Dakota
Version 6.20
Explore and Predict with Confidence
|
Derived approximation class for global basis polynomials. More...
Public Member Functions | |
SharedC3ApproxData () | |
default constructor | |
SharedC3ApproxData (ProblemDescDB &problem_db, size_t num_vars) | |
standard ProblemDescDB-driven constructor | |
SharedC3ApproxData (const String &approx_type, const UShortArray &approx_order, size_t num_vars, short data_order, short output_level) | |
on-the-fly constructor (no problem DB) | |
~SharedC3ApproxData () | |
destructor | |
size_t | regression_size () |
return number of FT unknowns using start_rank(), max_rank(), start_orders(), max_order() | |
size_t | max_rank_regression_size () |
return number of FT unknowns using maximum rank, start_orders(), max_order() | |
size_t | max_order_regression_size () |
return number of FT unknowns using start_rank(), max_rank(), and maximum basis order | |
size_t | max_regression_size () |
return number of FT unknowns using maxima for rank and basis order | |
void | set_parameter (String var, const UShortArray &val) |
set UShortArray attribute value based on identifier string | |
void | set_parameter (String var, unsigned short val) |
set unsigned short attribute value based on identifier string | |
void | set_parameter (String var, size_t val) |
set size_t attribute value based on identifier string | |
void | set_parameter (String var, bool val) |
set bool attribute value based on identifier string | |
void | set_parameter (String var, short val) |
set short attribute value based on identifier string | |
void | set_parameter (String var, double val) |
set double attribute value based on identifier string | |
void | set_parameter (String var, int val) |
set int attribute value based on identifier string | |
void | set_active_parameter (String var, const UShortArray &val) |
set active UShortArray attribute value based on identifier string | |
void | set_active_parameter (String var, unsigned short val) |
set active unsigned short attribute value based on identifier string | |
void | set_active_parameter (String var, size_t val) |
set active size_t attribute value based on identifier string | |
void | set_active_parameter (String var, int val) |
set active int attribute value based on identifier string | |
const UShortArray & | start_orders () const |
return active start value for basis order | |
UShortArray & | start_orders () |
return active start value for basis order (mutable) | |
unsigned short | max_order () const |
return active maximum value for basis order | |
unsigned short & | max_order () |
return active maximum value for basis order (mutable) | |
size_t | start_rank () const |
return active start value for expansion rank | |
size_t & | start_rank () |
return active start value for expansion rank (mutable) | |
size_t | max_rank () const |
return active maximum value for expansion rank | |
size_t & | max_rank () |
return active maximum value for expansion rank (mutable) | |
size_t | max_cross_validation_rank_candidates () const |
return maxCVRankCandidates | |
unsigned short | max_cross_validation_order_candidates () const |
return maxCVOrderCandidates | |
void | assign_start_ranks (SizetVector &start_ranks) const |
void | update_basis () |
update oneApproxOpts with active basis orders after an order change | |
void | update_basis (const UShortArray &start_orders, unsigned short max_order) |
update oneApproxOpts with passed basis orders after an order change | |
void | update_basis (size_t v, unsigned short start_order, unsigned short max_order) |
update oneApproxOpts for variable v with passed basis orders | |
Public Member Functions inherited from SharedApproxData | |
SharedApproxData () | |
default constructor More... | |
SharedApproxData (ProblemDescDB &problem_db, size_t num_vars) | |
standard constructor for envelope More... | |
SharedApproxData (const String &approx_type, const UShortArray &approx_order, size_t num_vars, short data_order, short output_level) | |
alternate constructor for envelope More... | |
SharedApproxData (const SharedApproxData &approx) | |
copy constructor More... | |
virtual | ~SharedApproxData () |
destructor | |
SharedApproxData | operator= (const SharedApproxData &approx) |
assignment operator | |
virtual void | clear_model_keys () |
reset initial state by clearing all model keys for an approximation | |
virtual void | integration_iterator (const Iterator &iterator) |
set integration driver for structured grid approximations | |
virtual void | rebuild () |
rebuilds the shared approximation data incrementally | |
virtual void | pre_push () |
push a previous state of the shared approximation data | |
virtual size_t | finalize_index (size_t i, const Pecos::ActiveKey &key) |
return index of i-th trial set within restorable bookkeeping sets | |
virtual void | pre_finalize () |
finalize the shared approximation data following a set of increments | |
virtual void | post_finalize () |
clean up popped bookkeeping following aggregation | |
virtual void | clear_inactive () |
clear inactive approximation data | |
virtual void | post_combine () |
clean up stored data sets after aggregation | |
virtual void | combined_to_active (bool clear_combined=true) |
promote aggregated data sets to active state | |
virtual void | update_basis_distribution_parameters (const Pecos::MultivariateDistribution &mvd) |
propagate updates to random variable distribution parameters to a polynomial basis | |
virtual void | configuration_options (const Pecos::ExpansionConfigOptions &ec_options) |
set ExpansionConfigOptions instance as a group specification | |
virtual void | configuration_options (const Pecos::BasisConfigOptions &bc_options) |
set BasisConfigOptions instance as a group specification | |
virtual void | configuration_options (const Pecos::RegressionConfigOptions &rc_options) |
set BasisConfigOptions instance as a group specification | |
virtual void | refinement_statistics_mode (short stats_mode) |
assign mode for statistics roll-up: {ACTIVE,COMBINED}_EXPANSION_STATS | |
virtual const Pecos::BitArrayULongMap & | sobol_index_map () const |
return set of Sobol indices that have been requested (e.g., as constrained by throttling) and are computable by a (sparse) expansion of limited order | |
const Pecos::ActiveKey & | active_model_key () const |
return active multi-index key | |
void | discrepancy_emulation_mode (short mode) |
set discrepancy emulation mode for approximating response differences | |
short | discrepancy_emulation_mode () const |
return discrepancy emulation mode for approximating response differences | |
bool | formulation_updated () const |
query whether the form of an approximation has been updated | |
void | formulation_updated (bool update) |
assign the status of approximation formulation updates | |
void | set_bounds (const RealVector &c_l_bnds, const RealVector &c_u_bnds, const IntVector &di_l_bnds, const IntVector &di_u_bnds, const RealVector &dr_l_bnds, const RealVector &dr_u_bnds) |
set approximation lower and upper bounds (currently only used by graphics) | |
std::shared_ptr< SharedApproxData > | data_rep () const |
returns dataRep for access to derived class member functions that are not mapped to the top SharedApproxData level | |
Static Public Member Functions | |
static size_t | regression_size (size_t num_v, size_t rank, size_t max_rank, const UShortArray &orders, unsigned short max_order) |
return number of FT unknowns given scalars: num vars, rank, order More... | |
Protected Member Functions | |
void | active_model_key (const Pecos::ActiveKey &key) |
activate an approximation state based on its multi-index key | |
void | construct_basis (const Pecos::MultivariateDistribution &mv_dist) |
construct the shared basis for an expansion-based approximation | |
short | discrepancy_reduction () const |
return the discrepancy type for approximations that support MLMF | |
void | random_variables_key (const BitArray &random_vars_key) |
assign key identifying a subset of variables that are to be treated as random for statistical purposes (e.g. expectation) | |
void | build () |
builds the shared approximation data from scratch | |
void | increment_order () |
increments polynomial expansion order (PCE, FT) | |
void | decrement_order () |
decrements polynomial expansion order (PCE, FT) | |
void | pop (bool save_surr_data) |
back out the previous increment to the shared approximation data | |
bool | push_available () |
queries availability of pushing data associated with a trial set | |
size_t | push_index (const Pecos::ActiveKey &key) |
return index for restoring trial set within stored data sets | |
void | post_push () |
clean up popped bookkeeping following push | |
void | pre_combine () |
aggregate the shared approximation data from current and stored states | |
bool | advancement_available () |
queries availability of advancing the approximation resolution | |
void | max_rank_advancement (bool r_advance) |
void | max_order_advancement (bool o_advance) |
bool | increment_max_rank () |
bool | increment_max_order () |
Protected Member Functions inherited from SharedApproxData | |
SharedApproxData (BaseConstructor, ProblemDescDB &problem_db, size_t num_vars) | |
constructor initializes the base class part of letter classes (BaseConstructor overloading avoids infinite recursion in the derived class constructors - Coplien, p. 139) More... | |
SharedApproxData (NoDBBaseConstructor, const String &approx_type, size_t num_vars, short data_order, short output_level) | |
constructor initializes the base class part of letter classes (BaseConstructor overloading avoids infinite recursion in the derived class constructors - Coplien, p. 139) More... | |
Protected Attributes | |
std::vector< OneApproxOpts * > | oneApproxOpts |
one-D approximation options (basis type, poly order, etc.) | |
MultiApproxOpts * | multiApproxOpts |
n-D approximation options, augmenting one-D options | |
bool | respScaling |
option to scale response data prior to regression | |
UShortArray | startOrders |
starting user specification for polynomial orders (from start_order scalar plus anisotropic dimension preference) | |
std::map< Pecos::ActiveKey, UShortArray > | startOrdersMap |
starting values for polynomial order (prior to adaptive refinement); for each model key, there is an array of polynomial orders per variable | |
unsigned short | kickOrder |
user specification for increment in order used within adapt_order | |
unsigned short | maxOrder |
maximum value for polynomial order from user spec | |
std::map< Pecos::ActiveKey, unsigned short > | maxOrderMap |
user specification for maximum order used within adapt_order; usually a scalar specification but can be adapted per model key for MAX_{ORDER,RANK_ORDER}_ADVANCEMENT refine types | |
bool | adaptOrder |
C3 FT can support CV over polynomial order in addition to adapt_rank. | |
UShortArray | combinedOrders |
polynomial basis order for combined expansion for each variable core | |
size_t | startRank |
starting user specification for rank (not augmented by dimension preference); Note: rank sequence spec is managed externally and becomes reflected in startRank model index mapping | |
std::map< Pecos::ActiveKey, size_t > | startRankMap |
starting values for rank (note: adapt_rank currently covers refinement); for each model index key, there is a scalar starting rank (recovered rank in C3FnrainPtrs can vary per core/variable and per QoI) | |
size_t | kickRank |
user specification for increment in rank used within adapt_rank | |
size_t | maxRank |
scalar user specification for maximum allowable rank when adapting | |
std::map< Pecos::ActiveKey, size_t > | maxRankMap |
user specification for maximum rank used within adapt_rank; usually a scalar specification but can be adapted per model key for MAX_{RANK,RANK_ORDER}_ADVANCEMENT refine types | |
bool | adaptRank |
internal C3 adaptation that identifies the best rank representation for a set of sample data based on cross validation | |
short | regressType |
type of regression solver for forming FT approximation | |
double | regressRegParam |
penalty parameter if regularized regression | |
double | solverTol |
tolerance on regression solver | |
double | solverRoundingTol |
tolerance for rounding (performing a truncation operation on a FT expansion) within the regression solver | |
double | statsRoundingTol |
tolerance for rounding (performing a truncation operation on a FT expansion) when post-processing an expansion: computing products for moments, combining expansions with c3axpy, etc. | |
size_t | maxSolverIterations |
maximum number of iterations for regression solver | |
int | crossMaxIter |
maximum number of iterations for (future) cross iteration solver | |
int | randomSeed |
C3 regression solver employs a random seed. | |
short | combineType |
type of discrepancy calculation: additive, multiplicative, or both | |
short | discrepReduction |
type of multilevel discrepancy emulation: distinct or recursive | |
short | allocControl |
type of multilevel strategy for sample allocation: ESTIMATOR_VARIANCE, RANK_SAMPLING, GREEDY | |
short | c3AdvancementType |
type of advancement strategy used in uniform refinement: {START_ORDER,START_RANK,MAX_ORDER,MAX_RANK,MAX_RANK_ORDER}_ADVANCEMENT | |
std::map< Pecos::ActiveKey, bool > | c3MaxRankAdvance |
flag indicating availability of rank advancement (accumulated from C3Approximation::advancement_available()) | |
std::map< Pecos::ActiveKey, bool > | c3MaxOrderAdvance |
flag indicating availability of order advancement (accumulated from C3Approximation::advancement_available()) | |
unsigned short | maxCVOrderCandidates |
restrict the number of candidates within cross validation for order (by increasing start order when needed as max order is advanced) | |
size_t | maxCVRankCandidates |
restrict the number of candidates within cross validation for rank (by increasing start rank when needed as max rank is advanced) | |
SizetArray | randomIndices |
indices for random subset when approximating in all-variables mode | |
std::map< Pecos::ActiveKey, size_t > | poppedCounts |
number of instances within the popped arrays (mostly a placeholder for supporting push_available()) | |
Protected Attributes inherited from SharedApproxData | |
size_t | numVars |
number of variables in the approximation | |
String | approxType |
approximation type identifier | |
short | buildDataOrder |
order of the data used for surrogate construction, in ActiveSet request vector 3-bit format. More... | |
short | outputLevel |
output verbosity level: {SILENT,QUIET,NORMAL,VERBOSE,DEBUG}_OUTPUT | |
Pecos::ActiveKey | activeKey |
key indicating the active model or model-pair used for approximation data | |
short | discrepEmulationMode |
approach for emulating response differences: {DEFAULT,DISTINCT,RECURSIVE}_EMULATION | |
String | modelExportPrefix |
Prefix for model export files. | |
unsigned short | modelExportFormat |
Bitmapped format request for exported models. | |
RealVector | approxCLowerBnds |
approximation continuous lower bounds (used by 3D graphics and Surfpack KrigingModel) | |
RealVector | approxCUpperBnds |
approximation continuous upper bounds (used by 3D graphics and Surfpack KrigingModel) | |
IntVector | approxDILowerBnds |
approximation continuous lower bounds | |
IntVector | approxDIUpperBnds |
approximation continuous upper bounds | |
RealVector | approxDRLowerBnds |
approximation continuous lower bounds | |
RealVector | approxDRUpperBnds |
approximation continuous upper bounds | |
std::map< Pecos::ActiveKey, bool > | formUpdated |
tracker for changes in order,rank configuration since last build (used by DataFitSurrModel::rebuild_approximation()) | |
Friends | |
class | C3Approximation |
Derived approximation class for global basis polynomials.
The SharedC3ApproxData class provides a global approximation based on basis polynomials. This includes orthogonal polynomials used for polynomial chaos expansions and interpolation polynomials used for stochastic collocation.
|
inlinestatic |
return number of FT unknowns given scalars: num vars, rank, order
simplified estimation for scalar-valued rank and order (e.g., from start rank/order user specification) inline size_t SharedC3ApproxData:: regression_size(size_t num_v, size_t rank, size_t order) { Each dimension has its own rank within the product of function cores. This fn estimates for the case where rank and order are either constant across dimensions or averaged into a scalar.
the first and last core contribute p*r terms the middle cores contribute r*r*p terms
size_t p = order+1.; switch (num_v) { case 1: return p; break; // collapses to a 1D PCE case 2: return 2.*p*rank; break; // first and last core, no middle default: return p*rank*(2. + (num_v-2)*rank); break; // first,last,middle } } simplified estimation for scalar-valued rank and vector-valued order (e.g., from start rank/start order/dimension pref user specification)
References SharedC3ApproxData::max_order(), SharedC3ApproxData::max_rank(), and Dakota::sum().