fixed_seed

Reuses the same seed value for multiple random sampling sets

Specification

• Alias: None

• Arguments: None

• Default: not fixed; pattern varies run-to-run

Description

The fixed_seed flag is relevant if multiple sampling sets will be generated over the coarse of a Dakota analysis. This occurs when using advance methods (e.g., surrogate-based optimization, optimization under uncertainty). The same seed value is reused for each of these multiple sampling sets, which can be important for reducing variability in the sampling results.

Default Behavior

The default behavior is to not use a fixed seed, as the repetition of the same sampling pattern can result in a modeling weakness that an optimizer could potentially exploit (resulting in actual reliabilities that are lower than the estimated reliabilities). For repeatable studies, the seed must also be specified.

Examples

method
  sampling
    sample_type lhs
    samples = 10
    fixed_seed