pyg-team/pytorch_geometric

Convert heterogeneous NetworkX graph to HeteroData without normalizing features across node types

Open

#7,340 opened on May 10, 2023

View on GitHub
 (7 comments) (0 reactions) (0 assignees)Python (3,514 forks)batch import
0 - Priority P0featurehelp wanted

Repository metrics

Stars
 (19,985 stars)
PR merge metrics
 (Avg merge 35d 1h) (14 merged PRs in 30d)

Description

🚀 The feature, motivation and pitch

I am working on converting a heterogeneous NetworkX graph to HeteroData. Here is an example that I would like to see working:

nodes_types = torch.tensor([1, 2, 3])
nx_graph = nx.MultiGraph()
nx_graph.add_node(0, type=1, foo=4.97, bar=torch.tensor([9, 8]))
nx_graph.add_node(1, type=2, foo=4.97)
nx_graph.add_node(2, type=3)
pyg.utils.from_networkx(nx_graph, group_node_attrs=['type', 'foo', 'bar'])

which currently throws raise ValueError('Not all nodes contain the same attributes').

I know that my graph is heterogeneous, but to convert from NetworkX to HeteroData, I need to manually normalize all of the node attributes. I would like to not have to write my own conversion layer.

Alternatives

N/A

Additional context

N/A

Contributor guide

Research direction
Investigate the `from networkx` function in `torch geometric.utils.convert`. Modify the `group node attrs` logic to handle heterogeneous node types by not requiring all nodes to have the same attributes. Instead, group attributes per node type and assign them to the corresponding node type in HeteroData.
Tech stack
pythonpytorch
Domain
machine learningdata
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
Half day
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
PythonPyTorchNetworkX
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
65