dgl.to_networkxΒΆ

dgl.to_networkx(g, node_attrs=None, edge_attrs=None, ntype_attr='ntype', etype_attr='etype', eid_attr='id')[source]ΒΆ

Convert a graph to a NetworkX graph and return.

The resulting NetworkX graph also contains the node/edge features of the input graph. Additionally, DGL saves the edge IDs as the 'id' edge attribute in the returned NetworkX graph.

Parameters
  • g (DGLGraph) – A homogeneous or heterogeneous graph.

  • node_attrs (iterable of str, optional) – The node attributes to copy from g.ndata. (Default: None)

  • edge_attrs (iterable of str, optional) – The edge attributes to copy from g.edata. (Default: None)

  • ntype_attr (str, optional) – The name of the node attribute to store the node types in the NetworkX object. (Default: β€œntype”)

  • etype_attr (str, optional) – The name of the edge attribute to store the edge canonical types in the NetworkX object. (Default: β€œetype”)

  • eid_attr (str, optional) – The name of the edge attribute to store the original edge ID in the NetworkX object. (Default: β€œid”)

Returns

The converted NetworkX graph.

Return type

networkx.DiGraph

Notes

The function only supports CPU graph input.

Examples

The following examples use the PyTorch backend.

>>> import dgl
>>> import torch

With a homogeneous graph:

>>> g = dgl.graph((torch.tensor([1, 2]), torch.tensor([1, 3])))
>>> g.ndata['h'] = torch.zeros(4, 1)
>>> g.edata['h1'] = torch.ones(2, 1)
>>> g.edata['h2'] = torch.zeros(2, 2)
>>> nx_g = dgl.to_networkx(g, node_attrs=['h'], edge_attrs=['h1', 'h2'])
>>> nx_g.nodes(data=True)
NodeDataView({
    0: {'h': tensor([0.])},
    1: {'h': tensor([0.])},
    2: {'h': tensor([0.])},
    3: {'h': tensor([0.])}
})
>>> nx_g.edges(data=True)
OutMultiEdgeDataView([
    (1, 1, {'id': 0, 'h1': tensor([1.]), 'h2': tensor([0., 0.])}),
    (2, 3, {'id': 1, 'h1': tensor([1.]), 'h2': tensor([0., 0.])})
])

With a heterogeneous graph:

>>> g = dgl.heterograph({
...     ('user', 'follows', 'user'): (torch.tensor([0, 1]), torch.tensor([1, 2])),
...     ('user', 'follows', 'topic'): (torch.tensor([1, 1]), torch.tensor([1, 2])),
...     ('user', 'plays', 'game'): (torch.tensor([0, 3]), torch.tensor([3, 4]))
... })
>>> g.ndata['n'] = {
...     'game': torch.zeros(5, 1),
...     'user': torch.ones(4, 1)
... }
>>> g.edata['e'] = {
...     ('user', 'follows', 'user'): torch.zeros(2, 1),
...     'plays': torch.ones(2, 1)
... }
>>> nx_g = dgl.to_networkx(g, node_attrs=['n'], edge_attrs=['e'])
>>> nx_g.nodes(data=True)
NodeDataView({
    0: {'ntype': 'game', 'n': tensor([0.])},
    1: {'ntype': 'game', 'n': tensor([0.])},
    2: {'ntype': 'game', 'n': tensor([0.])},
    3: {'ntype': 'game', 'n': tensor([0.])},
    4: {'ntype': 'game', 'n': tensor([0.])},
    5: {'ntype': 'topic'},
    6: {'ntype': 'topic'},
    7: {'ntype': 'topic'},
    8: {'ntype': 'user', 'n': tensor([1.])},
    9: {'ntype': 'user', 'n': tensor([1.])},
    10: {'ntype': 'user', 'n': tensor([1.])},
    11: {'ntype': 'user', 'n': tensor([1.])}
})
>>> nx_g.edges(data=True)
OutMultiEdgeDataView([
    (8, 9, {'id': 2, 'etype': ('user', 'follows', 'user'), 'e': tensor([0.])}),
    (8, 3, {'id': 4, 'etype': ('user', 'plays', 'game'), 'e': tensor([1.])}),
    (9, 6, {'id': 0, 'etype': ('user', 'follows', 'topic')}),
    (9, 7, {'id': 1, 'etype': ('user', 'follows', 'topic')}),
    (9, 10, {'id': 3, 'etype': ('user', 'follows', 'user'), 'e': tensor([0.])}),
    (11, 4, {'id': 5, 'etype': ('user', 'plays', 'game'), 'e': tensor([1.])})
])