Gnn's multi-hop neighbourhood sampling for each hop

[mahdis-saeedi]

for the gcn model in the phase of embedding generation:

• when we have +data.embedding.model.gnn.gcn.h=[ ] in command run, we see h in the folder name. ( better to see hNone)

• we expect to face an error with the command +data.embedding.model.gnn.gcn.nn=[10] +data.embedding.model.gnn.gcn.h=32
  because there is only 1-hop neighborhood sampling and there exists no hidden layer, but the code is running and generate the results in a folder its name includes h32.

[hosseinfani]

Hi @mahdis-saeedi
I fixed the folder naming as when h is empty or [], it becomes .hNone.

For the second item, seems there is no issue with pyg as it uses the last number for hops with no number -chatgpt says

https://pytorch-geometric.readthedocs.io/en/2.5.2/modules/loader.html

[mahdis-saeedi]

Thank you @hosseinfani .

[hosseinfani]

Hi @mahdis-saeedi
pyg-team/pytorch_geometric#10540