Tackling class imbalance in PyTorch Geometric

[JopVerbeek]

Hi,

I am currently building a graph classification network using PyTorch Geometric. The dataset I am using unfortunately has large class imbalances. I was wondering if there is a straightforward method to tackle this, like for example in regular PyTorch I can implement WeightedRandomSampler() in the DataLoader itself. Is this maybe also possible in the PyTorch Geometric DataLoader or do I need alter the DataLoader class in order to be able to use, for example, WeightedRandomSampler?

Thanks in advance!

[rusty1s]

Yes, you can also do that in PyG, too. We make use of the PyTorch DataLoader under the hood, hence, all arguments of the PyTorch Dataloader can be passed as well (expect for collate_fn).