Let’s not stop with the basic ones, how crazy can we get? How about HyperLogLog. HyperLogLog is based on a research paper from Google. It implements K-minimum value, bit observable patterns, stochastic averaging, and harmonic averaging. If you’re like me and had to google pretty much every one of those things the simple way to explain it: HyperLogLog is great for approximate distincts which can be stored in a really small disk space, then composed over time to find intersections/unions of uniques across various buckets. If you’re building a web analytics tool or an ad network HyperLogLog may end up being your best friend. You’ve also got more common approximation datatypes like TopN which is great for building leaderboards.