tc::autotune::GeneticSearch Class Reference

#include <genetic_search.h>

Collaboration diagram for tc::autotune::GeneticSearch:

Public Types
using	Population = std::vector< std::unique_ptr< CandidateConfiguration >>

Public Member Functions
	GeneticSearch (const TuningConfiguration &conf, size_t n, uint8_t crossOverRate, uint8_t mutationRate, size_t numberElites)
	GeneticSearch (const std::vector< TuningConfiguration > &confs, size_t n, uint8_t crossOverRate, uint8_t mutationRate, size_t numberElites)
void	updateParameters ()

Public Attributes
Population	population
TuningConfiguration	lastBestConf
const size_t	kMaxPopulationSize
const uint8_t	kCrossOverRate
const uint8_t	kMutationRate
const size_t	kNumberElites
std::mt19937_64	rng

Static Public Attributes
static constexpr int	kMutateIterations = 1000
static constexpr int	kMinCandidatesForBreeding = 3

Private Member Functions
void	breed ()
TuningConfiguration	crossover (TuningConfiguration &, TuningConfiguration &, TuningConfiguration &) const

Detailed Description

GeneticSearch operates on generations of candidate solutions. Each generation consists of multiple candidate solutions. Before a new generation can be generated all candidates of the current one are assigned a fitness value (1 / runtime (in us) in our case).

The top candidates are preserved intact across generation (elitism).

Each new candidate is generated with the following two mechanisms:

crossover: 3 parent candidates are selected probabilistically (influenced by their fitness, the higher it is the higher the chance of selection) and merged into a new candidate

mutation: parts of a candidate are randomly changed (mutated)

The crossover rate controls how many candidates survive across generations and how many new candidates are produced through crossover, e.g, 80% crossover rate would result in roughly 20% of the new generation to consist of candidates that survived from the previous one. Note, that this is different from elitism: elite candidates are not mutated but "survivors" are.

The mutation rate controls the probability with which mutation occurs.

Member Typedef Documentation

using tc::autotune::GeneticSearch::Population = std::vector<std::unique_ptr<CandidateConfiguration>>

Constructor & Destructor Documentation

tc::autotune::GeneticSearch::GeneticSearch	(	const TuningConfiguration &	conf,
		size_t	n,
		uint8_t	crossOverRate,
		uint8_t	mutationRate,
		size_t	numberElites
	)

conf is used to determine which are the tunable parameters, the selected values in conf are ignored and the population is randomized

n is the population size

crossoverRate is the probability ([0,100]) that a new candidate is produced through reproduction

mutationRate is the probability ([0,100]) that parameters are mutated (randomly changed) whenever a new generation is created

numberElites best candidates are preserved across generations (elitism), number Elites must be less than n

tc::autotune::GeneticSearch::GeneticSearch	(	const std::vector< TuningConfiguration > &	confs,
		size_t	n,
		uint8_t	crossOverRate,
		uint8_t	mutationRate,
		size_t	numberElites
	)

confs are used to seed the first generation, the rest of the population is randomly initialized

n is the population size

crossoverRate is the probability ([0,100]) that a new candidate is produced through reproduction

mutationRate is the probability ([0,100]) that parameters are mutated (randomly changed) whenever a new generation is created

numberElites best candidates are preserved across generations (elitism), number Elites must be less than n

Member Function Documentation

void tc::autotune::GeneticSearch::breed ( )

private

TuningConfiguration tc::autotune::GeneticSearch::crossover ( TuningConfiguration &  , TuningConfiguration &  , TuningConfiguration &    ) const

private

void tc::autotune::GeneticSearch::updateParameters

(

)

Member Data Documentation

const uint8_t tc::autotune::GeneticSearch::kCrossOverRate

const size_t tc::autotune::GeneticSearch::kMaxPopulationSize

constexpr int tc::autotune::GeneticSearch::kMinCandidatesForBreeding = 3

static

constexpr int tc::autotune::GeneticSearch::kMutateIterations = 1000

static

const uint8_t tc::autotune::GeneticSearch::kMutationRate

const size_t tc::autotune::GeneticSearch::kNumberElites

TuningConfiguration tc::autotune::GeneticSearch::lastBestConf

Population tc::autotune::GeneticSearch::population

std::mt19937_64 tc::autotune::GeneticSearch::rng

mutable

---

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

• include/tc/autotuner/genetic_search.h