export_sample_sequence

Enable export of multilevel/multifidelity sample sequences to individual files

Specification

  • Alias: None

  • Arguments: None

Child Keywords:

Required/Optional

Description of Group

Dakota Keyword

Dakota Keyword Description

Optional (Choose One)

Tabular Format

custom_annotated

Export sample sequences enabling file format customization

annotated

Export sample sequences with descriptors

freeform

Export sample sequences without heading descriptors

Description

When this option is active, separate output files are written for each unique sample increment and are tagged by algorithm type, simulation interface, iteration count, level count, and the number of samples as described below. The data content is comprised of the input variables only, without corresponding responses, as an intended use case is to support evaluation of these sample sets offline.

Default Behavior

If not specified, the annotated format is assumed.

Expected Output

Separate output files are generated according to the following format: {ml/cv}_{interface_id}_{iteration_number}_{level_number}_{number_of_samples}.dat.

With respect to the algorithm type, interface id, and level counter, the following definitions are employed:

  • Multilevel MC with hierarchical model: ml_ is pre-pended for all sample increments and sample sets are tagged with the interface id from the HF model.

  • Control Variate MC with hierarchical model (deprecated): cv_ is prepended for all sample increments, shared sample sets are tagged with the HF interface id, and LF-only refinements are tagged by the LF interface id.

  • Multilevel-Control Variate MC with hierarchical model: employs a combination of exports from the two cases above. ml_ is pre-pended for all HF sample increments and sample sets are tagged with the interface id from the HF model; cv_ is prepended for all LF increments and sample sets are tagged with the LF interface id. Note that the LF model shares the same ml_ sample sets, but a redundant file is not created for this data.

  • Non-hierarchical approaches (Multifidelity MC, Approximate Control Variate): cv_ is prepended for all sample increments and sample sets are tagged with the corresponding model interface id. The level number corresponds to the index of the approximation model for LF increments and to the id of the truth model (number of LF approximations + 1) for shared increments.

With respect to iteration count, pilot samples are tagged with iteration 0.