ultra::src::aggregate_evaluator< P, F, D, A, M > Class Template Reference

Inheritance diagram for ultra::src::aggregate_evaluator< P, F, D, A, M >:

Public Member Functions
	aggregate_evaluator (D &) noexcept
bool	set_parallel_evaluation (unsigned, std::size_t=0) noexcept
double	operator() (const P &) const
double	fast (const P &) const
auto	predictor (const P &) const

Static Public Attributes
static constexpr std::ptrdiff_t	fast_step = 5
Static Public Attributes inherited from ultra::src::evaluator< D >
static constexpr std::size_t	default_min_samples_per_thread = 512

Additional Inherited Members
Protected Member Functions inherited from ultra::src::evaluator< D >
	evaluator (D &) noexcept
bool	set_parallel_evaluation (unsigned, std::size_t=0) noexcept
const auto &	data () const noexcept
std::size_t	min_samples_per_thread () const noexcept
unsigned	parallel_helpers () const noexcept

Constructor & Destructor Documentation

aggregate_evaluator()

template<Individual P, class F, class D, aggregation_mode A, evaluation_mode M>
requires ExampleEvaluator<F, D, P>

aggregate_evaluator::aggregate_evaluator ( D & d )

explicitnoexcept

Constructs the evaluator.

Parameters

[in]

d

training dataset

Member Function Documentation

fast()

template<Individual P, class F, class D, aggregation_mode A, evaluation_mode M>
requires ExampleEvaluator<F, D, P>

double aggregate_evaluator::fast ( const P & prg ) const

nodiscard

Computes a (possibly approximate) fitness value using subsampling.

Parameters

[in]

prg

program to evaluate

Returns

fitness value (exact or approximate)

For sufficiently large datasets, the program is evaluated on a subset of the examples (one every step elements) to reduce computation time.

For smaller datasets, the full evaluation is performed, making this function equivalent to operator()().

For average-based evaluators, subsampling yields an estimate of the true average. For sum-based evaluators, the result is rescaled to remain comparable with operator()().

operator()()

template<Individual P, class F, class D, aggregation_mode A, evaluation_mode M>
requires ExampleEvaluator<F, D, P>

double aggregate_evaluator::operator() ( const P & prg ) const

nodiscard

Computes the fitness using all training examples.

Parameters

[in]

prg

program to evaluate

Returns

fitness value (greater is better)

predictor()

template<Individual P, class F, class D, aggregation_mode A, evaluation_mode M>
requires internal::has_regression_predictor_v<F, P>

auto aggregate_evaluator::predictor ( const P & prg ) const

nodiscard

Builds a regression predictor associated with a program.

Parameters

[in]

prg

program to transform into a predictor

Returns

predictor object associated with prg

Remarks

This member function is conditionally enabled only when the functor F used by the evaluator is known to support regression semantics (as indicated by has_regression_predictor<F, P>). It is not available for classification or generic score-based evaluators. Computes fitness from sampled examples.

Parameters

[in]	prg	program to evaluate
[in]	step	sampling step (one example every step)

Returns

fitness value

Values from valid examples are averaged. Missing values incur a heuristic penalty based on the ratio of missing to valid examples. A partially valid program may still outrank a fully valid one. If all values are missing, the function returns std::numeric_limits<double>::lowest().

For opted-in functors and forward-range datasets with enough sampled examples, evaluation is divided into strips across the calling thread and configured helper threads. Each strip constructs its own example functor. Their means are combined by valid-example count before the missing-value penalty and sum scaling are applied. Parallel evaluation may change floating-point rounding. Sets up the environment for error measurement.

Parameters

[in]

prg

the program to be measured

set_parallel_evaluation()

template<Individual P, class F, class D, aggregation_mode A, evaluation_mode M>
requires parallel_safe_example_evaluator<F>

bool aggregate_evaluator::set_parallel_evaluation ( unsigned n, std::size_t min_samples = 0 )

noexcept

Sets the number of helper threads and workload threshold for parallel evaluation.

Parameters

[in]	n	number of helper threads; zero disables parallel evaluation
[in]	min_samples	minimum number of samples per thread; zero uses the default threshold

Returns

true after applying the configuration

---

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

• gp/src/evaluator.h
• gp/src/evaluator.tcc