GAT runtime error

[salmon1802]

Thank you very much for your meaningful work! However, when I change the GCN to GAT in the config file, the following error is reported, I made attempts but still did not run successfully.

Below is my error log:

Processing...
Done!
seed for seed_everything(): 2024
2024-05-25 23:48:45,472 - Loaded dataset: ABIDE
0%| | 0/26 [00:00<?, ?it/s]edge_index shape: torch.Size([2, 1280000])
x shape: torch.Size([6400, 200])
Maximum index in edge_index: 6399
edge_attr: torch.Size([1280000])
0%| | 0/26 [00:00<?, ?it/s]
Traceback (most recent call last):
File "D:/GraduateStudent/code/BrainGB-master/examples/example_main.py", line 152, in
main(parser.parse_args())
File "D:/GraduateStudent/code/BrainGB-master/examples/example_main.py", line 78, in main
test_micro, test_auc, test_macro = train_and_evaluate(model, train_loader, test_loader,
File "D:\GraduateStudent\code\BrainGB-master\examples\train_and_evaluate.py", line 26, in train_and_evaluate
out = model(data)
File "D:\zj\miniconda\envs\Dual-HINet\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "D:\GraduateStudent\code\BrainGB-master\src\models\brainnn.py", line 18, in forward
g = self.gnn(x, edge_index, edge_attr, batch)
File "D:\zj\miniconda\envs\Dual-HINet\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "D:\GraduateStudent\code\BrainGB-master\src\models\gat.py", line 147, in forward
z = conv(z, edge_index, edge_attr)
File "D:\zj\miniconda\envs\Dual-HINet\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Users\28334\AppData\Local\Temp\torch_geometric.nn.sequential_f0afbb_syfjdr4d.py", line 23, in forward
x = self.module_0(x, edge_index, edge_attr)
File "D:\zj\miniconda\envs\Dual-HINet\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "D:\zj\miniconda\envs\Dual-HINet\lib\site-packages\torch_geometric\nn\conv\gat_conv.py", line 341, in forward
out = self.propagate(edge_index, x=x, alpha=alpha, size=size)
File "D:\zj\miniconda\envs\Dual-HINet\lib\site-packages\torch_geometric\nn\conv\message_passing.py", line 538, in propagate
coll_dict = self._collect(self._user_args, edge_index,
File "D:\zj\miniconda\envs\Dual-HINet\lib\site-packages\torch_geometric\nn\conv\message_passing.py", line 399, in _collect
self._set_size(size, dim, data)
File "D:\zj\miniconda\envs\Dual-HINet\lib\site-packages\torch_geometric\nn\conv\message_passing.py", line 298, in _set_size
raise ValueError(
ValueError: Encountered tensor with size 1280000 in dimension 0, but expected size 6400.

Process finished with exit code 1

[Gemaxis]

It's due to your environment not matching the version. I faced the same problem, and I changed my Torch version to fix it.
Btw, if you are using Windows, you can select torch=1.8.1+cu102 with the corresponding torch_geometric version.

[salmon1802]

Thanks for the reply @Gemaxis , I've run it successfully after changing the version as you said : )