Dakota Reference Manual  Version 6.12
Explore and Predict with Confidence
 All Pages
weights


Specify weights for each objective function

Specification

Alias: calibration_weights least_squares_weights

Argument(s): REALLIST

Default: equal weights

Description

Specifies relative emphasis through weights (multipliers) w on residual elements:

\[ f = \sum_{i=1}^{n} w_i R_i^2 = \sum_{i=1}^{n} w_i (y^{Model}_i - y^{Data}_i)^2 \]

Length: The weights must have length equal to calibration_terms. Thus, when scalar and/or field responses are specified, the number of weights must equal the number of scalars plus the number of fields, not the total elements in the fields.

Default Behavior If weights are not specified, then each residual is given equal weighting.

Usage Tips:

Weights are applied as multipliers, scales as charateristic values / divisors.

When scaling is active, it is applied to calibration terms after any residual formation (accounting for experimental data and optionally measurement error covariance), and before any weights. See the equations in calibration_terms.