baybe.acquisition.utils.make_partitioning¶
- baybe.acquisition.utils.make_partitioning(predictions: Tensor, ref_point: Tensor, alpha: float | None)[source]¶
Create a
BoxDecompositionobject for the given predictions and reference point.For details on the arguments, see
NondominatedPartitioning.- Parameters:
predictions (
Tensor) – The predictions tensor of shape (n_samples, n_outputs).ref_point (
Tensor) – The reference point tensor of shape (n_outputs,).alpha (
float|None) – Optional threshold parameter controlling the partitioning generation. Hypercells with a volume fraction (relative to the total Pareto set hypervolume) less than the specified value will be dropped, leading to more approximation but faster computation.
- Raises:
ValueError – If the predictions or reference point do not have the expected shapes.
- Return type:
- Returns:
A partitioning object for hypervolume acquisition functions.