GraphStore: how to split the data RandomLinkSplit

[denadai2]

Dear PyG community,

I am using the GraphStore and FeatureStore to scale up GNNs. Is there an easy way to split data with something like RandomLinkSplit? I do not understand where I can specify edge_label_index or something like that.

Moreover, are there plans for edge features?

Thaaaank you

[rusty1s]

Transforms currently do not know about the concept of GraphStore/FeatureStore, and I am afraid you would need to take care of splitting the data by yourself for now. edge_label_index would be then used as usual, i.e., as input argument to the LinkLoader.

Plans for adding edge features are in the making. FeatureStore can handle edge-level features by now, but they are not yet properly filtered during feature fetching.

[denadai2]

Ohh thanks! But then, where should I store edge_label_index? In the feature store?Thank you!Sent from my iPhoneOn 11 Jan 2024, at 09.12, Matthias Fey ***@***.***> wrote: Transforms currently do not know about the concept of GraphStore/FeatureStore, and I am afraid you would need to take care of splitting the data by yourself for now. edge_label_index would be then used as usual, i.e., as input argument to the LinkLoader. Plans for adding edge features are in the making. FeatureStore can handle edge-level features by now, but they are not yet properly filtered during feature fetching. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>

[rusty1s]

edge_label_index wouldn't be stored in the GraphStore IMO, it should be just an in-memory tensor.

[denadai2]

thanks!