Muhan Zhang

The .mat file contains two variables, 'net' and 'group'. The 'net' contains the adjacency matrix, and 'group' contains the node attributes. You may use MATLAB or Python scipy.io to open...

Can you try to set a breakpoint before `g_list` in `helper()` function to see what it prints? Seems the helper does not return any subgraphs, not even an empty list.

Hi! The pytorch_DGCNN implementation does not support windows. See [issue](https://github.com/muhanzhang/pytorch_DGCNN/issues/36). You can try to use the pytorch_geometric implementation of SEAL [here](https://github.com/facebookresearch/SEAL_OGB). It is faster and uses modern graph deep learning...

Seems the Word2Vec function has changed its API. Can you check the latest Word2Vec documentation? Maybe the "size" is changed to sth else.

Hi! I suggest using the latest implementation of [SEAL](https://github.com/facebookresearch/SEAL_OGB) using pytorch geometric. It is much more efficient on large graphs and support customized datasets, too. That repository also implements some...

Hi! It is reflected in this [section](https://github.com/muhanzhang/SEAL/blob/ca1f019a15fb0c21796042165b4e6bee73981dd3/Python/util_functions.py#L193-L198) of the code. When we compute all nodes' distance to target node 0, we temporarily remove the other target node 1 from the...

If you don't remove the positive link, the GNN model has already known there is a positive link there thus makes trivial predictions. Think about when you test, there are...

Thanks for the question. Yes, I have updated the README to include the link to the 12-layer model: https://drive.google.com/drive/folders/18GU9g5DNiHn2MOVKOiF1fCwNQMTA-mnH?usp=sharing

Sorry. I forgot to mention there are some necessary changes to enable 12-layer ENAS evaluation. First, change "outputs_6" to "outputs_12" in this [line](https://github.com/muhanzhang/D-VAE/blob/3569a4a3bf5ecf8e12ae548e0d22e72062d29e26/software/enas/src/cifar10/evaluation.py#L37). Second, change "child_num_layers" to 12 in this...

Hi! A batch contains multiple disconnected subgraphs, which are compressed into one single block-diagonal adjacency matrix `ADJ` by PyTorch Geometric. Your original `ACsr` is the observed graph from which you...