RandomLinksplit, GraphStore and FutureStore for heterogeneous link prediction model

[kadir-gunel]

Hello,

I am working on a heterogenous graph model for link prediction which uses RandomLinkSplit method. The problem that I am facing is that the RandomLinkSplit uses too much memory hence terminates the process which depends on the input graph data.
I read the docs and it suggests to use GraphStore and FeatureStore abstract classes for large datasets. Until here everything is fine. I managed to create dataset by using the LinkNeighborLoader. But I don't know how to split the dataset into train,val and test sets by applying negative_samples_ratio etc. for forming random links between nodes.

I found something similar on #7181, but o.p. has many graph data where in my case, there is only one graph.

Could someone shed light on it ?

B.R.