Forum
Elo (Skill) Rankings for Season 5
Created 2nd September 2009 @ 00:56
Add A Reply Pages: 1 2 3 Next »
Hi all,
I’ve been reading about game ranking systems and decided to have a play with the season 5 data. Specifically to see what the outcome of using Elo’s rating system would give. So, I decided all teams in division 1 would start with a rating of 2200, and teams in division 2a all start with a rating of 2000. I didn’t bother with other divisions for this test.
Next, I did Elo rating based on the outcome of each map played. So, each teams rating is effected based on losing one map and the winning another – not as if the outcome of the fixture was actually a draw. This may be a potential flaw in my approach, I’m not sure.
I did ranking on every map played against a team that did not drop out of the season early, and have not entered results of a default win, because I don’t think this is any type of indication of skill.
Anyway, without further ado, here are the results!
1. Team Dignitas (league rating 2266, per-round rating 2130)
2. Crack Clan (league rating 2252, per-round rating 2195)
3. Team Coolermaster (league rating 2248, per-round rating 2414)
4. Team YoYoTech (league rating 2237, per-round rating 2218)
5. I Don’t Know? (league rating 2161, per-round rating 2288)
6. The Imperial (league rating 2155, per-round rating 2102)
7. coolclan (league rating 2123, per-round rating 2154)
8. Fakkel Brigade (league rating 2120, per-round rating 2186)
9. rockit (league rating 2106, per-round rating 2240)
10. dmg. (league rating 2025, per-round rating 1987)
11. eSrael (league rating 2025, per-round rating 2030)
12. jAgarna (league rating 2002, per-round rating 2051)
13. Dead Cows Strike Back (league rating 1965, per-round rating 1895)
14. Euphoria (league rating 1960, per-round rating 1895)
15. veAre.black (league rating 1932, per-round rating 1839)
16. The Hampshire Heavies (league rating 1823, per-round rating 1777)
So, some observations – the top 3 differs from the actual top 3 when the league finished. It would appear that the default wins caused some sort of alterations to the data, as to be expected.
Next – it’s interesting to see idk?’s skill – bringing them right up into 5th place. This is due to the incredible performance in the season, losing only 2 maps in total I believe?
What do you think? Interesting figures, or pointless statistic drivel? :)
Forgot to mention, per round rating is the elo rating based on entering a win or loss everytime a round finished. So 6 – 0 means the Elo rating of each time is adjusted 6 times. Sorting based on per-round rating we get:
1. Team Coolermaster (league rating 2248, per-round rating 2414)
2. I Don’t Know? (league rating 2161, per-round rating 2288)
3. rockit (league rating 2106, per-round rating 2240)
4. Team YoYoTech (league rating 2237, per-round rating 2218)
5. Crack Clan (league rating 2252, per-round rating 2195)
6. Fakkel Brigade (league rating 2120, per-round rating 2186)
7. coolclan (league rating 2123, per-round rating 2154)
8. Team Dignitas (league rating 2266, per-round rating 2130)
9. The Imperial (league rating 2155, per-round rating 2102)
10. jAgarna (league rating 2002, per-round rating 2051)
11. eSrael (league rating 2025, per-round rating 2030)
12. dmg. (league rating 2025, per-round rating 1987)
13. Dead Cows Strike Back (league rating 1965, per-round rating 1895)
14. Euphoria (league rating 1960, per-round rating 1895)
15. veAre.black (league rating 1932, per-round rating 1839)
16. The Hampshire Heavies (league rating 1823, per-round rating 1777)
I guess this is a measure of how much a team is likely to completely annihilate another team ;) Or as I like to call “stompability”
right on
Chess, meet TF2. TF2, meet chess. Interesting experiment!
ELO is used to assign ratings in like any game actually
Yes, a bit interesting.
Chess and TF2 actually have quite a bit in common (Excpet in TF2 individual skill can make a worse piece/player win against a better).
There’s a guy who has done this for football: http://www.eloratings.net/ It includes historical data, seems like an awful lot of work (as does your rating job btw, good work!). His ELO-rating takes in account the actual score of the match by weighting the K-factor. This way you could take in account that a match consists of two (or three) maps.
jAgarna: not over 9000.
is it posible for a div2 team to exceed 2200 rating? cus then they could be rated higher than a team they’re worse than. through easier matches :)
Cycles is a winner, he also has too much free time.
One of those statements is true. In all sincerity hes been working on this for the past few days now and he has done an excellent job, could definately help in the seeding of already known teams in cups/ladders/leagues and what have you (in addition to being interesting)!
ELO is used to assign ratings in like any game actually
Indeed, I first heard about them when I was reading about how the QuakeLive skill system works, which has roots in Elo’s system.
There’s a guy who has done this for football: [Link] It includes historical data, seems like an awful lot of work (as does your rating job btw, good work!). His ELO-rating takes in account the actual score of the match by weighting the K-factor. This way you could take in account that a match consists of two (or three) maps.
Nice, I’d seen the site, but I didn’t know about how exactly it was implemented. I’d like to bring the K-factor into this as well somehow, but I need to give that a bit more thought. The trickiest thing is deciding the initial ratings though.
is it posible for a div2 team to exceed 2200 rating? cus then they could be rated higher than a team they’re worse than. through easier matches :)
For a division 2 team to really get over 2200, they would need to play division 1 players. Beating someone of a higher skill gives you a lot more points than beating someone below your skill. The k-factor controls this, to a degree, meaning that if you played a complete noob, you would gain very few points, but if they beat you they would score a massive amount (naturally).
could definately help in the seeding of already known teams in cups/ladders/leagues and what have you
What I really want is to get this into the ladder system, if ETF2L gets one, which should yes – give indications about where to seed people. In theory, creating divisions should be a case of splitting the ladder into equal sized sections, this way :)
is it posible for a div2 team to exceed 2200 rating? cus then they could be rated higher than a team they’re worse than. through easier matches :)
For a division 2 team to really get over 2200, they would need to play division 1 players. Beating someone of a higher skill gives you a lot more points than beating someone below your skill. The k-factor controls this, to a degree, meaning that if you played a complete noob, you would gain very few points, but if they beat you they would score a massive amount (naturally).
nice, seems like a perfect system to me then :) great find :)
Add A Reply Pages: 1 2 3 Next »
content rss