How to convert pyTorch-Geometric custom datasets into DeepSnap datasets?

[ajith01]

Hello,

I am trying to use a custom dataset for link prediction,
What i tried was

pyg_dataset = My_Own_Dataset()
graphs1 = GraphDataset.pyg_to_graphs(pyg_dataset) #error here

but i am getting an error: TypeError: zip argument #2 must support iteration

The full error is

/usr/local/lib/python3.7/dist-packages/deepsnap/dataset.py in pyg_to_graphs(dataset, verbose, fixed_split, tensor_backend, netlib)
   1280                     netlib=netlib
   1281                 )
-> 1282                 for data in dataset
   1283             ]
   1284 

/usr/local/lib/python3.7/dist-packages/deepsnap/dataset.py in <listcomp>(.0)
   1280                     netlib=netlib
   1281                 )
-> 1282                 for data in dataset
   1283             ]
   1284 

/usr/local/lib/python3.7/dist-packages/deepsnap/graph.py in pyg_to_graph(data, verbose, fixed_split, tensor_backend, netlib)
   2025             if Graph._is_node_attribute(key):
   2026                 if not tensor_backend:
-> 2027                     Graph.add_node_attr(G, key, value)
   2028                 else:
   2029                     attributes[key] = value

/usr/local/lib/python3.7/dist-packages/deepsnap/graph.py in add_node_attr(G, attr_name, node_attr)
   1909         # TODO: Better method here?
   1910         node_list = list(G.nodes)
-> 1911         attr_dict = dict(zip(node_list, node_attr))
   1912         deepsnap._netlib.set_node_attributes(G, attr_dict, name=attr_name)
   1913 

TypeError: zip argument #2 must support iteration

I have also tried to do this using networkx and converting it to a pyg graph and converting from there, in that case I get a different error .

This error doesn't happen when I am using a graph in Planetoid as in the Link prediction with DeepSnap example colab notebook.

What could be causing this problem? Is there a guide on how I can use custom data on DeepSnap?

Thank you for your help!

[Syzseisus]

This is not only for custom data, but also PyG data which has not node features: IMDB-b.

What I tried was:

pyg_dataset = torch_geometric.datasets.TUDataset('./imdb-b', 'IMDB-BINARY')
graphs = deepsnap.dataset.pyg_to_graphs(pyg_dataset)

And the full error is:

Traceback (most recent call last):
  File "cw224.py", line 39, in <module>
    graphs = GraphDataset.pyg_to_graphs(pyg_dataset)
  File "/opt/conda/envs/PR/lib/python3.8/site-packages/deepsnap/dataset.py", line 1276, in pyg_to_graphs
    return [
  File "/opt/conda/envs/PR/lib/python3.8/site-packages/deepsnap/dataset.py", line 1277, in <listcomp>
    Graph.pyg_to_graph(
  File "/opt/conda/envs/PR/lib/python3.8/site-packages/deepsnap/graph.py", line 2027, in pyg_to_graph
    Graph.add_node_attr(G, key, value)
  File "/opt/conda/envs/PR/lib/python3.8/site-packages/deepsnap/graph.py", line 1911, in add_node_attr
    attr_dict = dict(zip(node_list, node_attr))
TypeError: 'int' object is not iterable

I thought 'int' object is maybe 'node_attr', so I transformed dataset like:

transform = torch_geometric.transforms.Compose([T.Constant(value=-1), T.Constant(value=-2)])
pyg_dataset = torch_geometric.datasets.TUDataset('./imdb-b', 'IMDB-BINARY', transform=transform)
graphs = deepsnap.dataset.pyg_to_graphs(pyg_dataset)

but I got the same error.

Thank you for your help!