Can it be run on CPU? about shadow_gnn HOT 9 OPEN

Hi Parisa, thanks for your interest!

To run on CPU, please change the command line argument from --cpu to --gpu -1.

Please let me know if there is any other issues.

from shadow_gnn.

Thank you so much. That worked. However, I ran into a new problem, and I couldn't find a solution. I would be grateful if you could help me with it.

from shadow_gnn.

Maybe this issue is related:

https://stackoverflow.com/questions/68860621/ogb-dataset-i-can-not-import-pygnodeproppreddataset-from-ogb-nodeproppred

Could you please check if you have the packages mentioned in the above thread installed?

from shadow_gnn.

Thank you for your response. I went through the "Build and Run" section step by step, and also checked the installed packages, they are all installed. However, I still get the error.

Moreover, I tried to run another dataset (arxiv) as below:

>python -m shaDow.main --config_train\arxiv\vanilla\gcn_3_ppr.yml --dataset arxiv --gpu -1

and I got this error:

I'm not sure why I get these errors. Thank you for your help, in advance.

from shadow_gnn.

For your error on arxiv, I think you should provide the command line argument as --configs <file>.

So try this:

python -m shaDow.main --configs config_train\arxiv\vanilla\gcn_3_ppr.yml --dataset arxiv --gpu -1

I can continue looking into your issue in products once you confirm that the issue on arxiv is resolved.

from shadow_gnn.

Thank you. you were right. However, I get the same error for the arxiv dataset as well.

I also got another error when trying to run another dataset besides ogbn.

from shadow_gnn.

For the issue related with OGB, it should be some missing package / version mismatches for the PyG related packages. As a first step, could you please run the following in your terminal and let me know your outputs:

python -c "import torch; print(torch.__version__)"

python -c "import torch_geometric; print(torch_geometric.__version__)"

python -c "import torch_scatter; print(torch_scatter.__version__)"

python -c "import torch_sparse; print(torch_sparse.__version__)"

python -c "import ogb; print(ogb.__version__)"

For the non-ogb data like Flickr, you need to manually download the original files in the GraphSAINT repo to place them under the ./data/saint/flickr directory, as shown in the following sections of the README:

Can you please confirm if your ./data/saint/flickr directory contains the following files:

• adj_full.npz
• adj_train.npz
• class_map.json
• feats.npy
• role.json

from shadow_gnn.

Thank you for your help regarding non-ogb datasets. It worked and I got the results.

For the issue related to OGB, it seems the problem is with the torch_geometric. When I run python -c "import torch_geometric; print(torch_geometric.__version__)", I get the following error. Although when I run pip install torch_geometric I get Requirement already satisfied.

I also run the pip install torch_tb_profiler from this link and install torch_tb_profiler. But the problem still remains.

I appreciate your help and time.

from shadow_gnn.

I see. It's indeed a bit tricky to install it since all related packages need to be having compatible versions.

Sometimes, I also encounter issues when I set up a new env for torch geometric. I think uninstall torch geometric (e.g., pip uninstall) and then install again may resolve the issue.

After you uninstall torch geometric, maybe you can try going through these pip install steps here: https://pytorch-geometric.readthedocs.io/en/latest/notes/installation.html#installation-via-pip-wheels

In the worst case, you may need to create a new virtual env and reinstall all packages from scratches (sometimes I also have to do so, unfortunately).

You may also need to reinstall ogb after reinstalling torch geometric.

from shadow_gnn.