nlmixr2auto (version 1.0.0)

penaltyControl: Configure penalty settings for model evaluation

Description

Defines rules governing penalty assignment during model adequacy evaluation.

Usage

penaltyControl(
  penalty.value = 10000,
  step.penalties = list(rse = c(10, 10000), shrinkage = c(10, 10000), bsv = c(10, 10000),
    sigma = list(add = c(10, 10000), prop = c(10, 10000)), correlation = c(10, 10000)),
  bounds = param.bounds(),
  thresholds = list(),
  penalty.terms = c("total")
)

Value

A list containing the full penalty configuration for use in fitness().

Arguments

penalty.value

Numeric. Constant penalty assigned to binary violations and bound constraints.

step.penalties

A named list defining penalty magnitudes used in step-wise procedures. Each element must contain a numeric vector of length two representing penalty levels for moderate and critical deviations.

bounds

A list specifying lower and upper parameter limits, as returned by param.bounds(). The structure can include limits for theta, omega, sigma, and correlation terms.

thresholds

A named list describing evaluation rules for RSE and shrinkage. Each component must include a field named method, with value binary or step, together with the corresponding limit definition:

• If method = binary: a single cutoff value stored in threshold

• If method = step: two deviation boundaries stored in step.levels

penalty.terms

Character vector specifying which components are considered when penalties are reported. Recognized entries include: rse, shrinkage, theta, omega, sigma, correlation, covariance, and total. If total is included, penalties are aggregated across all components and any other entries are ignored.

Author

Zhonghui Huang

Details

Penalization may be triggered by exceeding predefined parameter bounds (fixed-effect and variance-covariance elements) or by surpassing thresholds for relative standard error (RSE) or shrinkage criteria. Binary and step-wise penalty procedures are supported.

See Also

param.bounds(), fitness().

Examples

# Default configuration
penaltyControl()

# Custom bounds for selected fixed-effect parameters
penaltyControl(bounds = param.bounds(
  theta = list(lower = list(cl = 0.01, vc = 0.01))
))

# Binary penalty method for RSE
penaltyControl(thresholds = list(
  rse = list(method = "binary", threshold = 40)
))