mirrors/valkey
1
0
Fork 0
mirror of http://github.com/valkey-io/valkey synced 2024-11-23 20:00:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

HyperLogLog: use LINEARCOUNTING up to 3m.

Browse Source 
The HyperLogLog original paper suggests using LINEARCOUNTING for
cardinalities < 2.5m, however for P=14 the median / max error
curves show that a value of '3' is the best pick for m = 16384.
This commit is contained in:
antirez 2014-03-31 09:27:33 +02:00
parent 7f9d289e10
commit 60e60f4ee0
1 changed files with 11 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
src/hyperloglog.c
View File

@ -339,7 +339,7 @@ int hllAdd(uint8_t *registers, unsigned char *ele, size_t elesize) {
uint64_t hllCount(uint8_t *registers) {
double m = REDIS_HLL_REGISTERS;
double alpha = 0.7213/(1+1.079/m);
double E = 0;
double E = 0, linearcounting_factor;
int ez = 0; /* Number of registers equal to 0. */
int j;
@ -407,8 +407,16 @@ uint64_t hllCount(uint8_t *registers) {
/* Muliply the inverse of E for alpha_m * m^2 to have the raw estimate. */
E = (1/E)*alpha*m*m;
/* Apply corrections for small cardinalities. */
if (E < m*2.5 && ez != 0) {
/* Use the LINEARCOUNTING algorithm for small cardinalities. Note that
* the HyperLogLog paper suggests using this correction for E < m*2.5
* while we are using it for E < m*3 since this was verified to have
* better median / max error rate in the 40000 - 50000 cardinality
* interval when P * is 14 (m = 16k).
*
* However for other values of P we resort to the paper's value of 2.5
* since no test was performed for other values. */
linearcounting_factor = (m == 16384) ? 3 : 2.5;
if (E < m*linearcounting_factor && ez != 0) {
E = m*log(m/ez); /* LINEARCOUNTING() */
}
/* We don't apply the correction for E > 1/30 of 2^32 since we use