Overrides, queries, or resets the runtime policy consulted by
edi_optimization_dispatch_policy() for which optimization algorithm
("newton_raphson", "lbfgs", or "irls") an inference
class's C++ model-fitting backend uses by default; see
get_optimization_dispatch_policy for the built-in default
table and the empirical rationale behind its per-family choices.
Value
Invisible NULL when policy is supplied (a mutation), or
invisibly the current policy configuration list when called for its
side-effect-free query/reset value.
Details
Call with no arguments (policy = NULL, reset = FALSE)
to retrieve the current configuration without changing it. Pass a named
list to policy to merge new/overriding entries into the current
configuration via modifyList (default_alg
and/or inference_class_overrides can each be supplied
independently). Pass reset = TRUE to discard any accumulated
overrides and restore the package's built-in default policy exactly as
returned by get_optimization_dispatch_policy. Note that this
only changes the default algorithm consulted when a fitting call
does not itself specify optimization_alg explicitly; an explicit
per-call argument still takes precedence.
See also
get_optimization_dispatch_policy for the policy
schema and built-in defaults; set_cold_start_dispatch_policy
and set_warm_start_dispatch_policy for the analogous setters
governing cold-start and warm-start behavior;
tune_EDI_for_this_machine, which calls this setter with
machine-measured, convergence-checked overrides rather than
hand-picked ones.