1 | /* ======================================================================== *\
|
---|
2 | !
|
---|
3 | ! *
|
---|
4 | ! * This file is part of MARS, the MAGIC Analysis and Reconstruction
|
---|
5 | ! * Software. It is distributed to you in the hope that it can be a useful
|
---|
6 | ! * and timesaving tool in analysing Data of imaging Cerenkov telescopes.
|
---|
7 | ! * It is distributed WITHOUT ANY WARRANTY.
|
---|
8 | ! *
|
---|
9 | ! * Permission to use, copy, modify and distribute this software and its
|
---|
10 | ! * documentation for any purpose is hereby granted without fee,
|
---|
11 | ! * provided that the above copyright notice appear in all copies and
|
---|
12 | ! * that both that copyright notice and this permission notice appear
|
---|
13 | ! * in supporting documentation. It is provided "as is" without express
|
---|
14 | ! * or implied warranty.
|
---|
15 | ! *
|
---|
16 | !
|
---|
17 | !
|
---|
18 | ! Author(s): Thomas Bretz, 3/2004 <mailto:tbretz@astro.uni-wuerzburg.de>
|
---|
19 | !
|
---|
20 | ! Copyright: MAGIC Software Development, 2000-2008
|
---|
21 | !
|
---|
22 | !
|
---|
23 | \* ======================================================================== */
|
---|
24 |
|
---|
25 | //////////////////////////////////////////////////////////////////////////////
|
---|
26 | //
|
---|
27 | // MAlphaFitter
|
---|
28 | //
|
---|
29 | // Create a single Alpha-Plot. The alpha-plot is fitted online. You can
|
---|
30 | // check the result when it is filles in the MStatusDisplay
|
---|
31 | // For more information see MHFalseSource::FitSignificance
|
---|
32 | //
|
---|
33 | // For convinience (fit) the output significance is stored in a
|
---|
34 | // container in the parlisrt
|
---|
35 | //
|
---|
36 | // Version 2:
|
---|
37 | // ----------
|
---|
38 | // + Double_t fSignificanceExc; // significance of a known excess
|
---|
39 | //
|
---|
40 | // Version 3:
|
---|
41 | // ----------
|
---|
42 | // + TArrayD fErrors; // errors of coefficients
|
---|
43 | //
|
---|
44 | // Version 4:
|
---|
45 | // ----------
|
---|
46 | // + Double_t fErrorExcess;
|
---|
47 | // - Double_t fSignificanceExc;
|
---|
48 | //
|
---|
49 | //
|
---|
50 | //////////////////////////////////////////////////////////////////////////////
|
---|
51 | #include "MAlphaFitter.h"
|
---|
52 |
|
---|
53 | #include <TF1.h>
|
---|
54 | #include <TH1.h>
|
---|
55 | #include <TH3.h>
|
---|
56 |
|
---|
57 | #include <TRandom.h>
|
---|
58 | #include <TFeldmanCousins.h>
|
---|
59 |
|
---|
60 | #include <TLine.h>
|
---|
61 | #include <TLatex.h>
|
---|
62 | #include <TVirtualPad.h>
|
---|
63 |
|
---|
64 | #include "MMath.h"
|
---|
65 | #include "MString.h"
|
---|
66 |
|
---|
67 | #include "MLogManip.h"
|
---|
68 |
|
---|
69 | ClassImp(MAlphaFitter);
|
---|
70 |
|
---|
71 | using namespace std;
|
---|
72 |
|
---|
73 | // --------------------------------------------------------------------------
|
---|
74 | //
|
---|
75 | // Default constructor.
|
---|
76 | //
|
---|
77 | MAlphaFitter::MAlphaFitter(const char *name, const char *title) : fSigInt(15),
|
---|
78 | fSigMax(75), fBgMin(45), fBgMax(85), fScaleMin(40), fScaleMax(80),
|
---|
79 | fPolynomOrder(2), fFitBackground(kTRUE), fFunc(0),
|
---|
80 | fScaleMode(kOffRegion), fScaleUser(1), fStrategy(kSignificance)
|
---|
81 | {
|
---|
82 | fName = name ? name : "MAlphaFitter";
|
---|
83 | fTitle = title ? title : "Fit alpha";
|
---|
84 |
|
---|
85 | SetSignalFunction(kGauss);
|
---|
86 |
|
---|
87 | Clear();
|
---|
88 | }
|
---|
89 |
|
---|
90 | // --------------------------------------------------------------------------
|
---|
91 | //
|
---|
92 | // Destructor
|
---|
93 | //
|
---|
94 | MAlphaFitter::~MAlphaFitter()
|
---|
95 | {
|
---|
96 | delete fFunc;
|
---|
97 | }
|
---|
98 |
|
---|
99 | // --------------------------------------------------------------------------
|
---|
100 | //
|
---|
101 | // Re-initializes fFunc either according to SignalFunc_t
|
---|
102 | //
|
---|
103 | void MAlphaFitter::SetSignalFunction(SignalFunc_t func)
|
---|
104 | {
|
---|
105 | if (gROOT->GetListOfFunctions()->FindObject(""))
|
---|
106 | {
|
---|
107 | gLog << err << "MAlphaFitter::SetSignalFunction -- '' found!" << endl;
|
---|
108 | return;
|
---|
109 | }
|
---|
110 |
|
---|
111 | delete fFunc;
|
---|
112 | fFunc = 0;
|
---|
113 |
|
---|
114 | switch (func)
|
---|
115 | {
|
---|
116 | case kGauss:
|
---|
117 | fFunc=new TF1("", MString::Format("gaus(0) + pol%d(3)", fPolynomOrder));
|
---|
118 | break;
|
---|
119 | case kThetaSq:
|
---|
120 | if (fPolynomOrder>0)
|
---|
121 | fPolynomOrder = 1;
|
---|
122 | fFunc=new TF1("", "[0]*exp(-0.5*((sqrt(x)-[1])/[2])^2) + expo(3)");
|
---|
123 | break;
|
---|
124 | }
|
---|
125 |
|
---|
126 | fSignalFunc=func;
|
---|
127 |
|
---|
128 | fFunc->SetName("Dummy");
|
---|
129 | gROOT->GetListOfFunctions()->Remove(fFunc);
|
---|
130 |
|
---|
131 | fCoefficients.Set(3+fPolynomOrder+1);
|
---|
132 | fCoefficients.Reset();
|
---|
133 |
|
---|
134 | fErrors.Set(3+fPolynomOrder+1);
|
---|
135 | fErrors.Reset();
|
---|
136 | }
|
---|
137 |
|
---|
138 | // --------------------------------------------------------------------------
|
---|
139 | //
|
---|
140 | // Reset variables which belong to results. Reset the arrays.
|
---|
141 | //
|
---|
142 | void MAlphaFitter::Clear(Option_t *o)
|
---|
143 | {
|
---|
144 | fSignificance=0;
|
---|
145 | fErrorExcess=0;
|
---|
146 | fEventsExcess=0;
|
---|
147 | fEventsSignal=0;
|
---|
148 | fEventsBackground=0;
|
---|
149 |
|
---|
150 | fChiSqSignal=0;
|
---|
151 | fChiSqBg=0;
|
---|
152 | fIntegralMax=0;
|
---|
153 | fScaleFactor=1;
|
---|
154 |
|
---|
155 | fCoefficients.Reset();
|
---|
156 | fErrors.Reset();
|
---|
157 | }
|
---|
158 |
|
---|
159 | // --------------------------------------------------------------------------
|
---|
160 | //
|
---|
161 | // Returns fFunc->Eval(d) or 0 if fFunc==NULL
|
---|
162 | //
|
---|
163 | Double_t MAlphaFitter::Eval(Double_t d) const
|
---|
164 | {
|
---|
165 | return fFunc ? fFunc->Eval(d) : 0;
|
---|
166 | }
|
---|
167 |
|
---|
168 | // --------------------------------------------------------------------------
|
---|
169 | //
|
---|
170 | // This function implementes the fit to the off-data as used in Fit()
|
---|
171 | //
|
---|
172 | Bool_t MAlphaFitter::FitOff(TH1D &h, Int_t paint)
|
---|
173 | {
|
---|
174 | if (h.GetEntries()==0)
|
---|
175 | return kFALSE;
|
---|
176 |
|
---|
177 | // First fit a polynom in the off region
|
---|
178 | fFunc->FixParameter(0, 0);
|
---|
179 | fFunc->FixParameter(1, 0);
|
---|
180 | fFunc->FixParameter(2, 1);
|
---|
181 | fFunc->ReleaseParameter(3);
|
---|
182 | if (fPolynomOrder!=1)
|
---|
183 | fFunc->FixParameter(4, 0);
|
---|
184 |
|
---|
185 | for (int i=5; i<fFunc->GetNpar(); i++)
|
---|
186 | if (fFitBackground)
|
---|
187 | fFunc->ReleaseParameter(i);
|
---|
188 | else
|
---|
189 | fFunc->SetParameter(i, 0);
|
---|
190 |
|
---|
191 | if (!fFitBackground)
|
---|
192 | return kTRUE;
|
---|
193 |
|
---|
194 | if (fSignalFunc==kThetaSq)
|
---|
195 | {
|
---|
196 | const Double_t sum = h.Integral(1, 3)/3;
|
---|
197 | const Double_t a = sum<=1 ? 0 : TMath::Log(sum);
|
---|
198 | const Double_t b = -1.7;
|
---|
199 |
|
---|
200 | // Do a best-guess
|
---|
201 | fFunc->SetParameter(3, a);
|
---|
202 | fFunc->SetParameter(4, b);
|
---|
203 | }
|
---|
204 |
|
---|
205 | // options : N do not store the function, do not draw
|
---|
206 | // I use integral of function in bin rather than value at bin center
|
---|
207 | // R use the range specified in the function range
|
---|
208 | // Q quiet mode
|
---|
209 | // E Perform better Errors estimation using Minos technique
|
---|
210 | h.Fit(fFunc, "NQI", "", fBgMin, fBgMax);
|
---|
211 | fChiSqBg = fFunc->GetChisquare()/fFunc->GetNDF();
|
---|
212 |
|
---|
213 | fCoefficients.Set(fFunc->GetNpar(), fFunc->GetParameters());
|
---|
214 | fErrors.Set(fFunc->GetNpar());
|
---|
215 | for (int i=3; i<fFunc->GetNpar(); i++)
|
---|
216 | fErrors[i] = fFunc->GetParError(i);
|
---|
217 |
|
---|
218 | // ------------------------------------
|
---|
219 |
|
---|
220 | if (paint)
|
---|
221 | {
|
---|
222 | if (paint==2)
|
---|
223 | {
|
---|
224 | fFunc->SetLineColor(kBlack);
|
---|
225 | fFunc->SetLineWidth(1);
|
---|
226 | }
|
---|
227 | else
|
---|
228 | {
|
---|
229 | fFunc->SetRange(0, 90);
|
---|
230 | fFunc->SetLineColor(kRed);
|
---|
231 | fFunc->SetLineWidth(2);
|
---|
232 | }
|
---|
233 | fFunc->Paint("same");
|
---|
234 | }
|
---|
235 |
|
---|
236 | return kTRUE;
|
---|
237 | }
|
---|
238 |
|
---|
239 | // --------------------------------------------------------------------------
|
---|
240 | //
|
---|
241 | // Calculate the result of the fit and set the corresponding data members
|
---|
242 | //
|
---|
243 | void MAlphaFitter::FitResult(const TH1D &h)
|
---|
244 | {
|
---|
245 | const Double_t alphaw = h.GetXaxis()->GetBinWidth(1);
|
---|
246 |
|
---|
247 | const Int_t bin = h.GetXaxis()->FindFixBin(fSigInt*0.999);
|
---|
248 |
|
---|
249 | fIntegralMax = h.GetBinLowEdge(bin+1);
|
---|
250 | fEventsBackground = fFunc->Integral(0, fIntegralMax)/alphaw;
|
---|
251 | fEventsSignal = h.Integral(1, bin);
|
---|
252 | fEventsExcess = fEventsSignal-fEventsBackground;
|
---|
253 | fSignificance = MMath::SignificanceLiMaSigned(fEventsSignal, fEventsBackground);
|
---|
254 | fErrorExcess = MMath::ErrorExc(fEventsSignal, fEventsBackground);
|
---|
255 |
|
---|
256 | // !Finitite includes IsNaN
|
---|
257 | if (!TMath::Finite(fSignificance))
|
---|
258 | fSignificance=0;
|
---|
259 | }
|
---|
260 |
|
---|
261 | // --------------------------------------------------------------------------
|
---|
262 | //
|
---|
263 | // This is a preliminary implementation of a alpha-fit procedure for
|
---|
264 | // all possible source positions. It will be moved into its own
|
---|
265 | // more powerfull class soon.
|
---|
266 | //
|
---|
267 | // The fit function is "gaus(0)+pol2(3)" which is equivalent to:
|
---|
268 | // [0]*exp(-0.5*((x-[1])/[2])^2) + [3] + [4]*x + [5]*x^2
|
---|
269 | // or
|
---|
270 | // A*exp(-0.5*((x-mu)/sigma)^2) + a + b*x + c*x^2
|
---|
271 | //
|
---|
272 | // Parameter [1] is fixed to 0 while the alpha peak should be
|
---|
273 | // symmetric around alpha=0.
|
---|
274 | //
|
---|
275 | // Parameter [4] is fixed to 0 because the first derivative at
|
---|
276 | // alpha=0 should be 0, too.
|
---|
277 | //
|
---|
278 | // In a first step the background is fitted between bgmin and bgmax,
|
---|
279 | // while the parameters [0]=0 and [2]=1 are fixed.
|
---|
280 | //
|
---|
281 | // In a second step the signal region (alpha<sigmax) is fittet using
|
---|
282 | // the whole function with parameters [1], [3], [4] and [5] fixed.
|
---|
283 | //
|
---|
284 | // The number of excess and background events are calculated as
|
---|
285 | // s = int(hist, 0, 1.25*sigint)
|
---|
286 | // b = int(pol2(3), 0, 1.25*sigint)
|
---|
287 | //
|
---|
288 | // The Significance is calculated using the Significance() member
|
---|
289 | // function.
|
---|
290 | //
|
---|
291 | Bool_t MAlphaFitter::Fit(TH1D &h, Bool_t paint)
|
---|
292 | {
|
---|
293 | Clear();
|
---|
294 |
|
---|
295 | // Check for the region which is not filled...
|
---|
296 | // if (alpha0==0)
|
---|
297 | // return kFALSE;
|
---|
298 |
|
---|
299 | // Perform fit to the off-data
|
---|
300 | if (!FitOff(h, paint))
|
---|
301 | return kFALSE;
|
---|
302 |
|
---|
303 | fFunc->ReleaseParameter(0); // It is also released by SetParLimits later on
|
---|
304 | //func.ReleaseParameter(1); // It is also released by SetParLimits later on
|
---|
305 | fFunc->ReleaseParameter(2);
|
---|
306 | for (int i=3; i<fFunc->GetNpar(); i++)
|
---|
307 | fFunc->FixParameter(i, fFunc->GetParameter(i));
|
---|
308 |
|
---|
309 |
|
---|
310 | // Do not allow signals smaller than the background
|
---|
311 | const Double_t alpha0 = h.GetBinContent(1);
|
---|
312 | const Double_t s = fSignalFunc==kGauss ? fFunc->GetParameter(3) : TMath::Exp(fFunc->GetParameter(3));
|
---|
313 | const Double_t A = alpha0-s;
|
---|
314 | //const Double_t dA = TMath::Abs(A);
|
---|
315 | //fFunc->SetParLimits(0, -dA*4, dA*4); // SetParLimits also releases the parameter
|
---|
316 | fFunc->SetParLimits(2, 0, 90); // SetParLimits also releases the parameter
|
---|
317 |
|
---|
318 | // Now fit a gaus in the on region on top of the polynom
|
---|
319 | fFunc->SetParameter(0, A);
|
---|
320 | fFunc->SetParameter(2, fSigMax*0.75);
|
---|
321 |
|
---|
322 | // options : N do not store the function, do not draw
|
---|
323 | // I use integral of function in bin rather than value at bin center
|
---|
324 | // R use the range specified in the function range
|
---|
325 | // Q quiet mode
|
---|
326 | // E Perform better Errors estimation using Minos technique
|
---|
327 | h.Fit(fFunc, "NQI", "", 0, fSigMax);
|
---|
328 |
|
---|
329 | fChiSqSignal = fFunc->GetChisquare()/fFunc->GetNDF();
|
---|
330 | fCoefficients.Set(fFunc->GetNpar(), fFunc->GetParameters());
|
---|
331 | for (int i=0; i<3; i++)
|
---|
332 | fErrors[i] = fFunc->GetParError(i);
|
---|
333 | //const Bool_t ok = NDF>0 && chi2<2.5*NDF;
|
---|
334 |
|
---|
335 | // ------------------------------------
|
---|
336 | if (paint)
|
---|
337 | {
|
---|
338 | fFunc->SetLineColor(kGreen);
|
---|
339 | fFunc->SetLineWidth(2);
|
---|
340 | fFunc->Paint("same");
|
---|
341 | }
|
---|
342 | // ------------------------------------
|
---|
343 |
|
---|
344 | //const Double_t s = fFunc->Integral(0, fSigInt)/alphaw;
|
---|
345 | fFunc->SetParameter(0, 0);
|
---|
346 | fFunc->SetParameter(2, 1);
|
---|
347 | //const Double_t b = fFunc->Integral(0, fSigInt)/alphaw;
|
---|
348 | //fSignificance = MMath::SignificanceLiMaSigned(s, b);
|
---|
349 |
|
---|
350 | // Calculate the fit result and set the corresponding data members
|
---|
351 | FitResult(h);
|
---|
352 |
|
---|
353 | return kTRUE;
|
---|
354 | }
|
---|
355 |
|
---|
356 | Double_t MAlphaFitter::DoOffFit(const TH1D &hon, const TH1D &hof, Bool_t paint)
|
---|
357 | {
|
---|
358 | if (fSignalFunc!=kThetaSq)
|
---|
359 | return 0;
|
---|
360 |
|
---|
361 | // ----------------------------------------------------------------------------
|
---|
362 |
|
---|
363 | const Int_t bin = hon.GetXaxis()->FindFixBin(fSigInt*0.999);
|
---|
364 |
|
---|
365 | MAlphaFitter fit(*this);
|
---|
366 | fit.EnableBackgroundFit();
|
---|
367 | fit.SetBackgroundFitMin(0);
|
---|
368 |
|
---|
369 | // produce a histogram containing the off-samples from on-source and
|
---|
370 | // off-source in the off-source region and the on-data in the source-region
|
---|
371 | TH1D h(hof);
|
---|
372 | h.SetDirectory(0);
|
---|
373 | h.Add(&hon);
|
---|
374 |
|
---|
375 | h.Scale(0.5);
|
---|
376 | for (int i=1; i<=bin+3; i++)
|
---|
377 | {
|
---|
378 | h.SetBinContent(i, hof.GetBinContent(i));
|
---|
379 | h.SetBinError( i, hof.GetBinError(i));
|
---|
380 | }
|
---|
381 |
|
---|
382 | // Now fit the off-data
|
---|
383 | if (!fit.FitOff(h, paint?2:0)) // FIXME: Show fit!
|
---|
384 | return -1;
|
---|
385 |
|
---|
386 | // Calculate fit-result
|
---|
387 | fit.FitResult(h);
|
---|
388 |
|
---|
389 | // Do a gaussian error propagation to calculated the error of
|
---|
390 | // the background estimated from the fit
|
---|
391 | const Double_t ea = fit.fErrors[3];
|
---|
392 | const Double_t eb = fit.fErrors[4];
|
---|
393 | const Double_t a = fit.fCoefficients[3];
|
---|
394 | const Double_t b = fit.fCoefficients[4];
|
---|
395 |
|
---|
396 | const Double_t t = fIntegralMax;
|
---|
397 |
|
---|
398 | const Double_t ex = TMath::Exp(t*b);
|
---|
399 | const Double_t eab = TMath::Exp(a)/b;
|
---|
400 |
|
---|
401 | const Double_t eA = ex-1;
|
---|
402 | const Double_t eB = t*ex - eA/b;
|
---|
403 |
|
---|
404 | const Double_t w = h.GetXaxis()->GetBinWidth(1);
|
---|
405 |
|
---|
406 | // Error of estimated background
|
---|
407 | const Double_t er = TMath::Abs(eab)*TMath::Hypot(eA*ea, eB*eb)/w;
|
---|
408 |
|
---|
409 | // Calculate arbitrary scale factor from propagated error from the
|
---|
410 | // condition: sqrt(alpha*background) = est.background/est.error
|
---|
411 | // const Double_t bg = hof.Integral(1, bin);
|
---|
412 | // const Double_t sc = bg * er*er / (fit2.GetEventsBackground()*fit2.GetEventsBackground());
|
---|
413 | // Assuming that bg and fit2.GetEventsBackground() are rather identical:
|
---|
414 | const Double_t sc = er*er / fit.fEventsBackground;
|
---|
415 |
|
---|
416 |
|
---|
417 | /*
|
---|
418 | cout << MMath::SignificanceLiMaSigned(hon.Integral(1, bin), fit.GetEventsBackground()/sc, sc) << " ";
|
---|
419 | cout << sc << " ";
|
---|
420 | cout << fit.fChiSqBg << endl;
|
---|
421 | */
|
---|
422 | return sc;
|
---|
423 | }
|
---|
424 |
|
---|
425 | Bool_t MAlphaFitter::Fit(const TH1D &hon, const TH1D &hof, Double_t alpha, Bool_t paint)
|
---|
426 | {
|
---|
427 | TH1D h(hon);
|
---|
428 | h.SetDirectory(0);
|
---|
429 | h.Add(&hof, -1); // substracts also number of entries!
|
---|
430 | h.SetEntries(hon.GetEntries());
|
---|
431 |
|
---|
432 | MAlphaFitter fit(*this);
|
---|
433 | fit.SetPolynomOrder(0);
|
---|
434 | if (alpha<=0 || !fit.Fit(h, paint))
|
---|
435 | return kFALSE;
|
---|
436 |
|
---|
437 | fChiSqSignal = fit.fChiSqSignal;
|
---|
438 | fChiSqBg = fit.fChiSqBg;
|
---|
439 | fCoefficients = fit.fCoefficients;
|
---|
440 | fErrors = fit.fErrors;
|
---|
441 |
|
---|
442 | // ----------------------------------------------------------------------------
|
---|
443 |
|
---|
444 | const Double_t scale = DoOffFit(hon, hof, paint);
|
---|
445 | if (scale<0)
|
---|
446 | return kFALSE;
|
---|
447 |
|
---|
448 | // ----------------------------------------------------------------------------
|
---|
449 |
|
---|
450 | const Int_t bin = hon.GetXaxis()->FindFixBin(fSigInt*0.999);
|
---|
451 |
|
---|
452 | fIntegralMax = hon.GetBinLowEdge(bin+1);
|
---|
453 | fEventsBackground = hof.Integral(1, bin);
|
---|
454 | fEventsSignal = hon.Integral(1, bin);
|
---|
455 | fEventsExcess = fEventsSignal-fEventsBackground;
|
---|
456 | fScaleFactor = alpha;
|
---|
457 | fSignificance = MMath::SignificanceLiMaSigned(fEventsSignal, fEventsBackground/alpha, alpha);
|
---|
458 | fErrorExcess = MMath::ErrorExc(fEventsSignal, fEventsBackground/alpha, alpha);
|
---|
459 |
|
---|
460 | // !Finitite includes IsNaN
|
---|
461 | if (!TMath::Finite(fSignificance))
|
---|
462 | fSignificance=0;
|
---|
463 |
|
---|
464 | return kTRUE;
|
---|
465 | }
|
---|
466 |
|
---|
467 | // --------------------------------------------------------------------------
|
---|
468 | //
|
---|
469 | // Calculate the upper limit for fEventsSignal number of observed events
|
---|
470 | // and fEventsBackground number of background events.
|
---|
471 | //
|
---|
472 | // Therefor TFeldmanCousin is used.
|
---|
473 | //
|
---|
474 | // The Feldman-Cousins method as described in PRD V57 #7, p3873-3889
|
---|
475 | //
|
---|
476 | Double_t MAlphaFitter::CalcUpperLimit() const
|
---|
477 | {
|
---|
478 | // get a FeldmanCousins calculation object with the default limits
|
---|
479 | // of calculating a 90% CL with the minimum signal value scanned
|
---|
480 | // = 0.0 and the maximum signal value scanned of 50.0
|
---|
481 | TFeldmanCousins f;
|
---|
482 | f.SetMuStep(0.05);
|
---|
483 | f.SetMuMax(100);
|
---|
484 | f.SetMuMin(0);
|
---|
485 | f.SetCL(90);
|
---|
486 |
|
---|
487 | return f.CalculateUpperLimit(fEventsSignal, fEventsBackground);
|
---|
488 | }
|
---|
489 |
|
---|
490 | void MAlphaFitter::PaintResult(Float_t x, Float_t y, Float_t size, Bool_t draw) const
|
---|
491 | {
|
---|
492 | const Double_t w = GetGausSigma();
|
---|
493 | const Double_t m = fIntegralMax;
|
---|
494 |
|
---|
495 | const Int_t l1 = w<=0 ? 0 : (Int_t)TMath::Ceil(-TMath::Log10(w));
|
---|
496 | const Int_t l2 = m<=0 ? 0 : (Int_t)TMath::Ceil(-TMath::Log10(m));
|
---|
497 | const TString fmt = MString::Format("\\sigma_{L/M}=%%.1f \\omega=%%.%df\\circ E=%%d B=%%d x<%%.%df \\tilde\\chi_{b}=%%.1f \\tilde\\chi_{s}=%%.1f c=%%.1f f=%%.2f",
|
---|
498 | l1<1?1:l1+1, l2<1?1:l2+1);
|
---|
499 | const TString txt = MString::Format(fmt.Data(), fSignificance, w, (int)fEventsExcess,
|
---|
500 | (int)fEventsBackground, m, fChiSqBg, fChiSqSignal,
|
---|
501 | fCoefficients[3], fScaleFactor);
|
---|
502 |
|
---|
503 | // This is totaly weired but the only way to get both options
|
---|
504 | // working with this nonsense implementation of TLatex
|
---|
505 | TLatex text(x, y, txt);
|
---|
506 | text.SetBit(TLatex::kTextNDC);
|
---|
507 | text.SetTextSize(size);
|
---|
508 | if (draw)
|
---|
509 | text.DrawLatex(x, y, txt);
|
---|
510 | else
|
---|
511 | text.Paint();
|
---|
512 |
|
---|
513 | TLine line;
|
---|
514 | line.SetLineColor(14);
|
---|
515 | if (draw)
|
---|
516 | line.DrawLine(m, gPad->GetUymin(), m, gPad->GetUymax());
|
---|
517 | else
|
---|
518 | line.PaintLine(m, gPad->GetUymin(), m, gPad->GetUymax());
|
---|
519 | }
|
---|
520 |
|
---|
521 | void MAlphaFitter::Copy(TObject &o) const
|
---|
522 | {
|
---|
523 | MAlphaFitter &f = static_cast<MAlphaFitter&>(o);
|
---|
524 |
|
---|
525 | // Setup
|
---|
526 | f.fSigInt = fSigInt;
|
---|
527 | f.fSigMax = fSigMax;
|
---|
528 | f.fBgMin = fBgMin;
|
---|
529 | f.fBgMax = fBgMax;
|
---|
530 | f.fScaleMin = fScaleMin;
|
---|
531 | f.fScaleMax = fScaleMax;
|
---|
532 | f.fPolynomOrder = fPolynomOrder;
|
---|
533 | f.fFitBackground= fFitBackground;
|
---|
534 | f.fSignalFunc = fSignalFunc;
|
---|
535 | f.fScaleMode = fScaleMode;
|
---|
536 | f.fScaleUser = fScaleUser;
|
---|
537 | f.fStrategy = fStrategy;
|
---|
538 | f.fCoefficients.Set(fCoefficients.GetSize());
|
---|
539 | f.fCoefficients.Reset();
|
---|
540 | f.fErrors.Set(fCoefficients.GetSize());
|
---|
541 | f.fErrors.Reset();
|
---|
542 |
|
---|
543 | // Result
|
---|
544 | f.fSignificance = fSignificance;
|
---|
545 | f.fErrorExcess = fErrorExcess;
|
---|
546 | f.fEventsExcess = fEventsExcess;
|
---|
547 | f.fEventsSignal = fEventsSignal;
|
---|
548 | f.fEventsBackground = fEventsBackground;
|
---|
549 | f.fChiSqSignal = fChiSqSignal;
|
---|
550 | f.fChiSqBg = fChiSqBg;
|
---|
551 | f.fIntegralMax = fIntegralMax;
|
---|
552 | f.fScaleFactor = fScaleFactor;
|
---|
553 |
|
---|
554 | // Function
|
---|
555 | delete f.fFunc;
|
---|
556 |
|
---|
557 | f.fFunc = new TF1(*fFunc);
|
---|
558 | f.fFunc->SetName("Dummy");
|
---|
559 | gROOT->GetListOfFunctions()->Remove(f.fFunc);
|
---|
560 | }
|
---|
561 |
|
---|
562 | void MAlphaFitter::Print(Option_t *o) const
|
---|
563 | {
|
---|
564 | *fLog << GetDescriptor() << ": Fitting..." << endl;
|
---|
565 | *fLog << " ...signal to " << fSigMax << " (integrate into bin at " << fSigInt << ")" << endl;
|
---|
566 | *fLog << " ...signal function: ";
|
---|
567 | switch (fSignalFunc)
|
---|
568 | {
|
---|
569 | case kGauss: *fLog << "gauss(x)/pol" << fPolynomOrder; break;
|
---|
570 | case kThetaSq: *fLog << "gauss(sqrt(x))/expo"; break;
|
---|
571 | }
|
---|
572 | *fLog << endl;
|
---|
573 | if (!fFitBackground)
|
---|
574 | *fLog << " ...no background." << endl;
|
---|
575 | else
|
---|
576 | {
|
---|
577 | *fLog << " ...background from " << fBgMin << " to " << fBgMax << endl;
|
---|
578 | *fLog << " ...polynom order " << fPolynomOrder << endl;
|
---|
579 | *fLog << " ...scale mode: ";
|
---|
580 | switch (fScaleMode)
|
---|
581 | {
|
---|
582 | case kNone: *fLog << "none."; break;
|
---|
583 | case kEntries: *fLog << "entries."; break;
|
---|
584 | case kIntegral: *fLog << "integral."; break;
|
---|
585 | case kOffRegion: *fLog << "off region (integral between " << fScaleMin << " and " << fScaleMax << ")"; break;
|
---|
586 | case kBackground: *fLog << "background (integral between " << fBgMin << " and " << fBgMax << ")"; break;
|
---|
587 | case kLeastSquare: *fLog << "least square (N/A)"; break;
|
---|
588 | case kUserScale: *fLog << "user def (" << fScaleUser << ")"; break;
|
---|
589 | }
|
---|
590 | *fLog << endl;
|
---|
591 | }
|
---|
592 |
|
---|
593 | if (TString(o).Contains("result"))
|
---|
594 | {
|
---|
595 | *fLog << "Result:" << endl;
|
---|
596 | *fLog << " - Significance (Li/Ma) " << fSignificance << endl;
|
---|
597 | *fLog << " - Excess Events " << fEventsExcess << endl;
|
---|
598 | *fLog << " - Signal Events " << fEventsSignal << endl;
|
---|
599 | *fLog << " - Background Events " << fEventsBackground << endl;
|
---|
600 | *fLog << " - Chi^2/ndf (Signal) " << fChiSqSignal << endl;
|
---|
601 | *fLog << " - Chi^2/ndf (Background) " << fChiSqBg << endl;
|
---|
602 | *fLog << " - Signal integrated up to " << fIntegralMax << "°" << endl;
|
---|
603 | *fLog << " - Scale Factor (Off) " << fScaleFactor << endl;
|
---|
604 | }
|
---|
605 | }
|
---|
606 |
|
---|
607 | Bool_t MAlphaFitter::FitEnergy(const TH3D &hon, UInt_t bin, Bool_t paint)
|
---|
608 | {
|
---|
609 | const TString name(MString::Format("TempAlphaEnergy%06d", gRandom->Integer(1000000)));
|
---|
610 |
|
---|
611 | TH1D *h = hon.ProjectionZ(name, 0, hon.GetNbinsX()+1, bin, bin, "E");
|
---|
612 | h->SetDirectory(0);
|
---|
613 |
|
---|
614 | const Bool_t rc = Fit(*h, paint);
|
---|
615 |
|
---|
616 | delete h;
|
---|
617 |
|
---|
618 | return rc;
|
---|
619 | }
|
---|
620 |
|
---|
621 | Bool_t MAlphaFitter::FitTheta(const TH3D &hon, UInt_t bin, Bool_t paint)
|
---|
622 | {
|
---|
623 | const TString name(MString::Format("TempAlphaTheta%06d", gRandom->Integer(1000000)));
|
---|
624 |
|
---|
625 | TH1D *h = hon.ProjectionZ(name, bin, bin, 0, hon.GetNbinsY()+1, "E");
|
---|
626 | h->SetDirectory(0);
|
---|
627 |
|
---|
628 | const Bool_t rc = Fit(*h, paint);
|
---|
629 |
|
---|
630 | delete h;
|
---|
631 |
|
---|
632 | return rc;
|
---|
633 | }
|
---|
634 | /*
|
---|
635 | Bool_t MAlphaFitter::FitTime(const TH3D &hon, UInt_t bin, Bool_t paint)
|
---|
636 | {
|
---|
637 | const TString name(Form("TempAlphaTime%06d", gRandom->Integer(1000000)));
|
---|
638 |
|
---|
639 | hon.GetZaxis()->SetRange(bin,bin);
|
---|
640 | TH1D *h = (TH1D*)hon.Project3D("ye");
|
---|
641 | hon.GetZaxis()->SetRange(-1,-1);
|
---|
642 |
|
---|
643 | h->SetDirectory(0);
|
---|
644 |
|
---|
645 | const Bool_t rc = Fit(*h, paint);
|
---|
646 | delete h;
|
---|
647 | return rc;
|
---|
648 | }
|
---|
649 | */
|
---|
650 | Bool_t MAlphaFitter::FitAlpha(const TH3D &hon, Bool_t paint)
|
---|
651 | {
|
---|
652 | const TString name(MString::Format("TempAlpha%06d", gRandom->Integer(1000000)));
|
---|
653 |
|
---|
654 | TH1D *h = hon.ProjectionZ(name, 0, hon.GetNbinsX()+1, 0, hon.GetNbinsY()+1, "E");
|
---|
655 | h->SetDirectory(0);
|
---|
656 |
|
---|
657 | const Bool_t rc = Fit(*h, paint);
|
---|
658 |
|
---|
659 | delete h;
|
---|
660 |
|
---|
661 | return rc;
|
---|
662 | }
|
---|
663 |
|
---|
664 | Bool_t MAlphaFitter::FitEnergy(const TH3D &hon, const TH3D &hof, UInt_t bin, Bool_t paint)
|
---|
665 | {
|
---|
666 | const TString name1(MString::Format("TempAlpha%06d_on", gRandom->Integer(1000000)));
|
---|
667 | const TString name0(MString::Format("TempAlpha%06d_off", gRandom->Integer(1000000)));
|
---|
668 |
|
---|
669 | TH1D *h1 = hon.ProjectionZ(name1, 0, hon.GetNbinsX()+1, bin, bin, "E");
|
---|
670 | h1->SetDirectory(0);
|
---|
671 |
|
---|
672 | TH1D *h0 = hof.ProjectionZ(name0, 0, hof.GetNbinsX()+1, bin, bin, "E");
|
---|
673 | h0->SetDirectory(0);
|
---|
674 |
|
---|
675 | const Bool_t rc = ScaleAndFit(*h1, h0, paint);
|
---|
676 |
|
---|
677 | delete h0;
|
---|
678 | delete h1;
|
---|
679 |
|
---|
680 | return rc;
|
---|
681 | }
|
---|
682 |
|
---|
683 | Bool_t MAlphaFitter::FitTheta(const TH3D &hon, const TH3D &hof, UInt_t bin, Bool_t paint)
|
---|
684 | {
|
---|
685 | const TString name1(MString::Format("TempAlpha%06d_on", gRandom->Integer(1000000)));
|
---|
686 | const TString name0(MString::Format("TempAlpha%06d_off", gRandom->Integer(1000000)));
|
---|
687 |
|
---|
688 | TH1D *h1 = hon.ProjectionZ(name1, bin, bin, 0, hon.GetNbinsY()+1, "E");
|
---|
689 | h1->SetDirectory(0);
|
---|
690 |
|
---|
691 | TH1D *h0 = hof.ProjectionZ(name0, bin, bin, 0, hof.GetNbinsY()+1, "E");
|
---|
692 | h0->SetDirectory(0);
|
---|
693 |
|
---|
694 | const Bool_t rc = ScaleAndFit(*h1, h0, paint);
|
---|
695 |
|
---|
696 | delete h0;
|
---|
697 | delete h1;
|
---|
698 |
|
---|
699 | return rc;
|
---|
700 | }
|
---|
701 | /*
|
---|
702 | Bool_t MAlphaFitter::FitTime(const TH3D &hon, const TH3D &hof, UInt_t bin, Bool_t paint)
|
---|
703 | {
|
---|
704 | const TString name1(Form("TempAlphaTime%06d_on", gRandom->Integer(1000000)));
|
---|
705 | const TString name0(Form("TempAlphaTime%06d_off", gRandom->Integer(1000000)));
|
---|
706 |
|
---|
707 | hon.GetZaxis()->SetRange(bin,bin);
|
---|
708 | TH1D *h1 = (TH1D*)hon.Project3D("ye");
|
---|
709 | hon.GetZaxis()->SetRange(-1,-1);
|
---|
710 | h1->SetDirectory(0);
|
---|
711 |
|
---|
712 | hof.GetZaxis()->SetRange(bin,bin);
|
---|
713 | TH1D *h0 = (TH1D*)hof.Project3D("ye");
|
---|
714 | hof.GetZaxis()->SetRange(-1,-1);
|
---|
715 | h0->SetDirectory(0);
|
---|
716 |
|
---|
717 | const Bool_t rc = ScaleAndFit(*h1, h0, paint);
|
---|
718 |
|
---|
719 | delete h0;
|
---|
720 | delete h1;
|
---|
721 |
|
---|
722 | return rc;
|
---|
723 | }
|
---|
724 | */
|
---|
725 |
|
---|
726 | Bool_t MAlphaFitter::FitAlpha(const TH3D &hon, const TH3D &hof, Bool_t paint)
|
---|
727 | {
|
---|
728 | const TString name1(MString::Format("TempAlpha%06d_on", gRandom->Integer(1000000)));
|
---|
729 | const TString name0(MString::Format("TempAlpha%06d_off", gRandom->Integer(1000000)));
|
---|
730 |
|
---|
731 | TH1D *h1 = hon.ProjectionZ(name1, 0, hon.GetNbinsX()+1, 0, hon.GetNbinsY()+1, "E");
|
---|
732 | h1->SetDirectory(0);
|
---|
733 |
|
---|
734 | TH1D *h0 = hof.ProjectionZ(name0, 0, hof.GetNbinsX()+1, 0, hof.GetNbinsY()+1, "E");
|
---|
735 | h0->SetDirectory(0);
|
---|
736 |
|
---|
737 | const Bool_t rc = ScaleAndFit(*h1, h0, paint);
|
---|
738 |
|
---|
739 | delete h0;
|
---|
740 | delete h1;
|
---|
741 |
|
---|
742 | return rc;
|
---|
743 | }
|
---|
744 |
|
---|
745 | Bool_t MAlphaFitter::ApplyScaling(const TH3D &hon, TH3D &hof, UInt_t bin) const
|
---|
746 | {
|
---|
747 | const TString name1(MString::Format("TempAlpha%06d_on", gRandom->Integer(1000000)));
|
---|
748 | const TString name0(MString::Format("TempAlpha%06d_off", gRandom->Integer(1000000)));
|
---|
749 |
|
---|
750 | TH1D *h1 = hon.ProjectionZ(name1, 0, hon.GetNbinsX()+1, bin, bin, "E");
|
---|
751 | h1->SetDirectory(0);
|
---|
752 |
|
---|
753 | TH1D *h0 = hof.ProjectionZ(name0, 0, hof.GetNbinsX()+1, bin, bin, "E");
|
---|
754 | h0->SetDirectory(0);
|
---|
755 |
|
---|
756 | const Double_t scale = Scale(*h0, *h1);
|
---|
757 |
|
---|
758 | delete h0;
|
---|
759 | delete h1;
|
---|
760 |
|
---|
761 | for (int x=0; x<=hof.GetNbinsX()+1; x++)
|
---|
762 | for (int z=0; z<=hof.GetNbinsZ()+1; z++)
|
---|
763 | {
|
---|
764 | hof.SetBinContent(x, bin, z, hof.GetBinContent(x, bin, z)*scale);
|
---|
765 | hof.SetBinError( x, bin, z, hof.GetBinError( x, bin, z)*scale);
|
---|
766 | }
|
---|
767 |
|
---|
768 | return scale>0;
|
---|
769 | }
|
---|
770 |
|
---|
771 | Bool_t MAlphaFitter::ApplyScaling(const TH3D &hon, TH3D &hof) const
|
---|
772 | {
|
---|
773 | for (int y=0; y<=hof.GetNbinsY()+1; y++)
|
---|
774 | ApplyScaling(hon, hof, y);
|
---|
775 |
|
---|
776 | return kTRUE;
|
---|
777 | }
|
---|
778 |
|
---|
779 | Double_t MAlphaFitter::Scale(TH1D &of, const TH1D &on) const
|
---|
780 | {
|
---|
781 | Float_t scaleon = 1;
|
---|
782 | Float_t scaleof = 1;
|
---|
783 | switch (fScaleMode)
|
---|
784 | {
|
---|
785 | case kNone:
|
---|
786 | return 1;
|
---|
787 |
|
---|
788 | case kEntries:
|
---|
789 | scaleon = on.GetEntries();
|
---|
790 | scaleof = of.GetEntries();
|
---|
791 | break;
|
---|
792 |
|
---|
793 | case kIntegral:
|
---|
794 | scaleon = on.Integral();
|
---|
795 | scaleof = of.Integral();
|
---|
796 | break;
|
---|
797 |
|
---|
798 | case kOffRegion:
|
---|
799 | {
|
---|
800 | const Int_t min = on.GetXaxis()->FindFixBin(fScaleMin);
|
---|
801 | const Int_t max = on.GetXaxis()->FindFixBin(fScaleMax);
|
---|
802 | scaleon = on.Integral(min, max);
|
---|
803 | scaleof = of.Integral(min, max);
|
---|
804 | }
|
---|
805 | break;
|
---|
806 |
|
---|
807 | case kBackground:
|
---|
808 | {
|
---|
809 | const Int_t min = on.GetXaxis()->FindFixBin(fBgMin);
|
---|
810 | const Int_t max = on.GetXaxis()->FindFixBin(fBgMax);
|
---|
811 | scaleon = on.Integral(min, max);
|
---|
812 | scaleof = of.Integral(min, max);
|
---|
813 | }
|
---|
814 | break;
|
---|
815 |
|
---|
816 | case kUserScale:
|
---|
817 | scaleon = fScaleUser;
|
---|
818 | break;
|
---|
819 |
|
---|
820 | // This is just to make some compiler happy
|
---|
821 | default:
|
---|
822 | return 1;
|
---|
823 | }
|
---|
824 |
|
---|
825 | if (scaleof!=0)
|
---|
826 | {
|
---|
827 | of.Scale(scaleon/scaleof);
|
---|
828 | return scaleon/scaleof;
|
---|
829 | }
|
---|
830 | else
|
---|
831 | {
|
---|
832 | of.Reset();
|
---|
833 | return 0;
|
---|
834 | }
|
---|
835 | }
|
---|
836 |
|
---|
837 | Double_t MAlphaFitter::GetMinimizationValue() const
|
---|
838 | {
|
---|
839 | switch (fStrategy)
|
---|
840 | {
|
---|
841 | case kSignificance:
|
---|
842 | return -GetSignificance();
|
---|
843 | case kSignificanceChi2:
|
---|
844 | return -GetSignificance()/GetChiSqSignal();
|
---|
845 | case kSignificanceLogExcess:
|
---|
846 | if (GetEventsExcess()<1)
|
---|
847 | return 0;
|
---|
848 | return -GetSignificance()*TMath::Log10(GetEventsExcess());
|
---|
849 | case kSignificanceExcess:
|
---|
850 | return -GetSignificance()*GetEventsExcess();
|
---|
851 | case kExcess:
|
---|
852 | return -GetEventsExcess();
|
---|
853 | case kGaussSigma:
|
---|
854 | return GetGausSigma();
|
---|
855 | case kWeakSource:
|
---|
856 | return GetEventsBackground()<1 ? -GetEventsExcess() : -GetEventsExcess()/TMath::Sqrt(GetEventsBackground());
|
---|
857 | case kWeakSourceLogExcess:
|
---|
858 | return GetEventsBackground()<1 ? -GetEventsExcess() : -GetEventsExcess()/TMath::Sqrt(GetEventsBackground())*TMath::Log10(GetEventsExcess());
|
---|
859 | }
|
---|
860 | return 0;
|
---|
861 | }
|
---|
862 |
|
---|
863 | Int_t MAlphaFitter::ReadEnv(const TEnv &env, TString prefix, Bool_t print)
|
---|
864 | {
|
---|
865 | Bool_t rc = kFALSE;
|
---|
866 |
|
---|
867 | //void SetScaleUser(Float_t scale) { fScaleUser = scale; fScaleMode=kUserScale; }
|
---|
868 | //void SetScaleMode(ScaleMode_t mode) { fScaleMode = mode; }
|
---|
869 |
|
---|
870 | if (IsEnvDefined(env, prefix, "SignalIntegralMax", print))
|
---|
871 | {
|
---|
872 | SetSignalIntegralMax(GetEnvValue(env, prefix, "SignalIntegralMax", fSigInt));
|
---|
873 | rc = kTRUE;
|
---|
874 | }
|
---|
875 | if (IsEnvDefined(env, prefix, "SignalFitMax", print))
|
---|
876 | {
|
---|
877 | SetSignalIntegralMax(GetEnvValue(env, prefix, "SignalFitMax", fSigMax));
|
---|
878 | rc = kTRUE;
|
---|
879 | }
|
---|
880 | if (IsEnvDefined(env, prefix, "BackgroundFitMax", print))
|
---|
881 | {
|
---|
882 | SetBackgroundFitMax(GetEnvValue(env, prefix, "BackgroundFitMax", fBgMax));
|
---|
883 | rc = kTRUE;
|
---|
884 | }
|
---|
885 | if (IsEnvDefined(env, prefix, "BackgroundFitMin", print))
|
---|
886 | {
|
---|
887 | SetBackgroundFitMin(GetEnvValue(env, prefix, "BackgroundFitMin", fBgMin));
|
---|
888 | rc = kTRUE;
|
---|
889 | }
|
---|
890 | if (IsEnvDefined(env, prefix, "ScaleMin", print))
|
---|
891 | {
|
---|
892 | SetScaleMin(GetEnvValue(env, prefix, "ScaleMin", fScaleMin));
|
---|
893 | rc = kTRUE;
|
---|
894 | }
|
---|
895 | if (IsEnvDefined(env, prefix, "ScaleMax", print))
|
---|
896 | {
|
---|
897 | SetScaleMax(GetEnvValue(env, prefix, "ScaleMax", fScaleMax));
|
---|
898 | rc = kTRUE;
|
---|
899 | }
|
---|
900 | if (IsEnvDefined(env, prefix, "PolynomOrder", print))
|
---|
901 | {
|
---|
902 | SetPolynomOrder(GetEnvValue(env, prefix, "PolynomOrder", fPolynomOrder));
|
---|
903 | rc = kTRUE;
|
---|
904 | }
|
---|
905 |
|
---|
906 | if (IsEnvDefined(env, prefix, "MinimizationStrategy", print))
|
---|
907 | {
|
---|
908 | TString txt = GetEnvValue(env, prefix, "MinimizationStrategy", "");
|
---|
909 | txt = txt.Strip(TString::kBoth);
|
---|
910 | txt.ToLower();
|
---|
911 | if (txt==(TString)"significance")
|
---|
912 | fStrategy = kSignificance;
|
---|
913 | if (txt==(TString)"significancechi2")
|
---|
914 | fStrategy = kSignificanceChi2;
|
---|
915 | if (txt==(TString)"significanceexcess")
|
---|
916 | fStrategy = kSignificanceExcess;
|
---|
917 | if (txt==(TString)"excess")
|
---|
918 | fStrategy = kExcess;
|
---|
919 | if (txt==(TString)"gausssigma" || txt==(TString)"gaussigma")
|
---|
920 | fStrategy = kGaussSigma;
|
---|
921 | if (txt==(TString)"weaksource")
|
---|
922 | fStrategy = kWeakSource;
|
---|
923 | rc = kTRUE;
|
---|
924 | }
|
---|
925 | if (IsEnvDefined(env, prefix, "Scale", print))
|
---|
926 | {
|
---|
927 | fScaleUser = GetEnvValue(env, prefix, "Scale", fScaleUser);
|
---|
928 | rc = kTRUE;
|
---|
929 | }
|
---|
930 | if (IsEnvDefined(env, prefix, "ScaleMode", print))
|
---|
931 | {
|
---|
932 | TString txt = GetEnvValue(env, prefix, "ScaleMode", "");
|
---|
933 | txt = txt.Strip(TString::kBoth);
|
---|
934 | txt.ToLower();
|
---|
935 | if (txt==(TString)"none")
|
---|
936 | fScaleMode = kNone;
|
---|
937 | if (txt==(TString)"entries")
|
---|
938 | fScaleMode = kEntries;
|
---|
939 | if (txt==(TString)"integral")
|
---|
940 | fScaleMode = kIntegral;
|
---|
941 | if (txt==(TString)"offregion")
|
---|
942 | fScaleMode = kOffRegion;
|
---|
943 | if (txt==(TString)"background")
|
---|
944 | fScaleMode = kBackground;
|
---|
945 | if (txt==(TString)"leastsquare")
|
---|
946 | fScaleMode = kLeastSquare;
|
---|
947 | if (txt==(TString)"userscale")
|
---|
948 | fScaleMode = kUserScale;
|
---|
949 | if (txt==(TString)"fixed")
|
---|
950 | FixScale();
|
---|
951 | rc = kTRUE;
|
---|
952 | }
|
---|
953 | if (IsEnvDefined(env, prefix, "SignalFunction", print))
|
---|
954 | {
|
---|
955 | TString txt = GetEnvValue(env, prefix, "SignalFunction", "");
|
---|
956 | txt = txt.Strip(TString::kBoth);
|
---|
957 | txt.ToLower();
|
---|
958 | if (txt==(TString)"gauss" || txt==(TString)"gaus")
|
---|
959 | SetSignalFunction(kGauss);
|
---|
960 | if (txt==(TString)"thetasq")
|
---|
961 | SetSignalFunction(kThetaSq);
|
---|
962 | rc = kTRUE;
|
---|
963 | }
|
---|
964 |
|
---|
965 | return rc;
|
---|
966 | }
|
---|