gp_opt Class Reference

#include <gp.hpp>

Inheritance diagram for gp_opt:

Public Member Functions
void	update_search_space (sspace::sspace_t *space) override
inst::set	generate_parameters (int param_id) override
void	receive_trial_results (int param_id, inst::set params, double value) override
Public Member Functions inherited from optk::optimiser
	optimiser (std::string name)
	~optimiser ()
virtual void	clear ()
std::string	get_name ()
void	accept (optimisers *o)

Private Attributes
uint	n_iters
sspace::sspace_t *	m_space

Additional Inherited Members
Protected Member Functions inherited from optk::optimiser
void	add_to_trials (int param_id, inst::node *n)
Protected Attributes inherited from optk::optimiser
std::unordered_map< int, inst::set >	trials

Detailed Description

gp_opt is a generic Gaussian process-based optimiser which is only implemented to provide a baseline against other methods.

Member Function Documentation

generate_parameters()

inst::set gp_opt::generate_parameters ( int  param_id )

overridevirtual

This is the method invoked to generate the next trial's parameters for the benchmark or program being optimised.

Parameters

param_id

An identifier for this parameter trial; this may or may not be used by the optimistaion algorithm.

Returns

A pointer to an inst::node, or a NULL pointer if there are no more parameters to return.

Reimplemented from optk::optimiser.

receive_trial_results()

void gp_opt::receive_trial_results ( int  param_id, inst::set  params, double  value  )

overridevirtual

Allows the benchmark or program being optimised to communicate back the value of the cost function for a parameter set.

Parameters

param_id	This parameter id is used to identify the parameters used to generate the result.
parameters	For increased flexibility, the actual parameters used to generate the result are also copied back.
value	The actual value of the objective function.

Reimplemented from optk::optimiser.

update_search_space()

void gp_opt::update_search_space ( sspace::sspace_t *  space )

overridevirtual

The GP optimiser is only compatible with continuous valued inputs (for the moment); conceretely an error will be raised if the input is not pt::uniform, pt::loguniform, pt::normal, pt::lognormal. Currently all parameter types are treated as pt::uniform, although in the future information about parameters' distributions may be taken into account.

Parameters

space

The new search space.

Reimplemented from optk::optimiser.

Member Data Documentation

m_space

sspace::sspace_t* gp_opt::m_space

private

A copy of the problem search space

n_iters

uint gp_opt::n_iters

private

Counts the number of iterations performed

---

The documentation for this class was generated from the following files:

• includes/optimisers/gp.hpp
• src/optimisers/gp.cpp