Dakota Reference Manual  Version 6.4
Large-Scale Engineering Optimization and Uncertainty Analysis
 All Pages
variable_neighborhood_search


Percentage of evaluations to do to escape local minima.

Specification

Alias: none

Argument(s): REAL

Default: 0.0

Description

The variable_neighborhood_search keyword is used to set the percentage (in decimal form) of function evaluations used to escape local minima. The mesh adaptive direct search method will try to perform a maximum of that percentage of the function evaluations within this more extensive search.

Default Behavior

By default, variable_neighborhood_search is not used.

Usage Tips

Using variable_neighborhood_search results in an increased number of function evaluations. If the desired result is a local minimum, the added cost is of little or no value, so the recommendation is not to use it. If the desired result is the best local minimum possible within a computational budget, then there is value in setting this parameter. Note that the higher the value, the greater the computational cost.

Examples

The following example shows the syntax used to set variable_neighborhood_search.

method
  mesh_adaptive_search
    seed = 1234
    variable_neighborhood_search = 0.1