Hyperparameter optimizer

fyt.core.hyperparameter_optimizer

Hyperparameter optimization using Optuna.

Extracted from Trainer to follow the Single Responsibility Principle.

HyperparameterOptimizer

Owns the Optuna study lifecycle for hyperparameter optimization.

Parameters:

Name	Type	Description	Default
optuna_config	OptunaConfig	Optuna configuration.	required
cross_validation_config	CrossValidationConfig	Cross-validation configuration.	required
model_type	str	Type of model to optimize.	required
model_params	dict[str, Any]	Base model parameters.	required
experiment_logger	BaseExperimentLogger | None	Experiment logger for logging trial results.	None
random_state	int	Random state for reproducibility.	42
validation_size	float	Validation fraction for the non-CV evaluation mode.	0.2
task	TaskType	Learning task type (drives CV strategy and stratification).	CLASSIFICATION
pipeline_builder	Callable[[], list[tuple[str, Any]]] | None	Callable returning fresh, unfitted (name, transformer) preprocessing steps. When provided and optuna_config.leakage_safe_cv is on, every trial is evaluated with a full sklearn Pipeline so preprocessing and feature selection are re-fit inside each CV fold (no fold leakage). When off, the steps are fit once on the whole training set (faster, optimistically biased).	None

best_params property

The best hyperparameters found by Optuna.

direction property

Optimization direction as a string ('maximize' or 'minimize').

metric_key property

Normalized name of the optimization metric (trial results column).

random_state property writable

The random state.

results_df property

The DataFrame of all trial results.

optimize(data_split)

Run Optuna hyperparameter optimization.

Parameters:

Name	Type	Description	Default
data_split	DataSplit	Data split containing training data for optimization.	required

Returns:

Type	Description
dict[str, Any]	Best hyperparameters found. Empty dict if optimization is disabled.