Fast approximate knn using random projections

Hi, C++ implementation of approximate nearest knn which doesn't suffer from problem of curse of dimentionality whilst giving theoretical bounds on approximation.