Minuit2Minimizer.cpp

Go to the documentation of this file.

// ************************************************************************** //
//
//  BornAgain: simulate and fit scattering at grazing incidence
//
//! @file      Fit/RootAdapter/Minuit2Minimizer.cpp
//! @brief     Declares class Minuit2Minimizer.
//!
//! @homepage  http://www.bornagainproject.org
//! @license   GNU General Public License v3 or higher (see COPYING)
//! @copyright Forschungszentrum Jülich GmbH 2018
//! @authors   Scientific Computing Group at MLZ (see CITATION, AUTHORS)
//
// ************************************************************************** //
 
#include "Fit/RootAdapter/Minuit2Minimizer.h"
#include "Fit/Tools/StringUtils.h"
#include <Minuit2/Minuit2Minimizer.h>
 
namespace
{
 
std::map<int, std::string> statusDescription()
{
    std::map<int, std::string> result;
    result[0] = "OK, valid minimum";
    result[1] = "Didn't converge, covariance was made pos defined";
    result[2] = "Didn't converge, Hessian is invalid";
    result[3] = "Didn't converge, Edm is above max";
    result[4] = "Didn't converge, reached call limit";
    result[5] = "Didn't converge, unknown failure";
    return result;
}
 
std::map<int, std::string> covmatrixStatusDescription()
{
    std::map<int, std::string> result;
    result[-1] = "Not available (inversion failed or Hessian failed)";
    result[0] = "Available but not positive defined";
    result[1] = "Covariance only approximate";
    result[2] = "Full matrix but forced pos def";
    result[3] = "Full accurate";
    return result;
}
 
} // namespace
 
Minuit2Minimizer::Minuit2Minimizer(const std::string& algorithmName)
    : RootMinimizerAdapter(MinimizerInfo::buildMinuit2Info(algorithmName)),
      m_minuit2_minimizer(new ROOT::Minuit2::Minuit2Minimizer(algorithmName.c_str()))
{
    addOption("Strategy", 1, "Minimization strategy (0-low, 1-medium, 2-high quality)");
    addOption("ErrorDef", 1.0, "Error definition factor for parameter error calculation");
    addOption("Tolerance", 0.01, "Tolerance on the function value at the minimum");
    addOption("Precision", -1.0, "Relative floating point arithmetic precision");
    addOption("PrintLevel", 0, "Minimizer internal print level");
    addOption("MaxFunctionCalls", 0, "Maximum number of function calls");
}
 
Minuit2Minimizer::~Minuit2Minimizer() = default;
 
void Minuit2Minimizer::setStrategy(int value)
{
    setOptionValue("Strategy", value);
}
 
int Minuit2Minimizer::strategy() const
{
    return optionValue<int>("Strategy");
}
 
void Minuit2Minimizer::setErrorDefinition(double value)
{
    setOptionValue("ErrorDef", value);
}
 
double Minuit2Minimizer::errorDefinition() const
{
    return optionValue<double>("ErrorDef");
}
 
void Minuit2Minimizer::setTolerance(double value)
{
    setOptionValue("Tolerance", value);
}
 
double Minuit2Minimizer::tolerance() const
{
    return optionValue<double>("Tolerance");
}
 
void Minuit2Minimizer::setPrecision(double value)
{
    setOptionValue("Precision", value);
}
 
double Minuit2Minimizer::precision() const
{
    return optionValue<double>("Precision");
}
 
void Minuit2Minimizer::setPrintLevel(int value)
{
    setOptionValue("PrintLevel", value);
}
 
int Minuit2Minimizer::printLevel() const
{
    return optionValue<int>("PrintLevel");
}
 
void Minuit2Minimizer::setMaxFunctionCalls(int value)
{
    setOptionValue("MaxFunctionCalls", value);
}
 
int Minuit2Minimizer::maxFunctionCalls() const
{
    return optionValue<int>("MaxFunctionCalls");
}
 
std::string Minuit2Minimizer::statusToString() const
{
    return statusDescription()[rootMinimizer()->Status()];
}
 
std::map<std::string, std::string> Minuit2Minimizer::statusMap() const
{
    auto result = RootMinimizerAdapter::statusMap();
    result["Edm"] = StringUtils::scientific(rootMinimizer()->Edm());
    result["CovMatrixStatus"] = covmatrixStatusDescription()[rootMinimizer()->CovMatrixStatus()];
    result["functionCalls"] = std::to_string(rootMinimizer()->NCalls());
    return result;
}
 
// Fumili algorithm can work only with gradient based objective function, while others can
// work with both, gradient based and chi2 based functions. Historically however, we use
// simplified approach: if not Fumili, then chi2 only. Think of refactoring TODO.
bool Minuit2Minimizer::requiresResiduals()
{
    return algorithmName() == "Fumili";
}
 
//! Propagate options down to ROOT's Minuit2Minimizer.
 
void Minuit2Minimizer::propagateOptions()
{
    m_minuit2_minimizer->SetStrategy(strategy());
    m_minuit2_minimizer->SetErrorDef(errorDefinition());
    m_minuit2_minimizer->SetTolerance(tolerance());
    m_minuit2_minimizer->SetPrecision(precision());
    m_minuit2_minimizer->SetPrintLevel(printLevel());
    m_minuit2_minimizer->SetMaxFunctionCalls(static_cast<unsigned int>(maxFunctionCalls()));
}
 
const RootMinimizerAdapter::root_minimizer_t* Minuit2Minimizer::rootMinimizer() const
{
    return m_minuit2_minimizer.get();
}