DiffPool example what is mask in TUDataset

[lisiq]

Hi everyone, I was trying to understand the code and the dataset in pytorch_geometric examples. I don't understand how is the mask in PROTEINS dataset generated, what does it represent and how is DiffPool making use of it?

Thank you very much!

[rusty1s]

The mask is generated by the ToDense transform, and indicates which nodes are valid nodes and which nodes were added due to padding reasons in order to create a dense adjacency matrix of shape [batch_size, max_num_nodes, max_num_nodes]. As such, mask is a [batch_size, max_num_nodes] matrix.

[lisiq]

Thank you very much that was helpful!

[lisiq]

@rusty1s Is there a tutorial available on how to create these datasets?

[rusty1s]

What do you mean exactly? This and this tutorial might be of interest to you!

[lisiq]

I have some graphs with different number of nodes, and I was wondering how to create a similar dataset. I found this tutorial https://colab.research.google.com/github/AntonioLonga/PytorchGeometricTutorial/blob/main/Tutorial15/Tutorial15.ipynb#scrollTo=69a46b4c so now I created similar textfiles and it seems to have solved my problem. Thank you!