dgl.graphbolt.fused_csc_sampling_graph

dgl.graphbolt.fused_csc_sampling_graph(csc_indptr: Tensor, indices: Tensor, node_type_offset: tensor | None = None, type_per_edge: tensor | None = None, node_type_to_id: Dict[str, int] | None = None, edge_type_to_id: Dict[str, int] | None = None, node_attributes: Dict[str, tensor] | None = None, edge_attributes: Dict[str, tensor] | None = None) β†’ FusedCSCSamplingGraph[source]

Create a FusedCSCSamplingGraph object from a CSC representation.

Parameters:

  • csc_indptr (torch.Tensor) – Pointer to the start of each row in the indices. An integer tensor with shape (total_num_nodes+1,).

  • indices (torch.Tensor) – Column indices of the non-zero elements in the CSC graph. An integer tensor with shape (total_num_edges,).

  • node_type_offset (Optional[torch.tensor], optional) – Offset of node types in the graph, by default None.

  • type_per_edge (Optional[torch.tensor], optional) – Type ids of each edge in the graph, by default None. If provided, it is required that the edge types in each vertex neighborhood are in sorted order. To be more precise, For each i in [0, csc_indptr.size(0) - 1), type_per_edge[indptr[i]: indptr[i + 1]] is expected to be monotonically nondecreasing.

  • node_type_to_id (Optional[Dict[str, int]], optional) – Map node types to ids, by default None.

  • edge_type_to_id (Optional[Dict[str, int]], optional) – Map edge types to ids, by default None.

  • node_attributes (Optional[Dict[str, torch.tensor]], optional) – Node attributes of the graph, by default None.

  • edge_attributes (Optional[Dict[str, torch.tensor]], optional) – Edge attributes of the graph, by default None.

Returns:

The created FusedCSCSamplingGraph object.

Return type:

FusedCSCSamplingGraph

Examples

>>> ntypes = {'n1': 0, 'n2': 1, 'n3': 2}
>>> etypes = {'n1:e1:n2': 0, 'n1:e2:n3': 1}
>>> csc_indptr = torch.tensor([0, 2, 5, 7, 8])
>>> indices = torch.tensor([1, 3, 0, 1, 2, 0, 3, 2])
>>> node_type_offset = torch.tensor([0, 1, 2, 4])
>>> type_per_edge = torch.tensor([0, 1, 0, 1, 1, 0, 0, 0])
>>> graph = graphbolt.fused_csc_sampling_graph(csc_indptr, indices,
...         node_type_offset=node_type_offset,
...         type_per_edge=type_per_edge,
...         node_type_to_id=ntypes, edge_type_to_id=etypes,
...         node_attributes=None, edge_attributes=None,)
>>> print(graph)
FusedCSCSamplingGraph(csc_indptr=tensor([0, 2, 5, 7, 8]),
                      indices=tensor([1, 3, 0, 1, 2, 0, 3, 2]),
                      total_num_nodes=4, num_edges={'n1:e1:n2': 5, 'n1:e2:n3': 3},
                      node_type_offset=tensor([0, 1, 2, 4]),
                      type_per_edge=tensor([0, 1, 0, 1, 1, 0, 0, 0]),
                      node_type_to_id={'n1': 0, 'n2': 1, 'n3': 2},
                      edge_type_to_id={'n1:e1:n2': 0, 'n1:e2:n3': 1},)