trueskill
trueskill copied to clipboard
sublee
Trueskill 2.0
The Trueskill 2.0 paper came out a few months ago. It appears to be taking in game per players stats into account and includes quit penalties. There claim is much higher predictive power. It however doesn't appear to come with any source code. Is it possible to infer that steps necessary to implement?
Probably possible. But I think if there was no Jeff Moser's paper, I could not implement this library.
Bumping this, just wondering if anyone has any knowledge of current open source implementations
I've just read the paper diagonally. It goes into much more detail of modeling game dynamics to approximate skill compared to the out-of-the-box style approach that this library has at the moment. It could be interesting to implement it, but I guess it depends on what this project aims for.
Is it desirable to implement TrueSkill 2.0? It doesn't seem to be a straight forward upgrade compared to this version.
@FirefoxMetzger It would be great if this library implements both TrueSkill 1.0 and 2.0 with convenient API. Do you attempt to implement the 2.0?
I can neither confirm nor deny this. I'm currently implementing a different paper that creates a bot for Hex, hence my need for a ranking system like trueskill to compare performance. This takes priority; afterward, I will be a little freer.
What exactly would "attempt to implement the 2.0" entail and how would you make both versions live in the same project? I read your code on a class/method level, and I, for example, saw that you are running your own matrix class. Is there a specific requirement not to use numpy or other libraries? I would need more information on the requirements from this project, and also read the linked paper more thoroughly to make any decision.
I misunderstood that you want to contribute this project by TrueSkill 2.0 implementation and you are already trying it. I don't like to make you feel any pressure. If you have interest in TrueSkill 2.0 implementation, to create trueskill2 would be a better decision.
Anyways, when I implemented this project, I chose to reduce dependencies. But now, I agree that numpy is better to handle matrices.
Hey, was anyone able to implement trueskill 2 (or found a library) that does? Would be of great help.
