Skip to content

givp

givp

`givp`¶

givp.givp ¶

givp(
    func: ObjectiveFn,
    bounds: BoundsLike,
    *,
    num_vars: int | None = None,
    minimize: bool | None = None,
    direction: str | None = None,
    config: GIVPConfig | None = None,
    initial_guess: Sequence[float] | None = None,
    initial_guesses: (
        Sequence[Sequence[float]] | None
    ) = None,
    iteration_callback: IterationCallback | None = None,
    seed: int | None = None,
    verbose: bool = False
) -> OptimizeResult

Minimize (or maximize) a scalar function with GRASP-ILS-VND-PR.

Parameters:

Name	Type	Description	Default
`func`	`ObjectiveFn`	Objective callable mapping a 1-D `np.ndarray` to a scalar.	required
`bounds`	`BoundsLike`	Either a sequence of `(low, high)` pairs (SciPy style) or a `(lower, upper)` tuple of two equally-sized sequences.	required
`num_vars`	`int \| None`	Optional explicit number of variables. Inferred from `bounds` when omitted.	`None`
`minimize`	`bool \| None`	Boolean flag for the optimization sense. `True` minimizes, `False` maximizes. Preferred over `direction` for new code.	`None`
`direction`	`str \| None`	`'minimize'` or `'maximize'` (SciPy/Optuna style). Kept for backward compatibility. Ignored when `minimize` is given. Defaults to `'minimize'` when neither flag is supplied.	`None`
`config`	`GIVPConfig \| None`	Algorithm hyper-parameters. `GIVPConfig()` is used when omitted. Any sense field on `config` is overridden by the explicit `minimize`/`direction` kwargs when provided.	`None`
`initial_guess`	`Sequence[float] \| None`	Optional warm-start vector, evaluated and inserted in the elite pool before the first iteration.	`None`
`initial_guesses`	`Sequence[Sequence[float]] \| None`	Optional collection of warm-start vectors to seed the elite pool with multiple known-good solutions.	`None`
`iteration_callback`	`IterationCallback \| None`	Optional callable invoked once per outer iteration with `(iteration, best_cost_in_core_sign, best_solution)`.	`None`
`seed`	`int \| None`	Optional integer seed for full reproducibility. When given, every internal RNG is derived deterministically from this seed, so two calls with the same inputs return the same result.	`None`
`verbose`	`bool`	If True, prints progress information to stdout.	`False`

Returns:

Name	Type	Description
`OptimizeResult`	`OptimizeResult`	Dataclass with `x` (best solution), `fun` (best
	`OptimizeResult`	objective value in the user's original sign) and metadata.

Raises:

Type	Description
`ValueError`	If both `minimize` and `direction` are passed with conflicting values, or if `direction` is not one of `'minimize'` / `'maximize'`.