How to copy a `torch_geometric.data.Data` object?

❓ Questions & Help

I like the torch_geometric.data.Data object a lot, especially in combination with all the other neat stuff and the general apply function. One thing I am missing and currently have a stupid work-around for is the copying of such a data object.

E.g. the apply function does generate new torch.Tensor objects but overwrites the attributes of the existing Data object. So when I pass a Data object into a function and want to process the data without affecting the input object, I need to make a copy of the Data object.

Currently I am doing this via data = torch_geometric.data.Data.from_dict(data.to_dict()). My question is, is there a better way of doing this or a reason why there does not exist a copy method? Thanks for your input on this 😃

[de-gozaru]

Hi @rusty1s

Thanks for your answer, so If I install now PyG using

pip3 install https://github.com/pyg-team/pytorch_geometric.git@master

It should be ok, right?

[rusty1s]

There is a Data.clone() to deep-copy a data object. Alternatively, copy.copy(data) and copy.deepcopy(data) should work just fine.