adjacency normalize problem

[Wwwwei]

hi,
I set normalize_adj=True in preprocess_input_graph function, then I use the normalized adj to train the GCN model and explain model. When I explain node, I find the loss will become very large and the explainer.masked_adj is too small for each edge to visualize.
Do I have that right? By the way, my graph is a weighted directed graphs.
I also want to konw why there is a self.coeffs param in ExplainModule?
Thanks.