I want write a minibatch train for GAT.
During full batch training, I write code as:
graph.apply_edges(fn.u_add_v('el', 'er', 'e'))
and in minibatch I create some nodeflow by sampler,so I need to update the edge by using apply_blocks,I write as:
nf.apply_block(block_id,fn.u_add_v('el', 'er', 'e'))
But when I run it, it told me that
I guess fn.u_add_v may not be used in nodeflow? So how can I replace it?
By the way,the how can I use edge_softmax function to update the edge value in nodeflow?