cugraph_pyg.loader.LinkLoader#

class cugraph_pyg.loader.LinkLoader(
data: Data | HeteroData | Tuple[FeatureStore, GraphStore],
link_sampler: BaseSampler,
edge_label_index: Tensor | None | Tuple[str, str, str] | Tuple[Tuple[str, str, str], Tensor | None] = None,
edge_label: Tensor | None = None,
edge_label_time: Tensor | None = None,
neg_sampling: NegativeSampling | None = None,
neg_sampling_ratio: int | float | None = None,
transform: Callable | None = None,
transform_sampler_output: Callable | None = None,
filter_per_worker: bool | None = None,
custom_cls: HeteroData | None = None,
input_id: Tensor | None = None,
batch_size: int = 1,
shuffle: bool = False,
drop_last: bool = False,
**kwargs,
)[source]#

Duck-typed version of torch_geometric.loader.LinkLoader. Loads samples from batches of input nodes using a ~cugraph_pyg.sampler.BaseSampler.sample_from_edges function.

__init__(
data: Data | HeteroData | Tuple[FeatureStore, GraphStore],
link_sampler: BaseSampler,
edge_label_index: Tensor | None | Tuple[str, str, str] | Tuple[Tuple[str, str, str], Tensor | None] = None,
edge_label: Tensor | None = None,
edge_label_time: Tensor | None = None,
neg_sampling: NegativeSampling | None = None,
neg_sampling_ratio: int | float | None = None,
transform: Callable | None = None,
transform_sampler_output: Callable | None = None,
filter_per_worker: bool | None = None,
custom_cls: HeteroData | None = None,
input_id: Tensor | None = None,
batch_size: int = 1,
shuffle: bool = False,
drop_last: bool = False,
**kwargs,
)[source]#
Parameters:
data: Data, HeteroData, or Tuple[FeatureStore, GraphStore]

See torch_geometric.loader.NodeLoader.

link_sampler: BaseSampler

See torch_geometric.loader.LinkLoader.

edge_label_index: InputEdges

See torch_geometric.loader.LinkLoader.

edge_label: OptTensor

See torch_geometric.loader.LinkLoader.

edge_label_time: OptTensor

See torch_geometric.loader.LinkLoader.

neg_sampling: Optional[NegativeSampling]

Type of negative sampling to perform, if desired. See torch_geometric.loader.LinkLoader.

neg_sampling_ratio: Optional[Union[int, float]]

Negative sampling ratio. Affects how many negative samples are generated. See torch_geometric.loader.LinkLoader.

transform: Callable (optional, default=None)

This argument currently has no effect.

transform_sampler_output: Callable (optional, default=None)

This argument currently has no effect.

filter_per_worker: bool (optional, default=False)

This argument currently has no effect.

custom_cls: HeteroData

This argument currently has no effect. This loader will always return a Data or HeteroData object.

input_id: OptTensor

See torch_geometric.loader.LinkLoader.

Methods

__init__(data, link_sampler[, ...])

On this page