44 print.
Warn(
"No variable parameters are defined! - Return current function value ");
50 print.
Debug(
"initial edm is ", edm);
56 print.
Error(
"Initial matrix not positive defined, edm = ",edm,
"\nExit minimization ");
60 std::vector<MinimumState>
result;
87 print.
Warn(
"FunctionMinimum is invalid");
96 print.
Debug(
"Approximate Edm", edm,
"npass",
ipass);
101 print.
Debug(
"FumiliBuilder will verify convergence and Error matrix; "
103 min.Error().Dcovar());
114 print.
Info(
"After Hessian");
121 print.
Debug(
"Edm", edm,
"State",
st);
126 print.
Warn(
"Stop iterations, no improvements after Hesse; current Edm", edm,
"previous value",
edmprev);
130 print.
Debug(
"Tolerance not sufficient, continue minimization; Edm", edm,
"Requested",
edmval);
134 if (min.IsAboveMaxEdm()) {
199 MnPrint print(
"FumiliBuilder");
221 double delta = 0.3 * std::max(1.0 ,
normX0);
222 const double eta = 0.1;
229 print.
Info(
"Using Fumili with a line search algorithm");
245 step = -1. *
s0.Error().InvHessian() *
s0.Gradient().Vec();
247 print.
Debug(
"Iteration -",
result.size(),
"\n Fval",
s0.Fval(),
"numOfCall",
fcn.NumOfCalls(),
248 "\n Internal Parameter values",
s0.Vec(),
"\n Newton step", step);
253 print.
Warn(
"Matrix not pos.def, gdel =",
gdel,
" > 0");
257 step = -1. *
s0.Error().InvHessian() *
s0.Gradient().Vec();
260 print.
Warn(
"After correction, gdel =",
gdel);
277 print.
Debug(
"Do a line search",
fcn.NumOfCalls());
281 if (std::fabs(pp.Y() -
s0.Fval()) <
prec.Eps()) {
286 print.
Debug(
"New point after Line Search :",
"\n FVAL ",
p.Fval(),
"\n Parameter",
p.Vec());
289 auto &
H =
s0.Error().Hessian();
290 unsigned int n = (
scaleTR) ?
H.Nrow() : 0;
297 for (
unsigned int i = 0; i <
n; i++){
298 double d = std::sqrt(
H(i,i));
309 print.
Debug(
"scaling Trust region with diagonal matrix D ",D);
327 print.
Debug(
"Accept full Newton step - it is inside TR ",delta);
335 auto gScaled = Dinv *
s0.Gradient().Grad();
339 for (
unsigned int i = 0; i <
n; i++) {
340 for (
unsigned int j = 0;
j <=i;
j++) {
356 step = - (delta/
normGrad) *
s0.Gradient().Grad();
358 print.
Debug(
"Use as new point the Cauchy point - along gradient with norm=delta ", delta);
368 print.
Debug(
"Use as new point the Cauchy point - along gradient with tau ", tau,
"delta = ", delta);
380 print.
Debug(
" dogleg equation",
a,
b,
c);
386 print.
Warn(
"a is equal to zero! a = ",
a);
387 print.
Info(
" delta ", delta,
" tau ", tau,
" gHg ",
gHg,
" normgrad2 ",
normGrad2);
390 double t1 = (-
b + sqrt(
b *
b - 4. *
a *
c)) / (2.0 *
a);
391 double t2 = (-
b - sqrt(
b *
b - 4. *
a *
c)) / (2.0 *
a);
393 print.
Debug(
" solution dogleg equation",
t1,
t2);
394 if (
t1 >= 0 &&
t1 <= 1.)
401 print.
Debug(
"New dogleg point is t = ", t);
403 print.
Debug(
"New accepted step is ",step);
420 if (rho > 0.75 &&
norm == delta) {
424 print.
Debug(
"New point after Trust region :",
"norm tr ",
norm,
" rho ", rho,
" delta ", delta,
425 " FVAL ",
p.Fval(),
"\n Parameter",
p.Vec());
432 print.
Debug(
"Trust region: accept new point p = x + step since rho is larger than eta");
436 print.
Debug(
"Trust region reject new point and repeat since rho is smaller than eta");
446 print.
Debug(
"Before Gradient - NCalls = ",
fcn.NumOfCalls());
450 print.
Debug(
"After Gradient - NCalls = ",
fcn.NumOfCalls());
461 print.
Debug(
"Updated new point:",
"\n FVAL ",
p.Fval(),
"\n Parameter",
p.Vec(),
"\n Gradient",
g.Vec(),
462 "\n InvHessian",
e.InvHessian(),
"\n Hessian",
e.Hessian(),
"\n Edm", edm);
465 print.
Warn(
"Matrix not pos.def., Edm < 0");
480 if (
p.Fval() <
s0.Fval())
491 print.
Debug(
"finish iteration -",
result.size(),
"lambda =", lambda,
"f1 =",
p.Fval(),
"f0 =",
s0.Fval(),
492 "num of calls =",
fcn.NumOfCalls(),
"edm =", edm);
500 edm *= (1. + 3. *
e.Dcovar());
514 if (edm < std::fabs(
prec.Eps2() *
result.back().Fval())) {
515 print.
Warn(
"Machine accuracy limits further improvement");
518 }
else if (edm < 10 *
edmval) {
522 print.
Warn(
"No convergence; Edm", edm,
"is above tolerance", 10 *
edmval);
528 print.
Debug(
"Exiting successfully",
"Ncalls",
fcn.NumOfCalls(),
"FCN",
result.back().Fval(),
"Edm", edm,
"Requested",
ROOT::Detail::TRangeCast< T, true > TRangeDynCast
TRangeDynCast is an adapter class that allows the typed iteration through a TCollection.
winID h TVirtualViewer3D TVirtualGLPainter p
Option_t Option_t TPoint TPoint const char GetTextMagnitude GetFillStyle GetLineColor GetLineWidth GetMarkerStyle GetTextAlign GetTextColor GetTextSize void char Point_t Rectangle_t WindowAttributes_t Float_t Float_t Float_t Int_t Int_t UInt_t UInt_t Rectangle_t result
Option_t Option_t TPoint TPoint const char GetTextMagnitude GetFillStyle GetLineColor GetLineWidth GetMarkerStyle GetTextAlign GetTextColor GetTextSize void gc
FunctionMinimum Minimum(const MnFcn &fMnFcn, const GradientCalculator &fGradienCalculator, const MinimumSeed &fMinimumSeed, const MnStrategy &fMnStrategy, unsigned int maxfcn, double edmval) const override
Class the member function calculating the Minimum and verifies the result depending on the strategy.
FumiliMethodType fMethodType
const FumiliErrorUpdator & ErrorUpdator() const
Accessor to the Error updator of the builder.
const VariableMetricEDMEstimator & Estimator() const
Accessor to the EDM (expected vertical distance to the Minimum) estimator.
class holding the full result of the minimization; both internal and external (MnUserParameterState) ...
interface class for gradient calculators
Class describing a symmetric matrix of size n.
unsigned int size() const
void TraceIteration(int iter, const MinimumState &state) const
MinimumError keeps the inv.
const FunctionGradient & Gradient() const
const MinimumError & Error() const
const MinimumParameters & Parameters() const
const MnMachinePrecision & Precision() const
const MinimumState & State() const
MinimumState keeps the information (position, Gradient, 2nd deriv, etc) after one minimization step (...
Wrapper class to FCNBase interface used internally by Minuit.
API class for calculating the numerical covariance matrix (== 2x Inverse Hessian == 2x Inverse 2nd de...
Implements a 1-dimensional minimization along a given direction (i.e.
Sets the relative floating point (double) arithmetic precision.
Force the covariance matrix to be positive defined by adding extra terms in the diagonal.
void Debug(const Ts &... args)
void Error(const Ts &... args)
void Info(const Ts &... args)
void Warn(const Ts &... args)
API class for defining four levels of strategies: low (0), medium (1), high (2), very high (>=3); act...
double similarity(const LAVector &, const LASymMatrix &)
double inner_product(const LAVector &, const LAVector &)
tbb::task_arena is an alias of tbb::interface7::task_arena, which doesn't allow to forward declare tb...