Logo ROOT   6.18/05
Reference Guide
BernsteinCorrection.cxx
Go to the documentation of this file.
1// @(#)root/roostats:$Id$
2// Author: Kyle Cranmer 28/07/2008
3
4/*************************************************************************
5 * Copyright (C) 1995-2008, Rene Brun and Fons Rademakers. *
6 * All rights reserved. *
7 * *
8 * For the licensing terms see $ROOTSYS/LICENSE. *
9 * For the list of contributors see $ROOTSYS/README/CREDITS. *
10 *************************************************************************/
11
12/** \class RooStats::BernsteinCorrection
13 \ingroup Roostats
14
15BernsteinCorrection is a utility in RooStats to augment a nominal PDF with a polynomial
16correction term. This is useful for incorporating systematic variations to the nominal PDF.
17The Bernstein basis polynomials are particularly appropriate because they are positive definite.
18
19This tool was inspired by the work of Glen Cowan together with Stephan Horner, Sascha Caron,
20Eilam Gross, and others.
21The initial implementation is independent work. The major step forward in the approach was
22to provide a well defined algorithm that specifies the order of polynomial to be included
23in the correction. This is an empirical algorithm, so in addition to the nominal model it
24needs either a real data set or a simulated one. In the early work, the nominal model was taken
25to be a histogram from Monte Carlo simulations, but in this implementation it is generalized to an
26arbitrary PDF (which includes a RooHistPdf). The algorithm basically consists of a
27hypothesis test of an nth-order correction (null) against a n+1-th order correction (alternate).
28The quantity q = -2 log LR is used to determine whether the n+1-th order correction is a major
29improvement to the n-th order correction. The distribution of q is expected to be roughly
30\f$\chi^2\f$ with one degree of freedom if the n-th order correction is a good model for the data.
31 Thus, one only moves to the n+1-th order correction of q is relatively large. The chance that
32one moves from the n-th to the n+1-th order correction when the n-th order correction
33(eg. a type 1 error) is sufficient is given by the Prob(\f$\chi^2_1\f$ > threshold). The constructor
34of this class allows you to directly set this tolerance (in terms of probability that the n+1-th
35 term is added unnecessarily).
36
37To do:
38Add another method to the utility that will make the sampling distribution for -2 log lambda
39for various m vs. m+1 order corrections using a nominal model and perhaps having two ways of
40generating the toys (either via a histogram or via an independent model that is supposed to
41 reflect reality). That will allow one to make plots like Glen has at the end of his DRAFT
42 very easily.
43
44*/
45
46
48
49#include "RooGlobalFunc.h"
50#include "RooDataSet.h"
51#include "RooRealVar.h"
52#include "RooAbsPdf.h"
53#include "RooFit.h"
54#include "RooFitResult.h"
55#include "TMath.h"
56#include <string>
57#include <vector>
58#include <stdio.h>
59#include <sstream>
60#include <iostream>
61
62#include "RooEffProd.h"
63#include "RooNLLVar.h"
64#include "RooWorkspace.h"
65#include "RooDataHist.h"
66#include "RooHistPdf.h"
67
68#include "RooBernstein.h"
69
71
72
74
75using namespace RooFit;
76using namespace RooStats;
77using namespace std;
78
79////////////////////////////////////////////////////////////////////////////////
80
81BernsteinCorrection::BernsteinCorrection(Double_t tolerance):
82 fMaxDegree(10), fMaxCorrection(100), fTolerance(tolerance){
83}
84
85
86////////////////////////////////////////////////////////////////////////////////
87/// Main method for Bernstein correction.
88
90 const char* nominalName,
91 const char* varName,
92 const char* dataName){
93 // get ingredients out of workspace
94 RooRealVar* x = wks->var(varName);
95 RooAbsPdf* nominal = wks->pdf(nominalName);
96 RooAbsData* data = wks->data(dataName);
97
98 if (!x || !nominal || !data) {
99 cout << "Error: wrong name for pdf or variable or dataset - return -1 " << std::endl;
100 return -1;
101 }
102
103 std::cout << "BernsteinCorrection::ImportCorrectedPdf - Doing initial Fit with nominal model " << std::endl;
104
105 // initialize alg, by checking how well nominal model fits
108
109 RooFitResult* nominalResult = nominal->fitTo(*data,Save(),Minos(kFALSE), Hesse(kFALSE),PrintLevel(printLevel),Minimizer(minimType));
110 Double_t lastNll= nominalResult->minNll();
111
112 if (nominalResult->status() != 0 ) {
113 std::cout << "BernsteinCorrection::ImportCorrectedPdf - Error fit with nominal model failed - exit" << std::endl;
114 return -1;
115 }
116
117 // setup a log
118 std::stringstream log;
119 log << "------ Begin Bernstein Correction Log --------" << endl;
120
121 // Local variables that we want to keep in scope after loop
122 RooArgList coeff;
123 vector<RooRealVar*> coefficients;
124 Double_t q = 1E6;
125 Int_t degree = -1;
126
127 // The while loop
128 bool keepGoing = true;
129 while( keepGoing ) {
130 degree++;
131
132 // we need to generate names for vars on the fly
133 std::stringstream str;
134 str<<"_"<<degree;
135
136 RooRealVar* newCoef = new RooRealVar(("c"+str.str()).c_str(),
137 "Bernstein basis poly coefficient",
138 1.0, 0., fMaxCorrection);
139 coeff.add(*newCoef);
140 coefficients.push_back(newCoef);
141 // Since pdf is normalized - coefficient for degree 0 is fixed to be 1
142 if (degree == 0) {
143 newCoef->setVal(1);
144 newCoef->setConstant(1);
145 continue;
146 }
147
148 // make the polynomial correction term
149 RooBernstein* poly = new RooBernstein("poly", "Bernstein poly", *x, coeff);
150
151 // make the corrected PDF = nominal * poly
152 RooEffProd* corrected = new RooEffProd("corrected","",*nominal,*poly);
153
154 // check to see how well this correction fits
155 RooFitResult* result = corrected->fitTo(*data,Save(),Minos(kFALSE), Hesse(kFALSE),PrintLevel(printLevel),Minimizer(minimType));
156
157 if (result->status() != 0) {
158 std::cout << "BernsteinCorrection::ImportCorrectedPdf - Error fit with corrected model failed" << std::endl;
159 return -1;
160 }
161
162
163 // Hypothesis test between previous correction (null)
164 // and this one (alternate). Use -2 log LR for test statistic
165 q = 2*(lastNll - result->minNll()); // -2 log lambda, goes like significance^2
166 // check if we should keep going based on rate of Type I error
167 keepGoing = (degree < 1 || TMath::Prob(q,1) < fTolerance );
168 if (degree >= fMaxDegree) keepGoing = false;
169
170 if(!keepGoing){
171 // terminate loop, import corrected PDF
172 //RooMsgService::instance().setGlobalKillBelow(RooFit::FATAL) ;
173 wks->import(*corrected);
174 //RooMsgService::instance().setGlobalKillBelow(RooFit::DEBUG) ;
175 } else {
176 // memory management
177 delete corrected;
178 delete poly;
179 }
180
181 // for the log
182 if(degree != 0){
183 log << "degree = " << degree
184 << " -log L("<<degree-1<<") = " << lastNll
185 << " -log L(" << degree <<") = " << result->minNll()
186 << " q = " << q
187 << " P(chi^2_1 > q) = " << TMath::Prob(q,1) << endl;;
188 }
189
190 // update last result for next iteration in loop
191 lastNll = result->minNll();
192
193 delete result;
194 }
195
196 log << "------ End Bernstein Correction Log --------" << endl;
197 cout << log.str();
198
199 return degree;
200}
201
202
203////////////////////////////////////////////////////////////////////////////////
204/// Create sampling distribution for q given degree-1 vs. degree corrections
205
207 const char* nominalName,
208 const char* varName,
209 const char* dataName,
210 TH1F* samplingDist,
211 TH1F* samplingDistExtra,
213 Int_t nToys){
214 // get ingredients out of workspace
215 RooRealVar* x = wks->var(varName);
216 RooAbsPdf* nominal = wks->pdf(nominalName);
217 RooAbsData* data = wks->data(dataName);
218
219 if (!x || !nominal || !data) {
220 cout << "Error: wrong name for pdf or variable or dataset ! " << std::endl;
221 return;
222 }
223
224 // setup a log
225 std::stringstream log;
226 log << "------ Begin Bernstein Correction Log --------" << endl;
227
228 // Local variables that we want to keep in scope after loop
229 RooArgList coeff; // n-th degree correction
230 RooArgList coeffNull; // n-1 correction
231 RooArgList coeffExtra; // n+1 correction
232 vector<RooRealVar*> coefficients;
233
234 //cout << "make coefs" << endl;
235 for(int i = 0; i<=degree+1; ++i) {
236 // we need to generate names for vars on the fly
237 std::stringstream str;
238 str<<"_"<<i;
239
240 RooRealVar* newCoef = new RooRealVar(("c"+str.str()).c_str(),
241 "Bernstein basis poly coefficient",
242 1., 0., fMaxCorrection);
243
244 // keep three sets of coefficients for n-1, n, n+1 corrections
245 if(i<degree) coeffNull.add(*newCoef);
246 if(i<=degree) coeff.add(*newCoef);
247 coeffExtra.add(*newCoef);
248 coefficients.push_back(newCoef);
249 }
250
251 // make the polynomial correction term
252 RooBernstein* poly
253 = new RooBernstein("poly", "Bernstein poly", *x, coeff);
254
255 // make the polynomial correction term
256 RooBernstein* polyNull
257 = new RooBernstein("polyNull", "Bernstein poly", *x, coeffNull);
258
259 // make the polynomial correction term
260 RooBernstein* polyExtra
261 = new RooBernstein("polyExtra", "Bernstein poly", *x, coeffExtra);
262
263 // make the corrected PDF = nominal * poly
264 RooEffProd* corrected
265 = new RooEffProd("corrected","",*nominal,*poly);
266
267 RooEffProd* correctedNull
268 = new RooEffProd("correctedNull","",*nominal,*polyNull);
269
270 RooEffProd* correctedExtra
271 = new RooEffProd("correctedExtra","",*nominal,*polyExtra);
272
273
274 cout << "made pdfs, make toy generator" << endl;
275
276 // make a PDF to generate the toys
277 RooDataHist dataHist("dataHist","",*x,*data);
278 RooHistPdf toyGen("toyGen","",*x,dataHist);
279
282
284 if (printLevel < 0) {
286 }
287
288
289 // TH1F* samplingDist = new TH1F("samplingDist","",20,0,10);
290 Double_t q = 0, qExtra = 0;
291 // do toys
292 for(int i=0; i<nToys; ++i){
293 cout << "on toy " << i << endl;
294
295 RooDataSet* tmpData = toyGen.generate(*x,data->numEntries());
296 // check to see how well this correction fits
297 RooFitResult* result
298 = corrected->fitTo(*tmpData,Save(),Minos(kFALSE),
299 Hesse(kFALSE),PrintLevel(printLevel),Minimizer(minimType));
300
301 RooFitResult* resultNull
302 = correctedNull->fitTo(*tmpData,Save(),Minos(kFALSE),
303 Hesse(kFALSE),PrintLevel(printLevel),Minimizer(minimType));
304
305
306 RooFitResult* resultExtra
307 = correctedExtra->fitTo(*tmpData,Save(),Minos(kFALSE),
308 Hesse(kFALSE),PrintLevel(printLevel),Minimizer(minimType));
309
310
311 // Hypothesis test between previous correction (null)
312 // and this one (alternate). Use -2 log LR for test statistic
313 q = 2*(resultNull->minNll() - result->minNll());
314
315 qExtra = 2*(result->minNll() - resultExtra->minNll());
316
317 samplingDist->Fill(q);
318 samplingDistExtra->Fill(qExtra);
319 if (printLevel > 0)
320 cout << "NLL Results: null " << resultNull->minNll() << " ref = " << result->minNll() << " extra" << resultExtra->minNll() << endl;
321
322
323 delete tmpData;
324 delete result;
325 delete resultNull;
326 delete resultExtra;
327 }
328
330
331 // return samplingDist;
332}
int Int_t
Definition: RtypesCore.h:41
const Bool_t kFALSE
Definition: RtypesCore.h:88
double Double_t
Definition: RtypesCore.h:55
#define ClassImp(name)
Definition: Rtypes.h:365
float * q
Definition: THbookFile.cxx:87
double log(double)
static const std::string & DefaultMinimizerType()
virtual Bool_t add(const RooAbsArg &var, Bool_t silent=kFALSE)
Add the specified argument to list.
RooAbsData is the common abstract base class for binned and unbinned datasets.
Definition: RooAbsData.h:37
virtual Int_t numEntries() const
Definition: RooAbsData.cxx:306
RooDataSet * generate(const RooArgSet &whatVars, Int_t nEvents, const RooCmdArg &arg1, const RooCmdArg &arg2=RooCmdArg::none(), const RooCmdArg &arg3=RooCmdArg::none(), const RooCmdArg &arg4=RooCmdArg::none(), const RooCmdArg &arg5=RooCmdArg::none())
See RooAbsPdf::generate(const RooArgSet&,const RooCmdArg&,const RooCmdArg&,const RooCmdArg&,...
Definition: RooAbsPdf.h:56
virtual RooFitResult * fitTo(RooAbsData &data, const RooCmdArg &arg1=RooCmdArg::none(), const RooCmdArg &arg2=RooCmdArg::none(), const RooCmdArg &arg3=RooCmdArg::none(), const RooCmdArg &arg4=RooCmdArg::none(), const RooCmdArg &arg5=RooCmdArg::none(), const RooCmdArg &arg6=RooCmdArg::none(), const RooCmdArg &arg7=RooCmdArg::none(), const RooCmdArg &arg8=RooCmdArg::none())
Fit PDF to given dataset.
Definition: RooAbsPdf.cxx:1119
void setConstant(Bool_t value=kTRUE)
RooArgList is a container object that can hold multiple RooAbsArg objects.
Definition: RooArgList.h:21
Bernstein basis polynomials are positive-definite in the range [0,1].
Definition: RooBernstein.h:23
The RooDataHist is a container class to hold N-dimensional binned data.
Definition: RooDataHist.h:40
RooDataSet is a container class to hold unbinned data.
Definition: RooDataSet.h:31
The class RooEffProd implements the product of a PDF with an efficiency function.
Definition: RooEffProd.h:20
RooFitResult is a container class to hold the input and output of a PDF fit to a dataset.
Definition: RooFitResult.h:40
Double_t minNll() const
Definition: RooFitResult.h:98
Int_t status() const
Definition: RooFitResult.h:77
RooHistPdf implements a probablity density function sampled from a multidimensional histogram.
Definition: RooHistPdf.h:28
static RooMsgService & instance()
Return reference to singleton instance.
void setGlobalKillBelow(RooFit::MsgLevel level)
RooFit::MsgLevel globalKillBelow() const
RooRealVar represents a fundamental (non-derived) real valued object.
Definition: RooRealVar.h:36
virtual void setVal(Double_t value)
Set value of variable to 'value'.
Definition: RooRealVar.cxx:233
BernsteinCorrection is a utility in RooStats to augment a nominal PDF with a polynomial correction te...
Int_t ImportCorrectedPdf(RooWorkspace *, const char *, const char *, const char *)
Main method for Bernstein correction.
void CreateQSamplingDist(RooWorkspace *wks, const char *nominalName, const char *varName, const char *dataName, TH1F *, TH1F *, Int_t degree, Int_t nToys=500)
Create sampling distribution for q given degree-1 vs. degree corrections.
The RooWorkspace is a persistable container for RooFit projects.
Definition: RooWorkspace.h:43
RooAbsData * data(const char *name) const
Retrieve dataset (binned or unbinned) with given name. A null pointer is returned if not found.
RooRealVar * var(const char *name) const
Retrieve real-valued variable (RooRealVar) with given name. A null pointer is returned if not found.
Bool_t import(const RooAbsArg &arg, const RooCmdArg &arg1=RooCmdArg(), const RooCmdArg &arg2=RooCmdArg(), const RooCmdArg &arg3=RooCmdArg(), const RooCmdArg &arg4=RooCmdArg(), const RooCmdArg &arg5=RooCmdArg(), const RooCmdArg &arg6=RooCmdArg(), const RooCmdArg &arg7=RooCmdArg(), const RooCmdArg &arg8=RooCmdArg(), const RooCmdArg &arg9=RooCmdArg())
Import a RooAbsArg object, e.g.
RooAbsPdf * pdf(const char *name) const
Retrieve p.d.f (RooAbsPdf) with given name. A null pointer is returned if not found.
1-D histogram with a float per channel (see TH1 documentation)}
Definition: TH1.h:571
virtual Int_t Fill(Double_t x)
Increment bin with abscissa X by 1.
Definition: TH1.cxx:3258
Basic string class.
Definition: TString.h:131
Double_t x[n]
Definition: legend1.C:17
Template specialisation used in RooAbsArg:
RooCmdArg Hesse(Bool_t flag=kTRUE)
RooCmdArg Save(Bool_t flag=kTRUE)
RooCmdArg PrintLevel(Int_t code)
RooCmdArg Minimizer(const char *type, const char *alg=0)
RooCmdArg Minos(Bool_t flag=kTRUE)
Namespace for the RooStats classes.
Definition: Asimov.h:20
static constexpr double degree
Double_t Prob(Double_t chi2, Int_t ndf)
Computation of the probability for a certain Chi-squared (chi2) and number of degrees of freedom (ndf...
Definition: TMath.cxx:621