Run post-optimisation convergence checks on a fitted SPoRC model

Evaluates four convergence criteria on the output of a fitted RTMB model: (1) finiteness of the joint negative log-likelihood, (2) maximum absolute gradient of fixed-effect parameters, (3) positive-definiteness of the Hessian, and (4) finiteness and magnitude of parameter standard errors and pairwise correlations. A diagnostic message is printed for each failed check identifying the offending parameter or parameter pair.

Usage

post_optim_sanity_checks(
  sd_rep,
  rep,
  gradient_tol = 0.001,
  se_tol = 100,
  corr_tol = 0.99
)

Arguments

sd_rep

sdreport object returned by RTMB::sdreport. Must contain $gradient.fixed, $par.fixed, $pdHess, and $cov.fixed.

rep

Report list returned by obj$rep after fitting a SPoRC model. Must contain $jnLL.

gradient_tol

Numeric. Maximum tolerated absolute gradient for any fixed-effect parameter. Default 1e-3; values above this threshold suggest the optimiser did not reach a local minimum.

se_tol

Numeric. Maximum tolerated parameter standard error. Default 100; very large SEs indicate poorly identified parameters.

corr_tol

Numeric. Maximum tolerated absolute pairwise parameter correlation. Default 0.99; values approaching 1 indicate near-redundant parameters.

Value

Logical. TRUE if all four checks pass; FALSE if any check fails. In either case, informative messages are printed via message.

See also

Other Utility: fit_model(), get_logistN_Sigma(), get_par_est_info(), rho_trans(), set_data_indicator_unused()