大型图的处理

当处理图级的超图分类问题，如果每个超图的节点和超边数量较多，batch加速处理对邻接矩阵进行类似dgl.batch的拼接方式时，似乎并不支持很大的batch_size，如何优化？或者在计算速度和显存之间取舍？

[LuShuaiqi-Geo]

I also encountered the same problem. When the number of graph vertices reaches about 20,000, there will be insufficient video memory and it is impossible to generate a supergraph from an ordinary graph. I am using a 3090 graphics card. This problem occurs when using larger memory. will be alleviated, but this limits the computational efficiency of the entire model and increases the dependence on memory.