dgl.DGLGraph.srcnodes¶

property DGLGraph.srcnodes¶

Return a node view for source nodes

If the graph is a uni-bipartite graph (see is_unibipartite() for reference), this is nodes() restricted to source node types. Otherwise, it is an alias for nodes().

One can use it for:

Getting the node IDs for a single node type.
Setting/getting features for all nodes of a single node type.

Examples

The following example uses PyTorch backend.

>>> import dgl
>>> import torch

Create a uni-bipartite graph.

>>> g = dgl.heterograph({
...     ('user', 'plays', 'game'): (torch.tensor([0]), torch.tensor([1])),
...     ('developer', 'develops', 'game'): (torch.tensor([1]), torch.tensor([2]))
... })

Get the node IDs for source node types.

>>> g.srcnodes('user')
tensor([0])
>>> g.srcnodes('developer')
tensor([0, 1])

Set/get features for source node types.

>>> g.srcnodes['user'].data['h'] = torch.ones(1, 1)
>>> g.srcnodes['user'].data['h']
tensor([[1.]])

Create a graph that is not uni-bipartite.

>>> g = dgl.heterograph({
...     ('user', 'follows', 'user'): (torch.tensor([0]), torch.tensor([1])),
...     ('developer', 'develops', 'game'): (torch.tensor([1]), torch.tensor([2]))
... })

dgl.DGLGraph.srcnodes() falls back to dgl.DGLGraph.nodes() and one can get the node IDs for both source and destination node types.

>>> g.srcnodes('game')
tensor([0, 1, 2])

One can also set/get features for destination node types in this case.

>>> g.srcnodes['game'].data['h'] = torch.ones(3, 1)
>>> g.srcnodes['game'].data['h']
tensor([[1.],
        [1.],
        [1.]])