How to Add Filter functions when aggregate neighbors

Hi team, is possible to add some filter functions in the heterogeneous graph message passing process.

For example, is possible to add some filter function In the multi_update_all function, like to filter some neighbors whose creation time is later than the current node to avoid data leakage?

Are you performing full-graph training or mini-batch training?

thanks for the help @mufeili . mini-batch training

Sorry for the late reply. I think you can define a custom message function for this purpose. For example,

def msg_func(edges):
    return {'m': edges.src['h'] * (edges.src['time'] < edges.dst['time'])}

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.