ultra::src::basic_search< ES, E > Class Template Reference

#include <search.h>

Inheritance diagram for ultra::src::basic_search< ES, E >:

Public Types
using	individual_t = typename ultra::basic_search<ES, E>::individual_t
using	fitness_t = typename ultra::basic_search<ES, E>::fitness_t
Public Types inherited from ultra::basic_search< ES, E >
using	individual_t = evaluator_individual_t<E>
using	fitness_t = evaluator_fitness_t<E>
using	after_generation_callback_t
using	refinement_callback_t
using	on_training_new_best_callback_t

Public Member Functions
	basic_search (problem &, metric_flags=metric_flags::accuracy)
auto	predictor (const individual_t &) const
bool	is_valid () const override
Public Member Functions inherited from ultra::basic_search< ES, E >
	basic_search (problem &, E)
search_stats< individual_t, fitness_t >	run (unsigned=1, const model_measurements< fitness_t > &={})
basic_search &	after_generation (after_generation_callback_t)
basic_search &	logger (search_log &) noexcept
basic_search &	refinement (refinement_callback_t)
basic_search &	on_training_new_best (on_training_new_best_callback_t)
basic_search &	messages (bool) noexcept
basic_search &	tag (const std::string &)
basic_search &	stop_source (std::stop_source)
template<ValidationStrategy, class... Args>
basic_search &	validation_strategy (Args &&...)
basic_search &	validation_strategy (const ultra::validation_strategy &)
template<ValidationStrategy V, class... Args>
basic_search< ES, E > &	validation_strategy (Args &&...args)
template<ValidationStrategy V, class... Args>
basic_search< ES, E > &	validation_strategy (Args &&...args)
template<template< class > class ES, Evaluator E>
basic_search< ES, E > &	validation_strategy (const ultra::validation_strategy &vs)

Protected Member Functions
model_measurements< fitness_t >	calculate_metrics (const individual_t &) const override
void	tune_parameters () override
Protected Member Functions inherited from ultra::basic_search< ES, E >
virtual void	after_evolution (unsigned, const std::vector< model_measurements< fitness_t > > &)

Friends
template<Individual>
class	search

Additional Inherited Members
Protected Attributes inherited from ultra::basic_search< ES, E >
evaluator_proxy< E >	eva_
std::unique_ptr< ultra::validation_strategy >	vs_ {std::make_unique<as_is_validation>()}
problem &	prob_
after_generation_callback_t	after_generation_callback_ {}
refinement_callback_t	refinement_callback_ {}
on_training_new_best_callback_t	on_training_new_best_callback_ {}

Detailed Description

template<template< class > class ES, Evaluator E>
class ultra::src::basic_search< ES, E >

Drives the search for solutions of symbolic regression / classification tasks.

Constructor & Destructor Documentation

basic_search()

template<template< class > class ES, Evaluator E>

basic_search::basic_search ( problem & p, metric_flags m = metric_flags::accuracy )

explicit

Builds a new search class, user chooses the evaluator.

Parameters

[in]	p	the problem we're working on. The lifetime of p must exceed the lifetime of this class
[in]	m	a bit field used to specify metrics we have to calculate while searching

Member Function Documentation

calculate_metrics()

template<template< class > class ES, Evaluator E>

model_measurements< typename basic_search< ES, E >::fitness_t > basic_search::calculate_metrics ( const individual_t & prg ) const

nodiscardoverrideprotectedvirtual

Calculates various performance metrics.

Parameters

[in]

prg

best individual from the evolution run just finished

Returns

measurements about the individual

Fitness and accuracy are calculated by default. Additional metrics must be explicitly requested in the basic_search constructor.

Warning

Can be very time consuming.

Reimplemented from ultra::basic_search< ES, E >.

is_valid()

template<template< class > class ES, Evaluator E>

bool ultra::src::basic_search< ES, E >::is_valid ( ) const

nodiscardoverridevirtual

Returns

true if the object passes the internal consistency check

Reimplemented from ultra::basic_search< ES, E >.

predictor()

template<template< class > class ES, Evaluator E>

auto basic_search::predictor ( const individual_t & ind ) const

nodiscard

Creates a predictor associated with a given individual.

Parameters

[in]

ind

individual used as building block for the predictor

Returns

the evaluator-specific predictor, or no_predictor if the active evaluator does not provide one

The returned object depends on the evaluator used during training:

• if the evaluator exposes a predictor() method, its result is forwarded;
• otherwise, a no_predictor sentinel is returned.

This function is always well-formed: lack of predictor support is handled at compile time via the no_predictor fallback rather than via runtime errors.

tune_parameters()

template<template< class > class ES, Evaluator E>

void ultra::src::basic_search< ES, E >::tune_parameters ( )

overrideprotectedvirtual

Tries to tune search parameters for the current problem.

Parameter tuning is a typical approach to algorithm design. Such tuning is done by experimenting with different values and selecting the ones that give the best results on the test problems at hand.

However, the number of possible parameters and their different values means that this is a very complex and time-consuming task; it is something we do not want users to worry about (power users can force many parameters, but our idea is "simple by default").

So if user sets an environment parameter he will force the search class to use it as is. Otherwise this function will try to guess a good starting point and changes its hint after every run. The code is a mix of black magic, experience, common logic and randomness but it seems reasonable.

Note

It has been formally proven, in the No-Free-Lunch theorem, that it is impossible to tune a search algorithm such that it will have optimal settings for all possible problems, but parameters can be properly set for a given problem.

See also

• https://github.com/morinim/ultra/wiki/bibliography#11
• https://github.com/morinim/ultra/wiki/bibliography#12

Reimplemented from ultra::basic_search< ES, E >.

---

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

• gp/src/search.h
• gp/src/search.tcc