search

acbull / pyHGT

Code for "Heterogeneous Graph Transformer" (WWW'20), which is based on pytorch_geometric
MIT License
775 stars 162 forks source link

Question on HAN implementation #28

Open shirinmous opened 3 years ago

shirinmous commented 3 years ago

Our implementation is a little bit different from the original HAN paper. We just add the meta-path when we create the graph dataset, and just treat them as another type of edges. The model architecture is the same as their original paper. Since we can still use sub-graph sampling to train over the large-scale graphs, the model can still be scaled to our dataset.

Due to a slightly different pipeline (as I stated above) with our HGT model, we didn't prepare HAN on this Github project. Sorry about that

Originally posted by @acbull in https://github.com/acbull/pyHGT/issues/18#issuecomment-698661667

shirinmous commented 3 years ago

Hi, thanks for sharing your work. In the above issue, you said that you treat meta-path as another type of edge to be able to use sub-graph sampling for HAN. I think that your graph for HAN is still a hetero graph. How you use HAN original implementation with a hetero graph ?

susierao commented 3 years ago

Hi @acbull , thank you for sharing the work. I would also be interested in the HetGNN and HAN implementation you had. Would it be possible to share them?

acbull commented 3 years ago

For these baselines, I just slightly modify them from their original code.

Our implementation is a little bit different from the original HAN paper. We just add the meta-path when we create the graph dataset, and just treat them as another type of edges. The model architecture is the same as their original paper. Since we can still use sub-graph sampling to train over the large-scale graphs, the model can still be scaled to our dataset.

Due to a slightly different pipeline (as I stated above) with our HGT model, we didn't prepare HAN on this Github project. Sorry about that. You can refer to this repo for this implementation (https://github.com/yangji9181/HNE/tree/master/Model) as reference.