How were these embeddings trained?

[bkj]

Is there software available to fit these kind of hyperbolic embeddings on our own datasets? If not, could you explain a little what the training procedure looks like?

Thanks! ~ Ben

[belizgunel]

Hi Ben,

Training was done using the combinatorial algorithm (under combinatorial/) followed by gradient-descent based optimization (under pytorch/), both publicly available in https://github.com/HazyResearch/hyperbolics. You can train embeddings on your dataset using those. Let me know if you have any further questions.