davidgwking/hll

a hyperloglog implementation for node

hll

hll implements HyperLogLog, a near-optimal distinct value (cardinality) estimator.

HyperLogLog boasts an easily derivable memory footprint and known standard error. As you increase the algorithm's memory footprint, the standard error of estimation results decreases dramatically.

This implementation accepts bit sample sizes that fall within the range [4, 12]. This allows you to customize the algorithm's standard error between 1.625% and 26%. Each register is an element of the Buffer class.

Bit Sample Size (b)	Number of Registers (m=2^b)	Standard Error (σ=1.04/√m)
4	16	26%
5	32	18.385%
6	64	13%
7	128	9.192%
8	256	6.5%
9	512	4.596%
10	1024	3.25%
11	2048	2.298%
12	4096	1.625%

Estimated values are expected to be Normally distributed and to fall within σ, 2σ, and 3σ of the exact count 65%, 95%, and 99% of the time, respectively.

This implementation uses MurmurHash3, a fast, non-cryptographic hash function and supports both 32-bit and 128-bit digest variants. The 32-bit MurmurHash3 variant is available for those that prefer performance over accuracy.

Usage

> var hll = require('hll');

// initialize a new hyperloglog data structure
> var h = hll();

// check out your standard error
> h.standardError
0.01625

// insert some values
> ['1', '2', '3', '4', '1', '2'].forEach(h.insert);

// crunch the numbers
h.estimate();
> 4

// then insert some more numbers and crunch them again!

API Specification

hll(opts)

Returns a new instance of a HyperLogLog data structure.

Option	Definition	Default
bitSampleSize	The number of bits that shall be sampled when determining register index. Integers that fall within the range [4, 12].	12
digestSize	The bit size of each input's MurmurHash3 digest. Must be one of {32, 128}.	128

If unacceptable input is provided, a RangeError is thrown.

Data Structure Methods

myHll.insert(value)

Insert a value. This value must be a string. Returns a summary of the operation.

If unacceptable input is provided, a TypeError is thrown.

myHll.estimate()

Iterates over the data structure's registers and returns the estimated cardinality of the data set.

myHll.standardError

Fetch the data structure's known standard error.