luz.datasets module

class ConcatDataset(datasets)

Bases: luz.datasets.BaseDataset, torch.utils.data.dataset.ConcatDataset

cumulative_sizes: List[int]

datasets: List[torch.utils.data.dataset.Dataset[torch.utils.data.dataset.T_co]]

class Data(x, y=None, **kwargs)

Bases: object

Data containing one or more tensors.

Parameters

• x (Tensor) – Primary data tensor.

• y (Optional[Tensor]) – Label tensor, by default None.

• **kwargs – Additional data tensors.

property keys: None

Return type

None

to(device)

Migrate tensors to device. Modifies Data object in-place.

Parameters

device (Union[str, device]) – Target device.

Returns

Migrated Data.

Return type

luz.Data

class Dataset(data)

Bases: torch.utils.data.dataset.Dataset, luz.datasets.BaseDataset

Object containing points from a domain, possibly with labels.

Parameters

data (Iterable[Data]) – Iterable of Data objects comprising the dataset.

class FolderDataset(root)

Bases: luz.datasets.BaseDataset, torch.utils.data.dataset.Dataset

Dataset which reads data from disk.

Parameters

root (str) – Root directory containing data stored in .pt files.

class Subset(dataset, indices)

Bases: luz.datasets.BaseDataset, torch.utils.data.dataset.Subset

dataset: torch.utils.data.dataset.Dataset[torch.utils.data.dataset.T_co]

indices: Sequence[int]

class TensorDataset(**tensors)

Bases: torch.utils.data.dataset.Dataset, luz.datasets.BaseDataset

class UnpackDataset(keys, dataset)

Bases: luz.datasets.BaseDataset, torch.utils.data.dataset.Dataset

class WrapperDataset(**datasets)

Bases: luz.datasets.BaseDataset, torch.utils.data.dataset.Dataset

default_collate(batch)

Collate multiple Data objects.

Parameters

batch (Iterable[luz.Data]) – Data objects to be collated

Returns

Collated Data object

Return type

luz.Data

graph_collate(batch)

Collate multiple Data objects containing graph data.

Parameters

batch (Iterable[luz.Data]) – Data objects to be collated

Returns

Collated Data object

Return type

luz.Data