A successor for Trueskill? Experts assemble!

[Brutus5000]

Hi folks,

I was looking around the current state around implementations of the Trueskill algorithm. While doing that I stumbled over an interesting project that evolved Trueskill : openskill.me

It's published alongside on its own scientific paper:

Joshy, V., (2024). OpenSkill: A faster asymmetric multi-team, multiplayer rating system.
Journal of Open Source Software, 9(93), 5901, https://doi.org/10.21105/joss.05901

With some excerpts:

OpenSkill boasts several advantages over implementations of proprietary models like TrueSkill. Notably, it delivers faster rating updates, with 3 times the performance of the popular Python open-source implementation of TrueSkill as seen in Lee (2018). OpenSkill also includes five distinct models, each with its unique characteristics and tradeoffs. While all the models are general purpose, the recommended model for most use cases is Plackett-Luce. This model extends the regular Plackett-Luce as described in Guiver & Snelson (2009) by incorporating variance parameters to account for the probability that a certain team is the winner among a set of competing teams.

and most importantly

OpenSkill does indeed support time decay, an important aspect of maintaining an accurate skill rating system. Over time, a player’s skill can decrease due to inactivity; our library allows users to adjust the sigma value accordingly. This feature ensures that our library maintains its adaptability and relevance even when faced with variable player engagement levels.

My statistics knowledge is not even partially capable of understanding the differences, so I'd like to ask you, the experts, to evaluate if this is something FAF could use over Trueskill.

[IndexLibrorum]

It mentions that OpenSkill is faster than TrueSkill. Was speed at any point an issue? What kind of times are we talking about for Trueskill?
If they say 3x the performance, but they're talking about bringing a total time of 0.1 second down to 0.03, I'm unsure how relevant this is. The paper doesn't really mention specific data, which is kinda sloppy.

Rating decay has been much requested and would be fairly valuable, I think.

[snoog]

Out of curiosity, what's the likelihood it could be implemented alongside the current Trueskill implementation but not actually used? So that people can see it, get used to it, compare it, etc before it is actually used.

[rampeer]

This post is deleted!

[Brutus5000]

Our system supports (in theory) running multiple rating systems in parallel - at least the database supports it. I don't know if we utilize it yet or if we would run into since unexpected issues.

Calculation speed is not an issue for current operations as we don't have many games per second. It might become relevant if we consider backtracking / recalculating existing games to get current ratings based on new algorithms or parameters.

[magge]

I can confirm that rating degradation over time has come up repeatedly in various discussions in the last years, and it is always a hot topic. Especially for players who return after months/years of inactivity, often have mixed feelings or are even frustrated to "perform at their last known rating" with the knowledge they will not keep up, until they lost enough matches to reflect their new rating

Having such a system would make it less punishing for returning players, and the players who are playing with them in custom matches. When it would be possible to have both rating system in place, it would also give us the opportunity to test the waters first and get feedback, before fully committing. (IF OpenSkill is truly a better alternate)

[nullptr]

https://en.wikipedia.org/wiki/Glicko_rating_system