Optuna

Optuna is a popular black-box optimizer. This page shows how to use Optuna with Hypster.

Installation

uv add 'hypster[optuna]'
# or
uv add optuna

Usage

See Advanced → Performing Hyperparameter Optimization for a full example. Summary:

1. Define your Hypster config (hpo_spec is optional; defaults are sensible).

2. In your objective, call hypster.hpo.optuna.suggest_values(trial, config=...) to get a values dict.

3. Instantiate your config with those values and evaluate.

import optuna
from hypster import HP, instantiate
from hypster.hpo.optuna import suggest_values


def model_cfg(hp: HP):
    kind = hp.select(["rf", "lr"], name="kind")  # hpo_spec omitted: linear/unordered defaults
    # ... conditional params ...
    return {"model": ("rf", 100, 10.0)}


def objective(trial: optuna.Trial) -> float:
    values = suggest_values(trial, config=model_cfg)
    cfg = instantiate(model_cfg, values=values)
    # ... train/evaluate with cfg["model"] ...
    return 0.0

Notes

• Supports scalar parameters: hp.int, hp.float, hp.select.

• Multi-* parameters are not expanded in the Optuna adapter (use booleans or explicit scalars if needed).

• It’s easy to add other backends (e.g., Ray Tune). Please open an issue or PR if you’re interested.