givp¶

GRASP-ILS-VND with Path Relinking — a direction-agnostic metaheuristic optimizer for continuous, integer, and mixed-variable problems, available in Python, Julia, Rust, C++, and R.

Highlights¶

SciPy-style API (Python) and idiomatic Julia API: pass an objective and bounds, get an OptimizeResult.
Minimize or maximize via the direction argument.
Composes GRASP construction, ILS perturbation, VND local search, and bidirectional Path Relinking with an elite pool.
Optional adaptive α, evaluation cache, convergence monitor, parallel candidate evaluation, and warm-start via initial_guess.
CLI entry point (givp run) — run the optimizer from the shell or pipe structured JSON output to AI agents and LLM tool calls (Python).
Typed result schema — OptimizeResult.to_dict() returns a JSON-safe dict with a closed TerminationReason enum, eliminating prompt-injection risk when output is passed to language models.
Fully typed package (ships py.typed for Python).

Installation¶

Python installation¶

pip install givp

Julia installation¶

cd julia
julia --project=. -e 'using Pkg; Pkg.instantiate()'

R installation¶

install.packages(
  "givp",
  repos = c("https://arnime.r-universe.dev", "https://cloud.r-project.org")
)

R package page: https://arnime.r-universe.dev/givp

Minimal example¶

Python example¶

import numpy as np
from givp import givp

def sphere(x: np.ndarray) -> float:
    return float(np.sum(x ** 2))

result = givp(sphere, [(-5.0, 5.0)] * 4)
print(result.x, result.fun)

Julia example¶

using GIVPOptimizer

sphere(x) = sum(x .^ 2)

result = givp(sphere, [(-5.0, 5.0) for _ in 1:4])
println(result.x, " ", result.fun)

See the Quickstart for richer examples and the API Reference for the full surface.