PlotUtils::GetTH1DFromRootFile(datafile, histname);
fDataHist->SetNameTitle((fSettings.GetName() + "_data").c_str(),
(fSettings.GetFullTitles()).c_str());
return;
};
//********************************************************************
void Measurement1D::SetEmptyData(){
//********************************************************************
fDataHist = new TH1D("EMPTY_DATA","EMPTY_DATA",1,0.0,1.0);
}
//********************************************************************
void Measurement1D::SetPoissonErrors() {
//********************************************************************
if (!fDataHist) {
ERR(FTL) << "Need a data hist to setup possion errors! " << std::endl;
ERR(FTL) << "Setup Data First!" << std::endl;
throw;
}
for (int i = 0; i < fDataHist->GetNbinsX() + 1; i++) {
fDataHist->SetBinError(i + 1, sqrt(fDataHist->GetBinContent(i + 1)));
}
}
//********************************************************************
void Measurement1D::SetCovarFromDiagonal(TH1D* data) {
//********************************************************************
if (!data and fDataHist) {
data = fDataHist;
}
if (data) {
LOG(SAM) << "Setting diagonal covariance for: " << data->GetName() << std::endl;
fFullCovar = StatUtils::MakeDiagonalCovarMatrix(data);
covar = StatUtils::GetInvert(fFullCovar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
} else {
ERR(FTL) << "No data input provided to set diagonal covar from!" << std::endl;
}
// if (!fIsDiag) {
// ERR(FTL) << "SetCovarMatrixFromDiag called for measurement "
// << "that is not set as diagonal." << std::endl;
// throw;
// }
}
//********************************************************************
void Measurement1D::SetCovarFromTextFile(std::string covfile, int dim) {
//********************************************************************
if (dim == -1) {
dim = fDataHist->GetNbinsX();
}
LOG(SAM) << "Reading covariance from text file: " << covfile << std::endl;
fFullCovar = StatUtils::GetCovarFromTextFile(covfile, dim);
covar = StatUtils::GetInvert(fFullCovar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
}
//********************************************************************
void Measurement1D::SetCovarFromMultipleTextFiles(std::string covfiles, int dim) {
//********************************************************************
if (dim == -1) {
dim = fDataHist->GetNbinsX();
}
std::vector<std::string> covList = GeneralUtils::ParseToStr(covfiles, ";");
fFullCovar = new TMatrixDSym(dim);
for (uint i = 0; i < covList.size(); ++i){
LOG(SAM) << "Reading covariance from text file: " << covList[i] << std::endl;
TMatrixDSym* temp_cov = StatUtils::GetCovarFromTextFile(covList[i], dim);
(*fFullCovar) += (*temp_cov);
delete temp_cov;
}
covar = StatUtils::GetInvert(fFullCovar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
}
//********************************************************************
void Measurement1D::SetCovarFromRootFile(std::string covfile, std::string histname) {
//********************************************************************
LOG(SAM) << "Reading covariance from text file: " << covfile << ";" << histname << std::endl;
fFullCovar = StatUtils::GetCovarFromRootFile(covfile, histname);
covar = StatUtils::GetInvert(fFullCovar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
}
//********************************************************************
void Measurement1D::SetCovarInvertFromTextFile(std::string covfile, int dim) {
//********************************************************************
if (dim == -1) {
dim = fDataHist->GetNbinsX();
}
LOG(SAM) << "Reading inverted covariance from text file: " << covfile << std::endl;
covar = StatUtils::GetCovarFromTextFile(covfile, dim);
fFullCovar = StatUtils::GetInvert(covar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
}
//********************************************************************
void Measurement1D::SetCovarInvertFromRootFile(std::string covfile, std::string histname) {
//********************************************************************
LOG(SAM) << "Reading inverted covariance from text file: " << covfile << ";" << histname << std::endl;
covar = StatUtils::GetCovarFromRootFile(covfile, histname);
fFullCovar = StatUtils::GetInvert(covar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
}
//********************************************************************
void Measurement1D::SetCorrelationFromTextFile(std::string covfile, int dim) {
//********************************************************************
if (dim == -1) dim = fDataHist->GetNbinsX();
LOG(SAM) << "Reading data correlations from text file: " << covfile << ";" << dim << std::endl;
TMatrixDSym* correlation = StatUtils::GetCovarFromTextFile(covfile, dim);
if (!fDataHist) {
ERR(FTL) << "Trying to set correlations from text file but there is no data to build it from. \n"
<< "In constructor make sure data is set before SetCorrelationFromTextFile is called. \n" << std::endl;
throw;
}
// Fill covar from data errors and correlations
fFullCovar = new TMatrixDSym(dim);
for (int i = 0; i < fDataHist->GetNbinsX(); i++) {
for (int j = 0; j < fDataHist->GetNbinsX(); j++) {
(*fFullCovar)(i, j) = (*correlation)(i, j) * fDataHist->GetBinError(i + 1) * fDataHist->GetBinError(j + 1) * 1.E76;
}
}
// Fill other covars.
covar = StatUtils::GetInvert(fFullCovar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
delete correlation;
}
//********************************************************************
void Measurement1D::SetCorrelationFromMultipleTextFiles(std::string corrfiles, int dim) {
//********************************************************************
if (dim == -1) {
dim = fDataHist->GetNbinsX();
}
std::vector<std::string> corrList = GeneralUtils::ParseToStr(corrfiles, ";");
fFullCovar = new TMatrixDSym(dim);
for (uint i = 0; i < corrList.size(); ++i){
LOG(SAM) << "Reading covariance from text file: " << corrList[i] << std::endl;
TMatrixDSym* temp_cov = StatUtils::GetCovarFromTextFile(corrList[i], dim);
for (int i = 0; i < fDataHist->GetNbinsX(); i++) {
for (int j = 0; j < fDataHist->GetNbinsX(); j++) {
(*temp_cov)(i, j) = (*temp_cov)(i, j) * fDataHist->GetBinError(i + 1) * fDataHist->GetBinError(j + 1) * 1.E76;
}
}
(*fFullCovar) += (*temp_cov);
delete temp_cov;
}
covar = StatUtils::GetInvert(fFullCovar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
}
//********************************************************************
void Measurement1D::SetCorrelationFromRootFile(std::string covfile, std::string histname) {
//********************************************************************
LOG(SAM) << "Reading data correlations from text file: " << covfile << ";" << histname << std::endl;
TMatrixDSym* correlation = StatUtils::GetCovarFromRootFile(covfile, histname);
if (!fDataHist) {
ERR(FTL) << "Trying to set correlations from text file but there is no data to build it from. \n"
<< "In constructor make sure data is set before SetCorrelationFromTextFile is called. \n" << std::endl;
throw;
}
// Fill covar from data errors and correlations
fFullCovar = new TMatrixDSym(fDataHist->GetNbinsX());
for (int i = 0; i < fDataHist->GetNbinsX(); i++) {
for (int j = 0; j < fDataHist->GetNbinsX(); j++) {
(*fFullCovar)(i, j) = (*correlation)(i, j) * fDataHist->GetBinError(i + 1) * fDataHist->GetBinError(j + 1) * 1.E76;
}
}
// Fill other covars.
covar = StatUtils::GetInvert(fFullCovar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
delete correlation;
}
//********************************************************************
void Measurement1D::SetCholDecompFromTextFile(std::string covfile, int dim) {
//********************************************************************
if (dim == -1) {
dim = fDataHist->GetNbinsX();
}
LOG(SAM) << "Reading cholesky from text file: " << covfile << std::endl;
TMatrixD* temp = StatUtils::GetMatrixFromTextFile(covfile, dim, dim);
TMatrixD* trans = (TMatrixD*)temp->Clone();
trans->T();
(*trans) *= (*temp);
fFullCovar = new TMatrixDSym(dim, trans->GetMatrixArray(), "");
covar = StatUtils::GetInvert(fFullCovar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
delete temp;
delete trans;
}
//********************************************************************
void Measurement1D::SetCholDecompFromRootFile(std::string covfile, std::string histname) {
//********************************************************************
LOG(SAM) << "Reading cholesky decomp from root file: " << covfile << ";" << histname << std::endl;
TMatrixD* temp = StatUtils::GetMatrixFromRootFile(covfile, histname);
TMatrixD* trans = (TMatrixD*)temp->Clone();
trans->T();
(*trans) *= (*temp);
fFullCovar = new TMatrixDSym(temp->GetNrows(), trans->GetMatrixArray(), "");
covar = StatUtils::GetInvert(fFullCovar);
fDecomp = StatUtils::GetDecomp(fFullCovar);
delete temp;
delete trans;
}
void Measurement1D::SetShapeCovar(){
// Return if this is missing any pre-requisites
if (!fFullCovar) return;
if (!fDataHist) return;
// Also return if it's bloody stupid under the circumstances
if (fIsDiag) return;
fShapeCovar = StatUtils::ExtractShapeOnlyCovar(fFullCovar, fDataHist);
return;
}
//********************************************************************
void Measurement1D::ScaleData(double scale) {
//********************************************************************
fDataHist->Scale(scale);
}
//********************************************************************
void Measurement1D::ScaleDataErrors(double scale) {
//********************************************************************
for (int i = 0; i < fDataHist->GetNbinsX(); i++) {
fDataHist->SetBinError(i + 1, fDataHist->GetBinError(i + 1) * scale);
}
}
//********************************************************************
void Measurement1D::ScaleCovar(double scale) {
//********************************************************************
(*fFullCovar) *= scale;
(*covar) *= 1.0 / scale;
(*fDecomp) *= sqrt(scale);
}
//********************************************************************
void Measurement1D::SetBinMask(std::string maskfile) {
//********************************************************************
if (!fIsMask) return;
LOG(SAM) << "Reading bin mask from file: " << maskfile << std::endl;
// Create a mask histogram with dim of data
int nbins = fDataHist->GetNbinsX();
fMaskHist =
new TH1I((fSettings.GetName() + "_BINMASK").c_str(),
(fSettings.GetName() + "_BINMASK; Bin; Mask?").c_str(), nbins, 0, nbins);
std::string line;
std::ifstream mask(maskfile.c_str(), std::ifstream::in);
if (!mask.is_open()) {
LOG(FTL) << " Cannot find mask file." << std::endl;
throw;
}
while (std::getline(mask >> std::ws, line, '\n')) {
std::vector<int> entries = GeneralUtils::ParseToInt(line, " ");
// Skip lines with poorly formatted lines
if (entries.size() < 2) {
LOG(WRN) << "Measurement1D::SetBinMask(), couldn't parse line: " << line
<< std::endl;
continue;
}
// The first index should be the bin number, the second should be the mask
// value.
int val = 0;
if (entries[1] > 0) val = 1;
fMaskHist->SetBinContent(entries[0], val);
}
// Apply masking by setting masked data bins to zero
PlotUtils::MaskBins(fDataHist, fMaskHist);
return;
}
//********************************************************************
void Measurement1D::FinaliseMeasurement() {
//********************************************************************
LOG(SAM) << "Finalising Measurement: " << fName << std::endl;
if (fSettings.GetB("onlymc")){
if (fDataHist) delete fDataHist;
fDataHist = new TH1D("empty_data","empty_data",1,0.0,1.0);
}
// Make sure data is setup
if (!fDataHist) {
ERR(FTL) << "No data has been setup inside " << fName << " constructor!" << std::endl;
throw;
}
// Make sure covariances are setup
if (!fFullCovar) {
fIsDiag = true;
SetCovarFromDiagonal(fDataHist);
}
if (!covar) {
covar = StatUtils::GetInvert(fFullCovar);
}
if (!fDecomp) {
fDecomp = StatUtils::GetDecomp(fFullCovar);
}
// Push the diagonals of fFullCovar onto the data histogram
// Comment this out until the covariance/data scaling is consistent!
StatUtils::SetDataErrorFromCov(fDataHist, fFullCovar, 1E-38);
// If shape only, set covar and fDecomp using the shape-only matrix (if set)
if (fIsShape && fShapeCovar && FitPar::Config().GetParB("UseShapeCovar")){
if (covar) delete covar;
covar = StatUtils::GetInvert(fShapeCovar);
if (fDecomp) delete fDecomp;
fDecomp = StatUtils::GetDecomp(fFullCovar);
}
// Setup fMCHist from data
fMCHist = (TH1D*)fDataHist->Clone();
fMCHist->SetNameTitle((fSettings.GetName() + "_MC").c_str(),
(fSettings.GetFullTitles()).c_str());
fMCHist->Reset();
// Setup fMCFine
fMCFine = new TH1D("mcfine", "mcfine", fDataHist->GetNbinsX() * 8,
fMCHist->GetBinLowEdge(1),
fMCHist->GetBinLowEdge(fDataHist->GetNbinsX() + 1));
fMCFine->SetNameTitle((fSettings.GetName() + "_MC_FINE").c_str(),
(fSettings.GetFullTitles()).c_str());
fMCFine->Reset();
// Setup MC Stat
fMCStat = (TH1D*)fMCHist->Clone();
fMCStat->Reset();
// Search drawopts for possible types to include by default
std::string drawopts = FitPar::Config().GetParS("drawopts");
if (drawopts.find("MODES") != std::string::npos) {
fMCHist_Modes = new TrueModeStack( (fSettings.GetName() + "_MODES").c_str(),
("True Channels"), fMCHist);
SetAutoProcessTH1(fMCHist_Modes, kCMD_Reset, kCMD_Norm, kCMD_Write);
}
// Setup bin masks using sample name
if (fIsMask) {
std::string curname = fName;
std::string origname = fSettings.GetS("originalname");
// Check rename.mask
std::string maskloc = FitPar::Config().GetParDIR(curname + ".mask");
// Check origname.mask
if (maskloc.empty()) maskloc = FitPar::Config().GetParDIR(origname + ".mask");
// Check database
if (maskloc.empty()) {
maskloc = FitPar::GetDataBase() + "/masks/" + origname + ".mask";
}
// Setup Bin Mask
SetBinMask(maskloc);
}
if (fScaleFactor < 0) {
ERR(FTL) << "I found a negative fScaleFactor in " << __FILE__ << ":" << __LINE__ << std::endl;
ERR(FTL) << "fScaleFactor = " << fScaleFactor << std::endl;
ERR(FTL) << "EXITING" << std::endl;
throw;
}
// Create and fill Weighted Histogram
if (!fMCWeighted) {
fMCWeighted = (TH1D*)fMCHist->Clone();
fMCWeighted->SetNameTitle((fName + "_MCWGHTS").c_str(),
(fName + "_MCWGHTS" + fPlotTitles).c_str());
fMCWeighted->GetYaxis()->SetTitle("Weighted Events");
}
}
//********************************************************************
void Measurement1D::SetFitOptions(std::string opt) {
//********************************************************************
// Do nothing if default given
if (opt == "DEFAULT") return;
// CHECK Conflicting Fit Options
std::vector<std::string> fit_option_allow =
GeneralUtils::ParseToStr(fAllowedTypes, "/");
for (UInt_t i = 0; i < fit_option_allow.size(); i++) {
std::vector<std::string> fit_option_section =
GeneralUtils::ParseToStr(fit_option_allow.at(i), ",");
bool found_option = false;
for (UInt_t j = 0; j < fit_option_section.size(); j++) {
std::string av_opt = fit_option_section.at(j);
if (!found_option and opt.find(av_opt) != std::string::npos) {
found_option = true;
} else if (found_option and opt.find(av_opt) != std::string::npos) {
ERR(FTL) << "ERROR: Conflicting fit options provided: "
<< opt << std::endl
<< "Conflicting group = " << fit_option_section.at(i) << std::endl
<< "You should only supply one of these options in card file." << std::endl;
throw;
}
}
}
// Check all options are allowed
std::vector<std::string> fit_options_input =
GeneralUtils::ParseToStr(opt, "/");
for (UInt_t i = 0; i < fit_options_input.size(); i++) {
if (fAllowedTypes.find(fit_options_input.at(i)) == std::string::npos) {
ERR(FTL) << "ERROR: Fit Option '" << fit_options_input.at(i)
<< "' Provided is not allowed for this measurement."
<< std::endl;
ERR(FTL) << "Fit Options should be provided as a '/' seperated list "
"(e.g. FREE/DIAG/NORM)"
<< std::endl;
ERR(FTL) << "Available options for " << fName << " are '" << fAllowedTypes
<< "'" << std::endl;
throw;
}
}
// Set TYPE
fFitType = opt;
// FIX,SHAPE,FREE
if (opt.find("FIX") != std::string::npos) {
fIsFree = fIsShape = false;
fIsFix = true;
} else if (opt.find("SHAPE") != std::string::npos) {
fIsFree = fIsFix = false;
fIsShape = true;
} else if (opt.find("FREE") != std::string::npos) {
fIsFix = fIsShape = false;
fIsFree = true;
}
// DIAG,FULL (or default to full)
if (opt.find("DIAG") != std::string::npos) {
fIsDiag = true;
fIsFull = false;
} else if (opt.find("FULL") != std::string::npos) {
fIsDiag = false;
fIsFull = true;
}
// CHI2/LL (OTHERS?)
if (opt.find("LOG") != std::string::npos) {
fIsChi2 = false;
ERR(FTL) << "No other LIKELIHOODS properly supported!" << std::endl;
ERR(FTL) << "Try to use a chi2!" << std::endl;
throw;
} else {
fIsChi2 = true;
}
// EXTRAS
if (opt.find("RAW") != std::string::npos) fIsRawEvents = true;
if (opt.find("NOWIDTH") != std::string::npos) fIsNoWidth = true;
if (opt.find("DIF") != std::string::npos) fIsDifXSec = true;
if (opt.find("ENU1D") != std::string::npos) fIsEnu1D = true;
if (opt.find("NORM") != std::string::npos) fAddNormPen = true;
if (opt.find("MASK") != std::string::npos) fIsMask = true;
return;
};
//********************************************************************
void Measurement1D::SetSmearingMatrix(std::string smearfile, int truedim,
int recodim) {
//********************************************************************
// The smearing matrix describes the migration from true bins (rows) to reco
// bins (columns)
// Counter over the true bins!
int row = 0;
std::string line;
std::ifstream smear(smearfile.c_str(), std::ifstream::in);
// Note that the smearing matrix may be rectangular.
fSmearMatrix = new TMatrixD(truedim, recodim);
if (smear.is_open())
LOG(SAM) << "Reading smearing matrix from file: " << smearfile << std::endl;
else
ERR(FTL) << "Smearing matrix provided is incorrect: " << smearfile
<< std::endl;
while (std::getline(smear >> std::ws, line, '\n')) {
int column = 0;
std::vector<double> entries = GeneralUtils::ParseToDbl(line, " ");
for (std::vector<double>::iterator iter = entries.begin();
iter != entries.end(); iter++) {
(*fSmearMatrix)(row, column) =
(*iter) / 100.; // Convert to fraction from
// percentage (this may not be
// general enough)
column++;
}
row++;
}
return;
}
//********************************************************************
void Measurement1D::ApplySmearingMatrix() {
//********************************************************************
if (!fSmearMatrix) {
ERR(WRN) << fName
<< ": attempted to apply smearing matrix, but none was set"
<< std::endl;
return;
}
TH1D* unsmeared = (TH1D*)fMCHist->Clone();
TH1D* smeared = (TH1D*)fMCHist->Clone();
smeared->Reset();
// Loop over reconstructed bins
// true = row; reco = column
for (int rbin = 0; rbin < fSmearMatrix->GetNcols(); ++rbin) {
// Sum up the constributions from all true bins
double rBinVal = 0;
// Loop over true bins
for (int tbin = 0; tbin < fSmearMatrix->GetNrows(); ++tbin) {
rBinVal +=
(*fSmearMatrix)(tbin, rbin) * unsmeared->GetBinContent(tbin + 1);
}
smeared->SetBinContent(rbin + 1, rBinVal);
}
fMCHist = (TH1D*)smeared->Clone();
return;
}
/*
Reconfigure LOOP
*/
//********************************************************************
void Measurement1D::ResetAll() {
//********************************************************************
fMCHist->Reset();
fMCFine->Reset();
fMCStat->Reset();
return;
};
//********************************************************************
void Measurement1D::FillHistograms() {
//********************************************************************
if (Signal) {
QLOG(DEB, "Fill MCHist: " << fXVar << ", " << Weight);
fMCHist->Fill(fXVar, Weight);
fMCFine->Fill(fXVar, Weight);
fMCStat->Fill(fXVar, 1.0);
if (fMCHist_Modes) fMCHist_Modes->Fill(Mode, fXVar, Weight);
}
return;
};
//********************************************************************
void Measurement1D::ScaleEvents() {
//********************************************************************
// Fill MCWeighted;
// for (int i = 0; i < fMCHist->GetNbinsX(); i++) {
// fMCWeighted->SetBinContent(i + 1, fMCHist->GetBinContent(i + 1));
// fMCWeighted->SetBinError(i + 1, fMCHist->GetBinError(i + 1));
// }
// Setup Stat ratios for MC and MC Fine
double* statratio = new double[fMCHist->GetNbinsX()];
for (int i = 0; i < fMCHist->GetNbinsX(); i++) {
if (fMCHist->GetBinContent(i + 1) != 0) {
statratio[i] = fMCHist->GetBinError(i + 1) / fMCHist->GetBinContent(i + 1);
} else {
statratio[i] = 0.0;
}
}
double* statratiofine = new double[fMCFine->GetNbinsX()];
for (int i = 0; i < fMCFine->GetNbinsX(); i++) {
if (fMCFine->GetBinContent(i + 1) != 0) {
statratiofine[i] = fMCFine->GetBinError(i + 1) / fMCFine->GetBinContent(i + 1);
} else {
statratiofine[i] = 0.0;
}
}
// Scaling for raw event rates
if (fIsRawEvents) {
double datamcratio = fDataHist->Integral() / fMCHist->Integral();
fMCHist->Scale(datamcratio);
fMCFine->Scale(datamcratio);
if (fMCHist_Modes) fMCHist_Modes->Scale(datamcratio);
// Scaling for XSec as function of Enu
} else if (fIsEnu1D) {
PlotUtils::FluxUnfoldedScaling(fMCHist, GetFluxHistogram(),
GetEventHistogram(), fScaleFactor,
fNEvents);
PlotUtils::FluxUnfoldedScaling(fMCFine, GetFluxHistogram(),
GetEventHistogram(), fScaleFactor,
fNEvents);
if (fMCHist_Modes) {
// Loop over the modes
fMCHist_Modes->FluxUnfold(GetFluxHistogram(), GetEventHistogram(), fScaleFactor, fNEvents);
//PlotUtils::FluxUnfoldedScaling(fMCHist_Modes, GetFluxHistogram(),
//GetEventHistogram(), fScaleFactor,
//fNEvents);
}
} else if (fIsNoWidth) {
fMCHist->Scale(fScaleFactor);
fMCFine->Scale(fScaleFactor);
if (fMCHist_Modes) fMCHist_Modes->Scale(fScaleFactor);
// Any other differential scaling
} else {
fMCHist->Scale(fScaleFactor, "width");
fMCFine->Scale(fScaleFactor, "width");
if (fMCHist_Modes) fMCHist_Modes->Scale(fScaleFactor, "width");
}
// Proper error scaling - ROOT Freaks out with xsec weights sometimes
for (int i = 0; i < fMCStat->GetNbinsX(); i++) {
fMCHist->SetBinError(i + 1, fMCHist->GetBinContent(i + 1) * statratio[i]);
}
for (int i = 0; i < fMCFine->GetNbinsX(); i++) {
fMCFine->SetBinError(i + 1, fMCFine->GetBinContent(i + 1) * statratiofine[i]);
}
// Clean up
delete[] statratio;
delete[] statratiofine;
return;
};
//********************************************************************
void Measurement1D::ApplyNormScale(double norm) {
//********************************************************************
fCurrentNorm = norm;
fMCHist->Scale(1.0 / norm);
fMCFine->Scale(1.0 / norm);
return;
};
/*
Statistic Functions - Outsources to StatUtils
*/
//********************************************************************
int Measurement1D::GetNDOF() {
//********************************************************************
int ndof = fDataHist->GetNbinsX();
if (fMaskHist and fIsMask) ndof -= fMaskHist->Integral();
return ndof;
}
//********************************************************************
double Measurement1D::GetLikelihood() {
//********************************************************************
// If this is for a ratio, there is no data histogram to compare to!
if (fNoData || !fDataHist) return 0.;
// Apply Masking to MC if Required.
if (fIsMask and fMaskHist) {
PlotUtils::MaskBins(fMCHist, fMaskHist);
}
// Sort Shape Scaling
double scaleF = 0.0;
// TODO Include !fIsRawEvents
if (fIsShape) {
if (fMCHist->Integral(1, fMCHist->GetNbinsX(), "width")) {
scaleF = fDataHist->Integral(1, fDataHist->GetNbinsX(), "width") /
fMCHist->Integral(1, fMCHist->GetNbinsX(), "width");
fMCHist->Scale(scaleF);
fMCFine->Scale(scaleF);
}
}
// Likelihood Calculation
double stat = 0.;
if (fIsChi2) {
if (fIsRawEvents) {
stat = StatUtils::GetChi2FromEventRate(fDataHist, fMCHist, fMaskHist);
} else if (fIsDiag) {
stat = StatUtils::GetChi2FromDiag(fDataHist, fMCHist, fMaskHist);
} else if (!fIsDiag and !fIsRawEvents) {
stat = StatUtils::GetChi2FromCov(fDataHist, fMCHist, covar, fMaskHist);
}
}
// Sort Penalty Terms
if (fAddNormPen) {
double penalty =
(1. - fCurrentNorm) * (1. - fCurrentNorm) / (fNormError * fNormError);
stat += penalty;
}
// Return to normal scaling
if (fIsShape) { // and !FitPar::Config().GetParB("saveshapescaling")) {
fMCHist->Scale(1. / scaleF);
fMCFine->Scale(1. / scaleF);
}
fLikelihood = stat;
return stat;
}
/*
Fake Data Functions
*/
//********************************************************************
void Measurement1D::SetFakeDataValues(std::string fakeOption) {
//********************************************************************
// Setup original/datatrue
TH1D* tempdata = (TH1D*) fDataHist->Clone();
if (!fIsFakeData) {
fIsFakeData = true;
// Make a copy of the original data histogram.
if (!fDataOrig) fDataOrig = (TH1D*)fDataHist->Clone((fName + "_data_original").c_str());
} else {
ResetFakeData();
}
// Setup Inputs
fFakeDataInput = fakeOption;
LOG(SAM) << "Setting fake data from : " << fFakeDataInput << std::endl;
// From MC
if (fFakeDataInput.compare("MC") == 0) {
fDataHist = (TH1D*)fMCHist->Clone((fName + "_MC").c_str());
// Fake File
} else {
if (!fFakeDataFile) fFakeDataFile = new TFile(fFakeDataInput.c_str(), "READ");
fDataHist = (TH1D*)fFakeDataFile->Get((fName + "_MC").c_str());
}
// Setup Data Hist
fDataHist->SetNameTitle((fName + "_FAKE").c_str(),
(fName + fPlotTitles).c_str());
// Replace Data True
if (fDataTrue) delete fDataTrue;
fDataTrue = (TH1D*)fDataHist->Clone();
fDataTrue->SetNameTitle((fName + "_FAKE_TRUE").c_str(),
(fName + fPlotTitles).c_str());
// Make a new covariance for fake data hist.
int nbins = fDataHist->GetNbinsX();
double alpha_i = 0.0;
double alpha_j = 0.0;
for (int i = 0; i < nbins; i++) {
for (int j = 0; j < nbins; j++) {
alpha_i = fDataHist->GetBinContent(i + 1) / tempdata->GetBinContent(i + 1);
alpha_j = fDataHist->GetBinContent(j + 1) / tempdata->GetBinContent(j + 1);
(*fFullCovar)(i, j) = alpha_i * alpha_j * (*fFullCovar)(i, j);
}
}
// Setup Covariances
if (covar) delete covar;
covar = StatUtils::GetInvert(fFullCovar);
if (fDecomp) delete fDecomp;
fDecomp = StatUtils::GetInvert(fFullCovar);
delete tempdata;
return;
};
//********************************************************************
void Measurement1D::ResetFakeData() {
//********************************************************************
if (fIsFakeData) {
if (fDataHist) delete fDataHist;
fDataHist = (TH1D*)fDataTrue->Clone((fSettings.GetName() + "_FKDAT").c_str());
}
}
//********************************************************************
void Measurement1D::ResetData() {
//********************************************************************
if (fIsFakeData) {
if (fDataHist) delete fDataHist;
fDataHist = (TH1D*)fDataOrig->Clone((fSettings.GetName() + "_data").c_str());
}
fIsFakeData = false;
}
//********************************************************************
void Measurement1D::ThrowCovariance() {
//********************************************************************
// Take a fDecomposition and use it to throw the current dataset.
// Requires fDataTrue also be set incase used repeatedly.
if (!fDataTrue) fDataTrue = (TH1D*) fDataHist->Clone();
if (fDataHist) delete fDataHist;
fDataHist = StatUtils::ThrowHistogram(fDataTrue, fFullCovar);
return;
};
//********************************************************************
void Measurement1D::ThrowDataToy(){
//********************************************************************
if (!fDataTrue) fDataTrue = (TH1D*) fDataHist->Clone();
if (fMCHist) delete fMCHist;
fMCHist = StatUtils::ThrowHistogram(fDataTrue, fFullCovar);
}
/*
Access Functions
*/
//********************************************************************
TH1D* Measurement1D::GetMCHistogram() {
//********************************************************************
if (!fMCHist) return fMCHist;
std::ostringstream chi2;
chi2 << std::setprecision(5) << this->GetLikelihood();
int linecolor = kRed;
int linestyle = 1;
int linewidth = 1;
int fillcolor = 0;
int fillstyle = 1001;
// if (fSettings.Has("linecolor")) linecolor = fSettings.GetI("linecolor");
// if (fSettings.Has("linestyle")) linestyle = fSettings.GetI("linestyle");
// if (fSettings.Has("linewidth")) linewidth = fSettings.GetI("linewidth");
// if (fSettings.Has("fillcolor")) fillcolor = fSettings.GetI("fillcolor");
// if (fSettings.Has("fillstyle")) fillstyle = fSettings.GetI("fillstyle");
fMCHist->SetTitle(chi2.str().c_str());
fMCHist->SetLineColor(linecolor);
fMCHist->SetLineStyle(linestyle);
fMCHist->SetLineWidth(linewidth);
fMCHist->SetFillColor(fillcolor);
fMCHist->SetFillStyle(fillstyle);
return fMCHist;
};
//********************************************************************
TH1D* Measurement1D::GetDataHistogram() {
//********************************************************************
if (!fDataHist) return fDataHist;
int datacolor = kBlack;
int datastyle = 1;
int datawidth = 1;
// if (fSettings.Has("datacolor")) datacolor = fSettings.GetI("datacolor");
// if (fSettings.Has("datastyle")) datastyle = fSettings.GetI("datastyle");
// if (fSettings.Has("datawidth")) datawidth = fSettings.GetI("datawidth");
fDataHist->SetLineColor(datacolor);
fDataHist->SetLineWidth(datawidth);
fDataHist->SetMarkerStyle(datastyle);
return fDataHist;
};
/*
Write Functions
*/
// Save all the histograms at once
//********************************************************************
void Measurement1D::Write(std::string drawOpt) {
//********************************************************************
// Get Draw Options
drawOpt = FitPar::Config().GetParS("drawopts");
// Write Settigns
if (drawOpt.find("SETTINGS") != std::string::npos){
fSettings.Set("#chi^{2}",fLikelihood);
fSettings.Set("NDOF", this->GetNDOF() );
fSettings.Set("#chi^{2}/NDOF", fLikelihood / this->GetNDOF() );
fSettings.Write();
}
// Write Data/MC
if (drawOpt.find("DATA") != std::string::npos) GetDataList().at(0)->Write();
if (drawOpt.find("MC") != std::string::npos) {
GetMCList().at(0)->Write();
if((fEvtRateScaleFactor != 0xdeadbeef) && GetMCList().at(0)){
TH1D * PredictedEvtRate = static_cast<TH1D *>(GetMCList().at(0)->Clone());
PredictedEvtRate->Scale(fEvtRateScaleFactor);
PredictedEvtRate->GetYaxis()->SetTitle("Predicted event rate");
PredictedEvtRate->Write();
}
}
// Write Fine Histogram
if (drawOpt.find("FINE") != std::string::npos)
GetFineList().at(0)->Write();
// Write Weighted Histogram
if (drawOpt.find("WEIGHTS") != std::string::npos && fMCWeighted)
fMCWeighted->Write();
// Save Flux/Evt if no event manager
if (!FitPar::Config().GetParB("EventManager")) {
if (drawOpt.find("FLUX") != std::string::npos && GetFluxHistogram())
GetFluxHistogram()->Write();
if (drawOpt.find("EVT") != std::string::npos && GetEventHistogram())
GetEventHistogram()->Write();
if (drawOpt.find("XSEC") != std::string::npos && GetEventHistogram())
GetXSecHistogram()->Write();
}
// Write Mask
if (fIsMask && (drawOpt.find("MASK") != std::string::npos)) {
fMaskHist->Write();
}
// Write Covariances
if (drawOpt.find("COV") != std::string::npos && fFullCovar) {
PlotUtils::GetFullCovarPlot(fFullCovar, fSettings.GetName());
}
if (drawOpt.find("INVCOV") != std::string::npos && covar) {
PlotUtils::GetInvCovarPlot(covar, fSettings.GetName());
}
if (drawOpt.find("DECOMP") != std::string::npos && fDecomp) {
PlotUtils::GetDecompCovarPlot(fDecomp, fSettings.GetName());
}
// // Likelihood residual plots
// if (drawOpt.find("RESIDUAL") != std::string::npos) {
// WriteResidualPlots();
// }
// Ratio and Shape Plots
if (drawOpt.find("RATIO") != std::string::npos) {
WriteRatioPlot();
}
if (drawOpt.find("SHAPE") != std::string::npos) {
WriteShapePlot();
if (drawOpt.find("RATIO") != std::string::npos)
WriteShapeRatioPlot();
}
// // RATIO
// if (drawOpt.find("CANVMC") != std::string::npos) {
// TCanvas* c1 = WriteMCCanvas(fDataHist, fMCHist);
// c1->Write();
// delete c1;
// }
// // PDG
// if (drawOpt.find("CANVPDG") != std::string::npos && fMCHist_Modes) {
// TCanvas* c2 = WritePDGCanvas(fDataHist, fMCHist, fMCHist_Modes);
// c2->Write();
// delete c2;
// }
// Write Extra Histograms
AutoWriteExtraTH1();
WriteExtraHistograms();
// Returning
LOG(SAM) << "Written Histograms: " << fName << std::endl;
return;
}
//********************************************************************
void Measurement1D::WriteRatioPlot() {
//********************************************************************
// Setup mc data ratios
TH1D* dataRatio = (TH1D*)fDataHist->Clone((fName + "_data_RATIO").c_str());
TH1D* mcRatio = (TH1D*)fMCHist->Clone((fName + "_MC_RATIO").c_str());
// Extra MC Data Ratios
for (int i = 0; i < mcRatio->GetNbinsX(); i++) {
dataRatio->SetBinContent(i + 1, fDataHist->GetBinContent(i + 1) / fMCHist->GetBinContent(i + 1));
dataRatio->SetBinError(i + 1, fDataHist->GetBinError(i + 1) / fMCHist->GetBinContent(i + 1));
mcRatio->SetBinContent(i + 1, fMCHist->GetBinContent(i + 1) / fMCHist->GetBinContent(i + 1));
mcRatio->SetBinError(i + 1, fMCHist->GetBinError(i + 1) / fMCHist->GetBinContent(i + 1));
}
// Write ratios
mcRatio->Write();
dataRatio->Write();
delete mcRatio;
delete dataRatio;
}
//********************************************************************
void Measurement1D::WriteShapePlot() {
//********************************************************************
TH1D* mcShape = (TH1D*)fMCHist->Clone((fName + "_MC_SHAPE").c_str());
TH1D* dataShape = (TH1D*)fDataHist->Clone((fName + "_data_SHAPE").c_str());
// Don't check error
if (fShapeCovar) StatUtils::SetDataErrorFromCov(dataShape, fShapeCovar, 1E-38, false);
double shapeScale = 1.0;
if (fIsRawEvents) {
shapeScale = fDataHist->Integral() / fMCHist->Integral();
} else {
shapeScale = fDataHist->Integral("width") / fMCHist->Integral("width");
}
mcShape->Scale(shapeScale);
std::stringstream ss;
ss << shapeScale;
mcShape->SetTitle(ss.str().c_str());
mcShape->SetLineWidth(3);
mcShape->SetLineStyle(7);
mcShape->Write();
dataShape->Write();
delete mcShape;
}
//********************************************************************
void Measurement1D::WriteShapeRatioPlot() {
//********************************************************************
// Get a mcshape histogram
TH1D* mcShape = (TH1D*)fMCHist->Clone((fName + "_MC_SHAPE").c_str());
double shapeScale = 1.0;
if (fIsRawEvents) {
shapeScale = fDataHist->Integral() / fMCHist->Integral();
} else {
shapeScale = fDataHist->Integral("width") / fMCHist->Integral("width");
}
mcShape->Scale(shapeScale);
// Create shape ratio histograms
TH1D* mcShapeRatio = (TH1D*)mcShape->Clone((fName + "_MC_SHAPE_RATIO").c_str());
TH1D* dataShapeRatio = (TH1D*)fDataHist->Clone((fName + "_data_SHAPE_RATIO").c_str());
// Divide the histograms
mcShapeRatio->Divide(mcShape);
dataShapeRatio->Divide(mcShape);
// Colour the shape ratio plots
mcShapeRatio->SetLineWidth(3);
mcShapeRatio->SetLineStyle(7);
mcShapeRatio->Write();
dataShapeRatio->Write();
delete mcShapeRatio;
delete dataShapeRatio;
}
//// CRAP TO BE REMOVED
//********************************************************************
void Measurement1D::SetupMeasurement(std::string inputfile, std::string type,
FitWeight * rw, std::string fkdt) {
//********************************************************************
nuiskey samplekey = Config::CreateKey("sample");
samplekey.Set("name", fName);
samplekey.Set("type",type);
samplekey.Set("input",inputfile);
fSettings = LoadSampleSettings(samplekey);
// Reset everything to NULL
// Init();
// Check if name contains Evt, indicating that it is a raw number of events
// measurements and should thus be treated as once
fIsRawEvents = false;
if ((fName.find("Evt") != std::string::npos) && fIsRawEvents == false) {
fIsRawEvents = true;
LOG(SAM) << "Found event rate measurement but fIsRawEvents == false!"
<< std::endl;
LOG(SAM) << "Overriding this and setting fIsRawEvents == true!"
<< std::endl;
}
fIsEnu1D = false;
if (fName.find("XSec_1DEnu") != std::string::npos) {
fIsEnu1D = true;
LOG(SAM) << "::" << fName << "::" << std::endl;
LOG(SAM) << "Found XSec Enu measurement, applying flux integrated scaling, "
"not flux averaged!"
<< std::endl;
}
if (fIsEnu1D && fIsRawEvents) {
LOG(SAM) << "Found 1D Enu XSec distribution AND fIsRawEvents, is this "
"really correct?!"
<< std::endl;
LOG(SAM) << "Check experiment constructor for " << fName
<< " and correct this!" << std::endl;
LOG(SAM) << "I live in " << __FILE__ << ":" << __LINE__ << std::endl;
throw;
}
fRW = rw;
if (!fInput and !fIsJoint) SetupInputs(inputfile);
// Set Default Options
SetFitOptions(fDefaultTypes);
// Set Passed Options
SetFitOptions(type);
// Still adding support for flat flux inputs
// // Set Enu Flux Scaling
// if (isFlatFluxFolding) this->Input()->ApplyFluxFolding(
// this->defaultFluxHist );
// FinaliseMeasurement();
}
//********************************************************************
void Measurement1D::SetupDefaultHist() {
//********************************************************************
// Setup fMCHist
fMCHist = (TH1D*)fDataHist->Clone();
fMCHist->SetNameTitle((fName + "_MC").c_str(),
(fName + "_MC" + fPlotTitles).c_str());
// Setup fMCFine
Int_t nBins = fMCHist->GetNbinsX();
fMCFine = new TH1D(
(fName + "_MC_FINE").c_str(), (fName + "_MC_FINE" + fPlotTitles).c_str(),
nBins * 6, fMCHist->GetBinLowEdge(1), fMCHist->GetBinLowEdge(nBins + 1));
fMCStat = (TH1D*)fMCHist->Clone();
fMCStat->Reset();
fMCHist->Reset();
fMCFine->Reset();
// Setup the NEUT Mode Array
PlotUtils::CreateNeutModeArray((TH1D*)fMCHist, (TH1**)fMCHist_PDG);
PlotUtils::ResetNeutModeArray((TH1**)fMCHist_PDG);
// Setup bin masks using sample name
if (fIsMask) {
std::string maskloc = FitPar::Config().GetParDIR(fName + ".mask");
if (maskloc.empty()) {
maskloc = FitPar::GetDataBase() + "/masks/" + fName + ".mask";
}
SetBinMask(maskloc);
}
fMCHist_Modes = new TrueModeStack( (fName + "_MODES").c_str(), ("True Channels"), fMCHist);
SetAutoProcessTH1(fMCHist_Modes, kCMD_Reset, kCMD_Norm, kCMD_Write);
return;
}
//********************************************************************
void Measurement1D::SetDataValues(std::string dataFile) {
//********************************************************************
// Override this function if the input file isn't in a suitable format
LOG(SAM) << "Reading data from: " << dataFile.c_str() << std::endl;
fDataHist =
PlotUtils::GetTH1DFromFile(dataFile, (fName + "_data"), fPlotTitles);
fDataTrue = (TH1D*)fDataHist->Clone();
// Number of data points is number of bins
fNDataPointsX = fDataHist->GetXaxis()->GetNbins();
return;
};
-
-
-
//********************************************************************
void Measurement1D::SetDataFromDatabase(std::string inhistfile,
std::string histname) {
//********************************************************************
LOG(SAM) << "Filling histogram from " << inhistfile << "->" << histname
<< std::endl;
fDataHist = PlotUtils::GetTH1DFromRootFile(
(GeneralUtils::GetTopLevelDir() + "/data/" + inhistfile), histname);
fDataHist->SetNameTitle((fName + "_data").c_str(), (fName + "_data").c_str());
return;
};
//********************************************************************
void Measurement1D::SetDataFromFile(std::string inhistfile,
std::string histname) {
//********************************************************************
LOG(SAM) << "Filling histogram from " << inhistfile << "->" << histname
<< std::endl;
fDataHist = PlotUtils::GetTH1DFromRootFile((inhistfile), histname);
fDataHist->SetNameTitle((fName + "_data").c_str(), (fName + "_data").c_str());
return;
};
//********************************************************************
void Measurement1D::SetCovarMatrix(std::string covarFile) {
//********************************************************************
// Covariance function, only really used when reading in the MB Covariances.
TFile* tempFile = new TFile(covarFile.c_str(), "READ");
TH2D* covarPlot = new TH2D();
TH2D* fFullCovarPlot = new TH2D();
std::string covName = "";
std::string covOption = FitPar::Config().GetParS("thrown_covariance");
if (fIsShape || fIsFree) covName = "shp_";
if (fIsDiag)
covName += "diag";
else
covName += "full";
covarPlot = (TH2D*)tempFile->Get((covName + "cov").c_str());
if (!covOption.compare("SUB"))
fFullCovarPlot = (TH2D*)tempFile->Get((covName + "cov").c_str());
else if (!covOption.compare("FULL"))
fFullCovarPlot = (TH2D*)tempFile->Get("fullcov");
else
ERR(WRN) << "Incorrect thrown_covariance option in parameters."
<< std::endl;
int dim = int(fDataHist->GetNbinsX()); //-this->masked->Integral());
int covdim = int(fDataHist->GetNbinsX());
this->covar = new TMatrixDSym(dim);
fFullCovar = new TMatrixDSym(dim);
fDecomp = new TMatrixDSym(dim);
int row, column = 0;
row = 0;
column = 0;
for (Int_t i = 0; i < covdim; i++) {
// if (this->masked->GetBinContent(i+1) > 0) continue;
for (Int_t j = 0; j < covdim; j++) {
// if (this->masked->GetBinContent(j+1) > 0) continue;
(*this->covar)(row, column) = covarPlot->GetBinContent(i + 1, j + 1);
(*fFullCovar)(row, column) = fFullCovarPlot->GetBinContent(i + 1, j + 1);
column++;
}
column = 0;
row++;
}
// Set bin errors on data
if (!fIsDiag) {
StatUtils::SetDataErrorFromCov(fDataHist, fFullCovar);
}
// Get Deteriminant and inverse matrix
// fCovDet = this->covar->Determinant();
TDecompSVD LU = TDecompSVD(*this->covar);
this->covar = new TMatrixDSym(dim, LU.Invert().GetMatrixArray(), "");
return;
};
//********************************************************************
// Sets the covariance matrix from a provided file in a text format
// scale is a multiplicative pre-factor to apply in the case where the
// covariance is given in some unit (e.g. 1E-38)
void Measurement1D::SetCovarMatrixFromText(std::string covarFile, int dim,
double scale) {
//********************************************************************
// Make a counter to track the line number
int row = 0;
std::string line;
std::ifstream covarread(covarFile.c_str(), std::ifstream::in);
this->covar = new TMatrixDSym(dim);
fFullCovar = new TMatrixDSym(dim);
if (covarread.is_open())
LOG(SAM) << "Reading covariance matrix from file: " << covarFile
<< std::endl;
else
ERR(FTL) << "Covariance matrix provided is incorrect: " << covarFile
<< std::endl;
// Loop over the lines in the file
while (std::getline(covarread >> std::ws, line, '\n')) {
int column = 0;
// Loop over entries and insert them into matrix
std::vector<double> entries = GeneralUtils::ParseToDbl(line, " ");
if (entries.size() <= 1) {
ERR(WRN) << "SetCovarMatrixFromText -> Covariance matrix only has <= 1 "
"entries on this line: "
<< row << std::endl;
}
for (std::vector<double>::iterator iter = entries.begin();
iter != entries.end(); iter++) {
(*covar)(row, column) = *iter;
(*fFullCovar)(row, column) = *iter;
column++;
}
row++;
}
covarread.close();
// Scale the actualy covariance matrix by some multiplicative factor
(*fFullCovar) *= scale;
// Robust matrix inversion method
TDecompSVD LU = TDecompSVD(*this->covar);
// THIS IS ACTUALLY THE INVERSE COVARIANCE MATRIXA AAAAARGH
delete this->covar;
this->covar = new TMatrixDSym(dim, LU.Invert().GetMatrixArray(), "");
// Now need to multiply by the scaling factor
// If the covariance
(*this->covar) *= 1. / (scale);
return;
};
//********************************************************************
void Measurement1D::SetCovarMatrixFromCorrText(std::string corrFile, int dim) {
//********************************************************************
// Make a counter to track the line number
int row = 0;
std::string line;
std::ifstream corr(corrFile.c_str(), std::ifstream::in);
this->covar = new TMatrixDSym(dim);
this->fFullCovar = new TMatrixDSym(dim);
if (corr.is_open())
LOG(SAM) << "Reading and converting correlation matrix from file: "
<< corrFile << std::endl;
else {
ERR(FTL) << "Correlation matrix provided is incorrect: " << corrFile
<< std::endl;
exit(-1);
}
while (std::getline(corr >> std::ws, line, '\n')) {
int column = 0;
// Loop over entries and insert them into matrix
// Multiply by the errors to get the covariance, rather than the correlation
// matrix
std::vector<double> entries = GeneralUtils::ParseToDbl(line, " ");
for (std::vector<double>::iterator iter = entries.begin();
iter != entries.end(); iter++) {
double val = (*iter) * this->fDataHist->GetBinError(row + 1) * 1E38 *
this->fDataHist->GetBinError(column + 1) * 1E38;
if (val == 0) {
ERR(FTL) << "Found a zero value in the covariance matrix, assuming "
"this is an error!"
<< std::endl;
exit(-1);
}
(*this->covar)(row, column) = val;
(*this->fFullCovar)(row, column) = val;
column++;
}
row++;
}
// Robust matrix inversion method
TDecompSVD LU = TDecompSVD(*this->covar);
delete this->covar;
this->covar = new TMatrixDSym(dim, LU.Invert().GetMatrixArray(), "");
return;
};
//********************************************************************
// FullUnits refers to if we have "real" unscaled units in the covariance matrix, e.g. 1E-76.
// If this is the case we need to scale it so that the chi2 contribution is correct
// NUISANCE internally assumes the covariance matrix has units of 1E76
void Measurement1D::SetCovarFromDataFile(std::string covarFile,
std::string covName, bool FullUnits) {
//********************************************************************
LOG(SAM) << "Getting covariance from " << covarFile << "->" << covName
<< std::endl;
TFile* tempFile = new TFile(covarFile.c_str(), "READ");
TH2D* covPlot = (TH2D*)tempFile->Get(covName.c_str());
covPlot->SetDirectory(0);
// Scale the covariance matrix if it comes in normal units
if (FullUnits) {
covPlot->Scale(1.E76);
}
int dim = covPlot->GetNbinsX();
fFullCovar = new TMatrixDSym(dim);
for (int i = 0; i < dim; i++) {
for (int j = 0; j < dim; j++) {
(*fFullCovar)(i, j) = covPlot->GetBinContent(i + 1, j + 1);
}
}
this->covar = (TMatrixDSym*)fFullCovar->Clone();
fDecomp = (TMatrixDSym*)fFullCovar->Clone();
TDecompSVD LU = TDecompSVD(*this->covar);
this->covar = new TMatrixDSym(dim, LU.Invert().GetMatrixArray(), "");
TDecompChol LUChol = TDecompChol(*fDecomp);
LUChol.Decompose();
fDecomp = new TMatrixDSym(dim, LU.GetU().GetMatrixArray(), "");
return;
};
// //********************************************************************
// void Measurement1D::SetBinMask(std::string maskFile) {
// //********************************************************************
// // Create a mask histogram.
// int nbins = fDataHist->GetNbinsX();
// fMaskHist =
// new TH1I((fName + "_fMaskHist").c_str(),
// (fName + "_fMaskHist; Bin; Mask?").c_str(), nbins, 0, nbins);
// std::string line;
// std::ifstream mask(maskFile.c_str(), std::ifstream::in);
// if (mask.is_open())
// LOG(SAM) << "Reading bin mask from file: " << maskFile << std::endl;
// else
// LOG(FTL) << " Cannot find mask file." << std::endl;
// while (std::getline(mask >> std::ws, line, '\n')) {
// std::vector<int> entries = GeneralUtils::ParseToInt(line, " ");
// // Skip lines with poorly formatted lines
// if (entries.size() < 2) {
// LOG(WRN) << "Measurement1D::SetBinMask(), couldn't parse line: " << line
// << std::endl;
// continue;
// }
// // The first index should be the bin number, the second should be the mask
// // value.
// fMaskHist->SetBinContent(entries[0], entries[1]);
// }
// // Set masked data bins to zero
// PlotUtils::MaskBins(fDataHist, fMaskHist);
// return;
// }
// //********************************************************************
// void Measurement1D::GetBinContents(std::vector<double>& cont,
// std::vector<double>& err) {
// //********************************************************************
// // Return a vector of the main bin contents
// for (int i = 0; i < fMCHist->GetNbinsX(); i++) {
// cont.push_back(fMCHist->GetBinContent(i + 1));
// err.push_back(fMCHist->GetBinError(i + 1));
// }
// return;
// };
/*
XSec Functions
*/
// //********************************************************************
// void Measurement1D::SetFluxHistogram(std::string fluxFile, int minE, int
// maxE,
// double fluxNorm) {
// //********************************************************************
// // Note this expects the flux bins to be given in terms of MeV
// LOG(SAM) << "Reading flux from file: " << fluxFile << std::endl;
// TGraph f(fluxFile.c_str(), "%lg %lg");
// fFluxHist =
// new TH1D((fName + "_flux").c_str(), (fName + "; E_{#nu} (GeV)").c_str(),
// f.GetN() - 1, minE, maxE);
// Double_t* yVal = f.GetY();
// for (int i = 0; i < fFluxHist->GetNbinsX(); ++i)
// fFluxHist->SetBinContent(i + 1, yVal[i] * fluxNorm);
// };
// //********************************************************************
// double Measurement1D::TotalIntegratedFlux(std::string intOpt, double low,
// double high) {
// //********************************************************************
// if (fInput->GetType() == kGiBUU) {
// return 1.0;
// }
// // The default case of low = -9999.9 and high = -9999.9
// if (low == -9999.9) low = this->EnuMin;
// if (high == -9999.9) high = this->EnuMax;
// int minBin = fFluxHist->GetXaxis()->FindBin(low);
// int maxBin = fFluxHist->GetXaxis()->FindBin(high);
// // Get integral over custom range
// double integral = fFluxHist->Integral(minBin, maxBin + 1, intOpt.c_str());
// return integral;
// };
File Metadata
Details
Attached
Mime Type
text/x-diff
Expires
Sat, May 3, 6:12 AM (1 d, 7 h)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
4982950
Default Alt Text
(59 KB)
Attached To
rNUISANCEGIT nuisancegit
Event Timeline
Log In to Comment