Closed zhanlingm closed 1 year ago
Hi - apologies for the late reply.
It sounds like your learning rate is too high. Can you copy/paste the config you're using and provide the number of networks you're integrating, as well as the number of nodes and edges in these networks?
Hi, I'm going to close this due to staleness. Feel free to reopen if necessary.
Dear Professor, hello! First of all, thank you for taking time out of your busy schedule to help me solve my doubts. Thank you very much! When I was training multiple networks using BIONIC 0.2.6 CPU version recently, the initial Loss value dropped normally, and suddenly increased to hundreds of thousands when running to 400 Epochs. I would like to ask you what is the cause of this? My guess is that there is no correlation between nodes of a certain network and other networks in the early stage of network processing, resulting in a large number of isolated nodes. When randomly sampling these nodes, neighbor nodes cannot be found, resulting in a very sparse batch network predicted. I don't know if my guess is correct, but I hope the professor can answer questions. Thank you very much!