search

divelab / DIG

A library for graph deep learning research
https://diveintographs.readthedocs.io/
GNU General Public License v3.0
1.84k stars 281 forks source link

Explanation for the Trained Models #105

Closed mahimanzum closed 2 years ago

mahimanzum commented 2 years ago

Hi, Thank you for sharing this very useful package. Are there any available explanations on how the models used for the Expandability (x-graphs) portion are trained? I mean the output shape or objective. I am trying to check If I can use all the explainable models on a single dataset like Tox-21. Please let me know if there are any available resources. Thanks.

CM-BF commented 2 years ago

Thank you for your issue. I'm not sure I've fully understood your question. We train our models with a general PyG pipeline and Adam optimizer. About the output shape, you can check it here: https://github.com/divelab/DIG/blob/dig/examples/xgraph/deeplift.ipynb.

mahimanzum commented 2 years ago

Hello, Sorry for not being specific. For example, I am trying to use pgExplainer in Tox21 dataset on the trained model. I am getting errors one after one. Is there any codes available for graph level explanations (for datasets like Tox21) with pgexplainer and subgraphX?

CM-BF commented 2 years ago

Hello, Sorry for not being specific. For example, I am trying to use pgExplainer in Tox21 dataset on the trained model. I am getting errors one after one. Is there any codes available for graph level explanations (for datasets like Tox21) with pgexplainer and subgraphX?

Sorry about the inconvenience, I think it may be the problem of the PyG version. We will check it right after and response to you as soon as possible.

CM-BF commented 2 years ago

I'm not sure whether you were facing the problem only with the PGExplainer or not. The current dig-stable branch is updated for the latest PyG version. May you try there? If there is any problem, can you provide a full error report so I can reproduce and locate the problem?

Oceanusity commented 2 years ago

Hello, there are example codes with PGExplainer and SubgraphX. Please refer to PGExplainer and SubgraphX. Also, you can find reference code in the benchmark directory providing examples of training models and explaining the model with multiple explainability methods.

CM-BF commented 2 years ago

I'll close this issue if no more discussion.