In the large-scale training post, at the end it states that three large graphs with 100 features/node were generated using RMAT. The link goes to the paper, but there does not seem to be an implementation referenced in the post. Will the implementation used also be open-sourced as part of DGL/part of DMLC? I think it would be useful for people working on the performance/scaling side of things to be able to also generate random graphs
