InSubgraphSampler

class dgl.graphbolt.InSubgraphSampler(datapipe, graph)

Bases: SubgraphSampler

Sample the subgraph induced on the inbound edges of the given nodes.

Functional name: sample_in_subgraph.

In-subgraph sampler is responsible for sampling a subgraph from given data, returning an induced subgraph along with compacted information.

Parameters:

• datapipe (DataPipe) – The datapipe.

• graph (FusedCSCSamplingGraph) – The graph on which to perform in_subgraph sampling.

Examples

>>> import dgl.graphbolt as gb
>>> import torch
>>> indptr = torch.LongTensor([0, 3, 5, 7, 9, 12, 14])
>>> indices = torch.LongTensor([0, 1, 4, 2, 3, 0, 5, 1, 2, 0, 3, 5, 1, 4])
>>> graph = gb.fused_csc_sampling_graph(indptr, indices)
>>> item_set = gb.ItemSet(len(indptr) - 1, names="seeds")
>>> item_sampler = gb.ItemSampler(item_set, batch_size=2)
>>> insubgraph_sampler = gb.InSubgraphSampler(item_sampler, graph)
>>> for _, data in enumerate(insubgraph_sampler):
...     print(data.sampled_subgraphs[0].sampled_csc)
...     print(data.sampled_subgraphs[0].original_row_node_ids)
...     print(data.sampled_subgraphs[0].original_column_node_ids)
CSCFormatBase(indptr=tensor([0, 3, 5]),
            indices=tensor([0, 1, 2, 3, 4]),
)
tensor([0, 1, 4, 2, 3])
tensor([0, 1])
CSCFormatBase(indptr=tensor([0, 2, 4]),
            indices=tensor([2, 3, 4, 0]),
)
tensor([2, 3, 0, 5, 1])
tensor([2, 3])
CSCFormatBase(indptr=tensor([0, 3, 5]),
            indices=tensor([2, 3, 1, 4, 0]),
)
tensor([4, 5, 0, 3, 1])
tensor([4, 5])

sample_subgraphs(seeds, seeds_timestamp)

Sample subgraphs from the given seeds, possibly with temporal constraints.

Any subclass of SubgraphSampler should implement this method.

Parameters:

• seeds (Union[torch.Tensor, Dict[str, torch.Tensor]]) – The seed nodes.

• seeds_timestamp (Union[torch.Tensor, Dict[str, torch.Tensor]]) – The timestamps of the seed nodes. If given, the sampled subgraphs should not contain any nodes or edges that are newer than the timestamps of the seed nodes. Default: None.

Returns:

• Union[torch.Tensor, Dict[str, torch.Tensor]] – The input nodes.

• List[SampledSubgraph] – The sampled subgraphs.

Examples

>>> @functional_datapipe("my_sample_subgraph")
>>> class MySubgraphSampler(SubgraphSampler):
>>>     def __init__(self, datapipe, graph, fanouts):
>>>         super().__init__(datapipe)
>>>         self.graph = graph
>>>         self.fanouts = fanouts
>>>     def sample_subgraphs(self, seeds):
>>>         # Sample subgraphs from the given seeds.
>>>         subgraphs = []
>>>         subgraphs_nodes = []
>>>         for fanout in reversed(self.fanouts):
>>>             subgraph = self.graph.sample_neighbors(seeds, fanout)
>>>             subgraphs.insert(0, subgraph)
>>>             subgraphs_nodes.append(subgraph.nodes)
>>>             seeds = subgraph.nodes
>>>         subgraphs_nodes = torch.unique(torch.cat(subgraphs_nodes))
>>>         return subgraphs_nodes, subgraphs