torch_geometric.data.LargeGraphIndexer

class LargeGraphIndexer(nodes: Iterable[Hashable], edges: Iterable[Tuple[Hashable, Hashable, Hashable]], node_attr: Optional[Dict[str, List[Any]]] = None, edge_attr: Optional[Dict[str, List[Any]]] = None)

Bases: object

For a dataset that consists of mulitiple subgraphs that are assumed to be part of a much larger graph, collate the values into a large graph store to save resources.

classmethod from_triplets(triplets: Iterable[Tuple[Hashable, Hashable, Hashable]], pre_transform: Optional[Callable[[Tuple[Hashable, Hashable, Hashable]], Tuple[Hashable, Hashable, Hashable]]] = None) → LargeGraphIndexer

Generate a new index from a series of triplets that represent edge relations between nodes. Formatted like (source_node, edge, dest_node).

Parameters:

• triplets (KnowledgeGraphLike) – Series of triplets representing knowledge graph relations.

• pre_transform (Optional[Callable[[TripletLike], TripletLike]]) – Optional preprocessing function to apply to triplets. Defaults to None.

Returns:

Index of unique nodes and edges.

Return type:

LargeGraphIndexer

classmethod collate(graphs: Iterable[LargeGraphIndexer]) → LargeGraphIndexer

Combines a series of large graph indexes into a single large graph index.

Parameters:

graphs (Iterable["LargeGraphIndexer"]) – Indices to be combined.

Returns:

Singular unique index for all nodes and edges

in input indices.

Return type:

LargeGraphIndexer

get_unique_node_features(feature_name: str = 'pid') → List[Hashable]

Get all the unique values for a specific node attribute.

Parameters:

feature_name (str, optional) – Name of feature to get. Defaults to NODE_PID.

Returns:

List of unique values for the specified feature.

Return type:

List[Hashable]

add_node_feature(new_feature_name: str, new_feature_vals: Union[Sequence[Any], Tensor], map_from_feature: str = 'pid') → None

Adds a new feature that corresponds to each unique node in

the graph.

Parameters:

• new_feature_name (str) – Name to call the new feature.

• new_feature_vals (FeatureValueType) – Values to map for that new feature.

• map_from_feature (str, optional) – Key of feature to map from. Size must match the number of feature values. Defaults to NODE_PID.

Return type:

None

get_node_features(feature_name: str = 'pid', pids: Optional[Iterable[Hashable]] = None) → List[Any]

Get node feature values for a given set of unique node ids.

Returned values are not necessarily unique.

Parameters:

• feature_name (str, optional) – Name of feature to fetch. Defaults to NODE_PID.

• pids (Optional[Iterable[Hashable]], optional) – Node ids to fetch for. Defaults to None, which fetches all nodes.

Returns:

Node features corresponding to the specified ids.

Return type:

List[Any]

get_node_features_iter(feature_name: str = 'pid', pids: Optional[Iterable[Hashable]] = None, index_only: bool = False) → Iterator[Any]

Iterator version of get_node_features. If index_only is True, yields indices instead of values.

Return type:

Iterator[Any]

get_unique_edge_features(feature_name: str = 'e_pid') → List[Hashable]

Get all the unique values for a specific edge attribute.

Parameters:

feature_name (str, optional) – Name of feature to get. Defaults to EDGE_PID.

Returns:

List of unique values for the specified feature.

Return type:

List[Hashable]

add_edge_feature(new_feature_name: str, new_feature_vals: Union[Sequence[Any], Tensor], map_from_feature: str = 'e_pid') → None

Adds a new feature that corresponds to each unique edge in the graph.

Parameters:

• new_feature_name (str) – Name to call the new feature.

• new_feature_vals (FeatureValueType) – Values to map for that new feature.

• map_from_feature (str, optional) – Key of feature to map from. Size must match the number of feature values. Defaults to EDGE_PID.

Return type:

None

get_edge_features(feature_name: str = 'e_pid', pids: Optional[Iterable[Hashable]] = None) → List[Any]

Get edge feature values for a given set of unique edge ids.

Returned values are not necessarily unique.

Parameters:

• feature_name (str, optional) – Name of feature to fetch. Defaults to EDGE_PID.

• pids (Optional[Iterable[Hashable]], optional) – Edge ids to fetch for. Defaults to None, which fetches all edges.

Returns:

Node features corresponding to the specified ids.

Return type:

List[Any]

get_edge_features_iter(feature_name: str = 'e_pid', pids: Optional[Iterable[Tuple[Hashable, Hashable, Hashable]]] = None, index_only: bool = False) → Iterator[Any]

Iterator version of get_edge_features. If index_only is True, yields indices instead of values.

Return type:

Iterator[Any]

to_data(node_feature_name: str, edge_feature_name: Optional[str] = None) → Data

Return a Data object containing all the specified node and

edge features and the graph.

Parameters:

• node_feature_name (str) – Feature to use for nodes

• edge_feature_name (Optional[str], optional) – Feature to use for edges. Defaults to None.

Returns:

Data object containing the specified node and

edge features and the graph.

Return type:

Data