MarkovChain.cxx

Go to the documentation of this file.

// @(#)root/roostats:$Id$
// Authors: Kevin Belasco        17/06/2009
// Authors: Kyle Cranmer         17/06/2009
/*************************************************************************
 * Copyright (C) 1995-2008, Rene Brun and Fons Rademakers.               *
 * All rights reserved.                                                  *
 *                                                                       *
 * For the licensing terms see $ROOTSYS/LICENSE.                         *
 * For the list of contributors see $ROOTSYS/README/CREDITS.             *
 *************************************************************************/
 
/** \class RooStats::MarkovChain
    \ingroup Roostats
 
Stores the steps in a Markov Chain of points.  Allows user to access the
weight and NLL value (if applicable) with which a point was added to the
MarkovChain.
 
*/
 
#include "Rtypes.h"
 
#include "TNamed.h"
#include "RooStats/MarkovChain.h"
#include "RooDataSet.h"
#include "RooArgSet.h"
#include "RooRealVar.h"
#include "RooStats/RooStatsUtils.h"
#include "RooDataHist.h"
#include "THnSparse.h"
 
ClassImp(RooStats::MarkovChain);
 
using namespace RooFit;
using namespace RooStats;
 
static const char* WEIGHT_NAME = "weight_MarkovChain_local_";
static const char* NLL_NAME = "nll_MarkovChain_local_";
static const char* DATASET_NAME = "dataset_MarkovChain_local_";
static const char* DEFAULT_NAME = "_markov_chain";
static const char* DEFAULT_TITLE = "Markov Chain";
 
MarkovChain::MarkovChain() : TNamed(DEFAULT_NAME, DEFAULT_TITLE) {}
 
MarkovChain::MarkovChain(RooArgSet &parameters) : TNamed(DEFAULT_NAME, DEFAULT_TITLE)
{
   SetParameters(parameters);
}
 
MarkovChain::MarkovChain(const char *name, const char *title, RooArgSet &parameters) : TNamed(name, title)
{
   SetParameters(parameters);
}
 
void MarkovChain::SetParameters(RooArgSet& parameters)
{
   delete fChain;
   delete fParameters;
   delete fDataEntry;
 
   fParameters = new RooArgSet();
   fParameters->addClone(parameters);
 
   // kbelasco: consider setting fDataEntry = fChain->get()
   // to see if that makes it possible to get values of variables without
   // doing string comparison
   RooRealVar nll(NLL_NAME, "-log Likelihood", 0);
 
   fDataEntry = new RooArgSet();
   fDataEntry->addClone(parameters);
   fDataEntry->addClone(nll);
   fNLL = static_cast<RooRealVar*>(fDataEntry->find(NLL_NAME));
 
   fChain = new RooDataSet(DATASET_NAME, "Markov Chain", *fDataEntry, RooFit::WeightVar(WEIGHT_NAME));
}
 
void MarkovChain::Add(RooArgSet& entry, double nllValue, double weight)
{
   if (fParameters == nullptr)
      SetParameters(entry);
   RooStats::SetParameters(&entry, fDataEntry);
   fNLL->setVal(nllValue);
   //kbelasco: this is stupid, but some things might require it, so be doubly sure
   fChain->add(*fDataEntry, weight);
   //fChain->add(*fDataEntry);
}
 
void MarkovChain::AddWithBurnIn(MarkovChain& otherChain, Int_t burnIn)
{
   // Discards the first n accepted points.
 
   if(fParameters == nullptr) SetParameters(*const_cast<RooArgSet*>(otherChain.Get()));
   int counter = 0;
   for( int i=0; i < otherChain.Size(); i++ ) {
      RooArgSet* entry = const_cast<RooArgSet*>(otherChain.Get(i));
      counter += 1;
      if( counter > burnIn ) {
         AddFast( *entry, otherChain.NLL(), otherChain.Weight() );
      }
   }
}
void MarkovChain::Add(MarkovChain& otherChain, double discardEntries)
{
   // Discards the first entries. This is different to the definition of
   // burn-in used in the Bayesian calculator where the first n accepted
   // terms from the proposal function are discarded.
 
   if(fParameters == nullptr) SetParameters(*const_cast<RooArgSet*>(otherChain.Get()));
   double counter = 0.0;
   for( int i=0; i < otherChain.Size(); i++ ) {
      RooArgSet* entry = const_cast<RooArgSet*>(otherChain.Get(i));
      counter += otherChain.Weight();
      if( counter > discardEntries ) {
         AddFast( *entry, otherChain.NLL(), otherChain.Weight() );
      }
   }
}
 
void MarkovChain::AddFast(RooArgSet& entry, double nllValue, double weight)
{
   RooStats::SetParameters(&entry, fDataEntry);
   fNLL->setVal(nllValue);
   //kbelasco: this is stupid, but some things might require it, so be doubly sure
   fChain->addFast(*fDataEntry, weight);
   //fChain->addFast(*fDataEntry);
}
 
RooFit::OwningPtr<RooDataSet> MarkovChain::GetAsDataSet(RooArgSet* whichVars) const
{
   RooArgSet args;
   if (whichVars == nullptr) {
      //args.add(*fParameters);
      //args.add(*fNLL);
      args.add(*fDataEntry);
   } else {
      args.add(*whichVars);
   }
 
   return RooFit::makeOwningPtr<RooDataSet>(std::unique_ptr<RooAbsData>{fChain->reduce(args)});
}
 
RooFit::OwningPtr<RooDataSet> MarkovChain::GetAsDataSet(const RooCmdArg &arg1, const RooCmdArg &arg2,
                                                        const RooCmdArg &arg3, const RooCmdArg &arg4,
                                                        const RooCmdArg &arg5, const RooCmdArg &arg6,
                                                        const RooCmdArg &arg7, const RooCmdArg &arg8) const
{
   return RooFit::makeOwningPtr<RooDataSet>(
      std::unique_ptr<RooAbsData>{fChain->reduce(arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8)});
}
 
RooFit::OwningPtr<RooDataHist> MarkovChain::GetAsDataHist(RooArgSet* whichVars) const
{
   RooArgSet args;
   if (whichVars == nullptr) {
      args.add(*fParameters);
      //args.add(*fNLL);
      //args.add(*fDataEntry);
   } else {
      args.add(*whichVars);
   }
 
   std::unique_ptr<RooAbsData> data{fChain->reduce(args)};
   return RooFit::makeOwningPtr(std::unique_ptr<RooDataHist>{static_cast<RooDataSet&>(*data).binnedClone()});
}
 
RooFit::OwningPtr<RooDataHist> MarkovChain::GetAsDataHist(const RooCmdArg &arg1, const RooCmdArg &arg2,
                                                          const RooCmdArg &arg3, const RooCmdArg &arg4,
                                                          const RooCmdArg &arg5, const RooCmdArg &arg6,
                                                          const RooCmdArg &arg7, const RooCmdArg &arg8) const
{
   std::unique_ptr<RooAbsData> data{fChain->reduce(arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8)};
   return RooFit::makeOwningPtr(std::unique_ptr<RooDataHist>{static_cast<RooDataSet &>(*data).binnedClone()});
}
 
THnSparse* MarkovChain::GetAsSparseHist(RooAbsCollection* whichVars) const
{
   RooArgList axes;
   if (whichVars == nullptr) {
      axes.add(*fParameters);
   } else {
      axes.add(*whichVars);
   }
 
   Int_t dim = axes.size();
   std::vector<double> min(dim);
   std::vector<double> max(dim);
   std::vector<Int_t> bins(dim);
   std::vector<const char *> names(dim);
   Int_t i = 0;
   for (auto const *var : static_range_cast<RooRealVar *>(axes)) {
      names[i] = var->GetName();
      min[i] = var->getMin();
      max[i] = var->getMax();
      bins[i] = var->numBins();
      ++i;
   }
 
   THnSparseF* sparseHist = new THnSparseF("posterior", "MCMC Posterior Histogram",
         dim, &bins[0], &min[0], &max[0]);
 
   // kbelasco: it appears we need to call Sumw2() just to get the
   // histogram to keep a running total of the weight so that Getsumw doesn't
   // just return 0
   sparseHist->Sumw2();
 
   // Fill histogram
   Int_t size = fChain->numEntries();
   const RooArgSet* entry;
   std::vector<double> x(dim);
   for ( i = 0; i < size; i++) {
      entry = fChain->get(i);
 
      for (Int_t ii = 0; ii < dim; ii++) {
         //LM:  doing this is probably quite slow
         x[ii] = entry->getRealValue( names[ii]);
         sparseHist->Fill(x.data(), fChain->weight());
      }
   }
 
   return sparseHist;
}
 
double MarkovChain::NLL(Int_t i) const
{
   // kbelasco: how to do this?
   //fChain->get(i);
   //return fNLL->getVal();
   return fChain->get(i)->getRealValue(NLL_NAME);
}
 
double MarkovChain::NLL() const
{
   // kbelasco: how to do this?
   //fChain->get();
   //return fNLL->getVal();
   return fChain->get()->getRealValue(NLL_NAME);
}
 
double MarkovChain::Weight() const
{
   return fChain->weight();
}
 
double MarkovChain::Weight(Int_t i) const
{
   fChain->get(i);
   return fChain->weight();
}