An implementation of the Glicko-2 rating algorithm written from scratch, with
the goal being to provide less conveniences than glicko2js
in favor of
faster execution time.
Basically glicko2-lite
does not keep track of anything for you. All it does
is calculate new ratings. Once I reached the point where I was trying to
calculate ratings for tens of thousands of players across tens of thousands of
matches, glicko2js
only got slower and slower as more matches were added. For
my case at least, I was already saving rating information to a database so I had
no need for the Player
classes that glicko2js
provides; all I really needed
was the actual calculation of new ratings itself.
Let's look at some benchmarks. First, the simplest case: create two players, play a match between them, and calculate the new rating, deviation, and volatility of both players:
$ node bench/single.js
Starting...
3 tests completed.
glicko2-lite x 1,719,122 ops/sec ±0.60% (92 runs sampled)
glicko2 x 1,406,286 ops/sec ±0.26% (94 runs sampled)
glicko2.ts x 755,440 ops/sec ±0.16% (100 runs sampled)
Now something larger: let's create 1,000 players, and have the first player play a match against each of the other 999 players. After each individual match, calculate the new ratings of the first player and their opponent:
$ node bench/many.js
Starting...
3 tests completed.
glicko2-lite x 1,457 ops/sec ±0.39% (94 runs sampled)
glicko2 x 33.92 ops/sec ±1.45% (59 runs sampled)
glicko2.ts x 0.12 ops/sec ±1.52% (5 runs sampled)
Some might say this is unfair, since new ratings are being calculated piecemeal after each match instead of in bulk after a series, or rating period, of matches. Let's try the same benchmark but only calculating the new ratings at the end of all of the matches:
$ node bench/batch.js
Starting...
3 tests completed.
glicko2-lite x 21,306 ops/sec ±1.20% (92 runs sampled)
glicko2 x 2,304 ops/sec ±0.78% (91 runs sampled)
glicko2.ts x 105 ops/sec ±0.72% (77 runs sampled)
const glicko2 = require('glicko2-lite');
// player A: 1500 rating, 350 rating deviation, and 0.06 volatility
// player B: 2000 rating, 70 rating deviation
// A loses to B
// new rating, rating deviation, and volatility:
glicko2(1500, 350, 0.06, [[2000, 70, 0]])
// => {
// => rating: 1467.5878493169462,
// => rd: 318.6617548537152,
// => vol: 0.059999457650202655
// => }
$ npm install glicko2-lite
const glicko2 = require('glicko2-lite');
rating
(Number)rd
(Number)vol
(Number)matches
(Array)options
(Object)