The experimental results are inconsistent with those in the paper.

[jweihe]

Hello, thank you for the work you've done. I have some questions regarding the experimental results in the paper "GOOD-D: On Unsupervised Graph Out-Of-Distribution Detection". I ran it locally with the given parameters and code, but the results differ from those in the paper. Why is this? My replication results are shown below. "GOOD-D(local)" is the result I got from running with the parameters in the GitHub code. The next two lines show the results from the paper. As can be seen, there's a significant discrepancy on datasets such as PTC-MR+MUTAG, FreeSolv+ToxCast, etc. The AUC metric differs by almost 10 points from your results, and others vary by 1-2 points. Can you explain these results or provide the correct way to replicate them? Thank you! I look forward to hearing from you.