Different dimension in node-level output

[tliu1023]

Hi,

I’m wondering is there a way to handle the node classification where the dimension of output is related to the number of edge connecting to the node?

For example, node 1 is connected to 2 edges (1, 2), so the classification logit will be [0.1, 0.9], but node 2 connected to 3 edges (2,3,4), so the classification logit will be [0.3, 0.3, 0.4].

I’m thinking to apply a mask to the final classification logit, where the output has the same dimension, but it’s surely not the best approach.

Thank you!

[Rhett-Ying]

How about saving such logits [0.3, 0.3, 0.4] on corresponding edges respectively?

as for the mask you mentioned, do you mean set all the tensor dimension as max of connected edges? set zero on those unused dimensions?

[tliu1023]

Thank you for your response! you are correct about the implementation of the mask. But I think your way to save on the corresponding edge is more concise.