About Environment For xgraph/PGExplainer

[divelab]

About Environment For xgraph/PGExplainer

[Oceanusity]

Hi, thanks for your issue. I have updated the install.sh file and add the required packages.

[Gori-LV]

Just a very quick question, I see most the xgraph papers already have their own implementation, have you compare your experimental result with theirs? regarding both GNN performance/behavior and explanation result, do they align with each other? thanks!

[Oceanusity]

Thank you for your questions.
First, some methods such as GNN_LRP haven't released an official implementation yet, and some experiment settings are not clear. Thus, we coherently follow the algorithm details to have reasonable comparable results with other algorithms.
Second, for some methods, we follow the official implementations in the packages or repositories and fit them into our framework. It's important that the explanation results are highly related to the parameters of GNN models, and some optimization-based methods may have different results with different initializations such as GNNExplainer. Therefore, we build the same GNN structures and transfer the provided model parameters to test some methods such as PGExplainer.
In a word, we can guarantee the correctness of algorithm implementations but can't promise the results are exactly the same.

[Gori-LV]

I see, thanks a lot for prompt answering, this is a great project, thanks for sharing!