Class for a utility class containing correlation calculations and variance-based decomposition. More...

Public Member Functions
	SensAnalysisGlobal ()
	constructor

	~SensAnalysisGlobal ()
	destructor

void	compute_correlations (const VariablesArray &vars_samples, const IntResponseMap &resp_samples, const StringSetArray &dss_vals)
	computes four correlation matrices for input and output data simple, partial, simple rank, and partial rank More...

void	compute_correlations (const RealMatrix &vars_samples, const IntResponseMap &resp_samples)
	computes four correlation matrices for input and output data simple, partial, simple rank, and partial rank More...

void	archive_correlations (const StrStrSizet &run_identifier, ResultsManager &iterator_results, const StringArray &var_labels, const StringArray &resp_labels, const size_t &inc_id=0) const
	save correlations to database

void	archive_std_regress_coeffs (const StrStrSizet &run_identifier, ResultsManager &iterator_results, const StringArray &var_labels, const StringArray &resp_labels, const size_t &inc_id=0) const
	save standardized regression coefficients to database

bool	correlations_computed () const
	returns corrComputed to indicate whether compute_correlations() has been invoked

void	print_correlations (std::ostream &s, const StringArray &var_labels, const StringArray &resp_labels) const
	prints the correlations computed in compute_correlations()

void	compute_std_regress_coeffs (const RealMatrix &vars_samples, const IntResponseMap &resp_samples)
	computes standardized regression coefficients and corresponding R^2 values for input and output data

void	print_std_regress_coeffs (std::ostream &s, StringArray var_labels, const StringArray &resp_labels) const
	prints the SRCs and R^2 values computed in compute_correlations()

void	compute_vbd_stats_via_sampling (const unsigned short method, const int num_bins, const size_t num_functions, const size_t num_continuous_vars, const size_t num_discrete_vars, const size_t num_samples, const RealMatrix &vars_samples, const IntResponseMap &resp_samples)
	compute VBD-based Sobol indices

void	print_sobol_indices (std::ostream &s, const StringArray &var_labels, const StringArray &resp_labels, const Real dropTol) const
	Printing of VBD results.

void	archive_sobol_indices (const StrStrSizet &run_identifier, ResultsManager &resultsDB, const StringArray &var_labels, const StringArray &resp_labels, const Real dropTol) const
	archive VBD-based Sobol indices

Protected Member Functions
void	compute_binned_sobol_indices_from_valid_samples (const RealMatrix &valid_samples, size_t num_bins)
	compute binned sobol indices from valid samples (having screened out samples with non-numeric response)

void	compute_binned_continuous_var_sobol_indices (RealMatrix &response_samples, RealVector &total_variances, IntMatrix &sorted_continuous_var_inds, size_t num_bins)
	compute binned sobol indices for continuous random variables

void	compute_binned_discrete_var_sobol_indices (RealMatrix &response_samples, RealVector &total_variances, IntMatrix &sorted_discrete_var_inds, std::vector< IntVector > &unique_counts)
	compute binned sobol indices for discrete random variables

RealMatrix	compute_binned_variances (RealMatrix &sorted_response_samples, IntVector &bin_counts)
	Given sorted response samples and a vector of length num_bins with bin counts per bin, computes the variance in each bin.

void	compute_Si_from_binned_variances_and_weights (RealVector &total_variances, RealMatrix &binned_variance_samples, RealVector &bin_weights, int i)
	Given binned variances, weights over bins, and the index of the variable,.

Protected Attributes
size_t	numFns
	Reorder. More...

size_t	numVars
	number of inputs

size_t	numContinuousVars
	number of continuous inputs

size_t	numDiscreteVars
	number of discrete inputs

RealVectorArray	indexSi
	VBD main effect indices.

RealVectorArray	indexTi
	VBD total effect indices.

Private Member Functions
void	check_num_samples (const size_t num_var_samples, const size_t num_response_samples, const char *method_name)
	Checks there are a nonzero number of samples and that the number of samples agree between variables and responses. Method name passed for error handling.

size_t	get_n_vars (const VariablesArray &vars_samples)
	Computes the number of variables for a full variable object.

size_t	find_valid_samples (const IntResponseMap &resp_samples, BoolDeque &valid_sample)
	find samples with finite response (any sample with any Nan or +/-Inf observation will be dropped)

void	valid_sample_matrix (const VariablesArray &vars_samples, const IntResponseMap &resp_samples, const StringSetArray &dss_vals, const BoolDeque is_valid_sample, RealMatrix &valid_data)
	extract a compact valid sample (vars/resp) matrix from the passed data

void	valid_sample_matrix (const RealMatrix &vars_samples, const IntResponseMap &resp_samples, const BoolDeque is_valid_sample, RealMatrix &valid_samples)
	extract a compact valid sample (vars/resp) matrix from the passed data

void	values_to_ranks (RealMatrix &valid_data)
	replace sample values with their ranks, in-place More...

void	correl_adjust (Real &corr_value)
	if result was NaN/Inf, preserve it, otherwise truncate to [-1.0, 1.0]

void	simple_corr (RealMatrix &total_data, const int &num_in, RealMatrix &corr_matrix)
	computes simple correlations, populating corr_matrix More...

void	partial_corr (RealMatrix &total_data, const int num_in, const RealMatrix &simple_corr_mat, RealMatrix &corr_matrix, bool &numerical_issues)
	computes partial correlations, populating corr_matrix and numerical_issues More...

bool	has_nan_or_inf (const RealMatrix &corr) const
	Return true if there are any NaN or Inf entries in the matrix.

void	check_correlations_for_nan_or_inf (std::ostream &s) const
	Check correlation matrices for nans or infs and print warning message.

void	print_simple_correlations (std::ostream &s, const StringArray &var_labels, const StringArray &resp_labels, bool rank) const
	print simple (rank = false) or simple rank (rank = true) correlations

void	print_partial_correlations (std::ostream &s, const StringArray &var_labels, const StringArray &resp_labels, bool rank) const
	print partial (rank = false) or partial rank (rank = true) correlations

void	archive_simple_correlations (const StrStrSizet &run_identifier, ResultsManager &iterator_results, const StringArray &var_labels, const StringArray &resp_labels, const std::vector< const char * > &combined_desc, const size_t &inc_id, bool rank) const
	archive simple (rank = false) or simple rank (rank = true) correlations

void	archive_partial_correlations (const StrStrSizet &run_identifier, ResultsManager &iterator_results, const StringArray &var_labels, const StringArray &resp_labels, const size_t &inc_id, bool rank) const
	archive partial (rank = false) or partial rank (rank = true) correlations

void	compute_pick_and_freeze_vbd_stats (const size_t num_functions, const size_t num_vars, const size_t num_samples, const IntResponseMap &resp_samples)

void	compute_binned_vbd_stats (const int num_bins, const size_t num_functions, const size_t num_continuous_vars, const size_t num_discrete_vars, const size_t num_samples, const RealMatrix &vars_samples, const IntResponseMap &resp_samples)

IntMatrix	get_continuous_var_samples_argsort (const RealMatrix &valid_data)
	Returns a matrix where columns are the indices that would sort each variable's samples smallest to largest.

void	get_discrete_var_samples_argsort_and_counts (const RealMatrix &valid_data, IntMatrix &sorted_var_indices, std::vector< IntVector > &unique_counts)
	Returns a matrix where columns are the indices that sort each variable into contiguous unique values and a list-of-lists containing the number of unique values from smallest to largest per variable.

void	compute_response_means_and_variances (const RealMatrix &response_samples, RealVector &total_means, RealVector &total_variances)

void	print_main_and_total_effects_indices (std::ostream &s, const StringArray &var_labels, const StringArray &resp_labels, const Real dropTol) const

void	print_main_effects_indices (std::ostream &s, const StringArray &var_labels, const StringArray &resp_labels, const Real dropTol) const

void	archive_main_effects_indices (const StrStrSizet &run_identifier, ResultsManager &resultsDB, const StringArray &var_labels, const StringArray &resp_labels, const Real dropTol) const
	archive VBD-based Sobol main effects indices

void	archive_total_effects_indices (const StrStrSizet &run_identifier, ResultsManager &resultsDB, const StringArray &var_labels, const StringArray &resp_labels, const Real dropTol) const
	archive VBD-based Sobol total effects indices

Static Private Member Functions
static bool	rank_sort (const int &x, const int &y)
	sort algorithm to compute ranks for rank correlations

Private Attributes
RealMatrix	simpleCorr
	matrix to hold simple raw correlations

RealMatrix	simpleRankCorr
	matrix to hold simple rank correlations

RealMatrix	partialCorr
	matrix to hold partial raw correlations

RealMatrix	partialRankCorr
	matrix to hold partial rank correlations

RealMatrix	stdRegressCoeffs
	matrix to hold standardized regression coefficients

RealVector	stdRegressCODs
	vector to hold coefficients of determination, eg R^2 values

bool	numericalIssuesRaw
	flag indicating numerical issues in partial raw correlation calculations

bool	numericalIssuesRank
	flag indicating numerical issues in partial rank correlation calculations

bool	corrComputed
	flag indicatng whether correlations have been computed

Static Private Attributes
static RealArray	rawData = RealArray()
	array to hold temporary data before sort

Detailed Description

Class for a utility class containing correlation calculations and variance-based decomposition.

This class provides code for several of the sampling methods both in the NonD branch and in the PStudyDACE branch. Currently, the utility functions provide global sensitivity analysis through correlation calculations (e.g. simple, partial, rank, raw) as well as variance-based decomposition.

Member Function Documentation

◆ compute_correlations() [1/2]

void compute_correlations	(	const VariablesArray &	vars_samples,
		const IntResponseMap &	resp_samples,
		const StringSetArray &	dss_vals
	)

computes four correlation matrices for input and output data simple, partial, simple rank, and partial rank

This version is used when full variables objects are being processed. Calculates simple correlation, partial correlation, simple rank correlation, and partial rank correlation coefficients.

Referenced by NonDSampling::compute_statistics(), DDACEDesignCompExp::post_run(), FSUDesignCompExp::post_run(), and ParamStudy::post_run().

◆ compute_correlations() [2/2]

void compute_correlations	(	const RealMatrix &	vars_samples,
		const IntResponseMap &	resp_samples
	)

computes four correlation matrices for input and output data simple, partial, simple rank, and partial rank

This version is used when compact samples matrix is being processed. Calculates simple correlation, partial correlation, simple rank correlation, and partial rank correlation coefficients.

◆ values_to_ranks()

void values_to_ranks ( RealMatrix & valid_data )

private

replace sample values with their ranks, in-place

When converting values to ranks, uses the average ranks of any tied values

Referenced by SensAnalysisGlobal::compute_correlations().

◆ simple_corr()

void simple_corr	(	RealMatrix &	total_data,
		const int &	num_in,
		RealMatrix &	corr_matrix
	)

private

computes simple correlations, populating corr_matrix

Calculates simple correlation coefficients from a matrix of data (oriented factors x observations):

num_corr is number of rows of total data
num_in indicates whether only pairs of correlations should be calculated between pairs of columns (num_in vs. num_corr-num_in); if num_in = num_corr, correlations are calculated between all columns

References Dakota::center_matrix_rows(), and SensAnalysisGlobal::correl_adjust().

Referenced by SensAnalysisGlobal::compute_correlations().

◆ partial_corr()

void partial_corr	(	RealMatrix &	total_data,
		const int	num_in,
		const RealMatrix &	simple_corr_mat,
		RealMatrix &	corr_matrix,
		bool &	numerical_issues
	)

private

computes partial correlations, populating corr_matrix and numerical_issues

Calculates partial correlation coefficients between num_in inputs and numRows() - num_in outputs.

References Dakota::abort_handler(), Dakota::center_matrix_rows(), SensAnalysisGlobal::correl_adjust(), Dakota::qr(), Dakota::qr_rsolve(), and Dakota::singular_value_decomp().

Referenced by SensAnalysisGlobal::compute_correlations().

Member Data Documentation

◆ numFns

size_t numFns

protected

Reorder.

number of responses

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

SensAnalysisGlobal.hpp
SensAnalysisGlobal.cpp

Public Member Functions

Protected Member Functions

Protected Attributes

Private Member Functions

Static Private Member Functions

Private Attributes

Static Private Attributes

Detailed Description

Member Function Documentation

◆ compute_correlations() [1/2]

◆ compute_correlations() [2/2]

◆ values_to_ranks()

◆ simple_corr()

◆ partial_corr()

Member Data Documentation

◆ numFns