hyperopt-samplers

Pluggable Sampler implementations for hyperopt-rs, the Optuna-shaped hyperparameter optimization framework. All are interchangeable through one Study API:

• RandomSampler — independent random draws; the baseline.
• GridSampler — exhaustive enumeration over a caller-provided grid.
• TpeSampler — adaptive Tree-structured Parzen Estimator (wraps the tpe crate).
• CmaEsSampler — Covariance Matrix Adaptation Evolution Strategy: a from-scratch (μ/μ_w, λ) implementation with its own symmetric eigensolver and reflective bound handling. On a 3-D sphere with an 80-trial budget it reaches a mean best of ~0.5 versus random's ~7.6.

Most users want the hyperopt-rs facade, which re-exports these. See the repository for the full guide.

License

MIT © mi7plus