Torch_geometric.data - something like pop function?

1

Hi everyone!

I have my own torch_geometric.data Dataset, that includes targets as well. What is the better way to “pop” the targets from it so I can have everything but targets in Dataset?

2

I’m not sure which Dataset you are using, but e.g. torch_geometric.data.Data uses an optional target attribute so I would assume you don’t need to pass it to the dataset while creating it.

3

I have something similar to gvp-pytorch/data.py at main · drorlab/gvp-pytorch · GitHub, but with targets

        data = torch_geometric.data.Data(x=X_ca, seq=seq, name=name,
                                         node_s=node_s, node_v=node_v,
                                         edge_s=edge_s, edge_v=edge_v,
                                         edge_index=edge_index, mask=mask, 
                                         targets=targets)

my question is I need to split this to data and targets, what is the best practice?
to get targets only i can use data[“targets”], but how to get everything else but targets?

        torch_geometric.data.Data(x=X_ca, seq=seq, name=name,
                                 node_s=node_s, node_v=node_v,
                                 edge_s=edge_s, edge_v=edge_v,
                                 edge_index=edge_index, mask=mask)
4

Solved it that way

targets = data["target"]
delattr(data,"target")
x = data