Class OptimizationReporter

Inheritance Relationships

Base Type

Derived Type

Class Documentation

class OptimizationReporter : public parpe::GradientFunction

The OptimizationReporter class is called from the optimizer and takes care of calling the actual objective function, thereby keeping track of iterations, computation time, logging intermediate results, timing and can tell the optimizer to exit.

This extra level of abstraction is added to avoid reimplementing timing, and other things for each supported optimizer. The indirection of cost function evaluation is added to allow caching previous cost function values.

Subclassed by parpe::HierarchicalOptimizationReporter

Public Functions

OptimizationReporter(GradientFunction *gradFun, std::unique_ptr<Logger> logger)
OptimizationReporter(GradientFunction *gradFun, std::unique_ptr<OptimizationResultWriter> rw, std::unique_ptr<Logger> logger)
virtual FunctionEvaluationStatus evaluate(gsl::span<double const> parameters, double &fval, gsl::span<double> gradient, Logger *logger = nullptr, double *cpuTime = nullptr) const override
virtual int numParameters() const override
virtual bool starting(gsl::span<const double> initialParameters) const

Is called just before the optimizer starts. Must be called before other functions.

Parameters:

initialParameters –

Returns:

Quit optimization?

virtual bool iterationFinished(gsl::span<const double> parameters, double objectiveFunctionValue, gsl::span<const double> objectiveFunctionGradient) const

Is called after each iteration except for the last one.

Parameters:

  • parameters –

  • objectiveFunctionValue –

  • objectiveFunctionGradient –

Returns:

Quit optimization?

virtual bool beforeCostFunctionCall(gsl::span<const double> parameters) const
virtual bool afterCostFunctionCall(gsl::span<const double> parameters, double objectiveFunctionValue, gsl::span<double const> objectiveFunctionGradient) const
virtual void finished(double optimalCost, gsl::span<const double> parameters, int exitStatus) const

Is called after optimization finished.

virtual double getFinalCost() const
virtual std::vector<double> const &getFinalParameters() const
void setGradientFunction(GradientFunction *gradFun) const
virtual std::vector<std::string> getParameterIds() const override

Public Members

std::unique_ptr<OptimizationResultWriter> result_writer_
mutable double cpu_time_total_sec_ = 0.0
mutable double cpu_time_iteration_sec_ = 0.0
std::unique_ptr<Logger> logger_

Protected Functions

void printObjectiveFunctionFailureMessage() const

Protected Attributes

mutable WallTimer wall_timer_
mutable int num_function_calls_ = 0
mutable int num_iterations_ = 0
mutable int num_parameters_ = 0
mutable bool started_ = false
mutable GradientFunction *grad_fun_ = nullptr
mutable bool have_cached_cost_ = false
mutable bool have_cached_gradient_ = false
mutable std::vector<double> cached_gradient_
mutable double cached_cost_ = std::numeric_limits<double>::infinity()
mutable FunctionEvaluationStatus cached_status_ = functionEvaluationSuccess
mutable double final_cost_ = std::numeric_limits<double>::quiet_NaN()
mutable std::vector<double> cached_parameters_
std::string default_logger_prefix_