How to add self loops for nodes without self-edges?

[wgeul]

I'm using HeteroData objects and I'm experimenting with self loops, adding the self loops like this:

data = T.AddSelfLoops()(data)

I notice the edge_index size expands for existing self-linked edges ('source', 'to', 'source'), but I don't see new edges being created for self-linked edges that were not yet in the data object, e.g. ('target', 'to', 'target'):

HeteroData(
source= { x=[..]},
target = { x =[..]},
('source', 'to', 'source') = { edge_index = [2, ..]}
)

So no addition of ('target', 'to', 'target')

What's the best way to add self loops for target?

[wgeul]

I presume this would do, please confirm:

data = T.AddSelfLoops()(data)
for node in data.metadata()[0]:
                if (node, "to", node) not in data.metadata()[1]:
                    dim = data[node].x.shape[0]
                    data[(node, "to", node)].edge_index = stack(
                        (arange(dim), arange(dim))
                    )

To be honest I expected to get self loops for all node types as default behavior.

[wsad1]

The edge type ('target', 'to', 'target') does not exist, so AddSelfLoops will no add the edge type and add self loops to it.
Your approach above looks good.
You could also try

data[('target', 'to', 'target')].edge_index = torch.empty(2,0)
data = T.AddSelfLoops()(data)

[wgeul]

Thanks @wsad1