All probabilistic structures are fascinating to me. I'm most enamored with HyperLogLog - estimating the number of distinct elements in extremely large sets of data without having to build a set of seen elements - but the entire field is pretty awesome.
The canonical use case for HLL is counting unique visitors. But any large dataset where you only need a good approximation is a valid target :)
The canonical use case for HLL is counting unique visitors. But any large dataset where you only need a good approximation is a valid target :)