reuse_points

Surrogate model training data reuse control

Topics

surrogate_models

Specification

• Alias: reuse_samples

• Arguments: None

• Default: all for import; none otherwise

Child Keywords:

Required/Optional	Description of Group	Dakota Keyword	Dakota Keyword Description
Required (Choose One)	Reuse Domain	all	Option for reuse_points
		region	Option for reuse_points
		none	Option for reuse_points

Description

Dakota’s global surrogate methods rely on training data, which can either come from evaluation of a “truth” model, which is generated by the method specified with model-surrogate-global-dace_method_pointer, from a file of existing training data, identified by model-surrogate-global-import_build_points_file, or both.

The reuse_points keyword controls the amount of training data used in building a surrogate model, either initially, or during iterative rebuild, as in surrogate-based optimization. If model-surrogate-global-import_build_points_file is specified, reuse_points controls how the file contents are used. If used during iterative rebuild, it controls what data from previous surrogate builds is reused in building the current model.

• all (default for file import) - use all points in the file or available from previous builds

• region - use only the points falling in the current trust region (see method-surrogate_based_local)

• none (default when no import) - ignore the contents of the file or previous build points, and gather new training data using the specified DACE method