MCMCProcessor Class Reference

Class responsible for processing MCMC chains, performing diagnostics, generating plots, and managing Bayesian analysis. More...

#include <Fitters/MCMCProcessor.h>

Inheritance diagram for MCMCProcessor:

Collaboration diagram for MCMCProcessor:

Public Member Functions
	MCMCProcessor (const std::string &InputFile)
	Constructs an MCMCProcessor object with the specified input file and options. More...
virtual	~MCMCProcessor ()
	Destroys the MCMCProcessor object. More...
void	Initialise ()
	Scan chain, what parameters we have and load information from covariance matrices. More...
void	MakePostfit (const std::map< std::string, std::pair< double, double >> &Edges={})
	Make 1D projection for each parameter and prepare structure. More...
void	MakeCovariance ()
	Calculate covariance by making 2D projection of each combination of parameters. More...
void	CacheSteps ()
	KS:By caching each step we use multithreading. More...
void	MakeCovariance_MP (const bool Mute=false)
	Calculate covariance by making 2D projection of each combination of parameters using multithreading. More...
void	MakeSubOptimality (const int NIntervals=10)
	Make and Draw SubOptimality [27]. More...
void	ResetHistograms ()
	Reset 2D posteriors, in case we would like to calculate in again with different BurnInCut. More...
void	DrawPostfit ()
	Draw the post-fit comparisons. More...
void	MakeViolin ()
	Make and Draw Violin. More...
void	MakeCredibleIntervals (const std::vector< double > &CredibleIntervals={0.99, 0.90, 0.68 }, const std::vector< Color_t > &CredibleIntervalsColours={kCyan+4, kCyan-2, kCyan-10}, const bool CredibleInSigmas=false)
	Make and Draw Credible intervals. More...
void	DrawCovariance ()
	Draw the post-fit covariances. More...
void	MakeCovarianceYAML (const std::string &OutputYAMLFile, const std::string &MeansMethod) const
	Make YAML file from post-fit covariance. More...
void	MakeCredibleRegions (const std::vector< double > &CredibleRegions={0.99, 0.90, 0.68}, const std::vector< Style_t > &CredibleRegionStyle={kDashed, kSolid, kDotted}, const std::vector< Color_t > &CredibleRegionColor={kGreen-3, kGreen-10, kGreen}, const bool CredibleInSigmas=false, const bool Draw2DPosterior=true, const bool DrawBestFit=true)
	Make and Draw Credible Regions. More...
void	MakeTrianglePlot (const std::vector< std::string > &ParNames, const std::vector< double > &CredibleIntervals={0.99, 0.90, 0.68 }, const std::vector< Color_t > &CredibleIntervalsColours={kCyan+4, kCyan-2, kCyan-10}, const std::vector< double > &CredibleRegions={0.99, 0.90, 0.68}, const std::vector< Style_t > &CredibleRegionStyle={kDashed, kSolid, kDotted}, const std::vector< Color_t > &CredibleRegionColor={kGreen-3, kGreen-10, kGreen}, const bool CredibleInSigmas=false)
	Make fancy triangle plot for selected parameters. More...
void	CheckCredibleIntervalsOrder (const std::vector< double > &CredibleIntervals, const std::vector< Color_t > &CredibleIntervalsColours) const
	Checks the order and size consistency of the CredibleIntervals and CredibleIntervalsColours vectors. More...
void	CheckCredibleRegionsOrder (const std::vector< double > &CredibleRegions, const std::vector< Style_t > &CredibleRegionStyle, const std::vector< Color_t > &CredibleRegionColor)
	Checks the order and size consistency of the CredibleRegions, CredibleRegionStyle, and CredibleRegionColor vectors. More...
void	GetPolarPlot (const std::vector< std::string > &ParNames)
	Make funny polar plot. More...
void	GetBayesFactor (const std::vector< std::string > &ParName, const std::vector< std::vector< double >> &Model1Bounds, const std::vector< std::vector< double >> &Model2Bounds, const std::vector< std::vector< std::string >> &ModelNames)
	Calculate Bayes factor for vector of params, and model boundaries. More...
void	GetSavageDickey (const std::vector< std::string > &ParName, const std::vector< double > &EvaluationPoint, const std::vector< std::vector< double >> &Bounds)
	Calculate Bayes factor for point like hypothesis using SavageDickey. More...
void	SavageDickeyPlot (std::unique_ptr< TH1D > &PriorHist, std::unique_ptr< TH1D > &PosteriorHist, const std::string &Title, const double EvaluationPoint) const
	Produce Savage Dickey plot. More...
void	ReweightPrior (const std::vector< std::string > &Names, const std::vector< double > &NewCentral, const std::vector< double > &NewError)
	Reweight Prior by giving new central value and new error. More...
void	SmearChain (const std::vector< std::string > &Names, const std::vector< double > &NewCentral, const bool &SaveBranch)
	Smear chain contours. More...
void	ParameterEvolution (const std::vector< std::string > &Names, const std::vector< int > &NIntervals)
	Make .gif of parameter evolution. More...
void	ThinMCMC (const int ThinningCut) const
	Thin MCMC Chain, to save space and maintain low autocorrelations. More...
void	DiagMCMC ()
	KS: Perform MCMC diagnostic including Autocorrelation, Trace etc. More...
int	GetNParams () const
	Get total number of used parameters. More...
int	GetNXSec () const
int	GetNND () const
int	GetNFD () const
YAML::Node	GetCovConfig (const int i) const
	Get Yaml config obtained from a Chain. More...
int	GetGroup (const std::string &name) const
	Number of params from a given group, for example flux. More...
TH1D *	GetHpost (const int i) const
	Get 1D posterior for a given parameter. More...
TH2D *	GetHpost2D (const int i, const int j) const
	Get 2D posterior for a given parameter combination. More...
TH2D *	GetViolin () const
	Get Violin plot for all parameters with posterior values. More...
TH2D *	GetViolinPrior () const
	Get Violin plot for all parameters with prior values. More...
std::vector< std::string >	GetXSecCov () const
std::string	GetNDCov () const
std::string	GetFDCov () const
void	GetPostfit (TVectorD *&Central, TVectorD *&Errors, TVectorD *&Central_Gauss, TVectorD *&Errors_Gauss, TVectorD *&Peaks)
	Get the post-fit results (arithmetic and Gaussian) More...
void	GetCovariance (TMatrixDSym *&Cov, TMatrixDSym *&Corr)
	Get the post-fit covariances and correlations. More...
void	GetPostfit_Ind (TVectorD *&Central, TVectorD *&Errors, TVectorD *&Peaks, ParameterEnum kParam)
	Or the individual post-fits. More...
const std::vector< TString > &	GetBranchNames () const
	Get the vector of branch names from root file. More...
void	GetNthParameter (const int param, double &Prior, double &PriorError, TString &Title) const
	Get properties of parameter by passing it number. More...
int	GetParamIndexFromName (const std::string &Name)
	Get parameter number based on name. More...
Long64_t	GetnEntries ()
	Get Number of entries that Chain has, for merged chains will not be the same Nsteps. More...
Long64_t	GetnSteps ()
	Get Number of Steps that Chain has, for merged chains will not be the same nEntries. More...
void	SetNBins (const int NewBins)
	Modify number of bins used for 1D and 2D Histograms. More...
void	SetEntries (const int NewEntries)
	Set number of entries to make potentially MCMC Processing faster. More...
void	SetStepCut (const std::string &Cuts)
	Set the step cutting by string. More...
void	SetStepCut (const int Cuts)
	Set the step cutting by int. More...
void	CheckStepCut () const
	Check if step cut isn't larger than highest values of step in a chain. More...
void	SetPlotRelativeToPrior (const bool PlotOrNot)
	You can set relative to prior or relative to generated. It is advised to use relate to prior. More...
void	SetPrintToPDF (const bool PlotOrNot)
void	SetPlotErrorForFlatPrior (const bool PlotOrNot)
	Set whether you want to plot error for parameters which have flat prior. More...
void	SetPlotBinValue (const bool PlotOrNot)
void	SetFancyNames (const bool PlotOrNot)
void	SetSmoothing (const bool PlotOrNot)
	Set whether want to use smoothing for histograms using ROOT algorithm. More...
void	SetPost2DPlotThreshold (const double Threshold)
	Code will only plot 2D posteriors if Correlation are larger than defined threshold. More...
void	SetUseFFTAutoCorrelation (const bool useFFT)
	Toggle using the FFT-based autocorrelation calculator. More...
void	SetExcludedTypes (std::vector< std::string > Name)
	Setter related what parameters we want to exclude from analysis, for example if cross-section parameters look like xsec_, then passing "xsec_" will. More...
void	SetExcludedNames (std::vector< std::string > Name)
void	SetExcludedGroups (std::vector< std::string > Name)
void	SetnBatches (const int Batches)
	Set value of Nbatches used for batched mean, this need to be done earlier as batches are made when reading tree. More...
void	SetnLags (const int nLags)
void	SetOutputSuffix (const std::string Suffix)
	Sett output suffix, this way jobs using the same file will have different names. More...
void	SetPosterior1DCut (const std::string Cut)
	Allow to set addtional cuts based on ROOT TBrowser cut, for to only affect one mass ordering. More...

Protected Member Functions
std::unique_ptr< TH1D >	MakePrefit ()
	Prepare prefit histogram for parameter overlay plot. More...
void	MakeOutputFile ()
	prepare output root file and canvas to which we will save EVERYTHING More...
void	DrawCorrelations1D ()
	Draw 1D correlations which might be more helpful than looking at huge 2D Corr matrix. More...
void	DrawCorrelationsGroup (const std::unique_ptr< TH2D > &CorrMatrix) const
	Produces correlation matrix but instead of giving name for each param it only give name for param group. More...
void	ReadInputCov ()
	CW: Read the input Covariance matrix entries. Get stuff like parameter input errors, names, and so on. More...
void	ReadInputCovLegacy ()
void	FindInputFiles ()
	Read the output MCMC file and find what inputs were used. More...
void	FindInputFilesLegacy ()
void	ReadModelFile ()
	Read the xsec file and get the input central values and errors. More...
virtual void	LoadAdditionalInfo ()
	allow loading additional info for example used for oscillation parameters More...
void	ReadNDFile ()
	Read the ND cov file and get the input central values and errors. More...
void	ReadFDFile ()
	Read the FD cov file and get the input central values and errors. More...
void	PrintInfo () const
	Print info like how many params have been loaded etc. More...
void	ScanInput ()
	Scan Input etc. More...
void	ScanParameterOrder ()
	Scan order of params from a different groups. More...
void	SetupOutput ()
	Prepare all objects used for output. More...
void	PrepareDiagMCMC ()
	CW: Prepare branches etc. for DiagMCMC. More...
void	ParamTraces ()
	CW: Draw trace plots of the parameters i.e. parameter vs step. More...
void	AutoCorrelation ()
	KS: Calculate autocorrelations supports both OpenMP and CUDA :) More...
void	AutoCorrelation_FFT ()
	MJR: Autocorrelation function using FFT algorithm for extra speed. More...
void	CalculateESS (const int nLags, const std::vector< std::vector< double >> &LagL)
	KS: calc Effective Sample Size. More...
void	BatchedAnalysis ()
	Get the batched means variance estimation and variable indicating if number of batches is sensible [4] [28]. More...
void	BatchedMeans ()
	CW: Batched means, literally read from an array and chuck into TH1D. More...
void	GewekeDiagnostic ()
	Geweke Diagnostic based on the methods described by Fang (2014) and Karlsbakk (2011). [8] [20]. More...
void	AcceptanceProbabilities ()
	Acceptance Probability. More...
void	PowerSpectrumAnalysis ()
	RC: Perform spectral analysis of MCMC [7]. More...
std::vector< double >	GetMargins (const std::unique_ptr< TCanvas > &Canv) const
	Get TCanvas margins, to be able to reset them if particular function need different margins. More...
void	SetMargins (std::unique_ptr< TCanvas > &Canv, const std::vector< double > &margins)
	Set TCanvas margins to specified values. More...
void	SetTLineStyle (TLine *Line, const Color_t Colour, const Width_t Width, const ELineStyle Style) const
	Configures a TLine object with the specified style parameters. More...
void	SetLegendStyle (TLegend *Legend, const double size) const
	Configures the style of a TLegend object. More...

Protected Attributes
std::string	MCMCFile
	Name of MCMC file. More...
std::string	OutputSuffix
	Output file suffix useful when running over same file with different settings. More...
std::vector< std::vector< std::string > >	CovPos
	Covariance matrix file name position. More...
std::vector< std::string >	CovNamePos
	Covariance matrix name position. More...
std::vector< YAML::Node >	CovConfig
	Covariance matrix config. More...
TChain *	Chain
	Main chain storing all steps etc. More...
std::string	StepCut
	BurnIn Cuts. More...
std::string	Posterior1DCut
	Cut used when making 1D Posterior distribution. More...
unsigned int	UpperCut
	KS: Used only for SubOptimality. More...
unsigned int	BurnInCut
	Value of burn in cut. More...
int	nBranches
	Number of branches in a TTree. More...
int	nEntries
	KS: For merged chains number of entries will be different from nSteps. More...
int	nSteps
	KS: For merged chains number of entries will be different from nSteps. More...
int	nSamples
	Number of sample PDF objects. More...
int	nSysts
	Number of covariance objects. More...
int	nDraw
	Number of all parameters used in the analysis. More...
std::vector< TString >	BranchNames
std::vector< std::string >	ExcludedTypes
std::vector< std::string >	ExcludedNames
std::vector< std::string >	ExcludedGroups
std::vector< bool >	IamVaried
	Is the ith parameter varied. More...
std::vector< std::vector< TString > >	ParamNames
	Name of parameters which we are going to analyse. More...
std::vector< std::vector< double > >	ParamCentral
	Parameters central values which we are going to analyse. More...
std::vector< std::vector< double > >	ParamNom
std::vector< std::vector< double > >	ParamErrors
	Uncertainty on a single parameter. More...
std::vector< std::vector< bool > >	ParamFlat
	Whether Param has flat prior or not. More...
std::vector< int >	nParam
	Number of parameters per type. More...
std::vector< ParameterEnum >	ParamType
	Make an enum for which class this parameter belongs to so we don't have to keep string comparing. More...
std::vector< int >	ParamTypeStartPos
std::vector< std::string >	ParameterGroup
std::vector< TString >	SampleName_v
	Vector of each systematic. More...
std::vector< TString >	SystName_v
	Vector of each sample PDF object. More...
std::string	OutputName
	Name of output files. More...
TString	CanvasName
	Name of canvas which help to save to the sample pdf. More...
bool	PlotFlatPrior
	Whether we plot flat prior or not, we usually provide error even for flat prior params. More...
bool	plotRelativeToPrior
	Whether we plot relative to prior or nominal, in most cases is prior. More...
bool	MadePostfit
	Sanity check if Postfit is already done to not make several times. More...
bool	printToPDF
	Will plot all plot to PDF not only to root file. More...
bool	FancyPlotNames
	Whether we want fancy plot names or not. More...
bool	plotBinValue
	If true it will print value on each bin of covariance matrix. More...
bool	ApplySmoothing
	Apply smoothing for 2D histos using root algorithm. More...
double	Post2DPlotThreshold
	KS: Set Threshold when to plot 2D posterior as by default we get a LOT of plots. More...
bool	useFFTAutoCorrelation
	MJR: Use FFT-based autocorrelation algorithm (save time & resources)? More...
std::vector< int >	NDSamplesBins
std::vector< std::string >	NDSamplesNames
std::unique_ptr< TF1 >	Gauss
	Gaussian fitter. More...
TFile *	OutputFile
	The output file. More...
std::unique_ptr< TCanvas >	Posterior
	Fancy canvas used for our beautiful plots. More...
TVectorD *	Central_Value
	Vector with central value for each parameter. More...
TVectorD *	Means
	Vector with mean values using Arithmetic Mean. More...
TVectorD *	Errors
	Vector with errors values using RMS. More...
TVectorD *	Means_Gauss
	Vector with mean values using Gaussian fit. More...
TVectorD *	Errors_Gauss
	Vector with error values using Gaussian fit. More...
TVectorD *	Means_HPD
	Vector with mean values using Highest Posterior Density. More...
TVectorD *	Errors_HPD
	Vector with error values using Highest Posterior Density. More...
TVectorD *	Errors_HPD_Positive
	Vector with positive error (right hand side) values using Highest Posterior Density. More...
TVectorD *	Errors_HPD_Negative
	Vector with negative error (left hand side) values using Highest Posterior Density. More...
TMatrixDSym *	Covariance
	Posterior Covariance Matrix. More...
TMatrixDSym *	Correlation
	Posterior Correlation Matrix. More...
std::vector< TH1D * >	hpost
	Holds 1D Posterior Distributions. More...
std::vector< std::vector< TH2D * > >	hpost2D
	Holds 2D Posterior Distributions. More...
std::unique_ptr< TH2D >	hviolin
	Holds violin plot for all dials. More...
std::unique_ptr< TH2D >	hviolin_prior
	Holds prior violin plot for all dials,. More...
double **	ParStep
	Array holding values for all parameters. More...
unsigned int *	StepNumber
	Step number for step, important if chains were merged. More...
int	nBins
	Number of bins. More...
double	DrawRange
	Drawrange for SetMaximum. More...
bool	CacheMCMC
	MCMC Chain has been cached. More...
bool	doDiagMCMC
	Doing MCMC Diagnostic. More...
int	nBatches
	Number of batches for Batched Mean. More...
int	AutoCorrLag
	LagL used in AutoCorrelation. More...
double *	ParamSums
	Total parameter sum for each param. More...
double **	BatchedAverages
	Values of batched average for every param and batch. More...
double **	SampleValues
	Holds the sample values. More...
double **	SystValues
	Holds the systs values. More...
double *	AccProbValues
	Holds all accProb. More...
double *	AccProbBatchedAverages
	Holds all accProb in batches. More...
bool	ReweightPosterior
	Whether to apply reweighting weight or not. More...
std::string	ReweightName
	Name of branch used for chain reweighting. More...
double *	WeightValue
	Stores value of weight for each step. More...

Detailed Description

Class responsible for processing MCMC chains, performing diagnostics, generating plots, and managing Bayesian analysis.

This class provides utilities to handle MCMC output generated by mcmc::runMCMC. It is particularly useful for extracting values from previous MCMC runs and initiating new MCMC runs with those values. Inspired by nd280_utils/DrawComp.cpp.

See also

For more details and examples, visit the Wiki.

Author

Clarence Wret

Kamil Skwarczynski

Definition at line 61 of file MCMCProcessor.h.

Constructor & Destructor Documentation

MCMCProcessor()

_MaCh3_Safe_Include_Start_ _MaCh3_Safe_Include_End_ MCMCProcessor::MCMCProcessor

(

const std::string &

InputFile

)

Constructs an MCMCProcessor object with the specified input file and options.

Parameters

InputFile

The path to the input file containing MCMC data.

Definition at line 17 of file MCMCProcessor.cpp.

                                                       :
  Chain(nullptr), StepCut(""), MadePostfit(false) {
// ****************************
  MCMCFile = InputFile;
 
  SetMaCh3LoggerFormat();
  MaCh3Utils::MaCh3Welcome();
  MACH3LOG_INFO("Making post-fit processor for: {}", MCMCFile);
 
  ParStep = nullptr;
  StepNumber = nullptr;
  ReweightPosterior = false;
  WeightValue = nullptr;
 
  Posterior = nullptr;
  hviolin = nullptr;
  hviolin_prior = nullptr;
 
  OutputFile = nullptr;
  
  ParamSums = nullptr;
  BatchedAverages = nullptr;
  SampleValues = nullptr;
  SystValues = nullptr;
  AccProbValues = nullptr;
  AccProbBatchedAverages = nullptr;
 
  //KS:Hardcoded should be a way to get it via config or something
  plotRelativeToPrior = false;
  printToPDF = false;
  plotBinValue = false;
  PlotFlatPrior = true;
  CacheMCMC = false;
  ApplySmoothing = true;
  FancyPlotNames = true;
  doDiagMCMC = false;
 
  // KS: ROOT can compile FFT code but it will crash during run time. Turn off FFT dynamically
#ifdef MaCh3_FFT
  useFFTAutoCorrelation = true;
#else
  useFFTAutoCorrelation = false;
#endif
  OutputSuffix = "_Process";
  Post2DPlotThreshold = 1.e-5;
 
  nDraw = 0;
  nEntries = 0;
  UpperCut = M3::_BAD_INT_;
  nSteps = 0;
  nBatches = 0;
  AutoCorrLag = 0;
  nSysts = 0;
  nSamples = 0;
  
  nBins = 70;
  DrawRange = 1.5;
  
  Posterior1DCut = "";
  //KS:Those keep basic information for ParameterEnum
  ParamNames.resize(kNParameterEnum);
  ParamCentral.resize(kNParameterEnum);
  ParamNom.resize(kNParameterEnum);
  ParamErrors.resize(kNParameterEnum);
  ParamFlat.resize(kNParameterEnum);
  ParamTypeStartPos.resize(kNParameterEnum);
  nParam.resize(kNParameterEnum);
  CovPos.resize(kNParameterEnum);
  CovNamePos.resize(kNParameterEnum);
  CovConfig.resize(kNParameterEnum);
 
  for(int i = 0; i < kNParameterEnum; i++)
  {
    ParamTypeStartPos[i] = 0;
    nParam[i] = 0;
  }
  //Only if GPU is enabled
  #ifdef MaCh3_CUDA
   ParStep_cpu = nullptr;
   NumeratorSum_cpu = nullptr;
   ParamSums_cpu = nullptr;
   DenomSum_cpu = nullptr;
 
   ParStep_gpu = nullptr;
   NumeratorSum_gpu = nullptr;
   ParamSums_gpu = nullptr;
   DenomSum_gpu = nullptr;
  #endif
}

~MCMCProcessor()

MCMCProcessor::~MCMCProcessor ( )

virtual

Destroys the MCMCProcessor object.

Definition at line 109 of file MCMCProcessor.cpp.

                              {
// ****************************
  // Close the pdf file
  MACH3LOG_INFO("Closing pdf in MCMCProcessor: {}", CanvasName.Data());
  CanvasName += "]";
  if(printToPDF) Posterior->Print(CanvasName);
 
  delete Covariance;
  delete Correlation;
  delete Central_Value;
  delete Means;
  delete Errors;
  delete Means_Gauss;
  delete Errors_Gauss;
  delete Means_HPD;
  delete Errors_HPD;
  delete Errors_HPD_Positive;
  delete Errors_HPD_Negative;
 
  if(WeightValue) delete[] WeightValue;
  for (int i = 0; i < nDraw; ++i)
  {
    if(hpost[i] != nullptr) delete hpost[i];
  }
  if(CacheMCMC)
  {
    for (int i = 0; i < nDraw; ++i) 
    {
      for (int j = 0; j < nDraw; ++j) 
      {
        delete hpost2D[i][j];
      }
      delete[] ParStep[i];
    }
    delete[] ParStep;
  }
  if(StepNumber != nullptr) delete[] StepNumber;
 
  if(OutputFile != nullptr) OutputFile->Close();
  if(OutputFile != nullptr) delete OutputFile;
  delete Chain;
}

Member Function Documentation

AcceptanceProbabilities()

void MCMCProcessor::AcceptanceProbabilities ( )

inlineprotected

Acceptance Probability.

Definition at line 4428 of file MCMCProcessor.cpp.

                                            {
// **************************
  if (AccProbBatchedAverages == nullptr) PrepareDiagMCMC();
 
  MACH3LOG_INFO("Making AccProb plots...");
 
  // Set the titles and limits for TH1Ds
  auto AcceptanceProbPlot = std::make_unique<TH1D>("AcceptanceProbability", "Acceptance Probability", nEntries, 0, nEntries);
  AcceptanceProbPlot->GetXaxis()->SetTitle("Step");
  AcceptanceProbPlot->GetYaxis()->SetTitle("Acceptance Probability");
 
  auto BatchedAcceptanceProblot = std::make_unique<TH1D>("AcceptanceProbability_Batch", "AcceptanceProbability_Batch", nBatches, 0, nBatches);
  BatchedAcceptanceProblot->GetYaxis()->SetTitle("Acceptance Probability");
  
  for (int i = 0; i < nBatches; ++i) {
    BatchedAcceptanceProblot->SetBinContent(i+1, AccProbBatchedAverages[i]);
    const int BatchRangeLow = double(i)*double(nEntries)/double(nBatches);
    const int BatchRangeHigh = double(i+1)*double(nEntries)/double(nBatches);
    std::stringstream ss;
    ss << BatchRangeLow << " - " << BatchRangeHigh;
    BatchedAcceptanceProblot->GetXaxis()->SetBinLabel(i+1, ss.str().c_str());
  }
  
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int i = 0; i < nEntries; ++i) {
    // Set bin content for the i-th bin to the parameter values
    AcceptanceProbPlot->SetBinContent(i, AccProbValues[i]);
  }
    
  TDirectory *probDir = OutputFile->mkdir("AccProb");
  probDir->cd();
  
  AcceptanceProbPlot->Write();
  BatchedAcceptanceProblot->Write();
  delete[] AccProbValues;
  delete[] AccProbBatchedAverages;
 
  probDir->Close();
  delete probDir;
 
  OutputFile->cd();
}

AutoCorrelation()

void MCMCProcessor::AutoCorrelation ( )

inlineprotected

KS: Calculate autocorrelations supports both OpenMP and CUDA :)

Definition at line 3685 of file MCMCProcessor.cpp.

                                    {
// *********************************
  if (ParStep == nullptr) PrepareDiagMCMC();
 
  TStopwatch clock;
  clock.Start();
  const int nLags = AutoCorrLag;
  MACH3LOG_INFO("Making auto-correlations for nLags = {}", nLags);
 
  // The sum of (Y-Ymean)^2 over all steps for each parameter
  std::vector<std::vector<double>> DenomSum(nDraw);
  std::vector<std::vector<double>> NumeratorSum(nDraw);
  std::vector<std::vector<double>> LagL(nDraw);
  for (int j = 0; j < nDraw; ++j) {
    DenomSum[j].resize(nLags);
    NumeratorSum[j].resize(nLags);
    LagL[j].resize(nLags);
  }
  std::vector<TH1D*> LagKPlots(nDraw);
  // Loop over the parameters of interest
  for (int j = 0; j < nDraw; ++j)
  {
    // Loop over each lag
    for (int k = 0; k < nLags; ++k) {
      NumeratorSum[j][k] = 0.0;
      DenomSum[j][k] = 0.0;
      LagL[j][k] = 0.0;
    }
 
    // Make TH1Ds for each parameter which hold the lag
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    
    GetNthParameter(j, Prior, PriorError, Title);
    std::string HistName = Form("%s_%s_Lag", Title.Data(), BranchNames[j].Data());
    LagKPlots[j] = new TH1D(HistName.c_str(), HistName.c_str(), nLags, 0.0, nLags);
    LagKPlots[j]->GetXaxis()->SetTitle("Lag");
    LagKPlots[j]->GetYaxis()->SetTitle("Auto-correlation function");
  }
//KS: If CUDA is not enabled do calculations on CPU
#ifndef MaCh3_CUDA
  // Loop over the lags
  //CW: Each lag is independent so might as well multi-thread them!
  #ifdef MULTITHREAD
  MACH3LOG_INFO("Using multi-threading...");
  #pragma omp parallel for collapse(2)
  #endif      // Loop over the number of parameters
  for (int j = 0; j < nDraw; ++j) {
    for (int k = 0; k < nLags; ++k) {
      // Loop over the number of entries
      for (int i = 0; i < nEntries; ++i) {
        const double Diff = ParStep[j][i]-ParamSums[j];
 
        // Only sum the numerator up to i = N-k
        if (i < nEntries-k) {
          const double LagTerm = ParStep[j][i+k]-ParamSums[j];
          const double Product = Diff*LagTerm;
          NumeratorSum[j][k] += Product;
        }
        // Square the difference to form the denominator
        const double Denom = Diff*Diff;
        DenomSum[j][k] += Denom;
      }
    }
  }
#else //NOW GPU specific code
  MACH3LOG_INFO("Using GPU");
  //KS: This allocates memory and copy data from CPU to GPU
  PrepareGPU_AutoCorr(nLags);
 
  //KS: This runs the main kernel and copy results back to CPU
  RunGPU_AutoCorr(
    ParStep_gpu,
    ParamSums_gpu,
    NumeratorSum_gpu,
    DenomSum_gpu,
    NumeratorSum_cpu,
    DenomSum_cpu);
 
  #ifdef MULTITHREAD
  #pragma omp parallel for collapse(2)
  #endif
  //KS: Now that that we received data from GPU convert it to CPU-like format
  for (int j = 0; j < nDraw; ++j)
  {
    for (int k = 0; k < nLags; ++k)
    {
      const int temp_index = j*nLags+k;
      NumeratorSum[j][k] = NumeratorSum_cpu[temp_index];
      DenomSum[j][k] = DenomSum_cpu[temp_index];
    }
  }
  //delete auxiliary variables
  delete[] NumeratorSum_cpu;
  delete[] DenomSum_cpu;
  delete[] ParStep_cpu;
  delete[] ParamSums_cpu;
 
  //KS: Delete stuff at GPU as well
  CleanupGPU_AutoCorr(
      ParStep_gpu,
      NumeratorSum_gpu,
      ParamSums_gpu,
      DenomSum_gpu);
 
//KS: End of GPU specific code
#endif
 
  OutputFile->cd();
  TDirectory *AutoCorrDir = OutputFile->mkdir("Auto_corr");
  // Now fill the LagK auto-correlation plots
  for (int j = 0; j < nDraw; ++j) {
    for (int k = 0; k < nLags; ++k) {
      LagL[j][k] = NumeratorSum[j][k]/DenomSum[j][k];
      LagKPlots[j]->SetBinContent(k, NumeratorSum[j][k]/DenomSum[j][k]);
    }
    AutoCorrDir->cd();
    LagKPlots[j]->Write();
    delete LagKPlots[j];
  }
 
  //KS: This is different diagnostic however it relies on calculated Lag, thus we call it before we delete LagKPlots
  CalculateESS(nLags, LagL);
 
  delete[] ParamSums;
 
  AutoCorrDir->Close();
  delete AutoCorrDir;
 
  OutputFile->cd();
 
  clock.Stop();
  MACH3LOG_INFO("Making auto-correlations took {:.2f}s", clock.RealTime());
}

AutoCorrelation_FFT()

void MCMCProcessor::AutoCorrelation_FFT ( )

inlineprotected

MJR: Autocorrelation function using FFT algorithm for extra speed.

Author

Michael Reh

Definition at line 3590 of file MCMCProcessor.cpp.

                                        {
// *********************************
  if (ParStep == nullptr) PrepareDiagMCMC();
 
  TStopwatch clock;
  clock.Start();
  const int nLags = AutoCorrLag;
  MACH3LOG_INFO("Making auto-correlations for nLags = {}", nLags);
 
  // Prep outputs
  OutputFile->cd();
  TDirectory* AutoCorrDir = OutputFile->mkdir("Auto_corr");
  std::vector<TH1D*> LagKPlots(nDraw);
  std::vector<std::vector<double>> LagL(nDraw);
 
  // Arrays needed to perform FFT using ROOT
  std::vector<double> ACFFT(nEntries, 0.0);  // Main autocorrelation array
  std::vector<double> ParVals(nEntries, 0.0);  // Param values for full chain
  std::vector<double> ParValsFFTR(nEntries, 0.0);  // FFT Real part
  std::vector<double> ParValsFFTI(nEntries, 0.0);  // FFT Imaginary part
  std::vector<double> ParValsFFTSquare(nEntries, 0.0);  // FFT Absolute square
  std::vector<double> ParValsComplex(nEntries, 0.0);  // Input Imaginary values (0)
 
  // Create forward and reverse FFT objects. I don't love using ROOT here,
  // but it works so I can't complain
  TVirtualFFT* fftf = TVirtualFFT::FFT(1, &nEntries, "C2CFORWARD");
  TVirtualFFT* fftb = TVirtualFFT::FFT(1, &nEntries, "C2CBACKWARD");
 
  // Loop over all pars and calculate the full autocorrelation function using FFT
  for (int j = 0; j < nDraw; ++j) {
    // Initialize
    LagL[j].resize(nLags);
    for (int i = 0; i < nEntries; ++i) {
      ParVals[i] = ParStep[j][i]-ParamSums[j]; // Subtract the mean to make it numerically tractable
      ParValsComplex[i] = 0.; // Reset dummy array
    }
 
    // Transform
    fftf->SetPointsComplex(ParVals.data(), ParValsComplex.data());
    fftf->Transform();
    fftf->GetPointsComplex(ParValsFFTR.data(), ParValsFFTI.data());
 
    // Square the results to get the power spectrum
    for (int i = 0; i < nEntries; ++i) {
      ParValsFFTSquare[i] = ParValsFFTR[i]*ParValsFFTR[i] + ParValsFFTI[i]*ParValsFFTI[i];
    }
 
    // Transforming back gives the autocovariance
    fftb->SetPointsComplex(ParValsFFTSquare.data(), ParValsComplex.data());
    fftb->Transform();
    fftb->GetPointsComplex(ACFFT.data(), ParValsComplex.data());
 
    // Divide by norm to get autocorrelation
    double normAC = ACFFT[0];
    for (int i = 0; i < nEntries; ++i) {
      ACFFT[i] /= normAC;
    }
 
    // Get plotting info
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(j, Prior, PriorError, Title);
    std::string HistName = Form("%s_%s_Lag", Title.Data(), BranchNames[j].Data());
 
    // Initialize Lag plot
    LagKPlots[j] = new TH1D(HistName.c_str(), HistName.c_str(), nLags, 0.0, nLags);
    LagKPlots[j]->GetXaxis()->SetTitle("Lag");
    LagKPlots[j]->GetYaxis()->SetTitle("Auto-correlation function");
 
    // Fill plot
    for (int k = 0; k < nLags; ++k) {
      LagL[j][k] = ACFFT[k];
      LagKPlots[j]->SetBinContent(k, ACFFT[k]);
    }
 
    // Write and clean up
    AutoCorrDir->cd();
    LagKPlots[j]->Write();
    delete LagKPlots[j];
  }
 
  //KS: This is different diagnostic however it relies on calculated Lag, thus we call it before we delete LagKPlots
  CalculateESS(nLags, LagL);
 
  AutoCorrDir->Close();
  delete AutoCorrDir;
 
  OutputFile->cd();
 
  clock.Stop();
  MACH3LOG_INFO("Making auto-correlations took {:.2f}s", clock.RealTime());
}

BatchedAnalysis()

void MCMCProcessor::BatchedAnalysis ( )

inlineprotected

Get the batched means variance estimation and variable indicating if number of batches is sensible [4] [28].

Definition at line 4059 of file MCMCProcessor.cpp.

                                    {
// **************************
  if(BatchedAverages == nullptr)
  {
    MACH3LOG_ERROR("BatchedAverages haven't been initialises or have been deleted something is wrong");
    MACH3LOG_ERROR("I need it and refuse to go further");
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
 
  // Calculate variance estimator using batched means following @cite chakraborty2019estimating see Eq. 1.2
  TVectorD* BatchedVariance = new TVectorD(nDraw);
  //KS: The hypothesis is rejected if C > z α for a given confidence level α. If the batch means do not pass the test, Correlated is reported for the half-width on the statistical reports following @cite rossetti2024batch alternatively for more old-school see Alexopoulos and Seila 1998 section 3.4.3
  TVectorD* C_Test_Statistics = new TVectorD(nDraw);
 
  std::vector<double> OverallBatchMean(nDraw);
  std::vector<double> C_Rho_Nominator(nDraw);
  std::vector<double> C_Rho_Denominator(nDraw);
  std::vector<double> C_Nominator(nDraw);
  std::vector<double> C_Denominator(nDraw);
  const int BatchLength = nEntries/nBatches+1;
//KS: Start parallel region
#ifdef MULTITHREAD
#pragma omp parallel
{
#endif
  #ifdef MULTITHREAD
  #pragma omp for
  #endif
  //KS: First calculate mean of batched means for each param and Initialise everything to 0
  for (int j = 0; j < nDraw; ++j)
  {
    OverallBatchMean[j] = 0.0;
    C_Rho_Nominator[j] = 0.0;
    C_Rho_Denominator[j] = 0.0;
    C_Nominator[j] = 0.0;
    C_Denominator[j] = 0.0;
 
    (*BatchedVariance)(j) = 0.0;
    (*C_Test_Statistics)(j) = 0.0;
    for (int i = 0; i < nBatches; ++i)
    {
      OverallBatchMean[j] += BatchedAverages[i][j];
    }
    OverallBatchMean[j] /= nBatches;
  }
 
  #ifdef MULTITHREAD
  #pragma omp for nowait
  #endif
  //KS: next loop is completely independent thus nowait clause
  for (int j = 0; j < nDraw; ++j)
  {
    for (int i = 0; i < nBatches; ++i)
    {
      (*BatchedVariance)(j) += (OverallBatchMean[j] - BatchedAverages[i][j])*(OverallBatchMean[j] - BatchedAverages[i][j]);
    }
    (*BatchedVariance)(j) = (BatchLength/(nBatches-1))* (*BatchedVariance)(j);
  }
  
  //KS: Now we focus on C test statistic, again use nowait as next is calculation is independent
  #ifdef MULTITHREAD
  #pragma omp for nowait
  #endif
  for (int j = 0; j < nDraw; ++j)
  {
    C_Nominator[j] = (OverallBatchMean[j] - BatchedAverages[0][j])*(OverallBatchMean[j] - BatchedAverages[0][j]) + 
                      (OverallBatchMean[j] - BatchedAverages[nBatches-1][j])*(OverallBatchMean[j] - BatchedAverages[nBatches-1][j]);
    for (int i = 0; i < nBatches; ++i)
    {
      C_Denominator[j] += (OverallBatchMean[j] - BatchedAverages[i][j])*(OverallBatchMean[j] - BatchedAverages[i][j]);
    }
    C_Denominator[j] = 2*C_Denominator[j];
  }
  
  //KS: We still calculate C and for this we need rho wee need autocorrelations between batches
  #ifdef MULTITHREAD
  #pragma omp for
  #endif
  for (int j = 0; j < nDraw; ++j)
  {
    for (int i = 0; i < nBatches-1; ++i)
    {
      C_Rho_Nominator[j] += (OverallBatchMean[j] - BatchedAverages[i][j])*(OverallBatchMean[j] - BatchedAverages[i+1][j]);
    }
    
    for (int i = 0; i < nBatches; ++i)
    {
      C_Rho_Denominator[j] += (OverallBatchMean[j] - BatchedAverages[i][j])*(OverallBatchMean[j] - BatchedAverages[i][j]);
    }
  }
  
  //KS: Final calculations of C
  #ifdef MULTITHREAD
  #pragma omp for
  #endif
  for (int j = 0; j < nDraw; ++j)
  {
    (*C_Test_Statistics)(j) = std::sqrt((nBatches*nBatches - 1)/(nBatches-2)) * ( C_Rho_Nominator[j]/C_Rho_Denominator[j] + C_Nominator[j]/ C_Denominator[j]);
  }
#ifdef MULTITHREAD
} //End parallel region
#endif
 
  //Save to file
  OutputFile->cd();
  BatchedVariance->Write("BatchedMeansVariance");
  C_Test_Statistics->Write("C_Test_Statistics");
 
  //Delete all variables
  delete BatchedVariance;
  delete C_Test_Statistics;
}

BatchedMeans()

void MCMCProcessor::BatchedMeans ( )

inlineprotected

CW: Batched means, literally read from an array and chuck into TH1D.

Definition at line 4001 of file MCMCProcessor.cpp.

                                 {
// **************************
  if (BatchedAverages == nullptr) PrepareDiagMCMC();
  MACH3LOG_INFO("Making BatchedMeans plots...");
  
  std::vector<TH1D*> BatchedParamPlots(nDraw);
  for (int j = 0; j < nDraw; ++j) {
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    
    GetNthParameter(j, Prior, PriorError, Title);
    
    std::string HistName = Form("%s_%s_batch", Title.Data(), BranchNames[j].Data());
    BatchedParamPlots[j] = new TH1D(HistName.c_str(), HistName.c_str(), nBatches, 0, nBatches);
  }
 
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int j = 0; j < nDraw; ++j) {
    for (int i = 0; i < nBatches; ++i) {
      BatchedParamPlots[j]->SetBinContent(i+1, BatchedAverages[i][j]);
      const int BatchRangeLow = double(i)*double(nEntries)/double(nBatches);
      const int BatchRangeHigh = double(i+1)*double(nEntries)/double(nBatches);
      std::stringstream ss;
      ss << BatchRangeLow << " - " << BatchRangeHigh;
      BatchedParamPlots[j]->GetXaxis()->SetBinLabel(i+1, ss.str().c_str());
    }
  }
 
  TDirectory *BatchDir = OutputFile->mkdir("Batched_means");
  BatchDir->cd();
  for (int j = 0; j < nDraw; ++j) {
    TF1 *Fitter = new TF1("Fitter","[0]", 0, nBatches);
    Fitter->SetLineColor(kRed);
    BatchedParamPlots[j]->Fit("Fitter","Rq");
    BatchedParamPlots[j]->Write();
    delete Fitter;
    delete BatchedParamPlots[j];
  }
 
  //KS: Get the batched means variance estimation and variable indicating if number of batches is sensible
  // We do this before deleting BatchedAverages
  BatchedAnalysis();
 
  for (int i = 0; i < nBatches; ++i) {
    delete BatchedAverages[i];
  }
  delete[] BatchedAverages;
 
  BatchDir->Close();
  delete BatchDir;
 
  OutputFile->cd();
}

CacheSteps()

void MCMCProcessor::CacheSteps

(

)

KS:By caching each step we use multithreading.

Definition at line 1072 of file MCMCProcessor.cpp.

                               {
// ***************
  if(CacheMCMC == true) return;
  
  CacheMCMC = true;
  
  if(ParStep != nullptr)
  {
    MACH3LOG_ERROR("It look like ParStep was already filled ");
    MACH3LOG_ERROR("Even though it is used for MakeCovariance_MP and for DiagMCMC ");
    MACH3LOG_ERROR("it has different structure in both for cache hits, sorry ");
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
 
  MACH3LOG_INFO("Caching input tree...");
  MACH3LOG_INFO("Allocating {:.2f} MB", double(sizeof(double)*nDraw*nEntries)/1.E6);
  TStopwatch clock;
  clock.Start();
  
  ParStep = new double*[nDraw];
  StepNumber = new unsigned int[nEntries];
  
  hpost2D.resize(nDraw);
  for (int i = 0; i < nDraw; ++i) 
  {
    ParStep[i] = new double[nEntries];
    hpost2D[i].resize(nDraw);
    for (int j = 0; j < nEntries; ++j)
    {
      ParStep[i][j] = -999.99;
      //KS: Set this only once
      if(i == 0) StepNumber[j] = 0;
    }
  }
 
  // Set all the branches to off
  Chain->SetBranchStatus("*", false);
  unsigned int stepBranch = 0;
  double* ParValBranch = new double[nDraw]();
  // Turn on the branches which we want for parameters
  for (int i = 0; i < nDraw; ++i)
  {
    Chain->SetBranchStatus(BranchNames[i].Data(), true);
    Chain->SetBranchAddress(BranchNames[i].Data(), &ParValBranch[i]);
  }
  Chain->SetBranchStatus("step", true);
  Chain->SetBranchAddress("step", &stepBranch);
 
  double ReweightWeight = 1.;
  if(ReweightPosterior)
  {
    WeightValue = new double[nEntries]();
    Chain->SetBranchStatus(ReweightName.c_str(), true);
    Chain->SetBranchAddress(ReweightName.c_str(), &ReweightWeight);
  }
 
  const Long64_t countwidth = nEntries/10;
 
  // Loop over the entries
  //KS: This is really a bottleneck right now, thus revisit with ROOT6 https://pep-root6.github.io/docs/analysis/parallell/root.html
  for (Long64_t j = 0; j < nEntries; ++j) 
  {
    if (j % countwidth == 0) {
        MaCh3Utils::PrintProgressBar(j, nEntries);
        MaCh3Utils::EstimateDataTransferRate(Chain, j);
    } else {
      Chain->GetEntry(j);
    }
    StepNumber[j] = stepBranch;
    // Set the branch addresses for params
    for (int i = 0; i < nDraw; ++i) {
      ParStep[i][j] = ParValBranch[i];
    }
 
    if(ReweightPosterior){
      WeightValue[j] = ReweightWeight;
    }
  }
  delete[] ParValBranch;
 
  // Set all the branches to on
  Chain->SetBranchStatus("*", true);
 
  // KS: Set temporary branch address to allow min/max, otherwise ROOT can segfaults
  double tempVal = 0.0;
  std::vector<double> Min_Chain(nDraw);
  std::vector<double> Max_Chain(nDraw);
  for (int i = 0; i < nDraw; ++i)
  {
    Chain->SetBranchAddress(BranchNames[i].Data(), &tempVal);
    Min_Chain[i] = Chain->GetMinimum(BranchNames[i]);
    Max_Chain[i] = Chain->GetMaximum(BranchNames[i]);
  }
 
  // Calculate the total number of TH2D objects
  size_t nHistograms = nDraw * (nDraw + 1) / 2;
  MACH3LOG_INFO("Caching 2D posterior histograms...");
  MACH3LOG_INFO("Allocating {:.2f} MB for {} 2D Posteriors (each {}x{} bins)",
                double(nHistograms * nBins * nBins * sizeof(double)) / 1.E6, nHistograms, nBins, nBins);
  // Cache max and min in chain for covariance matrix
  for (int i = 0; i < nDraw; ++i)
  {
    TString Title_i = "";
    double Prior_i, PriorError_i;
    GetNthParameter(i, Prior_i, PriorError_i, Title_i);
 
    for (int j = 0; j <= i; ++j)
    {
      TString Title_j = "";
      double Prior_j, PriorError_j;
      GetNthParameter(j, Prior_j, PriorError_j, Title_j);
 
      // TH2D to hold the Correlation
      hpost2D[i][j] = new TH2D((Title_i + "_" + Title_j).Data(), (Title_i + "_" + Title_j).Data(),
                               nBins, hpost[i]->GetXaxis()->GetXmin(), hpost[i]->GetXaxis()->GetXmax(),
                               nBins, hpost[j]->GetXaxis()->GetXmin(), hpost[j]->GetXaxis()->GetXmax());
      hpost2D[i][j]->SetMinimum(0);
      hpost2D[i][j]->GetXaxis()->SetTitle(Title_i);
      hpost2D[i][j]->GetYaxis()->SetTitle(Title_j);
      hpost2D[i][j]->GetZaxis()->SetTitle("Steps");
    }
  }
  clock.Stop();
  MACH3LOG_INFO("Caching steps took {:.2f}s to finish for {} steps", clock.RealTime(), nEntries );
}

CalculateESS()

void MCMCProcessor::CalculateESS ( const int  nLags, const std::vector< std::vector< double >> &  LagL  )

inlineprotected

KS: calc Effective Sample Size.

Parameters

nLags	Should be the same nLags as used in AutoCorrelation()
LagL	Value of LagL for each dial and each Lag

This function computes the Effective Sample Size (ESS) using the autocorrelations calculated by AutoCorrelation(). Ensure that the parameter nLags here matches the number of lags used in AutoCorrelation() to obtain accurate results. [30] [14] [9]

Definition at line 3904 of file MCMCProcessor.cpp.

                                                                                          {
// **************************
  if(LagL.size() == 0)
  {
    MACH3LOG_ERROR("Size of LagL is {}", LagL.size());
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  MACH3LOG_INFO("Making ESS plots...");
  TVectorD* EffectiveSampleSize = new TVectorD(nDraw);
  TVectorD* SamplingEfficiency = new TVectorD(nDraw);
  std::vector<double> TempDenominator(nDraw);
 
  constexpr int Nhists = 5;
  constexpr double Thresholds[Nhists + 1] = {1, 0.02, 0.005, 0.001, 0.0001, 0.0};
  constexpr Color_t ESSColours[Nhists] = {kGreen, kGreen + 2, kYellow, kOrange, kRed};
 
  //KS: This histogram is inspired by the following: @cite gabry2024visual
  std::vector<std::unique_ptr<TH1D>> EffectiveSampleSizeHist(Nhists);
  for(int i = 0; i < Nhists; ++i)
  {
    EffectiveSampleSizeHist[i] =
      std::make_unique<TH1D>(Form("EffectiveSampleSizeHist_%i", i), Form("EffectiveSampleSizeHist_%i", i), nDraw, 0, nDraw);
    EffectiveSampleSizeHist[i]->GetYaxis()->SetTitle("N_{eff}/N");
    EffectiveSampleSizeHist[i]->SetFillColor(ESSColours[i]);
    EffectiveSampleSizeHist[i]->SetLineColor(ESSColours[i]);
    EffectiveSampleSizeHist[i]->Sumw2();
    for (int j = 0; j < nDraw; ++j)
    {
      TString Title = "";
      double Prior = 1.0, PriorError = 1.0;
      GetNthParameter(j, Prior, PriorError, Title);
      EffectiveSampleSizeHist[i]->GetXaxis()->SetBinLabel(j+1, Title.Data());
    }
  }
 
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  //KS: Calculate ESS and MCMC efficiency for each parameter
  for (int j = 0; j < nDraw; ++j)
  {
    (*EffectiveSampleSize)(j) = M3::_BAD_DOUBLE_;
    (*SamplingEfficiency)(j) = M3::_BAD_DOUBLE_;
    TempDenominator[j] = 0.;
    //KS: Firs sum over all Calculated autocorrelations
    for (int k = 0; k < nLags; ++k)
    {
      TempDenominator[j] += LagL[j][k];
    }
    TempDenominator[j] = 1+2*TempDenominator[j];
    (*EffectiveSampleSize)(j) = double(nEntries)/TempDenominator[j];
    // 100 because we convert to percentage
    (*SamplingEfficiency)(j) = 100 * 1/TempDenominator[j];
 
    for(int i = 0; i < Nhists; ++i)
    {
      EffectiveSampleSizeHist[i]->SetBinContent(j+1, 0);
      EffectiveSampleSizeHist[i]->SetBinError(j+1, 0);
 
      const double TempEntry = std::fabs((*EffectiveSampleSize)(j)) / double(nEntries);
      if(Thresholds[i] >= TempEntry && TempEntry > Thresholds[i+1])
      {
        if( std::isnan((*EffectiveSampleSize)(j)) ) continue;
        EffectiveSampleSizeHist[i]->SetBinContent(j+1, TempEntry);
      }
    }
  }
 
  //KS Write to the output tree
  //Save to file
  OutputFile->cd();
  EffectiveSampleSize->Write("EffectiveSampleSize");
  SamplingEfficiency->Write("SamplingEfficiency");
 
  EffectiveSampleSizeHist[0]->SetTitle("Effective Sample Size");
  EffectiveSampleSizeHist[0]->Draw();
  for(int i = 1; i < Nhists; ++i)
  {
    EffectiveSampleSizeHist[i]->Draw("SAME");
  }
 
  auto leg = std::make_unique<TLegend>(0.2, 0.7, 0.6, 0.95);
  SetLegendStyle(leg.get(), 0.03);
  for(int i = 0; i < Nhists; ++i)
  {
    leg->AddEntry(EffectiveSampleSizeHist[i].get(), Form("%.4f >= N_{eff}/N > %.4f", Thresholds[i], Thresholds[i+1]), "f");
  }  leg->Draw("SAME");
 
  Posterior->Write("EffectiveSampleSizeCanvas");
 
  //Delete all variables
  delete EffectiveSampleSize;
  delete SamplingEfficiency;
}

CheckCredibleIntervalsOrder()

void MCMCProcessor::CheckCredibleIntervalsOrder	(	const std::vector< double > &	CredibleIntervals,
		const std::vector< Color_t > &	CredibleIntervalsColours
	)		const

Checks the order and size consistency of the CredibleIntervals and CredibleIntervalsColours vectors.

Parameters

CredibleIntervals	A vector of credible interval values.
CredibleIntervalsColours	A vector of colors associated with each credible interval.

Exceptions

MaCh3Exception

If the sizes are not equal or the intervals are not in decreasing order.

Definition at line 4474 of file MCMCProcessor.cpp.

                                                                                                                                                    {
// **************************
  if (CredibleIntervals.size() != CredibleIntervalsColours.size()) {
    MACH3LOG_ERROR("size of CredibleIntervals is not equal to size of CredibleIntervalsColours");
    throw MaCh3Exception(__FILE__, __LINE__);
  }
  if (CredibleIntervals.size() > 1) {
    for (unsigned int i = 1; i < CredibleIntervals.size(); i++) {
      if (CredibleIntervals[i] > CredibleIntervals[i - 1]) {
        MACH3LOG_ERROR("Interval {} is smaller than {}", i, i - 1);
        MACH3LOG_ERROR("{:.2f} {:.2f}", CredibleIntervals[i], CredibleIntervals[i - 1]);
        MACH3LOG_ERROR("They should be grouped in decreasing order");
        throw MaCh3Exception(__FILE__, __LINE__);
      }
    }
  }
}

CheckCredibleRegionsOrder()

void MCMCProcessor::CheckCredibleRegionsOrder	(	const std::vector< double > &	CredibleRegions,
		const std::vector< Style_t > &	CredibleRegionStyle,
		const std::vector< Color_t > &	CredibleRegionColor
	)

Checks the order and size consistency of the CredibleRegions, CredibleRegionStyle, and CredibleRegionColor vectors.

Parameters

CredibleRegions	A vector of credible region values.
CredibleRegionStyle	A vector of styles associated with each credible region.
CredibleRegionColor	A vector of colors associated with each credible region.

Exceptions

MaCh3Exception

If the sizes are not equal or the regions are not in decreasing order.

Definition at line 4493 of file MCMCProcessor.cpp.

                                                                              {
// **************************
  if ((CredibleRegions.size() != CredibleRegionStyle.size()) || (CredibleRegionStyle.size() != CredibleRegionColor.size())) {
    MACH3LOG_ERROR("size of CredibleRegions is not equal to size of CredibleRegionStyle or CredibleRegionColor");
    throw MaCh3Exception(__FILE__, __LINE__);
  }
  for (unsigned int i = 1; i < CredibleRegions.size(); i++) {
    if (CredibleRegions[i] > CredibleRegions[i - 1]) {
      MACH3LOG_ERROR("Interval {} is smaller than {}", i, i - 1);
      MACH3LOG_ERROR("{:.2f} {:.2f}", CredibleRegions[i], CredibleRegions[i - 1]);
      MACH3LOG_ERROR("They should be grouped in decreasing order");
      throw MaCh3Exception(__FILE__, __LINE__);
    }
  }
}

CheckStepCut()

void MCMCProcessor::CheckStepCut

(

)

const

Check if step cut isn't larger than highest values of step in a chain.

Definition at line 2685 of file MCMCProcessor.cpp.

                                       {
// ***************
  const unsigned int maxNsteps = Chain->GetMaximum("step");
  if(BurnInCut > maxNsteps){
    MACH3LOG_ERROR("StepCut({}) is larger than highest value of step({})", BurnInCut, maxNsteps);
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
}

DiagMCMC()

void MCMCProcessor::DiagMCMC

(

)

KS: Perform MCMC diagnostic including Autocorrelation, Trace etc.

Definition at line 3304 of file MCMCProcessor.cpp.

                             {
// **************************
  // Prepare branches etc for DiagMCMC
  PrepareDiagMCMC();
 
  // Draw the simple trace matrices
  ParamTraces();
 
  // Get the batched means
  BatchedMeans();
 
  // Draw the auto-correlations
  if (useFFTAutoCorrelation) {
    AutoCorrelation_FFT();
  } else {
    AutoCorrelation();
  }
 
  // Calculate Power Spectrum for each param
  PowerSpectrumAnalysis();
 
  // Get Geweke Z score helping select burn-in
  GewekeDiagnostic();
 
  // Draw acceptance Probability
  AcceptanceProbabilities();
}

DrawCorrelations1D()

void MCMCProcessor::DrawCorrelations1D ( )

inlineprotected

Draw 1D correlations which might be more helpful than looking at huge 2D Corr matrix.

Definition at line 1641 of file MCMCProcessor.cpp.

                                       {
// *********************
  //KS: Store it as we go back to them at the end
  const std::vector<double> Margins = GetMargins(Posterior);
  const int OptTitle = gStyle->GetOptTitle();
 
  Posterior->SetTopMargin(0.1);
  Posterior->SetBottomMargin(0.2);
  gStyle->SetOptTitle(1);
 
  constexpr int Nhists = 3;
  //KS: Highest value is just meant bo be sliglhy higher than 1 to catch >,
  constexpr double Thresholds[Nhists+1] = {0, 0.25, 0.5, 1.0001};
  constexpr Color_t CorrColours[Nhists] = {kRed-10, kRed-6,  kRed};
 
  //KS: This store necessary entries for stripped covariance which store only "meaningful correlations
  std::vector<std::vector<double>> CorrOfInterest;
  CorrOfInterest.resize(nDraw);
  std::vector<std::vector<std::string>> NameCorrOfInterest;
  NameCorrOfInterest.resize(nDraw);
 
  std::vector<std::vector<std::unique_ptr<TH1D>>> Corr1DHist(nDraw);
  //KS: Initialising ROOT objects is never safe in MP loop
  for(int i = 0; i < nDraw; ++i)
  {
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(i, Prior, PriorError, Title);
 
    Corr1DHist[i].resize(Nhists);
    for(int j = 0; j < Nhists; ++j)
    {
      Corr1DHist[i][j] = std::make_unique<TH1D>(Form("Corr1DHist_%i_%i", i, j), Form("Corr1DHist_%i_%i", i, j), nDraw, 0, nDraw);
      Corr1DHist[i][j]->SetTitle(Form("%s",Title.Data()));
      Corr1DHist[i][j]->GetYaxis()->SetTitle("Correlation");
      Corr1DHist[i][j]->SetFillColor(CorrColours[j]);
      Corr1DHist[i][j]->SetLineColor(kBlack);
 
      for (int k = 0; k < nDraw; ++k)
      {
        TString Title_y = "";
        double Prior_y = 1.0;
        double PriorError_y = 1.0;
        GetNthParameter(k, Prior_y, PriorError_y, Title_y);
        Corr1DHist[i][j]->GetXaxis()->SetBinLabel(k+1, Title_y.Data());
      }
    }
  }
 
  #ifdef MULTITHREAD
  #pragma omp parallel for collapse(2)
  #endif
  for(int i = 0; i < nDraw; ++i)
  {
    for(int j = 0; j < nDraw; ++j)
    {
      for(int k = 0; k < Nhists; ++k)
      {
        const double TempEntry = std::fabs((*Correlation)(i,j));
        if(Thresholds[k+1] > TempEntry && TempEntry >= Thresholds[k])
        {
          Corr1DHist[i][k]->SetBinContent(j+1, (*Correlation)(i,j));
        }
      }
      if(std::fabs((*Correlation)(i,j)) > Post2DPlotThreshold && i != j)
      {
        CorrOfInterest[i].push_back((*Correlation)(i,j));
        NameCorrOfInterest[i].push_back(Corr1DHist[i][0]->GetXaxis()->GetBinLabel(j+1));
      }
    }
  }
 
  TDirectory *CorrDir = OutputFile->mkdir("Corr1D");
  CorrDir->cd();
 
  for(int i = 0; i < nDraw; i++)
  {
    if (IamVaried[i] == false) continue;
 
    Corr1DHist[i][0]->GetXaxis()->LabelsOption("v");
    Corr1DHist[i][0]->SetMaximum(+1.);
    Corr1DHist[i][0]->SetMinimum(-1.);
    Corr1DHist[i][0]->Draw();
    for(int k = 1; k < Nhists; k++) {
      Corr1DHist[i][k]->Draw("SAME");
    }
 
    auto leg = std::make_unique<TLegend>(0.3, 0.75, 0.6, 0.90);
    SetLegendStyle(leg.get(), 0.02);
    for(int k = 0; k < Nhists; k++) {
      leg->AddEntry(Corr1DHist[i][k].get(), Form("%.2f > |Corr| >= %.2f", Thresholds[k+1], Thresholds[k]), "f");
    }
    leg->Draw("SAME");
 
    Posterior->Write(Corr1DHist[i][0]->GetTitle());
    if(printToPDF) Posterior->Print(CanvasName);
  }
 
  //KS: Plot only meaningful correlations
  for(int i = 0; i < nDraw; i++)
  {
    const int size = int(CorrOfInterest[i].size());
 
    if(size == 0) continue;
    auto Corr1DHist_Reduced = std::make_unique<TH1D>("Corr1DHist_Reduced", "Corr1DHist_Reduced", size, 0, size);
    Corr1DHist_Reduced->SetTitle(Corr1DHist[i][0]->GetTitle());
    Corr1DHist_Reduced->GetYaxis()->SetTitle("Correlation");
    Corr1DHist_Reduced->SetFillColor(kBlue);
    Corr1DHist_Reduced->SetLineColor(kBlue);
 
    for (int j = 0; j < size; ++j)
    {
      Corr1DHist_Reduced->GetXaxis()->SetBinLabel(j+1, NameCorrOfInterest[i][j].c_str());
      Corr1DHist_Reduced->SetBinContent(j+1, CorrOfInterest[i][j]);
    }
    Corr1DHist_Reduced->GetXaxis()->LabelsOption("v");
 
    Corr1DHist_Reduced->SetMaximum(+1.);
    Corr1DHist_Reduced->SetMinimum(-1.);
    Corr1DHist_Reduced->Draw();
 
    Posterior->Write(Form("%s_Red", Corr1DHist_Reduced->GetTitle()));
    if(printToPDF) Posterior->Print(CanvasName);
  }
 
  CorrDir->Close();
  delete CorrDir;
  OutputFile->cd();
 
  SetMargins(Posterior, Margins);
  gStyle->SetOptTitle(OptTitle);
}

DrawCorrelationsGroup()

void MCMCProcessor::DrawCorrelationsGroup ( const std::unique_ptr< TH2D > &  CorrMatrix ) const

inlineprotected

Produces correlation matrix but instead of giving name for each param it only give name for param group.

Parameters

CorrMatrix

correlation matrix that we are going to plot

Note

Inspired by plot in Ewan thesis see https://www.t2k.org/docs/thesis/152/Thesis#page=147

Definition at line 1531 of file MCMCProcessor.cpp.

                                                                                     {
// *********************
  MACH3LOG_INFO("Starting {}", __func__);
  const double RightMargin = Posterior->GetRightMargin();
  Posterior->SetRightMargin(0.15);
  auto MatrixCopy = M3::Clone(CorrMatrix.get());
 
  std::vector<std::string> GroupName;
  std::vector<int> GroupStart;
  std::vector<int> GroupEnd;
 
  // Loop over the Covariance matrix entries
  for (int iPar = 0; iPar < nDraw; ++iPar)
  {
    std::string GroupNameCurr;
    if(ParamType[iPar] == kXSecPar){
      const int InternalNumeration = iPar - ParamTypeStartPos[kXSecPar];
      GroupNameCurr = ParameterGroup[InternalNumeration];
    } else {
      GroupNameCurr = "Other"; // Use Other for all legacy params
    }
 
    if(iPar == 0) {
      GroupName.push_back(GroupNameCurr);
      GroupStart.push_back(0);
    } else if(GroupName.back() != GroupNameCurr ){
      GroupName.push_back(GroupNameCurr);
      GroupEnd.push_back(iPar);
      GroupStart.push_back(iPar);
    }
 
    MatrixCopy->GetXaxis()->SetBinLabel(iPar+1, "");
    MatrixCopy->GetYaxis()->SetBinLabel(iPar+1, "");
  }
  GroupEnd.push_back(nDraw);
 
  for(size_t iPar = 0; iPar < GroupName.size(); iPar++) {
    MACH3LOG_INFO("Group name {} from {} to {}", GroupName[iPar], GroupStart[iPar], GroupEnd[iPar]);
  }
  Posterior->cd();
  Posterior->Clear();
  MatrixCopy->Draw("colz");
 
  std::vector<std::unique_ptr<TLine>> groupLines; //((GroupStart.size() - 1) * 2);
 
  int nBinsX = MatrixCopy->GetNbinsX();
  int nBinsY = MatrixCopy->GetNbinsY();
 
  // Axis bounds from the histogram itself
  double xMin = MatrixCopy->GetXaxis()->GetBinLowEdge(1);
  double xMax = MatrixCopy->GetXaxis()->GetBinUpEdge(nBinsX);
  double yMin = MatrixCopy->GetYaxis()->GetBinLowEdge(1);
  double yMax = MatrixCopy->GetYaxis()->GetBinUpEdge(nBinsY);
 
  for (size_t g = 1; g < GroupStart.size(); ++g) {
    double posX = MatrixCopy->GetXaxis()->GetBinLowEdge(GroupStart[g] + 1);
    double posY = MatrixCopy->GetYaxis()->GetBinLowEdge(GroupStart[g] + 1);
 
    // Vertical line at group start
    auto vLine = std::make_unique<TLine>(posX, yMin, posX, yMax);
    vLine->SetLineColor(kBlack);
    vLine->SetLineWidth(2);
    vLine->Draw();
    groupLines.push_back(std::move(vLine));
 
    // Horizontal line at group start
    auto hLine = std::make_unique<TLine>(xMin, posY, xMax, posY);
    hLine->SetLineColor(kBlack);
    hLine->SetLineWidth(2);
    hLine->Draw();
    groupLines.push_back(std::move(hLine));
  }
 
  std::vector<std::unique_ptr<TText>> groupLabels(GroupName.size() * 2);
  double yOffsetBelow = 0.05 * (yMax - yMin); // space below x-axis
  double xOffsetRight = 0.02 * (xMax - xMin); // space right of y-axis
 
  for (size_t g = 0; g < GroupName.size(); ++g) {
    int startBin = GroupStart[g] + 1; // hist bins start at 1
    int endBin   = GroupEnd[g];
 
    double xStart = MatrixCopy->GetXaxis()->GetBinLowEdge(startBin);
    double xEnd   = MatrixCopy->GetXaxis()->GetBinUpEdge(endBin);
    double xMid   = 0.5 * (xStart + xEnd);
 
    double yStart = MatrixCopy->GetYaxis()->GetBinLowEdge(startBin);
    double yEnd   = MatrixCopy->GetYaxis()->GetBinUpEdge(endBin);
    double yMid   = 0.5 * (yStart + yEnd);
 
    // Label along X-axis (below histogram)
    auto labelX = std::make_unique<TText>(xMid, yMin - yOffsetBelow, GroupName[g].c_str());
    labelX->SetTextAlign(23); // center horizontally, top-aligned vertically
    labelX->SetTextSize(0.025);
    labelX->Draw();
    groupLabels.push_back(std::move(labelX));
 
    // Label along Y-axis (left of histogram)
    auto labelY = std::make_unique<TText>(xMin - xOffsetRight, yMid, GroupName[g].c_str());
    labelY->SetTextAlign(32); // right-aligned horizontally, center vertically
    labelY->SetTextSize(0.025);
    labelY->Draw();
    groupLabels.push_back(std::move(labelY));
  }
 
  if(printToPDF) Posterior->Print(CanvasName);
  Posterior->SetRightMargin(RightMargin);
}

DrawCovariance()

void MCMCProcessor::DrawCovariance

(

)

Draw the post-fit covariances.

Definition at line 1385 of file MCMCProcessor.cpp.

                                   {
// *********************
  const double RightMargin = Posterior->GetRightMargin();
  Posterior->SetRightMargin(0.15);
 
  // The Covariance matrix from the fit
  auto hCov = std::make_unique<TH2D>("hCov", "hCov", nDraw, 0, nDraw, nDraw, 0, nDraw);
  hCov->GetZaxis()->SetTitle("Covariance");
  // The Covariance matrix square root, with correct sign
  auto hCovSq = std::make_unique<TH2D>("hCovSq", "hCovSq", nDraw, 0, nDraw, nDraw, 0, nDraw);
  hCovSq->GetZaxis()->SetTitle("Covariance");
  // The Correlation
  auto hCorr = std::make_unique<TH2D>("hCorr", "hCorr", nDraw, 0, nDraw, nDraw, 0, nDraw);
  hCorr->GetZaxis()->SetTitle("Correlation");
  hCorr->SetMinimum(-1);
  hCorr->SetMaximum(1);
  hCov->GetXaxis()->SetLabelSize(0.015);
  hCov->GetYaxis()->SetLabelSize(0.015);
  hCovSq->GetXaxis()->SetLabelSize(0.015);
  hCovSq->GetYaxis()->SetLabelSize(0.015);
  hCorr->GetXaxis()->SetLabelSize(0.015);
  hCorr->GetYaxis()->SetLabelSize(0.015);
 
  // Loop over the Covariance matrix entries
  for (int i = 0; i < nDraw; ++i)
  {
    TString titlex = "";
    double nom, err;
    GetNthParameter(i, nom, err, titlex);
    
    hCov->GetXaxis()->SetBinLabel(i+1, titlex);
    hCovSq->GetXaxis()->SetBinLabel(i+1, titlex);
    hCorr->GetXaxis()->SetBinLabel(i+1, titlex);
 
    for (int j = 0; j < nDraw; ++j)
    {
      // The value of the Covariance
      const double cov = (*Covariance)(i,j);
      const double corr = (*Correlation)(i,j);
 
      hCov->SetBinContent(i+1, j+1, cov);
      hCovSq->SetBinContent(i+1, j+1, ((cov > 0) - (cov < 0))*std::sqrt(std::fabs(cov)));
      hCorr->SetBinContent(i+1, j+1, corr);
 
      TString titley = "";
      double nom_j, err_j;
      GetNthParameter(j, nom_j, err_j, titley);
 
      hCov->GetYaxis()->SetBinLabel(j+1, titley);
      hCovSq->GetYaxis()->SetBinLabel(j+1, titley);
      hCorr->GetYaxis()->SetBinLabel(j+1, titley);
    }
  }
 
  // Take away the stat box
  gStyle->SetOptStat(0);
  if(plotBinValue)gStyle->SetPaintTextFormat("4.1f"); //Precision of value in matrix element
  // Make pretty Correlation colors (red to blue)
  constexpr int NRGBs = 5;
  TColor::InitializeColors();
  Double_t stops[NRGBs] = { 0.00, 0.25, 0.50, 0.75, 1.00 };
  Double_t red[NRGBs]   = { 0.00, 0.25, 1.00, 1.00, 0.50 };
  Double_t green[NRGBs] = { 0.00, 0.25, 1.00, 0.25, 0.00 };
  Double_t blue[NRGBs]  = { 0.50, 1.00, 1.00, 0.25, 0.00 };
  TColor::CreateGradientColorTable(5, stops, red, green, blue, 255);
  gStyle->SetNumberContours(255);
 
  // cd into the correlation directory
  OutputFile->cd();
 
  Posterior->cd();
  Posterior->Clear();
  if(plotBinValue) hCov->Draw("colz text");
  else hCov->Draw("colz");
  if(printToPDF) Posterior->Print(CanvasName);
 
  Posterior->cd();
  Posterior->Clear();
  if(plotBinValue) hCorr->Draw("colz text");
  else hCorr->Draw("colz");
  if(printToPDF) Posterior->Print(CanvasName);
 
  hCov->Write("Covariance_plot");
  hCovSq->Write("Covariance_sq_plot");
  hCorr->Write("Correlation_plot");
  
  //Back to normal
  Posterior->SetRightMargin(RightMargin);
  DrawCorrelationsGroup(hCorr);
  DrawCorrelations1D();
}

DrawPostfit()

void MCMCProcessor::DrawPostfit

(

)

Draw the post-fit comparisons.

Definition at line 436 of file MCMCProcessor.cpp.

                                {
// *******************
  if (OutputFile == nullptr) MakeOutputFile();
 
  // Make the prefit plot
  std::unique_ptr<TH1D> prefit = MakePrefit();
 
  prefit->GetXaxis()->SetTitle("");
  // cd into the output file
  OutputFile->cd();
 
  std::string CutPosterior1D = "";
  if(Posterior1DCut != "")
  {
    CutPosterior1D = StepCut +" && " + Posterior1DCut;
  }
  else CutPosterior1D = StepCut;
 
  // Make a TH1D of the central values and the errors
  TH1D *paramPlot = new TH1D("paramPlot", "paramPlot", nDraw, 0, nDraw);
  paramPlot->SetName("mach3params");
  paramPlot->SetTitle(CutPosterior1D.c_str());
  paramPlot->SetFillStyle(3001);
  paramPlot->SetFillColor(kBlue-1);
  paramPlot->SetMarkerColor(paramPlot->GetFillColor());
  paramPlot->SetMarkerStyle(20);
  paramPlot->SetLineColor(paramPlot->GetFillColor());
  paramPlot->SetMarkerSize(prefit->GetMarkerSize());
  paramPlot->GetXaxis()->SetTitle("");
 
  // Same but with Gaussian output
  TH1D *paramPlot_Gauss = static_cast<TH1D*>(paramPlot->Clone());
  paramPlot_Gauss->SetMarkerColor(kOrange-5);
  paramPlot_Gauss->SetMarkerStyle(23);
  paramPlot_Gauss->SetLineWidth(2);
  paramPlot_Gauss->SetMarkerSize((prefit->GetMarkerSize())*0.75);
  paramPlot_Gauss->SetFillColor(paramPlot_Gauss->GetMarkerColor());
  paramPlot_Gauss->SetFillStyle(3244);
  paramPlot_Gauss->SetLineColor(paramPlot_Gauss->GetMarkerColor());
  paramPlot_Gauss->GetXaxis()->SetTitle("");
 
  // Same but with Gaussian output
  TH1D *paramPlot_HPD = static_cast<TH1D*>(paramPlot->Clone());
  paramPlot_HPD->SetMarkerColor(kBlack);
  paramPlot_HPD->SetMarkerStyle(25);
  paramPlot_HPD->SetLineWidth(2);
  paramPlot_HPD->SetMarkerSize((prefit->GetMarkerSize())*0.5);
  paramPlot_HPD->SetFillColor(0);
  paramPlot_HPD->SetFillStyle(0);
  paramPlot_HPD->SetLineColor(paramPlot_HPD->GetMarkerColor());
  paramPlot_HPD->GetXaxis()->SetTitle("");
 
  // Set labels and data
  for (int i = 0; i < nDraw; ++i)
  {
    //Those keep which parameter type we run currently and realtive number  
    int ParamEnu = ParamType[i];
    int ParamNo = i - ParamTypeStartPos[ParameterEnum(ParamEnu)];
 
    //KS: Slightly hacky way to get relative to prior or nominal as this is convention we use
    //This only applies for xsec for other systematic types doesn't matter
    double CentralValueTemp = 0;
    double Central, Central_gauss, Central_HPD;
    double Err, Err_Gauss, Err_HPD;
    
    if(plotRelativeToPrior)
    {
      CentralValueTemp = ParamCentral[ParamEnu][ParamNo];
      // Normalise the prior relative the nominal/prior, just the way we get our fit results in MaCh3
      if ( CentralValueTemp != 0) 
      {
        Central = (*Means)(i) / CentralValueTemp;
        Err = (*Errors)(i) / CentralValueTemp;
            
        Central_gauss = (*Means_Gauss)(i) / CentralValueTemp;
        Err_Gauss = (*Errors_Gauss)(i) / CentralValueTemp;
        
        Central_HPD = (*Means_HPD)(i) / CentralValueTemp;
        Err_HPD = (*Errors_HPD)(i) / CentralValueTemp;
      } 
      else {
        Central = 1+(*Means)(i);
        Err = (*Errors)(i);
            
        Central_gauss = 1+(*Means_Gauss)(i);
        Err_Gauss = (*Errors_Gauss)(i);
        
        Central_HPD = 1+(*Means_HPD)(i) ;
        Err_HPD = (*Errors_HPD)(i);
      }
    }
    //KS: Just get value of each parameter without dividing by prior
    else
    {
      Central = (*Means)(i);
      Err = (*Errors)(i);
          
      Central_gauss = (*Means_Gauss)(i);
      Err_Gauss = (*Errors_Gauss)(i);
      
      Central_HPD = (*Means_HPD)(i) ;
      Err_HPD = (*Errors_HPD)(i);
    }
    
    paramPlot->SetBinContent(i+1, Central);
    paramPlot->SetBinError(i+1, Err);
 
    paramPlot_Gauss->SetBinContent(i+1, Central_gauss);
    paramPlot_Gauss->SetBinError(i+1, Err_Gauss);
 
    paramPlot_HPD->SetBinContent(i+1, Central_HPD);
    paramPlot_HPD->SetBinError(i+1, Err_HPD);
    
    paramPlot->GetXaxis()->SetBinLabel(i+1, prefit->GetXaxis()->GetBinLabel(i+1));
    paramPlot_Gauss->GetXaxis()->SetBinLabel(i+1, prefit->GetXaxis()->GetBinLabel(i+1));
    paramPlot_HPD->GetXaxis()->SetBinLabel(i+1, prefit->GetXaxis()->GetBinLabel(i+1));
  }
  prefit->GetXaxis()->LabelsOption("v");
  paramPlot->GetXaxis()->LabelsOption("v");\
  paramPlot_Gauss->GetXaxis()->LabelsOption("v");
  paramPlot_HPD->GetXaxis()->LabelsOption("v");
 
  // Make a TLegend
  auto CompLeg = std::make_unique<TLegend>(0.33, 0.73, 0.76, 0.95);
  CompLeg->AddEntry(prefit.get(), "Prefit", "fp");
  CompLeg->AddEntry(paramPlot, "Postfit PDF", "fp");
  CompLeg->AddEntry(paramPlot_Gauss, "Postfit Gauss", "fp");
  CompLeg->AddEntry(paramPlot_HPD, "Postfit HPD", "lfep");
  CompLeg->SetFillColor(0);
  CompLeg->SetFillStyle(0);
  CompLeg->SetLineWidth(0);
  CompLeg->SetLineStyle(0);
  CompLeg->SetBorderSize(0);
 
  const std::vector<double> Margins = GetMargins(Posterior);
  Posterior->SetBottomMargin(0.2);
 
  OutputFile->cd();
 
  // Write the individual ones
  prefit->Write("param_xsec_prefit");
  paramPlot->Write("param_xsec");
  paramPlot_Gauss->Write("param_xsec_gaus");
  paramPlot_HPD->Write("param_xsec_HPD");
 
  // Plot the xsec parameters (0 to ~nXsec-nFlux) nXsec == xsec + flux, quite confusing I know
  // Have already looked through the branches earlier
  if(plotRelativeToPrior)  prefit->GetYaxis()->SetTitle("Variation rel. prior");
  else prefit->GetYaxis()->SetTitle("Parameter Value");
  prefit->GetYaxis()->SetRangeUser(-2.5, 2.5);
 
  // And the combined
  prefit->Draw("e2");
  paramPlot->Draw("e2, same");
  paramPlot_Gauss->Draw("e2, same");
  paramPlot_HPD->Draw("e1, same");
  CompLeg->Draw("same");
  Posterior->Write("param_xsec_canv");
 
  //KS: Tells how many parameters in one canvas we want
  constexpr int IntervalsSize = 20;
  const int NIntervals = nDraw/IntervalsSize;
 
  for (int i = 0; i < NIntervals+1; ++i)
  {
    int RangeMin = i*IntervalsSize;
    int RangeMax =RangeMin + IntervalsSize;
    if(i == NIntervals+1) {
      RangeMin = i*IntervalsSize;
      RangeMax = nDraw;
    }
    if(RangeMin >= nDraw) break;
 
    double ymin = std::numeric_limits<double>::max();
    double ymax = -std::numeric_limits<double>::max();
    for (int b = RangeMin; b <= RangeMax; ++b) {
      // prefit
      {
        double val = prefit->GetBinContent(b);
        double err = prefit->GetBinError(b);
        ymin = std::min(ymin, val - err);
        ymax = std::max(ymax, val + err);
      }
      // paramPlot_HPD
      {
        double val = paramPlot_HPD->GetBinContent(b);
        double err = paramPlot_HPD->GetBinError(b);
        ymin = std::min(ymin, val - err);
        ymax = std::max(ymax, val + err);
      }
    }
 
    double margin = 0.1 * (ymax - ymin);
    prefit->GetYaxis()->SetRangeUser(ymin - margin, ymax + margin);
 
    prefit->GetXaxis()->SetRangeUser(RangeMin, RangeMax);
    paramPlot->GetXaxis()->SetRangeUser(RangeMin, RangeMax);
    paramPlot_Gauss->GetXaxis()->SetRangeUser(RangeMin, RangeMax);
    paramPlot_HPD->GetXaxis()->SetRangeUser(RangeMin, RangeMax);
 
    // And the combined
    prefit->Draw("e2");
    paramPlot->Draw("e2, same");
    paramPlot_Gauss->Draw("e2, same");
    paramPlot_HPD->Draw("e1, same");
    CompLeg->Draw("same");
    if(printToPDF) Posterior->Print(CanvasName);
    Posterior->Clear();
  }
 
  if(nParam[kNDPar] > 0)
  {
    int Start = ParamTypeStartPos[kNDPar];
    int NDbinCounter = Start;
    //KS: Make prefit postfit for each ND sample, having all of them at the same plot is unreadable
    for(unsigned int i = 0; i < NDSamplesNames.size(); i++ )
    {
      std::string NDname = NDSamplesNames[i];
      NDbinCounter += NDSamplesBins[i];
      OutputFile->cd();
      prefit->GetYaxis()->SetTitle(("Variation for "+NDname).c_str());
      prefit->GetYaxis()->SetRangeUser(0.6, 1.4);
      prefit->GetXaxis()->SetRangeUser(Start, NDbinCounter);
 
      paramPlot->GetYaxis()->SetTitle(("Variation for "+NDname).c_str());
      paramPlot->GetYaxis()->SetRangeUser(0.6, 1.4);
      paramPlot->GetXaxis()->SetRangeUser(Start, NDbinCounter);
      paramPlot->SetTitle(CutPosterior1D.c_str());
 
      paramPlot_Gauss->GetYaxis()->SetTitle(("Variation for "+NDname).c_str());
      paramPlot_Gauss->GetYaxis()->SetRangeUser(0.6, 1.4);
      paramPlot_Gauss->GetXaxis()->SetRangeUser(Start, NDbinCounter);
      paramPlot_Gauss->SetTitle(CutPosterior1D.c_str());
 
      paramPlot_HPD->GetYaxis()->SetTitle(("Variation for "+NDname).c_str());
      paramPlot_HPD->GetYaxis()->SetRangeUser(0.6, 1.4);
      paramPlot_HPD->GetXaxis()->SetRangeUser(Start, NDbinCounter);
      paramPlot_HPD->SetTitle(CutPosterior1D.c_str());
 
      prefit->Write(("param_"+NDname+"_prefit").c_str());
      paramPlot->Write(("param_"+NDname).c_str());
      paramPlot_Gauss->Write(("param_"+NDname+"_gaus").c_str());
      paramPlot_HPD->Write(("param_"+NDname+"_HPD").c_str());
 
      prefit->Draw("e2");
      paramPlot->Draw("e2, same");
      paramPlot_Gauss->Draw("e1, same");
      paramPlot_HPD->Draw("e1, same");
      CompLeg->Draw("same");
      Posterior->Write(("param_"+NDname+"_canv").c_str());
      if(printToPDF) Posterior->Print(CanvasName);
      Posterior->Clear();
      Start += NDSamplesBins[i];
    }
  }
  delete paramPlot;
  delete paramPlot_Gauss;
  delete paramPlot_HPD;
 
  //KS: Return Margin to default one
  SetMargins(Posterior, Margins);
}

FindInputFiles()

void MCMCProcessor::FindInputFiles ( )

inlineprotected

Read the output MCMC file and find what inputs were used.

Definition at line 2411 of file MCMCProcessor.cpp.

                                   {
// **************************
  // Now read the MCMC file
  TFile *TempFile = new TFile(MCMCFile.c_str(), "open");
  TDirectory* CovarianceFolder = TempFile->Get<TDirectory>("CovarianceFolder");
 
  // Get the settings for the MCMC
  TMacro *Config = TempFile->Get<TMacro>("MaCh3_Config");
 
  if (Config == nullptr) {
    MACH3LOG_ERROR("Didn't find MaCh3_Config tree in MCMC file! {}", MCMCFile);
    TempFile->ls();
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  MACH3LOG_INFO("Loading YAML config from MCMC chain");
 
  YAML::Node Settings = TMacroToYAML(*Config);
 
  bool InputNotFound = false;
  //CW: Get the xsec Covariance matrix
  CovPos[kXSecPar] = GetFromManager<std::vector<std::string>>(Settings["General"]["Systematics"]["XsecCovFile"], {"none"});
  if(CovPos[kXSecPar].back() == "none")
  {
    MACH3LOG_WARN("Couldn't find XsecCov branch in output");
    InputNotFound = true;
  }
 
  TMacro *XsecConfig = M3::GetConfigMacroFromChain(CovarianceFolder);
  if (XsecConfig == nullptr) {
    MACH3LOG_WARN("Didn't find Config_xsec_cov tree in MCMC file! {}", MCMCFile);
  } else {
    CovConfig[kXSecPar] = TMacroToYAML(*XsecConfig);
  }
  if(InputNotFound) MaCh3Utils::PrintConfig(Settings);
 
  for(size_t i = 0; i < CovPos[kXSecPar].size(); i++)
    M3::AddPath(CovPos[kXSecPar][i]);
 
  // Delete the TTrees and the input file handle since we've now got the settings we need
  delete Config;
  delete XsecConfig;
 
  TMacro *ReweightConfig = TempFile->Get<TMacro>("Reweight_Config");
  if (ReweightConfig != nullptr) {
    YAML::Node ReweightSettings = TMacroToYAML(*ReweightConfig);
 
    if (ReweightSettings["Weight"]) {
      ReweightName = "Weight";
      ReweightPosterior = true;
      MACH3LOG_INFO("Enabling reweighting with following Config");
    } else {
      MACH3LOG_WARN("Found reweight config but without field ''Weight''");
    }
    MaCh3Utils::PrintConfig(ReweightSettings);
  }
 
  // Delete the MCMCFile pointer we're reading
  CovarianceFolder->Close();
  delete CovarianceFolder;
  TempFile->Close();
  delete TempFile;
}

FindInputFilesLegacy()

void MCMCProcessor::FindInputFilesLegacy ( )

inlineprotected

Warning

This will no longer be supported in future

Definition at line 2476 of file MCMCProcessor.cpp.

                                         {
// **************************
  // Now read the MCMC file
  TFile *TempFile = new TFile(MCMCFile.c_str(), "open");
 
  // Get the settings for the MCMC
  TMacro *Config = TempFile->Get<TMacro>("MaCh3_Config");
 
  if (Config == nullptr) {
    MACH3LOG_ERROR("Didn't find MaCh3_Config tree in MCMC file! {}", MCMCFile);
    TempFile->ls();
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  YAML::Node Settings = TMacroToYAML(*Config);
 
  //CW: And the ND Covariance matrix
  CovPos[kNDPar].push_back(GetFromManager<std::string>(Settings["General"]["Systematics"]["NDCovFile"], "none"));
 
  if(CovPos[kNDPar].back() == "none") {
    MACH3LOG_WARN("Couldn't find NDCov (legacy) branch in output");
  } else{
    //If the FD Cov is not none, then you need the name of the covariance object to grab
    CovNamePos[kNDPar] = GetFromManager<std::string>(Settings["General"]["Systematics"]["NDCovName"], "none");
    MACH3LOG_INFO("Given NDCovFile {} and NDCovName {}", CovPos[kNDPar].back(), CovNamePos[kNDPar]);
  }
 
  //CW: And the FD Covariance matrix
  CovPos[kFDDetPar].push_back(GetFromManager<std::string>(Settings["General"]["Systematics"]["FDCovFile"], "none"));
 
  if(CovPos[kFDDetPar].back() == "none") {
    MACH3LOG_WARN("Couldn't find FDCov (legacy) branch in output");
  } else {
    //If the FD Cov is not none, then you need the name of the covariance object to grab
    CovNamePos[kFDDetPar] = GetFromManager<std::string>(Settings["General"]["Systematics"]["FDCovName"], "none");
    MACH3LOG_INFO("Given FDCovFile {} and FDCovName {}", CovPos[kFDDetPar].back(), CovNamePos[kFDDetPar]);
  }
 
  for(size_t i = 0; i < CovPos[kNDPar].size(); i++)
    M3::AddPath(CovPos[kNDPar][i]);
 
  for(size_t i = 0; i < CovPos[kFDDetPar].size(); i++)
    M3::AddPath(CovPos[kFDDetPar][i]);
 
  TempFile->Close();
  delete TempFile;
}

GetBayesFactor()

void MCMCProcessor::GetBayesFactor	(	const std::vector< std::string > &	ParName,
		const std::vector< std::vector< double >> &	Model1Bounds,
		const std::vector< std::vector< double >> &	Model2Bounds,
		const std::vector< std::vector< std::string >> &	ModelNames
	)

Calculate Bayes factor for vector of params, and model boundaries.

Parameters

ParName	Vector with parameter names for which we calculate Bayes factor
Model1Bounds	Lower and upper bound for hypothesis 1. Within this bound we calculate integral used later for Bayes Factor
Model2Bounds	Lower and upper bound for hypothesis 2. Within this bound we calculate integral used later for Bayes Factor
ModelNames	Names for hypothesis 1 and 2

Definition at line 2816 of file MCMCProcessor.cpp.

                                                                                     {
// **************************
  if(hpost[0] == nullptr) MakePostfit();
 
  MACH3LOG_INFO("Calculating Bayes Factor");
  if((ParNames.size() != Model1Bounds.size()) || (Model2Bounds.size() != Model1Bounds.size())  || (Model2Bounds.size() != ModelNames.size()))
  {
    MACH3LOG_ERROR("Size doesn't match");
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  for(unsigned int k = 0; k < ParNames.size(); ++k)
  {
    //KS: First we need to find parameter number based on name
    int ParamNo = GetParamIndexFromName(ParNames[k]);
    if(ParamNo == M3::_BAD_INT_)
    {
      MACH3LOG_WARN("Couldn't find param {}. Will not calculate Bayes Factor", ParNames[k]);
      continue;
    }
 
    const double M1_min = Model1Bounds[k][0];
    const double M2_min = Model2Bounds[k][0];
    const double M1_max = Model1Bounds[k][1];
    const double M2_max = Model2Bounds[k][1];
 
    long double IntegralMode1 = hpost[ParamNo]->Integral(hpost[ParamNo]->FindFixBin(M1_min), hpost[ParamNo]->FindFixBin(M1_max));
    long double IntegralMode2 = hpost[ParamNo]->Integral(hpost[ParamNo]->FindFixBin(M2_min), hpost[ParamNo]->FindFixBin(M2_max));
 
    double BayesFactor = 0.;
    std::string Name = "";
    //KS: Calc Bayes Factor
    //If M1 is more likely
    if(IntegralMode1 >= IntegralMode2)
    {
      BayesFactor = IntegralMode1/IntegralMode2;
      Name = "\\mathfrak{B}(" + ModelNames[k][0]+ "/" + ModelNames[k][1] + ") = " + std::to_string(BayesFactor);
    }
    else //If M2 is more likely
    {
      BayesFactor = IntegralMode2/IntegralMode1;
      Name = "\\mathfrak{B}(" + ModelNames[k][1]+ "/" + ModelNames[k][0] + ") = " + std::to_string(BayesFactor);
    }
    std::string JeffreysScale = GetJeffreysScale(BayesFactor);
    std::string DunneKabothScale = GetDunneKaboth(BayesFactor);
 
    MACH3LOG_INFO("{} for {}", Name, ParNames[k]);
    MACH3LOG_INFO("Following Jeffreys Scale = {}", JeffreysScale);
    MACH3LOG_INFO("Following Dunne-Kaboth Scale = {}", DunneKabothScale);
    MACH3LOG_INFO("");
  }
}

GetBranchNames()

const std::vector<TString>& MCMCProcessor::GetBranchNames ( ) const

inline

Get the vector of branch names from root file.

Definition at line 248 of file MCMCProcessor.h.

{ return BranchNames;};

GetCovariance()

void MCMCProcessor::GetCovariance	(	TMatrixDSym *&	Cov,
		TMatrixDSym *&	Corr
	)

Get the post-fit covariances and correlations.

Parameters

Cov	Covariance matrix
Corr	Correlation matrix

Definition at line 196 of file MCMCProcessor.cpp.

                                                                       {
// ***************
  if (CacheMCMC) MakeCovariance_MP();
  else MakeCovariance();
  Cov = static_cast<TMatrixDSym*>(Covariance->Clone());
  Corr = static_cast<TMatrixDSym*>(Correlation->Clone());
}

GetCovConfig()

YAML::Node MCMCProcessor::GetCovConfig ( const int  i ) const

inline

Get Yaml config obtained from a Chain.

Definition at line 216 of file MCMCProcessor.h.

{return CovConfig.at(i); }

GetFDCov()

std::string MCMCProcessor::GetFDCov ( ) const

inline

Definition at line 236 of file MCMCProcessor.h.

{ return CovPos[kFDDetPar].back(); };

GetGroup()

int MCMCProcessor::GetGroup

(

const std::string &

name

)

const

Number of params from a given group, for example flux.

Definition at line 4512 of file MCMCProcessor.cpp.

                                                       {
// **************************
  // Lambda to compare strings case-insensitively
  auto caseInsensitiveCompare = [](const std::string& a, const std::string& b) {
    return std::equal(a.begin(), a.end(), b.begin(), b.end(),
                      [](char c1, char c2) { return std::tolower(c1) == std::tolower(c2); });
  };
  int numerator = 0;
  for (const auto& groupName : ParameterGroup) {
    if (caseInsensitiveCompare(groupName, name)) {
      numerator++;
    }
  }
  return numerator;
}

GetHpost()

TH1D* MCMCProcessor::GetHpost ( const int  i ) const

inline

Get 1D posterior for a given parameter.

Parameters

i	parameter index

Definition at line 223 of file MCMCProcessor.h.

{ return hpost[i]; };

GetHpost2D()

TH2D* MCMCProcessor::GetHpost2D ( const int  i, const int  j  ) const

inline

Get 2D posterior for a given parameter combination.

Parameters

i	parameter index X
j	parameter index Y

Definition at line 227 of file MCMCProcessor.h.

{ return hpost2D[i][j]; };

GetMargins()

std::vector< double > MCMCProcessor::GetMargins ( const std::unique_ptr< TCanvas > &  Canv ) const

protected

Get TCanvas margins, to be able to reset them if particular function need different margins.

Definition at line 4556 of file MCMCProcessor.cpp.

                                                                                    {
// **************************
  return std::vector<double>{Canv->GetTopMargin(), Canv->GetBottomMargin(),
                              Canv->GetLeftMargin(), Canv->GetRightMargin()};
}

GetNDCov()

std::string MCMCProcessor::GetNDCov ( ) const

inline

Definition at line 235 of file MCMCProcessor.h.

{ return CovPos[kNDPar].back(); };

GetnEntries()

Long64_t MCMCProcessor::GetnEntries ( )

inline

Get Number of entries that Chain has, for merged chains will not be the same Nsteps.

Definition at line 254 of file MCMCProcessor.h.

{return nEntries;};

GetNFD()

int MCMCProcessor::GetNFD ( ) const

inline

Definition at line 213 of file MCMCProcessor.h.

{ return nParam[kFDDetPar]; };

GetNND()

int MCMCProcessor::GetNND ( ) const

inline

Definition at line 212 of file MCMCProcessor.h.

{ return nParam[kNDPar]; };

GetNParams()

int MCMCProcessor::GetNParams ( ) const

inline

Get total number of used parameters.

Definition at line 210 of file MCMCProcessor.h.

{ return nDraw; };

GetnSteps()

Long64_t MCMCProcessor::GetnSteps ( )

inline

Get Number of Steps that Chain has, for merged chains will not be the same nEntries.

Definition at line 256 of file MCMCProcessor.h.

{return nSteps;};

GetNthParameter()

void MCMCProcessor::GetNthParameter	(	const int	param,
		double &	Prior,
		double &	PriorError,
		TString &	Title
	)		const

Get properties of parameter by passing it number.

Definition at line 2696 of file MCMCProcessor.cpp.

                                                                                                            {
// **************************
  ParameterEnum ParType = ParamType[param];
  int ParamNo = M3::_BAD_INT_;
  ParamNo = param - ParamTypeStartPos[ParType];
 
  Prior = ParamCentral[ParType][ParamNo];
  PriorError = ParamErrors[ParType][ParamNo];
  Title = ParamNames[ParType][ParamNo];
}

GetNXSec()

int MCMCProcessor::GetNXSec ( ) const

inline

Definition at line 211 of file MCMCProcessor.h.

{ return nParam[kXSecPar]; };

GetParamIndexFromName()

int MCMCProcessor::GetParamIndexFromName

(

const std::string &

Name

)

Get parameter number based on name.

Definition at line 2709 of file MCMCProcessor.cpp.

                                                             {
// **************************
  int ParamNo = M3::_BAD_INT_;
  for (int i = 0; i < nDraw; ++i)
  {
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(i, Prior, PriorError, Title);
 
    if(Name == Title)
    {
      ParamNo = i;
      break;
    }
  }
  return ParamNo;
}

GetPolarPlot()

void MCMCProcessor::GetPolarPlot

(

const std::vector< std::string > &

ParNames

)

Make funny polar plot.

Parameters

ParNames

Vector with parameter names for which Polar Plot will be made

Definition at line 2747 of file MCMCProcessor.cpp.

                                                                    {
// **************************
  if(hpost[0] == nullptr) MakePostfit();
 
  std::vector<double> Margins = GetMargins(Posterior);
 
  Posterior->SetTopMargin(0.1);
  Posterior->SetBottomMargin(0.1);
  Posterior->SetLeftMargin(0.1);
  Posterior->SetRightMargin(0.1);
  Posterior->Update();
 
  MACH3LOG_INFO("Calculating Polar Plot");
  TDirectory *PolarDir = OutputFile->mkdir("PolarDir");
  PolarDir->cd();
 
  for(unsigned int k = 0; k < ParNames.size(); ++k)
  {
    //KS: First we need to find parameter number based on name
    int ParamNo = GetParamIndexFromName(ParNames[k]);
    if(ParamNo == M3::_BAD_INT_)
    {
      MACH3LOG_WARN("Couldn't find param {}. Will not calculate Polar Plot", ParNames[k]);
      continue;
    }
 
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(ParamNo, Prior, PriorError, Title);
 
    std::vector<double> x_val(nBins);
    std::vector<double> y_val(nBins);
 
    constexpr double xmin = 0;
    constexpr double xmax = 2*TMath::Pi();
 
    double Integral = hpost[ParamNo]->Integral();
    for (Int_t ipt = 0; ipt < nBins; ipt++)
    {
      x_val[ipt] = ipt*(xmax-xmin)/nBins+xmin;
      y_val[ipt] = hpost[ParamNo]->GetBinContent(ipt+1)/Integral;
    }
 
    auto PolarGraph = std::make_unique<TGraphPolar>(nBins, x_val.data(), y_val.data());
    PolarGraph->SetLineWidth(2);
    PolarGraph->SetFillStyle(3001);
    PolarGraph->SetLineColor(kRed);
    PolarGraph->SetFillColor(kRed);
    PolarGraph->Draw("AFL");
 
    auto Text = std::make_unique<TText>(0.6, 0.1, Title);
    Text->SetTextSize(0.04);
    Text->SetNDC(true);
    Text->Draw("");
 
    Posterior->Print(CanvasName);
    Posterior->Write(Title);
  } //End loop over parameters
 
  PolarDir->Close();
  delete PolarDir;
 
  OutputFile->cd();
 
  SetMargins(Posterior, Margins);
}

GetPostfit()

void MCMCProcessor::GetPostfit	(	TVectorD *&	Central,
		TVectorD *&	Errors,
		TVectorD *&	Central_Gauss,
		TVectorD *&	Errors_Gauss,
		TVectorD *&	Peaks
	)

Get the post-fit results (arithmetic and Gaussian)

Definition at line 163 of file MCMCProcessor.cpp.

                                                                                                                                               {
// ***************
  // Make the post fit
  MakePostfit();
 
  // We now have the private members
  Central_PDF = Means;
  Errors_PDF = Errors;
  Central_G = Means_Gauss;
  Errors_G = Errors_Gauss;
  Peak_Values = Means_HPD;
}

GetPostfit_Ind()

void MCMCProcessor::GetPostfit_Ind	(	TVectorD *&	Central,
		TVectorD *&	Errors,
		TVectorD *&	Peaks,
		ParameterEnum	kParam
	)

Or the individual post-fits.

Definition at line 178 of file MCMCProcessor.cpp.

                                                                                                                              {
// ***************
  // Make the post fit
  MakePostfit();
 
  // Loop over the loaded param types
  const int ParamTypeSize = int(ParamType.size());
  int ParamNumber = 0;
  for (int i = 0; i < ParamTypeSize; ++i) {
    if (ParamType[i] != kParam) continue;
    (*PDF_Central)(ParamNumber) = (*Means)(i);
    (*PDF_Errors)(ParamNumber) = (*Errors)(i);
    (*Peak_Values)(ParamNumber) = (*Means_HPD)(i);
    ++ParamNumber;
  }
}

GetSavageDickey()

void MCMCProcessor::GetSavageDickey	(	const std::vector< std::string > &	ParName,
		const std::vector< double > &	EvaluationPoint,
		const std::vector< std::vector< double >> &	Bounds
	)

Calculate Bayes factor for point like hypothesis using SavageDickey.

Definition at line 2873 of file MCMCProcessor.cpp.

                                                                               {
// **************************
  if((ParNames.size() != EvaluationPoint.size()) || (Bounds.size() != EvaluationPoint.size()))
  {
    MACH3LOG_ERROR("Size doesn't match");
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  
  if(hpost[0] == nullptr) MakePostfit();
 
  MACH3LOG_INFO("Calculating Savage Dickey");
  TDirectory *SavageDickeyDir = OutputFile->mkdir("SavageDickey");
  SavageDickeyDir->cd();
  
  for(unsigned int k = 0; k < ParNames.size(); ++k)
  {
    //KS: First we need to find parameter number based on name
    int ParamNo = GetParamIndexFromName(ParNames[k]);
    if(ParamNo == M3::_BAD_INT_)
    {
      MACH3LOG_WARN("Couldn't find param {}. Will not calculate SavageDickey", ParNames[k]);
      continue;
    }
    
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    bool FlatPrior = false;
    GetNthParameter(ParamNo, Prior, PriorError, Title);
    
    ParameterEnum ParType = ParamType[ParamNo];
    int ParamTemp = ParamNo - ParamTypeStartPos[ParType];
    FlatPrior = ParamFlat[ParType][ParamTemp];
    
    auto PosteriorHist = M3::Clone<TH1D>(hpost[ParamNo], std::string(Title));
    RemoveFitter(PosteriorHist.get(), "Gauss");
            
    std::unique_ptr<TH1D> PriorHist;
    //KS: If flat prior we need to have well defined bounds otherwise Prior distribution will not make sense
    if(FlatPrior)
    {
      int NBins = PosteriorHist->GetNbinsX();
      if(Bounds[k][0] > Bounds[k][1])
      {
        MACH3LOG_ERROR("Lower bound is higher than upper bound");
        throw MaCh3Exception(__FILE__ , __LINE__ );
      }
      PriorHist = std::make_unique<TH1D>("PriorHist", Title, NBins, Bounds[k][0], Bounds[k][1]);
      
      double FlatProb = ( Bounds[k][1] - Bounds[k][0]) / NBins;
      for (int g = 0; g < NBins + 1; ++g) 
      {
        PriorHist->SetBinContent(g+1, FlatProb);
      }
    }
    else //KS: Otherwise throw from Gaussian
    {
      PriorHist = M3::Clone<TH1D>(PosteriorHist.get(), "Prior");
      PriorHist->Reset("");
      PriorHist->Fill(0.0, 0.0);
      
      auto rand = std::make_unique<TRandom3>(0);
      //KS: Throw nice gaussian, just need big number to have smooth distribution
      for(int g = 0; g < 1000000; ++g)
      {
        PriorHist->Fill(rand->Gaus(Prior, PriorError));
      }
    }
    SavageDickeyPlot(PriorHist, PosteriorHist, std::string(Title), EvaluationPoint[k]);
  } //End loop over parameters
 
  SavageDickeyDir->Close();
  delete SavageDickeyDir;
 
  OutputFile->cd();
}

GetViolin()

TH2D* MCMCProcessor::GetViolin ( ) const

inline

Get Violin plot for all parameters with posterior values.

Definition at line 229 of file MCMCProcessor.h.

{ return hviolin.get(); };

GetViolinPrior()

TH2D* MCMCProcessor::GetViolinPrior ( ) const

inline

Get Violin plot for all parameters with prior values.

Definition at line 231 of file MCMCProcessor.h.

{ return hviolin_prior.get(); };

GetXSecCov()

std::vector<std::string> MCMCProcessor::GetXSecCov ( ) const

inline

Definition at line 234 of file MCMCProcessor.h.

{ return CovPos[kXSecPar]; };

GewekeDiagnostic()

void MCMCProcessor::GewekeDiagnostic ( )

inlineprotected

Geweke Diagnostic based on the methods described by Fang (2014) and Karlsbakk (2011). [8] [20].

Definition at line 4296 of file MCMCProcessor.cpp.

                                     {
// **************************
  MACH3LOG_INFO("Making Geweke Diagnostic");
  //KS: Up refers to upper limit we check, it stays constant, in literature it is mostly 50% thus using 0.5 for threshold
  std::vector<double> MeanUp(nDraw, 0.0);
  std::vector<double> SpectralVarianceUp(nDraw, 0.0);
  std::vector<int> DenomCounterUp(nDraw, 0);
  const double Threshold = 0.5 * nSteps;
 
  //KS: Select values between which you want to scan, for example 0 means 0% burn in and 1 100% burn in.
  constexpr double LowerThreshold = 0;
  constexpr double UpperThreshold = 1.0;
  // Tells how many intervals between thresholds we want to check
  constexpr int NChecks = 100;
  constexpr double Division = (UpperThreshold - LowerThreshold)/NChecks;
 
  std::vector<std::unique_ptr<TH1D>> GewekePlots(nDraw);
  for (int j = 0; j < nDraw; ++j)
  {
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(j, Prior, PriorError, Title);
    std::string HistName = Form("%s_%s_Geweke", Title.Data(), BranchNames[j].Data());
    GewekePlots[j] = std::make_unique<TH1D>(HistName.c_str(), HistName.c_str(), NChecks, 0.0, 100 * UpperThreshold);
    GewekePlots[j]->GetXaxis()->SetTitle("Burn-In (%)");
    GewekePlots[j]->GetYaxis()->SetTitle("Geweke T score");
  }
 
//KS: Start parallel region
#ifdef MULTITHREAD
#pragma omp parallel
{
#endif
  //KS: First we calculate mean and spectral variance for the upper limit, this doesn't change and in literature is most often 50%
  #ifdef MULTITHREAD
  #pragma omp for
  #endif
  for (int j = 0; j < nDraw; ++j)
  {
    for(int i = 0; i < nEntries; ++i)
    {
      if(StepNumber[i] > Threshold)
      {
        MeanUp[j] += ParStep[j][i];
        DenomCounterUp[j]++;
      }
    }
    MeanUp[j] = MeanUp[j]/DenomCounterUp[j];
  }
 
  //KS: now Spectral variance which in this case is sample variance
  #ifdef MULTITHREAD
  #pragma omp for collapse(2)
  #endif
  for (int j = 0; j < nDraw; ++j)
  {
    for(int i = 0; i < nEntries; ++i)
    {
      if(StepNumber[i] > Threshold)
      {
        SpectralVarianceUp[j] += (ParStep[j][i] - MeanUp[j])*(ParStep[j][i] - MeanUp[j]);
      }
    }
  }
 
  //Loop over how many intervals we calculate
  #ifdef MULTITHREAD
  #pragma omp for
  #endif
  for (int k = 1; k < NChecks+1; ++k)
  {
    //KS each thread has it's own
    std::vector<double> MeanDown(nDraw, 0.0);
    std::vector<double> SpectralVarianceDown(nDraw, 0.0);
    std::vector<int> DenomCounterDown(nDraw, 0);
 
    const unsigned int ThresholsCheck = Division*k*nSteps;
    //KS: First mean
    for (int j = 0; j < nDraw; ++j)
    {
      for(int i = 0; i < nEntries; ++i)
      {
        if(StepNumber[i] < ThresholsCheck)
        {
          MeanDown[j] += ParStep[j][i];
          DenomCounterDown[j]++;
        }
      }
      MeanDown[j] = MeanDown[j]/DenomCounterDown[j];
    }
    //Now spectral variance
    for (int j = 0; j < nDraw; ++j)
    {
      for(int i = 0; i < nEntries; ++i)
      {
        if(StepNumber[i] < ThresholsCheck)
        {
          SpectralVarianceDown[j] += (ParStep[j][i] - MeanDown[j])*(ParStep[j][i] - MeanDown[j]);
        }
      }
    }
    //Lastly calc T score and fill histogram entry
    for (int j = 0; j < nDraw; ++j)
    {
      double T_score = std::fabs((MeanDown[j] - MeanUp[j])/std::sqrt(SpectralVarianceDown[j]/DenomCounterDown[j] + SpectralVarianceUp[j]/DenomCounterUp[j]));
      GewekePlots[j]->SetBinContent(k, T_score);
    }
  } //end loop over intervals
#ifdef MULTITHREAD
} //End parallel region
#endif
 
  //Finally save it to TFile
  OutputFile->cd();
  TDirectory *GewekeDir = OutputFile->mkdir("Geweke");
  for (int j = 0; j < nDraw; ++j)
  {
    GewekeDir->cd();
    GewekePlots[j]->Write();
  }
  for (int i = 0; i < nDraw; ++i) {
    delete[] ParStep[i];
  }
  delete[] ParStep;
 
  GewekeDir->Close();
  delete GewekeDir;
  OutputFile->cd();
}

Initialise()

void MCMCProcessor::Initialise

(

)

Scan chain, what parameters we have and load information from covariance matrices.

Definition at line 153 of file MCMCProcessor.cpp.

                              {
// ***************
  // Scan the ROOT file for useful branches
  ScanInput();
 
  // Setup the output
  SetupOutput();
}

LoadAdditionalInfo()

virtual void MCMCProcessor::LoadAdditionalInfo ( )

inlineprotectedvirtual

allow loading additional info for example used for oscillation parameters

Reimplemented in OscProcessor.

Definition at line 343 of file MCMCProcessor.h.

{};

MakeCovariance()

void MCMCProcessor::MakeCovariance

(

)

Calculate covariance by making 2D projection of each combination of parameters.

Warning

This is deprecated and slower, however it is less RAM intensive

Definition at line 959 of file MCMCProcessor.cpp.

                                   {
// *********************
  if (OutputFile == nullptr) MakeOutputFile();
 
  bool HaveMadeDiagonal = false;
  MACH3LOG_INFO("Making post-fit covariances...");
  // Check that the diagonal entries have been filled
  // i.e. MakePostfit() has been called
  for (int i = 0; i < nDraw; ++i) {
    if ((*Covariance)(i,i) == M3::_BAD_DOUBLE_) {
      HaveMadeDiagonal = false;
      MACH3LOG_INFO("Have not run diagonal elements in covariance, will do so now by calling MakePostfit()");
      break;
    } else {
      HaveMadeDiagonal = true;
    }
  }
 
  if (HaveMadeDiagonal == false) {
    MakePostfit();
  }
 
  TDirectory *PostHistDir = OutputFile->mkdir("Post_2d_hists");
  PostHistDir->cd();
  gStyle->SetPalette(55);
  // Now we are sure we have the diagonal elements, let's make the off-diagonals
  for (int i = 0; i < nDraw; ++i)
  {
    if (i % (nDraw/5) == 0)
      MaCh3Utils::PrintProgressBar(i, nDraw);
 
    TString Title_i = "";
    double Prior_i, PriorError;
 
    GetNthParameter(i, Prior_i, PriorError, Title_i);
    
    // Loop over the other parameters to get the correlations
    for (int j = 0; j <= i; ++j) {
      // Skip the diagonal elements which we've already done above
      if (j == i) continue;
 
      // If this parameter isn't varied
      if (IamVaried[j] == false) {
        (*Covariance)(i,j) = 0.0;
        (*Covariance)(j,i) = (*Covariance)(i,j);
        (*Correlation)(i,j) = 0.0;
        (*Correlation)(j,i) = (*Correlation)(i,j);
        continue;
      }
 
      TString Title_j = "";
      double Prior_j, PriorError_j;
      GetNthParameter(j, Prior_j, PriorError_j, Title_j);
 
      OutputFile->cd();
 
      // The draw which we want to perform
      TString DrawMe = BranchNames[j]+":"+BranchNames[i];
 
      // TH2F to hold the Correlation 
      auto hpost_2D = std::make_unique<TH2D>(DrawMe, DrawMe,
                      nBins, hpost[i]->GetXaxis()->GetXmin(), hpost[i]->GetXaxis()->GetXmax(),
                      nBins, hpost[j]->GetXaxis()->GetXmin(), hpost[j]->GetXaxis()->GetXmax());
      hpost_2D->SetMinimum(0);
      hpost_2D->GetXaxis()->SetTitle(Title_i);
      hpost_2D->GetYaxis()->SetTitle(Title_j);
      hpost_2D->GetZaxis()->SetTitle("Steps");
 
      std::string SelectionStr = StepCut;
      if (ReweightPosterior) {
        SelectionStr = "(" + StepCut + ")*(" + ReweightName + ")";
      }
      // The draw command we want, i.e. draw param j vs param i
      Chain->Project(DrawMe, DrawMe, SelectionStr.c_str());
      
      if(ApplySmoothing) hpost_2D->Smooth();
      // Get the Covariance for these two parameters
      (*Covariance)(i,j) = hpost_2D->GetCovariance();
      (*Covariance)(j,i) = (*Covariance)(i,j);
 
      (*Correlation)(i,j) = hpost_2D->GetCorrelationFactor();
      (*Correlation)(j,i) = (*Correlation)(i,j);
 
      if(printToPDF)
      {
        //KS: Skip Flux Params
        if(ParamType[i] == kXSecPar && ParamType[j] == kXSecPar)
        {
          if(std::fabs((*Correlation)(i,j)) > Post2DPlotThreshold)
          {
            Posterior->cd();
            hpost_2D->Draw("colz");
            Posterior->SetName(hpost_2D->GetName());
            Posterior->SetTitle(hpost_2D->GetTitle());
            Posterior->Print(CanvasName);
            hpost2D[i][j]->Write(hpost2D[i][j]->GetTitle());
          }
        }
      }
      // Write it to root file
      //OutputFile->cd();
      //if( std::fabs((*Correlation)(i,j)) > Post2DPlotThreshold ) hpost_2D->Write();
    } // End j loop
  } // End i loop
  PostHistDir->Close();
  delete PostHistDir;
  OutputFile->cd();
  Covariance->Write("Covariance");
  Correlation->Write("Correlation");
}

MakeCovariance_MP()

void MCMCProcessor::MakeCovariance_MP

(

const bool

Mute = false

)

Calculate covariance by making 2D projection of each combination of parameters using multithreading.

Parameters

Mute	Allow silencing many messages, especially important if we calculate matrix many times

Definition at line 1200 of file MCMCProcessor.cpp.

                                                     {
// *********************
  if (OutputFile == nullptr) MakeOutputFile();
    
  if(!CacheMCMC) CacheSteps();
  
  bool HaveMadeDiagonal = false;    
  // Check that the diagonal entries have been filled
  // i.e. MakePostfit() has been called
  for (int i = 0; i < nDraw; ++i) {
    if ((*Covariance)(i,i) == M3::_BAD_DOUBLE_) {
      HaveMadeDiagonal = false;
      MACH3LOG_WARN("Have not run diagonal elements in covariance, will do so now by calling MakePostfit()");
      break;
    } else {
      HaveMadeDiagonal = true;
    }
  }
    
  if (HaveMadeDiagonal == false) MakePostfit();
  TStopwatch clock;
  TDirectory *PostHistDir = nullptr;
  if(!Mute)
  {
    MACH3LOG_INFO("Calculating covariance matrix");
    clock.Start();
    PostHistDir = OutputFile->mkdir("Post_2d_hists");
    PostHistDir->cd();
  }
 
  if(!Mute)
 
  gStyle->SetPalette(55);
  // Now we are sure we have the diagonal elements, let's make the off-diagonals
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int i = 0; i < nDraw; ++i)
  {    
    for (int j = 0; j <= i; ++j)
    {
      // Skip the diagonal elements which we've already done above
      if (j == i) continue;
      
      // If this parameter isn't varied
      if (IamVaried[j] == false) {
        (*Covariance)(i,j) = 0.0;
        (*Covariance)(j,i) = (*Covariance)(i,j);
        (*Correlation)(i,j) = 0.0;
        (*Correlation)(j,i) = (*Correlation)(i,j);
        continue;
      }
      hpost2D[i][j]->SetMinimum(0);
          
      for (int k = 0; k < nEntries; ++k)
      {
        //KS: Burn in cut
        if(StepNumber[k] < BurnInCut) continue;
 
        const double Weight = ReweightPosterior ? WeightValue[i] : 1.;
        //KS: Fill histogram with cached steps
        hpost2D[i][j]->Fill(ParStep[i][k], ParStep[j][k], Weight);
      }
      if(ApplySmoothing) hpost2D[i][j]->Smooth();
      
      // Get the Covariance for these two parameters
      (*Covariance)(i,j) = hpost2D[i][j]->GetCovariance();
      (*Covariance)(j,i) = (*Covariance)(i,j);
 
      //KS: Since we already have covariance consider calculating correlation using it, right now we effectively calculate covariance twice
      //https://root.cern.ch/doc/master/TH2_8cxx_source.html#l01099
      (*Correlation)(i,j) = hpost2D[i][j]->GetCorrelationFactor();
      (*Correlation)(j,i) = (*Correlation)(i,j);
    }// End j loop
  }// End i loop
 
  if(!Mute) {
    clock.Stop();
    MACH3LOG_INFO("Making Covariance took {:.2f}s to finish for {} steps", clock.RealTime(), nEntries);
    if(printToPDF)
    {
      Posterior->cd();
      for (int i = 0; i < nDraw; ++i)
      {
        for (int j = 0; j <= i; ++j)
        {
          // Skip the diagonal elements which we've already done above
          if (j == i) continue;
          if (IamVaried[j] == false) continue;
 
          if(ParamType[i] == kXSecPar && ParamType[j] == kXSecPar)
          {
            if(std::fabs((*Correlation)(i,j)) > Post2DPlotThreshold)
            {
              hpost2D[i][j]->Draw("colz");
              Posterior->SetName(hpost2D[i][j]->GetName());
              Posterior->SetTitle(hpost2D[i][j]->GetTitle());
              Posterior->Print(CanvasName);
              hpost2D[i][j]->Write(hpost2D[i][j]->GetTitle());
            }
          }
          //if( std::fabs((*Correlation)(i,j)) > Post2DPlotThreshold) hpost2D[i][j]->Write();
        }// End j loop
      }// End i loop
    } //end if pdf
    PostHistDir->Close();
    delete PostHistDir;
    OutputFile->cd();
    Covariance->Write("Covariance");
    Correlation->Write("Correlation");
  } // end if not mute
}

MakeCovarianceYAML()

void MCMCProcessor::MakeCovarianceYAML	(	const std::string &	OutputYAMLFile,
		const std::string &	MeansMethod
	)		const

Make YAML file from post-fit covariance.

Definition at line 1478 of file MCMCProcessor.cpp.

                                                                                                          {
// *********************
  MACH3LOG_INFO("Making covariance matrix YAML file");
 
  if (ParamNames[kXSecPar].size() != static_cast<size_t>(nDraw)) {
    MACH3LOG_ERROR("Using Legacy Parameters i.e. not one from Parameter Handler Generic, this will not work");
    throw MaCh3Exception(__FILE__, __LINE__);
  }
  std::vector<double> MeanArray(nDraw);
  std::vector<double> ErrorArray(nDraw);
  std::vector<std::vector<double>> CorrelationMatrix(nDraw, std::vector<double>(nDraw, 0.0));
 
  TVectorD* means_vec;
  TVectorD* errors_vec;
 
  if (MeansMethod == "Arithmetic") {
    means_vec = Means;
    errors_vec = Errors;
  } else if (MeansMethod == "Gaussian") {
    means_vec = Means_Gauss;
    errors_vec = Errors_Gauss;
  } else if (MeansMethod == "HPD") {
    means_vec = Means_HPD;
    errors_vec = Errors_HPD;
  } else {
    MACH3LOG_ERROR("Unknown means method: {}, should be either 'Arithmetic', 'Gaussian', or 'HPD'.", MeansMethod);
    throw MaCh3Exception(__FILE__, __LINE__);
  }
 
  //Make vectors of mean, error, and correlations
  for (int i = 0; i < nDraw; i++)
  {
    MeanArray[i]  = (*means_vec)(i);
    ErrorArray[i] = (*errors_vec)(i);
    for (int j = 0; j <= i; j++)
    {
      CorrelationMatrix[i][j] = (*Correlation)(i,j);
      if(i != j) CorrelationMatrix[j][i] = (*Correlation)(i,j);
    }
  }
 
  //Make std::string param name vector
  std::vector<std::string> ParamStrings(ParamNames[kXSecPar].size());
  for (size_t i = 0; i < ParamNames[kXSecPar].size(); ++i) {
    ParamStrings[i] = static_cast<std::string>(ParamNames[kXSecPar][i]);
  }
 
  YAML::Node XSecFile = CovConfig[kXSecPar];
  M3::MakeCorrelationMatrix(XSecFile, MeanArray, ErrorArray, CorrelationMatrix, OutputYAMLFile, ParamStrings);
}

MakeCredibleIntervals()

void MCMCProcessor::MakeCredibleIntervals	(	const std::vector< double > &	CredibleIntervals = {0.99, 0.90, 0.68 },
		const std::vector< Color_t > &	CredibleIntervalsColours = {kCyan+4, kCyan-2, kCyan-10},
		const bool	CredibleInSigmas = false
	)

Make and Draw Credible intervals.

Parameters

CredibleIntervals	Vector with values of credible intervals, must be in descending order
CredibleIntervalsColours	Color_t telling what colour to use for each Interval line
CredibleInSigmas	Bool telling whether intervals are in percentage or in sigmas, then special conversions is used

Scale the histograms so it shows the posterior probability

Definition at line 701 of file MCMCProcessor.cpp.

                                                                       {
// *********************
  if(hpost[0] == nullptr) MakePostfit();
 
  MACH3LOG_INFO("Making Credible Intervals");
  const double LeftMargin = Posterior->GetLeftMargin();
  Posterior->SetLeftMargin(0.15);
 
  // KS: Sanity check of size and ordering is correct
  CheckCredibleIntervalsOrder(CredibleIntervals, CredibleIntervalsColours);
  const int nCredible = int(CredibleIntervals.size());
  std::vector<std::unique_ptr<TH1D>> hpost_copy(nDraw);
  std::vector<std::vector<std::unique_ptr<TH1D>>> hpost_cl(nDraw);
 
  //KS: Copy all histograms to be thread safe
  for (int i = 0; i < nDraw; ++i)
  {
    hpost_copy[i] = M3::Clone<TH1D>(hpost[i], Form("hpost_copy_%i", i));
    hpost_cl[i].resize(nCredible);
    for (int j = 0; j < nCredible; ++j)
    {
      hpost_cl[i][j] = M3::Clone<TH1D>(hpost[i], Form("hpost_copy_%i_CL_%f", i, CredibleIntervals[j]));
 
      //KS: Reset to get rid to TF1 otherwise we run into segfault :(
      hpost_cl[i][j]->Reset("");
      hpost_cl[i][j]->Fill(0.0, 0.0);
    }
  }
 
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int i = 0; i < nDraw; ++i)
  {
    hpost_copy[i]->Scale(1. / hpost_copy[i]->Integral());
    for (int j = 0; j < nCredible; ++j)
    {
      // Scale the histograms before gettindg credible intervals
      hpost_cl[i][j]->Scale(1. / hpost_cl[i][j]->Integral());
      GetCredibleIntervalSig(hpost_copy[i], hpost_cl[i][j], CredibleInSigmas, CredibleIntervals[j]);
 
      hpost_cl[i][j]->SetFillColor(CredibleIntervalsColours[j]);
      hpost_cl[i][j]->SetLineWidth(1);
    }
    hpost_copy[i]->GetYaxis()->SetTitleOffset(1.8);
    hpost_copy[i]->SetLineWidth(1);
    hpost_copy[i]->SetMaximum(hpost_copy[i]->GetMaximum()*1.2);
    hpost_copy[i]->SetLineWidth(2);
    hpost_copy[i]->SetLineColor(kBlack);
    hpost_copy[i]->GetYaxis()->SetTitle("Posterior Probability");
  }
 
  OutputFile->cd();
  TDirectory *CredibleDir = OutputFile->mkdir("Credible");
 
  for (int i = 0; i < nDraw; ++i)
  {
    if(!IamVaried[i]) continue;
 
    // Now make the TLine for the Asimov
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(i, Prior, PriorError, Title);
 
    auto Asimov = std::make_unique<TLine>(Prior, hpost_copy[i]->GetMinimum(), Prior, hpost_copy[i]->GetMaximum());
    SetTLineStyle(Asimov.get(), kRed-3, 2, kDashed);
 
    auto legend = std::make_unique<TLegend>(0.20, 0.7, 0.4, 0.92);
    SetLegendStyle(legend.get(), 0.03);
    hpost_copy[i]->Draw("HIST");
 
    for (int j = 0; j < nCredible; ++j)
        hpost_cl[i][j]->Draw("HIST SAME");
    for (int j = nCredible-1; j >= 0; --j)
    {
      if(CredibleInSigmas)
        legend->AddEntry(hpost_cl[i][j].get(), Form("%.0f#sigma Credible Interval", CredibleIntervals[j]), "f");
      else
        legend->AddEntry(hpost_cl[i][j].get(), Form("%.0f%% Credible Interval", CredibleIntervals[j]*100), "f");
    }
    legend->AddEntry(Asimov.get(), Form("#splitline{Prior}{x = %.2f , #sigma = %.2f}", Prior, PriorError), "l");
    legend->Draw("SAME");
    Asimov->Draw("SAME");
 
    // Write to file
    Posterior->SetName(hpost[i]->GetName());
    Posterior->SetTitle(hpost[i]->GetTitle());
 
    if(printToPDF) Posterior->Print(CanvasName);
    // cd into directory in root file
    CredibleDir->cd();
    Posterior->Write();
  }
  CredibleDir->Close();
  delete CredibleDir;
 
  OutputFile->cd();
 
  //Set back to normal
  Posterior->SetLeftMargin(LeftMargin);
}

MakeCredibleRegions()

void MCMCProcessor::MakeCredibleRegions	(	const std::vector< double > &	CredibleRegions = {0.99, 0.90, 0.68},
		const std::vector< Style_t > &	CredibleRegionStyle = {kDashed, kSolid, kDotted},
		const std::vector< Color_t > &	CredibleRegionColor = {kGreen-3, kGreen-10, kGreen},
		const bool	CredibleInSigmas = false,
		const bool	Draw2DPosterior = true,
		const bool	DrawBestFit = true
	)

Make and Draw Credible Regions.

Parameters

CredibleRegions	Vector with values of credible intervals, must be in descending order
CredibleRegionStyle	Style_t telling what line style to use for each Interval line
CredibleRegionColor	Color_t telling what colour to use for each Interval line
CredibleInSigmas	Bool telling whether intervals are in percentage or in sigmas, then special conversions is used

Definition at line 1776 of file MCMCProcessor.cpp.

                                                                {
// *********************
  if(hpost2D.size() == 0) MakeCovariance_MP();
  MACH3LOG_INFO("Making Credible Regions");
 
  CheckCredibleRegionsOrder(CredibleRegions, CredibleRegionStyle, CredibleRegionColor);
  const int nCredible = int(CredibleRegions.size());
 
  std::vector<std::vector<std::unique_ptr<TH2D>>> hpost_2D_copy(nDraw);
  std::vector<std::vector<std::vector<std::unique_ptr<TH2D>>>> hpost_2D_cl(nDraw);
  //KS: Copy all histograms to be thread safe
  for (int i = 0; i < nDraw; ++i)
  {
    hpost_2D_copy[i].resize(nDraw);
    hpost_2D_cl[i].resize(nDraw);
    for (int j = 0; j <= i; ++j)
    {
      hpost_2D_copy[i][j] = M3::Clone<TH2D>(hpost2D[i][j], Form("hpost_copy_%i_%i", i, j));
      hpost_2D_cl[i][j].resize(nCredible);
      for (int k = 0; k < nCredible; ++k)
      {
        hpost_2D_cl[i][j][k] = M3::Clone<TH2D>(hpost2D[i][j], Form("hpost_copy_%i_%i_CL_%f", i, j, CredibleRegions[k]));
      }
    }
  }
 
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  //Calculate credible histogram
  for (int i = 0; i < nDraw; ++i)
  {
    for (int j = 0; j <= i; ++j)
    {
      for (int k = 0; k < nCredible; ++k)
      {
        GetCredibleRegionSig(hpost_2D_cl[i][j][k], CredibleInSigmas, CredibleRegions[k]);
        hpost_2D_cl[i][j][k]->SetLineColor(CredibleRegionColor[k]);
        hpost_2D_cl[i][j][k]->SetLineWidth(2);
        hpost_2D_cl[i][j][k]->SetLineStyle(CredibleRegionStyle[k]);
      }
    }
  }
 
  gStyle->SetPalette(51);
  for (int i = 0; i < nDraw; ++i)
  {
    for (int j = 0; j <= i; ++j)
    {
      // Skip the diagonal elements which we've already done above
      if (j == i) continue;
      if (IamVaried[j] == false) continue;
 
      auto legend = std::make_unique<TLegend>(0.20, 0.7, 0.4, 0.92);
      legend->SetTextColor(kRed);
      SetLegendStyle(legend.get(), 0.03);
 
      //Get Best point
      auto bestfitM = std::make_unique<TGraph>(1);
      const int MaxBin = hpost_2D_copy[i][j]->GetMaximumBin();
      int Mbx, Mby, Mbz;
      hpost_2D_copy[i][j]->GetBinXYZ(MaxBin, Mbx, Mby, Mbz);
      const double Mx = hpost_2D_copy[i][j]->GetXaxis()->GetBinCenter(Mbx);
      const double My = hpost_2D_copy[i][j]->GetYaxis()->GetBinCenter(Mby);
 
      bestfitM->SetPoint(0, Mx, My);
      bestfitM->SetMarkerStyle(22);
      bestfitM->SetMarkerSize(1);
      bestfitM->SetMarkerColor(kMagenta);
    
      //Plot default 2D posterior
 
      if(Draw2DPosterior){
        hpost_2D_copy[i][j]->Draw("COLZ");
      } else{
        hpost_2D_copy[i][j]->Draw("AXIS");
      }
 
      //Now credible regions
      for (int k = 0; k < nCredible; ++k)
        hpost_2D_cl[i][j][k]->Draw("CONT3 SAME");
      for (int k = nCredible-1; k >= 0; --k)
      {
        if(CredibleInSigmas)
          legend->AddEntry(hpost_2D_cl[i][j][k].get(), Form("%.0f#sigma Credible Interval", CredibleRegions[k]), "l");
        else
          legend->AddEntry(hpost_2D_cl[i][j][k].get(), Form("%.0f%% Credible Region", CredibleRegions[k]*100), "l");
      }
      legend->Draw("SAME");
  
    if(DrawBestFit){
      legend->AddEntry(bestfitM.get(),"Best Fit","p");
      bestfitM->Draw("SAME.P");
     }
 
      // Write to file
      Posterior->SetName(hpost2D[i][j]->GetName());
      Posterior->SetTitle(hpost2D[i][j]->GetTitle());
 
      //KS: Print only regions with correlation greater than specified value, by default 0.2. This is done to avoid dumping thousands of plots
      if(printToPDF && std::fabs((*Correlation)(i,j)) > Post2DPlotThreshold) Posterior->Print(CanvasName);
      // Write it to root file
      //OutputFile->cd();
      //if( std::fabs((*Correlation)(i,j)) > Post2DPlotThreshold ) Posterior->Write();
    }
  }
 
  OutputFile->cd();
}

MakeOutputFile()

void MCMCProcessor::MakeOutputFile ( )

inlineprotected

prepare output root file and canvas to which we will save EVERYTHING

Definition at line 205 of file MCMCProcessor.cpp.

                                   {
// ***************
  //KS: ROOT hates me... but we can create several instances of MCMC Processor, each with own TCanvas ROOT is mad and will delete if there is more than one canvas with the same name, so we add random number to avoid issue
  auto rand = std::make_unique<TRandom3>(0);
  const int uniform = int(rand->Uniform(0, 10000));
  // Open a TCanvas to write the posterior onto
  Posterior = std::make_unique<TCanvas>(("Posterior" + std::to_string(uniform)).c_str(), ("Posterior" + std::to_string(uniform)).c_str(), 0, 0, 1024, 1024);
  //KS: No idea why but ROOT changed treatment of violin in R6. If you have non uniform binning this will results in very hard to see violin plots.
  TCandle::SetScaledViolin(false);
 
  Posterior->SetGrid();
  gStyle->SetOptStat(0);
  gStyle->SetOptTitle(0);
  Posterior->SetTickx();
  Posterior->SetTicky();
 
  Posterior->SetBottomMargin(0.1);
  Posterior->SetTopMargin(0.05);
  Posterior->SetRightMargin(0.03);
  Posterior->SetLeftMargin(0.15);
 
  //To avoid TCanvas::Print> messages
  gErrorIgnoreLevel = kWarning;
  
  // Output file to write to
  OutputName = MCMCFile + OutputSuffix +".root";
 
  // Output file
  OutputFile = new TFile(OutputName.c_str(), "recreate");
  OutputFile->cd();
}

MakePostfit()

void MCMCProcessor::MakePostfit

(

const std::map< std::string, std::pair< double, double >> &

Edges = {}

)

Make 1D projection for each parameter and prepare structure.

Definition at line 239 of file MCMCProcessor.cpp.

                                                                                       {
// ****************************
  // Check if we've already made post-fit
  if (MadePostfit == true) return;
  MadePostfit = true;
 
  // Check if the output file is ready
  if (OutputFile == nullptr) MakeOutputFile();
  
  MACH3LOG_INFO("MCMCProcessor is making post-fit plots...");
 
  int originalErrorLevel = gErrorIgnoreLevel;
  gErrorIgnoreLevel = kFatal;
 
  // Directory for posteriors
  TDirectory *PostDir = OutputFile->mkdir("Post");
  TDirectory *PostHistDir = OutputFile->mkdir("Post_1d_hists");
  
  //KS: Apply additional Cuts, like mass ordering
  std::string CutPosterior1D = "";
  if(Posterior1DCut != "") {
    CutPosterior1D = StepCut +" && " + Posterior1DCut;
  } else CutPosterior1D = StepCut;
 
  // Apply reweighting
  if (ReweightPosterior) {
    CutPosterior1D = "(" + CutPosterior1D + ")*(" + ReweightName + ")";
  }
  MACH3LOG_DEBUG("Using following cut {}", CutPosterior1D);
  
  // nDraw is number of draws we want to do
  for (int i = 0; i < nDraw; ++i)
  {
    if (i % (nDraw/5) == 0) {
      MaCh3Utils::PrintProgressBar(i, nDraw);
    }
    OutputFile->cd();
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(i, Prior, PriorError, Title);
    // Get bin edges for histograms
    double maxi, mini = M3::_BAD_DOUBLE_;
    if (Edges.find(Title.Data()) != Edges.end()) {
      mini = Edges.at(Title.Data()).first;
      maxi = Edges.at(Title.Data()).second;
    } else {
      maxi = Chain->GetMaximum(BranchNames[i]);
      mini = Chain->GetMinimum(BranchNames[i]);
    }
    MACH3LOG_DEBUG("Initialising histogram for {} with binning {:.4f}, {:.4f}", Title, mini, maxi);
    // This holds the posterior density
    hpost[i] = new TH1D(BranchNames[i], BranchNames[i], nBins, mini, maxi);
    hpost[i]->SetMinimum(0);
    hpost[i]->GetYaxis()->SetTitle("Steps");
    hpost[i]->GetYaxis()->SetNoExponent(false);
 
 
    // Project BranchNames[i] onto hpost, applying stepcut
    Chain->Project(BranchNames[i], BranchNames[i], CutPosterior1D.c_str());
 
    if(ApplySmoothing) hpost[i]->Smooth();
 
    (*Central_Value)(i) = Prior;
 
    double Mean, Err, Err_p, Err_m;
    GetArithmetic(hpost[i], Mean, Err);
    (*Means)(i) = Mean;
    (*Errors)(i) = Err;
 
    GetGaussian(hpost[i], Gauss.get(), Mean, Err);
    (*Means_Gauss)(i) = Mean;
    (*Errors_Gauss)(i) = Err;
 
    GetHPD(hpost[i], Mean, Err, Err_p, Err_m);
    (*Means_HPD)(i) = Mean;
    (*Errors_HPD)(i) = Err;
    (*Errors_HPD_Positive)(i) = Err_p;
    (*Errors_HPD_Negative)(i) = Err_m;
 
    // Write the results from the projection into the TVectors and TMatrices
    (*Covariance)(i,i) = (*Errors)(i)*(*Errors)(i);
    (*Correlation)(i,i) = 1.0;
 
    //KS: This need to be before SetMaximum(), this way plot is nicer as line end at the maximum
    auto hpd = std::make_unique<TLine>((*Means_HPD)(i), hpost[i]->GetMinimum(), (*Means_HPD)(i), hpost[i]->GetMaximum());
    SetTLineStyle(hpd.get(), kBlack, 2, kSolid);
    
    hpost[i]->SetLineWidth(2);
    hpost[i]->SetLineColor(kBlue-1);
    hpost[i]->SetMaximum(hpost[i]->GetMaximum()*DrawRange);
    hpost[i]->SetTitle(Title);
    hpost[i]->GetXaxis()->SetTitle(hpost[i]->GetTitle());
    
    // Now make the TLine for the Asimov
    auto Asimov = std::make_unique<TLine>(Prior, hpost[i]->GetMinimum(), Prior, hpost[i]->GetMaximum());
    SetTLineStyle(Asimov.get(), kRed-3, 2, kDashed);
 
    auto leg = std::make_unique<TLegend>(0.12, 0.6, 0.6, 0.97);
    SetLegendStyle(leg.get(), 0.04);
    leg->AddEntry(hpost[i], Form("#splitline{PDF}{#mu = %.2f, #sigma = %.2f}", hpost[i]->GetMean(), hpost[i]->GetRMS()), "l");
    leg->AddEntry(Gauss.get(), Form("#splitline{Gauss}{#mu = %.2f, #sigma = %.2f}", Gauss->GetParameter(1), Gauss->GetParameter(2)), "l");
    leg->AddEntry(hpd.get(), Form("#splitline{HPD}{#mu = %.2f, #sigma = %.2f (+%.2f-%.2f)}", (*Means_HPD)(i), (*Errors_HPD)(i), (*Errors_HPD_Positive)(i), (*Errors_HPD_Negative)(i)), "l");
    leg->AddEntry(Asimov.get(), Form("#splitline{Prior}{x = %.2f , #sigma = %.2f}", Prior, PriorError), "l");
 
    // Write to file
    Posterior->SetName(Title);
    Posterior->SetTitle(Title);
 
    //CW: Don't plot if this is a fixed histogram (i.e. the peak is the whole integral)
    if (hpost[i]->GetMaximum() == hpost[i]->Integral()*DrawRange) 
    {
      MACH3LOG_WARN("Found fixed parameter: {} ({}), moving on", Title, i);
      IamVaried[i] = false;
      //KS:Set mean and error to prior for fixed parameters, it looks much better when fixed parameter has mean on prior rather than on 0 with 0 error.
      (*Means_HPD)(i)  = Prior;
      (*Errors_HPD)(i) = PriorError;
      (*Errors_HPD_Positive)(i)  = PriorError;
      (*Errors_HPD_Negative)(i) = PriorError;
 
      (*Means_Gauss)(i)  = Prior;
      (*Errors_Gauss)(i) = PriorError;
 
      (*Means)(i)  = Prior;
      (*Errors)(i) = PriorError;
      continue;
    }
 
    // Store that this parameter is indeed being varied
    IamVaried[i] = true;
 
    // Draw onto the TCanvas
    hpost[i]->Draw();
    hpd->Draw("same");
    Asimov->Draw("same");
    leg->Draw("same");  
    
    if(printToPDF) Posterior->Print(CanvasName);
        
    // cd into params directory in root file
    PostDir->cd();
    Posterior->Write();
    
    hpost[i]->SetName(Title);
    hpost[i]->SetTitle(Title);
    PostHistDir->cd();
    hpost[i]->Write();
  } // end the for loop over nDraw
 
  OutputFile->cd();
  TTree *SettingsBranch = new TTree("Settings", "Settings");
  int NDParameters = nParam[kNDPar];
  SettingsBranch->Branch("NDParameters", &NDParameters);
  int NDParametersStartingPos = ParamTypeStartPos[kNDPar];
  SettingsBranch->Branch("NDParametersStartingPos", &NDParametersStartingPos);
 
  int FDParameters = nParam[kFDDetPar];
  SettingsBranch->Branch("FDParameters", &FDParameters);
  int FDParametersStartingPos = ParamTypeStartPos[kFDDetPar];
  SettingsBranch->Branch("FDParametersStartingPos", &FDParametersStartingPos);
  
  SettingsBranch->Branch("NDSamplesBins", &NDSamplesBins);
  SettingsBranch->Branch("NDSamplesNames", &NDSamplesNames);
 
  SettingsBranch->Fill();
  SettingsBranch->Write();
  delete SettingsBranch;
 
  TDirectory *Names = OutputFile->mkdir("Names");
  Names->cd();
  for (std::vector<TString>::iterator it = BranchNames.begin(); it != BranchNames.end(); ++it) {
    TObjString((*it)).Write();
  }
  Names->Close();
  delete Names;
 
  OutputFile->cd();
  Central_Value->Write("Central_Value");
  Means->Write("PDF_Means");
  Errors->Write("PDF_Error");
  Means_Gauss->Write("Gauss_Means");
  Errors_Gauss->Write("Gauss_Errors");
  Means_HPD->Write("Means_HPD");
  Errors_HPD->Write("Errors_HPD");
  Errors_HPD_Positive->Write("Errors_HPD_Positive");
  Errors_HPD_Negative->Write("Errors_HPD_Negative");
 
  PostDir->Close();
  delete PostDir;
  PostHistDir->Close();
  delete PostHistDir;
 
  // restore original warning setting
  gErrorIgnoreLevel = originalErrorLevel;
} // Have now written the postfit projections

MakePrefit()

std::unique_ptr< TH1D > MCMCProcessor::MakePrefit ( )

inlineprotected

Prepare prefit histogram for parameter overlay plot.

Definition at line 2331 of file MCMCProcessor.cpp.

                                              {
// *****************************
  if (OutputFile == nullptr) MakeOutputFile();
 
  auto PreFitPlot = std::make_unique<TH1D>("Prefit", "Prefit", nDraw, 0, nDraw);
  PreFitPlot->SetDirectory(nullptr);
  for (int i = 0; i < PreFitPlot->GetNbinsX() + 1; ++i) {
    PreFitPlot->SetBinContent(i+1, 0);
    PreFitPlot->SetBinError(i+1, 0);
  }
 
  //KS: Slightly hacky way to get relative to prior or nominal as this is convention we use,
  //Only applies for xsec, for other systematic it make no difference
  double CentralValueTemp, Central, Error;
 
  // Set labels and data
  for (int i = 0; i < nDraw; ++i)
  {
    //Those keep which parameter type we run currently and relative number
    int ParamEnum = ParamType[i];
    int ParamNo = i - ParamTypeStartPos[ParameterEnum(ParamEnum)];
    CentralValueTemp = ParamCentral[ParamEnum][ParamNo];
    if(plotRelativeToPrior) 
    {
      // Normalise the prior relative the nominal/prior, just the way we get our fit results in MaCh3
      if ( CentralValueTemp != 0) {
        Central = ParamCentral[ParamEnum][ParamNo] / CentralValueTemp;
        Error = ParamErrors[ParamEnum][ParamNo]/CentralValueTemp;
      } else {
        Central = CentralValueTemp + 1.0;
        Error = ParamErrors[ParamEnum][ParamNo];
      }
    }
    else
    {
      Central = CentralValueTemp;
      Error = ParamErrors[ParamEnum][ParamNo];
    }
    //KS: If plotting error for param with flat prior is turned off and given param really has flat prior set error to 0
    if(!PlotFlatPrior && ParamFlat[ParamEnum][ParamNo]) {
      Error = 0.;
    }
    PreFitPlot->SetBinContent(i+1, Central);
    PreFitPlot->SetBinError(i+1, Error);
    PreFitPlot->GetXaxis()->SetBinLabel(i+1, ParamNames[ParamEnum][ParamNo]);
  }
  PreFitPlot->SetDirectory(nullptr);
 
  PreFitPlot->SetFillStyle(1001);
  PreFitPlot->SetFillColor(kRed-3);
  PreFitPlot->SetMarkerStyle(21);
  PreFitPlot->SetMarkerSize(2.4);
  PreFitPlot->SetMarkerColor(kWhite);
  PreFitPlot->SetLineColor(PreFitPlot->GetFillColor());
  PreFitPlot->GetXaxis()->LabelsOption("v");
 
  return PreFitPlot;
}

MakeSubOptimality()

void MCMCProcessor::MakeSubOptimality

(

const int

NIntervals = 10

)

Make and Draw SubOptimality [27].

Author

Henry Wallace

Definition at line 1316 of file MCMCProcessor.cpp.

                                                          {
// *********************
  //Save burn in cut, at the end of the loop we will return to default values
  const int DefaultUpperCut = UpperCut;
  const int DefaultBurnInCut = BurnInCut;
  bool defaultPrintToPDF = printToPDF;
  BurnInCut = 0;
  UpperCut = 0;
  printToPDF = false;
 
  //Set via config in future
  int MaxStep = nSteps;
  int MinStep = 0;
  const int IntervalsSize = nSteps/NIntervals;
 
  MACH3LOG_INFO("Making Suboptimality");
  TStopwatch clock;
  clock.Start();
 
  std::unique_ptr<TH1D> SubOptimality = std::make_unique<TH1D>("Suboptimality", "Suboptimality", NIntervals, MinStep, MaxStep);
  SubOptimality->GetXaxis()->SetTitle("Step");
  SubOptimality->GetYaxis()->SetTitle("Suboptimality");
  SubOptimality->SetLineWidth(2);
  SubOptimality->SetLineColor(kBlue);
 
  for(int i = 0; i < NIntervals; ++i)
  {
    //Reset our cov matrix
    ResetHistograms();
 
    //Set threshold for calculating new matrix
    UpperCut = i*IntervalsSize;
    //Calculate cov matrix
    MakeCovariance_MP(true);
 
    //Calculate eigen values
    TMatrixDSymEigen eigen(*Covariance);
    TVectorD eigen_values;
    eigen_values.ResizeTo(eigen.GetEigenValues());
    eigen_values = eigen.GetEigenValues();
 
    //KS: Converting from ROOT to vector as to make using other libraires (Eigen) easier in future
    std::vector<double> EigenValues(eigen_values.GetNrows());
    for(unsigned int j = 0; j < EigenValues.size(); j++)
    {
      EigenValues[j] = eigen_values(j);
    }
    const double SubOptimalityValue = GetSubOptimality(EigenValues, nDraw);
    SubOptimality->SetBinContent(i+1, SubOptimalityValue);
  }
  clock.Stop();
  MACH3LOG_INFO("Making Suboptimality took {:.2f}s to finish for {} steps", clock.RealTime(), nEntries);
 
  UpperCut = DefaultUpperCut;
  BurnInCut = DefaultBurnInCut;
  printToPDF = defaultPrintToPDF;
 
  SubOptimality->Draw("l");
  Posterior->SetName(SubOptimality->GetName());
  Posterior->SetTitle(SubOptimality->GetTitle());
 
  if(printToPDF) Posterior->Print(CanvasName);
  // Write it to root file
  OutputFile->cd();
  Posterior->Write();
}

MakeTrianglePlot()

void MCMCProcessor::MakeTrianglePlot	(	const std::vector< std::string > &	ParNames,
		const std::vector< double > &	CredibleIntervals = {0.99, 0.90, 0.68 },
		const std::vector< Color_t > &	CredibleIntervalsColours = {kCyan+4, kCyan-2, kCyan-10},
		const std::vector< double > &	CredibleRegions = {0.99, 0.90, 0.68},
		const std::vector< Style_t > &	CredibleRegionStyle = {kDashed, kSolid, kDotted},
		const std::vector< Color_t > &	CredibleRegionColor = {kGreen-3, kGreen-10, kGreen},
		const bool	CredibleInSigmas = false
	)

Make fancy triangle plot for selected parameters.

Parameters

CredibleIntervals	Vector with values of credible intervals, must be in descending order
CredibleIntervalsColours	Color_t telling what colour to use for each Interval line
CredibleInSigmas	Bool telling whether intervals are in percentage or in sigmas, then special conversions is used
CredibleRegions	Vector with values of credible intervals, must be in descending order
CredibleRegionStyle	Style_t telling what line style to use for each Interval line
CredibleRegionColor	Color_t telling what colour to use for each Interval line

Scale the histograms so it shows the posterior probability

Definition at line 1893 of file MCMCProcessor.cpp.

                                                                  {
// *********************
  if(hpost2D.size() == 0) MakeCovariance_MP();
  MACH3LOG_INFO("Making Triangle Plot");
 
  const int nParamPlot = int(ParNames.size());
  std::vector<int> ParamNumber;
  for(int j = 0; j < nParamPlot; ++j)
  {
    int ParamNo = GetParamIndexFromName(ParNames[j]);
    if(ParamNo == M3::_BAD_INT_)
    {
      MACH3LOG_WARN("Couldn't find param {}. Will not plot Triangle plot", ParNames[j]);
      return;
    }
    ParamNumber.push_back(ParamNo);
  }
 
  //KS: Store it as we go back to them at the end
  const std::vector<double> Margins = GetMargins(Posterior);
  Posterior->SetTopMargin(0.001);
  Posterior->SetBottomMargin(0.001);
  Posterior->SetLeftMargin(0.001);
  Posterior->SetRightMargin(0.001);
 
  // KS: We later format hist several times so make one unfired lambda
  auto FormatHistogram = [](auto& hist) {
    hist->GetXaxis()->SetTitle("");
    hist->GetYaxis()->SetTitle("");
    hist->SetTitle("");
 
    hist->GetXaxis()->SetLabelSize(0.1);
    hist->GetYaxis()->SetLabelSize(0.1);
 
    hist->GetXaxis()->SetNdivisions(4);
    hist->GetYaxis()->SetNdivisions(4);
  };
 
  Posterior->cd();
  Posterior->Clear();
  Posterior->Update();
 
  //KS: We sort to have parameters from highest to lowest, this is related to how we make 2D projections in MakeCovariance_MP
  std::sort(ParamNumber.begin(), ParamNumber.end(),  std::greater<int>());
 
  //KS: Calculate how many pads/plots we need
  int Npad = 0;
  for(int j = 1; j < nParamPlot+1; j++) Npad += j;
  Posterior->cd();
  // KS: Sanity check of size and ordering is correct
  CheckCredibleIntervalsOrder(CredibleIntervals, CredibleIntervalsColours);
  CheckCredibleRegionsOrder(CredibleRegions, CredibleRegionStyle, CredibleRegionColor);
 
  const int nCredibleIntervals = int(CredibleIntervals.size());
  const int nCredibleRegions = int(CredibleRegions.size());
 
  //KS: Initialise Tpad histograms etc we will need
  std::vector<TPad*> TrianglePad(Npad);
  //KS: 1D copy of posterior, we need it as we modify them
  std::vector<std::unique_ptr<TH1D>> hpost_copy(nParamPlot);
  std::vector<std::vector<std::unique_ptr<TH1D>>> hpost_cl(nParamPlot);
  std::vector<std::unique_ptr<TText>> TriangleText(nParamPlot * 2);
  std::vector<std::unique_ptr<TH2D>> hpost_2D_copy(Npad-nParamPlot);
  std::vector<std::vector<std::unique_ptr<TH2D>>> hpost_2D_cl(Npad-nParamPlot);
  gStyle->SetPalette(51);
 
  //KS: Super convoluted way of calculating ranges for our pads, trust me it works...
  std::vector<double> X_Min(nParamPlot);
  std::vector<double> X_Max(nParamPlot);
  X_Min[0] = 0.10;
  double xScale = (0.95 - (X_Min[0]+0.05))/nParamPlot;
  //KS: 0.05 is because we need additional offset for labels
  X_Max[0] = X_Min[0]+xScale+0.05;
  for(int i = 1; i < nParamPlot; i++)
  {
    X_Min[i] = X_Max[i-1];
    X_Max[i] = X_Min[i]+xScale;
  }
  std::vector<double> Y_Min(nParamPlot);
  std::vector<double> Y_Max(nParamPlot);
  Y_Max[0] = 0.95;
  //KS: 0.10 is becasue we need additional offset for labels
  double yScale = std::fabs(0.10 - (Y_Max[0]))/nParamPlot;
  Y_Min[0] = Y_Max[0]-yScale;
  for(int i = 1; i < nParamPlot; i++)
  {
    Y_Max[i] = Y_Min[i-1];
    Y_Min[i] = Y_Max[i]-yScale;
  }
 
  //KS: We store as numbering of isn't straightforward
  int counterPad = 0, counterText = 0, counterPost = 0, counter2DPost = 0;
  //KS: We start from top of the plot, might be confusing but works very well
  for(int y = 0; y < nParamPlot; y++)
  {
    //KS: start from left and go right, depending on y
    for(int x = 0; x <= y; x++)
    {
      //KS: Need to go to canvas every time to have our pads in the same canvas, not pads in the pads
      Posterior->cd();
      TrianglePad[counterPad] = new TPad(Form("TPad_%i", counterPad), Form("TPad_%i", counterPad),
                                         X_Min[x], Y_Min[y], X_Max[x], Y_Max[y]);
 
      TrianglePad[counterPad]->SetTopMargin(0);
      TrianglePad[counterPad]->SetRightMargin(0);
 
      TrianglePad[counterPad]->SetGrid();
      TrianglePad[counterPad]->SetFrameBorderMode(0);
      TrianglePad[counterPad]->SetBorderMode(0);
      TrianglePad[counterPad]->SetBorderSize(0);
 
      //KS: Corresponds to bottom part of the plot, need margins for labels
      TrianglePad[counterPad]->SetBottomMargin(y == (nParamPlot - 1) ? 0.1 : 0);
      //KS: Corresponds to left part, need margins for labels
      TrianglePad[counterPad]->SetLeftMargin(x == 0 ? 0.15 : 0);
 
      TrianglePad[counterPad]->Draw();
      TrianglePad[counterPad]->cd();
 
      //KS:if diagonal plot main posterior
      if(x == y)
      {
        hpost_copy[counterPost] = M3::Clone<TH1D>(hpost[ParamNumber[x]], Form("hpost_copy_%i", ParamNumber[x]));
        hpost_cl[counterPost].resize(nCredibleIntervals);
        hpost_copy[counterPost]->Scale(1. / hpost_copy[counterPost]->Integral());
        for (int j = 0; j < nCredibleIntervals; ++j)
        {
          hpost_cl[counterPost][j] = M3::Clone<TH1D>(hpost[ParamNumber[x]], Form("hpost_copy_%i_CL_%f", ParamNumber[x], CredibleIntervals[j]));
          //KS: Reset to get rid to TF1 otherwise we run into segfault :(
          hpost_cl[counterPost][j]->Reset("");
          hpost_cl[counterPost][j]->Fill(0.0, 0.0);
 
          // Scale the histograms before gettindg credible intervals
          hpost_cl[counterPost][j]->Scale(1. / hpost_cl[counterPost][j]->Integral());
          GetCredibleIntervalSig(hpost_copy[counterPost], hpost_cl[counterPost][j], CredibleInSigmas, CredibleIntervals[j]);
 
          hpost_cl[counterPost][j]->SetFillColor(CredibleIntervalsColours[j]);
          hpost_cl[counterPost][j]->SetLineWidth(1);
        }
 
        hpost_copy[counterPost]->SetMaximum(hpost_copy[counterPost]->GetMaximum()*1.2);
        hpost_copy[counterPost]->SetLineWidth(2);
        hpost_copy[counterPost]->SetLineColor(kBlack);
 
        //KS: Don't want any titles
        FormatHistogram(hpost_copy[counterPost]);
 
        hpost_copy[counterPost]->Draw("HIST");
        for (int j = 0; j < nCredibleIntervals; ++j){
          hpost_cl[counterPost][j]->Draw("HIST SAME");
        }
        counterPost++;
      }
      //KS: Here we plot 2D credible regions
      else
      {
        hpost_2D_copy[counter2DPost] = M3::Clone<TH2D>(hpost2D[ParamNumber[x]][ParamNumber[y]],
                                                       Form("hpost_copy_%i_%i", ParamNumber[x], ParamNumber[y]));
        hpost_2D_cl[counter2DPost].resize(nCredibleRegions);
        //KS: Now copy for every credible region
        for (int k = 0; k < nCredibleRegions; ++k)
        {
          hpost_2D_cl[counter2DPost][k] = M3::Clone<TH2D>(hpost2D[ParamNumber[x]][ParamNumber[y]],
                                                          Form("hpost_copy_%i_%i_CL_%f", ParamNumber[x], ParamNumber[y], CredibleRegions[k]));
          GetCredibleRegionSig(hpost_2D_cl[counter2DPost][k], CredibleInSigmas, CredibleRegions[k]);
 
          hpost_2D_cl[counter2DPost][k]->SetLineColor(CredibleRegionColor[k]);
          hpost_2D_cl[counter2DPost][k]->SetLineWidth(2);
          hpost_2D_cl[counter2DPost][k]->SetLineStyle(CredibleRegionStyle[k]);
        }
        //KS: Don't want any titles
        FormatHistogram(hpost_2D_copy[counter2DPost]);
 
        hpost_2D_copy[counter2DPost]->Draw("COL");
        //Now credible regions
        for (int k = 0; k < nCredibleRegions; ++k){
          hpost_2D_cl[counter2DPost][k]->Draw("CONT3 SAME");
        }
        counter2DPost++;
      }
      //KS: Corresponds to bottom part of the plot
      if(y == (nParamPlot-1))
      {
        Posterior->cd();
        TriangleText[counterText] = std::make_unique<TText>(X_Min[x]+ (X_Max[x]-X_Min[x])/4, 0.04, hpost[ParamNumber[x]]->GetTitle());
        //KS: Unfortunately for many plots or long names this can go out of bounds :(
        TriangleText[counterText]->SetTextSize(0.015);
        TriangleText[counterText]->SetNDC(true);
        TriangleText[counterText]->Draw();
        counterText++;
      }
      //KS: Corresponds to left part
      if(x == 0)
      {
        Posterior->cd();
        TriangleText[counterText] = std::make_unique<TText>(0.04, Y_Min[y] + (Y_Max[y]-Y_Min[y])/4, hpost[ParamNumber[y]]->GetTitle());
        //KS: Rotate as this is y axis
        TriangleText[counterText]->SetTextAngle(90);
        //KS: Unfortunately for many plots or long names this can go out of bounds :(
        TriangleText[counterText]->SetTextSize(0.015);
        TriangleText[counterText]->SetNDC(true);
        TriangleText[counterText]->Draw();
        counterText++;
      }
      Posterior->Update();
      counterPad++;
    }
  }
 
  Posterior->cd();
  auto legend = std::make_unique<TLegend>(0.60, 0.7, 0.9, 0.9);
  SetLegendStyle(legend.get(), 0.03);
  //KS: Legend is shared so just take first histograms
  for (int j = nCredibleIntervals-1; j >= 0; --j)
  {
    if(CredibleInSigmas)
      legend->AddEntry(hpost_cl[0][j].get(), Form("%.0f#sigma Credible Interval", CredibleIntervals[j]), "f");
    else
      legend->AddEntry(hpost_cl[0][j].get(), Form("%.0f%% Credible Interval", CredibleRegions[j]*100), "f");
  }
  for (int k = nCredibleRegions-1; k >= 0; --k)
  {
    if(CredibleInSigmas)
      legend->AddEntry(hpost_2D_cl[0][k].get(), Form("%.0f#sigma Credible Region", CredibleRegions[k]), "l");
    else
      legend->AddEntry(hpost_2D_cl[0][k].get(), Form("%.0f%% Credible Region", CredibleRegions[k]*100), "l");
  }
  legend->Draw("SAME");
  Posterior->Update();
 
  // Write to file
  Posterior->SetName("TrianglePlot");
  Posterior->SetTitle("TrianglePlot");
 
  if(printToPDF) Posterior->Print(CanvasName);
  // Write it to root file
  OutputFile->cd();
  Posterior->Write();
 
  //KS: Remove allocated structures
  for(int i = 0; i < Npad; i++) delete TrianglePad[i];
 
  //KS: Restore margin
  SetMargins(Posterior, Margins);
}

MakeViolin()

void MCMCProcessor::MakeViolin

(

)

Make and Draw Violin.

Definition at line 808 of file MCMCProcessor.cpp.

                               {
// *********************
  //KS: Make sure we have steps
  if(!CacheMCMC) CacheSteps();
 
  MACH3LOG_INFO("Producing Violin Plot");
 
  // KS: Set temporary branch address to allow min/max, otherwise ROOT can segfaults
  double tempVal = 0.0;
  //KS: Find min and max to make histogram in range
  double maxi_y = -9999;
  double mini_y = +9999;
  for (int i = 0; i < nDraw; ++i)
  {
    Chain->SetBranchAddress(BranchNames[i].Data(), &tempVal);
    const double max_val = Chain->GetMaximum(BranchNames[i]);
    const double min_val = Chain->GetMinimum(BranchNames[i]);
  
    maxi_y = std::max(maxi_y, max_val);
    mini_y = std::min(mini_y, min_val);
  }
 
  const int vBins = (maxi_y-mini_y)*25;
  hviolin = std::make_unique<TH2D>("hviolin", "hviolin", nDraw, 0, nDraw, vBins, mini_y, maxi_y);
  hviolin->SetDirectory(nullptr);
  //KS: Prior has larger errors so we increase range and number of bins
  constexpr int PriorFactor = 4;
  hviolin_prior = std::make_unique<TH2D>("hviolin_prior", "hviolin_prior", nDraw, 0, nDraw, PriorFactor*vBins, PriorFactor*mini_y, PriorFactor*maxi_y);
  hviolin_prior->SetDirectory(nullptr);
 
  auto rand = std::make_unique<TRandom3>(0);
  std::vector<double> PriorVec(nDraw);
  std::vector<double> PriorErrorVec(nDraw);
  std::vector<bool> PriorFlatVec(nDraw);
 
  for (int x = 0; x < nDraw; ++x)
  {
    TString Title;
    double Prior, PriorError;
 
    GetNthParameter(x, Prior, PriorError, Title);
    //Set fancy labels
    hviolin->GetXaxis()->SetBinLabel(x+1, Title);
    hviolin_prior->GetXaxis()->SetBinLabel(x+1, Title);
    PriorVec[x] = Prior;
    PriorErrorVec[x] = PriorError;
 
    ParameterEnum ParType = ParamType[x];
    int ParamTemp = x - ParamTypeStartPos[ParType];
    PriorFlatVec[x] = ParamFlat[ParType][ParamTemp];
  }
 
  TStopwatch clock;
  clock.Start();
 
  // nDraw is number of draws we want to do
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int x = 0; x < nDraw; ++x)
  {
    //KS: Consider another treatment for fixed params
    //if (IamVaried[x] == false) continue;
    for (int k = 0; k < nEntries; ++k)
    {
      //KS: Burn in cut
      if(StepNumber[k] < BurnInCut) continue;
 
      //Only used for Suboptimatlity
      if(StepNumber[k] > UpperCut) continue;
 
      //KS: We know exactly which x bin we will end up, find y bin. This allow to avoid coslty Fill() and enable multithreading becasue I am master of faster
      const double y = hviolin->GetYaxis()->FindBin(ParStep[x][k]);
      hviolin->SetBinContent(x+1, y,  hviolin->GetBinContent(x+1, y)+1);
    }
 
    //KS: If we set option to not plot flat prior and param has flat prior then we skip this step
    if(!(!PlotFlatPrior && PriorFlatVec[x]))
    {
      for (int k = 0; k < nEntries; ++k)
      {
        const double Entry = rand->Gaus(PriorVec[x], PriorErrorVec[x]);
        const double y = hviolin_prior->GetYaxis()->FindBin(Entry);
        hviolin_prior->SetBinContent(x+1, y,  hviolin_prior->GetBinContent(x+1, y)+1);
      }
    }
  } // end the for loop over nDraw
  clock.Stop();
  MACH3LOG_INFO("Making Violin plot took {:.2f}s to finish for {} steps", clock.RealTime(), nEntries);
 
  //KS: Tells how many parameters in one canvas we want
  constexpr int IntervalsSize = 10;
  const int NIntervals = nDraw/IntervalsSize;
 
  hviolin->GetYaxis()->SetTitle("Parameter Value");
  hviolin->GetXaxis()->SetTitle();
  hviolin->GetXaxis()->LabelsOption("v");
  
  hviolin_prior->GetYaxis()->SetTitle("Parameter Value");
  hviolin_prior->GetXaxis()->SetTitle();
  hviolin_prior->GetXaxis()->LabelsOption("v");
 
  hviolin_prior->SetLineColor(kRed);
  hviolin_prior->SetMarkerColor(kRed);
  hviolin_prior->SetFillColorAlpha(kRed, 0.35);
  hviolin_prior->SetMarkerStyle(20);
  hviolin_prior->SetMarkerSize(0.5);
 
  // These control violin width, if you use larger then 1 they will most likely overlay, so be cautious
  hviolin_prior->SetBarWidth(1.0);
  hviolin_prior->SetBarOffset(0);
 
  hviolin->SetLineColor(kBlue);
  hviolin->SetMarkerColor(kBlue);
  hviolin->SetFillColorAlpha(kBlue, 0.35);
  hviolin->SetMarkerStyle(20);
  hviolin->SetMarkerSize(1.0);
  
  const double BottomMargin = Posterior->GetBottomMargin();
  Posterior->SetBottomMargin(0.2);
    
  OutputFile->cd();
  hviolin->Write("param_violin");
  hviolin_prior->Write("param_violin_prior");
  //KS: This is mostly for example plots, we have full file in the ROOT file so can do much better plot later
  hviolin->GetYaxis()->SetRangeUser(-1, +2);
  hviolin_prior->GetYaxis()->SetRangeUser(-1, +2);
  for (int i = 0; i < NIntervals+1; ++i)
  {
    int RangeMin = i*IntervalsSize;
    int RangeMax =RangeMin + IntervalsSize;
    if(i == NIntervals+1) {
      RangeMin = i*IntervalsSize;
      RangeMax = nDraw;
    }
    if(RangeMin >= nDraw) break;
 
    hviolin->GetXaxis()->SetRangeUser(RangeMin, RangeMax);
    hviolin_prior->GetXaxis()->SetRangeUser(RangeMin, RangeMax);
 
    //KS: ROOT6 has some additional options, consider updating it. more https://root.cern/doc/master/classTHistPainter.html#HP140b
    hviolin_prior->Draw("violinX(03100300)");
    hviolin->Draw("violinX(03100300) SAME");
    if(printToPDF) Posterior->Print(CanvasName);
  }
  //KS: Return Margin to default one
  Posterior->SetBottomMargin(BottomMargin);
}

ParameterEvolution()

void MCMCProcessor::ParameterEvolution	(	const std::vector< std::string > &	Names,
		const std::vector< int > &	NIntervals
	)

Make .gif of parameter evolution.

Parameters

Names	Parameter names for which we do .gif
NIntervals	Number of intervals for a gif

Definition at line 3222 of file MCMCProcessor.cpp.

                                                                         {
// **************************
  MACH3LOG_INFO("Parameter Evolution gif");
 
  //KS: First we need to find parameter number based on name
  for(unsigned int k = 0; k < Names.size(); ++k)
  {
    //KS: First we need to find parameter number based on name
    int ParamNo = GetParamIndexFromName(Names[k]);
    if(ParamNo == M3::_BAD_INT_)
    {
      MACH3LOG_WARN("Couldn't find param {}. Can't reweight Prior", Names[k]);
      continue;
    }
 
    const int IntervalsSize = nSteps/NIntervals[k];
    // ROOT won't overwrite gifs so we need to delete the file if it's there already
    std::string filename = Names[k] + ".gif";
    std::ifstream f(filename);
    if (f.good()) {
      f.close();
      int ret = system(fmt::format("rm {}", filename).c_str());
      if (ret != 0) {
        MACH3LOG_WARN("Error: system call to delete {} failed with code {}", filename, ret);
      }
    }
 
    int Counter = 0;
    for(int i = NIntervals[k]-1; i >= 0; --i)
    {
      // This holds the posterior density
      // KS: WARNING do not change to smart pointer, it breaks and I don't know why
      TH1D* EvePlot = new TH1D(BranchNames[ParamNo], BranchNames[ParamNo], nBins,
                               hpost[ParamNo]->GetXaxis()->GetXmin(), hpost[ParamNo]->GetXaxis()->GetXmax());
      EvePlot->SetMinimum(0);
      EvePlot->GetYaxis()->SetTitle("PDF");
      EvePlot->GetYaxis()->SetNoExponent(false);
 
      //KS: Apply additional Cuts, like mass ordering
      std::string CutPosterior1D = "step > " + std::to_string(i*IntervalsSize+IntervalsSize);
 
      // If Posterior1DCut is not empty, append it
      if (!Posterior1DCut.empty()) {
        CutPosterior1D += " && " + Posterior1DCut;
      }
 
      // Apply reweighting if requested
      if (ReweightPosterior) {
        CutPosterior1D = "(" + CutPosterior1D + ")*(" + ReweightName + ")";
      }
 
      std::string TextTitle = "Steps = 0 - "+std::to_string(Counter*IntervalsSize+IntervalsSize);
      // Project BranchNames[ParamNo] onto hpost, applying stepcut
      Chain->Project(BranchNames[ParamNo], BranchNames[ParamNo], CutPosterior1D.c_str());
 
      EvePlot->SetLineWidth(2);
      EvePlot->SetLineColor(kBlue-1);
      EvePlot->SetTitle(Names[k].c_str());
      EvePlot->GetXaxis()->SetTitle(EvePlot->GetTitle());
      EvePlot->GetYaxis()->SetLabelOffset(1000);
      if(ApplySmoothing) EvePlot->Smooth();
 
      EvePlot->Scale(1. / EvePlot->Integral());
      EvePlot->Draw("HIST");
 
      TText text(0.3, 0.8, TextTitle.c_str());
      text.SetTextFont (43);
      text.SetTextSize (40);
      text.SetNDC(true);
      text.Draw("SAME");
 
      if(i == 0) Posterior->Print((Names[k] + ".gif++20").c_str()); // produces infinite loop animated GIF
      else Posterior->Print((Names[k] + ".gif+20").c_str()); // add picture to .gif
      delete EvePlot;
      Counter++;
    }
  }
}

ParamTraces()

void MCMCProcessor::ParamTraces ( )

inlineprotected

CW: Draw trace plots of the parameters i.e. parameter vs step.

Definition at line 3497 of file MCMCProcessor.cpp.

                                {
// *****************
  if (ParStep == nullptr) PrepareDiagMCMC();
  MACH3LOG_INFO("Making trace plots...");
  // Make the TH1Ds
  std::vector<TH1D*> TraceParamPlots(nDraw);
  std::vector<TH1D*> TraceSamplePlots(nSamples);
  std::vector<TH1D*> TraceSystsPlots(nSysts);
 
  // Set the titles and limits for TH2Ds
  for (int j = 0; j < nDraw; ++j) {
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    
    GetNthParameter(j, Prior, PriorError, Title);
    std::string HistName = Form("%s_%s_Trace", Title.Data(), BranchNames[j].Data());
    TraceParamPlots[j] = new TH1D(HistName.c_str(), HistName.c_str(), nEntries, 0, nEntries);
    TraceParamPlots[j]->GetXaxis()->SetTitle("Step");
    TraceParamPlots[j]->GetYaxis()->SetTitle("Parameter Variation");
  }
 
  for (int j = 0; j < nSamples; ++j) {
    std::string HistName = SampleName_v[j].Data();
    TraceSamplePlots[j] = new TH1D(HistName.c_str(), HistName.c_str(), nEntries, 0, nEntries);
    TraceSamplePlots[j]->GetXaxis()->SetTitle("Step");
    TraceSamplePlots[j]->GetYaxis()->SetTitle("Sample -logL");
  }
 
  for (int j = 0; j < nSysts; ++j) {
    std::string HistName = SystName_v[j].Data();
    TraceSystsPlots[j] = new TH1D(HistName.c_str(), HistName.c_str(), nEntries, 0, nEntries);
    TraceSystsPlots[j]->GetXaxis()->SetTitle("Step");
    TraceSystsPlots[j]->GetYaxis()->SetTitle("Systematic -logL");
  }
 
  // Have now made the empty TH1Ds, now for writing content to them!
  // Loop over the number of parameters to draw their traces
  // Each histogram
#ifdef MULTITHREAD
  MACH3LOG_INFO("Using multi-threading...");
  #pragma omp parallel for
#endif
  for (int i = 0; i < nEntries; ++i) {
    // Set bin content for the ith bin to the parameter values
    for (int j = 0; j < nDraw; ++j) {
      TraceParamPlots[j]->SetBinContent(i, ParStep[j][i]);
    }
    for (int j = 0; j < nSamples; ++j) {
      TraceSamplePlots[j]->SetBinContent(i, SampleValues[i][j]);
    }
    for (int j = 0; j < nSysts; ++j) {
      TraceSystsPlots[j]->SetBinContent(i, SystValues[i][j]);
    }
  }
 
  // Write the output and delete the TH2Ds
  TDirectory *TraceDir = OutputFile->mkdir("Trace");
  TraceDir->cd();
  for (int j = 0; j < nDraw; ++j) {
    // Fit a linear function to the traces
    TF1 *Fitter = new TF1("Fitter","[0]", nEntries/2, nEntries);
    Fitter->SetLineColor(kRed);
    TraceParamPlots[j]->Fit("Fitter","Rq");
    TraceParamPlots[j]->Write();
    delete Fitter;
    delete TraceParamPlots[j];
  }
 
  TDirectory *LLDir = OutputFile->mkdir("LogL");
  LLDir->cd();
  for (int j = 0; j < nSamples; ++j) {
    TraceSamplePlots[j]->Write();
    delete TraceSamplePlots[j];
    delete[] SampleValues[j];
  }
  delete[] SampleValues;
 
  for (int j = 0; j < nSysts; ++j) {
    TraceSystsPlots[j]->Write();
    delete TraceSystsPlots[j];
    delete SystValues[j];
  }
  delete[] SystValues;
 
  TraceDir->Close();
  delete TraceDir;
 
  OutputFile->cd();
}

PowerSpectrumAnalysis()

void MCMCProcessor::PowerSpectrumAnalysis ( )

inlineprotected

RC: Perform spectral analysis of MCMC [7].

Author

Richard Calland

Todo:

KS: Code is awfully slow... I know how to make it faster (GPU scream in a distant) but for now just make it for two params, bit hacky sry...

Definition at line 4174 of file MCMCProcessor.cpp.

                                          {
// **************************
  TStopwatch clock;
  clock.Start();
 
  //KS: Store it as we go back to them at the end
  const double TopMargin  = Posterior->GetTopMargin();
  const int OptTitle = gStyle->GetOptTitle();
 
  Posterior->SetTopMargin(0.1);
  gStyle->SetOptTitle(1);
 
  MACH3LOG_INFO("Making Power Spectrum plots...");
 
  // This is only to reduce number of computations...
  const int N_Coeffs = std::min(10000, nEntries);
  const int start = -(N_Coeffs/2-1);
  const int end = N_Coeffs/2-1;
  const int v_size = end - start;
 
  int nPrams = nDraw;
  nPrams = 1;
 
  std::vector<std::vector<float>> k_j(nPrams, std::vector<float>(v_size, 0.0));
  std::vector<std::vector<float>> P_j(nPrams, std::vector<float>(v_size, 0.0));
 
  int _N = nEntries;
  if (_N % 2 != 0) _N -= 1; // N must be even
 
  //This is being used a lot so calculate it once to increase performance
  const double two_pi_over_N = 2 * TMath::Pi() / static_cast<double>(_N);
 
  // KS: This could be moved to GPU I guess
  #ifdef MULTITHREAD
  #pragma omp parallel for collapse(2)
  #endif
  // RC: equation 11: for each value of j coef, from range -N/2 -> N/2
  for (int j = 0; j < nPrams; ++j)
  {
    for (int jj = start; jj < end; ++jj)
    {
      std::complex<double> a_j = 0.0;
      for (int n = 0; n < _N; ++n)
      {
        //if(StepNumber[n] < BurnInCut) continue;
        std::complex<double> exp_temp(0, two_pi_over_N * jj * n);
        a_j += ParStep[j][n] * std::exp(exp_temp);
      }
      a_j /= std::sqrt(float(_N));
      const int _c = jj - start;
 
      k_j[j][_c] = two_pi_over_N * jj;
      // Equation 13
      P_j[j][_c] = std::norm(a_j);
    }
  }
 
  TDirectory *PowerDir = OutputFile->mkdir("PowerSpectrum");
  PowerDir->cd();
 
  TVectorD* PowerSpectrumStepSize = new TVectorD(nPrams);
  for (int j = 0; j < nPrams; ++j)
  {
    auto plot = std::make_unique<TGraph>(v_size, k_j[j].data(), P_j[j].data());
 
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(j, Prior, PriorError, Title);
 
    std::string name = Form("Power Spectrum of %s;k;P(k)", Title.Data());
 
    plot->SetTitle(name.c_str());
    name = Form("%s_power_spectrum", Title.Data());
    plot->SetName(name.c_str());
    plot->SetMarkerStyle(7);
 
    // Equation 18
    TF1 *func = new TF1("power_template", "[0]*( ([1] / x)^[2] / (([1] / x)^[2] +1) )", 0.0, 1.0);
    // P0 gives the amplitude of the white noise spectrum in the k → 0 limit
    func->SetParameter(0, 10.0);
    // k* indicates the position of the turnover to a different power law behaviour
    func->SetParameter(1, 0.1);
    // alpha free parameter
    func->SetParameter(2, 2.0);
 
    // Set parameter limits for stability
    func->SetParLimits(0, 0.0, 100.0); // Amplitude should be non-negative
    func->SetParLimits(1, 0.001, 1.0); // k* should be within a reasonable range
    func->SetParLimits(2, 0.0, 5.0);   // alpha should be positive
 
    plot->Fit("power_template","Rq");
 
    Posterior->SetLogx();
    Posterior->SetLogy();
    Posterior->SetGrid();
    plot->Write(plot->GetName());
    plot->Draw("AL");
    func->Draw("SAME");
    if(printToPDF) Posterior->Print(CanvasName);
 
    //KS: I have no clue what is the reason behind this. Found this in Rick Calland code...
    (*PowerSpectrumStepSize)(j) = std::sqrt(func->GetParameter(0)/float(v_size*0.5));
    delete func;
  }
 
  PowerSpectrumStepSize->Write("PowerSpectrumStepSize");
  delete PowerSpectrumStepSize;
  PowerDir->Close();
  delete PowerDir;
 
  clock.Stop();
  MACH3LOG_INFO("Making Power Spectrum took {:.2f}s", clock.RealTime());
 
  Posterior->SetTopMargin(TopMargin);
  gStyle->SetOptTitle(OptTitle);
}

PrepareDiagMCMC()

void MCMCProcessor::PrepareDiagMCMC ( )

inlineprotected

CW: Prepare branches etc. for DiagMCMC.

Definition at line 3334 of file MCMCProcessor.cpp.

                                    {
// **************************
  doDiagMCMC = true;
    
  if(ParStep != nullptr) {
    MACH3LOG_ERROR("It look like ParStep was already filled ");
    MACH3LOG_ERROR("Even though it is used for MakeCovariance_MP and for DiagMCMC");
    MACH3LOG_ERROR("it has different structure in both for cache hits, sorry ");
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  if(nBatches == 0) {
    MACH3LOG_ERROR("nBatches is equal to 0");
    MACH3LOG_ERROR("please use SetnBatches to set other value fore example 20");
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
    
  // Initialise ParStep
  ParStep = new double*[nDraw]();
  for (int j = 0; j < nDraw; ++j) {
    ParStep[j] = new double[nEntries]();
    for (int i = 0; i < nEntries; ++i) {
      ParStep[j][i] = -999.99;
    }
  }
 
  SampleValues = new double*[nEntries]();
  SystValues = new double*[nEntries]();
  AccProbValues = new double[nEntries]();
  StepNumber = new unsigned int[nEntries]();
  for (int i = 0; i < nEntries; ++i) {
    SampleValues[i] = new double[nSamples]();
    SystValues[i] = new double[nSysts]();
 
    for (int j = 0; j < nSamples; ++j) {
      SampleValues[i][j] = -999.99;
    }
    for (int j = 0; j < nSysts; ++j) {
      SystValues[i][j] = -999.99;
    }
    AccProbValues[i] = -999.99;
    StepNumber[i] = 0;
  }
 
  // Initialise the sums
  ParamSums = new double[nDraw]();
  for (int i = 0; i < nDraw; ++i) {
    ParamSums[i] = 0.0;
  }
  MACH3LOG_INFO("Reading input tree...");
  TStopwatch clock;
  clock.Start();
 
  // Set all the branches to off
  Chain->SetBranchStatus("*", false);
 
  // 10 entries output
  const int countwidth = nEntries/10;
 
  // Can also do the batched means here to minimize excessive loops
  // The length of each batch
  const int BatchLength = nEntries/nBatches+1;
  BatchedAverages = new double*[nBatches]();
  AccProbBatchedAverages = new double[nBatches]();
  for (int i = 0; i < nBatches; ++i) {
    BatchedAverages[i] = new double[nDraw];
    AccProbBatchedAverages[i] = 0;
    for (int j = 0; j < nDraw; ++j) {
      BatchedAverages[i][j] = 0.0;
    }
  }
  std::vector<double> ParStepBranch(nDraw);
  std::vector<double> SampleValuesBranch(nSamples);
  std::vector<double> SystValuesBranch(nSysts);
  int StepNumberBranch = 0;
  double AccProbValuesBranch = 0;
  // Set the branch addresses for params
  for (int j = 0; j < nDraw; ++j) {
    Chain->SetBranchStatus(BranchNames[j].Data(), true);
    Chain->SetBranchAddress(BranchNames[j].Data(), &ParStepBranch[j]);
  }
  // Set the branch addresses for samples
  for (int j = 0; j < nSamples; ++j) {
    Chain->SetBranchStatus(SampleName_v[j].Data(), true);
    Chain->SetBranchAddress(SampleName_v[j].Data(), &SampleValuesBranch[j]);
  }
  // Set the branch addresses for systematics
  for (int j = 0; j < nSysts; ++j) {
    Chain->SetBranchStatus(SystName_v[j].Data(), true);
    Chain->SetBranchAddress(SystName_v[j].Data(), &SystValuesBranch[j]);
  }
  // Only needed for Geweke right now
  Chain->SetBranchStatus("step", true);
  Chain->SetBranchAddress("step", &StepNumberBranch);
  // Turn on the branches which we want for acc prob
  Chain->SetBranchStatus("accProb", true);
  Chain->SetBranchAddress("accProb", &AccProbValuesBranch);
 
  // Loop over the entries
  //KS: This is really a bottleneck right now, thus revisit with ROOT6 https://pep-root6.github.io/docs/analysis/parallell/root.html
  for (int i = 0; i < nEntries; ++i) {
    // Fill up the arrays
    Chain->GetEntry(i);
 
    if (i % countwidth == 0)
      MaCh3Utils::PrintProgressBar(i, nEntries);
 
    // Set the branch addresses for params
    for (int j = 0; j < nDraw; ++j) {
      ParStep[j][i] = ParStepBranch[j];
    }
    // Set the branch addresses for samples
    for (int j = 0; j < nSamples; ++j) {
      SampleValues[i][j] = SampleValuesBranch[j];
    }
    // Set the branch addresses for systematics
    for (int j = 0; j < nSysts; ++j) {
      SystValues[i][j] = SystValuesBranch[j];
    }
      
    // Set the branch addresses for Acceptance Probability
    AccProbValues[i] = AccProbValuesBranch;
    StepNumber[i] = StepNumberBranch;
 
    // Find which batch the event belongs in
    int BatchNumber = -1;
    // I'm so lazy! But it's OK, the major overhead here is GetEntry: saved by ROOT!
    for (int j = 0; j < nBatches; ++j) {
      if (i < (j+1)*BatchLength) {
        BatchNumber = j;
        break;
      }
    }
    // Fill up the sum for each j param
    for (int j = 0; j < nDraw; ++j) {
      ParamSums[j] += ParStep[j][i];
      BatchedAverages[BatchNumber][j] += ParStep[j][i];
    }
    
    //KS: Could easily add this to above loop but I accProb is different beast so better keep it like this
    AccProbBatchedAverages[BatchNumber] += AccProbValues[i];
  }
  clock.Stop();
  MACH3LOG_INFO("Took {:.2f}s to finish caching statistic for Diag MCMC with {} steps", clock.RealTime(), nEntries);
 
  // Make the sums into average
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int i = 0; i < nDraw; ++i) {
    ParamSums[i] /= double(nEntries);
    for (int j = 0; j < nBatches; ++j) {
      // Divide by the total number of events in the batch
      BatchedAverages[j][i] /= BatchLength;
      if(i == 0) AccProbBatchedAverages[j] /= BatchLength; //KS: we have only one accProb, keep it like this for now
    }
  }
 
  // And make our sweet output file
  if (OutputFile == nullptr) MakeOutputFile();
}

PrintInfo()

void MCMCProcessor::PrintInfo ( ) const

inlineprotected

Print info like how many params have been loaded etc.

Definition at line 4529 of file MCMCProcessor.cpp.

                                    {
// **************************
  // KS: Create a map to store the counts of unique strings
  std::unordered_map<std::string, int> paramCounts;
  std::vector<std::string> orderedKeys;
 
  for (const std::string& param : ParameterGroup) {
    if (paramCounts[param] == 0) {
      orderedKeys.push_back(param);  // preserve order of first appearance
    }
    paramCounts[param]++;
  }
 
  MACH3LOG_INFO("************************************************");
  MACH3LOG_INFO("Scanning output branches...");
  MACH3LOG_INFO("# Useful entries in tree: \033[1;32m {} \033[0m ", nDraw);
  MACH3LOG_INFO("# Model params:  \033[1;32m {} starting at {} \033[0m ", nParam[kXSecPar], ParamTypeStartPos[kXSecPar]);
  MACH3LOG_INFO("# With following groups: ");
  for (const std::string& key : orderedKeys) {
    MACH3LOG_INFO(" # {} params: {}", key, paramCounts[key]);
  }
  MACH3LOG_INFO("# ND params (legacy):    \033[1;32m {} starting at {} \033[0m ", nParam[kNDPar], ParamTypeStartPos[kNDPar]);
  MACH3LOG_INFO("# FD params (legacy):    \033[1;32m {} starting at {} \033[0m ", nParam[kFDDetPar], ParamTypeStartPos[kFDDetPar]);
  MACH3LOG_INFO("************************************************");
}

ReadFDFile()

void MCMCProcessor::ReadFDFile ( )

inlineprotected

Read the FD cov file and get the input central values and errors.

Definition at line 2629 of file MCMCProcessor.cpp.

                               {
// ***************
  // Do the same for the FD
  TFile *FDdetFile = new TFile(CovPos[kFDDetPar].back().c_str(), "open");
  if (FDdetFile->IsZombie()) {
    MACH3LOG_ERROR("Couldn't find FDdetFile {}", CovPos[kFDDetPar].back());
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  FDdetFile->cd();
 
  TMatrixD *FDdetMatrix = FDdetFile->Get<TMatrixD>(CovNamePos[kFDDetPar].c_str());
 
  for (int i = 0; i < FDdetMatrix->GetNrows(); ++i)
  {
    //KS: FD parameters start at 1. in contrary to ND280
    ParamNom[kFDDetPar].push_back(1.);
    ParamCentral[kFDDetPar].push_back(1.);
 
    ParamErrors[kFDDetPar].push_back( std::sqrt((*FDdetMatrix)(i,i)) );
    ParamNames[kFDDetPar].push_back( Form("FD Det %i", i) );
 
    //KS: Currently we can only set it via config, change it in future
    ParamFlat[kFDDetPar].push_back( false );
  }
  //KS: The last parameter is p scale
  //ETA: we need to be careful here, this is only true for SK in the T2K beam analysis...
  if(FancyPlotNames) ParamNames[kFDDetPar].back() = "Momentum Scale";
 
  FDdetFile->Close();
  delete FDdetFile;
  delete FDdetMatrix;
}

ReadInputCov()

void MCMCProcessor::ReadInputCov ( )

inlineprotected

CW: Read the input Covariance matrix entries. Get stuff like parameter input errors, names, and so on.

Definition at line 2393 of file MCMCProcessor.cpp.

                                 {
// **************************
  FindInputFiles();
  if(CovPos[kXSecPar].back() != "none") ReadModelFile();
}

ReadInputCovLegacy()

void MCMCProcessor::ReadInputCovLegacy ( )

inlineprotected

Warning

This will no longer be supported in future

Definition at line 2402 of file MCMCProcessor.cpp.

                                       {
// **************************
  FindInputFilesLegacy();
  if(nParam[kNDPar] > 0)    ReadNDFile();
  if(nParam[kFDDetPar] > 0) ReadFDFile();
}

ReadModelFile()

void MCMCProcessor::ReadModelFile ( )

inlineprotected

Read the xsec file and get the input central values and errors.

Definition at line 2525 of file MCMCProcessor.cpp.

                                  {
// ***************
  YAML::Node XSecFile = CovConfig[kXSecPar];
 
  auto systematics = XSecFile["Systematics"];
  int paramIndex  = 0;
  for (auto it = systematics.begin(); it != systematics.end(); ++it, ++paramIndex )
  {
    auto const &param = *it;
    // Push back the name
    std::string ParName = (param["Systematic"]["Names"]["FancyName"].as<std::string>());
    std::string Group = param["Systematic"]["ParameterGroup"].as<std::string>();
 
    bool rejected = false;
    for (unsigned int ik = 0; ik < ExcludedNames.size(); ++ik)
    {
      if (ParName.rfind(ExcludedNames[ik], 0) == 0)
      {
        MACH3LOG_DEBUG("Excluding param {}, from group {}", ParName, Group);
        rejected = true;
        break;
      }
    }
    for (unsigned int ik = 0; ik < ExcludedGroups.size(); ++ik)
    {
      if (Group == ExcludedGroups[ik])
      {
        MACH3LOG_DEBUG("Excluding param {}, from group {}", ParName, Group);
        rejected = true;
        break;
      }
    }
    if(rejected) continue;
 
    ParamNames[kXSecPar].push_back(ParName);
    ParamCentral[kXSecPar].push_back(param["Systematic"]["ParameterValues"]["PreFitValue"].as<double>());
    ParamNom[kXSecPar].push_back(param["Systematic"]["ParameterValues"]["Generated"].as<double>());
    ParamErrors[kXSecPar].push_back(param["Systematic"]["Error"].as<double>() );
    ParamFlat[kXSecPar].push_back(GetFromManager<bool>(param["Systematic"]["FlatPrior"], false));
 
    ParameterGroup.push_back(Group);
 
    nParam[kXSecPar]++;
    ParamType.push_back(kXSecPar);
    // Params from osc group have branch name equal to fancy name while all others are basically xsec_0 for example
    if(ParameterGroup.back() == "Osc") {
      BranchNames.push_back(ParamNames[kXSecPar].back());
    } else {
      BranchNames.push_back("xsec_" + std::to_string(paramIndex));
    }
 
    // Check that the branch exists before setting address
    if (!Chain->GetBranch(BranchNames.back())) {
      MACH3LOG_ERROR("Couldn't find branch '{}'", BranchNames.back());
      throw MaCh3Exception(__FILE__, __LINE__);
    }
  }
}

ReadNDFile()

void MCMCProcessor::ReadNDFile ( )

inlineprotected

Read the ND cov file and get the input central values and errors.

Definition at line 2586 of file MCMCProcessor.cpp.

                               {
// ***************
  // Do the same for the ND280
  TFile *NDdetFile = new TFile(CovPos[kNDPar].back().c_str(), "open");
  if (NDdetFile->IsZombie()) {
    MACH3LOG_ERROR("Couldn't find NDdetFile {}", CovPos[kNDPar].back());
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  NDdetFile->cd();
 
  TMatrixDSym *NDdetMatrix = NDdetFile->Get<TMatrixDSym>(CovNamePos[kNDPar].c_str());
  TVectorD *NDdetNominal = NDdetFile->Get<TVectorD>("det_weights");
  TDirectory *BinningDirectory = NDdetFile->Get<TDirectory>("Binning");
 
  for (int i = 0; i < NDdetNominal->GetNrows(); ++i)
  {
    ParamNom[kNDPar].push_back( (*NDdetNominal)(i) );
    ParamCentral[kNDPar].push_back( (*NDdetNominal)(i) );
 
    ParamErrors[kNDPar].push_back( std::sqrt((*NDdetMatrix)(i,i)) );
    ParamNames[kNDPar].push_back( Form("ND Det %i", i) );
    //KS: Currently we can only set it via config, change it in future
    ParamFlat[kNDPar].push_back( false );
  }
 
  TIter next(BinningDirectory->GetListOfKeys());
  TKey *key = nullptr;
  // Loop through all entries
  while ((key = static_cast<TKey*>(next())))
  {
    std::string name = std::string(key->GetName());
    TH2Poly* RefPoly = BinningDirectory->Get<TH2Poly>((name).c_str());
    int size = RefPoly->GetNumberOfBins();
    NDSamplesBins.push_back(size);
    NDSamplesNames.push_back(RefPoly->GetTitle());
  }
 
  NDdetFile->Close();
  delete NDdetFile;
}

ResetHistograms()

void MCMCProcessor::ResetHistograms

(

)

Reset 2D posteriors, in case we would like to calculate in again with different BurnInCut.

Definition at line 2729 of file MCMCProcessor.cpp.

                                    {
// **************************************************
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int i = 0; i < nDraw; ++i) 
  {
    for (int j = 0; j <= i; ++j)
    {
      // TH2D to hold the Correlation 
      hpost2D[i][j]->Reset("");
      hpost2D[i][j]->Fill(0.0, 0.0, 0.0);
    }
  }
}

ReweightPrior()

void MCMCProcessor::ReweightPrior	(	const std::vector< std::string > &	Names,
		const std::vector< double > &	NewCentral,
		const std::vector< double > &	NewError
	)

Reweight Prior by giving new central value and new error.

Parameters

Names	Parameter names for which we do reweighting
NewCentral	New central value for which we reweight
NewError	New error used for calculating weight

Definition at line 3015 of file MCMCProcessor.cpp.

                                                                     {
// **************************
  MACH3LOG_INFO("Reweighting Prior");
 
  if( (Names.size() != NewCentral.size()) || (NewCentral.size() != NewError.size()))
  {
    MACH3LOG_ERROR("Size of passed vectors doesn't match in {}", __func__);
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  std::vector<int> Param;
  std::vector<double> OldCentral;
  std::vector<double> OldError;
  std::vector<bool> FlatPrior;
 
  //KS: First we need to find parameter number based on name
  for(unsigned int k = 0; k < Names.size(); ++k)
  {
    //KS: First we need to find parameter number based on name
    int ParamNo = GetParamIndexFromName(Names[k]);
    if(ParamNo == M3::_BAD_INT_)
    {
      MACH3LOG_WARN("Couldn't find param {}. Can't reweight Prior", Names[k]);
      return;
    }
 
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(ParamNo, Prior, PriorError, Title);
 
    Param.push_back(ParamNo);
    OldCentral.push_back(Prior);
    OldError.push_back(PriorError);
 
    ParameterEnum ParType = ParamType[ParamNo];
    int ParamTemp = ParamNo - ParamTypeStartPos[ParType];
 
    FlatPrior.push_back(ParamFlat[ParType][ParamTemp]);
  }
  std::vector<double> ParameterPos(Names.size());
 
  std::string InputFile = MCMCFile+".root";
  std::string OutputFilename = MCMCFile + "_reweighted.root";
 
  //KS: Simply create copy of file and add there new branch
  int ret = system(("cp " + InputFile + " " + OutputFilename).c_str());
  if (ret != 0)
    MACH3LOG_WARN("Error: system call to copy file failed with code {}", ret);
 
  TFile *OutputChain = new TFile(OutputFilename.c_str(), "UPDATE");
  OutputChain->cd();
  TTree *post = OutputChain->Get<TTree>("posteriors");
 
  double Weight = 1.;
 
  post->SetBranchStatus("*",false);
  // Set the branch addresses for params
  for (unsigned int j = 0; j < Names.size(); ++j) {
    post->SetBranchStatus(BranchNames[Param[j]].Data(), true);
    post->SetBranchAddress(BranchNames[Param[j]].Data(), &ParameterPos[j]);
  }
  TBranch *bpt = post->Branch("Weight", &Weight, "Weight/D");
  post->SetBranchStatus("Weight", true);
 
  for (int i = 0; i < nEntries; ++i)
  {
    post->GetEntry(i);
    Weight = 1.;
 
    //KS: Calculate reweight weight. Weights are multiplicative so we can do several reweights at once. FIXME Big limitation is that code only works for uncorrelated parameters :(
    for (unsigned int j = 0; j < Names.size(); ++j)
    {
      double new_chi = (ParameterPos[j] - NewCentral[j])/NewError[j];
      double new_prior = std::exp(-0.5 * new_chi * new_chi);
 
      double old_chi = -1;
      double old_prior = -1;
      if(FlatPrior[j]) {
        old_prior = 1.0;
      } else {
        old_chi = (ParameterPos[j] - OldCentral[j])/OldError[j];
        old_prior = std::exp(-0.5 * old_chi * old_chi);
      }
      Weight *= new_prior/old_prior;
    }
    bpt->Fill();
  }
  post->SetBranchStatus("*",true);
  OutputChain->cd();
  post->Write("posteriors", TObject::kOverwrite);
 
  // KS: Save reweight metadeta
  std::ostringstream yaml_stream;
  yaml_stream << "Weight:\n";
  for (size_t k = 0; k < Names.size(); ++k) {
    yaml_stream << "    " << Names[k] << ": [" << NewCentral[k] << ", " << NewError[k] << "]\n";
  }
  std::string yaml_string = yaml_stream.str();
  YAML::Node root = STRINGtoYAML(yaml_string);
  TMacro ConfigSave = YAMLtoTMacro(root, "Reweight_Config");
  ConfigSave.Write();
 
  OutputChain->Close();
  delete OutputChain;
 
  OutputFile->cd();
}

SavageDickeyPlot()

void MCMCProcessor::SavageDickeyPlot	(	std::unique_ptr< TH1D > &	PriorHist,
		std::unique_ptr< TH1D > &	PosteriorHist,
		const std::string &	Title,
		const double	EvaluationPoint
	)		const

Produce Savage Dickey plot.

Parameters

PriorHist	Histogram with prior distribution
PosteriorHist	Histogram with posterior distribution

Definition at line 2953 of file MCMCProcessor.cpp.

                                                                         {
// **************************
  // Area normalise the distributions
  PriorHist->Scale(1./PriorHist->Integral(), "width");
  PosteriorHist->Scale(1./PosteriorHist->Integral(), "width");
 
  PriorHist->SetLineColor(kRed);
  PriorHist->SetMarkerColor(kRed);
  PriorHist->SetFillColorAlpha(kRed, 0.35);
  PriorHist->SetFillStyle(1001);
  PriorHist->GetXaxis()->SetTitle(Title.c_str());
  PriorHist->GetYaxis()->SetTitle("Posterior Probability");
  PriorHist->SetMaximum(PosteriorHist->GetMaximum()*1.5);
  PriorHist->GetYaxis()->SetLabelOffset(999);
  PriorHist->GetYaxis()->SetLabelSize(0);
  PriorHist->SetLineWidth(2);
  PriorHist->SetLineStyle(kSolid);
 
  PosteriorHist->SetLineColor(kBlue);
  PosteriorHist->SetMarkerColor(kBlue);
  PosteriorHist->SetFillColorAlpha(kBlue, 0.35);
  PosteriorHist->SetFillStyle(1001);
 
  PriorHist->Draw("hist");
  PosteriorHist->Draw("hist same");
 
  double ProbPrior = PriorHist->GetBinContent(PriorHist->FindBin(EvaluationPoint));
  //KS: In case we go so far away that prior is 0, set this to small value to avoid dividing by 0
  if(ProbPrior < 0) ProbPrior = 0.00001;
  double ProbPosterior = PosteriorHist->GetBinContent(PosteriorHist->FindBin(EvaluationPoint));
  double SavageDickey = ProbPosterior/ProbPrior;
 
  std::string DunneKabothScale = GetDunneKaboth(SavageDickey);
  //Get Best point
  auto PostPoint = std::make_unique<TGraph>(1);
  PostPoint->SetPoint(0, EvaluationPoint, ProbPosterior);
  PostPoint->SetMarkerStyle(20);
  PostPoint->SetMarkerSize(1);
  PostPoint->Draw("P same");
 
  auto PriorPoint = std::make_unique<TGraph>(1);
  PriorPoint->SetPoint(0, EvaluationPoint, ProbPrior);
  PriorPoint->SetMarkerStyle(20);
  PriorPoint->SetMarkerSize(1);
  PriorPoint->Draw("P same");
 
  auto legend = std::make_unique<TLegend>(0.12, 0.6, 0.6, 0.97);
  SetLegendStyle(legend.get(), 0.04);
  legend->AddEntry(PriorHist.get(), "Prior", "l");
  legend->AddEntry(PosteriorHist.get(), "Posterior", "l");
  legend->AddEntry(PostPoint.get(), Form("SavageDickey = %.2f, (%s)", SavageDickey, DunneKabothScale.c_str()),"");
  legend->Draw("same");
 
  Posterior->Print(CanvasName);
  Posterior->Write(Title.c_str());
}

ScanInput()

void MCMCProcessor::ScanInput ( )

inlineprotected

Scan Input etc.

Definition at line 2151 of file MCMCProcessor.cpp.

                              {
// **************************
  // KS: This can reduce time necessary for caching even by half
  #ifdef MULTITHREAD
  //ROOT::EnableImplicitMT();
  #endif
 
  // Open the Chain
  Chain = new TChain("posteriors","posteriors");
  Chain->Add(MCMCFile.c_str());
 
  nEntries = int(Chain->GetEntries());
  
  //Only is suboptimality we might want to change it, therefore set it high enough so it doesn't affect other functionality
  UpperCut = nEntries+1;
 
  // Get the list of branches
  TObjArray* brlis = Chain->GetListOfBranches();
 
  // Get the number of branches
  nBranches = brlis->GetEntries();
 
  BranchNames.reserve(nBranches);
  ParamType.reserve(nBranches);
 
  // Read the input Covariances
  ReadInputCov();
 
  // Set all the branches to off
  Chain->SetBranchStatus("*", false);
 
  // Loop over the number of branches
  // Find the name and how many of each systematic we have
  for (int i = 0; i < nBranches; i++)
  {
    // Get the TBranch and its name
    TBranch* br = static_cast<TBranch*>(brlis->At(i));
    if(!br){
      MACH3LOG_ERROR("Invalid branch at position {}", i);
      throw MaCh3Exception(__FILE__,__LINE__);
    }
    TString bname = br->GetName();
 
    //KS: Exclude parameter types
    bool rejected = false;
    for(unsigned int ik = 0; ik < ExcludedTypes.size(); ++ik )
    {
      if(bname.BeginsWith(ExcludedTypes[ik]))
      {
        rejected = true;
        break;
      }
    }
    if(rejected) continue;
 
    // Turn on the branches which we want for parameters
    Chain->SetBranchStatus(bname.Data(), true);
 
    if (bname.BeginsWith("ndd_"))
    {
      BranchNames.push_back(bname);
      ParamType.push_back(kNDPar);
      nParam[kNDPar]++;
    }
    else if (bname.BeginsWith("skd_joint_"))
    {
      BranchNames.push_back(bname);
      ParamType.push_back(kFDDetPar);
      nParam[kFDDetPar]++;
    }
 
    //KS: as a bonus get LogL systematic
    if (bname.BeginsWith("LogL_sample_")) {
      SampleName_v.push_back(bname);
      nSamples++;
    }
    else if (bname.BeginsWith("LogL_systematic_")) {
      SystName_v.push_back(bname);
      nSysts++;
    }
  }
  nDraw = int(BranchNames.size());
 
  // Read the input Covariances
  ReadInputCovLegacy();
  
  // Check order of parameter types
  ScanParameterOrder();
 
  IamVaried.resize(nDraw, true);
 
  // Print useful Info
  PrintInfo();
 
  nSteps = Chain->GetMaximum("step");
  // Set the step cut to be 20%
  int cut = nSteps/5;
  SetStepCut(cut);
 
  // Basically allow loading oscillation parameters
  LoadAdditionalInfo();
}

ScanParameterOrder()

void MCMCProcessor::ScanParameterOrder ( )

inlineprotected

Scan order of params from a different groups.

Definition at line 2313 of file MCMCProcessor.cpp.

                                       {
// *****************************
  for(int i = 0; i < kNParameterEnum; i++)
  {
    for(unsigned int j = 0; j < ParamType.size(); j++)
    {
      if(ParamType[j] == ParameterEnum(i)) 
      {
        //KS: When we find that i-th parameter types start at j, save and move to the next parameter.
        ParamTypeStartPos[i] = j;
        break;
      }
    }
  }
}

SetEntries()

void MCMCProcessor::SetEntries ( const int  NewEntries )

inline

Set number of entries to make potentially MCMC Processing faster.

Warning

This option only sets an upper limit; burn-in events will NOT be discarded

Definition at line 262 of file MCMCProcessor.h.

                                                 {
      if (NewEntries > nEntries) {
        MACH3LOG_ERROR("Cannot increase entries from {} to {}. Only decreasing is allowed.", nEntries, NewEntries);
        throw MaCh3Exception(__FILE__, __LINE__);
      }
      if (NewEntries <= 0) {
        MACH3LOG_ERROR("Entries cannot be below 0, but {} was passed.", NewEntries);
        throw MaCh3Exception(__FILE__, __LINE__);
      }
 
      if (static_cast<int>(BurnInCut) > NewEntries) {
          MACH3LOG_ERROR("BurnInCut ({}) is larger than NewEntries ({})", BurnInCut, NewEntries);
          throw MaCh3Exception(__FILE__, __LINE__);
      }
 
      MACH3LOG_INFO("Setting entries to {} from {}.", NewEntries, nEntries);
      MACH3LOG_WARN("This may behave not as expected when using merged multiple chains");
      nEntries = NewEntries;
    }

SetExcludedGroups()

void MCMCProcessor::SetExcludedGroups ( std::vector< std::string >  Name )

inline

Definition at line 310 of file MCMCProcessor.h.

{ExcludedGroups = Name; };

SetExcludedNames()

void MCMCProcessor::SetExcludedNames ( std::vector< std::string >  Name )

inline

Definition at line 309 of file MCMCProcessor.h.

{ExcludedNames = Name; };

SetExcludedTypes()

void MCMCProcessor::SetExcludedTypes ( std::vector< std::string >  Name )

inline

Setter related what parameters we want to exclude from analysis, for example if cross-section parameters look like xsec_, then passing "xsec_" will.

Parameters

Batches

Vector with parameters type names we want to exclude

Definition at line 308 of file MCMCProcessor.h.

{ExcludedTypes = Name; };

SetFancyNames()

void MCMCProcessor::SetFancyNames ( const bool  PlotOrNot )

inline

Definition at line 297 of file MCMCProcessor.h.

{FancyPlotNames = PlotOrNot; };

SetLegendStyle()

void MCMCProcessor::SetLegendStyle ( TLegend *  Legend, const double  size  ) const

protected

Configures the style of a TLegend object.

Parameters

Legend	Pointer to the TLegend object to modify
size	The text size to set for the legend

Definition at line 4588 of file MCMCProcessor.cpp.

                                                                           {
// **************************
  Legend->SetTextSize(size);
  Legend->SetLineColor(0);
  Legend->SetLineStyle(0);
  Legend->SetFillColor(0);
  Legend->SetFillStyle(0);
  Legend->SetBorderSize(0);
}

SetMargins()

void MCMCProcessor::SetMargins ( std::unique_ptr< TCanvas > &  Canv, const std::vector< double > &  margins  )

protected

Set TCanvas margins to specified values.

Definition at line 4563 of file MCMCProcessor.cpp.

                                                                                             {
// **************************
  if (!Canv) {
    MACH3LOG_ERROR("Canv is nullptr");
    throw MaCh3Exception(__FILE__, __LINE__);
  }
  if (margins.size() != 4) {
    MACH3LOG_ERROR("Margin vector must have exactly 4 elements");
    throw MaCh3Exception(__FILE__, __LINE__);
  }
  Canv->SetTopMargin(margins[0]);
  Canv->SetBottomMargin(margins[1]);
  Canv->SetLeftMargin(margins[2]);
  Canv->SetRightMargin(margins[3]);
}

SetnBatches()

void MCMCProcessor::SetnBatches ( const int  Batches )

inline

Set value of Nbatches used for batched mean, this need to be done earlier as batches are made when reading tree.

Parameters

Batches

Number of batches, default is 20

Definition at line 314 of file MCMCProcessor.h.

{nBatches = Batches; };

SetNBins()

void MCMCProcessor::SetNBins ( const int  NewBins )

inline

Modify number of bins used for 1D and 2D Histograms.

Definition at line 258 of file MCMCProcessor.h.

{nBins = NewBins;};

SetnLags()

void MCMCProcessor::SetnLags ( const int  nLags )

inline

Definition at line 315 of file MCMCProcessor.h.

{AutoCorrLag = nLags; };

SetOutputSuffix()

void MCMCProcessor::SetOutputSuffix ( const std::string  Suffix )

inline

Sett output suffix, this way jobs using the same file will have different names.

Definition at line 318 of file MCMCProcessor.h.

{OutputSuffix = Suffix; };

SetPlotBinValue()

void MCMCProcessor::SetPlotBinValue ( const bool  PlotOrNot )

inline

Definition at line 296 of file MCMCProcessor.h.

{plotBinValue = PlotOrNot; };

SetPlotErrorForFlatPrior()

void MCMCProcessor::SetPlotErrorForFlatPrior ( const bool  PlotOrNot )

inline

Set whether you want to plot error for parameters which have flat prior.

Definition at line 295 of file MCMCProcessor.h.

{PlotFlatPrior = PlotOrNot; };

SetPlotRelativeToPrior()

void MCMCProcessor::SetPlotRelativeToPrior ( const bool  PlotOrNot )

inline

You can set relative to prior or relative to generated. It is advised to use relate to prior.

Parameters

PlotOrNot

bool controlling plotRelativeToPrior argument

Definition at line 292 of file MCMCProcessor.h.

{plotRelativeToPrior = PlotOrNot; };

SetPost2DPlotThreshold()

void MCMCProcessor::SetPost2DPlotThreshold ( const double  Threshold )

inline

Code will only plot 2D posteriors if Correlation are larger than defined threshold.

Parameters

Threshold

This threshold is compared with correlation value

Definition at line 302 of file MCMCProcessor.h.

{Post2DPlotThreshold = Threshold; };

SetPosterior1DCut()

void MCMCProcessor::SetPosterior1DCut ( const std::string  Cut )

inline

Allow to set addtional cuts based on ROOT TBrowser cut, for to only affect one mass ordering.

Definition at line 320 of file MCMCProcessor.h.

{Posterior1DCut = Cut; };

SetPrintToPDF()

void MCMCProcessor::SetPrintToPDF ( const bool  PlotOrNot )

inline

Definition at line 293 of file MCMCProcessor.h.

{printToPDF = PlotOrNot; };

SetSmoothing()

void MCMCProcessor::SetSmoothing ( const bool  PlotOrNot )

inline

Set whether want to use smoothing for histograms using ROOT algorithm.

Definition at line 299 of file MCMCProcessor.h.

{ApplySmoothing = PlotOrNot; };

SetStepCut() [1/2]

void MCMCProcessor::SetStepCut

(

const int

Cuts

)

Set the step cutting by int.

Parameters

Cuts	integer telling cut value

Definition at line 2674 of file MCMCProcessor.cpp.

                                             {
// ***************
  std::stringstream TempStream;
  TempStream << "step > " << Cuts;
  StepCut = TempStream.str();
  BurnInCut = Cuts;
  CheckStepCut();
}

SetStepCut() [2/2]

void MCMCProcessor::SetStepCut

(

const std::string &

Cuts

)

Set the step cutting by string.

Parameters

Cuts	string telling cut value

Definition at line 2664 of file MCMCProcessor.cpp.

                                                    {
// ***************
  StepCut = Cuts;
  BurnInCut = std::stoi( Cuts );
 
  CheckStepCut();
}

SetTLineStyle()

void MCMCProcessor::SetTLineStyle ( TLine *  Line, const Color_t  Colour, const Width_t  Width, const ELineStyle  Style  ) const

protected

Configures a TLine object with the specified style parameters.

Parameters

Line	Pointer to the TLine object to modify. Must not be nullptr.
Colour	The color to set for the line.
Width	The width of the line.
Style	The line style (e.g., solid, dashed, etc.).

Definition at line 4580 of file MCMCProcessor.cpp.

                                                                                                                      {
// **************************
  Line->SetLineColor(Colour);
  Line->SetLineWidth(Width);
  Line->SetLineStyle(Style);
}

SetupOutput()

void MCMCProcessor::SetupOutput ( )

inlineprotected

Prepare all objects used for output.

Definition at line 2256 of file MCMCProcessor.cpp.

                                {
// ****************************
  // Make sure we can read files located anywhere and strip the .root ending
  MCMCFile = MCMCFile.substr(0, MCMCFile.find(".root"));
 
  // Check if the output file is ready
  if (OutputFile == nullptr) MakeOutputFile();
  
  CanvasName = MCMCFile + OutputSuffix + ".pdf[";
  if(printToPDF) Posterior->Print(CanvasName);
 
  // Once the pdf file is open no longer need to bracket
  CanvasName.ReplaceAll("[","");
 
  // We fit with this Gaussian
  Gauss = std::make_unique<TF1>("Gauss", "[0]/sqrt(2.0*3.14159)/[2]*TMath::Exp(-0.5*pow(x-[1],2)/[2]/[2])", -5, 5);
  Gauss->SetLineWidth(2);
  Gauss->SetLineColor(kOrange-5);
 
  // Declare the TVectors
  Covariance = new TMatrixDSym(nDraw);
  Correlation = new TMatrixDSym(nDraw);
  Central_Value = new TVectorD(nDraw);
  Means = new TVectorD(nDraw);
  Errors = new TVectorD(nDraw);
  Means_Gauss = new TVectorD(nDraw);
  Errors_Gauss = new TVectorD(nDraw);
  Means_HPD    = new TVectorD(nDraw);
  Errors_HPD   = new TVectorD(nDraw);
  Errors_HPD_Positive = new TVectorD(nDraw);
  Errors_HPD_Negative = new TVectorD(nDraw);
 
  // Initialise to something silly
  #ifdef MULTITHREAD
  #pragma omp parallel for
  #endif
  for (int i = 0; i < nDraw; ++i)
  {
    (*Central_Value)(i) = M3::_BAD_DOUBLE_;
    (*Means)(i) = M3::_BAD_DOUBLE_;
    (*Errors)(i) = M3::_BAD_DOUBLE_;
    (*Means_Gauss)(i) = M3::_BAD_DOUBLE_;
    (*Errors_Gauss)(i) = M3::_BAD_DOUBLE_;
    (*Means_HPD)(i) = M3::_BAD_DOUBLE_;
    (*Errors_HPD)(i) = M3::_BAD_DOUBLE_;
    (*Errors_HPD_Positive)(i) = M3::_BAD_DOUBLE_;
    (*Errors_HPD_Negative)(i) = M3::_BAD_DOUBLE_;
    for (int j = 0; j < nDraw; ++j) {
      (*Covariance)(i, j) = M3::_BAD_DOUBLE_;
      (*Correlation)(i, j) = M3::_BAD_DOUBLE_;
    }
  }
  hpost.resize(nDraw);
}

SetUseFFTAutoCorrelation()

void MCMCProcessor::SetUseFFTAutoCorrelation ( const bool  useFFT )

inline

Toggle using the FFT-based autocorrelation calculator.

Definition at line 304 of file MCMCProcessor.h.

{useFFTAutoCorrelation = useFFT; };

SmearChain()

void MCMCProcessor::SmearChain	(	const std::vector< std::string > &	Names,
		const std::vector< double > &	NewCentral,
		const bool &	SaveBranch
	)

Smear chain contours.

Parameters

Names	Parameter names for which we do smearing
Error	Error based on which we smear
SaveBranch	Whether we save unsmeared branch or not

Definition at line 3127 of file MCMCProcessor.cpp.

                                                       {
// **************************
  MACH3LOG_INFO("Starting {}", __func__);
 
  if( (Names.size() != Error.size()))
  {
    MACH3LOG_ERROR("Size of passed vectors doesn't match in {}", __func__);
    throw MaCh3Exception(__FILE__ , __LINE__ );
  }
  std::vector<int> Param;
 
  //KS: First we need to find parameter number based on name
  for(unsigned int k = 0; k < Names.size(); ++k)
  {
    //KS: First we need to find parameter number based on name
    int ParamNo = GetParamIndexFromName(Names[k]);
    if(ParamNo == M3::_BAD_INT_)
    {
      MACH3LOG_WARN("Couldn't find param {}. Can't Smear", Names[k]);
      return;
    }
 
    TString Title = "";
    double Prior = 1.0, PriorError = 1.0;
    GetNthParameter(ParamNo, Prior, PriorError, Title);
 
    Param.push_back(ParamNo);
  }
  std::string InputFile = MCMCFile+".root";
  std::string OutputFilename = MCMCFile + "_smeared.root";
 
  //KS: Simply create copy of file and add there new branch
  int ret = system(("cp " + InputFile + " " + OutputFilename).c_str());
  if (ret != 0)
    MACH3LOG_WARN("Error: system call to copy file failed with code {}", ret);
 
  TFile *OutputChain = new TFile(OutputFilename.c_str(), "UPDATE");
  OutputChain->cd();
  TTree *post = OutputChain->Get<TTree>("posteriors");
  TTree *treeNew = post->CloneTree(0);
 
  std::vector<double> NewParameter(Names.size());
  for(size_t i = 0; i < Param.size(); i++) {
    post->SetBranchAddress(BranchNames[Param[i]], &NewParameter[i]);
  }
 
  std::vector<double> Unsmeared_Parameter;
  if(SaveBranch){
    Unsmeared_Parameter.resize(Param.size());
    for(size_t i = 0; i < Param.size(); i++) {
      treeNew->Branch((BranchNames[Param[i]] + "_unsmeared"), &Unsmeared_Parameter[i]);
    }
  }
 
  auto rand = std::make_unique<TRandom3>(0);
  Long64_t AllEntries = post->GetEntries();
  for (Long64_t i = 0; i < AllEntries; ++i) {
    // Entry from the old chain
    post->GetEntry(i);
 
    if(SaveBranch){
      for(size_t iPar = 0; iPar < Param.size(); iPar++) {
        Unsmeared_Parameter[iPar] = NewParameter[iPar];
      }
    }
    // Smear it
    for(size_t iPar = 0; iPar < Param.size(); iPar++) {
      NewParameter[iPar] = NewParameter[iPar] + rand->Gaus(0, Error[iPar]);
    }
    // Fill to the new chain
    treeNew->Fill();
  }
 
  OutputChain->cd();
  treeNew->Write("posteriors", TObject::kOverwrite);
 
  // KS: Save reweight metadeta
  std::ostringstream yaml_stream;
  yaml_stream << "Smearing:\n";
  for (size_t k = 0; k < Names.size(); ++k) {
    yaml_stream << "    " << Names[k] << ": [" << Error[k] << ", " << "Gauss" << "]\n";
  }
  std::string yaml_string = yaml_stream.str();
  YAML::Node root = STRINGtoYAML(yaml_string);
  TMacro ConfigSave = YAMLtoTMacro(root, "Smearing_Config");
  ConfigSave.Write();
 
  OutputChain->Close();
  delete OutputChain;
}

ThinMCMC()

void MCMCProcessor::ThinMCMC ( const int  ThinningCut ) const

inline

Thin MCMC Chain, to save space and maintain low autocorrelations.

Parameters

ThinningCut	every which entry you want to thin
Average	If true will perform MCMC averaging instead of thinning

Definition at line 203 of file MCMCProcessor.h.

{ ThinningMCMC(MCMCFile+".root", ThinningCut); };

Member Data Documentation

AccProbBatchedAverages

double* MCMCProcessor::AccProbBatchedAverages

protected

Holds all accProb in batches.

Definition at line 583 of file MCMCProcessor.h.

AccProbValues

double* MCMCProcessor::AccProbValues

protected

Holds all accProb.

Definition at line 581 of file MCMCProcessor.h.

ApplySmoothing

bool MCMCProcessor::ApplySmoothing

protected

Apply smoothing for 2D histos using root algorithm.

Definition at line 497 of file MCMCProcessor.h.

AutoCorrLag

int MCMCProcessor::AutoCorrLag

protected

LagL used in AutoCorrelation.

Definition at line 568 of file MCMCProcessor.h.

BatchedAverages

double** MCMCProcessor::BatchedAverages

protected

Values of batched average for every param and batch.

Definition at line 573 of file MCMCProcessor.h.

BranchNames

std::vector<TString> MCMCProcessor::BranchNames

protected

Definition at line 446 of file MCMCProcessor.h.

BurnInCut

unsigned int MCMCProcessor::BurnInCut

protected

Value of burn in cut.

Definition at line 431 of file MCMCProcessor.h.

CacheMCMC

bool MCMCProcessor::CacheMCMC

protected

MCMC Chain has been cached.

Definition at line 560 of file MCMCProcessor.h.

CanvasName

TString MCMCProcessor::CanvasName

protected

Name of canvas which help to save to the sample pdf.

Definition at line 480 of file MCMCProcessor.h.

Central_Value

TVectorD* MCMCProcessor::Central_Value

protected

Vector with central value for each parameter.

Definition at line 517 of file MCMCProcessor.h.

Chain

TChain* MCMCProcessor::Chain

protected

Main chain storing all steps etc.

Definition at line 423 of file MCMCProcessor.h.

Correlation

TMatrixDSym* MCMCProcessor::Correlation

protected

Posterior Correlation Matrix.

Definition at line 538 of file MCMCProcessor.h.

Covariance

TMatrixDSym* MCMCProcessor::Covariance

protected

Posterior Covariance Matrix.

Definition at line 536 of file MCMCProcessor.h.

CovConfig

std::vector<YAML::Node> MCMCProcessor::CovConfig

protected

Covariance matrix config.

Definition at line 420 of file MCMCProcessor.h.

CovNamePos

std::vector<std::string> MCMCProcessor::CovNamePos

protected

Covariance matrix name position.

Definition at line 418 of file MCMCProcessor.h.

CovPos

std::vector<std::vector<std::string> > MCMCProcessor::CovPos

protected

Covariance matrix file name position.

Definition at line 416 of file MCMCProcessor.h.

doDiagMCMC

bool MCMCProcessor::doDiagMCMC

protected

Doing MCMC Diagnostic.

Definition at line 562 of file MCMCProcessor.h.

DrawRange

double MCMCProcessor::DrawRange

protected

Drawrange for SetMaximum.

Definition at line 557 of file MCMCProcessor.h.

Errors

TVectorD* MCMCProcessor::Errors

protected

Vector with errors values using RMS.

Definition at line 521 of file MCMCProcessor.h.

Errors_Gauss

TVectorD* MCMCProcessor::Errors_Gauss

protected

Vector with error values using Gaussian fit.

Definition at line 525 of file MCMCProcessor.h.

Errors_HPD

TVectorD* MCMCProcessor::Errors_HPD

protected

Vector with error values using Highest Posterior Density.

Definition at line 529 of file MCMCProcessor.h.

Errors_HPD_Negative

TVectorD* MCMCProcessor::Errors_HPD_Negative

protected

Vector with negative error (left hand side) values using Highest Posterior Density.

Definition at line 533 of file MCMCProcessor.h.

Errors_HPD_Positive

TVectorD* MCMCProcessor::Errors_HPD_Positive

protected

Vector with positive error (right hand side) values using Highest Posterior Density.

Definition at line 531 of file MCMCProcessor.h.

ExcludedGroups

std::vector<std::string> MCMCProcessor::ExcludedGroups

protected

Definition at line 449 of file MCMCProcessor.h.

ExcludedNames

std::vector<std::string> MCMCProcessor::ExcludedNames

protected

Definition at line 448 of file MCMCProcessor.h.

ExcludedTypes

std::vector<std::string> MCMCProcessor::ExcludedTypes

protected

Definition at line 447 of file MCMCProcessor.h.

FancyPlotNames

bool MCMCProcessor::FancyPlotNames

protected

Whether we want fancy plot names or not.

Definition at line 493 of file MCMCProcessor.h.

Gauss

std::unique_ptr<TF1> MCMCProcessor::Gauss

protected

Gaussian fitter.

Definition at line 507 of file MCMCProcessor.h.

hpost

std::vector<TH1D*> MCMCProcessor::hpost

protected

Holds 1D Posterior Distributions.

Definition at line 541 of file MCMCProcessor.h.

hpost2D

std::vector<std::vector<TH2D*> > MCMCProcessor::hpost2D

protected

Holds 2D Posterior Distributions.

Definition at line 543 of file MCMCProcessor.h.

hviolin

std::unique_ptr<TH2D> MCMCProcessor::hviolin

protected

Holds violin plot for all dials.

Definition at line 545 of file MCMCProcessor.h.

hviolin_prior

std::unique_ptr<TH2D> MCMCProcessor::hviolin_prior

protected

Holds prior violin plot for all dials,.

Definition at line 547 of file MCMCProcessor.h.

IamVaried

std::vector<bool> MCMCProcessor::IamVaried

protected

Is the ith parameter varied.

Definition at line 452 of file MCMCProcessor.h.

MadePostfit

bool MCMCProcessor::MadePostfit

protected

Sanity check if Postfit is already done to not make several times.

Definition at line 489 of file MCMCProcessor.h.

MCMCFile

std::string MCMCProcessor::MCMCFile

protected

Name of MCMC file.

Definition at line 412 of file MCMCProcessor.h.

Means

TVectorD* MCMCProcessor::Means

protected

Vector with mean values using Arithmetic Mean.

Definition at line 519 of file MCMCProcessor.h.

Means_Gauss

TVectorD* MCMCProcessor::Means_Gauss

protected

Vector with mean values using Gaussian fit.

Definition at line 523 of file MCMCProcessor.h.

Means_HPD

TVectorD* MCMCProcessor::Means_HPD

protected

Vector with mean values using Highest Posterior Density.

Definition at line 527 of file MCMCProcessor.h.

nBatches

int MCMCProcessor::nBatches

protected

Number of batches for Batched Mean.

Definition at line 566 of file MCMCProcessor.h.

nBins

int MCMCProcessor::nBins

protected

Number of bins.

Definition at line 555 of file MCMCProcessor.h.

nBranches

int MCMCProcessor::nBranches

protected

Number of branches in a TTree.

Definition at line 433 of file MCMCProcessor.h.

nDraw

int MCMCProcessor::nDraw

protected

Number of all parameters used in the analysis.

Definition at line 443 of file MCMCProcessor.h.

NDSamplesBins

std::vector<int> MCMCProcessor::NDSamplesBins

protected

Definition at line 503 of file MCMCProcessor.h.

NDSamplesNames

std::vector<std::string> MCMCProcessor::NDSamplesNames

protected

Definition at line 504 of file MCMCProcessor.h.

nEntries

int MCMCProcessor::nEntries

protected

KS: For merged chains number of entries will be different from nSteps.

Definition at line 435 of file MCMCProcessor.h.

nParam

std::vector<int> MCMCProcessor::nParam

protected

Number of parameters per type.

Definition at line 463 of file MCMCProcessor.h.

nSamples

int MCMCProcessor::nSamples

protected

Number of sample PDF objects.

Definition at line 439 of file MCMCProcessor.h.

nSteps

int MCMCProcessor::nSteps

protected

KS: For merged chains number of entries will be different from nSteps.

Definition at line 437 of file MCMCProcessor.h.

nSysts

int MCMCProcessor::nSysts

protected

Number of covariance objects.

Definition at line 441 of file MCMCProcessor.h.

OutputFile

TFile* MCMCProcessor::OutputFile

protected

The output file.

Definition at line 510 of file MCMCProcessor.h.

OutputName

std::string MCMCProcessor::OutputName

protected

Name of output files.

Definition at line 478 of file MCMCProcessor.h.

OutputSuffix

std::string MCMCProcessor::OutputSuffix

protected

Output file suffix useful when running over same file with different settings.

Definition at line 414 of file MCMCProcessor.h.

ParamCentral

std::vector<std::vector<double> > MCMCProcessor::ParamCentral

protected

Parameters central values which we are going to analyse.

Definition at line 456 of file MCMCProcessor.h.

ParamErrors

std::vector<std::vector<double> > MCMCProcessor::ParamErrors

protected

Uncertainty on a single parameter.

Definition at line 459 of file MCMCProcessor.h.

ParameterGroup

std::vector<std::string> MCMCProcessor::ParameterGroup

protected

Definition at line 470 of file MCMCProcessor.h.

ParamFlat

std::vector<std::vector<bool> > MCMCProcessor::ParamFlat

protected

Whether Param has flat prior or not.

Definition at line 461 of file MCMCProcessor.h.

ParamNames

std::vector<std::vector<TString> > MCMCProcessor::ParamNames

protected

Name of parameters which we are going to analyse.

Definition at line 454 of file MCMCProcessor.h.

ParamNom

std::vector<std::vector<double> > MCMCProcessor::ParamNom

protected

Definition at line 457 of file MCMCProcessor.h.

ParamSums

double* MCMCProcessor::ParamSums

protected

Total parameter sum for each param.

Definition at line 571 of file MCMCProcessor.h.

ParamType

std::vector<ParameterEnum> MCMCProcessor::ParamType

protected

Make an enum for which class this parameter belongs to so we don't have to keep string comparing.

Definition at line 465 of file MCMCProcessor.h.

ParamTypeStartPos

std::vector<int> MCMCProcessor::ParamTypeStartPos

protected

KS: in MCMC output there is order of parameters so for example first goes xsec then nd det etc. Idea is that this parameter will keep track of it so code is flexible

Definition at line 468 of file MCMCProcessor.h.

ParStep

double** MCMCProcessor::ParStep

protected

Array holding values for all parameters.

Definition at line 550 of file MCMCProcessor.h.

plotBinValue

bool MCMCProcessor::plotBinValue

protected

If true it will print value on each bin of covariance matrix.

Definition at line 495 of file MCMCProcessor.h.

PlotFlatPrior

bool MCMCProcessor::PlotFlatPrior

protected

Whether we plot flat prior or not, we usually provide error even for flat prior params.

Definition at line 483 of file MCMCProcessor.h.

plotRelativeToPrior

bool MCMCProcessor::plotRelativeToPrior

protected

Whether we plot relative to prior or nominal, in most cases is prior.

Definition at line 487 of file MCMCProcessor.h.

Post2DPlotThreshold

double MCMCProcessor::Post2DPlotThreshold

protected

KS: Set Threshold when to plot 2D posterior as by default we get a LOT of plots.

Definition at line 499 of file MCMCProcessor.h.

Posterior

std::unique_ptr<TCanvas> MCMCProcessor::Posterior

protected

Fancy canvas used for our beautiful plots.

Definition at line 513 of file MCMCProcessor.h.

Posterior1DCut

std::string MCMCProcessor::Posterior1DCut

protected

Cut used when making 1D Posterior distribution.

Definition at line 427 of file MCMCProcessor.h.

printToPDF

bool MCMCProcessor::printToPDF

protected

Will plot all plot to PDF not only to root file.

Definition at line 491 of file MCMCProcessor.h.

ReweightName

std::string MCMCProcessor::ReweightName

protected

Name of branch used for chain reweighting.

Definition at line 588 of file MCMCProcessor.h.

ReweightPosterior

bool MCMCProcessor::ReweightPosterior

protected

Whether to apply reweighting weight or not.

Definition at line 586 of file MCMCProcessor.h.

SampleName_v

std::vector<TString> MCMCProcessor::SampleName_v

protected

Vector of each systematic.

Definition at line 473 of file MCMCProcessor.h.

SampleValues

double** MCMCProcessor::SampleValues

protected

Holds the sample values.

Definition at line 576 of file MCMCProcessor.h.

StepCut

std::string MCMCProcessor::StepCut

protected

BurnIn Cuts.

Definition at line 425 of file MCMCProcessor.h.

StepNumber

unsigned int* MCMCProcessor::StepNumber

protected

Step number for step, important if chains were merged.

Definition at line 552 of file MCMCProcessor.h.

SystName_v

std::vector<TString> MCMCProcessor::SystName_v

protected

Vector of each sample PDF object.

Definition at line 475 of file MCMCProcessor.h.

SystValues

double** MCMCProcessor::SystValues

protected

Holds the systs values.

Definition at line 578 of file MCMCProcessor.h.

UpperCut

unsigned int MCMCProcessor::UpperCut

protected

KS: Used only for SubOptimality.

Definition at line 429 of file MCMCProcessor.h.

useFFTAutoCorrelation

bool MCMCProcessor::useFFTAutoCorrelation

protected

MJR: Use FFT-based autocorrelation algorithm (save time & resources)?

Definition at line 501 of file MCMCProcessor.h.

WeightValue

double* MCMCProcessor::WeightValue

protected

Stores value of weight for each step.

Definition at line 590 of file MCMCProcessor.h.

---

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

• Fitters/MCMCProcessor.h
• Fitters/MCMCProcessor.cpp