Zuobai Zhang

Hi! You don't need to train the GIN first, since the InfoGraph itself defines a pretraining task. We wrap it with as a 'model' instead of a 'task' in TorchDrug...

Hi! Thanks for your advise! Yes, your understanding on `predict` is correct. The output of `predict` was originally designed for the `forward` function, where we define the loss function. However,...

Hi! Thanks for pointing out this issue. This bug is caused by a typo in the commit https://github.com/DeepGraphLearning/torchdrug/commit/f8d2de6fc91c1ff60a62b0279d61d1be8e1638ff, which slightly modified the `data.Graph.split()` method. I've fixed this typo in the...

Hi, Farhad! Thanks for your interest in contributions. I think it is a good idea to implement some equivariant graph neural networks in our library. GemNet is one of the...

Hi! This bug is because `SchNet` requires the 3D conformations of molecules to perform its convolution. But the `datasets.Clintox` dataset doesn't provide node positions for molecules. You can switch to...

For ProteinBert, you can find the configs in https://github.com/DeepGraphLearning/PEER_Benchmark.

Hi! Every class inherited from `core.Configurable` should has an attribute 'config_dict', e.g., `data.MoleculeDataset`. Could you show the code of our defined dataset? It will help us find the reason of...

Hi! Thanks very much for the useful suggestion! You're right that the memory can be saved by loading proteins and ligands once and use dictionaries to index them. We indeed...

Hi! This is because the transform `TruncateProtein` does not have a 'residue' argument. By default, the transform will truncate the protein in terms of the number of residues. For details,...

Hi! Thanks for raising this issue! The dataset pre-loading part is very time-consuming, so we'd like to only do it when we first process the dataset and save the file...