TrueSkill 2

[yovanoc]

Hello, thank you for doing this in TS !

Have you think to do the TrueSkill 2 implementation too?

[hakatashi]

I read the paper of TrueSkill2, and IMHO it is not something that should be implemented as a library. In this paper they proposed the way to make changes to the TrueSkill model for each game, and the way to adjust the parameters automatically using a large dataset of matchings.

They presented the several ways to modify the TrueSkill model using metrics. They include the inference based on the kill/death rate, squad size, and quit penalty. All of these changes are modelled based on assumptions specific to the game Halo 5. We can provide these model changes and the method to train the parameters as a library, but it will be not very useful. You should design your own rating model for your game to achieve meaningful results.

[yovanoc]

A lot of games have kill, death, quit, match duration, etc, we always can make this a library

[Wulfheart]

Are there any internal discussions about this in the background?

[yovanoc]

I don't think they will implement this one day

[Wulfheart]

Thanks for the clarification.