Open lh123cha opened 3 months ago
A bfs tranverse on a 40M nodes graph costs 20s.If we do a proximity for each minibatch training it will cost a lot.Does you do a proximity ordering for each minibatch? And how to implement a fifo cache on a single gpu?
A bfs tranverse on a 40M nodes graph costs 20s.If we do a proximity for each minibatch training it will cost a lot.Does you do a proximity ordering for each minibatch? And how to implement a fifo cache on a single gpu?