gigl.distributed.sampler

Attributes

EDGE_PACKED_FEATURES_METADATA_KEY
NEGATIVE_LABEL_METADATA_KEY
NODE_PACKED_FEATURES_METADATA_KEY
POSITIVE_LABEL_METADATA_KEY

Classes

ABLPNodeSamplerInput

Sampler input specific for ABLP use case. Contains additional information about positive labels, negative labels, and the corresponding

Module Contents

class gigl.distributed.sampler.ABLPNodeSamplerInput(node, input_type, positive_label_by_edge_types, negative_label_by_edge_types)

Bases: graphlearn_torch.sampler.NodeSamplerInput

Sampler input specific for ABLP use case. Contains additional information about positive labels, negative labels, and the corresponding supervision node type

Parameters:

• node (torch.Tensor) – Anchor nodes to fanout from

• input_type (Optional[Union[str, NodeType]]) – Node type of the anchor nodes

• positive_label_by_edge_types (dict[EdgeType, torch.Tensor]) – Positive label nodes to fanout from

• negative_label_by_edge_types (dict[EdgeType, torch.Tensor]) – Negative label nodes to fanout from

share_memory()

Return type:

ABLPNodeSamplerInput

property negative_label_by_edge_types: dict[gigl.src.common.types.graph_data.EdgeType, jaxtyping.Int64[torch.Tensor, anchors _negative_labels_per_anchor]]

Return type:

dict[gigl.src.common.types.graph_data.EdgeType, jaxtyping.Int64[torch.Tensor, anchors _negative_labels_per_anchor]]

property positive_label_by_edge_types: dict[gigl.src.common.types.graph_data.EdgeType, jaxtyping.Int64[torch.Tensor, anchors _positive_labels_per_anchor]]

Return type:

dict[gigl.src.common.types.graph_data.EdgeType, jaxtyping.Int64[torch.Tensor, anchors _positive_labels_per_anchor]]

gigl.distributed.sampler.EDGE_PACKED_FEATURES_METADATA_KEY: Final[str] = 'edge_packed_features'

gigl.distributed.sampler.NEGATIVE_LABEL_METADATA_KEY: Final[str] = 'gigl_negative_labels.'

gigl.distributed.sampler.NODE_PACKED_FEATURES_METADATA_KEY: Final[str] = 'node_packed_features'

gigl.distributed.sampler.POSITIVE_LABEL_METADATA_KEY: Final[str] = 'gigl_positive_labels.'