zacharychenpk / hetsann Goto Github PK

Hi,
Thank you for this PyTorch implementation of HetSANN. Is this code written in a way to support batch size > 1? Thank you.

Hi, seems like you are looking for datasets. The paper authors have uploaded Aminer dataset and you could see it in their git repo.https://github.com/didi/hetsann

Hello @ZacharyChenpk, I like your pytorch-version code to implement the HetSANN. But I'm a little bit confused about one line code here. It seems like this line is used to calculate something like the softmax probabilities for all neighbor nodes with same type. But I'm not sure because the sum of dim=1 of attn_o is not equal to 1 by using attn_o - attn_o.max(). Could you please help me explain why it has to calculate the attention by this way? Thank you so much!