Graph-convolutional
Graph-convolutional copied to clipboard
Issues on graph.py
I appreciate your neat Pytorch code, but I think you may have some issues regarding code in graph.py.
In the original paper, we are supposed to have three different weights for different edge categories (equation 2). However, your code only offers one single weight matrix (line 33 in graph.py).
@CHELSEA234 ,
Hi, it's been a long time since you open this issue, not sure if you can remind these details...
I think the transformation matrixes for the three kinds of edge were implemented via the self.V_in, self.V_out and self.W_self_loop, see line-33, line-46 and line-58 in graph.py. Are you on the master branch?