RooRandomizeParamMCSModule.cxx

Go to the documentation of this file.

/*****************************************************************************
 * Project: RooFit                                                           *
 * Package: RooFitCore                                                       *
 * @(#)root/roofitcore:$Id$
 * Authors:                                                                  *
 *   WV, Wouter Verkerke, UC Santa Barbara, verkerke@slac.stanford.edu       *
 *   DK, David Kirkby,    UC Irvine,         dkirkby@uci.edu                 *
 *                                                                           *
 * Copyright (c) 2000-2005, Regents of the University of California          *
 *                          and Stanford University. All rights reserved.    *
 *                                                                           *
 * Redistribution and use in source and binary forms,                        *
 * with or without modification, are permitted according to the terms        *
 * listed in LICENSE (http://roofit.sourceforge.net/license.txt)             *
 *****************************************************************************/
 
/**
\file RooRandomizeParamMCSModule.cxx
\class RooRandomizeParamMCSModule
\ingroup Roofitcore
 
Add-on module to RooMCStudy that
allows you to randomize input generation parameters. Randomized generation
parameters can be sampled from a uniform or Gaussian distribution.
For every randomized parameter, an extra variable is added to
RooMCStudy::fitParDataSet() named <tt>`<parname>`_gen</tt> that indicates the actual
value used for generation for each trial.
You can also choose to randomize the sum of N parameters, rather
than a single parameter. In that case common multiplicative scale
factor is applied to each component to bring the sum to the desired
target value taken from either uniform or Gaussian sampling. This
latter option is for example useful if you want to change the total
number of expected events of an extended p.d.f
**/
 
 
#include "Riostream.h"
#include "RooDataSet.h"
#include "RooRealVar.h"
#include "RooRandom.h"
#include "TString.h"
#include "RooFitResult.h"
#include "RooAddition.h"
#include "RooMsgService.h"
#include "RooRandomizeParamMCSModule.h"
 
using std::endl;
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Constructor
 
RooRandomizeParamMCSModule::RooRandomizeParamMCSModule() :
  RooAbsMCStudyModule("RooRandomizeParamMCSModule","RooRandomizeParamMCSModule")
{
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Copy constructor
 
RooRandomizeParamMCSModule::RooRandomizeParamMCSModule(const RooRandomizeParamMCSModule& other) :
  RooAbsMCStudyModule(other),
  _unifParams(other._unifParams),
  _gausParams(other._gausParams)
{
}
 
RooRandomizeParamMCSModule:: ~RooRandomizeParamMCSModule() = default;
 
////////////////////////////////////////////////////////////////////////////////
/// Request uniform smearing of param in range [lo,hi] in RooMCStudy
/// generation cycle
 
void RooRandomizeParamMCSModule::sampleUniform(RooRealVar& param, double lo, double hi)
{
  // If we're already attached to a RooMCStudy, check that given param is actual generator model parameter
  // If not attached, this check is repeated at the attachment moment
  if (genParams()) {
    RooRealVar* actualPar = static_cast<RooRealVar*>(genParams()->find(param.GetName())) ;
    if (!actualPar) {
      oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::initializeInstance: variable " << param.GetName() << " is not a parameter of RooMCStudy model and is ignored!" << std::endl ;
      return ;
    }
  }
 
  _unifParams.push_back(UniParam(&param,lo,hi)) ;
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Request Gaussian smearing of param in with mean 'mean' and width
/// 'sigma' in RooMCStudy generation cycle
 
void RooRandomizeParamMCSModule::sampleGaussian(RooRealVar& param, double mean, double sigma)
{
  // If we're already attached to a RooMCStudy, check that given param is actual generator model parameter
  // If not attached, this check is repeated at the attachment moment
  if (genParams()) {
    RooRealVar* actualPar = static_cast<RooRealVar*>(genParams()->find(param.GetName())) ;
    if (!actualPar) {
      oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::initializeInstance: variable " << param.GetName() << " is not a parameter of RooMCStudy model and is ignored!" << std::endl ;
      return ;
    }
  }
 
  _gausParams.push_back(GausParam(&param,mean,sigma)) ;
}
 
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Request uniform smearing of sum of parameters in paramSet uniform
/// smearing in range [lo,hi] in RooMCStudy generation cycle.  This
/// option applies a common multiplicative factor to each parameter
/// in paramSet to make the sum of the parameters add up to the
/// sampled value in the range [lo,hi]
 
void RooRandomizeParamMCSModule::sampleSumUniform(const RooArgSet& paramSet, double lo, double hi)
{
  // Check that all args are RooRealVars
  RooArgSet okset ;
  for(RooAbsArg * arg : paramSet) {
    // Check that arg is a RooRealVar
    RooRealVar* rrv = dynamic_cast<RooRealVar*>(arg) ;
    if (!rrv) {
      oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::sampleSumUniform() ERROR: input parameter " << arg->GetName() << " is not a RooRealVar and is ignored" << std::endl ;
      continue;
    }
    okset.add(*rrv) ;
  }
 
  // If we're already attached to a RooMCStudy, check that given param is actual generator model parameter
  // If not attached, this check is repeated at the attachment moment
  RooArgSet okset2 ;
  if (genParams()) {
    for(RooAbsArg * arg2 : okset) {
      RooRealVar* actualVar= static_cast<RooRealVar*>(genParams()->find(arg2->GetName())) ;
      if (!actualVar) {
   oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::sampleSumUniform: variable " << arg2->GetName() << " is not a parameter of RooMCStudy model and is ignored!" << std::endl ;
      } else {
   okset2.add(*actualVar) ;
      }
    }
  } else {
 
   // If genParams() are not available, skip this check for now
   okset2.add(okset) ;
 
  }
 
 
  _unifParamSets.push_back(UniParamSet(okset2,lo,hi)) ;
 
}
 
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Request gaussian smearing of sum of parameters in paramSet
/// uniform smearing with mean 'mean' and width 'sigma' in RooMCStudy
/// generation cycle.  This option applies a common multiplicative
/// factor to each parameter in paramSet to make the sum of the
/// parameters add up to the sampled value from the
/// gaussian(mean,sigma)
 
void RooRandomizeParamMCSModule::sampleSumGauss(const RooArgSet& paramSet, double mean, double sigma)
{
  // Check that all args are RooRealVars
  RooArgSet okset ;
  for(RooAbsArg * arg : paramSet) {
    // Check that arg is a RooRealVar
    RooRealVar* rrv = dynamic_cast<RooRealVar*>(arg) ;
    if (!rrv) {
      oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::sampleSumGauss() ERROR: input parameter " << arg->GetName() << " is not a RooRealVar and is ignored" << std::endl ;
      continue;
    }
    okset.add(*rrv) ;
  }
 
  // If we're already attached to a RooMCStudy, check that given param is actual generator model parameter
  // If not attached, this check is repeated at the attachment moment
  RooArgSet okset2 ;
  if (genParams()) {
    for(RooAbsArg * arg2 : okset) {
      RooRealVar* actualVar= static_cast<RooRealVar*>(genParams()->find(arg2->GetName())) ;
      if (!actualVar) {
   oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::sampleSumUniform: variable " << arg2->GetName() << " is not a parameter of RooMCStudy model and is ignored!" << std::endl ;
      } else {
   okset2.add(*actualVar) ;
      }
    }
  } else {
 
   // If genParams() are not available, skip this check for now
   okset2.add(okset) ;
 
  }
 
  _gausParamSets.push_back(GausParamSet(okset,mean,sigma)) ;
 
}
 
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Initialize module after attachment to RooMCStudy object
 
bool RooRandomizeParamMCSModule::initializeInstance()
{
  // Loop over all uniform smearing parameters
  std::list<UniParam>::iterator uiter ;
  for (uiter= _unifParams.begin() ; uiter!= _unifParams.end() ; ++uiter) {
 
    // Check that listed variable is actual generator model parameter
    RooRealVar* actualPar = static_cast<RooRealVar*>(genParams()->find(uiter->_param->GetName())) ;
    if (!actualPar) {
      oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::initializeInstance: variable " << uiter->_param->GetName() << " is not a parameter of RooMCStudy model and is ignored!" << std::endl ;
      uiter = _unifParams.erase(uiter) ;
      continue ;
    }
    uiter->_param = actualPar ;
 
    // Add variable to summary dataset to hold generator value
    std::string parName = std::string(uiter->_param->GetName()) + "_gen";
    std::string parTitle = std::string(uiter->_param->GetTitle()) + " as generated";
    _genParSet.addOwned(std::make_unique<RooRealVar>(parName.c_str(),parTitle.c_str(),0));
  }
 
  // Loop over all gaussian smearing parameters
  std::list<GausParam>::iterator giter ;
  for (giter= _gausParams.begin() ; giter!= _gausParams.end() ; ++giter) {
 
    // Check that listed variable is actual generator model parameter
    RooRealVar* actualPar = static_cast<RooRealVar*>(genParams()->find(giter->_param->GetName())) ;
    if (!actualPar) {
      oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::initializeInstance: variable " << giter->_param->GetName() << " is not a parameter of RooMCStudy model and is ignored!" << std::endl ;
      giter = _gausParams.erase(giter) ;
      continue ;
    }
    giter->_param = actualPar ;
 
    // Add variable to summary dataset to hold generator value
    std::string parName = std::string(giter->_param->GetName()) + "_gen";
    std::string parTitle = std::string(giter->_param->GetTitle()) + " as generated";
    _genParSet.addOwned(std::make_unique<RooRealVar>(parName.c_str(),parTitle.c_str(),0));
  }
 
 
  // Loop over all uniform smearing set of parameters
  std::list<UniParamSet>::iterator usiter ;
  for (usiter= _unifParamSets.begin() ; usiter!= _unifParamSets.end() ; ++usiter) {
 
    // Check that all listed variables are actual generator model parameters
    RooArgSet actualPSet ;
    for(RooAbsArg * arg : usiter->_pset) {
      RooRealVar* actualVar= static_cast<RooRealVar*>(genParams()->find(arg->GetName())) ;
      if (!actualVar) {
   oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::initializeInstance: variable " << arg->GetName() << " is not a parameter of RooMCStudy model and is ignored!" << std::endl ;
      } else {
   actualPSet.add(*actualVar) ;
      }
    }
    usiter->_pset.removeAll() ;
    usiter->_pset.add(actualPSet) ;
 
    // Add variables to summary dataset to hold generator values
    for(auto * param : static_range_cast<RooRealVar*>(usiter->_pset)) {
      std::string parName = std::string(param->GetName()) + "_gen";
      std::string parTitle = std::string(param->GetTitle()) + " as generated";
      _genParSet.addOwned(std::make_unique<RooRealVar>(parName.c_str(),parTitle.c_str(),0));
    }
  }
 
  // Loop over all gaussian smearing set of parameters
  std::list<GausParamSet>::iterator ugiter ;
  for (ugiter= _gausParamSets.begin() ; ugiter!= _gausParamSets.end() ; ++ugiter) {
 
    // Check that all listed variables are actual generator model parameters
    RooArgSet actualPSet ;
    for(RooAbsArg * arg : ugiter->_pset) {
      RooRealVar* actualVar= static_cast<RooRealVar*>(genParams()->find(arg->GetName())) ;
      if (!actualVar) {
   oocoutW(nullptr,InputArguments) << "RooRandomizeParamMCSModule::initializeInstance: variable " << arg->GetName() << " is not a parameter of RooMCStudy model and is ignored!" << std::endl ;
      } else {
   actualPSet.add(*actualVar) ;
      }
    }
 
    ugiter->_pset.removeAll() ;
    ugiter->_pset.add(actualPSet) ;
 
    // Add variables to summary dataset to hold generator values
    for(auto * param : static_range_cast<RooRealVar*>(ugiter->_pset)) {
      std::string parName = std::string(param->GetName()) + "_gen";
      std::string parTitle = std::string(param->GetTitle()) + " as generated";
      _genParSet.addOwned(std::make_unique<RooRealVar>(parName.c_str(),parTitle.c_str(),0));
    }
  }
 
  // Create new dataset to be merged with RooMCStudy::fitParDataSet
  _data = std::make_unique<RooDataSet>("DeltaLLSigData","Additional data for Delta(-log(L)) study",_genParSet) ;
 
  return true ;
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Initialize module at beginning of RooCMStudy run
 
bool RooRandomizeParamMCSModule::initializeRun(Int_t /*numSamples*/)
{
  // Clear dataset at beginning of run
  _data->reset() ;
  return true ;
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Apply all smearings to generator parameters
 
bool RooRandomizeParamMCSModule::processBeforeGen(Int_t /*sampleNum*/)
{
  // Apply uniform smearing to all generator parameters for which it is requested
  std::list<UniParam>::iterator uiter ;
  for (uiter= _unifParams.begin() ; uiter!= _unifParams.end() ; ++uiter) {
    double newVal = RooRandom::randomGenerator()->Uniform(uiter->_lo,uiter->_hi) ;
    oocoutE(nullptr,Generation) << "RooRandomizeParamMCSModule::processBeforeGen: applying uniform smearing to generator parameter "
    << uiter->_param->GetName() << " in range [" << uiter->_lo << "," << uiter->_hi << "], chosen value for this sample is " << newVal << std::endl ;
    uiter->_param->setVal(newVal) ;
 
    RooRealVar* genpar = static_cast<RooRealVar*>(_genParSet.find(Form("%s_gen",uiter->_param->GetName()))) ;
    genpar->setVal(newVal) ;
  }
 
  // Apply gaussian smearing to all generator parameters for which it is requested
  std::list<GausParam>::iterator giter ;
  for (giter= _gausParams.begin() ; giter!= _gausParams.end() ; ++giter) {
    double newVal = RooRandom::randomGenerator()->Gaus(giter->_mean,giter->_sigma) ;
    oocoutI(nullptr,Generation) << "RooRandomizeParamMCSModule::processBeforeGen: applying gaussian smearing to generator parameter "
    << giter->_param->GetName() << " with a mean of " << giter->_mean << " and a width of " << giter->_sigma << ", chosen value for this sample is " << newVal << std::endl ;
    giter->_param->setVal(newVal) ;
 
    RooRealVar* genpar = static_cast<RooRealVar*>(_genParSet.find(Form("%s_gen",giter->_param->GetName()))) ;
    genpar->setVal(newVal) ;
  }
 
  // Apply uniform smearing to all sets of generator parameters for which it is requested
  std::list<UniParamSet>::iterator usiter ;
  for (usiter= _unifParamSets.begin() ; usiter!= _unifParamSets.end() ; ++usiter) {
 
    // Calculate new value for sum
    double newVal = RooRandom::randomGenerator()->Uniform(usiter->_lo,usiter->_hi) ;
    oocoutI(nullptr,Generation) << "RooRandomizeParamMCSModule::processBeforeGen: applying uniform smearing to sum of set of generator parameters "
                <<  usiter->_pset
                << " in range [" << usiter->_lo << "," << usiter->_hi << "], chosen sum value for this sample is " << newVal << std::endl ;
 
    // Determine original value of sum and calculate per-component scale factor to obtain new value for sum
    RooAddition sumVal("sumVal","sumVal",usiter->_pset) ;
    double compScaleFactor = newVal/sumVal.getVal() ;
 
    // Apply multiplicative correction to each term of the sum
    for(auto * param : static_range_cast<RooRealVar*>(usiter->_pset)) {
      param->setVal(param->getVal()*compScaleFactor) ;
      RooRealVar* genpar = static_cast<RooRealVar*>(_genParSet.find(Form("%s_gen",param->GetName()))) ;
      genpar->setVal(param->getVal()) ;
    }
  }
 
  // Apply gaussian smearing to all sets of generator parameters for which it is requested
  std::list<GausParamSet>::iterator gsiter ;
  for (gsiter= _gausParamSets.begin() ; gsiter!= _gausParamSets.end() ; ++gsiter) {
 
    // Calculate new value for sum
    double newVal = RooRandom::randomGenerator()->Gaus(gsiter->_mean,gsiter->_sigma) ;
    oocoutI(nullptr,Generation) << "RooRandomizeParamMCSModule::processBeforeGen: applying gaussian smearing to sum of set of generator parameters "
                << gsiter->_pset
                << " with a mean of " << gsiter->_mean << " and a width of " << gsiter->_sigma
                << ", chosen value for this sample is " << newVal << std::endl ;
 
    // Determine original value of sum and calculate per-component scale factor to obtain new value for sum
    RooAddition sumVal("sumVal","sumVal",gsiter->_pset) ;
    double compScaleFactor = newVal/sumVal.getVal() ;
 
    // Apply multiplicative correction to each term of the sum
    for(auto * param : static_range_cast<RooRealVar*>(gsiter->_pset)) {
      param->setVal(param->getVal()*compScaleFactor) ;
      RooRealVar* genpar = static_cast<RooRealVar*>(_genParSet.find(Form("%s_gen",param->GetName()))) ;
      genpar->setVal(param->getVal()) ;
    }
  }
 
  // Store generator values later in processAfterFit if fit succeeded.
 
  return true ;
}
 
 
bool RooRandomizeParamMCSModule::processAfterFit(bool fitOk)
{
   if (!fitOk)
      return true;
   // Store generator values for all modified parameters
   _data->add(_genParSet);
   return true;
}
 
 
////////////////////////////////////////////////////////////////////////////////
/// Return auxiliary data of this module so that it is merged with
/// RooMCStudy::fitParDataSet()
 
RooDataSet* RooRandomizeParamMCSModule::finalizeRun()
{
  return _data.get();
}