why there is no option of cosine distance when initialize LSHConstructionParameters?

[dorioliu]

hello the distance measurement for what i using now is cosine distance , i know FALCONN has good performance in cosine distance, why there is no option of cosine distance when initializes LSHConstructionParameters? only exsisting Eulicdean distance and inner product. i install FALCONN through pip for python. thanks.@ludwigschmidt

[dorioliu]

I am sorry , i think i should be aware of the problem. squared Euclidean distance is proportional to the cosine distance when using l2-normalized euclidean distance. thanks anyway. so squared Euclidean distance would be ok.

[MrYxJ]

I am sorry , i think i should be aware of the problem. squared Euclidean distance is proportional to the cosine distance when using l2-normalized euclidean distance. thanks anyway. so squared Euclidean distance would be ok.

I have also encountered this doubt, thank you for your question to help me. And more precisely, the inner product is also proportional to the cosine similarity after normalization. What's the difference between the inner product and the Euclidean distance after normalization?