n_neighbors of point i includes i about umap HOT 3 CLOSED

In general I have been working with the (somewhat standard) view than the neighbors of a point includes itself. That does mean, as you correctly point out, that an n_neighbors value of less than 2 simply does not work. At the other end of the scale I believe using n_neighbors equal to the dataset size would be ... bad for many reasons. I am certainly happy to update the documentation to reflect the bounds on the values that n_neighbors can reasonably take, but I'm not sure I consider it a bug -- although I guess it certainly is if you interpret nearest neighbors to not include the point and get seemingly spurious results.

from umap.

Ah, I did wonder if I was just unfamiliar with the definition of neighbors. Perhaps extend the validity check on n_neighbors to be between 2 and X.shape[0] - 1? No big deal, though. Closing.

from umap.

I'll add a validity check -- I agree that that would be sensible and help alleviate confusion.

Thanks for the feedback/

from umap.