Module problem
Expand description
Standard black-box optimization problems and the best-value curve.
Every problem here is a minimization problem (kurobako’s convention, and
the direction the harness drives), continuous, with a
known global minimum. They are the classic synthetic test functions —
sphere, Rastrigin, Ackley, Branin — chosen so that a smooth one (sphere,
Branin) makes a learning sampler’s advantage over uniform random
unmistakable, while the multimodal ones (Rastrigin, Ackley) keep the
comparison honest.
The objective functions are pure and in-house (D4): the reference clones
under references/ were consulted for the canonical formulae and domains,
none of their code is used.
Structs§
- Cost
Problem - A cost-varying two-objective problem for the CARBS oracle: a bowl in the tuning parameters whose loss is bought down by a resource knob that also makes each evaluation more expensive.
- Curve
- The best objective value seen up to and including each trial.
- Problem
- A synthetic benchmark problem: a boxed continuous domain and a pure objective to minimize.