News for hackers. Stuff that matters.

A comment on Show HN: An improved version of HyperLogLog. Parent comment

Re:Show HN: An improved version of HyperLogLog by jbapple on Monday June 26, 2017 @10:57AM (#14637283)
HLL is a space/accuracy tradeoff. If you trim HLL to use the same space as if you were using HLL+TailCut but use the Ertl method for estimating cardinalities, which is more accurate: Ertl with a smaller `m` or HLL+TailCut with a larger `m`?
[ Parent | on HN ]
Re:Show HN: An improved version of HyperLogLog by seiflotfy on Monday June 26, 2017 @11:08AM (#14637380)
I think the latter. I will implement this next week. Nice call :D
[ Parent | on HN ]
Re:Show HN: An improved version of HyperLogLog by jbapple on Monday June 26, 2017 @11:13AM (#14637420)
I suspect you are right that HLL+TC will be more accurate. Based on Figure 5 vs. Figure 1 in the Ertl paper, I think Ert's method improves HLL estimation for low cardialities and for very high cardinalities, but not for the region where HLL is already at its most accurate.
[ Parent | on HN ]