E-means: 10x faster k-means at equal loss

Re:E-means: 10x faster k-means at equal loss by jdonaldson on Monday November 09, 2026 @11:43PM (#50436146)

Here's one of my own efforts on density-aware locality-sensitive hashing for embeddings. It uses k-means at one stage, I'll give this approach a shot.

https://dyf.io/

[ Parent | on HN ]

Re:E-means: 10x faster k-means at equal loss by swann_bessa on Tuesday November 10, 2026 @10:00AM (#50441331)

Hi, I'm the author of e-means. If you can send them, I would love to hear about the results you got !

[ Parent | on HN ]

Re:E-means: 10x faster k-means at equal loss by jdonaldson on Tuesday November 10, 2026 @02:10PM (#50444800)

Thanks for the attention! (It's all that matters) : https://jjd.io/posts/a-resolution-dial-for-e-means.html

Long story short... e-means produced better results than my dyf algorithm for the majority of the datasets I used as a comparison. But, dyf still managed to be much faster (which matters more for the embedding work that I do).

I did a bit of open heart surgery and transplanted the recursive partitioner from dyf into e-means, and found that it can work really well to auto-select the k-parameter (at least in everyone's favorite mnist example).

[ Parent | on HN ]