Logo ROOT   6.18/05
Reference Guide
GSLNLSMinimizer.cxx
Go to the documentation of this file.
1// @(#)root/mathmore:$Id$
2// Author: L. Moneta Wed Dec 20 17:16:32 2006
3
4/**********************************************************************
5 * *
6 * Copyright (c) 2006 LCG ROOT Math Team, CERN/PH-SFT *
7 * *
8 * *
9 **********************************************************************/
10
11// Implementation file for class GSLNLSMinimizer
12
14
17
18#include "Math/Error.h"
19#include "GSLMultiFit.h"
20#include "gsl/gsl_errno.h"
21
22
24//#include "Math/Derivator.h"
25
26#include <iostream>
27#include <iomanip>
28#include <cassert>
29
30namespace ROOT {
31
32 namespace Math {
33
34
35// class to implement transformation of chi2 function
36// in general could make template on the fit method function type
37
38class FitTransformFunction : public FitMethodFunction {
39
40public:
41
42 FitTransformFunction(const FitMethodFunction & f, const std::vector<EMinimVariableType> & types, const std::vector<double> & values,
43 const std::map<unsigned int, std::pair<double, double> > & bounds) :
44 FitMethodFunction( f.NDim(), f.NPoints() ),
45 fOwnTransformation(true),
46 fFunc(f),
47 fTransform(new MinimTransformFunction( new MultiNumGradFunction(f), types, values, bounds) ),
48 fGrad( std::vector<double>(f.NDim() ) )
49 {
50 // constructor
51 // need to pass to MinimTransformFunction a new pointer which will be managed by the class itself
52 // pass a gradient pointer although it will not be used byb the class
53 }
54
55 FitTransformFunction(const FitMethodFunction & f, MinimTransformFunction *transFunc ) :
56 FitMethodFunction( f.NDim(), f.NPoints() ),
57 fOwnTransformation(false),
58 fFunc(f),
59 fTransform(transFunc),
60 fGrad( std::vector<double>(f.NDim() ) )
61 {
62 // constructor from al already existing Transformation object. Ownership of the transformation onbect is passed to caller
63 }
64
65 ~FitTransformFunction() {
66 if (fOwnTransformation) {
67 assert(fTransform);
68 delete fTransform;
69 }
70 }
71
72 // re-implement data element
73 virtual double DataElement(const double * x, unsigned i, double * g = 0) const {
74 // transform from x internal to x external
75 const double * xExt = fTransform->Transformation(x);
76 if ( g == 0) return fFunc.DataElement( xExt, i );
77 // use gradient
78 double val = fFunc.DataElement( xExt, i, &fGrad[0]);
79 // transform gradient
80 fTransform->GradientTransformation( x, &fGrad.front(), g);
81 return val;
82 }
83
84
85 IMultiGenFunction * Clone() const {
86 // not supported
87 return 0;
88 }
89
90 // dimension (this is number of free dimensions)
91 unsigned int NDim() const {
92 return fTransform->NDim();
93 }
94
95 unsigned int NTot() const {
96 return fTransform->NTot();
97 }
98
99 // forward of transformation functions
100 const double * Transformation( const double * x) const { return fTransform->Transformation(x); }
101
102
103 /// inverse transformation (external -> internal)
104 void InvTransformation(const double * xext, double * xint) const { fTransform->InvTransformation(xext,xint); }
105
106 /// inverse transformation for steps (external -> internal) at external point x
107 void InvStepTransformation(const double * x, const double * sext, double * sint) const { fTransform->InvStepTransformation(x,sext,sint); }
108
109 ///transform gradient vector (external -> internal) at internal point x
110 void GradientTransformation(const double * x, const double *gext, double * gint) const { fTransform->GradientTransformation(x,gext,gint); }
111
112 void MatrixTransformation(const double * x, const double *cint, double * cext) const { fTransform->MatrixTransformation(x,cint,cext); }
113
114private:
115
116 // objects of this class are not meant for copying or assignment
117 FitTransformFunction(const FitTransformFunction& rhs);
118 FitTransformFunction& operator=(const FitTransformFunction& rhs);
119
120 double DoEval(const double * x) const {
121 return fFunc( fTransform->Transformation(x) );
122 }
123
124 bool fOwnTransformation;
125 const FitMethodFunction & fFunc; // pointer to original fit method function
126 MinimTransformFunction * fTransform; // pointer to transformation function
127 mutable std::vector<double> fGrad; // cached vector of gradient values
128
129};
130
131
132
133
134// GSLNLSMinimizer implementation
135
137 //fNFree(0),
138 fSize(0),
139 fChi2Func(0)
140{
141 // Constructor implementation : create GSLMultiFit wrapper object
142 const gsl_multifit_fdfsolver_type * gsl_type = 0; // use default type defined in GSLMultiFit
143 if (type == 1) gsl_type = gsl_multifit_fdfsolver_lmsder; // scaled lmder version
144 if (type == 2) gsl_type = gsl_multifit_fdfsolver_lmder; // unscaled version
145
146 fGSLMultiFit = new GSLMultiFit( gsl_type );
147
148 fEdm = -1;
149
150 // default tolerance and max iterations
152 if (niter <= 0) niter = 100;
153 SetMaxIterations(niter);
154
156 if (fLSTolerance <=0) fLSTolerance = 0.0001; // default internal value
157
159}
160
162 assert(fGSLMultiFit != 0);
163 delete fGSLMultiFit;
164}
165
166
167
169 // set the function to minimizer
170 // need to create vector of functions to be passed to GSL multifit
171 // support now only CHi2 implementation
172
173 // call base class method. It will clone the function and set ndimension
175 //need to check if function can be used
176 const ROOT::Math::FitMethodFunction * chi2Func = dynamic_cast<const ROOT::Math::FitMethodFunction *>(ObjFunction());
177 if (chi2Func == 0) {
178 if (PrintLevel() > 0) std::cout << "GSLNLSMinimizer: Invalid function set - only Chi2Func supported" << std::endl;
179 return;
180 }
181 fSize = chi2Func->NPoints();
182 fNFree = NDim();
183
184 // use vector by value
185 fResiduals.reserve(fSize);
186 for (unsigned int i = 0; i < fSize; ++i) {
187 fResiduals.push_back( LSResidualFunc(*chi2Func, i) );
188 }
189 // keep pointers to the chi2 function
190 fChi2Func = chi2Func;
191 }
192
194 // set the function to minimizer using gradient interface
195 // not supported yet, implemented using the other SetFunction
196 return SetFunction(static_cast<const ROOT::Math::IMultiGenFunction &>(func) );
197}
198
199
201 // set initial parameters of the minimizer
202 int debugLevel = PrintLevel();
203
204
205 assert (fGSLMultiFit != 0);
206 if (fResiduals.size() != fSize || fChi2Func == 0) {
207 MATH_ERROR_MSG("GSLNLSMinimizer::Minimize","Function has not been set");
208 return false;
209 }
210
211 unsigned int npar = NPar();
212 unsigned int ndim = NDim();
213 if (npar == 0 || npar < ndim) {
214 MATH_ERROR_MSGVAL("GSLNLSMinimizer::Minimize","Wrong number of parameters",npar);
215 return false;
216 }
217
218 // set residual functions and check if a transformation is needed
219 std::vector<double> startValues;
220
221 // transformation need a grad function. Delegate fChi2Func to given object
223 MinimTransformFunction * trFuncRaw = CreateTransformation(startValues, gradFunction);
224 // need to transform in a FitTransformFunction which is set in the residual functions
225 std::unique_ptr<FitTransformFunction> trFunc;
226 if (trFuncRaw) {
227 trFunc.reset(new FitTransformFunction(*fChi2Func, trFuncRaw) );
228 //FitTransformationFunction *trFunc = new FitTransformFunction(*fChi2Func, trFuncRaw);
229 for (unsigned int ires = 0; ires < fResiduals.size(); ++ires) {
230 fResiduals[ires] = LSResidualFunc(*trFunc, ires);
231 }
232
233 assert(npar == trFunc->NTot() );
234 }
235
236 if (debugLevel >=1 ) std::cout <<"Minimize using GSLNLSMinimizer " << std::endl;
237
238// // use a global step size = min (step vectors)
239// double stepSize = 1;
240// for (unsigned int i = 0; i < fSteps.size(); ++i)
241// //stepSize += fSteps[i];
242// if (fSteps[i] < stepSize) stepSize = fSteps[i];
243
244 int iret = fGSLMultiFit->Set( fResiduals, &startValues.front() );
245 if (iret) {
246 MATH_ERROR_MSGVAL("GSLNLSMinimizer::Minimize","Error setting the residual functions ",iret);
247 return false;
248 }
249
250 if (debugLevel >=1 ) std::cout <<"GSLNLSMinimizer: " << fGSLMultiFit->Name() << " - start iterating......... " << std::endl;
251
252 // start iteration
253 unsigned int iter = 0;
254 int status;
255 bool minFound = false;
256 do {
257 status = fGSLMultiFit->Iterate();
258
259 if (debugLevel >=1) {
260 std::cout << "----------> Iteration " << iter << " / " << MaxIterations() << " status " << gsl_strerror(status) << std::endl;
261 const double * x = fGSLMultiFit->X();
262 if (trFunc.get()) x = trFunc->Transformation(x);
263 int pr = std::cout.precision(18);
264 std::cout << " FVAL = " << (*fChi2Func)(x) << std::endl;
265 std::cout.precision(pr);
266 std::cout << " X Values : ";
267 for (unsigned int i = 0; i < NDim(); ++i)
268 std::cout << " " << VariableName(i) << " = " << X()[i];
269 std::cout << std::endl;
270 }
271
272 if (status) break;
273
274 // check also the delta in X()
275 status = fGSLMultiFit->TestDelta( Tolerance(), Tolerance() );
276 if (status == GSL_SUCCESS) {
277 minFound = true;
278 }
279
280 // double-check with the gradient
281 int status2 = fGSLMultiFit->TestGradient( Tolerance() );
282 if ( minFound && status2 != GSL_SUCCESS) {
283 // check now edm
284 fEdm = fGSLMultiFit->Edm();
285 if (fEdm > Tolerance() ) {
286 // continue the iteration
287 status = status2;
288 minFound = false;
289 }
290 }
291
292 if (debugLevel >=1) {
293 std::cout << " after Gradient and Delta tests: " << gsl_strerror(status);
294 if (fEdm > 0) std::cout << ", edm is: " << fEdm;
295 std::cout << std::endl;
296 }
297
298 iter++;
299
300 }
301 while (status == GSL_CONTINUE && iter < MaxIterations() );
302
303 // check edm
304 fEdm = fGSLMultiFit->Edm();
305 if ( fEdm < Tolerance() ) {
306 minFound = true;
307 }
308
309 // save state with values and function value
310 const double * x = fGSLMultiFit->X();
311 if (x == 0) return false;
312
314
315 SetMinValue( (*fChi2Func)(x) );
316 fStatus = status;
317
318 fErrors.resize(NDim());
319
320 // get errors from cov matrix
321 const double * cov = fGSLMultiFit->CovarMatrix();
322 if (cov) {
323
324 fCovMatrix.resize(ndim*ndim);
325
326 if (trFunc.get() ) {
327 trFunc->MatrixTransformation(x, fGSLMultiFit->CovarMatrix(), &fCovMatrix[0] );
328 }
329 else {
330 std::copy(cov, cov + ndim*ndim, fCovMatrix.begin() );
331 }
332
333 for (unsigned int i = 0; i < ndim; ++i)
334 fErrors[i] = std::sqrt(fCovMatrix[i*ndim + i]);
335 }
336
337 if (minFound) {
338
339 if (debugLevel >=1 ) {
340 std::cout << "GSLNLSMinimizer: Minimum Found" << std::endl;
341 int pr = std::cout.precision(18);
342 std::cout << "FVAL = " << MinValue() << std::endl;
343 std::cout << "Edm = " << fEdm << std::endl;
344 std::cout.precision(pr);
345 std::cout << "NIterations = " << iter << std::endl;
346 std::cout << "NFuncCalls = " << fChi2Func->NCalls() << std::endl;
347 for (unsigned int i = 0; i < NDim(); ++i)
348 std::cout << std::setw(12) << VariableName(i) << " = " << std::setw(12) << X()[i] << " +/- " << std::setw(12) << fErrors[i] << std::endl;
349 }
350
351 return true;
352 }
353 else {
354 if (debugLevel >=1 ) {
355 std::cout << "GSLNLSMinimizer: Minimization did not converge" << std::endl;
356 std::cout << "FVAL = " << MinValue() << std::endl;
357 std::cout << "Edm = " << fGSLMultiFit->Edm() << std::endl;
358 std::cout << "Niterations = " << iter << std::endl;
359 }
360 return false;
361 }
362 return false;
363}
364
365const double * GSLNLSMinimizer::MinGradient() const {
366 // return gradient (internal values)
367 return fGSLMultiFit->Gradient();
368}
369
370
371double GSLNLSMinimizer::CovMatrix(unsigned int i , unsigned int j ) const {
372 // return covariance matrix element
373 unsigned int ndim = NDim();
374 if ( fCovMatrix.size() == 0) return 0;
375 if (i > ndim || j > ndim) return 0;
376 return fCovMatrix[i*ndim + j];
377}
378
380 // return covariance matrix status = 0 not computed,
381 // 1 computed but is approximate because minimum is not valid, 3 is fine
382 if ( fCovMatrix.size() == 0) return 0;
383 // case minimization did not finished correctly
384 if (fStatus != GSL_SUCCESS) return 1;
385 return 3;
386}
387
388
389 } // end namespace Math
390
391} // end namespace ROOT
392
#define MATH_ERROR_MSGVAL(loc, txt, x)
Definition: Error.h:108
#define MATH_ERROR_MSG(loc, str)
Definition: Error.h:82
#define f(i)
Definition: RSha256.hxx:104
#define g(i)
Definition: RSha256.hxx:105
int type
Definition: TGX11.cxx:120
double sqrt(double)
Binding & operator=(OUT(*fun)(void))
FitMethodFunction class Interface for objective functions (like chi2 and likelihood used in the fit) ...
virtual double DataElement(const double *x, unsigned int i, double *g=0) const =0
method returning the data i-th contribution to the fit objective function For example the residual fo...
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)
virtual double MinValue() const
return minimum function value
virtual unsigned int NPar() const
total number of parameter defined
void SetMinValue(double val)
virtual void SetFunction(const ROOT::Math::IMultiGenFunction &func)
set the function to minimize
const ROOT::Math::IMultiGenFunction * ObjFunction() const
return pointer to used objective function
virtual unsigned int NDim() const
number of dimensions
virtual std::string VariableName(unsigned int ivar) const
get name of variables (override if minimizer support storing of variable names)
virtual const double * X() const
return pointer to X values at the minimum
void SetFinalValues(const double *x)
MinimTransformFunction * CreateTransformation(std::vector< double > &startValues, const ROOT::Math::IMultiGradFunction *func=0)
GSLMultiFit, internal class for implementing GSL non linear least square GSL fitting.
Definition: GSLMultiFit.h:52
int TestGradient(double absTol) const
test gradient (ask from solver gradient vector)
Definition: GSLMultiFit.h:195
int TestDelta(double absTol, double relTol) const
test using abs and relative tolerance |dx| < absTol + relTol*|x| for every component
Definition: GSLMultiFit.h:203
const double * Gradient() const
gradient value at the minimum
Definition: GSLMultiFit.h:170
std::string Name() const
Definition: GSLMultiFit.h:152
double Edm() const
Definition: GSLMultiFit.h:209
const double * CovarMatrix() const
return covariance matrix of the parameters
Definition: GSLMultiFit.h:181
const double * X() const
parameter values at the minimum
Definition: GSLMultiFit.h:163
int Set(const std::vector< Func > &funcVec, const double *x)
set the solver parameters
Definition: GSLMultiFit.h:123
const ROOT::Math::FitMethodFunction * fChi2Func
virtual bool Minimize()
method to perform the minimization
std::vector< double > fErrors
std::vector< LSResidualFunc > fResiduals
virtual double CovMatrix(unsigned int, unsigned int) const
return covariance matrices elements if the variable is fixed the matrix is zero The ordering of the v...
std::vector< double > fCovMatrix
virtual int CovMatrixStatus() const
return covariance matrix status
virtual const double * MinGradient() const
return pointer to gradient values at the minimum
GSLNLSMinimizer(int type=0)
Default constructor.
ROOT::Math::GSLMultiFit * fGSLMultiFit
virtual void SetFunction(const ROOT::Math::IMultiGenFunction &func)
set the function to minimize
~GSLNLSMinimizer()
Destructor (no operations)
Documentation for the abstract class IBaseFunctionMultiDim.
Definition: IFunction.h:62
Interface (abstract class) for multi-dimensional functions providing a gradient calculation.
Definition: IFunction.h:327
LSResidualFunc class description.
MinimTransformFunction class to perform a transformations on the variables to deal with fixed or limi...
double Tolerance() const
absolute tolerance
Definition: Minimizer.h:420
void SetMaxIterations(unsigned int maxiter)
set maximum iterations (one iteration can have many function calls)
Definition: Minimizer.h:451
unsigned int MaxIterations() const
max iterations
Definition: Minimizer.h:417
void SetPrintLevel(int level)
set print level
Definition: Minimizer.h:445
int PrintLevel() const
minimizer configuration parameters
Definition: Minimizer.h:411
MultiNumGradFunction class to wrap a normal function in a gradient function using numerical gradient ...
Double_t x[n]
Definition: legend1.C:17
Namespace for new Math classes and functions.
BasicFitMethodFunction< ROOT::Math::IMultiGenFunction > FitMethodFunction
Definition: Fitter.h:40
IMultiGenFunctionTempl< double > IMultiGenFunction
Definition: IFunctionfwd.h:32
Namespace for new ROOT classes and functions.
Definition: StringConv.hxx:21