Skip to contents

Calibrate a fixed-scale soft Dual-Anchor trade-off

DPprior_dual_soft.Rd

Calibrate a fixed-scale soft Dual-Anchor trade-off

Usage

DPprior_dual_soft(
  fit,
  target,
  lambda,
  max_iter = 100L,
  M_fit = .QUAD_NODES_DEFAULT,
  M_verify = NULL,
  log_bounds = .LOG_BOUNDS_DEFAULT,
  control = list(),
  allow_approximate = FALSE,
  start = NULL,
  ...
)

Arguments

fit

A K-only DPprior_fit object.

target

Named weight target with metric, relation, and value; threshold or probability is required when the metric needs it. Relations are "target", "at_most", or "at_least".

lambda

Mandatory finite scalar satisfying 0 < lambda <= 1.

max_iter

Maximum iterations for each optimizer attempt.

M_fit

Selected quadrature order.

M_verify

Independent verification order; by default at least twice M_fit and at least M_fit + 40.

log_bounds

Two finite log-parameter bounds.

control

Optimizer and verification controls.

allow_approximate

Return a finite but unusable approximate candidate when TRUE; otherwise signal a typed condition retaining the result.

start

Optional positive c(a,b) warm start.

...

Must be empty. Unknown or legacy hard-mode arguments are rejected as typed invalid input without evaluating their expressions.

Value

A canonical dpprior.result/1 DPprior_dual_soft object. The tradeoff extension records the fixed input-derived K scales, lambda, weight target, component losses, and total objective. Soft objects never contain constraint_satisfied. Unapproved approximate, infeasible, or failed outcomes are signalled as typed conditions whose condition$result retains the complete object; allow_approximate changes return policy, not scientific status.

Details

A soft result is decision-ready only when its mode-specific contract and status, usable, and verified fields permit that use. Lambda is a trade-off weight, not a constraint probability or a hard-satisfaction certificate.

See also

Other elicitation: DPprior_a1(), DPprior_a2_kl(), DPprior_a2_newton(), DPprior_dual(), DPprior_dual_hard(), DPprior_fit(), DPprior_target_K()