there is no declining games in the ladder.. and if you could decline people lower then you, no one would have games.. as everyone would either be declined, or declining...

Has anything about the ladder changed since this thread?

Here are some observations about the ladder ive made, please correct me if any are wrong.

1. In chess a 2000 rated player beats a 1600 player a very high % of the time so the huge change in ratings when the 1600 wins is justified. In warlight a 2000 does not beat a 1600 rated player close to the same % as chess. Mainly due to the rock/paper/scissor and luck aspects of warlight that are not present in chess.

This means you want to play players rated higher than you, so how do you do that? If youre at the top of the ladder you want to play as few games as possible. If youre at the bottom you want to play as many games as possible.

So its safe to assume if youre in the top 20 you are punished for being active and rewarded for playing few games very slowly.

Ill use heyheu as an example the average rating of his opponents is 1722. Way lower than all other top 10 players except sheriff another active player. If you take out all of hey's games against players rated lower than 1700 you bring his average up to par with the other top 10 players. hey went 45-4 vs these players take all those games out and his rating goes up 21 pts.

2. Top players rarely get to play each other, partly due to the disadvantages of playing more games. Some players play 20 games some play 140 this huge gap causes a problem with the matching system.

The solution

An adjustment to the rating system so youre not punished for being active.

A bias in the matching system to give you someone ranked closer to you more often, but not over and over. Something like: Players within 10% of your rank receives a minus 1 to their games played, so if they are 1 game over the median they are not eliminated.

from wiki
"Then, we count how many times you've played each remaining opponent, and take the median of this data set. We eliminate any opponents that we've played more than this median, since the ladder wants to simultaneously try and pair you with unplayed (or fewer played) opponents as much as possible"

Imo, being top of the 1v1 ladder is mainly an exercise in not losing to weaker players - sure I have to beat top 10 players a decent fraction of the time, BUT the main thing is to avoid taking the big rating hits associated with losing to a low-ranked player.

Maybe Fizzer could correct the theoretical probability distribution used in the rating algorithm? - I would guess that it needs much fatter tails compared to chess - can you compare the 'predicted' win fraction for each rating difference, against the actually observed ones? I would guess that they are very different. Then you could take the observed one, smooth it a bit, and use it instead?
A fair bit of work for fizzer I guess.

Btw, I am trying not to play too many games atm, so I would be just playing 2 games at a time, even if it were not the best strategy for me.


Also, maybe Fizzer could give an option button:
(1) [pair me mainly with similar ranked players]
(2) [pair me with first available players]

I am pretty much sure that the problem would automatically vanish if we had 1000 instead of ~100 players on the ladder... With only ~10 guys at a time that have 2approximately 2000 rating there is no other possibility to improve it without hampering the possibility to go up for other players... Unless we would want separate rules fir #1 (I personally wouldn't).

i thought vendetta was gunff. but now he talks like szewen. who is vendetta?

aside from that, any system that has had unknown and me on top of the 1v1 and 2v2 ladders the last few months is a good system!

I dont think more data or more players will solve this problem. We have data. Players with a rating of 1950+ beat players rated 1100-1700 90% of the time. HHH beats players rated 1100-1700 91.7% of the time. He is overachieving against these players so if you remove those games his rating should go down, but it actually goes up significantly.

If everyone played the same number of games, every ones opponents average ratings would be similar. So its not really a high rated vs low rated issue. Its an active/fast vs inactive/slow issue.

I dont know how difficult it would be to adjust the elo formula to correct this problem. Here is another approach that may be easier. Every time you play a game you get a .027% bonus to your rating-1000. A 2000 rated player would get a .27 rating bonus for each game played((2000-1000)*.027%=.27). At the end of the day the total bonus handed out is added up then deducted from the players ratings based on their rating compared to the total ratings. If 40 games were played about 14 points would have been handed out that day. if you have a 2000 rating and the total ratings on the ladder is 536,000 you would receive a .0819 rating deduction on that day. You would need to complete 1 game every 3.3 days to break even. ((2000-1000)/(536,000-365*1000)*14=.0819)

This does not give you an advantage if you play a lot of games quickly it just corrects the penalty you currently receive for playing a lot of games quickly, It also corrects the opposite problem for the people at the bottom.