| register_atomic_objective_internal | R Documentation |
Internal helper used by objective setter functions to define the active
single-objective configuration of a Problem object and, optionally,
register that objective as an atomic objective for multi-objective workflows.
x |
A |
alias |
Optional character scalar used to register the objective as an
atomic objective. If |
objective_id |
Character string giving the stable internal identifier of
the objective, for example |
model_type |
Character string giving the model-builder label associated
with this objective, for example |
objective_args |
A list of objective-specific arguments to be stored with the objective definition. |
sense |
Character string giving the optimization direction. Must be
either |
In multiscape, an atomic objective is a fully specified
objective definition that can later be reused by a multi-objective method
such as a weighted-sum formulation, an \epsilon-constraint method,
AUGMECON, or other objective-orchestration procedures.
Each atomic objective is identified by:
a stable internal identifier objective_id,
a solver-facing model label model_type,
a list of objective-specific arguments stored in
objective_args,
an optimization sense, either "min" or "max",
and, optionally, a user-facing identifier alias.
If alias is not NULL, the objective is stored in
x$data$objectives[[alias]]. This makes it possible to refer to the
same objective later by a stable user-facing name. For example, a user may
register objectives under aliases such as "cost",
"benefit", or "frag" and then pass those aliases to a
multi-objective method.
If alias is NULL, no atomic-objective entry is created in
x$data$objectives. In that case, the calling function still defines
the currently active single-objective configuration through
x$data$model_args, but no reusable multi-objective registration is
created.
Thus, this helper supports two complementary modes:
single-objective mode: only the active objective is stored,
multi-objective-ready mode: the active objective is stored and also registered under an alias for later reuse.
Conceptually, if an objective function is denoted by f(x), this helper
does not itself define the mathematical form of f; rather, it stores
the metadata required so that downstream code can reconstruct the correct
objective expression, its direction of optimization, and its user-visible
identity.
An updated Problem object.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.