Class Scine::Utils::Optimizer

class Optimizer

The base class for all Optimizers.

Given the fact that optimizers can function in many different ways, this base class is somewhat slim and only unifies the usage of auxiliary functions.

In general the optimizers are intended to be used for internal optimizations where a fixed optimizer and a fixed set of options for that optimizer are used, and also those cases where user settings at runtime shall be used.

To this end, the optimizers should have their options given as public member variables which can alternatively be set using the

Optimizer::applySettings function. In order to then allow the addition of the encoded options into a Settings object a Optimizer::addSettingsDescriptors() function has to be generated in any derived class.

Furthermore all optimizers shall track the number of cycles they run and also call the

Optimizer::triggerObservers() function once in each iteration.

The specific function handling the optimization is dependent on the particular type of optimizer, however it should have the following properties:

  • it should return the final cycle count upon completion

  • it should accept the (in this order):

    1. the initial parameters

    2. a template lambda function for the update of needed variables

    3. an object derived from ConvergenceCheck to signal the end of the optimization

Subclassed by Scine::Utils::Bfgs, Scine::Utils::Bofill, Scine::Utils::Dimer, Scine::Utils::EigenvectorFollowing, Scine::Utils::Lbfgs, Scine::Utils::LevenbergMarquardt, Scine::Utils::NewtonRaphson, Scine::Utils::SteepestDescent

Public Functions

Optimizer()

The Optimizer base class can be default constructed.

void addObserver(std::function<void(const int&, const double&, const Eigen::VectorXd&)> function)

Add an observer function that will be triggered in each iteration.

Parameters

  • function: A function to be executed in every loop of the optimization. The function will have access to the current cycle count, the current value and to a const reference of the current parameters.

void clearObservers()

Clear all existing observer functions.

For optimization problems with very fast evaluations of the underlying function the removal of all observers can increase performance as the observers are given as std::functions and can not be added via templates.

virtual void addSettingsDescriptors(UniversalSettings::DescriptorCollection &collection) const = 0

Adds all relevant options to the given UniversalSettings::DescriptorCollection thus expanding it to include these optimizer options.

This function has to be implemented in derived classes in order to allow for all optimizers to be used and configured at runtime by end users.

Parameters

  • collection: The DescriptorCollection to which new fields shall be added.

virtual void applySettings(const Settings &settings) = 0

Updates the optimizers options with those values given in the Settings.

This function has to be implemented in derived classes in order to allow for all optimizers to be used and configured at runtime by end users.

Note: Additional settings that are not used by the optimizer shall not generate errors, or warnings, they shall simply be ignored.

Parameters

  • settings: The settings to update the option of this optimizer with.

Protected Functions

void triggerObservers(const int &cycle, const double &value, const Eigen::VectorXd &parameters)

Triggers all registered observers in sequence.

Parameters

  • cycle: The current cycle counter.

  • value: The current value.

  • parameters: The current parameters as optimized thus far.