Fit a Gamma hyperprior from an occupied-cluster target

DPprior_fit() constructs one canonical $K_J$ target, or accepts one from DPprior_target_K, dispatches to A1, A2-MN, or A2-KL, and returns the numerical producer's native canonical result without flattening or reinterpreting its evidence.

Usage

DPprior_fit(
  J,
  mu_K = NULL,
  var_K = NULL,
  confidence = c("medium", "low", "high"),
  method = c("A2-MN", "A1", "A2-KL"),
  target_pmf = NULL,
  check_diagnostics = TRUE,
  warn_dominance = NULL,
  M = .QUAD_NODES_DEFAULT,
  verbose = FALSE,
  warning_policy = NULL,
  a1_projection = c("error", "nearest"),
  cv_K = NULL,
  K_interval = NULL,
  target_K = NULL,
  ...
)

Arguments

J: Integer design size. When target_K is supplied, J must match the target exactly.
mu_K, var_K: Optional direct target mean and variance.
confidence: Optional "low", "medium", or "high" qualitative uncertainty route.
method: One of "A2-MN", "A1", or "A2-KL".
target_pmf: Optional strict target PMF on $1,\ldots,J$.
check_diagnostics: Whether to attach freshly verified canonical diagnostics.
warn_dominance: Deprecated first-size-biased-weight warning adapter.
M: Selected quadrature order.
verbose: Whether the numerical producer prints progress.
warning_policy: Optional ordinary list with exact canonical fields estimand, direction, weight_threshold, and action_threshold. The legacy threshold field is accepted here and normalized immediately for compatibility.
a1_projection: A1 projection policy: "error" or an explicit opt-in to "nearest".
cv_K: Optional coefficient of variation $SD(K_J)/E(K_J)$.
K_interval: Optional canonical interval-target specification.
target_K: Optional validated dpprior.target/1 K target. It cannot be combined with scalar target arguments.
...: Must be empty.

Value

A canonical dpprior.result/1 DPprior_fit object. Use parameters$a and parameters$b for the Gamma shape and rate; target$K, achieved, and residuals for the scientific claim; and computation, verification, and provenance for its audit trail. A result is decision-ready only when its mode-specific contract and status, usable, and verified fields permit that use. If a backend signals a typed calibration condition, the canonical target or result is retained in condition$result.

Details

Supply exactly one uncertainty route: direct var_K, confidence, cv_K, K_interval, or target_pmf. A pre-built target_K is an alternative to all scalar target arguments. Compatibility aliases are non-authoritative; new code should use the canonical nested fields.

Examples

target <- DPprior_target_K(J = 50, mu_K = 5, var_K = 8)
fit <- DPprior_fit(
  J = 50, target_K = target, method = "A2-MN",
  check_diagnostics = FALSE
)
fit$parameters
#> $a
#> [1] 2.036093
#> 
#> $b
#> [1] 1.605054
#> 
#> $parameterization
#> [1] "log(shape), log(rate)"
#> 
fit[c("status", "usable", "verified")]
#> $status
#> [1] "converged"
#> 
#> $usable
#> [1] TRUE
#> 
#> $verified
#> [1] TRUE
#>