RooNumConvolution.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 RooNumConvolution.cxx
\class RooNumConvolution
\ingroup Roofitcore
 
Numeric 1-dimensional convolution operator PDF. This class can convolve any PDF
with any other PDF
This class should not be used blindly as numeric convolution is computing
intensive and prone to stability fitting problems. If an analytic convolution
can be calculated, you should use that or implement it if not available.
RooNumConvolution implements reasonable defaults that should convolve most
functions reasonably well, but results strongly depend on the shape of your
input PDFS so always check your result.
 
The default integration engine for the numeric convolution is the
adaptive Gauss-Kronrod method, which empirically seems the most robust
for this task. You can override the convolution integration settings via
the RooNumIntConfig object reference returned by the convIntConfig() member
function
By default the numeric convolution is integrated from -infinity to
+infinity through a <pre>x -> 1/x</pre> coordinate transformation of the
tails. For convolution with a very small bandwidth it may be
advantageous (for both CPU consumption and stability) if the
integration domain is limited to a finite range. The function
setConvolutionWindow(mean,width,scale) allows to set a sliding
window around the x value to be calculated taking a RooAbsReal
expression for an offset and a width to be taken around the x
value. These input expression can be RooFormulaVars or other
function objects although the 3d 'scale' argument 'scale'
multiplies the width RooAbsReal expression given in the 2nd
argument, allowing for an appropriate window definition for most
cases without need for a RooFormulaVar object: e.g. a Gaussian
resolution PDF do setConvolutionWindow(gaussMean,gaussSigma,5)
Note that for a 'wide' Gaussian the -inf to +inf integration
may converge more quickly than that over a finite range!
The default numeric precision is 1e-7, i.e. the global default for
numeric integration but you should experiment with this value to
see if it is sufficient for example by studying the number of function
calls that MINUIT needs to fit your function as function of the
convolution precision.
**/
 
#include "Riostream.h"
#include "TH2F.h"
#include "RooNumConvolution.h"
#include "RooArgList.h"
#include "RooRealVar.h"
#include "RooCustomizer.h"
#include "RooConvIntegrandBinding.h"
#include "RooNumIntFactory.h"
#include "RooGenContext.h"
#include "RooConvGenContext.h"
#include "RooMsgService.h"
 
 
using std::endl, std::ostream;
 
 
 
 
////////////////////////////////////////////////////////////////////////////////
 
RooNumConvolution::RooNumConvolution() :
  _init(false),
  _integrand(nullptr),
  _cloneVar(nullptr),
  _clonePdf(nullptr),
  _cloneModel(nullptr),
  _useWindow(false),
  _windowScale(1),
  _verboseThresh(2000),
  _doProf(false),
  _callHist(nullptr)
{
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Constructor of convolution operator PDF
///
/// convVar  :  convolution variable (on which both pdf and resmodel should depend)
/// pdf      :  input 'physics' pdf
/// resmodel :  input 'resolution' pdf
///
/// output is pdf(x) (X) resmodel(x) = Int [ pdf(x') resmodel (x-x') ] dx'
///
 
RooNumConvolution::RooNumConvolution(const char *name, const char *title, RooRealVar& convVar, RooAbsReal& inPdf, RooAbsReal& resmodel, const RooNumConvolution* proto) :
  RooAbsReal(name,title),
  _init(false),
  _convIntConfig(RooNumIntConfig::defaultConfig()),
  _integrand(nullptr),
  _origVar("origVar","Original Convolution variable",this,convVar),
  _origPdf("origPdf","Original Input PDF",this,inPdf),
  _origModel("origModel","Original Resolution model",this,resmodel),
  _ownedClonedPdfSet("ownedClonePdfSet"),
  _ownedClonedModelSet("ownedCloneModelSet"),
  _cloneVar(nullptr),
  _clonePdf(nullptr),
  _cloneModel(nullptr),
  _useWindow(false),
  _windowScale(1),
  _windowParam("windowParam","Convolution window parameter",this,false),
  _verboseThresh(2000),
  _doProf(false),
  _callHist(nullptr)
{
  // Use Adaptive Gauss-Kronrod integration by default for the convolution integral
  _convIntConfig.method1D().setLabel("RooAdaptiveGaussKronrodIntegrator1D") ;
  _convIntConfig.method1DOpen().setLabel("RooAdaptiveGaussKronrodIntegrator1D") ;
 
  if (proto) {
    convIntConfig() = proto->convIntConfig() ;
    if (proto->_useWindow) {
      setConvolutionWindow(static_cast<RooAbsReal&>(*proto->_windowParam.at(0)),static_cast<RooAbsReal&>(*proto->_windowParam.at(1)),proto->_windowScale) ;
    }
  }
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Copy constructor
 
RooNumConvolution::RooNumConvolution(const RooNumConvolution& other, const char* name) :
  RooAbsReal(other,name),
  _init(false),
  _convIntConfig(other._convIntConfig),
  _integrand(nullptr),
  _origVar("origVar",this,other._origVar),
  _origPdf("origPdf",this,other._origPdf),
  _origModel("origModel",this,other._origModel),
  _ownedClonedPdfSet("ownedClonePdfSet"),
  _ownedClonedModelSet("ownedCloneModelSet"),
  _cloneVar(nullptr),
  _clonePdf(nullptr),
  _cloneModel(nullptr),
  _useWindow(other._useWindow),
  _windowScale(other._windowScale),
  _windowParam("windowParam",this,other._windowParam),
  _verboseThresh(other._verboseThresh),
  _doProf(other._doProf),
  _callHist(other._callHist)
{
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// One-time initialization of object
 
void RooNumConvolution::initialize() const
{
  // Initialization function -- create clone of convVar (x') and deep-copy clones of pdf and
  // model that are connected to x' rather than x (convVar)
 
  // Start out clean
  _ownedClonedPdfSet.removeAll() ;
  _ownedClonedModelSet.removeAll() ;
 
  if (_cloneVar) delete _cloneVar ;
 
  // Customize a copy of origPdf that is connected to x' rather than x
  // store all cloned components in _clonePdfSet as well as x' itself
  _cloneVar = new RooRealVar(Form("%s_prime",_origVar.arg().GetName()),"Convolution Variable",0) ;
 
  RooCustomizer mgr1(pdf(),"NumConv_PdfClone") ;
  mgr1.setCloneBranchSet(_ownedClonedPdfSet) ;
  mgr1.replaceArg(var(),*_cloneVar) ;
  _clonePdf = static_cast<RooAbsReal*>(mgr1.build()) ;
 
  RooCustomizer mgr2(model(),"NumConv_ModelClone") ;
  mgr2.setCloneBranchSet(_ownedClonedModelSet) ;
  mgr2.replaceArg(var(),*_cloneVar) ;
  _cloneModel = static_cast<RooAbsReal*>(mgr2.build()) ;
 
  // Change name back to original name
  _cloneVar->SetName(var().GetName()) ;
 
  // Create Convolution integrand
  _integrand = new RooConvIntegrandBinding(*_clonePdf,*_cloneModel,*_cloneVar,var(),nullptr) ;
 
  // Instantiate integrator for convolution integrand
  _integrator = RooNumIntFactory::instance().createIntegrator(*_integrand,_convIntConfig,1);
  _integrator->setUseIntegrandLimits(false) ;
 
  _init = true ;
}
 
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Destructor
 
RooNumConvolution::~RooNumConvolution()
{
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Calculate convolution integral
 
double RooNumConvolution::evaluate() const
{
  // Check if deferred initialization has occurred
  if (!_init) initialize() ;
 
  // Retrieve current value of convolution variable
  double x = _origVar ;
 
  // Propagate current normalization set to integrand
  _integrand->setNormalizationSet(_origVar.nset()) ;
 
  // Adjust convolution integration window
  if (_useWindow) {
    double center = (static_cast<RooAbsReal*>(_windowParam.at(0)))->getVal() ;
    double width = _windowScale * (static_cast<RooAbsReal*>(_windowParam.at(1)))->getVal() ;
    _integrator->setLimits(x-center-width,x-center+width) ;
  } else {
    _integrator->setLimits(-RooNumber::infinity(),RooNumber::infinity()) ;
  }
 
  // Calculate convolution for present x
  if (_doProf) _integrand->resetNumCall() ;
  double ret = _integrator->integral(&x) ;
  if (_doProf) {
    _callHist->Fill(x,_integrand->numCall()) ;
    if (_integrand->numCall()>_verboseThresh) {
      coutW(Integration) << "RooNumConvolution::evaluate(" << GetName() << ") WARNING convolution integral at x=" << x
          << " required " << _integrand->numCall() << " function evaluations" << std::endl ;
    }
  }
 
  return ret ;
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Intercept server redirects. Throw away cache, as figuring out redirections on the cache is an unsolvable problem.
 
bool RooNumConvolution::redirectServersHook(const RooAbsCollection& newServerList, bool mustReplaceAll,
                     bool nameChange, bool isRecursive)
{
  _init = false ;
  return RooAbsReal::redirectServersHook(newServerList, mustReplaceAll, nameChange, isRecursive);
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Removes previously defined convolution window, reverting to convolution from -inf to +inf
 
void RooNumConvolution::clearConvolutionWindow()
{
  _useWindow = false ;
  _windowParam.removeAll() ;
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Restrict convolution integral to finite range [ x - C - S*W, x - C + S*W ]
/// where x is current value of convolution variablem, C = centerParam, W=widthParam and S = widthScaleFactor
/// Inputs centerParam and withParam can be function expressions (RooAbsReal, RooFormulaVar) etc.
 
void RooNumConvolution::setConvolutionWindow(RooAbsReal& centerParam, RooAbsReal& widthParam, double widthScaleFactor)
{
  _useWindow = true ;
  _windowParam.removeAll() ;
  _windowParam.add(centerParam) ;
  _windowParam.add(widthParam) ;
  _windowScale = widthScaleFactor ;
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Activate warning messages if number of function calls needed for evaluation of convolution integral
/// exceeds given threshold
 
void RooNumConvolution::setCallWarning(Int_t threshold)
{
  if (threshold<0) {
    coutE(InputArguments) << "RooNumConvolution::setCallWarning(" << GetName() << ") ERROR: threshold must be positive, value unchanged" << std::endl ;
    return ;
  }
  _verboseThresh = threshold ;
}
 
 
////////////////////////////////////////////////////////////////////////////////
/// Activate call profile if flag is set to true. A 2-D histogram is kept that stores the required number
/// of function calls versus the value of x, the convolution variable
///
/// All clones of RooNumConvolution objects will keep logging to the histogram of the original class
/// so that performance of temporary object clones, such as used in e.g. fitting, plotting and generating
/// are all logged in a single place.
///
/// Function caller should take ownership of profiling histogram as it is not deleted at the RooNumConvolution dtor
///
/// Calling this function with flag set to false will deactivate call profiling and delete the profiling histogram
 
void RooNumConvolution::setCallProfiling(bool flag, Int_t nbinX, Int_t nbinCall, Int_t nCallHigh)
{
  if (flag) {
    if (_doProf) {
      delete _callHist ;
    }
    _callHist = new TH2F(Form("callHist_%s",GetName()),Form("Call Profiling of RooNumConvolution %s",GetTitle()),
          nbinX,_origVar.min(),_origVar.max(),
          nbinCall,0,nCallHigh) ;
    _doProf=true ;
 
  } else if (_doProf) {
 
    delete _callHist ;
    _callHist = nullptr ;
    _doProf = false ;
  }
 
}
 
 
 
////////////////////////////////////////////////////////////////////////////////
/// Hook function to intercept printCompactTree() calls so that it can print out
/// the content of its private cache in the print sequence
 
void RooNumConvolution::printCompactTreeHook(ostream& os, const char* indent)
{
  os << indent << "RooNumConvolution begin cache" << std::endl ;
 
  if (_init) {
    _cloneVar->printCompactTree(os,Form("%s[Var]",indent)) ;
    _clonePdf->printCompactTree(os,Form("%s[Pdf]",indent)) ;
    _cloneModel->printCompactTree(os,Form("%s[Mod]",indent)) ;
  }
 
  os << indent << "RooNumConvolution end cache" << std::endl ;
}