HllAn implementation of HyperLogLog probabilistic cardinality estimator.
val make : error:float -> tCreate a new counter with error error rate. error should verify 0.0 < error && error < 1.0. 0.05 is a reasonable default.
Use estimate_memory to measure memory consumption and runtime of this function.
val add : t -> int64 -> unitadd t k counts item k in t.
k should be "random": it should be the output of some cryptographic hashing algorithm like SHA. It is not treated as an integer. This is key to getting proper results. No patterns should appear in the bits of the different items added.
Runtime is O(1).
Estimate the memory consumed in bytes by a counter with the specified error rate.
This ignores the constant overhead of the OCaml representation, around two words. It is a bytes of estimate_memory ~error + 1 length.
val card : t -> floatGet the cardinality estimation. Defaults to HyperLogLog++.
val card_hll : t -> floatval card_hllpp : t -> floatmerge ~into:t0 t' has the same effect as adding all items added to t' to t0.
t0 and t' must have been constructed with the same error rate!
val clear : t -> unitReset counter to 0.
The following algorithm provide a reasonable hashing function for integers, if you want to feed the HLL with "normal" integers.
val to_string : t -> stringReturns a string with the current state stored.
val of_string : string -> tRestore a HLL saved with to_string.
of_string (to_string t) is functionnally equivalent to copy t, except a bit more expensive.
It can raise Invalid_argument if the string provided was not saved by to_string.