MCMCProcessor.h

Go to the documentation of this file.

#pragma once
 
// C++ includes
#include <complex>
#include <cstdio>
 
// MaCh3 includes
#include "Fitters/StatisticalUtils.h"
#include "Samples/HistogramUtils.h"
#include "Parameters/ParameterHandlerUtils.h"
 
_MaCh3_Safe_Include_Start_ //{
// ROOT includes
#include "TFile.h"
#include "TBranch.h"
#include "TCanvas.h"
#include "TLine.h"
#include "TLegend.h"
#include "TString.h"
#include "TGraphErrors.h"
#include "TVectorD.h"
#include "TColor.h"
#include "TStyle.h"
#include "TStopwatch.h"
#include "TText.h"
#include "TGaxis.h"
#include "TTree.h"
#include "TROOT.h"
#include "TKey.h"
#include "TRandom3.h"
#include "TGraphPolar.h"
#include "TCandle.h"
#include "TMath.h"
#include "TMatrixDSymEigen.h"
_MaCh3_Safe_Include_End_ //}
 
 
//KS: Joy of forward declaration https://gieseanw.wordpress.com/2018/02/25/the-joys-of-forward-declarations-results-from-the-real-world/
class TChain;
class TF1;
class MCMCProcessorGPU;
 
enum ParameterEnum {
  kXSecPar  = 0,
  kNDPar    = 1,
  kFDDetPar = 2,
  
  kNParameterEnum = 3 //KS: keep it at the end to keep track of all parameters
};
 
class MCMCProcessor {
  public:
    MCMCProcessor(const std::string &InputFile);
    virtual ~MCMCProcessor();
 
    void Initialise();
    void MakePostfit(const std::map<std::string, std::pair<double, double>>& Edges = {});
    void MakeCovariance();
    void CacheSteps();
    void MakeCovariance_MP(const bool Mute = false);
    void MakeSubOptimality(const int NIntervals = 10);
 
    void Reset2DPosteriors();
        
    void DrawPostfit();
    void MakeViolin();
    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
                               );
    void DrawCovariance();
    void MakeCovarianceYAML(const std::string& OutputYAMLFile, const std::string& MeansMethod) const;
    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
                             );
    void MakeTrianglePlot(const std::vector<std::string>& ParNames,
                          // 1D
                          const std::vector<double>& CredibleIntervals = {0.99, 0.90, 0.68 },
                          const std::vector<Color_t>& CredibleIntervalsColours = {kCyan+4, kCyan-2, kCyan-10},
                          //2D
                          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},
                          // Other
                          const bool CredibleInSigmas = false
                          );
 
    void CheckCredibleIntervalsOrder(const std::vector<double>& CredibleIntervals, const std::vector<Color_t>& CredibleIntervalsColours) const;
 
    void CheckCredibleRegionsOrder(const std::vector<double>& CredibleRegions,
                                   const std::vector<Style_t>& CredibleRegionStyle,
                                   const std::vector<Color_t>& CredibleRegionColor);
 
    void GetPolarPlot(const std::vector<std::string>& ParNames);
 
    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);
    void GetSavageDickey(const std::vector<std::string>& ParName,
                         const std::vector<double>& EvaluationPoint,
                         const std::vector<std::vector<double>>& Bounds);
 
    void SavageDickeyPlot(std::unique_ptr<TH1D>& PriorHist,
                          std::unique_ptr<TH1D>& PosteriorHist,
                          const std::string& Title,
                          const double EvaluationPoint) const ;
 
    void ProduceChi2(const std::string& GroupName) const;
 
    void ReweightPrior(const std::vector<std::string>& Names,
                       const std::vector<double>& NewCentral,
                       const std::vector<double>& NewError);
    
 
    void SmearChain(const std::vector<std::string>& Names,
                    const std::vector<double>& Error,
                    const bool& SaveBranch) const;
 
    void ParameterEvolution(const std::vector<std::string>& Names,
                            const std::vector<int>& NIntervals);
 
    void ThinMCMC(const int ThinningCut) const { ThinningMCMC(MCMCFile+".root", ThinningCut); };
 
    void DiagMCMC();
    
    // Get the number of parameters
    int GetNParams() const { return nDraw; };
    int GetNXSec() const { return nParam[kXSecPar]; };
    int GetNND() const { return nParam[kNDPar]; };
    int GetNFD() const { return nParam[kFDDetPar]; };
 
    YAML::Node GetCovConfig(const int i) const {return CovConfig.at(i); }
 
    int GetGroup(const std::string& name) const;
 
    TH1D* GetHpost(const int i) const { return hpost[i]; };
    TH2D* GetHpost2D(const int i, const int j) const { return hpost2D[i][j]; };
    TH2D* GetViolin() const { return hviolin.get(); };
    TH2D* GetViolinPrior() const { return hviolin_prior.get(); };
 
    //Covariance getters
    std::vector<std::string> GetXSecCov()  const { return CovPos[kXSecPar]; };
    std::string GetNDCov() const { return CovPos[kNDPar].back(); };
    std::string GetFDCov() const { return CovPos[kFDDetPar].back(); };
 
    void GetPostfit(TVectorD *&Central, TVectorD *&Errors, TVectorD *&Central_Gauss, TVectorD *&Errors_Gauss, TVectorD *&Peaks);
    void GetCovariance(TMatrixDSym *&Cov, TMatrixDSym *&Corr);
    void GetPostfit_Ind(TVectorD *&Central, TVectorD *&Errors, TVectorD *&Peaks, ParameterEnum kParam);
    
    const std::vector<TString>& GetBranchNames() const { return BranchNames;};
    void GetNthParameter(const int param, double &Prior, double &PriorError, TString &Title) const;
    int GetParamIndexFromName(const std::string& Name) const;
    Long64_t GetnEntries(){return nEntries;};
    Long64_t GetnSteps(){return nSteps;};
    void SetNBins(const int NewBins) {nBins = NewBins;};
 
    void SetEntries(const int NewEntries) {
      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;
    }
    void SetStepCut(const std::string& Cuts);
    void SetStepCut(const int Cuts);
    void CheckStepCut() const;
 
    void SetPlotRelativeToPrior(const bool PlotOrNot){plotRelativeToPrior = PlotOrNot; };
    void SetPrintToPDF(const bool PlotOrNot){printToPDF = PlotOrNot; };
    void SetPlotErrorForFlatPrior(const bool PlotOrNot){PlotFlatPrior = PlotOrNot; };
    void SetPlotBinValue(const bool PlotOrNot){plotBinValue = PlotOrNot; };
    void SetFancyNames(const bool PlotOrNot){FancyPlotNames = PlotOrNot; };
    void SetSmoothing(const bool PlotOrNot){ApplySmoothing = PlotOrNot; };
    void SetPost2DPlotThreshold(const double Threshold){Post2DPlotThreshold = Threshold; };
    void SetUseFFTAutoCorrelation(const bool useFFT){useFFTAutoCorrelation = useFFT; };
 
    void SetExcludedTypes(std::vector<std::string> Name){ExcludedTypes = Name; };
    void SetExcludedNames(std::vector<std::string> Name){ExcludedNames = Name; };
    void SetExcludedGroups(std::vector<std::string> Name){ExcludedGroups = Name; };
 
    void SetnBatches(const int Batches){nBatches = Batches; };
    void SetnLags(const int nLags){AutoCorrLag = nLags; };
    
    void SetOutputSuffix(const std::string Suffix){OutputSuffix = Suffix; };
    void SetPosterior1DCut(const std::string Cut){Posterior1DCut = Cut; };
 
  protected:
      std::unique_ptr<TH1D> MakePrefit();
    void MakeOutputFile();
    void DrawCorrelations1D();
    void DrawCorrelationsGroup(const std::unique_ptr<TH2D>& CorrMatrix) const;
    void ReadInputCov();
    void ReadInputCovLegacy();
    void FindInputFiles();
    void FindInputFilesLegacy();
    void ReadModelFile();
    virtual void LoadAdditionalInfo() {};
    void ReadNDFile();
    void ReadFDFile();
    void PrintInfo() const;
 
    void ScanInput();
    void ScanParameterOrder();
    void SetupOutput();
 
    // MCMC Diagnostic
    void PrepareDiagMCMC();
    std::vector <double> GetParameterSums();
    void ParamTraces();
    void AutoCorrelation();
    void AutoCorrelation_FFT();
    void CalculateESS(const int nLags, const std::vector<std::vector<double>>& LagL);
    void BatchedAnalysis();
    void BatchedMeans();
    void GewekeDiagnostic();
    void AcceptanceProbabilities();
    void PowerSpectrumAnalysis();
 
    std::vector<double> GetMargins(const std::unique_ptr<TCanvas>& Canv) const;
    void SetMargins(std::unique_ptr<TCanvas>& Canv, const std::vector<double>& margins);
    void SetTLineStyle(TLine* Line, const Color_t Colour, const Width_t Width, const ELineStyle Style) const;
    void SetLegendStyle(TLegend* Legend, const double size) const;
 
    std::string MCMCFile;
    std::string OutputSuffix;
    std::vector<std::vector<std::string>> CovPos;
    std::vector<std::string> CovNamePos;
    std::vector<YAML::Node> CovConfig;
 
    TChain *Chain;
    std::string StepCut;
    std::string Posterior1DCut;
    unsigned int UpperCut;
    unsigned int BurnInCut;
    int nBranches;
    int nEntries;
    int nSteps;
    int nSampleHandlers;
    int nParameterHandlers;
    int nDraw;
    
    //Name of all branches as well as branches we don't want to include in the analysis
    std::vector<TString> BranchNames;
    std::vector<std::string> ExcludedTypes;
    std::vector<std::string> ExcludedNames;
    std::vector<std::string> ExcludedGroups;
 
    std::vector<bool> ParamVaried;
    std::vector<std::vector<TString>> ParamNames;
    std::vector<std::vector<double>>  ParamCentral;
    std::vector<std::vector<double>>  ParamErrors;
    std::vector<std::vector<bool>>    ParamFlat;
    std::vector<int> nParam;
    std::vector<ParameterEnum> ParamType;
    std::vector<int> ParamTypeStartPos;
    // KS: For example flux or detector within matrix
    std::vector<std::string> ParameterGroup;
 
    std::vector<TString> SampleName_v;
    std::vector<TString> SystName_v;
    
    std::string OutputName;
    TString CanvasName;
 
    bool PlotFlatPrior;
    
    //Even more flags
    bool plotRelativeToPrior;
    bool MadePostfit;
    bool printToPDF;
    bool FancyPlotNames;
    bool plotBinValue;
    bool ApplySmoothing;
    double Post2DPlotThreshold;
    bool useFFTAutoCorrelation;
 
    std::vector<int> NDSamplesBins;
    std::vector<std::string> NDSamplesNames;
 
    std::unique_ptr<TF1> Gauss;
 
    TFile *OutputFile;
    
    std::unique_ptr<TCanvas> Posterior;
 
    //Vector of best fit points and errors obtained with different methods
    TVectorD *Central_Value;
    TVectorD *Means;
    TVectorD *Errors;
    TVectorD *Means_Gauss;
    TVectorD *Errors_Gauss;
    TVectorD *Means_HPD;
    TVectorD *Errors_HPD;
    TVectorD *Errors_HPD_Positive;
    TVectorD *Errors_HPD_Negative;
 
    TMatrixDSym *Covariance;
    TMatrixDSym *Correlation;
 
    std::vector<TH1D*> hpost;
    std::vector<std::vector<TH2D*>> hpost2D;
    std::unique_ptr<TH2D> hviolin;
    std::unique_ptr<TH2D> hviolin_prior;
 
    M3::float_t** ParStep;
    unsigned int* StepNumber;
 
    int nBins;
    double DrawRange;
    
    bool CacheMCMC;
    bool doDiagMCMC;
    
    //Number of batches and LagL used in MCMC diagnostic
    int nBatches;
    int AutoCorrLag;
    
    double **BatchedAverages;
 
    double **SampleValues;
    double **SystValues;
 
    double *AccProbValues;
    double *AccProbBatchedAverages;
    
    bool ReweightPosterior;
    std::string ReweightName;
    double* WeightValue;
 
  //Only if GPU is enabled
  #ifdef MaCh3_CUDA
    void PrepareGPU_AutoCorr(const int nLags, const std::vector<double>& ParamSums, float*& ParStep_cpu,
                             float*& NumeratorSum_cpu, float*& ParamSums_cpu, float*& DenomSum_cpu);
    std::unique_ptr<MCMCProcessorGPU> GPUProcessor;
  #endif
};