search

PetarV- / GAT

Graph Attention Networks (https://arxiv.org/abs/1710.10903)
https://petar-v.com/GAT/
MIT License
3.15k stars 643 forks source link

Confusion about sp_gat's bias_mat #35

Open hapoyige opened 4 years ago

hapoyige commented 4 years ago

Hi, Petar

Thanks for your great works!

I have some questions about sp_gat model's bias_mat input. I notice that it is process from utils/process.py function preprocess_adj_bias()

indices = np.vstack((adj.col, adj.row)).transpose() # This is where I made a mistake, I used (adj.row, adj.col) instead

I just don't understand your comment here: the return values of the function is feed to tf.sparse_placeholder then, and (adj.row, adj.col) is standard input for tf.SparseTensor , So why you use (adj.col, adj.row) instead ? In other words, I will make the same mistake with you! Could you explain? Many thanks!

LeungH commented 4 years ago

I have the same confusion about this. Do you understand now?