class dgl.dataloading.ShaDowKHopSampler(fanouts, replace=False, prob=None, prefetch_node_feats=None, prefetch_edge_feats=None, output_device=None)[source]

Bases: dgl.dataloading.base.Sampler

K-hop subgraph sampler from Deep Graph Neural Networks with Shallow Subgraph Samplers.

It performs node-wise neighbor sampling and returns the subgraph induced by all the sampled nodes. The seed nodes from which the neighbors are sampled will appear the first in the induced nodes of the subgraph.

  • fanouts (list[int] or list[dict[etype, int]]) –

    List of neighbors to sample per edge type for each GNN layer, with the i-th element being the fanout for the i-th GNN layer.

    If only a single integer is provided, DGL assumes that every edge type will have the same fanout.

    If -1 is provided for one edge type on one layer, then all inbound edges of that edge type will be included.

  • replace (bool, default True) – Whether to sample with replacement

  • prob (str, optional) – If given, the probability of each neighbor being sampled is proportional to the edge feature value with the given name in g.edata. The feature must be a scalar on each edge.


Node classification

To train a 3-layer GNN for node classification on a set of nodes train_nid on a homogeneous graph where each node takes messages from 5, 10, 15 neighbors for the first, second, and third layer respectively (assuming the backend is PyTorch):

>>> g =[0]
>>> sampler = dgl.dataloading.ShaDowKHopSampler([5, 10, 15])
>>> dataloader = dgl.dataloading.DataLoader(
...     g, torch.arange(g.num_nodes()), sampler,
...     batch_size=5, shuffle=True, drop_last=False, num_workers=4)
>>> for input_nodes, output_nodes, subgraph in dataloader:
...     print(subgraph)
...     assert torch.equal(input_nodes, subgraph.ndata[dgl.NID])
...     assert torch.equal(input_nodes[:output_nodes.shape[0]], output_nodes)
...     break
Graph(num_nodes=529, num_edges=3796,
      ndata_schemes={'label': Scheme(shape=(), dtype=torch.int64),
                     'feat': Scheme(shape=(8710,), dtype=torch.float32),
                     '_ID': Scheme(shape=(), dtype=torch.int64)}
      edata_schemes={'_ID': Scheme(shape=(), dtype=torch.int64)})

If training on a heterogeneous graph and you want different number of neighbors for each edge type, one should instead provide a list of dicts. Each dict would specify the number of neighbors to pick per edge type.

>>> sampler = dgl.dataloading.ShaDowKHopSampler([
...     {('user', 'follows', 'user'): 5,
...      ('user', 'plays', 'game'): 4,
...      ('game', 'played-by', 'user'): 3}] * 3)

If you would like non-uniform neighbor sampling:

>>> g.edata['p'] = torch.rand(g.num_edges())   # any non-negative 1D vector works
>>> sampler = dgl.dataloading.ShaDowKHopSampler([5, 10, 15], prob='p')