heuropt

142 Commits 1 Branch 8 Tags

3 Commits

Include renames

Author	SHA1	Message	Date
swaits	4a59041d1a	style: apply rustfmt drift across the crate	2026-05-05 11:40:14 -06:00
swaits	aba9dbf469	build(bench): expand gungraun bench suite to cover every algorithm Goes from 6 benchmarks to ~25: Pareto utilities (existing): - non_dominated_sort_2d (n=50, 200) - crowding_distance_2d (n=50, 200) - hypervolume_2d (n=30, 100) - hypervolume_nd_3d (n=30, 100) Single-objective algorithms (all measured at "one short run"): - random_search, hill_climber, one_plus_one_es, simulated_annealing - genetic_algorithm, particle_swarm, differential_evolution, tlbo - cma_es, separable_nes, nelder_mead, bayesian_opt, tpe Multi-objective algorithms (one short run each): - nsga2, nsga3, spea2, moead, mopso, ibea, sms_emoa - hype, pesa2, epsilon_moea, age_moea, grea, knea, rvea Each uses a tiny problem with realistic-shape parameters (small pop, few generations, tight bounds) so the benchmark exercises each algorithm's inner loop cost rather than dominated by RNG init or config parsing.	2026-05-05 11:03:00 -06:00
swaits	0b31b266ef	build(deps): add gungraun (was iai-callgrind) instruction-count benches Wire `gungraun` 0.18 as a dev-dependency and a `benches/` directory with instruction-count benchmarks for the algorithmic hot paths. Why gungraun and not criterion: heuropt's hot paths are deterministic numerical loops where wall-clock noise dominates real differences. gungraun runs each benchmark under valgrind/callgrind once and reports exact instruction counts — stable across machines and CI runners, detects sub-microsecond regressions cleanly. Benchmarks added: - pareto::non_dominated_sort (the inner loop of every Pareto MOEA) - pareto::crowding_distance (NSGA-II survival selection) - metrics::hypervolume_nd (HSO recursion, used by SMS-EMOA) - internal::cholesky (BO's per-step posterior factorization) - algorithms::nsga2 single generation (end-to-end smoke check) - algorithms::cma_es single generation (eigendecomposition cost) Tracked size only — these aren't part of the regular CI matrix because they need valgrind installed. Run with `cargo bench` locally. Wired via the standard `[[bench]]` Cargo entries with `harness = false` so gungraun's main_macro does the dispatch.	2026-05-05 10:36:29 -06:00

Author

SHA1

Message

Date

swaits

4a59041d1a

style: apply rustfmt drift across the crate

2026-05-05 11:40:14 -06:00

swaits

aba9dbf469

build(bench): expand gungraun bench suite to cover every algorithm

Goes from 6 benchmarks to ~25:

Pareto utilities (existing):
- non_dominated_sort_2d (n=50, 200)
- crowding_distance_2d (n=50, 200)
- hypervolume_2d (n=30, 100)
- hypervolume_nd_3d (n=30, 100)

Single-objective algorithms (all measured at "one short run"):
- random_search, hill_climber, one_plus_one_es, simulated_annealing
- genetic_algorithm, particle_swarm, differential_evolution, tlbo
- cma_es, separable_nes, nelder_mead, bayesian_opt, tpe

Multi-objective algorithms (one short run each):
- nsga2, nsga3, spea2, moead, mopso, ibea, sms_emoa
- hype, pesa2, epsilon_moea, age_moea, grea, knea, rvea

Each uses a tiny problem with realistic-shape parameters (small pop,
few generations, tight bounds) so the benchmark exercises each
algorithm's *inner loop cost* rather than dominated by RNG init or
config parsing.

2026-05-05 11:03:00 -06:00

swaits

0b31b266ef

build(deps): add gungraun (was iai-callgrind) instruction-count benches

Wire `gungraun` 0.18 as a dev-dependency and a `benches/` directory
with instruction-count benchmarks for the algorithmic hot paths.

Why gungraun and not criterion: heuropt's hot paths are deterministic
numerical loops where wall-clock noise dominates real differences.
gungraun runs each benchmark under valgrind/callgrind once and reports
exact instruction counts — stable across machines and CI runners,
detects sub-microsecond regressions cleanly.

Benchmarks added:
- pareto::non_dominated_sort  (the inner loop of every Pareto MOEA)
- pareto::crowding_distance   (NSGA-II survival selection)
- metrics::hypervolume_nd     (HSO recursion, used by SMS-EMOA)
- internal::cholesky          (BO's per-step posterior factorization)
- algorithms::nsga2 single generation (end-to-end smoke check)
- algorithms::cma_es single generation (eigendecomposition cost)

Tracked size only — these aren't part of the regular CI matrix because
they need valgrind installed. Run with `cargo bench` locally.

Wired via the standard `[[bench]]` Cargo entries with `harness = false`
so gungraun's main_macro does the dispatch.

2026-05-05 10:36:29 -06:00