Making a custom Dataset

❓ Questions & Help

Hello all, I am fairly new to pytorch_gemoetric and this might be a noob question. I have an numpy ndarray of shape [22, 313] and I want to represent it as a graph, with 22 nodes connected to each other correlation between channels as edge weight. Using networkx, I create the graph and weight matrices and finally- data_g = from_networkx(G) data_final = Data(x=data, edge_index=data_g.edge_index, edge_attr=data_g.weight) Is this the correct way to go… Thanks in Advance

[rusty1s]

That’s correct. The easiest way to achieve that is to just pass a list of Data objects to the DataLoader (with. data.y set for each).

[srikanthmalla]

Hi @rusty1s , I did that (return seperately) in get method, not __getitem__ and I got the error mentioned in the starting of the post. But when I added separate data into Data object using get method, I was not getting any error. May be should also check this __getitem__ method.

Thanks Srikanth