truth_fixed_by_pilot
Option to suppress any increment to the number of truth samples
Specification
Alias: None
Arguments: None
Description
In some cases, it may be impractical to increment an initial set of
truth model samples, for example due to their extreme computational
cost or due to the use of legacy data sets that can no longer be
augmented. To fix the number of truth model evaluations to only those
prescribed in the pilot sample, the truth_fixed_by_pilot
keyword
can be used.
Default Behavior
The default is to allow the number of truth samples to advance beyond those specified in the pilot, while seeking an optimal sample profile.
Additional Discussion
The optimization problem is formulated using linear constraints and a reduced set of design variables, one per approximation model.
This option is most relevant in the default online_pilot
solution
mode, which would normally iterate to determine the optimal number of
truth samples. With truth_fixed_by_pilot
, the number of truth
samples is fixed within the optimization and no iteration occurs.
For the pilot_projection
solution mode, the optimization
formulation searches only over the approximation samples, providing
an optimal projection subject to the fixed truth constraint.
The offline_pilot
solution mode is not iterative and assumes that
the offline pilot has accurately estimated the correlations and
variances. In this case, the online profile is independent and it is
not necessary to refine the truth model samples beyond the initial
estimate of the optimal allocation. The truth_fixed_by_pilot
specification has no additional effect beyond these existing constraints.