I cannot run this method based on the current torch_geometric version (2.2.0)

[HelloWorldLTY]

Hi all, I can run successfually based on single GCN or GAT model, but if I try to call the GraphMAE model, for example, Premodel() function to generate a new model, I will meet segementation fault without any errors. Are there ant solutions towards this problem? Thanks a lot.

[HelloWorldLTY]

I install pytorch based on conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia I install pytorch_geometric based on pip install torch_geometric

[THINK2TRY]

@HelloWorldLTY Hi thanks for your attention! It seems that you are trying to run the pyg version of GraphMAE. And it is expected to follow the instruction in pyg for installation. In our experiments, the program runs successfully in Linux. Could you provide more details about the problem description?

[HelloWorldLTY]

Hi, I used jupyter notebook to run it and the directly error is kernel dead.

Here are my version details:

pytorch 1.12.1 cpu_py39h3439074_1 conda-forge pytorch-cuda 11.7 h778d358_3 pytorch pytorch-lightning 2.0.0 pypi_0 pypi pytorch-metric-learning 2.0.1 pypi_0 pypi pytorch-mutex 1.0 cuda pytorch

torch-geometric 2.3.0

[HelloWorldLTY]

Hi, there are two solutions I can guess for this problem. Approach 1 is using conda. Maybe you can upload your yaml file for your conda envrionment and I can set up such environment in my server. Approach 2 is using torch geometric. Sincer I only need the encoder part of your model, maybe you can consider to integrate your method into PYG. Thanks a lot!

[THINK2TRY]

Hi, below is a more detailed list of the environment information for PyTorch and PyG:

  - torch==1.12.0+cu113
  - torch-cluster==1.6.0
  - torch-geometric==2.2.0
  - torch-optimizer==0.3.0
  - torch-scatter==2.0.9
  - torch-sparse==0.6.15

It's possible that there may be conflicts in your current environment. To avoid this, you may consider creating a new Conda environment for your experiments. I have also uploaded the env.yml of my environment, which may be helpful.

[HelloWorldLTY]

Thanks a lot!