NGNN
NGNN copied to clipboard
Fashion Graph created but not used?
I've been working through this code, and found it slightly odd that the fashion graph created early on (aka: total_graph, G) does not get utilized during the training process.
The positive and negative training samples begin from a blank matrix and don't utilize any edge weights from total_graph, and while the main function cm_ggnn accepts the total_graph as an input it eventually goes un-used.
Is this desired behavior? What is the purpose of the total_graph?
I have the same question - this is extremely surprising, can someone explain?