<< Back to Ladder Forum   Search

Posts 1 - 19 of 19   
1v1 ladder is bad at rating players, who knew?: 1/5/2014 07:37:33


professor dead piggy 
Level 59
Report
Usually in an ELO rating system new players start out with a rating somewhere towards the lower-middle of the pack, playing against decentish opponents, and then are quickly sorted up or down based on their first ~5 games. This happens on chess.com, league of legends, starcraft etc (warlight's peers). However on WL you start riiiiight at the bottom, rated 0, playing against blowfly and dabird. This is inappropriate because a new ladder player is very often an experienced warlight player and could concievably be better than everyone else on the ladder at the time. Why are new ladder players rated as though this is their first game?

The negative effect this has is that 1.a potential #1's ladder run is torpedoed in their first couple of games by bad luck and theyre misrated for months thereafter because of a game that they would be too highly rated to ever get put in after their first few weeks. 2. The 15 game minimum to get ranked is even less accurate because ~5 of those games are trivially easy (assuming one gets decent luck in all of them) and do little to contribute to a players rating. Essentially this system exaggerated the luck factor inherent in each warlight game.

Its simple enough to avoid; one has to have a handful of unexpired games on their account when they start the ladder run. By the time theyre ranked the games have expired but they are allocated games with 1200~1800 rated players initially. This is how i and a few others got our #1 spots. It gives an advantage to people gaming the system and older players.

Fix it by starting new players off at ~1400.

Edited 1/5/2014 07:41:52
1v1 ladder is bad at rating players, who knew?: 1/5/2014 10:53:33

TeddyFSB 
Level 60
Report
I have a few disagreements.

1) I doubt that it's easier to reach #1 spot from a good initial ranking than from scratch. If you start from scratch, you typically need to go 14-1, with 7 games against bad players, and 8 games against good players. If you start from a good ranking, it might be sufficient to go 13-2 to reach #1, since you have a stronger average opponent rating. I think that winning 13 out of 15 against good players is a little more difficult than winning 7 out of 8.

2) Why would the system assume that a new player is good? That is usually not the case.

3) If I go with your assumption that the system makes it slightly more difficult for a new player to reach #1, that's a great thing! It's too easy to reach #1 from 15 games, with a little luck and a little selective stalling. So it's nice to find any feature that makes it less likely:)
1v1 ladder is bad at rating players, who knew?: 1/5/2014 11:36:58

Jehovah 
Level 58
Report
thats why its called the ladder right? you start at the very bottom, and climb your way up?
1v1 ladder is bad at rating players, who knew?: 1/5/2014 11:52:29


professor dead piggy 
Level 59
Report
If you are really really good your chance of beating a good player is ~80%, and your chance of beating a bad player is ~95%.

1) Your numbers are way off. If you lose even 1 of those 7 games against bad players then 14-1 wont cut it, you might not even make top 10. If there is a 5% chance of loss to a bad player due to luck and silly mistakes then 5*7=35. 35% chance of a potential number 1 being knocked out due to bad luck. JSA recently got number one with FAR more than 2/15 losses because he was only playing top 20 players.

2) I am saying assume a player new to the ladder is as good as a bit less than half the people already on the ladder (because that is infact true). I was and so were you. Just dont assume they suck, it stops them from being rated properly.

3) Haha this isnt a feature, the ladders job is to accurately rate players, and this is an example of it rating players inaccurately. Fuck it, fine, this is a good thing.
1v1 ladder is bad at rating players, who knew?: 1/5/2014 14:35:01

Qi 
Level 55
Report
I nominate dead piggy as ladder fixer upper. Fizzer should give dead piggy full powers to fix and upper. Here is the uservoice for my suggestion. Vote now! Save a child in Africa!

www.uservoice.com/deadpiggy
1v1 ladder is bad at rating players, who knew?: 1/5/2014 14:36:06

Qi 
Level 55
Report
Sorry for the double post.
1v1 ladder is bad at rating players, who knew?: 1/6/2014 21:21:21


Ⓖ. Ⓐrun 
Level 57
Report
It's not an ELO rating system is it? It isn't designed to be permanent, ranking players reliably by skill level. It's job isn't to "accurately rate players", I assume perposely by Fizzer in order to increase the appeal of it to more players (they have a chance to go top).

I think the ladder should be updated to be more accurate - either that or some form of ELO being introduced.

Edited 1/6/2014 21:25:02
1v1 ladder is bad at rating players, who knew?: 1/6/2014 22:35:27


ChrisCMU 
Level 60
Report
I think it is even more apparent on 2v2 ladder. Take Noobschool and I. We had a rating in the mid 1900s if you only counted our last game to expire. If we had joined in time, we would have gotten a bunch of games against top 10 teams. Instead we joined at a zero rating (he was currently with Frank at the time).

We made some mistakes in our first match in a LONG time and lost. So now all our games were against poorly rated teams. We had 5/6 games against teams with sub 1400 ratings before finally getting a match against a team over 1800 (which we won). Much of this is because Timinator/ultimaratio have been very slow against us (not sure if they are stalling or busy, currently one is on vacation). So we haven't had any real chance to get great matchups because we made a mistake in 1 game.
1v1 ladder is bad at rating players, who knew?: 1/7/2014 03:40:00

Jehovah 
Level 58
Report
^or perhaps you are just not very good, and dserve bad teams ;)

you lost two games, one gainst a team with 1500ish and one with a team with 1400ish, and you want good teams?

Edited 1/7/2014 03:46:34
1v1 ladder is bad at rating players, who knew?: 1/7/2014 07:12:06


professor dead piggy 
Level 59
Report
Arun, the ladder uses a bastardized Elo (not ELO. It is a name not an acronym) rating system.

Elo was designed to be used in games where luck is much less of a factor. The idea is that the difference in the ratings between two players is a predictor of the outcome of a match. A player whose rating is 100 points greater than his or her opponent's is expected to score 64%; if the difference is 200 points, then the expected score for the stronger player is 76%. In warlight we have people rated 2100 losing to people rated 1400 semi regularly. This is an example of the system not working as it should.

Also it was designed with the assumption that players do not start at a rating of 0.

Also it was designed with the understanding that the rating was inaccurate if it was based on too few games. Usually people are rated based on hundreds of games. In a population thats skill level is constantly changing all having ~30 games on their accounts the margin for error is hundreds of points.

Edited 1/7/2014 07:12:56
1v1 ladder is bad at rating players, who knew?: 1/7/2014 16:15:03


Ⓖ. Ⓐrun 
Level 57
Report
Cheers for the clarification.

How could the system be improved? No matter what rating system is used, the number of games for each player will generally always be low.
1v1 ladder is bad at rating players, who knew?: 1/7/2014 19:20:34


ChrisCMU 
Level 60
Report
@Brain-

Well, we were ranked 9th in our prior run, neither of us has gotten worse since then. We are not a 2,000 rated team, but also not a 1500 rated team either.

You are also looking at when the games resolved, and not when they started. Here are our opponents and when they were created:

11/22 - 1583 rating (lost on 12/4)
11/23 - 2020 rating (ongoing)
11/29 - 1231 rating (won on 12/16)
11/29 - 1185 rating (won on 12/21)
12/04 - 1292 rating (won on 12/10)
12/10 - 1426 rating (won on 12/16)
12/17 - 1478 rating (won on 12/23)
12/17 - 1364 rating (won on 12/24)
12/24 - 1745 rating (won on 12/27)

as you can see our first 4 matchups (all created before any win or loss) were 1 very good team, 1 very average team and 2 very bad teams. This is because the system rates us at a 0, not what we were before (even for matchup purposes). So the only real chance we have had to get good matchups is the game still in progress. Had we won the first game, we'd still be rated pretty poorly and still be getting 1500-1600 level matchups. Even as we won those other 7 games in a row, we still only got one team over 1500 after that. How exactly is this testing your skill level?

Sure, we lost two games out of 10 to teams rated around 1600. but the 7 wins over sub 1500 teams count for nothing (they are a punishment to have those matchups). I would much rather play all teams 1600+ and only win 6/10 than 8/10 with bad matchups.

Basically, if you lose your first match you might as well leave the ladder because you aren't going to get any real games for a long time.

Edited 1/7/2014 19:22:09
1v1 ladder is bad at rating players, who knew?: 1/7/2014 19:26:38


ChrisCMU 
Level 60
Report
@Arun-

It could be a lot better if when creating matchups, it assumed the team was the average rating on the ladder, not a 0. So it would match you up with teams about the average skill level on the ladder. From there you either prove you are better or worse, but at least you'd have some decent matchups to start out. In our case our average opponent rating was 1505, which would be in the high 30s on the ladder. I don't know how many teams were on the ladder in late November, but the average rating on the ladder right now is 1517, so not much different.

Ignoring our loss for a moment (say we won our first game), we'd still have zero shot at top 5 on the ladder even if we won 10 games in a row, because we started from 0 and got junk matchups. You can't improve your rating enough if it doesn't give you solid opponents.

This happens in the seasonal ladder too (which is a bigger problem because you have a limited number of games). If you don't get a decent matchup there, you'll never make up ground. Or you beat someone who is an elite player, and then they drop off the ladder the next day and surrender all their games. Well now your game is punishing you despite a win over a good opponent and you have no games to make up ground there with a set limit.

I suppose we should have just stalled our game we were losing, but I still think it wouldn't have gotten us very good matchups anyway without the 2000 rating game resolving. And as Dead Piggy said, then we'd be gaming the system, which is what it currently rewards. There is all kinds of stalling going on right now with some top teams because they know it works.

Edited 1/7/2014 19:35:00
1v1 ladder is bad at rating players, who knew?: 1/7/2014 23:55:52


[REGL] Pooh 
Level 60
Report
How does the Seasonal Ladder differ?

Looking at the games, I didn't play many players that were near my final ranking. I finished at (now) 3098, and I only played 5 players in the range of 3058 and up.

In those games, I'm 2-3 against those players.

Are matchups in the early season taking into account for previous season ranks?
1v1 ladder is bad at rating players, who knew?: 1/8/2014 04:13:11


ChrisCMU 
Level 60
Report
Seasonal ladder restricts number of games , so a lowsy matchup hurts even more
1v1 ladder is bad at rating players, who knew?: 1/8/2014 09:41:53


Math Wolf 
Level 63
Report
The main solution is TrueSkill with a small variance penalty over time.

As far as I am aware, there is no great solution for stalling, but a variance penalty in TrueSkill would give a rank advantage (not rating!) to players who play more (since stalling makes you play less games).

The problem is that adding this variance penalty may or may not be easy.

Additionally, TrueSkill (and Elo) are vulnerable to the order in which you finish games (L-W and W-L give a different result). It may be quite difficult to fix this, but it may actually be a feature in this case: L-W gives mostly a better rating than W-L so stalling would be counterproductive unless you have only losses going on.
1v1 ladder is bad at rating players, who knew?: 1/8/2014 12:19:26


professor dead piggy 
Level 59
Report
I didnt even know wolf read the forums, I am flattered that you graced my thread.
1v1 ladder is bad at rating players, who knew?: 1/8/2014 15:04:18


ChrisCMU 
Level 60
Report
Check this out:

http://warlight.net/LadderTeam?LadderTeamID=2448

Best win (ever) is over 28th ranked player, yet has a shot at #1
1v1 ladder is bad at rating players, who knew?: 1/9/2014 23:25:55


Ⓖ. Ⓐrun 
Level 57
Report
Lol I have a top 40 rating after about 5 wins :D
Posts 1 - 19 of 19