Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Surprised nobody's mentioned Hyper Log Log (or its more recent variants). A probabilistic constant-space unique cardinality estimator (i.e. if I see a billion events, how many are unique?) that also supports set operations.


The underlying reasoning behind PCSA/HLL (that if you squint, you can use the zeros and ones from a hash as analogues to coin-flipping, and can use the probability of the rarest event seen to estimate the total number of events seen) is excellent.

Count-min-sketch is also beautifully simple.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: