@blackyps said in Upcoming rating changes: Gaining global from playing matchmaker:
So basically, we now know that one or both ratings are horifically wrong, but we do NOT know which it is, nevertheless we always assume ladder is right and global is wrong.
We assume that ladder is right, because you gain ladder rating in a controlled environment that cannot be manipulated by the player, in contrast to global.
So we effectively do not trust ladder rating? Why do we still bother using it then? It is not needed (just remove it and use interpolated ladder rating instead).
But this is only the first problem. In addition, instead of taking into account global rating of both players when adjusting global rating of a player, it also replaces global rating with ladder rating when calculating the new global rating. This is nuts because it absolutely destroys the way in which deviation is supposed to affect the Trueskill algorithm.
I don't understand what you mean by this. It never changes only the mean, when inserting global rating to calculate the adjustment, it also uses the global deviation of course.
Every mean rating in a Trueskill context is only relative to other ratings inside the same context (500 mean in ladder is not the same as 500 in custom).
And in every match Trueskill tests its own (relative!) prediction of the player s skill to adjust its own knowledge. Inserting a prediction into the match that this Trueskill context never actually made (by replacing ONLY the opponents rating with their ladder rating) is bonkers! It makes no sense! The result of that calculation does not mean anything!
You COULD actually make use of ladder games by simply pretending that each ladder game was played both in ladder AND as a custom game. That would make sense, because now you are replacing ALL players rating with their global rating and then adjusting ALL players (whose relative ratings now make sense because all those ratings were produced in the same system). But that is not being proposed!
It would be the only thing that would make sense if you are hellbent on changing global rating through ladder games. Not sure if it can be exploited.
If we decide that global is so broken that we just fuck with it in any way we want, why do we not do the proper thing and remove global, rating and instead show an accumulation of various ladder ratings for custom games that cannot be affected by custom games?
We still need global so the communities that play on one map only and barely any matchmaker can keep a system to balance their games. What this change is trying to address is the problem of new players not being able to get into custom games, because they have no rating. In the future they can play matchmaker games and global will slowly float to their highest matchmaker rating.
This change does not intend to address the inconsistencies with global rating.
I hope you plan to remove global at some point then because if you do it as it is currently planned you are no longer doing things that make sense for having a predictable well-behaved rating system, you are deep inside band-aid mode already.
Some years ago i spent months in the forum on every rating related thread trying to convince people to reduce starting mean for new players. The conclusion was that we have no real way so simulate the effects of that change so wed rather not change anything.
That's not entirely true, because the rating interpolation was introduced. That was without simulating the effects and that is fine, because demanding every possible change be completely simulated beforehand seems over the top to me.
From what i understand, the rating interpolation does NOT interfere with the rating system in anyway? It is only used to determine which players are matched against each other. Matchmaking really is not mathematically connected to Trueskill! You can decide to let -100 players play against 2000 player all day long, Trueskill would not care!
Trueskill cares about having a prediction of a player skill that was built up in a non-changing context together with predictions of other players in that same context, and improving its own prediction of the player skills relative to each other. It is not really concerned with who is matched against who.
So if i am correct this is the first time we really try to fuck with Trueskill itself!
We don't use the interpolation because as I explained above we still need global to function as before for some players.
Interpolation would include global as one of the ratings being interpolated. Players that have no ladder games would have their global rating shown. I have no idea why you think interpolation would not work.
Also making this interpolation in lobby only leads to problems when you want to look up a players global rating in the client.
I don't see our way of adjusting global as such an egregious manipulation that we should at all costs avoid persisting it in the database.
Just use that exat same calculation to show people their "global" rating? Its not rocket science. Its just averaging by weights. Leave out the normalization step for all i care, its not essential.
The proposed change fucks with global in a way we have never done before, and i have proposed the interpolation solution for this exact problem before (i can try digging it up if you want), so i really do not see why we are know suddenly making this change when a better much less invasive solution could have been implemented.
And i think the argument in my last edit is kinda important unless death of global rating is planned:
The proposed change would also make simulating a change to global rating parameters with past data much harder i think, because global game outcomes are no longer enough to redo all global rating calculations.
Also, how about we (as a community) discuss such intrusive changes BEFORE they are implemented?