auto_refinement

Experimental auto-refinement of surrogate model

Topics

surrogate_models

Specification

• Alias: None

• Arguments: None

• Default: no refinement

Child Keywords:

Required/Optional	Description of Group	Dakota Keyword	Dakota Keyword Description
Optional		max_iterations	Number of iterations allowed for optimizers and adaptive UQ methods
Optional		max_function_evaluations	Number of function evaluations allowed for optimizers
Optional		convergence_tolerance	Cross-validation threshold for surrogate convergence
Optional		soft_convergence_limit	Maximum number of iterations without improvement in cross-validation
Optional		cross_validation_metric	Choice of error metric to satisfy

Description

(Experimental option) Automatically refine the surrogate model until desired cross-validation quality is achieved. Refinement is accomplished by iteratively adding more data to the training set until the cross-validation convergence_tolerance is achieved, or max_function_evaluations or max_iterations is exceeded.

The amount of new training data that is incorporated each iteration is specified in the DACE method that is referred to by the model’s dace_method_pointer. See method-sampling-refinement_samples for more information.