Table 4 Sensor/Actuator Percentages Reproducibility

[Karn-2000]

First of all, Thanks for your work and the provided code 😄
I am trying to recreate the actual sensor/actuator for each detected attack on SWaT data from your code.
For this I followed the given example.ipynb for SWaT dataset and loaded ADmodel.pt
As mentioned in the paper, the size of background dataset as 500 and parameter m was set to 20.
Still I'm having a little problem with reproducibility of Table 4 in the paper, the sensor/actuator percentages do no match with the table.
Could you please let me know if I am missing something order to get the reported scores in Table 4.
Thankyou

[ManuelFranco]

Hi!

Thank you very much for your interest in the work, and for taking the time to experiment with the code.

Unfortunately, the ADmodel.pt file included in that folder is not the original model used in the paper (we decided to upload a simplified model instead). Because of that, the exact percentages in Table 4 cannot be fully reproduced with the uploaded model.

That said, the results should remain qualitatively similar, especially regarding the comparison between KernelSHAP and ApproShaTS in terms of information loss.

If you’d like, feel free to email me your current experimental setup at manuel.francop@um.es. I’ll be happy to take a look and help you reproduce the reported results as closely as possible.

Thanks again for your interest, and please don’t hesitate to reach out!

Best regards,
Manuel