0.1.0-PRE29
- Switched to less-naive implementation of multiprobe L2 LSH. Specifically, uses algorithm 1 from Qin, et. al. to generate
perturbation sets lazily at query time instead of generating them exhaustively. This does not use the estimated
scoring optimization from that paper. - Performance optimizations for approximate queries. Specifically using a faster sorting method to sort the hashes before
retrieving matching docs from the shard.