Repeatability of embedding locations

[sgbaird]

@ancarnevali

I think this guide will help https://pytorch.org/docs/stable/notes/randomness.html#reproducibility

CrabNet is a bit tricky since it runs on a GPU and I don't know off the top of my head if it uses non-deterministic algorithms, but that shouldn't affect the embedding positions, only the target predictions.

Element Mover's distances should be reproducible as-is (i.e. it's calculated via an analytical formula). For UMAP (and by extension, DensMAP), see https://umap-learn.readthedocs.io/en/latest/reproducibility.html. The * in HDBSCAN* refers to the algorithm being deterministic if I understand correctly. So, you might get away with passing in {"random_state": 42} to umap_cluster_kwargs and umap_vis_kwargs.