For certain projections at least (RandomBinary for example), it doesn't seem to make a difference whether you normalize the elements for where they hash. But if you care about the actual distance in real space, it appears difficult to recover this since the vectors are stored normalized.
I admit I don't understand LSH at a deep level -- would it break the math to make the unitvec part an option?
For certain projections at least (RandomBinary for example), it doesn't seem to make a difference whether you normalize the elements for where they hash. But if you care about the actual distance in real space, it appears difficult to recover this since the vectors are stored normalized.
I admit I don't understand LSH at a deep level -- would it break the math to make the
unitvecpart an option?