transform = RandomNodeSplit(split='random') causes Colab to crash

[AdarshMJ]

Hi @rusty1s

I am training node classification on Cora. Before i train the model I am performing Ricci Flow using this library - Ricci Flow on the dataset.
That is, convert PyG to NetworkX -

from torch_geometric.utils import to_networkx,from_networkx
G1 = to_networkx(data,node_attrs=['x'],graph_attrs=['y'])
print(type(datanx))

Compute Ricci Flow -

orf = OllivierRicci(G1, alpha=0.5, base=1, exp_power=0, proc=4, verbose="INFO")
orf.compute_ricci_flow(iterations=5)
G_rf = orf.G.copy()

Convert back to PyG

train_data = from_networkx(G_rf)
print(train_data)
Data(x=[2485, 1433], edge_index=[2, 10138], y=[2708], ricciCurvature=[2485], weight=[10138], edge_ricciCurvature=[10138], original_RC=[10138])

I want the train/test/val mask so Im using the transforms -

from torch_geometric.transforms import RandomNodeSplit
transform = RandomNodeSplit(split='random')
train_data = transform(train_data)
print(train_data)

My Colab crashes saying - Crashed for unknown reason. Is it somehow using up all the RAM thats causing the issue? Thanks in advance :)

[rusty1s]

This is hard to tell, I am sorry. Where does it crash exactly? Is it in the RandomNodeSplit execution?

[AdarshMJ]

No actually its while applying the transforms -

train_data = transform(train_data)

[rusty1s]

Is it possible for you debug further, i.e, which statement inside RandomNodeSplit?

[AdarshMJ]

Hi I am sorry I am not sure I understand how to do that. Do you want me to check the source of RandomNodeSplit?

[rusty1s]

Basically doing the same as before: It would be great to see where the crash happens inside the computation in RandomNodeSplit, see here. You might be able to do this via a debugger or via print debugging inside the source code.

[AdarshMJ]

Okay I can check this out. Thank you!