intelligence.zenith_tune.strategies.optuna_sampler

Optuna sampler-based strategy for hyperparameter optimization.

OptunaStrategy Objects

@StrategyRegistry.register("optuna")
class OptunaStrategy(TuningStrategy)

Optuna sampler-based strategy.

Uses an Optuna study to drive hyperparameter sampling via Optuna's sampler API (default: TPESampler). By default the study is in-memory; pass a storage URL to persist it and resume across runs.

Runs indefinitely until eval_fn raises TrialExhausted.

Example:

strategy = OptunaStrategy() # defaults to TPESampler, in-memory strategy.optimize(eval_fn, search_space, Direction.MINIMIZE)

With a different sampler:

from optuna.samplers import CmaEsSampler strategy = OptunaStrategy(sampler=CmaEsSampler())

With persistent storage (resume across runs):

strategy = OptunaStrategy(storage="sqlite:///study.db", study_name="my_study")

__init__

def __init__(sampler: Optional[BaseSampler] = None,
             storage: Optional[str] = None,
             study_name: Optional[str] = None) -> None

Initialize the Optuna strategy.

Arguments:

• sampler - Optuna sampler instance. Defaults to TPESampler.
• storage - Optuna storage URL (e.g. "sqlite:///study.db"). If provided, the study is persisted and resumed across runs.
• study_name - Name of the Optuna study. Required when storage is provided and multiple studies exist in the same storage.