AttributeError: 'NoneType' object has no attribute 'origin'

rusty1s / pyg_autoscale

Implementation of "GNNAutoScale: Scalable and Expressive Graph Neural Networks via Historical Embeddings" in PyTorch

http://arxiv.org/abs/2106.05609

MIT License

159 stars 27 forks source link

AttributeError: 'NoneType' object has no attribute 'origin' #14

Open skycr4ft opened 2 years ago

skycr4ft commented 2 years ago

Hi @rusty1s, thank you for your great contribution to large-scale graph learning. These days I am facing a problem regarding the configuration of GAS. This issue occurs while importing the GAS library. I've tried many ways, shifting the version of torch, cuda, and torch-sparse, to fix it, but none of them make a difference. So I wanna ask you what can I do with it? issue

rusty1s commented 2 years ago

This has nothing to do with PyTorch or torch-sparse. Are you sure you installed the PyGAS library (there should exist some *.so files inside torch_geometric_autoscale. Otherwise, please run

pip uninstall torch_geometric_autoscale
pip install --verbose .

and report the log.

skycr4ft commented 2 years ago

The library is successfully installed (but I don't find some *.so files in the directory you said), and the problem still exists issue .

skycr4ft commented 2 years ago

Ohhh, there are indeed two *.so files, and it looks like these two are causing the problem. lib

rusty1s commented 2 years ago

Mmh, in case the *.so files exist, importlib should find them as well (in which case find_spec should not return None). Can you do some manual debugging on why these libraries are not found?

WendyQi-Guo commented 1 year ago

Ohhh, there are indeed two *.so files, and it looks like these two are causing the problem.

Hello, have you been able to resolve this issue? I've encountered the same problem as you.