Questions about the sampling from test data

[huaishengzhu]

Hi,

I'm wondering which result you utilize for your paper's result, sampling from perturbed test graph data or sampling from the noise as mentioned in your paper. Because I find that you utilize perturbed test graph data as the input of your model and then calculate MMD discrepancy between denoised graph data and true test data (in the train.py, function 'fit' for test data). I'm not sure whether I understand it well at that point and hope to get your response soon.

Thanks