I can not understand the following weight setting used in the implementation of the paper Variational Graph Auto-Encoders, see the weight seting in the foll