incerto.data.SubclassOOD

incerto.data.SubclassOOD#

class incerto.data.SubclassOOD(dataset_class, root='./data', id_classes=None, ood_classes=None, transform=None, **dataset_kwargs)[source]#

Bases: OODBenchmark

Create OOD benchmark by holding out specific classes.

Example: Train on classes 0-7 of CIFAR-10, test with classes 8-9 as OOD.

Parameters:
__init__(dataset_class, root='./data', id_classes=None, ood_classes=None, transform=None, **dataset_kwargs)[source]#

Initialize subclass OOD benchmark.

Parameters:

  • dataset_class – Dataset class (e.g., datasets.CIFAR10)

  • root (str | Path) – Data root directory

  • id_classes (Optional[List[int]]) – List of class indices to use as ID

  • ood_classes (Optional[List[int]]) – List of class indices to use as OOD

  • transform (Optional[Compose]) – Transform to apply (defaults to ToTensor())

  • **dataset_kwargs – Additional arguments for dataset

Methods

__init__(dataset_class[, root, id_classes, ...])

Initialize subclass OOD benchmark.

get_datasets()

Get ID and OOD subsets.
__init__(dataset_class, root='./data', id_classes=None, ood_classes=None, transform=None, **dataset_kwargs)[source]#

Initialize subclass OOD benchmark.

Parameters:

  • dataset_class – Dataset class (e.g., datasets.CIFAR10)

  • root (str | Path) – Data root directory

  • id_classes (Optional[List[int]]) – List of class indices to use as ID

  • ood_classes (Optional[List[int]]) – List of class indices to use as OOD

  • transform (Optional[Compose]) – Transform to apply (defaults to ToTensor())

  • **dataset_kwargs – Additional arguments for dataset

get_datasets()[source]#

Get ID and OOD subsets.

Return type:

Tuple[Dataset, Dataset]

Contents