Benchmark suite

pymetaheuristics.benchmarks provides Problem instances with known best values, independent of any heuristic. Use get(name), all_benchmarks() or BENCHMARKS; build custom instances with the factories knapsack, tsp, continuous (build instances; randomness comes from the heuristic's rng).

Name	Representation	Objective	Constraint	Known best
knapsack-3	list of 0/1 per item	maximize value	weight <= capacity (50)	220 (brute force)
knapsack-10	list of 0/1 per item	maximize value	weight <= capacity (269)	295 (brute force)
tsp-ring8	permutation of 8 cities, closed tour	minimize length	valid permutation	8.0 (brute force)
tsp-grid9	permutation of 3x3 grid points	minimize length	valid permutation	8 + sqrt(2) (brute force)
sphere-5	5 floats in [-5.12, 5.12]	minimize sum x^2	within bounds	0 (analytic)
rastrigin-5	5 floats in [-5.12, 5.12]	minimize Rastrigin	within bounds	0 (analytic)

Each Benchmark holds name, problem, known_best, known_best_solution (a feasible solution reaching it), source and criteria.

Evaluation criteria

Run a heuristic with a fixed budget and report gap(benchmark, value) for its best feasible value: |value - known_best| / max(|known_best|, 1), so 0 means optimal. Optionally also report evaluations needed to reach known_best (e.g. via target_value termination). Compare heuristics on the same benchmark and seed.