[yul-phaser] Add GenerationalElitistWithExclusivePools algorithm

This commit is contained in:
Kamil Śliwak 2020-02-06 00:51:11 +01:00
parent 7e80ac861f
commit fc4fedb214
3 changed files with 182 additions and 0 deletions

View File

@ -29,6 +29,7 @@
#include <boost/test/unit_test.hpp>
#include <boost/test/tools/output_test_stream.hpp>
#include <algorithm>
#include <vector>
using namespace std;
@ -132,6 +133,107 @@ BOOST_FIXTURE_TEST_CASE(runNextRound_should_not_replace_any_chromosomes_if_whole
BOOST_TEST((chromosomeLengths(algorithm.population()) == vector<size_t>{3, 3, 3, 3, 5, 5, 5, 5}));
}
BOOST_AUTO_TEST_SUITE_END()
BOOST_AUTO_TEST_SUITE(GenerationalElitistWithExclusivePoolsTest)
BOOST_FIXTURE_TEST_CASE(runNextRound_should_preserve_elite_and_regenerate_rest_of_population, GeneticAlgorithmFixture)
{
auto population = Population::makeRandom(m_fitnessMetric, 6, 3, 3) + Population::makeRandom(m_fitnessMetric, 4, 5, 5);
GenerationalElitistWithExclusivePools::Options options = {
/* mutationPoolSize = */ 0.2,
/* crossoverPoolSize = */ 0.2,
/* randomisationChance = */ 0.0,
/* deletionVsAdditionChance = */ 1.0,
/* percentGenesToRandomise = */ 0.0,
/* percentGenesToAddOrDelete = */ 1.0,
};
GenerationalElitistWithExclusivePools algorithm(population, m_output, options);
assert((chromosomeLengths(algorithm.population()) == vector<size_t>{3, 3, 3, 3, 3, 3, 5, 5, 5, 5}));
algorithm.runNextRound();
BOOST_TEST((chromosomeLengths(algorithm.population()) == vector<size_t>{0, 0, 3, 3, 3, 3, 3, 3, 3, 3}));
}
BOOST_FIXTURE_TEST_CASE(runNextRound_should_not_replace_elite_with_worse_individuals, GeneticAlgorithmFixture)
{
auto population = Population::makeRandom(m_fitnessMetric, 6, 3, 3) + Population::makeRandom(m_fitnessMetric, 4, 5, 5);
GenerationalElitistWithExclusivePools::Options options = {
/* mutationPoolSize = */ 0.2,
/* crossoverPoolSize = */ 0.2,
/* randomisationChance = */ 0.0,
/* deletionVsAdditionChance = */ 0.0,
/* percentGenesToRandomise = */ 0.0,
/* percentGenesToAddOrDelete = */ 1.0,
};
GenerationalElitistWithExclusivePools algorithm(population, m_output, options);
assert(chromosomeLengths(algorithm.population()) == (vector<size_t>{3, 3, 3, 3, 3, 3, 5, 5, 5, 5}));
algorithm.runNextRound();
BOOST_TEST((chromosomeLengths(algorithm.population()) == vector<size_t>{3, 3, 3, 3, 3, 3, 3, 3, 7, 7}));
}
BOOST_FIXTURE_TEST_CASE(runNextRound_should_generate_individuals_in_the_crossover_pool_by_mutating_the_elite, GeneticAlgorithmFixture)
{
auto population = Population::makeRandom(m_fitnessMetric, 20, 5, 5);
GenerationalElitistWithExclusivePools::Options options = {
/* mutationPoolSize = */ 0.8,
/* crossoverPoolSize = */ 0.0,
/* randomisationChance = */ 0.5,
/* deletionVsAdditionChance = */ 0.5,
/* percentGenesToRandomise = */ 1.0,
/* percentGenesToAddOrDelete = */ 1.0,
};
GenerationalElitistWithExclusivePools algorithm(population, m_output, options);
SimulationRNG::reset(1);
algorithm.runNextRound();
BOOST_TEST((
chromosomeLengths(algorithm.population()) ==
vector<size_t>{0, 0, 0, 0, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 11, 11, 11}
));
}
BOOST_FIXTURE_TEST_CASE(runNextRound_should_generate_individuals_in_the_crossover_pool_by_crossing_over_the_elite, GeneticAlgorithmFixture)
{
auto population = (
Population(m_fitnessMetric, {Chromosome("aa"), Chromosome("ff")}) +
Population::makeRandom(m_fitnessMetric, 8, 6, 6)
);
GenerationalElitistWithExclusivePools::Options options = {
/* mutationPoolSize = */ 0.0,
/* crossoverPoolSize = */ 0.8,
/* randomisationChance = */ 0.0,
/* deletionVsAdditionChance = */ 0.0,
/* percentGenesToRandomise = */ 0.0,
/* percentGenesToAddOrDelete = */ 0.0,
};
GenerationalElitistWithExclusivePools algorithm(population, m_output, options);
assert((chromosomeLengths(algorithm.population()) == vector<size_t>{2, 2, 6, 6, 6, 6, 6, 6, 6, 6}));
SimulationRNG::reset(1);
algorithm.runNextRound();
vector<Individual> const& newIndividuals = algorithm.population().individuals();
BOOST_TEST((chromosomeLengths(algorithm.population()) == vector<size_t>{2, 2, 2, 2, 2, 2, 2, 2, 2, 2}));
for (auto& individual: newIndividuals)
BOOST_TEST((
individual.chromosome == Chromosome("aa") ||
individual.chromosome == Chromosome("af") ||
individual.chromosome == Chromosome("fa") ||
individual.chromosome == Chromosome("ff")
));
BOOST_TEST(any_of(newIndividuals.begin() + 2, newIndividuals.end(), [](auto& individual){
return individual.chromosome != Chromosome("aa") && individual.chromosome != Chromosome("ff");
}));
}
BOOST_AUTO_TEST_SUITE_END()
BOOST_AUTO_TEST_SUITE_END()
BOOST_AUTO_TEST_SUITE_END()

View File

@ -16,7 +16,9 @@
*/
#include <tools/yulPhaser/GeneticAlgorithms.h>
#include <tools/yulPhaser/Mutations.h>
#include <tools/yulPhaser/Selections.h>
#include <tools/yulPhaser/PairSelections.h>
using namespace std;
using namespace solidity::phaser;
@ -48,3 +50,28 @@ void RandomAlgorithm::runNextRound()
m_options.maxChromosomeLength
);
}
void GenerationalElitistWithExclusivePools::runNextRound()
{
double elitePoolSize = 1.0 - (m_options.mutationPoolSize + m_options.crossoverPoolSize);
RangeSelection elite(0.0, elitePoolSize);
m_population =
m_population.select(elite) +
m_population.select(elite).mutate(
RandomSelection(m_options.mutationPoolSize / elitePoolSize),
alternativeMutations(
m_options.randomisationChance,
geneRandomisation(m_options.percentGenesToRandomise),
alternativeMutations(
m_options.deletionVsAdditionChance,
geneDeletion(m_options.percentGenesToAddOrDelete),
geneAddition(m_options.percentGenesToAddOrDelete)
)
)
) +
m_population.select(elite).crossover(
RandomPairSelection(m_options.crossoverPoolSize / elitePoolSize / 2),
randomPointCrossover()
);
}

View File

@ -112,4 +112,57 @@ private:
Options m_options;
};
/**
* A generational, elitist genetic algorithm that replaces the population by mutating and crossing
* over chromosomes from the elite.
*
* The elite consists of individuals not included in the crossover and mutation pools.
* The crossover operator used is @a randomPointCrossover. The mutation operator is randomly chosen
* from three possibilities: @a geneRandomisation, @a geneDeletion or @a geneAddition (with
* configurable probabilities). Each mutation also has a parameter determining the chance of a gene
* being affected by it.
*/
class GenerationalElitistWithExclusivePools: public GeneticAlgorithm
{
public:
struct Options
{
double mutationPoolSize; ///< Percentage of population to regenerate using mutations in each round.
double crossoverPoolSize; ///< Percentage of population to regenerate using crossover in each round.
double randomisationChance; ///< The chance of choosing @a geneRandomisation as the mutation to perform
double deletionVsAdditionChance; ///< The chance of choosing @a geneDeletion as the mutation if randomisation was not chosen.
double percentGenesToRandomise; ///< The chance of any given gene being mutated in gene randomisation.
double percentGenesToAddOrDelete; ///< The chance of a gene being added (or deleted) in gene addition (or deletion).
bool isValid() const
{
return (
0 <= mutationPoolSize && mutationPoolSize <= 1.0 &&
0 <= crossoverPoolSize && crossoverPoolSize <= 1.0 &&
0 <= randomisationChance && randomisationChance <= 1.0 &&
0 <= deletionVsAdditionChance && deletionVsAdditionChance <= 1.0 &&
0 <= percentGenesToRandomise && percentGenesToRandomise <= 1.0 &&
0 <= percentGenesToAddOrDelete && percentGenesToAddOrDelete <= 1.0 &&
mutationPoolSize + crossoverPoolSize <= 1.0
);
}
};
GenerationalElitistWithExclusivePools(
Population _initialPopulation,
std::ostream& _outputStream,
Options const& _options
):
GeneticAlgorithm(_initialPopulation, _outputStream),
m_options(_options)
{
assert(_options.isValid());
}
void runNextRound() override;
private:
Options m_options;
};
}