PredictiveThrower Class Reference

Implementation of Prior/Posterior Predictive and Bayesian p-Value calculations following the approach described in [11] , [12], [14]. More...

#include <Fitters/PredictiveThrower.h>

Inheritance diagram for PredictiveThrower:

Collaboration diagram for PredictiveThrower:

Public Member Functions
	PredictiveThrower (Manager *const fitMan)
	Constructor. More...
virtual	~PredictiveThrower ()
	Destructor. More...
void	ProduceToys ()
	Produce toys by throwing from MCMC. More...
void	RunPredictiveAnalysis ()
	Main routine responsible for producing posterior predictive distributions and $p$-value. More...
void	RunMCMC () override
	This is not used in this class. More...
Public Member Functions inherited from FitterBase
	FitterBase (Manager *const fitMan)
	Constructor. More...
virtual	~FitterBase ()
	Destructor for the FitterBase class. More...
void	AddSampleHandler (SampleHandlerInterface *sample)
	This function adds a sample PDF object to the analysis framework. The sample PDF object will be utilized in fitting procedures or likelihood scans. More...
void	AddSystObj (ParameterHandlerBase *cov)
	This function adds a Covariance object to the analysis framework. The Covariance object will be utilized in fitting procedures or likelihood scans. More...
void	DragRace (const int NLaps=100)
	Calculates the required time for each sample or covariance object in a drag race simulation. Inspired by Dan's feature. More...
void	RunLLHScan ()
	Perform a 1D likelihood scan. More...
void	RunLLHMap ()
	Perform a general multi-dimensional likelihood scan. More...
void	GetStepScaleBasedOnLLHScan (const std::string &filename="")
	LLH scan is good first estimate of step scale. More...
void	Run2DLLHScan ()
	Perform a 2D likelihood scan. More...
void	RunSigmaVar ()
	Perform a 1D/2D sigma var for all samples. More...
virtual void	StartFromPreviousFit (const std::string &FitName)
	Allow to start from previous fit/chain. More...
std::string	GetName () const
	Get name of class. More...

Private Member Functions
void	SetParamters (std::vector< std::string > &ParameterGroupsNotVaried, std::unordered_set< int > &ParameterOnlyToVary)
	This set some params to prior value this way you can evaluate errors from subset of errors. More...
void	SetupToyGeneration (std::vector< std::string > &ParameterGroupsNotVaried, std::unordered_set< int > &ParameterOnlyToVary, std::vector< const M3::float_t * > &BoundValuePointer, std::vector< std::pair< double, double >> &ParamBounds)
	Setup useful variables etc before stating toy generation. More...
bool	LoadToys ()
	Load existing toys. More...
void	WriteToy (TDirectory *ToyDirectory, TDirectory *Toy_1DDirectory, TDirectory *Toy_2DDirectory, const int iToy)
	Save histograms for a single MCMC Throw/Toy. More...
void	SetupSampleInformation ()
	Setup sample information. More...
std::vector< std::string >	GetStoredFancyName (ParameterHandlerBase *Systematics) const
	Get Fancy parameters stored in mcmc chains for passed ParameterHandler. More...
std::vector< std::unique_ptr< TH1 > >	MakePredictive (const std::vector< std::vector< std::unique_ptr< TH1 >>> &Toys, const std::vector< TDirectory * > &Director, const std::string &suffix, const bool DebugHistograms, const bool WriteHist)
	Produce posterior predictive distribution. More...
void	Study1DProjections (const std::vector< TDirectory * > &SampleDirectories) const
	Load 1D projections and later produce violin plots for each. More...
void	ProduceSpectra (const std::vector< std::vector< std::vector< std::unique_ptr< TH1D >>>> &Toys, const std::vector< TDirectory * > &Director, const std::string suffix) const
	Produce Violin style spectra. More...
void	MakeFluctuatedHistogram (TH1 *FluctHist, TH1 *PolyHist)
	Make Poisson fluctuation of TH1D hist. More...
void	PredictiveLLH (const std::vector< std::unique_ptr< TH1 >> &Data_histogram, const std::vector< std::unique_ptr< TH1 >> &PostPred_mc, const std::vector< std::unique_ptr< TH1 >> &PostPred_w, const std::vector< TDirectory * > &SampleDir)
	Calculate Posterior Predictive LLH. More...
void	PosteriorPredictivepValue (const std::vector< std::unique_ptr< TH1 >> &PostPred_mc, const std::vector< TDirectory * > &SampleDir)
	Calculate Posterior Predictive $p$-value Compares observed data to toy datasets generated from: More...
void	ExtractLLH (TH1 *DatHist, TH1 *MCHist, TH1 *W2Hist, const SampleHandlerInterface *SampleHandler) const
	Calculate the LLH for TH1, set the LLH to title of MCHist. More...
double	CalcLLH (const double data, const double mc, const double w2, const SampleHandlerInterface *SampleHandler) const
	Calculates the -2LLH (likelihood) for a single sample. More...
double	CalcLLH (const TH1 *DatHist, const TH1 *MCHist, const TH1 *W2Hist, const SampleHandlerInterface *SampleHandler) const
	Calculates the likelihood (-2LLH) for a single sample; dynamically casts to call the correct GetLLH overload. More...
double	GetLLH (const TH1D *DatHist, const TH1D *MCHist, const TH1D *W2Hist, const SampleHandlerInterface *SampleHandler) const
	Helper functions to calculate likelihoods using TH1D. More...
double	GetLLH (const TH2D *DatHist, const TH2D *MCHist, const TH2D *W2Hist, const SampleHandlerInterface *SampleHandler) const
	Helper functions to calculate likelihoods using TH2D. More...
double	GetLLH (const TH2Poly *DatHist, const TH2Poly *MCHist, const TH2Poly *W2Hist, const SampleHandlerInterface *SampleHandler) const
	Helper functions to calculate likelihoods using TH2Poly. More...
void	MakeChi2Plots (const std::vector< std::vector< double >> &Chi2_x, const std::string &Chi2_x_title, const std::vector< std::vector< double >> &Chi2_y, const std::string &Chi2_y_title, const std::vector< TDirectory * > &SampleDir, const std::string Title)
	Produce Chi2 plot for a single sample based on which $p$-value is calculated. More...
void	StudyInformationCriterion (M3::kInfCrit Criterion, const std::vector< std::unique_ptr< TH1 >> &PostPred_mc, const std::vector< std::unique_ptr< TH1 >> &PostPred_w)
	Information Criterion. More...
void	StudyBIC (const std::vector< std::unique_ptr< TH1 >> &PostPred_mc, const std::vector< std::unique_ptr< TH1 >> &PostPred_w)
	Study Bayesian Information Criterion (BIC) The BIC is defined as: More...
void	StudyDIC (const std::vector< std::unique_ptr< TH1 >> &PostPred_mc, const std::vector< std::unique_ptr< TH1 >> &PostPred_w)
	KS: Get the Deviance Information Criterion (DIC) The deviance is defined as: More...
void	StudyWAIC ()
	KS: Get the Watanabe-Akaike information criterion (WAIC) More...
std::string	GetBinName (TH1 *hist, const bool uniform, const int Dim, const std::vector< int > &bins) const
	Construct a human-readable label describing a specific analysis bin. More...
std::vector< std::unique_ptr< TH1D > >	PerBinHistogram (TH1 *hist, const int SampleId, const int Dim, const std::string &suffix) const
	Create per-bin posterior histograms for a given sample. More...
void	StudyBetaParameters (TDirectory *PredictiveDir)
	Evaluate prior/post predictive distribution for beta parameters (used for evaluating impact MC statistical uncertainty) More...
void	MakeCutEventRate (TH1D *Histogram, const double DataRate) const
	Make the 1D Event Rate Hist. More...
void	RateAnalysis (const std::vector< std::vector< std::unique_ptr< TH1 >>> &Toys, const std::vector< TDirectory * > &SampleDirectories) const
	Produce distribution of number of events for each sample. More...

Private Attributes
bool	FullLLH
	KS: Use Full LLH or only sample contribution based on discussion with Asher we almost always only want the sample likelihood. More...
int	NModelParams
	KS: Count total number of model parameters which can be used for stuff like BIC. More...
bool	Is_PriorPredictive
	Whether it is Prior or Posterior predictive. More...
int	TotalNumberOfSamples
	Number of toys we are generating analysing. More...
std::vector< PredictiveSample >	SampleInfo
	Handy struct for all sample info. More...
int	Ntoys
	Number of toys we are generating analysing. More...
ParameterHandlerGeneric *	ModelSystematic
	Pointer to El Generico. More...
std::vector< std::unique_ptr< TH1 > >	Data_Hist
	Vector of Data histograms. More...
std::vector< std::unique_ptr< TH1 > >	MC_Nom_Hist
	Vector of MC histograms. More...
std::vector< std::unique_ptr< TH1 > >	W2_Nom_Hist
	Vector of W2 histograms. More...
std::vector< std::vector< std::unique_ptr< TH1 > > >	MC_Hist_Toy
std::vector< std::vector< std::unique_ptr< TH1 > > >	W2_Hist_Toy
std::vector< double >	ReweightWeight
	Reweighting factors applied for each toy, by default 1. More...
std::vector< double >	PenaltyTerm
	Penalty term values for each toy by default 0. More...
bool	StandardFluctuation
	KS: We have two methods for Poissonian fluctuation. More...

Additional Inherited Members
Protected Member Functions inherited from FitterBase
void	ProcessMCMC ()
	Process MCMC output. More...
void	PrepareOutput ()
	Prepare the output file. More...
void	SaveOutput ()
	Save output and close files. More...
void	SanitiseInputs ()
	Remove obsolete memory and make other checks before fit starts. More...
void	SaveSettings ()
	Save the settings that the MCMC was run with. More...
bool	GetScanRange (std::map< std::string, std::vector< double >> &scanRanges) const
	YSP: Set up a mapping to store parameters with user-specified ranges, suggested by D. Barrow. More...
void	GetParameterScanRange (const ParameterHandlerBase *cov, const int i, double &CentralValue, double &lower, double &upper, const int n_points, const std::string &suffix="") const
	Helper function to get parameter scan range, central value. More...
bool	CheckSkipParameter (const std::vector< std::string > &SkipVector, const std::string &ParamName) const
	KS: Check whether we want to skip parameter using skip vector. More...
void	CustomRange (const std::string &ParName, const double sigma, double &ParamShiftValue) const
	For comparison with other fitting frameworks (like P-Theta) we usually have to apply different parameter values then usual 1, 3 sigma. More...
Protected Attributes inherited from FitterBase
Manager *	fitMan
	The manager for configuration handling. More...
unsigned int	step
	current state More...
double	logLCurr
	current likelihood More...
double	logLProp
	proposed likelihood More...
double	accProb
	current acceptance prob More...
int	accCount
	counts accepted steps More...
unsigned int	stepStart
	step start, by default 0 if we start from previous chain then it will be different More...
std::vector< double >	sample_llh
	store the llh breakdowns More...
std::vector< double >	syst_llh
	systematic llh breakdowns More...
std::vector< SampleHandlerInterface * >	samples
	Sample holder. More...
unsigned int	TotalNSamples
	Total number of samples used, single SampleHandler can store more than one analysis sample! More...
std::vector< ParameterHandlerBase * >	systematics
	Systematic holder. More...
std::unique_ptr< TStopwatch >	clock
	tells global time how long fit took More...
std::unique_ptr< TStopwatch >	stepClock
	tells how long single step/fit iteration took More...
double	stepTime
	Time of single step. More...
std::unique_ptr< TRandom3 >	random
	Random number. More...
TFile *	outputFile
	Output. More...
TDirectory *	CovFolder
	Output cov folder. More...
TDirectory *	SampleFolder
	Output sample folder. More...
TTree *	outTree
	Output tree with posteriors. More...
int	auto_save
	auto save every N steps More...
bool	fTestLikelihood
	Necessary for some fitting algorithms like PSO. More...
bool	FileSaved
	Checks if file saved not repeat some operations. More...
bool	SettingsSaved
	Checks if setting saved not repeat some operations. More...
bool	OutputPrepared
	Checks if output prepared not repeat some operations. More...
std::string	AlgorithmName
	Name of fitting algorithm that is being used. More...

Detailed Description

Implementation of Prior/Posterior Predictive and Bayesian p-Value calculations following the approach described in [11] , [12], [14].

For more information, visit the Posterior Predictive page.

Author

Asher Kaboth

Dan Barrow

Ed Atkin

Yashwanth S Prabhu

Kamil Skwarczynski

Patrick Dunne

Clarence Wret

Todo:

add BIC, DIC, WAIC

add Rate $p$-value

add plots by mode

Post Pred LLH

unify code with SampleSummary

Definition at line 38 of file PredictiveThrower.h.

Constructor & Destructor Documentation

PredictiveThrower()

PredictiveThrower::PredictiveThrower

(

Manager *const

fitMan

)

Constructor.

Parameters

fitMan

A pointer to a manager object, which will handle all settings.

Definition at line 10 of file PredictiveThrower.cpp.

                                                 : FitterBase(man) {
// *************************
  AlgorithmName = "PredictiveThrower";
  if(!CheckNodeExists(fitMan->raw(), "Predictive")) {
   MACH3LOG_ERROR("Predictive is missing in your main yaml config");
   throw MaCh3Exception(__FILE__ , __LINE__ );
  }
 
  StandardFluctuation = GetFromManager<bool>(fitMan->raw()["Predictive"]["StandardFluctuation"], true, __FILE__, __LINE__ );
 
  if(StandardFluctuation) MACH3LOG_INFO("Using standard method of statistical fluctuation");
  else MACH3LOG_INFO("Using alternative method of statistical fluctuation, which is much slower");
 
  ModelSystematic = nullptr;
  // Use the full likelihood for the Prior/Posterior predictive pvalue
  FullLLH = GetFromManager<bool>(fitMan->raw()["Predictive"]["FullLLH"], false, __FILE__, __LINE__ );
  NModelParams = 0;
 
  Is_PriorPredictive = Get<bool>(fitMan->raw()["Predictive"]["PriorPredictive"], __FILE__, __LINE__);
  Ntoys = Get<int>(fitMan->raw()["Predictive"]["Ntoy"], __FILE__, __LINE__);
 
  ReweightWeight.resize(Ntoys);
  PenaltyTerm.resize(Ntoys);
}

~PredictiveThrower()

PredictiveThrower::~PredictiveThrower ( )

virtual

Destructor.

Definition at line 37 of file PredictiveThrower.cpp.

                                      {
// *************************
 
}

Member Function Documentation

CalcLLH() [1/2]

double PredictiveThrower::CalcLLH ( const double  data, const double  mc, const double  w2, const SampleHandlerInterface *  SampleHandler  ) const

private

Calculates the -2LLH (likelihood) for a single sample.

Parameters

data	Data value for the sample.
mc	MC (Monte Carlo) value for the sample.
w2	W2 value for the sample.
SampleHandler	Pointer to SampleHandlerInterface providing the LLH test statistic.

Definition at line 1071 of file PredictiveThrower.cpp.

                                                                                     {
// *************************
  double llh = SampleHandler->GetTestStatLLH(data, mc, w2);
  //KS: do times 2 because banff reports chi2
  return 2*llh;
}

CalcLLH() [2/2]

double PredictiveThrower::CalcLLH ( const TH1 *  DatHist, const TH1 *  MCHist, const TH1 *  W2Hist, const SampleHandlerInterface *  SampleHandler  ) const

private

Calculates the likelihood (-2LLH) for a single sample; dynamically casts to call the correct GetLLH overload.

Parameters

DatHist	Data histogram with data distribution for a single sample
MCHist	MC histogram with MC distribution for a single sample
W2Hist	W2 histogram with W2 distribution for a single sample
SampleHandler	Pointer to SampleHandlerInterface providing LLH test statistic

Definition at line 1082 of file PredictiveThrower.cpp.

                                                                                     {
// *************************
  // 1D case
  if (auto h1 = dynamic_cast<const TH1D*>(DatHist)) {
    return GetLLH(h1,
                  static_cast<const TH1D*>(MCHist),
                  static_cast<const TH1D*>(W2Hist),
                  SampleHandler);
  }
 
  // 2D case
  if (auto h2 = dynamic_cast<const TH2D*>(DatHist)) {
    return GetLLH(h2,
                  static_cast<const TH2D*>(MCHist),
                  static_cast<const TH2D*>(W2Hist),
                  SampleHandler);
  }
 
  // 2D poly case
  if (auto h2p = dynamic_cast<const TH2Poly*>(DatHist)) {
    return GetLLH(h2p,
                  static_cast<const TH2Poly*>(MCHist),
                  static_cast<const TH2Poly*>(W2Hist),
                  SampleHandler);
  }
 
  MACH3LOG_ERROR("Unsupported histogram type in {}", __func__);
  throw MaCh3Exception(__FILE__ , __LINE__ );
}

ExtractLLH()

void PredictiveThrower::ExtractLLH ( TH1 *  DatHist, TH1 *  MCHist, TH1 *  W2Hist, const SampleHandlerInterface *  SampleHandler  ) const

private

Calculate the LLH for TH1, set the LLH to title of MCHist.

Parameters

DatHist	Data histogram with data distribution for a single sample
MCHist	MC histogram with MC distribution for a single sample
W2Hist	W2 histogram with W2 distribution for a single sample
SampleHandler	Pointer to SampleHandlerInterface providing LLH test statistic

Definition at line 1431 of file PredictiveThrower.cpp.

                                                                                                                             {
// ****************
  const double llh = CalcLLH(DatHist, MCHist, W2Hist, SampleHandler);
  std::stringstream ss;
  ss << "_2LLH=" << llh;
  MCHist->SetTitle((std::string(MCHist->GetTitle())+ss.str()).c_str());
  MACH3LOG_INFO("{:<55} {:<10.2f} {:<10.2f} {:<10.2f}", MCHist->GetName(), DatHist->Integral(), MCHist->Integral(), llh);
}

GetBinName()

std::string PredictiveThrower::GetBinName ( TH1 *  hist, const bool  uniform, const int  Dim, const std::vector< int > &  bins  ) const

private

Construct a human-readable label describing a specific analysis bin.

Parameters

hist	Histogram providing the binning definition.
uniform	Flag indicating whether the histogram uses regular axis binning (TH1/TH2) or irregular polygonal binning (e.g. TH2Poly).
Dim	Dimensionality of the original distribution.
bins	Vector of per-dimension bin indices in analysis coordinates.

Definition at line 825 of file PredictiveThrower.cpp.

                                                                            {
// *************************
  std::string BinName = "";
  if(Dim == 1) { // True 1D distribution using TH1D
    const int b = bins[0];
    const TAxis* ax = hist->GetXaxis();
    const double low = ax->GetBinLowEdge(b);
    const double up  = ax->GetBinUpEdge(b);
 
    BinName = fmt::format("Dim0 ({:g}, {:g})", low, up);
  } else if (Dim == 2) { // True 2D dsitrubitons
    if(uniform == true) { //using TH2D
      const int bx = bins[0];
      const int by = bins[1];
      const TAxis* ax = hist->GetXaxis();
      const TAxis* ay = hist->GetYaxis();
      BinName = fmt::format("Dim0 ({:g}, {:g}), ", ax->GetBinLowEdge(bx), ax->GetBinUpEdge(bx));
      BinName += fmt::format("Dim1 ({:g}, {:g})", ay->GetBinLowEdge(by), ay->GetBinUpEdge(by));
    } else { // using TH2Poly
      TH2PolyBin* bin = static_cast<TH2PolyBin*>(static_cast<TH2Poly*>(hist)->GetBins()->At(bins[0]-1));
      // Just make a little fancy name
      BinName += fmt::format("Dim{} ({:g}, {:g})", 0, bin->GetXMin(), bin->GetXMax());
      BinName += fmt::format("Dim{} ({:g}, {:g})", 1, bin->GetYMin(), bin->GetYMax());
    }
  } else { // N-dimensional distribution using flatten TH1D
    BinName = hist->GetXaxis()->GetBinLabel(bins[0]);
  }
  return BinName;
}

GetLLH() [1/3]

double PredictiveThrower::GetLLH ( const TH1D *  DatHist, const TH1D *  MCHist, const TH1D *  W2Hist, const SampleHandlerInterface *  SampleHandler  ) const

private

Helper functions to calculate likelihoods using TH1D.

Parameters

DatHist	Data histogram with data distribution for a single sample
MCHist	MC histogram with MC distribution for a single sample
W2Hist	W2 histogram with W2 distribution for a single sample
SampleHandler	Pointer to SampleHandlerInterface providing LLH test statistic

Definition at line 1116 of file PredictiveThrower.cpp.

                                                                                    {
// *************************
  double llh = 0.0;
  for (int i = 1; i <= DatHist->GetXaxis()->GetNbins(); ++i)
  {
    const double data = DatHist->GetBinContent(i);
    const double mc = MCHist->GetBinContent(i);
    const double w2 = W2Hist->GetBinContent(i);
    llh += SampleHandler->GetTestStatLLH(data, mc, w2);
  }
  //KS: do times 2 because banff reports chi2
  return 2*llh;
}

GetLLH() [2/3]

double PredictiveThrower::GetLLH ( const TH2D *  DatHist, const TH2D *  MCHist, const TH2D *  W2Hist, const SampleHandlerInterface *  SampleHandler  ) const

private

Helper functions to calculate likelihoods using TH2D.

Parameters

DatHist	Data 2D histogram with data distribution for a single sample
MCHist	MC 2D histogram with MC distribution for a single sample
W2Hist	W2 2D histogram with W2 distribution for a single sample
SampleHandler	Pointer to SampleHandlerInterface providing LLH test statistic

Definition at line 1152 of file PredictiveThrower.cpp.

                                                                                    {
// *************************
  double llh = 0.0;
 
  const int nBinsX = DatHist->GetXaxis()->GetNbins();
  const int nBinsY = DatHist->GetYaxis()->GetNbins();
 
  for (int i = 1; i <= nBinsX; ++i)
  {
    for (int j = 1; j <= nBinsY; ++j)
    {
      const double data = DatHist->GetBinContent(i, j);
      const double mc   = MCHist->GetBinContent(i, j);
      const double w2   = W2Hist->GetBinContent(i, j);
 
      llh += SampleHandler->GetTestStatLLH(data, mc, w2);
    }
  }
 
  // KS: do times 2 because banff reports chi2
  return 2 * llh;
}

GetLLH() [3/3]

double PredictiveThrower::GetLLH ( const TH2Poly *  DatHist, const TH2Poly *  MCHist, const TH2Poly *  W2Hist, const SampleHandlerInterface *  SampleHandler  ) const

private

Helper functions to calculate likelihoods using TH2Poly.

Parameters

DatHist	Data 2D poly histogram with data distribution for a single sample
MCHist	MC 2D poly histogram with MC distribution for a single sample
W2Hist	W2 2D poly histogram with W2 distribution for a single sample
SampleHandler	Pointer to SampleHandlerInterface providing LLH test statistic

Definition at line 1134 of file PredictiveThrower.cpp.

                                                                                    {
// *************************
  double llh = 0.0;
  for (int i = 1; i <= DatHist->GetNumberOfBins(); ++i)
  {
    const double data = DatHist->GetBinContent(i);
    const double mc = MCHist->GetBinContent(i);
    const double w2 = W2Hist->GetBinContent(i);
    llh += SampleHandler->GetTestStatLLH(data, mc, w2);
  }
  //KS: do times 2 because banff reports chi2
  return 2*llh;
}

GetStoredFancyName()

std::vector< std::string > PredictiveThrower::GetStoredFancyName ( ParameterHandlerBase *  Systematics ) const

private

Get Fancy parameters stored in mcmc chains for passed ParameterHandler.

Definition at line 328 of file PredictiveThrower.cpp.

                                                                                                  {
// *************************
  TDirectory * ogdir = gDirectory;
 
  std::vector<std::string> FancyNames;
  std::string Name = std::string("Config_") + Systematics->GetName();
  auto PosteriorFileName = Get<std::string>(fitMan->raw()["Predictive"]["PosteriorFile"], __FILE__, __LINE__);
 
  TFile* file = TFile::Open(PosteriorFileName.c_str(), "READ");
  TDirectory* CovarianceFolder = file->GetDirectory("CovarianceFolder");
 
  TMacro* FoundMacro = static_cast<TMacro*>(CovarianceFolder->Get(Name.c_str()));
  if(FoundMacro == nullptr) {
    file->Close();
    delete file;
    if(ogdir){ ogdir->cd(); }
 
    return FancyNames;
  }
  MACH3LOG_DEBUG("Found config for {}", Name);
  YAML::Node Settings = TMacroToYAML(*FoundMacro);
 
  int params = int(Settings["Systematics"].size());
  FancyNames.resize(params);
  int iPar = 0;
  for (auto const &param : Settings["Systematics"]) {
    FancyNames[iPar] = Get<std::string>(param["Systematic"]["Names"]["FancyName"], __FILE__ , __LINE__);
    iPar++;
  }
  file->Close();
  delete file;
  if(ogdir){ ogdir->cd(); }
  return FancyNames;
}

LoadToys()

bool PredictiveThrower::LoadToys ( )

private

Load existing toys.

Definition at line 235 of file PredictiveThrower.cpp.

                                 {
// *************************
  auto PosteriorFileName = Get<std::string>(fitMan->raw()["Predictive"]["PosteriorFile"], __FILE__, __LINE__);
  // Open the ROOT file
  int originalErrorWarning = gErrorIgnoreLevel;
  gErrorIgnoreLevel = kFatal;
  TFile* file = TFile::Open(PosteriorFileName.c_str(), "READ");
 
  gErrorIgnoreLevel = originalErrorWarning;
  TDirectory* ToyDir = nullptr;
  if (!file || file->IsZombie()) {
    return false;
  } else {
    // Check for the "toys" directory
    if ((ToyDir = file->GetDirectory("Toys"))) {
      MACH3LOG_INFO("Found toys in Posterior file will attempt toy reading");
    } else {
      file->Close();
      delete file;
      return false;
    }
  }
 
  // Finally get the TTree branch with the penalty vectors for each of the toy throws
  TTree* PenaltyTree = static_cast<TTree*>(file->Get("ToySummary"));
  if (!PenaltyTree) {
    MACH3LOG_WARN("ToySummary TTree not found in file.");
    file->Close();
    delete file;
    return false;
  }
 
  Ntoys = static_cast<int>(PenaltyTree->GetEntries());
  int ConfigNtoys = Get<int>(fitMan->raw()["Predictive"]["Ntoy"], __FILE__, __LINE__);;
  if (Ntoys != ConfigNtoys) {
    MACH3LOG_WARN("Found different number of toys in saved file than asked to run!");
    MACH3LOG_INFO("Will read _ALL_ toys in the file");
    MACH3LOG_INFO("Ntoys in file: {}", Ntoys);
    MACH3LOG_INFO("Ntoys specified: {}", ConfigNtoys);
  }
 
  PenaltyTerm.resize(Ntoys);
  ReweightWeight.resize(Ntoys);
 
  double Penalty = 0, Weight = 1;
  PenaltyTree->SetBranchAddress("Penalty", &Penalty);
  PenaltyTree->SetBranchAddress("Weight", &Weight);
  PenaltyTree->SetBranchAddress("NModelParams", &NModelParams);
 
  for (int i = 0; i < Ntoys; ++i) {
    PenaltyTree->GetEntry(i);
    if (FullLLH) {
      PenaltyTerm[i] = Penalty;
    } else {
      PenaltyTerm[i] = 0.0;
    }
 
    ReweightWeight[i] = Weight;
  }
  // Resize all vectors and get sample names
  SetupSampleInformation();
 
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    TH1* DataHist1D = static_cast<TH1*>(ToyDir->Get((SampleInfo[sample].Name + "_data").c_str()));
    Data_Hist[sample] = M3::Clone(DataHist1D);
 
    TH1* MCHist1D = static_cast<TH1*>(ToyDir->Get((SampleInfo[sample].Name + "_mc").c_str()));
    MC_Nom_Hist[sample] = M3::Clone(MCHist1D);
 
    TH1* W2Hist1D = static_cast<TH1*>(ToyDir->Get((SampleInfo[sample].Name + "_w2").c_str()));
    W2_Nom_Hist[sample] = M3::Clone(W2Hist1D);
  }
 
 
  for (int iToy = 0; iToy < Ntoys; ++iToy)
  {
    if (iToy % 100 == 0) MACH3LOG_INFO("   Loaded toy {}", iToy);
 
    for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
      TH1* MCHist1D = static_cast<TH1*>(ToyDir->Get((SampleInfo[sample].Name + "_mc_" + std::to_string(iToy)).c_str()));
      TH1* W2Hist1D = static_cast<TH1*>(ToyDir->Get((SampleInfo[sample].Name + "_w2_" + std::to_string(iToy)).c_str()));
 
      MC_Hist_Toy[sample][iToy] = M3::Clone(MCHist1D);
      W2_Hist_Toy[sample][iToy] = M3::Clone(W2Hist1D);
    }
  }
 
  file->Close();
  delete file;
  return true;
}

MakeChi2Plots()

void PredictiveThrower::MakeChi2Plots ( const std::vector< std::vector< double >> &  Chi2_x, const std::string &  Chi2_x_title, const std::vector< std::vector< double >> &  Chi2_y, const std::string &  Chi2_y_title, const std::vector< TDirectory * > &  SampleDir, const std::string  Title  )

private

Produce Chi2 plot for a single sample based on which $p$-value is calculated.

Definition at line 1295 of file PredictiveThrower.cpp.

                                                             {
// *************************
  for (int iSample = 0; iSample < TotalNumberOfSamples+1; ++iSample) {
    SampleDir[iSample]->cd();
 
    // Transpose to extract chi2 values for a given sample across all toys
    std::vector<double> chi2_y_sample(Ntoys);
    std::vector<double> chi2_x_per_sample(Ntoys);
 
    for (int iToy = 0; iToy < Ntoys; ++iToy) {
      chi2_y_sample[iToy] = Chi2_y[iSample][iToy];
      chi2_x_per_sample[iToy]  = Chi2_x[iSample][iToy];
    }
 
    const double min_val = std::min(*std::min_element(chi2_y_sample.begin(), chi2_y_sample.end()),
                              *std::min_element(chi2_x_per_sample.begin(), chi2_x_per_sample.end()));
    const double max_val = std::max(*std::max_element(chi2_y_sample.begin(), chi2_y_sample.end()),
                              *std::max_element(chi2_x_per_sample.begin(), chi2_x_per_sample.end()));
 
    auto chi2_hist = std::make_unique<TH2D>((SampleInfo[iSample].Name+ Title).c_str(),
                                            (SampleInfo[iSample].Name+ Title).c_str(),
                                            50, min_val, max_val, 50, min_val, max_val);
    chi2_hist->SetDirectory(nullptr);
    chi2_hist->GetXaxis()->SetTitle(Chi2_x_title.c_str());
    chi2_hist->GetYaxis()->SetTitle(Chi2_y_title.c_str());
 
    for (int iToy = 0; iToy < Ntoys; ++iToy) {
      chi2_hist->Fill(chi2_x_per_sample[iToy], chi2_y_sample[iToy]);
    }
 
    Get2DBayesianpValue(chi2_hist.get());
    chi2_hist->Write();
  }
}

MakeCutEventRate()

void PredictiveThrower::MakeCutEventRate ( TH1D *  Histogram, const double  DataRate  ) const

private

Make the 1D Event Rate Hist.

Definition at line 1442 of file PredictiveThrower.cpp.

                                                                                     {
// ****************
  // Open the ROOT file
  int originalErrorWarning = gErrorIgnoreLevel;
  gErrorIgnoreLevel = kFatal;
 
  // For the event rate histogram add a TLine to the data rate
  auto TempLine = std::make_unique<TLine>(DataRate, Histogram->GetMinimum(), DataRate, Histogram->GetMaximum());
  TempLine->SetLineColor(kRed);
  TempLine->SetLineWidth(2);
  // Also fit a Gaussian because why not?
  auto Fitter = std::make_unique<TF1>("Fit", "gaus", Histogram->GetBinLowEdge(1), Histogram->GetBinLowEdge(Histogram->GetNbinsX()+1));
  Histogram->Fit(Fitter.get(), "RQ");
  Fitter->SetLineColor(kRed-5);
  // Calculate a p-value
  double Above = 0.0;
  for (int z = 0; z < Histogram->GetNbinsX(); ++z) {
    const double xvalue = Histogram->GetBinCenter(z+1);
    if (xvalue >= DataRate) {
      Above += Histogram->GetBinContent(z+1);
    }
  }
  const double pvalue = Above/Histogram->Integral();
  TLegend Legend(0.4, 0.75, 0.98, 0.90);
  Legend.SetFillColor(0);
  Legend.SetFillStyle(0);
  Legend.SetLineWidth(0);
  Legend.SetLineColor(0);
  Legend.AddEntry(TempLine.get(), Form("Data, %.0f, p-value=%.2f", DataRate, pvalue), "l");
  Legend.AddEntry(Histogram, Form("MC, #mu=%.1f#pm%.1f", Histogram->GetMean(), Histogram->GetRMS()), "l");
  Legend.AddEntry(Fitter.get(), Form("Gauss, #mu=%.1f#pm%.1f", Fitter->GetParameter(1), Fitter->GetParameter(2)), "l");
  std::string TempTitle = std::string(Histogram->GetName());
  TempTitle += "_canv";
  TCanvas TempCanvas(TempTitle.c_str(), TempTitle.c_str(), 1024, 1024);
  TempCanvas.SetGridx();
  TempCanvas.SetGridy();
  TempCanvas.SetRightMargin(0.03);
  TempCanvas.SetBottomMargin(0.08);
  TempCanvas.SetLeftMargin(0.10);
  TempCanvas.SetTopMargin(0.06);
  TempCanvas.cd();
  Histogram->Draw();
  TempLine->Draw("same");
  Fitter->Draw("same");
  Legend.Draw("same");
  TempCanvas.Write();
  Histogram->Write();
  gErrorIgnoreLevel = originalErrorWarning;
}

MakeFluctuatedHistogram()

void PredictiveThrower::MakeFluctuatedHistogram ( TH1 *  FluctHist, TH1 *  PolyHist  )

private

Make Poisson fluctuation of TH1D hist.

Parameters

FluctHist	Histogram to store fluctuated values (must match Hist type)
Hist	Original histogram to fluctuate

Definition at line 1180 of file PredictiveThrower.cpp.

                                                                         {
// ****************
  // Determine which fluctuation function to call
  auto applyFluctuation = [&](auto* f, auto* h) {
    if (StandardFluctuation) {
      MakeFluctuatedHistogramStandard(f, h, random.get());
    } else {
      MakeFluctuatedHistogramAlternative(f, h, random.get());
    }
  };
 
  if (Hist->InheritsFrom(TH2Poly::Class())) {
    applyFluctuation(static_cast<TH2Poly*>(FluctHist), static_cast<TH2Poly*>(Hist));
  }
  else if (Hist->InheritsFrom(TH2D::Class())) {
    applyFluctuation(static_cast<TH2D*>(FluctHist), static_cast<TH2D*>(Hist));
  }
  else if (Hist->InheritsFrom(TH1D::Class())) {
    applyFluctuation(static_cast<TH1D*>(FluctHist), static_cast<TH1D*>(Hist));
  }
  else {
    MACH3LOG_ERROR("Unsupported histogram type");
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
}

MakePredictive()

std::vector< std::unique_ptr< TH1 > > PredictiveThrower::MakePredictive ( const std::vector< std::vector< std::unique_ptr< TH1 >>> &  Toys, const std::vector< TDirectory * > &  Director, const std::string &  suffix, const bool  DebugHistograms, const bool  WriteHist  )

private

Produce posterior predictive distribution.

1. Fill histograms, thread safe as all histograms are allocated before and we loop over samples

Definition at line 916 of file PredictiveThrower.cpp.

                                                                                          {
// *************************
  std::vector<std::unique_ptr<TH1>> PostPred(TotalNumberOfSamples);
  std::vector<std::vector<std::unique_ptr<TH1D>>> Posterior_hist(TotalNumberOfSamples);
  // 1.initialisation
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    const int nDims = SampleInfo[sample].Dimenstion;
    const std::string Sample_Name = SampleInfo[sample].Name;
    Posterior_hist[sample] = PerBinHistogram(Toys[sample][0].get(), sample, nDims, suffix);
    auto PredictiveHist = M3::Clone(Toys[sample][0].get());
    // Clear the bin contents
    PredictiveHist->Reset();
    PredictiveHist->SetName((Sample_Name + "_" + suffix + "_PostPred").c_str());
    PredictiveHist->SetTitle((Sample_Name + "_" + suffix + "_PostPred").c_str());
    PredictiveHist->SetDirectory(nullptr);
    PostPred[sample] = std::move(PredictiveHist);
  }
 
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    const int nDims = SampleInfo[sample].Dimenstion;
    auto& hist = Toys[sample][0];
    for (size_t iToy = 0; iToy < Toys[sample].size(); ++iToy) {
      if(nDims == 2) {
        if(std::string(hist->ClassName()) == "TH2Poly") {
          for (int i = 1; i <= static_cast<TH2Poly*>(hist.get())->GetNumberOfBins(); ++i) {
            double content = Toys[sample][iToy]->GetBinContent(i);
            Posterior_hist[sample][i-1]->Fill(content, ReweightWeight[iToy]);
          }
        } else {
          int nbinsx = hist->GetNbinsX();
          int nbinsy = hist->GetNbinsY();
          for (int iy = 1; iy <= nbinsy; ++iy) {
            for (int ix = 1; ix <= nbinsx; ++ix) {
              int Bin = (iy-1) * nbinsx + (ix-1);
              double content = Toys[sample][iToy]->GetBinContent(ix, iy);
              Posterior_hist[sample][Bin]->Fill(content, ReweightWeight[iToy]);
            } // end loop over X bins
          } // end loop over Y bins
        }
      } else {
        int nbinsx = hist->GetNbinsX();
          for (int i = 1; i <= nbinsx; ++i) {
            double content = Toys[sample][iToy]->GetBinContent(i);
            Posterior_hist[sample][i-1]->Fill(content, ReweightWeight[iToy]);
          } // end loop over bins
        } // end if over dimensions
    } // end loop over toys
  } // end loop over samples
 
  // 3.save
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    const int nDims = SampleInfo[sample].Dimenstion;
    auto& hist = Toys[sample][0];
    Directory[sample]->cd();
    if(nDims == 2) {
      if(std::string(hist->ClassName()) == "TH2Poly") {
        for (int i = 1; i <= static_cast<TH2Poly*>(hist.get())->GetNumberOfBins(); ++i) {
          PostPred[sample]->SetBinContent(i, Posterior_hist[sample][i-1]->GetMean());
          // KS: If ROOT below 6.18 one need -1 only for error due to stupid bug...
          PostPred[sample]->SetBinError(i, Posterior_hist[sample][i-1]->GetRMS());
          if (DebugHistograms) Posterior_hist[sample][i-1]->Write();
        } // end loop over poly bins
      } else {
        int nbinsx = hist->GetNbinsX();
        int nbinsy = hist->GetNbinsY();
        for (int iy = 1; iy <= nbinsy; ++iy) {
          for (int ix = 1; ix <= nbinsx; ++ix) {
            int Bin = (iy-1) * nbinsx + (ix-1);
            if (DebugHistograms) Posterior_hist[sample][Bin]->Write();
            PostPred[sample]->SetBinContent(ix, iy, Posterior_hist[sample][Bin]->GetMean());
            PostPred[sample]->SetBinError(ix, iy, Posterior_hist[sample][Bin]->GetRMS());
          } // end loop over x
        } // end loop over y
      }
    } else {
      int nbinsx = hist->GetNbinsX();
      for (int i = 1; i <= nbinsx; ++i) {
        PostPred[sample]->SetBinContent(i, Posterior_hist[sample][i-1]->GetMean());
        PostPred[sample]->SetBinError(i, Posterior_hist[sample][i-1]->GetRMS());
        if (DebugHistograms) Posterior_hist[sample][i-1]->Write();
      }
    }
    if(WriteHist) PostPred[sample]->Write();
  } // end loop over samples
  return PostPred;
}

PerBinHistogram()

std::vector< std::unique_ptr< TH1D > > PredictiveThrower::PerBinHistogram ( TH1 *  hist, const int  SampleId, const int  Dim, const std::string &  suffix  ) const

private

Create per-bin posterior histograms for a given sample.

For each analysis bin of the input histogram, this function allocates a new 1D histogram intended to accumulate the distribution of predicted event counts (e.g. across throws, toys, or posterior evaluations).

The number of output histograms therefore equals the number of physical bins:

• TH1 → N histograms
• TH2 → Nx × Ny histograms
• TH2Poly → one histogram per polygon bin

Parameters

hist	Input histogram defining the bin structure for this sample.
SampleId	Index identifying the sample in SampleInfo.
Dim	Dimensionality of the original distribution.
suffix	String appended to histogram names (e.g. to distinguish stages).

Definition at line 859 of file PredictiveThrower.cpp.

                                                                                                     {
// *************************
  std::vector<std::unique_ptr<TH1D>> PosteriorHistVec;
  constexpr int nBins = 100;
  const std::string Sample_Name = SampleInfo[SampleId].Name;
  if (Dim == 2) {
      if(std::string(hist->ClassName()) == "TH2Poly") {
        for (int i = 1; i <= static_cast<TH2Poly*>(hist)->GetNumberOfBins(); ++i) {
          std::string ProjName = fmt::format("{} {} Bin: {}",
                                             Sample_Name, suffix,
                                             GetBinName(hist, false, Dim, {i}));
          // KS: When a histogram is created with an axis lower limit greater or equal to its upper limit ROOT will automatically adjust histogram range
          // https://root.cern.ch/doc/master/classTH1.html#auto-bin
          auto PosteriorHist = std::make_unique<TH1D>(ProjName.c_str(), ProjName.c_str(), nBins, 1, -1);
          PosteriorHist->SetDirectory(nullptr);
          PosteriorHist->GetXaxis()->SetTitle("Events");
          PosteriorHistVec.push_back(std::move(PosteriorHist));
        } //end loop over bin
      } else {
        int nbinsx = hist->GetNbinsX();
        int nbinsy = hist->GetNbinsY();
        for (int iy = 1; iy <= nbinsy; ++iy) {
          for (int ix = 1; ix <= nbinsx; ++ix) {
            std::string ProjName = fmt::format("{} {} Bin: {}",
                                              Sample_Name, suffix,
                                              GetBinName(hist, true, Dim, {ix,iy}));
            //KS: When a histogram is created with an axis lower limit greater or equal to its upper limit ROOT will automatically adjust histogram range
            // https://root.cern.ch/doc/master/classTH1.html#auto-bin
            auto PosteriorHist = std::make_unique<TH1D>(ProjName.c_str(), ProjName.c_str(), nBins, 1, -1);
            PosteriorHist->SetDirectory(nullptr);
            PosteriorHist->GetXaxis()->SetTitle("Events");
            PosteriorHistVec.push_back(std::move(PosteriorHist));
          }
        }
      }
  } else {
    int nbinsx = hist->GetNbinsX();
    PosteriorHistVec.reserve(nbinsx);
    for (int i = 1; i <= nbinsx; ++i) {
      std::string ProjName = fmt::format("{} {} Bin: {}",
                                          Sample_Name, suffix,
                                          GetBinName(hist, true, Dim, {i}));
      //KS: When a histogram is created with an axis lower limit greater or equal to its upper limit ROOT will automatically adjust histogram range
      // https://root.cern.ch/doc/master/classTH1.html#auto-bin
      auto PosteriorHist = std::make_unique<TH1D>(ProjName.c_str(), ProjName.c_str(), nBins, 1, -1);
      PosteriorHist->SetDirectory(nullptr);
      PosteriorHist->GetXaxis()->SetTitle("Events");
      PosteriorHistVec.push_back(std::move(PosteriorHist));
    }
  }
  return PosteriorHistVec;
}

PosteriorPredictivepValue()

void PredictiveThrower::PosteriorPredictivepValue ( const std::vector< std::unique_ptr< TH1 >> &  PostPred_mc, const std::vector< TDirectory * > &  SampleDir  )

private

Calculate Posterior Predictive $p$-value Compares observed data to toy datasets generated from:

• fitted model fluctuations ("Draw")
• posterior predictive distribution ("Pred")

Computes two discrepancy metrics: • Shape+Rate : bin-by-bin likelihood • Rate-only : total event normalization

TODO This can be multithreaded but be careful for Clone!!!

Definition at line 1207 of file PredictiveThrower.cpp.

                                                                                           {
// *************************
  // Step 1: Initialize per-toy accumulators once
  // [Sample] [Toys]
  auto make_matrix = [&](double init = 0.0) {
    return std::vector<std::vector<double>>(
      TotalNumberOfSamples + 1,
      std::vector<double>(Ntoys, init));
  };
  auto chi2_dat       = make_matrix();
  auto chi2_mc        = make_matrix();
  auto chi2_pred      = make_matrix();
  auto chi2_rate_dat  = make_matrix();
  auto chi2_rate_mc   = make_matrix();
  auto chi2_rate_pred = make_matrix();
 
  // 2. Add penalty terms to global bin
  for (int iToy = 0; iToy < Ntoys; ++iToy) {
    chi2_dat[TotalNumberOfSamples][iToy] = PenaltyTerm[iToy];
    chi2_mc[TotalNumberOfSamples][iToy] = PenaltyTerm[iToy];
    chi2_pred[TotalNumberOfSamples][iToy] = PenaltyTerm[iToy];
 
    chi2_rate_dat[TotalNumberOfSamples][iToy] = PenaltyTerm[iToy];
    chi2_rate_mc[TotalNumberOfSamples][iToy] = PenaltyTerm[iToy];
    chi2_rate_pred[TotalNumberOfSamples][iToy] = PenaltyTerm[iToy];
  }
 
  for (int iSample = 0; iSample < TotalNumberOfSamples; ++iSample) {
    auto SampleHandler = SampleInfo[iSample].SamHandler;
    for (int iToy = 0; iToy < Ntoys; ++iToy) {
      // Clone histograms to avoid modifying originals
      auto DrawFluctHist = M3::Clone(MC_Hist_Toy[iSample][iToy].get());
      auto PredFluctHist = M3::Clone(PostPred_mc[iSample].get());
 
      // Apply fluctuations
      MakeFluctuatedHistogram(DrawFluctHist.get(), MC_Hist_Toy[iSample][iToy].get());
      MakeFluctuatedHistogram(PredFluctHist.get(), PostPred_mc[iSample].get());
 
      // I. SHAPE + RATE (bin-by-bin likelihood)
      chi2_dat[iSample][iToy]  = CalcLLH(Data_Hist[iSample].get(), MC_Hist_Toy[iSample][iToy].get(), W2_Hist_Toy[iSample][iToy].get(), SampleHandler);
      chi2_mc[iSample][iToy]   = CalcLLH(DrawFluctHist.get(), MC_Hist_Toy[iSample][iToy].get(), W2_Hist_Toy[iSample][iToy].get(), SampleHandler);
      chi2_pred[iSample][iToy] = CalcLLH(PredFluctHist.get(), MC_Hist_Toy[iSample][iToy].get(), W2_Hist_Toy[iSample][iToy].get(), SampleHandler);
 
      // II. RATE-ONLY (total normalization)
      chi2_rate_dat[iSample][iToy]  = CalcLLH(Data_Hist[iSample]->Integral(), MC_Hist_Toy[iSample][iToy]->Integral(), W2_Hist_Toy[iSample][iToy]->Integral(), SampleHandler);
      chi2_rate_mc[iSample][iToy]   = CalcLLH(DrawFluctHist->Integral(), MC_Hist_Toy[iSample][iToy]->Integral(), W2_Hist_Toy[iSample][iToy]->Integral(), SampleHandler);
      chi2_rate_pred[iSample][iToy] = CalcLLH(PredFluctHist->Integral(), MC_Hist_Toy[iSample][iToy]->Integral(), W2_Hist_Toy[iSample][iToy]->Integral(), SampleHandler);
 
      // III. accumulate global sums ---
      chi2_dat[TotalNumberOfSamples][iToy]  += chi2_dat[iSample][iToy];
      chi2_mc[TotalNumberOfSamples][iToy]   += chi2_mc[iSample][iToy];
      chi2_pred[TotalNumberOfSamples][iToy] += chi2_pred[iSample][iToy];
 
      chi2_rate_dat[TotalNumberOfSamples][iToy]  += chi2_rate_dat[iSample][iToy];
      chi2_rate_mc[TotalNumberOfSamples][iToy]   += chi2_rate_mc[iSample][iToy];
      chi2_rate_pred[TotalNumberOfSamples][iToy] += chi2_rate_pred[iSample][iToy];
    }
  }
 
  // 4. Produce pValue plots
  // Shape+rate posterior predictive checks
  MakeChi2Plots(chi2_mc,   "-2LLH (Draw Fluc, Draw)", chi2_dat, "-2LLH (Data, Draw)", SampleDir, "_drawfluc_draw");
  MakeChi2Plots(chi2_pred, "-2LLH (Pred Fluc, Draw)", chi2_dat, "-2LLH (Data, Draw)", SampleDir, "_predfluc_draw");
 
  // Rate-only posterior predictive checks
  MakeChi2Plots(chi2_rate_mc,   "-2LLH (Rate Draw Fluc, Draw)", chi2_rate_dat, "-2LLH (Rate Data, Draw)", SampleDir, "_rate_drawfluc_draw");
  MakeChi2Plots(chi2_rate_pred, "-2LLH (Rate Pred Fluc, Draw)", chi2_rate_dat, "-2LLH (Rate Data, Draw)", SampleDir, "_rate_predfluc_draw");
}

PredictiveLLH()

void PredictiveThrower::PredictiveLLH ( const std::vector< std::unique_ptr< TH1 >> &  Data_histogram, const std::vector< std::unique_ptr< TH1 >> &  PostPred_mc, const std::vector< std::unique_ptr< TH1 >> &  PostPred_w, const std::vector< TDirectory * > &  SampleDir  )

private

Calculate Posterior Predictive LLH.

Definition at line 1279 of file PredictiveThrower.cpp.

                                                                                        {
// *************************
  MACH3LOG_INFO("{:<55} {:<10} {:<10} {:<10}", "Sample", "DataInt", "MCInt", "-2LLH");
  MACH3LOG_INFO("{:-<55} {:-<10} {:-<10} {:-<10}", "", "", "", "");
  for (int iSample = 0; iSample < TotalNumberOfSamples; ++iSample) {
    SampleDir[iSample]->cd();
    ExtractLLH(Data_histogram[iSample].get(), PostPred_mc[iSample].get(), PostPred_w[iSample].get(), SampleInfo[iSample].SamHandler);
    PostPred_mc[iSample]->Write();
  }
}

ProduceSpectra()

void PredictiveThrower::ProduceSpectra ( const std::vector< std::vector< std::vector< std::unique_ptr< TH1D >>>> &  Toys, const std::vector< TDirectory * > &  Director, const std::string  suffix  ) const

private

Produce Violin style spectra.

Definition at line 732 of file PredictiveThrower.cpp.

                                                                     {
// *************************
  std::vector<std::vector<double>> MaxValue(TotalNumberOfSamples);
 
  // 1. Create Max value
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    const int nDims = SampleInfo[sample].Dimenstion;
    MaxValue[sample].assign(nDims, 0);
  }
 
  // 2. Find maximum entries over all toys
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    for (int toy = 0; toy < Ntoys; ++toy) {
      const int nDims = SampleInfo[sample].Dimenstion;
      for (int dim = 0; dim < nDims; dim++) {
        double max_val = Toys[sample][toy][dim]->GetMaximum();
        MaxValue[sample][dim] = std::max(MaxValue[sample][dim], max_val);
      }
    }
  }
 
  // 3. Make actual spectra histogram (this is because making ROOT histograms is not save)
  // And we now have actual max values
  std::vector<std::vector<std::unique_ptr<TH2D>>> Spectra(TotalNumberOfSamples);
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    const int nDims = SampleInfo[sample].Dimenstion;
    Spectra[sample].resize(nDims);
    for (int dim = 0; dim < nDims; dim++) {
      // Get MC histogram x-axis binning
      TH1D* refHist = Toys[sample][0][dim].get();
 
      const int n_bins_x = refHist->GetNbinsX();
      std::vector<double> x_bin_edges(n_bins_x + 1);
      for (int b = 0; b < n_bins_x; ++b) {
        x_bin_edges[b] = refHist->GetXaxis()->GetBinLowEdge(b + 1);
      }
      x_bin_edges[n_bins_x] = refHist->GetXaxis()->GetBinUpEdge(n_bins_x);
 
      constexpr int n_bins_y = 400;
      constexpr double y_min = 0.0;
      const double y_max = MaxValue[sample][dim] * 1.05;
 
      // Create TH2D with variable binning on x axis
      Spectra[sample][dim] = std::make_unique<TH2D>(
        (SampleInfo[sample].Name + "_" + suffix + "_dim" + std::to_string(dim)).c_str(),   // name
        (SampleInfo[sample].Name + "_" + suffix + "_dim" + std::to_string(dim)).c_str(),   // title
        n_bins_x, x_bin_edges.data(),                   // x axis bins
        n_bins_y, y_min, y_max                          // y axis bins
      );
 
      Spectra[sample][dim]->GetXaxis()->SetTitle(refHist->GetXaxis()->GetTitle());
      Spectra[sample][dim]->GetYaxis()->SetTitle("Events");
 
      Spectra[sample][dim]->SetDirectory(nullptr);
      Spectra[sample][dim]->Sumw2(true);
    }
  }
 
  // 4. now we can actually fill our projections
  #ifdef MULTITHREAD
  #pragma omp parallel for collapse(2)
  #endif
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    for (int toy = 0; toy < Ntoys; ++toy) {
      const int nDims = SampleInfo[sample].Dimenstion;
      for (int dim = 0; dim < nDims; dim++) {
        FastViolinFill(Spectra[sample][dim].get(), Toys[sample][toy][dim].get());
      }
    }
  }
 
  // 5. Save histograms which is not thread save
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    SampleDirectories[sample]->cd();
    const int nDims = SampleInfo[sample].Dimenstion;
    for (long unsigned int dim = 0; dim < Spectra[sample].size(); dim++) {
      Spectra[sample][dim]->Write();
      // For case of 2D make additional histograms
      if(nDims == 2) {
        const std::string name = SampleInfo[sample].Name + "_" + suffix+ "_PostPred_dim" + std::to_string(dim);
        auto Summary = MakeSummaryFromSpectra(Spectra[sample][dim].get(), name);
        Summary->Write();
      }
    }
  }
}

ProduceToys()

void PredictiveThrower::ProduceToys

(

)

Produce toys by throwing from MCMC.

KS: Names of parameter groups that will not be varied

KS: Index of parameters that will be varied

this store value of parameters sampled from a chain

Definition at line 439 of file PredictiveThrower.cpp.

                                    {
// *************************
  // If we found toys then skip process of making new toys
  if(LoadToys()) return;
 
  std::vector<std::string> ParameterGroupsNotVaried;
  std::unordered_set<int> ParameterOnlyToVary;
  // For study where one would like to apply bounds
  std::vector<const M3::float_t*> BoundValuePointer;
  std::vector<std::pair<double, double>> ParamBounds;
 
  // Setup useful information for toy generation
  SetupToyGeneration(ParameterGroupsNotVaried, ParameterOnlyToVary,
                     BoundValuePointer, ParamBounds);
 
  auto PosteriorFileName = Get<std::string>(fitMan->raw()["Predictive"]["PosteriorFile"], __FILE__, __LINE__);
 
  MACH3LOG_INFO("Starting {}", __func__);
 
  outputFile->cd();
  double Penalty = 0, Weight = 1;
  int Draw = 0;
 
  TTree *ToyTree = new TTree("ToySummary", "ToySummary");
  ToyTree->Branch("Penalty", &Penalty, "Penalty/D");
  ToyTree->Branch("Weight", &Weight, "Weight/D");
  ToyTree->Branch("Draw", &Draw, "Draw/I");
  ToyTree->Branch("NModelParams", &NModelParams, "NModelParams/I");
 
  // KS: define branches so we can keep track of what params we are throwing
  std::vector<double> ParamValues(NModelParams);
  std::vector<const M3::float_t*> ParampPointers(NModelParams);
  int ParamCounter = 0;
  for (size_t iSys = 0; iSys < systematics.size(); iSys++)
  {
    for (int iPar = 0; iPar < systematics[iSys]->GetNumParams(); iPar++)
    {
      ParampPointers[ParamCounter] = systematics[iSys]->RetPointer(iPar);
      std::string Name = systematics[iSys]->GetParFancyName(iPar);
      //CW: Also strip out - signs because it messes up TBranches
      while (Name.find("-") != std::string::npos) {
        Name.replace(Name.find("-"), 1, std::string("_"));
      }
      ToyTree->Branch(Name.c_str(), &ParamValues[ParamCounter], (Name + "/D").c_str());
      ParamCounter++;
    }
  }
  TDirectory* ToyDirectory = outputFile->mkdir("Toys");
  ToyDirectory->cd();
  int SampleCounter = 0;
  for (size_t iPDF = 0; iPDF < samples.size(); iPDF++)
  {
    auto* MaCh3Sample = samples[iPDF];
    for (int SampleIndex = 0; SampleIndex < MaCh3Sample->GetNSamples(); ++SampleIndex)
    {
      // Get nominal spectra and event rates
      const TH1* DataHist = MaCh3Sample->GetDataHist(SampleIndex);
      Data_Hist[SampleCounter] = M3::Clone(DataHist, MaCh3Sample->GetSampleTitle(SampleIndex) + "_data");
      Data_Hist[SampleCounter]->Write((MaCh3Sample->GetSampleTitle(SampleIndex) + "_data").c_str());
 
      const TH1* MCHist = MaCh3Sample->GetMCHist(SampleIndex);
      MC_Nom_Hist[SampleCounter] = M3::Clone(MCHist, MaCh3Sample->GetSampleTitle(SampleIndex) + "_mc");
      MC_Nom_Hist[SampleCounter]->Write((MaCh3Sample->GetSampleTitle(SampleIndex) + "_mc").c_str());
 
      const TH1* W2Hist = MaCh3Sample->GetW2Hist(SampleIndex);
      W2_Nom_Hist[SampleCounter] = M3::Clone(W2Hist, MaCh3Sample->GetSampleTitle(SampleIndex) + "_w2");
      W2_Nom_Hist[SampleCounter]->Write((MaCh3Sample->GetSampleTitle(SampleIndex) + "_w2").c_str());
      SampleCounter++;
    }
  }
 
  TDirectory* Toy_1DDirectory = outputFile->mkdir("Toys_1DHistVar");
  TDirectory* Toy_2DDirectory = outputFile->mkdir("Toys_2DHistVar");
  std::vector<std::vector<double>> branch_vals(systematics.size());
  std::vector<std::vector<std::string>> branch_name(systematics.size());
 
  TChain* PosteriorFile = nullptr;
  unsigned int burn_in = 0;
  unsigned int maxNsteps = 0;
  unsigned int Step = 0;
  if(!Is_PriorPredictive)
  {
    PosteriorFile = new TChain("posteriors");
    PosteriorFile->Add(PosteriorFileName.c_str());
 
    PosteriorFile->SetBranchAddress("step", &Step);
    if (PosteriorFile->GetBranch("Weight")) {
      PosteriorFile->SetBranchStatus("Weight", true);
      PosteriorFile->SetBranchAddress("Weight", &Weight);
    } else {
      MACH3LOG_WARN("Not applying reweighting weight");
      Weight = 1.0;
    }
 
    for (size_t s = 0; s < systematics.size(); ++s) {
      auto fancy_names = GetStoredFancyName(systematics[s]);
      systematics[s]->MatchMaCh3OutputBranches(PosteriorFile, branch_vals[s], branch_name[s], fancy_names);
    }
 
    //Get the burn-in from the config
    burn_in = Get<unsigned int>(fitMan->raw()["Predictive"]["BurnInSteps"], __FILE__, __LINE__);
 
    //DL: Adding sanity check for chains shorter than burn in
    maxNsteps = static_cast<unsigned int>(PosteriorFile->GetMaximum("step"));
    if(burn_in >= maxNsteps)
    {
      MACH3LOG_ERROR("You are running on a chain shorter than burn in cut");
      MACH3LOG_ERROR("Maximal value of nSteps: {}, burn in cut {}", maxNsteps, burn_in);
      MACH3LOG_ERROR("You will run into infinite loop");
      MACH3LOG_ERROR("You can make new chain or modify burn in cut");
      throw MaCh3Exception(__FILE__,__LINE__);
    }
  }
 
  TStopwatch TempClock;
  TempClock.Start();
  for(int i = 0; i < Ntoys; i++)
  {
    if(Ntoys >= 10 && i % (Ntoys/10) == 0) {
      M3::Utils::PrintProgressBar(i, Ntoys);
    }
    if(!Is_PriorPredictive){
      int entry = 0;
      Step = 0;
      // KS This allow to set additional bounds like mass ordering
      bool WithinBounds = false;
      //YSP: Ensures you get an entry from the mcmc even when burn_in is set to zero (Although not advised :p ).
      //Take 200k burn in steps, WP: Eb C in 1st peaky
      // If we have combined chains by hadd need to check the step in the chain
      // Note, entry is not necessarily same as step due to merged ROOT files, so can't choose entry in the range BurnIn - nEntries :(
      while(Step < burn_in || !WithinBounds) {
        entry = random->Integer(static_cast<unsigned int>(PosteriorFile->GetEntries()));
        PosteriorFile->GetEntry(entry);
        // KS: This might be bit hacky... but BoundValuePointer refer to values in ParameterHandler
        // so we need to update them
        if(BoundValuePointer.size() > 0) {
          for (size_t s = 0; s < systematics.size(); ++s) {
            systematics[s]->SetParameters(branch_vals[s]);
          }
        }
        WithinBounds = CheckBounds(BoundValuePointer, ParamBounds);
      }
      Draw = entry;
    }
    for (size_t s = 0; s < systematics.size(); ++s)
    {
      //KS: Below line can help you get prior predictive distributions which are helpful for getting pre and post ND fit spectra
      //YSP: If not set in the config, the code runs SK Posterior Predictive distributions by default. If true, then the code runs SK prior predictive.
      if(Is_PriorPredictive) {
        systematics[s]->ThrowParameters();
      } else {
        systematics[s]->SetParameters(branch_vals[s]);
      }
    }
 
    // This set some params to prior value this way you can evaluate errors from subset of errors
    SetParamters(ParameterGroupsNotVaried, ParameterOnlyToVary);
 
    Penalty = 0;
    if(FullLLH) {
      for (size_t s = 0; s < systematics.size(); ++s) {
        //KS: do times 2 because banff reports chi2
        Penalty = 2.0 * systematics[s]->GetLikelihood();
      }
    }
 
    PenaltyTerm[i] = Penalty;
    ReweightWeight[i] = Weight;
 
    for (size_t iPDF = 0; iPDF < samples.size(); iPDF++) {
      samples[iPDF]->Reweight();
    }
    // Save histograms to file
    WriteToy(ToyDirectory, Toy_1DDirectory, Toy_2DDirectory, i);
 
    // Fill parameter value so we know throw values
    for (size_t iPar = 0; iPar < ParamValues.size(); iPar++) {
      ParamValues[iPar] = *ParampPointers[iPar];
    }
 
    ToyTree->Fill();
  }//end of toys loop
  TempClock.Stop();
 
  if(PosteriorFile) delete PosteriorFile;
  ToyDirectory->Close(); delete ToyDirectory;
  Toy_1DDirectory->Close(); delete Toy_1DDirectory;
  Toy_2DDirectory->Close(); delete Toy_2DDirectory;
 
  outputFile->cd();
  ToyTree->Write(); delete ToyTree;
 
  MACH3LOG_INFO("{} took {:.2f}s to finish for {} toys", __func__, TempClock.RealTime(), Ntoys);
}

RateAnalysis()

void PredictiveThrower::RateAnalysis ( const std::vector< std::vector< std::unique_ptr< TH1 >>> &  Toys, const std::vector< TDirectory * > &  SampleDirectories  ) const

private

Produce distribution of number of events for each sample.

Definition at line 1493 of file PredictiveThrower.cpp.

                                                                                            {
// *************************
  std::vector<std::unique_ptr<TH1D>> EventHist(TotalNumberOfSamples+1);
  for (int iSample = 0; iSample < TotalNumberOfSamples+1; ++iSample) {
    std::string Title = "EventHist: ";
    if (iSample == TotalNumberOfSamples) {
      Title = "Total";
    } else {
      Title = SampleInfo[iSample].Name;
    }
    Title += "_sum";
    //KS: When a histogram is created with an axis lower limit greater or equal to its upper limit ROOT will automatically adjust histogram range
    // https://root.cern.ch/doc/master/classTH1.html#auto-bin
    EventHist[iSample] = std::make_unique<TH1D>(Title.c_str(), Title.c_str(), 100, 1, -1);
    EventHist[iSample]->SetDirectory(nullptr);
    EventHist[iSample]->GetXaxis()->SetTitle("Total event rate");
    EventHist[iSample]->GetYaxis()->SetTitle("Counts");
    EventHist[iSample]->SetLineWidth(2);
  }
 
  // First fill per-sample histograms
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int iSample = 0; iSample < TotalNumberOfSamples; ++iSample) {
    for (int iToy = 0; iToy < Ntoys; ++iToy) {
      double Count = Toys[iSample][iToy]->Integral();
      EventHist[iSample]->Fill(Count);
    }
  }
 
  // Now fill total histogram properly (per toy)
  for (int iToy = 0; iToy < Ntoys; ++iToy) {
    double TotalCount = 0.0;
    for (int iSample = 0; iSample < TotalNumberOfSamples; ++iSample) {
      TotalCount += Toys[iSample][iToy]->Integral();
    }
    EventHist[TotalNumberOfSamples]->Fill(TotalCount);
  }
 
  double DataRate = 0.0;
  std::vector<double> DataRates(TotalNumberOfSamples+1);
  #ifdef MULTITHREAD
  #pragma omp parallel for reduction(+:DataRate)
  #endif
  for (int i = 0; i < TotalNumberOfSamples; ++i) {
    DataRates[i] = Data_Hist[i]->Integral();
    DataRate += DataRates[i];
  }
  DataRates[TotalNumberOfSamples] = DataRate;
 
  for (int SampleNum = 0; SampleNum < TotalNumberOfSamples+1; ++SampleNum) {
    SampleDirectories[SampleNum]->cd();
    //Make fancy event rate histogram
    MakeCutEventRate(EventHist[SampleNum].get(), DataRates[SampleNum]);
  }
}

RunMCMC()

void PredictiveThrower::RunMCMC ( )

inlineoverridevirtual

This is not used in this class.

Implements FitterBase.

Definition at line 53 of file PredictiveThrower.h.

                          {
    MACH3LOG_ERROR("{} is not supported in {}", __func__, GetName());
    throw MaCh3Exception(__FILE__ , __LINE__ );
  };

RunPredictiveAnalysis()

void PredictiveThrower::RunPredictiveAnalysis

(

)

Main routine responsible for producing posterior predictive distributions and $p$-value.

Definition at line 1013 of file PredictiveThrower.cpp.

                                              {
// *************************
  // Remove not useful stuff
  SanitiseInputs();
 
  MACH3LOG_INFO("Starting {}", __func__);
  MACH3LOG_WARN("\033[0;31mCurrent Total RAM usage is {:.2f} GB\033[0m", M3::Utils::getValue("VmRSS") / 1048576.0);
  MACH3LOG_WARN("\033[0;31mOut of Total available RAM {:.2f} GB\033[0m", M3::Utils::getValue("MemTotal") / 1048576.0);
 
  TStopwatch TempClock;
  TempClock.Start();
 
  auto DebugHistograms = GetFromManager<bool>(fitMan->raw()["Predictive"]["DebugHistograms"], false, __FILE__, __LINE__);
  auto StudyBeta = GetFromManager<bool>(fitMan->raw()["Predictive"]["StudyBetaParameters"], true, __FILE__, __LINE__);
 
  TDirectory* PredictiveDir = outputFile->mkdir("Predictive");
  std::vector<TDirectory*> SampleDirectories;
  SampleDirectories.resize(TotalNumberOfSamples+1);
 
  // open directory for every sample
  for (int sample = 0; sample < TotalNumberOfSamples+1; ++sample) {
    SampleDirectories[sample] = PredictiveDir->mkdir(SampleInfo[sample].Name.c_str());
  }
 
  // Produce Violin style spectra
  Study1DProjections(SampleDirectories);
  // Produce posterior predictive distribution for mc
  auto PostPred_mc = MakePredictive(MC_Hist_Toy, SampleDirectories, "mc", DebugHistograms, false);
  // Produce posterior predictive distribution for w2
  auto PostPred_w2 = MakePredictive(W2_Hist_Toy, SampleDirectories, "w2", false, false);
  // Calculate Posterior Predictive LLH
  PredictiveLLH(Data_Hist, PostPred_mc, PostPred_w2, SampleDirectories);
  // Calculate Posterior Predictive $p$-value
  PosteriorPredictivepValue(PostPred_mc, SampleDirectories);
  // Check how number of events changed
  RateAnalysis(MC_Hist_Toy, SampleDirectories);
 
  // Studying information criterion
  StudyInformationCriterion(M3::kWAIC, PostPred_mc, PostPred_w2);
 
  // Close directories
  for (int sample = 0; sample < TotalNumberOfSamples+1; ++sample) {
    SampleDirectories[sample]->Close();
    delete SampleDirectories[sample];
  }
  // Perform beta analysis for mc statical uncertainty
  if(StudyBeta) StudyBetaParameters(PredictiveDir);
 
  PredictiveDir->Close();
  delete PredictiveDir;
 
  outputFile->cd();
 
  TempClock.Stop();
  MACH3LOG_INFO("{} took {:.2f}s to finish for {} toys", __func__, TempClock.RealTime(), Ntoys);
}

SetParamters()

void PredictiveThrower::SetParamters ( std::vector< std::string > &  ParameterGroupsNotVaried, std::unordered_set< int > &  ParameterOnlyToVary  )

private

This set some params to prior value this way you can evaluate errors from subset of errors.

Have ability to not throw legacy matrices

Alternatively vary only selected params

Definition at line 43 of file PredictiveThrower.cpp.

                                                                                 {
// *************************
  // WARNING This should be removed in the future
  auto DoNotThrowLegacyCov = GetFromManager<std::vector<std::string>>(fitMan->raw()["Predictive"]["DoNotThrowLegacyCov"], {}, __FILE__, __LINE__);
  for (size_t i = 0; i < DoNotThrowLegacyCov.size(); ++i) {
    for (size_t s = 0; s < systematics.size(); ++s) {
      if (systematics[s]->GetName() == DoNotThrowLegacyCov[i]) {
        systematics[s]->SetParameters();
        break;
      }
    }
  }
 
  // Set groups to prefit values if they were set to not be varies
  if(ModelSystematic && ParameterGroupsNotVaried.size() > 0) {
    ModelSystematic->SetGroupOnlyParameters(ParameterGroupsNotVaried);
  }
 
  if (ModelSystematic && !ParameterOnlyToVary.empty()) {
    for (int i = 0; i < ModelSystematic->GetNumParams(); ++i) {
      // KS: If parameter is in map then we are skipping this, otherwise for params that we don't want to vary we simply set it to prior
      if (ParameterOnlyToVary.find(i) == ParameterOnlyToVary.end()) {
        ModelSystematic->SetParProp(i, ModelSystematic->GetParInit(i));
      }
    }
  }
}

SetupSampleInformation()

void PredictiveThrower::SetupSampleInformation ( )

private

Setup sample information.

Definition at line 75 of file PredictiveThrower.cpp.

                                               {
// *************************
  TotalNumberOfSamples = 0;
  for (size_t iPDF = 0; iPDF < samples.size(); iPDF++) {
    TotalNumberOfSamples += samples[iPDF]->GetNSamples();
  }
 
  MC_Hist_Toy.resize(TotalNumberOfSamples);
  W2_Hist_Toy.resize(TotalNumberOfSamples);
  Data_Hist.resize(TotalNumberOfSamples);
  MC_Nom_Hist.resize(TotalNumberOfSamples);
  W2_Nom_Hist.resize(TotalNumberOfSamples);
 
  SampleInfo.resize(TotalNumberOfSamples+1);
 
 
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    MC_Hist_Toy[sample].resize(Ntoys);
    W2_Hist_Toy[sample].resize(Ntoys);
  }
  int counter = 0;
  for (size_t iPDF = 0; iPDF < samples.size(); iPDF++) {
    for (int SampleIndex = 0; SampleIndex < samples[iPDF]->GetNSamples(); ++SampleIndex) {
      SampleInfo[counter].Name = samples[iPDF]->GetSampleTitle(SampleIndex);
      SampleInfo[counter].LocalId = SampleIndex;
      SampleInfo[counter].SamHandler = samples[iPDF];
      SampleInfo[counter].Dimenstion = SampleInfo[counter].SamHandler->GetNDim(SampleIndex);
      counter++;
    }
  }
  SampleInfo[TotalNumberOfSamples].Name= "Total";
}

SetupToyGeneration()

void PredictiveThrower::SetupToyGeneration ( std::vector< std::string > &  ParameterGroupsNotVaried, std::unordered_set< int > &  ParameterOnlyToVary, std::vector< const M3::float_t * > &  BoundValuePointer, std::vector< std::pair< double, double >> &  ParamBounds  )

private

Setup useful variables etc before stating toy generation.

Let's ask the manager what are the file with covariance matrix

Definition at line 110 of file PredictiveThrower.cpp.

                                                                                            {
// *************************
  int counter = 0;
  for (size_t s = 0; s < systematics.size(); ++s) {
    auto* MaCh3Params = dynamic_cast<ParameterHandlerGeneric*>(systematics[s]);
    if(MaCh3Params) {
      ModelSystematic = MaCh3Params;
      counter++;
    }
  }
 
  SetupSampleInformation();
 
  if(Is_PriorPredictive) {
    MACH3LOG_INFO("You've chosen to run Prior Predictive Distribution");
  } else {
    auto PosteriorFileName = Get<std::string>(fitMan->raw()["Predictive"]["PosteriorFile"], __FILE__, __LINE__);
    //KS: We use MCMCProcessor to get names of covariances that were actually used to produce given chain
    MCMCProcessor Processor(PosteriorFileName);
    Processor.Initialise();
 
    // For throwing FD predictions from ND-only chain we have to allow having different yaml configs
    auto AllowDifferentConfigs = GetFromManager<bool>(fitMan->raw()["Predictive"]["AllowDifferentConfigs"], false, __FILE__, __LINE__);
 
    YAML::Node ConfigInChain = Processor.GetCovConfig(kXSecPar);
    if(ModelSystematic) {
      YAML::Node ConfigNow = ModelSystematic->GetConfig();
      if (!compareYAMLNodes(ConfigNow, ConfigInChain))
      {
        if(AllowDifferentConfigs){
          MACH3LOG_WARN("Yaml configs used for your ParameterHandler for chain you want sample from ({}) and one currently initialised are different", PosteriorFileName);
        } else {
          MACH3LOG_ERROR("Yaml configs used for your ParameterHandler for chain you want sample from ({}) and one currently initialised are different", PosteriorFileName);
          throw MaCh3Exception(__FILE__ , __LINE__ );
        }
      }
    }
  }
  if(counter > 1) {
    MACH3LOG_ERROR("Found {} ParmaterHandler inheriting from ParameterHandlerGeneric, I can accept at most 1", counter);
    throw MaCh3Exception(__FILE__, __LINE__);
  }
 
  for (size_t s = 0; s < systematics.size(); ++s) {
    NModelParams += systematics[s]->GetNumParams();
  }
 
  if (ModelSystematic) {
    auto ThrowParamGroupOnly = GetFromManager<std::vector<std::string>>(fitMan->raw()["Predictive"]["ThrowParamGroupOnly"], {}, __FILE__, __LINE__);
    auto UniqueParamGroup = ModelSystematic->GetUniqueParameterGroups();
    auto ParameterOnlyToVaryString = GetFromManager<std::vector<std::string>>(fitMan->raw()["Predictive"]["ThrowSinlgeParams"], {}, __FILE__, __LINE__);
 
    if (!ThrowParamGroupOnly.empty() && !ParameterOnlyToVaryString.empty()) {
      MACH3LOG_ERROR("Can't use ThrowParamGroupOnly and ThrowSinlgeParams at the same time");
      throw MaCh3Exception(__FILE__, __LINE__);
    }
 
    if (!ParameterOnlyToVaryString.empty()) {
      MACH3LOG_INFO("I will throw only: {}", fmt::join(ParameterOnlyToVaryString, ", "));
      std::vector<int> ParameterVary(ParameterOnlyToVaryString.size());
 
      for (size_t i = 0; i < ParameterOnlyToVaryString.size(); ++i) {
        ParameterVary[i] = ModelSystematic->GetParIndex(ParameterOnlyToVaryString[i]);
        if (ParameterVary[i] == M3::_BAD_INT_) {
          MACH3LOG_ERROR("Can't proceed if param {} is missing", ParameterOnlyToVaryString[i]);
          throw MaCh3Exception(__FILE__, __LINE__);
        }
      }
      ParameterOnlyToVary = std::unordered_set<int>(ParameterVary.begin(), ParameterVary.end());
    } else {
      MACH3LOG_INFO("I have following parameter groups: {}", fmt::join(UniqueParamGroup, ", "));
      if (ThrowParamGroupOnly.empty()) {
        MACH3LOG_INFO("I will vary all");
      } else {
        std::unordered_set<std::string> throwOnlySet(ThrowParamGroupOnly.begin(), ThrowParamGroupOnly.end());
        ParameterGroupsNotVaried.clear();
 
        for (const auto& group : UniqueParamGroup) {
          if (throwOnlySet.find(group) == throwOnlySet.end()) {
            ParameterGroupsNotVaried.push_back(group);
          }
        }
 
        MACH3LOG_INFO("I will vary: {}", fmt::join(ThrowParamGroupOnly, ", "));
        MACH3LOG_INFO("Exclude: {}", fmt::join(ParameterGroupsNotVaried, ", "));
      }
    }
  }
 
  auto paramNode = fitMan->raw()["Predictive"]["ParameterBounds"];
  for (const auto& p : paramNode) {
    // Extract name
    std::string name = p[0].as<std::string>();
 
    // Extract bounds: min and max
    double minVal = p[1][0].as<double>();
    double maxVal = p[1][1].as<double>();
    ParamBounds.emplace_back(minVal, maxVal);
 
    for (size_t s = 0; s < systematics.size(); ++s) {
      for(int iPar = 0; iPar < systematics[s]->GetNParameters(); iPar++){
        if(systematics[s]->GetParFancyName(iPar) == name){
          BoundValuePointer.push_back(systematics[s]->RetPointer(iPar));
          break;
        }
      }
    }
    if(ParamBounds.size() != BoundValuePointer.size()){
      MACH3LOG_ERROR("Ddin't find paramter {}", name);
      throw MaCh3Exception(__FILE__,__LINE__);
    }
    MACH3LOG_INFO("Parameter: {} with : [{}, {}]", name, minVal, maxVal);
  }
  if(Is_PriorPredictive && ParamBounds.size() > 0) {
    MACH3LOG_ERROR("Additional bounds not supported by prior predictive right now");
    throw MaCh3Exception(__FILE__,__LINE__);
  }
}

Study1DProjections()

void PredictiveThrower::Study1DProjections ( const std::vector< TDirectory * > &  SampleDirectories ) const

private

Load 1D projections and later produce violin plots for each.

Definition at line 638 of file PredictiveThrower.cpp.

                                                                                                {
// *************************
  MACH3LOG_INFO("Starting {}", __func__);
 
  TDirectory * ogdir = gDirectory;
  auto PosteriorFileName = Get<std::string>(fitMan->raw()["Predictive"]["PosteriorFile"], __FILE__, __LINE__);
  // Open the ROOT file
  int originalErrorWarning = gErrorIgnoreLevel;
  gErrorIgnoreLevel = kFatal;
 
  TFile* file = TFile::Open(PosteriorFileName.c_str(), "READ");
 
  gErrorIgnoreLevel = originalErrorWarning;
  TDirectory* ToyDir = file->GetDirectory("Toys_1DHistVar");
  // If toys not amiable in posterior file this means they must be in output file
  if(ToyDir == nullptr) {
    ToyDir = outputFile->GetDirectory("Toys_1DHistVar");
  }
  // [sample], [toy], [dim]
  std::vector<std::vector<std::vector<std::unique_ptr<TH1D>>>> ProjectionToys(TotalNumberOfSamples);
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    ProjectionToys[sample].resize(Ntoys);
    const int nDims = SampleInfo[sample].Dimenstion;
    for (int iToy = 0; iToy < Ntoys; ++iToy) {
      ProjectionToys[sample][iToy].resize(nDims);
    }
  }
 
  for (int iToy = 0; iToy < Ntoys; ++iToy) {
    if (iToy % 100 == 0) MACH3LOG_INFO("   Loaded Projection toys {}", iToy);
    for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
      const int nDims = SampleInfo[sample].Dimenstion;
      for(int iDim = 0; iDim < nDims; iDim ++){
        std::string ProjectionSuffix = "_1DProj" + std::to_string(iDim) + "_" + std::to_string(iToy);
        TH1D* MCHist1D = static_cast<TH1D*>(ToyDir->Get((SampleInfo[sample].Name + ProjectionSuffix).c_str()));
        ProjectionToys[sample][iToy][iDim] = M3::Clone(MCHist1D);
      }
    } // end loop over samples
  } // end loop over toys
  file->Close(); delete file;
  if(ogdir){ ogdir->cd(); }
 
  ProduceSpectra(ProjectionToys, SampleDirectories, "mc");
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    const int nDims = SampleInfo[sample].Dimenstion;
    // KS: We only care about doing projections for 2D, for 1D we have well 1D, for beyond 2D we have flattened TH1D
    if(nDims == 2){
      auto hist = Data_Hist[sample].get();
      SampleDirectories[sample]->cd();
 
      std::string nameX = "Data_" + SampleInfo[sample].Name + "_Dim0";
      std::string nameY = "Data_" + SampleInfo[sample].Name + "_Dim1";
 
      if(std::string(hist->ClassName()) == "TH2Poly") {
        TAxis* xax = ProjectionToys[sample][0][0]->GetXaxis();
        TAxis* yax = ProjectionToys[sample][0][1]->GetXaxis();
 
        std::vector<double> XBinning(xax->GetNbins()+1);
        std::vector<double> YBinning(yax->GetNbins()+1);
 
        for(int i=0;i<=xax->GetNbins();++i)
          XBinning[i] = xax->GetBinLowEdge(i+1);
 
        for(int i=0;i<=yax->GetNbins();++i)
          YBinning[i] = yax->GetBinLowEdge(i+1);
 
        TH1D* ProjectionX = PolyProjectionX(static_cast<TH2Poly*>(hist), nameX.c_str(), XBinning, false);
        TH1D* ProjectionY = PolyProjectionY(static_cast<TH2Poly*>(hist), nameY.c_str(), YBinning, false);
 
        ProjectionX->SetDirectory(nullptr);
        ProjectionY->SetDirectory(nullptr);
 
        ProjectionX->Write(nameX.c_str());
        ProjectionY->Write(nameY.c_str());
 
        delete ProjectionX;
        delete ProjectionY;
      } else { //TH2D
        TH1D* ProjectionX = static_cast<TH2D*>(hist)->ProjectionX(nameX.c_str());
        TH1D* ProjectionY = static_cast<TH2D*>(hist)->ProjectionY(nameY.c_str());
 
        ProjectionX->SetDirectory(nullptr);
        ProjectionY->SetDirectory(nullptr);
 
        ProjectionX->Write(nameX.c_str());
        ProjectionY->Write(nameY.c_str());
        delete ProjectionX;
        delete ProjectionY;
      }
    }
  }
}

StudyBetaParameters()

void PredictiveThrower::StudyBetaParameters ( TDirectory *  PredictiveDir )

private

Evaluate prior/post predictive distribution for beta parameters (used for evaluating impact MC statistical uncertainty)

1. Initialise Beta histogram
2. Fill histograms, thread safe as all histograms are allocated before and we loop over samples

ROOT enumerates from 1 while MaCh3 from 0

1. Write to file

Definition at line 1337 of file PredictiveThrower.cpp.

                                                                     {
// *************************
  bool StudyBeta = GetFromManager<bool>(fitMan->raw()["Predictive"]["StudyBetaParameters"], true, __FILE__, __LINE__ );
  if (StudyBeta == false) return;
 
  MACH3LOG_INFO("Starting {}", __func__);
  TDirectory* BetaDir = PredictiveDir->mkdir("BetaParameters");
  std::vector<std::vector<std::unique_ptr<TH1D>>> BetaHist(TotalNumberOfSamples);
  std::vector<TDirectory *> DirBeta(TotalNumberOfSamples);
  // initialise directory for each sample
  for (int sample = 0; sample < TotalNumberOfSamples; ++sample) {
    BetaDir->cd();
    DirBeta[sample] = BetaDir->mkdir(SampleInfo[sample].Name.c_str());
  }
 
  for (int iSample = 0; iSample < TotalNumberOfSamples; ++iSample) {
    const int nDims = SampleInfo[iSample].Dimenstion;
    // Use any histogram that defines the binning structure
    TH1* RefHist = Data_Hist[iSample].get();
    BetaHist[iSample] = PerBinHistogram(RefHist, iSample, nDims, "Beta_Parameter");
    // Change x-axis title
    for (size_t i = 0; i < BetaHist[iSample].size(); ++i) {
      BetaHist[iSample][i]->GetXaxis()->SetTitle("beta parameter");
    }
  }
 
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int iSample = 0; iSample < TotalNumberOfSamples; ++iSample) {
    const int nDims = SampleInfo[iSample].Dimenstion;
    const auto likelihood = SampleInfo[iSample].SamHandler->GetTestStatistic();
    for (int iToy = 0; iToy < Ntoys; ++iToy) {
      if (nDims == 2) {
        if(std::string(Data_Hist[iSample]->ClassName()) == "TH2Poly") {
          for (int i = 1; i <= static_cast<TH2Poly*>(Data_Hist[iSample].get())->GetNumberOfBins(); ++i) {
            const double Data = Data_Hist[iSample]->GetBinContent(i);
            const double MC   = MC_Hist_Toy[iSample][iToy]->GetBinContent(i);
            const double w2   = W2_Hist_Toy[iSample][iToy]->GetBinContent(i);
 
            const double BetaParam = GetBetaParameter(Data, MC, w2, likelihood);
            BetaHist[iSample][i-1]->Fill(BetaParam, ReweightWeight[iToy]);
          } // end loop over poly bins
        } else {
          const int nX = Data_Hist[iSample]->GetNbinsX();
          const int nY = Data_Hist[iSample]->GetNbinsY();
          for (int iy = 1; iy <= nY; ++iy) {
            for (int ix = 1; ix <= nX; ++ix) {
              const int FlatBin = (iy-1) * nX + (ix-1);
 
              const double Data = Data_Hist[iSample]->GetBinContent(ix, iy);
              const double MC   = MC_Hist_Toy[iSample][iToy]->GetBinContent(ix, iy);
              const double w2   = W2_Hist_Toy[iSample][iToy]->GetBinContent(ix, iy);
 
              const double BetaParam = GetBetaParameter(Data, MC, w2, likelihood);
              BetaHist[iSample][FlatBin]->Fill(BetaParam, ReweightWeight[iToy]);
            }
          } // end loop over x
        } // end loop over y
      } else {
        int nbinsx = Data_Hist[iSample]->GetNbinsX();
        for (int ix = 1; ix <= nbinsx; ++ix) {
          const double Data = Data_Hist[iSample]->GetBinContent(ix);
          const double MC = MC_Hist_Toy[iSample][iToy]->GetBinContent(ix);
          const double w2 = W2_Hist_Toy[iSample][iToy]->GetBinContent(ix);
 
          const double BetaParam = GetBetaParameter(Data, MC, w2, likelihood);
          BetaHist[iSample][ix-1]->Fill(BetaParam, ReweightWeight[iToy]);
        } // end loop over bins
      }
    } // end loop over toys
  } // end loop over samples
 
  for (int iSample = 0; iSample < TotalNumberOfSamples; ++iSample) {
    for (size_t iBin = 0; iBin < BetaHist[iSample].size(); iBin++) {
      DirBeta[iSample]->cd();
      BetaHist[iSample][iBin]->Write();
    }
    DirBeta[iSample]->Close();
    delete DirBeta[iSample];
  }
  BetaDir->Close();
  delete BetaDir;
 
  PredictiveDir->cd();
}

StudyBIC()

void PredictiveThrower::StudyBIC ( const std::vector< std::unique_ptr< TH1 >> &  PostPred_mc, const std::vector< std::unique_ptr< TH1 >> &  PostPred_w  )

private

Study Bayesian Information Criterion (BIC) The BIC is defined as:

\[ \mathrm{BIC} = -2 \log L + k \log(n) \]

where:

• \(L\) is the likelihood evaluated at the fitted model
• \(k\) is the number of model parameters
• \(n\) is the number of observations

[13]

Definition at line 1584 of file PredictiveThrower.cpp.

                                                                                  {
// ****************
  //make fancy event rate histogram
  double DataRate = 0.0;
  double BinsRate = 0.0;
  double TotalLLH = 0.0;
  #ifdef MULTITHREAD
  #pragma omp parallel for reduction(+:DataRate, BinsRate, TotalLLH)
  #endif
  for (int i = 0; i < TotalNumberOfSamples; ++i)
  {
    auto SampleHandler = SampleInfo[i].SamHandler;
    auto* h = Data_Hist[i].get();
    DataRate += h->Integral();
    if (auto h1 = dynamic_cast<TH1D*>(h)) {
      BinsRate += h1->GetNbinsX();
    } else if (auto h2 = dynamic_cast<TH2D*>(h)) {
      BinsRate += h2->GetNbinsX() * h2->GetNbinsY();
    } else if (auto h2poly = dynamic_cast<TH2Poly*>(h)) {
      BinsRate += h2poly->GetNumberOfBins();
    } else {
      MACH3LOG_WARN("Unknown histogram type in DataHist[{}]", i);
    }
    TotalLLH += CalcLLH(Data_Hist[i].get(), PostPred_mc[i].get(), PostPred_w[i].get(), SampleHandler);
  }
 
  const double EventRateBIC = GetBIC(TotalLLH, DataRate, NModelParams);
  const double BinBasedBIC = GetBIC(TotalLLH, BinsRate, NModelParams);
  MACH3LOG_INFO("Calculated Bayesian Information Criterion using global number of events: {:.2f}", EventRateBIC);
  MACH3LOG_INFO("Calculated Bayesian Information Criterion using global number of bins: {:.2f}", BinBasedBIC);
  MACH3LOG_INFO("Additional info: NModelParams: {}, DataRate: {:.2f}, BinsRate: {:.2f}", NModelParams, DataRate, BinsRate);
}

StudyDIC()

void PredictiveThrower::StudyDIC ( const std::vector< std::unique_ptr< TH1 >> &  PostPred_mc, const std::vector< std::unique_ptr< TH1 >> &  PostPred_w  )

private

KS: Get the Deviance Information Criterion (DIC) The deviance is defined as:

\[ D(\theta) = -2 \log L(\theta) \]

The DIC statistic is then:

\[ \mathrm{DIC} = D(\hat{\theta}) + 2p_D \]

where:

\[ p_D = \bar{D} - D(\hat{\theta}) \]

is the effective number of parameters, and

\[ \bar{D} = E_{\theta|y}[D(\theta)] \]

[32] [34]

Definition at line 1620 of file PredictiveThrower.cpp.

                                                                                  {
// ****************
  //The posterior mean of the deviance
  double Dbar = 0.;
  double TotalLLH = 0.0;
 
  #ifdef MULTITHREAD
  #pragma omp parallel for reduction(+:Dbar)
  #endif
  for (int iSample = 0; iSample < TotalNumberOfSamples; ++iSample)
  {
    auto SampleHandler = SampleInfo[iSample].SamHandler;
    TotalLLH += CalcLLH(Data_Hist[iSample].get(), PostPred_mc[iSample].get(), PostPred_w[iSample].get(), SampleHandler);
    double LLH_temp = 0.;
    for (int iToy = 0; iToy < Ntoys; ++iToy)
    {
      LLH_temp += CalcLLH(Data_Hist[iSample].get(), MC_Hist_Toy[iSample][iToy].get(), W2_Hist_Toy[iSample][iToy].get(), SampleHandler);
    }
    Dbar += LLH_temp;
  }
  Dbar = Dbar / Ntoys;
 
  // A point estimate of the deviance
  const double Dhat = TotalLLH;
 
  //Effective number of parameters
  const double p_D = std::fabs(Dbar - Dhat);
 
  //Actual test stat
  const double DIC_stat = Dhat + 2 * p_D;
  MACH3LOG_INFO("Effective number of parameters following DIC formalism is equal to: {:.2f}", p_D);
  MACH3LOG_INFO("DIC test statistic = {:.2f}", DIC_stat);
}

StudyInformationCriterion()

void PredictiveThrower::StudyInformationCriterion ( M3::kInfCrit  Criterion, const std::vector< std::unique_ptr< TH1 >> &  PostPred_mc, const std::vector< std::unique_ptr< TH1 >> &  PostPred_w  )

private

Information Criterion.

Definition at line 1554 of file PredictiveThrower.cpp.

                                                                                                   {
// ****************
  MACH3LOG_INFO("******************************");
  switch(Criterion) {
    case M3::kInfCrit::kBIC:
      // Study Bayesian Information Criterion
      StudyBIC(PostPred_mc, PostPred_w);
      break;
    case M3::kInfCrit::kDIC:
      // Study Deviance Information Criterion
      StudyDIC(PostPred_mc, PostPred_w);
      break;
    case M3::kInfCrit::kWAIC:
      // Study Watanabe-Akaike information criterion (WAIC)
      StudyWAIC();
      break;
    case M3::kInfCrit::kInfCrits:
      MACH3LOG_ERROR("kInfCrits is not a valid kInfCrit!");
      throw MaCh3Exception(__FILE__, __LINE__);
    default:
      MACH3LOG_ERROR("UNKNOWN Information Criterion SPECIFIED!");
      MACH3LOG_ERROR("You gave {}", static_cast<int>(Criterion));
      throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  MACH3LOG_INFO("******************************");
}

StudyWAIC()

void PredictiveThrower::StudyWAIC ( )

private

KS: Get the Watanabe-Akaike information criterion (WAIC)

WAIC is a fully Bayesian measure of model fit that estimates the predictive accuracy, taking into account the effective number of parameters in the model.

It is defined as:

\[ \text{WAIC} = -2 \left( \text{lppd} - p_\text{WAIC} \right), \]

where

\[ \text{lppd} = \sum_{i=1}^n \log \left( \frac{1}{S} \sum_{s=1}^S p(y_i \mid \theta_s) \right) \]

is the log pointwise predictive density and

\[ p_\text{WAIC} = \sum_{i=1}^n \text{Var}_{s=1,\dots,S} \left[ \log p(y_i \mid \theta_s) \right] \]

is the effective number of parameters (variance of log-likelihood over posterior samples).

[13] [18]

Definition at line 1691 of file PredictiveThrower.cpp.

                                  {
// ****************
  // log pointwise predictive density
  double lppd = 0.;
  // effective number of parameters
  double p_WAIC = 0.;
 
  #ifdef MULTITHREAD
  #pragma omp parallel for reduction(+:lppd, p_WAIC)
  #endif
  for (int iSample = 0; iSample < TotalNumberOfSamples; ++iSample) {
    auto SampleHandler = SampleInfo[iSample].SamHandler;
    auto* hData = Data_Hist[iSample].get();
 
    if (auto h2poly = dynamic_cast<TH2Poly*>(hData)) {
      // TH2Poly: irregular bins, linear indexing
      for (int i = 1; i <= h2poly->GetNumberOfBins(); ++i) {
        const double data = Data_Hist[iSample]->GetBinContent(i);
        double mean_llh = 0.;
        double sum_exp_llh = 0;
        double mean_llh_squared = 0.;
 
        for (int iToy = 0; iToy < Ntoys; ++iToy) {
          const double mc = MC_Hist_Toy[iSample][iToy]->GetBinContent(i);
          const double w2 = W2_Hist_Toy[iSample][iToy]->GetBinContent(i);
          // Get the -log-likelihood for this sample and bin
          double neg_LLH_temp = SampleHandler->GetTestStatLLH(data, mc, w2);
          AccumulateWAICToy(neg_LLH_temp, mean_llh, mean_llh_squared, sum_exp_llh);
        }
        AccumulateWAICBin(mean_llh, mean_llh_squared, sum_exp_llh, Ntoys, lppd, p_WAIC);
      }
    } else if (auto h2 = dynamic_cast<TH2D*>(hData)) {
      // TH2D: nested loops over X and Y
      for (int ix = 1; ix <= h2->GetNbinsX(); ++ix) {
        for (int iy = 1; iy <= h2->GetNbinsY(); ++iy) {
          const double data = hData->GetBinContent(ix, iy);
          double mean_llh = 0.;
          double mean_llh_squared = 0.;
          double sum_exp_llh = 0.;
          for (int iToy = 0; iToy < Ntoys; ++iToy) {
            const double mc = MC_Hist_Toy[iSample][iToy]->GetBinContent(ix, iy);
            const double w2 = W2_Hist_Toy[iSample][iToy]->GetBinContent(ix, iy);
            // Get the -log-likelihood for this sample and bin
            double neg_LLH_temp = SampleHandler->GetTestStatLLH(data, mc, w2);
            AccumulateWAICToy(neg_LLH_temp, mean_llh, mean_llh_squared, sum_exp_llh);
          }
          AccumulateWAICBin(mean_llh, mean_llh_squared, sum_exp_llh, Ntoys, lppd, p_WAIC);
        }
      }
    } else if (auto h1 = dynamic_cast<TH1D*>(hData)) {
      // TH1D: 1D histogram
      for (int iBin = 1; iBin <= h1->GetNbinsX(); ++iBin) {
        const double data = hData->GetBinContent(iBin);
        double mean_llh = 0.;
        double mean_llh_squared = 0.;
        double sum_exp_llh = 0.;
        for (int iToy = 0; iToy < Ntoys; ++iToy) {
          const double mc = MC_Hist_Toy[iSample][iToy]->GetBinContent(iBin);
          const double w2 = W2_Hist_Toy[iSample][iToy]->GetBinContent(iBin);
 
          // Get the -log-likelihood for this sample and bin
          double neg_LLH_temp = SampleHandler->GetTestStatLLH(data, mc, w2);
          AccumulateWAICToy(neg_LLH_temp, mean_llh, mean_llh_squared, sum_exp_llh);
        }
        AccumulateWAICBin(mean_llh, mean_llh_squared, sum_exp_llh, Ntoys, lppd, p_WAIC);
      }
    }
  }
 
  // Compute WAIC, see Eq. 13 in Gelman2014
  double WAIC = -2 * (lppd - p_WAIC);
  MACH3LOG_INFO("Effective number of parameters following WAIC formalism is equal to: {:.2f}", p_WAIC);
  MACH3LOG_INFO("WAIC = {:.2f}", WAIC);
}

WriteToy()

void PredictiveThrower::WriteToy ( TDirectory *  ToyDirectory, TDirectory *  Toy_1DDirectory, TDirectory *  Toy_2DDirectory, const int  iToy  )

private

Save histograms for a single MCMC Throw/Toy.

Definition at line 365 of file PredictiveThrower.cpp.

                                                 {
// *************************
  int SampleCounter = 0;
  for (size_t iPDF = 0; iPDF < samples.size(); iPDF++)
  {
    auto* SampleHandler = samples[iPDF];
    for (int iSample = 0; iSample < SampleHandler->GetNSamples(); ++iSample)
    {
      ToyDirectory->cd();
 
      auto SampleName = SampleHandler->GetSampleTitle(iSample);
      const TH1* MCHist = SampleHandler->GetMCHist(iSample);
      MC_Hist_Toy[SampleCounter][iToy] = M3::Clone(MCHist, SampleName + "_mc_" + std::to_string(iToy));
      MC_Hist_Toy[SampleCounter][iToy]->Write();
 
      const TH1* W2Hist = SampleHandler->GetW2Hist(iSample);
      W2_Hist_Toy[SampleCounter][iToy] = M3::Clone(W2Hist, SampleName + "_w2_" + std::to_string(iToy));
      W2_Hist_Toy[SampleCounter][iToy]->Write();
 
      // now get 1D projection for every dimension
      Toy_1DDirectory->cd();
      for(int iDim = 0; iDim < SampleHandler->GetNDim(iSample); iDim++) {
        std::string ProjectionName = SampleHandler->GetKinVarName(iSample, iDim);
        std::string ProjectionSuffix = "_1DProj" + std::to_string(iDim) + "_" + std::to_string(iToy);
 
        auto hist = SampleHandler->Get1DVarHist(iSample, ProjectionName);
        hist->SetTitle((SampleName + ProjectionSuffix).c_str());
        hist->SetName((SampleName + ProjectionSuffix).c_str());
        hist->Write();
      }
 
      Toy_2DDirectory->cd();
      // now get 2D projection for every combination
      for(int iDim1 = 0; iDim1 < SampleHandler->GetNDim(iSample); iDim1++) {
        for (int iDim2 = iDim1 + 1; iDim2 < SampleHandler->GetNDim(iSample); ++iDim2) {
          // Get the names for the two dimensions
          std::string XVarName = SampleHandler->GetKinVarName(iSample, iDim1);
          std::string YVarName = SampleHandler->GetKinVarName(iSample, iDim2);
 
          // Get the 2D histogram for this pair
          auto hist2D = SampleHandler->Get2DVarHist(iSample, XVarName, YVarName);
 
          // Write the histogram
          std::string suffix2D = "_2DProj_" + std::to_string(iDim1) + "_vs_" + std::to_string(iDim2) + "_" + std::to_string(iToy);
          hist2D->SetTitle((SampleName + suffix2D).c_str());
          hist2D->SetName((SampleName + suffix2D).c_str());
          hist2D->Write();
        }
      }
      SampleCounter++;
    }
  }
}

Member Data Documentation

Data_Hist

std::vector<std::unique_ptr<TH1> > PredictiveThrower::Data_Hist

private

Vector of Data histograms.

Definition at line 302 of file PredictiveThrower.h.

FullLLH

bool PredictiveThrower::FullLLH

private

KS: Use Full LLH or only sample contribution based on discussion with Asher we almost always only want the sample likelihood.

Definition at line 283 of file PredictiveThrower.h.

Is_PriorPredictive

bool PredictiveThrower::Is_PriorPredictive

private

Whether it is Prior or Posterior predictive.

Definition at line 287 of file PredictiveThrower.h.

MC_Hist_Toy

std::vector<std::vector<std::unique_ptr<TH1> > > PredictiveThrower::MC_Hist_Toy

private

Vector of MC histograms per sample and toy experiment. Indexed as [sample][toy].

Definition at line 310 of file PredictiveThrower.h.

MC_Nom_Hist

std::vector<std::unique_ptr<TH1> > PredictiveThrower::MC_Nom_Hist

private

Vector of MC histograms.

Definition at line 304 of file PredictiveThrower.h.

ModelSystematic

ParameterHandlerGeneric* PredictiveThrower::ModelSystematic

private

Pointer to El Generico.

Definition at line 299 of file PredictiveThrower.h.

NModelParams

int PredictiveThrower::NModelParams

private

KS: Count total number of model parameters which can be used for stuff like BIC.

Definition at line 285 of file PredictiveThrower.h.

Ntoys

int PredictiveThrower::Ntoys

private

Number of toys we are generating analysing.

Definition at line 296 of file PredictiveThrower.h.

PenaltyTerm

std::vector<double> PredictiveThrower::PenaltyTerm

private

Penalty term values for each toy by default 0.

Definition at line 318 of file PredictiveThrower.h.

ReweightWeight

std::vector<double> PredictiveThrower::ReweightWeight

private

Reweighting factors applied for each toy, by default 1.

Definition at line 316 of file PredictiveThrower.h.

SampleInfo

std::vector<PredictiveSample> PredictiveThrower::SampleInfo

private

Handy struct for all sample info.

Definition at line 293 of file PredictiveThrower.h.

StandardFluctuation

bool PredictiveThrower::StandardFluctuation

private

KS: We have two methods for Poissonian fluctuation.

Definition at line 321 of file PredictiveThrower.h.

TotalNumberOfSamples

int PredictiveThrower::TotalNumberOfSamples

private

Number of toys we are generating analysing.

Definition at line 290 of file PredictiveThrower.h.

W2_Hist_Toy

std::vector<std::vector<std::unique_ptr<TH1> > > PredictiveThrower::W2_Hist_Toy

private

Vector of W² histograms per sample and toy experiment. Indexed as [sample][toy]

Definition at line 313 of file PredictiveThrower.h.

W2_Nom_Hist

std::vector<std::unique_ptr<TH1> > PredictiveThrower::W2_Nom_Hist

private

Vector of W2 histograms.

Definition at line 306 of file PredictiveThrower.h.

---

The documentation for this class was generated from the following files:

• Fitters/PredictiveThrower.h
• Fitters/PredictiveThrower.cpp