Leland McInnes

The transform is stochastic, so unfortunately there is not way to remedy this. If you really need a consistent (and presumably fast) transform I would recommend looking at the ParametricUMAP...

It handles things in batches, so permuting, or adding rows will change things. It shouldn't be *completely* different (that may be an issue), but it will certainly change.

Okay, that means there is likely something different happening in the nearest neighbor search, or, simply, the point is effectively torn between a few options -- i.e. it's embedding is...

A PR would be welcome.

UMAP as an algorithm is stochastic. As an optimization process it also does not differentiate between rotations or reflections of the same result. Running on many threads changes the stochasticity...

It seems like numba added an __init__ file so structref now needs to be more explicit. As as immediate workaround you can change: ```python from numba.experimental import structref ``` to...

Yes, this seems reasonable. Sorry for being slow to get to it, and thanks for the thorough analysis, it makes it a lot easier to just merge.

Mostly I've been using trustworthiness over varying neighborhood sizes, which, I agree, is not tractable for large data sets. With enough compute time (and some judicious pre-compute and code tuning...

I have tried UMAP on fashion MNIST. It does not magically separate the 10 classes (at least not as tidily as it does with digits), but the classes that it...

I can dig up a visualization I think. It is closer to LargeVis in appearance, but still a little different. As to what to trust; I think you have to...