HAN
HAN copied to clipboard
About the code in DGL
The code in https://github.com/dmlc/dgl/blob/master/examples/pytorch/han/model.py Line 19,Line21.
beta = torch.softmax(w, dim=1)
return (beta * z).sum(1)
The dim should be 0 ?
The features are on dim=1 . So it is correct.
I found code in dgl obtains inferior performance. Do you have any idea? Please ref this issue! Thanks if you have time
@lipingcoding what doo you mean by performance? speed or final accuracy?