issue in running the code

[jeffreyhusc]

Traceback (most recent call last):
File "/Users/sirihu/.conda/envs/envcwn/lib/python3.8/runpy.py", line 194, in _run_module_as_main
return _run_code(code, main_globals, None,
File "/Users/sirihu/.conda/envs/envcwn/lib/python3.8/runpy.py", line 87, in _run_code
exec(code, run_globals)
File "/Users/sirihu/data/cwn/exp/run_mol_exp.py", line 105, in
exp_main(passed_args)
File "/Users/sirihu/data/cwn/exp/run_mol_exp.py", line 26, in exp_main
curves = main(parsed_args)
File "/Users/sirihu/data/cwn/exp/run_exp.py", line 90, in main
train_loader = DataLoader(dataset.get_split('train'), batch_size=args.batch_size,
File "/Users/sirihu/data/cwn/data/datasets/dataset.py", line 370, in get_split
if self.indices is not None:
AttributeError: 'OGBDataset' object has no attribute 'indices'

def get_split(self, split):
if split not in ['train', 'valid', 'test']:
raise ValueError(f'Unknown split {split}.')
idx = self.get_idx_split()[split]
if idx is None:
raise AssertionError("No split information found.")
370: if self.indices is not None:
raise AssertionError("Cannot get the split for a subset of the original dataset.")
return self[idx]

[crisbodnar]

Can you please provide the exact command you used and all the context for us to be able to debug? Are you using the version of the libraries specified in our requirements file? Did you make any modifications to the code? Based on the error, it seems you are trying to further split one of the train, test or validation splits.

For reference, I have just followed the README instructions on a clean machine and ran a couple of epochs of training on OGB-MOLHIV without any problems using sh exp/scripts/cwn-molhiv-small.sh.

[crisbodnar]

Hi @jeffreyhusc! We have spotted a problem with the dependencies when using CUDA-based libraries. This has been solved in PR #94 and might have caused the issues you observed. Please try running the code again by following the latest installation instructions from the README.

I will be closing this issue now, but feel free to reopen it if this problem shows up again.