dakota::surrogates Namespace Reference

namespace for new Dakota surrogates module More...

Classes
class	GaussianProcess
	The GaussianProcess constructs a Gaussian Process regressor surrogate given a matrix of data. More...
class	GP_Objective
	ROL objective function for the Gaussian Process (GP) surrogate. More...
class	Kernel
	Kernel functions for the Gaussian Process surrogate. More...
class	Matern32Kernel
	Stationary kernel with C^1 smooth realizations. More...
class	Matern52Kernel
	Stationary kernel with C^2 smooth realizations. More...
class	PolynomialRegression
	The PolynomialRegression class constructs a polynomial regressor using ordinary least squares. More...
class	SquaredExponentialKernel
	Stationary kernel with C^\infty smooth realizations. More...
class	Surrogate
	Parent class for surrogate models. More...

Typedefs
using	RolVec = ROL::Vector< double >
	Dakota alias for ROL Vector.
using	RolStdVec = ROL::StdVector< double >
	Dakota alias for ROL StdVector.
using	SCALER_TYPE = util::DataScaler::SCALER_TYPE
	alias for util SCALER_TYPE enum
using	SOLVER_TYPE = util::LinearSolverBase::SOLVER_TYPE
	alias for util SOLVER_TYPE enum

Functions
void	compute_next_combination (int num_dims, int level, VectorXi &index, bool &extend, int &h, int &t)
void	size_level_index_vector (int num_dims, int level, MatrixXi &indices)
	Compute a matrix of basis indices for given dimension and level. Each row of the matrix sums to level. More...
void	compute_hyperbolic_subdim_level_indices (int num_dims, int level, int num_active_dims, double p, MatrixXi &indices)
	Compute a matrix of indices for a submatrix (i.e. up to the active dimensions column) of indices produced by size_level_index_vector(num_dims, level, indices) where each feature has a component > 0 and respects the p-norm cutoff. More...
void	compute_hyperbolic_level_indices (int num_dims, int level, double p, MatrixXi &indices)
	Compute the hyperbolic cross indices for a given level. More...
void	compute_hyperbolic_indices (int num_dims, int level, double p, MatrixXi &indices)
	Compute the hyperbolic cross indices for all levels up to level. More...
void	compute_reduced_indices (int num_dims, int level, MatrixXi &indices)
	Compute the reduced indices for all levels up to level. More...
void	fd_check_gradient (Surrogate &surr, const MatrixXd &sample, MatrixXd &fd_error, const int num_steps=10)
	Perform a centered finite difference check of a Surrogate's gradient method. More...
void	fd_check_hessian (Surrogate &surr, const MatrixXd &sample, MatrixXd &fd_error, const int num_steps=10)
	Perform a centered finite difference check of a Surrogate's Hessian method. More...
std::vector< MatrixXd >	compute_cw_dists_squared (const std::vector< MatrixXd > &cw_dists)
	Compute a vector of component-wise squared distances from a vector of component-wise signed distances. More...
std::shared_ptr< Kernel >	kernel_factory (const std::string &kernel_type)
	Creates a derived Kernel class. More...

Detailed Description

namespace for new Dakota surrogates module

Function Documentation

compute_next_combination()

void compute_next_combination	(	int	num_dims,
		int	level,
		VectorXi &	index,
		bool &	extend,
		int &	h,
		int &	t
	)

Parameters

[in]	num_dims	Dimension of the feature space.
[in]	level	Total order in each row of indices. Should be >= 1.
[in,out]	index	Vector of ints that specifies the powers for each term in the basis.
[in,out]	extend	Bool for whether to continue with the computation of basis indices.
[in,out]	h	Working variable for basis enumeration.
[in,out]	t	Working variable for basis enumeration.

Referenced by size_level_index_vector().

size_level_index_vector()

void size_level_index_vector	(	int	num_dims,
		int	level,
		MatrixXi &	indices
	)

Compute a matrix of basis indices for given dimension and level. Each row of the matrix sums to level.

Parameters

[in]	num_dims	Dimension of the feature space.
[in]	level	Total order in each row of indices. Should be >= 1.
[out]	indices	Matrix of indices - (num_terms by num_dims).

References compute_next_combination(), and dakota::util::n_choose_k().

Referenced by compute_hyperbolic_level_indices(), and compute_hyperbolic_subdim_level_indices().

compute_hyperbolic_subdim_level_indices()

void compute_hyperbolic_subdim_level_indices	(	int	num_dims,
		int	level,
		int	num_active_dims,
		double	p,
		MatrixXi &	indices
	)

Compute a matrix of indices for a submatrix (i.e. up to the active dimensions column) of indices produced by size_level_index_vector(num_dims, level, indices) where each feature has a component > 0 and respects the p-norm cutoff.

Parameters

[in]	num_dims	Dimension of the feature space.
[in]	level	Total order in each row of indices. Should be >= 1.
[in]	num_active_dims	The # of active features and end index of the submatrix.
[in]	p	Real value for p-norm.
[out]	indices	Matrix of indices - (num_terms by num_active_dims)

References dakota::util::num_nonzeros(), dakota::util::p_norm(), dakota::silence_unused_args(), and size_level_index_vector().

Referenced by compute_hyperbolic_level_indices().

compute_hyperbolic_level_indices()

void compute_hyperbolic_level_indices	(	int	num_dims,
		int	level,
		double	p,
		MatrixXi &	indices
	)

Compute the hyperbolic cross indices for a given level.

Parameters

[in]	num_dims	Dimension of the feature space.
[in]	level	Total order in each column of indices. Should be >= 0.
[in]	p	Real value for p-norm.
[out]	indices	Matrix of indices - (num_dims by num_terms)

References dakota::util::append_columns(), compute_hyperbolic_subdim_level_indices(), dakota::util::nonzero(), dakota::util::num_nonzeros(), and size_level_index_vector().

Referenced by compute_hyperbolic_indices().

compute_hyperbolic_indices()

void compute_hyperbolic_indices	(	int	num_dims,
		int	level,
		double	p,
		MatrixXi &	indices
	)

Compute the hyperbolic cross indices for all levels up to level.

Parameters

[in]	num_dims	Dimension of the feature space.
[in]	level	Highest level to compute basis indices for.
[in]	p	Real value for p-norm.
[out]	indices	Matrix of indices - (num_dims by num_terms).

References dakota::util::append_columns(), and compute_hyperbolic_level_indices().

Referenced by PolynomialRegression::build().

compute_reduced_indices()

void compute_reduced_indices	(	int	num_dims,
		int	level,
		MatrixXi &	indices
	)

Compute the reduced indices for all levels up to level.

Parameters

[in]	num_dims	Dimension of the feature space.
[in]	level	Highest level to compute basis indices for.
[out]	indices	Matrix of indices - (num_dims by num_terms).

References dakota::util::append_columns().

Referenced by PolynomialRegression::build().

fd_check_gradient()

void fd_check_gradient	(	Surrogate &	surr,
		const MatrixXd &	sample,
		MatrixXd &	fd_error,
		const int	num_steps = 10
	)

Perform a centered finite difference check of a Surrogate's gradient method.

Parameters

[in]	surr	Reference to a Surrogate.
[in]	sample	Point to evaluate the gradient at - (1 by numVariables).
[out]	fd_error	Matrix of finite difference error for each component of the gradient - (num_steps by numVariables).
[in]	num_steps	Number of increments (N) for the finite difference. The increment vector h = 10**(-i), i = 1, ..., N.

References Surrogate::dataScaler, DataScaler::get_scaler_features_scale_factors(), Surrogate::gradient(), and Surrogate::value().

fd_check_hessian()

void fd_check_hessian	(	Surrogate &	surr,
		const MatrixXd &	sample,
		MatrixXd &	fd_error,
		const int	num_steps = 10
	)

Perform a centered finite difference check of a Surrogate's Hessian method.

Parameters

[in]	surr	Reference to a Surrogate.
[in]	sample	Point to evaluate the Hessian at - (1 by numVariables).
[out]	fd_error	Matrix of finite difference error for each independent component of the Hessian. There are numVariables*(numVariables+1)/2 = numInd components - (num_steps by numInd). of the Hessian - (num_steps by numVariables).
[in]	num_steps	Number of increments (N) for the finite difference. The increment vector h = 10**(-i), i = 1, ..., N.

References Surrogate::dataScaler, DataScaler::get_scaler_features_scale_factors(), Surrogate::hessian(), and Surrogate::value().

compute_cw_dists_squared()

std::vector< MatrixXd > compute_cw_dists_squared

(

const std::vector< MatrixXd > &

cw_dists

)

Compute a vector of component-wise squared distances from a vector of component-wise signed distances.

Parameters

[in]

dists2

Vector of signed component-wise distances.

Returns

Vector of squared distances.

Referenced by Matern32Kernel::compute_first_deriv_pred_gram(), Matern52Kernel::compute_first_deriv_pred_gram(), and Matern52Kernel::compute_second_deriv_pred_gram().

kernel_factory()

std::shared_ptr< Kernel > kernel_factory

(

const std::string &

kernel_type

)

Creates a derived Kernel class.

Parameters

[in]

kernel_type

Name of the kernel.

Returns

Pointer to specialized kernel class.

Referenced by GaussianProcess::build(), and GaussianProcess::serialize().