nat.profiler.parameter_optimization.parameter_selection#

Functions#

`_to_minimisation_matrix`(→ numpy.ndarray)	Return array (n_trials × n_objectives) where all objectives are ‘smaller-is-better’.
`pick_trial`(→ optuna.trial.FrozenTrial)	Collapse Optuna’s Pareto front (`study.best_trials`) to a single “best compromise”.

_to_minimisation_matrix( trials: collections.abc.Sequence[optuna.trial.FrozenTrial], directions: collections.abc.Sequence[optuna.study.StudyDirection], ) → numpy.ndarray#: Return array (n_trials × n_objectives) where all objectives are ‘smaller-is-better’.

pick_trial( study: optuna.study.Study, mode: str = 'harmonic', *, weights: collections.abc.Sequence[float] | None = None, ref_point: collections.abc.Sequence[float] | None = None, eps: float = 1e-12, ) → optuna.trial.FrozenTrial#: Collapse Optuna’s Pareto front (study.best_trials) to a single “best compromise”.

Parameters#

study : completed multi-objective Optuna study mode : {“harmonic”, “sum”, “chebyshev”, “hypervolume”} weights : per-objective weights (used only for “sum”) ref_point : reference point for hyper-volume (defaults to ones after normalisation) eps : tiny value to avoid division by zero

Returns#

optuna.trial.FrozenTrial