class AristoV4(**kwargs)[source]

Bases: PackedZipRemoteDataset

The Aristo-v4 dataset from [chen2021].


The dataset is based on the Aristo tuple KG from


While the original dataset is described as having 44,950 entities, after removing the entities not present in train, only 42,016 remain. Similarly, only 1,593 relations occur in training (from the original 1,605 ones). Consequently, some validation and testing triples are removed (originally: 20,000). Finally, only 242,567 of 242,594 original training triples are unique.

Initialize the Aristo-v4 dataset.


kwargs – keyword arguments passed to pykeen.datasets.base.ZipFileRemoteDataset.