This is a JavaScript/Node.js library for computing running (or rolling) statistics for one set of values.

Any up to date nodejs compatible runtime should properly work. (bun / nodejs).

Node 18.17.0 or more recent is required (specially for tests), or an up to date bun runtime.

Can be found on :

• npm
• github

const rollingArray = require("@fadoli/node-fast-running-stats");

// We will do stats on a maximum of 10 values (then last value override oldest one)
const myStats = new StatsArray(10);

myStats.append(0).getStats();
// { n: 1, min: 0, max: 0, sum: 0, mean: 0, variance: 0, standard_deviation: 0 }
myStats.append(1).getStats();
// { n: 2, min: 0, max: 1, sum: 1, mean: 0.5, variance: 0.25, standard_deviation: 0.5 }

you can run this on your machine npm run bench

For each size we run the computation 100 times. We add twice the size one per one and compute the stats each time. This means for size 10 we will compute stats on 1,2,3,4,5,6,7,8,9,10 entries at first, then on 10 entries 10 times.

simple rollingArray implem, size = 10: 2.903ms
fastStats implem, size = 10: 3.149ms

simple rollingArray implem, size = 100: 22.486ms
fastStats implem, size = 100: 3.946ms

simple rollingArray implem, size = 1000: 1.318s
fastStats implem, size = 1000: 10.619ms

simple rollingArray implem, size = 2000: 4.882s
fastStats implem, size = 2000: 12.948ms

simple rollingArray implem, size = 3000: 11.237s
fastStats implem, size = 3000: 21.145ms

Here is the difference in output when both methods are compared :

{
  simple: {
    n: 3000,
    min: 0.00009501596644567734,
    max: 0.9999407084813299,
    sum: 1514.788964278203,
    mean: 0.504929654759401,
    variance: 0.08372214898153563,
    standard_deviation: 0.28934779933764077
  },
  fastStats: {
    n: 3000,
    min: 0.00009501596644567734,
    max: 0.9999407084813299,
    sum: 1514.7889642781986,
    mean: 0.5049296547593995,
    variance: 0.08372328892386063,
    standard_deviation: 0.28934976917886185
  }
}

This module's approach creates more errors related to the float precision errors (more computation are done and re-used and as such float precision errors are added as times goes on). The result is recomputed after a certain amount of times (2 times the size of the array, or at least 25k) to prevent it from causing big errors.