buuface
Joined: Apr 23, 2014
|
  Posted:
Oct 21, 2014 - 09:36 |
|
Hello sorry if im missing the obvious but is there a thread or page somewhere which explains how CR is calculated?
Is it purely to do with your win/draw/loss ratio or does it also depend on factors such as which teams you play against and the CR of your opponent.
Thanks |
|
|
Wreckage
Joined: Aug 15, 2004
|
  Posted:
Oct 21, 2014 - 09:44 |
|
buuface wrote: | Hello sorry if im missing the obvious but is there a thread or page somewhere which explains how CR is calculated?
Is it purely to do with your win/draw/loss ratio or does it also depend on factors such as which teams you play against and the CR of your opponent.
Thanks |
Pretty sure it depends on all of those. TV too.
CR of opponent is probably the most import aspect.
The exact formular in its current version isn't publicly available i believe. |
|
|
Rabe
Joined: Jun 06, 2009
|
  Posted:
Oct 21, 2014 - 10:19 |
|
|
JimmyFantastic
Joined: Feb 06, 2007
|
  Posted:
Oct 21, 2014 - 10:22 |
|
Team strength? Poor Nurgz! |
_________________ Pull down the veil - actively bad for the hobby! |
|
Garion
Joined: Aug 19, 2009
|
  Posted:
Oct 21, 2014 - 10:24 |
|
Rabe wrote: | https://fumbbl.com/help:Ranking
Help -> Advanced Site Use |
That was the formula used during lrb4, it may have changed a little since then. But its probably not too dissimilar if it has. |
_________________
|
|
Wreckage
Joined: Aug 15, 2004
|
  Posted:
Oct 21, 2014 - 10:40 |
|
Garion wrote: | Rabe wrote: | https://fumbbl.com/help:Ranking
Help -> Advanced Site Use |
That was the formula used during lrb4, it may have changed a little since then. But its probably not too dissimilar if it has. |
Well, we had 3, 4 complete overhauls. The last one and most significant ended with the introduction of the titles (legend, super star) etc...
The old CR system on the display allowed a much wider gap in CR. The top 10 was usually comprised of 180+ CR coaches. 170 then was more like 166 now and most coaches on 155 did outragous mistakes while now we can sometimes see people just below 150 fight like this game is their life |
|
|
Christer
Joined: Aug 02, 2003
|
The CR formula described in the help page is a bit aged, and the full details of the current system are not public. That being said, it's based on the same system.
I have a review and redesign of the system on my todo list (low prio), which will change things on a more fundamental level. Roughly speaking, it'd be a change from the ELO based system we use now to something based on a more modern ranking system (Glicko, Trueskill, or some variation thereof).
The hard part about ranking on FUMBBL is that the system needs to be able to deal with the teams you play being uneven in some form and therefore pure implementations of ranking systems don't quite work. |
|
|
Rabe
Joined: Jun 06, 2009
|
  Posted:
Oct 21, 2014 - 12:18 |
|
Ah, sorry. Just looked for it quickly as I remembered there was something in the help section.
Interesting information, Christer. Looking forward to any change that improves things (even further), no matter how far in the distance. |
_________________ .
|
|
koadah
Joined: Mar 30, 2005
|
  Posted:
Oct 21, 2014 - 12:25 |
|
|
Christer
Joined: Aug 02, 2003
|
That's an interesting question Koadah, and the short answer is that the CR system interests me and I like to tinker with improvements (much like the way FFB now uses the Fortuna RNG rather than the more than adequate Mersenne Twister implementation). The long answer is as follows:
The current CR system is based on the ELO rating system. It's a fairly old and widely used system for various games and tournaments (chess being the big one, and the one which it was developed for in the first place).
ELO is designed as a zero-sum system (meaning that players effectively trade rating and none is added or lost), and generally speaking, players will end up on a normal distribution centered on the initial ranking (150.0).
If you look at the coach rankings we actually have on the site now, you'll see that this isn't the case. The peak of rankings is at roughly CR 146, and there's a fat tail on the upper end. This reduction of average CR is called rating deflation and causes a bit of a problem. Basically, a new coach will start out at CR 150 (above average) when in reality, they are likely below average due to the fact that FUMBBL is a very good learning ground to improve your game. Thus, players on the site are generally speaking better than average and newcomers are therefore often starting with a losing streak.
Lowering the starting CR isn't a solution either as it will further push down the average over time.
In addition to this problem, ELO doesn't really deal with the average skill level of the participants changing over time (again, something that happens on FUMBBL).
On top of that you have an issue where a coach who is away for some time and returns, their rating will be unchanged and could take quite a while to level out at its current (real) value.
Glicko, which is a likely formula I'd switch to, effectively introduces another value to the ranking: Rating Deviation (RD) which in a sense works like a dynamic K value in ELO. It also does away with the zero-sum constraint, which allows new coaches (or coaches returning after some time of inactivity) to find a better approximation of their rating without messing with existing users' rating (as a direct dynamic K value would do in a strict ELO based system).
Interestingly enough, there is a Glicko variant (Glicko Boost) that deals with inequalities in the sides of games (in Chess, White has an advantage), which might be worth exploring. Thus, it's on my todo list |
|
|
buuface
Joined: Apr 23, 2014
|
  Posted:
Oct 21, 2014 - 14:20 |
|
Thanks for the reply Christer!
I couple basic questions for you from someone who is tragically unfamiliar with the systems you mentioned in your post;
Does CR always decrease if you lose a match? Even if the opponent is a much higher CR?
If as you mentioned, team ST is taken into account, does this essentially mean that low Strength teams like WE, UW etc will get bigger bonuses to thier CR for winning than Orcs? |
|
|
Christer
Joined: Aug 02, 2003
|
As I stated, the details of the ranking system are not public.
But in terms of the ranking system, team strength isn't related to the strength stat on players. It's a reference to an alternative way that FUMBBL used to estimate how good a team is (basically, a custom team rating/team value type system). It's not in use anymore. |
|
|
Verminardo
Joined: Sep 27, 2006
|
  Posted:
Oct 21, 2014 - 14:40 |
|
I think "Strength" was not meant to refer to "Strength stats of players" but rather "effectiveness in winning games", which would make Wood Elves actually one of the strongest teams around.
Edit: Ninja'd! |
|
|
koadah
Joined: Mar 30, 2005
|
  Posted:
Oct 21, 2014 - 14:43 |
|
Christer wrote: |
ELO is designed as a zero-sum system (meaning that players effectively trade rating and none is added or lost), and generally speaking, players will end up on a normal distribution centered on the initial ranking (150.0).
If you look at the coach rankings we actually have on the site now, you'll see that this isn't the case. The peak of rankings is at roughly CR 146, and there's a fat tail on the upper end. This reduction of average CR is called rating deflation and causes a bit of a problem. Basically, a new coach will start out at CR 150 (above average) when in reality, they are likely below average due to the fact that FUMBBL is a very good learning ground to improve your game. Thus, players on the site are generally speaking better than average and newcomers are therefore often starting with a losing streak.
|
Does that graph only include 'current' coaches? It feels as though we have "rating inflation".
I would have thought that many weak coaches quit and the average CR of people who actually play generally goes up.
Maybe it is the case that higher rated coaches tend to play more games so we're more likely to run into them.
Personally I like it as it is and don't see it a much of a problem.
I think coaches feel a sense of achievement when they "make it back to 150"
"Average" kinda "feels like" around 155.
I was using "team strength" in a general way. Assuming that the team that you use and play against is factored in somehow. |
_________________
O[L]C 2016 Swiss! - 19th June! ---- All Star Bowl XII - Teams of Stars - Sign up NOW! |
|
Christer
Joined: Aug 02, 2003
|
The graph is of all coaches. The fat tail behaviour on the high end is probably there because some new coaches lose a couple of games and decide this place isn't for them (thus injecting ranking to high CR coaches)
And as I said, a large part of me wanting to revise the system is because I think it's fun, and ELO does have some genuine issues that are addressed by more modern systems such as Glicko. |
|
|
|