solver_metric
Metric employed during numerical solutions in sampling-based multifidelity UQ methods.
Specification
Alias: None
Arguments: None
Default: average_estimator_variance
Child Keywords:
Required/Optional |
Description of Group |
Dakota Keyword |
Dakota Keyword Description |
---|---|---|---|
Required (Choose One) |
Optimization Solver Metric |
Utilize the estimator variance averaged over the QoI as the solver metric for sampling-based multifidelity methods. |
|
Utilize a p-norm over the vector of QoI estimator variances as the solver metric for sampling-based multifidelity methods. |
|||
Utilize the maximum estimator variance as the solver metric for sampling-based multifidelity methods. |
Description
For sampling-based multifidelity UQ methods, this specification selects the metric to be used during numerical solutions for sample allocations per model. When there are multiple quantities of interest (QoI), the vector of QoI estimator variances needs to be converted to a scalar for use by the numerical solver. Options include the average, maximum, and p-norm of the QoI vector, with the default of average estimator variance.
Examples
method,
approximate_control_variate acv_mf
solver_metric norm_estimator_variance norm_order = 2 # objective fn for budget-constrained
solution_mode online_pilot
pilot_samples = 50 seed = 8674132
max_function_evaluations = 500