autopas::GaussianCluster Class Reference

Model to predicts the output of a blackbox function f(x) for given input x. More...

#include <GaussianCluster.h>

Public Types
enum	WeightFunction { evidenceMatchingProbabilityGM , evidenceMatchingScaledProbabilityGM , wasserstein2 }
	Different weight functions between clusters. More...

Public Member Functions
	GaussianCluster (const std::vector< int > &dimRestriction, size_t continuousDims, WeightFunction weightFun, double sigma, Random &rngRef, const GaussianModelTypes::VectorToStringFun &vectorToString=defaultVecToString, const std::string &outputSuffix="")
	Constructor.
const std::vector< int > &	getDimensions () const
	Get the number of clusters in each dimension.
void	setDimensions (const std::vector< int > &newValue)
	Change the number of cluster in all dimension.
const GaussianProcess &	getCluster (size_t index1D) const
	Get the underlying GaussianProcess of a cluster.
void	clear ()
	Discard all evidence.
size_t	numEvidence () const
	Get the number of evidence provided.
void	addEvidence (const GaussianModelTypes::VectorDiscrete &inputDiscrete, const GaussianModelTypes::VectorContinuous &inputContinuous, double output)
	Provide a input-output pair as evidence.
void	addEvidence (const GaussianModelTypes::VectorPairDiscreteContinuous &input, double output)
	Provide a input-output pair as evidence.
GaussianModelTypes::VectorPairDiscreteContinuous	getEvidenceMax () const
	Get the evidence with the highest output value.
std::vector< GaussianModelTypes::VectorAcquisition >	sampleAcquisition (AcquisitionFunctionOption af, const GaussianModelTypes::NeighbourFunction &neighbourFun, const std::vector< GaussianModelTypes::VectorContinuous > &continuousSamples) const
	Generate all possible combinations of discrete tuples and continuous tuples in samples and calculate their corresponding acquisition.
void	logDebugGraph (const GaussianModelTypes::NeighbourFunction &neighbourFun, const std::vector< GaussianModelTypes::VectorContinuous > &continuousSamples) const
	Generate all possible combinations of discrete tuples and continuous tuples in samples and calculate their weight to each other.
void	setVectorToStringFun (const GaussianModelTypes::VectorToStringFun &fun)
	Change the used function to convert from vector to string.
GaussianModelTypes::VectorAcquisition	sampleAcquisitionMax (AcquisitionFunctionOption af, const GaussianModelTypes::NeighbourFunction &neighbourFun, const std::vector< GaussianModelTypes::VectorContinuous > &continuousSamples) const
	Generate all possible combinations of discrete tuples and continuous tuples in samples and returns the vector with maximum acquisition.
std::vector< GaussianModelTypes::VectorPairDiscreteContinuous >	sampleOrderedByAcquisition (AcquisitionFunctionOption af, const GaussianModelTypes::NeighbourFunction &neighbourFun, const std::vector< GaussianModelTypes::VectorContinuous > &continuousSamples) const
	Generate all possible combinations of discrete tuples and continuous tuples in samples and order them by their acquisition.

Static Public Member Functions
static std::string	defaultVecToString (const GaussianModelTypes::VectorPairDiscreteContinuous &vec)
	Default function used to convert vectors to readable strings.

Detailed Description

Model to predicts the output of a blackbox function f(x) for given input x.

The model separates discrete and continuous dimensions of x. For each possible discrete tuple a Gaussian process is assigned to estimate f(x) if the tuple is fixed. Some sample input-output pairs (x,f(x)) should be provided as evidence.

Member Enumeration Documentation

WeightFunction

enum autopas::GaussianCluster::WeightFunction

Different weight functions between clusters.

Enumerator
evidenceMatchingProbabilityGM	geometric mean of probability densitiy over all evidence in neighbouring cluster if provided to the target cluster
evidenceMatchingScaledProbabilityGM	geometric mean of scaled probability densitiy over all evidence in neighbouring cluster if provided to the target cluster Probability densities are scaled such that the maximum is 1.
wasserstein2	Wasserstein-2 distance of normal distributions of cluster given a continuous cluster.

Constructor & Destructor Documentation

GaussianCluster()

autopas::GaussianCluster::GaussianCluster	(	const std::vector< int > &	dimRestriction,
		size_t	continuousDims,
		WeightFunction	weightFun,
		double	sigma,
		Random &	rngRef,
		const GaussianModelTypes::VectorToStringFun &	vectorToString = defaultVecToString,
		const std::string &	outputSuffix = ""
	)

Constructor.

Parameters

dimRestriction	restrict the i-th dimension to a integer between 0 and dimRestriction[i]-1
continuousDims	additional unrestricted dimensions
weightFun	function to calculate weight between clusters
sigma	fixed noise
rngRef	reference to random number generator
vectorToString	function to convert vectors to a readable string
outputSuffix	Suffix for all output files produced by this class.

Member Function Documentation

addEvidence() [1/2]

void autopas::GaussianCluster::addEvidence	(	const GaussianModelTypes::VectorDiscrete &	inputDiscrete,
		const GaussianModelTypes::VectorContinuous &	inputContinuous,
		double	output
	)

Provide a input-output pair as evidence.

Each evidence improve the quality of future predictions.

Parameters

inputDiscrete	x
inputContinuous	y
output	f((x,y))

addEvidence() [2/2]

void autopas::GaussianCluster::addEvidence	(	const GaussianModelTypes::VectorPairDiscreteContinuous &	input,
		double	output
	)

Provide a input-output pair as evidence.

Each evidence improves the quality of future predictions.

Parameters

input	(x,y)
output	f((x,y))

defaultVecToString()

std::string autopas::GaussianCluster::defaultVecToString ( const GaussianModelTypes::VectorPairDiscreteContinuous &  vec )

static

Default function used to convert vectors to readable strings.

Note

begin() and end() currently not available for Eigen::Vector, so AutoPas ArrayUtils cannot be used.

Parameters

vec

Returns

string with format (a,b,...,n) beginning with discrete values.

getCluster()

const autopas::GaussianProcess & autopas::GaussianCluster::getCluster

(

size_t

index1D

)

const

Get the underlying GaussianProcess of a cluster.

This function should only be used for testing purposes.

Parameters

index1D

Returns

getDimensions()

const std::vector< int > & autopas::GaussianCluster::getDimensions

(

)

const

Get the number of clusters in each dimension.

Returns

getEvidenceMax()

autopas::GaussianModelTypes::VectorPairDiscreteContinuous autopas::GaussianCluster::getEvidenceMax

(

)

const

Get the evidence with the highest output value.

Returns

input of max

logDebugGraph()

void autopas::GaussianCluster::logDebugGraph	(	const GaussianModelTypes::NeighbourFunction &	neighbourFun,
		const std::vector< GaussianModelTypes::VectorContinuous > &	continuousSamples
	)		const

Generate all possible combinations of discrete tuples and continuous tuples in samples and calculate their weight to each other.

Output graph in AutoPasLog Trace.

Parameters

neighbourFun	function which generates neighbours of given discrete tuple
continuousSamples	continuous tuples

numEvidence()

size_t autopas::GaussianCluster::numEvidence

(

)

const

Get the number of evidence provided.

Returns

sampleAcquisition()

std::vector< autopas::GaussianModelTypes::VectorAcquisition > autopas::GaussianCluster::sampleAcquisition	(	AcquisitionFunctionOption	af,
		const GaussianModelTypes::NeighbourFunction &	neighbourFun,
		const std::vector< GaussianModelTypes::VectorContinuous > &	continuousSamples
	)		const

Generate all possible combinations of discrete tuples and continuous tuples in samples and calculate their corresponding acquisition.

Parameters

af	function to optimize
neighbourFun	function which generates neighbours of given discrete tuple
continuousSamples	continuous tuples

Returns

all discrete-continuous tuples paired with their corresponding acquisition

sampleAcquisitionMax()

autopas::GaussianModelTypes::VectorAcquisition autopas::GaussianCluster::sampleAcquisitionMax	(	AcquisitionFunctionOption	af,
		const GaussianModelTypes::NeighbourFunction &	neighbourFun,
		const std::vector< GaussianModelTypes::VectorContinuous > &	continuousSamples
	)		const

Generate all possible combinations of discrete tuples and continuous tuples in samples and returns the vector with maximum acquisition.

Parameters

af	function to optimize
neighbourFun	function which generates neighbours with prior weight of given discrete tuple
continuousSamples	continuous tuples

Returns

pair of vector and corresponding maximum acquistion

sampleOrderedByAcquisition()

std::vector< autopas::GaussianModelTypes::VectorPairDiscreteContinuous > autopas::GaussianCluster::sampleOrderedByAcquisition	(	AcquisitionFunctionOption	af,
		const GaussianModelTypes::NeighbourFunction &	neighbourFun,
		const std::vector< GaussianModelTypes::VectorContinuous > &	continuousSamples
	)		const

Generate all possible combinations of discrete tuples and continuous tuples in samples and order them by their acquisition.

Parameters

af	function to optimize
neighbourFun	function which generates neighbours with prior weight of given discrete tuple
continuousSamples	continuous tuples

Returns

all discrete-continuous tuples ordered by their corresponding acquisition

setDimensions()

void autopas::GaussianCluster::setDimensions

(

const std::vector< int > &

newValue

)

Change the number of cluster in all dimension.

This will discard all evidence.

Parameters

newValue

new number of clusters in each dimension

setVectorToStringFun()

void autopas::GaussianCluster::setVectorToStringFun

(

const GaussianModelTypes::VectorToStringFun &

fun

)

Change the used function to convert from vector to string.

Parameters

fun	new converter

---

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

• autopas/tuning/tuningStrategy/GaussianModel/GaussianCluster.h
• autopas/tuning/tuningStrategy/GaussianModel/GaussianCluster.cpp