问题
I have a very specific question. I work on a project, were I need to find nearest neighbours (k and near). As I dont need the excat ones and want to be able to extend to high dimensions, I focused on LSH.
My data has a distance that is a metric, but non euclidean. I found many ways for vector space with euclidean metric (e.g. the p stable distribution), binary coding(via projections) or string based.
What I am searching are papers that present a LSH template for an arbitrary metric. Does anyone has some refernece to papers?
Thanks in advance Dan
回答1:
What you are looking for is quite new: I think this paper may help http://www.aaai.org/ocs/index.php/aaai/aaai10/paper/download/1839/2032
It suggests strategies for non-metric data, which is even worse than having a non-euclidean case.
来源：https://stackoverflow.com/questions/17875885/local-sensitive-hashing-using-a-arbitray-non-euclidean-metric