Compute the saddlepoint negative log-likelihood — compute.spa.negll • saddlepoint

Computes either the "standard" or "zeroth-order" saddlepoint approximation to the negative log–likelihood, with options for how the saddlepoint t-vector is obtained.

Usage

compute.spa.negll(
  parameter_vector,
  observed.data,
  cgf,
  tvec.hat = NULL,
  gradient = FALSE,
  hessian = FALSE,
  spa_method = "standard",
  tvec_source = c("auto", "user", "analytic", "newton", "solver_atomic"),
  solver_fun = saddlepoint.solve,
  newton_t_init = NULL,
  ...
)

Arguments

parameter_vector: Numeric vector of parameters.
observed.data: Numeric vector of observations.
cgf: A CGF object.
tvec.hat: Optional numeric t-vector (one-time tape).
gradient, hessian: Logical flags for derivatives.
spa_method: "standard" or "zeroth".
tvec_source: One of "auto","user","analytic","newton","solver_atomic".
solver_fun: Numeric solver used when tvec_source="solver_atomic". Default saddlepoint.solve.
newton_t_init: Optional start for t in the Newton solver path.
...: Passed to solver_fun (when used).

Value

list(vals=..., gradient=..., hessian=...)