Monday, September 26, 2011

Open Kaya

Short post this time.

On a my-friends-stood-me-up-to-go-out Saturday rage i wrote the code to run and compare different rating systems.

You can see some guidelines and the code here :

Anyone that desires to try out and implement a rating system and compare it to others can do so. I am contacting some mathematicians and programmers that enjoy the rating systems problem, and if anyone here wants to get involved he can. Im available for questions and even aid.
If you know someone that would be interested in doing such a thing, tell them! the more rating systems we have, the better.

By the time we lanuch i will implement a simple Elo based system which will be sufficient considering we wont have many players, but afterwards i want to make the best community-oriented decision on which rating system will be used in Kaya.

Tell your geek friends!


  1. I've had a fair experience with attempting to implement different rating systems, and settled for a long time on a relatively straightforward ELO system for OGS. An advantage of ELO is that it's simple, and caters for rating differences in an obvious albeit crude way. The disadvantage is that, even if you allow people to pick their starting rating, it quickly evolves to contain a range that is dependent on the size of the player base (which is why, for example, FIDE ELO ratings in chess are steadily expanding at the top end, going from 3-4 2700+ players 15 years ago to around 100+ now).

    With a small player base, the effect will be that your ratings will steadily compress (so high rated players will fall and low rated players will rise), although this will be somewhat mitigated by the variations in ranks meaning that the majority of results will probably be wins for the stronger player.

    I would recommend if you are planning on working on something at launch that you have a way of handling rating volatility (so a player that has played few games or not played for a long time has a bigger rating variance than someone who plays a bunch of games every day). To be honest, you can mathematically hack this into ELO anyway, but I think you'd be better placed to adopt a simplistic Glicko/Glicko2 system from the offset, as I now wish I'd done with OGS.

    Just my 2p from someone that's been there before :)


  2. Hey topazg. Check out the repository :) Yoyoma made a Glicko implementation, which is most likely the first rating system we will use.

    Thanks for the tip.