search

uoguelph-mlrg / LDG

PyTorch code for "Learning Temporal Attention in Dynamic Graphs with Bilinear Interactions"
https://arxiv.org/abs/1909.10367
Other
98 stars 24 forks source link

NaN embeddings #6

Open anonymous-papers opened 3 years ago

anonymous-papers commented 3 years ago

Hi, I have two questions:

  1. I tried some my own datasets, some of them gave me good embeddings (the z), but some of my datasets are all NaN embeddings. I am wondering what causes this problem (like too many or too less edges or nodes?) and is there a way to fix that?

  2. In the Dyrep paper, it seems that they do not need the whole [N_nodes*N_nodes] Adjacency Matrix for calculation, but it is required in your code. Is there a way to avoid that? Since it makes large networks impossible to be run.

Thank you!

NameIsAlreadyTakenY commented 3 years ago

hello： Can you send me a copy of the code environment if you run it successfully？ I didn't find an introduction to the code environment in readme file，and hope the author can make it up later