NonDAdaptImpSampling Class Reference

Class for the Adaptive Importance Sampling methods within DAKOTA. More...

Inheritance diagram for NonDAdaptImpSampling:

List of all members.

Public Member Functions
	NonDAdaptImpSampling (ProblemDescDB &problem_db, Model &model)
	standard constructor
	NonDAdaptImpSampling (Model &model, unsigned short sample_type, int samples, int seed, const String &rng, bool vary_pattern, unsigned short is_type, bool cdf_flag, bool x_space_model, bool use_model_bounds)
	alternate constructor for on-the-fly instantiations
	~NonDAdaptImpSampling ()
	destructor
void	derived_init_communicators (ParLevLIter pl_iter)
	derived class contributions to initializing the communicators associated with this Iterator instance
void	derived_set_communicators (ParLevLIter pl_iter)
	derived class contributions to setting the communicators associated with this Iterator instance
void	derived_free_communicators (ParLevLIter pl_iter)
	derived class contributions to freeing the communicators associated with this Iterator instance
void	quantify_uncertainty ()
	performs adaptive importance sampling and computes probability of failure
void	print_results (std::ostream &s)
	print the final statistics
void	initialize (const RealVectorArray &full_points, bool x_space_data, size_t resp_index, Real initial_prob, Real failure_threshold)
	initializes data needed for importance sampling: an initial set of points around which to sample, a failure threshold, an initial probability to refine, and flags to control transformations
void	initialize (const RealMatrix &full_points, bool x_space_data, size_t resp_index, Real initial_prob, Real failure_threshold)
	initializes data needed for importance sampling: an initial set of points around which to sample, a failure threshold, an initial probability to refine, and flags to control transformations
void	initialize (const RealVector &full_point, bool x_space_data, size_t resp_index, Real initial_prob, Real failure_threshold)
	initializes data needed for importance sampling: an initial point around which to sample, a failure threshold, an initial probability to refine, and flags to control transformations
Real	final_probability ()
	returns the final probability calculated by the importance sampling
Private Member Functions
void	select_rep_points (const RealVectorArray &var_samples_u, const RealVector &fn_samples)
	select representative points from a set of samples
void	converge_statistics (bool cov_flag)
	iteratively generate samples and select representative points until probability and (optionally) coefficient of variation converge
void	generate_samples (RealVectorArray &var_samples_u)
	generate a set of samples based on multimodal sampling density
void	evaluate_samples (const RealVectorArray &var_samples_u, RealVector &fn_samples)
	evaluate the model at the sample points and store the responses
void	calculate_statistics (const RealVectorArray &var_samples_u, const RealVector &fn_samples, size_t total_samples, Real &sum_prob, Real &prob, bool compute_cov, Real &sum_var, Real &cov)
	calculate the probability of exceeding the failure threshold and the coefficent of variation (if requested)
Real	distance (const RealVector &a, const RealVector &b)
	compute Euclidean distance between points a and b
Real	recentered_density (const RealVector &sample_point)
	compute density between a representative point and a sample point, assuming standard normal
Private Attributes
Model	uSpaceModel
	importance sampling is performed in standardized probability space. This u-space model is either passed in (alternate constructor for helper AIS) or constructed using transform_model() (standard constructor for stand-alone AIS)
unsigned short	importanceSamplingType
	integration type (is, ais, mmais) provided by input specification
bool	initLHS
	flag to identify if initial points are generated from an LHS sample
bool	useModelBounds
	flag to control if the sampler should respect the model bounds
bool	invertProb
	flag for inversion of probability values using 1.-p
int	refineSamples
	size of sample batch within each refinement iteration
size_t	respFnIndex
	the active response function index in the model to be sampled
RealVector	designPoint
	design subset for which uncertain subset is being sampled
RealVectorArray	initPointsU
	the original set of u-space samples passed in initialize()
RealVectorArray	repPointsU
	the set of representative points in u-space around which to sample
RealVector	repWeights
	the weight associated with each representative point
Real	probEstimate
	the probability estimate that is iteratively refined by importance sampling
Real	failThresh
	the failure threshold (z-bar) for the problem.

---

Detailed Description

Class for the Adaptive Importance Sampling methods within DAKOTA.

The NonDAdaptImpSampling implements the multi-modal adaptive importance sampling used for reliability calculations. (eventually we will want to broaden this). Need to add more detail to this description.

---

Constructor & Destructor Documentation

NonDAdaptImpSampling	(	ProblemDescDB &	problem_db,
		Model &	model
	)

standard constructor

This is the primary constructor. It accepts a Model reference. It will perform refinement for all response QOI and all probability levels.

References NonD::initialize_random_variables(), Iterator::iteratedModel, NonDSampling::numSamples, NonDAdaptImpSampling::refineSamples, NonDSampling::statsFlag, NonD::transform_model(), NonDAdaptImpSampling::useModelBounds, and NonDAdaptImpSampling::uSpaceModel.

NonDAdaptImpSampling	(	Model &	model,
		unsigned short	sample_type,
		int	refine_samples,
		int	refine_seed,
		const String &	rng,
		bool	vary_pattern,
		unsigned short	is_type,
		bool	cdf_flag,
		bool	x_space_model,
		bool	use_model_bounds
	)

alternate constructor for on-the-fly instantiations

This is an alternate constructor for instantiations on the fly using a Model but no ProblemDescDB. It will perform refinement for one response QOI and one probability level (passed in initialize()).

References NonD::cdfFlag, Iterator::maxEvalConcurrency, NonDAdaptImpSampling::refineSamples, NonD::transform_model(), NonDAdaptImpSampling::useModelBounds, and NonDAdaptImpSampling::uSpaceModel.

---

Member Function Documentation

void initialize	(	const RealVectorArray &	acv_points,
		bool	x_space_data,
		size_t	resp_index,
		Real	initial_prob,
		Real	failure_threshold
	)

initializes data needed for importance sampling: an initial set of points around which to sample, a failure threshold, an initial probability to refine, and flags to control transformations

Initializes data using a vector array of starting points.

References NonDAdaptImpSampling::designPoint, NonDAdaptImpSampling::failThresh, NonDAdaptImpSampling::initPointsU, NonDAdaptImpSampling::invertProb, NonD::natafTransform, NonD::numContDesVars, NonD::numUncertainVars, NonDAdaptImpSampling::probEstimate, and NonDAdaptImpSampling::respFnIndex.

Referenced by NonDExpansion::compute_statistics(), NonDGlobalReliability::importance_sampling(), and NonDAdaptImpSampling::quantify_uncertainty().

void initialize	(	const RealMatrix &	acv_points,
		bool	x_space_data,
		size_t	resp_index,
		Real	initial_prob,
		Real	failure_threshold
	)

initializes data needed for importance sampling: an initial set of points around which to sample, a failure threshold, an initial probability to refine, and flags to control transformations

Initializes data using a matrix of starting points.

References NonDAdaptImpSampling::designPoint, NonDAdaptImpSampling::failThresh, NonDAdaptImpSampling::initPointsU, NonDAdaptImpSampling::invertProb, NonD::natafTransform, NonD::numContDesVars, Analyzer::numContinuousVars, NonD::numUncertainVars, NonDAdaptImpSampling::probEstimate, and NonDAdaptImpSampling::respFnIndex.

void initialize	(	const RealVector &	acv_point,
		bool	x_space_data,
		size_t	resp_index,
		Real	initial_prob,
		Real	failure_threshold
	)

initializes data needed for importance sampling: an initial point around which to sample, a failure threshold, an initial probability to refine, and flags to control transformations

Initializes data using only one starting point.

References NonDAdaptImpSampling::designPoint, NonDAdaptImpSampling::failThresh, NonDAdaptImpSampling::initPointsU, NonDAdaptImpSampling::invertProb, NonD::natafTransform, NonD::numContDesVars, NonD::numUncertainVars, NonDAdaptImpSampling::probEstimate, and NonDAdaptImpSampling::respFnIndex.

---

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

• NonDAdaptImpSampling.hpp
• NonDAdaptImpSampling.cpp