autopas::FeatureVectorEncoder Class Reference

Encoder to convert FeatureVector from and to Eigen::Vector. More...

#include <FeatureVectorEncoder.h>

Public Member Functions
	FeatureVectorEncoder ()
	Default Constructor.
	FeatureVectorEncoder (const std::vector< FeatureVector::ContainerTraversalEstimatorOption > &containerTraversalEstimatorOptions, const std::vector< DataLayoutOption > &dataLayoutOptions, const std::vector< Newton3Option > &newton3Options, const NumberSet< double > &cellSizeFactors, const InteractionTypeOption &interactionType)
	Constructor.
void	setAllowedOptions (const std::vector< FeatureVector::ContainerTraversalEstimatorOption > &containerTraversalEstimatorOptions, const std::vector< DataLayoutOption > &dataLayoutOptions, const std::vector< Newton3Option > &newton3Options, const NumberSet< double > &cellSizeFactors)
	Set allowed options.
size_t	getOneHotDims () const
	Get the dimensions of a one-hot encoded vector.
const std::array< int, tunableDiscreteDims > &	getDiscreteRestrictions () const
	Get the number of allowed options of each discrete dimension.
Eigen::VectorXd	oneHotEncode (const FeatureVector &vec) const
	Encode FeatureVector to Eigen::VectorXd using one-hot-encoding.
FeatureVector	oneHotDecode (const Eigen::VectorXd &vec)
	Decode one-hot-encoded VectorXd to FeatureVector.
std::pair< Eigen::VectorXi, Eigen::VectorXd >	convertToCluster (const FeatureVector &vec, double iteration) const
	Convert Feature vector to cluster representation for GaussianCluster.
FeatureVector	convertFromCluster (const std::pair< Eigen::VectorXi, Eigen::VectorXd > &vec)
	Inverse of convertToCluster.
std::vector< std::pair< Eigen::VectorXi, double > >	clusterNeighboursManhattan1 (const Eigen::VectorXi &target)
	Get cluster-encoded neighbours of given target with fixed weight.
std::vector< std::pair< Eigen::VectorXi, double > >	clusterNeighboursManhattan1Container (const Eigen::VectorXi &target)
	Get cluster-encoded neighbours of given target.
std::vector< FeatureVector >	lhsSampleFeatures (size_t n, Random &rng) const
	Create n latin-hypercube-samples from given featureSpace.
std::vector< Eigen::VectorXd >	lhsSampleFeatureCluster (size_t n, Random &rng, double iteration) const
	Create n latin-hypercube-samples from the continuous featureSpace and append a value representing the current iteration to each sample.

Static Public Attributes
static constexpr size_t	tunableDiscreteDims {static_cast<size_t>(DiscreteIndices::TOTALNUMBER)}
	Number of tunable discrete dimensions.
static constexpr size_t	tunableContinuousDims {static_cast<size_t>(ContinuousIndices::TOTALNUMBER)}
	Number of tunable continuous dimensions.

Detailed Description

Encoder to convert FeatureVector from and to Eigen::Vector.

Constructor & Destructor Documentation

FeatureVectorEncoder()

autopas::FeatureVectorEncoder::FeatureVectorEncoder	(	const std::vector< FeatureVector::ContainerTraversalEstimatorOption > &	containerTraversalEstimatorOptions,
		const std::vector< DataLayoutOption > &	dataLayoutOptions,
		const std::vector< Newton3Option > &	newton3Options,
		const NumberSet< double > &	cellSizeFactors,
		const InteractionTypeOption &	interactionType
	)

Constructor.

Parameters

containerTraversalEstimatorOptions
dataLayoutOptions
newton3Options
cellSizeFactors
interactionType

Member Function Documentation

clusterNeighboursManhattan1()

std::vector< std::pair< Eigen::VectorXi, double > > autopas::FeatureVectorEncoder::clusterNeighboursManhattan1

(

const Eigen::VectorXi &

target

)

Get cluster-encoded neighbours of given target with fixed weight.

Neighbours are all configurations which differ in at most one configuration from target.

Parameters

target

Returns

clusterNeighboursManhattan1Container()

std::vector< std::pair< Eigen::VectorXi, double > > autopas::FeatureVectorEncoder::clusterNeighboursManhattan1Container

(

const Eigen::VectorXi &

target

)

Get cluster-encoded neighbours of given target.

Neighbours are all configurations which differ in at most one configuration from target. The weight is lowered if container is changed.

Parameters

target

Returns

convertFromCluster()

autopas::FeatureVector autopas::FeatureVectorEncoder::convertFromCluster

(

const std::pair< Eigen::VectorXi, Eigen::VectorXd > &

vec

)

Inverse of convertToCluster.

Convert cluster representation back to Feature vector while ignoring the iteration.

Parameters

vec	cluster encoded vector

Returns

decoded vector

convertToCluster()

std::pair< Eigen::VectorXi, Eigen::VectorXd > autopas::FeatureVectorEncoder::convertToCluster	(	const FeatureVector &	vec,
		double	iteration
	)		const

Convert Feature vector to cluster representation for GaussianCluster.

Discrete values are encoded using their index in given std::vector. Additionally, append current iteration to the continuous tuple.

Parameters

vec	vector to encode
iteration	current iteration which may be scaled by some factor

Returns

cluster encoded vector

getDiscreteRestrictions()

const std::array< int, autopas::FeatureVectorEncoder::tunableDiscreteDims > & autopas::FeatureVectorEncoder::getDiscreteRestrictions

(

)

const

Get the number of allowed options of each discrete dimension.

Returns

getOneHotDims()

size_t autopas::FeatureVectorEncoder::getOneHotDims

(

)

const

Get the dimensions of a one-hot encoded vector.

Returns

lhsSampleFeatureCluster()

std::vector< Eigen::VectorXd > autopas::FeatureVectorEncoder::lhsSampleFeatureCluster	(	size_t	n,
		autopas::Random &	rng,
		double	iteration
	)		const

Create n latin-hypercube-samples from the continuous featureSpace and append a value representing the current iteration to each sample.

Parameters

n	number of samples
rng
iteration	Current iteration which may be scaled by some factor.

Returns

vector of continuous feature samples

lhsSampleFeatures()

std::vector< autopas::FeatureVector > autopas::FeatureVectorEncoder::lhsSampleFeatures	(	size_t	n,
		autopas::Random &	rng
	)		const

Create n latin-hypercube-samples from given featureSpace.

Parameters

n	number of samples
rng

Returns

vector of sample featureVectors

oneHotDecode()

autopas::FeatureVector autopas::FeatureVectorEncoder::oneHotDecode

(

const Eigen::VectorXd &

vec

)

Decode one-hot-encoded VectorXd to FeatureVector.

Parameters

vec	one-hot-encoded vector

Returns

decoded FeatureVector

oneHotEncode()

Eigen::VectorXd autopas::FeatureVectorEncoder::oneHotEncode

(

const FeatureVector &

vec

)

const

Encode FeatureVector to Eigen::VectorXd using one-hot-encoding.

Parameters

vec	vector to encode

Returns

one-hot-encoded vector

setAllowedOptions()

void autopas::FeatureVectorEncoder::setAllowedOptions	(	const std::vector< FeatureVector::ContainerTraversalEstimatorOption > &	containerTraversalEstimatorOptions,
		const std::vector< DataLayoutOption > &	dataLayoutOptions,
		const std::vector< Newton3Option > &	newton3Options,
		const NumberSet< double > &	cellSizeFactors
	)

Set allowed options.

All previously encoded vector can not be decoded anymore.

Parameters

containerTraversalEstimatorOptions
dataLayoutOptions
newton3Options
cellSizeFactors

---

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

• autopas/tuning/utils/FeatureVectorEncoder.h
• autopas/tuning/utils/FeatureVectorEncoder.cpp