acbull/hgt-dgl

Code for "Heterogeneous Graph Transformer" (WWW'20), which is based on Deep Graph Library (DGL)

Heterogeneous Graph Transformer (HGT)

Alternative PyTorch-Geometric implementation

Heterogeneous Graph Transformer is a graph neural network architecture that can deal with large-scale heterogeneous and dynamic graphs.

You can see our WWW 2020 paper “Heterogeneous Graph Transformer” for more details.

This implementation of HGT is based on Deep Graph Library (DGL) API, following their tutorial

TODO: Current DGL example use random input node feature. We'll later add other benchmark using meaningful input node features.

hgt-dgl's Issues

RTE Module

Any tips on how best to implement the RTE module in DGL?

how to change node classification task to link prediction task?

You work is genuinely meaningful and helpful to my interest! I'm wondering how to change it from a node classification task to a link prediction task? Could you please give some kind suggestions?

batch training

@acbull @songjiang0909 @BanksyX @ShichangZh
Assuming we have a sampling strategy,
how we can use mini-batch training?

dgl.heterograph no longer supports graph construction from a SciPy sparse matrix

Hi, I encountered the following error at the point of creating G (train_acm.py line 15).

DGLError: dgl.heterograph no longer supports graph construction from a SciPy sparse matrix, use dgl.from_scipy instead.

Is there an updated version with from_script to create the graph instead?

the code‘s dataset is invalid, how to get the data?

I would like to ask "data_url = 'https://s3.us-east-2.amazonaws.com/dgl.ai/dataset/ACM.mat'" in the code is invalid, how to get the data.

sample_subgraph

Hi, bro,
this implementation of HGT is based on DGL.

But, I can not find the function sample_subgraph, which exit in HGT based on Pytorch Geometric.

So, you will inplement sample_subgrapg in this version in the future?

Invalid dataset url

https://s3.us-east-2.amazonaws.com/dgl.ai/dataset/ACM.mat