build new temporal dataset

[tliu1023]

Hi,

I'm new to PyG and I'm a little confused about the usage of it. I'm trying to solve a task like seq2seq prediction but on a series of graphs. That is f(G_t0, G_t1, ... G_tn) -> H_t0, H_t1, ..., H_t0, which is similar to DCRNN but not same. So I'm trying to build a new dataset from scratch using PyG and I'm confused which class should I use, Data or temporalData or Dataset?

Thank you

[rusty1s]

Have you also looked at https://github.com/benedekrozemberczki/pytorch_geometric_temporal? This might be the better tool for what you are trying to achieve.

Otherwise, I don't think TemporalData is the appropriate choice here (it expects a list of temporal events rather than graphs coming from different snapshots). As such, you could consider having a torch_geometric.data.Dataset that returns a sequence of graphs/torch_geometric.data.Data objects within __getitem__:

class MyDataset(Dataset):
    ...

def __getitem__(self, idx):
    # get data objects
    return G_t0, G_t1, G_t2


dataset = MyDataset()
loader = DataLoader(dataset, batch_size=8)

for batch_t0, batch_t1, batch_t2 in loader:
    pass