Logo ROOT  
Reference Guide
Fitter.cxx
Go to the documentation of this file.
1// @(#)root/mathcore:$Id$
2// Author: L. Moneta Mon Sep 4 17:00:10 2006
3
4/**********************************************************************
5 * *
6 * Copyright (c) 2006 LCG ROOT Math Team, CERN/PH-SFT *
7 * *
8 * *
9 **********************************************************************/
10
11// Implementation file for class Fitter
12
13
14#include "Fit/Fitter.h"
15#include "Fit/Chi2FCN.h"
18#include "Math/Minimizer.h"
21#include "Fit/BasicFCN.h"
22#include "Fit/BinData.h"
23#include "Fit/UnBinData.h"
24#include "Fit/FcnAdapter.h"
25#include "Fit/FitConfig.h"
26#include "Fit/FitResult.h"
27#include "Math/Error.h"
28
29#include <memory>
30
31#include "Math/IParamFunction.h"
32
34
35// #include "TMatrixDSym.h"
36// for debugging
37//#include "TMatrixD.h"
38// #include <iomanip>
39
40namespace ROOT {
41
42 namespace Fit {
43
44// use a static variable to get default minimizer options for error def
45// to see if user has changed it later on. If it has not been changed we set
46// for the likelihood method an error def of 0.5
47// t.b.d : multiply likelihood by 2 so have same error def definition as chi2
49
50
52 fUseGradient(false),
53 fBinFit(false),
54 fFitType(0),
55 fDataSize(0)
56{}
57
58Fitter::Fitter(const std::shared_ptr<FitResult> & result) :
59 fUseGradient(false),
60 fBinFit(false),
61 fFitType(0),
62 fDataSize(0),
63 fResult(result)
64{
65 if (result->fFitFunc) SetFunction(*fResult->fFitFunc); // this will create also the configuration
66 if (result->fObjFunc) fObjFunction = fResult->fObjFunc;
67 if (result->fFitData) fData = fResult->fFitData;
68}
69
71{
72 // Destructor implementation.
73
74 // nothing to do since we use shared_ptr now
75}
76
78{
79 // Implementation of copy constructor.
80 // copy FitResult, FitConfig and clone fit function
81 (*this) = rhs;
82}
83
85{
86 // Implementation of assignment operator.
87 // dummy implementation, since it is private
88 if (this == &rhs) return *this; // time saving self-test
89// fUseGradient = rhs.fUseGradient;
90// fBinFit = rhs.fBinFit;
91// fResult = rhs.fResult;
92// fConfig = rhs.fConfig;
93// // function is copied and managed by FitResult (maybe should use an unique_ptr)
94// fFunc = fResult.ModelFunction();
95// if (rhs.fFunc != 0 && fResult.ModelFunction() == 0) { // case no fit has been done yet - then clone
96// if (fFunc) delete fFunc;
97// fFunc = dynamic_cast<IModelFunction *>( (rhs.fFunc)->Clone() );
98// assert(fFunc != 0);
99// }
100 return *this;
101}
102
103void Fitter::SetFunction(const IModelFunction & func, bool useGradient)
104{
105
106 fUseGradient = useGradient;
107 if (fUseGradient) {
108 const IGradModelFunction * gradFunc = dynamic_cast<const IGradModelFunction*>(&func);
109 if (gradFunc) {
110 SetFunction(*gradFunc, true);
111 return;
112 }
113 else {
114 MATH_WARN_MSG("Fitter::SetFunction","Requested function does not provide gradient - use it as non-gradient function ");
115 }
116 }
117 fUseGradient = false;
118
119 // set the fit model function (clone the given one and keep a copy )
120 //std::cout << "set a non-grad function" << std::endl;
121
122 fFunc = std::shared_ptr<IModelFunction>(dynamic_cast<IModelFunction *>(func.Clone() ) );
123 assert(fFunc);
124
125 // creates the parameter settings
127 fFunc_v.reset();
128}
129
130void Fitter::SetFunction(const IModel1DFunction & func, bool useGradient)
131{
132 fUseGradient = useGradient;
133 if (fUseGradient) {
134 const IGradModel1DFunction * gradFunc = dynamic_cast<const IGradModel1DFunction*>(&func);
135 if (gradFunc) {
136 SetFunction(*gradFunc, true);
137 return;
138 }
139 else {
140 MATH_WARN_MSG("Fitter::SetFunction","Requested function does not provide gradient - use it as non-gradient function ");
141 }
142 }
143 fUseGradient = false;
144 //std::cout << "set a 1d function" << std::endl;
145
146 // function is cloned when creating the adapter
147 fFunc = std::shared_ptr<IModelFunction>(new ROOT::Math::MultiDimParamFunctionAdapter(func));
148
149 // creates the parameter settings
151 fFunc_v.reset();
152}
153
154void Fitter::SetFunction(const IGradModelFunction & func, bool useGradient)
155{
156 fUseGradient = useGradient;
157 //std::cout << "set a grad function" << std::endl;
158 // set the fit model function (clone the given one and keep a copy )
159 fFunc = std::shared_ptr<IModelFunction>( dynamic_cast<IGradModelFunction *> ( func.Clone() ) );
160 assert(fFunc);
161
162 // creates the parameter settings
164 fFunc_v.reset();
165}
166
167
168void Fitter::SetFunction(const IGradModel1DFunction & func, bool useGradient)
169{
170 //std::cout << "set a 1d grad function" << std::endl;
171 fUseGradient = useGradient;
172 // function is cloned when creating the adapter
173 fFunc = std::shared_ptr<IModelFunction>(new ROOT::Math::MultiDimParamGradFunctionAdapter(func));
174
175 // creates the parameter settings
177 fFunc_v.reset();
178}
179
180
181bool Fitter::SetFCN(const ROOT::Math::IMultiGenFunction & fcn, const double * params, unsigned int dataSize, bool chi2fit) {
182 // set the objective function for the fit
183 // if params is not NULL create the parameter settings
184 fUseGradient = false;
185 unsigned int npar = fcn.NDim();
186 if (npar == 0) {
187 MATH_ERROR_MSG("Fitter::SetFCN","FCN function has zero parameters ");
188 return false;
189 }
190 if (params != 0 )
191 fConfig.SetParamsSettings(npar, params);
192 else {
193 if ( fConfig.ParamsSettings().size() != npar) {
194 MATH_ERROR_MSG("Fitter::SetFCN","wrong fit parameter settings");
195 return false;
196 }
197 }
198
199 fBinFit = chi2fit;
200 fDataSize = dataSize;
201
202 // keep also a copy of FCN function and set this in minimizer so they will be managed together
203 // (remember that cloned copy will still depends on data and model function pointers)
204 fObjFunction = std::unique_ptr<ROOT::Math::IMultiGenFunction> ( fcn.Clone() );
205
206 // in case a model function and data exists from a previous fit - reset shared-ptr
207 if (fResult && fResult->FittedFunction() == 0 && fFunc) fFunc.reset();
208 if (fData) fData.reset();
209
210 return true;
211}
212
213bool Fitter::SetFCN(const ROOT::Math::IMultiGenFunction &fcn, const IModelFunction & func, const double *params, unsigned int dataSize, bool chi2fit) {
214 // set the objective function for the fit and a model function
215 if (!SetFCN(fcn, params, dataSize, chi2fit) ) return false;
216 // need to set fFunc afterwards because SetFCN could reset fFUnc
217 fFunc = std::shared_ptr<IModelFunction>(dynamic_cast<IModelFunction *>(func.Clone()));
218 return (fFunc != nullptr);
219}
220
221bool Fitter::SetFCN(const ROOT::Math::IMultiGradFunction &fcn, const double *params, unsigned int dataSize,
222 bool chi2fit)
223{
224 // set the objective function for the fit
225 // if params is not NULL create the parameter settings
226 if (!SetFCN(static_cast<const ROOT::Math::IMultiGenFunction &>(fcn), params, dataSize, chi2fit))
227 return false;
228 fUseGradient = true;
229 return true;
230}
231
232bool Fitter::SetFCN(const ROOT::Math::IMultiGradFunction &fcn, const IModelFunction &func, const double *params,
233 unsigned int dataSize, bool chi2fit)
234{
235 // set the objective function for the fit and a model function
236 if (!SetFCN(fcn, params, dataSize, chi2fit) ) return false;
237 fFunc = std::shared_ptr<IModelFunction>(dynamic_cast<IModelFunction *>(func.Clone()));
238 return (fFunc != nullptr);
239}
240
241bool Fitter::SetFCN(const ROOT::Math::FitMethodFunction &fcn, const double *params)
242{
243 // set the objective function for the fit
244 // if params is not NULL create the parameter settings
245 bool chi2fit = (fcn.Type() == ROOT::Math::FitMethodFunction::kLeastSquare);
246 if (!SetFCN(fcn, params, fcn.NPoints(), chi2fit))
247 return false;
248 fUseGradient = false;
249 fFitType = fcn.Type();
250 return true;
251}
252
253bool Fitter::SetFCN(const ROOT::Math::FitMethodGradFunction &fcn, const double *params)
254{
255 // set the objective function for the fit
256 // if params is not NULL create the parameter settings
258 if (!SetFCN(fcn, params, fcn.NPoints(), chi2fit))
259 return false;
260 fUseGradient = true;
261 fFitType = fcn.Type();
262 return true;
263}
264
265bool Fitter::FitFCN(const BaseFunc &fcn, const double *params, unsigned int dataSize, bool chi2fit)
266{
267 // fit a user provided FCN function
268 // create fit parameter settings
269 if (!SetFCN(fcn, params, dataSize, chi2fit))
270 return false;
271 return FitFCN();
272}
273
274bool Fitter::FitFCN(const BaseGradFunc &fcn, const double *params, unsigned int dataSize, bool chi2fit)
275{
276 // fit a user provided FCN gradient function
277
278 if (!SetFCN(fcn, params, dataSize, chi2fit))
279 return false;
280 return FitFCN();
281}
282
283bool Fitter::FitFCN(const ROOT::Math::FitMethodFunction &fcn, const double *params)
284{
285 // fit using the passed objective function for the fit
286 if (!SetFCN(fcn, params))
287 return false;
288 return FitFCN();
289}
290
291bool Fitter::FitFCN(const ROOT::Math::FitMethodGradFunction &fcn, const double *params)
292{
293 // fit using the passed objective function for the fit
294 if (!SetFCN(fcn, params))
295 return false;
296 return FitFCN();
297}
298
299bool Fitter::SetFCN(MinuitFCN_t fcn, int npar, const double *params, unsigned int dataSize, bool chi2fit)
300{
301 // set TMinuit style FCN type (global function pointer)
302 // create corresponfing objective function from that function
303
304 if (npar == 0) {
305 npar = fConfig.ParamsSettings().size();
306 if (npar == 0) {
307 MATH_ERROR_MSG("Fitter::FitFCN", "Fit Parameter settings have not been created ");
308 return false;
309 }
310 }
311
312 ROOT::Fit::FcnAdapter newFcn(fcn, npar);
313 return SetFCN(newFcn, params, dataSize, chi2fit);
314}
315
316bool Fitter::FitFCN(MinuitFCN_t fcn, int npar, const double *params, unsigned int dataSize, bool chi2fit)
317{
318 // fit using Minuit style FCN type (global function pointer)
319 // create corresponfing objective function from that function
320 if (!SetFCN(fcn, npar, params, dataSize, chi2fit))
321 return false;
322 fUseGradient = false;
323 return FitFCN();
324}
325
327{
328 // fit using the previously set FCN function
329
330 if (!fObjFunction) {
331 MATH_ERROR_MSG("Fitter::FitFCN", "Objective function has not been set");
332 return false;
333 }
334 // look if FCN s of a known type and we can get some modelfunction and data objects
335 if (!fFunc || !fData)
336 ExamineFCN();
337 // init the minimizer
338 if (!DoInitMinimizer())
339 return false;
340 // perform the minimization
341 return DoMinimization();
342}
343
345{
346 // evaluate the FCN using the stored values in fConfig
347
348 if (fFunc && fResult->FittedFunction() == 0)
349 fFunc.reset();
350
351 if (!fObjFunction) {
352 MATH_ERROR_MSG("Fitter::FitFCN", "Objective function has not been set");
353 return false;
354 }
355 // create a Fit result from the fit configuration
356 fResult = std::make_shared<ROOT::Fit::FitResult>(fConfig);
357 // evaluate one time the FCN
358 double fcnval = (*fObjFunction)(fResult->GetParams());
359 // update fit result
360 fResult->fVal = fcnval;
361 fResult->fNCalls++;
362 return true;
363}
364
366{
367
368 // perform a chi2 fit on a set of binned data
369 std::shared_ptr<BinData> data = std::dynamic_pointer_cast<BinData>(fData);
370 assert(data);
371
372 // check function
373 if (!fFunc && !fFunc_v) {
374 MATH_ERROR_MSG("Fitter::DoLeastSquareFit", "model function is not set");
375 return false;
376 } else {
377
378#ifdef DEBUG
379 std::cout << "Fitter ParamSettings " << Config().ParamsSettings()[3].IsBound() << " lower limit "
380 << Config().ParamsSettings()[3].LowerLimit() << " upper limit "
381 << Config().ParamsSettings()[3].UpperLimit() << std::endl;
382#endif
383
384 fBinFit = true;
385 fDataSize = data->Size();
386 // check if fFunc provides gradient
387 if (!fUseGradient) {
388 // do minimzation without using the gradient
389 if (fFunc_v) {
390 Chi2FCN<BaseFunc, IModelFunction_v> chi2(data, fFunc_v, executionPolicy);
391 fFitType = chi2.Type();
392 return DoMinimization(chi2);
393 } else {
394 Chi2FCN<BaseFunc> chi2(data, fFunc, executionPolicy);
395 fFitType = chi2.Type();
396 return DoMinimization(chi2);
397 }
398 } else {
399 // use gradient
401 MATH_INFO_MSG("Fitter::DoLeastSquareFit", "use gradient from model function");
402
403 if (fFunc_v) {
404 std::shared_ptr<IGradModelFunction_v> gradFun = std::dynamic_pointer_cast<IGradModelFunction_v>(fFunc_v);
405 if (gradFun) {
406 Chi2FCN<BaseGradFunc, IModelFunction_v> chi2(data, gradFun);
407 fFitType = chi2.Type();
408 return DoMinimization(chi2);
409 }
410 } else {
411 std::shared_ptr<IGradModelFunction> gradFun = std::dynamic_pointer_cast<IGradModelFunction>(fFunc);
412 if (gradFun) {
413 Chi2FCN<BaseGradFunc> chi2(data, gradFun);
414 fFitType = chi2.Type();
415 return DoMinimization(chi2);
416 }
417 }
418 MATH_ERROR_MSG("Fitter::DoLeastSquareFit", "wrong type of function - it does not provide gradient");
419 }
420 }
421 return false;
422}
423
424bool Fitter::DoBinnedLikelihoodFit(bool extended, const ROOT::Fit::ExecutionPolicy &executionPolicy)
425{
426 // perform a likelihood fit on a set of binned data
427 // The fit is extended (Poisson logl_ by default
428
429 std::shared_ptr<BinData> data = std::dynamic_pointer_cast<BinData>(fData);
430 assert(data);
431
432 bool useWeight = fConfig.UseWeightCorrection();
433
434 // check function
435 if (!fFunc && !fFunc_v) {
436 MATH_ERROR_MSG("Fitter::DoBinnedLikelihoodFit", "model function is not set");
437 return false;
438 }
439
440 // logl fit (error should be 0.5) set if different than default values (of 1)
443 }
444
445 if (useWeight && fConfig.MinosErrors()) {
446 MATH_INFO_MSG("Fitter::DoBinnedLikelihoodFit", "MINOS errors cannot be computed in weighted likelihood fits");
447 fConfig.SetMinosErrors(false);
448 }
449
450 fBinFit = true;
451 fDataSize = data->Size();
452
453 if (!fUseGradient) {
454 // do minimization without using the gradient
455 if (fFunc_v) {
456 // create a chi2 function to be used for the equivalent chi-square
458 PoissonLikelihoodFCN<BaseFunc, IModelFunction_v> logl(data, fFunc_v, useWeight, extended, executionPolicy);
459 fFitType = logl.Type();
460 // do minimization
461 if (!DoMinimization(logl, &chi2))
462 return false;
463 if (useWeight) {
465 if (!ApplyWeightCorrection(logl))
466 return false;
467 }
468 } else {
469 // create a chi2 function to be used for the equivalent chi-square
470 Chi2FCN<BaseFunc> chi2(data, fFunc);
471 PoissonLikelihoodFCN<BaseFunc> logl(data, fFunc, useWeight, extended, executionPolicy);
472 fFitType = logl.Type();
473 // do minimization
474 if (!DoMinimization(logl, &chi2))
475 return false;
476 if (useWeight) {
478 if (!ApplyWeightCorrection(logl))
479 return false;
480 }
481 }
482 } else {
483 if (fFunc_v) {
484 // create a chi2 function to be used for the equivalent chi-square
486 std::shared_ptr<IGradModelFunction_v> gradFun = std::dynamic_pointer_cast<IGradModelFunction_v>(fFunc_v);
487 if (!gradFun) {
488 MATH_ERROR_MSG("Fitter::DoBinnedLikelihoodFit", "wrong type of function - it does not provide gradient");
489 return false;
490 }
491 PoissonLikelihoodFCN<BaseGradFunc, IModelFunction_v> logl(data, gradFun, useWeight, true, executionPolicy);
492 fFitType = logl.Type();
493 // do minimization
494 if (!DoMinimization(logl, &chi2))
495 return false;
496 if (useWeight) {
498 if (!ApplyWeightCorrection(logl))
499 return false;
500 }
501 } else {
502 // create a chi2 function to be used for the equivalent chi-square
503 Chi2FCN<BaseFunc> chi2(data, fFunc);
505 MATH_INFO_MSG("Fitter::DoLikelihoodFit", "use gradient from model function");
506 // check if fFunc provides gradient
507 std::shared_ptr<IGradModelFunction> gradFun = std::dynamic_pointer_cast<IGradModelFunction>(fFunc);
508 if (!gradFun) {
509 MATH_ERROR_MSG("Fitter::DoBinnedLikelihoodFit", "wrong type of function - it does not provide gradient");
510 return false;
511 }
512 // use gradient for minimization
513 // not-extended is not impelemented in this case
514 if (!extended) {
515 MATH_WARN_MSG("Fitter::DoBinnedLikelihoodFit",
516 "Not-extended binned fit with gradient not yet supported - do an extended fit");
517 }
518 PoissonLikelihoodFCN<BaseGradFunc> logl(data, gradFun, useWeight, true, executionPolicy);
519 fFitType = logl.Type();
520 // do minimization
521 if (!DoMinimization(logl, &chi2))
522 return false;
523 if (useWeight) {
525 if (!ApplyWeightCorrection(logl))
526 return false;
527 }
528 }
529 }
530 return true;
531}
532
533bool Fitter::DoUnbinnedLikelihoodFit(bool extended, const ROOT::Fit::ExecutionPolicy &executionPolicy) {
534 // perform a likelihood fit on a set of unbinned data
535
536 std::shared_ptr<UnBinData> data = std::dynamic_pointer_cast<UnBinData>(fData);
537 assert(data);
538
539 bool useWeight = fConfig.UseWeightCorrection();
540
541 if (!fFunc && !fFunc_v) {
542 MATH_ERROR_MSG("Fitter::DoUnbinnedLikelihoodFit","model function is not set");
543 return false;
544 }
545
546 if (useWeight && fConfig.MinosErrors() ) {
547 MATH_INFO_MSG("Fitter::DoUnbinnedLikelihoodFit","MINOS errors cannot be computed in weighted likelihood fits");
548 fConfig.SetMinosErrors(false);
549 }
550
551
552 fBinFit = false;
553 fDataSize = data->Size();
554
555#ifdef DEBUG
556 int ipar = 0;
557 std::cout << "Fitter ParamSettings " << Config().ParamsSettings()[ipar].IsBound() << " lower limit " << Config().ParamsSettings()[ipar].LowerLimit() << " upper limit " << Config().ParamsSettings()[ipar].UpperLimit() << std::endl;
558#endif
559
560 // logl fit (error should be 0.5) set if different than default values (of 1)
563 }
564
565 if (!fUseGradient) {
566 // do minimization without using the gradient
567 if (fFunc_v ){
568 LogLikelihoodFCN<BaseFunc, IModelFunction_v> logl(data, fFunc_v, useWeight, extended, executionPolicy);
569 fFitType = logl.Type();
570 if (!DoMinimization (logl) ) return false;
571 if (useWeight) {
573 if (!ApplyWeightCorrection(logl) ) return false;
574 }
575 return true;
576 } else {
577 LogLikelihoodFCN<BaseFunc> logl(data, fFunc, useWeight, extended, executionPolicy);
578
579 fFitType = logl.Type();
580 if (!DoMinimization (logl) ) return false;
581 if (useWeight) {
583 if (!ApplyWeightCorrection(logl) ) return false;
584 }
585 return true;
586 }
587 } else {
588 // use gradient : check if fFunc provides gradient
589 if (fFunc_v) {
591 MATH_INFO_MSG("Fitter::DoUnbinnedLikelihoodFit", "use gradient from model function");
592 std::shared_ptr<IGradModelFunction_v> gradFun = std::dynamic_pointer_cast<IGradModelFunction_v>(fFunc_v);
593 if (gradFun) {
594 if (extended) {
595 MATH_WARN_MSG("Fitter::DoUnbinnedLikelihoodFit",
596 "Extended unbinned fit with gradient not yet supported - do a not-extended fit");
597 }
598 LogLikelihoodFCN<BaseGradFunc, IModelFunction_v> logl(data, gradFun, useWeight, extended);
599 fFitType = logl.Type();
600 if (!DoMinimization(logl))
601 return false;
602 if (useWeight) {
604 if (!ApplyWeightCorrection(logl))
605 return false;
606 }
607 return true;
608 }
609 MATH_ERROR_MSG("Fitter::DoUnbinnedLikelihoodFit", "wrong type of function - it does not provide gradient");
610
611 } else {
613 MATH_INFO_MSG("Fitter::DoUnbinnedLikelihoodFit", "use gradient from model function");
614 std::shared_ptr<IGradModelFunction> gradFun = std::dynamic_pointer_cast<IGradModelFunction>(fFunc);
615 if (gradFun) {
616 if (extended) {
617 MATH_WARN_MSG("Fitter::DoUnbinnedLikelihoodFit",
618 "Extended unbinned fit with gradient not yet supported - do a not-extended fit");
619 }
620 LogLikelihoodFCN<BaseGradFunc> logl(data, gradFun, useWeight, extended);
621 fFitType = logl.Type();
622 if (!DoMinimization(logl))
623 return false;
624 if (useWeight) {
626 if (!ApplyWeightCorrection(logl))
627 return false;
628 }
629 return true;
630 }
631 MATH_ERROR_MSG("Fitter::DoUnbinnedLikelihoodFit", "wrong type of function - it does not provide gradient");
632 }
633 }
634 return false;
635}
636
637
639
640 std::shared_ptr<BinData> data = std::dynamic_pointer_cast<BinData>(fData);
641 assert(data);
642
643 // perform a linear fit on a set of binned data
644 std::string prevminimizer = fConfig.MinimizerType();
645 fConfig.SetMinimizer("Linear");
646
647 fBinFit = true;
648
649 bool ret = DoLeastSquareFit();
650 fConfig.SetMinimizer(prevminimizer.c_str());
651 return ret;
652}
653
654
656 // compute the Hesse errors according to configuration
657 // set in the parameters and append value in fit result
658 if (!fObjFunction) {
659 MATH_ERROR_MSG("Fitter::CalculateHessErrors","Objective function has not been set");
660 return false;
661 }
662
663 // need a special treatment in case of weighted likelihood fit
664 // (not yet implemented)
665 if (fFitType == 2 && fConfig.UseWeightCorrection() ) {
666 MATH_ERROR_MSG("Fitter::CalculateHessErrors","Re-computation of Hesse errors not implemented for weighted likelihood fits");
667 MATH_INFO_MSG("Fitter::CalculateHessErrors","Do the Fit using configure option FitConfig::SetParabErrors()");
668 return false;
669 }
670 // if (!fUseGradient ) {
671 // ROOT::Math::FitMethodFunction * fcn = dynamic_cast< ROOT::Math::FitMethodFunction *>(fObjFunction.get());
672 // if (fcn && fcn->Type() == ROOT::Math::FitMethodFunction::kLogLikelihood) {
673 // if (!fBinFit) {
674 // ROOT::Math::LogLikelihoodFunction * nll = dynamic_cast< ROOT::Math::LogLikelihoodFunction *>(fcn);
675 // assert(nll);
676 // nll->UseSumOfWeightSquare(false);
677 // }
678 // else {
679 // ROOT::Math::PoissonLikelihoodFunction * nll = dynamic_cast< ROOT::Math::PoissonLikelihoodFunction *>(fcn);
680 // assert(nll);
681 // nll->UseSumOfWeightSquare(false);
682 // }
683 // // reset fcn in minimizer
684 // }
685
686 // a fit Result pointer must exist when a minimizer exists
687 if (fMinimizer && !fResult ) {
688 MATH_ERROR_MSG("Fitter::CalculateHessErrors", "FitResult has not been created");
689 return false;
690 }
691
692 // update minimizer (recreate if not done or if name has changed
694 MATH_ERROR_MSG("Fitter::CalculateHessErrors","Error re-initializing the minimizer");
695 return false;
696 }
697
698 if (!fMinimizer ) {
699 // this should not happen
700 MATH_ERROR_MSG("Fitter::CalculateHessErrors", "Need to do a fit before calculating the errors");
701 assert(false);
702 return false;
703 }
704
705 //run Hesse
706 bool ret = fMinimizer->Hesse();
707 if (!ret) MATH_WARN_MSG("Fitter::CalculateHessErrors","Error when calculating Hessian");
708
709
710 // update minimizer results with what comes out from Hesse
711 // in case is empty - create from a FitConfig
712 if (fResult->IsEmpty() )
713 fResult = std::unique_ptr<ROOT::Fit::FitResult>(new ROOT::Fit::FitResult(fConfig) );
714
715
716 // re-give a minimizer instance in case it has been changed
717 ret |= fResult->Update(fMinimizer, fConfig, ret);
718
719 // when possible get ncalls from FCN and set in fit result
721 fResult->fNCalls = GetNCallsFromFCN();
722 }
723
724 // set also new errors in FitConfig
726
727 return ret;
728}
729
730
732 // compute the Minos errors according to configuration
733 // set in the parameters and append value in fit result
734 // normally Minos errors are computed just after the minimization
735 // (in DoMinimization) aftewr minimizing if the
736 // FitConfig::MinosErrors() flag is set
737
738 if (!fMinimizer) {
739 MATH_ERROR_MSG("Fitter::CalculateMinosErrors","Minimizer does not exist - cannot calculate Minos errors");
740 return false;
741 }
742
743 if (!fResult || fResult->IsEmpty() ) {
744 MATH_ERROR_MSG("Fitter::CalculateMinosErrors","Invalid Fit Result - cannot calculate Minos errors");
745 return false;
746 }
747
748 if (fFitType == 2 && fConfig.UseWeightCorrection() ) {
749 MATH_ERROR_MSG("Fitter::CalculateMinosErrors","Computation of MINOS errors not implemented for weighted likelihood fits");
750 return false;
751 }
752
753 // update minimizer (but cannot re-create in this case). Must use an existing one
754 if (!DoUpdateMinimizerOptions(false)) {
755 MATH_ERROR_MSG("Fitter::CalculateHessErrors","Error re-initializing the minimizer");
756 return false;
757 }
758
759 // set flag to compute Minos error to false in FitConfig to avoid that
760 // following minimizaiton calls perform unwanted Minos error calculations
761 /// fConfig.SetMinosErrors(false);
762
763
764 const std::vector<unsigned int> & ipars = fConfig.MinosParams();
765 unsigned int n = (ipars.size() > 0) ? ipars.size() : fResult->Parameters().size();
766 bool ok = false;
767
768 int iparNewMin = 0;
769 int iparMax = n;
770 int iter = 0;
771 // rerun minos for the parameters run before a new Minimum has been found
772 do {
773 if (iparNewMin > 0)
774 MATH_INFO_MSG("Fitter::CalculateMinosErrors","Run again Minos for some parameters because a new Minimum has been found");
775 iparNewMin = 0;
776 for (int i = 0; i < iparMax; ++i) {
777 double elow, eup;
778 unsigned int index = (ipars.size() > 0) ? ipars[i] : i;
779 bool ret = fMinimizer->GetMinosError(index, elow, eup);
780 // flags case when a new minimum has been found
781 if ((fMinimizer->MinosStatus() & 8) != 0) {
782 iparNewMin = i;
783 }
784 if (ret)
785 fResult->SetMinosError(index, elow, eup);
786 ok |= ret;
787 }
788
789 iparMax = iparNewMin;
790 iter++; // to avoid infinite looping
791 }
792 while( iparNewMin > 0 && iter < 10);
793 if (!ok) {
794 MATH_ERROR_MSG("Fitter::CalculateMinosErrors","Minos error calculation failed for all the selected parameters");
795 }
796
797 // re-give a minimizer instance in case it has been changed
798 // but maintain previous valid status. Do not set result to false if minos failed
799
800 ok &= fResult->Update(fMinimizer, fConfig, fResult->IsValid());
801
802 return ok;
803}
804
805
806
807// traits for distinhuishing fit methods functions from generic objective functions
808template<class Func>
809struct ObjFuncTrait {
810 static unsigned int NCalls(const Func & ) { return 0; }
811 static int Type(const Func & ) { return -1; }
812 static bool IsGrad() { return false; }
813};
814template<>
815struct ObjFuncTrait<ROOT::Math::FitMethodFunction> {
816 static unsigned int NCalls(const ROOT::Math::FitMethodFunction & f ) { return f.NCalls(); }
817 static int Type(const ROOT::Math::FitMethodFunction & f) { return f.Type(); }
818 static bool IsGrad() { return false; }
819};
820template<>
821struct ObjFuncTrait<ROOT::Math::FitMethodGradFunction> {
822 static unsigned int NCalls(const ROOT::Math::FitMethodGradFunction & f ) { return f.NCalls(); }
823 static int Type(const ROOT::Math::FitMethodGradFunction & f) { return f.Type(); }
824 static bool IsGrad() { return true; }
825};
826
828 //initialize minimizer by creating it
829 // and set there the objective function
830 // obj function must have been copied before
831 assert(fObjFunction.get() );
832
833 // check configuration and objective function
834 if ( fConfig.ParamsSettings().size() != fObjFunction->NDim() ) {
835 MATH_ERROR_MSG("Fitter::DoInitMinimizer","wrong function dimension or wrong size for FitConfig");
836 return false;
837 }
838
839 // create first Minimizer
840 // using an auto_Ptr will delete the previous existing one
841 fMinimizer = std::shared_ptr<ROOT::Math::Minimizer> ( fConfig.CreateMinimizer() );
842 if (fMinimizer.get() == 0) {
843 MATH_ERROR_MSG("Fitter::DoInitMinimizer","Minimizer cannot be created");
844 return false;
845 }
846
847 // in case of gradient function one needs to downcast the pointer
848 if (fUseGradient) {
849 const ROOT::Math::IMultiGradFunction * gradfcn = dynamic_cast<const ROOT::Math::IMultiGradFunction *> (fObjFunction.get() );
850 if (!gradfcn) {
851 MATH_ERROR_MSG("Fitter::DoInitMinimizer","wrong type of function - it does not provide gradient");
852 return false;
853 }
854 fMinimizer->SetFunction( *gradfcn);
855 }
856 else
857 fMinimizer->SetFunction( *fObjFunction);
858
859
860 fMinimizer->SetVariables(fConfig.ParamsSettings().begin(), fConfig.ParamsSettings().end() );
861
862 // if requested parabolic error do correct error analysis by the minimizer (call HESSE)
863 if (fConfig.ParabErrors()) fMinimizer->SetValidError(true);
864
865 return true;
866
867}
868
869bool Fitter::DoUpdateMinimizerOptions(bool canDifferentMinim ) {
870 // update minimizer options when re-doing a Fit or computing Hesse or Minos errors
871
872
873 // create a new minimizer if it is different type
874 // minimizer type string stored in FitResult is "minimizer name" + " / " + minimizer algo
875 std::string newMinimType = fConfig.MinimizerName();
876 if (fMinimizer && fResult && newMinimType != fResult->MinimizerType()) {
877 // if a different minimizer is allowed (e.g. when calling Hesse)
878 if (canDifferentMinim) {
879 std::string msg = "Using now " + newMinimType;
880 MATH_INFO_MSG("Fitter::DoUpdateMinimizerOptions: ", msg.c_str());
881 if (!DoInitMinimizer() )
882 return false;
883 }
884 else {
885 std::string msg = "Cannot change minimizer. Continue using " + fResult->MinimizerType();
886 MATH_WARN_MSG("Fitter::DoUpdateMinimizerOptions",msg.c_str());
887 }
888 }
889
890 // create minimizer if it was not done before
891 if (!fMinimizer) {
892 if (!DoInitMinimizer())
893 return false;
894 }
895
896 // set new minimizer options (but not functions and parameters)
897 fMinimizer->SetOptions(fConfig.MinimizerOptions());
898 return true;
899}
900
902 // perform the minimization (assume we have already initialized the minimizer)
903
904 assert(fMinimizer );
905
906 bool isValid = fMinimizer->Minimize();
907
908 // unsigned int ncalls = ObjFuncTrait<ObjFunc>::NCalls(*fcn);
909 // int fitType = ObjFuncTrait<ObjFunc>::Type(objFunc);
910
911 if (!fResult) fResult = std::make_shared<FitResult>();
912
913 fResult->FillResult(fMinimizer,fConfig, fFunc, isValid, fDataSize, fBinFit, chi2func );
914
915 // if requested run Minos after minimization
916 if (isValid && fConfig.MinosErrors()) {
917 // minos error calculation will update also FitResult
919 }
920
921 // when possible get ncalls from FCN and set in fit result
923 fResult->fNCalls = GetNCallsFromFCN();
924 }
925
926 // fill information in fit result
927 fResult->fObjFunc = fObjFunction;
928 fResult->fFitData = fData;
929
930
931#ifdef DEBUG
932 std::cout << "ROOT::Fit::Fitter::DoMinimization : ncalls = " << fResult->fNCalls << " type of objfunc " << fFitFitResType << " typeid: " << typeid(*fObjFunction).name() << " use gradient " << fUseGradient << std::endl;
933#endif
934
936 fResult->NormalizeErrors();
937
938 // set also new parameter values and errors in FitConfig
939 if (fConfig.UpdateAfterFit() && isValid) DoUpdateFitConfig();
940
941 return isValid;
942}
943
944bool Fitter::DoMinimization(const BaseFunc & objFunc, const ROOT::Math::IMultiGenFunction * chi2func) {
945 // perform the minimization initializing the minimizer starting from a given obj function
946
947 // keep also a copy of FCN function and set this in minimizer so they will be managed together
948 // (remember that cloned copy will still depends on data and model function pointers)
949 fObjFunction = std::unique_ptr<ROOT::Math::IMultiGenFunction> ( objFunc.Clone() );
950 if (!DoInitMinimizer()) return false;
951 return DoMinimization(chi2func);
952}
953
954
956 // update the fit configuration after a fit using the obtained result
957 if (fResult->IsEmpty() || !fResult->IsValid() ) return;
958 for (unsigned int i = 0; i < fConfig.NPar(); ++i) {
960 par.SetValue( fResult->Value(i) );
961 if (fResult->Error(i) > 0) par.SetStepSize( fResult->Error(i) );
962 }
963}
964
966 // retrieve ncalls from the fit method functions
967 // this function is called when minimizer does not provide a way of returning the nnumber of function calls
968 int ncalls = 0;
969 if (!fUseGradient) {
970 const ROOT::Math::FitMethodFunction * fcn = dynamic_cast<const ROOT::Math::FitMethodFunction *>(fObjFunction.get());
971 if (fcn) ncalls = fcn->NCalls();
972 }
973 else {
975 if (fcn) ncalls = fcn->NCalls();
976 }
977 return ncalls;
978}
979
980
981bool Fitter::ApplyWeightCorrection(const ROOT::Math::IMultiGenFunction & loglw2, bool minimizeW2L) {
982 // apply correction for weight square
983 // Compute Hessian of the loglikelihood function using the sum of the weight squared
984 // This method assumes:
985 // - a fit has been done before and a covariance matrix exists
986 // - the objective function is a likelihood function and Likelihood::UseSumOfWeightSquare()
987 // has been called before
988
989 if (fMinimizer.get() == 0) {
990 MATH_ERROR_MSG("Fitter::ApplyWeightCorrection","Must perform first a fit before applying the correction");
991 return false;
992 }
993
994 unsigned int n = loglw2.NDim();
995 // correct errors for weight squared
996 std::vector<double> cov(n*n);
997 bool ret = fMinimizer->GetCovMatrix(&cov[0] );
998 if (!ret) {
999 MATH_ERROR_MSG("Fitter::ApplyWeightCorrection","Previous fit has no valid Covariance matrix");
1000 return false;
1001 }
1002 // need to re-init the minimizer and set w2
1003 fObjFunction = std::unique_ptr<ROOT::Math::IMultiGenFunction> ( loglw2.Clone() );
1004 // need to re-initialize the minimizer for the changes applied in the
1005 // objective functions
1006 if (!DoInitMinimizer()) return false;
1007
1008 //std::cout << "Running Hesse ..." << std::endl;
1009
1010 // run eventually before a minimization
1011 // ignore its error
1012 if (minimizeW2L) fMinimizer->Minimize();
1013 // run Hesse on the log-likelihood build using sum of weight squared
1014 ret = fMinimizer->Hesse();
1015 if (!ret) {
1016 MATH_ERROR_MSG("Fitter::ApplyWeightCorrection","Error running Hesse on weight2 likelihood - cannot compute errors");
1017 return false;
1018 }
1019
1020 if (fMinimizer->CovMatrixStatus() != 3) {
1021 MATH_WARN_MSG("Fitter::ApplyWeightCorrection","Covariance matrix for weighted likelihood is not accurate, the errors may be not reliable");
1022 if (fMinimizer->CovMatrixStatus() == 2)
1023 MATH_WARN_MSG("Fitter::ApplyWeightCorrection","Covariance matrix for weighted likelihood was forced to be defined positive");
1024 if (fMinimizer->CovMatrixStatus() <= 0)
1025 // probably should have failed before
1026 MATH_ERROR_MSG("Fitter::ApplyWeightCorrection","Covariance matrix for weighted likelihood is not valid !");
1027 }
1028
1029 // std::vector<double> c(n*n);
1030 // ret = fMinimizer->GetCovMatrix(&c2[0] );
1031 // if (!ret) std::cout << "Error reading cov matrix " << fMinimizer->Status() << std::endl;
1032 // TMatrixDSym cmat2(n,&c2[0]);
1033 // std::cout << "Cov matrix of w2 " << std::endl;
1034 // cmat2.Print();
1035 // cmat2.Invert();
1036 // std::cout << "Hessian of w2 " << std::endl;
1037 // cmat2.Print();
1038
1039 // get Hessian matrix from weight-square likelihood
1040 std::vector<double> hes(n*n);
1041 ret = fMinimizer->GetHessianMatrix(&hes[0] );
1042 if (!ret) {
1043 MATH_ERROR_MSG("Fitter::ApplyWeightCorrection","Error retrieving Hesse on weight2 likelihood - cannot compute errors");
1044 return false;
1045 }
1046
1047 // for debug
1048 // std::cout << "Hessian W2 matrix " << std::endl;
1049 // for (unsigned int i = 0; i < n; ++i) {
1050 // for (unsigned int j = 0; j < n; ++j) {
1051 // std::cout << std::setw(12) << hes[i*n + j] << " , ";
1052 // }
1053 // std::cout << std::endl;
1054 // }
1055
1056 // perform product of matvrix cov * hes * cov
1057 // since we do not want to add matrix dependence do product by hand
1058 // first do hes * cov
1059 std::vector<double> tmp(n*n);
1060 for (unsigned int i = 0; i < n; ++i) {
1061 for (unsigned int j = 0; j < n; ++j) {
1062 for (unsigned int k = 0; k < n; ++k)
1063 tmp[i*n+j] += hes[i*n + k] * cov[k*n + j];
1064 }
1065 }
1066 // do multiplication now cov * tmp save result
1067 std::vector<double> newCov(n*n);
1068 for (unsigned int i = 0; i < n; ++i) {
1069 for (unsigned int j = 0; j < n; ++j) {
1070 for (unsigned int k = 0; k < n; ++k)
1071 newCov[i*n+j] += cov[i*n + k] * tmp[k*n + j];
1072 }
1073 }
1074 // update fit result with new corrected covariance matrix
1075 unsigned int k = 0;
1076 for (unsigned int i = 0; i < n; ++i) {
1077 fResult->fErrors[i] = std::sqrt( newCov[i*(n+1)] );
1078 for (unsigned int j = 0; j <= i; ++j)
1079 fResult->fCovMatrix[k++] = newCov[i *n + j];
1080 }
1081 //fResult->PrintCovMatrix(std::cout);
1082
1083 return true;
1084}
1085
1086
1087
1089 // return a pointer to the binned data used in the fit
1090 // works only for chi2 or binned likelihood fits
1091 // thus when the objective function stored is a Chi2Func or a PoissonLikelihood
1092 // The funciton also set the model function correctly if it has not been set
1093
1096
1099
1100 //MATH_INFO_MSG("Fitter::ExamineFCN","Objective function is not of a known type - FitData and ModelFunction objects are not available");
1101 return;
1102}
1103
1104 } // end namespace Fit
1105
1106} // end namespace ROOT
#define MATH_INFO_MSG(loc, str)
Pre-processor macro to report messages which can be configured to use ROOT error or simply an std::io...
Definition: Error.h:76
#define MATH_ERROR_MSG(loc, str)
Definition: Error.h:82
#define MATH_WARN_MSG(loc, str)
Definition: Error.h:79
#define f(i)
Definition: RSha256.hxx:104
double sqrt(double)
BasicFCN class: base class for the objective functions used in the fits It has a reference to the dat...
Definition: BasicFCN.h:40
Chi2FCN class for binnned fits using the least square methods.
Definition: Chi2FCN.h:49
virtual BaseObjFunction::Type_t Type() const
get type of fit method function
Definition: Chi2FCN.h:134
const std::vector< unsigned int > & MinosParams() const
return vector of parameter indeces for which the Minos Error will be computed
Definition: FitConfig.h:219
void SetParamsSettings(unsigned int npar, const double *params, const double *vstep=0)
set the parameter settings from number of parameters and a vector of values and optionally step value...
Definition: FitConfig.cxx:136
bool UpdateAfterFit() const
Update configuration after a fit using the FitResult.
Definition: FitConfig.h:212
void SetMinosErrors(bool on=true)
set Minos erros computation to be performed after fitting
Definition: FitConfig.h:230
void SetMinimizer(const char *type, const char *algo=0)
set minimizer type
Definition: FitConfig.h:180
bool NormalizeErrors() const
flag to check if resulting errors are be normalized according to chi2/ndf
Definition: FitConfig.h:203
bool ParabErrors() const
do analysis for parabolic errors
Definition: FitConfig.h:206
unsigned int NPar() const
number of parameters settings
Definition: FitConfig.h:95
std::string MinimizerName() const
return Minimizer full name (type / algorithm)
Definition: FitConfig.cxx:238
bool UseWeightCorrection() const
Apply Weight correction for error matrix computation.
Definition: FitConfig.h:215
const std::vector< ROOT::Fit::ParameterSettings > & ParamsSettings() const
get the vector of parameter settings (const method)
Definition: FitConfig.h:85
ROOT::Math::Minimizer * CreateMinimizer()
create a new minimizer according to chosen configuration
Definition: FitConfig.cxx:174
void CreateParamsSettings(const ROOT::Math::IParamMultiFunctionTempl< T > &func)
set the parameter settings from a model function.
Definition: FitConfig.h:108
const std::string & MinimizerType() const
return type of minimizer package
Definition: FitConfig.h:188
const ParameterSettings & ParSettings(unsigned int i) const
get the parameter settings for the i-th parameter (const method)
Definition: FitConfig.h:75
ROOT::Math::MinimizerOptions & MinimizerOptions()
access to the minimizer control parameter (non const method)
Definition: FitConfig.h:166
bool MinosErrors() const
do minos errros analysis on the parameters
Definition: FitConfig.h:209
class containg the result of the fit and all the related information (fitted parameter values,...
Definition: FitResult.h:47
Fitter class, entry point for performing all type of fits.
Definition: Fitter.h:77
bool EvalFCN()
Perform a simple FCN evaluation.
Definition: Fitter.cxx:344
bool FitFCN()
Perform a fit with the previously set FCN function.
Definition: Fitter.cxx:326
void DoUpdateFitConfig()
Definition: Fitter.cxx:955
std::shared_ptr< ROOT::Math::Minimizer > fMinimizer
pointer to the object containing the result of the fit
Definition: Fitter.h:532
bool DoBinnedLikelihoodFit(bool extended=true, const ROOT::Fit::ExecutionPolicy &executionPolicy=ROOT::Fit::ExecutionPolicy::kSerial)
binned likelihood fit
Definition: Fitter.cxx:424
std::shared_ptr< ROOT::Fit::FitData > fData
pointer to used minimizer
Definition: Fitter.h:534
bool fUseGradient
Definition: Fitter.h:514
bool SetFCN(unsigned int npar, Function &fcn, const double *params=0, unsigned int dataSize=0, bool chi2fit=false)
Set a generic FCN function as a C++ callable object implementing double () (const double *) Note that...
Definition: Fitter.h:615
bool DoMinimization(const BaseFunc &f, const ROOT::Math::IMultiGenFunction *chifunc=0)
do minimization
Definition: Fitter.cxx:944
std::shared_ptr< ROOT::Math::IMultiGenFunction > fObjFunction
pointer to the fit data (binned or unbinned data)
Definition: Fitter.h:536
bool ApplyWeightCorrection(const ROOT::Math::IMultiGenFunction &loglw2, bool minimizeW2L=false)
apply correction in the error matrix for the weights for likelihood fits This method can be called on...
Definition: Fitter.cxx:981
void ExamineFCN()
look at the user provided FCN and get data and model function is they derive from ROOT::Fit FCN class...
Definition: Fitter.cxx:1088
const FitConfig & Config() const
access to the fit configuration (const method)
Definition: Fitter.h:412
std::shared_ptr< IModelFunction_v > fFunc_v
Definition: Fitter.h:526
bool DoUnbinnedLikelihoodFit(bool extended=false, const ROOT::Fit::ExecutionPolicy &executionPolicy=ROOT::Fit::ExecutionPolicy::kSerial)
un-binned likelihood fit
Definition: Fitter.cxx:533
Fitter & operator=(const Fitter &rhs)
Assignment operator (disabled, class is not copyable)
Definition: Fitter.cxx:84
std::shared_ptr< ROOT::Fit::FitResult > fResult
copy of the fitted function containing on output the fit result
Definition: Fitter.h:530
bool GetDataFromFCN()
internal functions to get data set and model function from FCN useful for fits done with customized F...
Definition: Fitter.h:544
bool CalculateMinosErrors()
perform an error analysis on the result using MINOS To be called only after fitting and when a minimi...
Definition: Fitter.cxx:731
~Fitter()
Destructor.
Definition: Fitter.cxx:70
bool DoUpdateMinimizerOptions(bool canDifferentMinim=true)
Definition: Fitter.cxx:869
void SetFunction(const IModelFunction &func, bool useGradient=false)
Set the fitted function (model function) from a parametric function interface.
Definition: Fitter.cxx:103
bool CalculateHessErrors()
perform an error analysis on the result using the Hessian Errors are obtaied from the inverse of the ...
Definition: Fitter.cxx:655
FitConfig fConfig
Definition: Fitter.h:524
Fitter()
Default constructor.
Definition: Fitter.cxx:51
std::shared_ptr< IModelFunction > fFunc
copy of the fitted function containing on output the fit result
Definition: Fitter.h:528
bool DoLinearFit()
linear least square fit
Definition: Fitter.cxx:638
bool DoInitMinimizer()
Definition: Fitter.cxx:827
int GetNCallsFromFCN()
Definition: Fitter.cxx:965
bool DoLeastSquareFit(const ROOT::Fit::ExecutionPolicy &executionPolicy=ROOT::Fit::ExecutionPolicy::kSerial)
least square fit
Definition: Fitter.cxx:365
LogLikelihoodFCN class for likelihood fits.
virtual BaseObjFunction::Type_t Type() const
get type of fit method function
void UseSumOfWeightSquare(bool on=true)
Class, describing value, limits and step size of the parameters Provides functionality also to set/re...
void SetValue(double val)
set the value
void SetStepSize(double err)
set the step size
class evaluating the log likelihood for binned Poisson likelihood fits it is template to distinguish ...
virtual BaseObjFunction::Type_t Type() const
get type of fit method function
FitMethodFunction class Interface for objective functions (like chi2 and likelihood used in the fit) ...
virtual Type_t Type() const
return the type of method, override if needed
virtual unsigned int NPoints() const
return the number of data points used in evaluating the function
virtual unsigned int NCalls() const
return the total number of function calls (overrided if needed)
Documentation for the abstract class IBaseFunctionMultiDim.
Definition: IFunction.h:62
virtual IBaseFunctionMultiDimTempl< T > * Clone() const =0
Clone a function.
virtual unsigned int NDim() const =0
Retrieve the dimension of the function.
Interface (abstract class) for multi-dimensional functions providing a gradient calculation.
Definition: IFunction.h:327
Specialized IParamFunction interface (abstract class) for one-dimensional parametric functions It is ...
Interface (abstract class) for parametric gradient multi-dimensional functions providing in addition ...
Interface (abstract class) for parametric one-dimensional gradient functions providing in addition to...
double ErrorDef() const
error definition
int PrintLevel() const
non-static methods for retrieving options
void SetErrorDef(double err)
set error def
MultiDimParamFunctionAdapter class to wrap a one-dimensional parametric function in a multi dimension...
MultiDimParamGradFunctionAdapter class to wrap a one-dimensional parametric gradient function in a mu...
Type
enumeration specifying the integration types.
const Int_t n
Definition: legend1.C:16
TFitResultPtr Fit(FitObject *h1, TF1 *f1, Foption_t &option, const ROOT::Math::MinimizerOptions &moption, const char *goption, ROOT::Fit::DataRange &range)
Definition: HFitImpl.cxx:134
double gDefaultErrorDef
Definition: Fitter.cxx:48
BasicFitMethodFunction< ROOT::Math::IMultiGenFunction > FitMethodFunction
Definition: Fitter.h:40
BasicFitMethodFunction< ROOT::Math::IMultiGradFunction > FitMethodGradFunction
Definition: Fitter.h:44
tbb::task_arena is an alias of tbb::interface7::task_arena, which doesn't allow to forward declare tb...
Definition: StringConv.hxx:21