JellyBelly
Joined: Jul 08, 2009
|
Posted:
Oct 01, 2016 - 13:45 |
|
So, I just won a rookie game with Slann against Dark Elves, and my High Elf ranking has gone up from 'Star' to 'Super Star'. Why's that then?
I mean, it's not really a big deal, but seems a bit odd .. |
_________________ "Opinions are like arseholes, everybody's got them and they all stink." - The protagonist, Fallout 2
"Go for the eyes, Boo! Go for the eyes!!" |
|
DukeTyrion
Joined: Feb 18, 2004
|
  Posted:
Oct 01, 2016 - 13:48 |
|
I think all the rankings get re-evaluated at the end of each month.
I haven't played my wood elves for a while, but every now and then it drops to super-star and then a month or two later back to legend. Depends purely where everyone elses rankings happen to be when it gets recalculated. |
|
|
Wreckage
Joined: Aug 15, 2004
|
  Posted:
Oct 01, 2016 - 13:56 |
|
Hmm, interesting.. I had a couple overall surprise drops to Superstar. Figured it was some mechanic that triggered because I didn't play enough games that I just never ran into in the earlier days. |
|
|
JellyBelly
Joined: Jul 08, 2009
|
  Posted:
Oct 01, 2016 - 14:04 |
|
Yeah, that is interesting. I wonder if there is a help page somewhere that explains how it works?
Also, I thought both teams had to have played some min. # of games (3 or 4) for a game to affect ratings. Is that not the case anymore? |
_________________ "Opinions are like arseholes, everybody's got them and they all stink." - The protagonist, Fallout 2
"Go for the eyes, Boo! Go for the eyes!!" |
|
Wreckage
Joined: Aug 15, 2004
|
  Posted:
Oct 01, 2016 - 14:25 |
|
JellyBelly wrote: | Yeah, that is interesting. I wonder if there is a help page somewhere that explains how it works? |
You can just check the old mechanics. At its core its probably still very similar. But the exact method Christer didn't make public this time around since he didn't want people to meta the system.
I did a number of mathematical experiments to get a closer look at the formular. It's been a while ago so I don't remember exactly. But the results suggested that:
a) The basic CR formular works the same as it always has. You basically will gain between 2 CR and nothing per game, depending on the difference in that formular.
b) There are some huge modifiers applied post match. I attributed them to racial matchmaking. They can easily push the reward loss beyond 2 CR.
c) In the first few games a coach plays CR gain/loss is amplified.
d) Ranks displayed depend on overall percentages. Each rank makes up roughly 15% starting with Legend. Ie: Legend are the top 15% and so on.
e) There is a slide area in between ranks that prevents you from dropping/increasing all of a sudden. Basically the system is built to keep you stable on a rank even when you are technically flopping between two. So if you're increasing from Superstar it will keep you there until you're a good bit higher and if you're Legend it will keep you there until you're a good bit lower in CR.
f) Apparently, as we concluded now, at the end of every month (or more often?) the ranks get adjusted based on where the 15% statistically relevant values reside at a certain time. |
|
|
JellyBelly
Joined: Jul 08, 2009
|
  Posted:
Oct 01, 2016 - 16:26 |
|
Makes sense, Wreckage. Thanks for the explanation, helps to clarify |
_________________ "Opinions are like arseholes, everybody's got them and they all stink." - The protagonist, Fallout 2
"Go for the eyes, Boo! Go for the eyes!!" |
|
Jedi_Knight_Amoeba
Joined: Oct 14, 2004
|
  Posted:
Oct 01, 2016 - 16:38 |
|
|
Purplegoo
Joined: Mar 23, 2006
|
  Posted:
Oct 02, 2016 - 11:07 |
|
Jedi_Knight_Amoeba wrote: | Does CR matter? At all? |
Nothing in BB or on FUMBBL matters apart from the stuff you choose to care about. If CR isn't on your list, you're not alone. |
|
|
tussock
Joined: May 29, 2011
|
  Posted:
Oct 02, 2016 - 12:02 |
|
CR lets you know if your opponent wins a lot more than they lose against the average coach here, when playing Ranked and/or Box, or the other way around, or somewhere in the middle. |
_________________
|
|
Lorebass
Joined: Jun 25, 2010
|
tussock wrote: | CR lets you know if your opponent wins a lot more than they lose against the average coach here, when playing Ranked and/or Box, or the other way around, or somewhere in the middle. |
Not quite, I have a Super Star ranking yet my overall win /loss is below .500 (I have lost more than won). A lot affects the ratings including the coaches you play and the teams you both use. All this within a certain amount of time range.
I have also met high win% coaches with lower rankings overall. |
|
|
DarthPhysicist
Joined: Jun 14, 2015
|
  Posted:
Oct 02, 2016 - 13:42 |
|
Haven't eaten cherry pie in 33 years after vomiting it up as a child. Very traumatic experience. |
_________________ Using derivative humor since 2005.
|
|
sheepycollins
Joined: Sep 21, 2015
|
  Posted:
Oct 02, 2016 - 13:43 |
|
Lorebass wrote: | tussock wrote: | CR lets you know if your opponent wins a lot more than they lose against the average coach here, when playing Ranked and/or Box, or the other way around, or somewhere in the middle. |
Not quite, I have a Super Star ranking yet my overall win /loss is below .500 (I have lost more than won). A lot affects the ratings including the coaches you play and the teams you both use. All this within a certain amount of time range.
I have also met high win% coaches with lower rankings overall. |
I have noticed it goes up or down more the higher/lower the CR of your opponent so if you're always playing rookies then your CR will go up alot slower than if your playing stars.
Same as it goes down faster if you lose to a rookie than a legend.
Makes sense really can't call yourself great unless you can beat good players |
_________________ Things I like and things I hate? I don't feel like telling you that. My dreams for the future? Never really thought about that. As for my hobbies... I have lots of hobbies. |
|
Wreckage
Joined: Aug 15, 2004
|
  Posted:
Oct 02, 2016 - 17:14 |
|
sheepycollins wrote: |
I have noticed it goes up or down more the higher/lower the CR of your opponent so if you're always playing rookies then your CR will go up alot slower than if your playing stars. |
https://fumbbl.com/help:Ranking
Just a few words since I think those formulars are never quite self explanatory:
S-p provides us with a value between 1 and -1. It has got basically 3 modes which is S is 1, S is 0.5 and S is 0. Depending on win/tie/loss.
K is what causes the value to go up and down between 0 and max 2. Since it is a simple two times multiplier.
Now p gets substracted from S regardless of whether it is a win, tie or loss. p will never be negative and p itself ranges between 1 and 0.
The reason for that is that the only variable in p is the power to the second summand of the divisor. Everything else is fixed. Now powers can get like really really high (easily approach infinity). And it's 10^x so a power of 10 is gonna approach infinity real fast.
What happens as infinity approaches, is because it is part of the divisor, the entire equation approaches 0.
When that happens that means the involved coach whose CR is calculated would either get 1 point for a win, 0.5 for a tie or 0 for a loss since it is s-p with p approaching 0. Multiplied by 2 means win gives 2 CR, tie 1 and loss 0.
You can already tell from the outcome that this describes a case where the game is very unbalanced in the other coaches favor.
But why is it that a power of 10 only increases close to infinity when the match is very unbalanced and not already when there is the slightest difference in team strength?
Simply because the power term that holds CR gets divided by 40. The power part that holds TV gets divided by 700.
If CR diff is sub 40, subsequently TV diff sub 700, only then 10 is actually to the power of 1.
Mind you even if it's just 10 to the power of 1, the divisor would be 1+10= 11 and p = 1/11, which means p is exceedingly small even without any power.
If 10 was to the power of 2 it would be 1/101, which is already very close to 0 but as you can see that would actually mean a constellation like CR difference being 80 or something. Efficiently any game like that whether 1/11 or 1/101 is a NO RISK ALL REWARD scenario - rather you would need just 1 win to make up 11 of those losses. Which has due to the randomness of dice very good odds to succeed.
So the reality is that really really bad games would make it the easiest to meta CR.
(Mind you you should't play these games simply because CR isn't a good reason to pick games when you lose in the process. Because you kinda wanna keep other things in mind like stability of the team and such.)
But now that we have looked at the extreme case lets look a bit more general. The general case will be that those power values are below 1. So if you have a CR diff of lets say 4. It would be 10 to the power 4/40 = ( 10 to the power of 1/10 ).
What happens if you have a value to the power of a sub 1 value? A sub 1 value can effectively be read as a root. 10 to the power of 1/10 equates 10 to the root of 10. Ie the root is straight up switched divident and divisor (in the power).
So that means in the much more common scenarios you will have to deal with a root of 10. In the scenario where CR diff is 4 and 10 to the root of 10, the outcome would be already that the term is roughly 1.26.
Making p about 1/2.26.
But rather than see what that is in coach rank lets rather look at what CR would look like if the teams are completely evenly matched.
If CR and TV is identical, it would be 10 to the power of 0 since regardless of the division by 40, 700 respectively, when the difference between to values is 0 it doesn't matter what multipliers/divisors you apply.
ANY value to the power of 0 is 1. So in an identical matchup p would be 1/(1+1) = 1/2.
From these two examples you can already see that the usual TV diff will cause a p value in the area of around 1/2.
1/2 is a half and the effect on S is quite obvious. If you got a win you'll have 1-1/2, a tie 1/2-1/2 and if you got a loss you'll have 0-1/2.
All these values will be multiplied by the K value of 2.
Resulting in giving you 1 CR on a win. 0 CR on a tie and -1 CR on a loss for identically strong opponents.
So there is only one value left to explore;
Which is:
How does all of this even cover for the case that I am STRONGER?
When you are stronger, your TV diff will be negative. Now you're dealing with a 10 to power value that is negative and might be below 1. Geez, right?
Negative power values switch divident and divisor of the base.
But since our 10 power to stuff is stuck on the divident with a + 1 you can't actually just conveniently write everything on the top and have to go:
Instead of 1/(1+10^(-stuff)) you now got 1/(1+(1/10)^(stuff)).
Math thats hard to follow I know. An unusual and disconcerting sight.
Well, not really.
Well, the bottom line is as you potentiate 1/10 instead of 10 it grows exponentially towards 0 instead of infinity, conversely root values will not make it go towards infinity but just 1.
So you got basically a reverse case from before. If the TV is very close, you'll have a large root value, 1/10 to power of 0.00001... will max out at 1 resulting in giving you around p = 1/2 again and if TV is very different 1/10 ^1 = 1/10 will result in something like p= 1/(1+1/10) = 10/11. Which is close to 1. So now you're substracting p values that range between 0.5 and 1 and subsequently when s=0 substracting -1 can get you down to overall -1.
As you can see CR lost and gained pretty much mirror each other. |
Last edited by Wreckage on %b %02, %2016 - %21:%Oct; edited 2 times in total |
|
spelledaren
Joined: Mar 06, 2004
|
  Posted:
Oct 02, 2016 - 18:57 |
|
|
_________________ FUMBBL! |
|
xnoelx
Joined: Jun 05, 2012
|
  Posted:
Oct 02, 2016 - 21:21 |
|
That might be the most understandable thing you've ever posted Wreckage. Good work. |
_________________ Nerf Ball 2014 |
|
|