Dakota Reference Manual  Version 6.15
Explore and Predict with Confidence
 All Pages
soft_convergence_limit


Maximum number of iterations without improvement in cross-validation

Specification

Alias: none

Argument(s): INTEGER

Default: 0

Description

The purpose of this control is to cease auto-refinement when improvement in the cross-validation score appears to have stalled or occurs too slowly.

A rolling average of improvement in the cross-validation score is computed over soft_convergence_limit iterations. Auto-refinement is halted if the average becomes smaller than the convergence_tolerance. The average is computed but ignored until the number of iterations has exceeded soft_convergence_limit.

The default setting of 0 disables soft convergence.