Open miselico opened 2 years ago
In earlier works on link prediction, we find different ways to create splits for inductive settings. They would make sure that all nodes occur in the training set, such that an embedding can always be learned.
We can reuse or draw inspiration from pykeen. E.g. https://pykeen.readthedocs.io/en/stable/reference/triples.html#pykeen.triples.CoreTriplesFactory.split
In earlier works on link prediction, we find different ways to create splits for inductive settings. They would make sure that all nodes occur in the training set, such that an embedding can always be learned.
We can reuse or draw inspiration from pykeen. E.g. https://pykeen.readthedocs.io/en/stable/reference/triples.html#pykeen.triples.CoreTriplesFactory.split