Minibatch training on graph


Can we train the models by minibatch? In other words, can we sample the neighborhood dynamically? I ask a similar question at How can I sample the neighborhood dynamically?. But nobody response. Thx!

How can I sample the neighborhood dynamically?

Hi, sorry for the late reply. We are working on sampling (mini-batch) based training now and hopefully we can bring them to you very soon. For the time being I guess you have to sample edges yourself and use send_and_recv for computing on the sampled neighbors (edges).