Rex Ying

I'm using Pytorch 0.4 in this repo. The other requirements are TensorboardX, networkx, and the standard python libraries (numpy, scipy, matplotlib etc.)

I think the tensorboardX interface has changed due to updates. An alternative is to use the new function: img = tensorboardX.utils.figure_to_image(fig) It's more convenient than manually convert fig to data.

i didn't use edge attributes here. should be easy to add. If categorical, you can have a different weight matrix for each edge type, and sum the messages for all...

> I tried using the PyTorch Geometric implementation, dense_diff_pool(). When A contains edge attributes as an additional dimension, the following computation (using torch.matmul()) throws an exception: > A(l+1)=S(l)^A(l)S(l), > where...

Thanks for the suggestions! For now i could think of using pytorch batch matmul A_perm = A.permute(2, 0, 1) S_perm = S.unsqueeze(0) S_perm_T = S.T.unsqueeze(0) A_next_level = S_perm_T @ A_perm...

Does this only use one of the edge type dimensions? You can also have weight matrices in graph conv layer to have an extra dimension corresponding to the edge attribute...

The difference from GCN is the DeepSet aggregation, and concatenation in GraphSAGE. But you are right, that this is doing whole graph convolution, rather than the minibatch updates in GraphSAGE.

It's for Proteins with node-label features.

Hi the soft-assign is the pooling method, base is the baseline, and set2set is the baseline with the set aggregation pooling for all node embeddings. The set2set method refers to...

Yes!