![]() |
Dakota Reference Manual
Version 6.16
Explore and Predict with Confidence
|
Portion of batch size dedicated to exploration in parallel EGO
Alias: none
Argument(s): INTEGER
Refinement candidates are generated by an acquisition function such as maximum expected improvement, which balances exploration and exploitation. Refinement candidates can also be generated by purely explorative metrics such as maximum prediction variance. For a specified batch_size, exploration specifies the subset of this total that will be dedicated to pure exploration of the parameter space.
Default Behavior All of the batch size is devoted to the standard acquisition approach, balancing exploration and exploitation.
method,
efficient_global
seed = 1237
batch_size = 8 # total
exploration = 2 # 2 out of 8