There are cases - which come up suprisingly often with embeddings - where a point will have a number of duplicate points > K.
It seems incorrect to simply leave a point as a nearest neighbor of only duplicate points, but not a nearest neighbor of other duplicates by random chance.
How are you handling duplicates?
There are cases - which come up suprisingly often with embeddings - where a point will have a number of duplicate points > K.
It seems incorrect to simply leave a point as a nearest neighbor of only duplicate points, but not a nearest neighbor of other duplicates by random chance.