Is it require to create an adjacency matrix?

[senemaktas]

Hello,

I have created a graph dataset with my own dataset. And it is consist of nodes and labels.
I do not see necessary to create an adjacency matrices for graphs (nodes consist of the features and
I would like to see that model extract the features between those nodes in graph level way). That part is done.

When i run general gnn model it is looking for adjacency matrices. Is it require to create
an adjacency matrix? I tried giving it as None, but the result is the same.
Here is the error:
File "C:....\general_gnn.py", line 172, in call
z = layer([out, a])
File "C:....\base_layer.py", line 1037, in call
outputs = call_fn(inputs, *args, **kwargs)
File "C:.....\general_conv.py", line 137, in call
x, a, _ = self.get_inputs(inputs)
File "C:....\message_passing.py", line 183, in get_inputs
assert K.is_sparse(a), "A must be a SparseTensor"
AssertionError: A must be a SparseTensor

Thanks for your time.

[danielegrattarola]

Short answer: yes, you need an adjacency matrix.
You can just make a fake adjacency matrix with self-loops only (sp.eye) or even all zeros, but at that point, I would question the use of a GNN at all and whether you should instead use a MLP.

Cheers

[senemaktas]

Yes exactly, I thought same thing and filled adj matrix with ones. I probably confused it with another parameter to define as None. I will try to find proper way to create the matrix. Thank you very much for your answer.