<< Back to Ladder Forum   Search

Posts 1 - 14 of 14   
Flaw in ratings: 12/9/2013 21:14:05

TeddyFSB 
Level 60
Report
So, it is possible to win a game and drop in ratings. Before my last win on seasonal ladder (against Dreamer):
Rank Name                           Elo    +    - games score oppo.
   1 TeddyFSB                      2886  214  214    10   90%  2584
   2 abr                           2884  256  256     8  100%  2526
   3 Mannerheim                    2857  235  235     9   89%  2505

After that win:
Rank Name                           Elo    +    - games score oppo.
   1 abr                           2884  256  256     8  100%  2527
   2 TeddyFSB                      2879  210  210    11   91%  2563
   3 Mannerheim                    2855  235  235     9   89%  2504

So a win cost me 7 points. Not that I mind, I think abr should be ranked above me at 8:0, but it is an interesting flaw.
Flaw in ratings: 12/9/2013 21:16:23


ChrisCMU 
Level 60
Report
This has been posted somewhere before, but yes you can win and drop. Depends on who you beat (and who they beat).
Flaw in ratings: 12/9/2013 21:19:00


x 
Level 58
Report
Isn't that because your previous 10 opponents had lost games during that period, making your wins against them less valuable? Rather than getting negative points for a win?

Edited 12/9/2013 21:19:27
Flaw in ratings: 12/9/2013 21:44:00


[REGL] Pooh 
Level 60
Report
What would it have been if you lost?
Flaw in ratings: 12/9/2013 22:09:42

TeddyFSB 
Level 60
Report
Isn't that because your previous 10 opponents had lost games during that period, making your wins against them less valuable?

No, I only removed that 1 win by removing line
addresult 171 59 2
from the log:
http://data.warlight.net/Data/BayeseloLog4011.txt

http://wiki.warlight.net/index.php/Ladder_Ranks_and_Ratings

By the way, correct offset = 2800-65*(20-N), where N is the number of games that have started. In this case N=13, and offset = 2345.

advantage is set to 0, although getting 1st pick is very valuable on these settings. 1st pick results in a win 57% of the time. That corresponds to advantage ~ 50 points. That's larger than the default chess advantage of 32.8 for playing white.

Although using advantage = 50, the top of the ratings is largely unchanged:
Rank Name                           Elo    +    - games score oppo.
   1 abr                           2887  258  258     8  100%  2523
   2 TeddyFSB                      2886  213  213    11   91%  2568
   3 Mannerheim                    2858  234  234     9   89%  2508
   4 Krzychu                       2799  219  219    10   90%  2470
   5 RuthlessBastard               2799  241  241     8   88%  2517
   6 Frankdeslimsteapex            2769  246  246     7   86%  2516

What would it have been if you lost?

I would drop 87 points and add up in 4th place.
Flaw in ratings: 12/10/2013 00:18:03

Jehovah 
Level 58
Report
I believe Fizzer said somewhere that it is impossible tolose points because of a win. From what i know, here ould be the reasons

1)Someone you lost to lost points
2)Someone you beat lost points

If you think about it, i think any game in the ladder can cuase an effect.

Say...

Player C beats Player A
2 days later....
Player B beats Player A


Say Player A lost to Player B, a very low ranked player. this means Player A will lose a lot of points. Player's C win over Player A 2 days ago, becomes less valuable. Player C now loses points as well.

Add more players...

Player C beats Player A
2 days later....
Player B beats Player A
2 days later...
Player E beats Player B
2 days later..
Player D beats Player E

Player C gains points from beating player A. Later Player B beats Player A, So Player C's win becomes less valuable and lose points. Another 2 days later player E beats Player B. This means Player B's rating decreases. This causes Player A's rating to decrease, this causing Player C's win the first day even less valuable. Player C loses points. Player D beats Player E. This means Player E and Player B's rating will drcreases. If Player B rating decreases, so will Player A. If Player A's rating decreases, Player C's win becomes less valuable and player C loses points

ta-da!

I think all games on ladder is connected in some way...
Want to win the ladder?

Simple.

Win everything.

Edited 12/10/2013 00:21:17
Flaw in ratings: 12/10/2013 00:36:16


Kenny • apex 
Level 59
Report
Teddy, because Dreamer lost to you, his rating directly drops and so the people who won vs him get less points, if you beat anyone Dreamer lost to, you may end up losing more points than winning due to the exchange. Also consider anyone who Dreamer BEAT loses points too, so anyone who Dreamer beat... it gets really complicated. The direct result is that you were awarded points, the indirect result is that you lost more points due to the wave that the game put out.

Also, consider the fact that Mannerheim's and abr's scores are LESS ACCURATE than your score due to less games.

Edited 12/10/2013 00:37:49
Flaw in ratings: 12/10/2013 07:54:20


[WM] Gnuffone 
Level 60
Report
@Teddy, all here will tell you that you lost point because opponent you already beat lost some game...
I have enough ladder experience for say, isn't true.
My last experience in ladder 2v2 was enough clear, i lose 35 points for beat guys with 1400 point only...
so yeah, if you are matched with bad guys with bad rating and you are winning, stall the game for don't lose point!!!!!

Edited 12/10/2013 11:07:11
Flaw in ratings: 12/10/2013 10:36:06

Yeon 
Level 61
Report
Gnuffone speaks the truth. Teddy lost the points because he played and won that game. Of course Teddy already knows that, but for the benefit of other people reading this: Teddy and Gnuffone are right.
Flaw in ratings: 12/10/2013 16:47:12


ChrisCMU 
Level 60
Report
It does seem silly to have a win over a bad player end up punishing you (sure, it is better than a loss, but still not beneficial). I am no math wiz, so not sure best way to fix it. you could tighten standard deviation on matchups, but then when someone has 5 games at once going, you are less likely to get a proper matchup without those outliers to test someone on a hot streak (could just stall a loss).
Flaw in ratings: 12/10/2013 23:17:36


Kenny • apex 
Level 59
Report
Here's a list of everyone affected by Dreamer's rating directly:

TeddyFSB
AA konglaide
Lao Tzu
BeerFridays
Krzychu
nusoos
Eagleblast
monkman

Then comes the first wave of people indirectly affected.. which is everyone who has played anyone on that list.

Then everyone who has played someone who has played someone on that list is also affected.

These 'waves' eventually pile up, and by beating Dreamer, you somehow affected the ratings to the point you received negative points from the indirectly affected people.
Flaw in ratings: 12/11/2013 08:24:33


{rp} GeneralGror
Level 58
Report
Surely it would be better to take the points at the time, like you are shown your current rank and rating in a new ladder game. The way things are, even if you beat someone, you can suffer from their losses. Doesn't seem like the fairest approach to me.

Edit: Seconding Chris.

Edited 12/11/2013 08:25:05
Flaw in ratings: 12/11/2013 22:53:23

TeddyFSB 
Level 60
Report
After playing with numbers, this effect makes sense to me. Here is a simple model.
Elo formula is pwin = 1/(1+10^(-rdiff/400)), where rdiff is the ratings difference between 2 players.

So knowing probability of winning over a rated opponent, our rating should be:
rating = opp.rating - LOG(1/pwin-1)*400.

Here is a copy of a spreadsheet that shows rating dropping after a win:
N games	10		new opp	1200
N wins	9			
pwin	0.9		pwin    0.909
sos	1500		sos	1472.7
rating	1882		rating	1873


sos is average strength of schedule. We start from winning 9 out of 10 games, and our average opponent is rated at 1500. Our rating should be 1882.

We add a win over a 1200-rated opponent. pwin goes up slightly to 10/11~91%, while average opponent rating goes down slightly to ~1473. And now the ELO formula says that given those two numbers, our rating should be 1873, 9 points lower than before. Drop in the strength of schedules has a bigger effect than increase in winning percentage.

So it's not an effect of propagating wins and losses through the rest of the field. This is directly the result of using the logistic function in the ELO formula, which has this flaw in it of dealing with opponents at very different ratings.
Flaw in ratings: 12/12/2013 15:18:45

Yeon 
Level 61
Report
Teddy: I remember from chess organization that ELO is never counted as more than 350 different from the opponent. So if a 2000 rated player defeats a 1400 rated player, he is rated as defeating a 1650 player, while the 1400 rated player is rated as losing to a 1750 player.
Posts 1 - 14 of 14