<< Back to General Forum   Search

Posts 1 - 30 of 40   1  2  Next >>   
Questions about rankings on the ladder: 4/11/2013 14:32:41


NoobSchool (AHoL) • apex 
Level 58
Report
Just a few questions about rankings on the ladder.

I understand that the more games you have on the ladder, generally the lower your average rating will be. Just how much is this? For example, I have over 100 unexpired games and I have always played 5 games at a time since I started the ladder in June of 2011. My average ranking for the last few months now has been about 1875. What would happen to this if I dropped off and waited until a bunch expired?

Where do you think my rating would (generally) go? If the ladder overestimates people who have less games, would this propel me higher? I don't particularly want to do this because I enjoy the feeling that I have worked hard for where I am, not just lucked my way higher by a compromised rating system.

Do you think that the players who are in the top 10-20 would have their ranking drop if they had more games? Would it be significant or insignificant in the long run? Would rankings generally stay the same with lower ratings or would everything change in the upper ranks?

A question which I've had on my mind for a while is about what players think about those who are highly ranked on the ladder. Is it respected that some people game the ladder to gain their #1? Would you rather see someone who hits #1 with 50+ games (such as HHH) or are you content seeing tons of players hitting #1 with 15 games then dropping off (like Beezlbub)?

If you are playing games on the ladder, do you fear playing people who are ranked higher with many games, or with few?


Long winded, but I've really been curious lately. Would really like to see what other people think about all this.
Questions about rankings on the ladder: 4/11/2013 14:48:15


[WM] Gnuffone 
Level 60
Report
Not necessarily dropped off and wait = when you join more point

I understand that the more games you have on the ladder, generally the lower your average rating will be


yeah, is true.
but not always. i finish 26 games that still count for my rating, and i still be 2nd.

you improve your rating if you don't lose more then 3/4 games, or you will have 1800/1900 again.

Is it respected that some people game the ladder to gain their #1? Would you rather see someone who hits #1 with 50+ games (such as HHH) or are you content seeing tons of players hitting #1 with 15 games then dropping off (like Beezlbub)?


is better see player with +35/30 game take 1st place, is too easy now take 1st place with 15 games. i think rating should be improved.

i don't take care about how many games my opponent play at same time.

I'm more afraid to play against players that I know are strong, regardless of their rating.
Questions about rankings on the ladder: 4/11/2013 14:56:23

Fizzer 
Level 58

Warzone Creator
Report
I understand that the more games you have on the ladder, generally the lower your average rating will be.

This is incorrect. I made this graph before, but I couldn't find it so I made it again. This data is accurate as of 4/11/2013 for the 1v1 ladder:



As you can see, there's no correlation between number of games played and rating. If anything, there's a positive correlation indicated by the linear trendline. This shows that, if anything, the more games you play the higher your rating will be (the opposite of your statement.)
Questions about rankings on the ladder: 4/11/2013 15:02:54

Qi 
Level 55
Report
fizzer once considered using something called real skill instead of the current system. with real skill, guys with more games were rated more fairly.
Questions about rankings on the ladder: 4/11/2013 15:04:53


Luxis • apex 
Level 51
Report
It's worth noticing that the spread lowers a lot also, so people with very high ranking are likely to end up getting a lower rating still.
Questions about rankings on the ladder: 4/11/2013 15:10:00


[WM] Gnuffone 
Level 60
Report
look above 2000 point (that is rating that we interest)

player that they take it, have less them 30 games :P
i don't take care of rating under 1850/1900...

Questions about rankings on the ladder: 4/11/2013 15:10:28


[WM] Gnuffone 
Level 60
Report
Questions about rankings on the ladder: 4/11/2013 15:29:19


professor dead piggy 
Level 59
Report
I only keep an eye on the top 4 or 5, and the answer is yes. If you want a higher rating, you need fewer unexpired games. Look at Niko, Zibik, ollie (who was delaying games), sze, gnuff, dunga, chris and watch as it happens to luxis. They get 15-20 games and peak, then on average the more games they complete the more their rating drops. How many number ones have you seen with more than 25 games unexpired? I dont think its happened as long as Ive been watching. Maybe sze? But most ladder players have 25 unexpired games. Look at where old number ones are now that they have more games, most drop out, but zibik, chris (losses from boot are expired now) and niko stuck around, and quickly dropped out of the top 10.

If fizzer wants to make graphs of individual players, and no of games vs rating then maybe he will see it.

I dont know if it works the same further down the ladder, I can see why it might not.
Questions about rankings on the ladder: 4/11/2013 15:35:29

Qi 
Level 55
Report
fizzer, great programming on wl. but bad logic with your graph. you ignore preferences, motivation, and the actual focus of the discussion, which determines how you interpret the graph

preferences: players who play more games on the ladder actually like 1v1 MEs. if you like something and do more of it, you get better. hence the expression "learning curve." hell, if you said your graph represents the learning curve of 1v1 ladder play in a rather basic way, i'd agree. but saying that and what you are trying to say are two different interpretations.

motivation: many top players join and then quit once they reach their goal. they have motivation to play as few games as possible to reach their goals. where is this shown in your graph?

focus: we are talking about inaccurate ratings in the top 20, not statistical modes. inaccurate here = inflated ratings bc your algorithm doesn't know any better until it is given more games. why do players keep bringing tthis up? bc we know better. we play these people and know who is better. the rating system's UPPER LIMIT is flawed and ONLY GOOD PLAYERS ARE ABLE TO EXPLOIT IT. so showing a graph including everyone means absolutely nothing when the conversation is about (1) incntives to (2) game a (3) flawed upper limit (4) by good players who have (5) completed less than 20 games.
Questions about rankings on the ladder: 4/11/2013 15:41:44


NoobSchool (AHoL) • apex 
Level 58
Report
http://warlight.net/LadderTeam?LadderTeamID=240

In response to Dead Piggy: The only player I have seen take number 1 with a large number of games was HHH. If you look at his graphs from 6/18/2012 to 8/17/12 he was number 1 for most of that time. If I'm not mistaken HHH has always played 5 games and achieved a rating that high with just a massive percent of wins. I believe 13CHRIS37 also did, maybe with not such a high number but more than 30.
Questions about rankings on the ladder: 4/11/2013 15:51:04


professor dead piggy 
Level 59
Report
I think chris had exactly 25, but I could be miscounting. Either way, his rating was higher when he had fewer games completed.
Questions about rankings on the ladder: 4/11/2013 15:57:35


NoobSchool (AHoL) • apex 
Level 58
Report
I won't argue that. Even looking at HHH's (what I believe is) incredible stats, he is obviously an extreme outlier. Players like him are extremely rare. Players who score extremely high ratings are much more common at lower game counts.
Questions about rankings on the ladder: 4/11/2013 16:24:03


[WM] Anonymous 
Level 52
Report
I am one of those who could have taken #1 having almost all the time 5 at the same time, and a ton of unexpired games. Gnuffone wanted to leave it so i'd take n1, but after 2 years of constant playing that would be so lame for me. No i dropped again when other players with 15-20 games unexpired joined like Luxis, PureMind and others.

Now i left and i'm thinking to wait 3 months as many other did, at least until something changes in the ranking.
Questions about rankings on the ladder: 4/11/2013 17:15:31

Qi 
Level 55
Report
if the incentives were not the current system (more games played is a disadvantage relative others in the top 10), i'm absolutely sure more of us would be playing the 1v1 ladder. why play if it's all about getting the most bang out of your first 20 games? after climaxing, do men keep making love?

adjust the algorithm and the ladder would have 5 to 10 more good players playing within a week. and since much of the 'community' side of the website revolves around the beloved 1v1 template, it's safe to say the forums would get a boost in activity.

who would rejoin if the ratings' upper limit were less inflated (in a staggered way) for games 15 to 25 and gave more incentive for players to play more games instead of less? i'd rejoin. i assume piggy and sze would. red probably would. and others would have less reason to quit.
Questions about rankings on the ladder: 4/11/2013 20:16:16


AquaHolic 
Level 55
Report
I think Zaeban also managed to get #1 (2100 rating) with a lot of games (at least 40)

http://warlight.net/LadderTeam?LadderTeamID=272

count it yourself, i might have a 10% error
Questions about rankings on the ladder: 4/12/2013 16:41:47


NoobSchool (AHoL) • apex 
Level 58
Report
Is there a way to make this more fair in the short run? Obviously the system works great with many games played and the only problem we have is when there are too few games.

Does anyone know of a way that this could be fixed? We can find two players who have reached #1 with more than 30 games and probably 15(?)+ people who have done it with less than 20 games.

I think if there was some way to fix this (besides raising the amount of games needed to be ranked because that would obviously deter people) there would be a lot more interest in the ladder in general.
Questions about rankings on the ladder: 4/12/2013 17:14:57


szeweningen 
Level 60
Report
Yes there is, let me explain why did we see that much fluctuation when it comes to #1.

First of all the ladder rating is always a performance rating over past x games, it does not reprecent global overall rating, but rather how well you can perform in a given time period. For example Carlsen managed to get performance ratings >2900 in some chess tournaments, but he never actually reached that rating himself.

By the law of large numbers we know that if X_{i} are some random variables (here let it be performance rating in game number i) then (X_{1}+...+X_{n})/n ---> E(X) when n gets bigger. That explains easily why a bigger statistical sample gives us a more accurate rating. It also shows how a small n can give you a huge rating which would not be adequate (for example suppose someone won only 1 game against piggy when he had his record breaking rating, he'd be rated on performance ~2400). Now n=15 is relatively good from a statistical point of view, so in general the number of games is not a problem by itself. Now we introduce the psychological factor which is magnified by game management. For example in chess no new game is finished untill the game before had ended. Here we can postpone our losses if we have a decent enough positional understanding when are we better and when are we worse. Now, getting to number one is a psychological goal, which in general is not connected at all with playing at your best, it means besting all the others currently participating in the ladder, which can be temporarily achieved by postponing your losses. One more remark, postponing losses is much more valuable for people with less games played, because it has a greater impact on win/loss ratio, which is most important (read on the algorithm for details).

So how to improve it? The solution is really very easy, the player should be ranked once he completes his FIRST 15 games. That way he will not be able to replace "older" losses with "newer" wins. It would be much less annoying than moving the game cutoff up...
Questions about rankings on the ladder: 4/12/2013 17:16:44


szeweningen 
Level 60
Report
Also I do not think that is a really big deal, people that postponed losses took 1st place for a few days at most, we can more/less say, that all the best players either did or at least had the opportunity to hold 1st place for much longer. I think JSA made a list like that a while ago.
Questions about rankings on the ladder: 4/12/2013 17:31:01


NoobSchool (AHoL) • apex 
Level 58
Report
Are you suggesting that losses never expire? I'm not sure I understand. For some players (such as myself) who have improved vastly since joining the ladder almost two years ago that would be detrimental. I feel like this would only help the newer players who join only when they have already become "good".
Questions about rankings on the ladder: 4/12/2013 17:43:34


NoobSchool (AHoL) • apex 
Level 58
Report
^^I understand what you were saying now. Play each of the first 15 games in order, one at a time. Not that they don't expire -.- I tend to make things up when I'm confused.

While that would work great if everyone played as fast as they could, for some people who play slow, take vacations, and draw out games, those first 15 could take months to finish, with the games expiring before they got ranked. You would have to average one game finished every 6 days to complete 15 before time is up. For someone who plays slow that could only be two turns of a 20+ turn game.
Questions about rankings on the ladder: 4/12/2013 18:29:00


Timinator • apex 
Level 64
Report
anyone encouraged enough to count the number of unexpired games i had wen i took #1? I'm too lazy to do, but i thought it was like 25-30.
Questions about rankings on the ladder: 4/12/2013 21:45:39

Qi 
Level 55
Report
Most practical solution:
- rank games in the order they start (sze explains this above)
- rate the first 20 games exactly like how the seasonal ladder works
- use True Skill instead of the current ELO

Seasonal ladder's wiki:

"The rankings are determined using the same ELO scale that the existing ladders use, with one small modification. Players who play fewer than 20 games (due to leaving the ladder early or joining late) get a penalty of 40 rating points per game." Thus, games 15 to 19 would be a provisional rating on the 1v1 ladder.

True Skill (how it works and how it influenced the ladder right before I became #1):

http://blog.warlight.net/index.php/2012/01/trueskill/

With these changes, (1) people with less than 20 games couldn't game the system, (2) players with more games would be rated better, (3) the incentives of the rating system would encourage us to play more games, and (4) more good players would participate in the ladder.
Questions about rankings on the ladder: 4/13/2013 16:58:36


hedja 
Level 58
Report
Well talking about someone hitting high with low game count. [REGL] nich http://warlight.net/LadderTeam?LadderTeamID=2129 is currently on a 2019 rating, if he beats Z-Dog and Red before Frank's game finishes he might have a shot of getting into first. He isn't gaming the system, but the system lets people do this. He is on 9 games and currently his rating would put him in 5th.
Questions about rankings on the ladder: 4/13/2013 20:27:22

JSA 
Level 59
Report
Yeah, and I don't know if you've looked at his games, but he does not deserve top 15 even in my opinion.

Of course, I don't think I deserve top 3 the way I have played on the ladder.
Questions about rankings on the ladder: 4/14/2013 00:23:24


[WM] Anonymous 
Level 52
Report
It's another Red, not me.
Questions about rankings on the ladder: 4/14/2013 15:11:45


hedja 
Level 58
Report
Yes, that is why i put Red, if not I would've put [WM] Red
He is lucky, if i am correct you are better than other Red
Questions about rankings on the ladder: 4/14/2013 18:53:07


[WM] Anonymous 
Level 52
Report
I was confused because i couldn't find the game against him :D
Questions about rankings on the ladder: 4/16/2013 16:39:53


ChrisCMU 
Level 60
Report
Here is an example how how a person can make a top ranking without many games:

http://warlight.net/LadderTeam?LadderTeamID=2129

I am not saying he is good or bad (haven't looked at games, never played against), just that a small sample size can really benefit you.

He/she has 1 top 10 win, no other quality wins. If he/she wins 1 more game, they'd be top 3 on ladder with only beating the 8th player on ladder.
Questions about rankings on the ladder: 4/16/2013 16:44:55


ChrisCMU 
Level 60
Report
Whoops, just saw this was posted a few above. Can't edit it though.
Questions about rankings on the ladder: 4/16/2013 17:29:35

The Duke of Ben 
Level 55
Report
A while back there was a guy called DEATH TACO who beat a really lackluster group of players (<1500 rating for the most part) and also beat HHH. He had over a 2000 rating when he finished his 15 games, then dropped off the ladder so no one could see a series of losses he took right after that. Before that particular run, he averaged something like a 1600 rating.

These particular examples seem to be more issues with winning every game you play. In the seasonal ladder Fizzer tightened the game matching some and increased the count to 20 games in order to reduce this problem. A similar fix might work in the 1v1 ladder. Even increasing the required game count to 20 would greatly reduce the chances of someone getting ranked without any losses.
Posts 1 - 30 of 40   1  2  Next >>