Logo ROOT  
Reference Guide
 
Loading...
Searching...
No Matches
HybridStandardForm.C
Go to the documentation of this file.
1/// \file
2/// \ingroup tutorial_roostats
3/// \notebook -js
4/// A hypothesis testing example based on number counting with background uncertainty.
5///
6/// A hypothesis testing example based on number counting
7/// with background uncertainty.
8///
9/// NOTE: This example is like HybridInstructional, but the model is more clearly
10/// generalizable to an analysis with shapes. There is a lot of flexibility
11/// for how one models a problem in RooFit/RooStats. Models come in a few
12/// common forms:
13/// - standard form: extended PDF of some discriminating variable m:
14/// eg: P(m) ~ S*fs(m) + B*fb(m), with S+B events expected
15/// in this case the dataset has N rows corresponding to N events
16/// and the extended term is Pois(N|S+B)
17///
18/// - fractional form: non-extended PDF of some discriminating variable m:
19/// eg: P(m) ~ s*fs(m) + (1-s)*fb(m), where s is a signal fraction
20/// in this case the dataset has N rows corresponding to N events
21/// and there is no extended term
22///
23/// - number counting form: in which there is no discriminating variable
24/// and the counts are modeled directly (see HybridInstructional)
25/// eg: P(N) = Pois(N|S+B)
26/// in this case the dataset has 1 row corresponding to N events
27/// and the extended term is the PDF itself.
28///
29/// Here we convert the number counting form into the standard form by
30/// introducing a dummy discriminating variable m with a uniform distribution.
31///
32/// This example:
33/// - demonstrates the usage of the HybridCalcultor (Part 4-6)
34/// - demonstrates the numerical integration of RooFit (Part 2)
35/// - validates the RooStats against an example with a known analytic answer
36/// - demonstrates usage of different test statistics
37/// - explains subtle choices in the prior used for hybrid methods
38/// - demonstrates usage of different priors for the nuisance parameters
39/// - demonstrates usage of PROOF
40///
41/// The basic setup here is that a main measurement has observed x events with an
42/// expectation of s+b. One can choose an ad hoc prior for the uncertainty on b,
43/// or try to base it on an auxiliary measurement. In this case, the auxiliary
44/// measurement (aka control measurement, sideband) is another counting experiment
45/// with measurement y and expectation tau*b. With an 'original prior' on b,
46/// called \f$ \eta(b) \f$ then one can obtain a posterior from the auxiliary measurement
47/// \f$ \pi(b) = \eta(b) * Pois(y|tau*b) \f$. This is a principled choice for a prior
48/// on b in the main measurement of x, which can then be treated in a hybrid
49/// Bayesian/Frequentist way. Additionally, one can try to treat the two
50/// measurements simultaneously, which is detailed in Part 6 of the tutorial.
51///
52/// This tutorial is related to the FourBin.C tutorial in the modeling, but
53/// focuses on hypothesis testing instead of interval estimation.
54///
55/// More background on this 'prototype problem' can be found in the
56/// following papers:
57///
58/// - Evaluation of three methods for calculating statistical significance
59/// when incorporating a systematic uncertainty into a test of the
60/// background-only hypothesis for a Poisson process
61/// Authors: Robert D. Cousins, James T. Linnemann, Jordan Tucker
62/// http://arxiv.org/abs/physics/0702156
63/// NIM A 595 (2008) 480--501
64///
65/// - Statistical Challenges for Searches for New Physics at the LHC
66/// Author: Kyle Cranmer
67/// http://arxiv.org/abs/physics/0511028
68///
69/// - Measures of Significance in HEP and Astrophysics
70/// Author: J. T. Linnemann
71/// http://arxiv.org/abs/physics/0312059
72///
73/// \macro_image
74/// \macro_output
75/// \macro_code
76///
77/// \authors Kyle Cranmer, Wouter Verkerke, and Sven Kreiss
78
79#include "RooGlobalFunc.h"
80#include "RooRealVar.h"
81#include "RooProdPdf.h"
82#include "RooWorkspace.h"
83#include "RooDataSet.h"
84#include "RooDataHist.h"
85#include "TCanvas.h"
86#include "TStopwatch.h"
87#include "TH1.h"
88#include "RooPlot.h"
89#include "RooMsgService.h"
90
92
96
102
103using namespace RooFit;
104using namespace RooStats;
105
106//-------------------------------------------------------
107// A New Test Statistic Class for this example.
108// It simply returns the sum of the values in a particular
109// column of a dataset.
110// You can ignore this class and focus on the macro below
111
112class BinCountTestStat : public TestStatistic {
113public:
114 BinCountTestStat(void) : fColumnName("tmp") {}
115 BinCountTestStat(string columnName) : fColumnName(columnName) {}
116
117 virtual Double_t Evaluate(RooAbsData &data, RooArgSet & /*nullPOI*/)
118 {
119 // This is the main method in the interface
120 Double_t value = 0.0;
121 for (int i = 0; i < data.numEntries(); i++) {
122 value += data.get(i)->getRealValue(fColumnName.c_str());
123 }
124 return value;
125 }
126 virtual const TString GetVarName() const { return fColumnName; }
127
128private:
129 string fColumnName;
130
131protected:
132 ClassDef(BinCountTestStat, 1)
133};
134
135ClassImp(BinCountTestStat)
136
137//-----------------------------
138// The Actual Tutorial Macro
139//-----------------------------
140
141void HybridStandardForm(int ntoys = 6000)
142{
143 double nToysRatio = 20; // ratio Ntoys Null/ntoys ALT
144
145 // This tutorial has 6 parts
146 // Table of Contents
147 // Setup
148 // 1. Make the model for the 'prototype problem'
149 // Special cases
150 // 2. NOT RELEVANT HERE
151 // 3. Use RooStats analytic solution for this problem
152 // RooStats HybridCalculator -- can be generalized
153 // 4. RooStats ToyMC version of 2. & 3.
154 // 5. RooStats ToyMC with an equivalent test statistic
155 // 6. RooStats ToyMC with simultaneous control & main measurement
156
157 // Part 4 takes ~4 min without PROOF.
158 // Part 5 takes about ~2 min with PROOF on 4 cores.
159 // Of course, everything looks nicer with more toys, which takes longer.
160
161 TStopwatch t;
162 t.Start();
163 TCanvas *c = new TCanvas;
164 c->Divide(2, 2);
165
166 //-----------------------------------------------------
167 // P A R T 1 : D I R E C T I N T E G R A T I O N
168 // ====================================================
169 // Make model for prototype on/off problem
170 // Pois(x | s+b) * Pois(y | tau b )
171 // for Z_Gamma, use uniform prior on b.
172 RooWorkspace *w = new RooWorkspace("w");
173
174 // replace the pdf in 'number counting form'
175 // w->factory("Poisson::px(x[150,0,500],sum::splusb(s[0,0,100],b[100,0,300]))");
176 // with one in standard form. Now x is encoded in event count
177 w->factory("Uniform::f(m[0,1])"); // m is a dummy discriminating variable
178 w->factory("ExtendPdf::px(f,sum::splusb(s[0,0,100],b[100,0.1,300]))");
179 w->factory("Poisson::py(y[100,0.1,500],prod::taub(tau[1.],b))");
180 w->factory("PROD::model(px,py)");
181 w->factory("Uniform::prior_b(b)");
182
183 // We will control the output level in a few places to avoid
184 // verbose progress messages. We start by keeping track
185 // of the current threshold on messages.
187
188 // Use PROOF-lite on multi-core machines
189 ProofConfig *pc = NULL;
190 // uncomment below if you want to use PROOF
191 pc = new ProofConfig(*w, 4, "workers=4", kFALSE); // machine with 4 cores
192 // pc = new ProofConfig(*w, 2, "workers=2", kFALSE); // machine with 2 cores
193
194 //-----------------------------------------------
195 // P A R T 3 : A N A L Y T I C R E S U L T
196 // ==============================================
197 // In this special case, the integrals are known analytically
198 // and they are implemented in RooStats::NumberCountingUtils
199
200 // analytic Z_Bi
201 double p_Bi = NumberCountingUtils::BinomialWithTauObsP(150, 100, 1);
202 double Z_Bi = NumberCountingUtils::BinomialWithTauObsZ(150, 100, 1);
203 cout << "-----------------------------------------" << endl;
204 cout << "Part 3" << endl;
205 std::cout << "Z_Bi p-value (analytic): " << p_Bi << std::endl;
206 std::cout << "Z_Bi significance (analytic): " << Z_Bi << std::endl;
207 t.Stop();
208 t.Print();
209 t.Reset();
210 t.Start();
211
212 //--------------------------------------------------------------
213 // P A R T 4 : U S I N G H Y B R I D C A L C U L A T O R
214 // ==============================================================
215 // Now we demonstrate the RooStats HybridCalculator.
216 //
217 // Like all RooStats calculators it needs the data and a ModelConfig
218 // for the relevant hypotheses. Since we are doing hypothesis testing
219 // we need a ModelConfig for the null (background only) and the alternate
220 // (signal+background) hypotheses. We also need to specify the PDF,
221 // the parameters of interest, and the observables. Furthermore, since
222 // the parameter of interest is floating, we need to specify which values
223 // of the parameter corresponds to the null and alternate (eg. s=0 and s=50)
224 //
225 // define some sets of variables obs={x} and poi={s}
226 // note here, x is the only observable in the main measurement
227 // and y is treated as a separate measurement, which is used
228 // to produce the prior that will be used in this calculation
229 // to randomize the nuisance parameters.
230 w->defineSet("obs", "m");
231 w->defineSet("poi", "s");
232
233 // create a toy dataset with the x=150
234 // RooDataSet *data = new RooDataSet("d", "d", *w->set("obs"));
235 // data->add(*w->set("obs"));
236 std::unique_ptr<RooDataSet> data{w->pdf("px")->generate(*w->set("obs"), 150)};
237
238 // Part 3a : Setup ModelConfigs
239 //-------------------------------------------------------
240 // create the null (background-only) ModelConfig with s=0
241 ModelConfig b_model("B_model", w);
242 b_model.SetPdf(*w->pdf("px"));
243 b_model.SetObservables(*w->set("obs"));
244 b_model.SetParametersOfInterest(*w->set("poi"));
245 w->var("s")->setVal(0.0); // important!
246 b_model.SetSnapshot(*w->set("poi"));
247
248 // create the alternate (signal+background) ModelConfig with s=50
249 ModelConfig sb_model("S+B_model", w);
250 sb_model.SetPdf(*w->pdf("px"));
251 sb_model.SetObservables(*w->set("obs"));
252 sb_model.SetParametersOfInterest(*w->set("poi"));
253 w->var("s")->setVal(50.0); // important!
254 sb_model.SetSnapshot(*w->set("poi"));
255
256 // Part 3b : Choose Test Statistic
257 //--------------------------------------------------------------
258 // To make an equivalent calculation we need to use x as the test
259 // statistic. This is not a built-in test statistic in RooStats
260 // so we define it above. The new class inherits from the
261 // RooStats::TestStatistic interface, and simply returns the value
262 // of x in the dataset.
263
264 NumEventsTestStat eventCount(*w->pdf("px"));
265
266 // Part 3c : Define Prior used to randomize nuisance parameters
267 //-------------------------------------------------------------
268 //
269 // The prior used for the hybrid calculator is the posterior
270 // from the auxiliary measurement y. The model for the aux.
271 // measurement is Pois(y|tau*b), thus the likelihood function
272 // is proportional to (has the form of) a Gamma distribution.
273 // if the 'original prior' $\eta(b)$ is uniform, then from
274 // Bayes's theorem we have the posterior:
275 // $\pi(b) = Pois(y|tau*b) * \eta(b)$
276 // If $\eta(b)$ is flat, then we arrive at a Gamma distribution.
277 // Since RooFit will normalize the PDF we can actually supply
278 // py=Pois(y,tau*b) that will be equivalent to multiplying by a uniform.
279 //
280 // Alternatively, we could explicitly use a gamma distribution:
281 //
282 // `w->factory("Gamma::gamma(b,sum::temp(y,1),1,0)");`
283 //
284 // or we can use some other ad hoc prior that do not naturally
285 // follow from the known form of the auxiliary measurement.
286 // The common choice is the equivalent Gaussian:
287 w->factory("Gaussian::gauss_prior(b,y, expr::sqrty('sqrt(y)',y))");
288 // this corresponds to the "Z_N" calculation.
289 //
290 // or one could use the analogous log-normal prior
291 w->factory("Lognormal::lognorm_prior(b,y, expr::kappa('1+1./sqrt(y)',y))");
292 //
293 // Ideally, the HybridCalculator would be able to inspect the full
294 // model Pois(x | s+b) * Pois(y | tau b ) and be given the original
295 // prior $\eta(b)$ to form $\pi(b) = Pois(y|tau*b) * \eta(b)$.
296 // This is not yet implemented because in the general case
297 // it is not easy to identify the terms in the PDF that correspond
298 // to the auxiliary measurement. So for now, it must be set
299 // explicitly with:
300 // - ForcePriorNuisanceNull()
301 // - ForcePriorNuisanceAlt()
302 // the name "ForcePriorNuisance" was chosen because we anticipate
303 // this to be auto-detected, but will leave the option open
304 // to force to a different prior for the nuisance parameters.
305
306 // Part 3d : Construct and configure the HybridCalculator
307 //-------------------------------------------------------
308
309 HybridCalculator hc1(*data, sb_model, b_model);
310 ToyMCSampler *toymcs1 = (ToyMCSampler *)hc1.GetTestStatSampler();
311 // toymcs1->SetNEventsPerToy(1); // because the model is in number counting form
312 toymcs1->SetTestStatistic(&eventCount); // set the test statistic
313 // toymcs1->SetGenerateBinned();
314 hc1.SetToys(ntoys, ntoys / nToysRatio);
315 hc1.ForcePriorNuisanceAlt(*w->pdf("py"));
316 hc1.ForcePriorNuisanceNull(*w->pdf("py"));
317 // if you wanted to use the ad hoc Gaussian prior instead
318 // ~~~
319 // hc1.ForcePriorNuisanceAlt(*w->pdf("gauss_prior"));
320 // hc1.ForcePriorNuisanceNull(*w->pdf("gauss_prior"));
321 // ~~~
322 // if you wanted to use the ad hoc log-normal prior instead
323 // ~~~
324 // hc1.ForcePriorNuisanceAlt(*w->pdf("lognorm_prior"));
325 // hc1.ForcePriorNuisanceNull(*w->pdf("lognorm_prior"));
326 // ~~~
327
328 // enable proof
329 // proof not enabled for this test statistic
330 // if(pc) toymcs1->SetProofConfig(pc);
331
332 // these lines save current msg level and then kill any messages below ERROR
334 // Get the result
335 HypoTestResult *r1 = hc1.GetHypoTest();
336 RooMsgService::instance().setGlobalKillBelow(msglevel); // set it back
337 cout << "-----------------------------------------" << endl;
338 cout << "Part 4" << endl;
339 r1->Print();
340 t.Stop();
341 t.Print();
342 t.Reset();
343 t.Start();
344
345 c->cd(2);
346 HypoTestPlot *p1 = new HypoTestPlot(*r1, 30); // 30 bins, TS is discrete
347 p1->Draw();
348
349 return; // keep the running time sort by default
350 //-------------------------------------------------------------------------
351 // # P A R T 5 : U S I N G H Y B R I D C A L C U L A T O R W I T H
352 // # A N A L T E R N A T I V E T E S T S T A T I S T I C
353 //
354 // A likelihood ratio test statistics should be 1-to-1 with the count x
355 // when the value of b is fixed in the likelihood. This is implemented
356 // by the SimpleLikelihoodRatioTestStat
357
358 SimpleLikelihoodRatioTestStat slrts(*b_model.GetPdf(), *sb_model.GetPdf());
359 slrts.SetNullParameters(*b_model.GetSnapshot());
360 slrts.SetAltParameters(*sb_model.GetSnapshot());
361
362 // HYBRID CALCULATOR
363 HybridCalculator hc2(*data, sb_model, b_model);
364 ToyMCSampler *toymcs2 = (ToyMCSampler *)hc2.GetTestStatSampler();
365 // toymcs2->SetNEventsPerToy(1);
366 toymcs2->SetTestStatistic(&slrts);
367 // toymcs2->SetGenerateBinned();
368 hc2.SetToys(ntoys, ntoys / nToysRatio);
369 hc2.ForcePriorNuisanceAlt(*w->pdf("py"));
370 hc2.ForcePriorNuisanceNull(*w->pdf("py"));
371 // if you wanted to use the ad hoc Gaussian prior instead
372 // ~~~
373 // hc2.ForcePriorNuisanceAlt(*w->pdf("gauss_prior"));
374 // hc2.ForcePriorNuisanceNull(*w->pdf("gauss_prior"));
375 // ~~~
376 // if you wanted to use the ad hoc log-normal prior instead
377 // ~~~
378 // hc2.ForcePriorNuisanceAlt(*w->pdf("lognorm_prior"));
379 // hc2.ForcePriorNuisanceNull(*w->pdf("lognorm_prior"));
380 // ~~~
381
382 // enable proof
383 if (pc)
384 toymcs2->SetProofConfig(pc);
385
386 // these lines save current msg level and then kill any messages below ERROR
388 // Get the result
389 HypoTestResult *r2 = hc2.GetHypoTest();
390 cout << "-----------------------------------------" << endl;
391 cout << "Part 5" << endl;
392 r2->Print();
393 t.Stop();
394 t.Print();
395 t.Reset();
396 t.Start();
398
399 c->cd(3);
400 HypoTestPlot *p2 = new HypoTestPlot(*r2, 30); // 30 bins
401 p2->Draw();
402
403 return; // so standard tutorial runs faster
404
405 //---------------------------------------------
406 // OUTPUT W/O PROOF (2.66 GHz Intel Core i7)
407 // ============================================
408
409 // -----------------------------------------
410 // Part 3
411 // Z_Bi p-value (analytic): 0.00094165
412 // Z_Bi significance (analytic): 3.10804
413 // Real time 0:00:00, CP time 0.610
414
415 // Results HybridCalculator_result:
416 // - Null p-value = 0.00103333 +/- 0.000179406
417 // - Significance = 3.08048 sigma
418 // - Number of S+B toys: 1000
419 // - Number of B toys: 30000
420 // - Test statistic evaluated on data: 150
421 // - CL_b: 0.998967 +/- 0.000185496
422 // - CL_s+b: 0.495 +/- 0.0158106
423 // - CL_s: 0.495512 +/- 0.0158272
424 // Real time 0:04:43, CP time 283.780
425
426 // With PROOF
427 // -----------------------------------------
428 // Part 5
429
430 // Results HybridCalculator_result:
431 // - Null p-value = 0.00105 +/- 0.000206022
432 // - Significance = 3.07571 sigma
433 // - Number of S+B toys: 1000
434 // - Number of B toys: 20000
435 // - Test statistic evaluated on data: 10.8198
436 // - CL_b: 0.99895 +/- 0.000229008
437 // - CL_s+b: 0.491 +/- 0.0158088
438 // - CL_s: 0.491516 +/- 0.0158258
439 // Real time 0:02:22, CP time 0.990
440
441 //-------------------------------------------------------
442 // Comparison
443 //-------------------------------------------------------
444 // LEPStatToolsForLHC
445 // https://plone4.fnal.gov:4430/P0/phystat/packages/0703002
446 // Uses Gaussian prior
447 // CL_b = 6.218476e-04, Significance = 3.228665 sigma
448 //
449 //-------------------------------------------------------
450 // Comparison
451 //-------------------------------------------------------
452 // Asymptotic
453 // From the value of the profile likelihood ratio (5.0338)
454 // The significance can be estimated using Wilks's theorem
455 // significance = sqrt(2*profileLR) = 3.1729 sigma
456}
#define c(i)
Definition RSha256.hxx:101
constexpr Bool_t kFALSE
Definition RtypesCore.h:101
double Double_t
Definition RtypesCore.h:59
#define ClassDef(name, id)
Definition Rtypes.h:337
#define ClassImp(name)
Definition Rtypes.h:377
Option_t Option_t TPoint TPoint const char GetTextMagnitude GetFillStyle GetLineColor GetLineWidth GetMarkerStyle GetTextAlign GetTextColor GetTextSize void data
Option_t Option_t TPoint TPoint const char GetTextMagnitude GetFillStyle GetLineColor GetLineWidth GetMarkerStyle GetTextAlign GetTextColor GetTextSize void value
Abstract base class for binned and unbinned datasets.
Definition RooAbsData.h:57
RooArgSet is a container object that can hold multiple RooAbsArg objects.
Definition RooArgSet.h:55
static RooMsgService & instance()
Return reference to singleton instance.
void setGlobalKillBelow(RooFit::MsgLevel level)
RooFit::MsgLevel globalKillBelow() const
Same purpose as HybridCalculatorOriginal, but different implementation.
This class provides the plots for the result of a study performed with any of the HypoTestCalculatorG...
HypoTestResult is a base class for results from hypothesis tests.
void Print(const Option_t *="") const override
Print out some information about the results Note: use Alt/Null labels for the hypotheses here as the...
ModelConfig is a simple class that holds configuration information specifying how a model should be u...
Definition ModelConfig.h:35
NumEventsTestStat is a simple implementation of the TestStatistic interface used for simple number co...
Holds configuration options for proof and proof-lite.
Definition ProofConfig.h:45
void Draw(Option_t *options=nullptr) override
Draw this plot and all of the elements it contains.
TestStatistic class that returns -log(L[null] / L[alt]) where L is the likelihood.
TestStatistic is an interface class to provide a facility for construction test statistics distributi...
ToyMCSampler is an implementation of the TestStatSampler interface.
void SetProofConfig(ProofConfig *pc=nullptr)
calling with argument or nullptr deactivates proof
virtual void SetTestStatistic(TestStatistic *testStatistic, unsigned int i)
Set the TestStatistic (want the argument to be a function of the data & parameter points.
Persistable container for RooFit projects.
The Canvas class.
Definition TCanvas.h:23
void Divide(Int_t nx=1, Int_t ny=1, Float_t xmargin=0.01, Float_t ymargin=0.01, Int_t color=0) override
Automatic pad generation by division.
Definition TPad.cxx:1153
Stopwatch class.
Definition TStopwatch.h:28
void Start(Bool_t reset=kTRUE)
Start the stopwatch.
void Stop()
Stop the stopwatch.
void Reset()
Definition TStopwatch.h:52
void Print(Option_t *option="") const override
Print the real and cpu time passed between the start and stop events.
Basic string class.
Definition TString.h:139
The namespace RooFit contains mostly switches that change the behaviour of functions of PDFs (or othe...
Definition JSONIO.h:26
MsgLevel
Verbosity level for RooMsgService::StreamConfig in RooMsgService.
Namespace for the RooStats classes.
Definition Asimov.h:19